[PATCH v2 0/51] updated patches for adding QIB driver

[PATCH v2 0/51] updated patches for adding QIB driver

[Ralph Campbell]

The following emails are in response to Rolands comments.
Since the patches are for whole files, its hard to see what
has changed but I describe it below.

A reminder that these depend on the patch I sent earlier which
introduces ib_sysfs_create_port_files() into the ib_core.

---

Changed the Makefile line-by-line to multi-line list of .o files and
removed the -D flags.

Removed qib_trace.[ch]. Debug messages use printk similar to ib_ipath.

Fixed compiles when CONFIG_PCI_MSI is not defined.

Removed #ifdefs and backport code for AER.

Reduced the size of the chip register definition files.

Changed global function names which conflict with ib_ipath to
be prefixed with qib_ (requires ib_ipath and ib_qib to be compiled
into the kernel instead of as separate modules).
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 01/51] IB/qib: Add Kconfig

[Ralph Campbell]

This creates the Kconfig file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/Kconfig |   11 +++++++++++
 1 files changed, 11 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/Kconfig

diff --git a/drivers/infiniband/hw/qib/Kconfig b/drivers/infiniband/hw/qib/Kconfig
new file mode 100644
index 0000000..1acfb54
--- /dev/null
+++ b/drivers/infiniband/hw/qib/Kconfig
@@ -0,0 +1,11 @@
+config INFINIBAND_QIB
+	tristate "QLogic PCIe HCA support"
+	depends on 64BIT && NET
+	---help---
+	This is a driver for QLogic PCIe QLE IB host channel adapters,
+	including InfiniBand verbs support.  This driver allows these
+	devices to be used with both kernel upper level protocols such
+	as IP-over-InfiniBand as well as with userspace applications
+	(in conjunction with InfiniBand userspace access).
+	This driver does not support the QLogic HyperTransport card
+	(model QHT7140).

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 01/51] IB/qib: Add Kconfig

[Roland Dreier]

Ralph, I'd like to look at merging this for 2.6.35.  Could you send your
latest code?

 - R.
-- 
Roland Dreier <rolandd-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> || For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 01/51] IB/qib: Add Kconfig

[Ralph Campbell]

On Wed, 2010-05-05 at 15:21 -0700, Roland Dreier wrote:
> Ralph, I'd like to look at merging this for 2.6.35.  Could you send your
> latest code?
> 
>  - R.

I haven't switched all the debug printks to event tracing.
I guess I will need to strip those out to get something
ready anytime soon.

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 01/51] IB/qib: Add Kconfig

[Roland Dreier]

 > I haven't switched all the debug printks to event tracing.
 > I guess I will need to strip those out to get something
 > ready anytime soon.

Sounds like a good plan.  I think we're better off adding something as
small as possible so that we can take a good look at each part as it
goes by.
-- 
Roland Dreier <rolandd-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> || For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 02/51] IB/qib: Add Makefile

[Ralph Campbell]

This creates the Makefile file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/Makefile |   15 +++++++++++++++
 1 files changed, 15 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/Makefile

diff --git a/drivers/infiniband/hw/qib/Makefile b/drivers/infiniband/hw/qib/Makefile
new file mode 100644
index 0000000..c6515a1
--- /dev/null
+++ b/drivers/infiniband/hw/qib/Makefile
@@ -0,0 +1,15 @@
+obj-$(CONFIG_INFINIBAND_QIB) += ib_qib.o
+
+ib_qib-y := qib_cq.o qib_diag.o qib_dma.o qib_driver.o qib_eeprom.o \
+	qib_file_ops.o qib_fs.o qib_init.o qib_intr.o qib_keys.o \
+	qib_mad.o qib_mmap.o qib_mr.o qib_pcie.o qib_pio_copy.o \
+	qib_qp.o qib_qsfp.o qib_rc.o qib_ruc.o qib_sdma.o qib_srq.o \
+	qib_sysfs.o qib_twsi.o qib_tx.o qib_uc.o qib_ud.o \
+	qib_user_pages.o qib_user_sdma.o qib_verbs_mcast.o qib_iba7220.o \
+	qib_sd7220.o qib_sd7220_img.o qib_iba7322.o qib_verbs.o
+
+# 6120 has no fallback if no MSI interrupts, others can do INTx
+ib_qib-$(CONFIG_PCI_MSI) += qib_iba6120.o
+
+ib_qib-$(CONFIG_X86_64) += qib_wc_x86_64.o
+ib_qib-$(CONFIG_PPC64) += qib_wc_ppc64.o

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 03/51] IB/qib: Add qib.h

[Ralph Campbell]

This creates the qib.h file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib.h | 1437 +++++++++++++++++++++++++++++++++++++++
 1 files changed, 1437 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib.h

diff --git a/drivers/infiniband/hw/qib/qib.h b/drivers/infiniband/hw/qib/qib.h
new file mode 100644
index 0000000..d715eeb
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib.h
@@ -0,0 +1,1437 @@
+#ifndef _QIB_KERNEL_H
+#define _QIB_KERNEL_H
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This header file is the base header file for qlogic_ib kernel code
+ * qib_user.h serves a similar purpose for user code.
+ */
+
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+#include <linux/dma-mapping.h>
+#include <linux/mutex.h>
+#include <linux/list.h>
+#include <linux/scatterlist.h>
+#include <linux/io.h>
+#include <linux/fs.h>
+#include <linux/completion.h>
+#include <linux/kref.h>
+#include <linux/sched.h>
+
+#include "qib_common.h"
+#include "qib_debug.h"
+#include "qib_verbs.h"
+
+/* only s/w major version of QLogic_IB we can handle */
+#define QIB_CHIP_VERS_MAJ 2U
+
+/* don't care about this except printing */
+#define QIB_CHIP_VERS_MIN 0U
+
+/* The Organization Unique Identifier (Mfg code), and its position in GUID */
+#define QIB_OUI 0x001175
+#define QIB_OUI_LSB 40
+
+/*
+ * per driver stats, either not device nor port-specific, or
+ * summed over all of the devices and ports.
+ * They are described by name via ipathfs filesystem, so layout
+ * and number of elements can change without breaking compatibility.
+ * If members are added or deleted qib_statnames[] in qib_fs.c must
+ * change to match.
+ */
+struct qlogic_ib_stats {
+	__u64 sps_ints; /* number of interrupts handled */
+	__u64 sps_errints; /* number of error interrupts */
+	__u64 sps_txerrs; /* tx-related packet errors */
+	__u64 sps_rcverrs; /* non-crc rcv packet errors */
+	__u64 sps_hwerrs; /* hardware errors reported (parity, etc.) */
+	__u64 sps_nopiobufs; /* no pio bufs avail from kernel */
+	__u64 sps_ctxts; /* number of contexts currently open */
+	__u64 sps_lenerrs; /* number of kernel packets where RHF != LRH len */
+	__u64 sps_buffull;
+	__u64 sps_hdrfull;
+};
+
+extern struct qlogic_ib_stats qib_stats;
+extern struct pci_error_handlers qib_pci_err_handler;
+extern struct pci_driver qib_driver;
+
+#define QIB_CHIP_SWVERSION QIB_CHIP_VERS_MAJ
+/*
+ * First-cut critierion for "device is active" is
+ * two thousand dwords combined Tx, Rx traffic per
+ * 5-second interval. SMA packets are 64 dwords,
+ * and occur "a few per second", presumably each way.
+ */
+#define QIB_TRAFFIC_ACTIVE_THRESHOLD (2000)
+
+/*
+ * Struct used to indicate which errors are logged in each of the
+ * error-counters that are logged to EEPROM. A counter is incremented
+ * _once_ (saturating at 255) for each event with any bits set in
+ * the error or hwerror register masks below.
+ */
+#define QIB_EEP_LOG_CNT (4)
+struct qib_eep_log_mask {
+	u64 errs_to_log;
+	u64 hwerrs_to_log;
+};
+
+/*
+ * Below contains all data related to a single context (formerly called port).
+ */
+struct qib_ctxtdata {
+	void **rcvegrbuf;
+	dma_addr_t *rcvegrbuf_phys;
+	/* rcvhdrq base, needs mmap before useful */
+	void *rcvhdrq;
+	/* kernel virtual address where hdrqtail is updated */
+	void *rcvhdrtail_kvaddr;
+	/*
+	 * temp buffer for expected send setup, allocated at open, instead
+	 * of each setup call
+	 */
+	void *tid_pg_list;
+	/*
+	 * Shared page for kernel to signal user processes that send buffers
+	 * need disarming.  The process should call QIB_CMD_DISARM_BUFS
+	 * or QIB_CMD_ACK_EVENT with IPATH_EVENT_DISARM_BUFS set.
+	 */
+	unsigned long *user_event_mask;
+	/* when waiting for rcv or pioavail */
+	wait_queue_head_t wait;
+	/*
+	 * rcvegr bufs base, physical, must fit
+	 * in 44 bits so 32 bit programs mmap64 44 bit works)
+	 */
+	dma_addr_t rcvegr_phys;
+	/* mmap of hdrq, must fit in 44 bits */
+	dma_addr_t rcvhdrq_phys;
+	dma_addr_t rcvhdrqtailaddr_phys;
+
+	/*
+	 * number of opens (including slave sub-contexts) on this instance
+	 * (ignoring forks, dup, etc. for now)
+	 */
+	int cnt;
+	/*
+	 * how much space to leave at start of eager TID entries for
+	 * protocol use, on each TID
+	 */
+	/* instead of calculating it */
+	unsigned ctxt;
+	/* non-zero if ctxt is being shared. */
+	u16 subctxt_cnt;
+	/* non-zero if ctxt is being shared. */
+	u16 subctxt_id;
+	/* number of eager TID entries. */
+	u16 rcvegrcnt;
+	/* index of first eager TID entry. */
+	u16 rcvegr_tid_base;
+	/* number of pio bufs for this ctxt (all procs, if shared) */
+	u32 piocnt;
+	/* first pio buffer for this ctxt */
+	u32 pio_base;
+	/* chip offset of PIO buffers for this ctxt */
+	u32 piobufs;
+	/* how many alloc_pages() chunks in rcvegrbuf_pages */
+	u32 rcvegrbuf_chunks;
+	/* how many egrbufs per chunk */
+	u32 rcvegrbufs_perchunk;
+	/* order for rcvegrbuf_pages */
+	size_t rcvegrbuf_size;
+	/* rcvhdrq size (for freeing) */
+	size_t rcvhdrq_size;
+	/* per-context flags for fileops/intr communication */
+	unsigned long flag;
+	/* next expected TID to check when looking for free */
+	u32 tidcursor;
+	/* WAIT_RCV that timed out, no interrupt */
+	u32 rcvwait_to;
+	/* WAIT_PIO that timed out, no interrupt */
+	u32 piowait_to;
+	/* WAIT_RCV already happened, no wait */
+	u32 rcvnowait;
+	/* WAIT_PIO already happened, no wait */
+	u32 pionowait;
+	/* total number of polled urgent packets */
+	u32 urgent;
+	/* saved total number of polled urgent packets for poll edge trigger */
+	u32 urgent_poll;
+	/* pid of process using this ctxt */
+	pid_t pid;
+	pid_t subpid[QLOGIC_IB_MAX_SUBCTXT];
+	/* same size as task_struct .comm[], command that opened context */
+	char comm[16];
+	/* pkeys set by this use of this ctxt */
+	u16 pkeys[4];
+	/* so file ops can get at unit */
+	struct qib_devdata *dd;
+	/* so funcs that need physical port can get it easily */
+	struct qib_pportdata *ppd;
+	/* A page of memory for rcvhdrhead, rcvegrhead, rcvegrtail * N */
+	void *subctxt_uregbase;
+	/* An array of pages for the eager receive buffers * N */
+	void *subctxt_rcvegrbuf;
+	/* An array of pages for the eager header queue entries * N */
+	void *subctxt_rcvhdr_base;
+	/* The version of the library which opened this ctxt */
+	u32 userversion;
+	/* Bitmask of active slaves */
+	u32 active_slaves;
+	/* Type of packets or conditions we want to poll for */
+	u16 poll_type;
+	/* receive packet sequence counter */
+	u8 seq_cnt;
+	u8 redirect_seq_cnt;
+	/* ctxt rcvhdrq head offset */
+	u32 head;
+	u32 pkt_count;
+	/* QPs waiting for context processing */
+	struct list_head qp_wait_list;
+};
+
+struct qib_sge_state;
+
+struct qib_sdma_txreq {
+	int                 flags;
+	int                 sg_count;
+	dma_addr_t          addr;
+	void              (*callback)(struct qib_sdma_txreq *, int);
+	u16                 start_idx;  /* sdma private */
+	u16                 next_descq_idx;  /* sdma private */
+	struct list_head    list;       /* sdma private */
+};
+
+struct qib_sdma_desc {
+	__le64 qw[2];
+};
+
+struct qib_verbs_txreq {
+	struct qib_sdma_txreq   txreq;
+	struct qib_qp           *qp;
+	struct qib_swqe         *wqe;
+	u32                     dwords;
+	u16                     hdr_dwords;
+	u16                     hdr_inx;
+	struct qib_pio_header	*align_buf;
+	struct qib_mregion	*mr;
+	struct qib_sge_state    *ss;
+};
+
+#define QIB_SDMA_TXREQ_F_USELARGEBUF  0x1
+#define QIB_SDMA_TXREQ_F_HEADTOHOST   0x2
+#define QIB_SDMA_TXREQ_F_INTREQ       0x4
+#define QIB_SDMA_TXREQ_F_FREEBUF      0x8
+#define QIB_SDMA_TXREQ_F_FREEDESC     0x10
+
+#define QIB_SDMA_TXREQ_S_OK        0
+#define QIB_SDMA_TXREQ_S_SENDERROR 1
+#define QIB_SDMA_TXREQ_S_ABORTED   2
+#define QIB_SDMA_TXREQ_S_SHUTDOWN  3
+
+/*
+ * Get/Set IB link-level config parameters for f_get/set_ib_cfg()
+ * Mostly for MADs that set or query link parameters, also ipath
+ * config interfaces
+ */
+#define QIB_IB_CFG_LIDLMC 0 /* LID (LS16b) and Mask (MS16b) */
+#define QIB_IB_CFG_LWID_ENB 2 /* allowed Link-width */
+#define QIB_IB_CFG_LWID 3 /* currently active Link-width */
+#define QIB_IB_CFG_SPD_ENB 4 /* allowed Link speeds */
+#define QIB_IB_CFG_SPD 5 /* current Link spd */
+#define QIB_IB_CFG_RXPOL_ENB 6 /* Auto-RX-polarity enable */
+#define QIB_IB_CFG_LREV_ENB 7 /* Auto-Lane-reversal enable */
+#define QIB_IB_CFG_LINKLATENCY 8 /* Link Latency (IB1.2 only) */
+#define QIB_IB_CFG_HRTBT 9 /* IB heartbeat off/enable/auto; DDR/QDR only */
+#define QIB_IB_CFG_OP_VLS 10 /* operational VLs */
+#define QIB_IB_CFG_VL_HIGH_CAP 11 /* num of VL high priority weights */
+#define QIB_IB_CFG_VL_LOW_CAP 12 /* num of VL low priority weights */
+#define QIB_IB_CFG_OVERRUN_THRESH 13 /* IB overrun threshold */
+#define QIB_IB_CFG_PHYERR_THRESH 14 /* IB PHY error threshold */
+#define QIB_IB_CFG_LINKDEFAULT 15 /* IB link default (sleep/poll) */
+#define QIB_IB_CFG_PKEYS 16 /* update partition keys */
+#define QIB_IB_CFG_MTU 17 /* update MTU in IBC */
+#define QIB_IB_CFG_LSTATE 18 /* update linkcmd and linkinitcmd in IBC */
+#define QIB_IB_CFG_VL_HIGH_LIMIT 19
+#define QIB_IB_CFG_PMA_TICKS 20 /* PMA sample tick resolution */
+#define QIB_IB_CFG_PORT 21 /* switch port we are connected to */
+
+/*
+ * for CFG_LSTATE: LINKCMD in upper 16 bits, LINKINITCMD in lower 16
+ * IB_LINKINITCMD_POLL and SLEEP are also used as set/get values for
+ * QIB_IB_CFG_LINKDEFAULT cmd
+ */
+#define   IB_LINKCMD_DOWN   (0 << 16)
+#define   IB_LINKCMD_ARMED  (1 << 16)
+#define   IB_LINKCMD_ACTIVE (2 << 16)
+#define   IB_LINKINITCMD_NOP     0
+#define   IB_LINKINITCMD_POLL    1
+#define   IB_LINKINITCMD_SLEEP   2
+#define   IB_LINKINITCMD_DISABLE 3
+
+/*
+ * valid states passed to qib_set_linkstate() user call
+ */
+#define QIB_IB_LINKDOWN         0
+#define QIB_IB_LINKARM          1
+#define QIB_IB_LINKACTIVE       2
+#define QIB_IB_LINKDOWN_ONLY    3
+#define QIB_IB_LINKDOWN_SLEEP   4
+#define QIB_IB_LINKDOWN_DISABLE 5
+
+/*
+ * These 7 values (SDR, DDR, and QDR may be ORed for auto-speed
+ * negotiation) are used for the 3rd argument to path_f_set_ib_cfg
+ * with cmd QIB_IB_CFG_SPD_ENB, by direct calls or via sysfs.  They
+ * are also the the possible values for qib_link_speed_enabled and active
+ * The values were chosen to match values used within the IB spec.
+ */
+#define QIB_IB_SDR 1
+#define QIB_IB_DDR 2
+#define QIB_IB_QDR 4
+
+#define QIB_DEFAULT_MTU 4096
+
+/*
+ * Possible IB config parameters for f_get/set_ib_table()
+ */
+#define QIB_IB_TBL_VL_HIGH_ARB 1 /* Get/set VL high priority weights */
+#define QIB_IB_TBL_VL_LOW_ARB 2 /* Get/set VL low priority weights */
+
+/*
+ * Possible "operations" for f_rcvctrl(ppd, op, ctxt)
+ * these are bits so they can be combined, e.g.
+ * QIB_RCVCTRL_INTRAVAIL_ENB | QIB_RCVCTRL_CTXT_ENB
+ */
+#define QIB_RCVCTRL_TAILUPD_ENB 0x01
+#define QIB_RCVCTRL_TAILUPD_DIS 0x02
+#define QIB_RCVCTRL_CTXT_ENB 0x04
+#define QIB_RCVCTRL_CTXT_DIS 0x08
+#define QIB_RCVCTRL_INTRAVAIL_ENB 0x10
+#define QIB_RCVCTRL_INTRAVAIL_DIS 0x20
+#define QIB_RCVCTRL_PKEY_ENB 0x40  /* Note, default is enabled */
+#define QIB_RCVCTRL_PKEY_DIS 0x80
+#define QIB_RCVCTRL_BP_ENB 0x0100
+#define QIB_RCVCTRL_BP_DIS 0x0200
+#define QIB_RCVCTRL_TIDFLOW_ENB 0x0400
+#define QIB_RCVCTRL_TIDFLOW_DIS 0x0800
+
+/*
+ * Possible "operations" for f_sendctrl(ppd, op, var)
+ * these are bits so they can be combined, e.g.
+ * QIB_SENDCTRL_BUFAVAIL_ENB | QIB_SENDCTRL_ENB
+ * Some operations (e.g. DISARM, ABORT) are known to
+ * be "one-shot", so do not modify shadow.
+ */
+#define QIB_SENDCTRL_DISARM       (0x1000)
+#define QIB_SENDCTRL_DISARM_BUF(bufn) ((bufn) | QIB_SENDCTRL_DISARM)
+	/* available (0x2000) */
+#define QIB_SENDCTRL_AVAIL_DIS    (0x4000)
+#define QIB_SENDCTRL_AVAIL_ENB    (0x8000)
+#define QIB_SENDCTRL_AVAIL_BLIP  (0x10000)
+#define QIB_SENDCTRL_SEND_DIS    (0x20000)
+#define QIB_SENDCTRL_SEND_ENB    (0x40000)
+#define QIB_SENDCTRL_FLUSH       (0x80000)
+#define QIB_SENDCTRL_CLEAR      (0x100000)
+#define QIB_SENDCTRL_DISARM_ALL (0x200000)
+
+/*
+ * These are the generic indices for requesting per-port
+ * counter values via the f_portcntr function.  They
+ * are always returned as 64 bit values, although most
+ * are 32 bit counters.
+ */
+/* send-related counters */
+#define QIBPORTCNTR_PKTSEND         0U
+#define QIBPORTCNTR_WORDSEND        1U
+#define QIBPORTCNTR_PSXMITDATA      2U
+#define QIBPORTCNTR_PSXMITPKTS      3U
+#define QIBPORTCNTR_PSXMITWAIT      4U
+#define QIBPORTCNTR_SENDSTALL       5U
+/* receive-related counters */
+#define QIBPORTCNTR_PKTRCV          6U
+#define QIBPORTCNTR_PSRCVDATA       7U
+#define QIBPORTCNTR_PSRCVPKTS       8U
+#define QIBPORTCNTR_RCVEBP          9U
+#define QIBPORTCNTR_RCVOVFL         10U
+#define QIBPORTCNTR_WORDRCV         11U
+/* IB link related error counters */
+#define QIBPORTCNTR_RXLOCALPHYERR   12U
+#define QIBPORTCNTR_RXVLERR         13U
+#define QIBPORTCNTR_ERRICRC         14U
+#define QIBPORTCNTR_ERRVCRC         15U
+#define QIBPORTCNTR_ERRLPCRC        16U
+#define QIBPORTCNTR_BADFORMAT       17U
+#define QIBPORTCNTR_ERR_RLEN        18U
+#define QIBPORTCNTR_IBSYMBOLERR     19U
+#define QIBPORTCNTR_INVALIDRLEN     20U
+#define QIBPORTCNTR_UNSUPVL         21U
+#define QIBPORTCNTR_EXCESSBUFOVFL   22U
+#define QIBPORTCNTR_ERRLINK         23U
+#define QIBPORTCNTR_IBLINKDOWN      24U
+#define QIBPORTCNTR_IBLINKERRRECOV  25U
+#define QIBPORTCNTR_LLI             26U
+/* other error counters */
+#define QIBPORTCNTR_RXDROPPKT       27U
+#define QIBPORTCNTR_VL15PKTDROP     28U
+#define QIBPORTCNTR_ERRPKEY         29U
+#define QIBPORTCNTR_KHDROVFL        30U
+/* sampling counters (these are actually control registers) */
+#define QIBPORTCNTR_PSINTERVAL      31U
+#define QIBPORTCNTR_PSSTART         32U
+#define QIBPORTCNTR_PSSTAT          33U
+
+/* how often we check for packet activity for "power on hours (in seconds) */
+#define ACTIVITY_TIMER 5
+
+/* Below is an opaque struct. Each chip (device) can maintain
+ * private data needed for its operation, but not germane to the
+ * rest of the driver.  For convenience, we define another that
+ * is chip-specific, per-port
+ */
+struct qib_chip_specific;
+struct qib_chipport_specific;
+
+enum qib_sdma_states {
+	qib_sdma_state_s00_hw_down,
+	qib_sdma_state_s10_hw_start_up_wait,
+	qib_sdma_state_s20_idle,
+	qib_sdma_state_s30_sw_clean_up_wait,
+	qib_sdma_state_s40_hw_clean_up_wait,
+	qib_sdma_state_s50_hw_halt_wait,
+	qib_sdma_state_s99_running,
+};
+
+enum qib_sdma_events {
+	qib_sdma_event_e00_go_hw_down,
+	qib_sdma_event_e10_go_hw_start,
+	qib_sdma_event_e20_hw_started,
+	qib_sdma_event_e30_go_running,
+	qib_sdma_event_e40_sw_cleaned,
+	qib_sdma_event_e50_hw_cleaned,
+	qib_sdma_event_e60_hw_halted,
+	qib_sdma_event_e70_go_idle,
+	qib_sdma_event_e7220_err_halted,
+	qib_sdma_event_e7322_err_halted,
+	qib_sdma_event_e90_timer_tick,
+};
+
+extern char *qib_sdma_state_names[];
+extern char *qib_sdma_event_names[];
+
+struct sdma_set_state_action {
+	unsigned op_enable:1;
+	unsigned op_intenable:1;
+	unsigned op_halt:1;
+	unsigned op_drain:1;
+	unsigned go_s99_running_tofalse:1;
+	unsigned go_s99_running_totrue:1;
+};
+
+struct qib_sdma_state {
+	struct kref          kref;
+	struct completion    comp;
+	enum qib_sdma_states current_state;
+	struct sdma_set_state_action *set_state_action;
+	unsigned             current_op;
+	unsigned             go_s99_running;
+	unsigned             first_sendbuf;
+	unsigned             last_sendbuf; /* really last +1 */
+	/* debugging/devel */
+	enum qib_sdma_states previous_state;
+	unsigned             previous_op;
+	enum qib_sdma_events last_event;
+};
+
+/*
+ * The structure below encapsulates data relevant to a physical IB Port.
+ * Current chips support only one such port, but the separation
+ * clarifies things a bit. Note that to conform to IB conventions,
+ * port-numbers are one-based. The first or only port is port1.
+ */
+struct qib_pportdata {
+	struct qib_ibport ibport_data;
+
+	struct qib_devdata *dd;
+	struct qib_chippport_specific *cpspec; /* chip-specific per-port */
+	struct kobject pport_kobj;
+	struct kobject sl2vl_kobj;
+
+	/* GUID for this interface, in network order */
+	__be64 guid;
+
+	/* QIB_POLL, etc. link-state specific flags, per port */
+	u32 lflags;
+	/* qib_lflags driver is waiting for */
+	u32 state_wanted;
+	spinlock_t lflags_lock;
+	/* number of (port-specific) interrupts for this port -- saturates... */
+	u32 int_counter;
+
+	/* ref count for each pkey */
+	atomic_t pkeyrefs[4];
+
+	/*
+	 * this address is mapped readonly into user processes so they can
+	 * get status cheaply, whenever they want.  One qword of status per port
+	 */
+	u64 *statusp;
+
+	/* SendDMA related entries */
+	spinlock_t            sdma_lock;
+	struct qib_sdma_state sdma_state;
+	unsigned long         sdma_buf_jiffies;
+	struct qib_sdma_desc *sdma_descq;
+	u64                   sdma_descq_added;
+	u64                   sdma_descq_removed;
+	u16                   sdma_descq_cnt;
+	u16                   sdma_descq_tail;
+	u16                   sdma_descq_head;
+	u16                   sdma_next_intr;
+	u16                   sdma_reset_wait;
+	u8                    sdma_generation;
+	struct tasklet_struct sdma_sw_clean_up_task;
+	struct list_head      sdma_activelist;
+
+	dma_addr_t       sdma_descq_phys;
+	volatile __le64 *sdma_head_dma; /* DMA'ed by chip */
+	dma_addr_t       sdma_head_phys;
+
+	wait_queue_head_t state_wait; /* for state_wanted */
+
+	/* HoL blocking for SMP replies */
+	unsigned          hol_state;
+	struct timer_list hol_timer;
+
+	/*
+	 * Shadow copies of registers; size indicates read access size.
+	 * Most of them are readonly, but some are write-only register,
+	 * where we manipulate the bits in the shadow copy, and then write
+	 * the shadow copy to qlogic_ib.
+	 *
+	 * We deliberately make most of these 32 bits, since they have
+	 * restricted range.  For any that we read, we won't to generate 32
+	 * bit accesses, since Opteron will generate 2 separate 32 bit HT
+	 * transactions for a 64 bit read, and we want to avoid unnecessary
+	 * bus transactions.
+	 */
+
+	/* This is the 64 bit group */
+	/* last ibcstatus.  opaque outside chip-specific code */
+	u64 lastibcstat;
+
+	/* these are the "32 bit" regs */
+
+	/*
+	 * the following two are 32-bit bitmasks, but {test,clear,set}_bit
+	 * all expect bit fields to be "unsigned long"
+	 */
+	unsigned long p_rcvctrl; /* shadow per-port rcvctrl */
+	unsigned long p_sendctrl; /* shadow per-port sendctrl */
+
+	u32 ibmtu; /* The MTU programmed for this unit */
+	/*
+	 * Current max size IB packet (in bytes) including IB headers, that
+	 * we can send. Changes when ibmtu changes.
+	 */
+	u32 ibmaxlen;
+	/*
+	 * ibmaxlen at init time, limited by chip and by receive buffer
+	 * size.  Not changed after init.
+	 */
+	u32 init_ibmaxlen;
+	/* LID programmed for this instance */
+	u16 lid;
+	/* list of pkeys programmed; 0 if not set */
+	u16 pkeys[4];
+	/* LID mask control */
+	u8 lmc;
+	u8 link_width_supported;
+	u8 link_speed_supported;
+	u8 link_width_enabled;
+	u8 link_speed_enabled;
+	u8 link_width_active;
+	u8 link_speed_active;
+	u8 vls_supported;
+	u8 vls_operational;
+	/* Rx Polarity inversion (compensate for ~tx on partner) */
+	u8 rx_pol_inv;
+
+	u8 hw_pidx;     /* physical port index */
+	u8 port;        /* IB port number and index into dd->pports - 1 */
+
+	u8 delay_mult;
+
+	/* used to override LED behavior */
+	u8 led_override;  /* Substituted for normal value, if non-zero */
+	u16 led_override_timeoff; /* delta to next timer event */
+	u8 led_override_vals[2]; /* Alternates per blink-frame */
+	u8 led_override_phase; /* Just counts, LSB picks from vals[] */
+	atomic_t led_override_timer_active;
+	/* Used to flash LEDs in override mode */
+	struct timer_list led_override_timer;
+};
+
+/* Observers. Not to be taken lightly, possibly not to ship. */
+/*
+ * If a diag read or write is to (bottom <= offset <= top),
+ * the "hoook" is called, allowing, e.g. shadows to be
+ * updated in sync with the driver. struct diag_observer
+ * is the "visible" part.
+ */
+struct diag_observer;
+
+typedef int (*diag_hook) (struct qib_devdata *dd,
+	const struct diag_observer *op,
+	u32 offs, u64 *data, u64 mask, int only_32);
+
+struct diag_observer {
+	diag_hook hook;
+	u32 bottom;
+	u32 top;
+};
+
+extern int qib_register_observer(struct qib_devdata *dd,
+	const struct diag_observer *op);
+
+/* Only declared here, not defined. Private to diags */
+struct diag_observer_list_elt;
+
+/* device data struct now contains only "general per-device" info.
+ * fields related to a physical IB port are in a qib_pportdata struct,
+ * described above) while fields only used by a particualr chip-type are in
+ * a qib_chipdata struct, whose contents are opaque to this file.
+ */
+struct qib_devdata {
+	struct qib_ibdev verbs_dev;     /* must be first */
+	struct list_head list;
+	/* pointers to related structs for this device */
+	/* pci access data structure */
+	struct pci_dev *pcidev;
+	struct cdev *user_cdev;
+	struct cdev *diag_cdev;
+	struct device *user_device;
+	struct device *diag_device;
+
+	/* mem-mapped pointer to base of chip regs */
+	u64 __iomem *kregbase;
+	/* end of mem-mapped chip space excluding sendbuf and user regs */
+	u64 __iomem *kregend;
+	/* physical address of chip for io_remap, etc. */
+	resource_size_t physaddr;
+	/* qib_cfgctxts pointers */
+	struct qib_ctxtdata **rcd; /* Receive Context Data */
+
+	/* qib_pportdata, points to array of (physical) port-specific
+	 * data structs, indexed by pidx (0..n-1)
+	 */
+	struct qib_pportdata *pport;
+	struct qib_chip_specific *cspec; /* chip-specific */
+
+	/* kvirt address of 1st 2k pio buffer */
+	void __iomem *pio2kbase;
+	/* kvirt address of 1st 4k pio buffer */
+	void __iomem *pio4kbase;
+	/* mem-mapped pointer to base of PIO buffers (if using WC PAT) */
+	void __iomem *piobase;
+	/* mem-mapped pointer to base of user chip regs (if using WC PAT) */
+	u64 __iomem *userbase;
+	/*
+	 * points to area where PIOavail registers will be DMA'ed.
+	 * Has to be on a page of it's own, because the page will be
+	 * mapped into user program space.  This copy is *ONLY* ever
+	 * written by DMA, not by the driver!  Need a copy per device
+	 * when we get to multiple devices
+	 */
+	volatile __le64 *pioavailregs_dma; /* DMA'ed by chip */
+	/* physical address where updates occur */
+	dma_addr_t pioavailregs_phys;
+
+	/* device-specific implementations of functions needed by
+	 * common code. Contrary to previous consensus, we can't
+	 * really just point to a device-specific table, because we
+	 * may need to "bend", e.g. *_f_put_tid
+	 */
+	/* fallback to alternate interrupt type if possible */
+	int (*f_intr_fallback)(struct qib_devdata *);
+	/* hard reset chip */
+	int (*f_reset)(struct qib_devdata *);
+	void (*f_quiet_serdes)(struct qib_pportdata *);
+	int (*f_bringup_serdes)(struct qib_pportdata *);
+	int (*f_early_init)(struct qib_devdata *);
+	void (*f_clear_tids)(struct qib_devdata *, struct qib_ctxtdata *);
+	void (*f_put_tid)(struct qib_devdata *, u64 __iomem*,
+				u32, unsigned long);
+	void (*f_cleanup)(struct qib_devdata *);
+	void (*f_setextled)(struct qib_pportdata *, u32);
+	/* fill out chip-specific fields */
+	int (*f_get_base_info)(struct qib_ctxtdata *, struct qib_base_info *);
+	/* free irq */
+	void (*f_free_irq)(struct qib_devdata *);
+	struct qib_message_header *(*f_get_msgheader)
+					(struct qib_devdata *, __le32 *);
+	void (*f_config_ctxts)(struct qib_devdata *);
+	int (*f_get_ib_cfg)(struct qib_pportdata *, int);
+	int (*f_set_ib_cfg)(struct qib_pportdata *, int, u32);
+	int (*f_set_ib_loopback)(struct qib_pportdata *, const char *);
+	int (*f_get_ib_table)(struct qib_pportdata *, int, void *);
+	int (*f_set_ib_table)(struct qib_pportdata *, int, void *);
+	u32 (*f_iblink_state)(u64);
+	u8 (*f_ibphys_portstate)(u64);
+	void (*f_xgxs_reset)(struct qib_pportdata *);
+	/* per chip actions needed for IB Link up/down changes */
+	int (*f_ib_updown)(struct qib_pportdata *, int, u64);
+	u32 __iomem *(*f_getsendbuf)(struct qib_pportdata *, u64, u32 *);
+	/* Read/modify/write of GPIO pins (potentially chip-specific */
+	int (*f_gpio_mod)(struct qib_devdata *dd, u32 out, u32 dir,
+		u32 mask);
+	/* Enable writes to config EEPROM (if supported) */
+	int (*f_eeprom_wen)(struct qib_devdata *dd, int wen);
+	/*
+	 * modify rcvctrl shadow[s] and write to appropriate chip-regs.
+	 * see above QIB_RCVCTRL_xxx_ENB/DIS for operations.
+	 * (ctxt == -1) means "all contexts", only meaningful for
+	 * clearing. Could remove if chip_spec shutdown properly done.
+	 */
+	void (*f_rcvctrl)(struct qib_pportdata *, unsigned int op,
+		int ctxt);
+	/* Read/modify/write sendctrl appropriately for op and port. */
+	void (*f_sendctrl)(struct qib_pportdata *, u32 op);
+	void (*f_set_intr_state)(struct qib_devdata *, u32);
+	void (*f_set_armlaunch)(struct qib_devdata *, u32);
+	void (*f_wantpiobuf_intr)(struct qib_devdata *, u32);
+	int (*f_late_initreg)(struct qib_devdata *);
+	void (*f_dump_sdma_state)(struct qib_pportdata *);
+	int (*f_init_sdma_regs)(struct qib_pportdata *);
+	u16 (*f_sdma_gethead)(struct qib_pportdata *);
+	int (*f_sdma_busy)(struct qib_pportdata *);
+	void (*f_sdma_update_tail)(struct qib_pportdata *, u16);
+	void (*f_sdma_set_desc_cnt)(struct qib_pportdata *, unsigned);
+	void (*f_sdma_sendctrl)(struct qib_pportdata *, unsigned);
+	void (*f_sdma_hw_clean_up)(struct qib_pportdata *);
+	void (*f_sdma_hw_start_up)(struct qib_pportdata *);
+	void (*f_sdma_init_early)(struct qib_pportdata *);
+	int (*f_sdma_flush_sends)(struct qib_pportdata *);
+	void (*f_set_cntr_sample)(struct qib_pportdata *, u32, u32);
+	void (*f_update_usrhead)(struct qib_ctxtdata *, u64, u32, u32);
+	u32 (*f_hdrqempty)(struct qib_ctxtdata *);
+	u64 (*f_portcntr)(struct qib_pportdata *, u32);
+	u32 (*f_read_cntrs)(struct qib_devdata *, loff_t, char **,
+		u64 **);
+	u32 (*f_read_portcntrs)(struct qib_devdata *, loff_t, u32,
+		char **, u64 **);
+	u32 (*f_setpbc_control)(struct qib_pportdata *, u32, u8, u8);
+	void (*f_initvl15_bufs)(struct qib_devdata *);
+	void (*f_init_ctxt)(struct qib_ctxtdata *);
+	void (*f_txchk_change)(struct qib_devdata *, u32, u32, u32,
+		struct qib_ctxtdata *);
+	void (*f_writescratch)(struct qib_devdata *, u32);
+	int (*f_tempsense_rd)(struct qib_devdata *, int regnum);
+
+	char *boardname; /* human readable board info */
+
+	/* template for writing TIDs  */
+	u64 tidtemplate;
+	/* value to write to free TIDs */
+	u64 tidinvalid;
+
+	/* number of registers used for pioavail */
+	u32 pioavregs;
+	/* device (not port) flags, basically device capabilities */
+	u32 flags;
+	/* last buffer for user use */
+	u32 lastctxt_piobuf;
+
+	/* saturating counter of (non-port-specific) device interrupts */
+	u32 int_counter;
+
+	/* pio bufs allocated per ctxt */
+	u32 pbufsctxt;
+	/* if remainder on bufs/ctxt, ctxts < extrabuf get 1 extra */
+	u32 ctxts_extrabuf;
+	/*
+	 * number of ctxts configured as max; zero is set to number chip
+	 * supports, less gives more pio bufs/ctxt, etc.
+	 */
+	u32 cfgctxts;
+
+	/*
+	 * hint that we should update pioavailshadow before
+	 * looking for a PIO buffer
+	 */
+	u32 upd_pio_shadow;
+
+	/* internal debugging stats */
+	u32 maxpkts_call;
+	u32 avgpkts_call;
+	u64 nopiobufs;
+
+	/* PCI Vendor ID (here for NodeInfo) */
+	u16 vendorid;
+	/* PCI Device ID (here for NodeInfo) */
+	u16 deviceid;
+	/* for write combining settings */
+	unsigned long wc_cookie;
+	unsigned long wc_base;
+	unsigned long wc_len;
+
+	/* shadow copy of struct page *'s for exp tid pages */
+	struct page **pageshadow;
+	/* shadow copy of dma handles for exp tid pages */
+	dma_addr_t *physshadow;
+	u64 __iomem *egrtidbase;
+	spinlock_t sendctrl_lock; /* protect changes to sendctrl shadow */
+	/* around rcd and (user ctxts) ctxt_cnt use (intr vs free) */
+	spinlock_t uctxt_lock; /* rcd and user context changes */
+	/*
+	 * per unit status, see also portdata statusp
+	 * mapped readonly into user processes so they can get unit and
+	 * IB link status cheaply
+	 */
+	u64 *devstatusp;
+	char *freezemsg; /* freeze msg if hw error put chip in freeze */
+	u32 freezelen; /* max length of freezemsg */
+	/* timer used to prevent stats overflow, error throttling, etc. */
+	struct timer_list stats_timer;
+
+	/* timer to verify interrupts work, and fallback if possible */
+	struct timer_list intrchk_timer;
+	unsigned long ureg_align; /* user register alignment */
+
+	/*
+	 * Protects pioavailshadow, pioavailkernel, pio_need_disarm, and
+	 * pio_writing.
+	 */
+	spinlock_t pioavail_lock;
+
+	/*
+	 * Shadow copies of registers; size indicates read access size.
+	 * Most of them are readonly, but some are write-only register,
+	 * where we manipulate the bits in the shadow copy, and then write
+	 * the shadow copy to qlogic_ib.
+	 *
+	 * We deliberately make most of these 32 bits, since they have
+	 * restricted range.  For any that we read, we won't to generate 32
+	 * bit accesses, since Opteron will generate 2 separate 32 bit HT
+	 * transactions for a 64 bit read, and we want to avoid unnecessary
+	 * bus transactions.
+	 */
+
+	/* This is the 64 bit group */
+
+	unsigned long pioavailshadow[6];
+	/* bitmap of send buffers available for the kernel to use with PIO. */
+	unsigned long pioavailkernel[6];
+	/* bitmap of send buffers which need to be disarmed. */
+	unsigned long pio_need_disarm[3];
+	/* bitmap of send buffers which are being written to. */
+	unsigned long pio_writing[3];
+	/* kr_revision shadow */
+	u64 revision;
+	/* Base GUID for device (from eeprom, network order) */
+	__be64 base_guid;
+
+	/*
+	 * kr_sendpiobufbase value (chip offset of pio buffers), and the
+	 * base of the 2KB buffer s(user processes only use 2K)
+	 */
+	u64 piobufbase;
+	u32 pio2k_bufbase;
+
+	/* these are the "32 bit" regs */
+
+	/* number of GUIDs in the flash for this interface */
+	u32 nguid;
+	/*
+	 * the following two are 32-bit bitmasks, but {test,clear,set}_bit
+	 * all expect bit fields to be "unsigned long"
+	 */
+	unsigned long rcvctrl; /* shadow per device rcvctrl */
+	unsigned long sendctrl; /* shadow per device sendctrl */
+
+	/* value we put in kr_rcvhdrcnt */
+	u32 rcvhdrcnt;
+	/* value we put in kr_rcvhdrsize */
+	u32 rcvhdrsize;
+	/* value we put in kr_rcvhdrentsize */
+	u32 rcvhdrentsize;
+	/* kr_ctxtcnt value */
+	u32 ctxtcnt;
+	/* kr_pagealign value */
+	u32 palign;
+	/* number of "2KB" PIO buffers */
+	u32 piobcnt2k;
+	/* size in bytes of "2KB" PIO buffers */
+	u32 piosize2k;
+	/* max usable size in dwords of a "2KB" PIO buffer before going "4KB" */
+	u32 piosize2kmax_dwords;
+	/* number of "4KB" PIO buffers */
+	u32 piobcnt4k;
+	/* size in bytes of "4KB" PIO buffers */
+	u32 piosize4k;
+	/* kr_rcvegrbase value */
+	u32 rcvegrbase;
+	/* kr_rcvtidbase value */
+	u32 rcvtidbase;
+	/* kr_rcvtidcnt value */
+	u32 rcvtidcnt;
+	/* kr_userregbase */
+	u32 uregbase;
+	/* shadow the control register contents */
+	u32 control;
+
+	/* chip address space used by 4k pio buffers */
+	u32 align4k;
+	/* size of each rcvegrbuffer */
+	u32 rcvegrbufsize;
+	/* localbus width (1, 2,4,8,16,32) from config space  */
+	u32 lbus_width;
+	/* localbus speed in MHz */
+	u32 lbus_speed;
+	int unit; /* unit # of this chip */
+
+	/* start of CHIP_SPEC move to chipspec, but need code changes */
+	/* low and high portions of MSI capability/vector */
+	u32 msi_lo;
+	/* saved after PCIe init for restore after reset */
+	u32 msi_hi;
+	/* MSI data (vector) saved for restore */
+	u16 msi_data;
+	/* so we can rewrite it after a chip reset */
+	u32 pcibar0;
+	/* so we can rewrite it after a chip reset */
+	u32 pcibar1;
+	u64 rhdrhead_intr_off;
+
+	/*
+	 * ASCII serial number, from flash, large enough for original
+	 * all digit strings, and longer QLogic serial number format
+	 */
+	u8 serial[16];
+	/* human readable board version */
+	u8 boardversion[96];
+	u8 lbus_info[32]; /* human readable localbus info */
+	/* chip major rev, from qib_revision */
+	u8 majrev;
+	/* chip minor rev, from qib_revision */
+	u8 minrev;
+
+	/* Misc small ints */
+	/* Number of physical ports available */
+	u8 num_pports;
+	/* Lowest context number which can be used by user processes */
+	u8 first_user_ctxt;
+	u8 n_krcv_queues;
+	u8 qpn_mask;
+	u8 skip_kctxt_mask;
+
+	u16 rhf_offset; /* offset of RHF within receive header entry */
+
+	/*
+	 * GPIO pins for twsi-connected devices, and device code for eeprom
+	 */
+	u8 gpio_sda_num;
+	u8 gpio_scl_num;
+	u8 twsi_eeprom_dev;
+
+	/* Support (including locks) for EEPROM logging of errors and time */
+	/* control access to actual counters, timer */
+	spinlock_t eep_st_lock;
+	/* control high-level access to EEPROM */
+	struct mutex eep_lock;
+	uint64_t traffic_wds;
+	/* active time is kept in seconds, but logged in hours */
+	atomic_t active_time;
+	/* Below are nominal shadow of EEPROM, new since last EEPROM update */
+	uint8_t eep_st_errs[QIB_EEP_LOG_CNT];
+	uint8_t eep_st_new_errs[QIB_EEP_LOG_CNT];
+	uint16_t eep_hrs;
+	/*
+	 * masks for which bits of errs, hwerrs that cause
+	 * each of the counters to increment.
+	 */
+	struct qib_eep_log_mask eep_st_masks[QIB_EEP_LOG_CNT];
+	struct qib_diag_client *diag_client;
+	spinlock_t qib_diag_trans_lock; /* protect diag observer ops */
+	struct diag_observer_list_elt *diag_observer_list;
+};
+
+/* hol_state values */
+#define QIB_HOL_UP       0
+#define QIB_HOL_INIT     1
+
+#define QIB_SDMA_SENDCTRL_OP_ENABLE    (1U << 0)
+#define QIB_SDMA_SENDCTRL_OP_INTENABLE (1U << 1)
+#define QIB_SDMA_SENDCTRL_OP_HALT      (1U << 2)
+#define QIB_SDMA_SENDCTRL_OP_CLEANUP   (1U << 3)
+#define QIB_SDMA_SENDCTRL_OP_DRAIN     (1U << 4)
+
+/* operation types for f_txchk_change() */
+#define TXCHK_CHG_TYPE_DIS1  3
+#define TXCHK_CHG_TYPE_ENAB1 2
+#define TXCHK_CHG_TYPE_KERN  1
+#define TXCHK_CHG_TYPE_USER  0
+
+#define QIB_CHASE_TIME msecs_to_jiffies(145)
+#define QIB_CHASE_DIS_TIME msecs_to_jiffies(160)
+
+/* Private data for file operations */
+struct qib_filedata {
+	struct qib_ctxtdata *rcd;
+	unsigned subctxt;
+	unsigned tidcursor;
+	struct qib_user_sdma_queue *pq;
+};
+
+extern struct list_head qib_dev_list;
+extern spinlock_t qib_devs_lock;
+extern struct qib_devdata *qib_lookup(int unit);
+
+extern unsigned qib_wc_pat;
+int qib_init(struct qib_devdata *, int);
+int init_chip_wc_pat(struct qib_devdata *dd, u32);
+int qib_enable_wc(struct qib_devdata *dd);
+void qib_disable_wc(struct qib_devdata *dd);
+int qib_count_units(int *npresentp, int *nupp);
+int qib_count_active_units(void);
+
+int qib_cdev_init(int minor, const char *name,
+		  const struct file_operations *fops,
+		  struct cdev **cdevp, struct device **devp);
+void qib_cdev_cleanup(struct cdev **cdevp, struct device **devp);
+int qib_dev_init(void);
+void qib_dev_cleanup(void);
+
+int qib_diag_add(struct qib_devdata *);
+void qib_diag_remove(struct qib_devdata *);
+void qib_handle_e_ibstatuschanged(struct qib_pportdata *, u64);
+void qib_sdma_update_tail(struct qib_pportdata *, u16); /* hold sdma_lock */
+
+int qib_decode_err(struct qib_devdata *dd, char *buf, size_t blen, u64 err);
+void qib_bad_intrstatus(struct qib_devdata *);
+void qib_handle_urcv(struct qib_devdata *, u64);
+
+/* clean up any per-chip chip-specific stuff */
+void qib_chip_cleanup(struct qib_devdata *);
+/* clean up any chip type-specific stuff */
+void qib_chip_done(void);
+
+/* check to see if we have to force ordering for write combining */
+int qib_unordered_wc(void);
+void qib_pio_copy(void __iomem *to, const void *from, size_t count);
+
+void qib_disarm_piobufs(struct qib_devdata *, unsigned, unsigned);
+int qib_disarm_piobufs_ifneeded(struct qib_ctxtdata *);
+void qib_disarm_piobufs_set(struct qib_devdata *, unsigned long *, unsigned);
+void qib_cancel_sends(struct qib_pportdata *);
+
+int qib_create_rcvhdrq(struct qib_devdata *, struct qib_ctxtdata *);
+int qib_setup_eagerbufs(struct qib_ctxtdata *);
+void qib_set_ctxtcnt(struct qib_devdata *);
+int qib_create_ctxts(struct qib_devdata *dd);
+struct qib_ctxtdata *qib_create_ctxtdata(struct qib_pportdata *, u32);
+void qib_init_pportdata(struct qib_pportdata *, struct qib_devdata *, u8, u8);
+void qib_free_ctxtdata(struct qib_devdata *, struct qib_ctxtdata *);
+
+u32 qib_kreceive(struct qib_ctxtdata *, u32 *, u32 *);
+int qib_reset_device(int);
+int qib_wait_linkstate(struct qib_pportdata *, u32, int);
+int qib_set_linkstate(struct qib_pportdata *, u8);
+int qib_set_mtu(struct qib_pportdata *, u16);
+int qib_set_lid(struct qib_pportdata *, u32, u8);
+void qib_hol_down(struct qib_pportdata *);
+void qib_hol_init(struct qib_pportdata *);
+void qib_hol_up(struct qib_pportdata *);
+void qib_hol_event(unsigned long);
+void qib_disable_after_error(struct qib_devdata *);
+int qib_set_uevent_bits(struct qib_pportdata *, const int);
+
+/* for use in system calls, where we want to know device type, etc. */
+#define ctxt_fp(fp) \
+	(((struct qib_filedata *)(fp)->private_data)->rcd)
+#define subctxt_fp(fp) \
+	(((struct qib_filedata *)(fp)->private_data)->subctxt)
+#define tidcursor_fp(fp) \
+	(((struct qib_filedata *)(fp)->private_data)->tidcursor)
+#define user_sdma_queue_fp(fp) \
+	(((struct qib_filedata *)(fp)->private_data)->pq)
+
+static inline struct qib_devdata *dd_from_ppd(struct qib_pportdata *ppd)
+{
+	return ppd->dd;
+}
+
+static inline struct qib_devdata *dd_from_dev(struct qib_ibdev *dev)
+{
+	return container_of(dev, struct qib_devdata, verbs_dev);
+}
+
+static inline struct qib_devdata *dd_from_ibdev(struct ib_device *ibdev)
+{
+	return dd_from_dev(to_idev(ibdev));
+}
+
+static inline struct qib_pportdata *ppd_from_ibp(struct qib_ibport *ibp)
+{
+	return container_of(ibp, struct qib_pportdata, ibport_data);
+}
+
+static inline struct qib_ibport *to_iport(struct ib_device *ibdev, u8 port)
+{
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
+
+	WARN_ON(pidx >= dd->num_pports);
+	return &dd->pport[pidx].ibport_data;
+}
+
+/*
+ * values for dd->flags (_device_ related flags) and
+ */
+#define QIB_HAS_LINK_LATENCY  0x1 /* supports link latency (IB 1.2) */
+#define QIB_INITTED           0x2 /* chip and driver up and initted */
+#define QIB_DOING_RESET       0x4  /* in the middle of doing chip reset */
+#define QIB_PRESENT           0x8  /* chip accesses can be done */
+#define QIB_PIO_FLUSH_WC      0x10 /* Needs Write combining flush for PIO */
+#define QIB_HAS_THRESH_UPDATE 0x40
+#define QIB_HAS_SDMA_TIMEOUT  0x80
+#define QIB_USE_SPCL_TRIG     0x100 /* SpecialTrigger launch enabled */
+#define QIB_NODMA_RTAIL       0x200 /* rcvhdrtail register DMA enabled */
+#define QIB_HAS_INTX          0x800 /* Supports INTx interrupts */
+#define QIB_HAS_SEND_DMA      0x1000 /* Supports Send DMA */
+#define QIB_HAS_VLSUPP        0x2000 /* Supports multiple VLs; PBC different */
+#define QIB_HAS_HDRSUPP       0x4000 /* Supports header suppression */
+#define QIB_BADINTR           0x8000 /* severe interrupt problems */
+#define QIB_DCA_ENABLED       0x10000 /* Direct Cache Access enabled */
+#define QIB_HAS_QSFP          0x20000 /* device (card instance) has QSFP */
+
+/*
+ * values for ppd->lflags (_ib_port_ related flags)
+ */
+#define QIBL_LINKV             0x1 /* IB link state valid */
+#define QIBL_LINKDOWN          0x8 /* IB link is down */
+#define QIBL_LINKINIT          0x10 /* IB link level is up */
+#define QIBL_LINKARMED         0x20 /* IB link is ARMED */
+#define QIBL_LINKACTIVE        0x40 /* IB link is ACTIVE */
+/* leave a gap for more IB-link state */
+#define QIBL_IB_AUTONEG_INPROG 0x1000 /* non-IBTA DDR/QDR neg active */
+#define QIBL_IB_AUTONEG_FAILED 0x2000 /* non-IBTA DDR/QDR neg failed */
+#define QIBL_IB_LINK_DISABLED  0x4000 /* Linkdown-disable forced,
+				       * Do not try to bring up */
+#define QIBL_IB_FORCE_NOTIFY   0x8000 /* force notify on next ib change */
+
+/* IB dword length mask in PBC (lower 11 bits); same for all chips */
+#define QIB_PBC_LENGTH_MASK                     ((1 << 11) - 1)
+
+
+/* ctxt_flag bit offsets */
+		/* waiting for a packet to arrive */
+#define QIB_CTXT_WAITING_RCV   2
+		/* master has not finished initializing */
+#define QIB_CTXT_MASTER_UNINIT 4
+		/* waiting for an urgent packet to arrive */
+#define QIB_CTXT_WAITING_URG 5
+
+/* free up any allocated data at closes */
+void qib_free_data(struct qib_ctxtdata *dd);
+void qib_chg_pioavailkernel(struct qib_devdata *, unsigned, unsigned,
+			    u32, struct qib_ctxtdata *);
+struct qib_devdata *qib_init_iba7322_funcs(struct pci_dev *,
+					   const struct pci_device_id *);
+struct qib_devdata *qib_init_iba7220_funcs(struct pci_dev *,
+					   const struct pci_device_id *);
+struct qib_devdata *qib_init_iba6120_funcs(struct pci_dev *,
+					   const struct pci_device_id *);
+void qib_free_devdata(struct qib_devdata *);
+struct qib_devdata *qib_alloc_devdata(struct pci_dev *pdev, size_t extra);
+
+#define QIB_TWSI_NO_DEV 0xFF
+/* Below qib_twsi_ functions must be called with eep_lock held */
+int qib_twsi_reset(struct qib_devdata *dd);
+int qib_twsi_blk_rd(struct qib_devdata *dd, int dev, int addr, void *buffer,
+		    int len);
+int qib_twsi_blk_wr(struct qib_devdata *dd, int dev, int addr,
+		    const void *buffer, int len);
+void qib_get_eeprom_info(struct qib_devdata *);
+int qib_update_eeprom_log(struct qib_devdata *dd);
+void qib_inc_eeprom_err(struct qib_devdata *dd, u32 eidx, u32 incr);
+void qib_dump_lookup_output_queue(struct qib_devdata *);
+void qib_force_pio_avail_update(struct qib_devdata *);
+
+/*
+ * Set LED override, only the two LSBs have "public" meaning, but
+ * any non-zero value substitutes them for the Link and LinkTrain
+ * LED states.
+ */
+#define QIB_LED_PHYS 1 /* Physical (linktraining) GREEN LED */
+#define QIB_LED_LOG 2  /* Logical (link) YELLOW LED */
+void qib_set_led_override(struct qib_pportdata *ppd, unsigned int val);
+
+/* send dma routines */
+int qib_setup_sdma(struct qib_pportdata *);
+void qib_teardown_sdma(struct qib_pportdata *);
+void __qib_sdma_intr(struct qib_pportdata *);
+void qib_sdma_intr(struct qib_pportdata *);
+int qib_sdma_verbs_send(struct qib_pportdata *, struct qib_sge_state *,
+			u32, struct qib_verbs_txreq *);
+/* ppd->sdma_lock should be locked before calling this. */
+int qib_sdma_make_progress(struct qib_pportdata *dd);
+
+/* must be called under qib_sdma_lock */
+static inline u16 qib_sdma_descq_freecnt(const struct qib_pportdata *ppd)
+{
+	return ppd->sdma_descq_cnt -
+		(ppd->sdma_descq_added - ppd->sdma_descq_removed) - 1;
+}
+
+static inline int __qib_sdma_running(struct qib_pportdata *ppd)
+{
+	return ppd->sdma_state.current_state == qib_sdma_state_s99_running;
+}
+int qib_sdma_running(struct qib_pportdata *);
+
+void dump_sdma_state(struct qib_pportdata *ppd);
+void __qib_sdma_process_event(struct qib_pportdata *, enum qib_sdma_events);
+void qib_sdma_process_event(struct qib_pportdata *, enum qib_sdma_events);
+
+/*
+ * number of words used for protocol header if not set by qib_userinit();
+ */
+#define QIB_DFLT_RCVHDRSIZE 9
+
+/*
+ * We need to be able to handle an IB header of at least 24 dwords.
+ * We need the rcvhdrq large enough to handle largest IB header, but
+ * still have room for a 2KB MTU standard IB packet.
+ * Additionally, some processor/memory controller combinations
+ * benefit quite strongly from having the DMA'ed data be cacheline
+ * aligned and a cacheline multiple, so we set the size to 32 dwords
+ * (2 64-byte primary cachelines for pretty much all processors of
+ * interest).  The alignment hurts nothing, other than using somewhat
+ * more memory.
+ */
+#define QIB_RCVHDR_ENTSIZE 32
+
+int qib_get_user_pages(unsigned long, size_t, struct page **);
+void qib_release_user_pages(struct page **, size_t);
+int qib_eeprom_read(struct qib_devdata *, u8, void *, int);
+int qib_eeprom_write(struct qib_devdata *, u8, const void *, int);
+u32 __iomem *qib_getsendbuf_range(struct qib_devdata *, u32 *, u32, u32);
+void qib_sendbuf_done(struct qib_devdata *, unsigned);
+
+static inline void qib_clear_rcvhdrtail(const struct qib_ctxtdata *rcd)
+{
+	*((u64 *) rcd->rcvhdrtail_kvaddr) = 0ULL;
+}
+
+static inline u32 qib_get_rcvhdrtail(const struct qib_ctxtdata *rcd)
+{
+	/*
+	 * volatile because it's a DMA target from the chip, routine is
+	 * inlined, and don't want register caching or reordering.
+	 */
+	return (u32) le64_to_cpu(
+		*((volatile __le64 *)rcd->rcvhdrtail_kvaddr)); /* DMA'ed */
+}
+
+static inline u32 qib_get_hdrqtail(const struct qib_ctxtdata *rcd)
+{
+	const struct qib_devdata *dd = rcd->dd;
+	u32 hdrqtail;
+
+	if (dd->flags & QIB_NODMA_RTAIL) {
+		__le32 *rhf_addr;
+		u32 seq;
+
+		rhf_addr = (__le32 *) rcd->rcvhdrq +
+			rcd->head + dd->rhf_offset;
+		seq = qib_hdrget_seq(rhf_addr);
+		hdrqtail = rcd->head;
+		if (seq == rcd->seq_cnt)
+			hdrqtail++;
+	} else
+		hdrqtail = qib_get_rcvhdrtail(rcd);
+
+	return hdrqtail;
+}
+
+/*
+ * sysfs interface.
+ */
+
+extern const char ib_qib_version[];
+
+int qib_device_create(struct qib_devdata *);
+void qib_device_remove(struct qib_devdata *);
+
+int qib_verbs_register_sysfs(struct qib_devdata *);
+void qib_verbs_unregister_sysfs(struct qib_devdata *);
+/* Hook for sysfs read of QSFP */
+extern int qib_qsfp_dump(struct qib_pportdata *ppd, char *buf, int len);
+
+int __init qib_init_qibfs(void);
+int __exit qib_exit_qibfs(void);
+
+int qibfs_add(struct qib_devdata *);
+int qibfs_remove(struct qib_devdata *);
+
+int qib_pcie_init(struct pci_dev *, const struct pci_device_id *);
+int qib_pcie_ddinit(struct qib_devdata *, struct pci_dev *,
+		    const struct pci_device_id *);
+void qib_pcie_ddcleanup(struct qib_devdata *);
+int qib_pcie_params(struct qib_devdata *, u32, u32 *, struct msix_entry *);
+int qib_reinit_intr(struct qib_devdata *);
+void qib_enable_intx(struct pci_dev *);
+void qib_nomsi(struct qib_devdata *);
+void qib_nomsix(struct qib_devdata *);
+void qib_pcie_getcmd(struct qib_devdata *, u16 *, u8 *, u8 *);
+void qib_pcie_reenable(struct qib_devdata *, u16, u8, u8);
+
+/*
+ * dma_addr wrappers - all 0's invalid for hw
+ */
+dma_addr_t qib_map_page(struct pci_dev *, struct page *, unsigned long,
+			  size_t, int);
+const char *qib_get_unit_name(int unit);
+
+/*
+ * Flush write combining store buffers (if present) and perform a write
+ * barrier.
+ */
+#if defined(CONFIG_X86_64)
+#define qib_flush_wc() asm volatile("sfence" : : : "memory")
+#else
+#define qib_flush_wc() wmb() /* no reorder around wc flush */
+#endif
+
+/* global module parameter variables */
+extern unsigned qib_debug; /* debugging bit mask */
+extern unsigned qib_ibmtu;
+extern ushort qib_cfgctxts;
+extern ushort qib_num_cfg_vls;
+extern ushort qib_mini_init; /* If set, do few (ideally 0) writes to chip */
+extern unsigned qib_n_krcv_queues;
+extern unsigned qib_sdma_fetch_arb;
+extern unsigned qib_compat_ddr_negotiate;
+extern int qib_special_trigger;
+
+extern struct mutex qib_mutex;
+
+/* Number of seconds before our card status check...  */
+#define STATUS_TIMEOUT 60
+
+#define QIB_DRV_NAME            "ib_qib"
+#define QIB_USER_MINOR_BASE     0
+#define QIB_TRACE_MINOR         127
+#define QIB_DIAGPKT_MINOR       128
+#define QIB_DIAG_MINOR_BASE     129
+#define QIB_NMINORS             255
+
+#define PCI_VENDOR_ID_PATHSCALE 0x1fc1
+#define PCI_VENDOR_ID_QLOGIC 0x1077
+#define PCI_DEVICE_ID_QLOGIC_IB_6120 0x10
+#define PCI_DEVICE_ID_QLOGIC_IB_7220 0x7220
+#define PCI_DEVICE_ID_QLOGIC_IB_7322 0x7322
+
+/*
+ * qib_early_err is used (only!) to print early errors before devdata is
+ * allocated, or when dd->pcidev may not be valid, and at the tail end of
+ * cleanup when devdata may have been freed, etc.  qib_dev_porterr is
+ * the same as qib_dev_err, but is used when the message really needs
+ * the IB port# to be definitive as to what's happening..
+ * All of these go to the trace log, and the trace log entry is done
+ * first to avoid possible serial port delays from printk.
+ */
+#define qib_early_err(dev, fmt, ...) \
+	do { \
+		dev_info(dev, KERN_ERR QIB_DRV_NAME ": " fmt, ##__VA_ARGS__); \
+	} while (0)
+
+#define qib_dev_err(dd, fmt, ...) \
+	do { \
+		dev_err(&(dd)->pcidev->dev, "%s: " fmt, \
+			qib_get_unit_name((dd)->unit), ##__VA_ARGS__); \
+	} while (0)
+
+#define qib_dev_porterr(dd, port, fmt, ...) \
+	do { \
+		dev_err(&(dd)->pcidev->dev, "%s: IB%u:%u " fmt, \
+			qib_get_unit_name((dd)->unit), (dd)->unit, (port), \
+			##__VA_ARGS__); \
+	} while (0)
+
+#define qib_devinfo(pcidev, fmt, ...) \
+	do { \
+		dev_info(&(pcidev)->dev, fmt, ##__VA_ARGS__); \
+	} while (0)
+
+#if _QIB_DEBUGGING
+
+# define __QIB_DBG_WHICH(which, fmt, ...) \
+	do { \
+		if (unlikely(qib_debug & (which))) \
+			printk(KERN_DEBUG QIB_DRV_NAME ": " fmt, \
+				##__VA_ARGS__); \
+	} while (0)
+
+# define qib_dbg(fmt, ...) \
+	__QIB_DBG_WHICH(__QIB_DBG, fmt, ##__VA_ARGS__)
+# define qib_cdbg(which, fmt, ...) \
+	__QIB_DBG_WHICH(__QIB_##which##DBG, fmt, ##__VA_ARGS__)
+
+#else /* ! _QIB_DEBUGGING */
+
+# define qib_dbg(fmt, ...)
+# define qib_cdbg(which, fmt, ...)
+
+#endif /* _QIB_DEBUGGING */
+
+/*
+ * this is used for formatting hw error messages...
+ */
+struct qib_hwerror_msgs {
+	u64 mask;
+	const char *msg;
+};
+
+#define QLOGIC_IB_HWE_MSG(a, b) { .mask = a, .msg = b }
+
+/* in qib_intr.c... */
+void qib_format_hwerrors(u64 hwerrs,
+			 const struct qib_hwerror_msgs *hwerrmsgs,
+			 size_t nhwerrmsgs, char *msg, size_t lmsg);
+#endif                          /* _QIB_KERNEL_H */

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 04/51] IB/qib: Add qib_6120_regs.h

[Ralph Campbell]

This creates the qib_6120_regs.h file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_6120_regs.h |  977 +++++++++++++++++++++++++++++
 1 files changed, 977 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_6120_regs.h

diff --git a/drivers/infiniband/hw/qib/qib_6120_regs.h b/drivers/infiniband/hw/qib/qib_6120_regs.h
new file mode 100644
index 0000000..8108b89
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_6120_regs.h
@@ -0,0 +1,977 @@
+/*
+ * Copyright (c) 2008, 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/* This file is mechanically generated from RTL. Any hand-edits will be lost! */
+
+#define QIB_6120_Revision_OFFS 0x0
+#define QIB_6120_Revision_R_Simulator_LSB 0x3F
+#define QIB_6120_Revision_R_Simulator_RMASK 0x1
+#define QIB_6120_Revision_Reserved_LSB 0x28
+#define QIB_6120_Revision_Reserved_RMASK 0x7FFFFF
+#define QIB_6120_Revision_BoardID_LSB 0x20
+#define QIB_6120_Revision_BoardID_RMASK 0xFF
+#define QIB_6120_Revision_R_SW_LSB 0x18
+#define QIB_6120_Revision_R_SW_RMASK 0xFF
+#define QIB_6120_Revision_R_Arch_LSB 0x10
+#define QIB_6120_Revision_R_Arch_RMASK 0xFF
+#define QIB_6120_Revision_R_ChipRevMajor_LSB 0x8
+#define QIB_6120_Revision_R_ChipRevMajor_RMASK 0xFF
+#define QIB_6120_Revision_R_ChipRevMinor_LSB 0x0
+#define QIB_6120_Revision_R_ChipRevMinor_RMASK 0xFF
+
+#define QIB_6120_Control_OFFS 0x8
+#define QIB_6120_Control_TxLatency_LSB 0x4
+#define QIB_6120_Control_TxLatency_RMASK 0x1
+#define QIB_6120_Control_PCIERetryBufDiagEn_LSB 0x3
+#define QIB_6120_Control_PCIERetryBufDiagEn_RMASK 0x1
+#define QIB_6120_Control_LinkEn_LSB 0x2
+#define QIB_6120_Control_LinkEn_RMASK 0x1
+#define QIB_6120_Control_FreezeMode_LSB 0x1
+#define QIB_6120_Control_FreezeMode_RMASK 0x1
+#define QIB_6120_Control_SyncReset_LSB 0x0
+#define QIB_6120_Control_SyncReset_RMASK 0x1
+
+#define QIB_6120_PageAlign_OFFS 0x10
+
+#define QIB_6120_PortCnt_OFFS 0x18
+
+#define QIB_6120_SendRegBase_OFFS 0x30
+
+#define QIB_6120_UserRegBase_OFFS 0x38
+
+#define QIB_6120_CntrRegBase_OFFS 0x40
+
+#define QIB_6120_Scratch_OFFS 0x48
+#define QIB_6120_Scratch_TopHalf_LSB 0x20
+#define QIB_6120_Scratch_TopHalf_RMASK 0xFFFFFFFF
+#define QIB_6120_Scratch_BottomHalf_LSB 0x0
+#define QIB_6120_Scratch_BottomHalf_RMASK 0xFFFFFFFF
+
+#define QIB_6120_IntBlocked_OFFS 0x60
+#define QIB_6120_IntBlocked_ErrorIntBlocked_LSB 0x1F
+#define QIB_6120_IntBlocked_ErrorIntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_PioSetIntBlocked_LSB 0x1E
+#define QIB_6120_IntBlocked_PioSetIntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_PioBufAvailIntBlocked_LSB 0x1D
+#define QIB_6120_IntBlocked_PioBufAvailIntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_assertGPIOIntBlocked_LSB 0x1C
+#define QIB_6120_IntBlocked_assertGPIOIntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_Reserved_LSB 0xF
+#define QIB_6120_IntBlocked_Reserved_RMASK 0x1FFF
+#define QIB_6120_IntBlocked_RcvAvail4IntBlocked_LSB 0x10
+#define QIB_6120_IntBlocked_RcvAvail4IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvAvail3IntBlocked_LSB 0xF
+#define QIB_6120_IntBlocked_RcvAvail3IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvAvail2IntBlocked_LSB 0xE
+#define QIB_6120_IntBlocked_RcvAvail2IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvAvail1IntBlocked_LSB 0xD
+#define QIB_6120_IntBlocked_RcvAvail1IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvAvail0IntBlocked_LSB 0xC
+#define QIB_6120_IntBlocked_RcvAvail0IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_Reserved1_LSB 0x5
+#define QIB_6120_IntBlocked_Reserved1_RMASK 0x7F
+#define QIB_6120_IntBlocked_RcvUrg4IntBlocked_LSB 0x4
+#define QIB_6120_IntBlocked_RcvUrg4IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvUrg3IntBlocked_LSB 0x3
+#define QIB_6120_IntBlocked_RcvUrg3IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvUrg2IntBlocked_LSB 0x2
+#define QIB_6120_IntBlocked_RcvUrg2IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvUrg1IntBlocked_LSB 0x1
+#define QIB_6120_IntBlocked_RcvUrg1IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvUrg0IntBlocked_LSB 0x0
+#define QIB_6120_IntBlocked_RcvUrg0IntBlocked_RMASK 0x1
+
+#define QIB_6120_IntMask_OFFS 0x68
+#define QIB_6120_IntMask_ErrorIntMask_LSB 0x1F
+#define QIB_6120_IntMask_ErrorIntMask_RMASK 0x1
+#define QIB_6120_IntMask_PioSetIntMask_LSB 0x1E
+#define QIB_6120_IntMask_PioSetIntMask_RMASK 0x1
+#define QIB_6120_IntMask_PioBufAvailIntMask_LSB 0x1D
+#define QIB_6120_IntMask_PioBufAvailIntMask_RMASK 0x1
+#define QIB_6120_IntMask_assertGPIOIntMask_LSB 0x1C
+#define QIB_6120_IntMask_assertGPIOIntMask_RMASK 0x1
+#define QIB_6120_IntMask_Reserved_LSB 0x11
+#define QIB_6120_IntMask_Reserved_RMASK 0x7FF
+#define QIB_6120_IntMask_RcvAvail4IntMask_LSB 0x10
+#define QIB_6120_IntMask_RcvAvail4IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvAvail3IntMask_LSB 0xF
+#define QIB_6120_IntMask_RcvAvail3IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvAvail2IntMask_LSB 0xE
+#define QIB_6120_IntMask_RcvAvail2IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvAvail1IntMask_LSB 0xD
+#define QIB_6120_IntMask_RcvAvail1IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvAvail0IntMask_LSB 0xC
+#define QIB_6120_IntMask_RcvAvail0IntMask_RMASK 0x1
+#define QIB_6120_IntMask_Reserved1_LSB 0x5
+#define QIB_6120_IntMask_Reserved1_RMASK 0x7F
+#define QIB_6120_IntMask_RcvUrg4IntMask_LSB 0x4
+#define QIB_6120_IntMask_RcvUrg4IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvUrg3IntMask_LSB 0x3
+#define QIB_6120_IntMask_RcvUrg3IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvUrg2IntMask_LSB 0x2
+#define QIB_6120_IntMask_RcvUrg2IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvUrg1IntMask_LSB 0x1
+#define QIB_6120_IntMask_RcvUrg1IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvUrg0IntMask_LSB 0x0
+#define QIB_6120_IntMask_RcvUrg0IntMask_RMASK 0x1
+
+#define QIB_6120_IntStatus_OFFS 0x70
+#define QIB_6120_IntStatus_Error_LSB 0x1F
+#define QIB_6120_IntStatus_Error_RMASK 0x1
+#define QIB_6120_IntStatus_PioSent_LSB 0x1E
+#define QIB_6120_IntStatus_PioSent_RMASK 0x1
+#define QIB_6120_IntStatus_PioBufAvail_LSB 0x1D
+#define QIB_6120_IntStatus_PioBufAvail_RMASK 0x1
+#define QIB_6120_IntStatus_assertGPIO_LSB 0x1C
+#define QIB_6120_IntStatus_assertGPIO_RMASK 0x1
+#define QIB_6120_IntStatus_Reserved_LSB 0xF
+#define QIB_6120_IntStatus_Reserved_RMASK 0x1FFF
+#define QIB_6120_IntStatus_RcvAvail4_LSB 0x10
+#define QIB_6120_IntStatus_RcvAvail4_RMASK 0x1
+#define QIB_6120_IntStatus_RcvAvail3_LSB 0xF
+#define QIB_6120_IntStatus_RcvAvail3_RMASK 0x1
+#define QIB_6120_IntStatus_RcvAvail2_LSB 0xE
+#define QIB_6120_IntStatus_RcvAvail2_RMASK 0x1
+#define QIB_6120_IntStatus_RcvAvail1_LSB 0xD
+#define QIB_6120_IntStatus_RcvAvail1_RMASK 0x1
+#define QIB_6120_IntStatus_RcvAvail0_LSB 0xC
+#define QIB_6120_IntStatus_RcvAvail0_RMASK 0x1
+#define QIB_6120_IntStatus_Reserved1_LSB 0x5
+#define QIB_6120_IntStatus_Reserved1_RMASK 0x7F
+#define QIB_6120_IntStatus_RcvUrg4_LSB 0x4
+#define QIB_6120_IntStatus_RcvUrg4_RMASK 0x1
+#define QIB_6120_IntStatus_RcvUrg3_LSB 0x3
+#define QIB_6120_IntStatus_RcvUrg3_RMASK 0x1
+#define QIB_6120_IntStatus_RcvUrg2_LSB 0x2
+#define QIB_6120_IntStatus_RcvUrg2_RMASK 0x1
+#define QIB_6120_IntStatus_RcvUrg1_LSB 0x1
+#define QIB_6120_IntStatus_RcvUrg1_RMASK 0x1
+#define QIB_6120_IntStatus_RcvUrg0_LSB 0x0
+#define QIB_6120_IntStatus_RcvUrg0_RMASK 0x1
+
+#define QIB_6120_IntClear_OFFS 0x78
+#define QIB_6120_IntClear_ErrorIntClear_LSB 0x1F
+#define QIB_6120_IntClear_ErrorIntClear_RMASK 0x1
+#define QIB_6120_IntClear_PioSetIntClear_LSB 0x1E
+#define QIB_6120_IntClear_PioSetIntClear_RMASK 0x1
+#define QIB_6120_IntClear_PioBufAvailIntClear_LSB 0x1D
+#define QIB_6120_IntClear_PioBufAvailIntClear_RMASK 0x1
+#define QIB_6120_IntClear_assertGPIOIntClear_LSB 0x1C
+#define QIB_6120_IntClear_assertGPIOIntClear_RMASK 0x1
+#define QIB_6120_IntClear_Reserved_LSB 0xF
+#define QIB_6120_IntClear_Reserved_RMASK 0x1FFF
+#define QIB_6120_IntClear_RcvAvail4IntClear_LSB 0x10
+#define QIB_6120_IntClear_RcvAvail4IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvAvail3IntClear_LSB 0xF
+#define QIB_6120_IntClear_RcvAvail3IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvAvail2IntClear_LSB 0xE
+#define QIB_6120_IntClear_RcvAvail2IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvAvail1IntClear_LSB 0xD
+#define QIB_6120_IntClear_RcvAvail1IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvAvail0IntClear_LSB 0xC
+#define QIB_6120_IntClear_RcvAvail0IntClear_RMASK 0x1
+#define QIB_6120_IntClear_Reserved1_LSB 0x5
+#define QIB_6120_IntClear_Reserved1_RMASK 0x7F
+#define QIB_6120_IntClear_RcvUrg4IntClear_LSB 0x4
+#define QIB_6120_IntClear_RcvUrg4IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvUrg3IntClear_LSB 0x3
+#define QIB_6120_IntClear_RcvUrg3IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvUrg2IntClear_LSB 0x2
+#define QIB_6120_IntClear_RcvUrg2IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvUrg1IntClear_LSB 0x1
+#define QIB_6120_IntClear_RcvUrg1IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvUrg0IntClear_LSB 0x0
+#define QIB_6120_IntClear_RcvUrg0IntClear_RMASK 0x1
+
+#define QIB_6120_ErrMask_OFFS 0x80
+#define QIB_6120_ErrMask_Reserved_LSB 0x34
+#define QIB_6120_ErrMask_Reserved_RMASK 0xFFF
+#define QIB_6120_ErrMask_HardwareErrMask_LSB 0x33
+#define QIB_6120_ErrMask_HardwareErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_ResetNegatedMask_LSB 0x32
+#define QIB_6120_ErrMask_ResetNegatedMask_RMASK 0x1
+#define QIB_6120_ErrMask_InvalidAddrErrMask_LSB 0x31
+#define QIB_6120_ErrMask_InvalidAddrErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_IBStatusChangedMask_LSB 0x30
+#define QIB_6120_ErrMask_IBStatusChangedMask_RMASK 0x1
+#define QIB_6120_ErrMask_Reserved1_LSB 0x26
+#define QIB_6120_ErrMask_Reserved1_RMASK 0x3FF
+#define QIB_6120_ErrMask_SendUnsupportedVLErrMask_LSB 0x25
+#define QIB_6120_ErrMask_SendUnsupportedVLErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendUnexpectedPktNumErrMask_LSB 0x24
+#define QIB_6120_ErrMask_SendUnexpectedPktNumErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendPioArmLaunchErrMask_LSB 0x23
+#define QIB_6120_ErrMask_SendPioArmLaunchErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendDroppedDataPktErrMask_LSB 0x22
+#define QIB_6120_ErrMask_SendDroppedDataPktErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendDroppedSmpPktErrMask_LSB 0x21
+#define QIB_6120_ErrMask_SendDroppedSmpPktErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendPktLenErrMask_LSB 0x20
+#define QIB_6120_ErrMask_SendPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendUnderRunErrMask_LSB 0x1F
+#define QIB_6120_ErrMask_SendUnderRunErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendMaxPktLenErrMask_LSB 0x1E
+#define QIB_6120_ErrMask_SendMaxPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendMinPktLenErrMask_LSB 0x1D
+#define QIB_6120_ErrMask_SendMinPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_Reserved2_LSB 0x12
+#define QIB_6120_ErrMask_Reserved2_RMASK 0x7FF
+#define QIB_6120_ErrMask_RcvIBLostLinkErrMask_LSB 0x11
+#define QIB_6120_ErrMask_RcvIBLostLinkErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvHdrErrMask_LSB 0x10
+#define QIB_6120_ErrMask_RcvHdrErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvHdrLenErrMask_LSB 0xF
+#define QIB_6120_ErrMask_RcvHdrLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvBadTidErrMask_LSB 0xE
+#define QIB_6120_ErrMask_RcvBadTidErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvHdrFullErrMask_LSB 0xD
+#define QIB_6120_ErrMask_RcvHdrFullErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvEgrFullErrMask_LSB 0xC
+#define QIB_6120_ErrMask_RcvEgrFullErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvBadVersionErrMask_LSB 0xB
+#define QIB_6120_ErrMask_RcvBadVersionErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvIBFlowErrMask_LSB 0xA
+#define QIB_6120_ErrMask_RcvIBFlowErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvEBPErrMask_LSB 0x9
+#define QIB_6120_ErrMask_RcvEBPErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvUnsupportedVLErrMask_LSB 0x8
+#define QIB_6120_ErrMask_RcvUnsupportedVLErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvUnexpectedCharErrMask_LSB 0x7
+#define QIB_6120_ErrMask_RcvUnexpectedCharErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvShortPktLenErrMask_LSB 0x6
+#define QIB_6120_ErrMask_RcvShortPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvLongPktLenErrMask_LSB 0x5
+#define QIB_6120_ErrMask_RcvLongPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvMaxPktLenErrMask_LSB 0x4
+#define QIB_6120_ErrMask_RcvMaxPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvMinPktLenErrMask_LSB 0x3
+#define QIB_6120_ErrMask_RcvMinPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvICRCErrMask_LSB 0x2
+#define QIB_6120_ErrMask_RcvICRCErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvVCRCErrMask_LSB 0x1
+#define QIB_6120_ErrMask_RcvVCRCErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvFormatErrMask_LSB 0x0
+#define QIB_6120_ErrMask_RcvFormatErrMask_RMASK 0x1
+
+#define QIB_6120_ErrStatus_OFFS 0x88
+#define QIB_6120_ErrStatus_Reserved_LSB 0x34
+#define QIB_6120_ErrStatus_Reserved_RMASK 0xFFF
+#define QIB_6120_ErrStatus_HardwareErr_LSB 0x33
+#define QIB_6120_ErrStatus_HardwareErr_RMASK 0x1
+#define QIB_6120_ErrStatus_ResetNegated_LSB 0x32
+#define QIB_6120_ErrStatus_ResetNegated_RMASK 0x1
+#define QIB_6120_ErrStatus_InvalidAddrErr_LSB 0x31
+#define QIB_6120_ErrStatus_InvalidAddrErr_RMASK 0x1
+#define QIB_6120_ErrStatus_IBStatusChanged_LSB 0x30
+#define QIB_6120_ErrStatus_IBStatusChanged_RMASK 0x1
+#define QIB_6120_ErrStatus_Reserved1_LSB 0x26
+#define QIB_6120_ErrStatus_Reserved1_RMASK 0x3FF
+#define QIB_6120_ErrStatus_SendUnsupportedVLErr_LSB 0x25
+#define QIB_6120_ErrStatus_SendUnsupportedVLErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendUnexpectedPktNumErr_LSB 0x24
+#define QIB_6120_ErrStatus_SendUnexpectedPktNumErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendPioArmLaunchErr_LSB 0x23
+#define QIB_6120_ErrStatus_SendPioArmLaunchErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendDroppedDataPktErr_LSB 0x22
+#define QIB_6120_ErrStatus_SendDroppedDataPktErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendDroppedSmpPktErr_LSB 0x21
+#define QIB_6120_ErrStatus_SendDroppedSmpPktErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendPktLenErr_LSB 0x20
+#define QIB_6120_ErrStatus_SendPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendUnderRunErr_LSB 0x1F
+#define QIB_6120_ErrStatus_SendUnderRunErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendMaxPktLenErr_LSB 0x1E
+#define QIB_6120_ErrStatus_SendMaxPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendMinPktLenErr_LSB 0x1D
+#define QIB_6120_ErrStatus_SendMinPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_Reserved2_LSB 0x12
+#define QIB_6120_ErrStatus_Reserved2_RMASK 0x7FF
+#define QIB_6120_ErrStatus_RcvIBLostLinkErr_LSB 0x11
+#define QIB_6120_ErrStatus_RcvIBLostLinkErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvHdrErr_LSB 0x10
+#define QIB_6120_ErrStatus_RcvHdrErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvHdrLenErr_LSB 0xF
+#define QIB_6120_ErrStatus_RcvHdrLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvBadTidErr_LSB 0xE
+#define QIB_6120_ErrStatus_RcvBadTidErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvHdrFullErr_LSB 0xD
+#define QIB_6120_ErrStatus_RcvHdrFullErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvEgrFullErr_LSB 0xC
+#define QIB_6120_ErrStatus_RcvEgrFullErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvBadVersionErr_LSB 0xB
+#define QIB_6120_ErrStatus_RcvBadVersionErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvIBFlowErr_LSB 0xA
+#define QIB_6120_ErrStatus_RcvIBFlowErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvEBPErr_LSB 0x9
+#define QIB_6120_ErrStatus_RcvEBPErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvUnsupportedVLErr_LSB 0x8
+#define QIB_6120_ErrStatus_RcvUnsupportedVLErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvUnexpectedCharErr_LSB 0x7
+#define QIB_6120_ErrStatus_RcvUnexpectedCharErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvShortPktLenErr_LSB 0x6
+#define QIB_6120_ErrStatus_RcvShortPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvLongPktLenErr_LSB 0x5
+#define QIB_6120_ErrStatus_RcvLongPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvMaxPktLenErr_LSB 0x4
+#define QIB_6120_ErrStatus_RcvMaxPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvMinPktLenErr_LSB 0x3
+#define QIB_6120_ErrStatus_RcvMinPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvICRCErr_LSB 0x2
+#define QIB_6120_ErrStatus_RcvICRCErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvVCRCErr_LSB 0x1
+#define QIB_6120_ErrStatus_RcvVCRCErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvFormatErr_LSB 0x0
+#define QIB_6120_ErrStatus_RcvFormatErr_RMASK 0x1
+
+#define QIB_6120_ErrClear_OFFS 0x90
+#define QIB_6120_ErrClear_Reserved_LSB 0x34
+#define QIB_6120_ErrClear_Reserved_RMASK 0xFFF
+#define QIB_6120_ErrClear_HardwareErrClear_LSB 0x33
+#define QIB_6120_ErrClear_HardwareErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_ResetNegatedClear_LSB 0x32
+#define QIB_6120_ErrClear_ResetNegatedClear_RMASK 0x1
+#define QIB_6120_ErrClear_InvalidAddrErrClear_LSB 0x31
+#define QIB_6120_ErrClear_InvalidAddrErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_IBStatusChangedClear_LSB 0x30
+#define QIB_6120_ErrClear_IBStatusChangedClear_RMASK 0x1
+#define QIB_6120_ErrClear_Reserved1_LSB 0x26
+#define QIB_6120_ErrClear_Reserved1_RMASK 0x3FF
+#define QIB_6120_ErrClear_SendUnsupportedVLErrClear_LSB 0x25
+#define QIB_6120_ErrClear_SendUnsupportedVLErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendUnexpectedPktNumErrClear_LSB 0x24
+#define QIB_6120_ErrClear_SendUnexpectedPktNumErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendPioArmLaunchErrClear_LSB 0x23
+#define QIB_6120_ErrClear_SendPioArmLaunchErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendDroppedDataPktErrClear_LSB 0x22
+#define QIB_6120_ErrClear_SendDroppedDataPktErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendDroppedSmpPktErrClear_LSB 0x21
+#define QIB_6120_ErrClear_SendDroppedSmpPktErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendPktLenErrClear_LSB 0x20
+#define QIB_6120_ErrClear_SendPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendUnderRunErrClear_LSB 0x1F
+#define QIB_6120_ErrClear_SendUnderRunErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendMaxPktLenErrClear_LSB 0x1E
+#define QIB_6120_ErrClear_SendMaxPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendMinPktLenErrClear_LSB 0x1D
+#define QIB_6120_ErrClear_SendMinPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_Reserved2_LSB 0x12
+#define QIB_6120_ErrClear_Reserved2_RMASK 0x7FF
+#define QIB_6120_ErrClear_RcvIBLostLinkErrClear_LSB 0x11
+#define QIB_6120_ErrClear_RcvIBLostLinkErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvHdrErrClear_LSB 0x10
+#define QIB_6120_ErrClear_RcvHdrErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvHdrLenErrClear_LSB 0xF
+#define QIB_6120_ErrClear_RcvHdrLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvBadTidErrClear_LSB 0xE
+#define QIB_6120_ErrClear_RcvBadTidErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvHdrFullErrClear_LSB 0xD
+#define QIB_6120_ErrClear_RcvHdrFullErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvEgrFullErrClear_LSB 0xC
+#define QIB_6120_ErrClear_RcvEgrFullErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvBadVersionErrClear_LSB 0xB
+#define QIB_6120_ErrClear_RcvBadVersionErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvIBFlowErrClear_LSB 0xA
+#define QIB_6120_ErrClear_RcvIBFlowErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvEBPErrClear_LSB 0x9
+#define QIB_6120_ErrClear_RcvEBPErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvUnsupportedVLErrClear_LSB 0x8
+#define QIB_6120_ErrClear_RcvUnsupportedVLErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvUnexpectedCharErrClear_LSB 0x7
+#define QIB_6120_ErrClear_RcvUnexpectedCharErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvShortPktLenErrClear_LSB 0x6
+#define QIB_6120_ErrClear_RcvShortPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvLongPktLenErrClear_LSB 0x5
+#define QIB_6120_ErrClear_RcvLongPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvMaxPktLenErrClear_LSB 0x4
+#define QIB_6120_ErrClear_RcvMaxPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvMinPktLenErrClear_LSB 0x3
+#define QIB_6120_ErrClear_RcvMinPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvICRCErrClear_LSB 0x2
+#define QIB_6120_ErrClear_RcvICRCErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvVCRCErrClear_LSB 0x1
+#define QIB_6120_ErrClear_RcvVCRCErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvFormatErrClear_LSB 0x0
+#define QIB_6120_ErrClear_RcvFormatErrClear_RMASK 0x1
+
+#define QIB_6120_HwErrMask_OFFS 0x98
+#define QIB_6120_HwErrMask_IBCBusFromSPCParityErrMask_LSB 0x3F
+#define QIB_6120_HwErrMask_IBCBusFromSPCParityErrMask_RMASK 0x1
+#define QIB_6120_HwErrMask_IBCBusToSPCParityErrMask_LSB 0x3E
+#define QIB_6120_HwErrMask_IBCBusToSPCParityErrMask_RMASK 0x1
+#define QIB_6120_HwErrMask_Reserved_LSB 0x3D
+#define QIB_6120_HwErrMask_Reserved_RMASK 0x1
+#define QIB_6120_HwErrMask_IBSerdesPClkNotDetectMask_LSB 0x3C
+#define QIB_6120_HwErrMask_IBSerdesPClkNotDetectMask_RMASK 0x1
+#define QIB_6120_HwErrMask_PCIESerdesQ0PClkNotDetectMask_LSB 0x3B
+#define QIB_6120_HwErrMask_PCIESerdesQ0PClkNotDetectMask_RMASK 0x1
+#define QIB_6120_HwErrMask_PCIESerdesQ1PClkNotDetectMask_LSB 0x3A
+#define QIB_6120_HwErrMask_PCIESerdesQ1PClkNotDetectMask_RMASK 0x1
+#define QIB_6120_HwErrMask_Reserved1_LSB 0x39
+#define QIB_6120_HwErrMask_Reserved1_RMASK 0x1
+#define QIB_6120_HwErrMask_IBPLLrfSlipMask_LSB 0x38
+#define QIB_6120_HwErrMask_IBPLLrfSlipMask_RMASK 0x1
+#define QIB_6120_HwErrMask_IBPLLfbSlipMask_LSB 0x37
+#define QIB_6120_HwErrMask_IBPLLfbSlipMask_RMASK 0x1
+#define QIB_6120_HwErrMask_PowerOnBISTFailedMask_LSB 0x36
+#define QIB_6120_HwErrMask_PowerOnBISTFailedMask_RMASK 0x1
+#define QIB_6120_HwErrMask_Reserved2_LSB 0x33
+#define QIB_6120_HwErrMask_Reserved2_RMASK 0x7
+#define QIB_6120_HwErrMask_RXEMemParityErrMask_LSB 0x2C
+#define QIB_6120_HwErrMask_RXEMemParityErrMask_RMASK 0x7F
+#define QIB_6120_HwErrMask_TXEMemParityErrMask_LSB 0x28
+#define QIB_6120_HwErrMask_TXEMemParityErrMask_RMASK 0xF
+#define QIB_6120_HwErrMask_Reserved3_LSB 0x22
+#define QIB_6120_HwErrMask_Reserved3_RMASK 0x3F
+#define QIB_6120_HwErrMask_PCIeBusParityErrMask_LSB 0x1F
+#define QIB_6120_HwErrMask_PCIeBusParityErrMask_RMASK 0x7
+#define QIB_6120_HwErrMask_PcieCplTimeoutMask_LSB 0x1E
+#define QIB_6120_HwErrMask_PcieCplTimeoutMask_RMASK 0x1
+#define QIB_6120_HwErrMask_PoisonedTLPMask_LSB 0x1D
+#define QIB_6120_HwErrMask_PoisonedTLPMask_RMASK 0x1
+#define QIB_6120_HwErrMask_Reserved4_LSB 0x6
+#define QIB_6120_HwErrMask_Reserved4_RMASK 0x7FFFFF
+#define QIB_6120_HwErrMask_PCIeMemParityErrMask_LSB 0x0
+#define QIB_6120_HwErrMask_PCIeMemParityErrMask_RMASK 0x3F
+
+#define QIB_6120_HwErrStatus_OFFS 0xA0
+#define QIB_6120_HwErrStatus_IBCBusFromSPCParityErr_LSB 0x3F
+#define QIB_6120_HwErrStatus_IBCBusFromSPCParityErr_RMASK 0x1
+#define QIB_6120_HwErrStatus_IBCBusToSPCParityErr_LSB 0x3E
+#define QIB_6120_HwErrStatus_IBCBusToSPCParityErr_RMASK 0x1
+#define QIB_6120_HwErrStatus_Reserved_LSB 0x3D
+#define QIB_6120_HwErrStatus_Reserved_RMASK 0x1
+#define QIB_6120_HwErrStatus_IBSerdesPClkNotDetect_LSB 0x3C
+#define QIB_6120_HwErrStatus_IBSerdesPClkNotDetect_RMASK 0x1
+#define QIB_6120_HwErrStatus_PCIESerdesQ0PClkNotDetect_LSB 0x3B
+#define QIB_6120_HwErrStatus_PCIESerdesQ0PClkNotDetect_RMASK 0x1
+#define QIB_6120_HwErrStatus_PCIESerdesQ1PClkNotDetect_LSB 0x3A
+#define QIB_6120_HwErrStatus_PCIESerdesQ1PClkNotDetect_RMASK 0x1
+#define QIB_6120_HwErrStatus_Reserved1_LSB 0x39
+#define QIB_6120_HwErrStatus_Reserved1_RMASK 0x1
+#define QIB_6120_HwErrStatus_IBPLLrfSlip_LSB 0x38
+#define QIB_6120_HwErrStatus_IBPLLrfSlip_RMASK 0x1
+#define QIB_6120_HwErrStatus_IBPLLfbSlip_LSB 0x37
+#define QIB_6120_HwErrStatus_IBPLLfbSlip_RMASK 0x1
+#define QIB_6120_HwErrStatus_PowerOnBISTFailed_LSB 0x36
+#define QIB_6120_HwErrStatus_PowerOnBISTFailed_RMASK 0x1
+#define QIB_6120_HwErrStatus_Reserved2_LSB 0x33
+#define QIB_6120_HwErrStatus_Reserved2_RMASK 0x7
+#define QIB_6120_HwErrStatus_RXEMemParity_LSB 0x2C
+#define QIB_6120_HwErrStatus_RXEMemParity_RMASK 0x7F
+#define QIB_6120_HwErrStatus_TXEMemParity_LSB 0x28
+#define QIB_6120_HwErrStatus_TXEMemParity_RMASK 0xF
+#define QIB_6120_HwErrStatus_Reserved3_LSB 0x22
+#define QIB_6120_HwErrStatus_Reserved3_RMASK 0x3F
+#define QIB_6120_HwErrStatus_PCIeBusParity_LSB 0x1F
+#define QIB_6120_HwErrStatus_PCIeBusParity_RMASK 0x7
+#define QIB_6120_HwErrStatus_PcieCplTimeout_LSB 0x1E
+#define QIB_6120_HwErrStatus_PcieCplTimeout_RMASK 0x1
+#define QIB_6120_HwErrStatus_PoisenedTLP_LSB 0x1D
+#define QIB_6120_HwErrStatus_PoisenedTLP_RMASK 0x1
+#define QIB_6120_HwErrStatus_Reserved4_LSB 0x6
+#define QIB_6120_HwErrStatus_Reserved4_RMASK 0x7FFFFF
+#define QIB_6120_HwErrStatus_PCIeMemParity_LSB 0x0
+#define QIB_6120_HwErrStatus_PCIeMemParity_RMASK 0x3F
+
+#define QIB_6120_HwErrClear_OFFS 0xA8
+#define QIB_6120_HwErrClear_IBCBusFromSPCParityErrClear_LSB 0x3F
+#define QIB_6120_HwErrClear_IBCBusFromSPCParityErrClear_RMASK 0x1
+#define QIB_6120_HwErrClear_IBCBusToSPCparityErrClear_LSB 0x3E
+#define QIB_6120_HwErrClear_IBCBusToSPCparityErrClear_RMASK 0x1
+#define QIB_6120_HwErrClear_Reserved_LSB 0x3D
+#define QIB_6120_HwErrClear_Reserved_RMASK 0x1
+#define QIB_6120_HwErrClear_IBSerdesPClkNotDetectClear_LSB 0x3C
+#define QIB_6120_HwErrClear_IBSerdesPClkNotDetectClear_RMASK 0x1
+#define QIB_6120_HwErrClear_PCIESerdesQ0PClkNotDetectClear_LSB 0x3B
+#define QIB_6120_HwErrClear_PCIESerdesQ0PClkNotDetectClear_RMASK 0x1
+#define QIB_6120_HwErrClear_PCIESerdesQ1PClkNotDetectClear_LSB 0x3A
+#define QIB_6120_HwErrClear_PCIESerdesQ1PClkNotDetectClear_RMASK 0x1
+#define QIB_6120_HwErrClear_Reserved1_LSB 0x39
+#define QIB_6120_HwErrClear_Reserved1_RMASK 0x1
+#define QIB_6120_HwErrClear_IBPLLrfSlipClear_LSB 0x38
+#define QIB_6120_HwErrClear_IBPLLrfSlipClear_RMASK 0x1
+#define QIB_6120_HwErrClear_IBPLLfbSlipClear_LSB 0x37
+#define QIB_6120_HwErrClear_IBPLLfbSlipClear_RMASK 0x1
+#define QIB_6120_HwErrClear_PowerOnBISTFailedClear_LSB 0x36
+#define QIB_6120_HwErrClear_PowerOnBISTFailedClear_RMASK 0x1
+#define QIB_6120_HwErrClear_Reserved2_LSB 0x33
+#define QIB_6120_HwErrClear_Reserved2_RMASK 0x7
+#define QIB_6120_HwErrClear_RXEMemParityClear_LSB 0x2C
+#define QIB_6120_HwErrClear_RXEMemParityClear_RMASK 0x7F
+#define QIB_6120_HwErrClear_TXEMemParityClear_LSB 0x28
+#define QIB_6120_HwErrClear_TXEMemParityClear_RMASK 0xF
+#define QIB_6120_HwErrClear_Reserved3_LSB 0x22
+#define QIB_6120_HwErrClear_Reserved3_RMASK 0x3F
+#define QIB_6120_HwErrClear_PCIeBusParityClr_LSB 0x1F
+#define QIB_6120_HwErrClear_PCIeBusParityClr_RMASK 0x7
+#define QIB_6120_HwErrClear_PcieCplTimeoutClear_LSB 0x1E
+#define QIB_6120_HwErrClear_PcieCplTimeoutClear_RMASK 0x1
+#define QIB_6120_HwErrClear_PoisonedTLPClear_LSB 0x1D
+#define QIB_6120_HwErrClear_PoisonedTLPClear_RMASK 0x1
+#define QIB_6120_HwErrClear_Reserved4_LSB 0x6
+#define QIB_6120_HwErrClear_Reserved4_RMASK 0x7FFFFF
+#define QIB_6120_HwErrClear_PCIeMemParityClr_LSB 0x0
+#define QIB_6120_HwErrClear_PCIeMemParityClr_RMASK 0x3F
+
+#define QIB_6120_HwDiagCtrl_OFFS 0xB0
+#define QIB_6120_HwDiagCtrl_ForceIBCBusFromSPCParityErr_LSB 0x3F
+#define QIB_6120_HwDiagCtrl_ForceIBCBusFromSPCParityErr_RMASK 0x1
+#define QIB_6120_HwDiagCtrl_ForceIBCBusToSPCParityErr_LSB 0x3E
+#define QIB_6120_HwDiagCtrl_ForceIBCBusToSPCParityErr_RMASK 0x1
+#define QIB_6120_HwDiagCtrl_CounterWrEnable_LSB 0x3D
+#define QIB_6120_HwDiagCtrl_CounterWrEnable_RMASK 0x1
+#define QIB_6120_HwDiagCtrl_CounterDisable_LSB 0x3C
+#define QIB_6120_HwDiagCtrl_CounterDisable_RMASK 0x1
+#define QIB_6120_HwDiagCtrl_Reserved_LSB 0x33
+#define QIB_6120_HwDiagCtrl_Reserved_RMASK 0x1FF
+#define QIB_6120_HwDiagCtrl_ForceRxMemParityErr_LSB 0x2C
+#define QIB_6120_HwDiagCtrl_ForceRxMemParityErr_RMASK 0x7F
+#define QIB_6120_HwDiagCtrl_ForceTxMemparityErr_LSB 0x28
+#define QIB_6120_HwDiagCtrl_ForceTxMemparityErr_RMASK 0xF
+#define QIB_6120_HwDiagCtrl_Reserved1_LSB 0x23
+#define QIB_6120_HwDiagCtrl_Reserved1_RMASK 0x1F
+#define QIB_6120_HwDiagCtrl_forcePCIeBusParity_LSB 0x1F
+#define QIB_6120_HwDiagCtrl_forcePCIeBusParity_RMASK 0xF
+#define QIB_6120_HwDiagCtrl_Reserved2_LSB 0x6
+#define QIB_6120_HwDiagCtrl_Reserved2_RMASK 0x1FFFFFF
+#define QIB_6120_HwDiagCtrl_forcePCIeMemParity_LSB 0x0
+#define QIB_6120_HwDiagCtrl_forcePCIeMemParity_RMASK 0x3F
+
+#define QIB_6120_IBCStatus_OFFS 0xC0
+#define QIB_6120_IBCStatus_TxCreditOk_LSB 0x1F
+#define QIB_6120_IBCStatus_TxCreditOk_RMASK 0x1
+#define QIB_6120_IBCStatus_TxReady_LSB 0x1E
+#define QIB_6120_IBCStatus_TxReady_RMASK 0x1
+#define QIB_6120_IBCStatus_Reserved_LSB 0x7
+#define QIB_6120_IBCStatus_Reserved_RMASK 0x7FFFFF
+#define QIB_6120_IBCStatus_LinkState_LSB 0x4
+#define QIB_6120_IBCStatus_LinkState_RMASK 0x7
+#define QIB_6120_IBCStatus_LinkTrainingState_LSB 0x0
+#define QIB_6120_IBCStatus_LinkTrainingState_RMASK 0xF
+
+#define QIB_6120_IBCCtrl_OFFS 0xC8
+#define QIB_6120_IBCCtrl_Loopback_LSB 0x3F
+#define QIB_6120_IBCCtrl_Loopback_RMASK 0x1
+#define QIB_6120_IBCCtrl_LinkDownDefaultState_LSB 0x3E
+#define QIB_6120_IBCCtrl_LinkDownDefaultState_RMASK 0x1
+#define QIB_6120_IBCCtrl_Reserved_LSB 0x2B
+#define QIB_6120_IBCCtrl_Reserved_RMASK 0x7FFFF
+#define QIB_6120_IBCCtrl_CreditScale_LSB 0x28
+#define QIB_6120_IBCCtrl_CreditScale_RMASK 0x7
+#define QIB_6120_IBCCtrl_OverrunThreshold_LSB 0x24
+#define QIB_6120_IBCCtrl_OverrunThreshold_RMASK 0xF
+#define QIB_6120_IBCCtrl_PhyerrThreshold_LSB 0x20
+#define QIB_6120_IBCCtrl_PhyerrThreshold_RMASK 0xF
+#define QIB_6120_IBCCtrl_Reserved1_LSB 0x1F
+#define QIB_6120_IBCCtrl_Reserved1_RMASK 0x1
+#define QIB_6120_IBCCtrl_MaxPktLen_LSB 0x14
+#define QIB_6120_IBCCtrl_MaxPktLen_RMASK 0x7FF
+#define QIB_6120_IBCCtrl_LinkCmd_LSB 0x12
+#define QIB_6120_IBCCtrl_LinkCmd_RMASK 0x3
+#define QIB_6120_IBCCtrl_LinkInitCmd_LSB 0x10
+#define QIB_6120_IBCCtrl_LinkInitCmd_RMASK 0x3
+#define QIB_6120_IBCCtrl_FlowCtrlWaterMark_LSB 0x8
+#define QIB_6120_IBCCtrl_FlowCtrlWaterMark_RMASK 0xFF
+#define QIB_6120_IBCCtrl_FlowCtrlPeriod_LSB 0x0
+#define QIB_6120_IBCCtrl_FlowCtrlPeriod_RMASK 0xFF
+
+#define QIB_6120_EXTStatus_OFFS 0xD0
+#define QIB_6120_EXTStatus_GPIOIn_LSB 0x30
+#define QIB_6120_EXTStatus_GPIOIn_RMASK 0xFFFF
+#define QIB_6120_EXTStatus_Reserved_LSB 0x20
+#define QIB_6120_EXTStatus_Reserved_RMASK 0xFFFF
+#define QIB_6120_EXTStatus_Reserved1_LSB 0x10
+#define QIB_6120_EXTStatus_Reserved1_RMASK 0xFFFF
+#define QIB_6120_EXTStatus_MemBISTFoundErr_LSB 0xF
+#define QIB_6120_EXTStatus_MemBISTFoundErr_RMASK 0x1
+#define QIB_6120_EXTStatus_MemBISTEndTest_LSB 0xE
+#define QIB_6120_EXTStatus_MemBISTEndTest_RMASK 0x1
+#define QIB_6120_EXTStatus_Reserved2_LSB 0x0
+#define QIB_6120_EXTStatus_Reserved2_RMASK 0x3FFF
+
+#define QIB_6120_EXTCtrl_OFFS 0xD8
+#define QIB_6120_EXTCtrl_GPIOOe_LSB 0x30
+#define QIB_6120_EXTCtrl_GPIOOe_RMASK 0xFFFF
+#define QIB_6120_EXTCtrl_GPIOInvert_LSB 0x20
+#define QIB_6120_EXTCtrl_GPIOInvert_RMASK 0xFFFF
+#define QIB_6120_EXTCtrl_Reserved_LSB 0x4
+#define QIB_6120_EXTCtrl_Reserved_RMASK 0xFFFFFFF
+#define QIB_6120_EXTCtrl_LEDPriPortGreenOn_LSB 0x3
+#define QIB_6120_EXTCtrl_LEDPriPortGreenOn_RMASK 0x1
+#define QIB_6120_EXTCtrl_LEDPriPortYellowOn_LSB 0x2
+#define QIB_6120_EXTCtrl_LEDPriPortYellowOn_RMASK 0x1
+#define QIB_6120_EXTCtrl_LEDGblOkGreenOn_LSB 0x1
+#define QIB_6120_EXTCtrl_LEDGblOkGreenOn_RMASK 0x1
+#define QIB_6120_EXTCtrl_LEDGblErrRedOff_LSB 0x0
+#define QIB_6120_EXTCtrl_LEDGblErrRedOff_RMASK 0x1
+
+#define QIB_6120_GPIOOut_OFFS 0xE0
+
+#define QIB_6120_GPIOMask_OFFS 0xE8
+
+#define QIB_6120_GPIOStatus_OFFS 0xF0
+
+#define QIB_6120_GPIOClear_OFFS 0xF8
+
+#define QIB_6120_RcvCtrl_OFFS 0x100
+#define QIB_6120_RcvCtrl_TailUpd_LSB 0x1F
+#define QIB_6120_RcvCtrl_TailUpd_RMASK 0x1
+#define QIB_6120_RcvCtrl_RcvPartitionKeyDisable_LSB 0x1E
+#define QIB_6120_RcvCtrl_RcvPartitionKeyDisable_RMASK 0x1
+#define QIB_6120_RcvCtrl_Reserved_LSB 0x15
+#define QIB_6120_RcvCtrl_Reserved_RMASK 0x1FF
+#define QIB_6120_RcvCtrl_IntrAvail_LSB 0x10
+#define QIB_6120_RcvCtrl_IntrAvail_RMASK 0x1F
+#define QIB_6120_RcvCtrl_Reserved1_LSB 0x9
+#define QIB_6120_RcvCtrl_Reserved1_RMASK 0x7F
+#define QIB_6120_RcvCtrl_Reserved2_LSB 0x5
+#define QIB_6120_RcvCtrl_Reserved2_RMASK 0xF
+#define QIB_6120_RcvCtrl_PortEnable_LSB 0x0
+#define QIB_6120_RcvCtrl_PortEnable_RMASK 0x1F
+
+#define QIB_6120_RcvBTHQP_OFFS 0x108
+#define QIB_6120_RcvBTHQP_BTHQP_Mask_LSB 0x1E
+#define QIB_6120_RcvBTHQP_BTHQP_Mask_RMASK 0x3
+#define QIB_6120_RcvBTHQP_Reserved_LSB 0x18
+#define QIB_6120_RcvBTHQP_Reserved_RMASK 0x3F
+#define QIB_6120_RcvBTHQP_RcvBTHQP_LSB 0x0
+#define QIB_6120_RcvBTHQP_RcvBTHQP_RMASK 0xFFFFFF
+
+#define QIB_6120_RcvHdrSize_OFFS 0x110
+
+#define QIB_6120_RcvHdrCnt_OFFS 0x118
+
+#define QIB_6120_RcvHdrEntSize_OFFS 0x120
+
+#define QIB_6120_RcvTIDBase_OFFS 0x128
+
+#define QIB_6120_RcvTIDCnt_OFFS 0x130
+
+#define QIB_6120_RcvEgrBase_OFFS 0x138
+
+#define QIB_6120_RcvEgrCnt_OFFS 0x140
+
+#define QIB_6120_RcvBufBase_OFFS 0x148
+
+#define QIB_6120_RcvBufSize_OFFS 0x150
+
+#define QIB_6120_RxIntMemBase_OFFS 0x158
+
+#define QIB_6120_RxIntMemSize_OFFS 0x160
+
+#define QIB_6120_RcvPartitionKey_OFFS 0x168
+
+#define QIB_6120_RcvPktLEDCnt_OFFS 0x178
+#define QIB_6120_RcvPktLEDCnt_ONperiod_LSB 0x20
+#define QIB_6120_RcvPktLEDCnt_ONperiod_RMASK 0xFFFFFFFF
+#define QIB_6120_RcvPktLEDCnt_OFFperiod_LSB 0x0
+#define QIB_6120_RcvPktLEDCnt_OFFperiod_RMASK 0xFFFFFFFF
+
+#define QIB_6120_SendCtrl_OFFS 0x1C0
+#define QIB_6120_SendCtrl_Disarm_LSB 0x1F
+#define QIB_6120_SendCtrl_Disarm_RMASK 0x1
+#define QIB_6120_SendCtrl_Reserved_LSB 0x17
+#define QIB_6120_SendCtrl_Reserved_RMASK 0xFF
+#define QIB_6120_SendCtrl_DisarmPIOBuf_LSB 0x10
+#define QIB_6120_SendCtrl_DisarmPIOBuf_RMASK 0x7F
+#define QIB_6120_SendCtrl_Reserved1_LSB 0x4
+#define QIB_6120_SendCtrl_Reserved1_RMASK 0xFFF
+#define QIB_6120_SendCtrl_PIOEnable_LSB 0x3
+#define QIB_6120_SendCtrl_PIOEnable_RMASK 0x1
+#define QIB_6120_SendCtrl_PIOBufAvailUpd_LSB 0x2
+#define QIB_6120_SendCtrl_PIOBufAvailUpd_RMASK 0x1
+#define QIB_6120_SendCtrl_PIOIntBufAvail_LSB 0x1
+#define QIB_6120_SendCtrl_PIOIntBufAvail_RMASK 0x1
+#define QIB_6120_SendCtrl_Abort_LSB 0x0
+#define QIB_6120_SendCtrl_Abort_RMASK 0x1
+
+#define QIB_6120_SendPIOBufBase_OFFS 0x1C8
+#define QIB_6120_SendPIOBufBase_Reserved_LSB 0x35
+#define QIB_6120_SendPIOBufBase_Reserved_RMASK 0x7FF
+#define QIB_6120_SendPIOBufBase_BaseAddr_LargePIO_LSB 0x20
+#define QIB_6120_SendPIOBufBase_BaseAddr_LargePIO_RMASK 0x1FFFFF
+#define QIB_6120_SendPIOBufBase_Reserved1_LSB 0x15
+#define QIB_6120_SendPIOBufBase_Reserved1_RMASK 0x7FF
+#define QIB_6120_SendPIOBufBase_BaseAddr_SmallPIO_LSB 0x0
+#define QIB_6120_SendPIOBufBase_BaseAddr_SmallPIO_RMASK 0x1FFFFF
+
+#define QIB_6120_SendPIOSize_OFFS 0x1D0
+#define QIB_6120_SendPIOSize_Reserved_LSB 0x2D
+#define QIB_6120_SendPIOSize_Reserved_RMASK 0xFFFFF
+#define QIB_6120_SendPIOSize_Size_LargePIO_LSB 0x20
+#define QIB_6120_SendPIOSize_Size_LargePIO_RMASK 0x1FFF
+#define QIB_6120_SendPIOSize_Reserved1_LSB 0xC
+#define QIB_6120_SendPIOSize_Reserved1_RMASK 0xFFFFF
+#define QIB_6120_SendPIOSize_Size_SmallPIO_LSB 0x0
+#define QIB_6120_SendPIOSize_Size_SmallPIO_RMASK 0xFFF
+
+#define QIB_6120_SendPIOBufCnt_OFFS 0x1D8
+#define QIB_6120_SendPIOBufCnt_Reserved_LSB 0x24
+#define QIB_6120_SendPIOBufCnt_Reserved_RMASK 0xFFFFFFF
+#define QIB_6120_SendPIOBufCnt_Num_LargePIO_LSB 0x20
+#define QIB_6120_SendPIOBufCnt_Num_LargePIO_RMASK 0xF
+#define QIB_6120_SendPIOBufCnt_Reserved1_LSB 0x9
+#define QIB_6120_SendPIOBufCnt_Reserved1_RMASK 0x7FFFFF
+#define QIB_6120_SendPIOBufCnt_Num_SmallPIO_LSB 0x0
+#define QIB_6120_SendPIOBufCnt_Num_SmallPIO_RMASK 0x1FF
+
+#define QIB_6120_SendPIOAvailAddr_OFFS 0x1E0
+#define QIB_6120_SendPIOAvailAddr_SendPIOAvailAddr_LSB 0x6
+#define QIB_6120_SendPIOAvailAddr_SendPIOAvailAddr_RMASK 0x3FFFFFFFF
+#define QIB_6120_SendPIOAvailAddr_Reserved_LSB 0x0
+#define QIB_6120_SendPIOAvailAddr_Reserved_RMASK 0x3F
+
+#define QIB_6120_SendBufErr0_OFFS 0x240
+#define QIB_6120_SendBufErr0_SendBufErrPIO_63_0_LSB 0x0
+#define QIB_6120_SendBufErr0_SendBufErrPIO_63_0_RMASK 0x0
+
+#define QIB_6120_RcvHdrAddr0_OFFS 0x280
+#define QIB_6120_RcvHdrAddr0_RcvHdrAddr0_LSB 0x2
+#define QIB_6120_RcvHdrAddr0_RcvHdrAddr0_RMASK 0x3FFFFFFFFF
+#define QIB_6120_RcvHdrAddr0_Reserved_LSB 0x0
+#define QIB_6120_RcvHdrAddr0_Reserved_RMASK 0x3
+
+#define QIB_6120_RcvHdrTailAddr0_OFFS 0x300
+#define QIB_6120_RcvHdrTailAddr0_RcvHdrTailAddr0_LSB 0x2
+#define QIB_6120_RcvHdrTailAddr0_RcvHdrTailAddr0_RMASK 0x3FFFFFFFFF
+#define QIB_6120_RcvHdrTailAddr0_Reserved_LSB 0x0
+#define QIB_6120_RcvHdrTailAddr0_Reserved_RMASK 0x3
+
+#define QIB_6120_SerdesCfg0_OFFS 0x3C0
+#define QIB_6120_SerdesCfg0_DisableIBTxIdleDetect_LSB 0x3F
+#define QIB_6120_SerdesCfg0_DisableIBTxIdleDetect_RMASK 0x1
+#define QIB_6120_SerdesCfg0_Reserved_LSB 0x38
+#define QIB_6120_SerdesCfg0_Reserved_RMASK 0x7F
+#define QIB_6120_SerdesCfg0_RxEqCtl_LSB 0x36
+#define QIB_6120_SerdesCfg0_RxEqCtl_RMASK 0x3
+#define QIB_6120_SerdesCfg0_TxTermAdj_LSB 0x34
+#define QIB_6120_SerdesCfg0_TxTermAdj_RMASK 0x3
+#define QIB_6120_SerdesCfg0_RxTermAdj_LSB 0x32
+#define QIB_6120_SerdesCfg0_RxTermAdj_RMASK 0x3
+#define QIB_6120_SerdesCfg0_TermAdj1_LSB 0x31
+#define QIB_6120_SerdesCfg0_TermAdj1_RMASK 0x1
+#define QIB_6120_SerdesCfg0_TermAdj0_LSB 0x30
+#define QIB_6120_SerdesCfg0_TermAdj0_RMASK 0x1
+#define QIB_6120_SerdesCfg0_LPBKA_LSB 0x2F
+#define QIB_6120_SerdesCfg0_LPBKA_RMASK 0x1
+#define QIB_6120_SerdesCfg0_LPBKB_LSB 0x2E
+#define QIB_6120_SerdesCfg0_LPBKB_RMASK 0x1
+#define QIB_6120_SerdesCfg0_LPBKC_LSB 0x2D
+#define QIB_6120_SerdesCfg0_LPBKC_RMASK 0x1
+#define QIB_6120_SerdesCfg0_LPBKD_LSB 0x2C
+#define QIB_6120_SerdesCfg0_LPBKD_RMASK 0x1
+#define QIB_6120_SerdesCfg0_PW_LSB 0x2B
+#define QIB_6120_SerdesCfg0_PW_RMASK 0x1
+#define QIB_6120_SerdesCfg0_RefSel_LSB 0x29
+#define QIB_6120_SerdesCfg0_RefSel_RMASK 0x3
+#define QIB_6120_SerdesCfg0_ParReset_LSB 0x28
+#define QIB_6120_SerdesCfg0_ParReset_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ParLPBK_LSB 0x27
+#define QIB_6120_SerdesCfg0_ParLPBK_RMASK 0x1
+#define QIB_6120_SerdesCfg0_OffsetEn_LSB 0x26
+#define QIB_6120_SerdesCfg0_OffsetEn_RMASK 0x1
+#define QIB_6120_SerdesCfg0_Offset_LSB 0x1E
+#define QIB_6120_SerdesCfg0_Offset_RMASK 0xFF
+#define QIB_6120_SerdesCfg0_L2PwrDn_LSB 0x1D
+#define QIB_6120_SerdesCfg0_L2PwrDn_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetPLL_LSB 0x1C
+#define QIB_6120_SerdesCfg0_ResetPLL_RMASK 0x1
+#define QIB_6120_SerdesCfg0_RxTermEnX_LSB 0x18
+#define QIB_6120_SerdesCfg0_RxTermEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_BeaconTxEnX_LSB 0x14
+#define QIB_6120_SerdesCfg0_BeaconTxEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_RxDetEnX_LSB 0x10
+#define QIB_6120_SerdesCfg0_RxDetEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_TxIdeEnX_LSB 0xC
+#define QIB_6120_SerdesCfg0_TxIdeEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_RxIdleEnX_LSB 0x8
+#define QIB_6120_SerdesCfg0_RxIdleEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_L1PwrDnA_LSB 0x7
+#define QIB_6120_SerdesCfg0_L1PwrDnA_RMASK 0x1
+#define QIB_6120_SerdesCfg0_L1PwrDnB_LSB 0x6
+#define QIB_6120_SerdesCfg0_L1PwrDnB_RMASK 0x1
+#define QIB_6120_SerdesCfg0_L1PwrDnC_LSB 0x5
+#define QIB_6120_SerdesCfg0_L1PwrDnC_RMASK 0x1
+#define QIB_6120_SerdesCfg0_L1PwrDnD_LSB 0x4
+#define QIB_6120_SerdesCfg0_L1PwrDnD_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetA_LSB 0x3
+#define QIB_6120_SerdesCfg0_ResetA_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetB_LSB 0x2
+#define QIB_6120_SerdesCfg0_ResetB_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetC_LSB 0x1
+#define QIB_6120_SerdesCfg0_ResetC_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetD_LSB 0x0
+#define QIB_6120_SerdesCfg0_ResetD_RMASK 0x1
+
+#define QIB_6120_SerdesStat_OFFS 0x3D0
+#define QIB_6120_SerdesStat_Reserved_LSB 0xC
+#define QIB_6120_SerdesStat_Reserved_RMASK 0xFFFFFFFFFFFFF
+#define QIB_6120_SerdesStat_BeaconDetA_LSB 0xB
+#define QIB_6120_SerdesStat_BeaconDetA_RMASK 0x1
+#define QIB_6120_SerdesStat_BeaconDetB_LSB 0xA
+#define QIB_6120_SerdesStat_BeaconDetB_RMASK 0x1
+#define QIB_6120_SerdesStat_BeaconDetC_LSB 0x9
+#define QIB_6120_SerdesStat_BeaconDetC_RMASK 0x1
+#define QIB_6120_SerdesStat_BeaconDetD_LSB 0x8
+#define QIB_6120_SerdesStat_BeaconDetD_RMASK 0x1
+#define QIB_6120_SerdesStat_RxDetA_LSB 0x7
+#define QIB_6120_SerdesStat_RxDetA_RMASK 0x1
+#define QIB_6120_SerdesStat_RxDetB_LSB 0x6
+#define QIB_6120_SerdesStat_RxDetB_RMASK 0x1
+#define QIB_6120_SerdesStat_RxDetC_LSB 0x5
+#define QIB_6120_SerdesStat_RxDetC_RMASK 0x1
+#define QIB_6120_SerdesStat_RxDetD_LSB 0x4
+#define QIB_6120_SerdesStat_RxDetD_RMASK 0x1
+#define QIB_6120_SerdesStat_TxIdleDetA_LSB 0x3
+#define QIB_6120_SerdesStat_TxIdleDetA_RMASK 0x1
+#define QIB_6120_SerdesStat_TxIdleDetB_LSB 0x2
+#define QIB_6120_SerdesStat_TxIdleDetB_RMASK 0x1
+#define QIB_6120_SerdesStat_TxIdleDetC_LSB 0x1
+#define QIB_6120_SerdesStat_TxIdleDetC_RMASK 0x1
+#define QIB_6120_SerdesStat_TxIdleDetD_LSB 0x0
+#define QIB_6120_SerdesStat_TxIdleDetD_RMASK 0x1
+
+#define QIB_6120_XGXSCfg_OFFS 0x3D8
+#define QIB_6120_XGXSCfg_ArmLaunchErrorDisable_LSB 0x3F
+#define QIB_6120_XGXSCfg_ArmLaunchErrorDisable_RMASK 0x1
+#define QIB_6120_XGXSCfg_Reserved_LSB 0x17
+#define QIB_6120_XGXSCfg_Reserved_RMASK 0xFFFFFFFFFF
+#define QIB_6120_XGXSCfg_polarity_inv_LSB 0x13
+#define QIB_6120_XGXSCfg_polarity_inv_RMASK 0xF
+#define QIB_6120_XGXSCfg_link_sync_mask_LSB 0x9
+#define QIB_6120_XGXSCfg_link_sync_mask_RMASK 0x3FF
+#define QIB_6120_XGXSCfg_port_addr_LSB 0x4
+#define QIB_6120_XGXSCfg_port_addr_RMASK 0x1F
+#define QIB_6120_XGXSCfg_mdd_30_LSB 0x3
+#define QIB_6120_XGXSCfg_mdd_30_RMASK 0x1
+#define QIB_6120_XGXSCfg_xcv_resetn_LSB 0x2
+#define QIB_6120_XGXSCfg_xcv_resetn_RMASK 0x1
+#define QIB_6120_XGXSCfg_Reserved1_LSB 0x1
+#define QIB_6120_XGXSCfg_Reserved1_RMASK 0x1
+#define QIB_6120_XGXSCfg_tx_rx_resetn_LSB 0x0
+#define QIB_6120_XGXSCfg_tx_rx_resetn_RMASK 0x1
+
+#define QIB_6120_LBIntCnt_OFFS 0x12000
+
+#define QIB_6120_LBFlowStallCnt_OFFS 0x12008
+
+#define QIB_6120_TxUnsupVLErrCnt_OFFS 0x12018
+
+#define QIB_6120_TxDataPktCnt_OFFS 0x12020
+
+#define QIB_6120_TxFlowPktCnt_OFFS 0x12028
+
+#define QIB_6120_TxDwordCnt_OFFS 0x12030
+
+#define QIB_6120_TxLenErrCnt_OFFS 0x12038
+
+#define QIB_6120_TxMaxMinLenErrCnt_OFFS 0x12040
+
+#define QIB_6120_TxUnderrunCnt_OFFS 0x12048
+
+#define QIB_6120_TxFlowStallCnt_OFFS 0x12050
+
+#define QIB_6120_TxDroppedPktCnt_OFFS 0x12058
+
+#define QIB_6120_RxDroppedPktCnt_OFFS 0x12060
+
+#define QIB_6120_RxDataPktCnt_OFFS 0x12068
+
+#define QIB_6120_RxFlowPktCnt_OFFS 0x12070
+
+#define QIB_6120_RxDwordCnt_OFFS 0x12078
+
+#define QIB_6120_RxLenErrCnt_OFFS 0x12080
+
+#define QIB_6120_RxMaxMinLenErrCnt_OFFS 0x12088
+
+#define QIB_6120_RxICRCErrCnt_OFFS 0x12090
+
+#define QIB_6120_RxVCRCErrCnt_OFFS 0x12098
+
+#define QIB_6120_RxFlowCtrlErrCnt_OFFS 0x120A0
+
+#define QIB_6120_RxBadFormatCnt_OFFS 0x120A8
+
+#define QIB_6120_RxLinkProblemCnt_OFFS 0x120B0
+
+#define QIB_6120_RxEBPCnt_OFFS 0x120B8
+
+#define QIB_6120_RxLPCRCErrCnt_OFFS 0x120C0
+
+#define QIB_6120_RxBufOvflCnt_OFFS 0x120C8
+
+#define QIB_6120_RxTIDFullErrCnt_OFFS 0x120D0
+
+#define QIB_6120_RxTIDValidErrCnt_OFFS 0x120D8
+
+#define QIB_6120_RxPKeyMismatchCnt_OFFS 0x120E0
+
+#define QIB_6120_RxP0HdrEgrOvflCnt_OFFS 0x120E8
+
+#define QIB_6120_IBStatusChangeCnt_OFFS 0x12140
+
+#define QIB_6120_IBLinkErrRecoveryCnt_OFFS 0x12148
+
+#define QIB_6120_IBLinkDownedCnt_OFFS 0x12150
+
+#define QIB_6120_IBSymbolErrCnt_OFFS 0x12158
+
+#define QIB_6120_PcieRetryBufDiagQwordCnt_OFFS 0x12170
+
+#define QIB_6120_RcvEgrArray0_OFFS 0x14000
+
+#define QIB_6120_RcvTIDArray0_OFFS 0x54000
+
+#define QIB_6120_PIOLaunchFIFO_OFFS 0x64000
+
+#define QIB_6120_SendPIOpbcCache_OFFS 0x64800
+
+#define QIB_6120_RcvBuf1_OFFS 0x72000
+
+#define QIB_6120_RcvBuf2_OFFS 0x75000
+
+#define QIB_6120_RcvFlags_OFFS 0x77000
+
+#define QIB_6120_RcvLookupBuf1_OFFS 0x79000
+
+#define QIB_6120_RcvDMABuf_OFFS 0x7B000
+
+#define QIB_6120_MiscRXEIntMem_OFFS 0x7C000
+
+#define QIB_6120_PCIERcvBuf_OFFS 0x80000
+
+#define QIB_6120_PCIERetryBuf_OFFS 0x82000
+
+#define QIB_6120_PCIERcvBufRdToWrAddr_OFFS 0x84000
+
+#define QIB_6120_PIOBuf0_MA_OFFS 0x100000

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 05/51] IB/qib: Add qib_7220.h

[Ralph Campbell]

creates the qib_7220.h file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_7220.h |  158 ++++++++++++++++++++++++++++++++++
 1 files changed, 158 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_7220.h

diff --git a/drivers/infiniband/hw/qib/qib_7220.h b/drivers/infiniband/hw/qib/qib_7220.h
new file mode 100644
index 0000000..a68230e
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_7220.h
@@ -0,0 +1,158 @@
+#ifndef _QIB_7220_H
+#define _QIB_7220_H
+/*
+ * Copyright (c) 2007, 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/* grab register-defs auto-generated by HW */
+#include "qib_7220_regs.h"
+
+/* The number of eager receive TIDs for context zero. */
+#define IBA7220_KRCVEGRCNT      2048U
+
+#define IB_7220_LT_STATE_CFGRCVFCFG      0x09
+#define IB_7220_LT_STATE_CFGWAITRMT      0x0a
+#define IB_7220_LT_STATE_TXREVLANES      0x0d
+#define IB_7220_LT_STATE_CFGENH          0x10
+
+struct qib_chip_specific {
+	u64 __iomem *cregbase;
+	u64 *cntrs;
+	u64 *portcntrs;
+	spinlock_t sdepb_lock; /* serdes EPB bus */
+	spinlock_t rcvmod_lock; /* protect rcvctrl shadow changes */
+	spinlock_t gpio_lock; /* RMW of shadows/regs for ExtCtrl and GPIO */
+	u64 hwerrmask;
+	u64 errormask;
+	u64 gpio_out; /* shadow of kr_gpio_out, for rmw ops */
+	u64 gpio_mask; /* shadow the gpio mask register */
+	u64 extctrl; /* shadow the gpio output enable, etc... */
+	u32 ncntrs;
+	u32 nportcntrs;
+	u32 cntrnamelen;
+	u32 portcntrnamelen;
+	u32 numctxts;
+	u32 rcvegrcnt;
+	u32 autoneg_tries;
+	u32 serdes_first_init_done;
+	u32 sdmabufcnt;
+	u32 lastbuf_for_pio;
+	u32 updthresh; /* current AvailUpdThld */
+	u32 updthresh_dflt; /* default AvailUpdThld */
+	int irq;
+	u8 presets_needed;
+	char emsgbuf[128];
+	char sdmamsgbuf[192];
+	char bitsmsgbuf[64];
+	struct qib_relock {
+		atomic_t relock_timer_active;
+		struct timer_list relock_timer;
+		unsigned int relock_interval; /* in jiffies */
+	} relock_st;
+};
+
+struct qib_chippport_specific {
+	struct qib_pportdata pportdata;
+	wait_queue_head_t autoneg_wait;
+	struct delayed_work autoneg_work;
+	struct timer_list chase_timer;
+	/*
+	 * these 5 fields are used to establish deltas for IB symbol
+	 * errors and linkrecovery errors.  They can be reported on
+	 * some chips during link negotiation prior to INIT, and with
+	 * DDR when faking DDR negotiations with non-IBTA switches.
+	 * The chip counters are adjusted at driver unload if there is
+	 * a non-zero delta.
+	 */
+	u64 ibdeltainprog;
+	u64 ibsymdelta;
+	u64 ibsymsnap;
+	u64 iblnkerrdelta;
+	u64 iblnkerrsnap;
+	u64 ibcctrl; /* kr_ibcctrl shadow */
+	u64 ibcddrctrl; /* kr_ibcddrctrl shadow */
+	u64 chase_end;
+	u32 last_delay_mult;
+};
+
+/*
+ * This header file provides the declarations and common definitions
+ * for (mostly) manipulation of the SerDes blocks within the IBA7220.
+ * the functions declared should only be called from within other
+ * 7220-related files such as qib_iba7220.c or qib_sd7220.c.
+ */
+int qib_sd7220_presets(struct qib_devdata *dd);
+int qib_sd7220_init(struct qib_devdata *dd);
+int qib_sd7220_prog_ld(struct qib_devdata *dd, int sdnum, u8 *img,
+		       int len, int offset);
+int qib_sd7220_prog_vfy(struct qib_devdata *dd, int sdnum, const u8 *img,
+			int len, int offset);
+void qib_sd7220_clr_ibpar(struct qib_devdata *);
+/*
+ * Below used for sdnum parameter, selecting one of the two sections
+ * used for PCIe, or the single SerDes used for IB, which is the
+ * only one currently used
+ */
+#define IB_7220_SERDES 2
+
+int qib_sd7220_ib_load(struct qib_devdata *dd);
+int qib_sd7220_ib_vfy(struct qib_devdata *dd);
+
+static inline u32 qib_read_kreg32(const struct qib_devdata *dd,
+				  const u16 regno)
+{
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return -1;
+	return readl((u32 __iomem *)&dd->kregbase[regno]);
+}
+
+static inline u64 qib_read_kreg64(const struct qib_devdata *dd,
+				  const u16 regno)
+{
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return -1;
+
+	return readq(&dd->kregbase[regno]);
+}
+
+static inline void qib_write_kreg(const struct qib_devdata *dd,
+				  const u16 regno, u64 value)
+{
+	if (dd->kregbase)
+		writeq(value, &dd->kregbase[regno]);
+}
+
+void set_7220_relock_poll(struct qib_devdata *, int);
+void shutdown_7220_relock_poll(struct qib_devdata *);
+void toggle_7220_rclkrls(struct qib_devdata *);
+
+
+#endif /* _QIB_7220_H */

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 06/51] IB/qib: Add qib_7220_regs.h

[Ralph Campbell]

creates the qib_7220_regs.h file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_7220_regs.h | 1496 +++++++++++++++++++++++++++++
 1 files changed, 1496 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_7220_regs.h

diff --git a/drivers/infiniband/hw/qib/qib_7220_regs.h b/drivers/infiniband/hw/qib/qib_7220_regs.h
new file mode 100644
index 0000000..81d86e3
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_7220_regs.h
@@ -0,0 +1,1496 @@
+/*
+ * Copyright (c) 2008, 2009 QLogic Corporation. All rights reserved.
+ *
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+/* This file is mechanically generated from RTL. Any hand-edits will be lost! */
+
+#define QIB_7220_Revision_OFFS 0x0
+#define QIB_7220_Revision_R_Simulator_LSB 0x3F
+#define QIB_7220_Revision_R_Simulator_RMASK 0x1
+#define QIB_7220_Revision_R_Emulation_LSB 0x3E
+#define QIB_7220_Revision_R_Emulation_RMASK 0x1
+#define QIB_7220_Revision_R_Emulation_Revcode_LSB 0x28
+#define QIB_7220_Revision_R_Emulation_Revcode_RMASK 0x3FFFFF
+#define QIB_7220_Revision_BoardID_LSB 0x20
+#define QIB_7220_Revision_BoardID_RMASK 0xFF
+#define QIB_7220_Revision_R_SW_LSB 0x18
+#define QIB_7220_Revision_R_SW_RMASK 0xFF
+#define QIB_7220_Revision_R_Arch_LSB 0x10
+#define QIB_7220_Revision_R_Arch_RMASK 0xFF
+#define QIB_7220_Revision_R_ChipRevMajor_LSB 0x8
+#define QIB_7220_Revision_R_ChipRevMajor_RMASK 0xFF
+#define QIB_7220_Revision_R_ChipRevMinor_LSB 0x0
+#define QIB_7220_Revision_R_ChipRevMinor_RMASK 0xFF
+
+#define QIB_7220_Control_OFFS 0x8
+#define QIB_7220_Control_SyncResetExceptPcieIRAMRST_LSB 0x7
+#define QIB_7220_Control_SyncResetExceptPcieIRAMRST_RMASK 0x1
+#define QIB_7220_Control_PCIECplQDiagEn_LSB 0x6
+#define QIB_7220_Control_PCIECplQDiagEn_RMASK 0x1
+#define QIB_7220_Control_Reserved_LSB 0x5
+#define QIB_7220_Control_Reserved_RMASK 0x1
+#define QIB_7220_Control_TxLatency_LSB 0x4
+#define QIB_7220_Control_TxLatency_RMASK 0x1
+#define QIB_7220_Control_PCIERetryBufDiagEn_LSB 0x3
+#define QIB_7220_Control_PCIERetryBufDiagEn_RMASK 0x1
+#define QIB_7220_Control_LinkEn_LSB 0x2
+#define QIB_7220_Control_LinkEn_RMASK 0x1
+#define QIB_7220_Control_FreezeMode_LSB 0x1
+#define QIB_7220_Control_FreezeMode_RMASK 0x1
+#define QIB_7220_Control_SyncReset_LSB 0x0
+#define QIB_7220_Control_SyncReset_RMASK 0x1
+
+#define QIB_7220_PageAlign_OFFS 0x10
+
+#define QIB_7220_PortCnt_OFFS 0x18
+
+#define QIB_7220_SendRegBase_OFFS 0x30
+
+#define QIB_7220_UserRegBase_OFFS 0x38
+
+#define QIB_7220_CntrRegBase_OFFS 0x40
+
+#define QIB_7220_Scratch_OFFS 0x48
+
+#define QIB_7220_IntMask_OFFS 0x68
+#define QIB_7220_IntMask_SDmaIntMask_LSB 0x3F
+#define QIB_7220_IntMask_SDmaIntMask_RMASK 0x1
+#define QIB_7220_IntMask_SDmaDisabledMasked_LSB 0x3E
+#define QIB_7220_IntMask_SDmaDisabledMasked_RMASK 0x1
+#define QIB_7220_IntMask_Reserved_LSB 0x31
+#define QIB_7220_IntMask_Reserved_RMASK 0x1FFF
+#define QIB_7220_IntMask_RcvUrg16IntMask_LSB 0x30
+#define QIB_7220_IntMask_RcvUrg16IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg15IntMask_LSB 0x2F
+#define QIB_7220_IntMask_RcvUrg15IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg14IntMask_LSB 0x2E
+#define QIB_7220_IntMask_RcvUrg14IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg13IntMask_LSB 0x2D
+#define QIB_7220_IntMask_RcvUrg13IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg12IntMask_LSB 0x2C
+#define QIB_7220_IntMask_RcvUrg12IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg11IntMask_LSB 0x2B
+#define QIB_7220_IntMask_RcvUrg11IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg10IntMask_LSB 0x2A
+#define QIB_7220_IntMask_RcvUrg10IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg9IntMask_LSB 0x29
+#define QIB_7220_IntMask_RcvUrg9IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg8IntMask_LSB 0x28
+#define QIB_7220_IntMask_RcvUrg8IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg7IntMask_LSB 0x27
+#define QIB_7220_IntMask_RcvUrg7IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg6IntMask_LSB 0x26
+#define QIB_7220_IntMask_RcvUrg6IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg5IntMask_LSB 0x25
+#define QIB_7220_IntMask_RcvUrg5IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg4IntMask_LSB 0x24
+#define QIB_7220_IntMask_RcvUrg4IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg3IntMask_LSB 0x23
+#define QIB_7220_IntMask_RcvUrg3IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg2IntMask_LSB 0x22
+#define QIB_7220_IntMask_RcvUrg2IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg1IntMask_LSB 0x21
+#define QIB_7220_IntMask_RcvUrg1IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg0IntMask_LSB 0x20
+#define QIB_7220_IntMask_RcvUrg0IntMask_RMASK 0x1
+#define QIB_7220_IntMask_ErrorIntMask_LSB 0x1F
+#define QIB_7220_IntMask_ErrorIntMask_RMASK 0x1
+#define QIB_7220_IntMask_PioSetIntMask_LSB 0x1E
+#define QIB_7220_IntMask_PioSetIntMask_RMASK 0x1
+#define QIB_7220_IntMask_PioBufAvailIntMask_LSB 0x1D
+#define QIB_7220_IntMask_PioBufAvailIntMask_RMASK 0x1
+#define QIB_7220_IntMask_assertGPIOIntMask_LSB 0x1C
+#define QIB_7220_IntMask_assertGPIOIntMask_RMASK 0x1
+#define QIB_7220_IntMask_IBSerdesTrimDoneIntMask_LSB 0x1B
+#define QIB_7220_IntMask_IBSerdesTrimDoneIntMask_RMASK 0x1
+#define QIB_7220_IntMask_JIntMask_LSB 0x1A
+#define QIB_7220_IntMask_JIntMask_RMASK 0x1
+#define QIB_7220_IntMask_Reserved1_LSB 0x11
+#define QIB_7220_IntMask_Reserved1_RMASK 0x1FF
+#define QIB_7220_IntMask_RcvAvail16IntMask_LSB 0x10
+#define QIB_7220_IntMask_RcvAvail16IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail15IntMask_LSB 0xF
+#define QIB_7220_IntMask_RcvAvail15IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail14IntMask_LSB 0xE
+#define QIB_7220_IntMask_RcvAvail14IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail13IntMask_LSB 0xD
+#define QIB_7220_IntMask_RcvAvail13IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail12IntMask_LSB 0xC
+#define QIB_7220_IntMask_RcvAvail12IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail11IntMask_LSB 0xB
+#define QIB_7220_IntMask_RcvAvail11IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail10IntMask_LSB 0xA
+#define QIB_7220_IntMask_RcvAvail10IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail9IntMask_LSB 0x9
+#define QIB_7220_IntMask_RcvAvail9IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail8IntMask_LSB 0x8
+#define QIB_7220_IntMask_RcvAvail8IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail7IntMask_LSB 0x7
+#define QIB_7220_IntMask_RcvAvail7IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail6IntMask_LSB 0x6
+#define QIB_7220_IntMask_RcvAvail6IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail5IntMask_LSB 0x5
+#define QIB_7220_IntMask_RcvAvail5IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail4IntMask_LSB 0x4
+#define QIB_7220_IntMask_RcvAvail4IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail3IntMask_LSB 0x3
+#define QIB_7220_IntMask_RcvAvail3IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail2IntMask_LSB 0x2
+#define QIB_7220_IntMask_RcvAvail2IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail1IntMask_LSB 0x1
+#define QIB_7220_IntMask_RcvAvail1IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail0IntMask_LSB 0x0
+#define QIB_7220_IntMask_RcvAvail0IntMask_RMASK 0x1
+
+#define QIB_7220_IntStatus_OFFS 0x70
+#define QIB_7220_IntStatus_SDmaInt_LSB 0x3F
+#define QIB_7220_IntStatus_SDmaInt_RMASK 0x1
+#define QIB_7220_IntStatus_SDmaDisabled_LSB 0x3E
+#define QIB_7220_IntStatus_SDmaDisabled_RMASK 0x1
+#define QIB_7220_IntStatus_Reserved_LSB 0x31
+#define QIB_7220_IntStatus_Reserved_RMASK 0x1FFF
+#define QIB_7220_IntStatus_RcvUrg16_LSB 0x30
+#define QIB_7220_IntStatus_RcvUrg16_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg15_LSB 0x2F
+#define QIB_7220_IntStatus_RcvUrg15_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg14_LSB 0x2E
+#define QIB_7220_IntStatus_RcvUrg14_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg13_LSB 0x2D
+#define QIB_7220_IntStatus_RcvUrg13_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg12_LSB 0x2C
+#define QIB_7220_IntStatus_RcvUrg12_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg11_LSB 0x2B
+#define QIB_7220_IntStatus_RcvUrg11_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg10_LSB 0x2A
+#define QIB_7220_IntStatus_RcvUrg10_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg9_LSB 0x29
+#define QIB_7220_IntStatus_RcvUrg9_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg8_LSB 0x28
+#define QIB_7220_IntStatus_RcvUrg8_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg7_LSB 0x27
+#define QIB_7220_IntStatus_RcvUrg7_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg6_LSB 0x26
+#define QIB_7220_IntStatus_RcvUrg6_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg5_LSB 0x25
+#define QIB_7220_IntStatus_RcvUrg5_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg4_LSB 0x24
+#define QIB_7220_IntStatus_RcvUrg4_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg3_LSB 0x23
+#define QIB_7220_IntStatus_RcvUrg3_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg2_LSB 0x22
+#define QIB_7220_IntStatus_RcvUrg2_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg1_LSB 0x21
+#define QIB_7220_IntStatus_RcvUrg1_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg0_LSB 0x20
+#define QIB_7220_IntStatus_RcvUrg0_RMASK 0x1
+#define QIB_7220_IntStatus_Error_LSB 0x1F
+#define QIB_7220_IntStatus_Error_RMASK 0x1
+#define QIB_7220_IntStatus_PioSent_LSB 0x1E
+#define QIB_7220_IntStatus_PioSent_RMASK 0x1
+#define QIB_7220_IntStatus_PioBufAvail_LSB 0x1D
+#define QIB_7220_IntStatus_PioBufAvail_RMASK 0x1
+#define QIB_7220_IntStatus_assertGPIO_LSB 0x1C
+#define QIB_7220_IntStatus_assertGPIO_RMASK 0x1
+#define QIB_7220_IntStatus_IBSerdesTrimDone_LSB 0x1B
+#define QIB_7220_IntStatus_IBSerdesTrimDone_RMASK 0x1
+#define QIB_7220_IntStatus_JInt_LSB 0x1A
+#define QIB_7220_IntStatus_JInt_RMASK 0x1
+#define QIB_7220_IntStatus_Reserved1_LSB 0x11
+#define QIB_7220_IntStatus_Reserved1_RMASK 0x1FF
+#define QIB_7220_IntStatus_RcvAvail16_LSB 0x10
+#define QIB_7220_IntStatus_RcvAvail16_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail15_LSB 0xF
+#define QIB_7220_IntStatus_RcvAvail15_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail14_LSB 0xE
+#define QIB_7220_IntStatus_RcvAvail14_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail13_LSB 0xD
+#define QIB_7220_IntStatus_RcvAvail13_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail12_LSB 0xC
+#define QIB_7220_IntStatus_RcvAvail12_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail11_LSB 0xB
+#define QIB_7220_IntStatus_RcvAvail11_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail10_LSB 0xA
+#define QIB_7220_IntStatus_RcvAvail10_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail9_LSB 0x9
+#define QIB_7220_IntStatus_RcvAvail9_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail8_LSB 0x8
+#define QIB_7220_IntStatus_RcvAvail8_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail7_LSB 0x7
+#define QIB_7220_IntStatus_RcvAvail7_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail6_LSB 0x6
+#define QIB_7220_IntStatus_RcvAvail6_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail5_LSB 0x5
+#define QIB_7220_IntStatus_RcvAvail5_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail4_LSB 0x4
+#define QIB_7220_IntStatus_RcvAvail4_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail3_LSB 0x3
+#define QIB_7220_IntStatus_RcvAvail3_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail2_LSB 0x2
+#define QIB_7220_IntStatus_RcvAvail2_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail1_LSB 0x1
+#define QIB_7220_IntStatus_RcvAvail1_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail0_LSB 0x0
+#define QIB_7220_IntStatus_RcvAvail0_RMASK 0x1
+
+#define QIB_7220_IntClear_OFFS 0x78
+#define QIB_7220_IntClear_SDmaIntClear_LSB 0x3F
+#define QIB_7220_IntClear_SDmaIntClear_RMASK 0x1
+#define QIB_7220_IntClear_SDmaDisabledClear_LSB 0x3E
+#define QIB_7220_IntClear_SDmaDisabledClear_RMASK 0x1
+#define QIB_7220_IntClear_Reserved_LSB 0x31
+#define QIB_7220_IntClear_Reserved_RMASK 0x1FFF
+#define QIB_7220_IntClear_RcvUrg16IntClear_LSB 0x30
+#define QIB_7220_IntClear_RcvUrg16IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg15IntClear_LSB 0x2F
+#define QIB_7220_IntClear_RcvUrg15IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg14IntClear_LSB 0x2E
+#define QIB_7220_IntClear_RcvUrg14IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg13IntClear_LSB 0x2D
+#define QIB_7220_IntClear_RcvUrg13IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg12IntClear_LSB 0x2C
+#define QIB_7220_IntClear_RcvUrg12IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg11IntClear_LSB 0x2B
+#define QIB_7220_IntClear_RcvUrg11IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg10IntClear_LSB 0x2A
+#define QIB_7220_IntClear_RcvUrg10IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg9IntClear_LSB 0x29
+#define QIB_7220_IntClear_RcvUrg9IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg8IntClear_LSB 0x28
+#define QIB_7220_IntClear_RcvUrg8IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg7IntClear_LSB 0x27
+#define QIB_7220_IntClear_RcvUrg7IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg6IntClear_LSB 0x26
+#define QIB_7220_IntClear_RcvUrg6IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg5IntClear_LSB 0x25
+#define QIB_7220_IntClear_RcvUrg5IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg4IntClear_LSB 0x24
+#define QIB_7220_IntClear_RcvUrg4IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg3IntClear_LSB 0x23
+#define QIB_7220_IntClear_RcvUrg3IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg2IntClear_LSB 0x22
+#define QIB_7220_IntClear_RcvUrg2IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg1IntClear_LSB 0x21
+#define QIB_7220_IntClear_RcvUrg1IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg0IntClear_LSB 0x20
+#define QIB_7220_IntClear_RcvUrg0IntClear_RMASK 0x1
+#define QIB_7220_IntClear_ErrorIntClear_LSB 0x1F
+#define QIB_7220_IntClear_ErrorIntClear_RMASK 0x1
+#define QIB_7220_IntClear_PioSetIntClear_LSB 0x1E
+#define QIB_7220_IntClear_PioSetIntClear_RMASK 0x1
+#define QIB_7220_IntClear_PioBufAvailIntClear_LSB 0x1D
+#define QIB_7220_IntClear_PioBufAvailIntClear_RMASK 0x1
+#define QIB_7220_IntClear_assertGPIOIntClear_LSB 0x1C
+#define QIB_7220_IntClear_assertGPIOIntClear_RMASK 0x1
+#define QIB_7220_IntClear_IBSerdesTrimDoneClear_LSB 0x1B
+#define QIB_7220_IntClear_IBSerdesTrimDoneClear_RMASK 0x1
+#define QIB_7220_IntClear_JIntClear_LSB 0x1A
+#define QIB_7220_IntClear_JIntClear_RMASK 0x1
+#define QIB_7220_IntClear_Reserved1_LSB 0x11
+#define QIB_7220_IntClear_Reserved1_RMASK 0x1FF
+#define QIB_7220_IntClear_RcvAvail16IntClear_LSB 0x10
+#define QIB_7220_IntClear_RcvAvail16IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail15IntClear_LSB 0xF
+#define QIB_7220_IntClear_RcvAvail15IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail14IntClear_LSB 0xE
+#define QIB_7220_IntClear_RcvAvail14IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail13IntClear_LSB 0xD
+#define QIB_7220_IntClear_RcvAvail13IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail12IntClear_LSB 0xC
+#define QIB_7220_IntClear_RcvAvail12IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail11IntClear_LSB 0xB
+#define QIB_7220_IntClear_RcvAvail11IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail10IntClear_LSB 0xA
+#define QIB_7220_IntClear_RcvAvail10IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail9IntClear_LSB 0x9
+#define QIB_7220_IntClear_RcvAvail9IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail8IntClear_LSB 0x8
+#define QIB_7220_IntClear_RcvAvail8IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail7IntClear_LSB 0x7
+#define QIB_7220_IntClear_RcvAvail7IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail6IntClear_LSB 0x6
+#define QIB_7220_IntClear_RcvAvail6IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail5IntClear_LSB 0x5
+#define QIB_7220_IntClear_RcvAvail5IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail4IntClear_LSB 0x4
+#define QIB_7220_IntClear_RcvAvail4IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail3IntClear_LSB 0x3
+#define QIB_7220_IntClear_RcvAvail3IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail2IntClear_LSB 0x2
+#define QIB_7220_IntClear_RcvAvail2IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail1IntClear_LSB 0x1
+#define QIB_7220_IntClear_RcvAvail1IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail0IntClear_LSB 0x0
+#define QIB_7220_IntClear_RcvAvail0IntClear_RMASK 0x1
+
+#define QIB_7220_ErrMask_OFFS 0x80
+#define QIB_7220_ErrMask_Reserved_LSB 0x36
+#define QIB_7220_ErrMask_Reserved_RMASK 0x3FF
+#define QIB_7220_ErrMask_InvalidEEPCmdMask_LSB 0x35
+#define QIB_7220_ErrMask_InvalidEEPCmdMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaDescAddrMisalignErrMask_LSB 0x34
+#define QIB_7220_ErrMask_SDmaDescAddrMisalignErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_HardwareErrMask_LSB 0x33
+#define QIB_7220_ErrMask_HardwareErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_ResetNegatedMask_LSB 0x32
+#define QIB_7220_ErrMask_ResetNegatedMask_RMASK 0x1
+#define QIB_7220_ErrMask_InvalidAddrErrMask_LSB 0x31
+#define QIB_7220_ErrMask_InvalidAddrErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_IBStatusChangedMask_LSB 0x30
+#define QIB_7220_ErrMask_IBStatusChangedMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaUnexpDataErrMask_LSB 0x2F
+#define QIB_7220_ErrMask_SDmaUnexpDataErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaMissingDwErrMask_LSB 0x2E
+#define QIB_7220_ErrMask_SDmaMissingDwErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaDwEnErrMask_LSB 0x2D
+#define QIB_7220_ErrMask_SDmaDwEnErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaRpyTagErrMask_LSB 0x2C
+#define QIB_7220_ErrMask_SDmaRpyTagErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDma1stDescErrMask_LSB 0x2B
+#define QIB_7220_ErrMask_SDma1stDescErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaBaseErrMask_LSB 0x2A
+#define QIB_7220_ErrMask_SDmaBaseErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaTailOutOfBoundErrMask_LSB 0x29
+#define QIB_7220_ErrMask_SDmaTailOutOfBoundErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaOutOfBoundErrMask_LSB 0x28
+#define QIB_7220_ErrMask_SDmaOutOfBoundErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaGenMismatchErrMask_LSB 0x27
+#define QIB_7220_ErrMask_SDmaGenMismatchErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendBufMisuseErrMask_LSB 0x26
+#define QIB_7220_ErrMask_SendBufMisuseErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendUnsupportedVLErrMask_LSB 0x25
+#define QIB_7220_ErrMask_SendUnsupportedVLErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendUnexpectedPktNumErrMask_LSB 0x24
+#define QIB_7220_ErrMask_SendUnexpectedPktNumErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendPioArmLaunchErrMask_LSB 0x23
+#define QIB_7220_ErrMask_SendPioArmLaunchErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendDroppedDataPktErrMask_LSB 0x22
+#define QIB_7220_ErrMask_SendDroppedDataPktErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendDroppedSmpPktErrMask_LSB 0x21
+#define QIB_7220_ErrMask_SendDroppedSmpPktErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendPktLenErrMask_LSB 0x20
+#define QIB_7220_ErrMask_SendPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendUnderRunErrMask_LSB 0x1F
+#define QIB_7220_ErrMask_SendUnderRunErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendMaxPktLenErrMask_LSB 0x1E
+#define QIB_7220_ErrMask_SendMaxPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendMinPktLenErrMask_LSB 0x1D
+#define QIB_7220_ErrMask_SendMinPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaDisabledErrMask_LSB 0x1C
+#define QIB_7220_ErrMask_SDmaDisabledErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendSpecialTriggerErrMask_LSB 0x1B
+#define QIB_7220_ErrMask_SendSpecialTriggerErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_Reserved1_LSB 0x12
+#define QIB_7220_ErrMask_Reserved1_RMASK 0x1FF
+#define QIB_7220_ErrMask_RcvIBLostLinkErrMask_LSB 0x11
+#define QIB_7220_ErrMask_RcvIBLostLinkErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvHdrErrMask_LSB 0x10
+#define QIB_7220_ErrMask_RcvHdrErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvHdrLenErrMask_LSB 0xF
+#define QIB_7220_ErrMask_RcvHdrLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvBadTidErrMask_LSB 0xE
+#define QIB_7220_ErrMask_RcvBadTidErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvHdrFullErrMask_LSB 0xD
+#define QIB_7220_ErrMask_RcvHdrFullErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvEgrFullErrMask_LSB 0xC
+#define QIB_7220_ErrMask_RcvEgrFullErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvBadVersionErrMask_LSB 0xB
+#define QIB_7220_ErrMask_RcvBadVersionErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvIBFlowErrMask_LSB 0xA
+#define QIB_7220_ErrMask_RcvIBFlowErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvEBPErrMask_LSB 0x9
+#define QIB_7220_ErrMask_RcvEBPErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvUnsupportedVLErrMask_LSB 0x8
+#define QIB_7220_ErrMask_RcvUnsupportedVLErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvUnexpectedCharErrMask_LSB 0x7
+#define QIB_7220_ErrMask_RcvUnexpectedCharErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvShortPktLenErrMask_LSB 0x6
+#define QIB_7220_ErrMask_RcvShortPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvLongPktLenErrMask_LSB 0x5
+#define QIB_7220_ErrMask_RcvLongPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvMaxPktLenErrMask_LSB 0x4
+#define QIB_7220_ErrMask_RcvMaxPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvMinPktLenErrMask_LSB 0x3
+#define QIB_7220_ErrMask_RcvMinPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvICRCErrMask_LSB 0x2
+#define QIB_7220_ErrMask_RcvICRCErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvVCRCErrMask_LSB 0x1
+#define QIB_7220_ErrMask_RcvVCRCErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvFormatErrMask_LSB 0x0
+#define QIB_7220_ErrMask_RcvFormatErrMask_RMASK 0x1
+
+#define QIB_7220_ErrStatus_OFFS 0x88
+#define QIB_7220_ErrStatus_Reserved_LSB 0x36
+#define QIB_7220_ErrStatus_Reserved_RMASK 0x3FF
+#define QIB_7220_ErrStatus_InvalidEEPCmdErr_LSB 0x35
+#define QIB_7220_ErrStatus_InvalidEEPCmdErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaDescAddrMisalignErr_LSB 0x34
+#define QIB_7220_ErrStatus_SDmaDescAddrMisalignErr_RMASK 0x1
+#define QIB_7220_ErrStatus_HardwareErr_LSB 0x33
+#define QIB_7220_ErrStatus_HardwareErr_RMASK 0x1
+#define QIB_7220_ErrStatus_ResetNegated_LSB 0x32
+#define QIB_7220_ErrStatus_ResetNegated_RMASK 0x1
+#define QIB_7220_ErrStatus_InvalidAddrErr_LSB 0x31
+#define QIB_7220_ErrStatus_InvalidAddrErr_RMASK 0x1
+#define QIB_7220_ErrStatus_IBStatusChanged_LSB 0x30
+#define QIB_7220_ErrStatus_IBStatusChanged_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaUnexpDataErr_LSB 0x2F
+#define QIB_7220_ErrStatus_SDmaUnexpDataErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaMissingDwErr_LSB 0x2E
+#define QIB_7220_ErrStatus_SDmaMissingDwErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaDwEnErr_LSB 0x2D
+#define QIB_7220_ErrStatus_SDmaDwEnErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaRpyTagErr_LSB 0x2C
+#define QIB_7220_ErrStatus_SDmaRpyTagErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDma1stDescErr_LSB 0x2B
+#define QIB_7220_ErrStatus_SDma1stDescErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaBaseErr_LSB 0x2A
+#define QIB_7220_ErrStatus_SDmaBaseErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaTailOutOfBoundErr_LSB 0x29
+#define QIB_7220_ErrStatus_SDmaTailOutOfBoundErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaOutOfBoundErr_LSB 0x28
+#define QIB_7220_ErrStatus_SDmaOutOfBoundErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaGenMismatchErr_LSB 0x27
+#define QIB_7220_ErrStatus_SDmaGenMismatchErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendBufMisuseErr_LSB 0x26
+#define QIB_7220_ErrStatus_SendBufMisuseErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendUnsupportedVLErr_LSB 0x25
+#define QIB_7220_ErrStatus_SendUnsupportedVLErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendUnexpectedPktNumErr_LSB 0x24
+#define QIB_7220_ErrStatus_SendUnexpectedPktNumErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendPioArmLaunchErr_LSB 0x23
+#define QIB_7220_ErrStatus_SendPioArmLaunchErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendDroppedDataPktErr_LSB 0x22
+#define QIB_7220_ErrStatus_SendDroppedDataPktErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendDroppedSmpPktErr_LSB 0x21
+#define QIB_7220_ErrStatus_SendDroppedSmpPktErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendPktLenErr_LSB 0x20
+#define QIB_7220_ErrStatus_SendPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendUnderRunErr_LSB 0x1F
+#define QIB_7220_ErrStatus_SendUnderRunErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendMaxPktLenErr_LSB 0x1E
+#define QIB_7220_ErrStatus_SendMaxPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendMinPktLenErr_LSB 0x1D
+#define QIB_7220_ErrStatus_SendMinPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaDisabledErr_LSB 0x1C
+#define QIB_7220_ErrStatus_SDmaDisabledErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendSpecialTriggerErr_LSB 0x1B
+#define QIB_7220_ErrStatus_SendSpecialTriggerErr_RMASK 0x1
+#define QIB_7220_ErrStatus_Reserved1_LSB 0x12
+#define QIB_7220_ErrStatus_Reserved1_RMASK 0x1FF
+#define QIB_7220_ErrStatus_RcvIBLostLinkErr_LSB 0x11
+#define QIB_7220_ErrStatus_RcvIBLostLinkErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvHdrErr_LSB 0x10
+#define QIB_7220_ErrStatus_RcvHdrErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvHdrLenErr_LSB 0xF
+#define QIB_7220_ErrStatus_RcvHdrLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvBadTidErr_LSB 0xE
+#define QIB_7220_ErrStatus_RcvBadTidErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvHdrFullErr_LSB 0xD
+#define QIB_7220_ErrStatus_RcvHdrFullErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvEgrFullErr_LSB 0xC
+#define QIB_7220_ErrStatus_RcvEgrFullErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvBadVersionErr_LSB 0xB
+#define QIB_7220_ErrStatus_RcvBadVersionErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvIBFlowErr_LSB 0xA
+#define QIB_7220_ErrStatus_RcvIBFlowErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvEBPErr_LSB 0x9
+#define QIB_7220_ErrStatus_RcvEBPErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvUnsupportedVLErr_LSB 0x8
+#define QIB_7220_ErrStatus_RcvUnsupportedVLErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvUnexpectedCharErr_LSB 0x7
+#define QIB_7220_ErrStatus_RcvUnexpectedCharErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvShortPktLenErr_LSB 0x6
+#define QIB_7220_ErrStatus_RcvShortPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvLongPktLenErr_LSB 0x5
+#define QIB_7220_ErrStatus_RcvLongPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvMaxPktLenErr_LSB 0x4
+#define QIB_7220_ErrStatus_RcvMaxPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvMinPktLenErr_LSB 0x3
+#define QIB_7220_ErrStatus_RcvMinPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvICRCErr_LSB 0x2
+#define QIB_7220_ErrStatus_RcvICRCErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvVCRCErr_LSB 0x1
+#define QIB_7220_ErrStatus_RcvVCRCErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvFormatErr_LSB 0x0
+#define QIB_7220_ErrStatus_RcvFormatErr_RMASK 0x1
+
+#define QIB_7220_ErrClear_OFFS 0x90
+#define QIB_7220_ErrClear_Reserved_LSB 0x36
+#define QIB_7220_ErrClear_Reserved_RMASK 0x3FF
+#define QIB_7220_ErrClear_InvalidEEPCmdErrClear_LSB 0x35
+#define QIB_7220_ErrClear_InvalidEEPCmdErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaDescAddrMisalignErrClear_LSB 0x34
+#define QIB_7220_ErrClear_SDmaDescAddrMisalignErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_HardwareErrClear_LSB 0x33
+#define QIB_7220_ErrClear_HardwareErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_ResetNegatedClear_LSB 0x32
+#define QIB_7220_ErrClear_ResetNegatedClear_RMASK 0x1
+#define QIB_7220_ErrClear_InvalidAddrErrClear_LSB 0x31
+#define QIB_7220_ErrClear_InvalidAddrErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_IBStatusChangedClear_LSB 0x30
+#define QIB_7220_ErrClear_IBStatusChangedClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaUnexpDataErrClear_LSB 0x2F
+#define QIB_7220_ErrClear_SDmaUnexpDataErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaMissingDwErrClear_LSB 0x2E
+#define QIB_7220_ErrClear_SDmaMissingDwErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaDwEnErrClear_LSB 0x2D
+#define QIB_7220_ErrClear_SDmaDwEnErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaRpyTagErrClear_LSB 0x2C
+#define QIB_7220_ErrClear_SDmaRpyTagErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDma1stDescErrClear_LSB 0x2B
+#define QIB_7220_ErrClear_SDma1stDescErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaBaseErrClear_LSB 0x2A
+#define QIB_7220_ErrClear_SDmaBaseErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaTailOutOfBoundErrClear_LSB 0x29
+#define QIB_7220_ErrClear_SDmaTailOutOfBoundErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaOutOfBoundErrClear_LSB 0x28
+#define QIB_7220_ErrClear_SDmaOutOfBoundErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaGenMismatchErrClear_LSB 0x27
+#define QIB_7220_ErrClear_SDmaGenMismatchErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendBufMisuseErrClear_LSB 0x26
+#define QIB_7220_ErrClear_SendBufMisuseErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendUnsupportedVLErrClear_LSB 0x25
+#define QIB_7220_ErrClear_SendUnsupportedVLErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendUnexpectedPktNumErrClear_LSB 0x24
+#define QIB_7220_ErrClear_SendUnexpectedPktNumErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendPioArmLaunchErrClear_LSB 0x23
+#define QIB_7220_ErrClear_SendPioArmLaunchErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendDroppedDataPktErrClear_LSB 0x22
+#define QIB_7220_ErrClear_SendDroppedDataPktErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendDroppedSmpPktErrClear_LSB 0x21
+#define QIB_7220_ErrClear_SendDroppedSmpPktErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendPktLenErrClear_LSB 0x20
+#define QIB_7220_ErrClear_SendPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendUnderRunErrClear_LSB 0x1F
+#define QIB_7220_ErrClear_SendUnderRunErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendMaxPktLenErrClear_LSB 0x1E
+#define QIB_7220_ErrClear_SendMaxPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendMinPktLenErrClear_LSB 0x1D
+#define QIB_7220_ErrClear_SendMinPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaDisabledErrClear_LSB 0x1C
+#define QIB_7220_ErrClear_SDmaDisabledErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendSpecialTriggerErrClear_LSB 0x1B
+#define QIB_7220_ErrClear_SendSpecialTriggerErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_Reserved1_LSB 0x12
+#define QIB_7220_ErrClear_Reserved1_RMASK 0x1FF
+#define QIB_7220_ErrClear_RcvIBLostLinkErrClear_LSB 0x11
+#define QIB_7220_ErrClear_RcvIBLostLinkErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvHdrErrClear_LSB 0x10
+#define QIB_7220_ErrClear_RcvHdrErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvHdrLenErrClear_LSB 0xF
+#define QIB_7220_ErrClear_RcvHdrLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvBadTidErrClear_LSB 0xE
+#define QIB_7220_ErrClear_RcvBadTidErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvHdrFullErrClear_LSB 0xD
+#define QIB_7220_ErrClear_RcvHdrFullErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvEgrFullErrClear_LSB 0xC
+#define QIB_7220_ErrClear_RcvEgrFullErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvBadVersionErrClear_LSB 0xB
+#define QIB_7220_ErrClear_RcvBadVersionErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvIBFlowErrClear_LSB 0xA
+#define QIB_7220_ErrClear_RcvIBFlowErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvEBPErrClear_LSB 0x9
+#define QIB_7220_ErrClear_RcvEBPErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvUnsupportedVLErrClear_LSB 0x8
+#define QIB_7220_ErrClear_RcvUnsupportedVLErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvUnexpectedCharErrClear_LSB 0x7
+#define QIB_7220_ErrClear_RcvUnexpectedCharErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvShortPktLenErrClear_LSB 0x6
+#define QIB_7220_ErrClear_RcvShortPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvLongPktLenErrClear_LSB 0x5
+#define QIB_7220_ErrClear_RcvLongPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvMaxPktLenErrClear_LSB 0x4
+#define QIB_7220_ErrClear_RcvMaxPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvMinPktLenErrClear_LSB 0x3
+#define QIB_7220_ErrClear_RcvMinPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvICRCErrClear_LSB 0x2
+#define QIB_7220_ErrClear_RcvICRCErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvVCRCErrClear_LSB 0x1
+#define QIB_7220_ErrClear_RcvVCRCErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvFormatErrClear_LSB 0x0
+#define QIB_7220_ErrClear_RcvFormatErrClear_RMASK 0x1
+
+#define QIB_7220_HwErrMask_OFFS 0x98
+#define QIB_7220_HwErrMask_IBCBusFromSPCParityErrMask_LSB 0x3F
+#define QIB_7220_HwErrMask_IBCBusFromSPCParityErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_IBCBusToSPCParityErrMask_LSB 0x3E
+#define QIB_7220_HwErrMask_IBCBusToSPCParityErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_Clk_uC_PLLNotLockedMask_LSB 0x3D
+#define QIB_7220_HwErrMask_Clk_uC_PLLNotLockedMask_RMASK 0x1
+#define QIB_7220_HwErrMask_IBSerdesPClkNotDetectMask_LSB 0x3C
+#define QIB_7220_HwErrMask_IBSerdesPClkNotDetectMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PCIESerdesQ3PClkNotDetectMask_LSB 0x3B
+#define QIB_7220_HwErrMask_PCIESerdesQ3PClkNotDetectMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PCIESerdesQ2PClkNotDetectMask_LSB 0x3A
+#define QIB_7220_HwErrMask_PCIESerdesQ2PClkNotDetectMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PCIESerdesQ1PClkNotDetectMask_LSB 0x39
+#define QIB_7220_HwErrMask_PCIESerdesQ1PClkNotDetectMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PCIESerdesQ0PClkNotDetectMask_LSB 0x38
+#define QIB_7220_HwErrMask_PCIESerdesQ0PClkNotDetectMask_RMASK 0x1
+#define QIB_7220_HwErrMask_Reserved_LSB 0x37
+#define QIB_7220_HwErrMask_Reserved_RMASK 0x1
+#define QIB_7220_HwErrMask_PowerOnBISTFailedMask_LSB 0x36
+#define QIB_7220_HwErrMask_PowerOnBISTFailedMask_RMASK 0x1
+#define QIB_7220_HwErrMask_Reserved1_LSB 0x33
+#define QIB_7220_HwErrMask_Reserved1_RMASK 0x7
+#define QIB_7220_HwErrMask_RXEMemParityErrMask_LSB 0x2C
+#define QIB_7220_HwErrMask_RXEMemParityErrMask_RMASK 0x7F
+#define QIB_7220_HwErrMask_TXEMemParityErrMask_LSB 0x28
+#define QIB_7220_HwErrMask_TXEMemParityErrMask_RMASK 0xF
+#define QIB_7220_HwErrMask_DDSRXEQMemoryParityErrMask_LSB 0x27
+#define QIB_7220_HwErrMask_DDSRXEQMemoryParityErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_IB_uC_MemoryParityErrMask_LSB 0x26
+#define QIB_7220_HwErrMask_IB_uC_MemoryParityErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PCIEOct1_uC_MemoryParityErrMask_LSB 0x25
+#define QIB_7220_HwErrMask_PCIEOct1_uC_MemoryParityErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PCIEOct0_uC_MemoryParityErrMask_LSB 0x24
+#define QIB_7220_HwErrMask_PCIEOct0_uC_MemoryParityErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_Reserved2_LSB 0x22
+#define QIB_7220_HwErrMask_Reserved2_RMASK 0x3
+#define QIB_7220_HwErrMask_PCIeBusParityErrMask_LSB 0x1F
+#define QIB_7220_HwErrMask_PCIeBusParityErrMask_RMASK 0x7
+#define QIB_7220_HwErrMask_PcieCplTimeoutMask_LSB 0x1E
+#define QIB_7220_HwErrMask_PcieCplTimeoutMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PoisonedTLPMask_LSB 0x1D
+#define QIB_7220_HwErrMask_PoisonedTLPMask_RMASK 0x1
+#define QIB_7220_HwErrMask_SDmaMemReadErrMask_LSB 0x1C
+#define QIB_7220_HwErrMask_SDmaMemReadErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_Reserved3_LSB 0x8
+#define QIB_7220_HwErrMask_Reserved3_RMASK 0xFFFFF
+#define QIB_7220_HwErrMask_PCIeMemParityErrMask_LSB 0x0
+#define QIB_7220_HwErrMask_PCIeMemParityErrMask_RMASK 0xFF
+
+#define QIB_7220_HwErrStatus_OFFS 0xA0
+#define QIB_7220_HwErrStatus_IBCBusFromSPCParityErr_LSB 0x3F
+#define QIB_7220_HwErrStatus_IBCBusFromSPCParityErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_IBCBusToSPCParityErr_LSB 0x3E
+#define QIB_7220_HwErrStatus_IBCBusToSPCParityErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_Clk_uC_PLLNotLocked_LSB 0x3D
+#define QIB_7220_HwErrStatus_Clk_uC_PLLNotLocked_RMASK 0x1
+#define QIB_7220_HwErrStatus_IBSerdesPClkNotDetect_LSB 0x3C
+#define QIB_7220_HwErrStatus_IBSerdesPClkNotDetect_RMASK 0x1
+#define QIB_7220_HwErrStatus_PCIESerdesQ3PClkNotDetect_LSB 0x3B
+#define QIB_7220_HwErrStatus_PCIESerdesQ3PClkNotDetect_RMASK 0x1
+#define QIB_7220_HwErrStatus_PCIESerdesQ2PClkNotDetect_LSB 0x3A
+#define QIB_7220_HwErrStatus_PCIESerdesQ2PClkNotDetect_RMASK 0x1
+#define QIB_7220_HwErrStatus_PCIESerdesQ1PClkNotDetect_LSB 0x39
+#define QIB_7220_HwErrStatus_PCIESerdesQ1PClkNotDetect_RMASK 0x1
+#define QIB_7220_HwErrStatus_PCIESerdesQ0PClkNotDetect_LSB 0x38
+#define QIB_7220_HwErrStatus_PCIESerdesQ0PClkNotDetect_RMASK 0x1
+#define QIB_7220_HwErrStatus_Reserved_LSB 0x37
+#define QIB_7220_HwErrStatus_Reserved_RMASK 0x1
+#define QIB_7220_HwErrStatus_PowerOnBISTFailed_LSB 0x36
+#define QIB_7220_HwErrStatus_PowerOnBISTFailed_RMASK 0x1
+#define QIB_7220_HwErrStatus_Reserved1_LSB 0x33
+#define QIB_7220_HwErrStatus_Reserved1_RMASK 0x7
+#define QIB_7220_HwErrStatus_RXEMemParity_LSB 0x2C
+#define QIB_7220_HwErrStatus_RXEMemParity_RMASK 0x7F
+#define QIB_7220_HwErrStatus_TXEMemParity_LSB 0x28
+#define QIB_7220_HwErrStatus_TXEMemParity_RMASK 0xF
+#define QIB_7220_HwErrStatus_DDSRXEQMemoryParityErr_LSB 0x27
+#define QIB_7220_HwErrStatus_DDSRXEQMemoryParityErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_IB_uC_MemoryParityErr_LSB 0x26
+#define QIB_7220_HwErrStatus_IB_uC_MemoryParityErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_PCIE_uC_Oct1MemoryParityErr_LSB 0x25
+#define QIB_7220_HwErrStatus_PCIE_uC_Oct1MemoryParityErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_PCIE_uC_Oct0MemoryParityErr_LSB 0x24
+#define QIB_7220_HwErrStatus_PCIE_uC_Oct0MemoryParityErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_Reserved2_LSB 0x22
+#define QIB_7220_HwErrStatus_Reserved2_RMASK 0x3
+#define QIB_7220_HwErrStatus_PCIeBusParity_LSB 0x1F
+#define QIB_7220_HwErrStatus_PCIeBusParity_RMASK 0x7
+#define QIB_7220_HwErrStatus_PcieCplTimeout_LSB 0x1E
+#define QIB_7220_HwErrStatus_PcieCplTimeout_RMASK 0x1
+#define QIB_7220_HwErrStatus_PoisenedTLP_LSB 0x1D
+#define QIB_7220_HwErrStatus_PoisenedTLP_RMASK 0x1
+#define QIB_7220_HwErrStatus_SDmaMemReadErr_LSB 0x1C
+#define QIB_7220_HwErrStatus_SDmaMemReadErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_Reserved3_LSB 0x8
+#define QIB_7220_HwErrStatus_Reserved3_RMASK 0xFFFFF
+#define QIB_7220_HwErrStatus_PCIeMemParity_LSB 0x0
+#define QIB_7220_HwErrStatus_PCIeMemParity_RMASK 0xFF
+
+#define QIB_7220_HwErrClear_OFFS 0xA8
+#define QIB_7220_HwErrClear_IBCBusFromSPCParityErrClear_LSB 0x3F
+#define QIB_7220_HwErrClear_IBCBusFromSPCParityErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_IBCBusToSPCparityErrClear_LSB 0x3E
+#define QIB_7220_HwErrClear_IBCBusToSPCparityErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_Clk_uC_PLLNotLockedClear_LSB 0x3D
+#define QIB_7220_HwErrClear_Clk_uC_PLLNotLockedClear_RMASK 0x1
+#define QIB_7220_HwErrClear_IBSerdesPClkNotDetectClear_LSB 0x3C
+#define QIB_7220_HwErrClear_IBSerdesPClkNotDetectClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PCIESerdesQ3PClkNotDetectClear_LSB 0x3B
+#define QIB_7220_HwErrClear_PCIESerdesQ3PClkNotDetectClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PCIESerdesQ2PClkNotDetectClear_LSB 0x3A
+#define QIB_7220_HwErrClear_PCIESerdesQ2PClkNotDetectClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PCIESerdesQ1PClkNotDetectClear_LSB 0x39
+#define QIB_7220_HwErrClear_PCIESerdesQ1PClkNotDetectClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PCIESerdesQ0PClkNotDetectClear_LSB 0x38
+#define QIB_7220_HwErrClear_PCIESerdesQ0PClkNotDetectClear_RMASK 0x1
+#define QIB_7220_HwErrClear_Reserved_LSB 0x37
+#define QIB_7220_HwErrClear_Reserved_RMASK 0x1
+#define QIB_7220_HwErrClear_PowerOnBISTFailedClear_LSB 0x36
+#define QIB_7220_HwErrClear_PowerOnBISTFailedClear_RMASK 0x1
+#define QIB_7220_HwErrClear_Reserved1_LSB 0x33
+#define QIB_7220_HwErrClear_Reserved1_RMASK 0x7
+#define QIB_7220_HwErrClear_RXEMemParityClear_LSB 0x2C
+#define QIB_7220_HwErrClear_RXEMemParityClear_RMASK 0x7F
+#define QIB_7220_HwErrClear_TXEMemParityClear_LSB 0x28
+#define QIB_7220_HwErrClear_TXEMemParityClear_RMASK 0xF
+#define QIB_7220_HwErrClear_DDSRXEQMemoryParityErrClear_LSB 0x27
+#define QIB_7220_HwErrClear_DDSRXEQMemoryParityErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_IB_uC_MemoryParityErrClear_LSB 0x26
+#define QIB_7220_HwErrClear_IB_uC_MemoryParityErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PCIE_uC_Oct1MemoryParityErrClear_LSB 0x25
+#define QIB_7220_HwErrClear_PCIE_uC_Oct1MemoryParityErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PCIE_uC_Oct0MemoryParityErrClear_LSB 0x24
+#define QIB_7220_HwErrClear_PCIE_uC_Oct0MemoryParityErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_Reserved2_LSB 0x22
+#define QIB_7220_HwErrClear_Reserved2_RMASK 0x3
+#define QIB_7220_HwErrClear_PCIeBusParityClr_LSB 0x1F
+#define QIB_7220_HwErrClear_PCIeBusParityClr_RMASK 0x7
+#define QIB_7220_HwErrClear_PcieCplTimeoutClear_LSB 0x1E
+#define QIB_7220_HwErrClear_PcieCplTimeoutClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PoisonedTLPClear_LSB 0x1D
+#define QIB_7220_HwErrClear_PoisonedTLPClear_RMASK 0x1
+#define QIB_7220_HwErrClear_SDmaMemReadErrClear_LSB 0x1C
+#define QIB_7220_HwErrClear_SDmaMemReadErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_Reserved3_LSB 0x8
+#define QIB_7220_HwErrClear_Reserved3_RMASK 0xFFFFF
+#define QIB_7220_HwErrClear_PCIeMemParityClr_LSB 0x0
+#define QIB_7220_HwErrClear_PCIeMemParityClr_RMASK 0xFF
+
+#define QIB_7220_HwDiagCtrl_OFFS 0xB0
+#define QIB_7220_HwDiagCtrl_ForceIBCBusFromSPCParityErr_LSB 0x3F
+#define QIB_7220_HwDiagCtrl_ForceIBCBusFromSPCParityErr_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_ForceIBCBusToSPCParityErr_LSB 0x3E
+#define QIB_7220_HwDiagCtrl_ForceIBCBusToSPCParityErr_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_CounterWrEnable_LSB 0x3D
+#define QIB_7220_HwDiagCtrl_CounterWrEnable_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_CounterDisable_LSB 0x3C
+#define QIB_7220_HwDiagCtrl_CounterDisable_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_Reserved_LSB 0x33
+#define QIB_7220_HwDiagCtrl_Reserved_RMASK 0x1FF
+#define QIB_7220_HwDiagCtrl_ForceRxMemParityErr_LSB 0x2C
+#define QIB_7220_HwDiagCtrl_ForceRxMemParityErr_RMASK 0x7F
+#define QIB_7220_HwDiagCtrl_ForceTxMemparityErr_LSB 0x28
+#define QIB_7220_HwDiagCtrl_ForceTxMemparityErr_RMASK 0xF
+#define QIB_7220_HwDiagCtrl_ForceDDSRXEQMemoryParityErr_LSB 0x27
+#define QIB_7220_HwDiagCtrl_ForceDDSRXEQMemoryParityErr_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_ForceIB_uC_MemoryParityErr_LSB 0x26
+#define QIB_7220_HwDiagCtrl_ForceIB_uC_MemoryParityErr_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_ForcePCIE_uC_Oct1MemoryParityErr_LSB 0x25
+#define QIB_7220_HwDiagCtrl_ForcePCIE_uC_Oct1MemoryParityErr_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_ForcePCIE_uC_Oct0MemoryParityErr_LSB 0x24
+#define QIB_7220_HwDiagCtrl_ForcePCIE_uC_Oct0MemoryParityErr_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_Reserved1_LSB 0x23
+#define QIB_7220_HwDiagCtrl_Reserved1_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_forcePCIeBusParity_LSB 0x1F
+#define QIB_7220_HwDiagCtrl_forcePCIeBusParity_RMASK 0xF
+#define QIB_7220_HwDiagCtrl_Reserved2_LSB 0x8
+#define QIB_7220_HwDiagCtrl_Reserved2_RMASK 0x7FFFFF
+#define QIB_7220_HwDiagCtrl_forcePCIeMemParity_LSB 0x0
+#define QIB_7220_HwDiagCtrl_forcePCIeMemParity_RMASK 0xFF
+
+#define QIB_7220_REG_0000B8_OFFS 0xB8
+
+#define QIB_7220_IBCStatus_OFFS 0xC0
+#define QIB_7220_IBCStatus_TxCreditOk_LSB 0x1F
+#define QIB_7220_IBCStatus_TxCreditOk_RMASK 0x1
+#define QIB_7220_IBCStatus_TxReady_LSB 0x1E
+#define QIB_7220_IBCStatus_TxReady_RMASK 0x1
+#define QIB_7220_IBCStatus_Reserved_LSB 0xE
+#define QIB_7220_IBCStatus_Reserved_RMASK 0xFFFF
+#define QIB_7220_IBCStatus_IBTxLaneReversed_LSB 0xD
+#define QIB_7220_IBCStatus_IBTxLaneReversed_RMASK 0x1
+#define QIB_7220_IBCStatus_IBRxLaneReversed_LSB 0xC
+#define QIB_7220_IBCStatus_IBRxLaneReversed_RMASK 0x1
+#define QIB_7220_IBCStatus_IB_SERDES_TRIM_DONE_LSB 0xB
+#define QIB_7220_IBCStatus_IB_SERDES_TRIM_DONE_RMASK 0x1
+#define QIB_7220_IBCStatus_DDS_RXEQ_FAIL_LSB 0xA
+#define QIB_7220_IBCStatus_DDS_RXEQ_FAIL_RMASK 0x1
+#define QIB_7220_IBCStatus_LinkWidthActive_LSB 0x9
+#define QIB_7220_IBCStatus_LinkWidthActive_RMASK 0x1
+#define QIB_7220_IBCStatus_LinkSpeedActive_LSB 0x8
+#define QIB_7220_IBCStatus_LinkSpeedActive_RMASK 0x1
+#define QIB_7220_IBCStatus_LinkState_LSB 0x5
+#define QIB_7220_IBCStatus_LinkState_RMASK 0x7
+#define QIB_7220_IBCStatus_LinkTrainingState_LSB 0x0
+#define QIB_7220_IBCStatus_LinkTrainingState_RMASK 0x1F
+
+#define QIB_7220_IBCCtrl_OFFS 0xC8
+#define QIB_7220_IBCCtrl_Loopback_LSB 0x3F
+#define QIB_7220_IBCCtrl_Loopback_RMASK 0x1
+#define QIB_7220_IBCCtrl_LinkDownDefaultState_LSB 0x3E
+#define QIB_7220_IBCCtrl_LinkDownDefaultState_RMASK 0x1
+#define QIB_7220_IBCCtrl_Reserved_LSB 0x2B
+#define QIB_7220_IBCCtrl_Reserved_RMASK 0x7FFFF
+#define QIB_7220_IBCCtrl_CreditScale_LSB 0x28
+#define QIB_7220_IBCCtrl_CreditScale_RMASK 0x7
+#define QIB_7220_IBCCtrl_OverrunThreshold_LSB 0x24
+#define QIB_7220_IBCCtrl_OverrunThreshold_RMASK 0xF
+#define QIB_7220_IBCCtrl_PhyerrThreshold_LSB 0x20
+#define QIB_7220_IBCCtrl_PhyerrThreshold_RMASK 0xF
+#define QIB_7220_IBCCtrl_MaxPktLen_LSB 0x15
+#define QIB_7220_IBCCtrl_MaxPktLen_RMASK 0x7FF
+#define QIB_7220_IBCCtrl_LinkCmd_LSB 0x13
+#define QIB_7220_IBCCtrl_LinkCmd_RMASK 0x3
+#define QIB_7220_IBCCtrl_LinkInitCmd_LSB 0x10
+#define QIB_7220_IBCCtrl_LinkInitCmd_RMASK 0x7
+#define QIB_7220_IBCCtrl_FlowCtrlWaterMark_LSB 0x8
+#define QIB_7220_IBCCtrl_FlowCtrlWaterMark_RMASK 0xFF
+#define QIB_7220_IBCCtrl_FlowCtrlPeriod_LSB 0x0
+#define QIB_7220_IBCCtrl_FlowCtrlPeriod_RMASK 0xFF
+
+#define QIB_7220_EXTStatus_OFFS 0xD0
+#define QIB_7220_EXTStatus_GPIOIn_LSB 0x30
+#define QIB_7220_EXTStatus_GPIOIn_RMASK 0xFFFF
+#define QIB_7220_EXTStatus_Reserved_LSB 0x20
+#define QIB_7220_EXTStatus_Reserved_RMASK 0xFFFF
+#define QIB_7220_EXTStatus_Reserved1_LSB 0x10
+#define QIB_7220_EXTStatus_Reserved1_RMASK 0xFFFF
+#define QIB_7220_EXTStatus_MemBISTDisabled_LSB 0xF
+#define QIB_7220_EXTStatus_MemBISTDisabled_RMASK 0x1
+#define QIB_7220_EXTStatus_MemBISTEndTest_LSB 0xE
+#define QIB_7220_EXTStatus_MemBISTEndTest_RMASK 0x1
+#define QIB_7220_EXTStatus_Reserved2_LSB 0x0
+#define QIB_7220_EXTStatus_Reserved2_RMASK 0x3FFF
+
+#define QIB_7220_EXTCtrl_OFFS 0xD8
+#define QIB_7220_EXTCtrl_GPIOOe_LSB 0x30
+#define QIB_7220_EXTCtrl_GPIOOe_RMASK 0xFFFF
+#define QIB_7220_EXTCtrl_GPIOInvert_LSB 0x20
+#define QIB_7220_EXTCtrl_GPIOInvert_RMASK 0xFFFF
+#define QIB_7220_EXTCtrl_Reserved_LSB 0x4
+#define QIB_7220_EXTCtrl_Reserved_RMASK 0xFFFFFFF
+#define QIB_7220_EXTCtrl_LEDPriPortGreenOn_LSB 0x3
+#define QIB_7220_EXTCtrl_LEDPriPortGreenOn_RMASK 0x1
+#define QIB_7220_EXTCtrl_LEDPriPortYellowOn_LSB 0x2
+#define QIB_7220_EXTCtrl_LEDPriPortYellowOn_RMASK 0x1
+#define QIB_7220_EXTCtrl_LEDGblOkGreenOn_LSB 0x1
+#define QIB_7220_EXTCtrl_LEDGblOkGreenOn_RMASK 0x1
+#define QIB_7220_EXTCtrl_LEDGblErrRedOff_LSB 0x0
+#define QIB_7220_EXTCtrl_LEDGblErrRedOff_RMASK 0x1
+
+#define QIB_7220_GPIOOut_OFFS 0xE0
+
+#define QIB_7220_GPIOMask_OFFS 0xE8
+
+#define QIB_7220_GPIOStatus_OFFS 0xF0
+
+#define QIB_7220_GPIOClear_OFFS 0xF8
+
+#define QIB_7220_RcvCtrl_OFFS 0x100
+#define QIB_7220_RcvCtrl_Reserved_LSB 0x27
+#define QIB_7220_RcvCtrl_Reserved_RMASK 0x1FFFFFF
+#define QIB_7220_RcvCtrl_RcvQPMapEnable_LSB 0x26
+#define QIB_7220_RcvCtrl_RcvQPMapEnable_RMASK 0x1
+#define QIB_7220_RcvCtrl_PortCfg_LSB 0x24
+#define QIB_7220_RcvCtrl_PortCfg_RMASK 0x3
+#define QIB_7220_RcvCtrl_TailUpd_LSB 0x23
+#define QIB_7220_RcvCtrl_TailUpd_RMASK 0x1
+#define QIB_7220_RcvCtrl_RcvPartitionKeyDisable_LSB 0x22
+#define QIB_7220_RcvCtrl_RcvPartitionKeyDisable_RMASK 0x1
+#define QIB_7220_RcvCtrl_IntrAvail_LSB 0x11
+#define QIB_7220_RcvCtrl_IntrAvail_RMASK 0x1FFFF
+#define QIB_7220_RcvCtrl_PortEnable_LSB 0x0
+#define QIB_7220_RcvCtrl_PortEnable_RMASK 0x1FFFF
+
+#define QIB_7220_RcvBTHQP_OFFS 0x108
+#define QIB_7220_RcvBTHQP_Reserved_LSB 0x18
+#define QIB_7220_RcvBTHQP_Reserved_RMASK 0xFF
+#define QIB_7220_RcvBTHQP_RcvBTHQP_LSB 0x0
+#define QIB_7220_RcvBTHQP_RcvBTHQP_RMASK 0xFFFFFF
+
+#define QIB_7220_RcvHdrSize_OFFS 0x110
+
+#define QIB_7220_RcvHdrCnt_OFFS 0x118
+
+#define QIB_7220_RcvHdrEntSize_OFFS 0x120
+
+#define QIB_7220_RcvTIDBase_OFFS 0x128
+
+#define QIB_7220_RcvTIDCnt_OFFS 0x130
+
+#define QIB_7220_RcvEgrBase_OFFS 0x138
+
+#define QIB_7220_RcvEgrCnt_OFFS 0x140
+
+#define QIB_7220_RcvBufBase_OFFS 0x148
+
+#define QIB_7220_RcvBufSize_OFFS 0x150
+
+#define QIB_7220_RxIntMemBase_OFFS 0x158
+
+#define QIB_7220_RxIntMemSize_OFFS 0x160
+
+#define QIB_7220_RcvPartitionKey_OFFS 0x168
+
+#define QIB_7220_RcvQPMulticastPort_OFFS 0x170
+#define QIB_7220_RcvQPMulticastPort_Reserved_LSB 0x5
+#define QIB_7220_RcvQPMulticastPort_Reserved_RMASK 0x7FFFFFFFFFFFFFF
+#define QIB_7220_RcvQPMulticastPort_RcvQpMcPort_LSB 0x0
+#define QIB_7220_RcvQPMulticastPort_RcvQpMcPort_RMASK 0x1F
+
+#define QIB_7220_RcvPktLEDCnt_OFFS 0x178
+#define QIB_7220_RcvPktLEDCnt_ONperiod_LSB 0x20
+#define QIB_7220_RcvPktLEDCnt_ONperiod_RMASK 0xFFFFFFFF
+#define QIB_7220_RcvPktLEDCnt_OFFperiod_LSB 0x0
+#define QIB_7220_RcvPktLEDCnt_OFFperiod_RMASK 0xFFFFFFFF
+
+#define QIB_7220_IBCDDRCtrl_OFFS 0x180
+#define QIB_7220_IBCDDRCtrl_IB_DLID_MASK_LSB 0x30
+#define QIB_7220_IBCDDRCtrl_IB_DLID_MASK_RMASK 0xFFFF
+#define QIB_7220_IBCDDRCtrl_IB_DLID_LSB 0x20
+#define QIB_7220_IBCDDRCtrl_IB_DLID_RMASK 0xFFFF
+#define QIB_7220_IBCDDRCtrl_Reserved_LSB 0x1B
+#define QIB_7220_IBCDDRCtrl_Reserved_RMASK 0x1F
+#define QIB_7220_IBCDDRCtrl_HRTBT_REQ_LSB 0x1A
+#define QIB_7220_IBCDDRCtrl_HRTBT_REQ_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_HRTBT_PORT_LSB 0x12
+#define QIB_7220_IBCDDRCtrl_HRTBT_PORT_RMASK 0xFF
+#define QIB_7220_IBCDDRCtrl_HRTBT_AUTO_LSB 0x11
+#define QIB_7220_IBCDDRCtrl_HRTBT_AUTO_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_HRTBT_ENB_LSB 0x10
+#define QIB_7220_IBCDDRCtrl_HRTBT_ENB_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_SD_DDS_LSB 0xC
+#define QIB_7220_IBCDDRCtrl_SD_DDS_RMASK 0xF
+#define QIB_7220_IBCDDRCtrl_SD_DDSV_LSB 0xB
+#define QIB_7220_IBCDDRCtrl_SD_DDSV_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_SD_ADD_ENB_LSB 0xA
+#define QIB_7220_IBCDDRCtrl_SD_ADD_ENB_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_SD_RX_EQUAL_ENABLE_LSB 0x9
+#define QIB_7220_IBCDDRCtrl_SD_RX_EQUAL_ENABLE_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_IB_LANE_REV_SUPPORTED_LSB 0x8
+#define QIB_7220_IBCDDRCtrl_IB_LANE_REV_SUPPORTED_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_IB_POLARITY_REV_SUPP_LSB 0x7
+#define QIB_7220_IBCDDRCtrl_IB_POLARITY_REV_SUPP_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_IB_NUM_CHANNELS_LSB 0x5
+#define QIB_7220_IBCDDRCtrl_IB_NUM_CHANNELS_RMASK 0x3
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_QDR_LSB 0x4
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_QDR_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_DDR_LSB 0x3
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_DDR_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_SDR_LSB 0x2
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_SDR_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_LSB 0x1
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_IB_ENHANCED_MODE_LSB 0x0
+#define QIB_7220_IBCDDRCtrl_IB_ENHANCED_MODE_RMASK 0x1
+
+#define QIB_7220_HRTBT_GUID_OFFS 0x188
+
+#define QIB_7220_IBCDDRCtrl2_OFFS 0x1A0
+#define QIB_7220_IBCDDRCtrl2_IB_BACK_PORCH_LSB 0x5
+#define QIB_7220_IBCDDRCtrl2_IB_BACK_PORCH_RMASK 0x1F
+#define QIB_7220_IBCDDRCtrl2_IB_FRONT_PORCH_LSB 0x0
+#define QIB_7220_IBCDDRCtrl2_IB_FRONT_PORCH_RMASK 0x1F
+
+#define QIB_7220_IBCDDRStatus_OFFS 0x1A8
+#define QIB_7220_IBCDDRStatus_heartbeat_timed_out_LSB 0x24
+#define QIB_7220_IBCDDRStatus_heartbeat_timed_out_RMASK 0x1
+#define QIB_7220_IBCDDRStatus_heartbeat_crosstalk_LSB 0x20
+#define QIB_7220_IBCDDRStatus_heartbeat_crosstalk_RMASK 0xF
+#define QIB_7220_IBCDDRStatus_RxEqLocalDevice_LSB 0x1E
+#define QIB_7220_IBCDDRStatus_RxEqLocalDevice_RMASK 0x3
+#define QIB_7220_IBCDDRStatus_ReqDDSLocalFromRmt_LSB 0x1A
+#define QIB_7220_IBCDDRStatus_ReqDDSLocalFromRmt_RMASK 0xF
+#define QIB_7220_IBCDDRStatus_LinkRoundTripLatency_LSB 0x0
+#define QIB_7220_IBCDDRStatus_LinkRoundTripLatency_RMASK 0x3FFFFFF
+
+#define QIB_7220_JIntReload_OFFS 0x1B0
+#define QIB_7220_JIntReload_J_limit_reload_LSB 0x10
+#define QIB_7220_JIntReload_J_limit_reload_RMASK 0xFFFF
+#define QIB_7220_JIntReload_J_reload_LSB 0x0
+#define QIB_7220_JIntReload_J_reload_RMASK 0xFFFF
+
+#define QIB_7220_IBNCModeCtrl_OFFS 0x1B8
+#define QIB_7220_IBNCModeCtrl_Reserved_LSB 0x1A
+#define QIB_7220_IBNCModeCtrl_Reserved_RMASK 0x3FFFFFFFFF
+#define QIB_7220_IBNCModeCtrl_TSMCode_TS2_LSB 0x11
+#define QIB_7220_IBNCModeCtrl_TSMCode_TS2_RMASK 0x1FF
+#define QIB_7220_IBNCModeCtrl_TSMCode_TS1_LSB 0x8
+#define QIB_7220_IBNCModeCtrl_TSMCode_TS1_RMASK 0x1FF
+#define QIB_7220_IBNCModeCtrl_Reserved1_LSB 0x3
+#define QIB_7220_IBNCModeCtrl_Reserved1_RMASK 0x1F
+#define QIB_7220_IBNCModeCtrl_TSMEnable_ignore_TSM_on_rx_LSB 0x2
+#define QIB_7220_IBNCModeCtrl_TSMEnable_ignore_TSM_on_rx_RMASK 0x1
+#define QIB_7220_IBNCModeCtrl_TSMEnable_send_TS2_LSB 0x1
+#define QIB_7220_IBNCModeCtrl_TSMEnable_send_TS2_RMASK 0x1
+#define QIB_7220_IBNCModeCtrl_TSMEnable_send_TS1_LSB 0x0
+#define QIB_7220_IBNCModeCtrl_TSMEnable_send_TS1_RMASK 0x1
+
+#define QIB_7220_SendCtrl_OFFS 0x1C0
+#define QIB_7220_SendCtrl_Disarm_LSB 0x1F
+#define QIB_7220_SendCtrl_Disarm_RMASK 0x1
+#define QIB_7220_SendCtrl_Reserved_LSB 0x1D
+#define QIB_7220_SendCtrl_Reserved_RMASK 0x3
+#define QIB_7220_SendCtrl_AvailUpdThld_LSB 0x18
+#define QIB_7220_SendCtrl_AvailUpdThld_RMASK 0x1F
+#define QIB_7220_SendCtrl_DisarmPIOBuf_LSB 0x10
+#define QIB_7220_SendCtrl_DisarmPIOBuf_RMASK 0xFF
+#define QIB_7220_SendCtrl_Reserved1_LSB 0xD
+#define QIB_7220_SendCtrl_Reserved1_RMASK 0x7
+#define QIB_7220_SendCtrl_SDmaHalt_LSB 0xC
+#define QIB_7220_SendCtrl_SDmaHalt_RMASK 0x1
+#define QIB_7220_SendCtrl_SDmaEnable_LSB 0xB
+#define QIB_7220_SendCtrl_SDmaEnable_RMASK 0x1
+#define QIB_7220_SendCtrl_SDmaSingleDescriptor_LSB 0xA
+#define QIB_7220_SendCtrl_SDmaSingleDescriptor_RMASK 0x1
+#define QIB_7220_SendCtrl_SDmaIntEnable_LSB 0x9
+#define QIB_7220_SendCtrl_SDmaIntEnable_RMASK 0x1
+#define QIB_7220_SendCtrl_Reserved2_LSB 0x5
+#define QIB_7220_SendCtrl_Reserved2_RMASK 0xF
+#define QIB_7220_SendCtrl_SSpecialTriggerEn_LSB 0x4
+#define QIB_7220_SendCtrl_SSpecialTriggerEn_RMASK 0x1
+#define QIB_7220_SendCtrl_SPioEnable_LSB 0x3
+#define QIB_7220_SendCtrl_SPioEnable_RMASK 0x1
+#define QIB_7220_SendCtrl_SendBufAvailUpd_LSB 0x2
+#define QIB_7220_SendCtrl_SendBufAvailUpd_RMASK 0x1
+#define QIB_7220_SendCtrl_SendIntBufAvail_LSB 0x1
+#define QIB_7220_SendCtrl_SendIntBufAvail_RMASK 0x1
+#define QIB_7220_SendCtrl_Abort_LSB 0x0
+#define QIB_7220_SendCtrl_Abort_RMASK 0x1
+
+#define QIB_7220_SendBufBase_OFFS 0x1C8
+#define QIB_7220_SendBufBase_Reserved_LSB 0x35
+#define QIB_7220_SendBufBase_Reserved_RMASK 0x7FF
+#define QIB_7220_SendBufBase_BaseAddr_LargePIO_LSB 0x20
+#define QIB_7220_SendBufBase_BaseAddr_LargePIO_RMASK 0x1FFFFF
+#define QIB_7220_SendBufBase_Reserved1_LSB 0x15
+#define QIB_7220_SendBufBase_Reserved1_RMASK 0x7FF
+#define QIB_7220_SendBufBase_BaseAddr_SmallPIO_LSB 0x0
+#define QIB_7220_SendBufBase_BaseAddr_SmallPIO_RMASK 0x1FFFFF
+
+#define QIB_7220_SendBufSize_OFFS 0x1D0
+#define QIB_7220_SendBufSize_Reserved_LSB 0x2D
+#define QIB_7220_SendBufSize_Reserved_RMASK 0xFFFFF
+#define QIB_7220_SendBufSize_Size_LargePIO_LSB 0x20
+#define QIB_7220_SendBufSize_Size_LargePIO_RMASK 0x1FFF
+#define QIB_7220_SendBufSize_Reserved1_LSB 0xC
+#define QIB_7220_SendBufSize_Reserved1_RMASK 0xFFFFF
+#define QIB_7220_SendBufSize_Size_SmallPIO_LSB 0x0
+#define QIB_7220_SendBufSize_Size_SmallPIO_RMASK 0xFFF
+
+#define QIB_7220_SendBufCnt_OFFS 0x1D8
+#define QIB_7220_SendBufCnt_Reserved_LSB 0x24
+#define QIB_7220_SendBufCnt_Reserved_RMASK 0xFFFFFFF
+#define QIB_7220_SendBufCnt_Num_LargeBuffers_LSB 0x20
+#define QIB_7220_SendBufCnt_Num_LargeBuffers_RMASK 0xF
+#define QIB_7220_SendBufCnt_Reserved1_LSB 0x9
+#define QIB_7220_SendBufCnt_Reserved1_RMASK 0x7FFFFF
+#define QIB_7220_SendBufCnt_Num_SmallBuffers_LSB 0x0
+#define QIB_7220_SendBufCnt_Num_SmallBuffers_RMASK 0x1FF
+
+#define QIB_7220_SendBufAvailAddr_OFFS 0x1E0
+#define QIB_7220_SendBufAvailAddr_SendBufAvailAddr_LSB 0x6
+#define QIB_7220_SendBufAvailAddr_SendBufAvailAddr_RMASK 0x3FFFFFFFF
+#define QIB_7220_SendBufAvailAddr_Reserved_LSB 0x0
+#define QIB_7220_SendBufAvailAddr_Reserved_RMASK 0x3F
+
+#define QIB_7220_TxIntMemBase_OFFS 0x1E8
+
+#define QIB_7220_TxIntMemSize_OFFS 0x1F0
+
+#define QIB_7220_SendDmaBase_OFFS 0x1F8
+#define QIB_7220_SendDmaBase_Reserved_LSB 0x30
+#define QIB_7220_SendDmaBase_Reserved_RMASK 0xFFFF
+#define QIB_7220_SendDmaBase_SendDmaBase_LSB 0x0
+#define QIB_7220_SendDmaBase_SendDmaBase_RMASK 0xFFFFFFFFFFFF
+
+#define QIB_7220_SendDmaLenGen_OFFS 0x200
+#define QIB_7220_SendDmaLenGen_Reserved_LSB 0x13
+#define QIB_7220_SendDmaLenGen_Reserved_RMASK 0x1FFFFFFFFFFF
+#define QIB_7220_SendDmaLenGen_Generation_LSB 0x10
+#define QIB_7220_SendDmaLenGen_Generation_MSB 0x12
+#define QIB_7220_SendDmaLenGen_Generation_RMASK 0x7
+#define QIB_7220_SendDmaLenGen_Length_LSB 0x0
+#define QIB_7220_SendDmaLenGen_Length_RMASK 0xFFFF
+
+#define QIB_7220_SendDmaTail_OFFS 0x208
+#define QIB_7220_SendDmaTail_Reserved_LSB 0x10
+#define QIB_7220_SendDmaTail_Reserved_RMASK 0xFFFFFFFFFFFF
+#define QIB_7220_SendDmaTail_SendDmaTail_LSB 0x0
+#define QIB_7220_SendDmaTail_SendDmaTail_RMASK 0xFFFF
+
+#define QIB_7220_SendDmaHead_OFFS 0x210
+#define QIB_7220_SendDmaHead_Reserved_LSB 0x30
+#define QIB_7220_SendDmaHead_Reserved_RMASK 0xFFFF
+#define QIB_7220_SendDmaHead_InternalSendDmaHead_LSB 0x20
+#define QIB_7220_SendDmaHead_InternalSendDmaHead_RMASK 0xFFFF
+#define QIB_7220_SendDmaHead_Reserved1_LSB 0x10
+#define QIB_7220_SendDmaHead_Reserved1_RMASK 0xFFFF
+#define QIB_7220_SendDmaHead_SendDmaHead_LSB 0x0
+#define QIB_7220_SendDmaHead_SendDmaHead_RMASK 0xFFFF
+
+#define QIB_7220_SendDmaHeadAddr_OFFS 0x218
+#define QIB_7220_SendDmaHeadAddr_Reserved_LSB 0x30
+#define QIB_7220_SendDmaHeadAddr_Reserved_RMASK 0xFFFF
+#define QIB_7220_SendDmaHeadAddr_SendDmaHeadAddr_LSB 0x0
+#define QIB_7220_SendDmaHeadAddr_SendDmaHeadAddr_RMASK 0xFFFFFFFFFFFF
+
+#define QIB_7220_SendDmaBufMask0_OFFS 0x220
+#define QIB_7220_SendDmaBufMask0_BufMask_63_0_LSB 0x0
+#define QIB_7220_SendDmaBufMask0_BufMask_63_0_RMASK 0x0
+
+#define QIB_7220_SendDmaStatus_OFFS 0x238
+#define QIB_7220_SendDmaStatus_ScoreBoardDrainInProg_LSB 0x3F
+#define QIB_7220_SendDmaStatus_ScoreBoardDrainInProg_RMASK 0x1
+#define QIB_7220_SendDmaStatus_AbortInProg_LSB 0x3E
+#define QIB_7220_SendDmaStatus_AbortInProg_RMASK 0x1
+#define QIB_7220_SendDmaStatus_InternalSDmaEnable_LSB 0x3D
+#define QIB_7220_SendDmaStatus_InternalSDmaEnable_RMASK 0x1
+#define QIB_7220_SendDmaStatus_ScbDescIndex_13_0_LSB 0x2F
+#define QIB_7220_SendDmaStatus_ScbDescIndex_13_0_RMASK 0x3FFF
+#define QIB_7220_SendDmaStatus_RpyLowAddr_6_0_LSB 0x28
+#define QIB_7220_SendDmaStatus_RpyLowAddr_6_0_RMASK 0x7F
+#define QIB_7220_SendDmaStatus_RpyTag_7_0_LSB 0x20
+#define QIB_7220_SendDmaStatus_RpyTag_7_0_RMASK 0xFF
+#define QIB_7220_SendDmaStatus_ScbFull_LSB 0x1F
+#define QIB_7220_SendDmaStatus_ScbFull_RMASK 0x1
+#define QIB_7220_SendDmaStatus_ScbEmpty_LSB 0x1E
+#define QIB_7220_SendDmaStatus_ScbEmpty_RMASK 0x1
+#define QIB_7220_SendDmaStatus_ScbEntryValid_LSB 0x1D
+#define QIB_7220_SendDmaStatus_ScbEntryValid_RMASK 0x1
+#define QIB_7220_SendDmaStatus_ScbFetchDescFlag_LSB 0x1C
+#define QIB_7220_SendDmaStatus_ScbFetchDescFlag_RMASK 0x1
+#define QIB_7220_SendDmaStatus_SplFifoReadyToGo_LSB 0x1B
+#define QIB_7220_SendDmaStatus_SplFifoReadyToGo_RMASK 0x1
+#define QIB_7220_SendDmaStatus_SplFifoDisarmed_LSB 0x1A
+#define QIB_7220_SendDmaStatus_SplFifoDisarmed_RMASK 0x1
+#define QIB_7220_SendDmaStatus_SplFifoEmpty_LSB 0x19
+#define QIB_7220_SendDmaStatus_SplFifoEmpty_RMASK 0x1
+#define QIB_7220_SendDmaStatus_SplFifoFull_LSB 0x18
+#define QIB_7220_SendDmaStatus_SplFifoFull_RMASK 0x1
+#define QIB_7220_SendDmaStatus_SplFifoBufNum_LSB 0x10
+#define QIB_7220_SendDmaStatus_SplFifoBufNum_RMASK 0xFF
+#define QIB_7220_SendDmaStatus_SplFifoDescIndex_LSB 0x0
+#define QIB_7220_SendDmaStatus_SplFifoDescIndex_RMASK 0xFFFF
+
+#define QIB_7220_SendBufErr0_OFFS 0x240
+#define QIB_7220_SendBufErr0_SendBufErr_63_0_LSB 0x0
+#define QIB_7220_SendBufErr0_SendBufErr_63_0_RMASK 0x0
+
+#define QIB_7220_RcvHdrAddr0_OFFS 0x270
+#define QIB_7220_RcvHdrAddr0_RcvHdrAddr0_LSB 0x2
+#define QIB_7220_RcvHdrAddr0_RcvHdrAddr0_RMASK 0x3FFFFFFFFF
+#define QIB_7220_RcvHdrAddr0_Reserved_LSB 0x0
+#define QIB_7220_RcvHdrAddr0_Reserved_RMASK 0x3
+
+#define QIB_7220_RcvHdrTailAddr0_OFFS 0x300
+#define QIB_7220_RcvHdrTailAddr0_RcvHdrTailAddr0_LSB 0x2
+#define QIB_7220_RcvHdrTailAddr0_RcvHdrTailAddr0_RMASK 0x3FFFFFFFFF
+#define QIB_7220_RcvHdrTailAddr0_Reserved_LSB 0x0
+#define QIB_7220_RcvHdrTailAddr0_Reserved_RMASK 0x3
+
+#define QIB_7220_ibsd_epb_access_ctrl_OFFS 0x3C0
+#define QIB_7220_ibsd_epb_access_ctrl_sw_ib_epb_req_granted_LSB 0x8
+#define QIB_7220_ibsd_epb_access_ctrl_sw_ib_epb_req_granted_RMASK 0x1
+#define QIB_7220_ibsd_epb_access_ctrl_Reserved_LSB 0x1
+#define QIB_7220_ibsd_epb_access_ctrl_Reserved_RMASK 0x7F
+#define QIB_7220_ibsd_epb_access_ctrl_sw_ib_epb_req_LSB 0x0
+#define QIB_7220_ibsd_epb_access_ctrl_sw_ib_epb_req_RMASK 0x1
+
+#define QIB_7220_ibsd_epb_transaction_reg_OFFS 0x3C8
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_rdy_LSB 0x1F
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_rdy_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_req_error_LSB 0x1E
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_req_error_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_Reserved_LSB 0x1D
+#define QIB_7220_ibsd_epb_transaction_reg_Reserved_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_mem_data_parity_LSB 0x1C
+#define QIB_7220_ibsd_epb_transaction_reg_mem_data_parity_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_Reserved1_LSB 0x1B
+#define QIB_7220_ibsd_epb_transaction_reg_Reserved1_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_cs_LSB 0x19
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_cs_RMASK 0x3
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_read_write_LSB 0x18
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_read_write_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_Reserved2_LSB 0x17
+#define QIB_7220_ibsd_epb_transaction_reg_Reserved2_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_address_LSB 0x8
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_address_RMASK 0x7FFF
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_data_LSB 0x0
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_data_RMASK 0xFF
+
+#define QIB_7220_XGXSCfg_OFFS 0x3D8
+#define QIB_7220_XGXSCfg_sel_link_down_for_fctrl_lane_sync_reset_LSB 0x3F
+#define QIB_7220_XGXSCfg_sel_link_down_for_fctrl_lane_sync_reset_RMASK 0x1
+#define QIB_7220_XGXSCfg_Reserved_LSB 0x13
+#define QIB_7220_XGXSCfg_Reserved_RMASK 0xFFFFFFFFFFF
+#define QIB_7220_XGXSCfg_link_sync_mask_LSB 0x9
+#define QIB_7220_XGXSCfg_link_sync_mask_RMASK 0x3FF
+#define QIB_7220_XGXSCfg_Reserved1_LSB 0x3
+#define QIB_7220_XGXSCfg_Reserved1_RMASK 0x3F
+#define QIB_7220_XGXSCfg_xcv_reset_LSB 0x2
+#define QIB_7220_XGXSCfg_xcv_reset_RMASK 0x1
+#define QIB_7220_XGXSCfg_Reserved2_LSB 0x1
+#define QIB_7220_XGXSCfg_Reserved2_RMASK 0x1
+#define QIB_7220_XGXSCfg_tx_rx_reset_LSB 0x0
+#define QIB_7220_XGXSCfg_tx_rx_reset_RMASK 0x1
+
+#define QIB_7220_IBSerDesCtrl_OFFS 0x3E0
+#define QIB_7220_IBSerDesCtrl_Reserved_LSB 0x2D
+#define QIB_7220_IBSerDesCtrl_Reserved_RMASK 0x7FFFF
+#define QIB_7220_IBSerDesCtrl_INT_uC_LSB 0x2C
+#define QIB_7220_IBSerDesCtrl_INT_uC_RMASK 0x1
+#define QIB_7220_IBSerDesCtrl_CKSEL_uC_LSB 0x2A
+#define QIB_7220_IBSerDesCtrl_CKSEL_uC_RMASK 0x3
+#define QIB_7220_IBSerDesCtrl_PLLN_LSB 0x28
+#define QIB_7220_IBSerDesCtrl_PLLN_RMASK 0x3
+#define QIB_7220_IBSerDesCtrl_PLLM_LSB 0x25
+#define QIB_7220_IBSerDesCtrl_PLLM_RMASK 0x7
+#define QIB_7220_IBSerDesCtrl_TXOBPD_LSB 0x24
+#define QIB_7220_IBSerDesCtrl_TXOBPD_RMASK 0x1
+#define QIB_7220_IBSerDesCtrl_TWC_LSB 0x23
+#define QIB_7220_IBSerDesCtrl_TWC_RMASK 0x1
+#define QIB_7220_IBSerDesCtrl_RXIDLE_LSB 0x22
+#define QIB_7220_IBSerDesCtrl_RXIDLE_RMASK 0x1
+#define QIB_7220_IBSerDesCtrl_RXINV_LSB 0x21
+#define QIB_7220_IBSerDesCtrl_RXINV_RMASK 0x1
+#define QIB_7220_IBSerDesCtrl_TXINV_LSB 0x20
+#define QIB_7220_IBSerDesCtrl_TXINV_RMASK 0x1
+#define QIB_7220_IBSerDesCtrl_Reserved1_LSB 0x12
+#define QIB_7220_IBSerDesCtrl_Reserved1_RMASK 0x3FFF
+#define QIB_7220_IBSerDesCtrl_NumSerDesRegsToWrForRXEQ_LSB 0xD
+#define QIB_7220_IBSerDesCtrl_NumSerDesRegsToWrForRXEQ_RMASK 0x1F
+#define QIB_7220_IBSerDesCtrl_NumSerDesRegsToWrForDDS_LSB 0x8
+#define QIB_7220_IBSerDesCtrl_NumSerDesRegsToWrForDDS_RMASK 0x1F
+#define QIB_7220_IBSerDesCtrl_Reserved2_LSB 0x1
+#define QIB_7220_IBSerDesCtrl_Reserved2_RMASK 0x7F
+#define QIB_7220_IBSerDesCtrl_ResetIB_uC_Core_LSB 0x0
+#define QIB_7220_IBSerDesCtrl_ResetIB_uC_Core_RMASK 0x1
+
+#define QIB_7220_pciesd_epb_access_ctrl_OFFS 0x400
+#define QIB_7220_pciesd_epb_access_ctrl_sw_pcie_epb_req_granted_LSB 0x8
+#define QIB_7220_pciesd_epb_access_ctrl_sw_pcie_epb_req_granted_RMASK 0x1
+#define QIB_7220_pciesd_epb_access_ctrl_Reserved_LSB 0x3
+#define QIB_7220_pciesd_epb_access_ctrl_Reserved_RMASK 0x1F
+#define QIB_7220_pciesd_epb_access_ctrl_sw_pcieepb_star_en_LSB 0x1
+#define QIB_7220_pciesd_epb_access_ctrl_sw_pcieepb_star_en_RMASK 0x3
+#define QIB_7220_pciesd_epb_access_ctrl_sw_pcie_epb_req_LSB 0x0
+#define QIB_7220_pciesd_epb_access_ctrl_sw_pcie_epb_req_RMASK 0x1
+
+#define QIB_7220_pciesd_epb_transaction_reg_OFFS 0x408
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_rdy_LSB 0x1F
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_rdy_RMASK 0x1
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_req_error_LSB 0x1E
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_req_error_RMASK 0x1
+#define QIB_7220_pciesd_epb_transaction_reg_Reserved_LSB 0x1D
+#define QIB_7220_pciesd_epb_transaction_reg_Reserved_RMASK 0x1
+#define QIB_7220_pciesd_epb_transaction_reg_mem_data_parity_LSB 0x1C
+#define QIB_7220_pciesd_epb_transaction_reg_mem_data_parity_RMASK 0x1
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_cs_LSB 0x19
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_cs_RMASK 0x7
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_read_write_LSB 0x18
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_read_write_RMASK 0x1
+#define QIB_7220_pciesd_epb_transaction_reg_Reserved1_LSB 0x17
+#define QIB_7220_pciesd_epb_transaction_reg_Reserved1_RMASK 0x1
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_address_LSB 0x8
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_address_RMASK 0x7FFF
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_data_LSB 0x0
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_data_RMASK 0xFF
+
+#define QIB_7220_SerDes_DDSRXEQ0_OFFS 0x500
+#define QIB_7220_SerDes_DDSRXEQ0_reg_addr_LSB 0x4
+#define QIB_7220_SerDes_DDSRXEQ0_reg_addr_RMASK 0x3F
+#define QIB_7220_SerDes_DDSRXEQ0_element_num_LSB 0x0
+#define QIB_7220_SerDes_DDSRXEQ0_element_num_RMASK 0xF
+
+#define QIB_7220_LBIntCnt_OFFS 0x13000
+
+#define QIB_7220_LBFlowStallCnt_OFFS 0x13008
+
+#define QIB_7220_TxSDmaDescCnt_OFFS 0x13010
+
+#define QIB_7220_TxUnsupVLErrCnt_OFFS 0x13018
+
+#define QIB_7220_TxDataPktCnt_OFFS 0x13020
+
+#define QIB_7220_TxFlowPktCnt_OFFS 0x13028
+
+#define QIB_7220_TxDwordCnt_OFFS 0x13030
+
+#define QIB_7220_TxLenErrCnt_OFFS 0x13038
+
+#define QIB_7220_TxMaxMinLenErrCnt_OFFS 0x13040
+
+#define QIB_7220_TxUnderrunCnt_OFFS 0x13048
+
+#define QIB_7220_TxFlowStallCnt_OFFS 0x13050
+
+#define QIB_7220_TxDroppedPktCnt_OFFS 0x13058
+
+#define QIB_7220_RxDroppedPktCnt_OFFS 0x13060
+
+#define QIB_7220_RxDataPktCnt_OFFS 0x13068
+
+#define QIB_7220_RxFlowPktCnt_OFFS 0x13070
+
+#define QIB_7220_RxDwordCnt_OFFS 0x13078
+
+#define QIB_7220_RxLenErrCnt_OFFS 0x13080
+
+#define QIB_7220_RxMaxMinLenErrCnt_OFFS 0x13088
+
+#define QIB_7220_RxICRCErrCnt_OFFS 0x13090
+
+#define QIB_7220_RxVCRCErrCnt_OFFS 0x13098
+
+#define QIB_7220_RxFlowCtrlViolCnt_OFFS 0x130A0
+
+#define QIB_7220_RxVersionErrCnt_OFFS 0x130A8
+
+#define QIB_7220_RxLinkMalformCnt_OFFS 0x130B0
+
+#define QIB_7220_RxEBPCnt_OFFS 0x130B8
+
+#define QIB_7220_RxLPCRCErrCnt_OFFS 0x130C0
+
+#define QIB_7220_RxBufOvflCnt_OFFS 0x130C8
+
+#define QIB_7220_RxTIDFullErrCnt_OFFS 0x130D0
+
+#define QIB_7220_RxTIDValidErrCnt_OFFS 0x130D8
+
+#define QIB_7220_RxPKeyMismatchCnt_OFFS 0x130E0
+
+#define QIB_7220_RxP0HdrEgrOvflCnt_OFFS 0x130E8
+
+#define QIB_7220_IBStatusChangeCnt_OFFS 0x13170
+
+#define QIB_7220_IBLinkErrRecoveryCnt_OFFS 0x13178
+
+#define QIB_7220_IBLinkDownedCnt_OFFS 0x13180
+
+#define QIB_7220_IBSymbolErrCnt_OFFS 0x13188
+
+#define QIB_7220_RxVL15DroppedPktCnt_OFFS 0x13190
+
+#define QIB_7220_RxOtherLocalPhyErrCnt_OFFS 0x13198
+
+#define QIB_7220_PcieRetryBufDiagQwordCnt_OFFS 0x131A0
+
+#define QIB_7220_ExcessBufferOvflCnt_OFFS 0x131A8
+
+#define QIB_7220_LocalLinkIntegrityErrCnt_OFFS 0x131B0
+
+#define QIB_7220_RxVlErrCnt_OFFS 0x131B8
+
+#define QIB_7220_RxDlidFltrCnt_OFFS 0x131C0
+
+#define QIB_7220_CNT_0131C8_OFFS 0x131C8
+
+#define QIB_7220_PSStat_OFFS 0x13200
+
+#define QIB_7220_PSStart_OFFS 0x13208
+
+#define QIB_7220_PSInterval_OFFS 0x13210
+
+#define QIB_7220_PSRcvDataCount_OFFS 0x13218
+
+#define QIB_7220_PSRcvPktsCount_OFFS 0x13220
+
+#define QIB_7220_PSXmitDataCount_OFFS 0x13228
+
+#define QIB_7220_PSXmitPktsCount_OFFS 0x13230
+
+#define QIB_7220_PSXmitWaitCount_OFFS 0x13238
+
+#define QIB_7220_CNT_013240_OFFS 0x13240
+
+#define QIB_7220_RcvEgrArray_OFFS 0x14000
+
+#define QIB_7220_MEM_038000_OFFS 0x38000
+
+#define QIB_7220_RcvTIDArray0_OFFS 0x53000
+
+#define QIB_7220_PIOLaunchFIFO_OFFS 0x64000
+
+#define QIB_7220_MEM_064480_OFFS 0x64480
+
+#define QIB_7220_SendPIOpbcCache_OFFS 0x64800
+
+#define QIB_7220_MEM_064C80_OFFS 0x64C80
+
+#define QIB_7220_PreLaunchFIFO_OFFS 0x65000
+
+#define QIB_7220_MEM_065080_OFFS 0x65080
+
+#define QIB_7220_ScoreBoard_OFFS 0x65400
+
+#define QIB_7220_MEM_065440_OFFS 0x65440
+
+#define QIB_7220_DescriptorFIFO_OFFS 0x65800
+
+#define QIB_7220_MEM_065880_OFFS 0x65880
+
+#define QIB_7220_RcvBuf1_OFFS 0x72000
+
+#define QIB_7220_MEM_074800_OFFS 0x74800
+
+#define QIB_7220_RcvBuf2_OFFS 0x75000
+
+#define QIB_7220_MEM_076400_OFFS 0x76400
+
+#define QIB_7220_RcvFlags_OFFS 0x77000
+
+#define QIB_7220_MEM_078400_OFFS 0x78400
+
+#define QIB_7220_RcvLookupBuf1_OFFS 0x79000
+
+#define QIB_7220_MEM_07A400_OFFS 0x7A400
+
+#define QIB_7220_RcvDMADatBuf_OFFS 0x7B000
+
+#define QIB_7220_RcvDMAHdrBuf_OFFS 0x7B800
+
+#define QIB_7220_MiscRXEIntMem_OFFS 0x7C000
+
+#define QIB_7220_MEM_07D400_OFFS 0x7D400
+
+#define QIB_7220_PCIERcvBuf_OFFS 0x80000
+
+#define QIB_7220_PCIERetryBuf_OFFS 0x84000
+
+#define QIB_7220_PCIERcvBufRdToWrAddr_OFFS 0x88000
+
+#define QIB_7220_PCIECplBuf_OFFS 0x90000
+
+#define QIB_7220_IBSerDesMappTable_OFFS 0x94000
+
+#define QIB_7220_MEM_095000_OFFS 0x95000
+
+#define QIB_7220_SendBuf0_MA_OFFS 0x100000
+
+#define QIB_7220_MEM_1A0000_OFFS 0x1A0000

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 08/51] IB/qib: Add qib_common.h

[Ralph Campbell]

creates the qib_common.h file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_common.h |  755 ++++++++++++++++++++++++++++++++
 1 files changed, 755 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_common.h

diff --git a/drivers/infiniband/hw/qib/qib_common.h b/drivers/infiniband/hw/qib/qib_common.h
new file mode 100644
index 0000000..51239aa
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_common.h
@@ -0,0 +1,755 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef _QIB_COMMON_H
+#define _QIB_COMMON_H
+
+/*
+ * This file contains defines, structures, etc. that are used
+ * to communicate between kernel and user code.
+ */
+
+/* This is the IEEE-assigned OUI for QLogic Inc. QLogic_IB */
+#define QIB_SRC_OUI_1 0x00
+#define QIB_SRC_OUI_2 0x11
+#define QIB_SRC_OUI_3 0x75
+
+/* version of protocol header (known to chip also). In the long run,
+ * we should be able to generate and accept a range of version numbers;
+ * for now we only accept one, and it's compiled in.
+ */
+#define IPS_PROTO_VERSION 2
+
+/*
+ * These are compile time constants that you may want to enable or disable
+ * if you are trying to debug problems with code or performance.
+ * QIB_VERBOSE_TRACING define as 1 if you want additional tracing in
+ * fastpath code
+ * QIB_TRACE_REGWRITES define as 1 if you want register writes to be
+ * traced in faspath code
+ * _QIB_TRACING define as 0 if you want to remove all tracing in a
+ * compilation unit
+ * _QIB_DEBUGGING define as 0 if you want to remove debug prints
+ */
+
+/*
+ * The value in the BTH QP field that QLogic_IB uses to differentiate
+ * an qlogic_ib protocol IB packet vs standard IB transport
+ * This it needs to be even (0x656b78), because the LSB is sometimes
+ * used for the MSB of context. The change may cause a problem
+ * interoperating with older software.
+ */
+#define QIB_KD_QP 0x656b78
+
+/*
+ * These are the status bits readable (in ascii form, 64bit value)
+ * from the "status" sysfs file.  For binary compatibility, values
+ * must remain as is; removed states can be reused for different
+ * purposes.
+ */
+#define QIB_STATUS_INITTED       0x1    /* basic initialization done */
+/* Chip has been found and initted */
+#define QIB_STATUS_CHIP_PRESENT 0x20
+/* IB link is at ACTIVE, usable for data traffic */
+#define QIB_STATUS_IB_READY     0x40
+/* link is configured, LID, MTU, etc. have been set */
+#define QIB_STATUS_IB_CONF      0x80
+/* A Fatal hardware error has occurred. */
+#define QIB_STATUS_HWERROR     0x200
+
+/*
+ * The list of usermode accessible registers.  Also see Reg_* later in file.
+ */
+enum qib_ureg {
+	/* (RO)  DMA RcvHdr to be used next. */
+	ur_rcvhdrtail = 0,
+	/* (RW)  RcvHdr entry to be processed next by host. */
+	ur_rcvhdrhead = 1,
+	/* (RO)  Index of next Eager index to use. */
+	ur_rcvegrindextail = 2,
+	/* (RW)  Eager TID to be processed next */
+	ur_rcvegrindexhead = 3,
+	/* For internal use only; max register number. */
+	_QIB_UregMax
+};
+
+/* bit values for spi_runtime_flags */
+#define QIB_RUNTIME_PCIE                0x0002
+#define QIB_RUNTIME_FORCE_WC_ORDER      0x0004
+#define QIB_RUNTIME_RCVHDR_COPY         0x0008
+#define QIB_RUNTIME_MASTER              0x0010
+#define QIB_RUNTIME_NODMA_RTAIL         0x0080
+#define QIB_RUNTIME_SPECIAL_TRIGGER     0x0100
+#define QIB_RUNTIME_SDMA                0x0200
+#define QIB_RUNTIME_FORCE_PIOAVAIL      0x0400
+#define QIB_RUNTIME_PIO_REGSWAPPED      0x0800
+#define QIB_RUNTIME_CTXT_MSB_IN_QP      0x1000
+#define QIB_RUNTIME_CTXT_REDIRECT       0x2000
+#define QIB_RUNTIME_HDRSUPP             0x4000
+
+/*
+ * This structure is returned by qib_userinit() immediately after
+ * open to get implementation-specific info, and info specific to this
+ * instance.
+ *
+ * This struct must have explict pad fields where type sizes
+ * may result in different alignments between 32 and 64 bit
+ * programs, since the 64 bit * bit kernel requires the user code
+ * to have matching offsets
+ */
+struct qib_base_info {
+	/* version of hardware, for feature checking. */
+	__u32 spi_hw_version;
+	/* version of software, for feature checking. */
+	__u32 spi_sw_version;
+	/* QLogic_IB context assigned, goes into sent packets */
+	__u16 spi_ctxt;
+	__u16 spi_subctxt;
+	/*
+	 * IB MTU, packets IB data must be less than this.
+	 * The MTU is in bytes, and will be a multiple of 4 bytes.
+	 */
+	__u32 spi_mtu;
+	/*
+	 * Size of a PIO buffer.  Any given packet's total size must be less
+	 * than this (in words).  Included is the starting control word, so
+	 * if 513 is returned, then total pkt size is 512 words or less.
+	 */
+	__u32 spi_piosize;
+	/* size of the TID cache in qlogic_ib, in entries */
+	__u32 spi_tidcnt;
+	/* size of the TID Eager list in qlogic_ib, in entries */
+	__u32 spi_tidegrcnt;
+	/* size of a single receive header queue entry in words. */
+	__u32 spi_rcvhdrent_size;
+	/*
+	 * Count of receive header queue entries allocated.
+	 * This may be less than the spu_rcvhdrcnt passed in!.
+	 */
+	__u32 spi_rcvhdr_cnt;
+
+	/* per-chip and other runtime features bitmap (QIB_RUNTIME_*) */
+	__u32 spi_runtime_flags;
+
+	/* address where hardware receive header queue is mapped */
+	__u64 spi_rcvhdr_base;
+
+	/* user program. */
+
+	/* base address of eager TID receive buffers used by hardware. */
+	__u64 spi_rcv_egrbufs;
+
+	/* Allocated by initialization code, not by protocol. */
+
+	/*
+	 * Size of each TID buffer in host memory, starting at
+	 * spi_rcv_egrbufs.  The buffers are virtually contiguous.
+	 */
+	__u32 spi_rcv_egrbufsize;
+	/*
+	 * The special QP (queue pair) value that identifies an qlogic_ib
+	 * protocol packet from standard IB packets.  More, probably much
+	 * more, to be added.
+	 */
+	__u32 spi_qpair;
+
+	/*
+	 * User register base for init code, not to be used directly by
+	 * protocol or applications.  Always points to chip registers,
+	 * for normal or shared context.
+	 */
+	__u64 spi_uregbase;
+	/*
+	 * Maximum buffer size in bytes that can be used in a single TID
+	 * entry (assuming the buffer is aligned to this boundary).  This is
+	 * the minimum of what the hardware and software support Guaranteed
+	 * to be a power of 2.
+	 */
+	__u32 spi_tid_maxsize;
+	/*
+	 * alignment of each pio send buffer (byte count
+	 * to add to spi_piobufbase to get to second buffer)
+	 */
+	__u32 spi_pioalign;
+	/*
+	 * The index of the first pio buffer available to this process;
+	 * needed to do lookup in spi_pioavailaddr; not added to
+	 * spi_piobufbase.
+	 */
+	__u32 spi_pioindex;
+	 /* number of buffers mapped for this process */
+	__u32 spi_piocnt;
+
+	/*
+	 * Base address of writeonly pio buffers for this process.
+	 * Each buffer has spi_piosize words, and is aligned on spi_pioalign
+	 * boundaries.  spi_piocnt buffers are mapped from this address
+	 */
+	__u64 spi_piobufbase;
+
+	/*
+	 * Base address of readonly memory copy of the pioavail registers.
+	 * There are 2 bits for each buffer.
+	 */
+	__u64 spi_pioavailaddr;
+
+	/*
+	 * Address where driver updates a copy of the interface and driver
+	 * status (QIB_STATUS_*) as a 64 bit value.  It's followed by a
+	 * link status qword (formerly combined with driver status), then a
+	 * string indicating hardware error, if there was one.
+	 */
+	__u64 spi_status;
+
+	/* number of chip ctxts available to user processes */
+	__u32 spi_nctxts;
+	__u16 spi_unit; /* unit number of chip we are using */
+	__u16 spi_port; /* IB port number we are using */
+	/* num bufs in each contiguous set */
+	__u32 spi_rcv_egrperchunk;
+	/* size in bytes of each contiguous set */
+	__u32 spi_rcv_egrchunksize;
+	/* total size of mmap to cover full rcvegrbuffers */
+	__u32 spi_rcv_egrbuftotlen;
+	__u32 spi_rhf_offset; /* dword offset in hdrqent for rcvhdr flags */
+	/* address of readonly memory copy of the rcvhdrq tail register. */
+	__u64 spi_rcvhdr_tailaddr;
+
+	/*
+	 * shared memory pages for subctxts if ctxt is shared; these cover
+	 * all the processes in the group sharing a single context.
+	 * all have enough space for the num_subcontexts value on this job.
+	 */
+	__u64 spi_subctxt_uregbase;
+	__u64 spi_subctxt_rcvegrbuf;
+	__u64 spi_subctxt_rcvhdr_base;
+
+	/* shared memory page for send buffer disarm status */
+	__u64 spi_sendbuf_status;
+} __attribute__ ((aligned(8)));
+
+/*
+ * This version number is given to the driver by the user code during
+ * initialization in the spu_userversion field of qib_user_info, so
+ * the driver can check for compatibility with user code.
+ *
+ * The major version changes when data structures
+ * change in an incompatible way.  The driver must be the same or higher
+ * for initialization to succeed.  In some cases, a higher version
+ * driver will not interoperate with older software, and initialization
+ * will return an error.
+ */
+#define QIB_USER_SWMAJOR 1
+
+/*
+ * Minor version differences are always compatible
+ * a within a major version, however if user software is larger
+ * than driver software, some new features and/or structure fields
+ * may not be implemented; the user code must deal with this if it
+ * cares, or it must abort after initialization reports the difference.
+ */
+#define QIB_USER_SWMINOR 9
+
+#define QIB_USER_SWVERSION ((QIB_USER_SWMAJOR << 16) | QIB_USER_SWMINOR)
+
+#ifndef QIB_KERN_TYPE
+#define QIB_KERN_TYPE 0
+#define QIB_IDSTR "QLogic kernel.org driver"
+#endif
+
+/*
+ * Similarly, this is the kernel version going back to the user.  It's
+ * slightly different, in that we want to tell if the driver was built as
+ * part of a QLogic release, or from the driver from openfabrics.org,
+ * kernel.org, or a standard distribution, for support reasons.
+ * The high bit is 0 for non-QLogic and 1 for QLogic-built/supplied.
+ *
+ * It's returned by the driver to the user code during initialization in the
+ * spi_sw_version field of qib_base_info, so the user code can in turn
+ * check for compatibility with the kernel.
+*/
+#define QIB_KERN_SWVERSION ((QIB_KERN_TYPE << 31) | QIB_USER_SWVERSION)
+
+/*
+ * This structure is passed to qib_userinit() to tell the driver where
+ * user code buffers are, sizes, etc.   The offsets and sizes of the
+ * fields must remain unchanged, for binary compatibility.  It can
+ * be extended, if userversion is changed so user code can tell, if needed
+ */
+struct qib_user_info {
+	/*
+	 * version of user software, to detect compatibility issues.
+	 * Should be set to QIB_USER_SWVERSION.
+	 */
+	__u32 spu_userversion;
+
+	__u32 _spu_unused2;
+
+	/* size of struct base_info to write to */
+	__u32 spu_base_info_size;
+
+	__u32 _spu_unused3;
+
+	/*
+	 * If two or more processes wish to share a context, each process
+	 * must set the spu_subctxt_cnt and spu_subctxt_id to the same
+	 * values.  The only restriction on the spu_subctxt_id is that
+	 * it be unique for a given node.
+	 */
+	__u16 spu_subctxt_cnt;
+	__u16 spu_subctxt_id;
+
+	__u32 spu_port; /* IB port requested by user if > 0 */
+
+	/*
+	 * address of struct base_info to write to
+	 */
+	__u64 spu_base_info;
+
+} __attribute__ ((aligned(8)));
+
+/* User commands. */
+
+/* 16 available, was: old set up userspace (for old user code) */
+#define QIB_CMD_CTXT_INFO       17      /* find out what resources we got */
+#define QIB_CMD_RECV_CTRL       18      /* control receipt of packets */
+#define QIB_CMD_TID_UPDATE      19      /* update expected TID entries */
+#define QIB_CMD_TID_FREE        20      /* free expected TID entries */
+#define QIB_CMD_SET_PART_KEY    21      /* add partition key */
+/* 22 available, was: return info on slave processes (for old user code) */
+#define QIB_CMD_ASSIGN_CTXT     23      /* allocate HCA and ctxt */
+#define QIB_CMD_USER_INIT       24      /* set up userspace */
+#define QIB_CMD_UNUSED_1        25
+#define QIB_CMD_UNUSED_2        26
+#define QIB_CMD_PIOAVAILUPD     27      /* force an update of PIOAvail reg */
+#define QIB_CMD_POLL_TYPE       28      /* set the kind of polling we want */
+#define QIB_CMD_ARMLAUNCH_CTRL  29      /* armlaunch detection control */
+/* 30 is unused */
+#define QIB_CMD_SDMA_INFLIGHT   31      /* sdma inflight counter request */
+#define QIB_CMD_SDMA_COMPLETE   32      /* sdma completion counter request */
+/* 33 available, was a testing feature  */
+#define QIB_CMD_DISARM_BUFS     34      /* disarm send buffers w/ errors */
+#define QIB_CMD_ACK_EVENT     35	/* ack & clear bits */
+
+/*
+ * QIB_CMD_ACK_EVENT obsoletes QIB_CMD_DISARM_BUFS, but we keep it for
+ * compatibility with libraries from previous release.   The ACK_EVENT
+ * will take appropriate driver action (if any, just DISARM for now),
+ * then clear the bits passed in as part of the mask.  These bits are
+ * in the first 64bit word at spi_sendbuf_status, and are passed to
+ * the driver in the event_mask union as well.
+ */
+#define _QIB_EVENT_DISARM_BUFS_BIT	0
+#define _QIB_EVENT_LINKDOWN_BIT		1
+#define _QIB_EVENT_LID_CHANGE_BIT	2
+#define _QIB_EVENT_LMC_CHANGE_BIT	3
+#define _QIB_EVENT_SL2VL_CHANGE_BIT	4
+#define _QIB_MAX_EVENT_BIT _QIB_EVENT_SL2VL_CHANGE_BIT
+
+#define QIB_EVENT_DISARM_BUFS_BIT	(1UL << _QIB_EVENT_DISARM_BUFS_BIT)
+#define QIB_EVENT_LINKDOWN_BIT		(1UL << _QIB_EVENT_LINKDOWN_BIT)
+#define QIB_EVENT_LID_CHANGE_BIT	(1UL << _QIB_EVENT_LID_CHANGE_BIT)
+#define QIB_EVENT_LMC_CHANGE_BIT	(1UL << _QIB_EVENT_LMC_CHANGE_BIT)
+#define QIB_EVENT_SL2VL_CHANGE_BIT	(1UL << _QIB_EVENT_SL2VL_CHANGE_BIT)
+
+
+/*
+ * Poll types
+ */
+#define QIB_POLL_TYPE_ANYRCV     0x0
+#define QIB_POLL_TYPE_URGENT     0x1
+
+struct qib_ctxt_info {
+	__u16 num_active;       /* number of active units */
+	__u16 unit;             /* unit (chip) assigned to caller */
+	__u16 port;             /* IB port assigned to caller (1-based) */
+	__u16 ctxt;             /* ctxt on unit assigned to caller */
+	__u16 subctxt;          /* subctxt on unit assigned to caller */
+	__u16 num_ctxts;        /* number of ctxts available on unit */
+	__u16 num_subctxts;     /* number of subctxts opened on ctxt */
+	__u16 __pad;            /* pad size out to qword multple */
+};
+
+struct qib_tid_info {
+	__u32 tidcnt;
+	/* make structure same size in 32 and 64 bit */
+	__u32 tid__unused;
+	/* virtual address of first page in transfer */
+	__u64 tidvaddr;
+	/* pointer (same size 32/64 bit) to __u16 tid array */
+	__u64 tidlist;
+
+	/*
+	 * pointer (same size 32/64 bit) to bitmap of TIDs used
+	 * for this call; checked for being large enough at open
+	 */
+	__u64 tidmap;
+};
+
+struct qib_cmd {
+	__u32 type;                     /* command type */
+	union {
+		struct qib_tid_info tid_info;
+		struct qib_user_info user_info;
+
+		/*
+		 * address in userspace where we should put the sdma
+		 * inflight counter
+		 */
+		__u64 sdma_inflight;
+		/*
+		 * address in userspace where we should put the sdma
+		 * completion counter
+		 */
+		__u64 sdma_complete;
+		/* address in userspace of struct qib_ctxt_info to
+		   write result to */
+		__u64 ctxt_info;
+		/* enable/disable receipt of packets */
+		__u32 recv_ctrl;
+		/* enable/disable armlaunch errors (non-zero to enable) */
+		__u32 armlaunch_ctrl;
+		/* partition key to set */
+		__u16 part_key;
+		/* user address of __u32 bitmask of active slaves */
+		__u64 slave_mask_addr;
+		/* type of polling we want */
+		__u16 poll_type;
+		/* back pressure enable bit for one particular context */
+		__u8 ctxt_bp;
+		/* qib_user_event_ack(), IPATH_EVENT_* bits */
+		__u64 event_mask;
+	} cmd;
+};
+
+struct qib_iovec {
+	/* Pointer to data, but same size 32 and 64 bit */
+	__u64 iov_base;
+
+	/*
+	 * Length of data; don't need 64 bits, but want
+	 * qib_sendpkt to remain same size as before 32 bit changes, so...
+	 */
+	__u64 iov_len;
+};
+
+/*
+ * Describes a single packet for send.  Each packet can have one or more
+ * buffers, but the total length (exclusive of IB headers) must be less
+ * than the MTU, and if using the PIO method, entire packet length,
+ * including IB headers, must be less than the qib_piosize value (words).
+ * Use of this necessitates including sys/uio.h
+ */
+struct __qib_sendpkt {
+	__u32 sps_flags;        /* flags for packet (TBD) */
+	__u32 sps_cnt;          /* number of entries to use in sps_iov */
+	/* array of iov's describing packet. TEMPORARY */
+	struct qib_iovec sps_iov[4];
+};
+
+/*
+ * Diagnostics can send a packet by "writing" the following
+ * structs to the diag data special file.
+ * This allows a custom
+ * pbc (+ static rate) qword, so that special modes and deliberate
+ * changes to CRCs can be used. The elements were also re-ordered
+ * for better alignment and to avoid padding issues.
+ */
+#define _DIAG_XPKT_VERS 3
+struct qib_diag_xpkt {
+	__u16 version;
+	__u16 unit;
+	__u16 port;
+	__u16 len;
+	__u64 data;
+	__u64 pbc_wd;
+};
+
+/*
+ * Data layout in I2C flash (for GUID, etc.)
+ * All fields are little-endian binary unless otherwise stated
+ */
+#define QIB_FLASH_VERSION 2
+struct qib_flash {
+	/* flash layout version (QIB_FLASH_VERSION) */
+	__u8 if_fversion;
+	/* checksum protecting if_length bytes */
+	__u8 if_csum;
+	/*
+	 * valid length (in use, protected by if_csum), including
+	 * if_fversion and if_csum themselves)
+	 */
+	__u8 if_length;
+	/* the GUID, in network order */
+	__u8 if_guid[8];
+	/* number of GUIDs to use, starting from if_guid */
+	__u8 if_numguid;
+	/* the (last 10 characters of) board serial number, in ASCII */
+	char if_serial[12];
+	/* board mfg date (YYYYMMDD ASCII) */
+	char if_mfgdate[8];
+	/* last board rework/test date (YYYYMMDD ASCII) */
+	char if_testdate[8];
+	/* logging of error counts, TBD */
+	__u8 if_errcntp[4];
+	/* powered on hours, updated at driver unload */
+	__u8 if_powerhour[2];
+	/* ASCII free-form comment field */
+	char if_comment[32];
+	/* Backwards compatible prefix for longer QLogic Serial Numbers */
+	char if_sprefix[4];
+	/* 82 bytes used, min flash size is 128 bytes */
+	__u8 if_future[46];
+};
+
+/*
+ * These are the counters implemented in the chip, and are listed in order.
+ * The InterCaps naming is taken straight from the chip spec.
+ */
+struct qlogic_ib_counters {
+	__u64 LBIntCnt;
+	__u64 LBFlowStallCnt;
+	__u64 TxSDmaDescCnt;    /* was Reserved1 */
+	__u64 TxUnsupVLErrCnt;
+	__u64 TxDataPktCnt;
+	__u64 TxFlowPktCnt;
+	__u64 TxDwordCnt;
+	__u64 TxLenErrCnt;
+	__u64 TxMaxMinLenErrCnt;
+	__u64 TxUnderrunCnt;
+	__u64 TxFlowStallCnt;
+	__u64 TxDroppedPktCnt;
+	__u64 RxDroppedPktCnt;
+	__u64 RxDataPktCnt;
+	__u64 RxFlowPktCnt;
+	__u64 RxDwordCnt;
+	__u64 RxLenErrCnt;
+	__u64 RxMaxMinLenErrCnt;
+	__u64 RxICRCErrCnt;
+	__u64 RxVCRCErrCnt;
+	__u64 RxFlowCtrlErrCnt;
+	__u64 RxBadFormatCnt;
+	__u64 RxLinkProblemCnt;
+	__u64 RxEBPCnt;
+	__u64 RxLPCRCErrCnt;
+	__u64 RxBufOvflCnt;
+	__u64 RxTIDFullErrCnt;
+	__u64 RxTIDValidErrCnt;
+	__u64 RxPKeyMismatchCnt;
+	__u64 RxP0HdrEgrOvflCnt;
+	__u64 RxP1HdrEgrOvflCnt;
+	__u64 RxP2HdrEgrOvflCnt;
+	__u64 RxP3HdrEgrOvflCnt;
+	__u64 RxP4HdrEgrOvflCnt;
+	__u64 RxP5HdrEgrOvflCnt;
+	__u64 RxP6HdrEgrOvflCnt;
+	__u64 RxP7HdrEgrOvflCnt;
+	__u64 RxP8HdrEgrOvflCnt;
+	__u64 RxP9HdrEgrOvflCnt;
+	__u64 RxP10HdrEgrOvflCnt;
+	__u64 RxP11HdrEgrOvflCnt;
+	__u64 RxP12HdrEgrOvflCnt;
+	__u64 RxP13HdrEgrOvflCnt;
+	__u64 RxP14HdrEgrOvflCnt;
+	__u64 RxP15HdrEgrOvflCnt;
+	__u64 RxP16HdrEgrOvflCnt;
+	__u64 IBStatusChangeCnt;
+	__u64 IBLinkErrRecoveryCnt;
+	__u64 IBLinkDownedCnt;
+	__u64 IBSymbolErrCnt;
+	__u64 RxVL15DroppedPktCnt;
+	__u64 RxOtherLocalPhyErrCnt;
+	__u64 PcieRetryBufDiagQwordCnt;
+	__u64 ExcessBufferOvflCnt;
+	__u64 LocalLinkIntegrityErrCnt;
+	__u64 RxVlErrCnt;
+	__u64 RxDlidFltrCnt;
+};
+
+/*
+ * The next set of defines are for packet headers, and chip register
+ * and memory bits that are visible to and/or used by user-mode software.
+ */
+
+/* RcvHdrFlags bits */
+#define QLOGIC_IB_RHF_LENGTH_MASK 0x7FF
+#define QLOGIC_IB_RHF_LENGTH_SHIFT 0
+#define QLOGIC_IB_RHF_RCVTYPE_MASK 0x7
+#define QLOGIC_IB_RHF_RCVTYPE_SHIFT 11
+#define QLOGIC_IB_RHF_EGRINDEX_MASK 0xFFF
+#define QLOGIC_IB_RHF_EGRINDEX_SHIFT 16
+#define QLOGIC_IB_RHF_SEQ_MASK 0xF
+#define QLOGIC_IB_RHF_SEQ_SHIFT 0
+#define QLOGIC_IB_RHF_HDRQ_OFFSET_MASK 0x7FF
+#define QLOGIC_IB_RHF_HDRQ_OFFSET_SHIFT 4
+#define QLOGIC_IB_RHF_H_ICRCERR   0x80000000
+#define QLOGIC_IB_RHF_H_VCRCERR   0x40000000
+#define QLOGIC_IB_RHF_H_PARITYERR 0x20000000
+#define QLOGIC_IB_RHF_H_LENERR    0x10000000
+#define QLOGIC_IB_RHF_H_MTUERR    0x08000000
+#define QLOGIC_IB_RHF_H_IHDRERR   0x04000000
+#define QLOGIC_IB_RHF_H_TIDERR    0x02000000
+#define QLOGIC_IB_RHF_H_MKERR     0x01000000
+#define QLOGIC_IB_RHF_H_IBERR     0x00800000
+#define QLOGIC_IB_RHF_H_ERR_MASK  0xFF800000
+#define QLOGIC_IB_RHF_L_USE_EGR   0x80000000
+#define QLOGIC_IB_RHF_L_SWA       0x00008000
+#define QLOGIC_IB_RHF_L_SWB       0x00004000
+
+/* qlogic_ib header fields */
+#define QLOGIC_IB_I_VERS_MASK 0xF
+#define QLOGIC_IB_I_VERS_SHIFT 28
+#define QLOGIC_IB_I_CTXT_MASK 0xF
+#define QLOGIC_IB_I_CTXT_SHIFT 24
+#define QLOGIC_IB_I_TID_MASK 0x7FF
+#define QLOGIC_IB_I_TID_SHIFT 13
+#define QLOGIC_IB_I_OFFSET_MASK 0x1FFF
+#define QLOGIC_IB_I_OFFSET_SHIFT 0
+
+/* K_PktFlags bits */
+#define QLOGIC_IB_KPF_INTR 0x1
+#define QLOGIC_IB_KPF_SUBCTXT_MASK 0x3
+#define QLOGIC_IB_KPF_SUBCTXT_SHIFT 1
+
+#define QLOGIC_IB_MAX_SUBCTXT   4
+
+/* SendPIO per-buffer control */
+#define QLOGIC_IB_SP_TEST    0x40
+#define QLOGIC_IB_SP_TESTEBP 0x20
+#define QLOGIC_IB_SP_TRIGGER_SHIFT  15
+
+/* SendPIOAvail bits */
+#define QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT 1
+#define QLOGIC_IB_SENDPIOAVAIL_CHECK_SHIFT 0
+
+/* qlogic_ib header format */
+struct qib_header {
+	/*
+	 * Version - 4 bits, Context - 4 bits, TID - 10 bits and Offset -
+	 * 14 bits before ECO change ~28 Dec 03.  After that, Vers 4,
+	 * Context 4, TID 11, offset 13.
+	 */
+	__le32 ver_ctxt_tid_offset;
+	__le16 chksum;
+	__le16 pkt_flags;
+};
+
+/*
+ * qlogic_ib user message header format.
+ * This structure contains the first 4 fields common to all protocols
+ * that employ qlogic_ib.
+ */
+struct qib_message_header {
+	__be16 lrh[4];
+	__be32 bth[3];
+	/* fields below this point are in host byte order */
+	struct qib_header iph;
+	__u8 sub_opcode;
+};
+
+/* IB - LRH header consts */
+#define QIB_LRH_GRH 0x0003      /* 1. word of IB LRH - next header: GRH */
+#define QIB_LRH_BTH 0x0002      /* 1. word of IB LRH - next header: BTH */
+
+/* misc. */
+#define SIZE_OF_CRC 1
+
+#define QIB_DEFAULT_P_KEY 0xFFFF
+#define QIB_PERMISSIVE_LID 0xFFFF
+#define QIB_AETH_CREDIT_SHIFT 24
+#define QIB_AETH_CREDIT_MASK 0x1F
+#define QIB_AETH_CREDIT_INVAL 0x1F
+#define QIB_PSN_MASK 0xFFFFFF
+#define QIB_MSN_MASK 0xFFFFFF
+#define QIB_QPN_MASK 0xFFFFFF
+#define QIB_MULTICAST_LID_BASE 0xC000
+#define QIB_EAGER_TID_ID QLOGIC_IB_I_TID_MASK
+#define QIB_MULTICAST_QPN 0xFFFFFF
+
+/* Receive Header Queue: receive type (from qlogic_ib) */
+#define RCVHQ_RCV_TYPE_EXPECTED  0
+#define RCVHQ_RCV_TYPE_EAGER     1
+#define RCVHQ_RCV_TYPE_NON_KD    2
+#define RCVHQ_RCV_TYPE_ERROR     3
+
+#define QIB_HEADER_QUEUE_WORDS 9
+
+/* functions for extracting fields from rcvhdrq entries for the driver.
+ */
+static inline __u32 qib_hdrget_err_flags(const __le32 *rbuf)
+{
+	return __le32_to_cpu(rbuf[1]) & QLOGIC_IB_RHF_H_ERR_MASK;
+}
+
+static inline __u32 qib_hdrget_rcv_type(const __le32 *rbuf)
+{
+	return (__le32_to_cpu(rbuf[0]) >> QLOGIC_IB_RHF_RCVTYPE_SHIFT) &
+		QLOGIC_IB_RHF_RCVTYPE_MASK;
+}
+
+static inline __u32 qib_hdrget_length_in_bytes(const __le32 *rbuf)
+{
+	return ((__le32_to_cpu(rbuf[0]) >> QLOGIC_IB_RHF_LENGTH_SHIFT) &
+		QLOGIC_IB_RHF_LENGTH_MASK) << 2;
+}
+
+static inline __u32 qib_hdrget_index(const __le32 *rbuf)
+{
+	return (__le32_to_cpu(rbuf[0]) >> QLOGIC_IB_RHF_EGRINDEX_SHIFT) &
+		QLOGIC_IB_RHF_EGRINDEX_MASK;
+}
+
+static inline __u32 qib_hdrget_seq(const __le32 *rbuf)
+{
+	return (__le32_to_cpu(rbuf[1]) >> QLOGIC_IB_RHF_SEQ_SHIFT) &
+		QLOGIC_IB_RHF_SEQ_MASK;
+}
+
+static inline __u32 qib_hdrget_offset(const __le32 *rbuf)
+{
+	return (__le32_to_cpu(rbuf[1]) >> QLOGIC_IB_RHF_HDRQ_OFFSET_SHIFT) &
+		QLOGIC_IB_RHF_HDRQ_OFFSET_MASK;
+}
+
+static inline __u32 qib_hdrget_use_egr_buf(const __le32 *rbuf)
+{
+	return __le32_to_cpu(rbuf[0]) & QLOGIC_IB_RHF_L_USE_EGR;
+}
+
+static inline __u32 qib_hdrget_qib_ver(__le32 hdrword)
+{
+	return (__le32_to_cpu(hdrword) >> QLOGIC_IB_I_VERS_SHIFT) &
+		QLOGIC_IB_I_VERS_MASK;
+}
+
+#endif                          /* _QIB_COMMON_H */

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 09/51] IB/qib: Add qib_cq.c

[Ralph Campbell]

creates the qib_cq.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_cq.c |  475 ++++++++++++++++++++++++++++++++++++
 1 files changed, 475 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_cq.c

diff --git a/drivers/infiniband/hw/qib/qib_cq.c b/drivers/infiniband/hw/qib/qib_cq.c
new file mode 100644
index 0000000..4aed485
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_cq.c
@@ -0,0 +1,475 @@
+/*
+ * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/err.h>
+#include <linux/vmalloc.h>
+
+#include "qib_verbs.h"
+
+/**
+ * qib_cq_enter - add a new entry to the completion queue
+ * @cq: completion queue
+ * @entry: work completion entry to add
+ * @sig: true if @entry is a solicitated entry
+ *
+ * This may be called with qp->s_lock held.
+ */
+void qib_cq_enter(struct qib_cq *cq, struct ib_wc *entry, int solicited)
+{
+	struct qib_cq_wc *wc;
+	unsigned long flags;
+	u32 head;
+	u32 next;
+
+	spin_lock_irqsave(&cq->lock, flags);
+
+	/*
+	 * Note that the head pointer might be writable by user processes.
+	 * Take care to verify it is a sane value.
+	 */
+	wc = cq->queue;
+	head = wc->head;
+	if (head >= (unsigned) cq->ibcq.cqe) {
+		head = cq->ibcq.cqe;
+		next = 0;
+	} else
+		next = head + 1;
+	if (unlikely(next == wc->tail)) {
+		spin_unlock_irqrestore(&cq->lock, flags);
+		if (cq->ibcq.event_handler) {
+			struct ib_event ev;
+
+			ev.device = cq->ibcq.device;
+			ev.element.cq = &cq->ibcq;
+			ev.event = IB_EVENT_CQ_ERR;
+			cq->ibcq.event_handler(&ev, cq->ibcq.cq_context);
+		}
+		return;
+	}
+	if (cq->ip) {
+		wc->uqueue[head].wr_id = entry->wr_id;
+		wc->uqueue[head].status = entry->status;
+		wc->uqueue[head].opcode = entry->opcode;
+		wc->uqueue[head].vendor_err = entry->vendor_err;
+		wc->uqueue[head].byte_len = entry->byte_len;
+		wc->uqueue[head].ex.imm_data =
+			(__u32 __force)entry->ex.imm_data;
+		wc->uqueue[head].qp_num = entry->qp->qp_num;
+		wc->uqueue[head].src_qp = entry->src_qp;
+		wc->uqueue[head].wc_flags = entry->wc_flags;
+		wc->uqueue[head].pkey_index = entry->pkey_index;
+		wc->uqueue[head].slid = entry->slid;
+		wc->uqueue[head].sl = entry->sl;
+		wc->uqueue[head].dlid_path_bits = entry->dlid_path_bits;
+		wc->uqueue[head].port_num = entry->port_num;
+		/* Make sure entry is written before the head index. */
+		smp_wmb();
+	} else
+		wc->kqueue[head] = *entry;
+	wc->head = next;
+
+	if (cq->notify == IB_CQ_NEXT_COMP ||
+	    (cq->notify == IB_CQ_SOLICITED && solicited)) {
+		cq->notify = IB_CQ_NONE;
+		cq->triggered++;
+		/*
+		 * This will cause send_complete() to be called in
+		 * another thread.
+		 */
+		tasklet_hi_schedule(&cq->comptask);
+	}
+
+	spin_unlock_irqrestore(&cq->lock, flags);
+}
+
+/**
+ * qib_poll_cq - poll for work completion entries
+ * @ibcq: the completion queue to poll
+ * @num_entries: the maximum number of entries to return
+ * @entry: pointer to array where work completions are placed
+ *
+ * Returns the number of completion entries polled.
+ *
+ * This may be called from interrupt context.  Also called by ib_poll_cq()
+ * in the generic verbs code.
+ */
+int qib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
+{
+	struct qib_cq *cq = to_icq(ibcq);
+	struct qib_cq_wc *wc;
+	unsigned long flags;
+	int npolled;
+	u32 tail;
+
+	/* The kernel can only poll a kernel completion queue */
+	if (cq->ip) {
+		npolled = -EINVAL;
+		goto bail;
+	}
+
+	spin_lock_irqsave(&cq->lock, flags);
+
+	wc = cq->queue;
+	tail = wc->tail;
+	if (tail > (u32) cq->ibcq.cqe)
+		tail = (u32) cq->ibcq.cqe;
+	for (npolled = 0; npolled < num_entries; ++npolled, ++entry) {
+		if (tail == wc->head)
+			break;
+		/* The kernel doesn't need a RMB since it has the lock. */
+		*entry = wc->kqueue[tail];
+		if (tail >= cq->ibcq.cqe)
+			tail = 0;
+		else
+			tail++;
+	}
+	wc->tail = tail;
+
+	spin_unlock_irqrestore(&cq->lock, flags);
+
+bail:
+	return npolled;
+}
+
+static void send_complete(unsigned long data)
+{
+	struct qib_cq *cq = (struct qib_cq *)data;
+
+	/*
+	 * The completion handler will most likely rearm the notification
+	 * and poll for all pending entries.  If a new completion entry
+	 * is added while we are in this routine, tasklet_hi_schedule()
+	 * won't call us again until we return so we check triggered to
+	 * see if we need to call the handler again.
+	 */
+	for (;;) {
+		u8 triggered = cq->triggered;
+
+		cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
+
+		if (cq->triggered == triggered)
+			return;
+	}
+}
+
+/**
+ * qib_create_cq - create a completion queue
+ * @ibdev: the device this completion queue is attached to
+ * @entries: the minimum size of the completion queue
+ * @context: unused by the QLogic_IB driver
+ * @udata: user data for libibverbs.so
+ *
+ * Returns a pointer to the completion queue or negative errno values
+ * for failure.
+ *
+ * Called by ib_create_cq() in the generic verbs code.
+ */
+struct ib_cq *qib_create_cq(struct ib_device *ibdev, int entries,
+			    int comp_vector, struct ib_ucontext *context,
+			    struct ib_udata *udata)
+{
+	struct qib_ibdev *dev = to_idev(ibdev);
+	struct qib_cq *cq;
+	struct qib_cq_wc *wc;
+	struct ib_cq *ret;
+	u32 sz;
+
+	if (entries < 1 || entries > ib_qib_max_cqes) {
+		ret = ERR_PTR(-EINVAL);
+		goto done;
+	}
+
+	/* Allocate the completion queue structure. */
+	cq = kmalloc(sizeof(*cq), GFP_KERNEL);
+	if (!cq) {
+		ret = ERR_PTR(-ENOMEM);
+		goto done;
+	}
+
+	/*
+	 * Allocate the completion queue entries and head/tail pointers.
+	 * This is allocated separately so that it can be resized and
+	 * also mapped into user space.
+	 * We need to use vmalloc() in order to support mmap and large
+	 * numbers of entries.
+	 */
+	sz = sizeof(*wc);
+	if (udata && udata->outlen >= sizeof(__u64))
+		sz += sizeof(struct ib_uverbs_wc) * (entries + 1);
+	else
+		sz += sizeof(struct ib_wc) * (entries + 1);
+	wc = vmalloc_user(sz);
+	if (!wc) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_cq;
+	}
+
+	/*
+	 * Return the address of the WC as the offset to mmap.
+	 * See qib_mmap() for details.
+	 */
+	if (udata && udata->outlen >= sizeof(__u64)) {
+		int err;
+
+		cq->ip = qib_create_mmap_info(dev, sz, context, wc);
+		if (!cq->ip) {
+			ret = ERR_PTR(-ENOMEM);
+			goto bail_wc;
+		}
+
+		err = ib_copy_to_udata(udata, &cq->ip->offset,
+				       sizeof(cq->ip->offset));
+		if (err) {
+			ret = ERR_PTR(err);
+			goto bail_ip;
+		}
+	} else
+		cq->ip = NULL;
+
+	spin_lock(&dev->n_cqs_lock);
+	if (dev->n_cqs_allocated == ib_qib_max_cqs) {
+		spin_unlock(&dev->n_cqs_lock);
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_ip;
+	}
+
+	dev->n_cqs_allocated++;
+	spin_unlock(&dev->n_cqs_lock);
+
+	if (cq->ip) {
+		spin_lock_irq(&dev->pending_lock);
+		list_add(&cq->ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
+	/*
+	 * ib_create_cq() will initialize cq->ibcq except for cq->ibcq.cqe.
+	 * The number of entries should be >= the number requested or return
+	 * an error.
+	 */
+	cq->ibcq.cqe = entries;
+	cq->notify = IB_CQ_NONE;
+	cq->triggered = 0;
+	spin_lock_init(&cq->lock);
+	tasklet_init(&cq->comptask, send_complete, (unsigned long)cq);
+	wc->head = 0;
+	wc->tail = 0;
+	cq->queue = wc;
+
+	ret = &cq->ibcq;
+
+	goto done;
+
+bail_ip:
+	kfree(cq->ip);
+bail_wc:
+	vfree(wc);
+bail_cq:
+	kfree(cq);
+done:
+	return ret;
+}
+
+/**
+ * qib_destroy_cq - destroy a completion queue
+ * @ibcq: the completion queue to destroy.
+ *
+ * Returns 0 for success.
+ *
+ * Called by ib_destroy_cq() in the generic verbs code.
+ */
+int qib_destroy_cq(struct ib_cq *ibcq)
+{
+	struct qib_ibdev *dev = to_idev(ibcq->device);
+	struct qib_cq *cq = to_icq(ibcq);
+
+	tasklet_kill(&cq->comptask);
+	spin_lock(&dev->n_cqs_lock);
+	dev->n_cqs_allocated--;
+	spin_unlock(&dev->n_cqs_lock);
+	if (cq->ip)
+		kref_put(&cq->ip->ref, qib_release_mmap_info);
+	else
+		vfree(cq->queue);
+	kfree(cq);
+
+	return 0;
+}
+
+/**
+ * qib_req_notify_cq - change the notification type for a completion queue
+ * @ibcq: the completion queue
+ * @notify_flags: the type of notification to request
+ *
+ * Returns 0 for success.
+ *
+ * This may be called from interrupt context.  Also called by
+ * ib_req_notify_cq() in the generic verbs code.
+ */
+int qib_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags)
+{
+	struct qib_cq *cq = to_icq(ibcq);
+	unsigned long flags;
+	int ret = 0;
+
+	spin_lock_irqsave(&cq->lock, flags);
+	/*
+	 * Don't change IB_CQ_NEXT_COMP to IB_CQ_SOLICITED but allow
+	 * any other transitions (see C11-31 and C11-32 in ch. 11.4.2.2).
+	 */
+	if (cq->notify != IB_CQ_NEXT_COMP)
+		cq->notify = notify_flags & IB_CQ_SOLICITED_MASK;
+
+	if ((notify_flags & IB_CQ_REPORT_MISSED_EVENTS) &&
+	    cq->queue->head != cq->queue->tail)
+		ret = 1;
+
+	spin_unlock_irqrestore(&cq->lock, flags);
+
+	return ret;
+}
+
+/**
+ * qib_resize_cq - change the size of the CQ
+ * @ibcq: the completion queue
+ *
+ * Returns 0 for success.
+ */
+int qib_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata)
+{
+	struct qib_cq *cq = to_icq(ibcq);
+	struct qib_cq_wc *old_wc;
+	struct qib_cq_wc *wc;
+	u32 head, tail, n;
+	int ret;
+	u32 sz;
+
+	if (cqe < 1 || cqe > ib_qib_max_cqes) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	/*
+	 * Need to use vmalloc() if we want to support large #s of entries.
+	 */
+	sz = sizeof(*wc);
+	if (udata && udata->outlen >= sizeof(__u64))
+		sz += sizeof(struct ib_uverbs_wc) * (cqe + 1);
+	else
+		sz += sizeof(struct ib_wc) * (cqe + 1);
+	wc = vmalloc_user(sz);
+	if (!wc) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	/* Check that we can write the offset to mmap. */
+	if (udata && udata->outlen >= sizeof(__u64)) {
+		__u64 offset = 0;
+
+		ret = ib_copy_to_udata(udata, &offset, sizeof(offset));
+		if (ret)
+			goto bail_free;
+	}
+
+	spin_lock_irq(&cq->lock);
+	/*
+	 * Make sure head and tail are sane since they
+	 * might be user writable.
+	 */
+	old_wc = cq->queue;
+	head = old_wc->head;
+	if (head > (u32) cq->ibcq.cqe)
+		head = (u32) cq->ibcq.cqe;
+	tail = old_wc->tail;
+	if (tail > (u32) cq->ibcq.cqe)
+		tail = (u32) cq->ibcq.cqe;
+	if (head < tail)
+		n = cq->ibcq.cqe + 1 + head - tail;
+	else
+		n = head - tail;
+	if (unlikely((u32)cqe < n)) {
+		ret = -EINVAL;
+		goto bail_unlock;
+	}
+	for (n = 0; tail != head; n++) {
+		if (cq->ip)
+			wc->uqueue[n] = old_wc->uqueue[tail];
+		else
+			wc->kqueue[n] = old_wc->kqueue[tail];
+		if (tail == (u32) cq->ibcq.cqe)
+			tail = 0;
+		else
+			tail++;
+	}
+	cq->ibcq.cqe = cqe;
+	wc->head = n;
+	wc->tail = 0;
+	cq->queue = wc;
+	spin_unlock_irq(&cq->lock);
+
+	vfree(old_wc);
+
+	if (cq->ip) {
+		struct qib_ibdev *dev = to_idev(ibcq->device);
+		struct qib_mmap_info *ip = cq->ip;
+
+		qib_update_mmap_info(dev, ip, sz, wc);
+
+		/*
+		 * Return the offset to mmap.
+		 * See qib_mmap() for details.
+		 */
+		if (udata && udata->outlen >= sizeof(__u64)) {
+			ret = ib_copy_to_udata(udata, &ip->offset,
+					       sizeof(ip->offset));
+			if (ret)
+				goto bail;
+		}
+
+		spin_lock_irq(&dev->pending_lock);
+		if (list_empty(&ip->pending_mmaps))
+			list_add(&ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
+	ret = 0;
+	goto bail;
+
+bail_unlock:
+	spin_unlock_irq(&cq->lock);
+bail_free:
+	vfree(wc);
+bail:
+	return ret;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 10/51] IB/qib: Add qib_debug.h

[Ralph Campbell]

creates the qib_debug.h file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_debug.h |   85 +++++++++++++++++++++++++++++++++
 1 files changed, 85 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_debug.h

diff --git a/drivers/infiniband/hw/qib/qib_debug.h b/drivers/infiniband/hw/qib/qib_debug.h
new file mode 100644
index 0000000..e6a77cd
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_debug.h
@@ -0,0 +1,85 @@
+/*
+ * Copyright (c) 2006, 2007, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef _QIB_DEBUG_H
+#define _QIB_DEBUG_H
+
+#ifndef _QIB_DEBUGGING  /* debugging enabled or not */
+#define _QIB_DEBUGGING 1
+#endif
+
+#if _QIB_DEBUGGING
+
+/*
+ * Mask values for debugging.  The scheme allows us to compile out any
+ * of the debug tracing stuff, and if compiled in, to enable or disable
+ * dynamically.  This can be set at modprobe time also:
+ *      modprobe qlogic_ib.ko qlogic_ib_debug=7
+ */
+
+#define __QIB_INFO        0x1   /* logging of dev_info and dev_err */
+#define __QIB_DBG         0x2   /* generic debug */
+/* leave some low verbosity spots open */
+#define __QIB_RVPKTDBG    0x10  /* verbose pktrcv debug */
+#define __QIB_INITDBG     0x20  /* init-level debug */
+#define __QIB_VERBDBG     0x40  /* very verbose debug */
+#define __QIB_PKTDBG      0x80  /* print packet data */
+#define __QIB_PROCDBG     0x100 /* process init/exit debug */
+#define __QIB_MMDBG       0x200 /* mmap, etc debug */
+#define __QIB_ERRPKTDBG   0x400 /* packet error debugging */
+#define __QIB_SDMADBG     0x800 /* Send DMA */
+#define __QIB_VPKTDBG     0x1000 /* Dump IB contents being copied to send buf */
+#define __QIB_LINKVERBDBG 0x200000      /* very verbose linkchange debug */
+
+#else                           /* _QIB_DEBUGGING */
+
+/*
+ * define all of these even with debugging off, for the few places that do
+ * if(qlogic_ib_debug & _QIB_xyzzy), but in a way that will make the
+ * compiler eliminate the code
+ */
+
+#define __QIB_DBG       0x0     /* generic debug */
+#define __QIB_VERBDBG   0x0     /* very verbose debug */
+#define __QIB_PKTDBG    0x0     /* print packet data */
+#define __QIB_PROCDBG   0x0      /* process init/exit debug */
+#define __QIB_MMDBG     0x0     /* mmap, etc debug */
+#define __QIB_ERRPKTDBG 0x0     /* packet error debugging */
+#define __QIB_SDMADBG   0x0     /* Send DMA */
+#define __QIB_LINKVERBDBG 0x0   /* very verbose linkchange debug */
+
+#endif                          /* _QIB_DEBUGGING */
+
+#define __QIB_VERBOSEDBG __QIB_VERBDBG
+
+#endif                          /* _QIB_DEBUG_H */

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 11/51] IB/qib: Add qib_diag.c

[Ralph Campbell]

creates the qib_diag.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_diag.c |  908 ++++++++++++++++++++++++++++++++++
 1 files changed, 908 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_diag.c

diff --git a/drivers/infiniband/hw/qib/qib_diag.c b/drivers/infiniband/hw/qib/qib_diag.c
new file mode 100644
index 0000000..f5cb4cc
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_diag.c
@@ -0,0 +1,908 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This file contains support for diagnostic functions.  It is accessed by
+ * opening the qib_diag device, normally minor number 129.  Diagnostic use
+ * of the QLogic_IB chip may render the chip or board unusable until the
+ * driver is unloaded, or in some cases, until the system is rebooted.
+ *
+ * Accesses to the chip through this interface are not similar to going
+ * through the /sys/bus/pci resource mmap interface.
+ */
+
+#include <linux/io.h>
+#include <linux/pci.h>
+#include <linux/poll.h>
+#include <linux/vmalloc.h>
+#include <linux/fs.h>
+#include <linux/uaccess.h>
+
+#include "qib.h"
+#include "qib_common.h"
+
+/*
+ * Each client that opens the diag device must read then write
+ * offset 0, to prevent lossage from random cat or od. diag_state
+ * sequences this "handshake".
+ */
+enum diag_state { UNUSED = 0, OPENED, INIT, READY };
+
+/* State for an individual client. PID so children cannot abuse handshake */
+static struct qib_diag_client {
+	struct qib_diag_client *next;
+	struct qib_devdata *dd;
+	pid_t pid;
+	enum diag_state state;
+} *client_pool;
+
+/*
+ * Get a client struct. Recycled if possible, else kmalloc.
+ * Must be called with qib_mutex held
+ */
+static struct qib_diag_client *get_client(struct qib_devdata *dd)
+{
+	struct qib_diag_client *dc;
+
+	dc = client_pool;
+	if (dc)
+		/* got from pool remove it and use */
+		client_pool = dc->next;
+	else
+		/* None in pool, alloc and init */
+		dc = kmalloc(sizeof *dc, GFP_KERNEL);
+
+	if (dc) {
+		dc->next = NULL;
+		dc->dd = dd;
+		dc->pid = current->pid;
+		dc->state = OPENED;
+	}
+	return dc;
+}
+
+/*
+ * Return to pool. Must be called with qib_mutex held
+ */
+static void return_client(struct qib_diag_client *dc)
+{
+	struct qib_devdata *dd = dc->dd;
+	struct qib_diag_client *tdc, *rdc;
+
+	rdc = NULL;
+	if (dc == dd->diag_client) {
+		dd->diag_client = dc->next;
+		rdc = dc;
+	} else {
+		tdc = dc->dd->diag_client;
+		while (tdc) {
+			if (dc == tdc->next) {
+				tdc->next = dc->next;
+				rdc = dc;
+				break;
+			}
+			tdc = tdc->next;
+		}
+	}
+	if (rdc) {
+		rdc->state = UNUSED;
+		rdc->dd = NULL;
+		rdc->pid = 0;
+		rdc->next = client_pool;
+		client_pool = rdc;
+	}
+}
+
+static int qib_diag_open(struct inode *in, struct file *fp);
+static int qib_diag_release(struct inode *in, struct file *fp);
+static ssize_t qib_diag_read(struct file *fp, char __user *data,
+			     size_t count, loff_t *off);
+static ssize_t qib_diag_write(struct file *fp, const char __user *data,
+			      size_t count, loff_t *off);
+
+static const struct file_operations diag_file_ops = {
+	.owner = THIS_MODULE,
+	.write = qib_diag_write,
+	.read = qib_diag_read,
+	.open = qib_diag_open,
+	.release = qib_diag_release
+};
+
+static atomic_t diagpkt_count = ATOMIC_INIT(0);
+static struct cdev *diagpkt_cdev;
+static struct device *diagpkt_device;
+
+static ssize_t qib_diagpkt_write(struct file *fp, const char __user *data,
+				 size_t count, loff_t *off);
+
+static const struct file_operations diagpkt_file_ops = {
+	.owner = THIS_MODULE,
+	.write = qib_diagpkt_write,
+};
+
+int qib_diag_add(struct qib_devdata *dd)
+{
+	char name[16];
+	int ret = 0;
+
+	if (atomic_inc_return(&diagpkt_count) == 1) {
+		ret = qib_cdev_init(QIB_DIAGPKT_MINOR, "ipath_diagpkt",
+				    &diagpkt_file_ops, &diagpkt_cdev,
+				    &diagpkt_device);
+		if (ret)
+			goto done;
+	}
+
+	snprintf(name, sizeof(name), "ipath_diag%d", dd->unit);
+	ret = qib_cdev_init(QIB_DIAG_MINOR_BASE + dd->unit, name,
+			    &diag_file_ops, &dd->diag_cdev,
+			    &dd->diag_device);
+done:
+	return ret;
+}
+
+static void qib_unregister_observers(struct qib_devdata *dd);
+
+void qib_diag_remove(struct qib_devdata *dd)
+{
+	struct qib_diag_client *dc;
+
+	if (atomic_dec_and_test(&diagpkt_count))
+		qib_cdev_cleanup(&diagpkt_cdev, &diagpkt_device);
+
+	qib_cdev_cleanup(&dd->diag_cdev, &dd->diag_device);
+
+	/*
+	 * Return all diag_clients of this device. There should be none,
+	 * as we are "guaranteed" that no clients are still open
+	 */
+	while (dd->diag_client)
+		return_client(dd->diag_client);
+
+	/* Now clean up all unused client structs */
+	while (client_pool) {
+		dc = client_pool;
+		client_pool = dc->next;
+		kfree(dc);
+	}
+	/* Clean up observer list */
+	qib_unregister_observers(dd);
+}
+
+/* qib_remap_ioaddr32 - remap an offset into chip address space to __iomem *
+ *
+ * @dd: the qlogic_ib device
+ * @offs: the offset in chip-space
+ * @cntp: Pointer to max (byte) count for transfer starting at offset
+ * This returns a u32 __iomem * so it can be used for both 64 and 32-bit
+ * mapping. It is needed because with the use of PAT for control of
+ * write-combining, the logically contiguous address-space of the chip
+ * may be split into virtually non-contiguous spaces, with different
+ * attributes, which are them mapped to contiguous physical space
+ * based from the first BAR.
+ *
+ * The code below makes the same assumptions as were made in
+ * init_chip_wc_pat() (qib_init.c), copied here:
+ * Assumes chip address space looks like:
+ *		- kregs + sregs + cregs + uregs (in any order)
+ *		- piobufs (2K and 4K bufs in either order)
+ *	or:
+ *		- kregs + sregs + cregs (in any order)
+ *		- piobufs (2K and 4K bufs in either order)
+ *		- uregs
+ *
+ * If cntp is non-NULL, returns how many bytes from offset can be accessed
+ * Returns 0 if the offset is not mapped.
+ */
+static u32 __iomem *qib_remap_ioaddr32(struct qib_devdata *dd, u32 offset,
+				       u32 *cntp)
+{
+	u32 kreglen;
+	u32 snd_bottom, snd_lim = 0;
+	u32 __iomem *krb32 = (u32 __iomem *)dd->kregbase;
+	u32 __iomem *map = NULL;
+	u32 cnt = 0;
+
+	/* First, simplest case, offset is within the first map. */
+	kreglen = (dd->kregend - dd->kregbase) * sizeof(u64);
+	if (offset < kreglen) {
+		map = krb32 + (offset / sizeof(u32));
+		cnt = kreglen - offset;
+		goto mapped;
+	}
+
+	/*
+	 * Next check for user regs, the next most common case,
+	 * and a cheap check because if they are not in the first map
+	 * they are last in chip.
+	 */
+	if (dd->userbase) {
+		/* If user regs mapped, they are after send, so set limit. */
+		u32 ulim = (dd->cfgctxts * dd->ureg_align) + dd->uregbase;
+		snd_lim = dd->uregbase;
+		krb32 = (u32 __iomem *)dd->userbase;
+		if (offset >= dd->uregbase && offset < ulim) {
+			map = krb32 + (offset - dd->uregbase) / sizeof(u32);
+			cnt = ulim - offset;
+			goto mapped;
+		}
+	}
+
+	/*
+	 * Lastly, check for offset within Send Buffers.
+	 * This is gnarly because struct devdata is deliberately vague
+	 * about things like 7322 VL15 buffers, and we are not in
+	 * chip-specific code here, so should not make many assumptions.
+	 * The one we _do_ make is that the only chip that has more sndbufs
+	 * than we admit is the 7322, and it has userregs above that, so
+	 * we know the snd_lim.
+	 */
+	/* Assume 2K buffers are first. */
+	snd_bottom = dd->pio2k_bufbase;
+	if (snd_lim == 0) {
+		u32 tot2k = dd->piobcnt2k * ALIGN(dd->piosize2k, dd->palign);
+		snd_lim = snd_bottom + tot2k;
+	}
+	/* If 4k buffers exist, account for them by bumping
+	 * appropriate limit.
+	 */
+	if (dd->piobcnt4k) {
+		u32 tot4k = dd->piobcnt4k * dd->align4k;
+		u32 offs4k = dd->piobufbase >> 32;
+		if (snd_bottom > offs4k)
+			snd_bottom = offs4k;
+		else {
+			/* 4k above 2k. Bump snd_lim, if needed*/
+			if (!dd->userbase)
+				snd_lim = offs4k + tot4k;
+		}
+	}
+	/*
+	 * Judgement call: can we ignore the space between SendBuffs and
+	 * UserRegs, where we would like to see vl15 buffs, but not more?
+	 */
+	if (offset >= snd_bottom && offset < snd_lim) {
+		offset -= snd_bottom;
+		map = (u32 __iomem *)dd->piobase + (offset / sizeof(u32));
+		cnt = snd_lim - offset;
+	}
+
+mapped:
+	if (cntp)
+		*cntp = cnt;
+	return map;
+}
+
+/*
+ * qib_read_umem64 - read a 64-bit quantity from the chip into user space
+ * @dd: the qlogic_ib device
+ * @uaddr: the location to store the data in user memory
+ * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
+ * @count: number of bytes to copy (multiple of 32 bits)
+ *
+ * This function also localizes all chip memory accesses.
+ * The copy should be written such that we read full cacheline packets
+ * from the chip.  This is usually used for a single qword
+ *
+ * NOTE:  This assumes the chip address is 64-bit aligned.
+ */
+static int qib_read_umem64(struct qib_devdata *dd, void __user *uaddr,
+			   u32 regoffs, size_t count)
+{
+	const u64 __iomem *reg_addr;
+	const u64 __iomem *reg_end;
+	u32 limit;
+	int ret;
+
+	reg_addr = (const u64 __iomem *)qib_remap_ioaddr32(dd, regoffs, &limit);
+	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (count >= limit)
+		count = limit;
+	reg_end = reg_addr + (count / sizeof(u64));
+
+	/* not very efficient, but it works for now */
+	while (reg_addr < reg_end) {
+		u64 data = readq(reg_addr);
+
+		if (copy_to_user(uaddr, &data, sizeof(u64))) {
+			ret = -EFAULT;
+			goto bail;
+		}
+		reg_addr++;
+		uaddr += sizeof(u64);
+	}
+	ret = 0;
+bail:
+	return ret;
+}
+
+/*
+ * qib_write_umem64 - write a 64-bit quantity to the chip from user space
+ * @dd: the qlogic_ib device
+ * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
+ * @uaddr: the source of the data in user memory
+ * @count: the number of bytes to copy (multiple of 32 bits)
+ *
+ * This is usually used for a single qword
+ * NOTE:  This assumes the chip address is 64-bit aligned.
+ */
+
+static int qib_write_umem64(struct qib_devdata *dd, u32 regoffs,
+			    const void __user *uaddr, size_t count)
+{
+	u64 __iomem *reg_addr;
+	const u64 __iomem *reg_end;
+	u32 limit;
+	int ret;
+
+	reg_addr = (u64 __iomem *)qib_remap_ioaddr32(dd, regoffs, &limit);
+	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (count >= limit)
+		count = limit;
+	reg_end = reg_addr + (count / sizeof(u64));
+
+	/* not very efficient, but it works for now */
+	while (reg_addr < reg_end) {
+		u64 data;
+		if (copy_from_user(&data, uaddr, sizeof(data))) {
+			ret = -EFAULT;
+			goto bail;
+		}
+		writeq(data, reg_addr);
+
+		reg_addr++;
+		uaddr += sizeof(u64);
+	}
+	ret = 0;
+bail:
+	return ret;
+}
+
+/*
+ * qib_read_umem32 - read a 32-bit quantity from the chip into user space
+ * @dd: the qlogic_ib device
+ * @uaddr: the location to store the data in user memory
+ * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
+ * @count: number of bytes to copy
+ *
+ * read 32 bit values, not 64 bit; for memories that only
+ * support 32 bit reads; usually a single dword.
+ */
+static int qib_read_umem32(struct qib_devdata *dd, void __user *uaddr,
+			   u32 regoffs, size_t count)
+{
+	const u32 __iomem *reg_addr;
+	const u32 __iomem *reg_end;
+	u32 limit;
+	int ret;
+
+	reg_addr = qib_remap_ioaddr32(dd, regoffs, &limit);
+	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (count >= limit)
+		count = limit;
+	reg_end = reg_addr + (count / sizeof(u32));
+
+	/* not very efficient, but it works for now */
+	while (reg_addr < reg_end) {
+		u32 data = readl(reg_addr);
+
+		if (copy_to_user(uaddr, &data, sizeof(data))) {
+			ret = -EFAULT;
+			goto bail;
+		}
+
+		reg_addr++;
+		uaddr += sizeof(u32);
+
+	}
+	ret = 0;
+bail:
+	return ret;
+}
+
+/*
+ * qib_write_umem32 - write a 32-bit quantity to the chip from user space
+ * @dd: the qlogic_ib device
+ * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
+ * @uaddr: the source of the data in user memory
+ * @count: number of bytes to copy
+ *
+ * write 32 bit values, not 64 bit; for memories that only
+ * support 32 bit write; usually a single dword.
+ */
+
+static int qib_write_umem32(struct qib_devdata *dd, u32 regoffs,
+			    const void __user *uaddr, size_t count)
+{
+	u32 __iomem *reg_addr;
+	const u32 __iomem *reg_end;
+	u32 limit;
+	int ret;
+
+	reg_addr = qib_remap_ioaddr32(dd, regoffs, &limit);
+	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (count >= limit)
+		count = limit;
+	reg_end = reg_addr + (count / sizeof(u32));
+
+	while (reg_addr < reg_end) {
+		u32 data;
+
+		if (copy_from_user(&data, uaddr, sizeof(data))) {
+			ret = -EFAULT;
+			goto bail;
+		}
+		writel(data, reg_addr);
+
+		reg_addr++;
+		uaddr += sizeof(u32);
+	}
+	ret = 0;
+bail:
+	return ret;
+}
+
+static int qib_diag_open(struct inode *in, struct file *fp)
+{
+	int unit = iminor(in) - QIB_DIAG_MINOR_BASE;
+	struct qib_devdata *dd;
+	struct qib_diag_client *dc;
+	int ret;
+
+	mutex_lock(&qib_mutex);
+
+	dd = qib_lookup(unit);
+
+	if (dd == NULL || !(dd->flags & QIB_PRESENT) ||
+	    !dd->kregbase) {
+		ret = -ENODEV;
+		goto bail;
+	}
+
+	dc = get_client(dd);
+	if (!dc) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+	dc->next = dd->diag_client;
+	dd->diag_client = dc;
+	fp->private_data = dc;
+	ret = 0;
+bail:
+	mutex_unlock(&qib_mutex);
+
+	return ret;
+}
+
+/**
+ * qib_diagpkt_write - write an IB packet
+ * @fp: the diag data device file pointer
+ * @data: qib_diag_pkt structure saying where to get the packet
+ * @count: size of data to write
+ * @off: unused by this code
+ */
+static ssize_t qib_diagpkt_write(struct file *fp,
+				 const char __user *data,
+				 size_t count, loff_t *off)
+{
+	u32 __iomem *piobuf;
+	u32 plen, clen, pbufn;
+	struct qib_diag_xpkt dp;
+	u32 *tmpbuf = NULL;
+	struct qib_devdata *dd;
+	struct qib_pportdata *ppd;
+	ssize_t ret = 0;
+
+	if (count != sizeof(dp)) {
+		qib_dbg("Invalid cmd length %u\n", (unsigned)count);
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (copy_from_user(&dp, data, sizeof(dp))) {
+		ret = -EFAULT;
+		goto bail;
+	}
+
+	dd = qib_lookup(dp.unit);
+	if (!dd || !(dd->flags & QIB_PRESENT) || !dd->kregbase) {
+		qib_cdbg(VERBOSE, "illegal unit %u for diag data send\n",
+			 dp.unit);
+		ret = -ENODEV;
+		goto bail;
+	}
+	if (!(dd->flags & QIB_INITTED)) {
+		/* no hardware, freeze, etc. */
+		qib_cdbg(VERBOSE, "unit %u not usable\n", dd->unit);
+		ret = -ENODEV;
+		goto bail;
+	}
+
+	if (dp.version != _DIAG_XPKT_VERS) {
+		qib_dev_err(dd, "Invalid version %u for diagpkt_write\n",
+			    dp.version);
+		ret = -EINVAL;
+		goto bail;
+	}
+	/* send count must be an exact number of dwords */
+	if (dp.len & 3) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (!dp.port || dp.port > dd->num_pports) {
+		qib_dbg("unit %u doesn't have physical port %u\n",
+			 dd->unit, dp.port);
+		ret = -EINVAL;
+		goto bail;
+	}
+	ppd = &dd->pport[dp.port - 1];
+
+	/* need total length before first word written */
+	/* +1 word is for the qword padding */
+	plen = sizeof(u32) + dp.len;
+	clen = dp.len >> 2;
+
+	if ((plen + 4) > ppd->ibmaxlen) {
+		qib_dbg("Pkt len 0x%x > ibmaxlen %x\n",
+			plen - 4, ppd->ibmaxlen);
+		ret = -EINVAL;
+		goto bail;      /* before writing pbc */
+	}
+	tmpbuf = vmalloc(plen);
+	if (!tmpbuf) {
+		qib_devinfo(dd->pcidev, "Unable to allocate tmp buffer, "
+			 "failing\n");
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	if (copy_from_user(tmpbuf,
+			   (const void __user *) (unsigned long) dp.data,
+			   dp.len)) {
+		ret = -EFAULT;
+		goto bail;
+	}
+
+	plen >>= 2;             /* in dwords */
+
+	if (dp.pbc_wd == 0)
+		dp.pbc_wd = plen;
+
+	piobuf = dd->f_getsendbuf(ppd, dp.pbc_wd, &pbufn);
+	if (!piobuf) {
+		qib_cdbg(VERBOSE, "No PIO buffers avail unit for %u\n",
+			   dd->unit);
+		ret = -EBUSY;
+		goto bail;
+	}
+	/* disarm it just to be extra sure */
+	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(pbufn));
+
+	/* disable header check on pbufn for this packet */
+	dd->f_txchk_change(dd, pbufn, 1, TXCHK_CHG_TYPE_DIS1, NULL);
+
+	if (qib_debug & __QIB_PKTDBG)
+		qib_cdbg(VERBOSE, "unit %u 0x%x+1w pio%d\n",
+			 dd->unit, plen - 1, pbufn);
+
+	writeq(dp.pbc_wd, piobuf);
+	/*
+	 * Copy all but the trigger word, then flush, so it's written
+	 * to chip before trigger word, then write trigger word, then
+	 * flush again, so packet is sent.
+	 */
+	if (dd->flags & QIB_PIO_FLUSH_WC) {
+		qib_flush_wc();
+		qib_pio_copy(piobuf + 2, tmpbuf, clen - 1);
+		qib_flush_wc();
+		__raw_writel(tmpbuf[clen - 1], piobuf + clen + 1);
+	} else
+		qib_pio_copy(piobuf + 2, tmpbuf, clen);
+
+	if (dd->flags & QIB_USE_SPCL_TRIG) {
+		u32 spcl_off = (pbufn >= dd->piobcnt2k) ? 2047 : 1023;
+
+		qib_flush_wc();
+		__raw_writel(0xaebecede, piobuf + spcl_off);
+	}
+
+	/*
+	 * Ensure buffer is written to the chip, then re-enable
+	 * header checks (if supported by chip).  The txchk
+	 * code will ensure seen by chip before returning.
+	 */
+	qib_flush_wc();
+	qib_sendbuf_done(dd, pbufn);
+	dd->f_txchk_change(dd, pbufn, 1, TXCHK_CHG_TYPE_ENAB1, NULL);
+
+	ret = sizeof(dp);
+
+bail:
+	vfree(tmpbuf);
+	return ret;
+}
+
+static int qib_diag_release(struct inode *in, struct file *fp)
+{
+	mutex_lock(&qib_mutex);
+	return_client(fp->private_data);
+	fp->private_data = NULL;
+	mutex_unlock(&qib_mutex);
+	return 0;
+}
+
+/*
+ * Chip-specific code calls to register its interest in
+ * a specific range.
+ */
+struct diag_observer_list_elt {
+	struct diag_observer_list_elt *next;
+	const struct diag_observer *op;
+};
+
+int qib_register_observer(struct qib_devdata *dd,
+			  const struct diag_observer *op)
+{
+	struct diag_observer_list_elt *olp;
+	int ret = -EINVAL;
+
+	if (!dd || !op)
+		goto bail;
+	ret = -ENOMEM;
+	olp = vmalloc(sizeof *olp);
+	if (!olp) {
+		printk(KERN_ERR QIB_DRV_NAME ": vmalloc for observer failed\n");
+		goto bail;
+	}
+	if (olp) {
+		unsigned long flags;
+
+		spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+		olp->op = op;
+		olp->next = dd->diag_observer_list;
+		dd->diag_observer_list = olp;
+		spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+		ret = 0;
+	}
+bail:
+	return ret;
+}
+
+/* Remove all registered observers when device is closed */
+static void qib_unregister_observers(struct qib_devdata *dd)
+{
+	struct diag_observer_list_elt *olp;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+	olp = dd->diag_observer_list;
+	while (olp) {
+		/* Pop one observer, let go of lock */
+		dd->diag_observer_list = olp->next;
+		spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+		vfree(olp);
+		/* try again. */
+		spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+		olp = dd->diag_observer_list;
+	}
+	spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+}
+
+/*
+ * Find the observer, if any, for the specified address. Initial implementation
+ * is simple stack of observers. This must be called with diag transaction
+ * lock held.
+ */
+static const struct diag_observer *diag_get_observer(struct qib_devdata *dd,
+						     u32 addr)
+{
+	struct diag_observer_list_elt *olp;
+	const struct diag_observer *op = NULL;
+
+	olp = dd->diag_observer_list;
+	while (olp) {
+		op = olp->op;
+		if (addr >= op->bottom && addr <= op->top)
+			break;
+		olp = olp->next;
+	}
+	if (!olp)
+		op = NULL;
+
+	return op;
+}
+
+static ssize_t qib_diag_read(struct file *fp, char __user *data,
+			     size_t count, loff_t *off)
+{
+	struct qib_diag_client *dc = fp->private_data;
+	struct qib_devdata *dd = dc->dd;
+	void __iomem *kreg_base;
+	ssize_t ret;
+
+	if (dc->pid != current->pid) {
+		ret = -EPERM;
+		goto bail;
+	}
+
+	kreg_base = dd->kregbase;
+
+	if (count == 0)
+		ret = 0;
+	else if ((count % 4) || (*off % 4))
+		/* address or length is not 32-bit aligned, hence invalid */
+		ret = -EINVAL;
+	else if (dc->state < READY && (*off || count != 8))
+		ret = -EINVAL;  /* prevent cat /dev/qib_diag* */
+	else {
+		unsigned long flags;
+		u64 data64 = 0;
+		int use_32;
+		const struct diag_observer *op;
+
+		use_32 = (count % 8) || (*off % 8);
+		ret = -1;
+		spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+		/*
+		 * Check for observer on this address range.
+		 * we only support a single 32 or 64-bit read
+		 * via observer, currently.
+		 */
+		op = diag_get_observer(dd, *off);
+		if (op) {
+			u32 offset = *off;
+			ret = op->hook(dd, op, offset, &data64, 0, use_32);
+		}
+		/*
+		 * We need to release lock before any copy_to_user(),
+		 * whether implicit in qib_read_umem* or explicit below.
+		 */
+		spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+		if (!op) {
+			if (use_32)
+				/*
+				 * Address or length is not 64-bit aligned;
+				 * do 32-bit rd
+				 */
+				ret = qib_read_umem32(dd, data, (u32) *off,
+						      count);
+			else
+				ret = qib_read_umem64(dd, data, (u32) *off,
+						      count);
+		} else if (ret == count) {
+			/* Below finishes case where observer existed */
+			ret = copy_to_user(data, &data64, use_32 ?
+					   sizeof(u32) : sizeof(u64));
+			if (ret)
+				ret = -EFAULT;
+		}
+	}
+
+	if (ret >= 0) {
+		*off += count;
+		ret = count;
+		if (dc->state == OPENED)
+			dc->state = INIT;
+	}
+bail:
+	return ret;
+}
+
+static ssize_t qib_diag_write(struct file *fp, const char __user *data,
+			      size_t count, loff_t *off)
+{
+	struct qib_diag_client *dc = fp->private_data;
+	struct qib_devdata *dd = dc->dd;
+	void __iomem *kreg_base;
+	ssize_t ret;
+
+	if (dc->pid != current->pid) {
+		ret = -EPERM;
+		goto bail;
+	}
+
+	kreg_base = dd->kregbase;
+
+	if (count == 0)
+		ret = 0;
+	else if ((count % 4) || (*off % 4))
+		/* address or length is not 32-bit aligned, hence invalid */
+		ret = -EINVAL;
+	else if (dc->state < READY &&
+		((*off || count != 8) || dc->state != INIT))
+		/* No writes except second-step of init seq */
+		ret = -EINVAL;  /* before any other write allowed */
+	else {
+		unsigned long flags;
+		const struct diag_observer *op = NULL;
+		int use_32 =  (count % 8) || (*off % 8);
+
+		/*
+		 * Check for observer on this address range.
+		 * We only support a single 32 or 64-bit write
+		 * via observer, currently. This helps, because
+		 * we would otherwise have to jump through hoops
+		 * to make "diag transaction" meaningful when we
+		 * cannot do a copy_from_user while holding the lock.
+		 */
+		if (count == 4 || count == 8) {
+			u64 data64;
+			u32 offset = *off;
+			ret = copy_from_user(&data64, data, count);
+			if (ret) {
+				ret = -EFAULT;
+				goto bail;
+			}
+			spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+			op = diag_get_observer(dd, *off);
+			if (op)
+				ret = op->hook(dd, op, offset, &data64, ~0Ull,
+					       use_32);
+			spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+		}
+
+		if (!op) {
+			if (use_32)
+				/*
+				 * Address or length is not 64-bit aligned;
+				 * do 32-bit write
+				 */
+				ret = qib_write_umem32(dd, (u32) *off, data,
+						       count);
+			else
+				ret = qib_write_umem64(dd, (u32) *off, data,
+						       count);
+		}
+	}
+
+	if (ret >= 0) {
+		*off += count;
+		ret = count;
+		if (dc->state == INIT)
+			dc->state = READY; /* all read/write OK now */
+	}
+bail:
+	return ret;
+}
+

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 12/51] IB/qib: Add qib_dma.c

[Ralph Campbell]

creates the qib_dma.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_dma.c |  182 +++++++++++++++++++++++++++++++++++
 1 files changed, 182 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_dma.c

diff --git a/drivers/infiniband/hw/qib/qib_dma.c b/drivers/infiniband/hw/qib/qib_dma.c
new file mode 100644
index 0000000..f84085e
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_dma.c
@@ -0,0 +1,182 @@
+/*
+ * Copyright (c) 2006, 2009 QLogic, Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/types.h>
+#include <linux/scatterlist.h>
+
+#include "qib_verbs.h"
+
+#define BAD_DMA_ADDRESS ((u64) 0)
+
+/*
+ * The following functions implement driver specific replacements
+ * for the ib_dma_*() functions.
+ *
+ * These functions return kernel virtual addresses instead of
+ * device bus addresses since the driver uses the CPU to copy
+ * data instead of using hardware DMA.
+ */
+
+static int qib_mapping_error(struct ib_device *dev, u64 dma_addr)
+{
+	return dma_addr == BAD_DMA_ADDRESS;
+}
+
+static u64 qib_dma_map_single(struct ib_device *dev, void *cpu_addr,
+			      size_t size, enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+	return (u64) cpu_addr;
+}
+
+static void qib_dma_unmap_single(struct ib_device *dev, u64 addr, size_t size,
+				 enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+}
+
+static u64 qib_dma_map_page(struct ib_device *dev, struct page *page,
+			    unsigned long offset, size_t size,
+			    enum dma_data_direction direction)
+{
+	u64 addr;
+
+	BUG_ON(!valid_dma_direction(direction));
+
+	if (offset + size > PAGE_SIZE) {
+		addr = BAD_DMA_ADDRESS;
+		goto done;
+	}
+
+	addr = (u64) page_address(page);
+	if (addr)
+		addr += offset;
+	/* TODO: handle highmem pages */
+
+done:
+	return addr;
+}
+
+static void qib_dma_unmap_page(struct ib_device *dev, u64 addr, size_t size,
+			       enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+}
+
+static int qib_map_sg(struct ib_device *dev, struct scatterlist *sgl,
+		      int nents, enum dma_data_direction direction)
+{
+	struct scatterlist *sg;
+	u64 addr;
+	int i;
+	int ret = nents;
+
+	BUG_ON(!valid_dma_direction(direction));
+
+	for_each_sg(sgl, sg, nents, i) {
+		addr = (u64) page_address(sg_page(sg));
+		/* TODO: handle highmem pages */
+		if (!addr) {
+			ret = 0;
+			break;
+		}
+	}
+	return ret;
+}
+
+static void qib_unmap_sg(struct ib_device *dev,
+			 struct scatterlist *sg, int nents,
+			 enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+}
+
+static u64 qib_sg_dma_address(struct ib_device *dev, struct scatterlist *sg)
+{
+	u64 addr = (u64) page_address(sg_page(sg));
+
+	if (addr)
+		addr += sg->offset;
+	return addr;
+}
+
+static unsigned int qib_sg_dma_len(struct ib_device *dev,
+				   struct scatterlist *sg)
+{
+	return sg->length;
+}
+
+static void qib_sync_single_for_cpu(struct ib_device *dev, u64 addr,
+				    size_t size, enum dma_data_direction dir)
+{
+}
+
+static void qib_sync_single_for_device(struct ib_device *dev, u64 addr,
+				       size_t size,
+				       enum dma_data_direction dir)
+{
+}
+
+static void *qib_dma_alloc_coherent(struct ib_device *dev, size_t size,
+				    u64 *dma_handle, gfp_t flag)
+{
+	struct page *p;
+	void *addr = NULL;
+
+	p = alloc_pages(flag, get_order(size));
+	if (p)
+		addr = page_address(p);
+	if (dma_handle)
+		*dma_handle = (u64) addr;
+	return addr;
+}
+
+static void qib_dma_free_coherent(struct ib_device *dev, size_t size,
+				  void *cpu_addr, u64 dma_handle)
+{
+	free_pages((unsigned long) cpu_addr, get_order(size));
+}
+
+struct ib_dma_mapping_ops qib_dma_mapping_ops = {
+	qib_mapping_error,
+	qib_dma_map_single,
+	qib_dma_unmap_single,
+	qib_dma_map_page,
+	qib_dma_unmap_page,
+	qib_map_sg,
+	qib_unmap_sg,
+	qib_sg_dma_address,
+	qib_sg_dma_len,
+	qib_sync_single_for_cpu,
+	qib_sync_single_for_device,
+	qib_dma_alloc_coherent,
+	qib_dma_free_coherent
+};

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 12/51] IB/qib: Add qib_dma.c

[Roland Dreier]

 > +struct ib_dma_mapping_ops qib_dma_mapping_ops = {
 > +	qib_mapping_error,
 > +	qib_dma_map_single,
 > +	qib_dma_unmap_single,
 > +	qib_dma_map_page,
 > +	qib_dma_unmap_page,
 > +	qib_map_sg,
 > +	qib_unmap_sg,
 > +	qib_sg_dma_address,
 > +	qib_sg_dma_len,
 > +	qib_sync_single_for_cpu,
 > +	qib_sync_single_for_device,
 > +	qib_dma_alloc_coherent,
 > +	qib_dma_free_coherent
 > +};

I think it would be a good idea to use a designated initializer -- eg

	.mapping_error		= qib_mapping_error, ...

instead to avoid silent breakage if we extend the mapping ops structure.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 12/51] IB/qib: Add qib_dma.c

[Roland Dreier]

 > +struct ib_dma_mapping_ops qib_dma_mapping_ops = {
 > +	qib_mapping_error,
 > +	qib_dma_map_single,
 > +	qib_dma_unmap_single,
 > +	qib_dma_map_page,
 > +	qib_dma_unmap_page,
 > +	qib_map_sg,
 > +	qib_unmap_sg,
 > +	qib_sg_dma_address,
 > +	qib_sg_dma_len,
 > +	qib_sync_single_for_cpu,
 > +	qib_sync_single_for_device,
 > +	qib_dma_alloc_coherent,
 > +	qib_dma_free_coherent
 > +};

I think it would be better to use designated initializers (".xxx = yyy,")
here to avoid problems if this struct ever changes layout.
-- 
Roland Dreier <rolandd-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> || For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 12/51] IB/qib: Add qib_dma.c

[Ralph Campbell]

OK. This will be in v3.

On Wed, 2010-03-31 at 16:18 -0700, Roland Dreier wrote:
> > +struct ib_dma_mapping_ops qib_dma_mapping_ops = {
>  > +	qib_mapping_error,
>  > +	qib_dma_map_single,
>  > +	qib_dma_unmap_single,
>  > +	qib_dma_map_page,
>  > +	qib_dma_unmap_page,
>  > +	qib_map_sg,
>  > +	qib_unmap_sg,
>  > +	qib_sg_dma_address,
>  > +	qib_sg_dma_len,
>  > +	qib_sync_single_for_cpu,
>  > +	qib_sync_single_for_device,
>  > +	qib_dma_alloc_coherent,
>  > +	qib_dma_free_coherent
>  > +};
> 
> I think it would be better to use designated initializers (".xxx = yyy,")
> here to avoid problems if this struct ever changes layout.


--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 13/51] IB/qib: Add qib_driver.c

[Ralph Campbell]

creates the qib_driver.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_driver.c |  793 ++++++++++++++++++++++++++++++++
 1 files changed, 793 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_driver.c

diff --git a/drivers/infiniband/hw/qib/qib_driver.c b/drivers/infiniband/hw/qib/qib_driver.c
new file mode 100644
index 0000000..a9a9896
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_driver.c
@@ -0,0 +1,793 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/spinlock.h>
+#include <linux/pci.h>
+#include <linux/io.h>
+#include <linux/delay.h>
+#include <linux/netdevice.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+
+/*
+ * The size has to be longer than this string, so we can append
+ * board/chip information to it in the init code.
+ */
+const char ib_qib_version[] = QIB_IDSTR "\n";
+
+DEFINE_SPINLOCK(qib_devs_lock);
+LIST_HEAD(qib_dev_list);
+DEFINE_MUTEX(qib_mutex);	/* general driver use */
+
+unsigned qib_debug;
+module_param_named(debug, qib_debug, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(debug, "mask for debug prints");
+
+unsigned qib_ibmtu;
+module_param_named(ibmtu, qib_ibmtu, uint, S_IRUGO);
+MODULE_PARM_DESC(ibmtu, "Set max IB MTU (0=2KB, 1=256, 2=512, ... 5=4096");
+
+unsigned qib_compat_ddr_negotiate = 1;
+module_param_named(compat_ddr_negotiate, qib_compat_ddr_negotiate, uint,
+		   S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(compat_ddr_negotiate,
+		 "Attempt pre-IBTA 1.2 DDR speed negotiation");
+
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_AUTHOR("QLogic <support-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>");
+MODULE_DESCRIPTION("QLogic IB driver");
+
+/*
+ * QIB_PIO_MAXIBHDR is the max IB header size allowed for in our
+ * PIO send buffers.  This is well beyond anything currently
+ * defined in the InfiniBand spec.
+ */
+#define QIB_PIO_MAXIBHDR 128
+
+struct qlogic_ib_stats qib_stats;
+
+const char *qib_get_unit_name(int unit)
+{
+	static char iname[16];
+
+	snprintf(iname, sizeof iname, "infinipath%u", unit);
+	return iname;
+}
+
+/*
+ * Return count of units with at least one port ACTIVE.
+ */
+int qib_count_active_units(void)
+{
+	struct qib_devdata *dd;
+	struct qib_pportdata *ppd;
+	unsigned long flags;
+	int pidx, nunits_active = 0;
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+	list_for_each_entry(dd, &qib_dev_list, list) {
+		if (!(dd->flags & QIB_PRESENT) || !dd->kregbase)
+			continue;
+		for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+			ppd = dd->pport + pidx;
+			if (ppd->lid && (ppd->lflags & (QIBL_LINKINIT |
+					 QIBL_LINKARMED | QIBL_LINKACTIVE))) {
+				nunits_active++;
+				break;
+			}
+		}
+	}
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+	return nunits_active;
+}
+
+/*
+ * Return count of all units, optionally return in arguments
+ * the number of usable (present) units, and the number of
+ * ports that are up.
+ */
+int qib_count_units(int *npresentp, int *nupp)
+{
+	int nunits = 0, npresent = 0, nup = 0;
+	struct qib_devdata *dd;
+	unsigned long flags;
+	int pidx;
+	struct qib_pportdata *ppd;
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+
+	list_for_each_entry(dd, &qib_dev_list, list) {
+		nunits++;
+		if ((dd->flags & QIB_PRESENT) && dd->kregbase)
+			npresent++;
+		for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+			ppd = dd->pport + pidx;
+			if (ppd->lid && (ppd->lflags & (QIBL_LINKINIT |
+					 QIBL_LINKARMED | QIBL_LINKACTIVE)))
+				nup++;
+		}
+	}
+
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+
+	if (npresentp)
+		*npresentp = npresent;
+	if (nupp)
+		*nupp = nup;
+
+	return nunits;
+}
+
+/**
+ * qib_wait_linkstate - wait for an IB link state change to occur
+ * @dd: the qlogic_ib device
+ * @state: the state to wait for
+ * @msecs: the number of milliseconds to wait
+ *
+ * wait up to msecs milliseconds for IB link state change to occur for
+ * now, take the easy polling route.  Currently used only by
+ * qib_set_linkstate.  Returns 0 if state reached, otherwise
+ * -ETIMEDOUT state can have multiple states set, for any of several
+ * transitions.
+ */
+int qib_wait_linkstate(struct qib_pportdata *ppd, u32 state, int msecs)
+{
+	int ret;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->lflags_lock, flags);
+	if (ppd->state_wanted) {
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		ret = -EBUSY;
+		goto bail;
+	}
+	ppd->state_wanted = state;
+	spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	wait_event_interruptible_timeout(ppd->state_wait,
+					 (ppd->lflags & state),
+					 msecs_to_jiffies(msecs));
+	spin_lock_irqsave(&ppd->lflags_lock, flags);
+	ppd->state_wanted = 0;
+	spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+
+	if (!(ppd->lflags & state))
+		ret = -ETIMEDOUT;
+	else
+		ret = 0;
+bail:
+	return ret;
+}
+
+int qib_set_linkstate(struct qib_pportdata *ppd, u8 newstate)
+{
+	u32 lstate;
+	int ret;
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+
+	switch (newstate) {
+	case QIB_IB_LINKDOWN_ONLY:
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_DOWN | IB_LINKINITCMD_NOP);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case QIB_IB_LINKDOWN:
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_DOWN | IB_LINKINITCMD_POLL);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case QIB_IB_LINKDOWN_SLEEP:
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_DOWN | IB_LINKINITCMD_SLEEP);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case QIB_IB_LINKDOWN_DISABLE:
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_DOWN | IB_LINKINITCMD_DISABLE);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case QIB_IB_LINKARM:
+		if (ppd->lflags & QIBL_LINKARMED) {
+			qib_dbg("Asked for ARM, already there, skip\n");
+			ret = 0;
+			goto bail;
+		}
+		if (!(ppd->lflags & (QIBL_LINKINIT | QIBL_LINKACTIVE))) {
+			qib_dbg("Asked for ARM, lflags %x, error\n",
+				ppd->lflags);
+			ret = -EINVAL;
+			goto bail;
+		}
+		/*
+		 * Since the port can be ACTIVE when we ask for ARMED,
+		 * clear QIBL_LINKV so we can wait for a transition.
+		 * If the link isn't ARMED, then something else happened
+		 * and there is no point waiting for ARMED.
+		 */
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~QIBL_LINKV;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_ARMED | IB_LINKINITCMD_NOP);
+		lstate = QIBL_LINKV;
+		break;
+
+	case QIB_IB_LINKACTIVE:
+		if (ppd->lflags & QIBL_LINKACTIVE) {
+			qib_dbg("Asked for ACTIVE, already there, skip\n");
+			ret = 0;
+			goto bail;
+		}
+		if (!(ppd->lflags & QIBL_LINKARMED)) {
+			qib_dbg("Asked for ACTIVE, lflags %x, error\n",
+				ppd->lflags);
+			ret = -EINVAL;
+			goto bail;
+		}
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_ACTIVE | IB_LINKINITCMD_NOP);
+		lstate = QIBL_LINKACTIVE;
+		break;
+
+	default:
+		qib_dbg("Invalid linkstate 0x%x requested\n", newstate);
+		ret = -EINVAL;
+		goto bail;
+	}
+	ret = qib_wait_linkstate(ppd, lstate, 10);
+
+bail:
+	return ret;
+}
+
+/*
+ * Get address of eager buffer from it's index (allocated in chunks, not
+ * contiguous).
+ */
+static inline void *qib_get_egrbuf(const struct qib_ctxtdata *rcd, u32 etail)
+{
+	const u32 chunk = etail / rcd->rcvegrbufs_perchunk;
+	const u32 idx =  etail % rcd->rcvegrbufs_perchunk;
+
+	return rcd->rcvegrbuf[chunk] + idx * rcd->dd->rcvegrbufsize;
+}
+
+/**
+ * get_rhf_errstring - decode RHF errors
+ * @err: the err number
+ * @msg: the output buffer
+ * @len: the length of the output buffer
+ *
+ * only used one place now, may want more later
+ */
+static void get_rhf_errstring(u32 err, char *msg, size_t len)
+{
+	/* if no errors, and so don't need to check what's first */
+	*msg = '\0';
+
+	if (err & QLOGIC_IB_RHF_H_ICRCERR)
+		strlcat(msg, "icrcerr ", len);
+	if (err & QLOGIC_IB_RHF_H_VCRCERR)
+		strlcat(msg, "vcrcerr ", len);
+	if (err & QLOGIC_IB_RHF_H_PARITYERR)
+		strlcat(msg, "parityerr ", len);
+	if (err & QLOGIC_IB_RHF_H_LENERR)
+		strlcat(msg, "lenerr ", len);
+	if (err & QLOGIC_IB_RHF_H_MTUERR)
+		strlcat(msg, "mtuerr ", len);
+	if (err & QLOGIC_IB_RHF_H_IHDRERR)
+		/* qlogic_ib hdr checksum error */
+		strlcat(msg, "qibhdrerr ", len);
+	if (err & QLOGIC_IB_RHF_H_TIDERR)
+		strlcat(msg, "tiderr ", len);
+	if (err & QLOGIC_IB_RHF_H_MKERR)
+		/* bad ctxt, offset, etc. */
+		strlcat(msg, "invalid qibhdr ", len);
+	if (err & QLOGIC_IB_RHF_H_IBERR)
+		strlcat(msg, "iberr ", len);
+	if (err & QLOGIC_IB_RHF_L_SWA)
+		strlcat(msg, "swA ", len);
+	if (err & QLOGIC_IB_RHF_L_SWB)
+		strlcat(msg, "swB ", len);
+}
+
+/*
+ * Returns 1 if error was a CRC, else 0.
+ * Needed for some chip's synthesized error counters.
+ */
+static u32 qib_rcv_hdrerr(struct qib_pportdata *ppd, u32 ctxt,
+			  u32 eflags, u32 l, u32 etail, __le32 *rhf_addr,
+			  struct qib_message_header *hdr)
+{
+	char emsg[128];
+	u32 ret = 0;
+
+	get_rhf_errstring(eflags, emsg, sizeof emsg);
+	qib_cdbg(ERRPKT, "IB%u:%u ctxt %u RHF %x qtail=%x typ=%u "
+		 "tlen=%x opcode=%x egridx=%x: %s\n", ppd->dd->unit,
+		 ppd->port, ctxt, eflags, l, qib_hdrget_rcv_type(rhf_addr),
+		 qib_hdrget_length_in_bytes(rhf_addr),
+		 be32_to_cpu(hdr->bth[0]) >> 24, etail, emsg);
+
+	if (eflags & (QLOGIC_IB_RHF_H_ICRCERR | QLOGIC_IB_RHF_H_VCRCERR))
+		ret = 1;
+	return ret;
+}
+
+/*
+ * qib_kreceive - receive a packet
+ * @rcd: the qlogic_ib context
+ * @llic: gets count of good packets needed to clear lli,
+ *          (used with chips that need need to track crcs for lli)
+ *
+ * called from interrupt handler for errors or receive interrupt
+ * Returns number of CRC error packets, needed by some chips for
+ * local link integrity tracking.   crcs are adjusted down by following
+ * good packets, if any, and count of good packets is also tracked.
+ */
+u32 qib_kreceive(struct qib_ctxtdata *rcd, u32 *llic, u32 *npkts)
+{
+	struct qib_devdata *dd = rcd->dd;
+	struct qib_pportdata *ppd = rcd->ppd;
+	__le32 *rhf_addr;
+	void *ebuf;
+	const u32 rsize = dd->rcvhdrentsize;        /* words */
+	const u32 maxcnt = dd->rcvhdrcnt * rsize;   /* words */
+	u32 etail = -1, l, hdrqtail;
+	struct qib_message_header *hdr;
+	u32 eflags, etype, tlen, i = 0, updegr = 0, crcs = 0;
+	int last;
+	u64 lval;
+	struct qib_qp *qp, *nqp;
+
+	l = rcd->head;
+	rhf_addr = (__le32 *) rcd->rcvhdrq + l + dd->rhf_offset;
+	if (dd->flags & QIB_NODMA_RTAIL) {
+		u32 seq = qib_hdrget_seq(rhf_addr);
+		if (seq != rcd->seq_cnt) {
+			qib_cdbg(PKT, "ctxt%u: hdrq seq diff @ hdrqhd %x\n",
+				 rcd->ctxt, rcd->head);
+			goto bail;
+		}
+		hdrqtail = 0;
+	} else {
+		hdrqtail = qib_get_rcvhdrtail(rcd);
+		if (l == hdrqtail) {
+			qib_cdbg(PKT, "ctxt%u: no pkts tail==hdrqhd %x\n",
+				 rcd->ctxt, rcd->head);
+			goto bail;
+		}
+		smp_rmb();  /* prevent speculative reads of dma'ed hdrq */
+	}
+
+	for (last = 0, i = 1; !last; i += !last) {
+		hdr = dd->f_get_msgheader(dd, rhf_addr);
+		eflags = qib_hdrget_err_flags(rhf_addr);
+		etype = qib_hdrget_rcv_type(rhf_addr);
+		/* total length */
+		tlen = qib_hdrget_length_in_bytes(rhf_addr);
+		ebuf = NULL;
+		if ((dd->flags & QIB_NODMA_RTAIL) ?
+		    qib_hdrget_use_egr_buf(rhf_addr) :
+		    (etype != RCVHQ_RCV_TYPE_EXPECTED)) {
+			etail = qib_hdrget_index(rhf_addr);
+			updegr = 1;
+			if (tlen > sizeof(*hdr) ||
+			    etype >= RCVHQ_RCV_TYPE_NON_KD)
+				ebuf = qib_get_egrbuf(rcd, etail);
+		}
+		if (!eflags) {
+			u16 lrh_len = be16_to_cpu(hdr->lrh[2]) << 2;
+
+			if (lrh_len != tlen) {
+				qib_dbg("IB%u:%u ctxt %u lrh_len %u "
+					"!= tlen %u\n",
+					dd->unit, ppd->port, rcd->ctxt,
+					lrh_len, tlen);
+				qib_stats.sps_lenerrs++;
+				goto move_along;
+			}
+		}
+		if (etype == RCVHQ_RCV_TYPE_NON_KD && !eflags &&
+		    ebuf == NULL &&
+		    tlen > (dd->rcvhdrentsize - 2 + 1 -
+				qib_hdrget_offset(rhf_addr)) << 2) {
+			qib_dbg("IB%d ctxt %u NULL data rhf %08x%08x tlen %u\n",
+				ppd->port, rcd->ctxt,
+				le32_to_cpu(rhf_addr[1]),
+				le32_to_cpu(rhf_addr[0]), tlen);
+			goto move_along;
+		}
+
+		/*
+		 * Both tiderr and qibhdrerr are set for all plain IB
+		 * packets; only qibhdrerr should be set.
+		 */
+
+		if (etype != RCVHQ_RCV_TYPE_NON_KD &&
+		    etype != RCVHQ_RCV_TYPE_ERROR &&
+		    qib_hdrget_qib_ver(hdr->iph.ver_ctxt_tid_offset) !=
+		    IPS_PROTO_VERSION)
+			qib_cdbg(ERRPKT, "Bad InfiniPath protocol version "
+				 "%x\n", etype);
+
+		if (unlikely(eflags))
+			crcs += qib_rcv_hdrerr(ppd, rcd->ctxt, eflags, l,
+					       etail, rhf_addr, hdr);
+		else if (etype == RCVHQ_RCV_TYPE_NON_KD) {
+			qib_ib_rcv(rcd, hdr, ebuf, tlen);
+			if (crcs)
+				crcs--;
+			else if (llic && *llic)
+				--*llic;
+		} else if (etype == RCVHQ_RCV_TYPE_EXPECTED)
+			qib_cdbg(ERRPKT, "type=Expected pkt, no err bits\n");
+		else if (etype == RCVHQ_RCV_TYPE_EAGER) {
+			u8 opcode = be32_to_cpu(hdr->bth[0]) >> 24;
+			u32 qpn = be32_to_cpu(hdr->bth[1]) & 0xffffff;
+
+			qib_cdbg(RVPKT, "typ %x, opcode %x (eager, "
+				 "qp=%x), len %x; ignored\n", etype,
+				 opcode, qpn, tlen);
+		} else {
+			/*
+			 * Error packet, type of error unknown.
+			 * Probably type 3, but we don't know, so don't
+			 * even try to print the opcode, etc.
+			 * Usually caused by a "bad packet", that has no
+			 * BTH, when the LRH says it should, or it's
+			 * a KD packet with an invalid KDETH.
+			 */
+			qib_cdbg(ERRPKT, "Error Pkt, but no eflags! egrbuf"
+				 " %x, len %x hdrq+%x rhf: %Lx\n", etail,
+				 tlen, l, le64_to_cpu(*(__le64 *) rhf_addr));
+			if (qib_debug & __QIB_ERRPKTDBG) {
+				u32 j, *d, dw = rsize - 2;
+
+				if (rsize > (tlen >> 2))
+					dw = tlen >> 2;
+				d = (u32 *)hdr;
+				printk(KERN_DEBUG "EPkt rcvhdr(%x dw):\n",
+				       dw);
+				for (j = 0; j < dw; j++)
+					printk(KERN_DEBUG "%8x%s", d[j],
+						(j%8) == 7 ? "\n" : " ");
+				printk(KERN_DEBUG ".\n");
+			}
+		}
+move_along:
+		l += rsize;
+		if (l >= maxcnt)
+			l = 0;
+		rhf_addr = (__le32 *) rcd->rcvhdrq + l + dd->rhf_offset;
+		if (dd->flags & QIB_NODMA_RTAIL) {
+			u32 seq = qib_hdrget_seq(rhf_addr);
+
+			if (++rcd->seq_cnt > 13)
+				rcd->seq_cnt = 1;
+			if (seq != rcd->seq_cnt)
+				last = 1;
+		} else if (l == hdrqtail)
+			last = 1;
+		/*
+		 * Update head regs etc., every 16 packets, if not last pkt,
+		 * to help prevent rcvhdrq overflows, when many packets
+		 * are processed and queue is nearly full.
+		 * Don't request an interrupt for intermediate updates.
+		 */
+		lval = l;
+		if (!last && !(i & 0xf)) {
+			dd->f_update_usrhead(rcd, lval, updegr, etail);
+			updegr = 0;
+		}
+	}
+
+	rcd->head = l;
+	rcd->pkt_count += i;
+
+	/*
+	 * Iterate over all QPs waiting to respond.
+	 * The list won't change since the IRQ is only run on one CPU.
+	 */
+	list_for_each_entry_safe(qp, nqp, &rcd->qp_wait_list, rspwait) {
+		list_del_init(&qp->rspwait);
+		if (qp->r_flags & QIB_R_RSP_NAK) {
+			qp->r_flags &= ~QIB_R_RSP_NAK;
+			qib_send_rc_ack(qp);
+		}
+		if (qp->r_flags & QIB_R_RSP_SEND) {
+			unsigned long flags;
+
+			qp->r_flags &= ~QIB_R_RSP_SEND;
+			spin_lock_irqsave(&qp->s_lock, flags);
+			if (ib_qib_state_ops[qp->state] &
+					QIB_PROCESS_OR_FLUSH_SEND)
+				qib_schedule_send(qp);
+			spin_unlock_irqrestore(&qp->s_lock, flags);
+		}
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	}
+
+	qib_cdbg(PKT, "IB%d ctxt %d handled %u packets\n",
+		 ppd->port, rcd->ctxt, i);
+bail:
+	/* Report number of packets consumed */
+	if (npkts)
+		*npkts = i;
+
+	/*
+	 * Always write head at end, and setup rcv interrupt, even
+	 * if no packets were processed.
+	 */
+	lval = (u64)rcd->head | dd->rhdrhead_intr_off;
+	dd->f_update_usrhead(rcd, lval, updegr, etail);
+	return crcs;
+}
+
+/**
+ * qib_set_mtu - set the MTU
+ * @ppd: the perport data
+ * @arg: the new MTU
+ *
+ * We can handle "any" incoming size, the issue here is whether we
+ * need to restrict our outgoing size.   For now, we don't do any
+ * sanity checking on this, and we don't deal with what happens to
+ * programs that are already running when the size changes.
+ * NOTE: changing the MTU will usually cause the IBC to go back to
+ * link INIT state...
+ */
+int qib_set_mtu(struct qib_pportdata *ppd, u16 arg)
+{
+	u32 piosize;
+	int ret, chk;
+
+	if (arg != 256 && arg != 512 && arg != 1024 && arg != 2048 &&
+	    arg != 4096) {
+		qib_dbg("Trying to set invalid mtu %u, failing\n", arg);
+		ret = -EINVAL;
+		goto bail;
+	}
+	chk = ib_mtu_enum_to_int(qib_ibmtu);
+	if (chk > 0 && arg > chk) {
+		qib_dbg("Trying to set mtu %u > ibmtu cap %u, failing\n",
+			arg, chk);
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	piosize = ppd->ibmaxlen;
+	ppd->ibmtu = arg;
+
+	if (arg >= (piosize - QIB_PIO_MAXIBHDR)) {
+		/* Only if it's not the initial value (or reset to it) */
+		if (piosize != ppd->init_ibmaxlen) {
+			if (arg > piosize && arg <= ppd->init_ibmaxlen)
+				piosize = ppd->init_ibmaxlen - 2 * sizeof(u32);
+			ppd->ibmaxlen = piosize;
+		}
+	} else if ((arg + QIB_PIO_MAXIBHDR) != ppd->ibmaxlen) {
+		piosize = arg + QIB_PIO_MAXIBHDR - 2 * sizeof(u32);
+		ppd->ibmaxlen = piosize;
+		qib_cdbg(VERBOSE, "ibmaxlen was 0x%x, setting to 0x%x "
+			   "(mtu 0x%x)\n", ppd->ibmaxlen, piosize,
+			   arg);
+	}
+
+	ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_MTU, 0);
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+int qib_set_lid(struct qib_pportdata *ppd, u32 lid, u8 lmc)
+{
+	struct qib_devdata *dd = ppd->dd;
+	ppd->lid = lid;
+	ppd->lmc = lmc;
+
+	dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LIDLMC,
+			 lid | (~((1U << lmc) - 1)) << 16);
+
+	qib_devinfo(dd->pcidev, "IB%u:%u got a lid: 0x%x\n",
+		    dd->unit, ppd->port, lid);
+
+	return 0;
+}
+
+/*
+ * Following deal with the "obviously simple" task of overriding the state
+ * of the LEDS, which normally indicate link physical and logical status.
+ * The complications arise in dealing with different hardware mappings
+ * and the board-dependent routine being called from interrupts.
+ * and then there's the requirement to _flash_ them.
+ */
+#define LED_OVER_FREQ_SHIFT 8
+#define LED_OVER_FREQ_MASK (0xFF<<LED_OVER_FREQ_SHIFT)
+/* Below is "non-zero" to force override, but both actual LEDs are off */
+#define LED_OVER_BOTH_OFF (8)
+
+static void qib_run_led_override(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *)opaque;
+	struct qib_devdata *dd = ppd->dd;
+	int timeoff;
+	int ph_idx;
+
+	if (!(dd->flags & QIB_INITTED))
+		return;
+
+	ph_idx = ppd->led_override_phase++ & 1;
+	ppd->led_override = ppd->led_override_vals[ph_idx];
+	timeoff = ppd->led_override_timeoff;
+
+	dd->f_setextled(ppd, 1);
+	/*
+	 * don't re-fire the timer if user asked for it to be off; we let
+	 * it fire one more time after they turn it off to simplify
+	 */
+	if (ppd->led_override_vals[0] || ppd->led_override_vals[1])
+		mod_timer(&ppd->led_override_timer, jiffies + timeoff);
+}
+
+void qib_set_led_override(struct qib_pportdata *ppd, unsigned int val)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int timeoff, freq;
+
+	if (!(dd->flags & QIB_INITTED))
+		return;
+
+	/* First check if we are blinking. If not, use 1HZ polling */
+	timeoff = HZ;
+	freq = (val & LED_OVER_FREQ_MASK) >> LED_OVER_FREQ_SHIFT;
+
+	if (freq) {
+		/* For blink, set each phase from one nybble of val */
+		ppd->led_override_vals[0] = val & 0xF;
+		ppd->led_override_vals[1] = (val >> 4) & 0xF;
+		timeoff = (HZ << 4)/freq;
+	} else {
+		/* Non-blink set both phases the same. */
+		ppd->led_override_vals[0] = val & 0xF;
+		ppd->led_override_vals[1] = val & 0xF;
+	}
+	ppd->led_override_timeoff = timeoff;
+
+	/*
+	 * If the timer has not already been started, do so. Use a "quick"
+	 * timeout so the function will be called soon, to look at our request.
+	 */
+	if (atomic_inc_return(&ppd->led_override_timer_active) == 1) {
+		/* Need to start timer */
+		init_timer(&ppd->led_override_timer);
+		ppd->led_override_timer.function = qib_run_led_override;
+		ppd->led_override_timer.data = (unsigned long) ppd;
+		ppd->led_override_timer.expires = jiffies + 1;
+		add_timer(&ppd->led_override_timer);
+	} else {
+		if (ppd->led_override_vals[0] || ppd->led_override_vals[1])
+			mod_timer(&ppd->led_override_timer, jiffies + 1);
+		atomic_dec(&ppd->led_override_timer_active);
+	}
+}
+
+/**
+ * qib_reset_device - reset the chip if possible
+ * @unit: the device to reset
+ *
+ * Whether or not reset is successful, we attempt to re-initialize the chip
+ * (that is, much like a driver unload/reload).  We clear the INITTED flag
+ * so that the various entry points will fail until we reinitialize.  For
+ * now, we only allow this if no user contexts are open that use chip resources
+ */
+int qib_reset_device(int unit)
+{
+	int ret, i;
+	struct qib_devdata *dd = qib_lookup(unit);
+	struct qib_pportdata *ppd;
+	unsigned long flags;
+	int pidx;
+
+	if (!dd) {
+		ret = -ENODEV;
+		goto bail;
+	}
+
+	qib_devinfo(dd->pcidev, "Reset on unit %u requested\n", unit);
+
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT)) {
+		qib_devinfo(dd->pcidev, "Invalid unit number %u or "
+			    "not initialized or not present\n", unit);
+		ret = -ENXIO;
+		goto bail;
+	}
+
+	spin_lock_irqsave(&dd->uctxt_lock, flags);
+	if (dd->rcd)
+		for (i = dd->first_user_ctxt; i < dd->cfgctxts; i++) {
+			if (!dd->rcd[i] || !dd->rcd[i]->cnt)
+				continue;
+			spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+			qib_dbg("unit %u ctxt %d is in use (PID %u cmd %s),"
+				" can't reset\n", unit, i,
+				dd->rcd[i]->pid,
+				dd->rcd[i]->comm);
+			ret = -EBUSY;
+			goto bail;
+		}
+	spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+		if (atomic_read(&ppd->led_override_timer_active)) {
+			/* Need to stop LED timer, _then_ shut off LEDs */
+			del_timer_sync(&ppd->led_override_timer);
+			atomic_set(&ppd->led_override_timer_active, 0);
+		}
+
+		/* Shut off LEDs after we are sure timer is not running */
+		ppd->led_override = LED_OVER_BOTH_OFF;
+		dd->f_setextled(ppd, 0);
+		if (dd->flags & QIB_HAS_SEND_DMA)
+			qib_teardown_sdma(ppd);
+	}
+
+	ret = dd->f_reset(dd);
+	if (ret == 1) {
+		qib_dbg("Reinitializing unit %u after reset attempt\n",
+			unit);
+		ret = qib_init(dd, 1);
+	} else
+		ret = -EAGAIN;
+	if (ret)
+		qib_dev_err(dd, "Reinitialize unit %u after "
+			    "reset failed with %d\n", unit, ret);
+	else
+		qib_devinfo(dd->pcidev, "Reinitialized unit %u after "
+			    "resetting\n", unit);
+
+bail:
+	return ret;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 13/51] IB/qib: Add qib_driver.c

[Roland Dreier]

 > +unsigned qib_debug;
 > +module_param_named(debug, qib_debug, uint, S_IWUSR | S_IRUGO);
 > +MODULE_PARM_DESC(debug, "mask for debug prints");

Did you look at using trace events for this stuff?  That gives you
extremely low overhead when tracing is turned off (dynamic patching to
NOP out the tracing when it's disabled) and also very fine-grained (per
trace site) control over what gets printed; plus you get dumping of the
trace buffer on crash, etc.

 - R.
-- 
Roland Dreier <rolandd-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> || For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 13/51] IB/qib: Add qib_driver.c

[Steve Wise]

Roland Dreier wrote:
>  > +unsigned qib_debug;
>  > +module_param_named(debug, qib_debug, uint, S_IWUSR | S_IRUGO);
>  > +MODULE_PARM_DESC(debug, "mask for debug prints");
>
> Did you look at using trace events for this stuff?  That gives you
> extremely low overhead when tracing is turned off (dynamic patching to
> NOP out the tracing when it's disabled) and also very fine-grained (per
> trace site) control over what gets printed; plus you get dumping of the
> trace buffer on crash, etc.
>
>  - R.
>   

Where can I find information on trace events?  Something in Documentation/*?

Thanks,


Steve.


--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 13/51] IB/qib: Add qib_driver.c

[Roland Dreier]

 > Where can I find information on trace events?  Something in Documentation/*?

Yep, Documentation/trace/events.txt.

Also look at eg include/trace/events/ext4.h and how the trace_ext4_{xxx}
functions are used in fs/ext4/*.

It's much better than anything you're going to hack up for a single
driver to use.

 - R.
-- 
Roland Dreier <rolandd-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> || For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 13/51] IB/qib: Add qib_driver.c

[John A. Gregor]

Roland Dreier <rdreier-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> wrote:
>  > Where can I find information on trace events?  Something in Documentation/*?
>
> Yep, Documentation/trace/events.txt.

LWN just did a really good writeup on using the TRACE_EVENT macro:

	http://lwn.net/Articles/379903/

Part 2 is still behind the paywall.

-John Gregor
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 13/51] IB/qib: Add qib_driver.c

[Roland Dreier]

 > +DEFINE_MUTEX(qib_mutex);	/* general driver use */

Rather than having this ill-defined mutex that I think is going to make
it hard to understand the locking and get the lock ordering right, would
it be better to have well-defined locking rules?  AFAICT this mutex is
used in only two places, qib_diag.c and qib_file_op.c.  Are those two
uses protecting the same thing?  Or could we have two static mutexes,
one in each file, that protects what each file needs protected?
-- 
Roland Dreier <rolandd-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> || For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 14/51] IB/qib: Add qib_eeprom.c

[Ralph Campbell]

creates the qib_eeprom.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_eeprom.c |  460 ++++++++++++++++++++++++++++++++
 1 files changed, 460 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_eeprom.c

diff --git a/drivers/infiniband/hw/qib/qib_eeprom.c b/drivers/infiniband/hw/qib/qib_eeprom.c
new file mode 100644
index 0000000..7a6fe41
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_eeprom.c
@@ -0,0 +1,460 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/delay.h>
+#include <linux/pci.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+
+/*
+ * Functions specific to the serial EEPROM on cards handled by ib_qib.
+ * The actual serail interface code is in qib_twsi.c. This file is a client
+ */
+
+/**
+ * qib_eeprom_read - receives bytes from the eeprom via I2C
+ * @dd: the qlogic_ib device
+ * @eeprom_offset: address to read from
+ * @buffer: where to store result
+ * @len: number of bytes to receive
+ */
+int qib_eeprom_read(struct qib_devdata *dd, u8 eeprom_offset,
+		    void *buff, int len)
+{
+	int ret;
+
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (!ret) {
+		ret = qib_twsi_reset(dd);
+		if (ret)
+			qib_dev_err(dd, "TWSI Reset failed, EEPROM Rd\n");
+		else
+			ret = qib_twsi_blk_rd(dd, dd->twsi_eeprom_dev,
+					      eeprom_offset, buff, len);
+		mutex_unlock(&dd->eep_lock);
+	}
+
+	return ret;
+}
+
+/*
+ * Actually update the eeprom, first doing write enable if
+ * needed, then restoring write enable state.
+ * Must be called with eep_lock held
+ */
+static int eeprom_write_with_enable(struct qib_devdata *dd, u8 offset,
+		     const void *buf, int len)
+{
+	int ret, pwen;
+
+	pwen = dd->f_eeprom_wen(dd, 1);
+	ret = qib_twsi_reset(dd);
+	if (ret)
+		qib_dev_err(dd, "TWSI Reset failed, EEPROM Wr\n");
+	else
+		ret = qib_twsi_blk_wr(dd, dd->twsi_eeprom_dev,
+				      offset, buf, len);
+	dd->f_eeprom_wen(dd, pwen);
+	return ret;
+}
+
+/**
+ * qib_eeprom_write - writes data to the eeprom via I2C
+ * @dd: the qlogic_ib device
+ * @eeprom_offset: where to place data
+ * @buffer: data to write
+ * @len: number of bytes to write
+ */
+int qib_eeprom_write(struct qib_devdata *dd, u8 eeprom_offset,
+		     const void *buff, int len)
+{
+	int ret;
+
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (!ret) {
+		ret = eeprom_write_with_enable(dd, eeprom_offset, buff, len);
+		mutex_unlock(&dd->eep_lock);
+	}
+
+	return ret;
+}
+
+static u8 flash_csum(struct qib_flash *ifp, int adjust)
+{
+	u8 *ip = (u8 *) ifp;
+	u8 csum = 0, len;
+
+	/*
+	 * Limit length checksummed to max length of actual data.
+	 * Checksum of erased eeprom will still be bad, but we avoid
+	 * reading past the end of the buffer we were passed.
+	 */
+	len = ifp->if_length;
+	if (len > sizeof(struct qib_flash))
+		len = sizeof(struct qib_flash);
+	while (len--)
+		csum += *ip++;
+	csum -= ifp->if_csum;
+	csum = ~csum;
+	if (adjust)
+		ifp->if_csum = csum;
+
+	return csum;
+}
+
+/**
+ * qib_get_eeprom_info- get the GUID et al. from the TSWI EEPROM device
+ * @dd: the qlogic_ib device
+ *
+ * We have the capability to use the nguid field, and get
+ * the guid from the first chip's flash, to use for all of them.
+ */
+void qib_get_eeprom_info(struct qib_devdata *dd)
+{
+	void *buf;
+	struct qib_flash *ifp;
+	__be64 guid;
+	int len, eep_stat;
+	u8 csum, *bguid;
+	int t = dd->unit;
+	struct qib_devdata *dd0 = qib_lookup(0);
+
+	if (t && dd0->nguid > 1 && t <= dd0->nguid) {
+		u8 oguid;
+		dd->base_guid = dd0->base_guid;
+		bguid = (u8 *) &dd->base_guid;
+
+		oguid = bguid[7];
+		bguid[7] += t;
+		if (oguid > bguid[7]) {
+			if (bguid[6] == 0xff) {
+				if (bguid[5] == 0xff) {
+					qib_dev_err(dd, "Can't set %s GUID"
+						    " from base, wraps to"
+						    " OUI!\n",
+						    qib_get_unit_name(t));
+					dd->base_guid = 0;
+					goto bail;
+				}
+				bguid[5]++;
+			}
+			bguid[6]++;
+		}
+		dd->nguid = 1;
+
+		qib_dbg("nguid %u, so adding %u to device 0 guid, "
+			"for %llx\n", dd0->nguid, t,
+			(unsigned long long) be64_to_cpu(dd->base_guid));
+		goto bail;
+	}
+
+	/*
+	 * Read full flash, not just currently used part, since it may have
+	 * been written with a newer definition.
+	 * */
+	len = sizeof(struct qib_flash);
+	buf = vmalloc(len);
+	if (!buf) {
+		qib_dev_err(dd, "Couldn't allocate memory to read %u "
+			    "bytes from eeprom for GUID\n", len);
+		goto bail;
+	}
+
+	/*
+	 * Use "public" eeprom read function, which does locking and
+	 * figures out device. This will migrate to chip-specific.
+	 */
+	eep_stat = qib_eeprom_read(dd, 0, buf, len);
+
+	if (eep_stat) {
+		qib_dev_err(dd, "Failed reading GUID from eeprom\n");
+		goto done;
+	}
+	ifp = (struct qib_flash *)buf;
+
+	csum = flash_csum(ifp, 0);
+	if (csum != ifp->if_csum) {
+		qib_devinfo(dd->pcidev, "Bad I2C flash checksum: "
+			 "0x%x, not 0x%x\n", csum, ifp->if_csum);
+		goto done;
+	}
+	if (*(__be64 *) ifp->if_guid == cpu_to_be64(0) ||
+	    *(__be64 *) ifp->if_guid == ~cpu_to_be64(0)) {
+		qib_dev_err(dd, "Invalid GUID %llx from flash; ignoring\n",
+			    *(unsigned long long *) ifp->if_guid);
+		/* don't allow GUID if all 0 or all 1's */
+		goto done;
+	}
+
+	/* complain, but allow it */
+	if (*(u64 *) ifp->if_guid == 0x100007511000000ULL)
+		qib_devinfo(dd->pcidev, "Warning, GUID %llx is "
+			 "default, probably not correct!\n",
+			 *(unsigned long long *) ifp->if_guid);
+
+	bguid = ifp->if_guid;
+	if (!bguid[0] && !bguid[1] && !bguid[2]) {
+		/*
+		 * Original incorrect GUID format in flash; fix in
+		 * core copy, by shifting up 2 octets; don't need to
+		 * change top octet, since both it and shifted are 0.
+		 */
+		bguid[1] = bguid[3];
+		bguid[2] = bguid[4];
+		bguid[3] = 0;
+		bguid[4] = 0;
+		guid = *(__be64 *) ifp->if_guid;
+		qib_cdbg(VERBOSE, "Old GUID format in flash, top 3 zero, "
+			 "shifting 2 octets\n");
+	} else
+		guid = *(__be64 *) ifp->if_guid;
+	dd->base_guid = guid;
+	dd->nguid = ifp->if_numguid;
+	/*
+	 * Things are slightly complicated by the desire to transparently
+	 * support both the Pathscale 10-digit serial number and the QLogic
+	 * 13-character version.
+	 */
+	if ((ifp->if_fversion > 1) && ifp->if_sprefix[0] &&
+	    ((u8 *) ifp->if_sprefix)[0] != 0xFF) {
+		char *snp = dd->serial;
+
+		/*
+		 * This board has a Serial-prefix, which is stored
+		 * elsewhere for backward-compatibility.
+		 */
+		memcpy(snp, ifp->if_sprefix, sizeof ifp->if_sprefix);
+		snp[sizeof ifp->if_sprefix] = '\0';
+		len = strlen(snp);
+		snp += len;
+		len = (sizeof dd->serial) - len;
+		if (len > sizeof ifp->if_serial)
+			len = sizeof ifp->if_serial;
+		memcpy(snp, ifp->if_serial, len);
+	} else
+		memcpy(dd->serial, ifp->if_serial,
+		       sizeof ifp->if_serial);
+	if (!strstr(ifp->if_comment, "Tested successfully"))
+		qib_dev_err(dd, "Board SN %s did not pass functional "
+			    "test: %s\n", dd->serial, ifp->if_comment);
+
+	qib_cdbg(VERBOSE, "Initted GUID to %llx from eeprom\n",
+		 (unsigned long long) be64_to_cpu(dd->base_guid));
+
+	memcpy(&dd->eep_st_errs, &ifp->if_errcntp, QIB_EEP_LOG_CNT);
+	/*
+	 * Power-on (actually "active") hours are kept as little-endian value
+	 * in EEPROM, but as seconds in a (possibly as small as 24-bit)
+	 * atomic_t while running.
+	 */
+	atomic_set(&dd->active_time, 0);
+	dd->eep_hrs = ifp->if_powerhour[0] | (ifp->if_powerhour[1] << 8);
+
+done:
+	vfree(buf);
+
+bail:;
+}
+
+/**
+ * qib_update_eeprom_log - copy active-time and error counters to eeprom
+ * @dd: the qlogic_ib device
+ *
+ * Although the time is kept as seconds in the qib_devdata struct, it is
+ * rounded to hours for re-write, as we have only 16 bits in EEPROM.
+ * First-cut code reads whole (expected) struct qib_flash, modifies,
+ * re-writes. Future direction: read/write only what we need, assuming
+ * that the EEPROM had to have been "good enough" for driver init, and
+ * if not, we aren't making it worse.
+ *
+ */
+int qib_update_eeprom_log(struct qib_devdata *dd)
+{
+	void *buf;
+	struct qib_flash *ifp;
+	int len, hi_water;
+	uint32_t new_time, new_hrs;
+	u8 csum;
+	int ret, idx;
+	unsigned long flags;
+
+	/* first, check if we actually need to do anything. */
+	ret = 0;
+	for (idx = 0; idx < QIB_EEP_LOG_CNT; ++idx) {
+		if (dd->eep_st_new_errs[idx]) {
+			ret = 1;
+			break;
+		}
+	}
+	new_time = atomic_read(&dd->active_time);
+
+	if (ret == 0 && new_time < 3600)
+		goto bail;
+
+	/*
+	 * The quick-check above determined that there is something worthy
+	 * of logging, so get current contents and do a more detailed idea.
+	 * read full flash, not just currently used part, since it may have
+	 * been written with a newer definition
+	 */
+	len = sizeof(struct qib_flash);
+	buf = vmalloc(len);
+	ret = 1;
+	if (!buf) {
+		qib_dev_err(dd, "Couldn't allocate memory to read %u "
+			    "bytes from eeprom for logging\n", len);
+		goto bail;
+	}
+
+	/* Grab semaphore and read current EEPROM. If we get an
+	 * error, let go, but if not, keep it until we finish write.
+	 */
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (ret) {
+		qib_dev_err(dd, "Unable to acquire EEPROM for logging\n");
+		goto free_bail;
+	}
+	ret = qib_twsi_blk_rd(dd, dd->twsi_eeprom_dev, 0, buf, len);
+	if (ret) {
+		mutex_unlock(&dd->eep_lock);
+		qib_dev_err(dd, "Unable read EEPROM for logging\n");
+		goto free_bail;
+	}
+	ifp = (struct qib_flash *)buf;
+
+	csum = flash_csum(ifp, 0);
+	if (csum != ifp->if_csum) {
+		mutex_unlock(&dd->eep_lock);
+		qib_dev_err(dd, "EEPROM cks err (0x%02X, S/B 0x%02X)\n",
+			    csum, ifp->if_csum);
+		ret = 1;
+		goto free_bail;
+	}
+	hi_water = 0;
+	spin_lock_irqsave(&dd->eep_st_lock, flags);
+	for (idx = 0; idx < QIB_EEP_LOG_CNT; ++idx) {
+		int new_val = dd->eep_st_new_errs[idx];
+		if (new_val) {
+			/*
+			 * If we have seen any errors, add to EEPROM values
+			 * We need to saturate at 0xFF (255) and we also
+			 * would need to adjust the checksum if we were
+			 * trying to minimize EEPROM traffic
+			 * Note that we add to actual current count in EEPROM,
+			 * in case it was altered while we were running.
+			 */
+			new_val += ifp->if_errcntp[idx];
+			if (new_val > 0xFF)
+				new_val = 0xFF;
+			if (ifp->if_errcntp[idx] != new_val) {
+				ifp->if_errcntp[idx] = new_val;
+				hi_water = offsetof(struct qib_flash,
+						    if_errcntp) + idx;
+			}
+			/*
+			 * update our shadow (used to minimize EEPROM
+			 * traffic), to match what we are about to write.
+			 */
+			dd->eep_st_errs[idx] = new_val;
+			dd->eep_st_new_errs[idx] = 0;
+		}
+	}
+	/*
+	 * Now update active-time. We would like to round to the nearest hour
+	 * but unless atomic_t are sure to be proper signed ints we cannot,
+	 * because we need to account for what we "transfer" to EEPROM and
+	 * if we log an hour at 31 minutes, then we would need to set
+	 * active_time to -29 to accurately count the _next_ hour.
+	 */
+	if (new_time >= 3600) {
+		new_hrs = new_time / 3600;
+		atomic_sub((new_hrs * 3600), &dd->active_time);
+		new_hrs += dd->eep_hrs;
+		if (new_hrs > 0xFFFF)
+			new_hrs = 0xFFFF;
+		dd->eep_hrs = new_hrs;
+		if ((new_hrs & 0xFF) != ifp->if_powerhour[0]) {
+			ifp->if_powerhour[0] = new_hrs & 0xFF;
+			hi_water = offsetof(struct qib_flash, if_powerhour);
+		}
+		if ((new_hrs >> 8) != ifp->if_powerhour[1]) {
+			ifp->if_powerhour[1] = new_hrs >> 8;
+			hi_water = offsetof(struct qib_flash, if_powerhour) + 1;
+		}
+	}
+	/*
+	 * There is a tiny possibility that we could somehow fail to write
+	 * the EEPROM after updating our shadows, but problems from holding
+	 * the spinlock too long are a much bigger issue.
+	 */
+	spin_unlock_irqrestore(&dd->eep_st_lock, flags);
+	if (hi_water) {
+		/* we made some change to the data, uopdate cksum and write */
+		csum = flash_csum(ifp, 1);
+		ret = eeprom_write_with_enable(dd, 0, buf, hi_water + 1);
+	}
+	mutex_unlock(&dd->eep_lock);
+	if (ret)
+		qib_dev_err(dd, "Failed updating EEPROM\n");
+
+free_bail:
+	vfree(buf);
+bail:
+	return ret;
+}
+
+/**
+ * qib_inc_eeprom_err - increment one of the four error counters
+ * that are logged to EEPROM.
+ * @dd: the qlogic_ib device
+ * @eidx: 0..3, the counter to increment
+ * @incr: how much to add
+ *
+ * Each counter is 8-bits, and saturates at 255 (0xFF). They
+ * are copied to the EEPROM (aka flash) whenever qib_update_eeprom_log()
+ * is called, but it can only be called in a context that allows sleep.
+ * This function can be called even at interrupt level.
+ */
+void qib_inc_eeprom_err(struct qib_devdata *dd, u32 eidx, u32 incr)
+{
+	uint new_val;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->eep_st_lock, flags);
+	new_val = dd->eep_st_new_errs[eidx] + incr;
+	if (new_val > 255)
+		new_val = 255;
+	dd->eep_st_new_errs[eidx] = new_val;
+	spin_unlock_irqrestore(&dd->eep_st_lock, flags);
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 15/51] IB/qib: Add qib_file_ops.c

[Ralph Campbell]

creates the qib_file_ops.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_file_ops.c | 2516 ++++++++++++++++++++++++++++++
 1 files changed, 2516 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_file_ops.c

diff --git a/drivers/infiniband/hw/qib/qib_file_ops.c b/drivers/infiniband/hw/qib/qib_file_ops.c
new file mode 100644
index 0000000..781c87f
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_file_ops.c
@@ -0,0 +1,2516 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/pci.h>
+#include <linux/poll.h>
+#include <linux/cdev.h>
+#include <linux/swap.h>
+#include <linux/vmalloc.h>
+#include <linux/highmem.h>
+#include <linux/io.h>
+#include <linux/uio.h>
+#include <linux/jiffies.h>
+#include <asm/pgtable.h>
+#include <linux/delay.h>
+
+#include "qib.h"
+#include "qib_common.h"
+#include "qib_user_sdma.h"
+
+static int qib_open(struct inode *, struct file *);
+static int qib_close(struct inode *, struct file *);
+static ssize_t qib_write(struct file *, const char __user *, size_t, loff_t *);
+static ssize_t qib_aio_write(struct kiocb *, const struct iovec *,
+			     unsigned long, loff_t);
+static unsigned int qib_poll(struct file *, struct poll_table_struct *);
+static int qib_mmapf(struct file *, struct vm_area_struct *);
+
+static const struct file_operations qib_file_ops = {
+	.owner = THIS_MODULE,
+	.write = qib_write,
+	.aio_write = qib_aio_write,
+	.open = qib_open,
+	.release = qib_close,
+	.poll = qib_poll,
+	.mmap = qib_mmapf
+};
+
+/*
+ * Convert kernel virtual addresses to physical addresses so they don't
+ * potentially conflict with the chip addresses used as mmap offsets.
+ * It doesn't really matter what mmap offset we use as long as we can
+ * interpret it correctly.
+ */
+static u64 cvt_kvaddr(void *p)
+{
+	struct page *page;
+	u64 paddr = 0;
+
+	page = vmalloc_to_page(p);
+	if (page)
+		paddr = page_to_pfn(page) << PAGE_SHIFT;
+
+	return paddr;
+}
+
+static int qib_get_base_info(struct file *fp, void __user *ubase,
+			     size_t ubase_size)
+{
+	struct qib_ctxtdata *rcd = ctxt_fp(fp);
+	int ret = 0;
+	struct qib_base_info *kinfo = NULL;
+	struct qib_devdata *dd = rcd->dd;
+	struct qib_pportdata *ppd = rcd->ppd;
+	unsigned subctxt_cnt;
+	int shared, master;
+	size_t sz;
+
+	subctxt_cnt = rcd->subctxt_cnt;
+	if (!subctxt_cnt) {
+		shared = 0;
+		master = 0;
+		subctxt_cnt = 1;
+	} else {
+		shared = 1;
+		master = !subctxt_fp(fp);
+	}
+
+	sz = sizeof(*kinfo);
+	/* If context sharing is not requested, allow the old size structure */
+	if (!shared)
+		sz -= 7 * sizeof(u64);
+	if (ubase_size < sz) {
+		qib_cdbg(PROC, "Base size %zu, need %zu (version mismatch?)\n",
+			 ubase_size, sz);
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	kinfo = kzalloc(sizeof(*kinfo), GFP_KERNEL);
+	if (kinfo == NULL) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	ret = dd->f_get_base_info(rcd, kinfo);
+	if (ret < 0)
+		goto bail;
+
+	kinfo->spi_rcvhdr_cnt = dd->rcvhdrcnt;
+	kinfo->spi_rcvhdrent_size = dd->rcvhdrentsize;
+	kinfo->spi_tidegrcnt = rcd->rcvegrcnt;
+	kinfo->spi_rcv_egrbufsize = dd->rcvegrbufsize;
+	/*
+	 * have to mmap whole thing
+	 */
+	kinfo->spi_rcv_egrbuftotlen =
+		rcd->rcvegrbuf_chunks * rcd->rcvegrbuf_size;
+	kinfo->spi_rcv_egrperchunk = rcd->rcvegrbufs_perchunk;
+	kinfo->spi_rcv_egrchunksize = kinfo->spi_rcv_egrbuftotlen /
+		rcd->rcvegrbuf_chunks;
+	kinfo->spi_tidcnt = dd->rcvtidcnt / subctxt_cnt;
+	if (master)
+		kinfo->spi_tidcnt += dd->rcvtidcnt % subctxt_cnt;
+	/*
+	 * for this use, may be cfgctxts summed over all chips that
+	 * are are configured and present
+	 */
+	kinfo->spi_nctxts = dd->cfgctxts;
+	/* unit (chip/board) our context is on */
+	kinfo->spi_unit = dd->unit;
+	kinfo->spi_port = ppd->port;
+	/* for now, only a single page */
+	kinfo->spi_tid_maxsize = PAGE_SIZE;
+
+	/*
+	 * Doing this per context, and based on the skip value, etc.  This has
+	 * to be the actual buffer size, since the protocol code treats it
+	 * as an array.
+	 *
+	 * These have to be set to user addresses in the user code via mmap.
+	 * These values are used on return to user code for the mmap target
+	 * addresses only.  For 32 bit, same 44 bit address problem, so use
+	 * the physical address, not virtual.  Before 2.6.11, using the
+	 * page_address() macro worked, but in 2.6.11, even that returns the
+	 * full 64 bit address (upper bits all 1's).  So far, using the
+	 * physical addresses (or chip offsets, for chip mapping) works, but
+	 * no doubt some future kernel release will change that, and we'll be
+	 * on to yet another method of dealing with this.
+	 * Normally only one of rcvhdr_tailaddr or rhf_offset is useful
+	 * since the chips with non-zero rhf_offset don't normally
+	 * enable tail register updates to host memory, but for testing,
+	 * both can be enabled and used.
+	 */
+	kinfo->spi_rcvhdr_base = (u64) rcd->rcvhdrq_phys;
+	kinfo->spi_rcvhdr_tailaddr = (u64) rcd->rcvhdrqtailaddr_phys;
+	kinfo->spi_rhf_offset = dd->rhf_offset;
+	kinfo->spi_rcv_egrbufs = (u64) rcd->rcvegr_phys;
+	kinfo->spi_pioavailaddr = (u64) dd->pioavailregs_phys;
+	/* setup per-unit (not port) status area for user programs */
+	kinfo->spi_status = (u64) kinfo->spi_pioavailaddr +
+		(char *) ppd->statusp -
+		(char *) dd->pioavailregs_dma;
+	kinfo->spi_uregbase = (u64) dd->uregbase + dd->ureg_align * rcd->ctxt;
+	if (!shared) {
+		kinfo->spi_piocnt = rcd->piocnt;
+		kinfo->spi_piobufbase = (u64) rcd->piobufs;
+		kinfo->spi_sendbuf_status = cvt_kvaddr(rcd->user_event_mask);
+	} else if (master) {
+		kinfo->spi_piocnt = (rcd->piocnt / subctxt_cnt) +
+				    (rcd->piocnt % subctxt_cnt);
+		/* Master's PIO buffers are after all the slave's */
+		kinfo->spi_piobufbase = (u64) rcd->piobufs +
+			dd->palign *
+			(rcd->piocnt - kinfo->spi_piocnt);
+	} else {
+		unsigned slave = subctxt_fp(fp) - 1;
+
+		kinfo->spi_piocnt = rcd->piocnt / subctxt_cnt;
+		kinfo->spi_piobufbase = (u64) rcd->piobufs +
+			dd->palign * kinfo->spi_piocnt * slave;
+	}
+
+	if (shared) {
+		kinfo->spi_sendbuf_status =
+			cvt_kvaddr(&rcd->user_event_mask[subctxt_fp(fp)]);
+		/* only spi_subctxt_* fields should be set in this block! */
+		kinfo->spi_subctxt_uregbase = cvt_kvaddr(rcd->subctxt_uregbase);
+
+		kinfo->spi_subctxt_rcvegrbuf =
+			cvt_kvaddr(rcd->subctxt_rcvegrbuf);
+		kinfo->spi_subctxt_rcvhdr_base =
+			cvt_kvaddr(rcd->subctxt_rcvhdr_base);
+		qib_cdbg(PROC, "ctxt %u:%u flags %x %llx %llx %llx %llx\n",
+			 rcd->ctxt, subctxt_fp(fp), kinfo->spi_runtime_flags,
+			 (unsigned long long) kinfo->spi_subctxt_uregbase,
+			 (unsigned long long) kinfo->spi_uregbase,
+			 (unsigned long long) kinfo->spi_subctxt_rcvegrbuf,
+			 (unsigned long long) kinfo->spi_subctxt_rcvhdr_base);
+	}
+
+	/*
+	 * All user buffers are 2KB buffers.  If we ever support
+	 * giving 4KB buffers to user processes, this will need some
+	 * work.  Can't use piobufbase directly, because it has
+	 * both 2K and 4K buffer base values.
+	 */
+	kinfo->spi_pioindex = (kinfo->spi_piobufbase - dd->pio2k_bufbase) /
+		dd->palign;
+	kinfo->spi_pioalign = dd->palign;
+	qib_cdbg(PROC, "%u:%u (pid%u) gets piobufs %u - %u (cnt %u)\n",
+		 dd->unit, rcd->ctxt, rcd->pid,
+		 kinfo->spi_pioindex,
+		 kinfo->spi_pioindex + kinfo->spi_piocnt - 1,
+		 kinfo->spi_piocnt);
+
+	kinfo->spi_qpair = QIB_KD_QP;
+	/*
+	 * user mode PIO buffers are always 2KB, even when 4KB can
+	 * be received, and sent via the kernel; this is ibmaxlen
+	 * for 2K MTU.
+	 */
+	kinfo->spi_piosize = dd->piosize2k - 2 * sizeof(u32);
+	kinfo->spi_mtu = ppd->ibmaxlen; /* maxlen, not ibmtu */
+	kinfo->spi_ctxt = rcd->ctxt;
+	kinfo->spi_subctxt = subctxt_fp(fp);
+	kinfo->spi_sw_version = QIB_KERN_SWVERSION;
+	kinfo->spi_sw_version |= 1U << 31; /* QLogic-built, not kernel.org */
+	kinfo->spi_hw_version = dd->revision;
+
+	if (master)
+		kinfo->spi_runtime_flags |= QIB_RUNTIME_MASTER;
+
+	sz = (ubase_size < sizeof(*kinfo)) ? ubase_size : sizeof(*kinfo);
+	if (copy_to_user(ubase, kinfo, sz))
+		ret = -EFAULT;
+bail:
+	kfree(kinfo);
+	return ret;
+}
+
+/**
+ * qib_tid_update - update a context TID
+ * @rcd: the context
+ * @fp: the qib device file
+ * @ti: the TID information
+ *
+ * The new implementation as of Oct 2004 is that the driver assigns
+ * the tid and returns it to the caller.   To reduce search time, we
+ * keep a cursor for each context, walking the shadow tid array to find
+ * one that's not in use.
+ *
+ * For now, if we can't allocate the full list, we fail, although
+ * in the long run, we'll allocate as many as we can, and the
+ * caller will deal with that by trying the remaining pages later.
+ * That means that when we fail, we have to mark the tids as not in
+ * use again, in our shadow copy.
+ *
+ * It's up to the caller to free the tids when they are done.
+ * We'll unlock the pages as they free them.
+ *
+ * Also, right now we are locking one page at a time, but since
+ * the intended use of this routine is for a single group of
+ * virtually contiguous pages, that should change to improve
+ * performance.
+ */
+static int qib_tid_update(struct qib_ctxtdata *rcd, struct file *fp,
+			  const struct qib_tid_info *ti)
+{
+	int ret = 0, ntids;
+	u32 tid, ctxttid, cnt, i, tidcnt, tidoff;
+	u16 *tidlist;
+	struct qib_devdata *dd = rcd->dd;
+	u64 physaddr;
+	unsigned long vaddr;
+	u64 __iomem *tidbase;
+	unsigned long tidmap[8];
+	struct page **pagep = NULL;
+	unsigned subctxt = subctxt_fp(fp);
+
+	if (!dd->pageshadow) {
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	cnt = ti->tidcnt;
+	if (!cnt) {
+		qib_dbg("After copyin, tidcnt 0, tidlist %llx\n",
+			(unsigned long long) ti->tidlist);
+		ret = -EFAULT;
+		goto done;
+	}
+	ctxttid = rcd->ctxt * dd->rcvtidcnt;
+	if (!rcd->subctxt_cnt) {
+		tidcnt = dd->rcvtidcnt;
+		tid = rcd->tidcursor;
+		tidoff = 0;
+	} else if (!subctxt) {
+		tidcnt = (dd->rcvtidcnt / rcd->subctxt_cnt) +
+			 (dd->rcvtidcnt % rcd->subctxt_cnt);
+		tidoff = dd->rcvtidcnt - tidcnt;
+		ctxttid += tidoff;
+		tid = tidcursor_fp(fp);
+	} else {
+		tidcnt = dd->rcvtidcnt / rcd->subctxt_cnt;
+		tidoff = tidcnt * (subctxt - 1);
+		ctxttid += tidoff;
+		tid = tidcursor_fp(fp);
+	}
+	if (cnt > tidcnt) {
+		/* make sure it all fits in tid_pg_list */
+		qib_devinfo(dd->pcidev, "Process tried to allocate %u "
+			 "TIDs, only trying max (%u)\n", cnt, tidcnt);
+		cnt = tidcnt;
+	}
+	pagep = (struct page **) rcd->tid_pg_list;
+	tidlist = (u16 *) &pagep[dd->rcvtidcnt];
+	pagep += tidoff;
+	tidlist += tidoff;
+
+	memset(tidmap, 0, sizeof(tidmap));
+	/* before decrement; chip actual # */
+	ntids = tidcnt;
+	tidbase = (u64 __iomem *) (((char __iomem *) dd->kregbase) +
+				   dd->rcvtidbase +
+				   ctxttid * sizeof(*tidbase));
+
+	qib_cdbg(VERBOSE, "ctxt%u %u tids, cursor %u, tidbase %p\n",
+		 rcd->ctxt, cnt, tid, tidbase);
+
+	/* virtual address of first page in transfer */
+	vaddr = ti->tidvaddr;
+	if (!access_ok(VERIFY_WRITE, (void __user *) vaddr,
+		       cnt * PAGE_SIZE)) {
+		qib_dbg("Fail vaddr %p, %u pages, !access_ok\n",
+			  (void *)vaddr, cnt);
+		ret = -EFAULT;
+		goto done;
+	}
+	ret = qib_get_user_pages(vaddr, cnt, pagep);
+	if (ret) {
+		/*
+		 * if (ret == -EBUSY)
+		 * We can't continue because the pagep array won't be
+		 * initialized. This should never happen,
+		 * unless perhaps the user has mpin'ed the pages
+		 * themselves.
+		 */
+		qib_devinfo(dd->pcidev,
+			 "Failed to lock addr %p, %u pages: "
+			 "errno %d\n", (void *) vaddr, cnt, -ret);
+		goto done;
+	}
+	for (i = 0; i < cnt; i++, vaddr += PAGE_SIZE) {
+		for (; ntids--; tid++) {
+			if (tid == tidcnt)
+				tid = 0;
+			if (!dd->pageshadow[ctxttid + tid])
+				break;
+		}
+		if (ntids < 0) {
+			/*
+			 * Oops, wrapped all the way through their TIDs,
+			 * and didn't have enough free; see comments at
+			 * start of routine
+			 */
+			qib_dbg("Not enough free TIDs for %u pages "
+				"(index %d), failing\n", cnt, i);
+			i--;    /* last tidlist[i] not filled in */
+			ret = -ENOMEM;
+			break;
+		}
+		tidlist[i] = tid + tidoff;
+		qib_cdbg(VERBOSE, "Updating idx %u to TID %u, "
+			 "vaddr %lx\n", i, tid + tidoff, vaddr);
+		/* we "know" system pages and TID pages are same size */
+		dd->pageshadow[ctxttid + tid] = pagep[i];
+		dd->physshadow[ctxttid + tid] =
+			qib_map_page(dd->pcidev, pagep[i], 0, PAGE_SIZE,
+				     PCI_DMA_FROMDEVICE);
+		/*
+		 * don't need atomic or it's overhead
+		 */
+		__set_bit(tid, tidmap);
+		physaddr = dd->physshadow[ctxttid + tid];
+		qib_cdbg(VERBOSE,
+			 "TID %u, vaddr %lx, physaddr %llx pgp %p\n",
+			 tid, vaddr, (unsigned long long) physaddr,
+			 pagep[i]);
+		/* PERFORMANCE: below should almost certainly be cached */
+		dd->f_put_tid(dd, &tidbase[tid],
+				  RCVHQ_RCV_TYPE_EXPECTED, physaddr);
+		/*
+		 * don't check this tid in qib_ctxtshadow, since we
+		 * just filled it in; start with the next one.
+		 */
+		tid++;
+	}
+
+	if (ret) {
+		u32 limit;
+cleanup:
+		/* jump here if copy out of updated info failed... */
+		qib_dbg("After failure (ret=%d), undo %d of %d entries\n",
+			-ret, i, cnt);
+		/* same code that's in qib_free_tid() */
+		limit = sizeof(tidmap) * BITS_PER_BYTE;
+		if (limit > tidcnt)
+			/* just in case size changes in future */
+			limit = tidcnt;
+		tid = find_first_bit((const unsigned long *)tidmap, limit);
+		for (; tid < limit; tid++) {
+			if (!test_bit(tid, tidmap))
+				continue;
+			if (dd->pageshadow[ctxttid + tid]) {
+				dma_addr_t phys;
+
+				qib_cdbg(VERBOSE, "Freeing TID %u\n", tid);
+				phys = dd->physshadow[ctxttid + tid];
+				dd->physshadow[ctxttid + tid] = dd->tidinvalid;
+				/* PERFORMANCE: below should almost certainly
+				 * be cached
+				 */
+				dd->f_put_tid(dd, &tidbase[tid],
+					      RCVHQ_RCV_TYPE_EXPECTED,
+					      dd->tidinvalid);
+				pci_unmap_page(dd->pcidev, phys, PAGE_SIZE,
+					       PCI_DMA_FROMDEVICE);
+				dd->pageshadow[ctxttid + tid] = NULL;
+			}
+		}
+		qib_release_user_pages(pagep, cnt);
+	} else {
+		/*
+		 * Copy the updated array, with qib_tid's filled in, back
+		 * to user.  Since we did the copy in already, this "should
+		 * never fail" If it does, we have to clean up...
+		 */
+		if (copy_to_user((void __user *)
+				 (unsigned long) ti->tidlist,
+				 tidlist, cnt * sizeof(*tidlist))) {
+			ret = -EFAULT;
+			goto cleanup;
+		}
+		if (copy_to_user((void __user *) (unsigned long) ti->tidmap,
+				 tidmap, sizeof tidmap)) {
+			ret = -EFAULT;
+			goto cleanup;
+		}
+		if (tid == tidcnt)
+			tid = 0;
+		if (!rcd->subctxt_cnt)
+			rcd->tidcursor = tid;
+		else
+			tidcursor_fp(fp) = tid;
+	}
+
+done:
+	if (ret)
+		qib_dbg("Failed to map %u TID pages, failing with %d\n",
+			ti->tidcnt, -ret);
+	return ret;
+}
+
+/**
+ * qib_tid_free - free a context TID
+ * @rcd: the context
+ * @subctxt: the subcontext
+ * @ti: the TID info
+ *
+ * right now we are unlocking one page at a time, but since
+ * the intended use of this routine is for a single group of
+ * virtually contiguous pages, that should change to improve
+ * performance.  We check that the TID is in range for this context
+ * but otherwise don't check validity; if user has an error and
+ * frees the wrong tid, it's only their own data that can thereby
+ * be corrupted.  We do check that the TID was in use, for sanity
+ * We always use our idea of the saved address, not the address that
+ * they pass in to us.
+ */
+static int qib_tid_free(struct qib_ctxtdata *rcd, unsigned subctxt,
+			const struct qib_tid_info *ti)
+{
+	int ret = 0;
+	u32 tid, ctxttid, cnt, limit, tidcnt;
+	struct qib_devdata *dd = rcd->dd;
+	u64 __iomem *tidbase;
+	unsigned long tidmap[8];
+
+	if (!dd->pageshadow) {
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	if (copy_from_user(tidmap, (void __user *)(unsigned long)ti->tidmap,
+			   sizeof tidmap)) {
+		ret = -EFAULT;
+		goto done;
+	}
+
+	ctxttid = rcd->ctxt * dd->rcvtidcnt;
+	if (!rcd->subctxt_cnt)
+		tidcnt = dd->rcvtidcnt;
+	else if (!subctxt) {
+		tidcnt = (dd->rcvtidcnt / rcd->subctxt_cnt) +
+			 (dd->rcvtidcnt % rcd->subctxt_cnt);
+		ctxttid += dd->rcvtidcnt - tidcnt;
+	} else {
+		tidcnt = dd->rcvtidcnt / rcd->subctxt_cnt;
+		ctxttid += tidcnt * (subctxt - 1);
+	}
+	tidbase = (u64 __iomem *) ((char __iomem *)(dd->kregbase) +
+				   dd->rcvtidbase +
+				   ctxttid * sizeof(*tidbase));
+
+	limit = sizeof(tidmap) * BITS_PER_BYTE;
+	if (limit > tidcnt)
+		/* just in case size changes in future */
+		limit = tidcnt;
+	tid = find_first_bit(tidmap, limit);
+	qib_cdbg(VERBOSE, "Context%u free %u tids; first bit (max=%d) "
+		 "set is %d, ctxttid %u\n", rcd->ctxt, ti->tidcnt,
+		 limit, tid, ctxttid);
+	for (cnt = 0; tid < limit; tid++) {
+		/*
+		 * small optimization; if we detect a run of 3 or so without
+		 * any set, use find_first_bit again.  That's mainly to
+		 * accelerate the case where we wrapped, so we have some at
+		 * the beginning, and some at the end, and a big gap
+		 * in the middle.
+		 */
+		if (!test_bit(tid, tidmap))
+			continue;
+		cnt++;
+		if (dd->pageshadow[ctxttid + tid]) {
+			struct page *p;
+			dma_addr_t phys;
+
+			p = dd->pageshadow[ctxttid + tid];
+			dd->pageshadow[ctxttid + tid] = NULL;
+			qib_cdbg(VERBOSE, "PID %u freeing TID %u\n",
+				 rcd->pid, tid);
+			phys = dd->physshadow[ctxttid + tid];
+			dd->physshadow[ctxttid + tid] = dd->tidinvalid;
+			/* PERFORMANCE: below should almost certainly be
+			 * cached
+			 */
+			dd->f_put_tid(dd, &tidbase[tid],
+				      RCVHQ_RCV_TYPE_EXPECTED, dd->tidinvalid);
+			pci_unmap_page(dd->pcidev, phys, PAGE_SIZE,
+				       PCI_DMA_FROMDEVICE);
+			qib_release_user_pages(&p, 1);
+		} else
+			qib_dbg("Unused tid %u, ignoring\n", tid);
+	}
+	if (cnt != ti->tidcnt)
+		qib_dbg("passed in tidcnt %d, only %d bits set in map\n",
+			ti->tidcnt, cnt);
+done:
+	if (ret)
+		qib_dbg("Failed to unmap %u TID pages, failing with %d\n",
+			ti->tidcnt, -ret);
+	return ret;
+}
+
+/**
+ * qib_set_part_key - set a partition key
+ * @rcd: the context
+ * @key: the key
+ *
+ * We can have up to 4 active at a time (other than the default, which is
+ * always allowed).  This is somewhat tricky, since multiple contexts may set
+ * the same key, so we reference count them, and clean up at exit.  All 4
+ * partition keys are packed into a single qlogic_ib register.  It's an
+ * error for a process to set the same pkey multiple times.  We provide no
+ * mechanism to de-allocate a pkey at this time, we may eventually need to
+ * do that.  I've used the atomic operations, and no locking, and only make
+ * a single pass through what's available.  This should be more than
+ * adequate for some time. I'll think about spinlocks or the like if and as
+ * it's necessary.
+ */
+static int qib_set_part_key(struct qib_ctxtdata *rcd, u16 key)
+{
+	struct qib_pportdata *ppd = rcd->ppd;
+	int i, any = 0, pidx = -1;
+	u16 lkey = key & 0x7FFF;
+	int ret;
+
+	if (lkey == (QIB_DEFAULT_P_KEY & 0x7FFF)) {
+		/* nothing to do; this key always valid */
+		ret = 0;
+		goto bail;
+	}
+
+	qib_cdbg(VERBOSE, "p%u try to set pkey %hx, current keys "
+		 "%hx:%x %hx:%x %hx:%x %hx:%x\n",
+		 rcd->ctxt, key, ppd->pkeys[0],
+		 atomic_read(&ppd->pkeyrefs[0]), ppd->pkeys[1],
+		 atomic_read(&ppd->pkeyrefs[1]), ppd->pkeys[2],
+		 atomic_read(&ppd->pkeyrefs[2]), ppd->pkeys[3],
+		 atomic_read(&ppd->pkeyrefs[3]));
+
+	if (!lkey) {
+		qib_cdbg(PROC, "p%u tries to set key 0, not allowed\n",
+			 rcd->ctxt);
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	/*
+	 * Set the full membership bit, because it has to be
+	 * set in the register or the packet, and it seems
+	 * cleaner to set in the register than to force all
+	 * callers to set it.
+	 */
+	key |= 0x8000;
+
+	for (i = 0; i < ARRAY_SIZE(rcd->pkeys); i++) {
+		if (!rcd->pkeys[i] && pidx == -1)
+			pidx = i;
+		if (rcd->pkeys[i] == key) {
+			qib_cdbg(VERBOSE, "p%u tries to set same pkey "
+				 "(%x) more than once\n",
+				 rcd->ctxt, key);
+			ret = -EEXIST;
+			goto bail;
+		}
+	}
+	if (pidx == -1) {
+		qib_dbg("All pkeys for context %u already in use, "
+			"can't set %x\n", rcd->ctxt, key);
+		ret = -EBUSY;
+		goto bail;
+	}
+	for (any = i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
+		if (!ppd->pkeys[i]) {
+			any++;
+			continue;
+		}
+		if (ppd->pkeys[i] == key) {
+			atomic_t *pkrefs = &ppd->pkeyrefs[i];
+
+			if (atomic_inc_return(pkrefs) > 1) {
+				rcd->pkeys[pidx] = key;
+				qib_cdbg(VERBOSE, "p%u set key %x "
+					 "matches #%d, count now %d\n",
+					 rcd->ctxt, key, i,
+					 atomic_read(pkrefs));
+				ret = 0;
+				goto bail;
+			} else {
+				/*
+				 * lost race, decrement count, catch below
+				 */
+				atomic_dec(pkrefs);
+				qib_cdbg(VERBOSE, "Lost race, count was "
+					 "0, after dec, it's %d\n",
+					 atomic_read(pkrefs));
+				any++;
+			}
+		}
+		if ((ppd->pkeys[i] & 0x7FFF) == lkey) {
+			/*
+			 * It makes no sense to have both the limited and
+			 * full membership PKEY set at the same time since
+			 * the unlimited one will disable the limited one.
+			 */
+			ret = -EEXIST;
+			goto bail;
+		}
+	}
+	if (!any) {
+		qib_dbg("context %u, all pkeys already in use, "
+			"can't set %x\n", rcd->ctxt, key);
+		ret = -EBUSY;
+		goto bail;
+	}
+	for (any = i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
+		if (!ppd->pkeys[i] &&
+		    atomic_inc_return(&ppd->pkeyrefs[i]) == 1) {
+			rcd->pkeys[pidx] = key;
+			ppd->pkeys[i] = key;
+			(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_PKEYS, 0);
+			ret = 0;
+			goto bail;
+		}
+	}
+	qib_dbg("ctxt %u, all pkeys already in use 2nd pass, "
+		"can't set %x\n", rcd->ctxt, key);
+	ret = -EBUSY;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_manage_rcvq - manage a context's receive queue
+ * @rcd: the context
+ * @subctxt: the subcontext
+ * @start_stop: action to carry out
+ *
+ * start_stop == 0 disables receive on the context, for use in queue
+ * overflow conditions.  start_stop==1 re-enables, to be used to
+ * re-init the software copy of the head register
+ */
+static int qib_manage_rcvq(struct qib_ctxtdata *rcd, unsigned subctxt,
+			   int start_stop)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned int rcvctrl_op;
+
+	qib_cdbg(PROC, "%sabling rcv for unit %u ctxt %u:%u\n",
+		 start_stop ? "en" : "dis", dd->unit,
+		 rcd->ctxt, subctxt);
+	if (subctxt)
+		goto bail;
+	/* atomically clear receive enable ctxt. */
+	if (start_stop) {
+		/*
+		 * On enable, force in-memory copy of the tail register to
+		 * 0, so that protocol code doesn't have to worry about
+		 * whether or not the chip has yet updated the in-memory
+		 * copy or not on return from the system call. The chip
+		 * always resets it's tail register back to 0 on a
+		 * transition from disabled to enabled.
+		 */
+		if (rcd->rcvhdrtail_kvaddr)
+			qib_clear_rcvhdrtail(rcd);
+		rcvctrl_op = QIB_RCVCTRL_CTXT_ENB;
+	} else
+		rcvctrl_op = QIB_RCVCTRL_CTXT_DIS;
+	dd->f_rcvctrl(rcd->ppd, rcvctrl_op, rcd->ctxt);
+	/* always; new head should be equal to new tail; see above */
+bail:
+	return 0;
+}
+
+static void qib_clean_part_key(struct qib_ctxtdata *rcd,
+			       struct qib_devdata *dd)
+{
+	int i, j, pchanged = 0;
+	u64 oldpkey;
+	struct qib_pportdata *ppd = rcd->ppd;
+
+	/* for debugging only */
+	oldpkey = (u64) ppd->pkeys[0] |
+		((u64) ppd->pkeys[1] << 16) |
+		((u64) ppd->pkeys[2] << 32) |
+		((u64) ppd->pkeys[3] << 48);
+
+	for (i = 0; i < ARRAY_SIZE(rcd->pkeys); i++) {
+		if (!rcd->pkeys[i])
+			continue;
+		qib_cdbg(VERBOSE, "look for key[%d] %hx in pkeys\n", i,
+			 rcd->pkeys[i]);
+		for (j = 0; j < ARRAY_SIZE(ppd->pkeys); j++) {
+			/* check for match independent of the global bit */
+			if ((ppd->pkeys[j] & 0x7fff) !=
+			    (rcd->pkeys[i] & 0x7fff))
+				continue;
+			if (atomic_dec_and_test(&ppd->pkeyrefs[j])) {
+				qib_cdbg(VERBOSE, "p%u clear key "
+					 "%x matches #%d\n",
+					 rcd->ctxt,
+					 rcd->pkeys[i], j);
+				ppd->pkeys[j] = 0;
+				pchanged++;
+			} else
+				qib_cdbg(VERBOSE, "p%u key %x matches #%d, "
+					 "but ref still %d\n", rcd->ctxt,
+					 rcd->pkeys[i], j,
+					 atomic_read(&ppd->pkeyrefs[j]));
+			break;
+		}
+		rcd->pkeys[i] = 0;
+	}
+	if (pchanged)
+		(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_PKEYS, 0);
+}
+
+/* common code for the mappings on dma_alloc_coherent mem */
+static int qib_mmap_mem(struct vm_area_struct *vma, struct qib_ctxtdata *rcd,
+			unsigned len, void *kvaddr, u32 write_ok, char *what)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned long pfn;
+	int ret;
+
+	if ((vma->vm_end - vma->vm_start) > len) {
+		qib_devinfo(dd->pcidev,
+			 "FAIL on %s: len %lx > %x\n", what,
+			 vma->vm_end - vma->vm_start, len);
+		ret = -EFAULT;
+		goto bail;
+	}
+
+	/*
+	 * shared context user code requires rcvhdrq mapped r/w, others
+	 * only allowed readonly mapping.
+	 */
+	if (!write_ok) {
+		if (vma->vm_flags & VM_WRITE) {
+			qib_devinfo(dd->pcidev,
+				 "%s must be mapped readonly\n", what);
+			ret = -EPERM;
+			goto bail;
+		}
+
+		/* don't allow them to later change with mprotect */
+		vma->vm_flags &= ~VM_MAYWRITE;
+	}
+
+	pfn = virt_to_phys(kvaddr) >> PAGE_SHIFT;
+	ret = remap_pfn_range(vma, vma->vm_start, pfn,
+			      len, vma->vm_page_prot);
+	if (ret)
+		qib_devinfo(dd->pcidev, "%s ctxt%u mmap of %lx, %x "
+			 "bytes failed: %d\n", what, rcd->ctxt,
+			 pfn, len, ret);
+	else
+		qib_cdbg(VERBOSE, "%s ctxt%u mmaped %lx, %x bytes\n",
+			 what, rcd->ctxt, pfn, len);
+bail:
+	return ret;
+}
+
+static int mmap_ureg(struct vm_area_struct *vma, struct qib_devdata *dd,
+		     u64 ureg)
+{
+	unsigned long phys;
+	unsigned long sz;
+	int ret;
+
+	/*
+	 * This is real hardware, so use io_remap.  This is the mechanism
+	 * for the user process to update the head registers for their ctxt
+	 * in the chip.
+	 */
+	sz = dd->flags & QIB_HAS_HDRSUPP ? 2 * PAGE_SIZE : PAGE_SIZE;
+	if ((vma->vm_end - vma->vm_start) > sz) {
+		qib_devinfo(dd->pcidev, "FAIL mmap userreg: reqlen "
+			 "%lx > PAGE\n", vma->vm_end - vma->vm_start);
+		ret = -EFAULT;
+	} else {
+		phys = dd->physaddr + ureg;
+		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+
+		vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
+		ret = io_remap_pfn_range(vma, vma->vm_start,
+					 phys >> PAGE_SHIFT,
+					 vma->vm_end - vma->vm_start,
+					 vma->vm_page_prot);
+	}
+	return ret;
+}
+
+static int mmap_piobufs(struct vm_area_struct *vma,
+			struct qib_devdata *dd,
+			struct qib_ctxtdata *rcd,
+			unsigned piobufs, unsigned piocnt)
+{
+	unsigned long phys;
+	int ret;
+
+	/*
+	 * When we map the PIO buffers in the chip, we want to map them as
+	 * writeonly, no read possible; unfortunately, x86 doesn't allow
+	 * for this in hardware, but we still prevent users from asking
+	 * for it.
+	 */
+	if ((vma->vm_end - vma->vm_start) > (piocnt * dd->palign)) {
+		qib_devinfo(dd->pcidev, "FAIL mmap piobufs: "
+			 "reqlen %lx > PAGE\n",
+			 vma->vm_end - vma->vm_start);
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	phys = dd->physaddr + piobufs;
+
+#if defined(__powerpc__)
+	/* There isn't a generic way to specify writethrough mappings */
+	pgprot_val(vma->vm_page_prot) |= _PAGE_NO_CACHE;
+	pgprot_val(vma->vm_page_prot) |= _PAGE_WRITETHRU;
+	pgprot_val(vma->vm_page_prot) &= ~_PAGE_GUARDED;
+#endif
+
+	/*
+	 * don't allow them to later change to readable with mprotect (for when
+	 * not initially mapped readable, as is normally the case)
+	 */
+	vma->vm_flags &= ~VM_MAYREAD;
+	vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
+
+	if (qib_wc_pat)
+		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
+
+	ret = io_remap_pfn_range(vma, vma->vm_start, phys >> PAGE_SHIFT,
+				 vma->vm_end - vma->vm_start,
+				 vma->vm_page_prot);
+bail:
+	return ret;
+}
+
+static int mmap_rcvegrbufs(struct vm_area_struct *vma,
+			   struct qib_ctxtdata *rcd)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned long start, size;
+	size_t total_size, i;
+	unsigned long pfn;
+	int ret;
+
+	size = rcd->rcvegrbuf_size;
+	total_size = rcd->rcvegrbuf_chunks * size;
+	if ((vma->vm_end - vma->vm_start) > total_size) {
+		qib_devinfo(dd->pcidev, "FAIL on egr bufs: "
+			 "reqlen %lx > actual %lx\n",
+			 vma->vm_end - vma->vm_start,
+			 (unsigned long) total_size);
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	if (vma->vm_flags & VM_WRITE) {
+		qib_devinfo(dd->pcidev, "Can't map eager buffers as "
+			 "writable (flags=%lx)\n", vma->vm_flags);
+		ret = -EPERM;
+		goto bail;
+	}
+	/* don't allow them to later change to writeable with mprotect */
+	vma->vm_flags &= ~VM_MAYWRITE;
+
+	start = vma->vm_start;
+
+	for (i = 0; i < rcd->rcvegrbuf_chunks; i++, start += size) {
+		pfn = virt_to_phys(rcd->rcvegrbuf[i]) >> PAGE_SHIFT;
+		ret = remap_pfn_range(vma, start, pfn, size,
+				      vma->vm_page_prot);
+		if (ret < 0)
+			goto bail;
+	}
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/*
+ * qib_file_vma_fault - handle a VMA page fault.
+ */
+static int qib_file_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
+{
+	struct page *page;
+
+	page = vmalloc_to_page((void *)(vmf->pgoff << PAGE_SHIFT));
+	if (!page)
+		return VM_FAULT_SIGBUS;
+
+	get_page(page);
+	vmf->page = page;
+
+	return 0;
+}
+
+static struct vm_operations_struct qib_file_vm_ops = {
+	.fault = qib_file_vma_fault,
+};
+
+static int mmap_kvaddr(struct vm_area_struct *vma, u64 pgaddr,
+		       struct qib_ctxtdata *rcd, unsigned subctxt)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned subctxt_cnt;
+	unsigned long len;
+	void *addr;
+	size_t size;
+	int ret = 0;
+
+	subctxt_cnt = rcd->subctxt_cnt;
+	size = rcd->rcvegrbuf_chunks * rcd->rcvegrbuf_size;
+
+	/*
+	 * Each process has all the subctxt uregbase, rcvhdrq, and
+	 * rcvegrbufs mmapped - as an array for all the processes,
+	 * and also separately for this process.
+	 */
+	if (pgaddr == cvt_kvaddr(rcd->subctxt_uregbase)) {
+		addr = rcd->subctxt_uregbase;
+		size = PAGE_SIZE * subctxt_cnt;
+	} else if (pgaddr == cvt_kvaddr(rcd->subctxt_rcvhdr_base)) {
+		addr = rcd->subctxt_rcvhdr_base;
+		size = rcd->rcvhdrq_size * subctxt_cnt;
+	} else if (pgaddr == cvt_kvaddr(rcd->subctxt_rcvegrbuf)) {
+		addr = rcd->subctxt_rcvegrbuf;
+		size *= subctxt_cnt;
+	} else if (pgaddr == cvt_kvaddr(rcd->subctxt_uregbase +
+					PAGE_SIZE * subctxt)) {
+		addr = rcd->subctxt_uregbase + PAGE_SIZE * subctxt;
+		size = PAGE_SIZE;
+	} else if (pgaddr == cvt_kvaddr(rcd->subctxt_rcvhdr_base +
+					rcd->rcvhdrq_size * subctxt)) {
+		addr = rcd->subctxt_rcvhdr_base +
+			rcd->rcvhdrq_size * subctxt;
+		size = rcd->rcvhdrq_size;
+	} else if (pgaddr == cvt_kvaddr(&rcd->user_event_mask[subctxt])) {
+		addr = rcd->user_event_mask;
+		size = PAGE_SIZE;
+	} else if (pgaddr == cvt_kvaddr(rcd->subctxt_rcvegrbuf +
+					size * subctxt)) {
+		addr = rcd->subctxt_rcvegrbuf + size * subctxt;
+		/* rcvegrbufs are read-only on the slave */
+		if (vma->vm_flags & VM_WRITE) {
+			qib_devinfo(dd->pcidev,
+				 "Can't map eager buffers as "
+				 "writable (flags=%lx)\n", vma->vm_flags);
+			ret = -EPERM;
+			goto bail;
+		}
+		/*
+		 * Don't allow permission to later change to writeable
+		 * with mprotect.
+		 */
+		vma->vm_flags &= ~VM_MAYWRITE;
+	} else
+		goto bail;
+	len = vma->vm_end - vma->vm_start;
+	if (len > size) {
+		qib_dbg("FAIL: reqlen %lx > %zx\n", len, size);
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	vma->vm_pgoff = (unsigned long) addr >> PAGE_SHIFT;
+	vma->vm_ops = &qib_file_vm_ops;
+	vma->vm_flags |= VM_RESERVED | VM_DONTEXPAND;
+	ret = 1;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_mmapf - mmap various structures into user space
+ * @fp: the file pointer
+ * @vma: the VM area
+ *
+ * We use this to have a shared buffer between the kernel and the user code
+ * for the rcvhdr queue, egr buffers, and the per-context user regs and pio
+ * buffers in the chip.  We have the open and close entries so we can bump
+ * the ref count and keep the driver from being unloaded while still mapped.
+ */
+static int qib_mmapf(struct file *fp, struct vm_area_struct *vma)
+{
+	struct qib_ctxtdata *rcd;
+	struct qib_devdata *dd;
+	u64 pgaddr, ureg;
+	unsigned piobufs, piocnt;
+	int ret, match = 1;
+
+	rcd = ctxt_fp(fp);
+	if (!rcd || !(vma->vm_flags & VM_SHARED)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	dd = rcd->dd;
+
+	/*
+	 * This is the qib_do_user_init() code, mapping the shared buffers
+	 * and per-context user registers into the user process. The address
+	 * referred to by vm_pgoff is the file offset passed via mmap().
+	 * For shared contexts, this is the kernel vmalloc() address of the
+	 * pages to share with the master.
+	 * For non-shared or master ctxts, this is a physical address.
+	 * We only do one mmap for each space mapped.
+	 */
+	pgaddr = vma->vm_pgoff << PAGE_SHIFT;
+
+	/*
+	 * Check for 0 in case one of the allocations failed, but user
+	 * called mmap anyway.
+	 */
+	if (!pgaddr)  {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	qib_cdbg(MM, "pgaddr %llx vm_start=%lx len %lx ctxt %u:%u:%u\n",
+		 (unsigned long long) pgaddr, vma->vm_start,
+		 vma->vm_end - vma->vm_start, dd->unit,
+		 rcd->ctxt, subctxt_fp(fp));
+
+	/*
+	 * Physical addresses must fit in 40 bits for our hardware.
+	 * Check for kernel virtual addresses first, anything else must
+	 * match a HW or memory address.
+	 */
+	ret = mmap_kvaddr(vma, pgaddr, rcd, subctxt_fp(fp));
+	if (ret) {
+		if (ret > 0)
+			ret = 0;
+		else
+			qib_cdbg(PROC, "ctxt%u:%x pgaddr %Lx mmapkv ret %d\n",
+				  rcd->ctxt, rcd->subctxt_id, pgaddr, ret);
+		goto bail;
+	}
+
+	ureg = dd->uregbase + dd->ureg_align * rcd->ctxt;
+	if (!rcd->subctxt_cnt) {
+		/* ctxt is not shared */
+		piocnt = rcd->piocnt;
+		piobufs = rcd->piobufs;
+	} else if (!subctxt_fp(fp)) {
+		/* caller is the master */
+		piocnt = (rcd->piocnt / rcd->subctxt_cnt) +
+			 (rcd->piocnt % rcd->subctxt_cnt);
+		piobufs = rcd->piobufs +
+			dd->palign * (rcd->piocnt - piocnt);
+	} else {
+		unsigned slave = subctxt_fp(fp) - 1;
+
+		/* caller is a slave */
+		piocnt = rcd->piocnt / rcd->subctxt_cnt;
+		piobufs = rcd->piobufs + dd->palign * piocnt * slave;
+	}
+
+	if (pgaddr == ureg)
+		ret = mmap_ureg(vma, dd, ureg);
+	else if (pgaddr == piobufs)
+		ret = mmap_piobufs(vma, dd, rcd, piobufs, piocnt);
+	else if (pgaddr == dd->pioavailregs_phys)
+		/* in-memory copy of pioavail registers */
+		ret = qib_mmap_mem(vma, rcd, PAGE_SIZE,
+				   (void *) dd->pioavailregs_dma, 0,
+				   "pioavail registers");
+	else if (pgaddr == rcd->rcvegr_phys)
+		ret = mmap_rcvegrbufs(vma, rcd);
+	else if (pgaddr == (u64) rcd->rcvhdrq_phys)
+		/*
+		 * The rcvhdrq itself; multiple pages, contiguous
+		 * from an i/o perspective.  Shared contexts need
+		 * to map r/w, so we allow writing.
+		 */
+		ret = qib_mmap_mem(vma, rcd, rcd->rcvhdrq_size,
+				   rcd->rcvhdrq, 1, "rcvhdrq");
+	else if (pgaddr == (u64) rcd->rcvhdrqtailaddr_phys)
+		/* in-memory copy of rcvhdrq tail register */
+		ret = qib_mmap_mem(vma, rcd, PAGE_SIZE,
+				   rcd->rcvhdrtail_kvaddr, 0,
+				   "rcvhdrq tail");
+	else
+		match = 0;
+	if (!match) {
+		ret = -EINVAL;
+		qib_dbg("ctxt%u:%u pgaddr %Lx matches no areas\n",
+			rcd->ctxt, rcd->subctxt_id, pgaddr);
+	} else if (ret)
+		qib_dbg("ctxt%u:%u pgaddr %Lx mmap fail %d\n",
+			rcd->ctxt, rcd->subctxt_id, pgaddr, ret);
+
+	vma->vm_private_data = NULL;
+
+	if (ret < 0)
+		qib_devinfo(dd->pcidev,
+			 "mmap Failure %d: off %llx len %lx\n",
+			 -ret, (unsigned long long)pgaddr,
+			 vma->vm_end - vma->vm_start);
+bail:
+	return ret;
+}
+
+static unsigned int qib_poll_urgent(struct qib_ctxtdata *rcd,
+				    struct file *fp,
+				    struct poll_table_struct *pt)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned pollflag;
+
+	poll_wait(fp, &rcd->wait, pt);
+
+	spin_lock_irq(&dd->uctxt_lock);
+	if (rcd->urgent != rcd->urgent_poll) {
+		pollflag = POLLIN | POLLRDNORM;
+		rcd->urgent_poll = rcd->urgent;
+	} else {
+		pollflag = 0;
+		set_bit(QIB_CTXT_WAITING_URG, &rcd->flag);
+	}
+	spin_unlock_irq(&dd->uctxt_lock);
+
+	return pollflag;
+}
+
+static unsigned int qib_poll_next(struct qib_ctxtdata *rcd,
+				  struct file *fp,
+				  struct poll_table_struct *pt)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned pollflag;
+
+	poll_wait(fp, &rcd->wait, pt);
+
+	spin_lock_irq(&dd->uctxt_lock);
+	if (dd->f_hdrqempty(rcd)) {
+		set_bit(QIB_CTXT_WAITING_RCV, &rcd->flag);
+		dd->f_rcvctrl(rcd->ppd, QIB_RCVCTRL_INTRAVAIL_ENB, rcd->ctxt);
+		pollflag = 0;
+	} else
+		pollflag = POLLIN | POLLRDNORM;
+	spin_unlock_irq(&dd->uctxt_lock);
+
+	return pollflag;
+}
+
+static unsigned int qib_poll(struct file *fp, struct poll_table_struct *pt)
+{
+	struct qib_ctxtdata *rcd;
+	unsigned pollflag;
+
+	rcd = ctxt_fp(fp);
+	if (!rcd)
+		pollflag = POLLERR;
+	else if (rcd->poll_type == QIB_POLL_TYPE_URGENT)
+		pollflag = qib_poll_urgent(rcd, fp, pt);
+	else  if (rcd->poll_type == QIB_POLL_TYPE_ANYRCV)
+		pollflag = qib_poll_next(rcd, fp, pt);
+	else /* invalid */
+		pollflag = POLLERR;
+
+	return pollflag;
+}
+
+/*
+ * Check that userland and driver are compatible for subcontexts.
+ */
+static int qib_compatible_subctxts(int user_swmajor, int user_swminor)
+{
+	/* this code is written long-hand for clarity */
+	if (QIB_USER_SWMAJOR != user_swmajor) {
+		/* no promise of compatibility if major mismatch */
+		return 0;
+	}
+	if (QIB_USER_SWMAJOR == 1) {
+		switch (QIB_USER_SWMINOR) {
+		case 0:
+		case 1:
+		case 2:
+			/* no subctxt implementation so cannot be compatible */
+			return 0;
+		case 3:
+			/* 3 is only compatible with itself */
+			return user_swminor == 3;
+		default:
+			/* >= 4 are compatible (or are expected to be) */
+			return user_swminor >= 4;
+		}
+	}
+	/* make no promises yet for future major versions */
+	return 0;
+}
+
+static int init_subctxts(struct qib_devdata *dd,
+			 struct qib_ctxtdata *rcd,
+			 const struct qib_user_info *uinfo)
+{
+	int ret = 0;
+	unsigned num_subctxts;
+	size_t size;
+
+	/*
+	 * If the user is requesting zero subctxts,
+	 * skip the subctxt allocation.
+	 */
+	if (uinfo->spu_subctxt_cnt <= 0)
+		goto bail;
+	num_subctxts = uinfo->spu_subctxt_cnt;
+
+	/* Check for subctxt compatibility */
+	if (!qib_compatible_subctxts(uinfo->spu_userversion >> 16,
+		uinfo->spu_userversion & 0xffff)) {
+		qib_devinfo(dd->pcidev,
+			 "Mismatched user version (%d.%d) and driver "
+			 "version (%d.%d) while context sharing. Ensure "
+			 "that driver and library are from the same "
+			 "release.\n",
+			 (int) (uinfo->spu_userversion >> 16),
+			 (int) (uinfo->spu_userversion & 0xffff),
+			 QIB_USER_SWMAJOR, QIB_USER_SWMINOR);
+		goto bail;
+	}
+	if (num_subctxts > QLOGIC_IB_MAX_SUBCTXT) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	rcd->subctxt_uregbase = vmalloc_user(PAGE_SIZE * num_subctxts);
+	if (!rcd->subctxt_uregbase) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+	/* Note: rcd->rcvhdrq_size isn't initialized yet. */
+	size = ALIGN(dd->rcvhdrcnt * dd->rcvhdrentsize *
+		     sizeof(u32), PAGE_SIZE) * num_subctxts;
+	rcd->subctxt_rcvhdr_base = vmalloc_user(size);
+	if (!rcd->subctxt_rcvhdr_base) {
+		ret = -ENOMEM;
+		goto bail_ureg;
+	}
+
+	rcd->subctxt_rcvegrbuf = vmalloc_user(rcd->rcvegrbuf_chunks *
+					      rcd->rcvegrbuf_size *
+					      num_subctxts);
+	if (!rcd->subctxt_rcvegrbuf) {
+		ret = -ENOMEM;
+		goto bail_rhdr;
+	}
+
+	rcd->subctxt_cnt = uinfo->spu_subctxt_cnt;
+	rcd->subctxt_id = uinfo->spu_subctxt_id;
+	rcd->active_slaves = 1;
+	rcd->redirect_seq_cnt = 1;
+	set_bit(QIB_CTXT_MASTER_UNINIT, &rcd->flag);
+	goto bail;
+
+bail_rhdr:
+	vfree(rcd->subctxt_rcvhdr_base);
+bail_ureg:
+	vfree(rcd->subctxt_uregbase);
+	rcd->subctxt_uregbase = NULL;
+bail:
+	return ret;
+}
+
+static int setup_ctxt(struct qib_pportdata *ppd, int ctxt,
+		      struct file *fp, const struct qib_user_info *uinfo)
+{
+	struct qib_devdata *dd = ppd->dd;
+	struct qib_ctxtdata *rcd;
+	void *ptmp = NULL;
+	int ret;
+
+	rcd = qib_create_ctxtdata(ppd, ctxt);
+
+	/*
+	 * Allocate memory for use in qib_tid_update() at open to
+	 * reduce cost of expected send setup per message segment
+	 */
+	if (rcd)
+		ptmp = kmalloc(dd->rcvtidcnt * sizeof(u16) +
+			       dd->rcvtidcnt * sizeof(struct page **),
+			       GFP_KERNEL);
+
+	if (!rcd || !ptmp) {
+		qib_dev_err(dd, "Unable to allocate ctxtdata "
+			    "memory, failing open\n");
+		ret = -ENOMEM;
+		goto bailerr;
+	}
+	rcd->userversion = uinfo->spu_userversion;
+	ret = init_subctxts(dd, rcd, uinfo);
+	if (ret)
+		goto bailerr;
+	rcd->tid_pg_list = ptmp;
+	rcd->pid = current->pid;
+	init_waitqueue_head(&dd->rcd[ctxt]->wait);
+	strlcpy(rcd->comm, current->comm, sizeof(rcd->comm));
+	ctxt_fp(fp) = rcd;
+	qib_cdbg(PROC, "%s[%u] opened u:p:c %u:%u:%u\n",
+		 rcd->comm, current->pid, dd->unit,
+		 rcd->ppd->port, ctxt);
+	qib_stats.sps_ctxts++;
+	ret = 0;
+	goto bail;
+
+bailerr:
+	dd->rcd[ctxt] = NULL;
+	kfree(rcd);
+	kfree(ptmp);
+bail:
+	return ret;
+}
+
+static inline int usable(struct qib_pportdata *ppd, int active_only)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u32 linkok = active_only ? QIBL_LINKACTIVE :
+		 (QIBL_LINKINIT | QIBL_LINKARMED | QIBL_LINKACTIVE);
+
+	return dd && (dd->flags & QIB_PRESENT) && dd->kregbase && ppd->lid &&
+		(ppd->lflags & linkok);
+}
+
+static int find_free_ctxt(int unit, struct file *fp,
+			  const struct qib_user_info *uinfo)
+{
+	struct qib_devdata *dd = qib_lookup(unit);
+	struct qib_pportdata *ppd = NULL;
+	int ret;
+	u32 ctxt;
+
+	if (!dd || (uinfo->spu_port && uinfo->spu_port > dd->num_pports)) {
+		if (dd)
+			qib_cdbg(PROC, "Skip invalid port %u on unit %u\n",
+				 uinfo->spu_port, unit);
+		ret = -ENODEV;
+		goto bail;
+	}
+
+	/*
+	 * If users requests specific port, only try that one port, else
+	 * select "best" port below, based on context.
+	 */
+	if (uinfo->spu_port) {
+		ppd = dd->pport + uinfo->spu_port - 1;
+		if (!usable(ppd, 0)) {
+			ret = -ENETDOWN;
+			goto bail;
+		}
+		qib_cdbg(PROC, "requested unit %u port %u\n", unit,
+			 uinfo->spu_port);
+	}
+
+	for (ctxt = dd->first_user_ctxt; ctxt < dd->cfgctxts; ctxt++) {
+		if (dd->rcd[ctxt])
+			continue;
+		/*
+		 * The setting and clearing of user context rcd[x] protected
+		 * by the qib_mutex
+		 */
+		if (!ppd) {
+			/* choose port based on ctxt, if up, else 1st up */
+			ppd = dd->pport + (ctxt % dd->num_pports);
+			if (!usable(ppd, 0)) {
+				int i;
+				for (i = 0; i < dd->num_pports; i++) {
+					ppd = dd->pport + i;
+					if (usable(ppd, 0))
+						break;
+				}
+				if (i == dd->num_pports) {
+					ret = -ENETDOWN;
+					goto bail;
+				}
+			}
+		}
+		ret = setup_ctxt(ppd, ctxt, fp, uinfo);
+		goto bail;
+	}
+	ret = -EBUSY;
+
+bail:
+	return ret;
+}
+
+static int find_best_unit(struct file *fp, const struct qib_user_info *uinfo)
+{
+	struct qib_pportdata *ppd;
+	int ret = 0, i, prefunit = -1, devmax;
+	int npresent, nup;
+	int ndev;
+	u32 port = uinfo->spu_port, ctxt;
+
+	devmax = qib_count_units(&npresent, &nup);
+
+	/*
+	 * This code is present to allow a knowledgeable person to
+	 * specify the layout of processes to processors before opening
+	 * this driver, and then we'll assign the process to the "closest"
+	 * QLogic_IB chip to that processor (we assume reasonable connectivity,
+	 * for now).  This code assumes that if affinity has been set
+	 * before this point, that at most one cpu is set; for now this
+	 * is reasonable.  I check for both cpumask_empty() and cpumask_full(),
+	 * in case some kernel variant sets none of the bits when no
+	 * affinity is set.
+	 * Some day we'll have to fix it up further to handle
+	 * a cpu subset.  Eventually this needs real topology
+	 * information.  There are issues with multi core numbering
+	 * as well.
+	 */
+	if (!cpumask_empty(tsk_cpumask(current)) &&
+	    !cpumask_full(tsk_cpumask(current))) {
+		int ncpus = num_online_cpus(), curcpu = -1, nset = 0;
+
+		for (i = 0; i < ncpus; i++)
+			if (cpu_isset(i, current->cpus_allowed)) {
+				if (qib_debug & __QIB_VERBDBG)
+					qib_cdbg(PROC, "%s[%u] cpuaff %d/%d"
+						 " set\n", current->comm,
+						 current->pid, i, ncpus);
+				curcpu = i;
+				nset++;
+			}
+		if (curcpu != -1 && nset != ncpus) {
+			if (npresent) {
+				prefunit = curcpu / (ncpus / npresent);
+				qib_cdbg(PROC, "%s[%u] %d chips, %d cpus, "
+					 "%d cpus/chip, select unit %d\n",
+					 current->comm, current->pid,
+					 npresent, ncpus, ncpus / npresent,
+					 prefunit);
+			}
+		}
+	}
+
+	/*
+	 * user contexts start at 1, kernel port is 0
+	 * For now, we do round-robin access across all chips
+	 */
+	if (prefunit != -1) {
+		devmax = prefunit + 1;
+		qib_cdbg(PROC, "user wants unit %u\n", prefunit);
+	}
+	if (port)
+		qib_cdbg(PROC, "user wants IB port %u\n", port);
+recheck:
+	for (ndev = prefunit != -1 ? prefunit : 0; ndev < devmax; ndev++) {
+		struct qib_devdata *dd = qib_lookup(ndev);
+
+		/* device portion of usable() */
+		if (!(dd && (dd->flags & QIB_PRESENT) && dd->kregbase))
+			continue;
+		for (ctxt = dd->first_user_ctxt; ctxt < dd->cfgctxts; ctxt++) {
+			if (dd->rcd[ctxt])
+				continue;
+			if (port) {
+				if (port > dd->num_pports) {
+					qib_cdbg(PROC,
+						 "skip invalid unit %u:%u\n",
+						 ndev, port);
+					continue;
+				}
+				ppd = dd->pport + port - 1;
+				if (!usable(ppd, 0))
+					continue;
+			} else {
+				/*
+				 * choose port based on ctxt, if up, else
+				 * first port that's up for multi-port HCA
+				 */
+				ppd = dd->pport + (ctxt % dd->num_pports);
+				if (!usable(ppd, 0)) {
+					int j;
+
+					ppd = NULL;
+					for (j = 0; j < dd->num_pports &&
+						!ppd; j++)
+						if (usable(dd->pport + j, 0))
+							ppd = dd->pport + j;
+					if (!ppd)
+						continue; /* to next unit */
+				}
+			}
+			ret = setup_ctxt(ppd, ctxt, fp, uinfo);
+			goto done;
+		}
+	}
+
+	if (npresent) {
+		if (nup == 0) {
+			ret = -ENETDOWN;
+			qib_dbg("No contexts available (none initialized "
+				"and ready)\n");
+		} else {
+			if (prefunit != -1) {
+				/*
+				 * if had prefunit, but it wasn't available,
+				 * restart, and try all units
+				 */
+				qib_cdbg(PROC, "%s[%u] no contexts on "
+					 "prefunit %d, clear and re-check\n",
+					 current->comm, current->pid,
+					 prefunit);
+				devmax = qib_count_units(NULL, NULL);
+				prefunit = -1;
+				goto recheck;
+			}
+			ret = -EBUSY;
+			qib_dbg("All contexts busy\n");
+		}
+	} else {
+		ret = -ENXIO;
+		qib_dbg("No boards found\n");
+	}
+
+done:
+	return ret;
+}
+
+static int find_shared_ctxt(struct file *fp,
+			    const struct qib_user_info *uinfo)
+{
+	int devmax, ndev, i;
+	int ret = 0;
+
+	devmax = qib_count_units(NULL, NULL);
+
+	for (ndev = 0; ndev < devmax; ndev++) {
+		struct qib_devdata *dd = qib_lookup(ndev);
+
+		/* device portion of usable() */
+		if (!(dd && (dd->flags & QIB_PRESENT) && dd->kregbase))
+			continue;
+		for (i = dd->first_user_ctxt; i < dd->cfgctxts; i++) {
+			struct qib_ctxtdata *rcd = dd->rcd[i];
+
+			/* Skip ctxts which are not yet open */
+			if (!rcd || !rcd->cnt)
+				continue;
+			/* Skip ctxt if it doesn't match the requested one */
+			if (rcd->subctxt_id != uinfo->spu_subctxt_id)
+				continue;
+			/* Verify the sharing process matches the master */
+			if (rcd->subctxt_cnt != uinfo->spu_subctxt_cnt ||
+			    rcd->userversion != uinfo->spu_userversion ||
+			    rcd->cnt >= rcd->subctxt_cnt) {
+				ret = -EINVAL;
+				goto done;
+			}
+			ctxt_fp(fp) = rcd;
+			subctxt_fp(fp) = rcd->cnt++;
+			rcd->subpid[subctxt_fp(fp)] = current->pid;
+			tidcursor_fp(fp) = 0;
+			rcd->active_slaves |= 1 << subctxt_fp(fp);
+			qib_cdbg(PROC,
+				 "%s[%u] %u sharing %s[%u] unit:ctxt %u:%u\n",
+				 current->comm, current->pid, subctxt_fp(fp),
+				 rcd->comm, rcd->pid, dd->unit,
+				 rcd->ctxt);
+			ret = 1;
+			goto done;
+		}
+	}
+
+done:
+	return ret;
+}
+
+static int qib_open(struct inode *in, struct file *fp)
+{
+	/* The real work is performed later in qib_assign_ctxt() */
+	fp->private_data = kzalloc(sizeof(struct qib_filedata), GFP_KERNEL);
+	return fp->private_data ? 0 : -ENOMEM;
+}
+
+/*
+ * Get ctxt early, so can set affinity prior to memory allocation.
+ */
+static int qib_assign_ctxt(struct file *fp, const struct qib_user_info *uinfo)
+{
+	int ret;
+	int i_minor;
+	unsigned swmajor, swminor;
+
+	/* Check to be sure we haven't already initialized this file */
+	if (ctxt_fp(fp)) {
+		ret = -EINVAL;
+		goto done;
+	}
+
+	/* for now, if major version is different, bail */
+	swmajor = uinfo->spu_userversion >> 16;
+	if (swmajor != QIB_USER_SWMAJOR) {
+		qib_dbg("User major version %d not same as driver "
+			"major %d\n", uinfo->spu_userversion >> 16,
+			QIB_USER_SWMAJOR);
+		ret = -ENODEV;
+		goto done;
+	}
+
+	swminor = uinfo->spu_userversion & 0xffff;
+	if (swminor != QIB_USER_SWMINOR)
+		qib_dbg("User minor version %d not same as driver "
+			"minor %d\n", swminor, QIB_USER_SWMINOR);
+
+	mutex_lock(&qib_mutex);
+
+	if (qib_compatible_subctxts(swmajor, swminor) &&
+	    uinfo->spu_subctxt_cnt) {
+		ret = find_shared_ctxt(fp, uinfo);
+		if (ret) {
+			if (ret > 0)
+				ret = 0;
+			goto done_chk_sdma;
+		}
+	}
+
+	i_minor = iminor(fp->f_dentry->d_inode) - QIB_USER_MINOR_BASE;
+	qib_cdbg(VERBOSE, "open on dev %lx (minor %d)\n",
+		 (long)fp->f_dentry->d_inode->i_rdev, i_minor);
+
+	if (i_minor)
+		ret = find_free_ctxt(i_minor - 1, fp, uinfo);
+	else
+		ret = find_best_unit(fp, uinfo);
+
+done_chk_sdma:
+	if (!ret) {
+		struct qib_filedata *fd = fp->private_data;
+		const struct qib_ctxtdata *rcd = fd->rcd;
+		const struct qib_devdata *dd = rcd->dd;
+
+		if (dd->flags & QIB_HAS_SEND_DMA) {
+			fd->pq = qib_user_sdma_queue_create(&dd->pcidev->dev,
+							    dd->unit,
+							    rcd->ctxt,
+							    fd->subctxt);
+			if (!fd->pq)
+				ret = -ENOMEM;
+		}
+	}
+
+	mutex_unlock(&qib_mutex);
+
+done:
+	return ret;
+}
+
+
+static int qib_do_user_init(struct file *fp,
+			    const struct qib_user_info *uinfo)
+{
+	int ret;
+	struct qib_ctxtdata *rcd = ctxt_fp(fp);
+	struct qib_devdata *dd;
+	unsigned uctxt;
+
+	/* Subctxts don't need to initialize anything since master did it. */
+	if (subctxt_fp(fp)) {
+		ret = wait_event_interruptible(rcd->wait,
+			!test_bit(QIB_CTXT_MASTER_UNINIT, &rcd->flag));
+		goto bail;
+	}
+
+	dd = rcd->dd;
+
+	/* some ctxts may get extra buffers, calculate that here */
+	uctxt = rcd->ctxt - dd->first_user_ctxt;
+	if (uctxt < dd->ctxts_extrabuf) {
+		rcd->piocnt = dd->pbufsctxt + 1;
+		rcd->pio_base = rcd->piocnt * uctxt;
+	} else {
+		rcd->piocnt = dd->pbufsctxt;
+		rcd->pio_base = rcd->piocnt * uctxt +
+			dd->ctxts_extrabuf;
+	}
+
+	/*
+	 * All user buffers are 2KB buffers.  If we ever support
+	 * giving 4KB buffers to user processes, this will need some
+	 * work.  Can't use piobufbase directly, because it has
+	 * both 2K and 4K buffer base values.  So check and handle.
+	 */
+	if ((rcd->pio_base + rcd->piocnt) > dd->piobcnt2k) {
+		if (rcd->pio_base >= dd->piobcnt2k) {
+			qib_dev_err(dd,
+				    "%u:ctxt%u: no 2KB buffers available\n",
+				    dd->unit, rcd->ctxt);
+			ret = -ENOBUFS;
+			goto bail;
+		}
+		rcd->piocnt = dd->piobcnt2k - rcd->pio_base;
+		qib_dev_err(dd, "Ctxt%u: would use 4KB bufs, using %u\n",
+			    rcd->ctxt, rcd->piocnt);
+	}
+
+	rcd->piobufs = dd->pio2k_bufbase + rcd->pio_base * dd->palign;
+	qib_chg_pioavailkernel(dd, rcd->pio_base, rcd->piocnt,
+			       TXCHK_CHG_TYPE_USER, rcd);
+	/*
+	 * try to ensure that processes start up with consistent avail update
+	 * for their own range, at least.   If system very quiet, it might
+	 * have the in-memory copy out of date at startup for this range of
+	 * buffers, when a context gets re-used.  Do after the chg_pioavail
+	 * and before the rest of setup, so it's "almost certain" the dma
+	 * will have occurred (can't 100% guarantee, but should be many
+	 * decimals of 9s, with this ordering), given how much else happens
+	 * after this.
+	 */
+	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_AVAIL_BLIP);
+
+	/*
+	 * Now allocate the rcvhdr Q and eager TIDs; skip the TID
+	 * array for time being.  If rcd->ctxt > chip-supported,
+	 * we need to do extra stuff here to handle by handling overflow
+	 * through ctxt 0, someday
+	 */
+	ret = qib_create_rcvhdrq(dd, rcd);
+	if (!ret)
+		ret = qib_setup_eagerbufs(rcd);
+	if (ret)
+		goto bail_pio;
+
+	rcd->tidcursor = 0; /* start at beginning after open */
+
+	/* initialize poll variables... */
+	rcd->urgent = 0;
+	rcd->urgent_poll = 0;
+
+	/*
+	 * Now enable the ctxt for receive.
+	 * For chips that are set to DMA the tail register to memory
+	 * when they change (and when the update bit transitions from
+	 * 0 to 1.  So for those chips, we turn it off and then back on.
+	 * This will (very briefly) affect any other open ctxts, but the
+	 * duration is very short, and therefore isn't an issue.  We
+	 * explictly set the in-memory tail copy to 0 beforehand, so we
+	 * don't have to wait to be sure the DMA update has happened
+	 * (chip resets head/tail to 0 on transition to enable).
+	 */
+	if (rcd->rcvhdrtail_kvaddr)
+		qib_clear_rcvhdrtail(rcd);
+
+	dd->f_rcvctrl(rcd->ppd, QIB_RCVCTRL_CTXT_ENB | QIB_RCVCTRL_TIDFLOW_ENB,
+		      rcd->ctxt);
+
+	/* Notify any waiting slaves */
+	if (rcd->subctxt_cnt) {
+		clear_bit(QIB_CTXT_MASTER_UNINIT, &rcd->flag);
+		wake_up(&rcd->wait);
+	}
+	return 0;
+
+bail_pio:
+	qib_chg_pioavailkernel(dd, rcd->pio_base, rcd->piocnt,
+			       TXCHK_CHG_TYPE_KERN, rcd);
+bail:
+	return ret;
+}
+
+/**
+ * unlock_exptid - unlock any expected TID entries context still had in use
+ * @rcd: ctxt
+ *
+ * We don't actually update the chip here, because we do a bulk update
+ * below, using f_clear_tids.
+ */
+static void unlock_expected_tids(struct qib_ctxtdata *rcd)
+{
+	struct qib_devdata *dd = rcd->dd;
+	int ctxt_tidbase = rcd->ctxt * dd->rcvtidcnt;
+	int i, cnt = 0, maxtid = ctxt_tidbase + dd->rcvtidcnt;
+
+	qib_cdbg(VERBOSE, "Context %u unlocking any locked expTID pages\n",
+		 rcd->ctxt);
+	for (i = ctxt_tidbase; i < maxtid; i++) {
+		struct page *p = dd->pageshadow[i];
+		dma_addr_t phys;
+
+		if (!p)
+			continue;
+
+		phys = dd->physshadow[i];
+		dd->physshadow[i] = dd->tidinvalid;
+		dd->pageshadow[i] = NULL;
+		pci_unmap_page(dd->pcidev, phys, PAGE_SIZE,
+			       PCI_DMA_FROMDEVICE);
+		qib_release_user_pages(&p, 1);
+		cnt++;
+	}
+	if (cnt)
+		qib_cdbg(VERBOSE, "Context %u locked %u expTID entries\n",
+			 rcd->ctxt, cnt);
+}
+
+static int qib_close(struct inode *in, struct file *fp)
+{
+	int ret = 0;
+	struct qib_filedata *fd;
+	struct qib_ctxtdata *rcd;
+	struct qib_devdata *dd;
+	unsigned long flags;
+	unsigned ctxt;
+	pid_t pid;
+
+	qib_cdbg(VERBOSE, "close on dev %lx, private data %p\n",
+		 (long)in->i_rdev, fp->private_data);
+
+	mutex_lock(&qib_mutex);
+
+	fd = (struct qib_filedata *) fp->private_data;
+	fp->private_data = NULL;
+	rcd = fd->rcd;
+	if (!rcd) {
+		mutex_unlock(&qib_mutex);
+		goto bail;
+	}
+
+	dd = rcd->dd;
+
+	/* ensure all pio buffer writes in progress are flushed */
+	qib_flush_wc();
+
+	/* drain user sdma queue */
+	if (fd->pq) {
+		qib_user_sdma_queue_drain(rcd->ppd, fd->pq);
+		qib_user_sdma_queue_destroy(fd->pq);
+	}
+
+	if (--rcd->cnt) {
+		/*
+		 * XXX If the master closes the context before the slave(s),
+		 * revoke the mmap for the eager receive queue so
+		 * the slave(s) don't wait for receive data forever.
+		 */
+		rcd->active_slaves &= ~(1 << fd->subctxt);
+		rcd->subpid[fd->subctxt] = 0;
+		mutex_unlock(&qib_mutex);
+		goto bail;
+	}
+
+	/* early; no interrupt users after this */
+	spin_lock_irqsave(&dd->uctxt_lock, flags);
+	ctxt = rcd->ctxt;
+	dd->rcd[ctxt] = NULL;
+	pid = rcd->pid;
+	rcd->pid = 0;
+	spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+
+	if (rcd->rcvwait_to || rcd->piowait_to ||
+	    rcd->rcvnowait || rcd->pionowait) {
+		qib_cdbg(VERBOSE, "ctxt%u, %u rcv, %u pio wait timeo; "
+			 "%u rcv %u, pio already\n",
+			 rcd->ctxt, rcd->rcvwait_to,
+			 rcd->piowait_to, rcd->rcvnowait,
+			 rcd->pionowait);
+		rcd->rcvwait_to = 0;
+		rcd->piowait_to = 0;
+		rcd->rcvnowait = 0;
+		rcd->pionowait = 0;
+	}
+	if (rcd->flag) {
+		qib_cdbg(PROC, "ctxt %u flag set: 0x%lx\n",
+			 rcd->ctxt, rcd->flag);
+		rcd->flag = 0;
+	}
+
+	if (dd->kregbase) {
+		/* atomically clear receive enable ctxt and intr avail. */
+		dd->f_rcvctrl(rcd->ppd, QIB_RCVCTRL_CTXT_DIS |
+				  QIB_RCVCTRL_INTRAVAIL_DIS, ctxt);
+
+		/* clean up the pkeys for this ctxt user */
+		qib_clean_part_key(rcd, dd);
+		qib_disarm_piobufs(dd, rcd->pio_base, rcd->piocnt);
+		qib_chg_pioavailkernel(dd, rcd->pio_base,
+				       rcd->piocnt, TXCHK_CHG_TYPE_KERN, NULL);
+
+		dd->f_clear_tids(dd, rcd);
+
+		if (dd->pageshadow)
+			unlock_expected_tids(rcd);
+		qib_stats.sps_ctxts--;
+		qib_cdbg(PROC, "%s[%u] closed ctxt %u:%u\n",
+			 rcd->comm, pid, dd->unit, ctxt);
+	}
+
+	mutex_unlock(&qib_mutex);
+	qib_free_ctxtdata(dd, rcd); /* after releasing the mutex */
+
+bail:
+	kfree(fd);
+	return ret;
+}
+
+static int qib_ctxt_info(struct qib_ctxtdata *rcd, u16 subctxt,
+			 struct qib_ctxt_info __user *uinfo)
+{
+	struct qib_ctxt_info info;
+	int ret;
+	size_t sz;
+
+	info.num_active = qib_count_active_units();
+	info.unit = rcd->dd->unit;
+	info.port = rcd->ppd->port;
+	info.ctxt = rcd->ctxt;
+	info.subctxt = subctxt;
+	/* Number of user ctxts available for this device. */
+	info.num_ctxts = rcd->dd->cfgctxts - rcd->dd->first_user_ctxt;
+	info.num_subctxts = rcd->subctxt_cnt;
+	sz = sizeof(info);
+
+	if (copy_to_user(uinfo, &info, sz)) {
+		ret = -EFAULT;
+		goto bail;
+	}
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+static int qib_sdma_get_inflight(struct qib_user_sdma_queue *pq,
+				 u32 __user *inflightp)
+{
+	const u32 val = qib_user_sdma_inflight_counter(pq);
+
+	if (put_user(val, inflightp))
+		return -EFAULT;
+
+	return 0;
+}
+
+static int qib_sdma_get_complete(struct qib_pportdata *ppd,
+				 struct qib_user_sdma_queue *pq,
+				 u32 __user *completep)
+{
+	u32 val;
+	int err;
+
+	if (!pq)
+		return -EINVAL;
+
+	err = qib_user_sdma_make_progress(ppd, pq);
+	if (err < 0)
+		return err;
+
+	val = qib_user_sdma_complete_counter(pq);
+	if (put_user(val, completep))
+		return -EFAULT;
+
+	return 0;
+}
+
+static int disarm_req_delay(struct qib_ctxtdata *rcd)
+{
+	int ret = 0;
+
+	if (!usable(rcd->ppd, 1)) {
+		int i;
+		/*
+		 * if link is down, or otherwise not usable, delay
+		 * the caller up to 30 seconds, so we don't thrash
+		 * in trying to get the chip back to ACTIVE, and
+		 * set flag so they make the call again.
+		 */
+		qib_cdbg(ERRPKT, "Ctxt%u, set disarm needed, "
+			 "link is down\n", rcd->ctxt);
+		if (rcd->user_event_mask) {
+			/*
+			 * subctxt_cnt is 0 if not shared, so do base
+			 * separately, first, then remaining subctxt, if any
+			 */
+			set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+				&rcd->user_event_mask[0]);
+			for (i = 1; i < rcd->subctxt_cnt; i++)
+				set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+					&rcd->user_event_mask[i]);
+		}
+		for (i = 0; !usable(rcd->ppd, 1) && i < 300; i++)
+			msleep(100);
+		ret = -ENETDOWN;
+	}
+	return ret;
+}
+
+/*
+ * Find all user contexts in use, and set the specified bit in their
+ * event mask.
+ * See also find_ctxt() for a similar use, that is specific to send buffers.
+ */
+int qib_set_uevent_bits(struct qib_pportdata *ppd, const int evtbit)
+{
+	struct qib_ctxtdata *rcd;
+	unsigned ctxt;
+	int ret = 0;
+
+	qib_cdbg(PROC, "IB%u:%u set event bit %d\n", ppd->dd->unit, ppd->port,
+		 evtbit);
+	spin_lock(&ppd->dd->uctxt_lock);
+	for (ctxt = ppd->dd->first_user_ctxt; ctxt < ppd->dd->cfgctxts;
+	     ctxt++) {
+		rcd = ppd->dd->rcd[ctxt];
+		if (!rcd)
+			continue;
+		if (rcd->user_event_mask) {
+			int i;
+			/*
+			 * subctxt_cnt is 0 if not shared, so do base
+			 * separately, first, then remaining subctxt, if any
+			 */
+			set_bit(evtbit, &rcd->user_event_mask[0]);
+			qib_cdbg(PROC, "qib%u,ctxt%u, uevent_mask %lx\n",
+				 ppd->dd->unit, rcd->ctxt,
+				rcd->user_event_mask[0]);
+			for (i = 1; i < rcd->subctxt_cnt; i++)
+				set_bit(evtbit, &rcd->user_event_mask[i]);
+		}
+		ret = 1;
+		break;
+	}
+	spin_unlock(&ppd->dd->uctxt_lock);
+
+	return ret;
+}
+
+/*
+ * clear the event notifier events for this context.
+ * For the DISARM_BUFS case, we also take action (this obsoletes
+ * the older QIB_CMD_DISARM_BUFS, but we keep it for backwards
+ * compatibility.
+ * Other bits don't currently require actions, just atomically clear.
+ * User process then performs actions appropriate to bit having been
+ * set, if desired, and checks again in future.
+ */
+static int qib_user_event_ack(struct qib_ctxtdata *rcd, int subctxt,
+			      unsigned long events)
+{
+	int ret = 0, i;
+
+	for (i = 0; i <= _QIB_MAX_EVENT_BIT; i++) {
+		if (!test_bit(i, &events))
+			continue;
+		if (i == _QIB_EVENT_DISARM_BUFS_BIT) {
+			(void)qib_disarm_piobufs_ifneeded(rcd);
+			ret = disarm_req_delay(rcd);
+			qib_cdbg(PROC, "qib%u,ctxt%u, disarmed bufs, "
+				 "uevent_mask %lx\n",
+				 rcd->ppd->dd->unit, rcd->ctxt,
+				 rcd->user_event_mask[subctxt]);
+		} else {
+			clear_bit(i, &rcd->user_event_mask[subctxt]);
+			qib_cdbg(PROC, "qib%u,ctxt%u, cleared uevent %d, "
+				 "uevent_mask now %lx\n",
+				 rcd->ppd->dd->unit, rcd->ctxt, i,
+				 rcd->user_event_mask[subctxt]);
+		}
+	}
+	return ret;
+}
+
+static ssize_t qib_write(struct file *fp, const char __user *data,
+			 size_t count, loff_t *off)
+{
+	const struct qib_cmd __user *ucmd;
+	struct qib_ctxtdata *rcd;
+	const void __user *src;
+	size_t consumed, copy = 0;
+	struct qib_cmd cmd;
+	ssize_t ret = 0;
+	void *dest;
+
+	if (count < sizeof(cmd.type)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	ucmd = (const struct qib_cmd __user *) data;
+
+	if (copy_from_user(&cmd.type, &ucmd->type, sizeof(cmd.type))) {
+		ret = -EFAULT;
+		goto bail;
+	}
+
+	consumed = sizeof(cmd.type);
+
+	switch (cmd.type) {
+	case QIB_CMD_ASSIGN_CTXT:
+	case QIB_CMD_USER_INIT:
+		copy = sizeof(cmd.cmd.user_info);
+		dest = &cmd.cmd.user_info;
+		src = &ucmd->cmd.user_info;
+		break;
+
+	case QIB_CMD_RECV_CTRL:
+		copy = sizeof(cmd.cmd.recv_ctrl);
+		dest = &cmd.cmd.recv_ctrl;
+		src = &ucmd->cmd.recv_ctrl;
+		break;
+
+	case QIB_CMD_CTXT_INFO:
+		copy = sizeof(cmd.cmd.ctxt_info);
+		dest = &cmd.cmd.ctxt_info;
+		src = &ucmd->cmd.ctxt_info;
+		break;
+
+	case QIB_CMD_TID_UPDATE:
+	case QIB_CMD_TID_FREE:
+		copy = sizeof(cmd.cmd.tid_info);
+		dest = &cmd.cmd.tid_info;
+		src = &ucmd->cmd.tid_info;
+		break;
+
+	case QIB_CMD_SET_PART_KEY:
+		copy = sizeof(cmd.cmd.part_key);
+		dest = &cmd.cmd.part_key;
+		src = &ucmd->cmd.part_key;
+		break;
+
+	case QIB_CMD_DISARM_BUFS:
+	case QIB_CMD_PIOAVAILUPD: /* force an update of PIOAvail reg */
+		copy = 0;
+		src = NULL;
+		dest = NULL;
+		break;
+
+	case QIB_CMD_POLL_TYPE:
+		copy = sizeof(cmd.cmd.poll_type);
+		dest = &cmd.cmd.poll_type;
+		src = &ucmd->cmd.poll_type;
+		break;
+
+	case QIB_CMD_ARMLAUNCH_CTRL:
+		copy = sizeof(cmd.cmd.armlaunch_ctrl);
+		dest = &cmd.cmd.armlaunch_ctrl;
+		src = &ucmd->cmd.armlaunch_ctrl;
+		break;
+
+	case QIB_CMD_SDMA_INFLIGHT:
+		copy = sizeof(cmd.cmd.sdma_inflight);
+		dest = &cmd.cmd.sdma_inflight;
+		src = &ucmd->cmd.sdma_inflight;
+		break;
+
+	case QIB_CMD_SDMA_COMPLETE:
+		copy = sizeof(cmd.cmd.sdma_complete);
+		dest = &cmd.cmd.sdma_complete;
+		src = &ucmd->cmd.sdma_complete;
+		break;
+
+	case QIB_CMD_ACK_EVENT:
+		copy = sizeof(cmd.cmd.event_mask);
+		dest = &cmd.cmd.event_mask;
+		src = &ucmd->cmd.event_mask;
+		break;
+
+	default:
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	if (copy) {
+		if ((count - consumed) < copy) {
+			ret = -EINVAL;
+			goto bail;
+		}
+		if (copy_from_user(dest, src, copy)) {
+			ret = -EFAULT;
+			goto bail;
+		}
+		consumed += copy;
+	}
+
+	rcd = ctxt_fp(fp);
+	if (!rcd && cmd.type != QIB_CMD_ASSIGN_CTXT) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	switch (cmd.type) {
+	case QIB_CMD_ASSIGN_CTXT:
+		ret = qib_assign_ctxt(fp, &cmd.cmd.user_info);
+		if (ret)
+			goto bail;
+		break;
+
+	case QIB_CMD_USER_INIT:
+		ret = qib_do_user_init(fp, &cmd.cmd.user_info);
+		if (ret)
+			goto bail;
+		ret = qib_get_base_info(fp, (void __user *) (unsigned long)
+					cmd.cmd.user_info.spu_base_info,
+					cmd.cmd.user_info.spu_base_info_size);
+		break;
+
+	case QIB_CMD_RECV_CTRL:
+		ret = qib_manage_rcvq(rcd, subctxt_fp(fp), cmd.cmd.recv_ctrl);
+		break;
+
+	case QIB_CMD_CTXT_INFO:
+		ret = qib_ctxt_info(rcd, subctxt_fp(fp),
+				    (struct qib_ctxt_info __user *)
+				    (unsigned long) cmd.cmd.ctxt_info);
+		break;
+
+	case QIB_CMD_TID_UPDATE:
+		ret = qib_tid_update(rcd, fp, &cmd.cmd.tid_info);
+		break;
+
+	case QIB_CMD_TID_FREE:
+		ret = qib_tid_free(rcd, subctxt_fp(fp), &cmd.cmd.tid_info);
+		break;
+
+	case QIB_CMD_SET_PART_KEY:
+		ret = qib_set_part_key(rcd, cmd.cmd.part_key);
+		break;
+
+	case QIB_CMD_DISARM_BUFS:
+		(void)qib_disarm_piobufs_ifneeded(rcd);
+		ret = disarm_req_delay(rcd);
+		break;
+
+	case QIB_CMD_PIOAVAILUPD:
+		qib_force_pio_avail_update(rcd->dd);
+		break;
+
+	case QIB_CMD_POLL_TYPE:
+		rcd->poll_type = cmd.cmd.poll_type;
+		break;
+
+	case QIB_CMD_ARMLAUNCH_CTRL:
+		rcd->dd->f_set_armlaunch(rcd->dd, cmd.cmd.armlaunch_ctrl);
+		break;
+
+	case QIB_CMD_SDMA_INFLIGHT:
+		ret = qib_sdma_get_inflight(user_sdma_queue_fp(fp),
+					    (u32 __user *) (unsigned long)
+					    cmd.cmd.sdma_inflight);
+		break;
+
+	case QIB_CMD_SDMA_COMPLETE:
+		ret = qib_sdma_get_complete(rcd->ppd,
+					    user_sdma_queue_fp(fp),
+					    (u32 __user *) (unsigned long)
+					    cmd.cmd.sdma_complete);
+		break;
+
+	case QIB_CMD_ACK_EVENT:
+		ret = qib_user_event_ack(rcd, subctxt_fp(fp),
+					 cmd.cmd.event_mask);
+		break;
+	}
+
+	if (ret >= 0)
+		ret = consumed;
+
+bail:
+	return ret;
+}
+
+static ssize_t qib_aio_write(struct kiocb *iocb, const struct iovec *iov,
+			     unsigned long dim, loff_t off)
+{
+	struct qib_filedata *fp = iocb->ki_filp->private_data;
+	struct qib_ctxtdata *rcd = ctxt_fp(iocb->ki_filp);
+	struct qib_user_sdma_queue *pq = fp->pq;
+
+	if (!dim || !pq)
+		return -EINVAL;
+
+	return qib_user_sdma_writev(rcd, pq, iov, dim);
+}
+
+static struct class *qib_class;
+static dev_t qib_dev;
+
+int qib_cdev_init(int minor, const char *name,
+		  const struct file_operations *fops,
+		  struct cdev **cdevp, struct device **devp)
+{
+	const dev_t dev = MKDEV(MAJOR(qib_dev), minor);
+	struct cdev *cdev;
+	struct device *device = NULL;
+	int ret;
+
+	cdev = cdev_alloc();
+	if (!cdev) {
+		printk(KERN_ERR QIB_DRV_NAME
+		       ": Could not allocate cdev for minor %d, %s\n",
+		       minor, name);
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	cdev->owner = THIS_MODULE;
+	cdev->ops = fops;
+	kobject_set_name(&cdev->kobj, name);
+
+	ret = cdev_add(cdev, dev, 1);
+	if (ret < 0) {
+		printk(KERN_ERR QIB_DRV_NAME
+		       ": Could not add cdev for minor %d, %s (err %d)\n",
+		       minor, name, -ret);
+		goto err_cdev;
+	}
+
+	device = device_create(qib_class, NULL, dev, NULL, name);
+	if (!IS_ERR(device))
+		goto done;
+	ret = PTR_ERR(device);
+	device = NULL;
+	printk(KERN_ERR QIB_DRV_NAME ": Could not create "
+	       "device for minor %d, %s (err %d)\n",
+	       minor, name, -ret);
+err_cdev:
+	cdev_del(cdev);
+	cdev = NULL;
+done:
+	*cdevp = cdev;
+	*devp = device;
+	return ret;
+}
+
+void qib_cdev_cleanup(struct cdev **cdevp, struct device **devp)
+{
+	struct device *device = *devp;
+
+	if (device) {
+		device_unregister(device);
+		*devp = NULL;
+	}
+
+	if (*cdevp) {
+		cdev_del(*cdevp);
+		*cdevp = NULL;
+	}
+}
+
+static struct cdev *wildcard_cdev;
+static struct device *wildcard_device;
+
+int __init qib_dev_init(void)
+{
+	int ret;
+
+	ret = alloc_chrdev_region(&qib_dev, 0, QIB_NMINORS, QIB_DRV_NAME);
+	if (ret < 0) {
+		printk(KERN_ERR QIB_DRV_NAME ": Could not allocate "
+		       "chrdev region (err %d)\n", -ret);
+		goto done;
+	}
+
+	qib_class = class_create(THIS_MODULE, QIB_DRV_NAME);
+	if (IS_ERR(qib_class)) {
+		ret = PTR_ERR(qib_class);
+		printk(KERN_ERR QIB_DRV_NAME ": Could not create "
+		       "device class (err %d)\n", -ret);
+		unregister_chrdev_region(qib_dev, QIB_NMINORS);
+	}
+
+done:
+	return ret;
+}
+
+void __exit qib_dev_cleanup(void)
+{
+	if (qib_class) {
+		class_destroy(qib_class);
+		qib_class = NULL;
+	}
+
+	unregister_chrdev_region(qib_dev, QIB_NMINORS);
+}
+
+static atomic_t user_count = ATOMIC_INIT(0);
+
+static void qib_user_remove(struct qib_devdata *dd)
+{
+	if (atomic_dec_return(&user_count) == 0)
+		qib_cdev_cleanup(&wildcard_cdev, &wildcard_device);
+
+	qib_cdev_cleanup(&dd->user_cdev, &dd->user_device);
+}
+
+static int qib_user_add(struct qib_devdata *dd)
+{
+	char name[10];
+	int ret;
+
+	if (atomic_inc_return(&user_count) == 1) {
+		ret = qib_cdev_init(0, "ipath", &qib_file_ops,
+				    &wildcard_cdev, &wildcard_device);
+		if (ret)
+			goto done;
+	}
+
+	snprintf(name, sizeof(name), "ipath%d", dd->unit);
+	ret = qib_cdev_init(dd->unit + 1, name, &qib_file_ops,
+			    &dd->user_cdev, &dd->user_device);
+	if (ret)
+		qib_user_remove(dd);
+done:
+	return ret;
+}
+
+/*
+ * Create per-unit files in /dev
+ */
+int qib_device_create(struct qib_devdata *dd)
+{
+	int r, ret;
+
+	r = qib_user_add(dd);
+	ret = qib_diag_add(dd);
+	if (r && !ret)
+		ret = r;
+	return ret;
+}
+
+/*
+ * Remove per-unit files in /dev
+ * void, core kernel returns no errors for this stuff
+ */
+void qib_device_remove(struct qib_devdata *dd)
+{
+	qib_user_remove(dd);
+	qib_diag_remove(dd);
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 16/51] IB/qib: Add qib_fs.c

[Ralph Campbell]

creates the qib_fs.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_fs.c |  549 ++++++++++++++++++++++++++++++++++++
 1 files changed, 549 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_fs.c

diff --git a/drivers/infiniband/hw/qib/qib_fs.c b/drivers/infiniband/hw/qib/qib_fs.c
new file mode 100644
index 0000000..629ff7e
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_fs.c
@@ -0,0 +1,549 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/mount.h>
+#include <linux/pagemap.h>
+#include <linux/init.h>
+#include <linux/namei.h>
+
+#include "qib.h"
+
+#define QIBFS_MAGIC 0x726a77
+
+static struct super_block *qib_super;
+
+#define private2dd(file) ((file)->f_dentry->d_inode->i_private)
+
+static int qibfs_mknod(struct inode *dir, struct dentry *dentry,
+		       int mode, const struct file_operations *fops,
+		       void *data)
+{
+	int error;
+	struct inode *inode = new_inode(dir->i_sb);
+
+	if (!inode) {
+		error = -EPERM;
+		goto bail;
+	}
+
+	inode->i_mode = mode;
+	inode->i_uid = 0;
+	inode->i_gid = 0;
+	inode->i_blocks = 0;
+	inode->i_atime = CURRENT_TIME;
+	inode->i_mtime = inode->i_atime;
+	inode->i_ctime = inode->i_atime;
+	inode->i_private = data;
+	if ((mode & S_IFMT) == S_IFDIR) {
+		inode->i_op = &simple_dir_inode_operations;
+		inc_nlink(inode);
+		inc_nlink(dir);
+	}
+
+	inode->i_fop = fops;
+
+	d_instantiate(dentry, inode);
+	error = 0;
+
+bail:
+	return error;
+}
+
+static int create_file(const char *name, mode_t mode,
+		       struct dentry *parent, struct dentry **dentry,
+		       const struct file_operations *fops, void *data)
+{
+	int error;
+
+	*dentry = NULL;
+	mutex_lock(&parent->d_inode->i_mutex);
+	*dentry = lookup_one_len(name, parent, strlen(name));
+	if (!IS_ERR(*dentry))
+		error = qibfs_mknod(parent->d_inode, *dentry,
+				    mode, fops, data);
+	else
+		error = PTR_ERR(*dentry);
+	mutex_unlock(&parent->d_inode->i_mutex);
+
+	return error;
+}
+
+static ssize_t driver_stats_read(struct file *file, char __user *buf,
+				 size_t count, loff_t *ppos)
+{
+	return simple_read_from_buffer(buf, count, ppos, &qib_stats,
+				       sizeof qib_stats);
+}
+
+/*
+ * driver stats field names, one line per stat, single string.  Used by
+ * programs like ipathstats to print the stats in a way which works for
+ * different versions of drivers, without changing program source.
+ * if qlogic_ib_stats changes, this needs to change.  Names need to be
+ * 12 chars or less (w/o newline), for proper display by ipathstats utility.
+ */
+static const char qib_statnames[] =
+	"KernIntr\n"
+	"ErrorIntr\n"
+	"Tx_Errs\n"
+	"Rcv_Errs\n"
+	"H/W_Errs\n"
+	"NoPIOBufs\n"
+	"CtxtsOpen\n"
+	"RcvLen_Errs\n"
+	"EgrBufFull\n"
+	"EgrHdrFull\n"
+	;
+
+static ssize_t driver_names_read(struct file *file, char __user *buf,
+				 size_t count, loff_t *ppos)
+{
+	return simple_read_from_buffer(buf, count, ppos, qib_statnames,
+		sizeof qib_statnames - 1); /* no null */
+}
+
+static const struct file_operations driver_ops[] = {
+	{ .read = driver_stats_read, },
+	{ .read = driver_names_read, },
+};
+
+/* read the per-device counters */
+static ssize_t dev_counters_read(struct file *file, char __user *buf,
+				 size_t count, loff_t *ppos)
+{
+	u64 *counters;
+	struct qib_devdata *dd = private2dd(file);
+
+	return simple_read_from_buffer(buf, count, ppos, counters,
+		dd->f_read_cntrs(dd, *ppos, NULL, &counters));
+}
+
+/* read the per-device counters */
+static ssize_t dev_names_read(struct file *file, char __user *buf,
+			      size_t count, loff_t *ppos)
+{
+	char *names;
+	struct qib_devdata *dd = private2dd(file);
+
+	return simple_read_from_buffer(buf, count, ppos, names,
+		dd->f_read_cntrs(dd, *ppos, &names, NULL));
+}
+
+static const struct file_operations cntr_ops[] = {
+	{ .read = dev_counters_read, },
+	{ .read = dev_names_read, },
+};
+
+/*
+ * Could use file->f_dentry->d_inode->i_ino to figure out which file,
+ * instead of separate routine for each, but for now, this works...
+ */
+
+/* read the per-port names (same for each port) */
+static ssize_t portnames_read(struct file *file, char __user *buf,
+			      size_t count, loff_t *ppos)
+{
+	char *names;
+	struct qib_devdata *dd = private2dd(file);
+
+	return simple_read_from_buffer(buf, count, ppos, names,
+		dd->f_read_portcntrs(dd, *ppos, 0, &names, NULL));
+}
+
+/* read the per-port counters for port 1 (pidx 0) */
+static ssize_t portcntrs_1_read(struct file *file, char __user *buf,
+				size_t count, loff_t *ppos)
+{
+	u64 *counters;
+	struct qib_devdata *dd = private2dd(file);
+
+	return simple_read_from_buffer(buf, count, ppos, counters,
+		dd->f_read_portcntrs(dd, *ppos, 0, NULL, &counters));
+}
+
+/* read the per-port counters for port 2 (pidx 1) */
+static ssize_t portcntrs_2_read(struct file *file, char __user *buf,
+				size_t count, loff_t *ppos)
+{
+	u64 *counters;
+	struct qib_devdata *dd = private2dd(file);
+
+	return simple_read_from_buffer(buf, count, ppos, counters,
+		dd->f_read_portcntrs(dd, *ppos, 1, NULL, &counters));
+}
+
+static const struct file_operations portcntr_ops[] = {
+	{ .read = portnames_read, },
+	{ .read = portcntrs_1_read, },
+	{ .read = portcntrs_2_read, },
+};
+
+static ssize_t flash_read(struct file *file, char __user *buf,
+			  size_t count, loff_t *ppos)
+{
+	struct qib_devdata *dd;
+	ssize_t ret;
+	loff_t pos;
+	char *tmp;
+
+	pos = *ppos;
+
+	if (pos < 0) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	if (pos >= sizeof(struct qib_flash)) {
+		ret = 0;
+		goto bail;
+	}
+
+	if (count > sizeof(struct qib_flash) - pos)
+		count = sizeof(struct qib_flash) - pos;
+
+	tmp = kmalloc(count, GFP_KERNEL);
+	if (!tmp) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	dd = private2dd(file);
+	if (qib_eeprom_read(dd, pos, tmp, count)) {
+		qib_dev_err(dd, "failed to read from flash\n");
+		ret = -ENXIO;
+		goto bail_tmp;
+	}
+
+	if (copy_to_user(buf, tmp, count)) {
+		ret = -EFAULT;
+		goto bail_tmp;
+	}
+
+	*ppos = pos + count;
+	ret = count;
+
+bail_tmp:
+	kfree(tmp);
+
+bail:
+	return ret;
+}
+
+static ssize_t flash_write(struct file *file, const char __user *buf,
+			   size_t count, loff_t *ppos)
+{
+	struct qib_devdata *dd;
+	ssize_t ret;
+	loff_t pos;
+	char *tmp;
+
+	pos = *ppos;
+
+	if (pos != 0) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	if (count != sizeof(struct qib_flash)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	tmp = kmalloc(count, GFP_KERNEL);
+	if (!tmp) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	if (copy_from_user(tmp, buf, count)) {
+		ret = -EFAULT;
+		goto bail_tmp;
+	}
+
+	dd = private2dd(file);
+	if (qib_eeprom_write(dd, pos, tmp, count)) {
+		ret = -ENXIO;
+		qib_dev_err(dd, "failed to write to flash\n");
+		goto bail_tmp;
+	}
+
+	*ppos = pos + count;
+	ret = count;
+
+bail_tmp:
+	kfree(tmp);
+
+bail:
+	return ret;
+}
+
+static const struct file_operations flash_ops = {
+	.read = flash_read,
+	.write = flash_write,
+};
+
+static int add_cntr_files(struct super_block *sb, struct qib_devdata *dd)
+{
+	struct dentry *dir, *tmp;
+	char unit[10];
+	int ret, i;
+
+	/* create the per-unit directory */
+	snprintf(unit, sizeof unit, "%u", dd->unit);
+	ret = create_file(unit, S_IFDIR|S_IRUGO|S_IXUGO, sb->s_root, &dir,
+			  &simple_dir_operations, dd);
+	if (ret) {
+		printk(KERN_ERR "create_file(%s) failed: %d\n", unit, ret);
+		goto bail;
+	}
+
+	/* create the files in the new directory */
+	ret = create_file("counters", S_IFREG|S_IRUGO, dir, &tmp,
+			  &cntr_ops[0], dd);
+	if (ret) {
+		printk(KERN_ERR "create_file(%s/counters) failed: %d\n",
+		       unit, ret);
+		goto bail;
+	}
+	ret = create_file("counter_names", S_IFREG|S_IRUGO, dir, &tmp,
+			  &cntr_ops[1], dd);
+	if (ret) {
+		printk(KERN_ERR "create_file(%s/counter_names) failed: %d\n",
+		       unit, ret);
+		goto bail;
+	}
+	ret = create_file("portcounter_names", S_IFREG|S_IRUGO, dir, &tmp,
+			  &portcntr_ops[0], dd);
+	if (ret) {
+		printk(KERN_ERR "create_file(%s/%s) failed: %d\n",
+		       unit, "portcounter_names", ret);
+		goto bail;
+	}
+	for (i = 1; i <= dd->num_pports; i++) {
+		char portcntr[24];
+
+		sprintf(portcntr, "port%dcounters", i);
+		/* create the files in the new directory */
+		ret = create_file(portcntr, S_IFREG|S_IRUGO, dir, &tmp,
+				  &portcntr_ops[i], dd);
+		if (ret) {
+			printk(KERN_ERR "create_file(%s/%s) failed: %d\n",
+				unit, portcntr, ret);
+			goto bail;
+		}
+	}
+
+	ret = create_file("flash", S_IFREG|S_IWUSR|S_IRUGO, dir, &tmp,
+			  &flash_ops, dd);
+	if (ret)
+		printk(KERN_ERR "create_file(%s/flash) failed: %d\n",
+			unit, ret);
+bail:
+	return ret;
+}
+
+static int remove_file(struct dentry *parent, char *name)
+{
+	struct dentry *tmp;
+	int ret;
+
+	tmp = lookup_one_len(name, parent, strlen(name));
+
+	if (IS_ERR(tmp)) {
+		ret = PTR_ERR(tmp);
+		goto bail;
+	}
+
+	spin_lock(&dcache_lock);
+	spin_lock(&tmp->d_lock);
+	if (!(d_unhashed(tmp) && tmp->d_inode)) {
+		dget_locked(tmp);
+		__d_drop(tmp);
+		spin_unlock(&tmp->d_lock);
+		spin_unlock(&dcache_lock);
+		simple_unlink(parent->d_inode, tmp);
+	} else {
+		spin_unlock(&tmp->d_lock);
+		spin_unlock(&dcache_lock);
+	}
+
+	ret = 0;
+bail:
+	/*
+	 * We don't expect clients to care about the return value, but
+	 * it's there if they need it.
+	 */
+	return ret;
+}
+
+static int remove_device_files(struct super_block *sb,
+			       struct qib_devdata *dd)
+{
+	struct dentry *dir, *root;
+	char unit[10];
+	int ret, i;
+
+	root = dget(sb->s_root);
+	mutex_lock(&root->d_inode->i_mutex);
+	snprintf(unit, sizeof unit, "%u", dd->unit);
+	dir = lookup_one_len(unit, root, strlen(unit));
+
+	if (IS_ERR(dir)) {
+		ret = PTR_ERR(dir);
+		printk(KERN_ERR "Lookup of %s failed\n", unit);
+		goto bail;
+	}
+
+	remove_file(dir, "counters");
+	remove_file(dir, "counter_names");
+	remove_file(dir, "portcounter_names");
+	for (i = 0; i < dd->num_pports; i++) {
+		char portcntr[24];
+
+		sprintf(portcntr, "port%dcounters", i + 1);
+		remove_file(dir, portcntr);
+	}
+	remove_file(dir, "flash");
+	d_delete(dir);
+	ret = simple_rmdir(root->d_inode, dir);
+
+bail:
+	mutex_unlock(&root->d_inode->i_mutex);
+	dput(root);
+	return ret;
+}
+
+/*
+ * This fills everything in when the fs is mounted, to handle umount/mount
+ * after device init.  The direct add_cntr_files() call handles adding
+ * them from the init code, when the fs is already mounted.
+ */
+static int qibfs_fill_super(struct super_block *sb, void *data, int silent)
+{
+	struct qib_devdata *dd, *tmp;
+	unsigned long flags;
+	int ret;
+
+	static struct tree_descr files[] = {
+		[2] = {"driver_stats", &driver_ops[0], S_IRUGO},
+		[3] = {"driver_stats_names", &driver_ops[1], S_IRUGO},
+		{""},
+	};
+
+	ret = simple_fill_super(sb, QIBFS_MAGIC, files);
+	if (ret) {
+		printk(KERN_ERR "simple_fill_super failed: %d\n", ret);
+		goto bail;
+	}
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+
+	list_for_each_entry_safe(dd, tmp, &qib_dev_list, list) {
+		spin_unlock_irqrestore(&qib_devs_lock, flags);
+		ret = add_cntr_files(sb, dd);
+		if (ret) {
+			deactivate_super(sb);
+			goto bail;
+		}
+		spin_lock_irqsave(&qib_devs_lock, flags);
+	}
+
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+
+bail:
+	return ret;
+}
+
+static int qibfs_get_sb(struct file_system_type *fs_type, int flags,
+			const char *dev_name, void *data, struct vfsmount *mnt)
+{
+	int ret = get_sb_single(fs_type, flags, data,
+				qibfs_fill_super, mnt);
+	if (ret >= 0)
+		qib_super = mnt->mnt_sb;
+	return ret;
+}
+
+static void qibfs_kill_super(struct super_block *s)
+{
+	kill_litter_super(s);
+	qib_super = NULL;
+}
+
+int qibfs_add(struct qib_devdata *dd)
+{
+	int ret;
+
+	/*
+	 * On first unit initialized, qib_super will not yet exist
+	 * because nobody has yet tried to mount the filesystem, so
+	 * we can't consider that to be an error; if an error occurs
+	 * during the mount, that will get a complaint, so this is OK.
+	 * add_cntr_files() for all units is done at mount from
+	 * qibfs_fill_super(), so one way or another, everything works.
+	 */
+	if (qib_super == NULL)
+		ret = 0;
+	else
+		ret = add_cntr_files(qib_super, dd);
+	return ret;
+}
+
+int qibfs_remove(struct qib_devdata *dd)
+{
+	int ret = 0;
+
+	if (qib_super)
+		ret = remove_device_files(qib_super, dd);
+
+	return ret;
+}
+
+static struct file_system_type qibfs_fs_type = {
+	.owner =        THIS_MODULE,
+	.name =         "ipathfs",
+	.get_sb =       qibfs_get_sb,
+	.kill_sb =      qibfs_kill_super,
+};
+
+int __init qib_init_qibfs(void)
+{
+	return register_filesystem(&qibfs_fs_type);
+}
+
+int __exit qib_exit_qibfs(void)
+{
+	return unregister_filesystem(&qibfs_fs_type);
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 20/51] IB/qib: Add qib_init.c

[Ralph Campbell]

creates the qib_init.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_init.c | 1618 ++++++++++++++++++++++++++++++++++
 1 files changed, 1618 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_init.c

diff --git a/drivers/infiniband/hw/qib/qib_init.c b/drivers/infiniband/hw/qib/qib_init.c
new file mode 100644
index 0000000..6ce1128
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_init.c
@@ -0,0 +1,1618 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/pci.h>
+#include <linux/netdevice.h>
+#include <linux/vmalloc.h>
+#include <linux/delay.h>
+#include <linux/idr.h>
+
+#include "qib.h"
+#include "qib_common.h"
+
+/*
+ * min buffers we want to have per context, after driver
+ */
+#define QIB_MIN_USER_CTXT_BUFCNT 7
+
+#define QLOGIC_IB_R_SOFTWARE_MASK 0xFF
+#define QLOGIC_IB_R_SOFTWARE_SHIFT 24
+#define QLOGIC_IB_R_EMULATOR_MASK (1ULL<<62)
+
+/*
+ * Number of ctxts we are configured to use (to allow for more pio
+ * buffers per ctxt, etc.)  Zero means use chip value.
+ */
+ushort qib_cfgctxts;
+module_param_named(cfgctxts, qib_cfgctxts, ushort, S_IRUGO);
+MODULE_PARM_DESC(cfgctxts, "Set max number of contexts to use");
+
+/*
+ * If set, do not write to any regs if avoidable, hack to allow
+ * check for deranged default register values.
+ */
+ushort qib_mini_init;
+module_param_named(mini_init, qib_mini_init, ushort, S_IRUGO);
+MODULE_PARM_DESC(mini_init, "If set, do minimal diag init");
+
+unsigned qib_n_krcv_queues;
+module_param_named(krcvqs, qib_n_krcv_queues, uint, S_IRUGO);
+MODULE_PARM_DESC(krcvqs, "number of kernel receive queues per IB port");
+
+/*
+ * qib_wc_pat parameter:
+ *      0 is WC via MTRR
+ *      1 is WC via PAT
+ *      If PAT initialization fails, code reverts back to MTRR
+ */
+unsigned qib_wc_pat = 1; /* default (1) is to use PAT, not MTRR */
+module_param_named(wc_pat, qib_wc_pat, uint, S_IRUGO);
+MODULE_PARM_DESC(wc_pat, "enable write-combining via PAT mechanism");
+
+static void verify_interrupt(unsigned long);
+
+static struct idr qib_unit_table;
+
+/* set number of contexts we'll actually use */
+void qib_set_ctxtcnt(struct qib_devdata *dd)
+{
+	if (!qib_cfgctxts)
+		dd->cfgctxts = dd->ctxtcnt;
+	else if (qib_cfgctxts < dd->num_pports) {
+		dd->cfgctxts = dd->ctxtcnt;
+		qib_dbg("Configured to use too few ctxts (%u); using %u\n",
+			qib_cfgctxts, dd->cfgctxts);
+	} else if (qib_cfgctxts <= dd->ctxtcnt) {
+		dd->cfgctxts = qib_cfgctxts;
+		qib_cdbg(INIT, "Configured to use %u ctxts\n", dd->cfgctxts);
+	} else {
+		dd->cfgctxts = dd->ctxtcnt;
+		qib_dbg("Configured to use too many ctxts (%u); using %u\n",
+			qib_cfgctxts, dd->cfgctxts);
+	}
+}
+
+/*
+ * Common code for creating the receive context array.
+ */
+int qib_create_ctxts(struct qib_devdata *dd)
+{
+	unsigned i;
+	int ret;
+
+	/*
+	 * Allocate full ctxtcnt array, rather than just cfgctxts, because
+	 * cleanup iterates across all possible ctxts.
+	 */
+	dd->rcd = kzalloc(sizeof(*dd->rcd) * dd->ctxtcnt, GFP_KERNEL);
+	if (!dd->rcd) {
+		qib_dev_err(dd, "Unable to allocate ctxtdata array, "
+			    "failing\n");
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	/* create (one or more) kctxt */
+	for (i = 0; i < dd->first_user_ctxt; ++i) {
+		struct qib_pportdata *ppd;
+		struct qib_ctxtdata *rcd;
+
+		if (dd->skip_kctxt_mask & (1 << i))
+			continue;
+
+		ppd = dd->pport + (i % dd->num_pports);
+		rcd = qib_create_ctxtdata(ppd, i);
+		if (!rcd) {
+			qib_dev_err(dd, "Unable to allocate ctxtdata"
+				    " for Kernel ctxt, failing\n");
+			ret = -ENOMEM;
+			goto done;
+		}
+		rcd->pkeys[0] = QIB_DEFAULT_P_KEY;
+		rcd->seq_cnt = 1;
+	}
+	ret = 0;
+done:
+	return ret;
+}
+
+/*
+ * Common code for user and kernel context setup.
+ */
+struct qib_ctxtdata *qib_create_ctxtdata(struct qib_pportdata *ppd, u32 ctxt)
+{
+	struct qib_devdata *dd = ppd->dd;
+	struct qib_ctxtdata *rcd;
+
+	rcd = kzalloc(sizeof(*rcd), GFP_KERNEL);
+	if (rcd) {
+		INIT_LIST_HEAD(&rcd->qp_wait_list);
+		rcd->ppd = ppd;
+		rcd->dd = dd;
+		rcd->cnt = 1;
+		rcd->ctxt = ctxt;
+		dd->rcd[ctxt] = rcd;
+
+		dd->f_init_ctxt(rcd);
+
+		/*
+		 * To avoid wasting a lot of memory, we allocate 32KB chunks
+		 * of physically contiguous memory, advance through it until
+		 * used up and then allocate more.  Of course, we need
+		 * memory to store those extra pointers, now.  32KB seems to
+		 * be the most that is "safe" under memory pressure
+		 * (creating large files and then copying them over
+		 * NFS while doing lots of MPI jobs).  The OOM killer can
+		 * get invoked, even though we say we can sleep and this can
+		 * cause significant system problems....
+		 */
+		rcd->rcvegrbuf_size = 0x8000;
+		rcd->rcvegrbufs_perchunk =
+			rcd->rcvegrbuf_size / dd->rcvegrbufsize;
+		rcd->rcvegrbuf_chunks = (rcd->rcvegrcnt +
+			rcd->rcvegrbufs_perchunk - 1) /
+			rcd->rcvegrbufs_perchunk;
+	}
+	return rcd;
+}
+
+/*
+ * Common code for initializing the physical port structure.
+ */
+void qib_init_pportdata(struct qib_pportdata *ppd, struct qib_devdata *dd,
+			u8 hw_pidx, u8 port)
+{
+	ppd->dd = dd;
+	ppd->hw_pidx = hw_pidx;
+	ppd->port = port; /* IB port number, not index */
+
+	spin_lock_init(&ppd->sdma_lock);
+	spin_lock_init(&ppd->lflags_lock);
+	init_waitqueue_head(&ppd->state_wait);
+}
+
+static int init_pioavailregs(struct qib_devdata *dd)
+{
+	int ret, pidx;
+	u64 *status_page;
+
+	dd->pioavailregs_dma = dma_alloc_coherent(
+		&dd->pcidev->dev, PAGE_SIZE, &dd->pioavailregs_phys,
+		GFP_KERNEL);
+	if (!dd->pioavailregs_dma) {
+		qib_dev_err(dd, "failed to allocate PIOavail reg area "
+			    "in memory\n");
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	/*
+	 * We really want L2 cache aligned, but for current CPUs of
+	 * interest, they are the same.
+	 */
+	status_page = (u64 *)
+		((char *) dd->pioavailregs_dma +
+		 ((2 * L1_CACHE_BYTES +
+		   dd->pioavregs * sizeof(u64)) & ~L1_CACHE_BYTES));
+	/* device status comes first, for backwards compatibility */
+	dd->devstatusp = status_page;
+	*status_page++ = 0;
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		dd->pport[pidx].statusp = status_page;
+		*status_page++ = 0;
+	}
+
+	/*
+	 * Setup buffer to hold freeze and other messages, accessible to
+	 * apps, following statusp.  This is per-unit, not per port.
+	 */
+	dd->freezemsg = (char *) status_page;
+	*dd->freezemsg = 0;
+	/* length of msg buffer is "whatever is left" */
+	ret = (char *) status_page - (char *) dd->pioavailregs_dma;
+	dd->freezelen = PAGE_SIZE - ret;
+
+	ret = 0;
+
+done:
+	return ret;
+}
+
+/**
+ * init_shadow_tids - allocate the shadow TID array
+ * @dd: the qlogic_ib device
+ *
+ * allocate the shadow TID array, so we can qib_munlock previous
+ * entries.  It may make more sense to move the pageshadow to the
+ * ctxt data structure, so we only allocate memory for ctxts actually
+ * in use, since we at 8k per ctxt, now.
+ * We don't want failures here to prevent use of the driver/chip,
+ * so no return value.
+ */
+static void init_shadow_tids(struct qib_devdata *dd)
+{
+	struct page **pages;
+	dma_addr_t *addrs;
+
+	pages = vmalloc(dd->cfgctxts * dd->rcvtidcnt * sizeof(struct page *));
+	if (!pages) {
+		qib_dev_err(dd, "failed to allocate shadow page * "
+			    "array, no expected sends!\n");
+		goto bail;
+	}
+
+	addrs = vmalloc(dd->cfgctxts * dd->rcvtidcnt * sizeof(dma_addr_t));
+	if (!addrs) {
+		qib_dev_err(dd, "failed to allocate shadow dma handle "
+			    "array, no expected sends!\n");
+		goto bail_free;
+	}
+
+	memset(pages, 0, dd->cfgctxts * dd->rcvtidcnt * sizeof(struct page *));
+	memset(addrs, 0, dd->cfgctxts * dd->rcvtidcnt * sizeof(dma_addr_t));
+
+	dd->pageshadow = pages;
+	dd->physshadow = addrs;
+	return;
+
+bail_free:
+	vfree(pages);
+bail:
+	dd->pageshadow = NULL;
+}
+
+/*
+ * Do initialization for device that is only needed on
+ * first detect, not on resets.
+ */
+static int loadtime_init(struct qib_devdata *dd)
+{
+	int ret = 0;
+
+	qib_cdbg(VERBOSE, "Revision %Lx\n", dd->revision);
+
+	if (((dd->revision >> QLOGIC_IB_R_SOFTWARE_SHIFT) &
+	     QLOGIC_IB_R_SOFTWARE_MASK) != QIB_CHIP_SWVERSION) {
+		qib_dev_err(dd, "Driver only handles version %d, "
+			    "chip swversion is %d (%llx), failng\n",
+			    QIB_CHIP_SWVERSION,
+			    (int)(dd->revision >>
+				QLOGIC_IB_R_SOFTWARE_SHIFT) &
+			    QLOGIC_IB_R_SOFTWARE_MASK,
+			    (unsigned long long) dd->revision);
+		ret = -ENOSYS;
+		goto done;
+	}
+
+	if (dd->revision & QLOGIC_IB_R_EMULATOR_MASK)
+		qib_devinfo(dd->pcidev, "%s", dd->boardversion);
+	else
+		qib_cdbg(INIT, "%s", dd->boardversion);
+
+	spin_lock_init(&dd->pioavail_lock);
+	spin_lock_init(&dd->sendctrl_lock);
+	spin_lock_init(&dd->uctxt_lock);
+	spin_lock_init(&dd->qib_diag_trans_lock);
+	spin_lock_init(&dd->eep_st_lock);
+	mutex_init(&dd->eep_lock);
+
+	if (qib_mini_init)
+		goto done;
+
+	ret = init_pioavailregs(dd);
+	init_shadow_tids(dd);
+
+	qib_get_eeprom_info(dd);
+
+	/* setup time (don't start yet) to verify we got interrupt */
+	init_timer(&dd->intrchk_timer);
+	dd->intrchk_timer.function = verify_interrupt;
+	dd->intrchk_timer.data = (unsigned long) dd;
+
+done:
+	return ret;
+}
+
+/**
+ * init_after_reset - re-initialize after a reset
+ * @dd: the qlogic_ib device
+ *
+ * sanity check at least some of the values after reset, and
+ * ensure no receive or transmit (explictly, in case reset
+ * failed
+ */
+static int init_after_reset(struct qib_devdata *dd)
+{
+	int i;
+
+	/*
+	 * Ensure chip does no sends or receives, tail updates, or
+	 * pioavail updates while we re-initialize.  This is mostly
+	 * for the driver data structures, not chip registers.
+	 */
+	for (i = 0; i < dd->num_pports; ++i) {
+		/*
+		 * ctxt == -1 means "all contexts". Only really safe for
+		 * _dis_abling things, as here.
+		 */
+		dd->f_rcvctrl(dd->pport + i, QIB_RCVCTRL_CTXT_DIS |
+				  QIB_RCVCTRL_INTRAVAIL_DIS |
+				  QIB_RCVCTRL_TAILUPD_DIS, -1);
+		/* Redundant across ports for some, but no big deal.  */
+		dd->f_sendctrl(dd->pport + i, QIB_SENDCTRL_SEND_DIS |
+			QIB_SENDCTRL_AVAIL_DIS);
+	}
+
+	return 0;
+}
+
+static void enable_chip(struct qib_devdata *dd)
+{
+	u64 rcvmask;
+	int i;
+
+	/*
+	 * Enable PIO send, and update of PIOavail regs to memory.
+	 */
+	for (i = 0; i < dd->num_pports; ++i)
+		dd->f_sendctrl(dd->pport + i, QIB_SENDCTRL_SEND_ENB |
+			QIB_SENDCTRL_AVAIL_ENB);
+	/*
+	 * Enable kernel ctxts' receive and receive interrupt.
+	 * Other ctxts done as user opens and inits them.
+	 */
+	rcvmask = QIB_RCVCTRL_CTXT_ENB | QIB_RCVCTRL_INTRAVAIL_ENB;
+	rcvmask |= (dd->flags & QIB_NODMA_RTAIL) ?
+		  QIB_RCVCTRL_TAILUPD_DIS : QIB_RCVCTRL_TAILUPD_ENB;
+	for (i = 0; dd->rcd && i < dd->first_user_ctxt; ++i) {
+		struct qib_ctxtdata *rcd = dd->rcd[i];
+
+		if (rcd)
+			dd->f_rcvctrl(rcd->ppd, rcvmask, i);
+	}
+}
+
+static void verify_interrupt(unsigned long opaque)
+{
+	struct qib_devdata *dd = (struct qib_devdata *) opaque;
+
+	if (!dd)
+		return; /* being torn down */
+
+	/*
+	 * If we don't have a lid or any interrupts, let the user know and
+	 * don't bother checking again.
+	 */
+	if (dd->int_counter == 0) {
+		if (!dd->f_intr_fallback(dd))
+			dev_err(&dd->pcidev->dev, "No interrupts detected, "
+				"not usable.\n");
+		else /* re-arm the timer to see if fallback works */
+			mod_timer(&dd->intrchk_timer, jiffies + HZ/2);
+	} else
+		qib_cdbg(INIT, "%u interrupts at timer check\n",
+			 dd->int_counter);
+}
+
+static void init_piobuf_state(struct qib_devdata *dd)
+{
+	int i, pidx;
+	u32 uctxts;
+
+	/*
+	 * Ensure all buffers are free, and fifos empty.  Buffers
+	 * are common, so only do once for port 0.
+	 *
+	 * After enable and qib_chg_pioavailkernel so we can safely
+	 * enable pioavail updates and PIOENABLE.  After this, packets
+	 * are ready and able to go out.
+	 */
+	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_ALL);
+	for (pidx = 0; pidx < dd->num_pports; ++pidx)
+		dd->f_sendctrl(dd->pport + pidx, QIB_SENDCTRL_FLUSH);
+
+	/*
+	 * If not all sendbufs are used, add the one to each of the lower
+	 * numbered contexts.  pbufsctxt and lastctxt_piobuf are
+	 * calculated in chip-specific code because it may cause some
+	 * chip-specific adjustments to be made.
+	 */
+	uctxts = dd->cfgctxts - dd->first_user_ctxt;
+	qib_cdbg(INIT, "%d sendbufs for each of %u user ctxts\n",
+		 dd->pbufsctxt, uctxts);
+	dd->ctxts_extrabuf = dd->pbufsctxt ?
+		dd->lastctxt_piobuf - (dd->pbufsctxt * uctxts) : 0;
+	if (dd->ctxts_extrabuf)
+		qib_cdbg(INIT, "%u pbufs/ctxt leaves some unused, add a "
+			 "buffer to ctxts %u - %u\n", dd->pbufsctxt,
+			 dd->first_user_ctxt,
+			 dd->first_user_ctxt + dd->ctxts_extrabuf - 1);
+
+	/*
+	 * Set up the shadow copies of the piobufavail registers,
+	 * which we compare against the chip registers for now, and
+	 * the in memory DMA'ed copies of the registers.
+	 * By now pioavail updates to memory should have occurred, so
+	 * copy them into our working/shadow registers; this is in
+	 * case something went wrong with abort, but mostly to get the
+	 * initial values of the generation bit correct.
+	 */
+	for (i = 0; i < dd->pioavregs; i++) {
+		__le64 tmp;
+
+		tmp = dd->pioavailregs_dma[i];
+		/*
+		 * Don't need to worry about pioavailkernel here
+		 * because we will call qib_chg_pioavailkernel() later
+		 * in initialization, to busy out buffers as needed.
+		 */
+		dd->pioavailshadow[i] = le64_to_cpu(tmp);
+	}
+	while (i < ARRAY_SIZE(dd->pioavailshadow))
+		dd->pioavailshadow[i++] = 0; /* for debugging sanity */
+
+	/* after pioavailshadow is setup */
+	qib_chg_pioavailkernel(dd, 0, dd->piobcnt2k + dd->piobcnt4k,
+			       TXCHK_CHG_TYPE_KERN, NULL);
+	dd->f_initvl15_bufs(dd);
+}
+
+/**
+ * qib_init - do the actual initialization sequence on the chip
+ * @dd: the qlogic_ib device
+ * @reinit: reinitializing, so don't allocate new memory
+ *
+ * Do the actual initialization sequence on the chip.  This is done
+ * both from the init routine called from the PCI infrastructure, and
+ * when we reset the chip, or detect that it was reset internally,
+ * or it's administratively re-enabled.
+ *
+ * Memory allocation here and in called routines is only done in
+ * the first case (reinit == 0).  We have to be careful, because even
+ * without memory allocation, we need to re-write all the chip registers
+ * TIDs, etc. after the reset or enable has completed.
+ */
+int qib_init(struct qib_devdata *dd, int reinit)
+{
+	int ret = 0, pidx, lastfail = 0;
+	u32 portok = 0;
+	unsigned i;
+	struct qib_ctxtdata *rcd;
+	struct qib_pportdata *ppd;
+	unsigned long flags;
+
+	/* Set linkstate to unknown, so we can watch for a transition. */
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~(QIBL_LINKACTIVE | QIBL_LINKARMED |
+				 QIBL_LINKDOWN | QIBL_LINKINIT |
+				 QIBL_LINKV);
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	}
+
+	if (reinit)
+		ret = init_after_reset(dd);
+	else
+		ret = loadtime_init(dd);
+	if (ret)
+		goto done;
+
+	/* Bypass most chip-init, to get to device creation */
+	if (qib_mini_init)
+		return 0;
+
+	ret = dd->f_late_initreg(dd);
+	if (ret)
+		goto done;
+
+	/* dd->rcd can be NULL if early init failed */
+	for (i = 0; dd->rcd && i < dd->first_user_ctxt; ++i) {
+		/*
+		 * Set up the (kernel) rcvhdr queue and egr TIDs.  If doing
+		 * re-init, the simplest way to handle this is to free
+		 * existing, and re-allocate.
+		 * Need to re-create rest of ctxt 0 ctxtdata as well.
+		 */
+		rcd = dd->rcd[i];
+		if (!rcd)
+			continue;
+
+		lastfail = qib_create_rcvhdrq(dd, rcd);
+		if (!lastfail)
+			lastfail = qib_setup_eagerbufs(rcd);
+		if (lastfail) {
+			qib_dev_err(dd, "failed to allocate kernel ctxt's "
+				    "rcvhdrq and/or egr bufs\n");
+			continue;
+		}
+	}
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		int mtu;
+		if (lastfail)
+			ret = lastfail;
+		ppd = dd->pport + pidx;
+		mtu = ib_mtu_enum_to_int(qib_ibmtu);
+		if (mtu == -1) {
+			mtu = QIB_DEFAULT_MTU;
+			qib_ibmtu = 0; /* don't leave invalid value */
+		}
+		/* set max we can ever have for this driver load */
+		ppd->init_ibmaxlen = min(mtu > 2048 ?
+					 dd->piosize4k : dd->piosize2k,
+					 dd->rcvegrbufsize +
+					 (dd->rcvhdrentsize << 2));
+		/*
+		 * Have to initialize ibmaxlen, but this will normally
+		 * change immediately in qib_set_mtu().
+		 */
+		ppd->ibmaxlen = ppd->init_ibmaxlen;
+		qib_set_mtu(ppd, mtu);
+
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags |= QIBL_IB_LINK_DISABLED;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+
+		lastfail = dd->f_bringup_serdes(ppd);
+		if (lastfail) {
+			qib_devinfo(dd->pcidev,
+				 "Failed to bringup IB port %u\n", ppd->port);
+			lastfail = -ENETDOWN;
+			continue;
+		}
+
+		/* let link come up, and enable IBC */
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~QIBL_IB_LINK_DISABLED;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		portok++;
+	}
+
+	if (!portok) {
+		/* none of the ports initialized */
+		if (!ret && lastfail)
+			ret = lastfail;
+		else if (!ret)
+			ret = -ENETDOWN;
+		/* but continue on, so we can debug cause */
+	}
+
+	enable_chip(dd);
+
+	init_piobuf_state(dd);
+
+done:
+	if (!ret) {
+		/* chip is OK for user apps; mark it as initialized */
+		for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+			ppd = dd->pport + pidx;
+			/*
+			 * Set status even if port serdes is not initialized
+			 * so that diags will work.
+			 */
+			*ppd->statusp |= QIB_STATUS_CHIP_PRESENT |
+				QIB_STATUS_INITTED;
+			if (!ppd->link_speed_enabled)
+				continue;
+			if (dd->flags & QIB_HAS_SEND_DMA)
+				ret = qib_setup_sdma(ppd);
+			init_timer(&ppd->hol_timer);
+			ppd->hol_timer.function = qib_hol_event;
+			ppd->hol_timer.data = (unsigned long)ppd;
+			ppd->hol_state = QIB_HOL_UP;
+		}
+
+		/* now we can enable all interrupts from the chip */
+		dd->f_set_intr_state(dd, 1);
+
+		/*
+		 * Setup to verify we get an interrupt, and fallback
+		 * to an alternate if necessary and possible.
+		 */
+		mod_timer(&dd->intrchk_timer, jiffies + HZ/2);
+		/* start stats retrieval timer */
+		mod_timer(&dd->stats_timer, jiffies + HZ * ACTIVITY_TIMER);
+	} else
+		qib_dbg("Failed (%d) to initialize unit %u\n",
+			ret, dd->unit);
+
+	/* if ret is non-zero, we probably should do some cleanup here... */
+	return ret;
+}
+
+/*
+ * These next two routines are placeholders in case we don't have per-arch
+ * code for controlling write combining.  If explicit control of write
+ * combining is not available, performance will probably be awful.
+ */
+
+int __attribute__((weak)) qib_enable_wc(struct qib_devdata *dd)
+{
+	return -EOPNOTSUPP;
+}
+
+void __attribute__((weak)) qib_disable_wc(struct qib_devdata *dd)
+{
+}
+
+static inline struct qib_devdata *__qib_lookup(int unit)
+{
+	return idr_find(&qib_unit_table, unit);
+}
+
+struct qib_devdata *qib_lookup(int unit)
+{
+	struct qib_devdata *dd;
+	unsigned long flags;
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+	dd = __qib_lookup(unit);
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+
+	return dd;
+}
+
+/*
+ * Stop the timers during unit shutdown, or after an error late
+ * in initialization.
+ */
+static void qib_stop_timers(struct qib_devdata *dd)
+{
+	struct qib_pportdata *ppd;
+	int pidx;
+
+	if (dd->stats_timer.data) {
+		del_timer_sync(&dd->stats_timer);
+		dd->stats_timer.data = 0;
+	}
+	if (dd->intrchk_timer.data) {
+		del_timer_sync(&dd->intrchk_timer);
+		dd->intrchk_timer.data = 0;
+	}
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+		if (ppd->hol_timer.data)
+			del_timer_sync(&ppd->hol_timer);
+		if (ppd->led_override_timer.data) {
+			del_timer_sync(&ppd->led_override_timer);
+			atomic_set(&ppd->led_override_timer_active, 0);
+		}
+	}
+}
+
+/**
+ * qib_shutdown_device - shut down a device
+ * @dd: the qlogic_ib device
+ *
+ * This is called to make the device quiet when we are about to
+ * unload the driver, and also when the device is administratively
+ * disabled.   It does not free any data structures.
+ * Everything it does has to be setup again by qib_init(dd, 1)
+ */
+static void qib_shutdown_device(struct qib_devdata *dd)
+{
+	struct qib_pportdata *ppd;
+	unsigned pidx;
+
+	qib_dbg("Shutting down the device\n");
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+
+		spin_lock_irq(&ppd->lflags_lock);
+		ppd->lflags &= ~(QIBL_LINKDOWN | QIBL_LINKINIT |
+				 QIBL_LINKARMED | QIBL_LINKACTIVE |
+				 QIBL_LINKV);
+		spin_unlock_irq(&ppd->lflags_lock);
+		*ppd->statusp &= ~(QIB_STATUS_IB_CONF | QIB_STATUS_IB_READY);
+	}
+	dd->flags &= ~QIB_INITTED;
+
+	/* mask interrupts, but not errors */
+	dd->f_set_intr_state(dd, 0);
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+		dd->f_rcvctrl(ppd, QIB_RCVCTRL_TAILUPD_DIS |
+				   QIB_RCVCTRL_CTXT_DIS |
+				   QIB_RCVCTRL_INTRAVAIL_DIS |
+				   QIB_RCVCTRL_PKEY_ENB, -1);
+		/*
+		 * Gracefully stop all sends allowing any in progress to
+		 * trickle out first.
+		 */
+		dd->f_sendctrl(ppd, QIB_SENDCTRL_CLEAR);
+	}
+
+	/*
+	 * Enough for anything that's going to trickle out to have actually
+	 * done so.
+	 */
+	udelay(20);
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+		dd->f_setextled(ppd, 0); /* make sure LEDs are off */
+
+		if (dd->flags & QIB_HAS_SEND_DMA)
+			qib_teardown_sdma(ppd);
+
+		dd->f_sendctrl(ppd, QIB_SENDCTRL_AVAIL_DIS |
+				    QIB_SENDCTRL_SEND_DIS);
+		/*
+		 * Clear SerdesEnable.
+		 * We can't count on interrupts since we are stopping.
+		 */
+		dd->f_quiet_serdes(ppd);
+	}
+
+	qib_cdbg(VERBOSE, "Flush time and errors to EEPROM\n");
+	qib_update_eeprom_log(dd);
+}
+
+/**
+ * qib_free_ctxtdata - free a context's allocated data
+ * @dd: the qlogic_ib device
+ * @rcd: the ctxtdata structure
+ *
+ * free up any allocated data for a context
+ * This should not touch anything that would affect a simultaneous
+ * re-allocation of context data, because it is called after qib_mutex
+ * is released (and can be called from reinit as well).
+ * It should never change any chip state, or global driver state.
+ */
+void qib_free_ctxtdata(struct qib_devdata *dd, struct qib_ctxtdata *rcd)
+{
+	if (!rcd)
+		return;
+
+	if (rcd->rcvhdrq) {
+		qib_cdbg(VERBOSE, "free closed ctxt %d rcvhdrq @ %p "
+			 "(size=%lu)\n", rcd->ctxt, rcd->rcvhdrq,
+			 (unsigned long) rcd->rcvhdrq_size);
+		dma_free_coherent(&dd->pcidev->dev, rcd->rcvhdrq_size,
+				  rcd->rcvhdrq, rcd->rcvhdrq_phys);
+		rcd->rcvhdrq = NULL;
+		if (rcd->rcvhdrtail_kvaddr) {
+			dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
+					  rcd->rcvhdrtail_kvaddr,
+					  rcd->rcvhdrqtailaddr_phys);
+			rcd->rcvhdrtail_kvaddr = NULL;
+		}
+	}
+	if (rcd->rcvegrbuf) {
+		unsigned e;
+
+		for (e = 0; e < rcd->rcvegrbuf_chunks; e++) {
+			void *base = rcd->rcvegrbuf[e];
+			size_t size = rcd->rcvegrbuf_size;
+
+			qib_cdbg(VERBOSE, "egrbuf free(%p, %lu), "
+				 "chunk %u/%u\n", base,
+				 (unsigned long) size,
+				 e, rcd->rcvegrbuf_chunks);
+			dma_free_coherent(&dd->pcidev->dev, size,
+					  base, rcd->rcvegrbuf_phys[e]);
+		}
+		kfree(rcd->rcvegrbuf);
+		rcd->rcvegrbuf = NULL;
+		kfree(rcd->rcvegrbuf_phys);
+		rcd->rcvegrbuf_phys = NULL;
+		rcd->rcvegrbuf_chunks = 0;
+	}
+
+	kfree(rcd->tid_pg_list);
+	vfree(rcd->user_event_mask);
+	vfree(rcd->subctxt_uregbase);
+	vfree(rcd->subctxt_rcvegrbuf);
+	vfree(rcd->subctxt_rcvhdr_base);
+	kfree(rcd);
+}
+
+/*
+ * Perform a PIO buffer bandwidth write test, to verify proper system
+ * configuration.  Even when all the setup calls work, occasionally
+ * BIOS or other issues can prevent write combining from working, or
+ * can cause other bandwidth problems to the chip.
+ *
+ * This test simply writes the same buffer over and over again, and
+ * measures close to the peak bandwidth to the chip (not testing
+ * data bandwidth to the wire).   On chips that use an address-based
+ * trigger to send packets to the wire, this is easy.  On chips that
+ * use a count to trigger, we want to make sure that the packet doesn't
+ * go out on the wire, or trigger flow control checks.
+ */
+static void qib_verify_pioperf(struct qib_devdata *dd)
+{
+	u32 pbnum, cnt, lcnt;
+	u32 __iomem *piobuf;
+	u32 *addr;
+	u64 msecs, emsecs;
+
+	piobuf = dd->f_getsendbuf(dd->pport, 0ULL, &pbnum);
+	if (!piobuf) {
+		qib_devinfo(dd->pcidev,
+			 "No PIObufs for checking perf, skipping\n");
+		return;
+	}
+
+	/*
+	 * Enough to give us a reasonable test, less than piobuf size, and
+	 * likely multiple of store buffer length.
+	 */
+	cnt = 1024;
+
+	addr = vmalloc(cnt);
+	if (!addr) {
+		qib_devinfo(dd->pcidev,
+			 "Couldn't get memory for checking PIO perf,"
+			 " skipping\n");
+		goto done;
+	}
+
+	preempt_disable();  /* we want reasonably accurate elapsed time */
+	msecs = 1 + jiffies_to_msecs(jiffies);
+	for (lcnt = 0; lcnt < 10000U; lcnt++) {
+		/* wait until we cross msec boundary */
+		if (jiffies_to_msecs(jiffies) >= msecs)
+			break;
+		udelay(1);
+	}
+
+	dd->f_set_armlaunch(dd, 0);
+
+	/*
+	 * length 0, no dwords actually sent
+	 */
+	writeq(0, piobuf);
+	qib_flush_wc();
+
+	/*
+	 * This is only roughly accurate, since even with preempt we
+	 * still take interrupts that could take a while.   Running for
+	 * >= 5 msec seems to get us "close enough" to accurate values.
+	 */
+	msecs = jiffies_to_msecs(jiffies);
+	for (emsecs = lcnt = 0; emsecs <= 5UL; lcnt++) {
+		qib_pio_copy(piobuf + 64, addr, cnt >> 2);
+		emsecs = jiffies_to_msecs(jiffies) - msecs;
+	}
+
+	/* 1 GiB/sec, slightly over IB SDR line rate */
+	if (lcnt < (emsecs * 1024U))
+		qib_dev_err(dd,
+			    "Performance problem: bandwidth to PIO buffers is "
+			    "only %u MiB/sec\n",
+			    lcnt / (u32) emsecs);
+	else
+		qib_cdbg(INIT, "PIO buffer bandwidth %u MiB/sec is OK\n",
+			 lcnt / (u32) emsecs);
+
+	preempt_enable();
+
+	vfree(addr);
+
+done:
+	/* disarm piobuf, so it's available again */
+	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(pbnum));
+	qib_sendbuf_done(dd, pbnum);
+	dd->f_set_armlaunch(dd, 1);
+}
+
+
+void qib_free_devdata(struct qib_devdata *dd)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+	idr_remove(&qib_unit_table, dd->unit);
+	list_del(&dd->list);
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+
+	ib_dealloc_device(&dd->verbs_dev.ibdev);
+}
+
+/*
+ * Allocate our primary per-unit data structure.  Must be done via verbs
+ * allocator, because the verbs cleanup process both does cleanup and
+ * free of the data structure.
+ * "extra" is for chip-specific data.
+ *
+ * Use the idr mechanism to get a unit number for this unit.
+ */
+struct qib_devdata *qib_alloc_devdata(struct pci_dev *pdev, size_t extra)
+{
+	unsigned long flags;
+	struct qib_devdata *dd;
+	int ret;
+
+	if (!idr_pre_get(&qib_unit_table, GFP_KERNEL)) {
+		dd = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	dd = (struct qib_devdata *) ib_alloc_device(sizeof(*dd) + extra);
+	if (!dd) {
+		dd = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+	ret = idr_get_new(&qib_unit_table, dd, &dd->unit);
+	if (ret >= 0)
+		list_add(&dd->list, &qib_dev_list);
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+
+	if (ret < 0) {
+		qib_early_err(&pdev->dev,
+			      "Could not allocate unit ID: error %d\n", -ret);
+		ib_dealloc_device(&dd->verbs_dev.ibdev);
+		dd = ERR_PTR(ret);
+	}
+bail:
+	return dd;
+}
+
+/*
+ * Called from freeze mode handlers, and from PCI error
+ * reporting code.  Should be paranoid about state of
+ * system and data structures.
+ */
+void qib_disable_after_error(struct qib_devdata *dd)
+{
+	if (dd->flags & QIB_INITTED) {
+		u32 pidx;
+
+		dd->flags &= ~QIB_INITTED;
+		if (dd->pport)
+			for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+				struct qib_pportdata *ppd;
+
+				ppd = dd->pport + pidx;
+				if (dd->flags & QIB_PRESENT) {
+					qib_set_linkstate(ppd,
+						QIB_IB_LINKDOWN_DISABLE);
+					dd->f_setextled(ppd, 0);
+				}
+				*ppd->statusp &= ~QIB_STATUS_IB_READY;
+			}
+	}
+
+	/*
+	 * Mark as having had an error for driver, and also
+	 * for /sys and status word mapped to user programs.
+	 * This marks unit as not usable, until reset.
+	 */
+	if (dd->devstatusp)
+		*dd->devstatusp |= QIB_STATUS_HWERROR;
+}
+
+static void __devexit qib_remove_one(struct pci_dev *);
+static int __devinit qib_init_one(struct pci_dev *,
+				  const struct pci_device_id *);
+
+#define DRIVER_LOAD_MSG "QLogic " QIB_DRV_NAME " loaded: "
+#define PFX QIB_DRV_NAME ": "
+
+static const struct pci_device_id qib_pci_tbl[] = {
+	{ PCI_DEVICE(PCI_VENDOR_ID_PATHSCALE, PCI_DEVICE_ID_QLOGIC_IB_6120) },
+	{ PCI_DEVICE(PCI_VENDOR_ID_QLOGIC, PCI_DEVICE_ID_QLOGIC_IB_7220) },
+	{ PCI_DEVICE(PCI_VENDOR_ID_QLOGIC, PCI_DEVICE_ID_QLOGIC_IB_7322) },
+	{ 0, }
+};
+
+MODULE_DEVICE_TABLE(pci, qib_pci_tbl);
+
+struct pci_driver qib_driver = {
+	.name = QIB_DRV_NAME,
+	.probe = qib_init_one,
+	.remove = __devexit_p(qib_remove_one),
+	.id_table = qib_pci_tbl,
+	.err_handler = &qib_pci_err_handler,
+};
+
+/*
+ * Do all the generic driver unit- and chip-independent memory
+ * allocation and initialization.
+ */
+static int __init qlogic_ib_init(void)
+{
+	int ret;
+
+	if (qib_debug & __QIB_DBG)
+		printk(KERN_INFO DRIVER_LOAD_MSG "%s", ib_qib_version);
+
+	ret = qib_dev_init();
+	if (ret)
+		goto bail;
+
+	/*
+	 * These must be called before the driver is registered with
+	 * the PCI subsystem.
+	 */
+	idr_init(&qib_unit_table);
+	if (!idr_pre_get(&qib_unit_table, GFP_KERNEL)) {
+		printk(KERN_ERR QIB_DRV_NAME ": idr_pre_get() failed\n");
+		ret = -ENOMEM;
+		goto bail_dev;
+	}
+
+	ret = pci_register_driver(&qib_driver);
+	if (ret < 0) {
+		printk(KERN_ERR QIB_DRV_NAME
+		       ": Unable to register driver: error %d\n", -ret);
+		goto bail_unit;
+	}
+
+	/* not fatal if it doesn't work */
+	if (qib_init_qibfs())
+		printk(KERN_ERR QIB_DRV_NAME ": Unable to register ipathfs\n");
+	goto bail; /* all OK */
+
+bail_unit:
+	idr_destroy(&qib_unit_table);
+bail_dev:
+	qib_dev_cleanup();
+bail:
+	return ret;
+}
+
+module_init(qlogic_ib_init);
+
+/*
+ * Do the non-unit driver cleanup, memory free, etc. at unload.
+ */
+static void __exit qlogic_ib_cleanup(void)
+{
+	int ret;
+
+	ret = qib_exit_qibfs();
+	if (ret)
+		printk(KERN_ERR QIB_DRV_NAME ": "
+			"Unable to cleanup counter filesystem: "
+			"error %d\n", -ret);
+
+	pci_unregister_driver(&qib_driver);
+
+	idr_destroy(&qib_unit_table);
+	qib_dev_cleanup();
+}
+
+module_exit(qlogic_ib_cleanup);
+
+/* this can only be called after a successful initialization */
+static void cleanup_device_data(struct qib_devdata *dd)
+{
+	int ctxt;
+	int pidx;
+	struct qib_ctxtdata **tmp;
+	unsigned long flags;
+
+	/* users can't do anything more with chip */
+	for (pidx = 0; pidx < dd->num_pports; ++pidx)
+		if (dd->pport[pidx].statusp)
+			*dd->pport[pidx].statusp &= ~QIB_STATUS_CHIP_PRESENT;
+
+	if (!qib_wc_pat)
+		qib_disable_wc(dd);
+
+	if (dd->pioavailregs_dma) {
+		dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
+				  (void *) dd->pioavailregs_dma,
+				  dd->pioavailregs_phys);
+		dd->pioavailregs_dma = NULL;
+	}
+
+	if (dd->pageshadow) {
+		struct page **tmpp = dd->pageshadow;
+		dma_addr_t *tmpd = dd->physshadow;
+		int i, cnt = 0;
+
+		qib_cdbg(VERBOSE, "Unlocking any expTID pages still "
+			 "locked\n");
+		for (ctxt = 0; ctxt < dd->cfgctxts; ctxt++) {
+			int ctxt_tidbase = ctxt * dd->rcvtidcnt;
+			int maxtid = ctxt_tidbase + dd->rcvtidcnt;
+
+			for (i = ctxt_tidbase; i < maxtid; i++) {
+				if (!tmpp[i])
+					continue;
+				pci_unmap_page(dd->pcidev, tmpd[i],
+					       PAGE_SIZE, PCI_DMA_FROMDEVICE);
+				qib_release_user_pages(&tmpp[i], 1);
+				tmpp[i] = NULL;
+				cnt++;
+			}
+		}
+		if (cnt)
+			qib_cdbg(VERBOSE, "There were still %u expTID "
+				 "entries locked\n", cnt);
+
+		qib_cdbg(VERBOSE, "Free shadow page tid array at %p\n",
+			 dd->pageshadow);
+		tmpp = dd->pageshadow;
+		dd->pageshadow = NULL;
+		vfree(tmpp);
+	}
+
+	/*
+	 * Free any resources still in use (usually just kernel contexts)
+	 * at unload; we do for ctxtcnt, because that's what we allocate.
+	 * We acquire lock to be really paranoid that rcd isn't being
+	 * accessed from some interrupt-related code (that should not happen,
+	 * but best to be sure).
+	 */
+	spin_lock_irqsave(&dd->uctxt_lock, flags);
+	tmp = dd->rcd;
+	dd->rcd = NULL;
+	spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+	for (ctxt = 0; tmp && ctxt < dd->ctxtcnt; ctxt++) {
+		struct qib_ctxtdata *rcd = tmp[ctxt];
+
+		tmp[ctxt] = NULL; /* debugging paranoia */
+		qib_free_ctxtdata(dd, rcd);
+	}
+	kfree(tmp);
+	kfree(dd->boardname);
+}
+
+/*
+ * Clean up on unit shutdown, or error during unit load after
+ * successful initialization.
+ */
+static void qib_postinit_cleanup(struct qib_devdata *dd)
+{
+	/*
+	 * Clean up chip-specific stuff.
+	 * We check for NULL here, because it's outside
+	 * the kregbase check, and we need to call it
+	 * after the free_irq.  Thus it's possible that
+	 * the function pointers were never initialized.
+	 */
+	if (dd->f_cleanup)
+		dd->f_cleanup(dd);
+
+	qib_pcie_ddcleanup(dd);
+
+	cleanup_device_data(dd);
+
+	qib_free_devdata(dd);
+}
+
+static int __devinit qib_init_one(struct pci_dev *pdev,
+				  const struct pci_device_id *ent)
+{
+	int ret, j, pidx, initfail;
+	struct qib_devdata *dd = NULL;
+
+	ret = qib_pcie_init(pdev, ent);
+	if (ret)
+		goto bail;
+
+	/*
+	 * Do device-specific initialiation, function table setup, dd
+	 * allocation, etc.
+	 */
+	switch (ent->device) {
+	case PCI_DEVICE_ID_QLOGIC_IB_6120:
+#ifndef CONFIG_PCI_MSI
+		qib_early_err(&pdev->dev, "QLogic PCIE device 0x%x cannot "
+			      "work if CONFIG_PCI_MSI is not enabled\n",
+			      ent->device);
+		ret = -ENODEV;
+#else
+		dd = qib_init_iba6120_funcs(pdev, ent);
+#endif
+		break;
+
+	case PCI_DEVICE_ID_QLOGIC_IB_7220:
+		dd = qib_init_iba7220_funcs(pdev, ent);
+		break;
+
+	case PCI_DEVICE_ID_QLOGIC_IB_7322:
+		dd = qib_init_iba7322_funcs(pdev, ent);
+		break;
+
+	default:
+		qib_early_err(&pdev->dev, "Failing on unknown QLogic "
+			      "deviceid 0x%x\n", ent->device);
+		ret = -ENODEV;
+	}
+
+	if (IS_ERR(dd))
+		ret = PTR_ERR(dd);
+	if (ret)
+		goto bail; /* error already printed */
+
+	/* do the generic initialization */
+	initfail = qib_init(dd, 0);
+
+	ret = qib_register_ib_device(dd);
+
+	/*
+	 * Now ready for use.  this should be cleared whenever we
+	 * detect a reset, or initiate one.  If earlier failure,
+	 * we still create devices, so diags, etc. can be used
+	 * to determine cause of problem.
+	 */
+	if (!qib_mini_init && !initfail && !ret)
+		dd->flags |= QIB_INITTED;
+
+	j = qib_device_create(dd);
+	if (j)
+		qib_dev_err(dd, "Failed to create /dev devices: %d\n", -j);
+	j = qibfs_add(dd);
+	if (j)
+		qib_dev_err(dd, "Failed filesystem setup for counters: %d\n",
+			    -j);
+
+	if (qib_mini_init || initfail || ret) {
+		qib_stop_timers(dd);
+		for (pidx = 0; pidx < dd->num_pports; ++pidx)
+			dd->f_quiet_serdes(dd->pport + pidx);
+		if (initfail)
+			ret = initfail;
+		goto bail;
+	}
+
+	if (!qib_wc_pat) {
+		ret = qib_enable_wc(dd);
+		if (ret) {
+			qib_dev_err(dd, "Write combining not enabled "
+				    "(err %d): performance may be poor\n",
+				    -ret);
+			ret = 0;
+		}
+	}
+
+	qib_verify_pioperf(dd);
+bail:
+	return ret;
+}
+
+static void __devexit qib_remove_one(struct pci_dev *pdev)
+{
+	struct qib_devdata *dd = pci_get_drvdata(pdev);
+	int ret;
+
+	qib_cdbg(VERBOSE, "removing, pdev=%p, dd=%p\n", pdev, dd);
+
+	/* unregister from IB core */
+	qib_unregister_ib_device(dd);
+
+	/*
+	 * Disable the IB link, disable interrupts on the device,
+	 * clear dma engines, etc.
+	 */
+	if (!qib_mini_init)
+		qib_shutdown_device(dd);
+
+	qib_stop_timers(dd);
+
+	/* wait until all of our (qsfp) schedule_work() calls complete */
+	flush_scheduled_work();
+
+	ret = qibfs_remove(dd);
+	if (ret)
+		qib_dev_err(dd, "Failed counters filesystem cleanup: %d\n",
+			    -ret);
+
+	qib_device_remove(dd);
+
+	qib_postinit_cleanup(dd);
+}
+
+/**
+ * qib_create_rcvhdrq - create a receive header queue
+ * @dd: the qlogic_ib device
+ * @rcd: the context data
+ *
+ * This must be contiguous memory (from an i/o perspective), and must be
+ * DMA'able (which means for some systems, it will go through an IOMMU,
+ * or be forced into a low address range).
+ */
+int qib_create_rcvhdrq(struct qib_devdata *dd, struct qib_ctxtdata *rcd)
+{
+	unsigned amt;
+
+	if (!rcd->rcvhdrq) {
+		dma_addr_t phys_hdrqtail;
+		gfp_t gfp_flags;
+
+		amt = ALIGN(dd->rcvhdrcnt * dd->rcvhdrentsize *
+			    sizeof(u32), PAGE_SIZE);
+		gfp_flags = (rcd->ctxt >= dd->first_user_ctxt) ?
+			GFP_USER : GFP_KERNEL;
+		rcd->rcvhdrq = dma_alloc_coherent(
+			&dd->pcidev->dev, amt, &rcd->rcvhdrq_phys,
+			gfp_flags | __GFP_COMP);
+
+		if (!rcd->rcvhdrq) {
+			qib_dev_err(dd, "attempt to allocate %d bytes "
+				    "for ctxt %u rcvhdrq failed\n",
+				    amt, rcd->ctxt);
+			goto bail;
+		}
+
+		if (rcd->ctxt >= dd->first_user_ctxt) {
+			rcd->user_event_mask = vmalloc_user(PAGE_SIZE);
+			if (!rcd->user_event_mask)
+				goto bail_free_hdrq;
+		}
+
+		if (!(dd->flags & QIB_NODMA_RTAIL)) {
+			rcd->rcvhdrtail_kvaddr = dma_alloc_coherent(
+				&dd->pcidev->dev, PAGE_SIZE, &phys_hdrqtail,
+				gfp_flags);
+			if (!rcd->rcvhdrtail_kvaddr)
+				goto bail_free;
+			rcd->rcvhdrqtailaddr_phys = phys_hdrqtail;
+			qib_cdbg(INIT, "ctxt %d hdrtailaddr, %llx "
+				 "physical\n", rcd->ctxt,
+				 (unsigned long long) phys_hdrqtail);
+		}
+
+		rcd->rcvhdrq_size = amt;
+
+		qib_cdbg(PROC, "%d pages at %p (phys %lx) size=%lu "
+			 "for ctxt %u rcvhdr Q\n",
+			 amt >> PAGE_SHIFT, rcd->rcvhdrq,
+			 (unsigned long) rcd->rcvhdrq_phys,
+			 (unsigned long) rcd->rcvhdrq_size,
+			 rcd->ctxt);
+	} else
+		qib_cdbg(PROC, "reuse ctxt %d rcvhdrq @%p %llx phys; "
+			 "hdrtailaddr@%p %llx physical\n",
+			 rcd->ctxt, rcd->rcvhdrq,
+			 (unsigned long long) rcd->rcvhdrq_phys,
+			 rcd->rcvhdrtail_kvaddr, (unsigned long long)
+			 rcd->rcvhdrqtailaddr_phys);
+
+	/* clear for security and sanity on each use */
+	memset(rcd->rcvhdrq, 0, rcd->rcvhdrq_size);
+	if (rcd->rcvhdrtail_kvaddr)
+		memset(rcd->rcvhdrtail_kvaddr, 0, PAGE_SIZE);
+	return 0;
+
+bail_free:
+	qib_dev_err(dd, "attempt to allocate 1 page for ctxt %u "
+		    "rcvhdrqtailaddr failed\n", rcd->ctxt);
+	vfree(rcd->user_event_mask);
+	rcd->user_event_mask = NULL;
+bail_free_hdrq:
+	dma_free_coherent(&dd->pcidev->dev, amt, rcd->rcvhdrq,
+			  rcd->rcvhdrq_phys);
+	rcd->rcvhdrq = NULL;
+bail:
+	return -ENOMEM;
+}
+
+/**
+ * allocate eager buffers, both kernel and user contexts.
+ * @rcd: the context we are setting up.
+ *
+ * Allocate the eager TID buffers and program them into hip.
+ * They are no longer completely contiguous, we do multiple allocation
+ * calls.  Otherwise we get the OOM code involved, by asking for too
+ * much per call, with disastrous results on some kernels.
+ */
+int qib_setup_eagerbufs(struct qib_ctxtdata *rcd)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned e, egrcnt, egrperchunk, chunk, egrsize, egroff;
+	size_t size;
+	gfp_t gfp_flags;
+
+	/*
+	 * GFP_USER, but without GFP_FS, so buffer cache can be
+	 * coalesced (we hope); otherwise, even at order 4,
+	 * heavy filesystem activity makes these fail, and we can
+	 * use compound pages.
+	 */
+	gfp_flags = __GFP_WAIT | __GFP_IO | __GFP_COMP;
+
+	egrcnt = rcd->rcvegrcnt;
+	egroff = rcd->rcvegr_tid_base;
+	egrsize = dd->rcvegrbufsize;
+	qib_cdbg(PROC, "ctxt%d: %d egr buffers, at egrtid "
+		 "offset %x, egrsize %u\n", rcd->ctxt, egrcnt, egroff,
+		 egrsize);
+
+	chunk = rcd->rcvegrbuf_chunks;
+	egrperchunk = rcd->rcvegrbufs_perchunk;
+	size = rcd->rcvegrbuf_size;
+	if (!rcd->rcvegrbuf) {
+		rcd->rcvegrbuf =
+			kzalloc(chunk * sizeof(rcd->rcvegrbuf[0]),
+				GFP_KERNEL);
+		if (!rcd->rcvegrbuf)
+			goto bail;
+	}
+	if (!rcd->rcvegrbuf_phys) {
+		rcd->rcvegrbuf_phys =
+			kmalloc(chunk * sizeof(rcd->rcvegrbuf_phys[0]),
+				GFP_KERNEL);
+		if (!rcd->rcvegrbuf_phys)
+			goto bail_rcvegrbuf;
+	}
+	for (e = 0; e < rcd->rcvegrbuf_chunks; e++) {
+		if (rcd->rcvegrbuf[e])
+			continue;
+		rcd->rcvegrbuf[e] =
+			dma_alloc_coherent(&dd->pcidev->dev, size,
+					   &rcd->rcvegrbuf_phys[e],
+					   gfp_flags);
+		if (!rcd->rcvegrbuf[e])
+			goto bail_rcvegrbuf_phys;
+	}
+
+	rcd->rcvegr_phys = rcd->rcvegrbuf_phys[0];
+
+	for (e = chunk = 0; chunk < rcd->rcvegrbuf_chunks; chunk++) {
+		dma_addr_t pa = rcd->rcvegrbuf_phys[chunk];
+		unsigned i;
+
+		for (i = 0; e < egrcnt && i < egrperchunk; e++, i++) {
+			dd->f_put_tid(dd, e + egroff +
+					  (u64 __iomem *)
+					  ((char __iomem *)
+					   dd->kregbase +
+					   dd->rcvegrbase),
+					  RCVHQ_RCV_TYPE_EAGER, pa);
+			pa += egrsize;
+		}
+		cond_resched(); /* don't hog the cpu */
+	}
+
+	return 0;
+
+bail_rcvegrbuf_phys:
+	for (e = 0; e < rcd->rcvegrbuf_chunks && rcd->rcvegrbuf[e]; e++)
+		dma_free_coherent(&dd->pcidev->dev, size,
+				  rcd->rcvegrbuf[e], rcd->rcvegrbuf_phys[e]);
+	kfree(rcd->rcvegrbuf_phys);
+	rcd->rcvegrbuf_phys = NULL;
+bail_rcvegrbuf:
+	kfree(rcd->rcvegrbuf);
+	rcd->rcvegrbuf = NULL;
+bail:
+	return -ENOMEM;
+}
+
+int init_chip_wc_pat(struct qib_devdata *dd, u32 vl15buflen)
+{
+	u64 __iomem *qib_kregbase = NULL;
+	void __iomem *qib_piobase = NULL;
+	u64 __iomem *qib_userbase = NULL;
+	u64 qib_kreglen;
+	u64 qib_pio2koffset = dd->piobufbase & 0xffffffff;
+	u64 qib_pio4koffset = dd->piobufbase >> 32;
+	u64 qib_pio2klen = dd->piobcnt2k * dd->palign;
+	u64 qib_pio4klen = dd->piobcnt4k * dd->align4k;
+	u64 qib_physaddr = dd->physaddr;
+	u64 qib_piolen;
+	u64 qib_userlen = 0;
+
+	/*
+	 * Free the old mapping because the kernel will try to reuse the
+	 * old mapping and not create a new mapping with the
+	 * write combining attribute.
+	 */
+	iounmap(dd->kregbase);
+	dd->kregbase = NULL;
+
+	/*
+	 * Assumes chip address space looks like:
+	 *	- kregs + sregs + cregs + uregs (in any order)
+	 *	- piobufs (2K and 4K bufs in either order)
+	 * or:
+	 *	- kregs + sregs + cregs (in any order)
+	 *	- piobufs (2K and 4K bufs in either order)
+	 *	- uregs
+	 */
+	if (dd->piobcnt4k == 0) {
+		qib_kreglen = qib_pio2koffset;
+		qib_piolen = qib_pio2klen;
+	} else if (qib_pio2koffset < qib_pio4koffset) {
+		qib_kreglen = qib_pio2koffset;
+		qib_piolen = qib_pio4koffset + qib_pio4klen - qib_kreglen;
+	} else {
+		qib_kreglen = qib_pio4koffset;
+		qib_piolen = qib_pio2koffset + qib_pio2klen - qib_kreglen;
+	}
+	qib_piolen += vl15buflen;
+	/* Map just the configured ports (not all hw ports) */
+	if (dd->uregbase > qib_kreglen)
+		qib_userlen = dd->ureg_align * dd->cfgctxts;
+
+	/* Sanity checks passed, now create the new mappings */
+	qib_kregbase = ioremap_nocache(qib_physaddr, qib_kreglen);
+	if (!qib_kregbase) {
+		qib_dbg("Unable to remap io addr %llx to kvirt\n",
+			qib_physaddr);
+		goto bail;
+	}
+	qib_cdbg(VERBOSE, "WC PAT remapped io addr %llx"
+		 " to kregbase %p for %llx bytes\n",
+		 qib_physaddr, qib_kregbase, qib_kreglen);
+
+	qib_piobase = ioremap_wc(qib_physaddr + qib_kreglen, qib_piolen);
+	if (!qib_piobase) {
+		qib_dbg("Unable to remap io addr %llx to kvirt\n",
+			qib_physaddr + qib_kreglen);
+		goto bail_kregbase;
+	}
+	qib_cdbg(VERBOSE, "WC PAT remapped io addr %llx"
+		 " to piobase %p for %llx bytes\n",
+		 qib_physaddr + qib_kreglen, qib_piobase, qib_piolen);
+
+	if (qib_userlen) {
+		qib_userbase = ioremap_nocache(qib_physaddr + dd->uregbase,
+					       qib_userlen);
+		if (!qib_userbase) {
+			qib_dbg("Unable to remap io addr %llx to kvirt\n",
+				qib_physaddr + dd->uregbase);
+			goto bail_piobase;
+		}
+		qib_cdbg(VERBOSE, "WC PAT remapped io addr %llx"
+			 " to userbase %p for %llx bytes\n",
+			 qib_physaddr + dd->uregbase,
+			 qib_userbase, qib_userlen);
+	}
+
+	dd->kregbase = qib_kregbase;
+	dd->kregend = (u64 __iomem *)
+		((char __iomem *) qib_kregbase + qib_kreglen);
+	dd->piobase = qib_piobase;
+	dd->pio2kbase = (void __iomem *)
+		(((char __iomem *) dd->piobase) +
+		 qib_pio2koffset - qib_kreglen);
+	if (dd->piobcnt4k)
+		dd->pio4kbase = (void __iomem *)
+			(((char __iomem *) dd->piobase) +
+			 qib_pio4koffset - qib_kreglen);
+	if (qib_userlen)
+		/* ureg will now be accessed relative to dd->userbase */
+		dd->userbase = qib_userbase;
+	qib_cdbg(VERBOSE, "New addrs: kreg %p pio2k %p pio4k %p user %p\n",
+		 dd->kregbase, dd->pio2kbase, dd->pio4kbase, dd->userbase);
+	return 0;
+
+bail_piobase:
+	iounmap(qib_piobase);
+bail_kregbase:
+	iounmap(qib_kregbase);
+bail:
+	return -ENOMEM;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 21/51] IB/qib: Add qib_intr.c

[Ralph Campbell]

creates the qib_intr.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_intr.c |  217 ++++++++++++++++++++++++++++++++++
 1 files changed, 217 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_intr.c

diff --git a/drivers/infiniband/hw/qib/qib_intr.c b/drivers/infiniband/hw/qib/qib_intr.c
new file mode 100644
index 0000000..46e128a
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_intr.c
@@ -0,0 +1,217 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/pci.h>
+#include <linux/delay.h>
+
+#include "qib.h"
+#include "qib_common.h"
+
+/**
+ * qib_format_hwmsg - format a single hwerror message
+ * @msg message buffer
+ * @msgl length of message buffer
+ * @hwmsg message to add to message buffer
+ */
+static void qib_format_hwmsg(char *msg, size_t msgl, const char *hwmsg)
+{
+	strlcat(msg, "[", msgl);
+	strlcat(msg, hwmsg, msgl);
+	strlcat(msg, "]", msgl);
+}
+
+/**
+ * qib_format_hwerrors - format hardware error messages for display
+ * @hwerrs hardware errors bit vector
+ * @hwerrmsgs hardware error descriptions
+ * @nhwerrmsgs number of hwerrmsgs
+ * @msg message buffer
+ * @msgl message buffer length
+ */
+void qib_format_hwerrors(u64 hwerrs, const struct qib_hwerror_msgs *hwerrmsgs,
+			 size_t nhwerrmsgs, char *msg, size_t msgl)
+{
+	int i;
+
+	for (i = 0; i < nhwerrmsgs; i++)
+		if (hwerrs & hwerrmsgs[i].mask)
+			qib_format_hwmsg(msg, msgl, hwerrmsgs[i].msg);
+}
+
+static void signal_ib_event(struct qib_pportdata *ppd, enum ib_event_type ev)
+{
+	struct ib_event event;
+	struct qib_devdata *dd = ppd->dd;
+
+	event.device = &dd->verbs_dev.ibdev;
+	event.element.port_num = ppd->port;
+	event.event = ev;
+	ib_dispatch_event(&event);
+}
+
+void qib_handle_e_ibstatuschanged(struct qib_pportdata *ppd, u64 ibcs)
+{
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+	u32 lstate;
+	u8 ltstate;
+
+	lstate = dd->f_iblink_state(ibcs); /* linkstate */
+	ltstate = dd->f_ibphys_portstate(ibcs);
+
+	/*
+	 * If linkstate transitions into INIT from any of the various down
+	 * states, or if it transitions from any of the up (INIT or better)
+	 * states into any of the down states (except link recovery), then
+	 * call the chip-specific code to take appropriate actions.
+	 */
+	if (lstate >= IB_PORT_INIT && (ppd->lflags & QIBL_LINKDOWN) &&
+	    ltstate == IB_PHYSPORTSTATE_LINKUP) {
+		/* transitioned to UP */
+		if (dd->f_ib_updown(ppd, 1, ibcs))
+			goto skip_ibchange; /* chip-code handled */
+	} else if (ppd->lflags & (QIBL_LINKINIT | QIBL_LINKARMED |
+		   QIBL_LINKACTIVE | QIBL_IB_FORCE_NOTIFY)) {
+		if (ltstate != IB_PHYSPORTSTATE_LINKUP &&
+		    ltstate <= IB_PHYSPORTSTATE_CFG_TRAIN &&
+		    dd->f_ib_updown(ppd, 0, ibcs))
+			goto skip_ibchange; /* chip-code handled */
+		qib_set_uevent_bits(ppd, _QIB_EVENT_LINKDOWN_BIT);
+	}
+
+	if (lstate != IB_PORT_DOWN) {
+		/* lstate is INIT, ARMED, or ACTIVE */
+		if (lstate != IB_PORT_ACTIVE) {
+			*ppd->statusp &= ~QIB_STATUS_IB_READY;
+			if (ppd->lflags & QIBL_LINKACTIVE)
+				signal_ib_event(ppd, IB_EVENT_PORT_ERR);
+			spin_lock_irqsave(&ppd->lflags_lock, flags);
+			if (lstate == IB_PORT_ARMED) {
+				ppd->lflags |= QIBL_LINKARMED | QIBL_LINKV;
+				ppd->lflags &= ~(QIBL_LINKINIT |
+					QIBL_LINKDOWN | QIBL_LINKACTIVE);
+			} else {
+				ppd->lflags |= QIBL_LINKINIT | QIBL_LINKV;
+				ppd->lflags &= ~(QIBL_LINKARMED |
+					QIBL_LINKDOWN | QIBL_LINKACTIVE);
+			}
+			spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		} else if (ltstate == IB_PHYSPORTSTATE_LINKUP) {
+			/* active, but not active defered */
+			*ppd->statusp |=
+				QIB_STATUS_IB_READY | QIB_STATUS_IB_CONF;
+			spin_lock_irqsave(&ppd->lflags_lock, flags);
+			ppd->lflags |= QIBL_LINKACTIVE | QIBL_LINKV;
+			ppd->lflags &= ~(QIBL_LINKINIT |
+				QIBL_LINKDOWN | QIBL_LINKARMED);
+			spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+			if (dd->flags & QIB_HAS_SEND_DMA)
+				qib_sdma_process_event(ppd,
+					qib_sdma_event_e30_go_running);
+			signal_ib_event(ppd, IB_EVENT_PORT_ACTIVE);
+			dd->f_setextled(ppd, 1);
+			qib_hol_up(ppd);
+		}
+	} else { /* down */
+		if (ppd->lflags & QIBL_LINKACTIVE)
+			signal_ib_event(ppd, IB_EVENT_PORT_ERR);
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags |= QIBL_LINKDOWN | QIBL_LINKV;
+		ppd->lflags &= ~(QIBL_LINKINIT |
+				 QIBL_LINKACTIVE | QIBL_LINKARMED);
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		*ppd->statusp &= ~QIB_STATUS_IB_READY;
+	}
+
+skip_ibchange:
+	ppd->lastibcstat = ibcs;
+	return;
+}
+
+/*
+ * Handle receive interrupts for user ctxts; this means a user
+ * process was waiting for a packet to arrive, and didn't want
+ * to poll.
+ */
+void qib_handle_urcv(struct qib_devdata *dd, u64 ctxtr)
+{
+	struct qib_ctxtdata *rcd;
+	unsigned long flags;
+	int i;
+
+	spin_lock_irqsave(&dd->uctxt_lock, flags);
+	for (i = dd->first_user_ctxt; dd->rcd && i < dd->cfgctxts; i++) {
+		if (!(ctxtr & (1ULL << i)))
+			continue;
+		rcd = dd->rcd[i];
+		if (!rcd || !rcd->cnt)
+			continue;
+
+		if (test_and_clear_bit(QIB_CTXT_WAITING_RCV, &rcd->flag)) {
+			wake_up_interruptible(&rcd->wait);
+			dd->f_rcvctrl(rcd->ppd, QIB_RCVCTRL_INTRAVAIL_DIS,
+				      rcd->ctxt);
+		} else if (test_and_clear_bit(QIB_CTXT_WAITING_URG,
+					      &rcd->flag)) {
+			rcd->urgent++;
+			wake_up_interruptible(&rcd->wait);
+		}
+	}
+	spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+}
+
+void qib_bad_intrstatus(struct qib_devdata *dd)
+{
+	static int allbits;
+
+	/* separate routine, for better optimization of qib_intr() */
+
+	/*
+	 * We print the message and disable interrupts, in hope of
+	 * having a better chance of debugging the problem.
+	 */
+	qib_dev_err(dd, "Read of chip interrupt status failed"
+		    " disabling interrupts\n");
+	if (allbits++) {
+		/* disable interrupt delivery, something is very wrong */
+		if (allbits == 2)
+			dd->f_set_intr_state(dd, 0);
+		if (allbits == 3) {
+			qib_dev_err(dd, "2nd bad interrupt status, "
+				    "unregistering interrupts\n");
+			dd->flags |= QIB_BADINTR;
+			dd->flags &= ~QIB_INITTED;
+			dd->f_free_irq(dd);
+		}
+	}
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 22/51] IB/qib: Add qib_keys.c

[Ralph Campbell]

creates the qib_keys.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_keys.c |  278 ++++++++++++++++++++++++++++++++++
 1 files changed, 278 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_keys.c

diff --git a/drivers/infiniband/hw/qib/qib_keys.c b/drivers/infiniband/hw/qib/qib_keys.c
new file mode 100644
index 0000000..e2b42e5
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_keys.c
@@ -0,0 +1,278 @@
+/*
+ * Copyright (c) 2006, 2007, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "qib.h"
+
+/**
+ * qib_alloc_lkey - allocate an lkey
+ * @rkt: lkey table in which to allocate the lkey
+ * @mr: memory region that this lkey protects
+ *
+ * Returns 1 if successful, otherwise returns 0.
+ */
+
+int qib_alloc_lkey(struct qib_lkey_table *rkt, struct qib_mregion *mr)
+{
+	unsigned long flags;
+	u32 r;
+	u32 n;
+	int ret;
+
+	spin_lock_irqsave(&rkt->lock, flags);
+
+	/* Find the next available LKEY */
+	r = rkt->next;
+	n = r;
+	for (;;) {
+		if (rkt->table[r] == NULL)
+			break;
+		r = (r + 1) & (rkt->max - 1);
+		if (r == n) {
+			spin_unlock_irqrestore(&rkt->lock, flags);
+			qib_dbg("LKEY table full\n");
+			ret = 0;
+			goto bail;
+		}
+	}
+	rkt->next = (r + 1) & (rkt->max - 1);
+	/*
+	 * Make sure lkey is never zero which is reserved to indicate an
+	 * unrestricted LKEY.
+	 */
+	rkt->gen++;
+	mr->lkey = (r << (32 - ib_qib_lkey_table_size)) |
+		((((1 << (24 - ib_qib_lkey_table_size)) - 1) & rkt->gen)
+		 << 8);
+	if (mr->lkey == 0) {
+		mr->lkey |= 1 << 8;
+		rkt->gen++;
+	}
+	rkt->table[r] = mr;
+	spin_unlock_irqrestore(&rkt->lock, flags);
+
+	ret = 1;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_free_lkey - free an lkey
+ * @rkt: table from which to free the lkey
+ * @lkey: lkey id to free
+ */
+int qib_free_lkey(struct qib_ibdev *dev, struct qib_mregion *mr)
+{
+	unsigned long flags;
+	u32 lkey = mr->lkey;
+	u32 r;
+	int ret;
+
+	spin_lock_irqsave(&dev->lk_table.lock, flags);
+	if (lkey == 0) {
+		if (dev->dma_mr && dev->dma_mr == mr) {
+			ret = atomic_read(&dev->dma_mr->refcount);
+			if (!ret)
+				dev->dma_mr = NULL;
+		} else
+			ret = 0;
+	} else {
+		r = lkey >> (32 - ib_qib_lkey_table_size);
+		ret = atomic_read(&dev->lk_table.table[r]->refcount);
+		if (!ret)
+			dev->lk_table.table[r] = NULL;
+	}
+	spin_unlock_irqrestore(&dev->lk_table.lock, flags);
+
+	if (ret) {
+		qib_dbg("MR busy (LKEY %x cnt %u)\n", lkey, ret);
+		ret = -EBUSY;
+	}
+	return ret;
+}
+
+/**
+ * qib_lkey_ok - check IB SGE for validity and initialize
+ * @rkt: table containing lkey to check SGE against
+ * @isge: outgoing internal SGE
+ * @sge: SGE to check
+ * @acc: access flags
+ *
+ * Return 1 if valid and successful, otherwise returns 0.
+ *
+ * Check the IB SGE for validity and initialize our internal version
+ * of it.
+ */
+int qib_lkey_ok(struct qib_lkey_table *rkt, struct qib_pd *pd,
+		struct qib_sge *isge, struct ib_sge *sge, int acc)
+{
+	struct qib_mregion *mr;
+	unsigned n, m;
+	size_t off;
+	int ret = 0;
+	unsigned long flags;
+
+	/*
+	 * We use LKEY == zero for kernel virtual addresses
+	 * (see qib_get_dma_mr and qib_dma.c).
+	 */
+	spin_lock_irqsave(&rkt->lock, flags);
+	if (sge->lkey == 0) {
+		struct qib_ibdev *dev = to_idev(pd->ibpd.device);
+
+		if (pd->user)
+			goto bail;
+		if (!dev->dma_mr)
+			goto bail;
+		atomic_inc(&dev->dma_mr->refcount);
+		isge->mr = dev->dma_mr;
+		isge->vaddr = (void *) sge->addr;
+		isge->length = sge->length;
+		isge->sge_length = sge->length;
+		isge->m = 0;
+		isge->n = 0;
+		goto ok;
+	}
+	mr = rkt->table[(sge->lkey >> (32 - ib_qib_lkey_table_size))];
+	if (unlikely(mr == NULL || mr->lkey != sge->lkey ||
+		     mr->pd != &pd->ibpd))
+		goto bail;
+
+	off = sge->addr - mr->user_base;
+	if (unlikely(sge->addr < mr->user_base ||
+		     off + sge->length > mr->length ||
+		     (mr->access_flags & acc) != acc))
+		goto bail;
+
+	off += mr->offset;
+	m = 0;
+	n = 0;
+	while (off >= mr->map[m]->segs[n].length) {
+		off -= mr->map[m]->segs[n].length;
+		n++;
+		if (n >= QIB_SEGSZ) {
+			m++;
+			n = 0;
+		}
+	}
+	atomic_inc(&mr->refcount);
+	isge->mr = mr;
+	isge->vaddr = mr->map[m]->segs[n].vaddr + off;
+	isge->length = mr->map[m]->segs[n].length - off;
+	isge->sge_length = sge->length;
+	isge->m = m;
+	isge->n = n;
+ok:
+	ret = 1;
+bail:
+	spin_unlock_irqrestore(&rkt->lock, flags);
+	return ret;
+}
+
+/**
+ * qib_rkey_ok - check the IB virtual address, length, and RKEY
+ * @dev: infiniband device
+ * @ss: SGE state
+ * @len: length of data
+ * @vaddr: virtual address to place data
+ * @rkey: rkey to check
+ * @acc: access flags
+ *
+ * Return 1 if successful, otherwise 0.
+ */
+int qib_rkey_ok(struct qib_qp *qp, struct qib_sge *sge,
+		u32 len, u64 vaddr, u32 rkey, int acc)
+{
+	struct qib_lkey_table *rkt = &to_idev(qp->ibqp.device)->lk_table;
+	struct qib_mregion *mr;
+	unsigned n, m;
+	size_t off;
+	int ret = 0;
+	unsigned long flags;
+
+	/*
+	 * We use RKEY == zero for kernel virtual addresses
+	 * (see qib_get_dma_mr and qib_dma.c).
+	 */
+	spin_lock_irqsave(&rkt->lock, flags);
+	if (rkey == 0) {
+		struct qib_pd *pd = to_ipd(qp->ibqp.pd);
+		struct qib_ibdev *dev = to_idev(pd->ibpd.device);
+
+		if (pd->user)
+			goto bail;
+		if (!dev->dma_mr)
+			goto bail;
+		atomic_inc(&dev->dma_mr->refcount);
+		sge->mr = dev->dma_mr;
+		sge->vaddr = (void *) vaddr;
+		sge->length = len;
+		sge->sge_length = len;
+		sge->m = 0;
+		sge->n = 0;
+		goto ok;
+	}
+
+	mr = rkt->table[(rkey >> (32 - ib_qib_lkey_table_size))];
+	if (unlikely(mr == NULL || mr->lkey != rkey || qp->ibqp.pd != mr->pd))
+		goto bail;
+
+	off = vaddr - mr->iova;
+	if (unlikely(vaddr < mr->iova || off + len > mr->length ||
+		     (mr->access_flags & acc) == 0))
+		goto bail;
+
+	off += mr->offset;
+	m = 0;
+	n = 0;
+	while (off >= mr->map[m]->segs[n].length) {
+		off -= mr->map[m]->segs[n].length;
+		n++;
+		if (n >= QIB_SEGSZ) {
+			m++;
+			n = 0;
+		}
+	}
+	atomic_inc(&mr->refcount);
+	sge->mr = mr;
+	sge->vaddr = mr->map[m]->segs[n].vaddr + off;
+	sge->length = mr->map[m]->segs[n].length - off;
+	sge->sge_length = len;
+	sge->m = m;
+	sge->n = n;
+ok:
+	ret = 1;
+bail:
+	spin_unlock_irqrestore(&rkt->lock, flags);
+	return ret;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 23/51] IB/qib: Add qib_mad.c

[Ralph Campbell]

creates the qib_mad.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_mad.c | 1867 +++++++++++++++++++++++++++++++++++
 1 files changed, 1867 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_mad.c

diff --git a/drivers/infiniband/hw/qib/qib_mad.c b/drivers/infiniband/hw/qib/qib_mad.c
new file mode 100644
index 0000000..d6fd6e1
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_mad.c
@@ -0,0 +1,1867 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_smi.h>
+
+#include "qib.h"
+#include "qib_mad.h"
+
+static int reply(struct ib_smp *smp)
+{
+	/*
+	 * The verbs framework will handle the directed/LID route
+	 * packet changes.
+	 */
+	smp->method = IB_MGMT_METHOD_GET_RESP;
+	if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
+		smp->status |= IB_SMP_DIRECTION;
+	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
+}
+
+static void qib_send_trap(struct qib_ibport *ibp, void *data, unsigned len)
+{
+	struct ib_mad_send_buf *send_buf;
+	struct ib_mad_agent *agent;
+	struct ib_smp *smp;
+	int ret;
+	unsigned long flags;
+	unsigned long timeout;
+
+	agent = ibp->send_agent;
+	if (!agent)
+		return;
+
+	/* o14-3.2.1 */
+	if (!(ppd_from_ibp(ibp)->lflags & QIBL_LINKACTIVE))
+		return;
+
+	/* o14-2 */
+	if (ibp->trap_timeout && time_before(jiffies, ibp->trap_timeout))
+		return;
+
+	send_buf = ib_create_send_mad(agent, 0, 0, 0, IB_MGMT_MAD_HDR,
+				      IB_MGMT_MAD_DATA, GFP_ATOMIC);
+	if (IS_ERR(send_buf))
+		return;
+
+	smp = send_buf->mad;
+	smp->base_version = IB_MGMT_BASE_VERSION;
+	smp->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	smp->class_version = 1;
+	smp->method = IB_MGMT_METHOD_TRAP;
+	ibp->tid++;
+	smp->tid = cpu_to_be64(ibp->tid);
+	smp->attr_id = IB_SMP_ATTR_NOTICE;
+	/* o14-1: smp->mkey = 0; */
+	memcpy(smp->data, data, len);
+
+	spin_lock_irqsave(&ibp->lock, flags);
+	if (!ibp->sm_ah) {
+		if (ibp->sm_lid != be16_to_cpu(IB_LID_PERMISSIVE)) {
+			struct ib_ah *ah;
+			struct ib_ah_attr attr;
+
+			memset(&attr, 0, sizeof attr);
+			attr.dlid = ibp->sm_lid;
+			attr.port_num = ppd_from_ibp(ibp)->port;
+			ah = ib_create_ah(ibp->qp0->ibqp.pd, &attr);
+			if (IS_ERR(ah))
+				ret = -EINVAL;
+			else {
+				send_buf->ah = ah;
+				ibp->sm_ah = to_iah(ah);
+				ret = 0;
+			}
+		} else
+			ret = -EINVAL;
+	} else {
+		send_buf->ah = &ibp->sm_ah->ibah;
+		ret = 0;
+	}
+	spin_unlock_irqrestore(&ibp->lock, flags);
+
+	if (!ret)
+		ret = ib_post_send_mad(send_buf, NULL);
+	if (!ret) {
+		/* 4.096 usec. */
+		timeout = (4096 * (1UL << ibp->subnet_timeout)) / 1000;
+		ibp->trap_timeout = jiffies + usecs_to_jiffies(timeout);
+	} else {
+		ib_free_send_mad(send_buf);
+		ibp->trap_timeout = 0;
+	}
+}
+
+/*
+ * Send a bad [PQ]_Key trap (ch. 14.3.8).
+ */
+void qib_bad_pqkey(struct qib_ibport *ibp, __be16 trap_num, u32 key, u32 sl,
+		   u32 qp1, u32 qp2, __be16 lid1, __be16 lid2)
+{
+	struct ib_mad_notice_attr data;
+
+	if (trap_num == IB_NOTICE_TRAP_BAD_PKEY)
+		ibp->pkey_violations++;
+	else
+		ibp->qkey_violations++;
+	ibp->n_pkt_drops++;
+
+	/* Send violation trap */
+	data.generic_type = IB_NOTICE_TYPE_SECURITY;
+	data.prod_type_msb = 0;
+	data.prod_type_lsb = IB_NOTICE_PROD_CA;
+	data.trap_num = trap_num;
+	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
+	data.toggle_count = 0;
+	memset(&data.details, 0, sizeof data.details);
+	data.details.ntc_257_258.lid1 = lid1;
+	data.details.ntc_257_258.lid2 = lid2;
+	data.details.ntc_257_258.key = cpu_to_be32(key);
+	data.details.ntc_257_258.sl_qp1 = cpu_to_be32((sl << 28) | qp1);
+	data.details.ntc_257_258.qp2 = cpu_to_be32(qp2);
+
+	qib_send_trap(ibp, &data, sizeof data);
+}
+
+/*
+ * Send a bad M_Key trap (ch. 14.3.9).
+ */
+static void qib_bad_mkey(struct qib_ibport *ibp, struct ib_smp *smp)
+{
+	struct ib_mad_notice_attr data;
+
+	/* Send violation trap */
+	data.generic_type = IB_NOTICE_TYPE_SECURITY;
+	data.prod_type_msb = 0;
+	data.prod_type_lsb = IB_NOTICE_PROD_CA;
+	data.trap_num = IB_NOTICE_TRAP_BAD_MKEY;
+	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
+	data.toggle_count = 0;
+	memset(&data.details, 0, sizeof data.details);
+	data.details.ntc_256.lid = data.issuer_lid;
+	data.details.ntc_256.method = smp->method;
+	data.details.ntc_256.attr_id = smp->attr_id;
+	data.details.ntc_256.attr_mod = smp->attr_mod;
+	data.details.ntc_256.mkey = smp->mkey;
+	if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
+		u8 hop_cnt;
+
+		data.details.ntc_256.dr_slid = smp->dr_slid;
+		data.details.ntc_256.dr_trunc_hop = IB_NOTICE_TRAP_DR_NOTICE;
+		hop_cnt = smp->hop_cnt;
+		if (hop_cnt > ARRAY_SIZE(data.details.ntc_256.dr_rtn_path)) {
+			data.details.ntc_256.dr_trunc_hop |=
+				IB_NOTICE_TRAP_DR_TRUNC;
+			hop_cnt = ARRAY_SIZE(data.details.ntc_256.dr_rtn_path);
+		}
+		data.details.ntc_256.dr_trunc_hop |= hop_cnt;
+		memcpy(data.details.ntc_256.dr_rtn_path, smp->return_path,
+		       hop_cnt);
+	}
+
+	qib_send_trap(ibp, &data, sizeof data);
+}
+
+/*
+ * Send a Port Capability Mask Changed trap (ch. 14.3.11).
+ */
+void qib_cap_mask_chg(struct qib_ibport *ibp)
+{
+	struct ib_mad_notice_attr data;
+
+	data.generic_type = IB_NOTICE_TYPE_INFO;
+	data.prod_type_msb = 0;
+	data.prod_type_lsb = IB_NOTICE_PROD_CA;
+	data.trap_num = IB_NOTICE_TRAP_CAP_MASK_CHG;
+	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
+	data.toggle_count = 0;
+	memset(&data.details, 0, sizeof data.details);
+	data.details.ntc_144.lid = data.issuer_lid;
+	data.details.ntc_144.new_cap_mask = cpu_to_be32(ibp->port_cap_flags);
+
+	qib_send_trap(ibp, &data, sizeof data);
+}
+
+/*
+ * Send a System Image GUID Changed trap (ch. 14.3.12).
+ */
+void qib_sys_guid_chg(struct qib_ibport *ibp)
+{
+	struct ib_mad_notice_attr data;
+
+	data.generic_type = IB_NOTICE_TYPE_INFO;
+	data.prod_type_msb = 0;
+	data.prod_type_lsb = IB_NOTICE_PROD_CA;
+	data.trap_num = IB_NOTICE_TRAP_SYS_GUID_CHG;
+	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
+	data.toggle_count = 0;
+	memset(&data.details, 0, sizeof data.details);
+	data.details.ntc_145.lid = data.issuer_lid;
+	data.details.ntc_145.new_sys_guid = ib_qib_sys_image_guid;
+
+	qib_send_trap(ibp, &data, sizeof data);
+}
+
+/*
+ * Send a Node Description Changed trap (ch. 14.3.13).
+ */
+void qib_node_desc_chg(struct qib_ibport *ibp)
+{
+	struct ib_mad_notice_attr data;
+
+	data.generic_type = IB_NOTICE_TYPE_INFO;
+	data.prod_type_msb = 0;
+	data.prod_type_lsb = IB_NOTICE_PROD_CA;
+	data.trap_num = IB_NOTICE_TRAP_CAP_MASK_CHG;
+	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
+	data.toggle_count = 0;
+	memset(&data.details, 0, sizeof data.details);
+	data.details.ntc_144.lid = data.issuer_lid;
+	data.details.ntc_144.local_changes = 1;
+	data.details.ntc_144.change_flags = IB_NOTICE_TRAP_NODE_DESC_CHG;
+
+	qib_send_trap(ibp, &data, sizeof data);
+}
+
+static int subn_get_nodedescription(struct ib_smp *smp,
+				    struct ib_device *ibdev)
+{
+	if (smp->attr_mod)
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	memcpy(smp->data, ibdev->node_desc, sizeof(smp->data));
+
+	return reply(smp);
+}
+
+static int subn_get_nodeinfo(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct ib_node_info *nip = (struct ib_node_info *)&smp->data;
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	u32 vendor, majrev, minrev;
+	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
+
+	/* GUID 0 is illegal */
+	if (smp->attr_mod || pidx >= dd->num_pports ||
+	    dd->pport[pidx].guid == 0)
+		smp->status |= IB_SMP_INVALID_FIELD;
+	else
+		nip->port_guid = dd->pport[pidx].guid;
+
+	nip->base_version = 1;
+	nip->class_version = 1;
+	nip->node_type = 1;     /* channel adapter */
+	nip->num_ports = ibdev->phys_port_cnt;
+	/* This is already in network order */
+	nip->sys_guid = ib_qib_sys_image_guid;
+	nip->node_guid = dd->pport->guid; /* Use first-port GUID as node */
+	nip->partition_cap = cpu_to_be16(qib_get_npkeys(dd));
+	nip->device_id = cpu_to_be16(dd->deviceid);
+	majrev = dd->majrev;
+	minrev = dd->minrev;
+	nip->revision = cpu_to_be32((majrev << 16) | minrev);
+	nip->local_port_num = port;
+	vendor = dd->vendorid;
+	nip->vendor_id[0] = QIB_SRC_OUI_1;
+	nip->vendor_id[1] = QIB_SRC_OUI_2;
+	nip->vendor_id[2] = QIB_SRC_OUI_3;
+
+	return reply(smp);
+}
+
+static int subn_get_guidinfo(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	u32 startgx = 8 * be32_to_cpu(smp->attr_mod);
+	__be64 *p = (__be64 *) smp->data;
+	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
+
+	/* 32 blocks of 8 64-bit GUIDs per block */
+
+	memset(smp->data, 0, sizeof(smp->data));
+
+	if (startgx == 0 && pidx < dd->num_pports) {
+		struct qib_pportdata *ppd = dd->pport + pidx;
+		struct qib_ibport *ibp = &ppd->ibport_data;
+		__be64 g = ppd->guid;
+		unsigned i;
+
+		/* GUID 0 is illegal */
+		if (g == 0)
+			smp->status |= IB_SMP_INVALID_FIELD;
+		else {
+			/* The first is a copy of the read-only HW GUID. */
+			p[0] = g;
+			for (i = 1; i < QIB_GUIDS_PER_PORT; i++)
+				p[i] = ibp->guids[i - 1];
+		}
+	} else
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	return reply(smp);
+}
+
+static void set_link_width_enabled(struct qib_pportdata *ppd, u32 w)
+{
+	(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LWID_ENB, w);
+}
+
+static void set_link_speed_enabled(struct qib_pportdata *ppd, u32 s)
+{
+	(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_SPD_ENB, s);
+}
+
+static int get_overrunthreshold(struct qib_pportdata *ppd)
+{
+	return ppd->dd->f_get_ib_cfg(ppd, QIB_IB_CFG_OVERRUN_THRESH);
+}
+
+/**
+ * set_overrunthreshold - set the overrun threshold
+ * @ppd: the physical port data
+ * @n: the new threshold
+ *
+ * Note that this will only take effect when the link state changes.
+ */
+static int set_overrunthreshold(struct qib_pportdata *ppd, unsigned n)
+{
+	(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_OVERRUN_THRESH,
+					 (u32)n);
+	return 0;
+}
+
+static int get_phyerrthreshold(struct qib_pportdata *ppd)
+{
+	return ppd->dd->f_get_ib_cfg(ppd, QIB_IB_CFG_PHYERR_THRESH);
+}
+
+/**
+ * set_phyerrthreshold - set the physical error threshold
+ * @ppd: the physical port data
+ * @n: the new threshold
+ *
+ * Note that this will only take effect when the link state changes.
+ */
+static int set_phyerrthreshold(struct qib_pportdata *ppd, unsigned n)
+{
+	(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_PHYERR_THRESH,
+					 (u32)n);
+	return 0;
+}
+
+/**
+ * get_linkdowndefaultstate - get the default linkdown state
+ * @ppd: the physical port data
+ *
+ * Returns zero if the default is POLL, 1 if the default is SLEEP.
+ */
+static int get_linkdowndefaultstate(struct qib_pportdata *ppd)
+{
+	return ppd->dd->f_get_ib_cfg(ppd, QIB_IB_CFG_LINKDEFAULT) ==
+		IB_LINKINITCMD_SLEEP;
+}
+
+static int check_mkey(struct qib_ibport *ibp, struct ib_smp *smp, int mad_flags)
+{
+	int ret = 0;
+
+	/* Is the mkey in the process of expiring? */
+	if (ibp->mkey_lease_timeout &&
+	    time_after_eq(jiffies, ibp->mkey_lease_timeout)) {
+		/* Clear timeout and mkey protection field. */
+		ibp->mkey_lease_timeout = 0;
+		ibp->mkeyprot = 0;
+	}
+
+	/* M_Key checking depends on Portinfo:M_Key_protect_bits */
+	if ((mad_flags & IB_MAD_IGNORE_MKEY) == 0 && ibp->mkey != 0 &&
+	    ibp->mkey != smp->mkey &&
+	    (smp->method == IB_MGMT_METHOD_SET ||
+	     smp->method == IB_MGMT_METHOD_TRAP_REPRESS ||
+	     (smp->method == IB_MGMT_METHOD_GET && ibp->mkeyprot >= 2))) {
+		if (ibp->mkey_violations != 0xFFFF)
+			++ibp->mkey_violations;
+		if (!ibp->mkey_lease_timeout && ibp->mkey_lease_period)
+			ibp->mkey_lease_timeout = jiffies +
+				ibp->mkey_lease_period * HZ;
+		/* Generate a trap notice. */
+		qib_bad_mkey(ibp, smp);
+		ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+	} else if (ibp->mkey_lease_timeout)
+		ibp->mkey_lease_timeout = 0;
+
+	return ret;
+}
+
+static int subn_get_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct qib_devdata *dd;
+	struct qib_pportdata *ppd;
+	struct qib_ibport *ibp;
+	struct ib_port_info *pip = (struct ib_port_info *)smp->data;
+	u16 lid;
+	u8 mtu;
+	int ret;
+	u32 state;
+	u32 port_num = be32_to_cpu(smp->attr_mod);
+
+	if (port_num == 0)
+		port_num = port;
+	else {
+		if (port_num > ibdev->phys_port_cnt) {
+			smp->status |= IB_SMP_INVALID_FIELD;
+			ret = reply(smp);
+			goto bail;
+		}
+		if (port_num != port) {
+			ibp = to_iport(ibdev, port_num);
+			ret = check_mkey(ibp, smp, 0);
+			if (ret)
+				goto bail;
+		}
+	}
+
+	dd = dd_from_ibdev(ibdev);
+	/* IB numbers ports from 1, hdw from 0 */
+	ppd = dd->pport + (port_num - 1);
+	ibp = &ppd->ibport_data;
+
+	/* Clear all fields.  Only set the non-zero fields. */
+	memset(smp->data, 0, sizeof(smp->data));
+
+	/* Only return the mkey if the protection field allows it. */
+	if (smp->method == IB_MGMT_METHOD_SET || ibp->mkey == smp->mkey ||
+	    ibp->mkeyprot == 0)
+		pip->mkey = ibp->mkey;
+	pip->gid_prefix = ibp->gid_prefix;
+	lid = ppd->lid;
+	pip->lid = lid ? cpu_to_be16(lid) : IB_LID_PERMISSIVE;
+	pip->sm_lid = cpu_to_be16(ibp->sm_lid);
+	pip->cap_mask = cpu_to_be32(ibp->port_cap_flags);
+	/* pip->diag_code; */
+	pip->mkey_lease_period = cpu_to_be16(ibp->mkey_lease_period);
+	pip->local_port_num = port;
+	pip->link_width_enabled = ppd->link_width_enabled;
+	pip->link_width_supported = ppd->link_width_supported;
+	pip->link_width_active = ppd->link_width_active;
+	state = dd->f_iblink_state(ppd->lastibcstat);
+	pip->linkspeed_portstate = ppd->link_speed_supported << 4 | state;
+
+	pip->portphysstate_linkdown =
+		(dd->f_ibphys_portstate(ppd->lastibcstat) << 4) |
+		(get_linkdowndefaultstate(ppd) ? 1 : 2);
+	pip->mkeyprot_resv_lmc = (ibp->mkeyprot << 6) | ppd->lmc;
+	pip->linkspeedactive_enabled = (ppd->link_speed_active << 4) |
+		ppd->link_speed_enabled;
+	switch (ppd->ibmtu) {
+	default: /* something is wrong; fall through */
+	case 4096:
+		mtu = IB_MTU_4096;
+		break;
+	case 2048:
+		mtu = IB_MTU_2048;
+		break;
+	case 1024:
+		mtu = IB_MTU_1024;
+		break;
+	case 512:
+		mtu = IB_MTU_512;
+		break;
+	case 256:
+		mtu = IB_MTU_256;
+		break;
+	}
+	pip->neighbormtu_mastersmsl = (mtu << 4) | ibp->sm_sl;
+	pip->vlcap_inittype = ppd->vls_supported << 4;  /* InitType = 0 */
+	pip->vl_high_limit = ibp->vl_high_limit;
+	pip->vl_arb_high_cap =
+		dd->f_get_ib_cfg(ppd, QIB_IB_CFG_VL_HIGH_CAP);
+	pip->vl_arb_low_cap =
+		dd->f_get_ib_cfg(ppd, QIB_IB_CFG_VL_LOW_CAP);
+	/* InitTypeReply = 0 */
+	pip->inittypereply_mtucap = qib_ibmtu ? qib_ibmtu : IB_MTU_4096;
+	/* HCAs ignore VLStallCount and HOQLife */
+	/* pip->vlstallcnt_hoqlife; */
+	pip->operationalvl_pei_peo_fpi_fpo =
+		dd->f_get_ib_cfg(ppd, QIB_IB_CFG_OP_VLS) << 4;
+	pip->mkey_violations = cpu_to_be16(ibp->mkey_violations);
+	/* P_KeyViolations are counted by hardware. */
+	pip->pkey_violations = cpu_to_be16(ibp->pkey_violations);
+	pip->qkey_violations = cpu_to_be16(ibp->qkey_violations);
+	/* Only the hardware GUID is supported for now */
+	pip->guid_cap = QIB_GUIDS_PER_PORT;
+	pip->clientrereg_resv_subnetto = ibp->subnet_timeout;
+	/* 32.768 usec. response time (guessing) */
+	pip->resv_resptimevalue = 3;
+	pip->localphyerrors_overrunerrors =
+		(get_phyerrthreshold(ppd) << 4) |
+		get_overrunthreshold(ppd);
+	/* pip->max_credit_hint; */
+	if (ibp->port_cap_flags & IB_PORT_LINK_LATENCY_SUP) {
+		u32 v;
+
+		v = dd->f_get_ib_cfg(ppd, QIB_IB_CFG_LINKLATENCY);
+		pip->link_roundtrip_latency[0] = v >> 16;
+		pip->link_roundtrip_latency[1] = v >> 8;
+		pip->link_roundtrip_latency[2] = v;
+	}
+
+	ret = reply(smp);
+
+bail:
+	return ret;
+}
+
+/**
+ * get_pkeys - return the PKEY table
+ * @dd: the qlogic_ib device
+ * @port: the IB port number
+ * @pkeys: the pkey table is placed here
+ */
+static int get_pkeys(struct qib_devdata *dd, u8 port, u16 *pkeys)
+{
+	struct qib_pportdata *ppd = dd->pport + port - 1;
+	/*
+	 * always a kernel context, no locking needed.
+	 * If we get here with ppd setup, no need to check
+	 * that pd is valid.
+	 */
+	struct qib_ctxtdata *rcd = dd->rcd[ppd->hw_pidx];
+
+	memcpy(pkeys, rcd->pkeys, sizeof(rcd->pkeys));
+
+	return 0;
+}
+
+static int subn_get_pkeytable(struct ib_smp *smp, struct ib_device *ibdev,
+			      u8 port)
+{
+	u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
+	u16 *p = (u16 *) smp->data;
+	__be16 *q = (__be16 *) smp->data;
+
+	/* 64 blocks of 32 16-bit P_Key entries */
+
+	memset(smp->data, 0, sizeof(smp->data));
+	if (startpx == 0) {
+		struct qib_devdata *dd = dd_from_ibdev(ibdev);
+		unsigned i, n = qib_get_npkeys(dd);
+
+		get_pkeys(dd, port, p);
+
+		for (i = 0; i < n; i++)
+			q[i] = cpu_to_be16(p[i]);
+	} else
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	return reply(smp);
+}
+
+static int subn_set_guidinfo(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	u32 startgx = 8 * be32_to_cpu(smp->attr_mod);
+	__be64 *p = (__be64 *) smp->data;
+	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
+
+	/* 32 blocks of 8 64-bit GUIDs per block */
+
+	if (startgx == 0 && pidx < dd->num_pports) {
+		struct qib_pportdata *ppd = dd->pport + pidx;
+		struct qib_ibport *ibp = &ppd->ibport_data;
+		unsigned i;
+
+		/* The first entry is read-only. */
+		for (i = 1; i < QIB_GUIDS_PER_PORT; i++)
+			ibp->guids[i - 1] = p[i];
+	} else
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	/* The only GUID we support is the first read-only entry. */
+	return subn_get_guidinfo(smp, ibdev, port);
+}
+
+/**
+ * subn_set_portinfo - set port information
+ * @smp: the incoming SM packet
+ * @ibdev: the infiniband device
+ * @port: the port on the device
+ *
+ * Set Portinfo (see ch. 14.2.5.6).
+ */
+static int subn_set_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct ib_port_info *pip = (struct ib_port_info *)smp->data;
+	struct ib_event event;
+	struct qib_devdata *dd;
+	struct qib_pportdata *ppd;
+	struct qib_ibport *ibp;
+	char clientrereg = 0;
+	unsigned long flags;
+	u16 lid, smlid;
+	u8 lwe;
+	u8 lse;
+	u8 state;
+	u8 vls;
+	u16 lstate;
+	int ret, ore, mtu;
+	u32 port_num = be32_to_cpu(smp->attr_mod);
+
+	if (port_num == 0)
+		port_num = port;
+	else {
+		if (port_num > ibdev->phys_port_cnt)
+			goto err;
+		/* Port attributes can only be set on the receiving port */
+		if (port_num != port)
+			goto get_only;
+	}
+
+	dd = dd_from_ibdev(ibdev);
+	/* IB numbers ports from 1, hdw from 0 */
+	ppd = dd->pport + (port_num - 1);
+	ibp = &ppd->ibport_data;
+	event.device = ibdev;
+	event.element.port_num = port;
+
+	ibp->mkey = pip->mkey;
+	ibp->gid_prefix = pip->gid_prefix;
+	ibp->mkey_lease_period = be16_to_cpu(pip->mkey_lease_period);
+
+	lid = be16_to_cpu(pip->lid);
+	/* Must be a valid unicast LID address. */
+	if (lid == 0 || lid >= QIB_MULTICAST_LID_BASE)
+		goto err;
+	if (ppd->lid != lid || ppd->lmc != (pip->mkeyprot_resv_lmc & 7)) {
+		if (ppd->lid != lid)
+			qib_set_uevent_bits(ppd, _QIB_EVENT_LID_CHANGE_BIT);
+		if (ppd->lmc != (pip->mkeyprot_resv_lmc & 7))
+			qib_set_uevent_bits(ppd, _QIB_EVENT_LMC_CHANGE_BIT);
+		qib_set_lid(ppd, lid, pip->mkeyprot_resv_lmc & 7);
+		event.event = IB_EVENT_LID_CHANGE;
+		ib_dispatch_event(&event);
+	}
+
+	smlid = be16_to_cpu(pip->sm_lid);
+	/* Must be a valid unicast LID address. */
+	if (smlid == 0 || smlid >= QIB_MULTICAST_LID_BASE)
+		goto err;
+	if (smlid != ibp->sm_lid) {
+		spin_lock_irqsave(&ibp->lock, flags);
+		if (ibp->sm_ah)
+			ibp->sm_ah->attr.dlid = smlid;
+		spin_unlock_irqrestore(&ibp->lock, flags);
+		ibp->sm_lid = smlid;
+		event.event = IB_EVENT_SM_CHANGE;
+		ib_dispatch_event(&event);
+	}
+
+	/* Allow 1x or 4x to be set (see 14.2.6.6). */
+	lwe = pip->link_width_enabled;
+	if (lwe) {
+		if (lwe == 0xFF)
+			lwe = ppd->link_width_supported;
+		else if (lwe >= 16 || (lwe & ~ppd->link_width_supported))
+			goto err;
+		set_link_width_enabled(ppd, lwe);
+	}
+
+	lse = pip->linkspeedactive_enabled & 0xF;
+	if (lse) {
+		/*
+		 * The IB 1.2 spec. only allows link speed values
+		 * 1, 3, 5, 7, 15.  1.2.1 extended to allow specific
+		 * speeds.
+		 */
+		if (lse == 15)
+			lse = ppd->link_speed_supported;
+		else if (lse >= 8 || (lse & ~ppd->link_speed_supported))
+			goto err;
+		set_link_speed_enabled(ppd, lse);
+	}
+
+	/* Set link down default state. */
+	switch (pip->portphysstate_linkdown & 0xF) {
+	case 0: /* NOP */
+		break;
+	case 1: /* SLEEP */
+		(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LINKDEFAULT,
+					IB_LINKINITCMD_SLEEP);
+		break;
+	case 2: /* POLL */
+		(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LINKDEFAULT,
+					IB_LINKINITCMD_POLL);
+		break;
+	default:
+		goto err;
+	}
+
+	ibp->mkeyprot = pip->mkeyprot_resv_lmc >> 6;
+	ibp->vl_high_limit = pip->vl_high_limit;
+	(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_VL_HIGH_LIMIT,
+				    ibp->vl_high_limit);
+
+	mtu = ib_mtu_enum_to_int((pip->neighbormtu_mastersmsl >> 4) & 0xF);
+	if (mtu == -1)
+		goto err;
+	qib_set_mtu(ppd, mtu);
+
+	ibp->sm_sl = pip->neighbormtu_mastersmsl & 0xF;
+
+	/* Set operational VLs */
+	vls = (pip->operationalvl_pei_peo_fpi_fpo >> 4) & 0xF;
+	if (vls) {
+		if (vls > ppd->vls_supported)
+			goto err;
+		(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_OP_VLS, vls);
+	}
+
+	if (pip->mkey_violations == 0)
+		ibp->mkey_violations = 0;
+
+	if (pip->pkey_violations == 0)
+		ibp->pkey_violations = 0;
+
+	if (pip->qkey_violations == 0)
+		ibp->qkey_violations = 0;
+
+	ore = pip->localphyerrors_overrunerrors;
+	if (set_phyerrthreshold(ppd, (ore >> 4) & 0xF))
+		goto err;
+
+	if (set_overrunthreshold(ppd, (ore & 0xF)))
+		goto err;
+
+	ibp->subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F;
+
+	if (pip->clientrereg_resv_subnetto & 0x80) {
+		clientrereg = 1;
+		event.event = IB_EVENT_CLIENT_REREGISTER;
+		ib_dispatch_event(&event);
+	}
+
+	/*
+	 * Do the port state change now that the other link parameters
+	 * have been set.
+	 * Changing the port physical state only makes sense if the link
+	 * is down or is being set to down.
+	 */
+	state = pip->linkspeed_portstate & 0xF;
+	lstate = (pip->portphysstate_linkdown >> 4) & 0xF;
+	if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP))
+		goto err;
+
+	qib_dbg("IB%u:%u state %u lstate %u lflags %x\n",
+		dd->unit, ppd->port, state, lstate, ppd->lflags);
+	/*
+	 * Only state changes of DOWN, ARM, and ACTIVE are valid
+	 * and must be in the correct state to take effect (see 7.2.6).
+	 */
+	switch (state) {
+	case IB_PORT_NOP:
+		if (lstate == 0)
+			break;
+		/* FALLTHROUGH */
+	case IB_PORT_DOWN:
+		if (lstate == 0)
+			lstate = QIB_IB_LINKDOWN_ONLY;
+		else if (lstate == 1)
+			lstate = QIB_IB_LINKDOWN_SLEEP;
+		else if (lstate == 2)
+			lstate = QIB_IB_LINKDOWN;
+		else if (lstate == 3)
+			lstate = QIB_IB_LINKDOWN_DISABLE;
+		else
+			goto err;
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~QIBL_LINKV;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		qib_set_linkstate(ppd, lstate);
+		/*
+		 * Don't send a reply if the response would be sent
+		 * through the disabled port.
+		 */
+		if (lstate == QIB_IB_LINKDOWN_DISABLE && smp->hop_cnt) {
+			ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+			goto done;
+		}
+		qib_wait_linkstate(ppd, QIBL_LINKV, 10);
+		break;
+	case IB_PORT_ARMED:
+		qib_set_linkstate(ppd, QIB_IB_LINKARM);
+		break;
+	case IB_PORT_ACTIVE:
+		qib_set_linkstate(ppd, QIB_IB_LINKACTIVE);
+		break;
+	default:
+		/* XXX We have already partially updated our state! */
+		goto err;
+	}
+
+	ret = subn_get_portinfo(smp, ibdev, port);
+
+	if (clientrereg)
+		pip->clientrereg_resv_subnetto |= 0x80;
+
+	goto done;
+
+err:
+	qib_dbg("invalid field\n");
+	smp->status |= IB_SMP_INVALID_FIELD;
+get_only:
+	ret = subn_get_portinfo(smp, ibdev, port);
+done:
+	return ret;
+}
+
+/**
+ * rm_pkey - decrecment the reference count for the given PKEY
+ * @dd: the qlogic_ib device
+ * @key: the PKEY index
+ *
+ * Return true if this was the last reference and the hardware table entry
+ * needs to be changed.
+ */
+static int rm_pkey(struct qib_pportdata *ppd, u16 key)
+{
+	int i;
+	int ret;
+
+	for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
+		if (ppd->pkeys[i] != key)
+			continue;
+		if (atomic_dec_and_test(&ppd->pkeyrefs[i])) {
+			ppd->pkeys[i] = 0;
+			ret = 1;
+			goto bail;
+		}
+		break;
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * add_pkey - add the given PKEY to the hardware table
+ * @dd: the qlogic_ib device
+ * @key: the PKEY
+ *
+ * Return an error code if unable to add the entry, zero if no change,
+ * or 1 if the hardware PKEY register needs to be updated.
+ */
+static int add_pkey(struct qib_pportdata *ppd, u16 key)
+{
+	int i;
+	u16 lkey = key & 0x7FFF;
+	int any = 0;
+	int ret;
+
+	if (lkey == 0x7FFF) {
+		ret = 0;
+		goto bail;
+	}
+
+	/* Look for an empty slot or a matching PKEY. */
+	for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
+		if (!ppd->pkeys[i]) {
+			any++;
+			continue;
+		}
+		/* If it matches exactly, try to increment the ref count */
+		if (ppd->pkeys[i] == key) {
+			if (atomic_inc_return(&ppd->pkeyrefs[i]) > 1) {
+				ret = 0;
+				goto bail;
+			}
+			/* Lost the race. Look for an empty slot below. */
+			atomic_dec(&ppd->pkeyrefs[i]);
+			any++;
+		}
+		/*
+		 * It makes no sense to have both the limited and unlimited
+		 * PKEY set at the same time since the unlimited one will
+		 * disable the limited one.
+		 */
+		if ((ppd->pkeys[i] & 0x7FFF) == lkey) {
+			ret = -EEXIST;
+			goto bail;
+		}
+	}
+	if (!any) {
+		ret = -EBUSY;
+		goto bail;
+	}
+	for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
+		if (!ppd->pkeys[i] &&
+		    atomic_inc_return(&ppd->pkeyrefs[i]) == 1) {
+			/* for qibstats, etc. */
+			ppd->pkeys[i] = key;
+			ret = 1;
+			goto bail;
+		}
+	}
+	ret = -EBUSY;
+
+bail:
+	return ret;
+}
+
+/**
+ * set_pkeys - set the PKEY table for ctxt 0
+ * @dd: the qlogic_ib device
+ * @port: the IB port number
+ * @pkeys: the PKEY table
+ */
+static int set_pkeys(struct qib_devdata *dd, u8 port, u16 *pkeys)
+{
+	struct qib_pportdata *ppd;
+	struct qib_ctxtdata *rcd;
+	int i;
+	int changed = 0;
+
+	/*
+	 * IB port one/two always maps to context zero/one,
+	 * always a kernel context, no locking needed
+	 * If we get here with ppd setup, no need to check
+	 * that rcd is valid.
+	 */
+	ppd = dd->pport + (port - 1);
+	rcd = dd->rcd[ppd->hw_pidx];
+
+	for (i = 0; i < ARRAY_SIZE(rcd->pkeys); i++) {
+		u16 key = pkeys[i];
+		u16 okey = rcd->pkeys[i];
+
+		if (key == okey)
+			continue;
+		/*
+		 * The value of this PKEY table entry is changing.
+		 * Remove the old entry in the hardware's array of PKEYs.
+		 */
+		if (okey & 0x7FFF)
+			changed |= rm_pkey(ppd, okey);
+		if (key & 0x7FFF) {
+			int ret = add_pkey(ppd, key);
+
+			if (ret < 0)
+				key = 0;
+			else
+				changed |= ret;
+		}
+		rcd->pkeys[i] = key;
+	}
+	if (changed) {
+		struct ib_event event;
+
+		(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_PKEYS, 0);
+
+		event.event = IB_EVENT_PKEY_CHANGE;
+		event.device = &dd->verbs_dev.ibdev;
+		event.element.port_num = 1;
+		ib_dispatch_event(&event);
+	}
+	return 0;
+}
+
+static int subn_set_pkeytable(struct ib_smp *smp, struct ib_device *ibdev,
+			      u8 port)
+{
+	u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
+	__be16 *p = (__be16 *) smp->data;
+	u16 *q = (u16 *) smp->data;
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	unsigned i, n = qib_get_npkeys(dd);
+
+	for (i = 0; i < n; i++)
+		q[i] = be16_to_cpu(p[i]);
+
+	if (startpx != 0 || set_pkeys(dd, port, q) != 0)
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	return subn_get_pkeytable(smp, ibdev, port);
+}
+
+static int subn_get_sl_to_vl(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	u8 *p = (u8 *) smp->data;
+	unsigned i;
+
+	memset(smp->data, 0, sizeof(smp->data));
+
+	if (!(ibp->port_cap_flags & IB_PORT_SL_MAP_SUP))
+		smp->status |= IB_SMP_UNSUP_METHOD;
+	else
+		for (i = 0; i < ARRAY_SIZE(ibp->sl_to_vl); i += 2)
+			*p++ = (ibp->sl_to_vl[i] << 4) | ibp->sl_to_vl[i + 1];
+
+	return reply(smp);
+}
+
+static int subn_set_sl_to_vl(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	u8 *p = (u8 *) smp->data;
+	unsigned i;
+
+	if (!(ibp->port_cap_flags & IB_PORT_SL_MAP_SUP)) {
+		smp->status |= IB_SMP_UNSUP_METHOD;
+		return reply(smp);
+	}
+
+	for (i = 0; i < ARRAY_SIZE(ibp->sl_to_vl); i += 2, p++) {
+		ibp->sl_to_vl[i] = *p >> 4;
+		ibp->sl_to_vl[i + 1] = *p & 0xF;
+	}
+	qib_set_uevent_bits(ppd_from_ibp(to_iport(ibdev, port)),
+			    _QIB_EVENT_SL2VL_CHANGE_BIT);
+
+	return subn_get_sl_to_vl(smp, ibdev, port);
+}
+
+static int subn_get_vl_arb(struct ib_smp *smp, struct ib_device *ibdev,
+			   u8 port)
+{
+	unsigned which = be32_to_cpu(smp->attr_mod) >> 16;
+	struct qib_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
+
+	memset(smp->data, 0, sizeof(smp->data));
+
+	if (ppd->vls_supported == IB_VL_VL0)
+		smp->status |= IB_SMP_UNSUP_METHOD;
+	else if (which == IB_VLARB_LOWPRI_0_31)
+		(void) ppd->dd->f_get_ib_table(ppd, QIB_IB_TBL_VL_LOW_ARB,
+						   smp->data);
+	else if (which == IB_VLARB_HIGHPRI_0_31)
+		(void) ppd->dd->f_get_ib_table(ppd, QIB_IB_TBL_VL_HIGH_ARB,
+						   smp->data);
+	else
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	return reply(smp);
+}
+
+static int subn_set_vl_arb(struct ib_smp *smp, struct ib_device *ibdev,
+			   u8 port)
+{
+	unsigned which = be32_to_cpu(smp->attr_mod) >> 16;
+	struct qib_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
+
+	if (ppd->vls_supported == IB_VL_VL0)
+		smp->status |= IB_SMP_UNSUP_METHOD;
+	else if (which == IB_VLARB_LOWPRI_0_31)
+		(void) ppd->dd->f_set_ib_table(ppd, QIB_IB_TBL_VL_LOW_ARB,
+						   smp->data);
+	else if (which == IB_VLARB_HIGHPRI_0_31)
+		(void) ppd->dd->f_set_ib_table(ppd, QIB_IB_TBL_VL_HIGH_ARB,
+						   smp->data);
+	else
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	return subn_get_vl_arb(smp, ibdev, port);
+}
+
+static int subn_trap_repress(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	/*
+	 * For now, we only send the trap once so no need to process this.
+	 * o13-6, o13-7,
+	 * o14-3.a4 The SMA shall not send any message in response to a valid
+	 * SubnTrapRepress() message.
+	 */
+	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+}
+
+static int pma_get_classportinfo(struct ib_perf *pmp)
+{
+	struct ib_pma_classportinfo *p =
+		(struct ib_pma_classportinfo *)pmp->data;
+
+	memset(pmp->data, 0, sizeof(pmp->data));
+
+	if (pmp->attr_mod != 0)
+		pmp->status |= IB_SMP_INVALID_FIELD;
+
+	/* Note that AllPortSelect is not valid */
+	p->base_version = 1;
+	p->class_version = 1;
+	p->cap_mask = IB_PMA_CLASS_CAP_EXT_WIDTH;
+	/*
+	 * Expected response time is 4.096 usec. * 2^18 == 1.073741824 sec.
+	 */
+	p->resp_time_value = 18;
+
+	return reply((struct ib_smp *) pmp);
+}
+
+static int pma_get_portsamplescontrol(struct ib_perf *pmp,
+				      struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portsamplescontrol *p =
+		(struct ib_pma_portsamplescontrol *)pmp->data;
+	struct qib_ibdev *dev = to_idev(ibdev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	unsigned long flags;
+	u8 port_select = p->port_select;
+
+	memset(pmp->data, 0, sizeof(pmp->data));
+
+	p->port_select = port_select;
+	if (pmp->attr_mod != 0 || port_select != port) {
+		pmp->status |= IB_SMP_INVALID_FIELD;
+		goto bail;
+	}
+	spin_lock_irqsave(&ibp->lock, flags);
+	p->tick = dd->f_get_ib_cfg(ppd, QIB_IB_CFG_PMA_TICKS);
+	p->sample_status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
+	p->counter_width = 4;   /* 32 bit counters */
+	p->counter_mask0_9 = COUNTER_MASK0_9;
+	p->sample_start = cpu_to_be32(ibp->pma_sample_start);
+	p->sample_interval = cpu_to_be32(ibp->pma_sample_interval);
+	p->tag = cpu_to_be16(ibp->pma_tag);
+	p->counter_select[0] = ibp->pma_counter_select[0];
+	p->counter_select[1] = ibp->pma_counter_select[1];
+	p->counter_select[2] = ibp->pma_counter_select[2];
+	p->counter_select[3] = ibp->pma_counter_select[3];
+	p->counter_select[4] = ibp->pma_counter_select[4];
+	spin_unlock_irqrestore(&ibp->lock, flags);
+
+bail:
+	return reply((struct ib_smp *) pmp);
+}
+
+static int pma_set_portsamplescontrol(struct ib_perf *pmp,
+				      struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portsamplescontrol *p =
+		(struct ib_pma_portsamplescontrol *)pmp->data;
+	struct qib_ibdev *dev = to_idev(ibdev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	unsigned long flags;
+	u8 status;
+	int ret;
+
+	if (pmp->attr_mod != 0 || p->port_select != port) {
+		pmp->status |= IB_SMP_INVALID_FIELD;
+		ret = reply((struct ib_smp *) pmp);
+		goto bail;
+	}
+
+	spin_lock_irqsave(&ibp->lock, flags);
+	status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
+	if (status == IB_PMA_SAMPLE_STATUS_DONE) {
+		ibp->pma_sample_start = be32_to_cpu(p->sample_start);
+		ibp->pma_sample_interval = be32_to_cpu(p->sample_interval);
+		ibp->pma_tag = be16_to_cpu(p->tag);
+		ibp->pma_counter_select[0] = p->counter_select[0];
+		ibp->pma_counter_select[1] = p->counter_select[1];
+		ibp->pma_counter_select[2] = p->counter_select[2];
+		ibp->pma_counter_select[3] = p->counter_select[3];
+		ibp->pma_counter_select[4] = p->counter_select[4];
+		dd->f_set_cntr_sample(ppd, ibp->pma_sample_interval,
+				      ibp->pma_sample_start);
+	}
+	spin_unlock_irqrestore(&ibp->lock, flags);
+
+	ret = pma_get_portsamplescontrol(pmp, ibdev, port);
+
+bail:
+	return ret;
+}
+
+static u64 get_counter(struct qib_ibport *ibp, struct qib_pportdata *ppd,
+		       __be16 sel)
+{
+	u64 ret;
+
+	switch (sel) {
+	case IB_PMA_PORT_XMIT_DATA:
+		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSXMITDATA);
+		break;
+	case IB_PMA_PORT_RCV_DATA:
+		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSRCVDATA);
+		break;
+	case IB_PMA_PORT_XMIT_PKTS:
+		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSXMITPKTS);
+		break;
+	case IB_PMA_PORT_RCV_PKTS:
+		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSRCVPKTS);
+		break;
+	case IB_PMA_PORT_XMIT_WAIT:
+		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSXMITWAIT);
+		break;
+	default:
+		ret = 0;
+	}
+
+	return ret;
+}
+
+static int pma_get_portsamplesresult(struct ib_perf *pmp,
+				     struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portsamplesresult *p =
+		(struct ib_pma_portsamplesresult *)pmp->data;
+	struct qib_ibdev *dev = to_idev(ibdev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	unsigned long flags;
+	u8 status;
+	int i;
+
+	memset(pmp->data, 0, sizeof(pmp->data));
+	spin_lock_irqsave(&ibp->lock, flags);
+	p->tag = cpu_to_be16(ibp->pma_tag);
+	status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
+	p->sample_status = cpu_to_be16(status);
+	for (i = 0; i < ARRAY_SIZE(ibp->pma_counter_select); i++)
+		p->counter[i] = (status != IB_PMA_SAMPLE_STATUS_DONE) ? 0 :
+		    cpu_to_be32(
+			get_counter(ibp, ppd, ibp->pma_counter_select[i]));
+	spin_unlock_irqrestore(&ibp->lock, flags);
+
+	return reply((struct ib_smp *) pmp);
+}
+
+static int pma_get_portsamplesresult_ext(struct ib_perf *pmp,
+					 struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portsamplesresult_ext *p =
+		(struct ib_pma_portsamplesresult_ext *)pmp->data;
+	struct qib_ibdev *dev = to_idev(ibdev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	u8 status;
+	int i;
+
+	memset(pmp->data, 0, sizeof(pmp->data));
+	p->tag = cpu_to_be16(ibp->pma_tag);
+	status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
+	p->sample_status = cpu_to_be16(status);
+	/* 64 bits */
+	p->extended_width = cpu_to_be32(0x80000000);
+	for (i = 0; i < ARRAY_SIZE(ibp->pma_counter_select); i++)
+		p->counter[i] = (status != IB_PMA_SAMPLE_STATUS_DONE) ? 0 :
+		    cpu_to_be64(
+			get_counter(ibp, ppd, ibp->pma_counter_select[i]));
+
+	return reply((struct ib_smp *) pmp);
+}
+
+static int pma_get_portcounters(struct ib_perf *pmp,
+				struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
+		pmp->data;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct qib_verbs_counters cntrs;
+	u8 port_select = p->port_select;
+
+	qib_get_counters(ppd, &cntrs);
+
+	/* Adjust counters for any resets done. */
+	cntrs.symbol_error_counter -= ibp->z_symbol_error_counter;
+	cntrs.link_error_recovery_counter -=
+		ibp->z_link_error_recovery_counter;
+	cntrs.link_downed_counter -= ibp->z_link_downed_counter;
+	cntrs.port_rcv_errors -= ibp->z_port_rcv_errors;
+	cntrs.port_rcv_remphys_errors -= ibp->z_port_rcv_remphys_errors;
+	cntrs.port_xmit_discards -= ibp->z_port_xmit_discards;
+	cntrs.port_xmit_data -= ibp->z_port_xmit_data;
+	cntrs.port_rcv_data -= ibp->z_port_rcv_data;
+	cntrs.port_xmit_packets -= ibp->z_port_xmit_packets;
+	cntrs.port_rcv_packets -= ibp->z_port_rcv_packets;
+	cntrs.local_link_integrity_errors -=
+		ibp->z_local_link_integrity_errors;
+	cntrs.excessive_buffer_overrun_errors -=
+		ibp->z_excessive_buffer_overrun_errors;
+	cntrs.vl15_dropped -= ibp->z_vl15_dropped;
+	cntrs.vl15_dropped += ibp->n_vl15_dropped;
+
+	memset(pmp->data, 0, sizeof(pmp->data));
+
+	p->port_select = port_select;
+	if (pmp->attr_mod != 0 || port_select != port)
+		pmp->status |= IB_SMP_INVALID_FIELD;
+
+	if (cntrs.symbol_error_counter > 0xFFFFUL)
+		p->symbol_error_counter = cpu_to_be16(0xFFFF);
+	else
+		p->symbol_error_counter =
+			cpu_to_be16((u16)cntrs.symbol_error_counter);
+	if (cntrs.link_error_recovery_counter > 0xFFUL)
+		p->link_error_recovery_counter = 0xFF;
+	else
+		p->link_error_recovery_counter =
+			(u8)cntrs.link_error_recovery_counter;
+	if (cntrs.link_downed_counter > 0xFFUL)
+		p->link_downed_counter = 0xFF;
+	else
+		p->link_downed_counter = (u8)cntrs.link_downed_counter;
+	if (cntrs.port_rcv_errors > 0xFFFFUL)
+		p->port_rcv_errors = cpu_to_be16(0xFFFF);
+	else
+		p->port_rcv_errors =
+			cpu_to_be16((u16) cntrs.port_rcv_errors);
+	if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
+		p->port_rcv_remphys_errors = cpu_to_be16(0xFFFF);
+	else
+		p->port_rcv_remphys_errors =
+			cpu_to_be16((u16)cntrs.port_rcv_remphys_errors);
+	if (cntrs.port_xmit_discards > 0xFFFFUL)
+		p->port_xmit_discards = cpu_to_be16(0xFFFF);
+	else
+		p->port_xmit_discards =
+			cpu_to_be16((u16)cntrs.port_xmit_discards);
+	if (cntrs.local_link_integrity_errors > 0xFUL)
+		cntrs.local_link_integrity_errors = 0xFUL;
+	if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
+		cntrs.excessive_buffer_overrun_errors = 0xFUL;
+	p->lli_ebor_errors = (cntrs.local_link_integrity_errors << 4) |
+		cntrs.excessive_buffer_overrun_errors;
+	if (cntrs.vl15_dropped > 0xFFFFUL)
+		p->vl15_dropped = cpu_to_be16(0xFFFF);
+	else
+		p->vl15_dropped = cpu_to_be16((u16)cntrs.vl15_dropped);
+	if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
+		p->port_xmit_data = cpu_to_be32(0xFFFFFFFF);
+	else
+		p->port_xmit_data = cpu_to_be32((u32)cntrs.port_xmit_data);
+	if (cntrs.port_rcv_data > 0xFFFFFFFFUL)
+		p->port_rcv_data = cpu_to_be32(0xFFFFFFFF);
+	else
+		p->port_rcv_data = cpu_to_be32((u32)cntrs.port_rcv_data);
+	if (cntrs.port_xmit_packets > 0xFFFFFFFFUL)
+		p->port_xmit_packets = cpu_to_be32(0xFFFFFFFF);
+	else
+		p->port_xmit_packets =
+			cpu_to_be32((u32)cntrs.port_xmit_packets);
+	if (cntrs.port_rcv_packets > 0xFFFFFFFFUL)
+		p->port_rcv_packets = cpu_to_be32(0xFFFFFFFF);
+	else
+		p->port_rcv_packets =
+			cpu_to_be32((u32) cntrs.port_rcv_packets);
+
+	return reply((struct ib_smp *) pmp);
+}
+
+static int pma_get_portcounters_ext(struct ib_perf *pmp,
+				    struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portcounters_ext *p =
+		(struct ib_pma_portcounters_ext *)pmp->data;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	u64 swords, rwords, spkts, rpkts, xwait;
+	u8 port_select = p->port_select;
+
+	memset(pmp->data, 0, sizeof(pmp->data));
+
+	p->port_select = port_select;
+	if (pmp->attr_mod != 0 || port_select != port) {
+		pmp->status |= IB_SMP_INVALID_FIELD;
+		goto bail;
+	}
+
+	qib_snapshot_counters(ppd, &swords, &rwords, &spkts, &rpkts, &xwait);
+
+	/* Adjust counters for any resets done. */
+	swords -= ibp->z_port_xmit_data;
+	rwords -= ibp->z_port_rcv_data;
+	spkts -= ibp->z_port_xmit_packets;
+	rpkts -= ibp->z_port_rcv_packets;
+
+	p->port_xmit_data = cpu_to_be64(swords);
+	p->port_rcv_data = cpu_to_be64(rwords);
+	p->port_xmit_packets = cpu_to_be64(spkts);
+	p->port_rcv_packets = cpu_to_be64(rpkts);
+	p->port_unicast_xmit_packets = cpu_to_be64(ibp->n_unicast_xmit);
+	p->port_unicast_rcv_packets = cpu_to_be64(ibp->n_unicast_rcv);
+	p->port_multicast_xmit_packets = cpu_to_be64(ibp->n_multicast_xmit);
+	p->port_multicast_rcv_packets = cpu_to_be64(ibp->n_multicast_rcv);
+
+bail:
+	return reply((struct ib_smp *) pmp);
+}
+
+static int pma_set_portcounters(struct ib_perf *pmp,
+				struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
+		pmp->data;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct qib_verbs_counters cntrs;
+
+	/*
+	 * Since the HW doesn't support clearing counters, we save the
+	 * current count and subtract it from future responses.
+	 */
+	qib_get_counters(ppd, &cntrs);
+
+	if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR)
+		ibp->z_symbol_error_counter = cntrs.symbol_error_counter;
+
+	if (p->counter_select & IB_PMA_SEL_LINK_ERROR_RECOVERY)
+		ibp->z_link_error_recovery_counter =
+			cntrs.link_error_recovery_counter;
+
+	if (p->counter_select & IB_PMA_SEL_LINK_DOWNED)
+		ibp->z_link_downed_counter = cntrs.link_downed_counter;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_RCV_ERRORS)
+		ibp->z_port_rcv_errors = cntrs.port_rcv_errors;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS)
+		ibp->z_port_rcv_remphys_errors =
+			cntrs.port_rcv_remphys_errors;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
+		ibp->z_port_xmit_discards = cntrs.port_xmit_discards;
+
+	if (p->counter_select & IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS)
+		ibp->z_local_link_integrity_errors =
+			cntrs.local_link_integrity_errors;
+
+	if (p->counter_select & IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS)
+		ibp->z_excessive_buffer_overrun_errors =
+			cntrs.excessive_buffer_overrun_errors;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_VL15_DROPPED) {
+		ibp->n_vl15_dropped = 0;
+		ibp->z_vl15_dropped = cntrs.vl15_dropped;
+	}
+
+	if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
+		ibp->z_port_xmit_data = cntrs.port_xmit_data;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_RCV_DATA)
+		ibp->z_port_rcv_data = cntrs.port_rcv_data;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_XMIT_PACKETS)
+		ibp->z_port_xmit_packets = cntrs.port_xmit_packets;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_RCV_PACKETS)
+		ibp->z_port_rcv_packets = cntrs.port_rcv_packets;
+
+	return pma_get_portcounters(pmp, ibdev, port);
+}
+
+static int pma_set_portcounters_ext(struct ib_perf *pmp,
+				    struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
+		pmp->data;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	u64 swords, rwords, spkts, rpkts, xwait;
+
+	qib_snapshot_counters(ppd, &swords, &rwords, &spkts, &rpkts, &xwait);
+
+	if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA)
+		ibp->z_port_xmit_data = swords;
+
+	if (p->counter_select & IB_PMA_SELX_PORT_RCV_DATA)
+		ibp->z_port_rcv_data = rwords;
+
+	if (p->counter_select & IB_PMA_SELX_PORT_XMIT_PACKETS)
+		ibp->z_port_xmit_packets = spkts;
+
+	if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
+		ibp->z_port_rcv_packets = rpkts;
+
+	if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
+		ibp->n_unicast_xmit = 0;
+
+	if (p->counter_select & IB_PMA_SELX_PORT_UNI_RCV_PACKETS)
+		ibp->n_unicast_rcv = 0;
+
+	if (p->counter_select & IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS)
+		ibp->n_multicast_xmit = 0;
+
+	if (p->counter_select & IB_PMA_SELX_PORT_MULTI_RCV_PACKETS)
+		ibp->n_multicast_rcv = 0;
+
+	return pma_get_portcounters_ext(pmp, ibdev, port);
+}
+
+static int process_subn(struct ib_device *ibdev, int mad_flags,
+			u8 port, struct ib_mad *in_mad,
+			struct ib_mad *out_mad)
+{
+	struct ib_smp *smp = (struct ib_smp *)out_mad;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	int ret;
+
+	*out_mad = *in_mad;
+	if (smp->class_version != 1) {
+		smp->status |= IB_SMP_UNSUP_VERSION;
+		ret = reply(smp);
+		goto bail;
+	}
+
+	ret = check_mkey(ibp, smp, mad_flags);
+	if (ret) {
+		u32 port_num = be32_to_cpu(smp->attr_mod);
+
+		/*
+		 * If this is a get/set portinfo, we already check the
+		 * M_Key if the MAD is for another port and the M_Key
+		 * is OK on the receiving port. This check is needed
+		 * to increment the error counters when the M_Key
+		 * fails to match on *both* ports.
+		 */
+		if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO &&
+		    (smp->method == IB_MGMT_METHOD_GET ||
+		     smp->method == IB_MGMT_METHOD_SET) &&
+		    port_num && port_num <= ibdev->phys_port_cnt &&
+		    port != port_num)
+			(void) check_mkey(to_iport(ibdev, port_num), smp, 0);
+		goto bail;
+	}
+
+	switch (smp->method) {
+	case IB_MGMT_METHOD_GET:
+		switch (smp->attr_id) {
+		case IB_SMP_ATTR_NODE_DESC:
+			ret = subn_get_nodedescription(smp, ibdev);
+			goto bail;
+		case IB_SMP_ATTR_NODE_INFO:
+			ret = subn_get_nodeinfo(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_GUID_INFO:
+			ret = subn_get_guidinfo(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_PORT_INFO:
+			ret = subn_get_portinfo(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_PKEY_TABLE:
+			ret = subn_get_pkeytable(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_SL_TO_VL_TABLE:
+			ret = subn_get_sl_to_vl(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_VL_ARB_TABLE:
+			ret = subn_get_vl_arb(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_SM_INFO:
+			if (ibp->port_cap_flags & IB_PORT_SM_DISABLED) {
+				ret = IB_MAD_RESULT_SUCCESS |
+					IB_MAD_RESULT_CONSUMED;
+				goto bail;
+			}
+			if (ibp->port_cap_flags & IB_PORT_SM) {
+				ret = IB_MAD_RESULT_SUCCESS;
+				goto bail;
+			}
+			/* FALLTHROUGH */
+		default:
+			smp->status |= IB_SMP_UNSUP_METH_ATTR;
+			ret = reply(smp);
+			goto bail;
+		}
+
+	case IB_MGMT_METHOD_SET:
+		switch (smp->attr_id) {
+		case IB_SMP_ATTR_GUID_INFO:
+			ret = subn_set_guidinfo(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_PORT_INFO:
+			ret = subn_set_portinfo(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_PKEY_TABLE:
+			ret = subn_set_pkeytable(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_SL_TO_VL_TABLE:
+			ret = subn_set_sl_to_vl(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_VL_ARB_TABLE:
+			ret = subn_set_vl_arb(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_SM_INFO:
+			if (ibp->port_cap_flags & IB_PORT_SM_DISABLED) {
+				ret = IB_MAD_RESULT_SUCCESS |
+					IB_MAD_RESULT_CONSUMED;
+				goto bail;
+			}
+			if (ibp->port_cap_flags & IB_PORT_SM) {
+				ret = IB_MAD_RESULT_SUCCESS;
+				goto bail;
+			}
+			/* FALLTHROUGH */
+		default:
+			smp->status |= IB_SMP_UNSUP_METH_ATTR;
+			ret = reply(smp);
+			goto bail;
+		}
+
+	case IB_MGMT_METHOD_TRAP_REPRESS:
+		if (smp->attr_id == IB_SMP_ATTR_NOTICE)
+			ret = subn_trap_repress(smp, ibdev, port);
+		else {
+			smp->status |= IB_SMP_UNSUP_METH_ATTR;
+			ret = reply(smp);
+		}
+		goto bail;
+
+	case IB_MGMT_METHOD_TRAP:
+	case IB_MGMT_METHOD_REPORT:
+	case IB_MGMT_METHOD_REPORT_RESP:
+	case IB_MGMT_METHOD_GET_RESP:
+		/*
+		 * The ib_mad module will call us to process responses
+		 * before checking for other consumers.
+		 * Just tell the caller to process it normally.
+		 */
+		ret = IB_MAD_RESULT_SUCCESS;
+		goto bail;
+
+	case IB_MGMT_METHOD_SEND:
+		if (ib_get_smp_direction(smp) &&
+		    smp->attr_id == QIB_VENDOR_IPG) {
+			ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_PORT,
+					      smp->data[0]);
+			ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+		} else
+			ret = IB_MAD_RESULT_SUCCESS;
+		goto bail;
+
+	default:
+		smp->status |= IB_SMP_UNSUP_METHOD;
+		ret = reply(smp);
+	}
+
+bail:
+	return ret;
+}
+
+static int process_perf(struct ib_device *ibdev, u8 port,
+			struct ib_mad *in_mad,
+			struct ib_mad *out_mad)
+{
+	struct ib_perf *pmp = (struct ib_perf *)out_mad;
+	int ret;
+
+	*out_mad = *in_mad;
+	if (pmp->class_version != 1) {
+		pmp->status |= IB_SMP_UNSUP_VERSION;
+		ret = reply((struct ib_smp *) pmp);
+		goto bail;
+	}
+
+	switch (pmp->method) {
+	case IB_MGMT_METHOD_GET:
+		switch (pmp->attr_id) {
+		case IB_PMA_CLASS_PORT_INFO:
+			ret = pma_get_classportinfo(pmp);
+			goto bail;
+		case IB_PMA_PORT_SAMPLES_CONTROL:
+			ret = pma_get_portsamplescontrol(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_SAMPLES_RESULT:
+			ret = pma_get_portsamplesresult(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_SAMPLES_RESULT_EXT:
+			ret = pma_get_portsamplesresult_ext(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_COUNTERS:
+			ret = pma_get_portcounters(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_COUNTERS_EXT:
+			ret = pma_get_portcounters_ext(pmp, ibdev, port);
+			goto bail;
+		default:
+			pmp->status |= IB_SMP_UNSUP_METH_ATTR;
+			ret = reply((struct ib_smp *) pmp);
+			goto bail;
+		}
+
+	case IB_MGMT_METHOD_SET:
+		switch (pmp->attr_id) {
+		case IB_PMA_PORT_SAMPLES_CONTROL:
+			ret = pma_set_portsamplescontrol(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_COUNTERS:
+			ret = pma_set_portcounters(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_COUNTERS_EXT:
+			ret = pma_set_portcounters_ext(pmp, ibdev, port);
+			goto bail;
+		default:
+			pmp->status |= IB_SMP_UNSUP_METH_ATTR;
+			ret = reply((struct ib_smp *) pmp);
+			goto bail;
+		}
+
+	case IB_MGMT_METHOD_TRAP:
+	case IB_MGMT_METHOD_GET_RESP:
+		/*
+		 * The ib_mad module will call us to process responses
+		 * before checking for other consumers.
+		 * Just tell the caller to process it normally.
+		 */
+		ret = IB_MAD_RESULT_SUCCESS;
+		goto bail;
+
+	default:
+		pmp->status |= IB_SMP_UNSUP_METHOD;
+		ret = reply((struct ib_smp *) pmp);
+	}
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_process_mad - process an incoming MAD packet
+ * @ibdev: the infiniband device this packet came in on
+ * @mad_flags: MAD flags
+ * @port: the port number this packet came in on
+ * @in_wc: the work completion entry for this packet
+ * @in_grh: the global route header for this packet
+ * @in_mad: the incoming MAD
+ * @out_mad: any outgoing MAD reply
+ *
+ * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
+ * interested in processing.
+ *
+ * Note that the verbs framework has already done the MAD sanity checks,
+ * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
+ * MADs.
+ *
+ * This is called by the ib_mad module.
+ */
+int qib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
+		    struct ib_wc *in_wc, struct ib_grh *in_grh,
+		    struct ib_mad *in_mad, struct ib_mad *out_mad)
+{
+	int ret;
+
+	switch (in_mad->mad_hdr.mgmt_class) {
+	case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
+	case IB_MGMT_CLASS_SUBN_LID_ROUTED:
+		ret = process_subn(ibdev, mad_flags, port, in_mad, out_mad);
+		goto bail;
+
+	case IB_MGMT_CLASS_PERF_MGMT:
+		ret = process_perf(ibdev, port, in_mad, out_mad);
+		goto bail;
+
+	default:
+		ret = IB_MAD_RESULT_SUCCESS;
+	}
+
+bail:
+	return ret;
+}
+
+static void send_handler(struct ib_mad_agent *agent,
+			 struct ib_mad_send_wc *mad_send_wc)
+{
+	ib_free_send_mad(mad_send_wc->send_buf);
+}
+
+int qib_create_agents(struct qib_ibdev *dev)
+{
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct ib_mad_agent *agent;
+	struct qib_ibport *ibp;
+	int p;
+	int ret;
+
+	for (p = 0; p < dd->num_pports; p++) {
+		ibp = &dd->pport[p].ibport_data;
+		agent = ib_register_mad_agent(&dev->ibdev, p + 1, IB_QPT_SMI,
+					      NULL, 0, send_handler,
+					      NULL, NULL);
+		if (IS_ERR(agent)) {
+			ret = PTR_ERR(agent);
+			goto err;
+		}
+		ibp->send_agent = agent;
+	}
+
+	return 0;
+
+err:
+	for (p = 0; p < dd->num_pports; p++) {
+		ibp = &dd->pport[p].ibport_data;
+		if (ibp->send_agent) {
+			agent = ibp->send_agent;
+			ibp->send_agent = NULL;
+			ib_unregister_mad_agent(agent);
+		}
+	}
+
+	return ret;
+}
+
+void qib_free_agents(struct qib_ibdev *dev)
+{
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct ib_mad_agent *agent;
+	struct qib_ibport *ibp;
+	int p;
+
+	for (p = 0; p < dd->num_pports; p++) {
+		ibp = &dd->pport[p].ibport_data;
+		if (ibp->send_agent) {
+			agent = ibp->send_agent;
+			ibp->send_agent = NULL;
+			ib_unregister_mad_agent(agent);
+		}
+		if (ibp->sm_ah) {
+			ib_destroy_ah(&ibp->sm_ah->ibah);
+			ibp->sm_ah = NULL;
+		}
+	}
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 24/51] IB/qib: Add qib_mad.h

[Ralph Campbell]

creates the qib_mad.h file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_mad.h |  333 +++++++++++++++++++++++++++++++++++
 1 files changed, 333 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_mad.h

diff --git a/drivers/infiniband/hw/qib/qib_mad.h b/drivers/infiniband/hw/qib/qib_mad.h
new file mode 100644
index 0000000..57cab0d
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_mad.h
@@ -0,0 +1,333 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#define IB_SMP_UNSUP_VERSION    cpu_to_be16(0x0004)
+#define IB_SMP_UNSUP_METHOD     cpu_to_be16(0x0008)
+#define IB_SMP_UNSUP_METH_ATTR  cpu_to_be16(0x000C)
+#define IB_SMP_INVALID_FIELD    cpu_to_be16(0x001C)
+
+struct ib_node_info {
+	u8 base_version;
+	u8 class_version;
+	u8 node_type;
+	u8 num_ports;
+	__be64 sys_guid;
+	__be64 node_guid;
+	__be64 port_guid;
+	__be16 partition_cap;
+	__be16 device_id;
+	__be32 revision;
+	u8 local_port_num;
+	u8 vendor_id[3];
+} __attribute__ ((packed));
+
+struct ib_mad_notice_attr {
+	u8 generic_type;
+	u8 prod_type_msb;
+	__be16 prod_type_lsb;
+	__be16 trap_num;
+	__be16 issuer_lid;
+	__be16 toggle_count;
+
+	union {
+		struct {
+			u8	details[54];
+		} raw_data;
+
+		struct {
+			__be16	reserved;
+			__be16	lid;		/* where violation happened */
+			u8	port_num;	/* where violation happened */
+		} __attribute__ ((packed)) ntc_129_131;
+
+		struct {
+			__be16	reserved;
+			__be16	lid;		/* LID where change occured */
+			u8	reserved2;
+			u8	local_changes;	/* low bit - local changes */
+			__be32	new_cap_mask;	/* new capability mask */
+			u8	reserved3;
+			u8	change_flags;	/* low 3 bits only */
+		} __attribute__ ((packed)) ntc_144;
+
+		struct {
+			__be16	reserved;
+			__be16	lid;		/* lid where sys guid changed */
+			__be16	reserved2;
+			__be64	new_sys_guid;
+		} __attribute__ ((packed)) ntc_145;
+
+		struct {
+			__be16	reserved;
+			__be16	lid;
+			__be16	dr_slid;
+			u8	method;
+			u8	reserved2;
+			__be16	attr_id;
+			__be32	attr_mod;
+			__be64	mkey;
+			u8	reserved3;
+			u8	dr_trunc_hop;
+			u8	dr_rtn_path[30];
+		} __attribute__ ((packed)) ntc_256;
+
+		struct {
+			__be16		reserved;
+			__be16		lid1;
+			__be16		lid2;
+			__be32		key;
+			__be32		sl_qp1;	/* SL: high 4 bits */
+			__be32		qp2;	/* high 8 bits reserved */
+			union ib_gid	gid1;
+			union ib_gid	gid2;
+		} __attribute__ ((packed)) ntc_257_258;
+
+	} details;
+};
+
+/*
+ * Generic trap/notice types
+ */
+#define IB_NOTICE_TYPE_FATAL	0x80
+#define IB_NOTICE_TYPE_URGENT	0x81
+#define IB_NOTICE_TYPE_SECURITY	0x82
+#define IB_NOTICE_TYPE_SM	0x83
+#define IB_NOTICE_TYPE_INFO	0x84
+
+/*
+ * Generic trap/notice producers
+ */
+#define IB_NOTICE_PROD_CA		cpu_to_be16(1)
+#define IB_NOTICE_PROD_SWITCH		cpu_to_be16(2)
+#define IB_NOTICE_PROD_ROUTER		cpu_to_be16(3)
+#define IB_NOTICE_PROD_CLASS_MGR	cpu_to_be16(4)
+
+/*
+ * Generic trap/notice numbers
+ */
+#define IB_NOTICE_TRAP_LLI_THRESH	cpu_to_be16(129)
+#define IB_NOTICE_TRAP_EBO_THRESH	cpu_to_be16(130)
+#define IB_NOTICE_TRAP_FLOW_UPDATE	cpu_to_be16(131)
+#define IB_NOTICE_TRAP_CAP_MASK_CHG	cpu_to_be16(144)
+#define IB_NOTICE_TRAP_SYS_GUID_CHG	cpu_to_be16(145)
+#define IB_NOTICE_TRAP_BAD_MKEY		cpu_to_be16(256)
+#define IB_NOTICE_TRAP_BAD_PKEY		cpu_to_be16(257)
+#define IB_NOTICE_TRAP_BAD_QKEY		cpu_to_be16(258)
+
+/*
+ * Repress trap/notice flags
+ */
+#define IB_NOTICE_REPRESS_LLI_THRESH	(1 << 0)
+#define IB_NOTICE_REPRESS_EBO_THRESH	(1 << 1)
+#define IB_NOTICE_REPRESS_FLOW_UPDATE	(1 << 2)
+#define IB_NOTICE_REPRESS_CAP_MASK_CHG	(1 << 3)
+#define IB_NOTICE_REPRESS_SYS_GUID_CHG	(1 << 4)
+#define IB_NOTICE_REPRESS_BAD_MKEY	(1 << 5)
+#define IB_NOTICE_REPRESS_BAD_PKEY	(1 << 6)
+#define IB_NOTICE_REPRESS_BAD_QKEY	(1 << 7)
+
+/*
+ * Generic trap/notice other local changes flags (trap 144).
+ */
+#define IB_NOTICE_TRAP_LSE_CHG		0x04	/* Link Speed Enable changed */
+#define IB_NOTICE_TRAP_LWE_CHG		0x02	/* Link Width Enable changed */
+#define IB_NOTICE_TRAP_NODE_DESC_CHG	0x01
+
+/*
+ * Generic trap/notice M_Key volation flags in dr_trunc_hop (trap 256).
+ */
+#define IB_NOTICE_TRAP_DR_NOTICE	0x80
+#define IB_NOTICE_TRAP_DR_TRUNC		0x40
+
+struct ib_vl_weight_elem {
+	u8      vl;     /* Only low 4 bits, upper 4 bits reserved */
+	u8      weight;
+};
+
+#define IB_VLARB_LOWPRI_0_31    1
+#define IB_VLARB_LOWPRI_32_63   2
+#define IB_VLARB_HIGHPRI_0_31   3
+#define IB_VLARB_HIGHPRI_32_63  4
+
+/*
+ * PMA class portinfo capability mask bits
+ */
+#define IB_PMA_CLASS_CAP_EXT_WIDTH      cpu_to_be16(1 << 9)
+#define IB_PMA_CLASS_CAP_XMIT_WAIT      cpu_to_be16(1 << 12)
+
+#define IB_PMA_CLASS_PORT_INFO          cpu_to_be16(0x0001)
+#define IB_PMA_PORT_SAMPLES_CONTROL     cpu_to_be16(0x0010)
+#define IB_PMA_PORT_SAMPLES_RESULT      cpu_to_be16(0x0011)
+#define IB_PMA_PORT_COUNTERS            cpu_to_be16(0x0012)
+#define IB_PMA_PORT_COUNTERS_EXT        cpu_to_be16(0x001D)
+#define IB_PMA_PORT_SAMPLES_RESULT_EXT  cpu_to_be16(0x001E)
+
+struct ib_perf {
+	u8 base_version;
+	u8 mgmt_class;
+	u8 class_version;
+	u8 method;
+	__be16 status;
+	__be16 unused;
+	__be64 tid;
+	__be16 attr_id;
+	__be16 resv;
+	__be32 attr_mod;
+	u8 reserved[40];
+	u8 data[192];
+} __attribute__ ((packed));
+
+struct ib_pma_classportinfo {
+	u8 base_version;
+	u8 class_version;
+	__be16 cap_mask;
+	u8 reserved[3];
+	u8 resp_time_value;     /* only lower 5 bits */
+	union ib_gid redirect_gid;
+	__be32 redirect_tc_sl_fl;       /* 8, 4, 20 bits respectively */
+	__be16 redirect_lid;
+	__be16 redirect_pkey;
+	__be32 redirect_qp;     /* only lower 24 bits */
+	__be32 redirect_qkey;
+	union ib_gid trap_gid;
+	__be32 trap_tc_sl_fl;   /* 8, 4, 20 bits respectively */
+	__be16 trap_lid;
+	__be16 trap_pkey;
+	__be32 trap_hl_qp;      /* 8, 24 bits respectively */
+	__be32 trap_qkey;
+} __attribute__ ((packed));
+
+struct ib_pma_portsamplescontrol {
+	u8 opcode;
+	u8 port_select;
+	u8 tick;
+	u8 counter_width;       /* only lower 3 bits */
+	__be32 counter_mask0_9; /* 2, 10 * 3, bits */
+	__be16 counter_mask10_14;       /* 1, 5 * 3, bits */
+	u8 sample_mechanisms;
+	u8 sample_status;       /* only lower 2 bits */
+	__be64 option_mask;
+	__be64 vendor_mask;
+	__be32 sample_start;
+	__be32 sample_interval;
+	__be16 tag;
+	__be16 counter_select[15];
+} __attribute__ ((packed));
+
+struct ib_pma_portsamplesresult {
+	__be16 tag;
+	__be16 sample_status;   /* only lower 2 bits */
+	__be32 counter[15];
+} __attribute__ ((packed));
+
+struct ib_pma_portsamplesresult_ext {
+	__be16 tag;
+	__be16 sample_status;   /* only lower 2 bits */
+	__be32 extended_width;  /* only upper 2 bits */
+	__be64 counter[15];
+} __attribute__ ((packed));
+
+struct ib_pma_portcounters {
+	u8 reserved;
+	u8 port_select;
+	__be16 counter_select;
+	__be16 symbol_error_counter;
+	u8 link_error_recovery_counter;
+	u8 link_downed_counter;
+	__be16 port_rcv_errors;
+	__be16 port_rcv_remphys_errors;
+	__be16 port_rcv_switch_relay_errors;
+	__be16 port_xmit_discards;
+	u8 port_xmit_constraint_errors;
+	u8 port_rcv_constraint_errors;
+	u8 reserved1;
+	u8 lli_ebor_errors;     /* 4, 4, bits */
+	__be16 reserved2;
+	__be16 vl15_dropped;
+	__be32 port_xmit_data;
+	__be32 port_rcv_data;
+	__be32 port_xmit_packets;
+	__be32 port_rcv_packets;
+} __attribute__ ((packed));
+
+#define IB_PMA_SEL_SYMBOL_ERROR                 cpu_to_be16(0x0001)
+#define IB_PMA_SEL_LINK_ERROR_RECOVERY          cpu_to_be16(0x0002)
+#define IB_PMA_SEL_LINK_DOWNED                  cpu_to_be16(0x0004)
+#define IB_PMA_SEL_PORT_RCV_ERRORS              cpu_to_be16(0x0008)
+#define IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS      cpu_to_be16(0x0010)
+#define IB_PMA_SEL_PORT_XMIT_DISCARDS           cpu_to_be16(0x0040)
+#define IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS  cpu_to_be16(0x0200)
+#define IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS    cpu_to_be16(0x0400)
+#define IB_PMA_SEL_PORT_VL15_DROPPED            cpu_to_be16(0x0800)
+#define IB_PMA_SEL_PORT_XMIT_DATA               cpu_to_be16(0x1000)
+#define IB_PMA_SEL_PORT_RCV_DATA                cpu_to_be16(0x2000)
+#define IB_PMA_SEL_PORT_XMIT_PACKETS            cpu_to_be16(0x4000)
+#define IB_PMA_SEL_PORT_RCV_PACKETS             cpu_to_be16(0x8000)
+
+struct ib_pma_portcounters_ext {
+	u8 reserved;
+	u8 port_select;
+	__be16 counter_select;
+	__be32 reserved1;
+	__be64 port_xmit_data;
+	__be64 port_rcv_data;
+	__be64 port_xmit_packets;
+	__be64 port_rcv_packets;
+	__be64 port_unicast_xmit_packets;
+	__be64 port_unicast_rcv_packets;
+	__be64 port_multicast_xmit_packets;
+	__be64 port_multicast_rcv_packets;
+} __attribute__ ((packed));
+
+#define IB_PMA_SELX_PORT_XMIT_DATA              cpu_to_be16(0x0001)
+#define IB_PMA_SELX_PORT_RCV_DATA               cpu_to_be16(0x0002)
+#define IB_PMA_SELX_PORT_XMIT_PACKETS           cpu_to_be16(0x0004)
+#define IB_PMA_SELX_PORT_RCV_PACKETS            cpu_to_be16(0x0008)
+#define IB_PMA_SELX_PORT_UNI_XMIT_PACKETS       cpu_to_be16(0x0010)
+#define IB_PMA_SELX_PORT_UNI_RCV_PACKETS        cpu_to_be16(0x0020)
+#define IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS     cpu_to_be16(0x0040)
+#define IB_PMA_SELX_PORT_MULTI_RCV_PACKETS      cpu_to_be16(0x0080)
+
+/*
+ * The PortSamplesControl.CounterMasks field is an array of 3 bit fields
+ * which specify the N'th counter's capabilities. See ch. 16.1.3.2.
+ * We support 5 counters which only count the mandatory quantities.
+ */
+#define COUNTER_MASK(q, n) (q << ((9 - n) * 3))
+#define COUNTER_MASK0_9 \
+	cpu_to_be32(COUNTER_MASK(1, 0) | \
+		    COUNTER_MASK(1, 1) | \
+		    COUNTER_MASK(1, 2) | \
+		    COUNTER_MASK(1, 3) | \
+		    COUNTER_MASK(1, 4))
+

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 25/51] IB/qib: Add qib_mmap.c

[Ralph Campbell]

creates the qib_mmap.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_mmap.c |  173 ++++++++++++++++++++++++++++++++++
 1 files changed, 173 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_mmap.c

diff --git a/drivers/infiniband/hw/qib/qib_mmap.c b/drivers/infiniband/hw/qib/qib_mmap.c
new file mode 100644
index 0000000..ae15d38
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_mmap.c
@@ -0,0 +1,173 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/vmalloc.h>
+#include <linux/mm.h>
+#include <linux/errno.h>
+#include <asm/pgtable.h>
+
+#include "qib_verbs.h"
+
+/**
+ * qib_release_mmap_info - free mmap info structure
+ * @ref: a pointer to the kref within struct qib_mmap_info
+ */
+void qib_release_mmap_info(struct kref *ref)
+{
+	struct qib_mmap_info *ip =
+		container_of(ref, struct qib_mmap_info, ref);
+	struct qib_ibdev *dev = to_idev(ip->context->device);
+
+	spin_lock_irq(&dev->pending_lock);
+	list_del(&ip->pending_mmaps);
+	spin_unlock_irq(&dev->pending_lock);
+
+	vfree(ip->obj);
+	kfree(ip);
+}
+
+/*
+ * open and close keep track of how many times the CQ is mapped,
+ * to avoid releasing it.
+ */
+static void qib_vma_open(struct vm_area_struct *vma)
+{
+	struct qib_mmap_info *ip = vma->vm_private_data;
+
+	kref_get(&ip->ref);
+}
+
+static void qib_vma_close(struct vm_area_struct *vma)
+{
+	struct qib_mmap_info *ip = vma->vm_private_data;
+
+	kref_put(&ip->ref, qib_release_mmap_info);
+}
+
+static struct vm_operations_struct qib_vm_ops = {
+	.open =     qib_vma_open,
+	.close =    qib_vma_close,
+};
+
+/**
+ * qib_mmap - create a new mmap region
+ * @context: the IB user context of the process making the mmap() call
+ * @vma: the VMA to be initialized
+ * Return zero if the mmap is OK. Otherwise, return an errno.
+ */
+int qib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
+{
+	struct qib_ibdev *dev = to_idev(context->device);
+	unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
+	unsigned long size = vma->vm_end - vma->vm_start;
+	struct qib_mmap_info *ip, *pp;
+	int ret = -EINVAL;
+
+	/*
+	 * Search the device's list of objects waiting for a mmap call.
+	 * Normally, this list is very short since a call to create a
+	 * CQ, QP, or SRQ is soon followed by a call to mmap().
+	 */
+	spin_lock_irq(&dev->pending_lock);
+	list_for_each_entry_safe(ip, pp, &dev->pending_mmaps,
+				 pending_mmaps) {
+		/* Only the creator is allowed to mmap the object */
+		if (context != ip->context || (__u64) offset != ip->offset)
+			continue;
+		/* Don't allow a mmap larger than the object. */
+		if (size > ip->size)
+			break;
+
+		list_del_init(&ip->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+
+		ret = remap_vmalloc_range(vma, ip->obj, 0);
+		if (ret)
+			goto done;
+		vma->vm_ops = &qib_vm_ops;
+		vma->vm_private_data = ip;
+		qib_vma_open(vma);
+		goto done;
+	}
+	spin_unlock_irq(&dev->pending_lock);
+done:
+	return ret;
+}
+
+/*
+ * Allocate information for qib_mmap
+ */
+struct qib_mmap_info *qib_create_mmap_info(struct qib_ibdev *dev,
+					   u32 size,
+					   struct ib_ucontext *context,
+					   void *obj) {
+	struct qib_mmap_info *ip;
+
+	ip = kmalloc(sizeof *ip, GFP_KERNEL);
+	if (!ip)
+		goto bail;
+
+	size = PAGE_ALIGN(size);
+
+	spin_lock_irq(&dev->mmap_offset_lock);
+	if (dev->mmap_offset == 0)
+		dev->mmap_offset = PAGE_SIZE;
+	ip->offset = dev->mmap_offset;
+	dev->mmap_offset += size;
+	spin_unlock_irq(&dev->mmap_offset_lock);
+
+	INIT_LIST_HEAD(&ip->pending_mmaps);
+	ip->size = size;
+	ip->context = context;
+	ip->obj = obj;
+	kref_init(&ip->ref);
+
+bail:
+	return ip;
+}
+
+void qib_update_mmap_info(struct qib_ibdev *dev, struct qib_mmap_info *ip,
+			  u32 size, void *obj)
+{
+	size = PAGE_ALIGN(size);
+
+	spin_lock_irq(&dev->mmap_offset_lock);
+	if (dev->mmap_offset == 0)
+		dev->mmap_offset = PAGE_SIZE;
+	ip->offset = dev->mmap_offset;
+	dev->mmap_offset += size;
+	spin_unlock_irq(&dev->mmap_offset_lock);
+
+	ip->size = size;
+	ip->obj = obj;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 26/51] IB/qib: Add qib_mr.c

[Ralph Campbell]

creates the qib_mr.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_mr.c |  455 ++++++++++++++++++++++++++++++++++++
 1 files changed, 455 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_mr.c

diff --git a/drivers/infiniband/hw/qib/qib_mr.c b/drivers/infiniband/hw/qib/qib_mr.c
new file mode 100644
index 0000000..f283c3b
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_mr.c
@@ -0,0 +1,455 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_umem.h>
+#include <rdma/ib_smi.h>
+
+#include "qib.h"
+
+/* Fast memory region */
+struct qib_fmr {
+	struct ib_fmr ibfmr;
+	u8 page_shift;
+	struct qib_mregion mr;        /* must be last */
+};
+
+static inline struct qib_fmr *to_ifmr(struct ib_fmr *ibfmr)
+{
+	return container_of(ibfmr, struct qib_fmr, ibfmr);
+}
+
+/**
+ * qib_get_dma_mr - get a DMA memory region
+ * @pd: protection domain for this memory region
+ * @acc: access flags
+ *
+ * Returns the memory region on success, otherwise returns an errno.
+ * Note that all DMA addresses should be created via the
+ * struct ib_dma_mapping_ops functions (see qib_dma.c).
+ */
+struct ib_mr *qib_get_dma_mr(struct ib_pd *pd, int acc)
+{
+	struct qib_ibdev *dev = to_idev(pd->device);
+	struct qib_mr *mr;
+	struct ib_mr *ret;
+	unsigned long flags;
+
+	if (to_ipd(pd)->user) {
+		ret = ERR_PTR(-EPERM);
+		goto bail;
+	}
+
+	mr = kzalloc(sizeof *mr, GFP_KERNEL);
+	if (!mr) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	mr->mr.access_flags = acc;
+	atomic_set(&mr->mr.refcount, 0);
+
+	spin_lock_irqsave(&dev->lk_table.lock, flags);
+	if (!dev->dma_mr)
+		dev->dma_mr = &mr->mr;
+	spin_unlock_irqrestore(&dev->lk_table.lock, flags);
+
+	ret = &mr->ibmr;
+
+bail:
+	return ret;
+}
+
+static struct qib_mr *alloc_mr(int count, struct qib_lkey_table *lk_table)
+{
+	struct qib_mr *mr;
+	int m, i = 0;
+
+	/* Allocate struct plus pointers to first level page tables. */
+	m = (count + QIB_SEGSZ - 1) / QIB_SEGSZ;
+	mr = kmalloc(sizeof *mr + m * sizeof mr->mr.map[0], GFP_KERNEL);
+	if (!mr)
+		goto done;
+
+	/* Allocate first level page tables. */
+	for (; i < m; i++) {
+		mr->mr.map[i] = kmalloc(sizeof *mr->mr.map[0], GFP_KERNEL);
+		if (!mr->mr.map[i])
+			goto bail;
+	}
+	mr->mr.mapsz = m;
+
+	/*
+	 * ib_reg_phys_mr() will initialize mr->ibmr except for
+	 * lkey and rkey.
+	 */
+	if (!qib_alloc_lkey(lk_table, &mr->mr))
+		goto bail;
+	mr->ibmr.lkey = mr->mr.lkey;
+	mr->ibmr.rkey = mr->mr.lkey;
+
+	atomic_set(&mr->mr.refcount, 0);
+	goto done;
+
+bail:
+	while (i)
+		kfree(mr->mr.map[--i]);
+	kfree(mr);
+	mr = NULL;
+
+done:
+	return mr;
+}
+
+/**
+ * qib_reg_phys_mr - register a physical memory region
+ * @pd: protection domain for this memory region
+ * @buffer_list: pointer to the list of physical buffers to register
+ * @num_phys_buf: the number of physical buffers to register
+ * @iova_start: the starting address passed over IB which maps to this MR
+ *
+ * Returns the memory region on success, otherwise returns an errno.
+ */
+struct ib_mr *qib_reg_phys_mr(struct ib_pd *pd,
+			      struct ib_phys_buf *buffer_list,
+			      int num_phys_buf, int acc, u64 *iova_start)
+{
+	struct qib_mr *mr;
+	int n, m, i;
+	struct ib_mr *ret;
+
+	mr = alloc_mr(num_phys_buf, &to_idev(pd->device)->lk_table);
+	if (mr == NULL) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	mr->mr.pd = pd;
+	mr->mr.user_base = *iova_start;
+	mr->mr.iova = *iova_start;
+	mr->mr.length = 0;
+	mr->mr.offset = 0;
+	mr->mr.access_flags = acc;
+	mr->mr.max_segs = num_phys_buf;
+	mr->umem = NULL;
+
+	m = 0;
+	n = 0;
+	for (i = 0; i < num_phys_buf; i++) {
+		mr->mr.map[m]->segs[n].vaddr = (void *) buffer_list[i].addr;
+		mr->mr.map[m]->segs[n].length = buffer_list[i].size;
+		mr->mr.length += buffer_list[i].size;
+		n++;
+		if (n == QIB_SEGSZ) {
+			m++;
+			n = 0;
+		}
+	}
+
+	ret = &mr->ibmr;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_reg_user_mr - register a userspace memory region
+ * @pd: protection domain for this memory region
+ * @start: starting userspace address
+ * @length: length of region to register
+ * @virt_addr: virtual address to use (from HCA's point of view)
+ * @mr_access_flags: access flags for this memory region
+ * @udata: unused by the QLogic_IB driver
+ *
+ * Returns the memory region on success, otherwise returns an errno.
+ */
+struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+			      u64 virt_addr, int mr_access_flags,
+			      struct ib_udata *udata)
+{
+	struct qib_mr *mr;
+	struct ib_umem *umem;
+	struct ib_umem_chunk *chunk;
+	int n, m, i;
+	struct ib_mr *ret;
+
+	if (length == 0) {
+		ret = ERR_PTR(-EINVAL);
+		goto bail;
+	}
+
+	umem = ib_umem_get(pd->uobject->context, start, length,
+			   mr_access_flags, 0);
+	if (IS_ERR(umem))
+		return (void *) umem;
+
+	n = 0;
+	list_for_each_entry(chunk, &umem->chunk_list, list)
+		n += chunk->nents;
+
+	mr = alloc_mr(n, &to_idev(pd->device)->lk_table);
+	if (!mr) {
+		ret = ERR_PTR(-ENOMEM);
+		ib_umem_release(umem);
+		goto bail;
+	}
+
+	mr->mr.pd = pd;
+	mr->mr.user_base = start;
+	mr->mr.iova = virt_addr;
+	mr->mr.length = length;
+	mr->mr.offset = umem->offset;
+	mr->mr.access_flags = mr_access_flags;
+	mr->mr.max_segs = n;
+	mr->umem = umem;
+
+	m = 0;
+	n = 0;
+	list_for_each_entry(chunk, &umem->chunk_list, list) {
+		for (i = 0; i < chunk->nents; i++) {
+			void *vaddr;
+
+			vaddr = page_address(sg_page(&chunk->page_list[i]));
+			if (!vaddr) {
+				ret = ERR_PTR(-EINVAL);
+				goto bail;
+			}
+			mr->mr.map[m]->segs[n].vaddr = vaddr;
+			mr->mr.map[m]->segs[n].length = umem->page_size;
+			n++;
+			if (n == QIB_SEGSZ) {
+				m++;
+				n = 0;
+			}
+		}
+	}
+	ret = &mr->ibmr;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_dereg_mr - unregister and free a memory region
+ * @ibmr: the memory region to free
+ *
+ * Returns 0 on success.
+ *
+ * Note that this is called to free MRs created by qib_get_dma_mr()
+ * or qib_reg_user_mr().
+ */
+int qib_dereg_mr(struct ib_mr *ibmr)
+{
+	struct qib_mr *mr = to_imr(ibmr);
+	struct qib_ibdev *dev = to_idev(ibmr->device);
+	int ret;
+	int i;
+
+	ret = qib_free_lkey(dev, &mr->mr);
+	if (ret)
+		return ret;
+
+	i = mr->mr.mapsz;
+	while (i)
+		kfree(mr->mr.map[--i]);
+	if (mr->umem)
+		ib_umem_release(mr->umem);
+	kfree(mr);
+	return 0;
+}
+
+/**
+ * qib_alloc_fmr - allocate a fast memory region
+ * @pd: the protection domain for this memory region
+ * @mr_access_flags: access flags for this memory region
+ * @fmr_attr: fast memory region attributes
+ *
+ * Returns the memory region on success, otherwise returns an errno.
+ */
+struct ib_fmr *qib_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
+			     struct ib_fmr_attr *fmr_attr)
+{
+	struct qib_fmr *fmr;
+	int m, i = 0;
+	struct ib_fmr *ret;
+
+	/* Allocate struct plus pointers to first level page tables. */
+	m = (fmr_attr->max_pages + QIB_SEGSZ - 1) / QIB_SEGSZ;
+	fmr = kmalloc(sizeof *fmr + m * sizeof fmr->mr.map[0], GFP_KERNEL);
+	if (!fmr)
+		goto bail;
+
+	/* Allocate first level page tables. */
+	for (; i < m; i++) {
+		fmr->mr.map[i] = kmalloc(sizeof *fmr->mr.map[0],
+					 GFP_KERNEL);
+		if (!fmr->mr.map[i])
+			goto bail;
+	}
+	fmr->mr.mapsz = m;
+
+	/*
+	 * ib_alloc_fmr() will initialize fmr->ibfmr except for lkey &
+	 * rkey.
+	 */
+	if (!qib_alloc_lkey(&to_idev(pd->device)->lk_table, &fmr->mr))
+		goto bail;
+	fmr->ibfmr.rkey = fmr->mr.lkey;
+	fmr->ibfmr.lkey = fmr->mr.lkey;
+	/*
+	 * Resources are allocated but no valid mapping (RKEY can't be
+	 * used).
+	 */
+	fmr->mr.pd = pd;
+	fmr->mr.user_base = 0;
+	fmr->mr.iova = 0;
+	fmr->mr.length = 0;
+	fmr->mr.offset = 0;
+	fmr->mr.access_flags = mr_access_flags;
+	fmr->mr.max_segs = fmr_attr->max_pages;
+	fmr->page_shift = fmr_attr->page_shift;
+
+	atomic_set(&fmr->mr.refcount, 0);
+	ret = &fmr->ibfmr;
+	goto done;
+
+bail:
+	while (i)
+		kfree(fmr->mr.map[--i]);
+	kfree(fmr);
+	ret = ERR_PTR(-ENOMEM);
+
+done:
+	return ret;
+}
+
+/**
+ * qib_map_phys_fmr - set up a fast memory region
+ * @ibmfr: the fast memory region to set up
+ * @page_list: the list of pages to associate with the fast memory region
+ * @list_len: the number of pages to associate with the fast memory region
+ * @iova: the virtual address of the start of the fast memory region
+ *
+ * This may be called from interrupt context.
+ */
+
+int qib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
+		     int list_len, u64 iova)
+{
+	struct qib_fmr *fmr = to_ifmr(ibfmr);
+	struct qib_lkey_table *rkt;
+	unsigned long flags;
+	int m, n, i;
+	u32 ps;
+	int ret;
+
+	if (atomic_read(&fmr->mr.refcount)) {
+		qib_dbg("FMR modified when busy (LKEY %x cnt %u)\n",
+			fmr->mr.lkey, atomic_read(&fmr->mr.refcount));
+		return -EBUSY;
+	}
+
+	if (list_len > fmr->mr.max_segs) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	rkt = &to_idev(ibfmr->device)->lk_table;
+	spin_lock_irqsave(&rkt->lock, flags);
+	fmr->mr.user_base = iova;
+	fmr->mr.iova = iova;
+	ps = 1 << fmr->page_shift;
+	fmr->mr.length = list_len * ps;
+	m = 0;
+	n = 0;
+	for (i = 0; i < list_len; i++) {
+		fmr->mr.map[m]->segs[n].vaddr = (void *) page_list[i];
+		fmr->mr.map[m]->segs[n].length = ps;
+		if (++n == QIB_SEGSZ) {
+			m++;
+			n = 0;
+		}
+	}
+	spin_unlock_irqrestore(&rkt->lock, flags);
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_unmap_fmr - unmap fast memory regions
+ * @fmr_list: the list of fast memory regions to unmap
+ *
+ * Returns 0 on success.
+ */
+int qib_unmap_fmr(struct list_head *fmr_list)
+{
+	struct qib_fmr *fmr;
+	struct qib_lkey_table *rkt;
+	unsigned long flags;
+
+	list_for_each_entry(fmr, fmr_list, ibfmr.list) {
+		rkt = &to_idev(fmr->ibfmr.device)->lk_table;
+		spin_lock_irqsave(&rkt->lock, flags);
+		if (atomic_read(&fmr->mr.refcount))
+			qib_dbg("FMR busy (LKEY %x cnt %u)\n",
+				fmr->mr.lkey, atomic_read(&fmr->mr.refcount));
+		fmr->mr.user_base = 0;
+		fmr->mr.iova = 0;
+		fmr->mr.length = 0;
+		spin_unlock_irqrestore(&rkt->lock, flags);
+	}
+	return 0;
+}
+
+/**
+ * qib_dealloc_fmr - deallocate a fast memory region
+ * @ibfmr: the fast memory region to deallocate
+ *
+ * Returns 0 on success.
+ */
+int qib_dealloc_fmr(struct ib_fmr *ibfmr)
+{
+	struct qib_fmr *fmr = to_ifmr(ibfmr);
+	int ret;
+	int i;
+
+	ret = qib_free_lkey(to_idev(ibfmr->device), &fmr->mr);
+	if (ret)
+		return ret;
+
+	i = fmr->mr.mapsz;
+	while (i)
+		kfree(fmr->mr.map[--i]);
+	kfree(fmr);
+	return 0;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 26/51] IB/qib: Add qib_mr.c

[Roland Dreier]

 > +struct ib_fmr *qib_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
 > +			     struct ib_fmr_attr *fmr_attr)

Instead of the crufty old FMR API, I think it would be preferable to
implement the real IB spec fast registration through a work queue
stuff.  I understand that it might take some effort to do that but if we
end up with our newest driver stuck with FMRs, we'll never be able to
move away from that.
-- 
Roland Dreier <rolandd-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> || For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 26/51] IB/qib: Add qib_mr.c

[Ralph Campbell]

On Tue, 2010-04-06 at 21:15 -0700, Roland Dreier wrote:
> > +struct ib_fmr *qib_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
>  > +			     struct ib_fmr_attr *fmr_attr)
> 
> Instead of the crufty old FMR API, I think it would be preferable to
> implement the real IB spec fast registration through a work queue
> stuff.  I understand that it might take some effort to do that but if we
> end up with our newest driver stuck with FMRs, we'll never be able to
> move away from that.

It looks like the IB_WR_FAST_REG_MR operation isn't too hard to
implement so I will add this support.

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 27/51] IB/qib: Add qib_pcie.c

[Ralph Campbell]

creates the qib_pcie.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_pcie.c |  810 ++++++++++++++++++++++++++++++++++
 1 files changed, 810 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_pcie.c

diff --git a/drivers/infiniband/hw/qib/qib_pcie.c b/drivers/infiniband/hw/qib/qib_pcie.c
new file mode 100644
index 0000000..714cad4
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_pcie.c
@@ -0,0 +1,810 @@
+/*
+ * Copyright (c) 2008, 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/pci.h>
+#include <linux/io.h>
+#include <linux/delay.h>
+#include <linux/vmalloc.h>
+#include <linux/aer.h>
+
+#include "qib.h"
+
+/*
+ * This file contains PCIe utility routines that are common to the
+ * various QLogic InfiniPath adapters
+ */
+
+/*
+ * Code to adjust PCIe capabilities.
+ * To minimize the change footprint, we call it
+ * from qib_pcie_params, which every chip-specific
+ * file calls, even though this violates some
+ * expectations of harmlessness.
+ */
+static int qib_tune_pcie_caps(struct qib_devdata *);
+static int qib_tune_pcie_coalesce(struct qib_devdata *);
+
+/*
+ * Do all the common PCIe setup and initialization.
+ * devdata is not yet allocated, and is not allocated until after this
+ * routine returns success.  Therefore qib_dev_err() can't be used for error
+ * printing.
+ */
+int qib_pcie_init(struct pci_dev *pdev, const struct pci_device_id *ent)
+{
+	int ret;
+
+	ret = pci_enable_device(pdev);
+	if (ret) {
+		/*
+		 * This can happen (in theory) iff:
+		 * We did a chip reset, and then failed to reprogram the
+		 * BAR, or the chip reset due to an internal error.  We then
+		 * unloaded the driver and reloaded it.
+		 *
+		 * Both reset cases set the BAR back to initial state.  For
+		 * the latter case, the AER sticky error bit at offset 0x718
+		 * should be set, but the Linux kernel doesn't yet know
+		 * about that, it appears.  If the original BAR was retained
+		 * in the kernel data structures, this may be OK.
+		 */
+		qib_early_err(&pdev->dev, "pci enable failed: error %d\n",
+			      -ret);
+		goto done;
+	}
+
+	ret = pci_request_regions(pdev, QIB_DRV_NAME);
+	if (ret) {
+		qib_devinfo(pdev, "pci_request_regions fails: err %d\n", -ret);
+		goto bail;
+	}
+
+	ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
+	if (ret) {
+		/*
+		 * If the 64 bit setup fails, try 32 bit.  Some systems
+		 * do not setup 64 bit maps on systems with 2GB or less
+		 * memory installed.
+		 */
+		ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
+		if (ret) {
+			qib_devinfo(pdev, "Unable to set DMA mask: %d\n", ret);
+			goto bail;
+		}
+		qib_dbg("No 64bit DMA mask, used 32 bit mask\n");
+		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
+	} else
+		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
+	if (ret)
+		qib_early_err(&pdev->dev,
+			      "Unable to set DMA consistent mask: %d\n", ret);
+
+	pci_set_master(pdev);
+	ret = pci_enable_pcie_error_reporting(pdev);
+	if (ret)
+		qib_early_err(&pdev->dev,
+			      "Unable to enable pcie error reporting: %d\n",
+			      ret);
+	goto done;
+
+bail:
+	pci_disable_device(pdev);
+	pci_release_regions(pdev);
+done:
+	return ret;
+}
+
+/*
+ * Do remaining PCIe setup, once dd is allocated, and save away
+ * fields required to re-initialize after a chip reset, or for
+ * various other purposes
+ */
+int qib_pcie_ddinit(struct qib_devdata *dd, struct pci_dev *pdev,
+		    const struct pci_device_id *ent)
+{
+	unsigned long len;
+	resource_size_t addr;
+
+	dd->pcidev = pdev;
+	pci_set_drvdata(pdev, dd);
+
+	addr = pci_resource_start(pdev, 0);
+	len = pci_resource_len(pdev, 0);
+	qib_cdbg(VERBOSE, "regbase (0) %llx len %ld vend %x/%x "
+		 "driver_data %p\n", (unsigned long long)addr, len,
+		 ent->vendor, ent->device, pci_get_drvdata(pdev));
+
+#if defined(__powerpc__)
+	/* There isn't a generic way to specify writethrough mappings */
+	dd->kregbase = __ioremap(addr, len, _PAGE_NO_CACHE | _PAGE_WRITETHRU);
+#else
+	dd->kregbase = ioremap_nocache(addr, len);
+#endif
+
+	if (!dd->kregbase) {
+		qib_dbg("Unable to map io addr %llx to kvirt, failing\n",
+			(unsigned long long)addr);
+		return -ENOMEM;
+	}
+
+	dd->kregend = (u64 __iomem *)((void __iomem *) dd->kregbase + len);
+	dd->physaddr = addr;        /* used for io_remap, etc. */
+	/* for user mmap */
+	qib_cdbg(VERBOSE, "mapped io addr %llx to kregbase %p through %p\n",
+		 (unsigned long long)addr, dd->kregbase, dd->kregend);
+
+	/*
+	 * Save BARs to rewrite after device reset.  Save all 64 bits of
+	 * BAR, just in case.
+	 */
+	dd->pcibar0 = addr;
+	dd->pcibar1 = addr >> 32;
+	dd->deviceid = ent->device; /* save for later use */
+	dd->vendorid = ent->vendor;
+
+	return 0;
+}
+
+/*
+ * Do PCIe cleanup, after chip-specific cleanup, etc.  Just prior
+ * to releasing the dd memory.
+ * void because none of the core pcie cleanup returns are void
+ */
+void qib_pcie_ddcleanup(struct qib_devdata *dd)
+{
+	u64 __iomem *base = (void __iomem *) dd->kregbase;
+
+	qib_cdbg(VERBOSE, "Unmapping kregbase %p\n", dd->kregbase);
+	dd->kregbase = NULL;
+	iounmap(base);
+	if (dd->piobase) {
+		qib_cdbg(VERBOSE, "Unmapping piobase %p\n", dd->piobase);
+		iounmap(dd->piobase);
+	}
+	if (dd->userbase) {
+		qib_cdbg(VERBOSE, "Unmapping userbase %p\n", dd->userbase);
+		iounmap(dd->userbase);
+	}
+
+	qib_cdbg(VERBOSE, "calling pci_disable_device\n");
+	pci_disable_device(dd->pcidev);
+	pci_release_regions(dd->pcidev);
+
+	pci_set_drvdata(dd->pcidev, NULL);
+}
+
+static void qib_msix_setup(struct qib_devdata *dd, int pos, u32 *msixcnt,
+			   struct msix_entry *msix_entry)
+{
+	int ret;
+	u32 tabsize = 0;
+	u16 msix_flags;
+
+	pci_read_config_word(dd->pcidev, pos + PCI_MSIX_FLAGS, &msix_flags);
+	tabsize = 1 + (msix_flags & PCI_MSIX_FLAGS_QSIZE);
+	if (tabsize > *msixcnt)
+		tabsize = *msixcnt;
+	ret = pci_enable_msix(dd->pcidev, msix_entry, tabsize);
+	if (ret > 0) {
+		qib_dbg("Couldn't get %d MSIx vectors, "
+			"trying to get %d\n", tabsize, ret);
+		tabsize = ret;
+		ret = pci_enable_msix(dd->pcidev, msix_entry, tabsize);
+	}
+	if (ret) {
+		qib_dev_err(dd, "pci_enable_msix %d vectors failed: %d, "
+			    "falling back to INTx\n", tabsize, ret);
+		tabsize = 0;
+	}
+	*msixcnt = tabsize;
+
+	if (ret)
+		qib_enable_intx(dd->pcidev);
+
+}
+
+/**
+ * We save the msi lo and hi values, so we can restore them after
+ * chip reset (the kernel PCI infrastructure doesn't yet handle that
+ * correctly.
+ */
+static int qib_msi_setup(struct qib_devdata *dd, int pos)
+{
+	struct pci_dev *pdev = dd->pcidev;
+	u16 control;
+	int ret;
+
+	ret = pci_enable_msi(pdev);
+	if (ret)
+		qib_dev_err(dd, "pci_enable_msi failed: %d, "
+			    "interrupts may not work\n", ret);
+	/* continue even if it fails, we may still be OK... */
+
+	pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_LO,
+			      &dd->msi_lo);
+	pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_HI,
+			      &dd->msi_hi);
+	pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &control);
+	/* now save the data (vector) info */
+	pci_read_config_word(pdev, pos + ((control & PCI_MSI_FLAGS_64BIT)
+				    ? 12 : 8),
+			     &dd->msi_data);
+	qib_cdbg(VERBOSE, "Read msi data 0x%x from config offset "
+		 "0x%x, control=0x%x\n", dd->msi_data,
+		 pos + ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8),
+		 control);
+	return ret;
+}
+
+int qib_pcie_params(struct qib_devdata *dd, u32 minw, u32 *nent,
+		    struct msix_entry *entry)
+{
+	u16 linkstat, speed;
+	int pos = 0, pose, ret = 1;
+
+	pose = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP);
+	if (!pose) {
+		qib_dev_err(dd, "Can't find PCI Express capability!\n");
+		/* set up something... */
+		dd->lbus_width = 1;
+		dd->lbus_speed = 2500; /* Gen1, 2.5GHz */
+		goto bail;
+	}
+
+	pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSIX);
+	if (nent && *nent && pos) {
+		qib_msix_setup(dd, pos, nent, entry);
+		ret = 0; /* did it, either MSIx or INTx */
+	} else {
+		pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI);
+		if (pos)
+			ret = qib_msi_setup(dd, pos);
+		else
+			qib_dev_err(dd, "No PCI MSI or MSIx capability!\n");
+	}
+	if (!pos)
+		qib_enable_intx(dd->pcidev);
+
+	pci_read_config_word(dd->pcidev, pose + PCI_EXP_LNKSTA, &linkstat);
+	/*
+	 * speed is bits 0-3, linkwidth is bits 4-8
+	 * no defines for them in headers
+	 */
+	speed = linkstat & 0xf;
+	linkstat >>= 4;
+	linkstat &= 0x1f;
+	dd->lbus_width = linkstat;
+
+	switch (speed) {
+	case 1:
+		dd->lbus_speed = 2500; /* Gen1, 2.5GHz */
+		break;
+	case 2:
+		dd->lbus_speed = 5000; /* Gen1, 5GHz */
+		break;
+	default: /* not defined, assume gen1 */
+		qib_dbg("PCIe unexpected link speed %u\n", speed);
+		dd->lbus_speed = 2500;
+		break;
+	}
+
+	/*
+	 * Check against expected pcie width and complain if "wrong"
+	 * on first initialization, not afterwards (i.e., reset).
+	 */
+	if (minw && linkstat < minw)
+		qib_dev_err(dd,
+			    "PCIe width %u (x%u HCA), performance reduced\n",
+			    linkstat, minw);
+	else
+		qib_cdbg(VERBOSE, "PCIe speed %u width %u (x8 HCA)\n",
+			 dd->lbus_speed, linkstat);
+
+	qib_tune_pcie_caps(dd);
+
+	qib_tune_pcie_coalesce(dd);
+
+bail:
+	/* fill in string, even on errors */
+	snprintf(dd->lbus_info, sizeof(dd->lbus_info),
+		 "PCIe,%uMHz,x%u\n", dd->lbus_speed, dd->lbus_width);
+	return ret;
+}
+
+/*
+ * Setup pcie interrupt stuff again after a reset.  I'd like to just call
+ * pci_enable_msi() again for msi, but when I do that,
+ * the MSI enable bit doesn't get set in the command word, and
+ * we switch to to a different interrupt vector, which is confusing,
+ * so I instead just do it all inline.  Perhaps somehow can tie this
+ * into the PCIe hotplug support at some point
+ */
+int qib_reinit_intr(struct qib_devdata *dd)
+{
+	int pos;
+	u16 control;
+	int ret = 0;
+
+	/* If we aren't using MSI, don't restore it */
+	if (!dd->msi_lo)
+		goto bail;
+
+	pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI);
+	if (!pos) {
+		qib_dev_err(dd, "Can't find MSI capability, "
+			    "can't restore MSI settings\n");
+		ret = 0;
+		/* nothing special for MSIx, just MSI */
+		goto bail;
+	}
+	qib_cdbg(VERBOSE, "Writing msi_lo 0x%x to config offset 0x%x\n",
+		 dd->msi_lo, pos + PCI_MSI_ADDRESS_LO);
+	pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO,
+			       dd->msi_lo);
+	qib_cdbg(VERBOSE, "Writing msi_lo 0x%x to config offset 0x%x\n",
+		 dd->msi_hi, pos + PCI_MSI_ADDRESS_HI);
+	pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI,
+			       dd->msi_hi);
+	pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS, &control);
+	if (!(control & PCI_MSI_FLAGS_ENABLE)) {
+		qib_cdbg(INIT, "MSI control at off %x was %x, "
+			 "setting MSI enable (%x)\n", pos + PCI_MSI_FLAGS,
+			 control, control | PCI_MSI_FLAGS_ENABLE);
+		control |= PCI_MSI_FLAGS_ENABLE;
+		pci_write_config_word(dd->pcidev, pos + PCI_MSI_FLAGS,
+				      control);
+	}
+	/* now rewrite the data (vector) info */
+	pci_write_config_word(dd->pcidev, pos +
+			      ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8),
+			      dd->msi_data);
+	ret = 1;
+bail:
+	if (!ret && (dd->flags & QIB_HAS_INTX)) {
+		qib_cdbg(INIT, "Using INTx, MSI disabled or not configured\n");
+		qib_enable_intx(dd->pcidev);
+		ret = 1;
+	}
+
+	/* and now set the pci master bit again */
+	pci_set_master(dd->pcidev);
+
+	return ret;
+}
+
+/*
+ * Disable msi interrupt if enabled, and clear msi_lo.
+ * This is used primarily for the fallback to INTx, but
+ * is also used in reinit after reset, and during cleanup.
+ */
+void qib_nomsi(struct qib_devdata *dd)
+{
+	dd->msi_lo = 0;
+	qib_cdbg(INIT, "disable device MSI interrupts\n");
+	pci_disable_msi(dd->pcidev);
+}
+
+/*
+ * Same as qib_nosmi, but for MSIx.
+ */
+void qib_nomsix(struct qib_devdata *dd)
+{
+	qib_cdbg(INIT, "disable device MSIx interrupts\n");
+	pci_disable_msix(dd->pcidev);
+}
+
+/*
+ * Similar to pci_intx(pdev, 1), except that we make sure
+ * msi(x) is off.
+ */
+void qib_enable_intx(struct pci_dev *pdev)
+{
+	u16 cw, new;
+	int pos;
+
+	qib_cdbg(INIT, "Using INTx interrupts\n");
+
+	/* first, turn on INTx */
+	pci_read_config_word(pdev, PCI_COMMAND, &cw);
+	new = cw & ~PCI_COMMAND_INTX_DISABLE;
+	if (new != cw)
+		pci_write_config_word(pdev, PCI_COMMAND, new);
+
+	pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
+	if (pos) {
+		/* then turn off MSI */
+		pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &cw);
+		new = cw & ~PCI_MSI_FLAGS_ENABLE;
+		if (new != cw)
+			pci_write_config_word(pdev, pos + PCI_MSI_FLAGS, new);
+	}
+	pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
+	if (pos) {
+		/* then turn off MSIx */
+		pci_read_config_word(pdev, pos + PCI_MSIX_FLAGS, &cw);
+		new = cw & ~PCI_MSIX_FLAGS_ENABLE;
+		if (new != cw)
+			pci_write_config_word(pdev, pos + PCI_MSIX_FLAGS, new);
+	}
+}
+
+/*
+ * These two routines are helper routines for the device reset code
+ * to move all the pcie code out of the chip-specific driver code.
+ */
+void qib_pcie_getcmd(struct qib_devdata *dd, u16 *cmd, u8 *iline, u8 *cline)
+{
+	pci_read_config_word(dd->pcidev, PCI_COMMAND, cmd);
+	pci_read_config_byte(dd->pcidev, PCI_INTERRUPT_LINE, iline);
+	pci_read_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE, cline);
+}
+
+void qib_pcie_reenable(struct qib_devdata *dd, u16 cmd, u8 iline, u8 cline)
+{
+	int r;
+	r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
+				   dd->pcibar0);
+	if (r)
+		qib_dev_err(dd, "rewrite of BAR0 failed: %d\n", r);
+	r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
+				   dd->pcibar1);
+	if (r)
+		qib_dev_err(dd, "rewrite of BAR1 failed: %d\n", r);
+	/* now re-enable memory access, and restore cosmetic settings */
+	pci_write_config_word(dd->pcidev, PCI_COMMAND, cmd);
+	pci_write_config_byte(dd->pcidev, PCI_INTERRUPT_LINE, iline);
+	pci_write_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE, cline);
+	r = pci_enable_device(dd->pcidev);
+	if (r)
+		qib_dev_err(dd, "pci_enable_device failed after "
+			    "reset: %d\n", r);
+}
+
+/* code to adjust PCIe capabilities. */
+
+static int fld2val(int wd, int mask)
+{
+	int lsbmask;
+
+	if (!mask)
+		return 0;
+	wd &= mask;
+	lsbmask = mask ^ (mask & (mask - 1));
+	wd /= lsbmask;
+	return wd;
+}
+
+static int val2fld(int wd, int mask)
+{
+	int lsbmask;
+
+	if (!mask)
+		return 0;
+	lsbmask = mask ^ (mask & (mask - 1));
+	wd *= lsbmask;
+	return wd;
+}
+
+static int qib_pcie_coalesce;
+module_param_named(pcie_coalesce, qib_pcie_coalesce, int, S_IRUGO);
+MODULE_PARM_DESC(pcie_coalesce, "tune PCIe colescing on some Intel chipsets");
+
+/*
+ * Enable PCIe completion and data coalescing, on Intel 5x00 and 7300
+ * chipsets.   This is known to be unsafe for some revisions of some
+ * of these chipsets, with some BIOS settings, and enabling it on those
+ * systems may result in the system crashing, and/or data corruption.
+ */
+static int qib_tune_pcie_coalesce(struct qib_devdata *dd)
+{
+	int r;
+	struct pci_dev *parent;
+	int ppos;
+	u16 devid;
+	u32 mask, bits, val;
+
+	if (!qib_pcie_coalesce)
+		return 0;
+
+	/* Find out supported and configured values for parent (root) */
+	parent = dd->pcidev->bus->self;
+	if (parent->bus->parent) {
+		qib_devinfo(dd->pcidev, "Parent not root\n");
+		return 1;
+	}
+	ppos = pci_find_capability(parent, PCI_CAP_ID_EXP);
+	if (!ppos) {
+		qib_dbg("parent not PCIe root complex!?\n");
+		return 1;
+	}
+	if (parent->vendor != 0x8086) {
+		qib_cdbg(VERBOSE, "VendorID 0x%x isn't Intel, skip\n",
+			 parent->vendor);
+		return 1;
+	}
+
+	/*
+	 *  - bit 12: Max_rdcmp_Imt_EN: need to set to 1
+	 *  - bit 11: COALESCE_FORCE: need to set to 0
+	 *  - bit 10: COALESCE_EN: need to set to 1
+	 *  (but limitations on some on some chipsets)
+	 *
+	 *  On the Intel 5000, 5100, and 7300 chipsets, there is
+	 *  also: - bit 25:24: COALESCE_MODE, need to set to 0
+	 */
+	devid = parent->device;
+	if (devid >= 0x25e2 && devid <= 0x25fa) {
+		u8 rev;
+
+		/* 5000 P/V/X/Z */
+		pci_read_config_byte(parent, PCI_REVISION_ID, &rev);
+		if (rev <= 0xb2) {
+			bits = 1U << 10;
+			qib_cdbg(INIT, "Old rev 5000* (0x%x), enable-only\n",
+				 rev);
+		} else
+			bits = 7U << 10;
+		mask = (3U << 24) | (7U << 10);
+	} else if (devid >= 0x65e2 && devid <= 0x65fa) {
+		/* 5100 */
+		bits = 1U << 10;
+		mask = (3U << 24) | (7U << 10);
+	} else if (devid >= 0x4021 && devid <= 0x402e) {
+		/* 5400 */
+		bits = 7U << 10;
+		mask = 7U << 10;
+	} else if (devid >= 0x3604 && devid <= 0x360a) {
+		/* 7300 */
+		bits = 7U << 10;
+		mask = (3U << 24) | (7U << 10);
+	} else {
+		/* not one of the chipsets that we know about */
+		qib_cdbg(VERBOSE, "DeviceID 0x%x isn't one we know, skip\n",
+			 devid);
+		return 1;
+	}
+	pci_read_config_dword(parent, 0x48, &val);
+	qib_cdbg(VERBOSE, "Read initial value 0x%x at 0x48, deviceid 0x%x\n",
+		val, devid);
+	val &= ~mask;
+	val |= bits;
+	r = pci_write_config_dword(parent, 0x48, val);
+	if (r)
+		qib_dev_err(dd, "Unable to update deviceid 0x%x to val 0x%x"
+			    " for PCIe coalescing\n", devid, val);
+	else
+		qib_dbg("Updated deviceid 0x%x to val 0x%x"
+			 " for PCIe coalescing\n", devid, val);
+	return 0;
+}
+
+/*
+ * BIOS may not set PCIe bus-utilization parameters for best performance.
+ * Check and optionally adjust them to maximize our throughput.
+ */
+static int qib_pcie_caps;
+module_param_named(pcie_caps, qib_pcie_caps, int, S_IRUGO);
+MODULE_PARM_DESC(pcie_caps, "Max PCIe tuning: Payload (4lsb), ReadReq (D4..7)");
+
+static int qib_tune_pcie_caps(struct qib_devdata *dd)
+{
+	int ret = 1; /* Assume the worst */
+	struct pci_dev *parent;
+	int ppos, epos;
+	u16 pcaps, pctl, ecaps, ectl;
+	int rc_sup, ep_sup;
+	int rc_cur, ep_cur;
+
+	/* Find out supported and configured values for parent (root) */
+	parent = dd->pcidev->bus->self;
+	if (parent->bus->parent) {
+		qib_devinfo(dd->pcidev, "Parent not root\n");
+		goto bail;
+	}
+	ppos = pci_find_capability(parent, PCI_CAP_ID_EXP);
+	if (ppos) {
+		pci_read_config_word(parent, ppos + PCI_EXP_DEVCAP, &pcaps);
+		pci_read_config_word(parent, ppos + PCI_EXP_DEVCTL, &pctl);
+		qib_cdbg(VERBOSE, "Root caps %04X, ctl %04X\n", pcaps, pctl);
+	} else
+		goto bail;
+	/* Find out supported and configured values for endpoint (us) */
+	epos = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP);
+	if (epos) {
+		pci_read_config_word(dd->pcidev, epos + PCI_EXP_DEVCAP, &ecaps);
+		pci_read_config_word(dd->pcidev, epos + PCI_EXP_DEVCTL, &ectl);
+		qib_cdbg(VERBOSE, "our caps %04X, ctl %04X\n", ecaps, ectl);
+	} else
+		goto bail;
+	ret = 0;
+	/* Find max payload supported by root, endpoint */
+	rc_sup = fld2val(pcaps, PCI_EXP_DEVCAP_PAYLOAD);
+	ep_sup = fld2val(ecaps, PCI_EXP_DEVCAP_PAYLOAD);
+	qib_cdbg(VERBOSE, "Root supports %d payload, we support %d\n",
+		(1U << (rc_sup + 7)), (1U << (ep_sup + 7)));
+	if (rc_sup > ep_sup)
+		rc_sup = ep_sup;
+
+	rc_cur = fld2val(pctl, PCI_EXP_DEVCTL_PAYLOAD);
+	ep_cur = fld2val(ectl, PCI_EXP_DEVCTL_PAYLOAD);
+	qib_cdbg(VERBOSE, "Root set to %d payload, we are %d\n",
+		 (1U << (rc_cur + 7)), (1U << (ep_cur + 7)));
+
+	/* If Supported greater than limit in module param, limit it */
+	if (rc_sup > (qib_pcie_caps & 7))
+		rc_sup = qib_pcie_caps & 7;
+	/* If less than (allowed, supported), bump root payload */
+	if (rc_sup > rc_cur) {
+		rc_cur = rc_sup;
+		qib_cdbg(INIT, "Increase Root to %d Payload\n",
+			 (1U << (rc_cur + 7)));
+		pctl = (pctl & ~PCI_EXP_DEVCTL_PAYLOAD) |
+			val2fld(rc_cur, PCI_EXP_DEVCTL_PAYLOAD);
+		pci_write_config_word(parent, ppos + PCI_EXP_DEVCTL, pctl);
+	}
+	/* If less than (allowed, supported), bump endpoint payload */
+	if (rc_sup > ep_cur) {
+		ep_cur = rc_sup;
+		qib_cdbg(INIT, "Increase Endpoint to %d Payload\n",
+			 (1U << (ep_cur + 7)));
+		ectl = (ectl & ~PCI_EXP_DEVCTL_PAYLOAD) |
+			val2fld(ep_cur, PCI_EXP_DEVCTL_PAYLOAD);
+		pci_write_config_word(dd->pcidev, epos + PCI_EXP_DEVCTL, ectl);
+	}
+
+	/*
+	 * Now the Read Request size.
+	 * No field for max supported, but PCIe spec limits it to 4096,
+	 * which is code '5' (log2(4096) - 7)
+	 */
+	rc_sup = 5;
+	if (rc_sup > ((qib_pcie_caps >> 4) & 7))
+		rc_sup = (qib_pcie_caps >> 4) & 7;
+	rc_cur = fld2val(pctl, PCI_EXP_DEVCTL_READRQ);
+	ep_cur = fld2val(ectl, PCI_EXP_DEVCTL_READRQ);
+
+	qib_cdbg(VERBOSE, "Root set to %d rdreq, we are %d\n",
+		 1U << (rc_cur + 7), 1U << (ep_cur + 7));
+
+	if (rc_sup > rc_cur) {
+		rc_cur = rc_sup;
+		qib_cdbg(INIT, "Increase Root to %d RdReq\n",
+			 (1U << (rc_cur + 7)));
+		pctl = (pctl & ~PCI_EXP_DEVCTL_READRQ) |
+			val2fld(rc_cur, PCI_EXP_DEVCTL_READRQ);
+		pci_write_config_word(parent, ppos + PCI_EXP_DEVCTL, pctl);
+	}
+	if (rc_sup > ep_cur) {
+		ep_cur = rc_sup;
+		qib_cdbg(INIT, "Increase Endpoint to %d RdReq\n",
+			 (1U << (ep_cur + 7)));
+		ectl = (ectl & ~PCI_EXP_DEVCTL_READRQ) |
+			val2fld(ep_cur, PCI_EXP_DEVCTL_READRQ);
+		pci_write_config_word(dd->pcidev, epos + PCI_EXP_DEVCTL, ectl);
+	}
+bail:
+	return ret;
+}
+/* End of PCIe capability tuning */
+
+/*
+ * From here through qib_pci_err_handler definition is invoked via
+ * PCI error infrastructure, registered via pci
+ */
+static pci_ers_result_t
+qib_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
+{
+	struct qib_devdata *dd = pci_get_drvdata(pdev);
+	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
+
+	switch (state) {
+	case pci_channel_io_normal:
+		qib_devinfo(pdev, "State Normal, ignoring\n");
+		break;
+
+	case pci_channel_io_frozen:
+		qib_devinfo(pdev, "State Frozen, requesting reset\n");
+		pci_disable_device(pdev);
+		ret = PCI_ERS_RESULT_NEED_RESET;
+		break;
+
+	case pci_channel_io_perm_failure:
+		qib_devinfo(pdev, "State Permanent Failure, disabling\n");
+		if (dd) {
+			/* no more register accesses! */
+			dd->flags &= ~QIB_PRESENT;
+			qib_disable_after_error(dd);
+		}
+		 /* else early, or other problem */
+		ret =  PCI_ERS_RESULT_DISCONNECT;
+		break;
+
+	default: /* shouldn't happen */
+		qib_devinfo(pdev, "QIB PCI errors detected (state %d)\n",
+			state);
+		break;
+	}
+	return ret;
+}
+
+static pci_ers_result_t
+qib_pci_mmio_enabled(struct pci_dev *pdev)
+{
+	u64 words = 0U;
+	struct qib_devdata *dd = pci_get_drvdata(pdev);
+	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
+
+	if (dd && dd->pport) {
+		words = dd->f_portcntr(dd->pport, QIBPORTCNTR_WORDRCV);
+		if (words == ~0ULL)
+			ret = PCI_ERS_RESULT_NEED_RESET;
+	}
+	qib_devinfo(pdev, "QIB mmio_enabled function called, "
+		 "read wordscntr %Lx, returning %d\n", words, ret);
+	return  ret;
+}
+
+static pci_ers_result_t
+qib_pci_slot_reset(struct pci_dev *pdev)
+{
+	qib_devinfo(pdev, "QIB link_reset function called, ignored\n");
+	return PCI_ERS_RESULT_CAN_RECOVER;
+}
+
+static pci_ers_result_t
+qib_pci_link_reset(struct pci_dev *pdev)
+{
+	qib_devinfo(pdev, "QIB link_reset function called, ignored\n");
+	return PCI_ERS_RESULT_CAN_RECOVER;
+}
+
+static void
+qib_pci_resume(struct pci_dev *pdev)
+{
+	struct qib_devdata *dd = pci_get_drvdata(pdev);
+	qib_devinfo(pdev, "QIB resume function called\n");
+	pci_cleanup_aer_uncorrect_error_status(pdev);
+	/*
+	 * Running jobs will fail, since it's asynchronous
+	 * unlike sysfs-requested reset.   Better than
+	 * doing nothing.
+	 */
+	qib_init(dd, 1); /* same as re-init after reset */
+}
+
+struct pci_error_handlers qib_pci_err_handler = {
+	.error_detected = qib_pci_error_detected,
+	.mmio_enabled = qib_pci_mmio_enabled,
+	.link_reset = qib_pci_link_reset,
+	.slot_reset = qib_pci_slot_reset,
+	.resume = qib_pci_resume,
+};

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 28/51] IB/qib: Add qib_pio_copy.c

[Ralph Campbell]

creates the qib_pio_copy.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_pio_copy.c |   64 ++++++++++++++++++++++++++++++
 1 files changed, 64 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_pio_copy.c

diff --git a/drivers/infiniband/hw/qib/qib_pio_copy.c b/drivers/infiniband/hw/qib/qib_pio_copy.c
new file mode 100644
index 0000000..10b8c44
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_pio_copy.c
@@ -0,0 +1,64 @@
+/*
+ * Copyright (c) 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "qib.h"
+
+/**
+ * qib_pio_copy - copy data to MMIO space, in multiples of 32-bits
+ * @to: destination, in MMIO space (must be 64-bit aligned)
+ * @from: source (must be 64-bit aligned)
+ * @count: number of 32-bit quantities to copy
+ *
+ * Copy data from kernel space to MMIO space, in multiples of 32 bits at a
+ * time.  Order of access is not guaranteed, nor is a memory barrier
+ * performed afterwards.
+ */
+void qib_pio_copy(void __iomem *to, const void *from, size_t count)
+{
+#ifdef CONFIG_64BIT
+	u64 __iomem *dst = to;
+	const u64 *src = from;
+	const u64 *end = src + (count >> 1);
+
+	while (src < end)
+		__raw_writeq(*src++, dst++);
+	if (count & 1)
+		__raw_writel(*(const u32 *)src, dst);
+#else
+	u32 __iomem *dst = to;
+	const u32 *src = from;
+	const u32 *end = src + count;
+
+	while (src < end)
+		__raw_writel(*src++, dst++);
+#endif
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 29/51] IB/qib: Add qib_qp.c

[Ralph Campbell]

creates the qib_qp.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_qp.c | 1241 ++++++++++++++++++++++++++++++++++++
 1 files changed, 1241 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_qp.c

diff --git a/drivers/infiniband/hw/qib/qib_qp.c b/drivers/infiniband/hw/qib/qib_qp.c
new file mode 100644
index 0000000..d0ccbf5
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_qp.c
@@ -0,0 +1,1241 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/err.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+
+#define BITS_PER_PAGE           (PAGE_SIZE*BITS_PER_BYTE)
+#define BITS_PER_PAGE_MASK      (BITS_PER_PAGE-1)
+
+static inline unsigned mk_qpn(struct qib_qpn_table *qpt,
+			      struct qpn_map *map, unsigned off)
+{
+	return (map - qpt->map) * BITS_PER_PAGE + off;
+}
+
+static inline unsigned find_next_offset(struct qib_qpn_table *qpt,
+					struct qpn_map *map, unsigned off,
+					unsigned r)
+{
+	if (qpt->mask) {
+		off++;
+		if ((off & qpt->mask) >> 1 != r)
+			off = ((off & qpt->mask) ?
+				(off | qpt->mask) + 1 : off) | (r << 1);
+	} else
+		off = find_next_zero_bit(map->page, BITS_PER_PAGE, off);
+	return off;
+}
+
+/*
+ * Convert the AETH credit code into the number of credits.
+ */
+static u32 credit_table[31] = {
+	0,                      /* 0 */
+	1,                      /* 1 */
+	2,                      /* 2 */
+	3,                      /* 3 */
+	4,                      /* 4 */
+	6,                      /* 5 */
+	8,                      /* 6 */
+	12,                     /* 7 */
+	16,                     /* 8 */
+	24,                     /* 9 */
+	32,                     /* A */
+	48,                     /* B */
+	64,                     /* C */
+	96,                     /* D */
+	128,                    /* E */
+	192,                    /* F */
+	256,                    /* 10 */
+	384,                    /* 11 */
+	512,                    /* 12 */
+	768,                    /* 13 */
+	1024,                   /* 14 */
+	1536,                   /* 15 */
+	2048,                   /* 16 */
+	3072,                   /* 17 */
+	4096,                   /* 18 */
+	6144,                   /* 19 */
+	8192,                   /* 1A */
+	12288,                  /* 1B */
+	16384,                  /* 1C */
+	24576,                  /* 1D */
+	32768                   /* 1E */
+};
+
+static void get_map_page(struct qib_qpn_table *qpt, struct qpn_map *map)
+{
+	unsigned long page = get_zeroed_page(GFP_KERNEL);
+
+	/*
+	 * Free the page if someone raced with us installing it.
+	 */
+
+	spin_lock(&qpt->lock);
+	if (map->page)
+		free_page(page);
+	else
+		map->page = (void *)page;
+	spin_unlock(&qpt->lock);
+}
+
+/*
+ * Allocate the next available QPN or
+ * zero/one for QP type IB_QPT_SMI/IB_QPT_GSI.
+ */
+static int alloc_qpn(struct qib_devdata *dd, struct qib_qpn_table *qpt,
+		     enum ib_qp_type type, u8 port)
+{
+	u32 i, offset, max_scan, qpn;
+	struct qpn_map *map;
+	u32 ret;
+	int r;
+
+	if (type == IB_QPT_SMI || type == IB_QPT_GSI) {
+		unsigned n;
+
+		ret = type == IB_QPT_GSI;
+		n = 1 << (ret + 2 * (port - 1));
+		spin_lock(&qpt->lock);
+		if (qpt->flags & n)
+			ret = -EINVAL;
+		else
+			qpt->flags |= n;
+		spin_unlock(&qpt->lock);
+		goto bail;
+	}
+
+	r = smp_processor_id();
+	if (r >= dd->n_krcv_queues)
+		r %= dd->n_krcv_queues;
+	qpn = qpt->last + 1;
+	if (qpn >= QPN_MAX)
+		qpn = 2;
+	if (qpt->mask && ((qpn & qpt->mask) >> 1) != r)
+		qpn = ((qpn & qpt->mask) ? (qpn | qpt->mask) + 1 : qpn) |
+			(r << 1);
+	offset = qpn & BITS_PER_PAGE_MASK;
+	map = &qpt->map[qpn / BITS_PER_PAGE];
+	max_scan = qpt->nmaps - !offset;
+	for (i = 0;;) {
+		if (unlikely(!map->page)) {
+			get_map_page(qpt, map);
+			if (unlikely(!map->page))
+				break;
+		}
+		do {
+			if (!test_and_set_bit(offset, map->page)) {
+				qpt->last = qpn;
+				ret = qpn;
+				goto bail;
+			}
+			offset = find_next_offset(qpt, map, offset, r);
+			qpn = mk_qpn(qpt, map, offset);
+			/*
+			 * This test differs from alloc_pidmap().
+			 * If find_next_offset() does find a zero
+			 * bit, we don't need to check for QPN
+			 * wrapping around past our starting QPN.
+			 * We just need to be sure we don't loop
+			 * forever.
+			 */
+		} while (offset < BITS_PER_PAGE && qpn < QPN_MAX);
+		/*
+		 * In order to keep the number of pages allocated to a
+		 * minimum, we scan the all existing pages before increasing
+		 * the size of the bitmap table.
+		 */
+		if (++i > max_scan) {
+			if (qpt->nmaps == QPNMAP_ENTRIES)
+				break;
+			map = &qpt->map[qpt->nmaps++];
+			offset = qpt->mask ? (r << 1) : 0;
+		} else if (map < &qpt->map[qpt->nmaps]) {
+			++map;
+			offset = qpt->mask ? (r << 1) : 0;
+		} else {
+			map = &qpt->map[0];
+			offset = qpt->mask ? (r << 1) : 2;
+		}
+		qpn = mk_qpn(qpt, map, offset);
+	}
+
+	ret = -ENOMEM;
+
+bail:
+	return ret;
+}
+
+static void free_qpn(struct qib_qpn_table *qpt, u32 qpn)
+{
+	struct qpn_map *map;
+
+	map = qpt->map + qpn / BITS_PER_PAGE;
+	if (map->page)
+		clear_bit(qpn & BITS_PER_PAGE_MASK, map->page);
+}
+
+/*
+ * Put the QP into the hash table.
+ * The hash table holds a reference to the QP.
+ */
+static void insert_qp(struct qib_ibdev *dev, struct qib_qp *qp)
+{
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	unsigned n = qp->ibqp.qp_num % dev->qp_table_size;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dev->qpt_lock, flags);
+
+	if (qp->ibqp.qp_num == 0)
+		ibp->qp0 = qp;
+	else if (qp->ibqp.qp_num == 1)
+		ibp->qp1 = qp;
+	else {
+		qp->next = dev->qp_table[n];
+		dev->qp_table[n] = qp;
+	}
+	atomic_inc(&qp->refcount);
+
+	spin_unlock_irqrestore(&dev->qpt_lock, flags);
+}
+
+/*
+ * Remove the QP from the table so it can't be found asynchronously by
+ * the receive interrupt routine.
+ */
+static void remove_qp(struct qib_ibdev *dev, struct qib_qp *qp)
+{
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	struct qib_qp *q, **qpp;
+	unsigned long flags;
+
+	qpp = &dev->qp_table[qp->ibqp.qp_num % dev->qp_table_size];
+
+	spin_lock_irqsave(&dev->qpt_lock, flags);
+
+	if (ibp->qp0 == qp) {
+		ibp->qp0 = NULL;
+		atomic_dec(&qp->refcount);
+	} else if (ibp->qp1 == qp) {
+		ibp->qp1 = NULL;
+		atomic_dec(&qp->refcount);
+	} else
+		for (; (q = *qpp) != NULL; qpp = &q->next)
+			if (q == qp) {
+				*qpp = qp->next;
+				qp->next = NULL;
+				atomic_dec(&qp->refcount);
+				break;
+			}
+
+	spin_unlock_irqrestore(&dev->qpt_lock, flags);
+}
+
+/**
+ * qib_free_all_qps - check for QPs still in use
+ * @qpt: the QP table to empty
+ *
+ * There should not be any QPs still in use.
+ * Free memory for table.
+ */
+unsigned qib_free_all_qps(struct qib_devdata *dd)
+{
+	struct qib_ibdev *dev = &dd->verbs_dev;
+	unsigned long flags;
+	struct qib_qp *qp;
+	unsigned n, qp_inuse = 0;
+
+	for (n = 0; n < dd->num_pports; n++) {
+		struct qib_ibport *ibp = &dd->pport[n].ibport_data;
+
+		if (!qib_mcast_tree_empty(ibp))
+			qp_inuse++;
+		if (ibp->qp0)
+			qp_inuse++;
+		if (ibp->qp1)
+			qp_inuse++;
+	}
+
+	spin_lock_irqsave(&dev->qpt_lock, flags);
+	for (n = 0; n < dev->qp_table_size; n++) {
+		qp = dev->qp_table[n];
+		dev->qp_table[n] = NULL;
+
+		for (; qp; qp = qp->next)
+			qp_inuse++;
+	}
+	spin_unlock_irqrestore(&dev->qpt_lock, flags);
+
+	return qp_inuse;
+}
+
+/**
+ * qib_lookup_qpn - return the QP with the given QPN
+ * @qpt: the QP table
+ * @qpn: the QP number to look up
+ *
+ * The caller is responsible for decrementing the QP reference count
+ * when done.
+ */
+struct qib_qp *qib_lookup_qpn(struct qib_ibport *ibp, u32 qpn)
+{
+	struct qib_ibdev *dev = &ppd_from_ibp(ibp)->dd->verbs_dev;
+	unsigned long flags;
+	struct qib_qp *qp;
+
+	spin_lock_irqsave(&dev->qpt_lock, flags);
+
+	if (qpn == 0)
+		qp = ibp->qp0;
+	else if (qpn == 1)
+		qp = ibp->qp1;
+	else
+		for (qp = dev->qp_table[qpn % dev->qp_table_size]; qp;
+		     qp = qp->next)
+			if (qp->ibqp.qp_num == qpn)
+				break;
+	if (qp)
+		atomic_inc(&qp->refcount);
+
+	spin_unlock_irqrestore(&dev->qpt_lock, flags);
+	return qp;
+}
+
+/**
+ * qib_reset_qp - initialize the QP state to the reset state
+ * @qp: the QP to reset
+ * @type: the QP type
+ */
+static void qib_reset_qp(struct qib_qp *qp, enum ib_qp_type type)
+{
+	qp->remote_qpn = 0;
+	qp->qkey = 0;
+	qp->qp_access_flags = 0;
+	atomic_set(&qp->s_dma_busy, 0);
+	qp->s_flags &= QIB_S_SIGNAL_REQ_WR;
+	qp->s_hdrwords = 0;
+	qp->s_wqe = NULL;
+	qp->s_draining = 0;
+	qp->s_next_psn = 0;
+	qp->s_last_psn = 0;
+	qp->s_sending_psn = 0;
+	qp->s_sending_hpsn = 0;
+	qp->s_psn = 0;
+	qp->r_psn = 0;
+	qp->r_msn = 0;
+	if (type == IB_QPT_RC) {
+		qp->s_state = IB_OPCODE_RC_SEND_LAST;
+		qp->r_state = IB_OPCODE_RC_SEND_LAST;
+	} else {
+		qp->s_state = IB_OPCODE_UC_SEND_LAST;
+		qp->r_state = IB_OPCODE_UC_SEND_LAST;
+	}
+	qp->s_ack_state = IB_OPCODE_RC_ACKNOWLEDGE;
+	qp->r_nak_state = 0;
+	qp->r_aflags = 0;
+	qp->r_flags = 0;
+	qp->s_head = 0;
+	qp->s_tail = 0;
+	qp->s_cur = 0;
+	qp->s_acked = 0;
+	qp->s_last = 0;
+	qp->s_ssn = 1;
+	qp->s_lsn = 0;
+	qp->s_mig_state = IB_MIG_MIGRATED;
+	memset(qp->s_ack_queue, 0, sizeof(qp->s_ack_queue));
+	qp->r_head_ack_queue = 0;
+	qp->s_tail_ack_queue = 0;
+	qp->s_num_rd_atomic = 0;
+	if (qp->r_rq.wq) {
+		qp->r_rq.wq->head = 0;
+		qp->r_rq.wq->tail = 0;
+	}
+	qp->r_sge.num_sge = 0;
+}
+
+static void clear_mr_refs(struct qib_qp *qp, int clr_sends)
+{
+	unsigned n;
+
+	while (qp->r_sge.num_sge) {
+		atomic_dec(&qp->r_sge.sge.mr->refcount);
+		if (--qp->r_sge.num_sge)
+			qp->r_sge.sge = *qp->r_sge.sg_list++;
+	}
+
+	if (clr_sends) {
+		while (qp->s_last != qp->s_head) {
+			struct qib_swqe *wqe = get_swqe_ptr(qp, qp->s_last);
+			unsigned i;
+
+			for (i = 0; i < wqe->wr.num_sge; i++) {
+				struct qib_sge *sge = &wqe->sg_list[i];
+
+				atomic_dec(&sge->mr->refcount);
+			}
+			if (qp->ibqp.qp_type == IB_QPT_UD ||
+			    qp->ibqp.qp_type == IB_QPT_SMI ||
+			    qp->ibqp.qp_type == IB_QPT_GSI)
+				atomic_dec(&to_iah(wqe->wr.wr.ud.ah)->refcount);
+			if (++qp->s_last >= qp->s_size)
+				qp->s_last = 0;
+		}
+		if (qp->s_rdma_mr) {
+			atomic_dec(&qp->s_rdma_mr->refcount);
+			qp->s_rdma_mr = NULL;
+		}
+	}
+
+	if (qp->ibqp.qp_type != IB_QPT_RC)
+		return;
+
+	for (n = 0; n < ARRAY_SIZE(qp->s_ack_queue); n++) {
+		struct qib_ack_entry *e = &qp->s_ack_queue[n];
+
+		if (e->opcode == IB_OPCODE_RC_RDMA_READ_REQUEST &&
+		    e->rdma_sge.mr) {
+			atomic_dec(&e->rdma_sge.mr->refcount);
+			e->rdma_sge.mr = NULL;
+		}
+	}
+}
+
+/**
+ * qib_error_qp - put a QP into the error state
+ * @qp: the QP to put into the error state
+ * @err: the receive completion error to signal if a RWQE is active
+ *
+ * Flushes both send and receive work queues.
+ * Returns true if last WQE event should be generated.
+ * The QP s_lock should be held and interrupts disabled.
+ * If we are already in error state, just return.
+ */
+int qib_error_qp(struct qib_qp *qp, enum ib_wc_status err)
+{
+	struct qib_ibdev *dev = to_idev(qp->ibqp.device);
+	struct ib_wc wc;
+	int ret = 0;
+
+	if (qp->state == IB_QPS_ERR)
+		goto bail;
+
+	qp->state = IB_QPS_ERR;
+
+	if (qp->s_flags & (QIB_S_TIMER | QIB_S_WAIT_RNR)) {
+		qp->s_flags &= ~(QIB_S_TIMER | QIB_S_WAIT_RNR);
+		del_timer(&qp->s_timer);
+	}
+	spin_lock(&dev->pending_lock);
+	if (!list_empty(&qp->iowait) && !(qp->s_flags & QIB_S_BUSY)) {
+		qp->s_flags &= ~QIB_S_ANY_WAIT_IO;
+		list_del_init(&qp->iowait);
+	}
+	spin_unlock(&dev->pending_lock);
+
+	if (!(qp->s_flags & QIB_S_BUSY)) {
+		qp->s_hdrwords = 0;
+		if (qp->s_rdma_mr) {
+			atomic_dec(&qp->s_rdma_mr->refcount);
+			qp->s_rdma_mr = NULL;
+		}
+		if (qp->s_tx) {
+			qib_put_txreq(qp->s_tx);
+			qp->s_tx = NULL;
+		}
+	}
+
+	/* Schedule the sending tasklet to drain the send work queue. */
+	if (qp->s_last != qp->s_head)
+		qib_schedule_send(qp);
+
+	clear_mr_refs(qp, 0);
+
+	memset(&wc, 0, sizeof(wc));
+	wc.qp = &qp->ibqp;
+	wc.opcode = IB_WC_RECV;
+
+	if (test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags)) {
+		wc.wr_id = qp->r_wr_id;
+		wc.status = err;
+		qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1);
+	}
+	wc.status = IB_WC_WR_FLUSH_ERR;
+
+	if (qp->r_rq.wq) {
+		struct qib_rwq *wq;
+		u32 head;
+		u32 tail;
+
+		spin_lock(&qp->r_rq.lock);
+
+		/* sanity check pointers before trusting them */
+		wq = qp->r_rq.wq;
+		head = wq->head;
+		if (head >= qp->r_rq.size)
+			head = 0;
+		tail = wq->tail;
+		if (tail >= qp->r_rq.size)
+			tail = 0;
+		while (tail != head) {
+			wc.wr_id = get_rwqe_ptr(&qp->r_rq, tail)->wr_id;
+			if (++tail >= qp->r_rq.size)
+				tail = 0;
+			qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1);
+		}
+		wq->tail = tail;
+
+		spin_unlock(&qp->r_rq.lock);
+	} else if (qp->ibqp.event_handler)
+		ret = 1;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_modify_qp - modify the attributes of a queue pair
+ * @ibqp: the queue pair who's attributes we're modifying
+ * @attr: the new attributes
+ * @attr_mask: the mask of attributes to modify
+ * @udata: user data for libibverbs.so
+ *
+ * Returns 0 on success, otherwise returns an errno.
+ */
+int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		  int attr_mask, struct ib_udata *udata)
+{
+	struct qib_ibdev *dev = to_idev(ibqp->device);
+	struct qib_qp *qp = to_iqp(ibqp);
+	enum ib_qp_state cur_state, new_state;
+	struct ib_event ev;
+	int lastwqe = 0;
+	int mig = 0;
+	int ret;
+	u32 pmtu = 0; /* for gcc warning only */
+
+	spin_lock_irq(&qp->s_lock);
+
+	cur_state = attr_mask & IB_QP_CUR_STATE ?
+		attr->cur_qp_state : qp->state;
+	new_state = attr_mask & IB_QP_STATE ? attr->qp_state : cur_state;
+
+	if (!ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type,
+				attr_mask))
+		goto inval;
+
+	if (attr_mask & IB_QP_AV) {
+		if (attr->ah_attr.dlid >= QIB_MULTICAST_LID_BASE)
+			goto inval;
+		if (qib_check_ah(qp->ibqp.device, &attr->ah_attr))
+			goto inval;
+	}
+
+	if (attr_mask & IB_QP_ALT_PATH) {
+		if (attr->alt_ah_attr.dlid >= QIB_MULTICAST_LID_BASE)
+			goto inval;
+		if (qib_check_ah(qp->ibqp.device, &attr->alt_ah_attr))
+			goto inval;
+		if (attr->alt_pkey_index >= qib_get_npkeys(dd_from_dev(dev)))
+			goto inval;
+	}
+
+	if (attr_mask & IB_QP_PKEY_INDEX)
+		if (attr->pkey_index >= qib_get_npkeys(dd_from_dev(dev)))
+			goto inval;
+
+	if (attr_mask & IB_QP_MIN_RNR_TIMER)
+		if (attr->min_rnr_timer > 31)
+			goto inval;
+
+	if (attr_mask & IB_QP_PORT)
+		if (qp->ibqp.qp_type == IB_QPT_SMI ||
+		    qp->ibqp.qp_type == IB_QPT_GSI ||
+		    attr->port_num == 0 ||
+		    attr->port_num > ibqp->device->phys_port_cnt)
+			goto inval;
+
+	if (attr_mask & IB_QP_DEST_QPN)
+		if (attr->dest_qp_num > QIB_QPN_MASK)
+			goto inval;
+
+	if (attr_mask & IB_QP_RETRY_CNT)
+		if (attr->retry_cnt > 7)
+			goto inval;
+
+	if (attr_mask & IB_QP_RNR_RETRY)
+		if (attr->rnr_retry > 7)
+			goto inval;
+
+	/*
+	 * Don't allow invalid path_mtu values.  OK to set greater
+	 * than the active mtu (or even the max_cap, if we have tuned
+	 * that to a small mtu.  We'll set qp->path_mtu
+	 * to the lesser of requested attribute mtu and active,
+	 * for packetizing messages.
+	 * Note that the QP port has to be set in INIT and MTU in RTR.
+	 */
+	if (attr_mask & IB_QP_PATH_MTU) {
+		struct qib_devdata *dd = dd_from_dev(dev);
+		int mtu, pidx = qp->port_num - 1;
+
+		mtu = ib_mtu_enum_to_int(attr->path_mtu);
+		if (mtu == -1)
+			goto inval;
+		if (mtu > dd->pport[pidx].ibmtu) {
+			switch (dd->pport[pidx].ibmtu) {
+			case 4096:
+				pmtu = IB_MTU_4096;
+				break;
+			case 2048:
+				pmtu = IB_MTU_2048;
+				break;
+			case 1024:
+				pmtu = IB_MTU_1024;
+				break;
+			case 512:
+				pmtu = IB_MTU_512;
+				break;
+			case 256:
+				pmtu = IB_MTU_256;
+				break;
+			default:
+				pmtu = IB_MTU_2048;
+			}
+		} else
+			pmtu = attr->path_mtu;
+	}
+
+	if (attr_mask & IB_QP_PATH_MIG_STATE) {
+		if (attr->path_mig_state == IB_MIG_REARM) {
+			if (qp->s_mig_state == IB_MIG_ARMED)
+				goto inval;
+			if (new_state != IB_QPS_RTS)
+				goto inval;
+		} else if (attr->path_mig_state == IB_MIG_MIGRATED) {
+			if (qp->s_mig_state == IB_MIG_REARM)
+				goto inval;
+			if (new_state != IB_QPS_RTS && new_state != IB_QPS_SQD)
+				goto inval;
+			if (qp->s_mig_state == IB_MIG_ARMED)
+				mig = 1;
+		} else
+			goto inval;
+	}
+
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
+		if (attr->max_dest_rd_atomic > QIB_MAX_RDMA_ATOMIC)
+			goto inval;
+
+	switch (new_state) {
+	case IB_QPS_RESET:
+		if (qp->state != IB_QPS_RESET) {
+			qp->state = IB_QPS_RESET;
+			spin_lock(&dev->pending_lock);
+			if (!list_empty(&qp->iowait))
+				list_del_init(&qp->iowait);
+			spin_unlock(&dev->pending_lock);
+			qp->s_flags &= ~(QIB_S_TIMER | QIB_S_ANY_WAIT);
+			spin_unlock_irq(&qp->s_lock);
+			/* Stop the sending work queue and retry timer */
+			cancel_work_sync(&qp->s_work);
+			del_timer_sync(&qp->s_timer);
+			wait_event(qp->wait_dma, !atomic_read(&qp->s_dma_busy));
+			if (qp->s_tx) {
+				qib_put_txreq(qp->s_tx);
+				qp->s_tx = NULL;
+			}
+			remove_qp(dev, qp);
+			wait_event(qp->wait, !atomic_read(&qp->refcount));
+			spin_lock_irq(&qp->s_lock);
+			clear_mr_refs(qp, 1);
+			qib_reset_qp(qp, ibqp->qp_type);
+		}
+		break;
+
+	case IB_QPS_RTR:
+		/* Allow event to retrigger if QP set to RTR more than once */
+		qp->r_flags &= ~QIB_R_COMM_EST;
+		qp->state = new_state;
+		break;
+
+	case IB_QPS_SQD:
+		qp->s_draining = qp->s_last != qp->s_cur;
+		qp->state = new_state;
+		break;
+
+	case IB_QPS_SQE:
+		if (qp->ibqp.qp_type == IB_QPT_RC)
+			goto inval;
+		qp->state = new_state;
+		break;
+
+	case IB_QPS_ERR:
+		lastwqe = qib_error_qp(qp, IB_WC_WR_FLUSH_ERR);
+		break;
+
+	default:
+		qp->state = new_state;
+		break;
+	}
+
+	if (attr_mask & IB_QP_PKEY_INDEX)
+		qp->s_pkey_index = attr->pkey_index;
+
+	if (attr_mask & IB_QP_PORT)
+		qp->port_num = attr->port_num;
+
+	if (attr_mask & IB_QP_DEST_QPN)
+		qp->remote_qpn = attr->dest_qp_num;
+
+	if (attr_mask & IB_QP_SQ_PSN) {
+		qp->s_next_psn = attr->sq_psn & QIB_PSN_MASK;
+		qp->s_psn = qp->s_next_psn;
+		qp->s_sending_psn = qp->s_next_psn;
+		qp->s_last_psn = qp->s_next_psn - 1;
+		qp->s_sending_hpsn = qp->s_last_psn;
+	}
+
+	if (attr_mask & IB_QP_RQ_PSN)
+		qp->r_psn = attr->rq_psn & QIB_PSN_MASK;
+
+	if (attr_mask & IB_QP_ACCESS_FLAGS)
+		qp->qp_access_flags = attr->qp_access_flags;
+
+	if (attr_mask & IB_QP_AV) {
+		qp->remote_ah_attr = attr->ah_attr;
+		qp->s_srate = attr->ah_attr.static_rate;
+	}
+
+	if (attr_mask & IB_QP_ALT_PATH) {
+		qp->alt_ah_attr = attr->alt_ah_attr;
+		qp->s_alt_pkey_index = attr->alt_pkey_index;
+	}
+
+	if (attr_mask & IB_QP_PATH_MIG_STATE) {
+		qp->s_mig_state = attr->path_mig_state;
+		if (mig) {
+			qp->remote_ah_attr = qp->alt_ah_attr;
+			qp->port_num = qp->alt_ah_attr.port_num;
+			qp->s_pkey_index = qp->s_alt_pkey_index;
+		}
+	}
+
+	if (attr_mask & IB_QP_PATH_MTU)
+		qp->path_mtu = pmtu;
+
+	if (attr_mask & IB_QP_RETRY_CNT) {
+		qp->s_retry_cnt = attr->retry_cnt;
+		qp->s_retry = attr->retry_cnt;
+	}
+
+	if (attr_mask & IB_QP_RNR_RETRY) {
+		qp->s_rnr_retry_cnt = attr->rnr_retry;
+		qp->s_rnr_retry = attr->rnr_retry;
+	}
+
+	if (attr_mask & IB_QP_MIN_RNR_TIMER)
+		qp->r_min_rnr_timer = attr->min_rnr_timer;
+
+	if (attr_mask & IB_QP_TIMEOUT)
+		qp->timeout = attr->timeout;
+
+	if (attr_mask & IB_QP_QKEY)
+		qp->qkey = attr->qkey;
+
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
+		qp->r_max_rd_atomic = attr->max_dest_rd_atomic;
+
+	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC)
+		qp->s_max_rd_atomic = attr->max_rd_atomic;
+
+	spin_unlock_irq(&qp->s_lock);
+
+	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+		insert_qp(dev, qp);
+
+	if (lastwqe) {
+		ev.device = qp->ibqp.device;
+		ev.element.qp = &qp->ibqp;
+		ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
+		qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+	}
+	if (mig) {
+		ev.device = qp->ibqp.device;
+		ev.element.qp = &qp->ibqp;
+		ev.event = IB_EVENT_PATH_MIG;
+		qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+	}
+	ret = 0;
+	goto bail;
+
+inval:
+	spin_unlock_irq(&qp->s_lock);
+	ret = -EINVAL;
+
+bail:
+	return ret;
+}
+
+int qib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		 int attr_mask, struct ib_qp_init_attr *init_attr)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+
+	attr->qp_state = qp->state;
+	attr->cur_qp_state = attr->qp_state;
+	attr->path_mtu = qp->path_mtu;
+	attr->path_mig_state = qp->s_mig_state;
+	attr->qkey = qp->qkey;
+	attr->rq_psn = qp->r_psn & QIB_PSN_MASK;
+	attr->sq_psn = qp->s_next_psn & QIB_PSN_MASK;
+	attr->dest_qp_num = qp->remote_qpn;
+	attr->qp_access_flags = qp->qp_access_flags;
+	attr->cap.max_send_wr = qp->s_size - 1;
+	attr->cap.max_recv_wr = qp->ibqp.srq ? 0 : qp->r_rq.size - 1;
+	attr->cap.max_send_sge = qp->s_max_sge;
+	attr->cap.max_recv_sge = qp->r_rq.max_sge;
+	attr->cap.max_inline_data = 0;
+	attr->ah_attr = qp->remote_ah_attr;
+	attr->alt_ah_attr = qp->alt_ah_attr;
+	attr->pkey_index = qp->s_pkey_index;
+	attr->alt_pkey_index = qp->s_alt_pkey_index;
+	attr->en_sqd_async_notify = 0;
+	attr->sq_draining = qp->s_draining;
+	attr->max_rd_atomic = qp->s_max_rd_atomic;
+	attr->max_dest_rd_atomic = qp->r_max_rd_atomic;
+	attr->min_rnr_timer = qp->r_min_rnr_timer;
+	attr->port_num = qp->port_num;
+	attr->timeout = qp->timeout;
+	attr->retry_cnt = qp->s_retry_cnt;
+	attr->rnr_retry = qp->s_rnr_retry_cnt;
+	attr->alt_port_num = qp->alt_ah_attr.port_num;
+	attr->alt_timeout = qp->alt_timeout;
+
+	init_attr->event_handler = qp->ibqp.event_handler;
+	init_attr->qp_context = qp->ibqp.qp_context;
+	init_attr->send_cq = qp->ibqp.send_cq;
+	init_attr->recv_cq = qp->ibqp.recv_cq;
+	init_attr->srq = qp->ibqp.srq;
+	init_attr->cap = attr->cap;
+	if (qp->s_flags & QIB_S_SIGNAL_REQ_WR)
+		init_attr->sq_sig_type = IB_SIGNAL_REQ_WR;
+	else
+		init_attr->sq_sig_type = IB_SIGNAL_ALL_WR;
+	init_attr->qp_type = qp->ibqp.qp_type;
+	init_attr->port_num = qp->port_num;
+	return 0;
+}
+
+/**
+ * qib_compute_aeth - compute the AETH (syndrome + MSN)
+ * @qp: the queue pair to compute the AETH for
+ *
+ * Returns the AETH.
+ */
+__be32 qib_compute_aeth(struct qib_qp *qp)
+{
+	u32 aeth = qp->r_msn & QIB_MSN_MASK;
+
+	if (qp->ibqp.srq) {
+		/*
+		 * Shared receive queues don't generate credits.
+		 * Set the credit field to the invalid value.
+		 */
+		aeth |= QIB_AETH_CREDIT_INVAL << QIB_AETH_CREDIT_SHIFT;
+	} else {
+		u32 min, max, x;
+		u32 credits;
+		struct qib_rwq *wq = qp->r_rq.wq;
+		u32 head;
+		u32 tail;
+
+		/* sanity check pointers before trusting them */
+		head = wq->head;
+		if (head >= qp->r_rq.size)
+			head = 0;
+		tail = wq->tail;
+		if (tail >= qp->r_rq.size)
+			tail = 0;
+		/*
+		 * Compute the number of credits available (RWQEs).
+		 * XXX Not holding the r_rq.lock here so there is a small
+		 * chance that the pair of reads are not atomic.
+		 */
+		credits = head - tail;
+		if ((int)credits < 0)
+			credits += qp->r_rq.size;
+		/*
+		 * Binary search the credit table to find the code to
+		 * use.
+		 */
+		min = 0;
+		max = 31;
+		for (;;) {
+			x = (min + max) / 2;
+			if (credit_table[x] == credits)
+				break;
+			if (credit_table[x] > credits)
+				max = x;
+			else if (min == x)
+				break;
+			else
+				min = x;
+		}
+		aeth |= x << QIB_AETH_CREDIT_SHIFT;
+	}
+	return cpu_to_be32(aeth);
+}
+
+/**
+ * qib_create_qp - create a queue pair for a device
+ * @ibpd: the protection domain who's device we create the queue pair for
+ * @init_attr: the attributes of the queue pair
+ * @udata: user data for libibverbs.so
+ *
+ * Returns the queue pair on success, otherwise returns an errno.
+ *
+ * Called by the ib_create_qp() core verbs function.
+ */
+struct ib_qp *qib_create_qp(struct ib_pd *ibpd,
+			    struct ib_qp_init_attr *init_attr,
+			    struct ib_udata *udata)
+{
+	struct qib_qp *qp;
+	int err;
+	struct qib_swqe *swq = NULL;
+	struct qib_ibdev *dev;
+	struct qib_devdata *dd;
+	size_t sz;
+	size_t sg_list_sz;
+	struct ib_qp *ret;
+
+	if (init_attr->cap.max_send_sge > ib_qib_max_sges ||
+	    init_attr->cap.max_send_wr > ib_qib_max_qp_wrs) {
+		ret = ERR_PTR(-EINVAL);
+		goto bail;
+	}
+
+	/* Check receive queue parameters if no SRQ is specified. */
+	if (!init_attr->srq) {
+		if (init_attr->cap.max_recv_sge > ib_qib_max_sges ||
+		    init_attr->cap.max_recv_wr > ib_qib_max_qp_wrs) {
+			ret = ERR_PTR(-EINVAL);
+			goto bail;
+		}
+		if (init_attr->cap.max_send_sge +
+		    init_attr->cap.max_send_wr +
+		    init_attr->cap.max_recv_sge +
+		    init_attr->cap.max_recv_wr == 0) {
+			ret = ERR_PTR(-EINVAL);
+			goto bail;
+		}
+	}
+
+	switch (init_attr->qp_type) {
+	case IB_QPT_SMI:
+	case IB_QPT_GSI:
+		if (init_attr->port_num == 0 ||
+		    init_attr->port_num > ibpd->device->phys_port_cnt) {
+			ret = ERR_PTR(-EINVAL);
+			goto bail;
+		}
+	case IB_QPT_UC:
+	case IB_QPT_RC:
+	case IB_QPT_UD:
+		sz = sizeof(struct qib_sge) *
+			init_attr->cap.max_send_sge +
+			sizeof(struct qib_swqe);
+		swq = vmalloc((init_attr->cap.max_send_wr + 1) * sz);
+		if (swq == NULL) {
+			ret = ERR_PTR(-ENOMEM);
+			goto bail;
+		}
+		sz = sizeof(*qp);
+		sg_list_sz = 0;
+		if (init_attr->srq) {
+			struct qib_srq *srq = to_isrq(init_attr->srq);
+
+			if (srq->rq.max_sge > 1)
+				sg_list_sz = sizeof(*qp->r_sg_list) *
+					(srq->rq.max_sge - 1);
+		} else if (init_attr->cap.max_recv_sge > 1)
+			sg_list_sz = sizeof(*qp->r_sg_list) *
+				(init_attr->cap.max_recv_sge - 1);
+		qp = kzalloc(sz + sg_list_sz, GFP_KERNEL);
+		if (!qp) {
+			ret = ERR_PTR(-ENOMEM);
+			goto bail_swq;
+		}
+		if (init_attr->srq)
+			sz = 0;
+		else {
+			qp->r_rq.size = init_attr->cap.max_recv_wr + 1;
+			qp->r_rq.max_sge = init_attr->cap.max_recv_sge;
+			sz = (sizeof(struct ib_sge) * qp->r_rq.max_sge) +
+				sizeof(struct qib_rwqe);
+			qp->r_rq.wq = vmalloc_user(sizeof(struct qib_rwq) +
+						   qp->r_rq.size * sz);
+			if (!qp->r_rq.wq) {
+				ret = ERR_PTR(-ENOMEM);
+				goto bail_qp;
+			}
+		}
+
+		/*
+		 * ib_create_qp() will initialize qp->ibqp
+		 * except for qp->ibqp.qp_num.
+		 */
+		spin_lock_init(&qp->r_lock);
+		spin_lock_init(&qp->s_lock);
+		spin_lock_init(&qp->r_rq.lock);
+		atomic_set(&qp->refcount, 0);
+		init_waitqueue_head(&qp->wait);
+		init_waitqueue_head(&qp->wait_dma);
+		init_timer(&qp->s_timer);
+		qp->s_timer.data = (unsigned long)qp;
+		INIT_WORK(&qp->s_work, qib_do_send);
+		INIT_LIST_HEAD(&qp->iowait);
+		INIT_LIST_HEAD(&qp->rspwait);
+		qp->state = IB_QPS_RESET;
+		qp->s_wq = swq;
+		qp->s_size = init_attr->cap.max_send_wr + 1;
+		qp->s_max_sge = init_attr->cap.max_send_sge;
+		if (init_attr->sq_sig_type == IB_SIGNAL_REQ_WR)
+			qp->s_flags = QIB_S_SIGNAL_REQ_WR;
+		dev = to_idev(ibpd->device);
+		dd = dd_from_dev(dev);
+		err = alloc_qpn(dd, &dev->qpn_table, init_attr->qp_type,
+				init_attr->port_num);
+		if (err < 0) {
+			ret = ERR_PTR(err);
+			vfree(qp->r_rq.wq);
+			goto bail_qp;
+		}
+		qp->ibqp.qp_num = err;
+		qp->port_num = init_attr->port_num;
+		qp->processor_id = smp_processor_id();
+		qib_reset_qp(qp, init_attr->qp_type);
+		break;
+
+	default:
+		/* Don't support raw QPs */
+		ret = ERR_PTR(-ENOSYS);
+		goto bail;
+	}
+
+	init_attr->cap.max_inline_data = 0;
+
+	/*
+	 * Return the address of the RWQ as the offset to mmap.
+	 * See qib_mmap() for details.
+	 */
+	if (udata && udata->outlen >= sizeof(__u64)) {
+		if (!qp->r_rq.wq) {
+			__u64 offset = 0;
+
+			err = ib_copy_to_udata(udata, &offset,
+					       sizeof(offset));
+			if (err) {
+				ret = ERR_PTR(err);
+				goto bail_ip;
+			}
+		} else {
+			u32 s = sizeof(struct qib_rwq) + qp->r_rq.size * sz;
+
+			qp->ip = qib_create_mmap_info(dev, s,
+						      ibpd->uobject->context,
+						      qp->r_rq.wq);
+			if (!qp->ip) {
+				ret = ERR_PTR(-ENOMEM);
+				goto bail_ip;
+			}
+
+			err = ib_copy_to_udata(udata, &(qp->ip->offset),
+					       sizeof(qp->ip->offset));
+			if (err) {
+				ret = ERR_PTR(err);
+				goto bail_ip;
+			}
+		}
+	}
+
+	spin_lock(&dev->n_qps_lock);
+	if (dev->n_qps_allocated == ib_qib_max_qps) {
+		spin_unlock(&dev->n_qps_lock);
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_ip;
+	}
+
+	dev->n_qps_allocated++;
+	spin_unlock(&dev->n_qps_lock);
+
+	if (qp->ip) {
+		spin_lock_irq(&dev->pending_lock);
+		list_add(&qp->ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
+	ret = &qp->ibqp;
+	goto bail;
+
+bail_ip:
+	if (qp->ip)
+		kref_put(&qp->ip->ref, qib_release_mmap_info);
+	else
+		vfree(qp->r_rq.wq);
+	free_qpn(&dev->qpn_table, qp->ibqp.qp_num);
+bail_qp:
+	kfree(qp);
+bail_swq:
+	vfree(swq);
+bail:
+	return ret;
+}
+
+/**
+ * qib_destroy_qp - destroy a queue pair
+ * @ibqp: the queue pair to destroy
+ *
+ * Returns 0 on success.
+ *
+ * Note that this can be called while the QP is actively sending or
+ * receiving!
+ */
+int qib_destroy_qp(struct ib_qp *ibqp)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+	struct qib_ibdev *dev = to_idev(ibqp->device);
+
+	/* Make sure HW and driver activity is stopped. */
+	spin_lock_irq(&qp->s_lock);
+	if (qp->state != IB_QPS_RESET) {
+		qp->state = IB_QPS_RESET;
+		spin_lock(&dev->pending_lock);
+		if (!list_empty(&qp->iowait))
+			list_del_init(&qp->iowait);
+		spin_unlock(&dev->pending_lock);
+		qp->s_flags &= ~(QIB_S_TIMER | QIB_S_ANY_WAIT);
+		spin_unlock_irq(&qp->s_lock);
+		cancel_work_sync(&qp->s_work);
+		del_timer_sync(&qp->s_timer);
+		wait_event(qp->wait_dma, !atomic_read(&qp->s_dma_busy));
+		if (qp->s_tx) {
+			qib_put_txreq(qp->s_tx);
+			qp->s_tx = NULL;
+		}
+		remove_qp(dev, qp);
+		wait_event(qp->wait, !atomic_read(&qp->refcount));
+		clear_mr_refs(qp, 1);
+	} else
+		spin_unlock_irq(&qp->s_lock);
+
+	/* all user's cleaned up, mark it available */
+	free_qpn(&dev->qpn_table, qp->ibqp.qp_num);
+	spin_lock(&dev->n_qps_lock);
+	dev->n_qps_allocated--;
+	spin_unlock(&dev->n_qps_lock);
+
+	if (qp->ip)
+		kref_put(&qp->ip->ref, qib_release_mmap_info);
+	else
+		vfree(qp->r_rq.wq);
+	vfree(qp->s_wq);
+	kfree(qp);
+	return 0;
+}
+
+/**
+ * qib_init_qpn_table - initialize the QP number table for a device
+ * @qpt: the QPN table
+ */
+void qib_init_qpn_table(struct qib_devdata *dd, struct qib_qpn_table *qpt)
+{
+	spin_lock_init(&qpt->lock);
+	qpt->last = 1;          /* start with QPN 2 */
+	qpt->nmaps = 1;
+	qpt->mask = dd->qpn_mask;
+}
+
+/**
+ * qib_free_qpn_table - free the QP number table for a device
+ * @qpt: the QPN table
+ */
+void qib_free_qpn_table(struct qib_qpn_table *qpt)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(qpt->map); i++)
+		if (qpt->map[i].page)
+			free_page((unsigned long) qpt->map[i].page);
+}
+
+/**
+ * qib_get_credit - flush the send work queue of a QP
+ * @qp: the qp who's send work queue to flush
+ * @aeth: the Acknowledge Extended Transport Header
+ *
+ * The QP s_lock should be held.
+ */
+void qib_get_credit(struct qib_qp *qp, u32 aeth)
+{
+	u32 credit = (aeth >> QIB_AETH_CREDIT_SHIFT) & QIB_AETH_CREDIT_MASK;
+
+	/*
+	 * If the credit is invalid, we can send
+	 * as many packets as we like.  Otherwise, we have to
+	 * honor the credit field.
+	 */
+	if (credit == QIB_AETH_CREDIT_INVAL) {
+		if (!(qp->s_flags & QIB_S_UNLIMITED_CREDIT)) {
+			qp->s_flags |= QIB_S_UNLIMITED_CREDIT;
+			if (qp->s_flags & QIB_S_WAIT_SSN_CREDIT) {
+				qp->s_flags &= ~QIB_S_WAIT_SSN_CREDIT;
+				qib_schedule_send(qp);
+			}
+		}
+	} else if (!(qp->s_flags & QIB_S_UNLIMITED_CREDIT)) {
+		/* Compute new LSN (i.e., MSN + credit) */
+		credit = (aeth + credit_table[credit]) & QIB_MSN_MASK;
+		if (qib_cmp24(credit, qp->s_lsn) > 0) {
+			qp->s_lsn = credit;
+			if (qp->s_flags & QIB_S_WAIT_SSN_CREDIT) {
+				qp->s_flags &= ~QIB_S_WAIT_SSN_CREDIT;
+				qib_schedule_send(qp);
+			}
+		}
+	}
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 30/51] IB/qib: Add qib_qsfp.c

[Ralph Campbell]

creates the qib_qsfp.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_qsfp.c |  593 ++++++++++++++++++++++++++++++++++
 1 files changed, 593 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_qsfp.c

diff --git a/drivers/infiniband/hw/qib/qib_qsfp.c b/drivers/infiniband/hw/qib/qib_qsfp.c
new file mode 100644
index 0000000..5ffeddf
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_qsfp.c
@@ -0,0 +1,593 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/delay.h>
+#include <linux/pci.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+#include "qib_qsfp.h"
+
+/*
+ * QSFP support for ib_qib driver, using "Two Wire Serial Interface" driver
+ * in qib_twsi.c
+ */
+#define QSFP_MAX_RETRY 4
+
+static int qsfp_read(struct qib_pportdata *ppd, int addr, void *bp, int len)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u32 out, mask;
+	int ret, cnt, pass = 0;
+	int stuck = 0;
+	u8 *buff = bp;
+
+	qib_cdbg(VERBOSE, "Grabbing Mutex for QSFP in %d:%d\n", dd->unit,
+		 ppd->port);
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (ret)
+		goto no_unlock;
+
+	if (dd->twsi_eeprom_dev == QIB_TWSI_NO_DEV) {
+		qib_dbg("QSFP read on board without QSFP\n");
+		ret = -ENXIO;
+		goto bail;
+	}
+
+	/*
+	 * We presume, if we are called at all, that this board has
+	 * QSFP. This is on the same i2c chain as the legacy parts,
+	 * but only responds if the module is selected via GPIO pins.
+	 * Further, there are very long setup and hold requirements
+	 * on MODSEL.
+	 */
+	mask = QSFP_GPIO_MOD_SEL_N | QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	out = QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	if (ppd->hw_pidx) {
+		mask <<= QSFP_GPIO_PORT2_SHIFT;
+		out <<= QSFP_GPIO_PORT2_SHIFT;
+	}
+
+	dd->f_gpio_mod(dd, out, mask, mask);
+
+	/*
+	 * Module could take up to 2 Msec to respond to MOD_SEL, and there
+	 * is no way to tell if it is ready, so we must wait.
+	 */
+	msleep(2);
+
+	/* Make sure TWSI bus is in sane state. */
+	ret = qib_twsi_reset(dd);
+	if (ret) {
+		qib_dev_porterr(dd, ppd->port, "TWSI Reset failed, QSFP Rd\n");
+		ret = -EIO;
+		stuck = 1;
+		goto deselect;
+	}
+
+	/* All QSFP modules are at A0 */
+
+	cnt = 0;
+	while (cnt < len) {
+		unsigned in_page;
+		int wlen = len - cnt;
+		in_page = addr % QSFP_PAGESIZE;
+		if ((in_page + wlen) > QSFP_PAGESIZE)
+			wlen = QSFP_PAGESIZE - in_page;
+		ret = qib_twsi_blk_rd(dd, QSFP_DEV, addr, buff + cnt, wlen);
+		/* Some QSFP's fail first try. Retry as experiment */
+		if (ret && cnt == 0 && ++pass < QSFP_MAX_RETRY)
+			continue;
+		if (ret) {
+			/* qib_twsi_blk_rd() 1 for error, else 0 */
+			ret = -EIO;
+			goto deselect;
+		}
+		addr += wlen;
+		cnt += wlen;
+	}
+	ret = cnt;
+
+deselect:
+	/*
+	 * Module could take up to 10 uSec after transfer before
+	 * ready to respond to MOD_SEL negation, and there is no way
+	 * to tell if it is ready, so we must wait.
+	 */
+	udelay(10);
+	/* set QSFP MODSEL, RST. LP all high */
+	dd->f_gpio_mod(dd, mask, mask, mask);
+
+	/*
+	 * Module could take up to 2 Msec to respond to MOD_SEL
+	 * going away, and there is no way to tell if it is ready.
+	 * so we must wait.
+	 */
+	if (stuck)
+		qib_dev_err(dd, "TWSI bus stuck non-idle\n");
+
+	if (pass >= QSFP_MAX_RETRY && ret)
+		qib_dev_porterr(dd, ppd->port, "QSFP failed even retrying\n");
+	else if (pass)
+		qib_dev_porterr(dd, ppd->port, "QSFP retries: %d\n", pass);
+
+	msleep(2);
+
+bail:
+	mutex_unlock(&dd->eep_lock);
+	qib_cdbg(VERBOSE, "Released Mutex for QSFP %d:%d, ret %d\n", dd->unit,
+		ppd->port, ret);
+
+no_unlock:
+	return ret;
+}
+
+/*
+ * qsfp_write
+ * We do not ordinarily write the QSFP, but this is needed to select
+ * the page on non-flat QSFPs, and possibly later unusual cases
+ */
+static int qib_qsfp_write(struct qib_pportdata *ppd, int addr, void *bp,
+			  int len)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u32 out, mask;
+	int ret, cnt;
+	u8 *buff = bp;
+
+	qib_cdbg(VERBOSE, "Grabbing Mutex for QSFP %d:%d Wr\n", dd->unit,
+		 ppd->port);
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (ret)
+		goto no_unlock;
+
+	if (dd->twsi_eeprom_dev == QIB_TWSI_NO_DEV) {
+		qib_dbg("QSFP write on board (%d:%d) without QSFP\n", dd->unit,
+			ppd->port);
+		ret = -ENXIO;
+		goto bail;
+	}
+
+	/*
+	 * We presume, if we are called at all, that this board has
+	 * QSFP. This is on the same i2c chain as the legacy parts,
+	 * but only responds if the module is selected via GPIO pins.
+	 * Further, there are very long setup and hold requirements
+	 * on MODSEL.
+	 */
+	mask = QSFP_GPIO_MOD_SEL_N | QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	out = QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	if (ppd->hw_pidx) {
+		mask <<= QSFP_GPIO_PORT2_SHIFT;
+		out <<= QSFP_GPIO_PORT2_SHIFT;
+	}
+	dd->f_gpio_mod(dd, out, mask, mask);
+
+	/*
+	 * Module could take up to 2 Msec to respond to MOD_SEL,
+	 * and there is no way to tell if it is ready, so we must wait.
+	 */
+	msleep(2);
+
+	/* Make sure TWSI bus is in sane state. */
+	ret = qib_twsi_reset(dd);
+	if (ret) {
+		qib_dev_porterr(dd, ppd->port, "TWSI Reset failed, QSFP Wr\n");
+		ret = -EIO;
+		goto deselect;
+	}
+
+	/* All QSFP modules are at A0 */
+
+	cnt = 0;
+	while (cnt < len) {
+		unsigned in_page;
+		int wlen = len - cnt;
+		in_page = addr % QSFP_PAGESIZE;
+		if ((in_page + wlen) > QSFP_PAGESIZE)
+			wlen = QSFP_PAGESIZE - in_page;
+		ret = qib_twsi_blk_wr(dd, QSFP_DEV, addr, buff + cnt, wlen);
+		if (ret) {
+			/* qib_twsi_blk_wr() 1 for error, else 0 */
+			qib_dbg("qib_twsi_blk_wr(%d:%d, %d, , %d) failed\n",
+				dd->unit, ppd->port, addr, wlen);
+			ret = -EIO;
+			goto deselect;
+		}
+		addr += wlen;
+		cnt += wlen;
+	}
+	ret = cnt;
+
+deselect:
+	/*
+	 * Module could take up to 10 uSec after transfer before
+	 * ready to respond to MOD_SEL negation, and there is no way
+	 * to tell if it is ready, so we must wait.
+	 */
+	udelay(10);
+	/* set QSFP MODSEL, RST, LP high */
+	dd->f_gpio_mod(dd, mask, mask, mask);
+	/*
+	 * Module could take up to 2 Msec to respond to MOD_SEL
+	 * going away, and there is no way to tell if it is ready.
+	 * so we must wait.
+	 */
+	msleep(2);
+
+bail:
+	mutex_unlock(&dd->eep_lock);
+	qib_cdbg(VERBOSE, "Released Mutex for QSFP %d:%d, ret %d\n", dd->unit,
+		 ppd->port, ret);
+
+no_unlock:
+	return ret;
+}
+
+/*
+ * For validation, we want to check the checksums, even of the
+ * fields we do not otherwise use. This function reads the bytes from
+ * <first> to <next-1> and returns the 8lsbs of the sum, or <0 for errors
+ */
+static int qsfp_cks(struct qib_pportdata *ppd, int first, int next)
+{
+	int ret;
+	u16 cks;
+	u8 bval;
+
+	cks = 0;
+	while (first < next) {
+		ret = qsfp_read(ppd, first, &bval, 1);
+		if (ret < 0)
+			goto bail;
+		cks += bval;
+		++first;
+	}
+	ret = cks & 0xFF;
+bail:
+	return ret;
+
+}
+
+int qib_refresh_qsfp_cache(struct qib_pportdata *ppd, struct qib_qsfp_cache *cp)
+{
+	int ret;
+	int idx;
+	u16 cks;
+	u32 mask;
+	u8 peek[4];
+
+	mask = QSFP_GPIO_MOD_PRS_N;
+	if (ppd->hw_pidx)
+		mask <<= QSFP_GPIO_PORT2_SHIFT;
+
+	ret = ppd->dd->f_gpio_mod(ppd->dd, 0, 0, 0);
+	if (ret & mask) {
+		qib_dbg("No QSFP module in %d:%d\n", ppd->dd->unit, ppd->port);
+		ret = -ENODEV;
+		goto bail;
+	}
+
+	ret = qsfp_read(ppd, 0, peek, 3);
+	if (ret < 0)
+		goto bail;
+	if ((peek[0] & 0xFE) != 0x0C)
+		qib_dev_porterr(ppd->dd, ppd->port,
+				"QSFP byte0 is 0x%02X, S/B 0x0C/D\n", peek[0]);
+
+	if ((peek[2] & 2) == 0) {
+		/*
+		 * If cable is paged, rather than "flat memory", we need to
+		 * set the page to zero, Even if it already appears to be zero.
+		 */
+		u8 poke = 0;
+		ret = qib_qsfp_write(ppd, 127, &poke, 1);
+		udelay(50);
+		if (ret != 1) {
+			qib_dev_porterr(ppd->dd, ppd->port,
+					"Failed QSFP Page set\n");
+			goto bail;
+		}
+	}
+
+	ret = qsfp_read(ppd, QSFP_MOD_ID_OFFS, &cp->id, 1);
+	if (ret < 0)
+		goto bail;
+	if ((cp->id & 0xFE) != 0x0C)
+		qib_dev_porterr(ppd->dd, ppd->port,
+				"QSFP ID byte is 0x%02X, S/B 0x0C/D\n", cp->id);
+	cks = cp->id;
+
+	ret = qsfp_read(ppd, QSFP_MOD_PWR_OFFS, &cp->pwr, 1);
+	if (ret < 0)
+		goto bail;
+	cks += cp->pwr;
+
+	ret = qsfp_cks(ppd, QSFP_MOD_PWR_OFFS + 1, QSFP_MOD_LEN_OFFS);
+	if (ret < 0)
+		goto bail;
+	cks += ret;
+
+	ret = qsfp_read(ppd, QSFP_MOD_LEN_OFFS, &cp->len, 1);
+	if (ret < 0)
+		goto bail;
+	cks += cp->len;
+
+	ret = qsfp_read(ppd, QSFP_MOD_TECH_OFFS, &cp->tech, 1);
+	if (ret < 0)
+		goto bail;
+	cks += cp->tech;
+
+	ret = qsfp_read(ppd, QSFP_VEND_OFFS, &cp->vendor, QSFP_VEND_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_VEND_LEN; ++idx)
+		cks += cp->vendor[idx];
+
+	ret = qsfp_read(ppd, QSFP_IBXCV_OFFS, &cp->xt_xcv, 1);
+	if (ret < 0)
+		goto bail;
+	cks += cp->xt_xcv;
+
+	ret = qsfp_read(ppd, QSFP_VOUI_OFFS, &cp->oui, QSFP_VOUI_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_VOUI_LEN; ++idx)
+		cks += cp->oui[idx];
+
+	ret = qsfp_read(ppd, QSFP_PN_OFFS, &cp->partnum, QSFP_PN_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_PN_LEN; ++idx)
+		cks += cp->partnum[idx];
+
+	ret = qsfp_read(ppd, QSFP_REV_OFFS, &cp->rev, QSFP_REV_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_REV_LEN; ++idx)
+		cks += cp->rev[idx];
+
+	ret = qsfp_read(ppd, QSFP_ATTEN_OFFS, &cp->atten, QSFP_ATTEN_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_ATTEN_LEN; ++idx)
+		cks += cp->atten[idx];
+
+	ret = qsfp_cks(ppd, QSFP_ATTEN_OFFS + QSFP_ATTEN_LEN, QSFP_CC_OFFS);
+	if (ret < 0)
+		goto bail;
+	cks += ret;
+
+	cks &= 0xFF;
+	ret = qsfp_read(ppd, QSFP_CC_OFFS, &cp->cks1, 1);
+	if (ret < 0)
+		goto bail;
+	if (cks != cp->cks1)
+		qib_dev_porterr(ppd->dd, ppd->port,
+				"QSFP cks1 is %02X, computed %02X\n", cp->cks1,
+				cks);
+
+	/* Second checksum covers 192 to (serial, date, lot) */
+	ret = qsfp_cks(ppd, QSFP_CC_OFFS + 1, QSFP_SN_OFFS);
+	if (ret < 0)
+		goto bail;
+	cks = ret;
+
+	ret = qsfp_read(ppd, QSFP_SN_OFFS, &cp->serial, QSFP_SN_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_SN_LEN; ++idx)
+		cks += cp->serial[idx];
+
+	ret = qsfp_read(ppd, QSFP_DATE_OFFS, &cp->date, QSFP_DATE_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_DATE_LEN; ++idx)
+		cks += cp->date[idx];
+
+	ret = qsfp_read(ppd, QSFP_LOT_OFFS, &cp->lot, QSFP_LOT_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_LOT_LEN; ++idx)
+		cks += cp->lot[idx];
+
+	ret = qsfp_cks(ppd, QSFP_LOT_OFFS + QSFP_LOT_LEN, QSFP_CC_EXT_OFFS);
+	if (ret < 0)
+		goto bail;
+	cks += ret;
+
+	ret = qsfp_read(ppd, QSFP_CC_EXT_OFFS, &cp->cks2, 1);
+	if (ret < 0)
+		goto bail;
+	cks &= 0xFF;
+	if (cks != cp->cks2)
+		qib_dev_porterr(ppd->dd, ppd->port,
+				"QSFP cks2 is %02X, computed %02X\n", cp->cks2,
+				cks);
+	ret = 0;
+
+bail:
+	return ret;
+
+}
+
+const char * const qib_qsfp_devtech[16] = {
+	"850nm VCSEL", "1310nm VCSEL", "1550nm VCSEL", "1310nm FP",
+	"1310nm DFB", "1550nm DFB", "1310nm EML", "1550nm EML",
+	"Cu Misc", "1490nm DFB", "Cu NoEq", "Cu Eq",
+	"Undef", "Cu Active BothEq", "Cu FarEq", "Cu NearEq"
+};
+
+#define QSFP_DUMP_CHUNK 16 /* Holds longest string */
+#define QSFP_DEFAULT_HDR_CNT 224
+
+static const char *pwr_codes = "1.5W2.0W2.5W3.5W";
+
+void qib_qsfp_short_msg(struct qib_qsfp_data *qd)
+{
+	char lenstr[6];
+	lenstr[0] = ' ';
+	lenstr[1] = '\0';
+	if (QSFP_IS_CU(qd->cache.tech))
+		sprintf(lenstr, "%dM ", qd->cache.len);
+	qib_cdbg(INIT, "IB%u:%u QSFP %.*s %.*s %.3s%s SN: %.*s\n",
+			qd->ppd->dd->unit, qd->ppd->port, QSFP_VEND_LEN,
+			qd->cache.vendor, QSFP_PN_LEN, qd->cache.partnum,
+			lenstr, qib_qsfp_devtech[qd->cache.tech >> 4],
+			QSFP_SN_LEN, qd->cache.serial);
+}
+
+/*
+ * Initialize structures that control access to QSFP. Called once per port
+ * on cards that support QSFP.
+ */
+void qib_qsfp_init(struct qib_qsfp_data *qd,
+		   void (*fevent)(struct work_struct *))
+{
+	u32 mask, highs;
+	int pins;
+
+	struct qib_devdata *dd = qd->ppd->dd;
+
+	/* Initialize work struct for later QSFP events */
+	INIT_WORK(&qd->work, fevent);
+
+	/*
+	 * Later, we may want more validation. For now, just set up pins and
+	 * blip reset. If module is present, call qib_refresh_qsfp_cache(),
+	 * to do further init.
+	 */
+	mask = QSFP_GPIO_MOD_SEL_N | QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	highs = mask - QSFP_GPIO_MOD_RST_N;
+	if (qd->ppd->hw_pidx) {
+		mask <<= QSFP_GPIO_PORT2_SHIFT;
+		highs <<= QSFP_GPIO_PORT2_SHIFT;
+	}
+	dd->f_gpio_mod(dd, highs, mask, mask);
+	udelay(20); /* Generous RST dwell */
+
+	dd->f_gpio_mod(dd, mask, mask, mask);
+	/* Spec says module can take up to two seconds! */
+	mask = QSFP_GPIO_MOD_PRS_N;
+	if (qd->ppd->hw_pidx)
+		mask <<= QSFP_GPIO_PORT2_SHIFT;
+
+	/* Do not try to wait here. Better to let event handle it */
+	pins = dd->f_gpio_mod(dd, 0, 0, 0);
+	if (pins & mask) {
+		qib_dbg("IB%u:%u has no QSFP module present\n",
+			qd->ppd->dd->unit, qd->ppd->port);
+		goto bail;
+	}
+	/* We see a Mdoule, but it may be unwise to look yet. Just schedule */
+	qd->t_insert = get_jiffies_64();
+	schedule_work(&qd->work);
+bail:
+	return;
+}
+
+void qib_qsfp_deinit(struct qib_qsfp_data *qd)
+{
+	/*
+	 * There is nothing to do here for now.  our
+	 * work is scheduled with schedule_work(), and
+	 * flush_scheduled_work() from remove_one will
+	 * block until all work ssetup with schedule_work()
+	 * completes.
+	 */
+}
+
+int qib_qsfp_dump(struct qib_pportdata *ppd, char *buf, int len)
+{
+	struct qib_qsfp_cache cd;
+	u8 bin_buff[QSFP_DUMP_CHUNK];
+	char lenstr[6];
+	int sofar, ret;
+	int bidx = 0;
+
+	sofar = 0;
+	ret = qib_refresh_qsfp_cache(ppd, &cd);
+	if (ret < 0) {
+		qib_dbg("IB%u:%u QSFP cache-refresh failed\n", ppd->dd->unit,
+			ppd->port);
+		goto bail;
+	}
+
+	lenstr[0] = ' ';
+	lenstr[1] = '\0';
+	if (QSFP_IS_CU(cd.tech))
+		sprintf(lenstr, "%dM ", cd.len);
+
+	sofar += scnprintf(buf + sofar, len - sofar, "PWR:%.3sW\n", pwr_codes +
+			   (QSFP_PWR(cd.pwr) * 4));
+
+	sofar += scnprintf(buf + sofar, len - sofar, "TECH:%s%s\n", lenstr,
+			   qib_qsfp_devtech[cd.tech >> 4]);
+
+	sofar += scnprintf(buf + sofar, len - sofar, "Vendor:%.*s\n",
+			   QSFP_VEND_LEN, cd.vendor);
+
+	sofar += scnprintf(buf + sofar, len - sofar, "OUI:%06X\n",
+			   QSFP_OUI(cd.oui));
+
+	sofar += scnprintf(buf + sofar, len - sofar, "Part#:%.*s\n",
+			   QSFP_PN_LEN, cd.partnum);
+	sofar += scnprintf(buf + sofar, len - sofar, "Rev:%.*s\n",
+			   QSFP_REV_LEN, cd.rev);
+	if (QSFP_IS_CU(cd.tech))
+		sofar += scnprintf(buf + sofar, len - sofar, "Atten:%d, %d\n",
+				   QSFP_ATTEN_SDR(cd.atten),
+				   QSFP_ATTEN_DDR(cd.atten));
+	sofar += scnprintf(buf + sofar, len - sofar, "Serial:%.*s\n",
+			   QSFP_SN_LEN, cd.serial);
+	sofar += scnprintf(buf + sofar, len - sofar, "Date:%.*s\n",
+			   QSFP_DATE_LEN, cd.date);
+	sofar += scnprintf(buf + sofar, len - sofar, "Lot:%.*s\n",
+			   QSFP_LOT_LEN, cd.date);
+
+	while (bidx < QSFP_DEFAULT_HDR_CNT) {
+		int iidx;
+		ret = qsfp_read(ppd, bidx, bin_buff, QSFP_DUMP_CHUNK);
+		if (ret < 0)
+			goto bail;
+		for (iidx = 0; iidx < ret; ++iidx) {
+			sofar += scnprintf(buf + sofar, len-sofar, " %02X",
+				bin_buff[iidx]);
+		}
+		sofar += scnprintf(buf + sofar, len - sofar, "\n");
+		bidx += QSFP_DUMP_CHUNK;
+	}
+	ret = sofar;
+bail:
+	return ret;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 31/51] IB/qib: Add qib_qsfp.h

[Ralph Campbell]

creates the qib_qsfp.h file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_qsfp.h |  183 ++++++++++++++++++++++++++++++++++
 1 files changed, 183 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_qsfp.h

diff --git a/drivers/infiniband/hw/qib/qib_qsfp.h b/drivers/infiniband/hw/qib/qib_qsfp.h
new file mode 100644
index 0000000..01ea507
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_qsfp.h
@@ -0,0 +1,183 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+/* QSFP support common definitions, for ib_qib driver */
+
+#define QSFP_DEV 0xA0
+#define QSFP_PWR_LAG_MSEC 2000
+
+/*
+ * Below are masks for various QSFP signals, for Port 1.
+ * Port2 equivalents are shifted by QSFP_GPIO_PORT2_SHIFT.
+ * _N means asserted low
+ */
+#define QSFP_GPIO_MOD_SEL_N (4)
+#define QSFP_GPIO_MOD_PRS_N (8)
+#define QSFP_GPIO_INT_N (0x10)
+#define QSFP_GPIO_MOD_RST_N (0x20)
+#define QSFP_GPIO_LP_MODE (0x40)
+#define QSFP_GPIO_PORT2_SHIFT 5
+
+#define QSFP_PAGESIZE 128
+/* Defined fields that QLogic requires of qualified cables */
+/* Byte 0 is Identifier, not checked */
+/* Byte 1 is reserved "status MSB" */
+/* Byte 2 is "status LSB" We only care that D2 "Flat Mem" is set. */
+/*
+ * Rest of first 128 not used, although 127 is reserved for page select
+ * if module is not "Flat memory".
+ */
+/* Byte 128 is Identifier: must be 0x0c for QSFP, or 0x0d for QSFP+ */
+#define QSFP_MOD_ID_OFFS 128
+/*
+ * Byte 129 is "Extended Identifier". We only care about D7,D6: Power class
+ *  0:1.5W, 1:2.0W, 2:2.5W, 3:3.5W
+ */
+#define QSFP_MOD_PWR_OFFS 129
+/* Byte 130 is Connector type. Not QLogic req'd */
+/* Bytes 131..138 are Transceiver types, bit maps for various tech, none IB */
+/* Byte 139 is encoding. code 0x01 is 8b10b. Not QLogic req'd */
+/* byte 140 is nominal bit-rate, in units of 100Mbits/sec Not QLogic req'd */
+/* Byte 141 is Extended Rate Select. Not QLogic req'd */
+/* Bytes 142..145 are lengths for various fiber types. Not QLogic req'd */
+/* Byte 146 is length for Copper. Units of 1 meter */
+#define QSFP_MOD_LEN_OFFS 146
+/*
+ * Byte 147 is Device technology. D0..3 not Qlogc req'd
+ * D4..7 select from 15 choices, translated by table:
+ */
+#define QSFP_MOD_TECH_OFFS 147
+extern const char *const qib_qsfp_devtech[16];
+/* Length is only valid if technology is "copper" */
+#define QSFP_IS_CU(tech) ((0xED00 >> ((tech) >> 4)) & 1)
+#define QSFP_TECH_1490 9
+
+#define QSFP_OUI(oui) (((unsigned)oui[0] << 16) | ((unsigned)oui[1] << 8) | \
+			oui[2])
+#define QSFP_OUI_AMPHENOL 0x415048
+#define QSFP_OUI_FINISAR  0x009065
+#define QSFP_OUI_GORE     0x002177
+
+/* Bytes 148..163 are Vendor Name, Left-justified Blank-filled */
+#define QSFP_VEND_OFFS 148
+#define QSFP_VEND_LEN 16
+/* Byte 164 is IB Extended tranceiver codes Bits D0..3 are SDR,DDR,QDR,EDR */
+#define QSFP_IBXCV_OFFS 164
+/* Bytes 165..167 are Vendor OUI number */
+#define QSFP_VOUI_OFFS 165
+#define QSFP_VOUI_LEN 3
+/* Bytes 168..183 are Vendor Part Number, string */
+#define QSFP_PN_OFFS 168
+#define QSFP_PN_LEN 16
+/* Bytes 184,185 are Vendor Rev. Left Justified, Blank-filled */
+#define QSFP_REV_OFFS 184
+#define QSFP_REV_LEN 2
+/*
+ * Bytes 186,187 are Wavelength, if Optical. Not Qlogic req'd
+ *  If copper, they are attenuation in dB:
+ * Byte 186 is at 2.5Gb/sec (SDR), Byte 187 at 5.0Gb/sec (DDR)
+ */
+#define QSFP_ATTEN_OFFS 186
+#define QSFP_ATTEN_LEN 2
+/* Bytes 188,189 are Wavelength tolerance, not QLogic req'd */
+/* Byte 190 is Max Case Temp. Not QLogic req'd */
+/* Byte 191 is LSB of sum of bytes 128..190. Not QLogic req'd */
+#define QSFP_CC_OFFS 191
+/* Bytes 192..195 are Options implemented in qsfp. Not Qlogic req'd */
+/* Bytes 196..211 are Serial Number, String */
+#define QSFP_SN_OFFS 196
+#define QSFP_SN_LEN 16
+/* Bytes 212..219 are date-code YYMMDD (MM==1 for Jan) */
+#define QSFP_DATE_OFFS 212
+#define QSFP_DATE_LEN 6
+/* Bytes 218,219 are optional lot-code, string */
+#define QSFP_LOT_OFFS 218
+#define QSFP_LOT_LEN 2
+/* Bytes 220, 221 indicate monitoring options, Not QLogic req'd */
+/* Byte 223 is LSB of sum of bytes 192..222 */
+#define QSFP_CC_EXT_OFFS 223
+
+/*
+ * struct qib_qsfp_data encapsulates state of QSFP device for one port.
+ * it will be part of port-chip-specific data if a board supports QSFP.
+ *
+ * Since multiple board-types use QSFP, and their pport_data structs
+ * differ (in the chip-specific section), we need a pointer to its head.
+ *
+ * Avoiding premature optimization, we will have one work_struct per port,
+ * and let the (increasingly inaccurately named) eep_lock arbitrate
+ * access to common resources.
+ *
+ */
+
+/*
+ * Hold the parts of the onboard EEPROM that we care about, so we aren't
+ * coonstantly bit-boffing
+ */
+struct qib_qsfp_cache {
+	u8 id;	/* must be 0x0C or 0x0D */
+	u8 pwr; /* in D6,7 */
+	u8 len;	/* in meters, Cu only */
+	u8 tech;
+	char vendor[QSFP_VEND_LEN];
+	u8 xt_xcv; /* Ext. tranceiver codes, 4 lsbs are IB speed supported */
+	u8 oui[QSFP_VOUI_LEN];
+	u8 partnum[QSFP_PN_LEN];
+	u8 rev[QSFP_REV_LEN];
+	u8 atten[QSFP_ATTEN_LEN];
+	u8 cks1;	/* Checksum of bytes 128..190 */
+	u8 serial[QSFP_SN_LEN];
+	u8 date[QSFP_DATE_LEN];
+	u8 lot[QSFP_LOT_LEN];
+	u8 cks2;	/* Checsum of bytes 192..222 */
+};
+
+#define QSFP_PWR(pbyte) (((pbyte) >> 6) & 3)
+#define QSFP_ATTEN_SDR(attenarray) (attenarray[0])
+#define QSFP_ATTEN_DDR(attenarray) (attenarray[1])
+
+struct qib_qsfp_data {
+	/* Helps to find our way */
+	struct qib_pportdata *ppd;
+	struct work_struct work;
+	struct qib_qsfp_cache cache;
+	u64 t_insert;
+};
+
+extern int qib_refresh_qsfp_cache(struct qib_pportdata *ppd,
+				  struct qib_qsfp_cache *cp);
+extern void qib_qsfp_short_msg(struct qib_qsfp_data *qd);
+
+extern void qib_qsfp_init(struct qib_qsfp_data *qd,
+			  void (*fevent)(struct work_struct *));
+extern void qib_qsfp_deinit(struct qib_qsfp_data *qd);
+

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 32/51] IB/qib: Add qib_rc.c

[Ralph Campbell]

creates the qib_rc.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_rc.c | 2293 ++++++++++++++++++++++++++++++++++++
 1 files changed, 2293 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_rc.c

diff --git a/drivers/infiniband/hw/qib/qib_rc.c b/drivers/infiniband/hw/qib/qib_rc.c
new file mode 100644
index 0000000..8d31f1c
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_rc.c
@@ -0,0 +1,2293 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/io.h>
+
+#include "qib.h"
+
+/* cut down ridiculously long IB macro names */
+#define OP(x) IB_OPCODE_RC_##x
+
+static void rc_timeout(unsigned long arg);
+
+static u32 restart_sge(struct qib_sge_state *ss, struct qib_swqe *wqe,
+		       u32 psn, u32 pmtu)
+{
+	u32 len;
+
+	len = ((psn - wqe->psn) & QIB_PSN_MASK) * pmtu;
+	ss->sge = wqe->sg_list[0];
+	ss->sg_list = wqe->sg_list + 1;
+	ss->num_sge = wqe->wr.num_sge;
+	ss->total_len = wqe->length;
+	qib_skip_sge(ss, len, 0);
+	return wqe->length - len;
+}
+
+static void start_timer(struct qib_qp *qp)
+{
+	qp->s_flags |= QIB_S_TIMER;
+	qp->s_timer.function = rc_timeout;
+	/* 4.096 usec. * (1 << qp->timeout) */
+	qp->s_timer.expires = jiffies +
+		usecs_to_jiffies((4096UL * (1UL << qp->timeout)) / 1000UL);
+	add_timer(&qp->s_timer);
+}
+
+/**
+ * qib_make_rc_ack - construct a response packet (ACK, NAK, or RDMA read)
+ * @dev: the device for this QP
+ * @qp: a pointer to the QP
+ * @ohdr: a pointer to the IB header being constructed
+ * @pmtu: the path MTU
+ *
+ * Return 1 if constructed; otherwise, return 0.
+ * Note that we are in the responder's side of the QP context.
+ * Note the QP s_lock must be held.
+ */
+static int qib_make_rc_ack(struct qib_ibdev *dev, struct qib_qp *qp,
+			   struct qib_other_headers *ohdr, u32 pmtu)
+{
+	struct qib_ack_entry *e;
+	u32 hwords;
+	u32 len;
+	u32 bth0;
+	u32 bth2;
+
+	/* Don't send an ACK if we aren't supposed to. */
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK))
+		goto bail;
+
+	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
+	hwords = 5;
+
+	switch (qp->s_ack_state) {
+	case OP(RDMA_READ_RESPONSE_LAST):
+	case OP(RDMA_READ_RESPONSE_ONLY):
+		e = &qp->s_ack_queue[qp->s_tail_ack_queue];
+		if (e->rdma_sge.mr) {
+			atomic_dec(&e->rdma_sge.mr->refcount);
+			e->rdma_sge.mr = NULL;
+		}
+		/* FALLTHROUGH */
+	case OP(ATOMIC_ACKNOWLEDGE):
+		/*
+		 * We can increment the tail pointer now that the last
+		 * response has been sent instead of only being
+		 * constructed.
+		 */
+		if (++qp->s_tail_ack_queue > QIB_MAX_RDMA_ATOMIC)
+			qp->s_tail_ack_queue = 0;
+		/* FALLTHROUGH */
+	case OP(SEND_ONLY):
+	case OP(ACKNOWLEDGE):
+		/* Check for no next entry in the queue. */
+		if (qp->r_head_ack_queue == qp->s_tail_ack_queue) {
+			if (qp->s_flags & QIB_S_ACK_PENDING)
+				goto normal;
+			goto bail;
+		}
+
+		e = &qp->s_ack_queue[qp->s_tail_ack_queue];
+		if (e->opcode == OP(RDMA_READ_REQUEST)) {
+			/*
+			 * If a RDMA read response is being resent and
+			 * we haven't seen the duplicate request yet,
+			 * then stop sending the remaining responses the
+			 * responder has seen until the requester resends it.
+			 */
+			len = e->rdma_sge.sge_length;
+			if (len && !e->rdma_sge.mr) {
+				qp->s_tail_ack_queue = qp->r_head_ack_queue;
+				goto bail;
+			}
+			/* Copy SGE state in case we need to resend */
+			qp->s_rdma_mr = e->rdma_sge.mr;
+			if (qp->s_rdma_mr)
+				atomic_inc(&qp->s_rdma_mr->refcount);
+			qp->s_ack_rdma_sge.sge = e->rdma_sge;
+			qp->s_ack_rdma_sge.num_sge = 1;
+			qp->s_cur_sge = &qp->s_ack_rdma_sge;
+			if (len > pmtu) {
+				len = pmtu;
+				qp->s_ack_state = OP(RDMA_READ_RESPONSE_FIRST);
+			} else {
+				qp->s_ack_state = OP(RDMA_READ_RESPONSE_ONLY);
+				e->sent = 1;
+			}
+			ohdr->u.aeth = qib_compute_aeth(qp);
+			hwords++;
+			qp->s_ack_rdma_psn = e->psn;
+			bth2 = qp->s_ack_rdma_psn++ & QIB_PSN_MASK;
+		} else {
+			/* COMPARE_SWAP or FETCH_ADD */
+			qp->s_cur_sge = NULL;
+			len = 0;
+			qp->s_ack_state = OP(ATOMIC_ACKNOWLEDGE);
+			ohdr->u.at.aeth = qib_compute_aeth(qp);
+			ohdr->u.at.atomic_ack_eth[0] =
+				cpu_to_be32(e->atomic_data >> 32);
+			ohdr->u.at.atomic_ack_eth[1] =
+				cpu_to_be32(e->atomic_data);
+			hwords += sizeof(ohdr->u.at) / sizeof(u32);
+			bth2 = e->psn & QIB_PSN_MASK;
+			e->sent = 1;
+		}
+		bth0 = qp->s_ack_state << 24;
+		break;
+
+	case OP(RDMA_READ_RESPONSE_FIRST):
+		qp->s_ack_state = OP(RDMA_READ_RESPONSE_MIDDLE);
+		/* FALLTHROUGH */
+	case OP(RDMA_READ_RESPONSE_MIDDLE):
+		qp->s_cur_sge = &qp->s_ack_rdma_sge;
+		qp->s_rdma_mr = qp->s_ack_rdma_sge.sge.mr;
+		if (qp->s_rdma_mr)
+			atomic_inc(&qp->s_rdma_mr->refcount);
+		len = qp->s_ack_rdma_sge.sge.sge_length;
+		if (len > pmtu)
+			len = pmtu;
+		else {
+			ohdr->u.aeth = qib_compute_aeth(qp);
+			hwords++;
+			qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST);
+			e = &qp->s_ack_queue[qp->s_tail_ack_queue];
+			e->sent = 1;
+		}
+		bth0 = qp->s_ack_state << 24;
+		bth2 = qp->s_ack_rdma_psn++ & QIB_PSN_MASK;
+		break;
+
+	default:
+normal:
+		/*
+		 * Send a regular ACK.
+		 * Set the s_ack_state so we wait until after sending
+		 * the ACK before setting s_ack_state to ACKNOWLEDGE
+		 * (see above).
+		 */
+		qp->s_ack_state = OP(SEND_ONLY);
+		qp->s_flags &= ~QIB_S_ACK_PENDING;
+		qp->s_cur_sge = NULL;
+		if (qp->s_nak_state)
+			ohdr->u.aeth =
+				cpu_to_be32((qp->r_msn & QIB_MSN_MASK) |
+					    (qp->s_nak_state <<
+					     QIB_AETH_CREDIT_SHIFT));
+		else
+			ohdr->u.aeth = qib_compute_aeth(qp);
+		hwords++;
+		len = 0;
+		bth0 = OP(ACKNOWLEDGE) << 24;
+		bth2 = qp->s_ack_psn & QIB_PSN_MASK;
+	}
+	qp->s_rdma_ack_cnt++;
+	qp->s_hdrwords = hwords;
+	qp->s_cur_size = len;
+	qib_make_ruc_header(qp, ohdr, bth0, bth2);
+	return 1;
+
+bail:
+	qp->s_ack_state = OP(ACKNOWLEDGE);
+	qp->s_flags &= ~(QIB_S_RESP_PENDING | QIB_S_ACK_PENDING);
+	return 0;
+}
+
+/**
+ * qib_make_rc_req - construct a request packet (SEND, RDMA r/w, ATOMIC)
+ * @qp: a pointer to the QP
+ *
+ * Return 1 if constructed; otherwise, return 0.
+ */
+int qib_make_rc_req(struct qib_qp *qp)
+{
+	struct qib_ibdev *dev = to_idev(qp->ibqp.device);
+	struct qib_other_headers *ohdr;
+	struct qib_sge_state *ss;
+	struct qib_swqe *wqe;
+	u32 hwords;
+	u32 len;
+	u32 bth0;
+	u32 bth2;
+	u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
+	char newreq;
+	unsigned long flags;
+	int ret = 0;
+	int delta;
+
+	ohdr = &qp->s_hdr.u.oth;
+	if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
+		ohdr = &qp->s_hdr.u.l.oth;
+
+	/*
+	 * The lock is needed to synchronize between the sending tasklet,
+	 * the receive interrupt handler, and timeout resends.
+	 */
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	/* Sending responses has higher priority over sending requests. */
+	if ((qp->s_flags & QIB_S_RESP_PENDING) &&
+	    qib_make_rc_ack(dev, qp, ohdr, pmtu))
+		goto done;
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_SEND_OK)) {
+		if (!(ib_qib_state_ops[qp->state] & QIB_FLUSH_SEND))
+			goto bail;
+		/* We are in the error state, flush the work request. */
+		if (qp->s_last == qp->s_head)
+			goto bail;
+		/* If DMAs are in progress, we can't flush immediately. */
+		if (atomic_read(&qp->s_dma_busy)) {
+			qp->s_flags |= QIB_S_WAIT_DMA;
+			goto bail;
+		}
+		wqe = get_swqe_ptr(qp, qp->s_last);
+		while (qp->s_last != qp->s_acked) {
+			qib_send_complete(qp, wqe, IB_WC_SUCCESS);
+			if (++qp->s_last >= qp->s_size)
+				qp->s_last = 0;
+			wqe = get_swqe_ptr(qp, qp->s_last);
+		}
+		qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
+		goto done;
+	}
+
+	if (qp->s_flags & (QIB_S_WAIT_RNR | QIB_S_WAIT_ACK))
+		goto bail;
+
+	if (qib_cmp24(qp->s_psn, qp->s_sending_hpsn) <= 0) {
+		if (qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) <= 0) {
+			qp->s_flags |= QIB_S_WAIT_PSN;
+			goto bail;
+		}
+		qp->s_sending_psn = qp->s_psn;
+		qp->s_sending_hpsn = qp->s_psn - 1;
+	}
+
+	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
+	hwords = 5;
+	bth0 = 0;
+
+	/* Send a request. */
+	wqe = get_swqe_ptr(qp, qp->s_cur);
+	switch (qp->s_state) {
+	default:
+		if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_NEXT_SEND_OK))
+			goto bail;
+		/*
+		 * Resend an old request or start a new one.
+		 *
+		 * We keep track of the current SWQE so that
+		 * we don't reset the "furthest progress" state
+		 * if we need to back up.
+		 */
+		newreq = 0;
+		if (qp->s_cur == qp->s_tail) {
+			/* Check if send work queue is empty. */
+			if (qp->s_tail == qp->s_head)
+				goto bail;
+			/*
+			 * If a fence is requested, wait for previous
+			 * RDMA read and atomic operations to finish.
+			 */
+			if ((wqe->wr.send_flags & IB_SEND_FENCE) &&
+			    qp->s_num_rd_atomic) {
+				qp->s_flags |= QIB_S_WAIT_FENCE;
+				goto bail;
+			}
+			wqe->psn = qp->s_next_psn;
+			newreq = 1;
+		}
+		/*
+		 * Note that we have to be careful not to modify the
+		 * original work request since we may need to resend
+		 * it.
+		 */
+		len = wqe->length;
+		ss = &qp->s_sge;
+		bth2 = qp->s_psn & QIB_PSN_MASK;
+		switch (wqe->wr.opcode) {
+		case IB_WR_SEND:
+		case IB_WR_SEND_WITH_IMM:
+			/* If no credit, return. */
+			if (!(qp->s_flags & QIB_S_UNLIMITED_CREDIT) &&
+			    qib_cmp24(wqe->ssn, qp->s_lsn + 1) > 0) {
+				qp->s_flags |= QIB_S_WAIT_SSN_CREDIT;
+				goto bail;
+			}
+			wqe->lpsn = wqe->psn;
+			if (len > pmtu) {
+				wqe->lpsn += (len - 1) / pmtu;
+				qp->s_state = OP(SEND_FIRST);
+				len = pmtu;
+				break;
+			}
+			if (wqe->wr.opcode == IB_WR_SEND)
+				qp->s_state = OP(SEND_ONLY);
+			else {
+				qp->s_state = OP(SEND_ONLY_WITH_IMMEDIATE);
+				/* Immediate data comes after the BTH */
+				ohdr->u.imm_data = wqe->wr.ex.imm_data;
+				hwords += 1;
+			}
+			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+				bth0 |= IB_BTH_SOLICITED;
+			bth2 |= IB_BTH_REQ_ACK;
+			if (++qp->s_cur == qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		case IB_WR_RDMA_WRITE:
+			if (newreq && !(qp->s_flags & QIB_S_UNLIMITED_CREDIT))
+				qp->s_lsn++;
+			/* FALLTHROUGH */
+		case IB_WR_RDMA_WRITE_WITH_IMM:
+			/* If no credit, return. */
+			if (!(qp->s_flags & QIB_S_UNLIMITED_CREDIT) &&
+			    qib_cmp24(wqe->ssn, qp->s_lsn + 1) > 0) {
+				qp->s_flags |= QIB_S_WAIT_SSN_CREDIT;
+				goto bail;
+			}
+			ohdr->u.rc.reth.vaddr =
+				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+			ohdr->u.rc.reth.rkey =
+				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+			ohdr->u.rc.reth.length = cpu_to_be32(len);
+			hwords += sizeof(struct ib_reth) / sizeof(u32);
+			wqe->lpsn = wqe->psn;
+			if (len > pmtu) {
+				wqe->lpsn += (len - 1) / pmtu;
+				qp->s_state = OP(RDMA_WRITE_FIRST);
+				len = pmtu;
+				break;
+			}
+			if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
+				qp->s_state = OP(RDMA_WRITE_ONLY);
+			else {
+				qp->s_state =
+					OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE);
+				/* Immediate data comes after RETH */
+				ohdr->u.rc.imm_data = wqe->wr.ex.imm_data;
+				hwords += 1;
+				if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+					bth0 |= IB_BTH_SOLICITED;
+			}
+			bth2 |= IB_BTH_REQ_ACK;
+			if (++qp->s_cur == qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		case IB_WR_RDMA_READ:
+			/*
+			 * Don't allow more operations to be started
+			 * than the QP limits allow.
+			 */
+			if (newreq) {
+				if (qp->s_num_rd_atomic >=
+				    qp->s_max_rd_atomic) {
+					qp->s_flags |= QIB_S_WAIT_RDMAR;
+					goto bail;
+				}
+				qp->s_num_rd_atomic++;
+				if (!(qp->s_flags & QIB_S_UNLIMITED_CREDIT))
+					qp->s_lsn++;
+				/*
+				 * Adjust s_next_psn to count the
+				 * expected number of responses.
+				 */
+				if (len > pmtu)
+					qp->s_next_psn += (len - 1) / pmtu;
+				wqe->lpsn = qp->s_next_psn++;
+			}
+			ohdr->u.rc.reth.vaddr =
+				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+			ohdr->u.rc.reth.rkey =
+				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+			ohdr->u.rc.reth.length = cpu_to_be32(len);
+			qp->s_state = OP(RDMA_READ_REQUEST);
+			hwords += sizeof(ohdr->u.rc.reth) / sizeof(u32);
+			ss = NULL;
+			len = 0;
+			bth2 |= IB_BTH_REQ_ACK;
+			if (++qp->s_cur == qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		case IB_WR_ATOMIC_CMP_AND_SWP:
+		case IB_WR_ATOMIC_FETCH_AND_ADD:
+			/*
+			 * Don't allow more operations to be started
+			 * than the QP limits allow.
+			 */
+			if (newreq) {
+				if (qp->s_num_rd_atomic >=
+				    qp->s_max_rd_atomic) {
+					qp->s_flags |= QIB_S_WAIT_RDMAR;
+					goto bail;
+				}
+				qp->s_num_rd_atomic++;
+				if (!(qp->s_flags & QIB_S_UNLIMITED_CREDIT))
+					qp->s_lsn++;
+				wqe->lpsn = wqe->psn;
+			}
+			if (wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
+				qp->s_state = OP(COMPARE_SWAP);
+				ohdr->u.atomic_eth.swap_data = cpu_to_be64(
+					wqe->wr.wr.atomic.swap);
+				ohdr->u.atomic_eth.compare_data = cpu_to_be64(
+					wqe->wr.wr.atomic.compare_add);
+			} else {
+				qp->s_state = OP(FETCH_ADD);
+				ohdr->u.atomic_eth.swap_data = cpu_to_be64(
+					wqe->wr.wr.atomic.compare_add);
+				ohdr->u.atomic_eth.compare_data = 0;
+			}
+			ohdr->u.atomic_eth.vaddr[0] = cpu_to_be32(
+				wqe->wr.wr.atomic.remote_addr >> 32);
+			ohdr->u.atomic_eth.vaddr[1] = cpu_to_be32(
+				wqe->wr.wr.atomic.remote_addr);
+			ohdr->u.atomic_eth.rkey = cpu_to_be32(
+				wqe->wr.wr.atomic.rkey);
+			hwords += sizeof(struct ib_atomic_eth) / sizeof(u32);
+			ss = NULL;
+			len = 0;
+			bth2 |= IB_BTH_REQ_ACK;
+			if (++qp->s_cur == qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		default:
+			goto bail;
+		}
+		qp->s_sge.sge = wqe->sg_list[0];
+		qp->s_sge.sg_list = wqe->sg_list + 1;
+		qp->s_sge.num_sge = wqe->wr.num_sge;
+		qp->s_sge.total_len = wqe->length;
+		qp->s_len = wqe->length;
+		if (newreq) {
+			qp->s_tail++;
+			if (qp->s_tail >= qp->s_size)
+				qp->s_tail = 0;
+		}
+		if (wqe->wr.opcode == IB_WR_RDMA_READ)
+			qp->s_psn = wqe->lpsn + 1;
+		else {
+			qp->s_psn++;
+			if (qib_cmp24(qp->s_psn, qp->s_next_psn) > 0)
+				qp->s_next_psn = qp->s_psn;
+		}
+		break;
+
+	case OP(RDMA_READ_RESPONSE_FIRST):
+		/*
+		 * qp->s_state is normally set to the opcode of the
+		 * last packet constructed for new requests and therefore
+		 * is never set to RDMA read response.
+		 * RDMA_READ_RESPONSE_FIRST is used by the ACK processing
+		 * thread to indicate a SEND needs to be restarted from an
+		 * earlier PSN without interferring with the sending thread.
+		 * See qib_restart_rc().
+		 */
+		qp->s_len = restart_sge(&qp->s_sge, wqe, qp->s_psn, pmtu);
+		/* FALLTHROUGH */
+	case OP(SEND_FIRST):
+		qp->s_state = OP(SEND_MIDDLE);
+		/* FALLTHROUGH */
+	case OP(SEND_MIDDLE):
+		bth2 = qp->s_psn++ & QIB_PSN_MASK;
+		if (qib_cmp24(qp->s_psn, qp->s_next_psn) > 0)
+			qp->s_next_psn = qp->s_psn;
+		ss = &qp->s_sge;
+		len = qp->s_len;
+		if (len > pmtu) {
+			len = pmtu;
+			break;
+		}
+		if (wqe->wr.opcode == IB_WR_SEND)
+			qp->s_state = OP(SEND_LAST);
+		else {
+			qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE);
+			/* Immediate data comes after the BTH */
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
+			hwords += 1;
+		}
+		if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+			bth0 |= IB_BTH_SOLICITED;
+		bth2 |= IB_BTH_REQ_ACK;
+		qp->s_cur++;
+		if (qp->s_cur >= qp->s_size)
+			qp->s_cur = 0;
+		break;
+
+	case OP(RDMA_READ_RESPONSE_LAST):
+		/*
+		 * qp->s_state is normally set to the opcode of the
+		 * last packet constructed for new requests and therefore
+		 * is never set to RDMA read response.
+		 * RDMA_READ_RESPONSE_LAST is used by the ACK processing
+		 * thread to indicate a RDMA write needs to be restarted from
+		 * an earlier PSN without interferring with the sending thread.
+		 * See qib_restart_rc().
+		 */
+		qp->s_len = restart_sge(&qp->s_sge, wqe, qp->s_psn, pmtu);
+		/* FALLTHROUGH */
+	case OP(RDMA_WRITE_FIRST):
+		qp->s_state = OP(RDMA_WRITE_MIDDLE);
+		/* FALLTHROUGH */
+	case OP(RDMA_WRITE_MIDDLE):
+		bth2 = qp->s_psn++ & QIB_PSN_MASK;
+		if (qib_cmp24(qp->s_psn, qp->s_next_psn) > 0)
+			qp->s_next_psn = qp->s_psn;
+		ss = &qp->s_sge;
+		len = qp->s_len;
+		if (len > pmtu) {
+			len = pmtu;
+			break;
+		}
+		if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
+			qp->s_state = OP(RDMA_WRITE_LAST);
+		else {
+			qp->s_state = OP(RDMA_WRITE_LAST_WITH_IMMEDIATE);
+			/* Immediate data comes after the BTH */
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
+			hwords += 1;
+			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+				bth0 |= IB_BTH_SOLICITED;
+		}
+		bth2 |= IB_BTH_REQ_ACK;
+		qp->s_cur++;
+		if (qp->s_cur >= qp->s_size)
+			qp->s_cur = 0;
+		break;
+
+	case OP(RDMA_READ_RESPONSE_MIDDLE):
+		/*
+		 * qp->s_state is normally set to the opcode of the
+		 * last packet constructed for new requests and therefore
+		 * is never set to RDMA read response.
+		 * RDMA_READ_RESPONSE_MIDDLE is used by the ACK processing
+		 * thread to indicate a RDMA read needs to be restarted from
+		 * an earlier PSN without interferring with the sending thread.
+		 * See qib_restart_rc().
+		 */
+		len = ((qp->s_psn - wqe->psn) & QIB_PSN_MASK) * pmtu;
+		ohdr->u.rc.reth.vaddr =
+			cpu_to_be64(wqe->wr.wr.rdma.remote_addr + len);
+		ohdr->u.rc.reth.rkey =
+			cpu_to_be32(wqe->wr.wr.rdma.rkey);
+		ohdr->u.rc.reth.length = cpu_to_be32(wqe->length - len);
+		qp->s_state = OP(RDMA_READ_REQUEST);
+		hwords += sizeof(ohdr->u.rc.reth) / sizeof(u32);
+		bth2 = (qp->s_psn & QIB_PSN_MASK) | IB_BTH_REQ_ACK;
+		qp->s_psn = wqe->lpsn + 1;
+		ss = NULL;
+		len = 0;
+		qp->s_cur++;
+		if (qp->s_cur == qp->s_size)
+			qp->s_cur = 0;
+		break;
+	}
+	qp->s_sending_hpsn = bth2;
+	delta = (((int) bth2 - (int) wqe->psn) << 8) >> 8;
+	if (delta && delta % QIB_PSN_CREDIT == 0)
+		bth2 |= IB_BTH_REQ_ACK;
+	if (qp->s_flags & QIB_S_SEND_ONE) {
+		qp->s_flags &= ~QIB_S_SEND_ONE;
+		qp->s_flags |= QIB_S_WAIT_ACK;
+		bth2 |= IB_BTH_REQ_ACK;
+	}
+	qp->s_len -= len;
+	qp->s_hdrwords = hwords;
+	qp->s_cur_sge = ss;
+	qp->s_cur_size = len;
+	qib_make_ruc_header(qp, ohdr, bth0 | (qp->s_state << 24), bth2);
+done:
+	ret = 1;
+	goto unlock;
+
+bail:
+	qp->s_flags &= ~QIB_S_BUSY;
+unlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+	return ret;
+}
+
+/**
+ * qib_send_rc_ack - Construct an ACK packet and send it
+ * @qp: a pointer to the QP
+ *
+ * This is called from qib_rc_rcv() and qib_kreceive().
+ * Note that RDMA reads and atomics are handled in the
+ * send side QP state and tasklet.
+ */
+void qib_send_rc_ack(struct qib_qp *qp)
+{
+	struct qib_devdata *dd = dd_from_ibdev(qp->ibqp.device);
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	u64 pbc;
+	u16 lrh0;
+	u32 bth0;
+	u32 hwords;
+	u32 pbufn;
+	u32 __iomem *piobuf;
+	struct qib_ib_header hdr;
+	struct qib_other_headers *ohdr;
+	u32 control;
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	/* Don't send ACK or NAK if a RDMA read or atomic is pending. */
+	if ((qp->s_flags & QIB_S_RESP_PENDING) || qp->s_rdma_ack_cnt)
+		goto queue_ack;
+
+	/* Construct the header with s_lock held so APM doesn't change it. */
+	ohdr = &hdr.u.oth;
+	lrh0 = QIB_LRH_BTH;
+	/* header size in 32-bit words LRH+BTH+AETH = (8+12+4)/4. */
+	hwords = 6;
+	if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
+		hwords += qib_make_grh(ibp, &hdr.u.l.grh,
+				       &qp->remote_ah_attr.grh, hwords, 0);
+		ohdr = &hdr.u.l.oth;
+		lrh0 = QIB_LRH_GRH;
+	}
+	/* read pkey_index w/o lock (its atomic) */
+	bth0 = qib_get_pkey(ibp, qp->s_pkey_index) | (OP(ACKNOWLEDGE) << 24);
+	if (qp->s_mig_state == IB_MIG_MIGRATED)
+		bth0 |= IB_BTH_MIG_REQ;
+	if (qp->r_nak_state)
+		ohdr->u.aeth = cpu_to_be32((qp->r_msn & QIB_MSN_MASK) |
+					    (qp->r_nak_state <<
+					     QIB_AETH_CREDIT_SHIFT));
+	else
+		ohdr->u.aeth = qib_compute_aeth(qp);
+	lrh0 |= ibp->sl_to_vl[qp->remote_ah_attr.sl] << 12 |
+		qp->remote_ah_attr.sl << 4;
+	hdr.lrh[0] = cpu_to_be16(lrh0);
+	hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
+	hdr.lrh[2] = cpu_to_be16(hwords + SIZE_OF_CRC);
+	hdr.lrh[3] = cpu_to_be16(ppd->lid | qp->remote_ah_attr.src_path_bits);
+	ohdr->bth[0] = cpu_to_be32(bth0);
+	ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
+	ohdr->bth[2] = cpu_to_be32(qp->r_ack_psn & QIB_PSN_MASK);
+
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+
+	/* Don't try to send ACKs if the link isn't ACTIVE */
+	if (!(ppd->lflags & QIBL_LINKACTIVE))
+		goto done;
+
+	control = dd->f_setpbc_control(ppd, hwords + SIZE_OF_CRC,
+				       qp->s_srate, lrh0 >> 12);
+	/* length is + 1 for the control dword */
+	pbc = ((u64) control << 32) | (hwords + 1);
+
+	piobuf = dd->f_getsendbuf(ppd, pbc, &pbufn);
+	if (!piobuf) {
+		/*
+		 * We are out of PIO buffers at the moment.
+		 * Pass responsibility for sending the ACK to the
+		 * send tasklet so that when a PIO buffer becomes
+		 * available, the ACK is sent ahead of other outgoing
+		 * packets.
+		 */
+		spin_lock_irqsave(&qp->s_lock, flags);
+		goto queue_ack;
+	}
+
+	/*
+	 * Write the pbc.
+	 * We have to flush after the PBC for correctness
+	 * on some cpus or WC buffer can be written out of order.
+	 */
+	writeq(pbc, piobuf);
+
+	if (dd->flags & QIB_PIO_FLUSH_WC) {
+		u32 *hdrp = (u32 *) &hdr;
+
+		qib_flush_wc();
+		qib_pio_copy(piobuf + 2, hdrp, hwords - 1);
+		qib_flush_wc();
+		__raw_writel(hdrp[hwords - 1], piobuf + hwords + 1);
+	} else
+		qib_pio_copy(piobuf + 2, (u32 *) &hdr, hwords);
+
+	if (dd->flags & QIB_USE_SPCL_TRIG) {
+		u32 spcl_off = (pbufn >= dd->piobcnt2k) ? 2047 : 1023;
+
+		qib_flush_wc();
+		__raw_writel(0xaebecede, piobuf + spcl_off);
+	}
+
+	qib_flush_wc();
+	qib_sendbuf_done(dd, pbufn);
+
+	ibp->n_unicast_xmit++;
+	goto done;
+
+queue_ack:
+	if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK) {
+		ibp->n_rc_qacks++;
+		qp->s_flags |= QIB_S_ACK_PENDING | QIB_S_RESP_PENDING;
+		qp->s_nak_state = qp->r_nak_state;
+		qp->s_ack_psn = qp->r_ack_psn;
+
+		/* Schedule the send tasklet. */
+		qib_schedule_send(qp);
+	}
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+done:
+	return;
+}
+
+/**
+ * reset_psn - reset the QP state to send starting from PSN
+ * @qp: the QP
+ * @psn: the packet sequence number to restart at
+ *
+ * This is called from qib_rc_rcv() to process an incoming RC ACK
+ * for the given QP.
+ * Called at interrupt level with the QP s_lock held.
+ */
+static void reset_psn(struct qib_qp *qp, u32 psn)
+{
+	u32 n = qp->s_acked;
+	struct qib_swqe *wqe = get_swqe_ptr(qp, n);
+	u32 opcode;
+
+	qp->s_cur = n;
+
+	/*
+	 * If we are starting the request from the beginning,
+	 * let the normal send code handle initialization.
+	 */
+	if (qib_cmp24(psn, wqe->psn) <= 0) {
+		qp->s_state = OP(SEND_LAST);
+		goto done;
+	}
+
+	/* Find the work request opcode corresponding to the given PSN. */
+	opcode = wqe->wr.opcode;
+	for (;;) {
+		int diff;
+
+		if (++n == qp->s_size)
+			n = 0;
+		if (n == qp->s_tail)
+			break;
+		wqe = get_swqe_ptr(qp, n);
+		diff = qib_cmp24(psn, wqe->psn);
+		if (diff < 0)
+			break;
+		qp->s_cur = n;
+		/*
+		 * If we are starting the request from the beginning,
+		 * let the normal send code handle initialization.
+		 */
+		if (diff == 0) {
+			qp->s_state = OP(SEND_LAST);
+			goto done;
+		}
+		opcode = wqe->wr.opcode;
+	}
+
+	/*
+	 * Set the state to restart in the middle of a request.
+	 * Don't change the s_sge, s_cur_sge, or s_cur_size.
+	 * See qib_make_rc_req().
+	 */
+	switch (opcode) {
+	case IB_WR_SEND:
+	case IB_WR_SEND_WITH_IMM:
+		qp->s_state = OP(RDMA_READ_RESPONSE_FIRST);
+		break;
+
+	case IB_WR_RDMA_WRITE:
+	case IB_WR_RDMA_WRITE_WITH_IMM:
+		qp->s_state = OP(RDMA_READ_RESPONSE_LAST);
+		break;
+
+	case IB_WR_RDMA_READ:
+		qp->s_state = OP(RDMA_READ_RESPONSE_MIDDLE);
+		break;
+
+	default:
+		/*
+		 * This case shouldn't happen since its only
+		 * one PSN per req.
+		 */
+		qp->s_state = OP(SEND_LAST);
+	}
+done:
+	qp->s_psn = psn;
+	/*
+	 * Set QIB_S_WAIT_PSN as qib_rc_complete() may start the timer
+	 * asynchronously before the send tasklet can get scheduled.
+	 * Doing it in qib_make_rc_req() is too late.
+	 */
+	if ((qib_cmp24(qp->s_psn, qp->s_sending_hpsn) <= 0) &&
+	    (qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) <= 0))
+		qp->s_flags |= QIB_S_WAIT_PSN;
+}
+
+/*
+ * Back up requester to resend the last un-ACKed request.
+ * The QP s_lock should be held and interrupts disabled.
+ */
+static void qib_restart_rc(struct qib_qp *qp, u32 psn, int wait)
+{
+	struct qib_swqe *wqe = get_swqe_ptr(qp, qp->s_acked);
+	struct qib_ibport *ibp;
+
+	if (qp->s_retry == 0) {
+		if (qp->s_mig_state == IB_MIG_ARMED) {
+			qib_migrate_qp(qp);
+			qp->s_retry = qp->s_retry_cnt;
+		} else if (qp->s_last == qp->s_acked) {
+			qib_send_complete(qp, wqe, IB_WC_RETRY_EXC_ERR);
+			qib_error_qp(qp, IB_WC_WR_FLUSH_ERR);
+			return;
+		} else {
+			/* XXX need to handle delayed completion */
+			qib_dbg("Delayed too many retries\n");
+			return;
+		}
+	} else
+		qp->s_retry--;
+
+	ibp = to_iport(qp->ibqp.device, qp->port_num);
+	if (wqe->wr.opcode == IB_WR_RDMA_READ)
+		ibp->n_rc_resends++;
+	else
+		ibp->n_rc_resends += (qp->s_psn - psn) & QIB_PSN_MASK;
+
+	qp->s_flags &= ~(QIB_S_WAIT_FENCE | QIB_S_WAIT_RDMAR |
+			 QIB_S_WAIT_SSN_CREDIT | QIB_S_WAIT_PSN |
+			 QIB_S_WAIT_ACK);
+	if (wait)
+		qp->s_flags |= QIB_S_SEND_ONE;
+	reset_psn(qp, psn);
+}
+
+/*
+ * This is called from s_timer for missing responses.
+ */
+static void rc_timeout(unsigned long arg)
+{
+	struct qib_qp *qp = (struct qib_qp *)arg;
+	struct qib_ibport *ibp;
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+	if (qp->s_flags & QIB_S_TIMER) {
+		ibp = to_iport(qp->ibqp.device, qp->port_num);
+		ibp->n_timeouts++;
+		qp->s_flags &= ~QIB_S_TIMER;
+		del_timer(&qp->s_timer);
+		qib_restart_rc(qp, qp->s_last_psn + 1, 1);
+		qib_schedule_send(qp);
+	}
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+}
+
+/*
+ * This is called from s_timer for RNR timeouts.
+ */
+void qib_rc_rnr_retry(unsigned long arg)
+{
+	struct qib_qp *qp = (struct qib_qp *)arg;
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+	if (qp->s_flags & QIB_S_WAIT_RNR) {
+		qp->s_flags &= ~QIB_S_WAIT_RNR;
+		del_timer(&qp->s_timer);
+		qib_schedule_send(qp);
+	}
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+}
+
+/*
+ * Set qp->s_sending_psn to the next PSN after the given one.
+ * This would be psn+1 except when RDMA reads are present.
+ */
+static void reset_sending_psn(struct qib_qp *qp, u32 psn)
+{
+	struct qib_swqe *wqe;
+	u32 n = qp->s_last;
+
+	/* Find the work request corresponding to the given PSN. */
+	for (;;) {
+		wqe = get_swqe_ptr(qp, n);
+		if (qib_cmp24(psn, wqe->lpsn) <= 0) {
+			if (wqe->wr.opcode == IB_WR_RDMA_READ)
+				qp->s_sending_psn = wqe->lpsn + 1;
+			else
+				qp->s_sending_psn = psn + 1;
+			break;
+		}
+		if (++n == qp->s_size)
+			n = 0;
+		if (n == qp->s_tail)
+			break;
+	}
+}
+
+/*
+ * This should be called with the QP s_lock held and interrupts disabled.
+ */
+void qib_rc_send_complete(struct qib_qp *qp, struct qib_ib_header *hdr)
+{
+	struct qib_other_headers *ohdr;
+	struct qib_swqe *wqe;
+	struct ib_wc wc;
+	unsigned i;
+	u32 opcode;
+	u32 psn;
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_OR_FLUSH_SEND))
+		return;
+
+	/* Find out where the BTH is */
+	if ((be16_to_cpu(hdr->lrh[0]) & 3) == QIB_LRH_BTH)
+		ohdr = &hdr->u.oth;
+	else
+		ohdr = &hdr->u.l.oth;
+
+	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
+	if (opcode >= OP(RDMA_READ_RESPONSE_FIRST) &&
+	    opcode <= OP(ATOMIC_ACKNOWLEDGE)) {
+		WARN_ON(!qp->s_rdma_ack_cnt);
+		qp->s_rdma_ack_cnt--;
+		return;
+	}
+
+	psn = be32_to_cpu(ohdr->bth[2]);
+	reset_sending_psn(qp, psn);
+
+	/*
+	 * Start timer after a packet requesting an ACK has been sent and
+	 * there are still requests that haven't been acked.
+	 */
+	if ((psn & IB_BTH_REQ_ACK) && qp->s_acked != qp->s_tail &&
+	    !(qp->s_flags & (QIB_S_TIMER | QIB_S_WAIT_RNR | QIB_S_WAIT_PSN)))
+		start_timer(qp);
+
+	while (qp->s_last != qp->s_acked) {
+		wqe = get_swqe_ptr(qp, qp->s_last);
+		if (qib_cmp24(wqe->lpsn, qp->s_sending_psn) >= 0 &&
+		    qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) <= 0)
+			break;
+		for (i = 0; i < wqe->wr.num_sge; i++) {
+			struct qib_sge *sge = &wqe->sg_list[i];
+
+			atomic_dec(&sge->mr->refcount);
+		}
+		/* Post a send completion queue entry if requested. */
+		if (!(qp->s_flags & QIB_S_SIGNAL_REQ_WR) ||
+		    (wqe->wr.send_flags & IB_SEND_SIGNALED)) {
+			memset(&wc, 0, sizeof wc);
+			wc.wr_id = wqe->wr.wr_id;
+			wc.status = IB_WC_SUCCESS;
+			wc.opcode = ib_qib_wc_opcode[wqe->wr.opcode];
+			wc.byte_len = wqe->length;
+			wc.qp = &qp->ibqp;
+			qib_cq_enter(to_icq(qp->ibqp.send_cq), &wc, 0);
+		}
+		if (++qp->s_last >= qp->s_size)
+			qp->s_last = 0;
+	}
+	/*
+	 * If we were waiting for sends to complete before resending,
+	 * and they are now complete, restart sending.
+	 */
+	if (qp->s_flags & QIB_S_WAIT_PSN &&
+	    qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) > 0) {
+		qp->s_flags &= ~QIB_S_WAIT_PSN;
+		qp->s_sending_psn = qp->s_psn;
+		qp->s_sending_hpsn = qp->s_psn - 1;
+		qib_schedule_send(qp);
+	}
+}
+
+static inline void update_last_psn(struct qib_qp *qp, u32 psn)
+{
+	qp->s_last_psn = psn;
+}
+
+/*
+ * Generate a SWQE completion.
+ * This is similar to qib_send_complete but has to check to be sure
+ * that the SGEs are not being referenced if the SWQE is being resent.
+ */
+static struct qib_swqe *do_rc_completion(struct qib_qp *qp,
+					 struct qib_swqe *wqe,
+					 struct qib_ibport *ibp)
+{
+	struct ib_wc wc;
+	unsigned i;
+
+	/*
+	 * Don't decrement refcount and don't generate a
+	 * completion if the SWQE is being resent until the send
+	 * is finished.
+	 */
+	if (qib_cmp24(wqe->lpsn, qp->s_sending_psn) < 0 ||
+	    qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) > 0) {
+		for (i = 0; i < wqe->wr.num_sge; i++) {
+			struct qib_sge *sge = &wqe->sg_list[i];
+
+			atomic_dec(&sge->mr->refcount);
+		}
+		/* Post a send completion queue entry if requested. */
+		if (!(qp->s_flags & QIB_S_SIGNAL_REQ_WR) ||
+		    (wqe->wr.send_flags & IB_SEND_SIGNALED)) {
+			memset(&wc, 0, sizeof wc);
+			wc.wr_id = wqe->wr.wr_id;
+			wc.status = IB_WC_SUCCESS;
+			wc.opcode = ib_qib_wc_opcode[wqe->wr.opcode];
+			wc.byte_len = wqe->length;
+			wc.qp = &qp->ibqp;
+			qib_cq_enter(to_icq(qp->ibqp.send_cq), &wc, 0);
+		}
+		if (++qp->s_last >= qp->s_size)
+			qp->s_last = 0;
+	} else
+		ibp->n_rc_delayed_comp++;
+
+	qp->s_retry = qp->s_retry_cnt;
+	update_last_psn(qp, wqe->lpsn);
+
+	/*
+	 * If we are completing a request which is in the process of
+	 * being resent, we can stop resending it since we know the
+	 * responder has already seen it.
+	 */
+	if (qp->s_acked == qp->s_cur) {
+		if (++qp->s_cur >= qp->s_size)
+			qp->s_cur = 0;
+		qp->s_acked = qp->s_cur;
+		wqe = get_swqe_ptr(qp, qp->s_cur);
+		if (qp->s_acked != qp->s_tail) {
+			qp->s_state = OP(SEND_LAST);
+			qp->s_psn = wqe->psn;
+		}
+	} else {
+		if (++qp->s_acked >= qp->s_size)
+			qp->s_acked = 0;
+		if (qp->state == IB_QPS_SQD && qp->s_acked == qp->s_cur)
+			qp->s_draining = 0;
+		wqe = get_swqe_ptr(qp, qp->s_acked);
+	}
+	return wqe;
+}
+
+/**
+ * do_rc_ack - process an incoming RC ACK
+ * @qp: the QP the ACK came in on
+ * @psn: the packet sequence number of the ACK
+ * @opcode: the opcode of the request that resulted in the ACK
+ *
+ * This is called from qib_rc_rcv_resp() to process an incoming RC ACK
+ * for the given QP.
+ * Called at interrupt level with the QP s_lock held.
+ * Returns 1 if OK, 0 if current operation should be aborted (NAK).
+ */
+static int do_rc_ack(struct qib_qp *qp, u32 aeth, u32 psn, int opcode,
+		     u64 val, struct qib_ctxtdata *rcd)
+{
+	struct qib_ibport *ibp;
+	enum ib_wc_status status;
+	struct qib_swqe *wqe;
+	int ret = 0;
+	u32 ack_psn;
+	int diff;
+
+	/* Remove QP from retry timer */
+	if (qp->s_flags & (QIB_S_TIMER | QIB_S_WAIT_RNR)) {
+		qp->s_flags &= ~(QIB_S_TIMER | QIB_S_WAIT_RNR);
+		del_timer(&qp->s_timer);
+	}
+
+	/*
+	 * Note that NAKs implicitly ACK outstanding SEND and RDMA write
+	 * requests and implicitly NAK RDMA read and atomic requests issued
+	 * before the NAK'ed request.  The MSN won't include the NAK'ed
+	 * request but will include an ACK'ed request(s).
+	 */
+	ack_psn = psn;
+	if (aeth >> 29)
+		ack_psn--;
+	wqe = get_swqe_ptr(qp, qp->s_acked);
+	ibp = to_iport(qp->ibqp.device, qp->port_num);
+
+	/*
+	 * The MSN might be for a later WQE than the PSN indicates so
+	 * only complete WQEs that the PSN finishes.
+	 */
+	while ((diff = qib_cmp24(ack_psn, wqe->lpsn)) >= 0) {
+		/*
+		 * RDMA_READ_RESPONSE_ONLY is a special case since
+		 * we want to generate completion events for everything
+		 * before the RDMA read, copy the data, then generate
+		 * the completion for the read.
+		 */
+		if (wqe->wr.opcode == IB_WR_RDMA_READ &&
+		    opcode == OP(RDMA_READ_RESPONSE_ONLY) &&
+		    diff == 0) {
+			ret = 1;
+			goto bail;
+		}
+		/*
+		 * If this request is a RDMA read or atomic, and the ACK is
+		 * for a later operation, this ACK NAKs the RDMA read or
+		 * atomic.  In other words, only a RDMA_READ_LAST or ONLY
+		 * can ACK a RDMA read and likewise for atomic ops.  Note
+		 * that the NAK case can only happen if relaxed ordering is
+		 * used and requests are sent after an RDMA read or atomic
+		 * is sent but before the response is received.
+		 */
+		if ((wqe->wr.opcode == IB_WR_RDMA_READ &&
+		     (opcode != OP(RDMA_READ_RESPONSE_LAST) || diff != 0)) ||
+		    ((wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+		      wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) &&
+		     (opcode != OP(ATOMIC_ACKNOWLEDGE) || diff != 0))) {
+			/* Retry this request. */
+			if (!(qp->r_flags & QIB_R_RDMAR_SEQ)) {
+				qp->r_flags |= QIB_R_RDMAR_SEQ;
+				qib_restart_rc(qp, qp->s_last_psn + 1, 0);
+				if (list_empty(&qp->rspwait)) {
+					qp->r_flags |= QIB_R_RSP_SEND;
+					atomic_inc(&qp->refcount);
+					list_add_tail(&qp->rspwait,
+						      &rcd->qp_wait_list);
+				}
+			}
+			/*
+			 * No need to process the ACK/NAK since we are
+			 * restarting an earlier request.
+			 */
+			goto bail;
+		}
+		if (wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+		    wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
+			u64 *vaddr = wqe->sg_list[0].vaddr;
+			*vaddr = val;
+		}
+		if (qp->s_num_rd_atomic &&
+		    (wqe->wr.opcode == IB_WR_RDMA_READ ||
+		     wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+		     wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD)) {
+			qp->s_num_rd_atomic--;
+			/* Restart sending task if fence is complete */
+			if ((qp->s_flags & QIB_S_WAIT_FENCE) &&
+			    !qp->s_num_rd_atomic) {
+				qp->s_flags &= ~(QIB_S_WAIT_FENCE |
+						 QIB_S_WAIT_ACK);
+				qib_schedule_send(qp);
+			} else if (qp->s_flags & QIB_S_WAIT_RDMAR) {
+				qp->s_flags &= ~(QIB_S_WAIT_RDMAR |
+						 QIB_S_WAIT_ACK);
+				qib_schedule_send(qp);
+			}
+		}
+		wqe = do_rc_completion(qp, wqe, ibp);
+		if (qp->s_acked == qp->s_tail)
+			break;
+	}
+
+	switch (aeth >> 29) {
+	case 0:         /* ACK */
+		ibp->n_rc_acks++;
+		if (qp->s_acked != qp->s_tail) {
+			/*
+			 * We are expecting more ACKs so
+			 * reset the retransmit timer.
+			 */
+			start_timer(qp);
+			/*
+			 * We can stop resending the earlier packets and
+			 * continue with the next packet the receiver wants.
+			 */
+			if (qib_cmp24(qp->s_psn, psn) <= 0)
+				reset_psn(qp, psn + 1);
+		} else if (qib_cmp24(qp->s_psn, psn) <= 0) {
+			qp->s_state = OP(SEND_LAST);
+			qp->s_psn = psn + 1;
+		}
+		if (qp->s_flags & QIB_S_WAIT_ACK) {
+			qp->s_flags &= ~QIB_S_WAIT_ACK;
+			qib_schedule_send(qp);
+		}
+		qib_get_credit(qp, aeth);
+		qp->s_rnr_retry = qp->s_rnr_retry_cnt;
+		qp->s_retry = qp->s_retry_cnt;
+		update_last_psn(qp, psn);
+		ret = 1;
+		goto bail;
+
+	case 1:         /* RNR NAK */
+		ibp->n_rnr_naks++;
+		if (qp->s_acked == qp->s_tail)
+			goto bail;
+		if (qp->s_flags & QIB_S_WAIT_RNR)
+			goto bail;
+		if (qp->s_rnr_retry == 0) {
+			status = IB_WC_RNR_RETRY_EXC_ERR;
+			goto class_b;
+		}
+		if (qp->s_rnr_retry_cnt < 7)
+			qp->s_rnr_retry--;
+
+		/* The last valid PSN is the previous PSN. */
+		update_last_psn(qp, psn - 1);
+
+		ibp->n_rc_resends += (qp->s_psn - psn) & QIB_PSN_MASK;
+
+		reset_psn(qp, psn);
+
+		qp->s_flags &= ~(QIB_S_WAIT_SSN_CREDIT | QIB_S_WAIT_ACK);
+		qp->s_flags |= QIB_S_WAIT_RNR;
+		qp->s_timer.function = qib_rc_rnr_retry;
+		qp->s_timer.expires = jiffies + usecs_to_jiffies(
+			ib_qib_rnr_table[(aeth >> QIB_AETH_CREDIT_SHIFT) &
+					   QIB_AETH_CREDIT_MASK]);
+		add_timer(&qp->s_timer);
+		goto bail;
+
+	case 3:         /* NAK */
+		if (qp->s_acked == qp->s_tail)
+			goto bail;
+		/* The last valid PSN is the previous PSN. */
+		update_last_psn(qp, psn - 1);
+		switch ((aeth >> QIB_AETH_CREDIT_SHIFT) &
+			QIB_AETH_CREDIT_MASK) {
+		case 0: /* PSN sequence error */
+			ibp->n_seq_naks++;
+			/*
+			 * Back up to the responder's expected PSN.
+			 * Note that we might get a NAK in the middle of an
+			 * RDMA READ response which terminates the RDMA
+			 * READ.
+			 */
+			qib_restart_rc(qp, psn, 0);
+			qib_schedule_send(qp);
+			break;
+
+		case 1: /* Invalid Request */
+			status = IB_WC_REM_INV_REQ_ERR;
+			ibp->n_other_naks++;
+			goto class_b;
+
+		case 2: /* Remote Access Error */
+			status = IB_WC_REM_ACCESS_ERR;
+			ibp->n_other_naks++;
+			goto class_b;
+
+		case 3: /* Remote Operation Error */
+			status = IB_WC_REM_OP_ERR;
+			ibp->n_other_naks++;
+class_b:
+			if (qp->s_last == qp->s_acked) {
+				qib_send_complete(qp, wqe, status);
+				qib_error_qp(qp, IB_WC_WR_FLUSH_ERR);
+			} else {
+				/* XXX need to handle delayed completion */
+				qib_dbg("Delayed too many retries\n");
+			}
+			break;
+
+		default:
+			/* Ignore other reserved NAK error codes */
+			goto reserved;
+		}
+		qp->s_retry = qp->s_retry_cnt;
+		qp->s_rnr_retry = qp->s_rnr_retry_cnt;
+		goto bail;
+
+	default:                /* 2: reserved */
+reserved:
+		/* Ignore reserved NAK codes. */
+		goto bail;
+	}
+
+bail:
+	return ret;
+}
+
+/*
+ * We have seen an out of sequence RDMA read middle or last packet.
+ * This ACKs SENDs and RDMA writes up to the first RDMA read or atomic SWQE.
+ */
+static void rdma_seq_err(struct qib_qp *qp, struct qib_ibport *ibp, u32 psn,
+			 struct qib_ctxtdata *rcd)
+{
+	struct qib_swqe *wqe;
+
+	/* Remove QP from retry timer */
+	if (qp->s_flags & (QIB_S_TIMER | QIB_S_WAIT_RNR)) {
+		qp->s_flags &= ~(QIB_S_TIMER | QIB_S_WAIT_RNR);
+		del_timer(&qp->s_timer);
+	}
+
+	wqe = get_swqe_ptr(qp, qp->s_acked);
+
+	while (qib_cmp24(psn, wqe->lpsn) > 0) {
+		if (wqe->wr.opcode == IB_WR_RDMA_READ ||
+		    wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+		    wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
+			break;
+		wqe = do_rc_completion(qp, wqe, ibp);
+	}
+
+	ibp->n_rdma_seq++;
+	qp->r_flags |= QIB_R_RDMAR_SEQ;
+	qib_restart_rc(qp, qp->s_last_psn + 1, 0);
+	if (list_empty(&qp->rspwait)) {
+		qp->r_flags |= QIB_R_RSP_SEND;
+		atomic_inc(&qp->refcount);
+		list_add_tail(&qp->rspwait, &rcd->qp_wait_list);
+	}
+}
+
+/**
+ * qib_rc_rcv_resp - process an incoming RC response packet
+ * @ibp: the port this packet came in on
+ * @ohdr: the other headers for this packet
+ * @data: the packet data
+ * @tlen: the packet length
+ * @qp: the QP for this packet
+ * @opcode: the opcode for this packet
+ * @psn: the packet sequence number for this packet
+ * @hdrsize: the header length
+ * @pmtu: the path MTU
+ *
+ * This is called from qib_rc_rcv() to process an incoming RC response
+ * packet for the given QP.
+ * Called at interrupt level.
+ */
+static void qib_rc_rcv_resp(struct qib_ibport *ibp,
+			    struct qib_other_headers *ohdr,
+			    void *data, u32 tlen,
+			    struct qib_qp *qp,
+			    u32 opcode,
+			    u32 psn, u32 hdrsize, u32 pmtu,
+			    struct qib_ctxtdata *rcd)
+{
+	struct qib_swqe *wqe;
+	enum ib_wc_status status;
+	unsigned long flags;
+	int diff;
+	u32 pad;
+	u32 aeth;
+	u64 val;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	/* Double check we can process this now that we hold the s_lock. */
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK))
+		goto ack_done;
+
+	/* Ignore invalid responses. */
+	if (qib_cmp24(psn, qp->s_next_psn) >= 0)
+		goto ack_done;
+
+	/* Ignore duplicate responses. */
+	diff = qib_cmp24(psn, qp->s_last_psn);
+	if (unlikely(diff <= 0)) {
+		/* Update credits for "ghost" ACKs */
+		if (diff == 0 && opcode == OP(ACKNOWLEDGE)) {
+			aeth = be32_to_cpu(ohdr->u.aeth);
+			if ((aeth >> 29) == 0)
+				qib_get_credit(qp, aeth);
+		}
+		goto ack_done;
+	}
+
+	/*
+	 * Skip everything other than the PSN we expect, if we are waiting
+	 * for a reply to a restarted RDMA read or atomic op.
+	 */
+	if (qp->r_flags & QIB_R_RDMAR_SEQ) {
+		if (qib_cmp24(psn, qp->s_last_psn + 1) != 0)
+			goto ack_done;
+		qp->r_flags &= ~QIB_R_RDMAR_SEQ;
+	}
+
+	if (unlikely(qp->s_acked == qp->s_tail))
+		goto ack_done;
+	wqe = get_swqe_ptr(qp, qp->s_acked);
+	status = IB_WC_SUCCESS;
+
+	switch (opcode) {
+	case OP(ACKNOWLEDGE):
+	case OP(ATOMIC_ACKNOWLEDGE):
+	case OP(RDMA_READ_RESPONSE_FIRST):
+		aeth = be32_to_cpu(ohdr->u.aeth);
+		if (opcode == OP(ATOMIC_ACKNOWLEDGE)) {
+			__be32 *p = ohdr->u.at.atomic_ack_eth;
+
+			val = ((u64) be32_to_cpu(p[0]) << 32) |
+				be32_to_cpu(p[1]);
+		} else
+			val = 0;
+		if (!do_rc_ack(qp, aeth, psn, opcode, val, rcd) ||
+		    opcode != OP(RDMA_READ_RESPONSE_FIRST))
+			goto ack_done;
+		hdrsize += 4;
+		wqe = get_swqe_ptr(qp, qp->s_acked);
+		if (unlikely(wqe->wr.opcode != IB_WR_RDMA_READ))
+			goto ack_op_err;
+		/*
+		 * If this is a response to a resent RDMA read, we
+		 * have to be careful to copy the data to the right
+		 * location.
+		 */
+		qp->s_rdma_read_len = restart_sge(&qp->s_rdma_read_sge,
+						  wqe, psn, pmtu);
+		goto read_middle;
+
+	case OP(RDMA_READ_RESPONSE_MIDDLE):
+		/* no AETH, no ACK */
+		if (unlikely(qib_cmp24(psn, qp->s_last_psn + 1)))
+			goto ack_seq_err;
+		if (unlikely(wqe->wr.opcode != IB_WR_RDMA_READ))
+			goto ack_op_err;
+read_middle:
+		if (unlikely(tlen != (hdrsize + pmtu + 4)))
+			goto ack_len_err;
+		if (unlikely(pmtu >= qp->s_rdma_read_len))
+			goto ack_len_err;
+
+		/*
+		 * We got a response so update the timeout.
+		 * 4.096 usec. * (1 << qp->timeout)
+		 */
+		qp->s_flags |= QIB_S_TIMER;
+		mod_timer(&qp->s_timer, jiffies +
+			usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /
+					 1000UL));
+		if (qp->s_flags & QIB_S_WAIT_ACK) {
+			qp->s_flags &= ~QIB_S_WAIT_ACK;
+			qib_schedule_send(qp);
+		}
+
+		if (opcode == OP(RDMA_READ_RESPONSE_MIDDLE))
+			qp->s_retry = qp->s_retry_cnt;
+
+		/*
+		 * Update the RDMA receive state but do the copy w/o
+		 * holding the locks and blocking interrupts.
+		 */
+		qp->s_rdma_read_len -= pmtu;
+		update_last_psn(qp, psn);
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+		qib_copy_sge(&qp->s_rdma_read_sge, data, pmtu, 0);
+		goto bail;
+
+	case OP(RDMA_READ_RESPONSE_ONLY):
+		aeth = be32_to_cpu(ohdr->u.aeth);
+		if (!do_rc_ack(qp, aeth, psn, opcode, 0, rcd))
+			goto ack_done;
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/*
+		 * Check that the data size is >= 0 && <= pmtu.
+		 * Remember to account for the AETH header (4) and
+		 * ICRC (4).
+		 */
+		if (unlikely(tlen < (hdrsize + pad + 8)))
+			goto ack_len_err;
+		/*
+		 * If this is a response to a resent RDMA read, we
+		 * have to be careful to copy the data to the right
+		 * location.
+		 */
+		wqe = get_swqe_ptr(qp, qp->s_acked);
+		qp->s_rdma_read_len = restart_sge(&qp->s_rdma_read_sge,
+						  wqe, psn, pmtu);
+		goto read_last;
+
+	case OP(RDMA_READ_RESPONSE_LAST):
+		/* ACKs READ req. */
+		if (unlikely(qib_cmp24(psn, qp->s_last_psn + 1)))
+			goto ack_seq_err;
+		if (unlikely(wqe->wr.opcode != IB_WR_RDMA_READ))
+			goto ack_op_err;
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/*
+		 * Check that the data size is >= 1 && <= pmtu.
+		 * Remember to account for the AETH header (4) and
+		 * ICRC (4).
+		 */
+		if (unlikely(tlen <= (hdrsize + pad + 8)))
+			goto ack_len_err;
+read_last:
+		tlen -= hdrsize + pad + 8;
+		if (unlikely(tlen != qp->s_rdma_read_len))
+			goto ack_len_err;
+		aeth = be32_to_cpu(ohdr->u.aeth);
+		qib_copy_sge(&qp->s_rdma_read_sge, data, tlen, 0);
+		WARN_ON(qp->s_rdma_read_sge.num_sge);
+		(void) do_rc_ack(qp, aeth, psn,
+				 OP(RDMA_READ_RESPONSE_LAST), 0, rcd);
+		goto ack_done;
+	}
+
+ack_op_err:
+	status = IB_WC_LOC_QP_OP_ERR;
+	goto ack_err;
+
+ack_seq_err:
+	rdma_seq_err(qp, ibp, psn, rcd);
+	goto ack_done;
+
+ack_len_err:
+	status = IB_WC_LOC_LEN_ERR;
+ack_err:
+	if (qp->s_last == qp->s_acked) {
+		qib_send_complete(qp, wqe, status);
+		qib_error_qp(qp, IB_WC_WR_FLUSH_ERR);
+	} else {
+		/* XXX need to handle delayed completion */
+		qib_dbg("Delayed too many retries\n");
+	}
+ack_done:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+bail:
+	return;
+}
+
+/**
+ * qib_rc_rcv_error - process an incoming duplicate or error RC packet
+ * @ohdr: the other headers for this packet
+ * @data: the packet data
+ * @qp: the QP for this packet
+ * @opcode: the opcode for this packet
+ * @psn: the packet sequence number for this packet
+ * @diff: the difference between the PSN and the expected PSN
+ *
+ * This is called from qib_rc_rcv() to process an unexpected
+ * incoming RC packet for the given QP.
+ * Called at interrupt level.
+ * Return 1 if no more processing is needed; otherwise return 0 to
+ * schedule a response to be sent.
+ */
+static int qib_rc_rcv_error(struct qib_other_headers *ohdr,
+			    void *data,
+			    struct qib_qp *qp,
+			    u32 opcode,
+			    u32 psn,
+			    int diff,
+			    struct qib_ctxtdata *rcd)
+{
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	struct qib_ack_entry *e;
+	unsigned long flags;
+	u8 i, prev;
+	int old_req;
+
+	if (diff > 0) {
+		/*
+		 * Packet sequence error.
+		 * A NAK will ACK earlier sends and RDMA writes.
+		 * Don't queue the NAK if we already sent one.
+		 */
+		if (!qp->r_nak_state) {
+			ibp->n_rc_seqnak++;
+			qp->r_nak_state = IB_NAK_PSN_ERROR;
+			/* Use the expected PSN. */
+			qp->r_ack_psn = qp->r_psn;
+			/*
+			 * Wait to send the sequence NAK until all packets
+			 * in the receive queue have been processed.
+			 * Otherwise, we end up propagating congestion.
+			 */
+			if (list_empty(&qp->rspwait)) {
+				qp->r_flags |= QIB_R_RSP_NAK;
+				atomic_inc(&qp->refcount);
+				list_add_tail(&qp->rspwait, &rcd->qp_wait_list);
+			}
+		}
+		goto done;
+	}
+
+	/*
+	 * Handle a duplicate request.  Don't re-execute SEND, RDMA
+	 * write or atomic op.  Don't NAK errors, just silently drop
+	 * the duplicate request.  Note that r_sge, r_len, and
+	 * r_rcv_len may be in use so don't modify them.
+	 *
+	 * We are supposed to ACK the earliest duplicate PSN but we
+	 * can coalesce an outstanding duplicate ACK.  We have to
+	 * send the earliest so that RDMA reads can be restarted at
+	 * the requester's expected PSN.
+	 *
+	 * First, find where this duplicate PSN falls within the
+	 * ACKs previously sent.
+	 * old_req is true if there is an older response that is scheduled
+	 * to be sent before sending this one.
+	 */
+	e = NULL;
+	old_req = 1;
+	ibp->n_rc_dupreq++;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+	/* Double check we can process this now that we hold the s_lock. */
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK))
+		goto unlock_done;
+
+	for (i = qp->r_head_ack_queue; ; i = prev) {
+		if (i == qp->s_tail_ack_queue)
+			old_req = 0;
+		if (i)
+			prev = i - 1;
+		else
+			prev = QIB_MAX_RDMA_ATOMIC;
+		if (prev == qp->r_head_ack_queue) {
+			e = NULL;
+			break;
+		}
+		e = &qp->s_ack_queue[prev];
+		if (!e->opcode) {
+			e = NULL;
+			break;
+		}
+		if (qib_cmp24(psn, e->psn) >= 0) {
+			if (prev == qp->s_tail_ack_queue &&
+			    qib_cmp24(psn, e->lpsn) <= 0)
+				old_req = 0;
+			break;
+		}
+	}
+	switch (opcode) {
+	case OP(RDMA_READ_REQUEST): {
+		struct ib_reth *reth;
+		u32 offset;
+		u32 len;
+
+		/*
+		 * If we didn't find the RDMA read request in the ack queue,
+		 * we can ignore this request.
+		 */
+		if (!e || e->opcode != OP(RDMA_READ_REQUEST))
+			goto unlock_done;
+		/* RETH comes after BTH */
+		reth = &ohdr->u.rc.reth;
+		/*
+		 * Address range must be a subset of the original
+		 * request and start on pmtu boundaries.
+		 * We reuse the old ack_queue slot since the requester
+		 * should not back up and request an earlier PSN for the
+		 * same request.
+		 */
+		offset = ((psn - e->psn) & QIB_PSN_MASK) *
+			ib_mtu_enum_to_int(qp->path_mtu);
+		len = be32_to_cpu(reth->length);
+		if (unlikely(offset + len != e->rdma_sge.sge_length))
+			goto unlock_done;
+		if (e->rdma_sge.mr) {
+			atomic_dec(&e->rdma_sge.mr->refcount);
+			e->rdma_sge.mr = NULL;
+		}
+		if (len != 0) {
+			u32 rkey = be32_to_cpu(reth->rkey);
+			u64 vaddr = be64_to_cpu(reth->vaddr);
+			int ok;
+
+			ok = qib_rkey_ok(qp, &e->rdma_sge, len, vaddr, rkey,
+					 IB_ACCESS_REMOTE_READ);
+			if (unlikely(!ok))
+				goto unlock_done;
+		} else {
+			e->rdma_sge.vaddr = NULL;
+			e->rdma_sge.length = 0;
+			e->rdma_sge.sge_length = 0;
+		}
+		e->psn = psn;
+		if (old_req)
+			goto unlock_done;
+		qp->s_tail_ack_queue = prev;
+		break;
+	}
+
+	case OP(COMPARE_SWAP):
+	case OP(FETCH_ADD): {
+		/*
+		 * If we didn't find the atomic request in the ack queue
+		 * or the send tasklet is already backed up to send an
+		 * earlier entry, we can ignore this request.
+		 */
+		if (!e || e->opcode != (u8) opcode || old_req)
+			goto unlock_done;
+		qp->s_tail_ack_queue = prev;
+		break;
+	}
+
+	default:
+		/*
+		 * Ignore this operation if it doesn't request an ACK
+		 * or an earlier RDMA read or atomic is going to be resent.
+		 */
+		if (!(psn & IB_BTH_REQ_ACK) || old_req)
+			goto unlock_done;
+		/*
+		 * Resend the most recent ACK if this request is
+		 * after all the previous RDMA reads and atomics.
+		 */
+		if (i == qp->r_head_ack_queue) {
+			spin_unlock_irqrestore(&qp->s_lock, flags);
+			qp->r_nak_state = 0;
+			qp->r_ack_psn = qp->r_psn - 1;
+			goto send_ack;
+		}
+		/*
+		 * Try to send a simple ACK to work around a Mellanox bug
+		 * which doesn't accept a RDMA read response or atomic
+		 * response as an ACK for earlier SENDs or RDMA writes.
+		 */
+		if (!(qp->s_flags & QIB_S_RESP_PENDING)) {
+			spin_unlock_irqrestore(&qp->s_lock, flags);
+			qp->r_nak_state = 0;
+			qp->r_ack_psn = qp->s_ack_queue[i].psn - 1;
+			goto send_ack;
+		}
+		/*
+		 * Resend the RDMA read or atomic op which
+		 * ACKs this duplicate request.
+		 */
+		qp->s_tail_ack_queue = i;
+		break;
+	}
+	qp->s_ack_state = OP(ACKNOWLEDGE);
+	qp->s_flags |= QIB_S_RESP_PENDING;
+	qp->r_nak_state = 0;
+	qib_schedule_send(qp);
+
+unlock_done:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+done:
+	return 1;
+
+send_ack:
+	return 0;
+}
+
+void qib_rc_error(struct qib_qp *qp, enum ib_wc_status err)
+{
+	unsigned long flags;
+	int lastwqe;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+	lastwqe = qib_error_qp(qp, err);
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+
+	if (lastwqe) {
+		struct ib_event ev;
+
+		ev.device = qp->ibqp.device;
+		ev.element.qp = &qp->ibqp;
+		ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
+		qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+	}
+}
+
+static inline void qib_update_ack_queue(struct qib_qp *qp, unsigned n)
+{
+	unsigned next;
+
+	next = n + 1;
+	if (next > QIB_MAX_RDMA_ATOMIC)
+		next = 0;
+	qp->s_tail_ack_queue = next;
+	qp->s_ack_state = OP(ACKNOWLEDGE);
+}
+
+/**
+ * qib_rc_rcv - process an incoming RC packet
+ * @rcd: the context pointer
+ * @hdr: the header of this packet
+ * @has_grh: true if the header has a GRH
+ * @data: the packet data
+ * @tlen: the packet length
+ * @qp: the QP for this packet
+ *
+ * This is called from qib_qp_rcv() to process an incoming RC packet
+ * for the given QP.
+ * Called at interrupt level.
+ */
+void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp)
+{
+	struct qib_ibport *ibp = &rcd->ppd->ibport_data;
+	struct qib_other_headers *ohdr;
+	u32 opcode;
+	u32 hdrsize;
+	u32 psn;
+	u32 pad;
+	struct ib_wc wc;
+	u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
+	int diff;
+	struct ib_reth *reth;
+	unsigned long flags;
+	int ret;
+
+	/* Check for GRH */
+	if (!has_grh) {
+		ohdr = &hdr->u.oth;
+		hdrsize = 8 + 12;       /* LRH + BTH */
+	} else {
+		ohdr = &hdr->u.l.oth;
+		hdrsize = 8 + 40 + 12;  /* LRH + GRH + BTH */
+	}
+
+	opcode = be32_to_cpu(ohdr->bth[0]);
+	spin_lock_irqsave(&qp->s_lock, flags);
+	if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode))
+		goto sunlock;
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+
+	psn = be32_to_cpu(ohdr->bth[2]);
+	opcode >>= 24;
+
+	/* Prevent simultaneous processing after APM on different CPUs */
+	spin_lock(&qp->r_lock);
+
+	/*
+	 * Process responses (ACKs) before anything else.  Note that the
+	 * packet sequence number will be for something in the send work
+	 * queue rather than the expected receive packet sequence number.
+	 * In other words, this QP is the requester.
+	 */
+	if (opcode >= OP(RDMA_READ_RESPONSE_FIRST) &&
+	    opcode <= OP(ATOMIC_ACKNOWLEDGE)) {
+		qib_rc_rcv_resp(ibp, ohdr, data, tlen, qp, opcode, psn,
+				hdrsize, pmtu, rcd);
+		goto runlock;
+	}
+
+	/* Compute 24 bits worth of difference. */
+	diff = qib_cmp24(psn, qp->r_psn);
+	if (unlikely(diff)) {
+		if (qib_rc_rcv_error(ohdr, data, qp, opcode, psn, diff, rcd))
+			goto runlock;
+		goto send_ack;
+	}
+
+	/* Check for opcode sequence errors. */
+	switch (qp->r_state) {
+	case OP(SEND_FIRST):
+	case OP(SEND_MIDDLE):
+		if (opcode == OP(SEND_MIDDLE) ||
+		    opcode == OP(SEND_LAST) ||
+		    opcode == OP(SEND_LAST_WITH_IMMEDIATE))
+			break;
+		goto nack_inv;
+
+	case OP(RDMA_WRITE_FIRST):
+	case OP(RDMA_WRITE_MIDDLE):
+		if (opcode == OP(RDMA_WRITE_MIDDLE) ||
+		    opcode == OP(RDMA_WRITE_LAST) ||
+		    opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
+			break;
+		goto nack_inv;
+
+	default:
+		if (opcode == OP(SEND_MIDDLE) ||
+		    opcode == OP(SEND_LAST) ||
+		    opcode == OP(SEND_LAST_WITH_IMMEDIATE) ||
+		    opcode == OP(RDMA_WRITE_MIDDLE) ||
+		    opcode == OP(RDMA_WRITE_LAST) ||
+		    opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
+			goto nack_inv;
+		/*
+		 * Note that it is up to the requester to not send a new
+		 * RDMA read or atomic operation before receiving an ACK
+		 * for the previous operation.
+		 */
+		break;
+	}
+
+	memset(&wc, 0, sizeof wc);
+
+	if (qp->state == IB_QPS_RTR && !(qp->r_flags & QIB_R_COMM_EST)) {
+		qp->r_flags |= QIB_R_COMM_EST;
+		if (qp->ibqp.event_handler) {
+			struct ib_event ev;
+
+			ev.device = qp->ibqp.device;
+			ev.element.qp = &qp->ibqp;
+			ev.event = IB_EVENT_COMM_EST;
+			qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+		}
+	}
+
+	/* OK, process the packet. */
+	switch (opcode) {
+	case OP(SEND_FIRST):
+		ret = qib_get_rwqe(qp, 0);
+		if (ret < 0)
+			goto nack_op_err;
+		if (!ret)
+			goto rnr_nak;
+		qp->r_rcv_len = 0;
+		/* FALLTHROUGH */
+	case OP(SEND_MIDDLE):
+	case OP(RDMA_WRITE_MIDDLE):
+send_middle:
+		/* Check for invalid length PMTU or posted rwqe len. */
+		if (unlikely(tlen != (hdrsize + pmtu + 4)))
+			goto nack_inv;
+		qp->r_rcv_len += pmtu;
+		if (unlikely(qp->r_rcv_len > qp->r_len))
+			goto nack_inv;
+		qib_copy_sge(&qp->r_sge, data, pmtu, 1);
+		break;
+
+	case OP(RDMA_WRITE_LAST_WITH_IMMEDIATE):
+		/* consume RWQE */
+		ret = qib_get_rwqe(qp, 1);
+		if (ret < 0)
+			goto nack_op_err;
+		if (!ret)
+			goto rnr_nak;
+		goto send_last_imm;
+
+	case OP(SEND_ONLY):
+	case OP(SEND_ONLY_WITH_IMMEDIATE):
+		ret = qib_get_rwqe(qp, 0);
+		if (ret < 0)
+			goto nack_op_err;
+		if (!ret)
+			goto rnr_nak;
+		qp->r_rcv_len = 0;
+		if (opcode == OP(SEND_ONLY))
+			goto send_last;
+		/* FALLTHROUGH */
+	case OP(SEND_LAST_WITH_IMMEDIATE):
+send_last_imm:
+		wc.ex.imm_data = ohdr->u.imm_data;
+		hdrsize += 4;
+		wc.wc_flags = IB_WC_WITH_IMM;
+		/* FALLTHROUGH */
+	case OP(SEND_LAST):
+	case OP(RDMA_WRITE_LAST):
+send_last:
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/* Check for invalid length. */
+		/* XXX LAST len should be >= 1 */
+		if (unlikely(tlen < (hdrsize + pad + 4)))
+			goto nack_inv;
+		/* Don't count the CRC. */
+		tlen -= (hdrsize + pad + 4);
+		wc.byte_len = tlen + qp->r_rcv_len;
+		if (unlikely(wc.byte_len > qp->r_len))
+			goto nack_inv;
+		qib_copy_sge(&qp->r_sge, data, tlen, 1);
+		while (qp->r_sge.num_sge) {
+			atomic_dec(&qp->r_sge.sge.mr->refcount);
+			if (--qp->r_sge.num_sge)
+				qp->r_sge.sge = *qp->r_sge.sg_list++;
+		}
+		qp->r_msn++;
+		if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
+			break;
+		wc.wr_id = qp->r_wr_id;
+		wc.status = IB_WC_SUCCESS;
+		if (opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE) ||
+		    opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
+			wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
+		else
+			wc.opcode = IB_WC_RECV;
+		wc.qp = &qp->ibqp;
+		wc.src_qp = qp->remote_qpn;
+		wc.slid = qp->remote_ah_attr.dlid;
+		wc.sl = qp->remote_ah_attr.sl;
+		/* Signal completion event if the solicited bit is set. */
+		qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+			     (ohdr->bth[0] &
+			      cpu_to_be32(IB_BTH_SOLICITED)) != 0);
+		break;
+
+	case OP(RDMA_WRITE_FIRST):
+	case OP(RDMA_WRITE_ONLY):
+	case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE):
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_WRITE)))
+			goto nack_inv;
+		/* consume RWQE */
+		reth = &ohdr->u.rc.reth;
+		hdrsize += sizeof(*reth);
+		qp->r_len = be32_to_cpu(reth->length);
+		qp->r_rcv_len = 0;
+		qp->r_sge.sg_list = NULL;
+		if (qp->r_len != 0) {
+			u32 rkey = be32_to_cpu(reth->rkey);
+			u64 vaddr = be64_to_cpu(reth->vaddr);
+			int ok;
+
+			/* Check rkey & NAK */
+			ok = qib_rkey_ok(qp, &qp->r_sge.sge, qp->r_len, vaddr,
+					 rkey, IB_ACCESS_REMOTE_WRITE);
+			if (unlikely(!ok))
+				goto nack_acc;
+			qp->r_sge.num_sge = 1;
+		} else {
+			qp->r_sge.num_sge = 0;
+			qp->r_sge.sge.mr = NULL;
+			qp->r_sge.sge.vaddr = NULL;
+			qp->r_sge.sge.length = 0;
+			qp->r_sge.sge.sge_length = 0;
+		}
+		if (opcode == OP(RDMA_WRITE_FIRST))
+			goto send_middle;
+		else if (opcode == OP(RDMA_WRITE_ONLY))
+			goto send_last;
+		ret = qib_get_rwqe(qp, 1);
+		if (ret < 0)
+			goto nack_op_err;
+		if (!ret)
+			goto rnr_nak;
+		goto send_last_imm;
+
+	case OP(RDMA_READ_REQUEST): {
+		struct qib_ack_entry *e;
+		u32 len;
+		u8 next;
+
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ)))
+			goto nack_inv;
+		next = qp->r_head_ack_queue + 1;
+		/* s_ack_queue is size QIB_MAX_RDMA_ATOMIC+1 so use > not >= */
+		if (next > QIB_MAX_RDMA_ATOMIC)
+			next = 0;
+		spin_lock_irqsave(&qp->s_lock, flags);
+		/* Double check we can process this while holding the s_lock. */
+		if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK))
+			goto srunlock;
+		if (unlikely(next == qp->s_tail_ack_queue)) {
+			if (!qp->s_ack_queue[next].sent)
+				goto nack_inv_unlck;
+			qib_update_ack_queue(qp, next);
+		}
+		e = &qp->s_ack_queue[qp->r_head_ack_queue];
+		if (e->opcode == OP(RDMA_READ_REQUEST) && e->rdma_sge.mr) {
+			atomic_dec(&e->rdma_sge.mr->refcount);
+			e->rdma_sge.mr = NULL;
+		}
+		reth = &ohdr->u.rc.reth;
+		len = be32_to_cpu(reth->length);
+		if (len) {
+			u32 rkey = be32_to_cpu(reth->rkey);
+			u64 vaddr = be64_to_cpu(reth->vaddr);
+			int ok;
+
+			/* Check rkey & NAK */
+			ok = qib_rkey_ok(qp, &e->rdma_sge, len, vaddr,
+					 rkey, IB_ACCESS_REMOTE_READ);
+			if (unlikely(!ok))
+				goto nack_acc_unlck;
+			/*
+			 * Update the next expected PSN.  We add 1 later
+			 * below, so only add the remainder here.
+			 */
+			if (len > pmtu)
+				qp->r_psn += (len - 1) / pmtu;
+		} else {
+			e->rdma_sge.mr = NULL;
+			e->rdma_sge.vaddr = NULL;
+			e->rdma_sge.length = 0;
+			e->rdma_sge.sge_length = 0;
+		}
+		e->opcode = opcode;
+		e->sent = 0;
+		e->psn = psn;
+		e->lpsn = qp->r_psn;
+		/*
+		 * We need to increment the MSN here instead of when we
+		 * finish sending the result since a duplicate request would
+		 * increment it more than once.
+		 */
+		qp->r_msn++;
+		qp->r_psn++;
+		qp->r_state = opcode;
+		qp->r_nak_state = 0;
+		qp->r_head_ack_queue = next;
+
+		/* Schedule the send tasklet. */
+		qp->s_flags |= QIB_S_RESP_PENDING;
+		qib_schedule_send(qp);
+
+		goto srunlock;
+	}
+
+	case OP(COMPARE_SWAP):
+	case OP(FETCH_ADD): {
+		struct ib_atomic_eth *ateth;
+		struct qib_ack_entry *e;
+		u64 vaddr;
+		atomic64_t *maddr;
+		u64 sdata;
+		u32 rkey;
+		u8 next;
+
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC)))
+			goto nack_inv;
+		next = qp->r_head_ack_queue + 1;
+		if (next > QIB_MAX_RDMA_ATOMIC)
+			next = 0;
+		spin_lock_irqsave(&qp->s_lock, flags);
+		/* Double check we can process this while holding the s_lock. */
+		if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK))
+			goto srunlock;
+		if (unlikely(next == qp->s_tail_ack_queue)) {
+			if (!qp->s_ack_queue[next].sent)
+				goto nack_inv_unlck;
+			qib_update_ack_queue(qp, next);
+		}
+		e = &qp->s_ack_queue[qp->r_head_ack_queue];
+		if (e->opcode == OP(RDMA_READ_REQUEST) && e->rdma_sge.mr) {
+			atomic_dec(&e->rdma_sge.mr->refcount);
+			e->rdma_sge.mr = NULL;
+		}
+		ateth = &ohdr->u.atomic_eth;
+		vaddr = ((u64) be32_to_cpu(ateth->vaddr[0]) << 32) |
+			be32_to_cpu(ateth->vaddr[1]);
+		if (unlikely(vaddr & (sizeof(u64) - 1)))
+			goto nack_inv_unlck;
+		rkey = be32_to_cpu(ateth->rkey);
+		/* Check rkey & NAK */
+		if (unlikely(!qib_rkey_ok(qp, &qp->r_sge.sge, sizeof(u64),
+					  vaddr, rkey,
+					  IB_ACCESS_REMOTE_ATOMIC)))
+			goto nack_acc_unlck;
+		/* Perform atomic OP and save result. */
+		maddr = (atomic64_t *) qp->r_sge.sge.vaddr;
+		sdata = be64_to_cpu(ateth->swap_data);
+		e->atomic_data = (opcode == OP(FETCH_ADD)) ?
+			(u64) atomic64_add_return(sdata, maddr) - sdata :
+			(u64) cmpxchg((u64 *) qp->r_sge.sge.vaddr,
+				      be64_to_cpu(ateth->compare_data),
+				      sdata);
+		atomic_dec(&qp->r_sge.sge.mr->refcount);
+		qp->r_sge.num_sge = 0;
+		e->opcode = opcode;
+		e->sent = 0;
+		e->psn = psn;
+		e->lpsn = psn;
+		qp->r_msn++;
+		qp->r_psn++;
+		qp->r_state = opcode;
+		qp->r_nak_state = 0;
+		qp->r_head_ack_queue = next;
+
+		/* Schedule the send tasklet. */
+		qp->s_flags |= QIB_S_RESP_PENDING;
+		qib_schedule_send(qp);
+
+		goto srunlock;
+	}
+
+	default:
+		/* NAK unknown opcodes. */
+		goto nack_inv;
+	}
+	qp->r_psn++;
+	qp->r_state = opcode;
+	qp->r_ack_psn = psn;
+	qp->r_nak_state = 0;
+	/* Send an ACK if requested or required. */
+	if (psn & (1 << 31))
+		goto send_ack;
+	goto runlock;
+
+rnr_nak:
+	qp->r_nak_state = IB_RNR_NAK | qp->r_min_rnr_timer;
+	qp->r_ack_psn = qp->r_psn;
+	/* Queue RNR NAK for later */
+	if (list_empty(&qp->rspwait)) {
+		qp->r_flags |= QIB_R_RSP_NAK;
+		atomic_inc(&qp->refcount);
+		list_add_tail(&qp->rspwait, &rcd->qp_wait_list);
+	}
+	goto runlock;
+
+nack_op_err:
+	qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
+	qp->r_nak_state = IB_NAK_REMOTE_OPERATIONAL_ERROR;
+	qp->r_ack_psn = qp->r_psn;
+	/* Queue NAK for later */
+	if (list_empty(&qp->rspwait)) {
+		qp->r_flags |= QIB_R_RSP_NAK;
+		atomic_inc(&qp->refcount);
+		list_add_tail(&qp->rspwait, &rcd->qp_wait_list);
+	}
+	goto runlock;
+
+nack_inv_unlck:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+nack_inv:
+	qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
+	qp->r_nak_state = IB_NAK_INVALID_REQUEST;
+	qp->r_ack_psn = qp->r_psn;
+	/* Queue NAK for later */
+	if (list_empty(&qp->rspwait)) {
+		qp->r_flags |= QIB_R_RSP_NAK;
+		atomic_inc(&qp->refcount);
+		list_add_tail(&qp->rspwait, &rcd->qp_wait_list);
+	}
+	goto runlock;
+
+nack_acc_unlck:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+nack_acc:
+	qib_rc_error(qp, IB_WC_LOC_PROT_ERR);
+	qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR;
+	qp->r_ack_psn = qp->r_psn;
+send_ack:
+	qib_send_rc_ack(qp);
+runlock:
+	spin_unlock(&qp->r_lock);
+	return;
+
+srunlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+	spin_unlock(&qp->r_lock);
+	return;
+
+sunlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 33/51] IB/qib: Add qib_ruc.c

[Ralph Campbell]

creates the qib_ruc.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_ruc.c |  817 +++++++++++++++++++++++++++++++++++
 1 files changed, 817 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_ruc.c

diff --git a/drivers/infiniband/hw/qib/qib_ruc.c b/drivers/infiniband/hw/qib/qib_ruc.c
new file mode 100644
index 0000000..eb78d93
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_ruc.c
@@ -0,0 +1,817 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/spinlock.h>
+
+#include "qib.h"
+#include "qib_mad.h"
+
+/*
+ * Convert the AETH RNR timeout code into the number of microseconds.
+ */
+const u32 ib_qib_rnr_table[32] = {
+	655360,	/* 00: 655.36 */
+	10,	/* 01:    .01 */
+	20,	/* 02     .02 */
+	30,	/* 03:    .03 */
+	40,	/* 04:    .04 */
+	60,	/* 05:    .06 */
+	80,	/* 06:    .08 */
+	120,	/* 07:    .12 */
+	160,	/* 08:    .16 */
+	240,	/* 09:    .24 */
+	320,	/* 0A:    .32 */
+	480,	/* 0B:    .48 */
+	640,	/* 0C:    .64 */
+	960,	/* 0D:    .96 */
+	1280,	/* 0E:   1.28 */
+	1920,	/* 0F:   1.92 */
+	2560,	/* 10:   2.56 */
+	3840,	/* 11:   3.84 */
+	5120,	/* 12:   5.12 */
+	7680,	/* 13:   7.68 */
+	10240,	/* 14:  10.24 */
+	15360,	/* 15:  15.36 */
+	20480,	/* 16:  20.48 */
+	30720,	/* 17:  30.72 */
+	40960,	/* 18:  40.96 */
+	61440,	/* 19:  61.44 */
+	81920,	/* 1A:  81.92 */
+	122880,	/* 1B: 122.88 */
+	163840,	/* 1C: 163.84 */
+	245760,	/* 1D: 245.76 */
+	327680,	/* 1E: 327.68 */
+	491520	/* 1F: 491.52 */
+};
+
+/*
+ * Validate a RWQE and fill in the SGE state.
+ * Return 1 if OK.
+ */
+static int qib_init_sge(struct qib_qp *qp, struct qib_rwqe *wqe)
+{
+	int i, j, ret;
+	struct ib_wc wc;
+	struct qib_lkey_table *rkt;
+	struct qib_pd *pd;
+	struct qib_sge_state *ss;
+
+	rkt = &to_idev(qp->ibqp.device)->lk_table;
+	pd = to_ipd(qp->ibqp.srq ? qp->ibqp.srq->pd : qp->ibqp.pd);
+	ss = &qp->r_sge;
+	ss->sg_list = qp->r_sg_list;
+	qp->r_len = 0;
+	for (i = j = 0; i < wqe->num_sge; i++) {
+		if (wqe->sg_list[i].length == 0)
+			continue;
+		/* Check LKEY */
+		if (!qib_lkey_ok(rkt, pd, j ? &ss->sg_list[j - 1] : &ss->sge,
+				 &wqe->sg_list[i], IB_ACCESS_LOCAL_WRITE))
+			goto bad_lkey;
+		qp->r_len += wqe->sg_list[i].length;
+		j++;
+	}
+	ss->num_sge = j;
+	ss->total_len = qp->r_len;
+	ret = 1;
+	goto bail;
+
+bad_lkey:
+	while (j) {
+		struct qib_sge *sge = --j ? &ss->sg_list[j - 1] : &ss->sge;
+
+		atomic_dec(&sge->mr->refcount);
+	}
+	ss->num_sge = 0;
+	memset(&wc, 0, sizeof(wc));
+	wc.wr_id = wqe->wr_id;
+	wc.status = IB_WC_LOC_PROT_ERR;
+	wc.opcode = IB_WC_RECV;
+	wc.qp = &qp->ibqp;
+	/* Signal solicited completion event. */
+	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1);
+	ret = 0;
+bail:
+	return ret;
+}
+
+/**
+ * qib_get_rwqe - copy the next RWQE into the QP's RWQE
+ * @qp: the QP
+ * @wr_id_only: update qp->r_wr_id only, not qp->r_sge
+ *
+ * Return -1 if there is a local error, 0 if no RWQE is available,
+ * otherwise return 1.
+ *
+ * Can be called from interrupt level.
+ */
+int qib_get_rwqe(struct qib_qp *qp, int wr_id_only)
+{
+	unsigned long flags;
+	struct qib_rq *rq;
+	struct qib_rwq *wq;
+	struct qib_srq *srq;
+	struct qib_rwqe *wqe;
+	void (*handler)(struct ib_event *, void *);
+	u32 tail;
+	int ret;
+
+	if (qp->ibqp.srq) {
+		srq = to_isrq(qp->ibqp.srq);
+		handler = srq->ibsrq.event_handler;
+		rq = &srq->rq;
+	} else {
+		srq = NULL;
+		handler = NULL;
+		rq = &qp->r_rq;
+	}
+
+	spin_lock_irqsave(&rq->lock, flags);
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)) {
+		ret = 0;
+		goto unlock;
+	}
+
+	wq = rq->wq;
+	tail = wq->tail;
+	/* Validate tail before using it since it is user writable. */
+	if (tail >= rq->size)
+		tail = 0;
+	if (unlikely(tail == wq->head)) {
+		ret = 0;
+		goto unlock;
+	}
+	/* Make sure entry is read after head index is read. */
+	smp_rmb();
+	wqe = get_rwqe_ptr(rq, tail);
+	/*
+	 * Even though we update the tail index in memory, the verbs
+	 * consumer is not supposed to post more entries until a
+	 * completion is generated.
+	 */
+	if (++tail >= rq->size)
+		tail = 0;
+	wq->tail = tail;
+	if (!wr_id_only && !qib_init_sge(qp, wqe)) {
+		ret = -1;
+		goto unlock;
+	}
+	qp->r_wr_id = wqe->wr_id;
+
+	ret = 1;
+	set_bit(QIB_R_WRID_VALID, &qp->r_aflags);
+	if (handler) {
+		u32 n;
+
+		/*
+		 * Validate head pointer value and compute
+		 * the number of remaining WQEs.
+		 */
+		n = wq->head;
+		if (n >= rq->size)
+			n = 0;
+		if (n < tail)
+			n += rq->size - tail;
+		else
+			n -= tail;
+		if (n < srq->limit) {
+			struct ib_event ev;
+
+			srq->limit = 0;
+			spin_unlock_irqrestore(&rq->lock, flags);
+			ev.device = qp->ibqp.device;
+			ev.element.srq = qp->ibqp.srq;
+			ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
+			handler(&ev, srq->ibsrq.srq_context);
+			goto bail;
+		}
+	}
+unlock:
+	spin_unlock_irqrestore(&rq->lock, flags);
+bail:
+	return ret;
+}
+
+/*
+ * Switch to alternate path.
+ * The QP s_lock should be held and interrupts disabled.
+ */
+void qib_migrate_qp(struct qib_qp *qp)
+{
+	struct ib_event ev;
+
+	qp->s_mig_state = IB_MIG_MIGRATED;
+	qp->remote_ah_attr = qp->alt_ah_attr;
+	qp->port_num = qp->alt_ah_attr.port_num;
+	qp->s_pkey_index = qp->s_alt_pkey_index;
+
+	ev.device = qp->ibqp.device;
+	ev.element.qp = &qp->ibqp;
+	ev.event = IB_EVENT_PATH_MIG;
+	qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+}
+
+static __be64 get_sguid(struct qib_ibport *ibp, unsigned index)
+{
+	if (!index) {
+		struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+
+		return ppd->guid;
+	} else
+		return ibp->guids[index - 1];
+}
+
+static int gid_ok(union ib_gid *gid, __be64 gid_prefix, __be64 id)
+{
+	return (gid->global.interface_id == id &&
+		(gid->global.subnet_prefix == gid_prefix ||
+		 gid->global.subnet_prefix == IB_DEFAULT_GID_PREFIX));
+}
+
+/*
+ *
+ * This should be called with the QP s_lock held.
+ */
+int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		      int has_grh, struct qib_qp *qp, u32 bth0)
+{
+	__be64 guid;
+
+	if (qp->s_mig_state == IB_MIG_ARMED && (bth0 & IB_BTH_MIG_REQ)) {
+		if (!has_grh) {
+			if (qp->alt_ah_attr.ah_flags & IB_AH_GRH)
+				goto err;
+		} else {
+			if (!(qp->alt_ah_attr.ah_flags & IB_AH_GRH))
+				goto err;
+			guid = get_sguid(ibp, qp->alt_ah_attr.grh.sgid_index);
+			if (!gid_ok(&hdr->u.l.grh.dgid, ibp->gid_prefix, guid))
+				goto err;
+			if (!gid_ok(&hdr->u.l.grh.sgid,
+			    qp->alt_ah_attr.grh.dgid.global.subnet_prefix,
+			    qp->alt_ah_attr.grh.dgid.global.interface_id))
+				goto err;
+		}
+		if (!qib_pkey_ok((u16)bth0,
+				 qib_get_pkey(ibp, qp->s_alt_pkey_index))) {
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY,
+				      (u16)bth0,
+				      (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
+				      0, qp->ibqp.qp_num,
+				      hdr->lrh[3], hdr->lrh[1]);
+			goto err;
+		}
+		/* Validate the SLID. See Ch. 9.6.1.5 and 17.2.8 */
+		if (be16_to_cpu(hdr->lrh[3]) != qp->alt_ah_attr.dlid ||
+		    ppd_from_ibp(ibp)->port != qp->alt_ah_attr.port_num)
+			goto err;
+		qib_migrate_qp(qp);
+	} else {
+		if (!has_grh) {
+			if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
+				goto err;
+		} else {
+			if (!(qp->remote_ah_attr.ah_flags & IB_AH_GRH))
+				goto err;
+			guid = get_sguid(ibp,
+					 qp->remote_ah_attr.grh.sgid_index);
+			if (!gid_ok(&hdr->u.l.grh.dgid, ibp->gid_prefix, guid))
+				goto err;
+			if (!gid_ok(&hdr->u.l.grh.sgid,
+			    qp->remote_ah_attr.grh.dgid.global.subnet_prefix,
+			    qp->remote_ah_attr.grh.dgid.global.interface_id))
+				goto err;
+		}
+		if (!qib_pkey_ok((u16)bth0,
+				 qib_get_pkey(ibp, qp->s_pkey_index))) {
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY,
+				      (u16)bth0,
+				      (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
+				      0, qp->ibqp.qp_num,
+				      hdr->lrh[3], hdr->lrh[1]);
+			goto err;
+		}
+		/* Validate the SLID. See Ch. 9.6.1.5 */
+		if (be16_to_cpu(hdr->lrh[3]) != qp->remote_ah_attr.dlid ||
+		    ppd_from_ibp(ibp)->port != qp->port_num)
+			goto err;
+		if (qp->s_mig_state == IB_MIG_REARM &&
+		    !(bth0 & IB_BTH_MIG_REQ))
+			qp->s_mig_state = IB_MIG_ARMED;
+	}
+
+	return 0;
+
+err:
+	return 1;
+}
+
+/**
+ * qib_ruc_loopback - handle UC and RC lookback requests
+ * @sqp: the sending QP
+ *
+ * This is called from qib_do_send() to
+ * forward a WQE addressed to the same HCA.
+ * Note that although we are single threaded due to the tasklet, we still
+ * have to protect against post_send().  We don't have to worry about
+ * receive interrupts since this is a connected protocol and all packets
+ * will pass through here.
+ */
+static void qib_ruc_loopback(struct qib_qp *sqp)
+{
+	struct qib_ibport *ibp = to_iport(sqp->ibqp.device, sqp->port_num);
+	struct qib_qp *qp;
+	struct qib_swqe *wqe;
+	struct qib_sge *sge;
+	unsigned long flags;
+	struct ib_wc wc;
+	u64 sdata;
+	atomic64_t *maddr;
+	enum ib_wc_status send_status;
+	int release;
+	int ret;
+
+	/*
+	 * Note that we check the responder QP state after
+	 * checking the requester's state.
+	 */
+	qp = qib_lookup_qpn(ibp, sqp->remote_qpn);
+
+	spin_lock_irqsave(&sqp->s_lock, flags);
+
+	/* Return if we are already busy processing a work request. */
+	if ((sqp->s_flags & (QIB_S_BUSY | QIB_S_ANY_WAIT)) ||
+	    !(ib_qib_state_ops[sqp->state] & QIB_PROCESS_OR_FLUSH_SEND))
+		goto unlock;
+
+	sqp->s_flags |= QIB_S_BUSY;
+
+again:
+	if (sqp->s_last == sqp->s_head)
+		goto clr_busy;
+	wqe = get_swqe_ptr(sqp, sqp->s_last);
+
+	/* Return if it is not OK to start a new work reqeust. */
+	if (!(ib_qib_state_ops[sqp->state] & QIB_PROCESS_NEXT_SEND_OK)) {
+		if (!(ib_qib_state_ops[sqp->state] & QIB_FLUSH_SEND))
+			goto clr_busy;
+		/* We are in the error state, flush the work request. */
+		send_status = IB_WC_WR_FLUSH_ERR;
+		goto flush_send;
+	}
+
+	/*
+	 * We can rely on the entry not changing without the s_lock
+	 * being held until we update s_last.
+	 * We increment s_cur to indicate s_last is in progress.
+	 */
+	if (sqp->s_last == sqp->s_cur) {
+		if (++sqp->s_cur >= sqp->s_size)
+			sqp->s_cur = 0;
+	}
+	spin_unlock_irqrestore(&sqp->s_lock, flags);
+
+	if (!qp || !(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK) ||
+	    qp->ibqp.qp_type != sqp->ibqp.qp_type) {
+		ibp->n_pkt_drops++;
+		/*
+		 * For RC, the requester would timeout and retry so
+		 * shortcut the timeouts and just signal too many retries.
+		 */
+		if (sqp->ibqp.qp_type == IB_QPT_RC)
+			send_status = IB_WC_RETRY_EXC_ERR;
+		else
+			send_status = IB_WC_SUCCESS;
+		goto serr;
+	}
+
+	memset(&wc, 0, sizeof wc);
+	send_status = IB_WC_SUCCESS;
+
+	release = 1;
+	sqp->s_sge.sge = wqe->sg_list[0];
+	sqp->s_sge.sg_list = wqe->sg_list + 1;
+	sqp->s_sge.num_sge = wqe->wr.num_sge;
+	sqp->s_len = wqe->length;
+	switch (wqe->wr.opcode) {
+	case IB_WR_SEND_WITH_IMM:
+		wc.wc_flags = IB_WC_WITH_IMM;
+		wc.ex.imm_data = wqe->wr.ex.imm_data;
+		/* FALLTHROUGH */
+	case IB_WR_SEND:
+		ret = qib_get_rwqe(qp, 0);
+		if (ret < 0)
+			goto op_err;
+		if (!ret)
+			goto rnr_nak;
+		break;
+
+	case IB_WR_RDMA_WRITE_WITH_IMM:
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_WRITE)))
+			goto inv_err;
+		wc.wc_flags = IB_WC_WITH_IMM;
+		wc.ex.imm_data = wqe->wr.ex.imm_data;
+		ret = qib_get_rwqe(qp, 1);
+		if (ret < 0)
+			goto op_err;
+		if (!ret)
+			goto rnr_nak;
+		/* FALLTHROUGH */
+	case IB_WR_RDMA_WRITE:
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_WRITE)))
+			goto inv_err;
+		if (wqe->length == 0)
+			break;
+		if (unlikely(!qib_rkey_ok(qp, &qp->r_sge.sge, wqe->length,
+					  wqe->wr.wr.rdma.remote_addr,
+					  wqe->wr.wr.rdma.rkey,
+					  IB_ACCESS_REMOTE_WRITE)))
+			goto acc_err;
+		qp->r_sge.sg_list = NULL;
+		qp->r_sge.num_sge = 1;
+		qp->r_sge.total_len = wqe->length;
+		break;
+
+	case IB_WR_RDMA_READ:
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ)))
+			goto inv_err;
+		if (unlikely(!qib_rkey_ok(qp, &sqp->s_sge.sge, wqe->length,
+					  wqe->wr.wr.rdma.remote_addr,
+					  wqe->wr.wr.rdma.rkey,
+					  IB_ACCESS_REMOTE_READ)))
+			goto acc_err;
+		release = 0;
+		sqp->s_sge.sg_list = NULL;
+		sqp->s_sge.num_sge = 1;
+		qp->r_sge.sge = wqe->sg_list[0];
+		qp->r_sge.sg_list = wqe->sg_list + 1;
+		qp->r_sge.num_sge = wqe->wr.num_sge;
+		qp->r_sge.total_len = wqe->length;
+		break;
+
+	case IB_WR_ATOMIC_CMP_AND_SWP:
+	case IB_WR_ATOMIC_FETCH_AND_ADD:
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC)))
+			goto inv_err;
+		if (unlikely(!qib_rkey_ok(qp, &qp->r_sge.sge, sizeof(u64),
+					  wqe->wr.wr.atomic.remote_addr,
+					  wqe->wr.wr.atomic.rkey,
+					  IB_ACCESS_REMOTE_ATOMIC)))
+			goto acc_err;
+		/* Perform atomic OP and save result. */
+		maddr = (atomic64_t *) qp->r_sge.sge.vaddr;
+		sdata = wqe->wr.wr.atomic.compare_add;
+		*(u64 *) sqp->s_sge.sge.vaddr =
+			(wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) ?
+			(u64) atomic64_add_return(sdata, maddr) - sdata :
+			(u64) cmpxchg((u64 *) qp->r_sge.sge.vaddr,
+				      sdata, wqe->wr.wr.atomic.swap);
+		atomic_dec(&qp->r_sge.sge.mr->refcount);
+		qp->r_sge.num_sge = 0;
+		goto send_comp;
+
+	default:
+		send_status = IB_WC_LOC_QP_OP_ERR;
+		goto serr;
+	}
+
+	sge = &sqp->s_sge.sge;
+	while (sqp->s_len) {
+		u32 len = sqp->s_len;
+
+		if (len > sge->length)
+			len = sge->length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		qib_copy_sge(&qp->r_sge, sge->vaddr, len, release);
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (!release)
+				atomic_dec(&sge->mr->refcount);
+			if (--sqp->s_sge.num_sge)
+				*sge = *sqp->s_sge.sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+		sqp->s_len -= len;
+	}
+	if (release)
+		while (qp->r_sge.num_sge) {
+			atomic_dec(&qp->r_sge.sge.mr->refcount);
+			if (--qp->r_sge.num_sge)
+				qp->r_sge.sge = *qp->r_sge.sg_list++;
+		}
+
+	if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
+		goto send_comp;
+
+	if (wqe->wr.opcode == IB_WR_RDMA_WRITE_WITH_IMM)
+		wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
+	else
+		wc.opcode = IB_WC_RECV;
+	wc.wr_id = qp->r_wr_id;
+	wc.status = IB_WC_SUCCESS;
+	wc.byte_len = wqe->length;
+	wc.qp = &qp->ibqp;
+	wc.src_qp = qp->remote_qpn;
+	wc.slid = qp->remote_ah_attr.dlid;
+	wc.sl = qp->remote_ah_attr.sl;
+	wc.port_num = 1;
+	/* Signal completion event if the solicited bit is set. */
+	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+		       wqe->wr.send_flags & IB_SEND_SOLICITED);
+
+send_comp:
+	spin_lock_irqsave(&sqp->s_lock, flags);
+	ibp->n_loop_pkts++;
+flush_send:
+	sqp->s_rnr_retry = sqp->s_rnr_retry_cnt;
+	qib_send_complete(sqp, wqe, send_status);
+	goto again;
+
+rnr_nak:
+	/* Handle RNR NAK */
+	if (qp->ibqp.qp_type == IB_QPT_UC)
+		goto send_comp;
+	ibp->n_rnr_naks++;
+	/*
+	 * Note: we don't need the s_lock held since the BUSY flag
+	 * makes this single threaded.
+	 */
+	if (sqp->s_rnr_retry == 0) {
+		send_status = IB_WC_RNR_RETRY_EXC_ERR;
+		goto serr;
+	}
+	if (sqp->s_rnr_retry_cnt < 7)
+		sqp->s_rnr_retry--;
+	spin_lock_irqsave(&sqp->s_lock, flags);
+	if (!(ib_qib_state_ops[sqp->state] & QIB_PROCESS_RECV_OK))
+		goto clr_busy;
+	sqp->s_flags |= QIB_S_WAIT_RNR;
+	sqp->s_timer.function = qib_rc_rnr_retry;
+	sqp->s_timer.expires = jiffies +
+		usecs_to_jiffies(ib_qib_rnr_table[qp->r_min_rnr_timer]);
+	add_timer(&sqp->s_timer);
+	goto clr_busy;
+
+op_err:
+	send_status = IB_WC_REM_OP_ERR;
+	wc.status = IB_WC_LOC_QP_OP_ERR;
+	goto err;
+
+inv_err:
+	send_status = IB_WC_REM_INV_REQ_ERR;
+	wc.status = IB_WC_LOC_QP_OP_ERR;
+	goto err;
+
+acc_err:
+	send_status = IB_WC_REM_ACCESS_ERR;
+	wc.status = IB_WC_LOC_PROT_ERR;
+err:
+	/* responder goes to error state */
+	qib_rc_error(qp, wc.status);
+
+serr:
+	spin_lock_irqsave(&sqp->s_lock, flags);
+	qib_send_complete(sqp, wqe, send_status);
+	if (sqp->ibqp.qp_type == IB_QPT_RC) {
+		int lastwqe = qib_error_qp(sqp, IB_WC_WR_FLUSH_ERR);
+
+		sqp->s_flags &= ~QIB_S_BUSY;
+		spin_unlock_irqrestore(&sqp->s_lock, flags);
+		if (lastwqe) {
+			struct ib_event ev;
+
+			ev.device = sqp->ibqp.device;
+			ev.element.qp = &sqp->ibqp;
+			ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
+			sqp->ibqp.event_handler(&ev, sqp->ibqp.qp_context);
+		}
+		goto done;
+	}
+clr_busy:
+	sqp->s_flags &= ~QIB_S_BUSY;
+unlock:
+	spin_unlock_irqrestore(&sqp->s_lock, flags);
+done:
+	if (qp && atomic_dec_and_test(&qp->refcount))
+		wake_up(&qp->wait);
+}
+
+/**
+ * qib_make_grh - construct a GRH header
+ * @ibp: a pointer to the IB port
+ * @hdr: a pointer to the GRH header being constructed
+ * @grh: the global route address to send to
+ * @hwords: the number of 32 bit words of header being sent
+ * @nwords: the number of 32 bit words of data being sent
+ *
+ * Return the size of the header in 32 bit words.
+ */
+u32 qib_make_grh(struct qib_ibport *ibp, struct ib_grh *hdr,
+		 struct ib_global_route *grh, u32 hwords, u32 nwords)
+{
+	hdr->version_tclass_flow =
+		cpu_to_be32((IB_GRH_VERSION << IB_GRH_VERSION_SHIFT) |
+			    (grh->traffic_class << IB_GRH_TCLASS_SHIFT) |
+			    (grh->flow_label << IB_GRH_FLOW_SHIFT));
+	hdr->paylen = cpu_to_be16((hwords - 2 + nwords + SIZE_OF_CRC) << 2);
+	/* next_hdr is defined by C8-7 in ch. 8.4.1 */
+	hdr->next_hdr = IB_GRH_NEXT_HDR;
+	hdr->hop_limit = grh->hop_limit;
+	/* The SGID is 32-bit aligned. */
+	hdr->sgid.global.subnet_prefix = ibp->gid_prefix;
+	hdr->sgid.global.interface_id = grh->sgid_index ?
+		ibp->guids[grh->sgid_index - 1] : ppd_from_ibp(ibp)->guid;
+	hdr->dgid = grh->dgid;
+
+	/* GRH header size in 32-bit words. */
+	return sizeof(struct ib_grh) / sizeof(u32);
+}
+
+void qib_make_ruc_header(struct qib_qp *qp, struct qib_other_headers *ohdr,
+			 u32 bth0, u32 bth2)
+{
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	u16 lrh0;
+	u32 nwords;
+	u32 extra_bytes;
+
+	/* Construct the header. */
+	extra_bytes = -qp->s_cur_size & 3;
+	nwords = (qp->s_cur_size + extra_bytes) >> 2;
+	lrh0 = QIB_LRH_BTH;
+	if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
+		qp->s_hdrwords += qib_make_grh(ibp, &qp->s_hdr.u.l.grh,
+					       &qp->remote_ah_attr.grh,
+					       qp->s_hdrwords, nwords);
+		lrh0 = QIB_LRH_GRH;
+	}
+	lrh0 |= ibp->sl_to_vl[qp->remote_ah_attr.sl] << 12 |
+		qp->remote_ah_attr.sl << 4;
+	qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
+	qp->s_hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
+	qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
+	qp->s_hdr.lrh[3] = cpu_to_be16(ppd_from_ibp(ibp)->lid |
+				       qp->remote_ah_attr.src_path_bits);
+	bth0 |= qib_get_pkey(ibp, qp->s_pkey_index);
+	bth0 |= extra_bytes << 20;
+	if (qp->s_mig_state == IB_MIG_MIGRATED)
+		bth0 |= IB_BTH_MIG_REQ;
+	ohdr->bth[0] = cpu_to_be32(bth0);
+	ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
+	ohdr->bth[2] = cpu_to_be32(bth2);
+}
+
+/**
+ * qib_do_send - perform a send on a QP
+ * @work: contains a pointer to the QP
+ *
+ * Process entries in the send work queue until credit or queue is
+ * exhausted.  Only allow one CPU to send a packet per QP (tasklet).
+ * Otherwise, two threads could send packets out of order.
+ */
+void qib_do_send(struct work_struct *work)
+{
+	struct qib_qp *qp = container_of(work, struct qib_qp, s_work);
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	int (*make_req)(struct qib_qp *qp);
+	unsigned long flags;
+
+	if ((qp->ibqp.qp_type == IB_QPT_RC ||
+	     qp->ibqp.qp_type == IB_QPT_UC) &&
+	    (qp->remote_ah_attr.dlid & ~((1 << ppd->lmc) - 1)) == ppd->lid) {
+		qib_ruc_loopback(qp);
+		return;
+	}
+
+	if (qp->ibqp.qp_type == IB_QPT_RC)
+		make_req = qib_make_rc_req;
+	else if (qp->ibqp.qp_type == IB_QPT_UC)
+		make_req = qib_make_uc_req;
+	else
+		make_req = qib_make_ud_req;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	/* Return if we are already busy processing a work request. */
+	if (!qib_send_ok(qp)) {
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+		return;
+	}
+
+	qp->s_flags |= QIB_S_BUSY;
+
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+
+	do {
+		/* Check for a constructed packet to be sent. */
+		if (qp->s_hdrwords != 0) {
+			/*
+			 * If the packet cannot be sent now, return and
+			 * the send tasklet will be woken up later.
+			 */
+			if (qib_verbs_send(qp, &qp->s_hdr, qp->s_hdrwords,
+					   qp->s_cur_sge, qp->s_cur_size))
+				break;
+			/* Record that s_hdr is empty. */
+			qp->s_hdrwords = 0;
+		}
+	} while (make_req(qp));
+}
+
+/*
+ * This should be called with s_lock held.
+ */
+void qib_send_complete(struct qib_qp *qp, struct qib_swqe *wqe,
+		       enum ib_wc_status status)
+{
+	u32 old_last, last;
+	unsigned i;
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_OR_FLUSH_SEND))
+		return;
+
+	for (i = 0; i < wqe->wr.num_sge; i++) {
+		struct qib_sge *sge = &wqe->sg_list[i];
+
+		atomic_dec(&sge->mr->refcount);
+	}
+	if (qp->ibqp.qp_type == IB_QPT_UD ||
+	    qp->ibqp.qp_type == IB_QPT_SMI ||
+	    qp->ibqp.qp_type == IB_QPT_GSI)
+		atomic_dec(&to_iah(wqe->wr.wr.ud.ah)->refcount);
+
+	/* See ch. 11.2.4.1 and 10.7.3.1 */
+	if (!(qp->s_flags & QIB_S_SIGNAL_REQ_WR) ||
+	    (wqe->wr.send_flags & IB_SEND_SIGNALED) ||
+	    status != IB_WC_SUCCESS) {
+		struct ib_wc wc;
+
+		memset(&wc, 0, sizeof wc);
+		wc.wr_id = wqe->wr.wr_id;
+		wc.status = status;
+		wc.opcode = ib_qib_wc_opcode[wqe->wr.opcode];
+		wc.qp = &qp->ibqp;
+		if (status == IB_WC_SUCCESS)
+			wc.byte_len = wqe->length;
+		qib_cq_enter(to_icq(qp->ibqp.send_cq), &wc,
+			     status != IB_WC_SUCCESS);
+	}
+
+	last = qp->s_last;
+	old_last = last;
+	if (++last >= qp->s_size)
+		last = 0;
+	qp->s_last = last;
+	if (qp->s_acked == old_last)
+		qp->s_acked = last;
+	if (qp->s_cur == old_last)
+		qp->s_cur = last;
+	if (qp->s_tail == old_last)
+		qp->s_tail = last;
+	if (qp->state == IB_QPS_SQD && last == qp->s_cur)
+		qp->s_draining = 0;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 34/51] IB/qib: Add qib_sd7220.c

[Ralph Campbell]

creates the qib_sd7220.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_sd7220.c | 1442 ++++++++++++++++++++++++++++++++
 1 files changed, 1442 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_sd7220.c

diff --git a/drivers/infiniband/hw/qib/qib_sd7220.c b/drivers/infiniband/hw/qib/qib_sd7220.c
new file mode 100644
index 0000000..2230a09
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_sd7220.c
@@ -0,0 +1,1442 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+/*
+ * This file contains all of the code that is specific to the SerDes
+ * on the QLogic_IB 7220 chip.
+ */
+
+#include <linux/pci.h>
+#include <linux/delay.h>
+
+#include "qib.h"
+#include "qib_7220.h"
+
+/*
+ * Same as in qib_iba7220.c, but just the registers needed here.
+ * Could move whole set to qib_7220.h, but decided better to keep
+ * local.
+ */
+#define KREG_IDX(regname) (QIB_7220_##regname##_OFFS / sizeof(u64))
+#define kr_hwerrclear KREG_IDX(HwErrClear)
+#define kr_hwerrmask KREG_IDX(HwErrMask)
+#define kr_hwerrstatus KREG_IDX(HwErrStatus)
+#define kr_ibcstatus KREG_IDX(IBCStatus)
+#define kr_ibserdesctrl KREG_IDX(IBSerDesCtrl)
+#define kr_scratch KREG_IDX(Scratch)
+#define kr_xgxs_cfg KREG_IDX(XGXSCfg)
+/* these are used only here, not in qib_iba7220.c */
+#define kr_ibsd_epb_access_ctrl KREG_IDX(ibsd_epb_access_ctrl)
+#define kr_ibsd_epb_transaction_reg KREG_IDX(ibsd_epb_transaction_reg)
+#define kr_pciesd_epb_transaction_reg KREG_IDX(pciesd_epb_transaction_reg)
+#define kr_pciesd_epb_access_ctrl KREG_IDX(pciesd_epb_access_ctrl)
+#define kr_serdes_ddsrxeq0 KREG_IDX(SerDes_DDSRXEQ0)
+
+/*
+ * The IBSerDesMappTable is a memory that holds values to be stored in
+ * various SerDes registers by IBC.
+ */
+#define kr_serdes_maptable KREG_IDX(IBSerDesMappTable)
+
+/*
+ * Below used for sdnum parameter, selecting one of the two sections
+ * used for PCIe, or the single SerDes used for IB.
+ */
+#define PCIE_SERDES0 0
+#define PCIE_SERDES1 1
+
+/*
+ * The EPB requires addressing in a particular form. EPB_LOC() is intended
+ * to make #definitions a little more readable.
+ */
+#define EPB_ADDR_SHF 8
+#define EPB_LOC(chn, elt, reg) \
+	(((elt & 0xf) | ((chn & 7) << 4) | ((reg & 0x3f) << 9)) << \
+	 EPB_ADDR_SHF)
+#define EPB_IB_QUAD0_CS_SHF (25)
+#define EPB_IB_QUAD0_CS (1U <<  EPB_IB_QUAD0_CS_SHF)
+#define EPB_IB_UC_CS_SHF (26)
+#define EPB_PCIE_UC_CS_SHF (27)
+#define EPB_GLOBAL_WR (1U << (EPB_ADDR_SHF + 8))
+
+/* Forward declarations. */
+static int qib_sd7220_reg_mod(struct qib_devdata *dd, int sdnum, u32 loc,
+			      u32 data, u32 mask);
+static int ibsd_mod_allchnls(struct qib_devdata *dd, int loc, int val,
+			     int mask);
+static int qib_sd_trimdone_poll(struct qib_devdata *dd);
+static void qib_sd_trimdone_monitor(struct qib_devdata *dd, const char *where);
+static int qib_sd_setvals(struct qib_devdata *dd);
+static int qib_sd_early(struct qib_devdata *dd);
+static int qib_sd_dactrim(struct qib_devdata *dd);
+static int qib_internal_presets(struct qib_devdata *dd);
+/* Tweak the register (CMUCTRL5) that contains the TRIMSELF controls */
+static int qib_sd_trimself(struct qib_devdata *dd, int val);
+static int epb_access(struct qib_devdata *dd, int sdnum, int claim);
+
+/*
+ * Below keeps track of whether the "once per power-on" initialization has
+ * been done, because uC code Version 1.32.17 or higher allows the uC to
+ * be reset at will, and Automatic Equalization may require it. So the
+ * state of the reset "pin", is no longer valid. Instead, we check for the
+ * actual uC code having been loaded.
+ */
+static int qib_ibsd_ucode_loaded(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	if (!dd->cspec->serdes_first_init_done && (qib_sd7220_ib_vfy(dd) > 0))
+		dd->cspec->serdes_first_init_done = 1;
+	return dd->cspec->serdes_first_init_done;
+}
+
+/* repeat #define for local use. "Real" #define is in qib_iba7220.c */
+#define QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR      0x0000004000000000ULL
+#define IB_MPREG5 (EPB_LOC(6, 0, 0xE) | (1L << EPB_IB_UC_CS_SHF))
+#define IB_MPREG6 (EPB_LOC(6, 0, 0xF) | (1U << EPB_IB_UC_CS_SHF))
+#define UC_PAR_CLR_D 8
+#define UC_PAR_CLR_M 0xC
+#define IB_CTRL2(chn) (EPB_LOC(chn, 7, 3) | EPB_IB_QUAD0_CS)
+#define START_EQ1(chan) EPB_LOC(chan, 7, 0x27)
+
+void qib_sd7220_clr_ibpar(struct qib_devdata *dd)
+{
+	int ret;
+
+	/* clear, then re-enable parity errs */
+	ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6,
+		UC_PAR_CLR_D, UC_PAR_CLR_M);
+	if (ret < 0) {
+		qib_dev_err(dd, "Failed clearing IBSerDes Parity err\n");
+		goto bail;
+	}
+	ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6, 0,
+		UC_PAR_CLR_M);
+
+	qib_read_kreg32(dd, kr_scratch);
+	udelay(4);
+	qib_write_kreg(dd, kr_hwerrclear,
+		QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR);
+	qib_read_kreg32(dd, kr_scratch);
+bail:
+	return;
+}
+
+/*
+ * After a reset or other unusual event, the epb interface may need
+ * to be re-synchronized, between the host and the uC.
+ * returns <0 for failure to resync within IBSD_RESYNC_TRIES (not expected)
+ */
+#define IBSD_RESYNC_TRIES 3
+#define IB_PGUDP(chn) (EPB_LOC((chn), 2, 1) | EPB_IB_QUAD0_CS)
+#define IB_CMUDONE(chn) (EPB_LOC((chn), 7, 0xF) | EPB_IB_QUAD0_CS)
+
+static int qib_resync_ibepb(struct qib_devdata *dd)
+{
+	int ret, pat, tries, chn;
+	u32 loc;
+
+	ret = -1;
+	chn = 0;
+	for (tries = 0; tries < (4 * IBSD_RESYNC_TRIES); ++tries) {
+		loc = IB_PGUDP(chn);
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, loc, 0, 0);
+		if (ret < 0) {
+			qib_dev_err(dd, "Failed read in resync\n");
+			continue;
+		}
+		if (ret != 0xF0 && ret != 0x55 && tries == 0)
+			qib_dev_err(dd, "unexpected pattern in resync\n");
+		pat = ret ^ 0xA5; /* alternate F0 and 55 */
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, loc, pat, 0xFF);
+		if (ret < 0) {
+			qib_dev_err(dd, "Failed write in resync\n");
+			continue;
+		}
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, loc, 0, 0);
+		if (ret < 0) {
+			qib_dev_err(dd, "Failed re-read in resync\n");
+			continue;
+		}
+		if (ret != pat) {
+			qib_dev_err(dd, "Failed compare1 in resync\n");
+			continue;
+		}
+		loc = IB_CMUDONE(chn);
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, loc, 0, 0);
+		if (ret < 0) {
+			qib_dev_err(dd, "Failed CMUDONE rd in resync\n");
+			continue;
+		}
+		if ((ret & 0x70) != ((chn << 4) | 0x40)) {
+			qib_dev_err(dd, "Bad CMUDONE value %02X, chn %d\n",
+				    ret, chn);
+			continue;
+		}
+		if (++chn == 4)
+			break;  /* Success */
+	}
+	qib_cdbg(VERBOSE, "Resync in %d tries\n", tries);
+	return (ret > 0) ? 0 : ret;
+}
+
+/*
+ * Localize the stuff that should be done to change IB uC reset
+ * returns <0 for errors.
+ */
+static int qib_ibsd_reset(struct qib_devdata *dd, int assert_rst)
+{
+	u64 rst_val;
+	int ret = 0;
+	unsigned long flags;
+
+	rst_val = qib_read_kreg64(dd, kr_ibserdesctrl);
+	if (assert_rst) {
+		/*
+		 * Vendor recommends "interrupting" uC before reset, to
+		 * minimize possible glitches.
+		 */
+		spin_lock_irqsave(&dd->cspec->sdepb_lock, flags);
+		epb_access(dd, IB_7220_SERDES, 1);
+		rst_val |= 1ULL;
+		/* Squelch possible parity error from _asserting_ reset */
+		qib_write_kreg(dd, kr_hwerrmask,
+			       dd->cspec->hwerrmask &
+			       ~QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR);
+		qib_write_kreg(dd, kr_ibserdesctrl, rst_val);
+		/* flush write, delay to ensure it took effect */
+		qib_read_kreg32(dd, kr_scratch);
+		udelay(2);
+		/* once it's reset, can remove interrupt */
+		epb_access(dd, IB_7220_SERDES, -1);
+		spin_unlock_irqrestore(&dd->cspec->sdepb_lock, flags);
+	} else {
+		/*
+		 * Before we de-assert reset, we need to deal with
+		 * possible glitch on the Parity-error line.
+		 * Suppress it around the reset, both in chip-level
+		 * hwerrmask and in IB uC control reg. uC will allow
+		 * it again during startup.
+		 */
+		u64 val;
+		rst_val &= ~(1ULL);
+		qib_write_kreg(dd, kr_hwerrmask,
+			       dd->cspec->hwerrmask &
+			       ~QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR);
+
+		ret = qib_resync_ibepb(dd);
+		if (ret < 0)
+			qib_dev_err(dd, "unable to re-sync IB EPB\n");
+
+		/* set uC control regs to suppress parity errs */
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG5, 1, 1);
+		if (ret < 0)
+			goto bail;
+		/* IB uC code past Version 1.32.17 allow suppression of wdog */
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6, 0x80,
+			0x80);
+		if (ret < 0) {
+			qib_dev_err(dd, "Failed to set WDOG disable\n");
+			goto bail;
+		}
+		qib_write_kreg(dd, kr_ibserdesctrl, rst_val);
+		/* flush write, delay for startup */
+		qib_read_kreg32(dd, kr_scratch);
+		udelay(1);
+		/* clear, then re-enable parity errs */
+		qib_sd7220_clr_ibpar(dd);
+		val = qib_read_kreg64(dd, kr_hwerrstatus);
+		if (val & QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR) {
+			qib_dev_err(dd, "IBUC Parity still set after RST\n");
+			dd->cspec->hwerrmask &=
+				~QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR;
+		}
+		qib_write_kreg(dd, kr_hwerrmask,
+			dd->cspec->hwerrmask);
+	}
+
+bail:
+	return ret;
+}
+
+static void qib_sd_trimdone_monitor(struct qib_devdata *dd,
+       const char *where)
+{
+	int ret, chn, baduns;
+	u64 val;
+
+	if (!where)
+		where = "?";
+
+	/* give time for reset to settle out in EPB */
+	udelay(2);
+
+	ret = qib_resync_ibepb(dd);
+	if (ret < 0)
+		qib_dev_err(dd, "not able to re-sync IB EPB (%s)\n", where);
+
+	/* Do "sacrificial read" to get EPB in sane state after reset */
+	ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, IB_CTRL2(0), 0, 0);
+	if (ret < 0)
+		qib_dev_err(dd, "Failed TRIMDONE 1st read, (%s)\n", where);
+
+	/* Check/show "summary" Trim-done bit in IBCStatus */
+	val = qib_read_kreg64(dd, kr_ibcstatus);
+	if (val & (1ULL << 11))
+		qib_cdbg(VERBOSE, "IBCS TRIMDONE set (%s)\n", where);
+	else
+		qib_dev_err(dd, "IBCS TRIMDONE clear (%s)\n", where);
+	/*
+	 * Do "dummy read/mod/wr" to get EPB in sane state after reset
+	 * The default value for MPREG6 is 0.
+	 */
+	udelay(2);
+
+	ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6, 0x80, 0x80);
+	if (ret < 0)
+		qib_dev_err(dd, "Failed Dummy RMW, (%s)\n", where);
+	udelay(10);
+
+	baduns = 0;
+
+	for (chn = 3; chn >= 0; --chn) {
+		/* Read CTRL reg for each channel to check TRIMDONE */
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES,
+			IB_CTRL2(chn), 0, 0);
+		if (ret < 0)
+			qib_dev_err(dd, "Failed checking TRIMDONE, chn %d"
+				    " (%s)\n", chn, where);
+
+		if (!(ret & 0x10)) {
+			int probe;
+
+			baduns |= (1 << chn);
+			qib_dev_err(dd, "TRIMDONE cleared on chn %d (%02X)."
+				" (%s)\n", chn, ret, where);
+			probe = qib_sd7220_reg_mod(dd, IB_7220_SERDES,
+				IB_PGUDP(0), 0, 0);
+			qib_dev_err(dd, "probe is %d (%02X)\n",
+				probe, probe);
+			probe = qib_sd7220_reg_mod(dd, IB_7220_SERDES,
+				IB_CTRL2(chn), 0, 0);
+			qib_dev_err(dd, "re-read: %d (%02X)\n",
+				probe, probe);
+			ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES,
+				IB_CTRL2(chn), 0x10, 0x10);
+			if (ret < 0)
+				qib_dev_err(dd,
+					"Err on TRIMDONE rewrite1\n");
+		}
+	}
+	for (chn = 3; chn >= 0; --chn) {
+		/* Read CTRL reg for each channel to check TRIMDONE */
+		if (baduns & (1 << chn)) {
+			qib_dev_err(dd,
+				"Reseting TRIMDONE on chn %d (%s)\n",
+				chn, where);
+			ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES,
+				IB_CTRL2(chn), 0x10, 0x10);
+			if (ret < 0)
+				qib_dev_err(dd, "Failed re-setting "
+					"TRIMDONE, chn %d (%s)\n",
+					chn, where);
+		}
+	}
+}
+
+/*
+ * Below is portion of IBA7220-specific bringup_serdes() that actually
+ * deals with registers and memory within the SerDes itself.
+ * Post IB uC code version 1.32.17, was_reset being 1 is not really
+ * informative, so we double-check.
+ */
+int qib_sd7220_init(struct qib_devdata *dd)
+{
+	int ret = 1; /* default to failure */
+	int first_reset, was_reset;
+
+	/* SERDES MPU reset recorded in D0 */
+	was_reset = (qib_read_kreg64(dd, kr_ibserdesctrl) & 1);
+	qib_cdbg(INIT, "IBReset %s xgxsconfig %llx\n",
+		 was_reset ? "Asserted" : "Negated", (unsigned long long)
+		 qib_read_kreg64(dd, kr_xgxs_cfg));
+	if (!was_reset) {
+		/* entered with reset not asserted, we need to do it */
+		qib_ibsd_reset(dd, 1);
+		qib_sd_trimdone_monitor(dd, "Driver-reload");
+	}
+	/* Substitute our deduced value for was_reset */
+	ret = qib_ibsd_ucode_loaded(dd->pport);
+	if (ret < 0)
+		goto bail;
+
+	first_reset = !ret; /* First reset if IBSD uCode not yet loaded */
+	/*
+	 * Alter some regs per vendor latest doc, reset-defaults
+	 * are not right for IB.
+	 */
+	ret = qib_sd_early(dd);
+	if (ret < 0) {
+		qib_dev_err(dd, "Failed to set IB SERDES early defaults\n");
+		goto bail;
+	}
+	/*
+	 * Set DAC manual trim IB.
+	 * We only do this once after chip has been reset (usually
+	 * same as once per system boot).
+	 */
+	if (first_reset) {
+		ret = qib_sd_dactrim(dd);
+		if (ret < 0) {
+			qib_dev_err(dd, "Failed IB SERDES DAC trim\n");
+			goto bail;
+		}
+	}
+	/*
+	 * Set various registers (DDS and RXEQ) that will be
+	 * controlled by IBC (in 1.2 mode) to reasonable preset values
+	 * Calling the "internal" version avoids the "check for needed"
+	 * and "trimdone monitor" that might be counter-productive.
+	 */
+	ret = qib_internal_presets(dd);
+	if (ret < 0) {
+		qib_dev_err(dd, "Failed to set IB SERDES presets\n");
+		goto bail;
+	}
+	ret = qib_sd_trimself(dd, 0x80);
+	if (ret < 0) {
+		qib_dev_err(dd, "Failed to set IB SERDES TRIMSELF\n");
+		goto bail;
+	}
+
+	/* Load image, then try to verify */
+	ret = 0;        /* Assume success */
+	if (first_reset) {
+		int vfy;
+		int trim_done;
+
+		qib_dbg("SerDes uC was reset, reloading PRAM\n");
+		ret = qib_sd7220_ib_load(dd);
+		if (ret < 0) {
+			qib_dev_err(dd, "Failed to load IB SERDES image\n");
+			goto bail;
+		} else {
+			/* Loaded image, try to verify */
+			vfy = qib_sd7220_ib_vfy(dd);
+			if (vfy != ret) {
+				qib_dev_err(dd, "SERDES PRAM VFY failed\n");
+				goto bail;
+			} /* end if verified */
+		} /* end if loaded */
+
+		/*
+		 * Loaded and verified. Almost good...
+		 * hold "success" in ret
+		 */
+		ret = 0;
+		/*
+		 * Prev steps all worked, continue bringup
+		 * De-assert RESET to uC, only in first reset, to allow
+		 * trimming.
+		 *
+		 * Since our default setup sets START_EQ1 to
+		 * PRESET, we need to clear that for this very first run.
+		 */
+		ret = ibsd_mod_allchnls(dd, START_EQ1(0), 0, 0x38);
+		if (ret < 0) {
+			qib_dev_err(dd, "Failed clearing START_EQ1\n");
+			goto bail;
+		}
+
+		qib_ibsd_reset(dd, 0);
+		/*
+		 * If this is not the first reset, trimdone should be set
+		 * already. We may need to check about this.
+		 */
+		trim_done = qib_sd_trimdone_poll(dd);
+		/*
+		 * Whether or not trimdone succeeded, we need to put the
+		 * uC back into reset to avoid a possible fight with the
+		 * IBC state-machine.
+		 */
+		qib_ibsd_reset(dd, 1);
+
+		if (!trim_done) {
+			qib_dev_err(dd, "No TRIMDONE seen\n");
+			goto bail;
+		}
+		/*
+		 * DEBUG: check each time we reset if trimdone bits have
+		 * gotten cleared, and re-set them.
+		 */
+		qib_sd_trimdone_monitor(dd, "First-reset");
+		/* Remember so we do not re-do the load, dactrim, etc. */
+		dd->cspec->serdes_first_init_done = 1;
+	}
+	/*
+	 * setup for channel training and load values for
+	 * RxEq and DDS in tables used by IBC in IB1.2 mode
+	 */
+	ret = 0;
+	if (qib_sd_setvals(dd) >= 0)
+		goto done;
+bail:
+	ret = 1;
+done:
+	/* start relock timer regardless, but start at 1 second */
+	set_7220_relock_poll(dd, -1);
+	return ret;
+}
+
+#define EPB_ACC_REQ 1
+#define EPB_ACC_GNT 0x100
+#define EPB_DATA_MASK 0xFF
+#define EPB_RD (1ULL << 24)
+#define EPB_TRANS_RDY (1ULL << 31)
+#define EPB_TRANS_ERR (1ULL << 30)
+#define EPB_TRANS_TRIES 5
+
+/*
+ * query, claim, release ownership of the EPB (External Parallel Bus)
+ * for a specified SERDES.
+ * the "claim" parameter is >0 to claim, <0 to release, 0 to query.
+ * Returns <0 for errors, >0 if we had ownership, else 0.
+ */
+static int epb_access(struct qib_devdata *dd, int sdnum, int claim)
+{
+	u16 acc;
+	u64 accval;
+	int owned = 0;
+	u64 oct_sel = 0;
+
+	switch (sdnum) {
+	case IB_7220_SERDES:
+		/*
+		 * The IB SERDES "ownership" is fairly simple. A single each
+		 * request/grant.
+		 */
+		acc = kr_ibsd_epb_access_ctrl;
+		break;
+
+	case PCIE_SERDES0:
+	case PCIE_SERDES1:
+		/* PCIe SERDES has two "octants", need to select which */
+		acc = kr_pciesd_epb_access_ctrl;
+		oct_sel = (2 << (sdnum - PCIE_SERDES0));
+		break;
+
+	default:
+		return 0;
+	}
+
+	/* Make sure any outstanding transaction was seen */
+	qib_read_kreg32(dd, kr_scratch);
+	udelay(15);
+
+	accval = qib_read_kreg32(dd, acc);
+
+	owned = !!(accval & EPB_ACC_GNT);
+	if (claim < 0) {
+		/* Need to release */
+		u64 pollval;
+		/*
+		 * The only writeable bits are the request and CS.
+		 * Both should be clear
+		 */
+		u64 newval = 0;
+		qib_write_kreg(dd, acc, newval);
+		/* First read after write is not trustworthy */
+		pollval = qib_read_kreg32(dd, acc);
+		udelay(5);
+		pollval = qib_read_kreg32(dd, acc);
+		if (pollval & EPB_ACC_GNT)
+			owned = -1;
+	} else if (claim > 0) {
+		/* Need to claim */
+		u64 pollval;
+		u64 newval = EPB_ACC_REQ | oct_sel;
+		qib_write_kreg(dd, acc, newval);
+		/* First read after write is not trustworthy */
+		pollval = qib_read_kreg32(dd, acc);
+		udelay(5);
+		pollval = qib_read_kreg32(dd, acc);
+		if (!(pollval & EPB_ACC_GNT))
+			owned = -1;
+	}
+	return owned;
+}
+
+/*
+ * Lemma to deal with race condition of write..read to epb regs
+ */
+static int epb_trans(struct qib_devdata *dd, u16 reg, u64 i_val, u64 *o_vp)
+{
+	int tries;
+	u64 transval;
+
+	qib_write_kreg(dd, reg, i_val);
+	/* Throw away first read, as RDY bit may be stale */
+	transval = qib_read_kreg64(dd, reg);
+
+	for (tries = EPB_TRANS_TRIES; tries; --tries) {
+		transval = qib_read_kreg32(dd, reg);
+		if (transval & EPB_TRANS_RDY)
+			break;
+		udelay(5);
+	}
+	if (transval & EPB_TRANS_ERR)
+		return -1;
+	if (tries > 0 && o_vp)
+		*o_vp = transval;
+	return tries;
+}
+
+/**
+ * qib_sd7220_reg_mod - modify SERDES register
+ * @dd: the qlogic_ib device
+ * @sdnum: which SERDES to access
+ * @loc: location - channel, element, register, as packed by EPB_LOC() macro.
+ * @wd: Write Data - value to set in register
+ * @mask: ones where data should be spliced into reg.
+ *
+ * Basic register read/modify/write, with un-needed acesses elided. That is,
+ * a mask of zero will prevent write, while a mask of 0xFF will prevent read.
+ * returns current (presumed, if a write was done) contents of selected
+ * register, or <0 if errors.
+ */
+static int qib_sd7220_reg_mod(struct qib_devdata *dd, int sdnum, u32 loc,
+			      u32 wd, u32 mask)
+{
+	u16 trans;
+	u64 transval;
+	int owned;
+	int tries, ret;
+	unsigned long flags;
+
+	switch (sdnum) {
+	case IB_7220_SERDES:
+		trans = kr_ibsd_epb_transaction_reg;
+		break;
+
+	case PCIE_SERDES0:
+	case PCIE_SERDES1:
+		trans = kr_pciesd_epb_transaction_reg;
+		break;
+
+	default:
+		return -1;
+	}
+
+	/*
+	 * All access is locked in software (vs other host threads) and
+	 * hardware (vs uC access).
+	 */
+	spin_lock_irqsave(&dd->cspec->sdepb_lock, flags);
+
+	owned = epb_access(dd, sdnum, 1);
+	if (owned < 0) {
+		spin_unlock_irqrestore(&dd->cspec->sdepb_lock, flags);
+		return -1;
+	}
+	ret = 0;
+	for (tries = EPB_TRANS_TRIES; tries; --tries) {
+		transval = qib_read_kreg32(dd, trans);
+		if (transval & EPB_TRANS_RDY)
+			break;
+		udelay(5);
+	}
+
+	if (tries > 0) {
+		tries = 1;      /* to make read-skip work */
+		if (mask != 0xFF) {
+			/*
+			 * Not a pure write, so need to read.
+			 * loc encodes chip-select as well as address
+			 */
+			transval = loc | EPB_RD;
+			tries = epb_trans(dd, trans, transval, &transval);
+		}
+		if (tries > 0 && mask != 0) {
+			/*
+			 * Not a pure read, so need to write.
+			 */
+			wd = (wd & mask) | (transval & ~mask);
+			transval = loc | (wd & EPB_DATA_MASK);
+			tries = epb_trans(dd, trans, transval, &transval);
+		}
+	}
+	/* else, failed to see ready, what error-handling? */
+
+	/*
+	 * Release bus. Failure is an error.
+	 */
+	if (epb_access(dd, sdnum, -1) < 0)
+		ret = -1;
+	else
+		ret = transval & EPB_DATA_MASK;
+
+	spin_unlock_irqrestore(&dd->cspec->sdepb_lock, flags);
+	if (tries <= 0)
+		ret = -1;
+	return ret;
+}
+
+#define EPB_ROM_R (2)
+#define EPB_ROM_W (1)
+/*
+ * Below, all uC-related, use appropriate UC_CS, depending
+ * on which SerDes is used.
+ */
+#define EPB_UC_CTL EPB_LOC(6, 0, 0)
+#define EPB_MADDRL EPB_LOC(6, 0, 2)
+#define EPB_MADDRH EPB_LOC(6, 0, 3)
+#define EPB_ROMDATA EPB_LOC(6, 0, 4)
+#define EPB_RAMDATA EPB_LOC(6, 0, 5)
+
+/* Transfer date to/from uC Program RAM of IB or PCIe SerDes */
+static int qib_sd7220_ram_xfer(struct qib_devdata *dd, int sdnum, u32 loc,
+			       u8 *buf, int cnt, int rd_notwr)
+{
+	u16 trans;
+	u64 transval;
+	u64 csbit;
+	int owned;
+	int tries;
+	int sofar;
+	int addr;
+	int ret;
+	unsigned long flags;
+	const char *op;
+
+	/* Pick appropriate transaction reg and "Chip select" for this serdes */
+	switch (sdnum) {
+	case IB_7220_SERDES:
+		csbit = 1ULL << EPB_IB_UC_CS_SHF;
+		trans = kr_ibsd_epb_transaction_reg;
+		break;
+
+	case PCIE_SERDES0:
+	case PCIE_SERDES1:
+		/* PCIe SERDES has uC "chip select" in different bit, too */
+		csbit = 1ULL << EPB_PCIE_UC_CS_SHF;
+		trans = kr_pciesd_epb_transaction_reg;
+		break;
+
+	default:
+		return -1;
+	}
+
+	op = rd_notwr ? "Rd" : "Wr";
+	spin_lock_irqsave(&dd->cspec->sdepb_lock, flags);
+
+	owned = epb_access(dd, sdnum, 1);
+	if (owned < 0) {
+		spin_unlock_irqrestore(&dd->cspec->sdepb_lock, flags);
+		qib_dbg("Could not get %s access to %s EPB: %X, loc %X\n",
+			op, (sdnum == IB_7220_SERDES) ? "IB" : "PCIe",
+			owned, loc);
+		return -1;
+	}
+
+	/*
+	 * In future code, we may need to distinguish several address ranges,
+	 * and select various memories based on this. For now, just trim
+	 * "loc" (location including address and memory select) to
+	 * "addr" (address within memory). we will only support PRAM
+	 * The memory is 8KB.
+	 */
+	addr = loc & 0x1FFF;
+	for (tries = EPB_TRANS_TRIES; tries; --tries) {
+		transval = qib_read_kreg32(dd, trans);
+		if (transval & EPB_TRANS_RDY)
+			break;
+		udelay(5);
+	}
+
+	sofar = 0;
+	if (tries <= 0)
+		qib_dbg("No initial RDY on EPB access request\n");
+	else {
+		/*
+		 * Every "memory" access is doubly-indirect.
+		 * We set two bytes of address, then read/write
+		 * one or mores bytes of data.
+		 */
+
+		/* First, we set control to "Read" or "Write" */
+		transval = csbit | EPB_UC_CTL |
+			(rd_notwr ? EPB_ROM_R : EPB_ROM_W);
+		tries = epb_trans(dd, trans, transval, &transval);
+		if (tries <= 0)
+			qib_dbg("No EPB response to uC %s cmd\n", op);
+		while (tries > 0 && sofar < cnt) {
+			if (!sofar) {
+				/* Only set address at start of chunk */
+				int addrbyte = (addr + sofar) >> 8;
+				transval = csbit | EPB_MADDRH | addrbyte;
+				tries = epb_trans(dd, trans, transval,
+						  &transval);
+				if (tries <= 0) {
+					qib_dbg("No EPB response ADDRH\n");
+					break;
+				}
+				addrbyte = (addr + sofar) & 0xFF;
+				transval = csbit | EPB_MADDRL | addrbyte;
+				tries = epb_trans(dd, trans, transval,
+						 &transval);
+				if (tries <= 0) {
+					qib_dbg("No EPB response ADDRL\n");
+					break;
+				}
+			}
+
+			if (rd_notwr)
+				transval = csbit | EPB_ROMDATA | EPB_RD;
+			else
+				transval = csbit | EPB_ROMDATA | buf[sofar];
+			tries = epb_trans(dd, trans, transval, &transval);
+			if (tries <= 0) {
+				qib_dbg("No EPB response DATA\n");
+				break;
+			}
+			if (rd_notwr)
+				buf[sofar] = transval & EPB_DATA_MASK;
+			++sofar;
+		}
+		/* Finally, clear control-bit for Read or Write */
+		transval = csbit | EPB_UC_CTL;
+		tries = epb_trans(dd, trans, transval, &transval);
+		if (tries <= 0)
+			qib_dbg("No EPB response to drop of uC %s cmd\n", op);
+	}
+
+	ret = sofar;
+	/* Release bus. Failure is an error */
+	if (epb_access(dd, sdnum, -1) < 0)
+		ret = -1;
+
+	spin_unlock_irqrestore(&dd->cspec->sdepb_lock, flags);
+	if (tries <= 0) {
+		qib_dbg("SERDES PRAM %s failed after %d bytes\n", op, sofar);
+		ret = -1;
+	}
+	return ret;
+}
+
+#define PROG_CHUNK 64
+
+int qib_sd7220_prog_ld(struct qib_devdata *dd, int sdnum,
+		       u8 *img, int len, int offset)
+{
+	int cnt, sofar, req;
+
+	sofar = 0;
+	while (sofar < len) {
+		req = len - sofar;
+		if (req > PROG_CHUNK)
+			req = PROG_CHUNK;
+		cnt = qib_sd7220_ram_xfer(dd, sdnum, offset + sofar,
+					  img + sofar, req, 0);
+		if (cnt < req) {
+			sofar = -1;
+			break;
+		}
+		sofar += req;
+	}
+	return sofar;
+}
+
+#define VFY_CHUNK 64
+#define SD_PRAM_ERROR_LIMIT 42
+
+int qib_sd7220_prog_vfy(struct qib_devdata *dd, int sdnum,
+			const u8 *img, int len, int offset)
+{
+	int cnt, sofar, req, idx, errors;
+	unsigned char readback[VFY_CHUNK];
+
+	errors = 0;
+	sofar = 0;
+	while (sofar < len) {
+		req = len - sofar;
+		if (req > VFY_CHUNK)
+			req = VFY_CHUNK;
+		cnt = qib_sd7220_ram_xfer(dd, sdnum, sofar + offset,
+					  readback, req, 1);
+		if (cnt < req) {
+			/* failed in read itself */
+			sofar = -1;
+			break;
+		}
+		for (idx = 0; idx < cnt; ++idx) {
+			if (readback[idx] != img[idx+sofar])
+				++errors;
+		}
+		sofar += cnt;
+	}
+	return errors ? -errors : sofar;
+}
+
+/*
+ * IRQ not set up at this point in init, so we poll.
+ */
+#define IB_SERDES_TRIM_DONE (1ULL << 11)
+#define TRIM_TMO (30)
+
+static int qib_sd_trimdone_poll(struct qib_devdata *dd)
+{
+	int trim_tmo, ret;
+	uint64_t val;
+
+	/*
+	 * Default to failure, so IBC will not start
+	 * without IB_SERDES_TRIM_DONE.
+	 */
+	ret = 0;
+	for (trim_tmo = 0; trim_tmo < TRIM_TMO; ++trim_tmo) {
+		val = qib_read_kreg64(dd, kr_ibcstatus);
+		if (val & IB_SERDES_TRIM_DONE) {
+			qib_cdbg(VERBOSE, "TRIMDONE after %d\n", trim_tmo);
+			ret = 1;
+			break;
+		}
+		msleep(10);
+	}
+	if (trim_tmo >= TRIM_TMO) {
+		qib_dev_err(dd, "No TRIMDONE in %d tries\n", trim_tmo);
+		ret = 0;
+	}
+	return ret;
+}
+
+#define TX_FAST_ELT (9)
+
+/*
+ * Set the "negotiation" values for SERDES. These are used by the IB1.2
+ * link negotiation. Macros below are attempt to keep the values a
+ * little more human-editable.
+ * First, values related to Drive De-emphasis Settings.
+ */
+
+#define NUM_DDS_REGS 6
+#define DDS_REG_MAP 0x76A910 /* LSB-first list of regs (in elt 9) to mod */
+
+#define DDS_VAL(amp_d, main_d, ipst_d, ipre_d, amp_s, main_s, ipst_s, ipre_s) \
+	{ { ((amp_d & 0x1F) << 1) | 1, ((amp_s & 0x1F) << 1) | 1, \
+	  (main_d << 3) | 4 | (ipre_d >> 2), \
+	  (main_s << 3) | 4 | (ipre_s >> 2), \
+	  ((ipst_d & 0xF) << 1) | ((ipre_d & 3) << 6) | 0x21, \
+	  ((ipst_s & 0xF) << 1) | ((ipre_s & 3) << 6) | 0x21 } }
+
+static struct dds_init {
+	uint8_t reg_vals[NUM_DDS_REGS];
+} dds_init_vals[] = {
+	/*       DDR(FDR)       SDR(HDR)   */
+	/* Vendor recommends below for 3m cable */
+#define DDS_3M 0
+	DDS_VAL(31, 19, 12, 0, 29, 22,  9, 0),
+	DDS_VAL(31, 12, 15, 4, 31, 15, 15, 1),
+	DDS_VAL(31, 13, 15, 3, 31, 16, 15, 0),
+	DDS_VAL(31, 14, 15, 2, 31, 17, 14, 0),
+	DDS_VAL(31, 15, 15, 1, 31, 18, 13, 0),
+	DDS_VAL(31, 16, 15, 0, 31, 19, 12, 0),
+	DDS_VAL(31, 17, 14, 0, 31, 20, 11, 0),
+	DDS_VAL(31, 18, 13, 0, 30, 21, 10, 0),
+	DDS_VAL(31, 20, 11, 0, 28, 23,  8, 0),
+	DDS_VAL(31, 21, 10, 0, 27, 24,  7, 0),
+	DDS_VAL(31, 22,  9, 0, 26, 25,  6, 0),
+	DDS_VAL(30, 23,  8, 0, 25, 26,  5, 0),
+	DDS_VAL(29, 24,  7, 0, 23, 27,  4, 0),
+	/* Vendor recommends below for 1m cable */
+#define DDS_1M 13
+	DDS_VAL(28, 25,  6, 0, 21, 28,  3, 0),
+	DDS_VAL(27, 26,  5, 0, 19, 29,  2, 0),
+	DDS_VAL(25, 27,  4, 0, 17, 30,  1, 0)
+};
+
+/*
+ * Now the RXEQ section of the table.
+ */
+/* Hardware packs an element number and register address thus: */
+#define RXEQ_INIT_RDESC(elt, addr) (((elt) & 0xF) | ((addr) << 4))
+#define RXEQ_VAL(elt, adr, val0, val1, val2, val3) \
+	{RXEQ_INIT_RDESC((elt), (adr)), {(val0), (val1), (val2), (val3)} }
+
+#define RXEQ_VAL_ALL(elt, adr, val)  \
+	{RXEQ_INIT_RDESC((elt), (adr)), {(val), (val), (val), (val)} }
+
+#define RXEQ_SDR_DFELTH 0
+#define RXEQ_SDR_TLTH 0
+#define RXEQ_SDR_G1CNT_Z1CNT 0x11
+#define RXEQ_SDR_ZCNT 23
+
+static struct rxeq_init {
+	u16 rdesc;      /* in form used in SerDesDDSRXEQ */
+	u8  rdata[4];
+} rxeq_init_vals[] = {
+	/* Set Rcv Eq. to Preset node */
+	RXEQ_VAL_ALL(7, 0x27, 0x10),
+	/* Set DFELTHFDR/HDR thresholds */
+	RXEQ_VAL(7, 8,    0, 0, 0, 0), /* FDR, was 0, 1, 2, 3 */
+	RXEQ_VAL(7, 0x21, 0, 0, 0, 0), /* HDR */
+	/* Set TLTHFDR/HDR theshold */
+	RXEQ_VAL(7, 9,    2, 2, 2, 2), /* FDR, was 0, 2, 4, 6 */
+	RXEQ_VAL(7, 0x23, 2, 2, 2, 2), /* HDR, was  0, 1, 2, 3 */
+	/* Set Preamp setting 2 (ZFR/ZCNT) */
+	RXEQ_VAL(7, 0x1B, 12, 12, 12, 12), /* FDR, was 12, 16, 20, 24 */
+	RXEQ_VAL(7, 0x1C, 12, 12, 12, 12), /* HDR, was 12, 16, 20, 24 */
+	/* Set Preamp DC gain and Setting 1 (GFR/GHR) */
+	RXEQ_VAL(7, 0x1E, 16, 16, 16, 16), /* FDR, was 16, 17, 18, 20 */
+	RXEQ_VAL(7, 0x1F, 16, 16, 16, 16), /* HDR, was 16, 17, 18, 20 */
+	/* Toggle RELOCK (in VCDL_CTRL0) to lock to data */
+	RXEQ_VAL_ALL(6, 6, 0x20), /* Set D5 High */
+	RXEQ_VAL_ALL(6, 6, 0), /* Set D5 Low */
+};
+
+/* There are 17 values from vendor, but IBC only accesses the first 16 */
+#define DDS_ROWS (16)
+#define RXEQ_ROWS ARRAY_SIZE(rxeq_init_vals)
+
+static int qib_sd_setvals(struct qib_devdata *dd)
+{
+	int idx, midx;
+	int min_idx;     /* Minimum index for this portion of table */
+	uint32_t dds_reg_map;
+	u64 __iomem *taddr, *iaddr;
+	uint64_t data;
+	uint64_t sdctl;
+
+	taddr = dd->kregbase + kr_serdes_maptable;
+	iaddr = dd->kregbase + kr_serdes_ddsrxeq0;
+
+	/*
+	 * Init the DDS section of the table.
+	 * Each "row" of the table provokes NUM_DDS_REG writes, to the
+	 * registers indicated in DDS_REG_MAP.
+	 */
+	sdctl = qib_read_kreg64(dd, kr_ibserdesctrl);
+	sdctl = (sdctl & ~(0x1f << 8)) | (NUM_DDS_REGS << 8);
+	sdctl = (sdctl & ~(0x1f << 13)) | (RXEQ_ROWS << 13);
+	qib_write_kreg(dd, kr_ibserdesctrl, sdctl);
+
+	/*
+	 * Iterate down table within loop for each register to store.
+	 */
+	dds_reg_map = DDS_REG_MAP;
+	for (idx = 0; idx < NUM_DDS_REGS; ++idx) {
+		data = ((dds_reg_map & 0xF) << 4) | TX_FAST_ELT;
+		writeq(data, iaddr + idx);
+		mmiowb();
+		qib_read_kreg32(dd, kr_scratch);
+		dds_reg_map >>= 4;
+		for (midx = 0; midx < DDS_ROWS; ++midx) {
+			u64 __iomem *daddr = taddr + ((midx << 4) + idx);
+			data = dds_init_vals[midx].reg_vals[idx];
+			writeq(data, daddr);
+			mmiowb();
+			qib_read_kreg32(dd, kr_scratch);
+		} /* End inner for (vals for this reg, each row) */
+	} /* end outer for (regs to be stored) */
+
+	/*
+	 * Init the RXEQ section of the table.
+	 * This runs in a different order, as the pattern of
+	 * register references is more complex, but there are only
+	 * four "data" values per register.
+	 */
+	min_idx = idx; /* RXEQ indices pick up where DDS left off */
+	taddr += 0x100; /* RXEQ data is in second half of table */
+	/* Iterate through RXEQ register addresses */
+	for (idx = 0; idx < RXEQ_ROWS; ++idx) {
+		int didx; /* "destination" */
+		int vidx;
+
+		/* didx is offset by min_idx to address RXEQ range of regs */
+		didx = idx + min_idx;
+		/* Store the next RXEQ register address */
+		writeq(rxeq_init_vals[idx].rdesc, iaddr + didx);
+		mmiowb();
+		qib_read_kreg32(dd, kr_scratch);
+		/* Iterate through RXEQ values */
+		for (vidx = 0; vidx < 4; vidx++) {
+			data = rxeq_init_vals[idx].rdata[vidx];
+			writeq(data, taddr + (vidx << 6) + idx);
+			mmiowb();
+			qib_read_kreg32(dd, kr_scratch);
+		}
+	} /* end outer for (Reg-writes for RXEQ) */
+	return 0;
+}
+
+#define CMUCTRL5 EPB_LOC(7, 0, 0x15)
+#define RXHSCTRL0(chan) EPB_LOC(chan, 6, 0)
+#define VCDL_DAC2(chan) EPB_LOC(chan, 6, 5)
+#define VCDL_CTRL0(chan) EPB_LOC(chan, 6, 6)
+#define VCDL_CTRL2(chan) EPB_LOC(chan, 6, 8)
+#define START_EQ2(chan) EPB_LOC(chan, 7, 0x28)
+
+/*
+ * Repeat a "store" across all channels of the IB SerDes.
+ * Although nominally it inherits the "read value" of the last
+ * channel it modified, the only really useful return is <0 for
+ * failure, >= 0 for success. The parameter 'loc' is assumed to
+ * be the location in some channel of the register to be modified
+ * The caller can specify use of the "gang write" option of EPB,
+ * in which case we use the specified channel data for any fields
+ * not explicitely written.
+ */
+static int ibsd_mod_allchnls(struct qib_devdata *dd, int loc, int val,
+			     int mask)
+{
+	int ret = -1;
+	int chnl;
+
+	if (loc & EPB_GLOBAL_WR) {
+		/*
+		 * Our caller has assured us that we can set all four
+		 * channels at once. Trust that. If mask is not 0xFF,
+		 * we will read the _specified_ channel for our starting
+		 * value.
+		 */
+		loc |= (1U << EPB_IB_QUAD0_CS_SHF);
+		chnl = (loc >> (4 + EPB_ADDR_SHF)) & 7;
+		if (mask != 0xFF) {
+			ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES,
+						 loc & ~EPB_GLOBAL_WR, 0, 0);
+			if (ret < 0) {
+				int sloc = loc >> EPB_ADDR_SHF;
+
+				qib_dev_err(dd, "pre-read failed: elt %d,"
+					    " addr 0x%X, chnl %d\n",
+					    (sloc & 0xF),
+					    (sloc >> 9) & 0x3f, chnl);
+				return ret;
+			}
+			val = (ret & ~mask) | (val & mask);
+		}
+		loc &=  ~(7 << (4+EPB_ADDR_SHF));
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, loc, val, 0xFF);
+		if (ret < 0) {
+			int sloc = loc >> EPB_ADDR_SHF;
+
+			qib_dev_err(dd, "Global WR failed: elt %d,"
+				    " addr 0x%X, val %02X\n",
+				    (sloc & 0xF), (sloc >> 9) & 0x3f, val);
+		}
+		return ret;
+	}
+	/* Clear "channel" and set CS so we can simply iterate */
+	loc &=  ~(7 << (4+EPB_ADDR_SHF));
+	loc |= (1U << EPB_IB_QUAD0_CS_SHF);
+	for (chnl = 0; chnl < 4; ++chnl) {
+		int cloc = loc | (chnl << (4+EPB_ADDR_SHF));
+
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, cloc, val, mask);
+		if (ret < 0) {
+			int sloc = loc >> EPB_ADDR_SHF;
+
+			qib_dev_err(dd, "Write failed: elt %d,"
+				    " addr 0x%X, chnl %d, val 0x%02X,"
+				    " mask 0x%02X\n",
+				    (sloc & 0xF), (sloc >> 9) & 0x3f, chnl,
+				    val & 0xFF, mask & 0xFF);
+			break;
+		}
+	}
+	return ret;
+}
+
+/*
+ * Set the Tx values normally modified by IBC in IB1.2 mode to default
+ * values, as gotten from first row of init table.
+ */
+static int set_dds_vals(struct qib_devdata *dd, struct dds_init *ddi)
+{
+	int ret;
+	int idx, reg, data;
+	uint32_t regmap;
+
+	regmap = DDS_REG_MAP;
+	for (idx = 0; idx < NUM_DDS_REGS; ++idx) {
+		reg = (regmap & 0xF);
+		regmap >>= 4;
+		data = ddi->reg_vals[idx];
+		/* Vendor says RMW not needed for these regs, use 0xFF mask */
+		ret = ibsd_mod_allchnls(dd, EPB_LOC(0, 9, reg), data, 0xFF);
+		if (ret < 0)
+			break;
+	}
+	return ret;
+}
+
+/*
+ * Set the Rx values normally modified by IBC in IB1.2 mode to default
+ * values, as gotten from selected column of init table.
+ */
+static int set_rxeq_vals(struct qib_devdata *dd, int vsel)
+{
+	int ret;
+	int ridx;
+	int cnt = ARRAY_SIZE(rxeq_init_vals);
+
+	for (ridx = 0; ridx < cnt; ++ridx) {
+		int elt, reg, val, loc;
+
+		elt = rxeq_init_vals[ridx].rdesc & 0xF;
+		reg = rxeq_init_vals[ridx].rdesc >> 4;
+		loc = EPB_LOC(0, elt, reg);
+		val = rxeq_init_vals[ridx].rdata[vsel];
+		/* mask of 0xFF, because hardware does full-byte store. */
+		ret = ibsd_mod_allchnls(dd, loc, val, 0xFF);
+		if (ret < 0)
+			break;
+	}
+	return ret;
+}
+
+/*
+ * Set the default values (row 0) for DDR Driver Demphasis.
+ * we do this initially and whenever we turn off IB-1.2
+ *
+ * The "default" values for Rx equalization are also stored to
+ * SerDes registers. Formerly (and still default), we used set 2.
+ * For experimenting with cables and link-partners, we allow changing
+ * that via a module parameter.
+ */
+static unsigned qib_rxeq_set = 2;
+module_param_named(rxeq_default_set, qib_rxeq_set, uint,
+		   S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(rxeq_default_set,
+		 "Which set [0..3] of Rx Equalization values is default");
+
+static int qib_internal_presets(struct qib_devdata *dd)
+{
+	int ret = 0;
+
+	ret = set_dds_vals(dd, dds_init_vals + DDS_3M);
+
+	if (ret < 0)
+		qib_dev_err(dd, "Failed to set default DDS values\n");
+	ret = set_rxeq_vals(dd, qib_rxeq_set & 3);
+	if (ret < 0)
+		qib_dev_err(dd, "Failed to set default RXEQ values\n");
+	return ret;
+}
+
+int qib_sd7220_presets(struct qib_devdata *dd)
+{
+	int ret = 0;
+
+	if (!dd->cspec->presets_needed)
+		return ret;
+	dd->cspec->presets_needed = 0;
+	/* Assert uC reset, so we don't clash with it. */
+	qib_ibsd_reset(dd, 1);
+	udelay(2);
+	qib_sd_trimdone_monitor(dd, "link-down");
+
+	ret = qib_internal_presets(dd);
+	return ret;
+}
+
+static int qib_sd_trimself(struct qib_devdata *dd, int val)
+{
+	int loc = CMUCTRL5 | (1U << EPB_IB_QUAD0_CS_SHF);
+
+	return qib_sd7220_reg_mod(dd, IB_7220_SERDES, loc, val, 0xFF);
+}
+
+static int qib_sd_early(struct qib_devdata *dd)
+{
+	int ret;
+
+	ret = ibsd_mod_allchnls(dd, RXHSCTRL0(0) | EPB_GLOBAL_WR, 0xD4, 0xFF);
+	if (ret < 0)
+		goto bail;
+	ret = ibsd_mod_allchnls(dd, START_EQ1(0) | EPB_GLOBAL_WR, 0x10, 0xFF);
+	if (ret < 0)
+		goto bail;
+	ret = ibsd_mod_allchnls(dd, START_EQ2(0) | EPB_GLOBAL_WR, 0x30, 0xFF);
+bail:
+	return ret;
+}
+
+#define BACTRL(chnl) EPB_LOC(chnl, 6, 0x0E)
+#define LDOUTCTRL1(chnl) EPB_LOC(chnl, 7, 6)
+#define RXHSSTATUS(chnl) EPB_LOC(chnl, 6, 0xF)
+
+static int qib_sd_dactrim(struct qib_devdata *dd)
+{
+	int ret;
+
+	ret = ibsd_mod_allchnls(dd, VCDL_DAC2(0) | EPB_GLOBAL_WR, 0x2D, 0xFF);
+	if (ret < 0)
+		goto bail;
+
+	/* more fine-tuning of what will be default */
+	ret = ibsd_mod_allchnls(dd, VCDL_CTRL2(0), 3, 0xF);
+	if (ret < 0)
+		goto bail;
+
+	ret = ibsd_mod_allchnls(dd, BACTRL(0) | EPB_GLOBAL_WR, 0x40, 0xFF);
+	if (ret < 0)
+		goto bail;
+
+	ret = ibsd_mod_allchnls(dd, LDOUTCTRL1(0) | EPB_GLOBAL_WR, 0x04, 0xFF);
+	if (ret < 0)
+		goto bail;
+
+	ret = ibsd_mod_allchnls(dd, RXHSSTATUS(0) | EPB_GLOBAL_WR, 0x04, 0xFF);
+	if (ret < 0)
+		goto bail;
+
+	/*
+	 * Delay for max possible number of steps, with slop.
+	 * Each step is about 4usec.
+	 */
+	udelay(415);
+
+	ret = ibsd_mod_allchnls(dd, LDOUTCTRL1(0) | EPB_GLOBAL_WR, 0x00, 0xFF);
+
+bail:
+	return ret;
+}
+
+#define RELOCK_FIRST_MS 3
+#define RXLSPPM(chan) EPB_LOC(chan, 0, 2)
+void toggle_7220_rclkrls(struct qib_devdata *dd)
+{
+	int loc = RXLSPPM(0) | EPB_GLOBAL_WR;
+	int ret;
+
+	ret = ibsd_mod_allchnls(dd, loc, 0, 0x80);
+	if (ret < 0)
+		qib_dev_err(dd, "RCLKRLS failed to clear D7\n");
+	else {
+		udelay(1);
+		ibsd_mod_allchnls(dd, loc, 0x80, 0x80);
+	}
+	/* And again for good measure */
+	udelay(1);
+	ret = ibsd_mod_allchnls(dd, loc, 0, 0x80);
+	if (ret < 0)
+		qib_dev_err(dd, "RCLKRLS failed to clear D7\n");
+	else {
+		udelay(1);
+		ibsd_mod_allchnls(dd, loc, 0x80, 0x80);
+	}
+	/* Now reset xgxs and IBC to complete the recovery */
+	dd->f_xgxs_reset(dd->pport);
+}
+
+/*
+ * Shut down the timer that polls for relock occasions, if needed
+ * this is "hooked" from qib_7220_quiet_serdes(), which is called
+ * just before qib_shutdown_device() in qib_driver.c shuts down all
+ * the other timers
+ */
+void shutdown_7220_relock_poll(struct qib_devdata *dd)
+{
+	struct qib_relock *irp = &dd->cspec->relock_st;
+
+	if (atomic_read(&irp->relock_timer_active)) {
+		del_timer_sync(&irp->relock_timer);
+		atomic_set(&irp->relock_timer_active, 0);
+	}
+}
+
+static unsigned qib_relock_by_timer = 1;
+module_param_named(relock_by_timer, qib_relock_by_timer, uint,
+		   S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(relock_by_timer, "Allow relock attempt if link not up");
+
+static void qib_run_relock(unsigned long opaque)
+{
+	struct qib_devdata *dd = (struct qib_devdata *)opaque;
+	struct qib_pportdata *ppd = dd->pport;
+	struct qib_relock *irp = &dd->cspec->relock_st;
+	int timeoff;
+
+	/*
+	 * Check link-training state for "stuck" state, when down.
+	 * if found, try relock and schedule another try at
+	 * exponentially growing delay, maxed at one second.
+	 * if not stuck, our work is done.
+	 */
+	if ((dd->flags & QIB_INITTED) && !(ppd->lflags &
+	    (QIBL_IB_AUTONEG_INPROG | QIBL_LINKINIT | QIBL_LINKARMED |
+	     QIBL_LINKACTIVE))) {
+		if (qib_relock_by_timer) {
+			if (!(ppd->lflags & QIBL_IB_LINK_DISABLED)) {
+				qib_cdbg(VERBOSE, "RELOCK\n");
+				toggle_7220_rclkrls(dd);
+			}
+		}
+		/* re-set timer for next check */
+		timeoff = irp->relock_interval << 1;
+		if (timeoff > HZ)
+			timeoff = HZ;
+		irp->relock_interval = timeoff;
+	} else
+		timeoff = HZ;
+	mod_timer(&irp->relock_timer, jiffies + timeoff);
+}
+
+void set_7220_relock_poll(struct qib_devdata *dd, int ibup)
+{
+	struct qib_relock *irp = &dd->cspec->relock_st;
+
+	if (ibup > 0) {
+		/* We are now up, relax timer to 1 second interval */
+		if (atomic_read(&irp->relock_timer_active))
+			mod_timer(&irp->relock_timer, jiffies + HZ);
+	} else {
+		/* Transition to down, (re-)set timer to short interval. */
+		int timeout;
+
+		timeout = (HZ * ((ibup == -1) ? 1000 : RELOCK_FIRST_MS)) / 1000;
+		if (timeout == 0)
+			timeout = 1;
+		/* If timer has not yet been started, do so. */
+		if (atomic_inc_return(&irp->relock_timer_active) == 1) {
+			init_timer(&irp->relock_timer);
+			irp->relock_timer.function = qib_run_relock;
+			irp->relock_timer.data = (unsigned long) dd;
+			irp->relock_interval = timeout;
+			irp->relock_timer.expires = jiffies + timeout;
+			add_timer(&irp->relock_timer);
+		} else {
+			irp->relock_interval = timeout;
+			mod_timer(&irp->relock_timer, jiffies + timeout);
+			atomic_dec(&irp->relock_timer_active);
+		}
+	}
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 35/51] IB/qib: Add qib_sd7220_img.c

[Ralph Campbell]

creates the qib_sd7220_img.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_sd7220_img.c | 1081 ++++++++++++++++++++++++++++
 1 files changed, 1081 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_sd7220_img.c

diff --git a/drivers/infiniband/hw/qib/qib_sd7220_img.c b/drivers/infiniband/hw/qib/qib_sd7220_img.c
new file mode 100644
index 0000000..a1118fb
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_sd7220_img.c
@@ -0,0 +1,1081 @@
+/*
+ * Copyright (c) 2007, 2008 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This file contains the memory image from the vendor, to be copied into
+ * the IB SERDES of the IBA7220 during initialization.
+ * The file also includes the two functions which use this image.
+ */
+#include <linux/pci.h>
+#include <linux/delay.h>
+
+#include "qib.h"
+#include "qib_7220.h"
+
+static unsigned char qib_sd7220_ib_img[] = {
+/*0000*/0x02, 0x0A, 0x29, 0x02, 0x0A, 0x87, 0xE5, 0xE6,
+	0x30, 0xE6, 0x04, 0x7F, 0x01, 0x80, 0x02, 0x7F,
+/*0010*/0x00, 0xE5, 0xE2, 0x30, 0xE4, 0x04, 0x7E, 0x01,
+	0x80, 0x02, 0x7E, 0x00, 0xEE, 0x5F, 0x60, 0x08,
+/*0020*/0x53, 0xF9, 0xF7, 0xE4, 0xF5, 0xFE, 0x80, 0x08,
+	0x7F, 0x0A, 0x12, 0x17, 0x31, 0x12, 0x0E, 0xA2,
+/*0030*/0x75, 0xFC, 0x08, 0xE4, 0xF5, 0xFD, 0xE5, 0xE7,
+	0x20, 0xE7, 0x03, 0x43, 0xF9, 0x08, 0x22, 0x00,
+/*0040*/0x01, 0x20, 0x11, 0x00, 0x04, 0x20, 0x00, 0x75,
+	0x51, 0x01, 0xE4, 0xF5, 0x52, 0xF5, 0x53, 0xF5,
+/*0050*/0x52, 0xF5, 0x7E, 0x7F, 0x04, 0x02, 0x04, 0x38,
+	0xC2, 0x36, 0x05, 0x52, 0xE5, 0x52, 0xD3, 0x94,
+/*0060*/0x0C, 0x40, 0x05, 0x75, 0x52, 0x01, 0xD2, 0x36,
+	0x90, 0x07, 0x0C, 0x74, 0x07, 0xF0, 0xA3, 0x74,
+/*0070*/0xFF, 0xF0, 0xE4, 0xF5, 0x0C, 0xA3, 0xF0, 0x90,
+	0x07, 0x14, 0xF0, 0xA3, 0xF0, 0x75, 0x0B, 0x20,
+/*0080*/0xF5, 0x09, 0xE4, 0xF5, 0x08, 0xE5, 0x08, 0xD3,
+	0x94, 0x30, 0x40, 0x03, 0x02, 0x04, 0x04, 0x12,
+/*0090*/0x00, 0x06, 0x15, 0x0B, 0xE5, 0x08, 0x70, 0x04,
+	0x7F, 0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5, 0x09,
+/*00A0*/0x70, 0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00,
+	0xEE, 0x5F, 0x60, 0x05, 0x12, 0x18, 0x71, 0xD2,
+/*00B0*/0x35, 0x53, 0xE1, 0xF7, 0xE5, 0x08, 0x45, 0x09,
+	0xFF, 0xE5, 0x0B, 0x25, 0xE0, 0x25, 0xE0, 0x24,
+/*00C0*/0x83, 0xF5, 0x82, 0xE4, 0x34, 0x07, 0xF5, 0x83,
+	0xEF, 0xF0, 0x85, 0xE2, 0x20, 0xE5, 0x52, 0xD3,
+/*00D0*/0x94, 0x01, 0x40, 0x0D, 0x12, 0x19, 0xF3, 0xE0,
+	0x54, 0xA0, 0x64, 0x40, 0x70, 0x03, 0x02, 0x03,
+/*00E0*/0xFB, 0x53, 0xF9, 0xF8, 0x90, 0x94, 0x70, 0xE4,
+	0xF0, 0xE0, 0xF5, 0x10, 0xAF, 0x09, 0x12, 0x1E,
+/*00F0*/0xB3, 0xAF, 0x08, 0xEF, 0x44, 0x08, 0xF5, 0x82,
+	0x75, 0x83, 0x80, 0xE0, 0xF5, 0x29, 0xEF, 0x44,
+/*0100*/0x07, 0x12, 0x1A, 0x3C, 0xF5, 0x22, 0x54, 0x40,
+	0xD3, 0x94, 0x00, 0x40, 0x1E, 0xE5, 0x29, 0x54,
+/*0110*/0xF0, 0x70, 0x21, 0x12, 0x19, 0xF3, 0xE0, 0x44,
+	0x80, 0xF0, 0xE5, 0x22, 0x54, 0x30, 0x65, 0x08,
+/*0120*/0x70, 0x09, 0x12, 0x19, 0xF3, 0xE0, 0x54, 0xBF,
+	0xF0, 0x80, 0x09, 0x12, 0x19, 0xF3, 0x74, 0x40,
+/*0130*/0xF0, 0x02, 0x03, 0xFB, 0x12, 0x1A, 0x12, 0x75,
+	0x83, 0xAE, 0x74, 0xFF, 0xF0, 0xAF, 0x08, 0x7E,
+/*0140*/0x00, 0xEF, 0x44, 0x07, 0xF5, 0x82, 0xE0, 0xFD,
+	0xE5, 0x0B, 0x25, 0xE0, 0x25, 0xE0, 0x24, 0x81,
+/*0150*/0xF5, 0x82, 0xE4, 0x34, 0x07, 0xF5, 0x83, 0xED,
+	0xF0, 0x90, 0x07, 0x0E, 0xE0, 0x04, 0xF0, 0xEF,
+/*0160*/0x44, 0x07, 0xF5, 0x82, 0x75, 0x83, 0x98, 0xE0,
+	0xF5, 0x28, 0x12, 0x1A, 0x23, 0x40, 0x0C, 0x12,
+/*0170*/0x19, 0xF3, 0xE0, 0x44, 0x01, 0x12, 0x1A, 0x32,
+	0x02, 0x03, 0xF6, 0xAF, 0x08, 0x7E, 0x00, 0x74,
+/*0180*/0x80, 0xCD, 0xEF, 0xCD, 0x8D, 0x82, 0xF5, 0x83,
+	0xE0, 0x30, 0xE0, 0x0A, 0x12, 0x19, 0xF3, 0xE0,
+/*0190*/0x44, 0x20, 0xF0, 0x02, 0x03, 0xFB, 0x12, 0x19,
+	0xF3, 0xE0, 0x54, 0xDF, 0xF0, 0xEE, 0x44, 0xAE,
+/*01A0*/0x12, 0x1A, 0x43, 0x30, 0xE4, 0x03, 0x02, 0x03,
+	0xFB, 0x74, 0x9E, 0x12, 0x1A, 0x05, 0x20, 0xE0,
+/*01B0*/0x03, 0x02, 0x03, 0xFB, 0x8F, 0x82, 0x8E, 0x83,
+	0xE0, 0x20, 0xE0, 0x03, 0x02, 0x03, 0xFB, 0x12,
+/*01C0*/0x19, 0xF3, 0xE0, 0x44, 0x10, 0xF0, 0xE5, 0xE3,
+	0x20, 0xE7, 0x08, 0xE5, 0x08, 0x12, 0x1A, 0x3A,
+/*01D0*/0x44, 0x04, 0xF0, 0xAF, 0x08, 0x7E, 0x00, 0xEF,
+	0x12, 0x1A, 0x3A, 0x20, 0xE2, 0x34, 0x12, 0x19,
+/*01E0*/0xF3, 0xE0, 0x44, 0x08, 0xF0, 0xE5, 0xE4, 0x30,
+	0xE6, 0x04, 0x7D, 0x01, 0x80, 0x02, 0x7D, 0x00,
+/*01F0*/0xE5, 0x7E, 0xC3, 0x94, 0x04, 0x50, 0x04, 0x7C,
+	0x01, 0x80, 0x02, 0x7C, 0x00, 0xEC, 0x4D, 0x60,
+/*0200*/0x05, 0xC2, 0x35, 0x02, 0x03, 0xFB, 0xEE, 0x44,
+	0xD2, 0x12, 0x1A, 0x43, 0x44, 0x40, 0xF0, 0x02,
+/*0210*/0x03, 0xFB, 0x12, 0x19, 0xF3, 0xE0, 0x54, 0xF7,
+	0xF0, 0x12, 0x1A, 0x12, 0x75, 0x83, 0xD2, 0xE0,
+/*0220*/0x54, 0xBF, 0xF0, 0x90, 0x07, 0x14, 0xE0, 0x04,
+	0xF0, 0xE5, 0x7E, 0x70, 0x03, 0x75, 0x7E, 0x01,
+/*0230*/0xAF, 0x08, 0x7E, 0x00, 0x12, 0x1A, 0x23, 0x40,
+	0x12, 0x12, 0x19, 0xF3, 0xE0, 0x44, 0x01, 0x12,
+/*0240*/0x19, 0xF2, 0xE0, 0x54, 0x02, 0x12, 0x1A, 0x32,
+	0x02, 0x03, 0xFB, 0x12, 0x19, 0xF3, 0xE0, 0x44,
+/*0250*/0x02, 0x12, 0x19, 0xF2, 0xE0, 0x54, 0xFE, 0xF0,
+	0xC2, 0x35, 0xEE, 0x44, 0x8A, 0x8F, 0x82, 0xF5,
+/*0260*/0x83, 0xE0, 0xF5, 0x17, 0x54, 0x8F, 0x44, 0x40,
+	0xF0, 0x74, 0x90, 0xFC, 0xE5, 0x08, 0x44, 0x07,
+/*0270*/0xFD, 0xF5, 0x82, 0x8C, 0x83, 0xE0, 0x54, 0x3F,
+	0x90, 0x07, 0x02, 0xF0, 0xE0, 0x54, 0xC0, 0x8D,
+/*0280*/0x82, 0x8C, 0x83, 0xF0, 0x74, 0x92, 0x12, 0x1A,
+	0x05, 0x90, 0x07, 0x03, 0x12, 0x1A, 0x19, 0x74,
+/*0290*/0x82, 0x12, 0x1A, 0x05, 0x90, 0x07, 0x04, 0x12,
+	0x1A, 0x19, 0x74, 0xB4, 0x12, 0x1A, 0x05, 0x90,
+/*02A0*/0x07, 0x05, 0x12, 0x1A, 0x19, 0x74, 0x94, 0xFE,
+	0xE5, 0x08, 0x44, 0x06, 0x12, 0x1A, 0x0A, 0xF5,
+/*02B0*/0x10, 0x30, 0xE0, 0x04, 0xD2, 0x37, 0x80, 0x02,
+	0xC2, 0x37, 0xE5, 0x10, 0x54, 0x7F, 0x8F, 0x82,
+/*02C0*/0x8E, 0x83, 0xF0, 0x30, 0x44, 0x30, 0x12, 0x1A,
+	0x03, 0x54, 0x80, 0xD3, 0x94, 0x00, 0x40, 0x04,
+/*02D0*/0xD2, 0x39, 0x80, 0x02, 0xC2, 0x39, 0x8F, 0x82,
+	0x8E, 0x83, 0xE0, 0x44, 0x80, 0xF0, 0x12, 0x1A,
+/*02E0*/0x03, 0x54, 0x40, 0xD3, 0x94, 0x00, 0x40, 0x04,
+	0xD2, 0x3A, 0x80, 0x02, 0xC2, 0x3A, 0x8F, 0x82,
+/*02F0*/0x8E, 0x83, 0xE0, 0x44, 0x40, 0xF0, 0x74, 0x92,
+	0xFE, 0xE5, 0x08, 0x44, 0x06, 0x12, 0x1A, 0x0A,
+/*0300*/0x30, 0xE7, 0x04, 0xD2, 0x38, 0x80, 0x02, 0xC2,
+	0x38, 0x8F, 0x82, 0x8E, 0x83, 0xE0, 0x54, 0x7F,
+/*0310*/0xF0, 0x12, 0x1E, 0x46, 0xE4, 0xF5, 0x0A, 0x20,
+	0x03, 0x02, 0x80, 0x03, 0x30, 0x43, 0x03, 0x12,
+/*0320*/0x19, 0x95, 0x20, 0x02, 0x02, 0x80, 0x03, 0x30,
+	0x42, 0x03, 0x12, 0x0C, 0x8F, 0x30, 0x30, 0x06,
+/*0330*/0x12, 0x19, 0x95, 0x12, 0x0C, 0x8F, 0x12, 0x0D,
+	0x47, 0x12, 0x19, 0xF3, 0xE0, 0x54, 0xFB, 0xF0,
+/*0340*/0xE5, 0x0A, 0xC3, 0x94, 0x01, 0x40, 0x46, 0x43,
+	0xE1, 0x08, 0x12, 0x19, 0xF3, 0xE0, 0x44, 0x04,
+/*0350*/0xF0, 0xE5, 0xE4, 0x20, 0xE7, 0x2A, 0x12, 0x1A,
+	0x12, 0x75, 0x83, 0xD2, 0xE0, 0x54, 0x08, 0xD3,
+/*0360*/0x94, 0x00, 0x40, 0x04, 0x7F, 0x01, 0x80, 0x02,
+	0x7F, 0x00, 0xE5, 0x0A, 0xC3, 0x94, 0x01, 0x40,
+/*0370*/0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEF,
+	0x5E, 0x60, 0x05, 0x12, 0x1D, 0xD7, 0x80, 0x17,
+/*0380*/0x12, 0x1A, 0x12, 0x75, 0x83, 0xD2, 0xE0, 0x44,
+	0x08, 0xF0, 0x02, 0x03, 0xFB, 0x12, 0x1A, 0x12,
+/*0390*/0x75, 0x83, 0xD2, 0xE0, 0x54, 0xF7, 0xF0, 0x12,
+	0x1E, 0x46, 0x7F, 0x08, 0x12, 0x17, 0x31, 0x74,
+/*03A0*/0x8E, 0xFE, 0x12, 0x1A, 0x12, 0x8E, 0x83, 0xE0,
+	0xF5, 0x10, 0x54, 0xFE, 0xF0, 0xE5, 0x10, 0x44,
+/*03B0*/0x01, 0xFF, 0xE5, 0x08, 0xFD, 0xED, 0x44, 0x07,
+	0xF5, 0x82, 0xEF, 0xF0, 0xE5, 0x10, 0x54, 0xFE,
+/*03C0*/0xFF, 0xED, 0x44, 0x07, 0xF5, 0x82, 0xEF, 0x12,
+	0x1A, 0x11, 0x75, 0x83, 0x86, 0xE0, 0x44, 0x10,
+/*03D0*/0x12, 0x1A, 0x11, 0xE0, 0x44, 0x10, 0xF0, 0x12,
+	0x19, 0xF3, 0xE0, 0x54, 0xFD, 0x44, 0x01, 0xFF,
+/*03E0*/0x12, 0x19, 0xF3, 0xEF, 0x12, 0x1A, 0x32, 0x30,
+	0x32, 0x0C, 0xE5, 0x08, 0x44, 0x08, 0xF5, 0x82,
+/*03F0*/0x75, 0x83, 0x82, 0x74, 0x05, 0xF0, 0xAF, 0x0B,
+	0x12, 0x18, 0xD7, 0x74, 0x10, 0x25, 0x08, 0xF5,
+/*0400*/0x08, 0x02, 0x00, 0x85, 0x05, 0x09, 0xE5, 0x09,
+	0xD3, 0x94, 0x07, 0x50, 0x03, 0x02, 0x00, 0x82,
+/*0410*/0xE5, 0x7E, 0xD3, 0x94, 0x00, 0x40, 0x04, 0x7F,
+	0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5, 0x7E, 0xC3,
+/*0420*/0x94, 0xFA, 0x50, 0x04, 0x7E, 0x01, 0x80, 0x02,
+	0x7E, 0x00, 0xEE, 0x5F, 0x60, 0x02, 0x05, 0x7E,
+/*0430*/0x30, 0x35, 0x0B, 0x43, 0xE1, 0x01, 0x7F, 0x09,
+	0x12, 0x17, 0x31, 0x02, 0x00, 0x58, 0x53, 0xE1,
+/*0440*/0xFE, 0x02, 0x00, 0x58, 0x8E, 0x6A, 0x8F, 0x6B,
+	0x8C, 0x6C, 0x8D, 0x6D, 0x75, 0x6E, 0x01, 0x75,
+/*0450*/0x6F, 0x01, 0x75, 0x70, 0x01, 0xE4, 0xF5, 0x73,
+	0xF5, 0x74, 0xF5, 0x75, 0x90, 0x07, 0x2F, 0xF0,
+/*0460*/0xF5, 0x3C, 0xF5, 0x3E, 0xF5, 0x46, 0xF5, 0x47,
+	0xF5, 0x3D, 0xF5, 0x3F, 0xF5, 0x6F, 0xE5, 0x6F,
+/*0470*/0x70, 0x0F, 0xE5, 0x6B, 0x45, 0x6A, 0x12, 0x07,
+	0x2A, 0x75, 0x83, 0x80, 0x74, 0x3A, 0xF0, 0x80,
+/*0480*/0x09, 0x12, 0x07, 0x2A, 0x75, 0x83, 0x80, 0x74,
+	0x1A, 0xF0, 0xE4, 0xF5, 0x6E, 0xC3, 0x74, 0x3F,
+/*0490*/0x95, 0x6E, 0xFF, 0x12, 0x08, 0x65, 0x75, 0x83,
+	0x82, 0xEF, 0xF0, 0x12, 0x1A, 0x4D, 0x12, 0x08,
+/*04A0*/0xC6, 0xE5, 0x33, 0xF0, 0x12, 0x08, 0xFA, 0x12,
+	0x08, 0xB1, 0x40, 0xE1, 0xE5, 0x6F, 0x70, 0x0B,
+/*04B0*/0x12, 0x07, 0x2A, 0x75, 0x83, 0x80, 0x74, 0x36,
+	0xF0, 0x80, 0x09, 0x12, 0x07, 0x2A, 0x75, 0x83,
+/*04C0*/0x80, 0x74, 0x16, 0xF0, 0x75, 0x6E, 0x01, 0x12,
+	0x07, 0x2A, 0x75, 0x83, 0xB4, 0xE5, 0x6E, 0xF0,
+/*04D0*/0x12, 0x1A, 0x4D, 0x74, 0x3F, 0x25, 0x6E, 0xF5,
+	0x82, 0xE4, 0x34, 0x00, 0xF5, 0x83, 0xE5, 0x33,
+/*04E0*/0xF0, 0x74, 0xBF, 0x25, 0x6E, 0xF5, 0x82, 0xE4,
+	0x34, 0x00, 0x12, 0x08, 0xB1, 0x40, 0xD8, 0xE4,
+/*04F0*/0xF5, 0x70, 0xF5, 0x46, 0xF5, 0x47, 0xF5, 0x6E,
+	0x12, 0x08, 0xFA, 0xF5, 0x83, 0xE0, 0xFE, 0x12,
+/*0500*/0x08, 0xC6, 0xE0, 0x7C, 0x00, 0x24, 0x00, 0xFF,
+	0xEC, 0x3E, 0xFE, 0xAD, 0x3B, 0xD3, 0xEF, 0x9D,
+/*0510*/0xEE, 0x9C, 0x50, 0x04, 0x7B, 0x01, 0x80, 0x02,
+	0x7B, 0x00, 0xE5, 0x70, 0x70, 0x04, 0x7A, 0x01,
+/*0520*/0x80, 0x02, 0x7A, 0x00, 0xEB, 0x5A, 0x60, 0x06,
+	0x85, 0x6E, 0x46, 0x75, 0x70, 0x01, 0xD3, 0xEF,
+/*0530*/0x9D, 0xEE, 0x9C, 0x50, 0x04, 0x7F, 0x01, 0x80,
+	0x02, 0x7F, 0x00, 0xE5, 0x70, 0xB4, 0x01, 0x04,
+/*0540*/0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEF, 0x5E,
+	0x60, 0x03, 0x85, 0x6E, 0x47, 0x05, 0x6E, 0xE5,
+/*0550*/0x6E, 0x64, 0x7F, 0x70, 0xA3, 0xE5, 0x46, 0x60,
+	0x05, 0xE5, 0x47, 0xB4, 0x7E, 0x03, 0x85, 0x46,
+/*0560*/0x47, 0xE5, 0x6F, 0x70, 0x08, 0x85, 0x46, 0x76,
+	0x85, 0x47, 0x77, 0x80, 0x0E, 0xC3, 0x74, 0x7F,
+/*0570*/0x95, 0x46, 0xF5, 0x78, 0xC3, 0x74, 0x7F, 0x95,
+	0x47, 0xF5, 0x79, 0xE5, 0x6F, 0x70, 0x37, 0xE5,
+/*0580*/0x46, 0x65, 0x47, 0x70, 0x0C, 0x75, 0x73, 0x01,
+	0x75, 0x74, 0x01, 0xF5, 0x3C, 0xF5, 0x3D, 0x80,
+/*0590*/0x35, 0xE4, 0xF5, 0x4E, 0xC3, 0xE5, 0x47, 0x95,
+	0x46, 0xF5, 0x3C, 0xC3, 0x13, 0xF5, 0x71, 0x25,
+/*05A0*/0x46, 0xF5, 0x72, 0xC3, 0x94, 0x3F, 0x40, 0x05,
+	0xE4, 0xF5, 0x3D, 0x80, 0x40, 0xC3, 0x74, 0x3F,
+/*05B0*/0x95, 0x72, 0xF5, 0x3D, 0x80, 0x37, 0xE5, 0x46,
+	0x65, 0x47, 0x70, 0x0F, 0x75, 0x73, 0x01, 0x75,
+/*05C0*/0x75, 0x01, 0xF5, 0x3E, 0xF5, 0x3F, 0x75, 0x4E,
+	0x01, 0x80, 0x22, 0xE4, 0xF5, 0x4E, 0xC3, 0xE5,
+/*05D0*/0x47, 0x95, 0x46, 0xF5, 0x3E, 0xC3, 0x13, 0xF5,
+	0x71, 0x25, 0x46, 0xF5, 0x72, 0xD3, 0x94, 0x3F,
+/*05E0*/0x50, 0x05, 0xE4, 0xF5, 0x3F, 0x80, 0x06, 0xE5,
+	0x72, 0x24, 0xC1, 0xF5, 0x3F, 0x05, 0x6F, 0xE5,
+/*05F0*/0x6F, 0xC3, 0x94, 0x02, 0x50, 0x03, 0x02, 0x04,
+	0x6E, 0xE5, 0x6D, 0x45, 0x6C, 0x70, 0x02, 0x80,
+/*0600*/0x04, 0xE5, 0x74, 0x45, 0x75, 0x90, 0x07, 0x2F,
+	0xF0, 0x7F, 0x01, 0xE5, 0x3E, 0x60, 0x04, 0xE5,
+/*0610*/0x3C, 0x70, 0x14, 0xE4, 0xF5, 0x3C, 0xF5, 0x3D,
+	0xF5, 0x3E, 0xF5, 0x3F, 0x12, 0x08, 0xD2, 0x70,
+/*0620*/0x04, 0xF0, 0x02, 0x06, 0xA4, 0x80, 0x7A, 0xE5,
+	0x3C, 0xC3, 0x95, 0x3E, 0x40, 0x07, 0xE5, 0x3C,
+/*0630*/0x95, 0x3E, 0xFF, 0x80, 0x06, 0xC3, 0xE5, 0x3E,
+	0x95, 0x3C, 0xFF, 0xE5, 0x76, 0xD3, 0x95, 0x79,
+/*0640*/0x40, 0x05, 0x85, 0x76, 0x7A, 0x80, 0x03, 0x85,
+	0x79, 0x7A, 0xE5, 0x77, 0xC3, 0x95, 0x78, 0x50,
+/*0650*/0x05, 0x85, 0x77, 0x7B, 0x80, 0x03, 0x85, 0x78,
+	0x7B, 0xE5, 0x7B, 0xD3, 0x95, 0x7A, 0x40, 0x30,
+/*0660*/0xE5, 0x7B, 0x95, 0x7A, 0xF5, 0x3C, 0xF5, 0x3E,
+	0xC3, 0xE5, 0x7B, 0x95, 0x7A, 0x90, 0x07, 0x19,
+/*0670*/0xF0, 0xE5, 0x3C, 0xC3, 0x13, 0xF5, 0x71, 0x25,
+	0x7A, 0xF5, 0x72, 0xC3, 0x94, 0x3F, 0x40, 0x05,
+/*0680*/0xE4, 0xF5, 0x3D, 0x80, 0x1F, 0xC3, 0x74, 0x3F,
+	0x95, 0x72, 0xF5, 0x3D, 0xF5, 0x3F, 0x80, 0x14,
+/*0690*/0xE4, 0xF5, 0x3C, 0xF5, 0x3E, 0x90, 0x07, 0x19,
+	0xF0, 0x12, 0x08, 0xD2, 0x70, 0x03, 0xF0, 0x80,
+/*06A0*/0x03, 0x74, 0x01, 0xF0, 0x12, 0x08, 0x65, 0x75,
+	0x83, 0xD0, 0xE0, 0x54, 0x0F, 0xFE, 0xAD, 0x3C,
+/*06B0*/0x70, 0x02, 0x7E, 0x07, 0xBE, 0x0F, 0x02, 0x7E,
+	0x80, 0xEE, 0xFB, 0xEF, 0xD3, 0x9B, 0x74, 0x80,
+/*06C0*/0xF8, 0x98, 0x40, 0x1F, 0xE4, 0xF5, 0x3C, 0xF5,
+	0x3E, 0x12, 0x08, 0xD2, 0x70, 0x03, 0xF0, 0x80,
+/*06D0*/0x12, 0x74, 0x01, 0xF0, 0xE5, 0x08, 0xFB, 0xEB,
+	0x44, 0x07, 0xF5, 0x82, 0x75, 0x83, 0xD2, 0xE0,
+/*06E0*/0x44, 0x10, 0xF0, 0xE5, 0x08, 0xFB, 0xEB, 0x44,
+	0x09, 0xF5, 0x82, 0x75, 0x83, 0x9E, 0xED, 0xF0,
+/*06F0*/0xEB, 0x44, 0x07, 0xF5, 0x82, 0x75, 0x83, 0xCA,
+	0xED, 0xF0, 0x12, 0x08, 0x65, 0x75, 0x83, 0xCC,
+/*0700*/0xEF, 0xF0, 0x22, 0xE5, 0x08, 0x44, 0x07, 0xF5,
+	0x82, 0x75, 0x83, 0xBC, 0xE0, 0x54, 0xF0, 0xF0,
+/*0710*/0xE5, 0x08, 0x44, 0x07, 0xF5, 0x82, 0x75, 0x83,
+	0xBE, 0xE0, 0x54, 0xF0, 0xF0, 0xE5, 0x08, 0x44,
+/*0720*/0x07, 0xF5, 0x82, 0x75, 0x83, 0xC0, 0xE0, 0x54,
+	0xF0, 0xF0, 0xE5, 0x08, 0x44, 0x07, 0xF5, 0x82,
+/*0730*/0x22, 0xF0, 0x90, 0x07, 0x28, 0xE0, 0xFE, 0xA3,
+	0xE0, 0xF5, 0x82, 0x8E, 0x83, 0x22, 0x85, 0x42,
+/*0740*/0x42, 0x85, 0x41, 0x41, 0x85, 0x40, 0x40, 0x74,
+	0xC0, 0x2F, 0xF5, 0x82, 0x74, 0x02, 0x3E, 0xF5,
+/*0750*/0x83, 0xE5, 0x42, 0xF0, 0x74, 0xE0, 0x2F, 0xF5,
+	0x82, 0x74, 0x02, 0x3E, 0xF5, 0x83, 0x22, 0xE5,
+/*0760*/0x42, 0x29, 0xFD, 0xE4, 0x33, 0xFC, 0xE5, 0x3C,
+	0xC3, 0x9D, 0xEC, 0x64, 0x80, 0xF8, 0x74, 0x80,
+/*0770*/0x98, 0x22, 0xF5, 0x83, 0xE0, 0x90, 0x07, 0x22,
+	0x54, 0x1F, 0xFD, 0xE0, 0xFA, 0xA3, 0xE0, 0xF5,
+/*0780*/0x82, 0x8A, 0x83, 0xED, 0xF0, 0x22, 0x90, 0x07,
+	0x22, 0xE0, 0xFC, 0xA3, 0xE0, 0xF5, 0x82, 0x8C,
+/*0790*/0x83, 0x22, 0x90, 0x07, 0x24, 0xFF, 0xED, 0x44,
+	0x07, 0xCF, 0xF0, 0xA3, 0xEF, 0xF0, 0x22, 0x85,
+/*07A0*/0x38, 0x38, 0x85, 0x39, 0x39, 0x85, 0x3A, 0x3A,
+	0x74, 0xC0, 0x2F, 0xF5, 0x82, 0x74, 0x02, 0x3E,
+/*07B0*/0xF5, 0x83, 0x22, 0x90, 0x07, 0x26, 0xFF, 0xED,
+	0x44, 0x07, 0xCF, 0xF0, 0xA3, 0xEF, 0xF0, 0x22,
+/*07C0*/0xF0, 0x74, 0xA0, 0x2F, 0xF5, 0x82, 0x74, 0x02,
+	0x3E, 0xF5, 0x83, 0x22, 0x74, 0xC0, 0x25, 0x11,
+/*07D0*/0xF5, 0x82, 0xE4, 0x34, 0x01, 0xF5, 0x83, 0x22,
+	0x74, 0x00, 0x25, 0x11, 0xF5, 0x82, 0xE4, 0x34,
+/*07E0*/0x02, 0xF5, 0x83, 0x22, 0x74, 0x60, 0x25, 0x11,
+	0xF5, 0x82, 0xE4, 0x34, 0x03, 0xF5, 0x83, 0x22,
+/*07F0*/0x74, 0x80, 0x25, 0x11, 0xF5, 0x82, 0xE4, 0x34,
+	0x03, 0xF5, 0x83, 0x22, 0x74, 0xE0, 0x25, 0x11,
+/*0800*/0xF5, 0x82, 0xE4, 0x34, 0x03, 0xF5, 0x83, 0x22,
+	0x74, 0x40, 0x25, 0x11, 0xF5, 0x82, 0xE4, 0x34,
+/*0810*/0x06, 0xF5, 0x83, 0x22, 0x74, 0x80, 0x2F, 0xF5,
+	0x82, 0x74, 0x02, 0x3E, 0xF5, 0x83, 0x22, 0xAF,
+/*0820*/0x08, 0x7E, 0x00, 0xEF, 0x44, 0x07, 0xF5, 0x82,
+	0x22, 0xF5, 0x83, 0xE5, 0x82, 0x44, 0x07, 0xF5,
+/*0830*/0x82, 0xE5, 0x40, 0xF0, 0x22, 0x74, 0x40, 0x25,
+	0x11, 0xF5, 0x82, 0xE4, 0x34, 0x02, 0xF5, 0x83,
+/*0840*/0x22, 0x74, 0xC0, 0x25, 0x11, 0xF5, 0x82, 0xE4,
+	0x34, 0x03, 0xF5, 0x83, 0x22, 0x74, 0x00, 0x25,
+/*0850*/0x11, 0xF5, 0x82, 0xE4, 0x34, 0x06, 0xF5, 0x83,
+	0x22, 0x74, 0x20, 0x25, 0x11, 0xF5, 0x82, 0xE4,
+/*0860*/0x34, 0x06, 0xF5, 0x83, 0x22, 0xE5, 0x08, 0xFD,
+	0xED, 0x44, 0x07, 0xF5, 0x82, 0x22, 0xE5, 0x41,
+/*0870*/0xF0, 0xE5, 0x65, 0x64, 0x01, 0x45, 0x64, 0x22,
+	0x7E, 0x00, 0xFB, 0x7A, 0x00, 0xFD, 0x7C, 0x00,
+/*0880*/0x22, 0x74, 0x20, 0x25, 0x11, 0xF5, 0x82, 0xE4,
+	0x34, 0x02, 0x22, 0x74, 0xA0, 0x25, 0x11, 0xF5,
+/*0890*/0x82, 0xE4, 0x34, 0x03, 0x22, 0x85, 0x3E, 0x42,
+	0x85, 0x3F, 0x41, 0x8F, 0x40, 0x22, 0x85, 0x3C,
+/*08A0*/0x42, 0x85, 0x3D, 0x41, 0x8F, 0x40, 0x22, 0x75,
+	0x45, 0x3F, 0x90, 0x07, 0x20, 0xE4, 0xF0, 0xA3,
+/*08B0*/0x22, 0xF5, 0x83, 0xE5, 0x32, 0xF0, 0x05, 0x6E,
+	0xE5, 0x6E, 0xC3, 0x94, 0x40, 0x22, 0xF0, 0xE5,
+/*08C0*/0x08, 0x44, 0x06, 0xF5, 0x82, 0x22, 0x74, 0x00,
+	0x25, 0x6E, 0xF5, 0x82, 0xE4, 0x34, 0x00, 0xF5,
+/*08D0*/0x83, 0x22, 0xE5, 0x6D, 0x45, 0x6C, 0x90, 0x07,
+	0x2F, 0x22, 0xE4, 0xF9, 0xE5, 0x3C, 0xD3, 0x95,
+/*08E0*/0x3E, 0x22, 0x74, 0x80, 0x2E, 0xF5, 0x82, 0xE4,
+	0x34, 0x02, 0xF5, 0x83, 0xE0, 0x22, 0x74, 0xA0,
+/*08F0*/0x2E, 0xF5, 0x82, 0xE4, 0x34, 0x02, 0xF5, 0x83,
+	0xE0, 0x22, 0x74, 0x80, 0x25, 0x6E, 0xF5, 0x82,
+/*0900*/0xE4, 0x34, 0x00, 0x22, 0x25, 0x42, 0xFD, 0xE4,
+	0x33, 0xFC, 0x22, 0x85, 0x42, 0x42, 0x85, 0x41,
+/*0910*/0x41, 0x85, 0x40, 0x40, 0x22, 0xED, 0x4C, 0x60,
+	0x03, 0x02, 0x09, 0xE5, 0xEF, 0x4E, 0x70, 0x37,
+/*0920*/0x90, 0x07, 0x26, 0x12, 0x07, 0x89, 0xE0, 0xFD,
+	0x12, 0x07, 0xCC, 0xED, 0xF0, 0x90, 0x07, 0x28,
+/*0930*/0x12, 0x07, 0x89, 0xE0, 0xFD, 0x12, 0x07, 0xD8,
+	0xED, 0xF0, 0x12, 0x07, 0x86, 0xE0, 0x54, 0x1F,
+/*0940*/0xFD, 0x12, 0x08, 0x81, 0xF5, 0x83, 0xED, 0xF0,
+	0x90, 0x07, 0x24, 0x12, 0x07, 0x89, 0xE0, 0x54,
+/*0950*/0x1F, 0xFD, 0x12, 0x08, 0x35, 0xED, 0xF0, 0xEF,
+	0x64, 0x04, 0x4E, 0x70, 0x37, 0x90, 0x07, 0x26,
+/*0960*/0x12, 0x07, 0x89, 0xE0, 0xFD, 0x12, 0x07, 0xE4,
+	0xED, 0xF0, 0x90, 0x07, 0x28, 0x12, 0x07, 0x89,
+/*0970*/0xE0, 0xFD, 0x12, 0x07, 0xF0, 0xED, 0xF0, 0x12,
+	0x07, 0x86, 0xE0, 0x54, 0x1F, 0xFD, 0x12, 0x08,
+/*0980*/0x8B, 0xF5, 0x83, 0xED, 0xF0, 0x90, 0x07, 0x24,
+	0x12, 0x07, 0x89, 0xE0, 0x54, 0x1F, 0xFD, 0x12,
+/*0990*/0x08, 0x41, 0xED, 0xF0, 0xEF, 0x64, 0x01, 0x4E,
+	0x70, 0x04, 0x7D, 0x01, 0x80, 0x02, 0x7D, 0x00,
+/*09A0*/0xEF, 0x64, 0x02, 0x4E, 0x70, 0x04, 0x7F, 0x01,
+	0x80, 0x02, 0x7F, 0x00, 0xEF, 0x4D, 0x60, 0x78,
+/*09B0*/0x90, 0x07, 0x26, 0x12, 0x07, 0x35, 0xE0, 0xFF,
+	0x12, 0x07, 0xFC, 0xEF, 0x12, 0x07, 0x31, 0xE0,
+/*09C0*/0xFF, 0x12, 0x08, 0x08, 0xEF, 0xF0, 0x90, 0x07,
+	0x22, 0x12, 0x07, 0x35, 0xE0, 0x54, 0x1F, 0xFF,
+/*09D0*/0x12, 0x08, 0x4D, 0xEF, 0xF0, 0x90, 0x07, 0x24,
+	0x12, 0x07, 0x35, 0xE0, 0x54, 0x1F, 0xFF, 0x12,
+/*09E0*/0x08, 0x59, 0xEF, 0xF0, 0x22, 0x12, 0x07, 0xCC,
+	0xE4, 0xF0, 0x12, 0x07, 0xD8, 0xE4, 0xF0, 0x12,
+/*09F0*/0x08, 0x81, 0xF5, 0x83, 0xE4, 0xF0, 0x12, 0x08,
+	0x35, 0x74, 0x14, 0xF0, 0x12, 0x07, 0xE4, 0xE4,
+/*0A00*/0xF0, 0x12, 0x07, 0xF0, 0xE4, 0xF0, 0x12, 0x08,
+	0x8B, 0xF5, 0x83, 0xE4, 0xF0, 0x12, 0x08, 0x41,
+/*0A10*/0x74, 0x14, 0xF0, 0x12, 0x07, 0xFC, 0xE4, 0xF0,
+	0x12, 0x08, 0x08, 0xE4, 0xF0, 0x12, 0x08, 0x4D,
+/*0A20*/0xE4, 0xF0, 0x12, 0x08, 0x59, 0x74, 0x14, 0xF0,
+	0x22, 0x53, 0xF9, 0xF7, 0x75, 0xFC, 0x10, 0xE4,
+/*0A30*/0xF5, 0xFD, 0x75, 0xFE, 0x30, 0xF5, 0xFF, 0xE5,
+	0xE7, 0x20, 0xE7, 0x03, 0x43, 0xF9, 0x08, 0xE5,
+/*0A40*/0xE6, 0x20, 0xE7, 0x0B, 0x78, 0xFF, 0xE4, 0xF6,
+	0xD8, 0xFD, 0x53, 0xE6, 0xFE, 0x80, 0x09, 0x78,
+/*0A50*/0x08, 0xE4, 0xF6, 0xD8, 0xFD, 0x53, 0xE6, 0xFE,
+	0x75, 0x81, 0x80, 0xE4, 0xF5, 0xA8, 0xD2, 0xA8,
+/*0A60*/0xC2, 0xA9, 0xD2, 0xAF, 0xE5, 0xE2, 0x20, 0xE5,
+	0x05, 0x20, 0xE6, 0x02, 0x80, 0x03, 0x43, 0xE1,
+/*0A70*/0x02, 0xE5, 0xE2, 0x20, 0xE0, 0x0E, 0x90, 0x00,
+	0x00, 0x7F, 0x00, 0x7E, 0x08, 0xE4, 0xF0, 0xA3,
+/*0A80*/0xDF, 0xFC, 0xDE, 0xFA, 0x02, 0x0A, 0xDB, 0x43,
+	0xFA, 0x01, 0xC0, 0xE0, 0xC0, 0xF0, 0xC0, 0x83,
+/*0A90*/0xC0, 0x82, 0xC0, 0xD0, 0x12, 0x1C, 0xE7, 0xD0,
+	0xD0, 0xD0, 0x82, 0xD0, 0x83, 0xD0, 0xF0, 0xD0,
+/*0AA0*/0xE0, 0x53, 0xFA, 0xFE, 0x32, 0x02, 0x1B, 0x55,
+	0xE4, 0x93, 0xA3, 0xF8, 0xE4, 0x93, 0xA3, 0xF6,
+/*0AB0*/0x08, 0xDF, 0xF9, 0x80, 0x29, 0xE4, 0x93, 0xA3,
+	0xF8, 0x54, 0x07, 0x24, 0x0C, 0xC8, 0xC3, 0x33,
+/*0AC0*/0xC4, 0x54, 0x0F, 0x44, 0x20, 0xC8, 0x83, 0x40,
+	0x04, 0xF4, 0x56, 0x80, 0x01, 0x46, 0xF6, 0xDF,
+/*0AD0*/0xE4, 0x80, 0x0B, 0x01, 0x02, 0x04, 0x08, 0x10,
+	0x20, 0x40, 0x80, 0x90, 0x00, 0x3F, 0xE4, 0x7E,
+/*0AE0*/0x01, 0x93, 0x60, 0xC1, 0xA3, 0xFF, 0x54, 0x3F,
+	0x30, 0xE5, 0x09, 0x54, 0x1F, 0xFE, 0xE4, 0x93,
+/*0AF0*/0xA3, 0x60, 0x01, 0x0E, 0xCF, 0x54, 0xC0, 0x25,
+	0xE0, 0x60, 0xAD, 0x40, 0xB8, 0x80, 0xFE, 0x8C,
+/*0B00*/0x64, 0x8D, 0x65, 0x8A, 0x66, 0x8B, 0x67, 0xE4,
+	0xF5, 0x69, 0xEF, 0x4E, 0x70, 0x03, 0x02, 0x1D,
+/*0B10*/0x55, 0xE4, 0xF5, 0x68, 0xE5, 0x67, 0x45, 0x66,
+	0x70, 0x32, 0x12, 0x07, 0x2A, 0x75, 0x83, 0x90,
+/*0B20*/0xE4, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC2, 0xE4,
+	0x12, 0x07, 0x29, 0x75, 0x83, 0xC4, 0xE4, 0x12,
+/*0B30*/0x08, 0x70, 0x70, 0x29, 0x12, 0x07, 0x2A, 0x75,
+	0x83, 0x92, 0xE4, 0x12, 0x07, 0x29, 0x75, 0x83,
+/*0B40*/0xC6, 0xE4, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC8,
+	0xE4, 0xF0, 0x80, 0x11, 0x90, 0x07, 0x26, 0x12,
+/*0B50*/0x07, 0x35, 0xE4, 0x12, 0x08, 0x70, 0x70, 0x05,
+	0x12, 0x07, 0x32, 0xE4, 0xF0, 0x12, 0x1D, 0x55,
+/*0B60*/0x12, 0x1E, 0xBF, 0xE5, 0x67, 0x45, 0x66, 0x70,
+	0x33, 0x12, 0x07, 0x2A, 0x75, 0x83, 0x90, 0xE5,
+/*0B70*/0x41, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC2, 0xE5,
+	0x41, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC4, 0x12,
+/*0B80*/0x08, 0x6E, 0x70, 0x29, 0x12, 0x07, 0x2A, 0x75,
+	0x83, 0x92, 0xE5, 0x40, 0x12, 0x07, 0x29, 0x75,
+/*0B90*/0x83, 0xC6, 0xE5, 0x40, 0x12, 0x07, 0x29, 0x75,
+	0x83, 0xC8, 0x80, 0x0E, 0x90, 0x07, 0x26, 0x12,
+/*0BA0*/0x07, 0x35, 0x12, 0x08, 0x6E, 0x70, 0x06, 0x12,
+	0x07, 0x32, 0xE5, 0x40, 0xF0, 0xAF, 0x69, 0x7E,
+/*0BB0*/0x00, 0xAD, 0x67, 0xAC, 0x66, 0x12, 0x04, 0x44,
+	0x12, 0x07, 0x2A, 0x75, 0x83, 0xCA, 0xE0, 0xD3,
+/*0BC0*/0x94, 0x00, 0x50, 0x0C, 0x05, 0x68, 0xE5, 0x68,
+	0xC3, 0x94, 0x05, 0x50, 0x03, 0x02, 0x0B, 0x14,
+/*0BD0*/0x22, 0x8C, 0x60, 0x8D, 0x61, 0x12, 0x08, 0xDA,
+	0x74, 0x20, 0x40, 0x0D, 0x2F, 0xF5, 0x82, 0x74,
+/*0BE0*/0x03, 0x3E, 0xF5, 0x83, 0xE5, 0x3E, 0xF0, 0x80,
+	0x0B, 0x2F, 0xF5, 0x82, 0x74, 0x03, 0x3E, 0xF5,
+/*0BF0*/0x83, 0xE5, 0x3C, 0xF0, 0xE5, 0x3C, 0xD3, 0x95,
+	0x3E, 0x40, 0x3C, 0xE5, 0x61, 0x45, 0x60, 0x70,
+/*0C00*/0x10, 0xE9, 0x12, 0x09, 0x04, 0xE5, 0x3E, 0x12,
+	0x07, 0x68, 0x40, 0x3B, 0x12, 0x08, 0x95, 0x80,
+/*0C10*/0x18, 0xE5, 0x3E, 0xC3, 0x95, 0x38, 0x40, 0x1D,
+	0x85, 0x3E, 0x38, 0xE5, 0x3E, 0x60, 0x05, 0x85,
+/*0C20*/0x3F, 0x39, 0x80, 0x03, 0x85, 0x39, 0x39, 0x8F,
+	0x3A, 0x12, 0x08, 0x14, 0xE5, 0x3E, 0x12, 0x07,
+/*0C30*/0xC0, 0xE5, 0x3F, 0xF0, 0x22, 0x80, 0x43, 0xE5,
+	0x61, 0x45, 0x60, 0x70, 0x19, 0x12, 0x07, 0x5F,
+/*0C40*/0x40, 0x05, 0x12, 0x08, 0x9E, 0x80, 0x27, 0x12,
+	0x09, 0x0B, 0x12, 0x08, 0x14, 0xE5, 0x42, 0x12,
+/*0C50*/0x07, 0xC0, 0xE5, 0x41, 0xF0, 0x22, 0xE5, 0x3C,
+	0xC3, 0x95, 0x38, 0x40, 0x1D, 0x85, 0x3C, 0x38,
+/*0C60*/0xE5, 0x3C, 0x60, 0x05, 0x85, 0x3D, 0x39, 0x80,
+	0x03, 0x85, 0x39, 0x39, 0x8F, 0x3A, 0x12, 0x08,
+/*0C70*/0x14, 0xE5, 0x3C, 0x12, 0x07, 0xC0, 0xE5, 0x3D,
+	0xF0, 0x22, 0x85, 0x38, 0x38, 0x85, 0x39, 0x39,
+/*0C80*/0x85, 0x3A, 0x3A, 0x12, 0x08, 0x14, 0xE5, 0x38,
+	0x12, 0x07, 0xC0, 0xE5, 0x39, 0xF0, 0x22, 0x7F,
+/*0C90*/0x06, 0x12, 0x17, 0x31, 0x12, 0x1D, 0x23, 0x12,
+	0x0E, 0x04, 0x12, 0x0E, 0x33, 0xE0, 0x44, 0x0A,
+/*0CA0*/0xF0, 0x74, 0x8E, 0xFE, 0x12, 0x0E, 0x04, 0x12,
+	0x0E, 0x0B, 0xEF, 0xF0, 0xE5, 0x28, 0x30, 0xE5,
+/*0CB0*/0x03, 0xD3, 0x80, 0x01, 0xC3, 0x40, 0x05, 0x75,
+	0x14, 0x20, 0x80, 0x03, 0x75, 0x14, 0x08, 0x12,
+/*0CC0*/0x0E, 0x04, 0x75, 0x83, 0x8A, 0xE5, 0x14, 0xF0,
+	0xB4, 0xFF, 0x05, 0x75, 0x12, 0x80, 0x80, 0x06,
+/*0CD0*/0xE5, 0x14, 0xC3, 0x13, 0xF5, 0x12, 0xE4, 0xF5,
+	0x16, 0xF5, 0x7F, 0x12, 0x19, 0x36, 0x12, 0x13,
+/*0CE0*/0xA3, 0xE5, 0x0A, 0xC3, 0x94, 0x01, 0x50, 0x09,
+	0x05, 0x16, 0xE5, 0x16, 0xC3, 0x94, 0x14, 0x40,
+/*0CF0*/0xEA, 0xE5, 0xE4, 0x20, 0xE7, 0x28, 0x12, 0x0E,
+	0x04, 0x75, 0x83, 0xD2, 0xE0, 0x54, 0x08, 0xD3,
+/*0D00*/0x94, 0x00, 0x40, 0x04, 0x7F, 0x01, 0x80, 0x02,
+	0x7F, 0x00, 0xE5, 0x0A, 0xC3, 0x94, 0x01, 0x40,
+/*0D10*/0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEF,
+	0x5E, 0x60, 0x03, 0x12, 0x1D, 0xD7, 0xE5, 0x7F,
+/*0D20*/0xC3, 0x94, 0x11, 0x40, 0x14, 0x12, 0x0E, 0x04,
+	0x75, 0x83, 0xD2, 0xE0, 0x44, 0x80, 0xF0, 0xE5,
+/*0D30*/0xE4, 0x20, 0xE7, 0x0F, 0x12, 0x1D, 0xD7, 0x80,
+	0x0A, 0x12, 0x0E, 0x04, 0x75, 0x83, 0xD2, 0xE0,
+/*0D40*/0x54, 0x7F, 0xF0, 0x12, 0x1D, 0x23, 0x22, 0x74,
+	0x8A, 0x85, 0x08, 0x82, 0xF5, 0x83, 0xE5, 0x17,
+/*0D50*/0xF0, 0x12, 0x0E, 0x3A, 0xE4, 0xF0, 0x90, 0x07,
+	0x02, 0xE0, 0x12, 0x0E, 0x17, 0x75, 0x83, 0x90,
+/*0D60*/0xEF, 0xF0, 0x74, 0x92, 0xFE, 0xE5, 0x08, 0x44,
+	0x07, 0xFF, 0xF5, 0x82, 0x8E, 0x83, 0xE0, 0x54,
+/*0D70*/0xC0, 0xFD, 0x90, 0x07, 0x03, 0xE0, 0x54, 0x3F,
+	0x4D, 0x8F, 0x82, 0x8E, 0x83, 0xF0, 0x90, 0x07,
+/*0D80*/0x04, 0xE0, 0x12, 0x0E, 0x17, 0x75, 0x83, 0x82,
+	0xEF, 0xF0, 0x90, 0x07, 0x05, 0xE0, 0xFF, 0xED,
+/*0D90*/0x44, 0x07, 0xF5, 0x82, 0x75, 0x83, 0xB4, 0xEF,
+	0x12, 0x0E, 0x03, 0x75, 0x83, 0x80, 0xE0, 0x54,
+/*0DA0*/0xBF, 0xF0, 0x30, 0x37, 0x0A, 0x12, 0x0E, 0x91,
+	0x75, 0x83, 0x94, 0xE0, 0x44, 0x80, 0xF0, 0x30,
+/*0DB0*/0x38, 0x0A, 0x12, 0x0E, 0x91, 0x75, 0x83, 0x92,
+	0xE0, 0x44, 0x80, 0xF0, 0xE5, 0x28, 0x30, 0xE4,
+/*0DC0*/0x1A, 0x20, 0x39, 0x0A, 0x12, 0x0E, 0x04, 0x75,
+	0x83, 0x88, 0xE0, 0x54, 0x7F, 0xF0, 0x20, 0x3A,
+/*0DD0*/0x0A, 0x12, 0x0E, 0x04, 0x75, 0x83, 0x88, 0xE0,
+	0x54, 0xBF, 0xF0, 0x74, 0x8C, 0xFE, 0x12, 0x0E,
+/*0DE0*/0x04, 0x8E, 0x83, 0xE0, 0x54, 0x0F, 0x12, 0x0E,
+	0x03, 0x75, 0x83, 0x86, 0xE0, 0x54, 0xBF, 0xF0,
+/*0DF0*/0xE5, 0x08, 0x44, 0x06, 0x12, 0x0D, 0xFD, 0x75,
+	0x83, 0x8A, 0xE4, 0xF0, 0x22, 0xF5, 0x82, 0x75,
+/*0E00*/0x83, 0x82, 0xE4, 0xF0, 0xE5, 0x08, 0x44, 0x07,
+	0xF5, 0x82, 0x22, 0x8E, 0x83, 0xE0, 0xF5, 0x10,
+/*0E10*/0x54, 0xFE, 0xF0, 0xE5, 0x10, 0x44, 0x01, 0xFF,
+	0xE5, 0x08, 0xFD, 0xED, 0x44, 0x07, 0xF5, 0x82,
+/*0E20*/0x22, 0xE5, 0x15, 0xC4, 0x54, 0x07, 0xFF, 0xE5,
+	0x08, 0xFD, 0xED, 0x44, 0x08, 0xF5, 0x82, 0x75,
+/*0E30*/0x83, 0x82, 0x22, 0x75, 0x83, 0x80, 0xE0, 0x44,
+	0x40, 0xF0, 0xE5, 0x08, 0x44, 0x08, 0xF5, 0x82,
+/*0E40*/0x75, 0x83, 0x8A, 0x22, 0xE5, 0x16, 0x25, 0xE0,
+	0x25, 0xE0, 0x24, 0xAF, 0xF5, 0x82, 0xE4, 0x34,
+/*0E50*/0x1A, 0xF5, 0x83, 0xE4, 0x93, 0xF5, 0x0D, 0x22,
+	0x43, 0xE1, 0x10, 0x43, 0xE1, 0x80, 0x53, 0xE1,
+/*0E60*/0xFD, 0x85, 0xE1, 0x10, 0x22, 0xE5, 0x16, 0x25,
+	0xE0, 0x25, 0xE0, 0x24, 0xB2, 0xF5, 0x82, 0xE4,
+/*0E70*/0x34, 0x1A, 0xF5, 0x83, 0xE4, 0x93, 0x22, 0x85,
+	0x55, 0x82, 0x85, 0x54, 0x83, 0xE5, 0x15, 0xF0,
+/*0E80*/0x22, 0xE5, 0xE2, 0x54, 0x20, 0xD3, 0x94, 0x00,
+	0x22, 0xE5, 0xE2, 0x54, 0x40, 0xD3, 0x94, 0x00,
+/*0E90*/0x22, 0xE5, 0x08, 0x44, 0x06, 0xF5, 0x82, 0x22,
+	0xFD, 0xE5, 0x08, 0xFB, 0xEB, 0x44, 0x07, 0xF5,
+/*0EA0*/0x82, 0x22, 0x53, 0xF9, 0xF7, 0x75, 0xFE, 0x30,
+	0x22, 0xEF, 0x4E, 0x70, 0x26, 0x12, 0x07, 0xCC,
+/*0EB0*/0xE0, 0xFD, 0x90, 0x07, 0x26, 0x12, 0x07, 0x7B,
+	0x12, 0x07, 0xD8, 0xE0, 0xFD, 0x90, 0x07, 0x28,
+/*0EC0*/0x12, 0x07, 0x7B, 0x12, 0x08, 0x81, 0x12, 0x07,
+	0x72, 0x12, 0x08, 0x35, 0xE0, 0x90, 0x07, 0x24,
+/*0ED0*/0x12, 0x07, 0x78, 0xEF, 0x64, 0x04, 0x4E, 0x70,
+	0x29, 0x12, 0x07, 0xE4, 0xE0, 0xFD, 0x90, 0x07,
+/*0EE0*/0x26, 0x12, 0x07, 0x7B, 0x12, 0x07, 0xF0, 0xE0,
+	0xFD, 0x90, 0x07, 0x28, 0x12, 0x07, 0x7B, 0x12,
+/*0EF0*/0x08, 0x8B, 0x12, 0x07, 0x72, 0x12, 0x08, 0x41,
+	0xE0, 0x54, 0x1F, 0xFD, 0x90, 0x07, 0x24, 0x12,
+/*0F00*/0x07, 0x7B, 0xEF, 0x64, 0x01, 0x4E, 0x70, 0x04,
+	0x7D, 0x01, 0x80, 0x02, 0x7D, 0x00, 0xEF, 0x64,
+/*0F10*/0x02, 0x4E, 0x70, 0x04, 0x7F, 0x01, 0x80, 0x02,
+	0x7F, 0x00, 0xEF, 0x4D, 0x60, 0x35, 0x12, 0x07,
+/*0F20*/0xFC, 0xE0, 0xFF, 0x90, 0x07, 0x26, 0x12, 0x07,
+	0x89, 0xEF, 0xF0, 0x12, 0x08, 0x08, 0xE0, 0xFF,
+/*0F30*/0x90, 0x07, 0x28, 0x12, 0x07, 0x89, 0xEF, 0xF0,
+	0x12, 0x08, 0x4D, 0xE0, 0x54, 0x1F, 0xFF, 0x12,
+/*0F40*/0x07, 0x86, 0xEF, 0xF0, 0x12, 0x08, 0x59, 0xE0,
+	0x54, 0x1F, 0xFF, 0x90, 0x07, 0x24, 0x12, 0x07,
+/*0F50*/0x89, 0xEF, 0xF0, 0x22, 0xE4, 0xF5, 0x53, 0x12,
+	0x0E, 0x81, 0x40, 0x04, 0x7F, 0x01, 0x80, 0x02,
+/*0F60*/0x7F, 0x00, 0x12, 0x0E, 0x89, 0x40, 0x04, 0x7E,
+	0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE, 0x4F, 0x70,
+/*0F70*/0x03, 0x02, 0x0F, 0xF6, 0x85, 0xE1, 0x10, 0x43,
+	0xE1, 0x02, 0x53, 0xE1, 0x0F, 0x85, 0xE1, 0x10,
+/*0F80*/0xE4, 0xF5, 0x51, 0xE5, 0xE3, 0x54, 0x3F, 0xF5,
+	0x52, 0x12, 0x0E, 0x89, 0x40, 0x1D, 0xAD, 0x52,
+/*0F90*/0xAF, 0x51, 0x12, 0x11, 0x18, 0xEF, 0x60, 0x08,
+	0x85, 0xE1, 0x10, 0x43, 0xE1, 0x40, 0x80, 0x0B,
+/*0FA0*/0x53, 0xE1, 0xBF, 0x12, 0x0E, 0x58, 0x12, 0x00,
+	0x06, 0x80, 0xFB, 0xE5, 0xE3, 0x54, 0x3F, 0xF5,
+/*0FB0*/0x51, 0xE5, 0xE4, 0x54, 0x3F, 0xF5, 0x52, 0x12,
+	0x0E, 0x81, 0x40, 0x1D, 0xAD, 0x52, 0xAF, 0x51,
+/*0FC0*/0x12, 0x11, 0x18, 0xEF, 0x60, 0x08, 0x85, 0xE1,
+	0x10, 0x43, 0xE1, 0x20, 0x80, 0x0B, 0x53, 0xE1,
+/*0FD0*/0xDF, 0x12, 0x0E, 0x58, 0x12, 0x00, 0x06, 0x80,
+	0xFB, 0x12, 0x0E, 0x81, 0x40, 0x04, 0x7F, 0x01,
+/*0FE0*/0x80, 0x02, 0x7F, 0x00, 0x12, 0x0E, 0x89, 0x40,
+	0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE,
+/*0FF0*/0x4F, 0x60, 0x03, 0x12, 0x0E, 0x5B, 0x22, 0x12,
+	0x0E, 0x21, 0xEF, 0xF0, 0x12, 0x10, 0x91, 0x22,
+/*1000*/0x02, 0x11, 0x00, 0x02, 0x10, 0x40, 0x02, 0x10,
+	0x90, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1010*/0x01, 0x20, 0x01, 0x20, 0xE4, 0xF5, 0x57, 0x12,
+	0x16, 0xBD, 0x12, 0x16, 0x44, 0xE4, 0x12, 0x10,
+/*1020*/0x56, 0x12, 0x14, 0xB7, 0x90, 0x07, 0x26, 0x12,
+	0x07, 0x35, 0xE4, 0x12, 0x07, 0x31, 0xE4, 0xF0,
+/*1030*/0x12, 0x10, 0x56, 0x12, 0x14, 0xB7, 0x90, 0x07,
+	0x26, 0x12, 0x07, 0x35, 0xE5, 0x41, 0x12, 0x07,
+/*1040*/0x31, 0xE5, 0x40, 0xF0, 0xAF, 0x57, 0x7E, 0x00,
+	0xAD, 0x56, 0x7C, 0x00, 0x12, 0x04, 0x44, 0xAF,
+/*1050*/0x56, 0x7E, 0x00, 0x02, 0x11, 0xEE, 0xFF, 0x90,
+	0x07, 0x20, 0xA3, 0xE0, 0xFD, 0xE4, 0xF5, 0x56,
+/*1060*/0xF5, 0x40, 0xFE, 0xFC, 0xAB, 0x56, 0xFA, 0x12,
+	0x11, 0x51, 0x7F, 0x0F, 0x7D, 0x18, 0xE4, 0xF5,
+/*1070*/0x56, 0xF5, 0x40, 0xFE, 0xFC, 0xAB, 0x56, 0xFA,
+	0x12, 0x15, 0x41, 0xAF, 0x56, 0x7E, 0x00, 0x12,
+/*1080*/0x1A, 0xFF, 0xE4, 0xFF, 0xF5, 0x56, 0x7D, 0x1F,
+	0xF5, 0x40, 0xFE, 0xFC, 0xAB, 0x56, 0xFA, 0x22,
+/*1090*/0x22, 0xE4, 0xF5, 0x55, 0xE5, 0x08, 0xFD, 0x74,
+	0xA0, 0xF5, 0x56, 0xED, 0x44, 0x07, 0xF5, 0x57,
+/*10A0*/0xE5, 0x28, 0x30, 0xE5, 0x03, 0xD3, 0x80, 0x01,
+	0xC3, 0x40, 0x05, 0x7F, 0x28, 0xEF, 0x80, 0x04,
+/*10B0*/0x7F, 0x14, 0xEF, 0xC3, 0x13, 0xF5, 0x54, 0xE4,
+	0xF9, 0x12, 0x0E, 0x18, 0x75, 0x83, 0x8E, 0xE0,
+/*10C0*/0xF5, 0x10, 0xCE, 0xEF, 0xCE, 0xEE, 0xD3, 0x94,
+	0x00, 0x40, 0x26, 0xE5, 0x10, 0x54, 0xFE, 0x12,
+/*10D0*/0x0E, 0x98, 0x75, 0x83, 0x8E, 0xED, 0xF0, 0xE5,
+	0x10, 0x44, 0x01, 0xFD, 0xEB, 0x44, 0x07, 0xF5,
+/*10E0*/0x82, 0xED, 0xF0, 0x85, 0x57, 0x82, 0x85, 0x56,
+	0x83, 0xE0, 0x30, 0xE3, 0x01, 0x09, 0x1E, 0x80,
+/*10F0*/0xD4, 0xC2, 0x34, 0xE9, 0xC3, 0x95, 0x54, 0x40,
+	0x02, 0xD2, 0x34, 0x22, 0x02, 0x00, 0x06, 0x22,
+/*1100*/0x30, 0x30, 0x11, 0x90, 0x10, 0x00, 0xE4, 0x93,
+	0xF5, 0x10, 0x90, 0x10, 0x10, 0xE4, 0x93, 0xF5,
+/*1110*/0x10, 0x12, 0x10, 0x90, 0x12, 0x11, 0x50, 0x22,
+	0xE4, 0xFC, 0xC3, 0xED, 0x9F, 0xFA, 0xEF, 0xF5,
+/*1120*/0x83, 0x75, 0x82, 0x00, 0x79, 0xFF, 0xE4, 0x93,
+	0xCC, 0x6C, 0xCC, 0xA3, 0xD9, 0xF8, 0xDA, 0xF6,
+/*1130*/0xE5, 0xE2, 0x30, 0xE4, 0x02, 0x8C, 0xE5, 0xED,
+	0x24, 0xFF, 0xFF, 0xEF, 0x75, 0x82, 0xFF, 0xF5,
+/*1140*/0x83, 0xE4, 0x93, 0x6C, 0x70, 0x03, 0x7F, 0x01,
+	0x22, 0x7F, 0x00, 0x22, 0x22, 0x11, 0x00, 0x00,
+/*1150*/0x22, 0x8E, 0x58, 0x8F, 0x59, 0x8C, 0x5A, 0x8D,
+	0x5B, 0x8A, 0x5C, 0x8B, 0x5D, 0x75, 0x5E, 0x01,
+/*1160*/0xE4, 0xF5, 0x5F, 0xF5, 0x60, 0xF5, 0x62, 0x12,
+	0x07, 0x2A, 0x75, 0x83, 0xD0, 0xE0, 0xFF, 0xC4,
+/*1170*/0x54, 0x0F, 0xF5, 0x61, 0x12, 0x1E, 0xA5, 0x85,
+	0x59, 0x5E, 0xD3, 0xE5, 0x5E, 0x95, 0x5B, 0xE5,
+/*1180*/0x5A, 0x12, 0x07, 0x6B, 0x50, 0x4B, 0x12, 0x07,
+	0x03, 0x75, 0x83, 0xBC, 0xE0, 0x45, 0x5E, 0x12,
+/*1190*/0x07, 0x29, 0x75, 0x83, 0xBE, 0xE0, 0x45, 0x5E,
+	0x12, 0x07, 0x29, 0x75, 0x83, 0xC0, 0xE0, 0x45,
+/*11A0*/0x5E, 0xF0, 0xAF, 0x5F, 0xE5, 0x60, 0x12, 0x08,
+	0x78, 0x12, 0x0A, 0xFF, 0xAF, 0x62, 0x7E, 0x00,
+/*11B0*/0xAD, 0x5D, 0xAC, 0x5C, 0x12, 0x04, 0x44, 0xE5,
+	0x61, 0xAF, 0x5E, 0x7E, 0x00, 0xB4, 0x03, 0x05,
+/*11C0*/0x12, 0x1E, 0x21, 0x80, 0x07, 0xAD, 0x5D, 0xAC,
+	0x5C, 0x12, 0x13, 0x17, 0x05, 0x5E, 0x02, 0x11,
+/*11D0*/0x7A, 0x12, 0x07, 0x03, 0x75, 0x83, 0xBC, 0xE0,
+	0x45, 0x40, 0x12, 0x07, 0x29, 0x75, 0x83, 0xBE,
+/*11E0*/0xE0, 0x45, 0x40, 0x12, 0x07, 0x29, 0x75, 0x83,
+	0xC0, 0xE0, 0x45, 0x40, 0xF0, 0x22, 0x8E, 0x58,
+/*11F0*/0x8F, 0x59, 0x75, 0x5A, 0x01, 0x79, 0x01, 0x75,
+	0x5B, 0x01, 0xE4, 0xFB, 0x12, 0x07, 0x2A, 0x75,
+/*1200*/0x83, 0xAE, 0xE0, 0x54, 0x1A, 0xFF, 0x12, 0x08,
+	0x65, 0xE0, 0xC4, 0x13, 0x54, 0x07, 0xFE, 0xEF,
+/*1210*/0x70, 0x0C, 0xEE, 0x65, 0x35, 0x70, 0x07, 0x90,
+	0x07, 0x2F, 0xE0, 0xB4, 0x01, 0x0D, 0xAF, 0x35,
+/*1220*/0x7E, 0x00, 0x12, 0x0E, 0xA9, 0xCF, 0xEB, 0xCF,
+	0x02, 0x1E, 0x60, 0xE5, 0x59, 0x64, 0x02, 0x45,
+/*1230*/0x58, 0x70, 0x04, 0x7F, 0x01, 0x80, 0x02, 0x7F,
+	0x00, 0xE5, 0x59, 0x45, 0x58, 0x70, 0x04, 0x7E,
+/*1240*/0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE, 0x4F, 0x60,
+	0x23, 0x85, 0x41, 0x49, 0x85, 0x40, 0x4B, 0xE5,
+/*1250*/0x59, 0x45, 0x58, 0x70, 0x2C, 0xAF, 0x5A, 0xFE,
+	0xCD, 0xE9, 0xCD, 0xFC, 0xAB, 0x59, 0xAA, 0x58,
+/*1260*/0x12, 0x0A, 0xFF, 0xAF, 0x5B, 0x7E, 0x00, 0x12,
+	0x1E, 0x60, 0x80, 0x15, 0xAF, 0x5B, 0x7E, 0x00,
+/*1270*/0x12, 0x1E, 0x60, 0x90, 0x07, 0x26, 0x12, 0x07,
+	0x35, 0xE5, 0x49, 0x12, 0x07, 0x31, 0xE5, 0x4B,
+/*1280*/0xF0, 0xE4, 0xFD, 0xAF, 0x35, 0xFE, 0xFC, 0x12,
+	0x09, 0x15, 0x22, 0x8C, 0x64, 0x8D, 0x65, 0x12,
+/*1290*/0x08, 0xDA, 0x40, 0x3C, 0xE5, 0x65, 0x45, 0x64,
+	0x70, 0x10, 0x12, 0x09, 0x04, 0xC3, 0xE5, 0x3E,
+/*12A0*/0x12, 0x07, 0x69, 0x40, 0x3B, 0x12, 0x08, 0x95,
+	0x80, 0x18, 0xE5, 0x3E, 0xC3, 0x95, 0x38, 0x40,
+/*12B0*/0x1D, 0x85, 0x3E, 0x38, 0xE5, 0x3E, 0x60, 0x05,
+	0x85, 0x3F, 0x39, 0x80, 0x03, 0x85, 0x39, 0x39,
+/*12C0*/0x8F, 0x3A, 0x12, 0x07, 0xA8, 0xE5, 0x3E, 0x12,
+	0x07, 0x53, 0xE5, 0x3F, 0xF0, 0x22, 0x80, 0x3B,
+/*12D0*/0xE5, 0x65, 0x45, 0x64, 0x70, 0x11, 0x12, 0x07,
+	0x5F, 0x40, 0x05, 0x12, 0x08, 0x9E, 0x80, 0x1F,
+/*12E0*/0x12, 0x07, 0x3E, 0xE5, 0x41, 0xF0, 0x22, 0xE5,
+	0x3C, 0xC3, 0x95, 0x38, 0x40, 0x1D, 0x85, 0x3C,
+/*12F0*/0x38, 0xE5, 0x3C, 0x60, 0x05, 0x85, 0x3D, 0x39,
+	0x80, 0x03, 0x85, 0x39, 0x39, 0x8F, 0x3A, 0x12,
+/*1300*/0x07, 0xA8, 0xE5, 0x3C, 0x12, 0x07, 0x53, 0xE5,
+	0x3D, 0xF0, 0x22, 0x12, 0x07, 0x9F, 0xE5, 0x38,
+/*1310*/0x12, 0x07, 0x53, 0xE5, 0x39, 0xF0, 0x22, 0x8C,
+	0x63, 0x8D, 0x64, 0x12, 0x08, 0xDA, 0x40, 0x3C,
+/*1320*/0xE5, 0x64, 0x45, 0x63, 0x70, 0x10, 0x12, 0x09,
+	0x04, 0xC3, 0xE5, 0x3E, 0x12, 0x07, 0x69, 0x40,
+/*1330*/0x3B, 0x12, 0x08, 0x95, 0x80, 0x18, 0xE5, 0x3E,
+	0xC3, 0x95, 0x38, 0x40, 0x1D, 0x85, 0x3E, 0x38,
+/*1340*/0xE5, 0x3E, 0x60, 0x05, 0x85, 0x3F, 0x39, 0x80,
+	0x03, 0x85, 0x39, 0x39, 0x8F, 0x3A, 0x12, 0x07,
+/*1350*/0xA8, 0xE5, 0x3E, 0x12, 0x07, 0x53, 0xE5, 0x3F,
+	0xF0, 0x22, 0x80, 0x3B, 0xE5, 0x64, 0x45, 0x63,
+/*1360*/0x70, 0x11, 0x12, 0x07, 0x5F, 0x40, 0x05, 0x12,
+	0x08, 0x9E, 0x80, 0x1F, 0x12, 0x07, 0x3E, 0xE5,
+/*1370*/0x41, 0xF0, 0x22, 0xE5, 0x3C, 0xC3, 0x95, 0x38,
+	0x40, 0x1D, 0x85, 0x3C, 0x38, 0xE5, 0x3C, 0x60,
+/*1380*/0x05, 0x85, 0x3D, 0x39, 0x80, 0x03, 0x85, 0x39,
+	0x39, 0x8F, 0x3A, 0x12, 0x07, 0xA8, 0xE5, 0x3C,
+/*1390*/0x12, 0x07, 0x53, 0xE5, 0x3D, 0xF0, 0x22, 0x12,
+	0x07, 0x9F, 0xE5, 0x38, 0x12, 0x07, 0x53, 0xE5,
+/*13A0*/0x39, 0xF0, 0x22, 0xE5, 0x0D, 0xFE, 0xE5, 0x08,
+	0x8E, 0x54, 0x44, 0x05, 0xF5, 0x55, 0x75, 0x15,
+/*13B0*/0x0F, 0xF5, 0x82, 0x12, 0x0E, 0x7A, 0x12, 0x17,
+	0xA3, 0x20, 0x31, 0x05, 0x75, 0x15, 0x03, 0x80,
+/*13C0*/0x03, 0x75, 0x15, 0x0B, 0xE5, 0x0A, 0xC3, 0x94,
+	0x01, 0x50, 0x38, 0x12, 0x14, 0x20, 0x20, 0x31,
+/*13D0*/0x06, 0x05, 0x15, 0x05, 0x15, 0x80, 0x04, 0x15,
+	0x15, 0x15, 0x15, 0xE5, 0x0A, 0xC3, 0x94, 0x01,
+/*13E0*/0x50, 0x21, 0x12, 0x14, 0x20, 0x20, 0x31, 0x04,
+	0x05, 0x15, 0x80, 0x02, 0x15, 0x15, 0xE5, 0x0A,
+/*13F0*/0xC3, 0x94, 0x01, 0x50, 0x0E, 0x12, 0x0E, 0x77,
+	0x12, 0x17, 0xA3, 0x20, 0x31, 0x05, 0x05, 0x15,
+/*1400*/0x12, 0x0E, 0x77, 0xE5, 0x15, 0xB4, 0x08, 0x04,
+	0x7F, 0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5, 0x15,
+/*1410*/0xB4, 0x07, 0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E,
+	0x00, 0xEE, 0x4F, 0x60, 0x02, 0x05, 0x7F, 0x22,
+/*1420*/0x85, 0x55, 0x82, 0x85, 0x54, 0x83, 0xE5, 0x15,
+	0xF0, 0x12, 0x17, 0xA3, 0x22, 0x12, 0x07, 0x2A,
+/*1430*/0x75, 0x83, 0xAE, 0x74, 0xFF, 0x12, 0x07, 0x29,
+	0xE0, 0x54, 0x1A, 0xF5, 0x34, 0xE0, 0xC4, 0x13,
+/*1440*/0x54, 0x07, 0xF5, 0x35, 0x24, 0xFE, 0x60, 0x24,
+	0x24, 0xFE, 0x60, 0x3C, 0x24, 0x04, 0x70, 0x63,
+/*1450*/0x75, 0x31, 0x2D, 0xE5, 0x08, 0xFD, 0x74, 0xB6,
+	0x12, 0x07, 0x92, 0x74, 0xBC, 0x90, 0x07, 0x22,
+/*1460*/0x12, 0x07, 0x95, 0x74, 0x90, 0x12, 0x07, 0xB3,
+	0x74, 0x92, 0x80, 0x3C, 0x75, 0x31, 0x3A, 0xE5,
+/*1470*/0x08, 0xFD, 0x74, 0xBA, 0x12, 0x07, 0x92, 0x74,
+	0xC0, 0x90, 0x07, 0x22, 0x12, 0x07, 0xB6, 0x74,
+/*1480*/0xC4, 0x12, 0x07, 0xB3, 0x74, 0xC8, 0x80, 0x20,
+	0x75, 0x31, 0x35, 0xE5, 0x08, 0xFD, 0x74, 0xB8,
+/*1490*/0x12, 0x07, 0x92, 0x74, 0xBE, 0xFF, 0xED, 0x44,
+	0x07, 0x90, 0x07, 0x22, 0xCF, 0xF0, 0xA3, 0xEF,
+/*14A0*/0xF0, 0x74, 0xC2, 0x12, 0x07, 0xB3, 0x74, 0xC6,
+	0xFF, 0xED, 0x44, 0x07, 0xA3, 0xCF, 0xF0, 0xA3,
+/*14B0*/0xEF, 0xF0, 0x22, 0x75, 0x34, 0x01, 0x22, 0x8E,
+	0x58, 0x8F, 0x59, 0x8C, 0x5A, 0x8D, 0x5B, 0x8A,
+/*14C0*/0x5C, 0x8B, 0x5D, 0x75, 0x5E, 0x01, 0xE4, 0xF5,
+	0x5F, 0x12, 0x1E, 0xA5, 0x85, 0x59, 0x5E, 0xD3,
+/*14D0*/0xE5, 0x5E, 0x95, 0x5B, 0xE5, 0x5A, 0x12, 0x07,
+	0x6B, 0x50, 0x57, 0xE5, 0x5D, 0x45, 0x5C, 0x70,
+/*14E0*/0x30, 0x12, 0x07, 0x2A, 0x75, 0x83, 0x92, 0xE5,
+	0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC6, 0xE5,
+/*14F0*/0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC8, 0xE5,
+	0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0x90, 0xE5,
+/*1500*/0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC2, 0xE5,
+	0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC4, 0x80,
+/*1510*/0x03, 0x12, 0x07, 0x32, 0xE5, 0x5E, 0xF0, 0xAF,
+	0x5F, 0x7E, 0x00, 0xAD, 0x5D, 0xAC, 0x5C, 0x12,
+/*1520*/0x04, 0x44, 0xAF, 0x5E, 0x7E, 0x00, 0xAD, 0x5D,
+	0xAC, 0x5C, 0x12, 0x0B, 0xD1, 0x05, 0x5E, 0x02,
+/*1530*/0x14, 0xCF, 0xAB, 0x5D, 0xAA, 0x5C, 0xAD, 0x5B,
+	0xAC, 0x5A, 0xAF, 0x59, 0xAE, 0x58, 0x02, 0x1B,
+/*1540*/0xFB, 0x8C, 0x5C, 0x8D, 0x5D, 0x8A, 0x5E, 0x8B,
+	0x5F, 0x75, 0x60, 0x01, 0xE4, 0xF5, 0x61, 0xF5,
+/*1550*/0x62, 0xF5, 0x63, 0x12, 0x1E, 0xA5, 0x8F, 0x60,
+	0xD3, 0xE5, 0x60, 0x95, 0x5D, 0xE5, 0x5C, 0x12,
+/*1560*/0x07, 0x6B, 0x50, 0x61, 0xE5, 0x5F, 0x45, 0x5E,
+	0x70, 0x27, 0x12, 0x07, 0x2A, 0x75, 0x83, 0xB6,
+/*1570*/0xE5, 0x60, 0x12, 0x07, 0x29, 0x75, 0x83, 0xB8,
+	0xE5, 0x60, 0x12, 0x07, 0x29, 0x75, 0x83, 0xBA,
+/*1580*/0xE5, 0x60, 0xF0, 0xAF, 0x61, 0x7E, 0x00, 0xE5,
+	0x62, 0x12, 0x08, 0x7A, 0x12, 0x0A, 0xFF, 0x80,
+/*1590*/0x19, 0x90, 0x07, 0x24, 0x12, 0x07, 0x35, 0xE5,
+	0x60, 0x12, 0x07, 0x29, 0x75, 0x83, 0x8E, 0xE4,
+/*15A0*/0x12, 0x07, 0x29, 0x74, 0x01, 0x12, 0x07, 0x29,
+	0xE4, 0xF0, 0xAF, 0x63, 0x7E, 0x00, 0xAD, 0x5F,
+/*15B0*/0xAC, 0x5E, 0x12, 0x04, 0x44, 0xAF, 0x60, 0x7E,
+	0x00, 0xAD, 0x5F, 0xAC, 0x5E, 0x12, 0x12, 0x8B,
+/*15C0*/0x05, 0x60, 0x02, 0x15, 0x58, 0x22, 0x90, 0x11,
+	0x4D, 0xE4, 0x93, 0x90, 0x07, 0x2E, 0xF0, 0x12,
+/*15D0*/0x08, 0x1F, 0x75, 0x83, 0xAE, 0xE0, 0x54, 0x1A,
+	0xF5, 0x34, 0x70, 0x67, 0xEF, 0x44, 0x07, 0xF5,
+/*15E0*/0x82, 0x75, 0x83, 0xCE, 0xE0, 0xFF, 0x13, 0x13,
+	0x13, 0x54, 0x07, 0xF5, 0x36, 0x54, 0x0F, 0xD3,
+/*15F0*/0x94, 0x00, 0x40, 0x06, 0x12, 0x14, 0x2D, 0x12,
+	0x1B, 0xA9, 0xE5, 0x36, 0x54, 0x0F, 0x24, 0xFE,
+/*1600*/0x60, 0x0C, 0x14, 0x60, 0x0C, 0x14, 0x60, 0x19,
+	0x24, 0x03, 0x70, 0x37, 0x80, 0x10, 0x02, 0x1E,
+/*1610*/0x91, 0x12, 0x1E, 0x91, 0x12, 0x07, 0x2A, 0x75,
+	0x83, 0xCE, 0xE0, 0x54, 0xEF, 0xF0, 0x02, 0x1D,
+/*1620*/0xAE, 0x12, 0x10, 0x14, 0xE4, 0xF5, 0x55, 0x12,
+	0x1D, 0x85, 0x05, 0x55, 0xE5, 0x55, 0xC3, 0x94,
+/*1630*/0x05, 0x40, 0xF4, 0x12, 0x07, 0x2A, 0x75, 0x83,
+	0xCE, 0xE0, 0x54, 0xC7, 0x12, 0x07, 0x29, 0xE0,
+/*1640*/0x44, 0x08, 0xF0, 0x22, 0xE4, 0xF5, 0x58, 0xF5,
+	0x59, 0xAF, 0x08, 0xEF, 0x44, 0x07, 0xF5, 0x82,
+/*1650*/0x75, 0x83, 0xD0, 0xE0, 0xFD, 0xC4, 0x54, 0x0F,
+	0xF5, 0x5A, 0xEF, 0x44, 0x07, 0xF5, 0x82, 0x75,
+/*1660*/0x83, 0x80, 0x74, 0x01, 0xF0, 0x12, 0x08, 0x21,
+	0x75, 0x83, 0x82, 0xE5, 0x45, 0xF0, 0xEF, 0x44,
+/*1670*/0x07, 0xF5, 0x82, 0x75, 0x83, 0x8A, 0x74, 0xFF,
+	0xF0, 0x12, 0x1A, 0x4D, 0x12, 0x07, 0x2A, 0x75,
+/*1680*/0x83, 0xBC, 0xE0, 0x54, 0xEF, 0x12, 0x07, 0x29,
+	0x75, 0x83, 0xBE, 0xE0, 0x54, 0xEF, 0x12, 0x07,
+/*1690*/0x29, 0x75, 0x83, 0xC0, 0xE0, 0x54, 0xEF, 0x12,
+	0x07, 0x29, 0x75, 0x83, 0xBC, 0xE0, 0x44, 0x10,
+/*16A0*/0x12, 0x07, 0x29, 0x75, 0x83, 0xBE, 0xE0, 0x44,
+	0x10, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC0, 0xE0,
+/*16B0*/0x44, 0x10, 0xF0, 0xAF, 0x58, 0xE5, 0x59, 0x12,
+	0x08, 0x78, 0x02, 0x0A, 0xFF, 0xE4, 0xF5, 0x58,
+/*16C0*/0x7D, 0x01, 0xF5, 0x59, 0xAF, 0x35, 0xFE, 0xFC,
+	0x12, 0x09, 0x15, 0x12, 0x07, 0x2A, 0x75, 0x83,
+/*16D0*/0xB6, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83,
+	0xB8, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83,
+/*16E0*/0xBA, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83,
+	0xBC, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83,
+/*16F0*/0xBE, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83,
+	0xC0, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83,
+/*1700*/0x90, 0xE4, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC2,
+	0xE4, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC4, 0xE4,
+/*1710*/0x12, 0x07, 0x29, 0x75, 0x83, 0x92, 0xE4, 0x12,
+	0x07, 0x29, 0x75, 0x83, 0xC6, 0xE4, 0x12, 0x07,
+/*1720*/0x29, 0x75, 0x83, 0xC8, 0xE4, 0xF0, 0xAF, 0x58,
+	0xFE, 0xE5, 0x59, 0x12, 0x08, 0x7A, 0x02, 0x0A,
+/*1730*/0xFF, 0xE5, 0xE2, 0x30, 0xE4, 0x6C, 0xE5, 0xE7,
+	0x54, 0xC0, 0x64, 0x40, 0x70, 0x64, 0xE5, 0x09,
+/*1740*/0xC4, 0x54, 0x30, 0xFE, 0xE5, 0x08, 0x25, 0xE0,
+	0x25, 0xE0, 0x54, 0xC0, 0x4E, 0xFE, 0xEF, 0x54,
+/*1750*/0x3F, 0x4E, 0xFD, 0xE5, 0x2B, 0xAE, 0x2A, 0x78,
+	0x02, 0xC3, 0x33, 0xCE, 0x33, 0xCE, 0xD8, 0xF9,
+/*1760*/0xF5, 0x82, 0x8E, 0x83, 0xED, 0xF0, 0xE5, 0x2B,
+	0xAE, 0x2A, 0x78, 0x02, 0xC3, 0x33, 0xCE, 0x33,
+/*1770*/0xCE, 0xD8, 0xF9, 0xFF, 0xF5, 0x82, 0x8E, 0x83,
+	0xA3, 0xE5, 0xFE, 0xF0, 0x8F, 0x82, 0x8E, 0x83,
+/*1780*/0xA3, 0xA3, 0xE5, 0xFD, 0xF0, 0x8F, 0x82, 0x8E,
+	0x83, 0xA3, 0xA3, 0xA3, 0xE5, 0xFC, 0xF0, 0xC3,
+/*1790*/0xE5, 0x2B, 0x94, 0xFA, 0xE5, 0x2A, 0x94, 0x00,
+	0x50, 0x08, 0x05, 0x2B, 0xE5, 0x2B, 0x70, 0x02,
+/*17A0*/0x05, 0x2A, 0x22, 0xE4, 0xFF, 0xE4, 0xF5, 0x58,
+	0xF5, 0x56, 0xF5, 0x57, 0x74, 0x82, 0xFC, 0x12,
+/*17B0*/0x0E, 0x04, 0x8C, 0x83, 0xE0, 0xF5, 0x10, 0x54,
+	0x7F, 0xF0, 0xE5, 0x10, 0x44, 0x80, 0x12, 0x0E,
+/*17C0*/0x98, 0xED, 0xF0, 0x7E, 0x0A, 0x12, 0x0E, 0x04,
+	0x75, 0x83, 0xA0, 0xE0, 0x20, 0xE0, 0x26, 0xDE,
+/*17D0*/0xF4, 0x05, 0x57, 0xE5, 0x57, 0x70, 0x02, 0x05,
+	0x56, 0xE5, 0x14, 0x24, 0x01, 0xFD, 0xE4, 0x33,
+/*17E0*/0xFC, 0xD3, 0xE5, 0x57, 0x9D, 0xE5, 0x56, 0x9C,
+	0x40, 0xD9, 0xE5, 0x0A, 0x94, 0x20, 0x50, 0x02,
+/*17F0*/0x05, 0x0A, 0x43, 0xE1, 0x08, 0xC2, 0x31, 0x12,
+	0x0E, 0x04, 0x75, 0x83, 0xA6, 0xE0, 0x55, 0x12,
+/*1800*/0x65, 0x12, 0x70, 0x03, 0xD2, 0x31, 0x22, 0xC2,
+	0x31, 0x22, 0x90, 0x07, 0x26, 0xE0, 0xFA, 0xA3,
+/*1810*/0xE0, 0xF5, 0x82, 0x8A, 0x83, 0xE0, 0xF5, 0x41,
+	0xE5, 0x39, 0xC3, 0x95, 0x41, 0x40, 0x26, 0xE5,
+/*1820*/0x39, 0x95, 0x41, 0xC3, 0x9F, 0xEE, 0x12, 0x07,
+	0x6B, 0x40, 0x04, 0x7C, 0x01, 0x80, 0x02, 0x7C,
+/*1830*/0x00, 0xE5, 0x41, 0x64, 0x3F, 0x60, 0x04, 0x7B,
+	0x01, 0x80, 0x02, 0x7B, 0x00, 0xEC, 0x5B, 0x60,
+/*1840*/0x29, 0x05, 0x41, 0x80, 0x28, 0xC3, 0xE5, 0x41,
+	0x95, 0x39, 0xC3, 0x9F, 0xEE, 0x12, 0x07, 0x6B,
+/*1850*/0x40, 0x04, 0x7F, 0x01, 0x80, 0x02, 0x7F, 0x00,
+	0xE5, 0x41, 0x60, 0x04, 0x7E, 0x01, 0x80, 0x02,
+/*1860*/0x7E, 0x00, 0xEF, 0x5E, 0x60, 0x04, 0x15, 0x41,
+	0x80, 0x03, 0x85, 0x39, 0x41, 0x85, 0x3A, 0x40,
+/*1870*/0x22, 0xE5, 0xE2, 0x30, 0xE4, 0x60, 0xE5, 0xE1,
+	0x30, 0xE2, 0x5B, 0xE5, 0x09, 0x70, 0x04, 0x7F,
+/*1880*/0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5, 0x08, 0x70,
+	0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE,
+/*1890*/0x5F, 0x60, 0x43, 0x53, 0xF9, 0xF8, 0xE5, 0xE2,
+	0x30, 0xE4, 0x3B, 0xE5, 0xE1, 0x30, 0xE2, 0x2E,
+/*18A0*/0x43, 0xFA, 0x02, 0x53, 0xFA, 0xFB, 0xE4, 0xF5,
+	0x10, 0x90, 0x94, 0x70, 0xE5, 0x10, 0xF0, 0xE5,
+/*18B0*/0xE1, 0x30, 0xE2, 0xE7, 0x90, 0x94, 0x70, 0xE0,
+	0x65, 0x10, 0x60, 0x03, 0x43, 0xFA, 0x04, 0x05,
+/*18C0*/0x10, 0x90, 0x94, 0x70, 0xE5, 0x10, 0xF0, 0x70,
+	0xE6, 0x12, 0x00, 0x06, 0x80, 0xE1, 0x53, 0xFA,
+/*18D0*/0xFD, 0x53, 0xFA, 0xFB, 0x80, 0xC0, 0x22, 0x8F,
+	0x54, 0x12, 0x00, 0x06, 0xE5, 0xE1, 0x30, 0xE0,
+/*18E0*/0x04, 0x7F, 0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5,
+	0x7E, 0xD3, 0x94, 0x05, 0x40, 0x04, 0x7E, 0x01,
+/*18F0*/0x80, 0x02, 0x7E, 0x00, 0xEE, 0x4F, 0x60, 0x3D,
+	0x85, 0x54, 0x11, 0xE5, 0xE2, 0x20, 0xE1, 0x32,
+/*1900*/0x74, 0xCE, 0x12, 0x1A, 0x05, 0x30, 0xE7, 0x04,
+	0x7D, 0x01, 0x80, 0x02, 0x7D, 0x00, 0x8F, 0x82,
+/*1910*/0x8E, 0x83, 0xE0, 0x30, 0xE6, 0x04, 0x7F, 0x01,
+	0x80, 0x02, 0x7F, 0x00, 0xEF, 0x5D, 0x70, 0x15,
+/*1920*/0x12, 0x15, 0xC6, 0x74, 0xCE, 0x12, 0x1A, 0x05,
+	0x30, 0xE6, 0x07, 0xE0, 0x44, 0x80, 0xF0, 0x43,
+/*1930*/0xF9, 0x80, 0x12, 0x18, 0x71, 0x22, 0x12, 0x0E,
+	0x44, 0xE5, 0x16, 0x25, 0xE0, 0x25, 0xE0, 0x24,
+/*1940*/0xB0, 0xF5, 0x82, 0xE4, 0x34, 0x1A, 0xF5, 0x83,
+	0xE4, 0x93, 0xF5, 0x0F, 0xE5, 0x16, 0x25, 0xE0,
+/*1950*/0x25, 0xE0, 0x24, 0xB1, 0xF5, 0x82, 0xE4, 0x34,
+	0x1A, 0xF5, 0x83, 0xE4, 0x93, 0xF5, 0x0E, 0x12,
+/*1960*/0x0E, 0x65, 0xF5, 0x10, 0xE5, 0x0F, 0x54, 0xF0,
+	0x12, 0x0E, 0x17, 0x75, 0x83, 0x8C, 0xEF, 0xF0,
+/*1970*/0xE5, 0x0F, 0x30, 0xE0, 0x0C, 0x12, 0x0E, 0x04,
+	0x75, 0x83, 0x86, 0xE0, 0x44, 0x40, 0xF0, 0x80,
+/*1980*/0x0A, 0x12, 0x0E, 0x04, 0x75, 0x83, 0x86, 0xE0,
+	0x54, 0xBF, 0xF0, 0x12, 0x0E, 0x91, 0x75, 0x83,
+/*1990*/0x82, 0xE5, 0x0E, 0xF0, 0x22, 0x7F, 0x05, 0x12,
+	0x17, 0x31, 0x12, 0x0E, 0x04, 0x12, 0x0E, 0x33,
+/*19A0*/0x74, 0x02, 0xF0, 0x74, 0x8E, 0xFE, 0x12, 0x0E,
+	0x04, 0x12, 0x0E, 0x0B, 0xEF, 0xF0, 0x75, 0x15,
+/*19B0*/0x70, 0x12, 0x0F, 0xF7, 0x20, 0x34, 0x05, 0x75,
+	0x15, 0x10, 0x80, 0x03, 0x75, 0x15, 0x50, 0x12,
+/*19C0*/0x0F, 0xF7, 0x20, 0x34, 0x04, 0x74, 0x10, 0x80,
+	0x02, 0x74, 0xF0, 0x25, 0x15, 0xF5, 0x15, 0x12,
+/*19D0*/0x0E, 0x21, 0xEF, 0xF0, 0x12, 0x10, 0x91, 0x20,
+	0x34, 0x17, 0xE5, 0x15, 0x64, 0x30, 0x60, 0x0C,
+/*19E0*/0x74, 0x10, 0x25, 0x15, 0xF5, 0x15, 0xB4, 0x80,
+	0x03, 0xE4, 0xF5, 0x15, 0x12, 0x0E, 0x21, 0xEF,
+/*19F0*/0xF0, 0x22, 0xF0, 0xE5, 0x0B, 0x25, 0xE0, 0x25,
+	0xE0, 0x24, 0x82, 0xF5, 0x82, 0xE4, 0x34, 0x07,
+/*1A00*/0xF5, 0x83, 0x22, 0x74, 0x88, 0xFE, 0xE5, 0x08,
+	0x44, 0x07, 0xFF, 0xF5, 0x82, 0x8E, 0x83, 0xE0,
+/*1A10*/0x22, 0xF0, 0xE5, 0x08, 0x44, 0x07, 0xF5, 0x82,
+	0x22, 0xF0, 0xE0, 0x54, 0xC0, 0x8F, 0x82, 0x8E,
+/*1A20*/0x83, 0xF0, 0x22, 0xEF, 0x44, 0x07, 0xF5, 0x82,
+	0x75, 0x83, 0x86, 0xE0, 0x54, 0x10, 0xD3, 0x94,
+/*1A30*/0x00, 0x22, 0xF0, 0x90, 0x07, 0x15, 0xE0, 0x04,
+	0xF0, 0x22, 0x44, 0x06, 0xF5, 0x82, 0x75, 0x83,
+/*1A40*/0x9E, 0xE0, 0x22, 0xFE, 0xEF, 0x44, 0x07, 0xF5,
+	0x82, 0x8E, 0x83, 0xE0, 0x22, 0xE4, 0x90, 0x07,
+/*1A50*/0x2A, 0xF0, 0xA3, 0xF0, 0x12, 0x07, 0x2A, 0x75,
+	0x83, 0x82, 0xE0, 0x54, 0x7F, 0x12, 0x07, 0x29,
+/*1A60*/0xE0, 0x44, 0x80, 0xF0, 0x12, 0x10, 0xFC, 0x12,
+	0x08, 0x1F, 0x75, 0x83, 0xA0, 0xE0, 0x20, 0xE0,
+/*1A70*/0x1A, 0x90, 0x07, 0x2B, 0xE0, 0x04, 0xF0, 0x70,
+	0x06, 0x90, 0x07, 0x2A, 0xE0, 0x04, 0xF0, 0x90,
+/*1A80*/0x07, 0x2A, 0xE0, 0xB4, 0x10, 0xE1, 0xA3, 0xE0,
+	0xB4, 0x00, 0xDC, 0xEE, 0x44, 0xA6, 0xFC, 0xEF,
+/*1A90*/0x44, 0x07, 0xF5, 0x82, 0x8C, 0x83, 0xE0, 0xF5,
+	0x32, 0xEE, 0x44, 0xA8, 0xFE, 0xEF, 0x44, 0x07,
+/*1AA0*/0xF5, 0x82, 0x8E, 0x83, 0xE0, 0xF5, 0x33, 0x22,
+	0x01, 0x20, 0x11, 0x00, 0x04, 0x20, 0x00, 0x90,
+/*1AB0*/0x00, 0x20, 0x0F, 0x92, 0x00, 0x21, 0x0F, 0x94,
+	0x00, 0x22, 0x0F, 0x96, 0x00, 0x23, 0x0F, 0x98,
+/*1AC0*/0x00, 0x24, 0x0F, 0x9A, 0x00, 0x25, 0x0F, 0x9C,
+	0x00, 0x26, 0x0F, 0x9E, 0x00, 0x27, 0x0F, 0xA0,
+/*1AD0*/0x01, 0x20, 0x01, 0xA2, 0x01, 0x21, 0x01, 0xA4,
+	0x01, 0x22, 0x01, 0xA6, 0x01, 0x23, 0x01, 0xA8,
+/*1AE0*/0x01, 0x24, 0x01, 0xAA, 0x01, 0x25, 0x01, 0xAC,
+	0x01, 0x26, 0x01, 0xAE, 0x01, 0x27, 0x01, 0xB0,
+/*1AF0*/0x01, 0x28, 0x01, 0xB4, 0x00, 0x28, 0x0F, 0xB6,
+	0x40, 0x28, 0x0F, 0xB8, 0x61, 0x28, 0x01, 0xCB,
+/*1B00*/0xEF, 0xCB, 0xCA, 0xEE, 0xCA, 0x7F, 0x01, 0xE4,
+	0xFD, 0xEB, 0x4A, 0x70, 0x24, 0xE5, 0x08, 0xF5,
+/*1B10*/0x82, 0x74, 0xB6, 0x12, 0x08, 0x29, 0xE5, 0x08,
+	0xF5, 0x82, 0x74, 0xB8, 0x12, 0x08, 0x29, 0xE5,
+/*1B20*/0x08, 0xF5, 0x82, 0x74, 0xBA, 0x12, 0x08, 0x29,
+	0x7E, 0x00, 0x7C, 0x00, 0x12, 0x0A, 0xFF, 0x80,
+/*1B30*/0x12, 0x90, 0x07, 0x26, 0x12, 0x07, 0x35, 0xE5,
+	0x41, 0xF0, 0x90, 0x07, 0x24, 0x12, 0x07, 0x35,
+/*1B40*/0xE5, 0x40, 0xF0, 0x12, 0x07, 0x2A, 0x75, 0x83,
+	0x8E, 0xE4, 0x12, 0x07, 0x29, 0x74, 0x01, 0x12,
+/*1B50*/0x07, 0x29, 0xE4, 0xF0, 0x22, 0xE4, 0xF5, 0x26,
+	0xF5, 0x27, 0x53, 0xE1, 0xFE, 0xF5, 0x2A, 0x75,
+/*1B60*/0x2B, 0x01, 0xF5, 0x08, 0x7F, 0x01, 0x12, 0x17,
+	0x31, 0x30, 0x30, 0x1C, 0x90, 0x1A, 0xA9, 0xE4,
+/*1B70*/0x93, 0xF5, 0x10, 0x90, 0x1F, 0xF9, 0xE4, 0x93,
+	0xF5, 0x10, 0x90, 0x00, 0x41, 0xE4, 0x93, 0xF5,
+/*1B80*/0x10, 0x90, 0x1E, 0xCA, 0xE4, 0x93, 0xF5, 0x10,
+	0x7F, 0x02, 0x12, 0x17, 0x31, 0x12, 0x0F, 0x54,
+/*1B90*/0x7F, 0x03, 0x12, 0x17, 0x31, 0x12, 0x00, 0x06,
+	0xE5, 0xE2, 0x30, 0xE7, 0x09, 0x12, 0x10, 0x00,
+/*1BA0*/0x30, 0x30, 0x03, 0x12, 0x11, 0x00, 0x02, 0x00,
+	0x47, 0x12, 0x08, 0x1F, 0x75, 0x83, 0xD0, 0xE0,
+/*1BB0*/0xC4, 0x54, 0x0F, 0xFD, 0x75, 0x43, 0x01, 0x75,
+	0x44, 0xFF, 0x12, 0x08, 0xAA, 0x74, 0x04, 0xF0,
+/*1BC0*/0x75, 0x3B, 0x01, 0xED, 0x14, 0x60, 0x0C, 0x14,
+	0x60, 0x0B, 0x14, 0x60, 0x0F, 0x24, 0x03, 0x70,
+/*1BD0*/0x0B, 0x80, 0x09, 0x80, 0x00, 0x12, 0x08, 0xA7,
+	0x04, 0xF0, 0x80, 0x06, 0x12, 0x08, 0xA7, 0x74,
+/*1BE0*/0x04, 0xF0, 0xEE, 0x44, 0x82, 0xFE, 0xEF, 0x44,
+	0x07, 0xF5, 0x82, 0x8E, 0x83, 0xE5, 0x45, 0x12,
+/*1BF0*/0x08, 0xBE, 0x75, 0x83, 0x82, 0xE5, 0x31, 0xF0,
+	0x02, 0x11, 0x4C, 0x8E, 0x60, 0x8F, 0x61, 0x12,
+/*1C00*/0x1E, 0xA5, 0xE4, 0xFF, 0xCE, 0xED, 0xCE, 0xEE,
+	0xD3, 0x95, 0x61, 0xE5, 0x60, 0x12, 0x07, 0x6B,
+/*1C10*/0x40, 0x39, 0x74, 0x20, 0x2E, 0xF5, 0x82, 0xE4,
+	0x34, 0x03, 0xF5, 0x83, 0xE0, 0x70, 0x03, 0xFF,
+/*1C20*/0x80, 0x26, 0x12, 0x08, 0xE2, 0xFD, 0xC3, 0x9F,
+	0x40, 0x1E, 0xCF, 0xED, 0xCF, 0xEB, 0x4A, 0x70,
+/*1C30*/0x0B, 0x8D, 0x42, 0x12, 0x08, 0xEE, 0xF5, 0x41,
+	0x8E, 0x40, 0x80, 0x0C, 0x12, 0x08, 0xE2, 0xF5,
+/*1C40*/0x38, 0x12, 0x08, 0xEE, 0xF5, 0x39, 0x8E, 0x3A,
+	0x1E, 0x80, 0xBC, 0x22, 0x75, 0x58, 0x01, 0xE5,
+/*1C50*/0x35, 0x70, 0x0C, 0x12, 0x07, 0xCC, 0xE0, 0xF5,
+	0x4A, 0x12, 0x07, 0xD8, 0xE0, 0xF5, 0x4C, 0xE5,
+/*1C60*/0x35, 0xB4, 0x04, 0x0C, 0x12, 0x07, 0xE4, 0xE0,
+	0xF5, 0x4A, 0x12, 0x07, 0xF0, 0xE0, 0xF5, 0x4C,
+/*1C70*/0xE5, 0x35, 0xB4, 0x01, 0x04, 0x7F, 0x01, 0x80,
+	0x02, 0x7F, 0x00, 0xE5, 0x35, 0xB4, 0x02, 0x04,
+/*1C80*/0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE, 0x4F,
+	0x60, 0x0C, 0x12, 0x07, 0xFC, 0xE0, 0xF5, 0x4A,
+/*1C90*/0x12, 0x08, 0x08, 0xE0, 0xF5, 0x4C, 0x85, 0x41,
+	0x49, 0x85, 0x40, 0x4B, 0x22, 0x75, 0x5B, 0x01,
+/*1CA0*/0x90, 0x07, 0x24, 0x12, 0x07, 0x35, 0xE0, 0x54,
+	0x1F, 0xFF, 0xD3, 0x94, 0x02, 0x50, 0x04, 0x8F,
+/*1CB0*/0x58, 0x80, 0x05, 0xEF, 0x24, 0xFE, 0xF5, 0x58,
+	0xEF, 0xC3, 0x94, 0x18, 0x40, 0x05, 0x75, 0x59,
+/*1CC0*/0x18, 0x80, 0x04, 0xEF, 0x04, 0xF5, 0x59, 0x85,
+	0x43, 0x5A, 0xAF, 0x58, 0x7E, 0x00, 0xAD, 0x59,
+/*1CD0*/0x7C, 0x00, 0xAB, 0x5B, 0x7A, 0x00, 0x12, 0x15,
+	0x41, 0xAF, 0x5A, 0x7E, 0x00, 0x12, 0x18, 0x0A,
+/*1CE0*/0xAF, 0x5B, 0x7E, 0x00, 0x02, 0x1A, 0xFF, 0xE5,
+	0xE2, 0x30, 0xE7, 0x0E, 0x12, 0x10, 0x03, 0xC2,
+/*1CF0*/0x30, 0x30, 0x30, 0x03, 0x12, 0x10, 0xFF, 0x20,
+	0x33, 0x28, 0xE5, 0xE7, 0x30, 0xE7, 0x05, 0x12,
+/*1D00*/0x0E, 0xA2, 0x80, 0x0D, 0xE5, 0xFE, 0xC3, 0x94,
+	0x20, 0x50, 0x06, 0x12, 0x0E, 0xA2, 0x43, 0xF9,
+/*1D10*/0x08, 0xE5, 0xF2, 0x30, 0xE7, 0x03, 0x53, 0xF9,
+	0x7F, 0xE5, 0xF1, 0x54, 0x70, 0xD3, 0x94, 0x00,
+/*1D20*/0x50, 0xD8, 0x22, 0x12, 0x0E, 0x04, 0x75, 0x83,
+	0x80, 0xE4, 0xF0, 0xE5, 0x08, 0x44, 0x07, 0x12,
+/*1D30*/0x0D, 0xFD, 0x75, 0x83, 0x84, 0x12, 0x0E, 0x02,
+	0x75, 0x83, 0x86, 0x12, 0x0E, 0x02, 0x75, 0x83,
+/*1D40*/0x8C, 0xE0, 0x54, 0xF3, 0x12, 0x0E, 0x03, 0x75,
+	0x83, 0x8E, 0x12, 0x0E, 0x02, 0x75, 0x83, 0x94,
+/*1D50*/0xE0, 0x54, 0xFB, 0xF0, 0x22, 0x12, 0x07, 0x2A,
+	0x75, 0x83, 0x8E, 0xE4, 0x12, 0x07, 0x29, 0x74,
+/*1D60*/0x01, 0x12, 0x07, 0x29, 0xE4, 0x12, 0x08, 0xBE,
+	0x75, 0x83, 0x8C, 0xE0, 0x44, 0x20, 0x12, 0x08,
+/*1D70*/0xBE, 0xE0, 0x54, 0xDF, 0xF0, 0x74, 0x84, 0x85,
+	0x08, 0x82, 0xF5, 0x83, 0xE0, 0x54, 0x7F, 0xF0,
+/*1D80*/0xE0, 0x44, 0x80, 0xF0, 0x22, 0x75, 0x56, 0x01,
+	0xE4, 0xFD, 0xF5, 0x57, 0xAF, 0x35, 0xFE, 0xFC,
+/*1D90*/0x12, 0x09, 0x15, 0x12, 0x1C, 0x9D, 0x12, 0x1E,
+	0x7A, 0x12, 0x1C, 0x4C, 0xAF, 0x57, 0x7E, 0x00,
+/*1DA0*/0xAD, 0x56, 0x7C, 0x00, 0x12, 0x04, 0x44, 0xAF,
+	0x56, 0x7E, 0x00, 0x02, 0x11, 0xEE, 0x75, 0x56,
+/*1DB0*/0x01, 0xE4, 0xFD, 0xF5, 0x57, 0xAF, 0x35, 0xFE,
+	0xFC, 0x12, 0x09, 0x15, 0x12, 0x1C, 0x9D, 0x12,
+/*1DC0*/0x1E, 0x7A, 0x12, 0x1C, 0x4C, 0xAF, 0x57, 0x7E,
+	0x00, 0xAD, 0x56, 0x7C, 0x00, 0x12, 0x04, 0x44,
+/*1DD0*/0xAF, 0x56, 0x7E, 0x00, 0x02, 0x11, 0xEE, 0xE4,
+	0xF5, 0x16, 0x12, 0x0E, 0x44, 0xFE, 0xE5, 0x08,
+/*1DE0*/0x44, 0x05, 0xFF, 0x12, 0x0E, 0x65, 0x8F, 0x82,
+	0x8E, 0x83, 0xF0, 0x05, 0x16, 0xE5, 0x16, 0xC3,
+/*1DF0*/0x94, 0x14, 0x40, 0xE6, 0xE5, 0x08, 0x12, 0x0E,
+	0x2B, 0xE4, 0xF0, 0x22, 0xE4, 0xF5, 0x58, 0xF5,
+/*1E00*/0x59, 0xF5, 0x5A, 0xFF, 0xFE, 0xAD, 0x58, 0xFC,
+	0x12, 0x09, 0x15, 0x7F, 0x04, 0x7E, 0x00, 0xAD,
+/*1E10*/0x58, 0x7C, 0x00, 0x12, 0x09, 0x15, 0x7F, 0x02,
+	0x7E, 0x00, 0xAD, 0x58, 0x7C, 0x00, 0x02, 0x09,
+/*1E20*/0x15, 0xE5, 0x3C, 0x25, 0x3E, 0xFC, 0xE5, 0x42,
+	0x24, 0x00, 0xFB, 0xE4, 0x33, 0xFA, 0xEC, 0xC3,
+/*1E30*/0x9B, 0xEA, 0x12, 0x07, 0x6B, 0x40, 0x0B, 0x8C,
+	0x42, 0xE5, 0x3D, 0x25, 0x3F, 0xF5, 0x41, 0x8F,
+/*1E40*/0x40, 0x22, 0x12, 0x09, 0x0B, 0x22, 0x74, 0x84,
+	0xF5, 0x18, 0x85, 0x08, 0x19, 0x85, 0x19, 0x82,
+/*1E50*/0x85, 0x18, 0x83, 0xE0, 0x54, 0x7F, 0xF0, 0xE0,
+	0x44, 0x80, 0xF0, 0xE0, 0x44, 0x80, 0xF0, 0x22,
+/*1E60*/0xEF, 0x4E, 0x70, 0x0B, 0x12, 0x07, 0x2A, 0x75,
+	0x83, 0xD2, 0xE0, 0x54, 0xDF, 0xF0, 0x22, 0x12,
+/*1E70*/0x07, 0x2A, 0x75, 0x83, 0xD2, 0xE0, 0x44, 0x20,
+	0xF0, 0x22, 0x75, 0x58, 0x01, 0x90, 0x07, 0x26,
+/*1E80*/0x12, 0x07, 0x35, 0xE0, 0x54, 0x3F, 0xF5, 0x41,
+	0x12, 0x07, 0x32, 0xE0, 0x54, 0x3F, 0xF5, 0x40,
+/*1E90*/0x22, 0x75, 0x56, 0x02, 0xE4, 0xF5, 0x57, 0x12,
+	0x1D, 0xFC, 0xAF, 0x57, 0x7E, 0x00, 0xAD, 0x56,
+/*1EA0*/0x7C, 0x00, 0x02, 0x04, 0x44, 0xE4, 0xF5, 0x42,
+	0xF5, 0x41, 0xF5, 0x40, 0xF5, 0x38, 0xF5, 0x39,
+/*1EB0*/0xF5, 0x3A, 0x22, 0xEF, 0x54, 0x07, 0xFF, 0xE5,
+	0xF9, 0x54, 0xF8, 0x4F, 0xF5, 0xF9, 0x22, 0x7F,
+/*1EC0*/0x01, 0xE4, 0xFE, 0x0F, 0x0E, 0xBE, 0xFF, 0xFB,
+	0x22, 0x01, 0x20, 0x00, 0x01, 0x04, 0x20, 0x00,
+/*1ED0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1EE0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1EF0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1F00*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1F10*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1F20*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1F30*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1F40*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1F50*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1F60*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1F70*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1F80*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1F90*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1FA0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1FB0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1FC0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1FD0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1FE0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+/*1FF0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x01, 0x20, 0x11, 0x00, 0x04, 0x20, 0x00, 0x81
+};
+
+int qib_sd7220_ib_load(struct qib_devdata *dd)
+{
+	return qib_sd7220_prog_ld(dd, IB_7220_SERDES, qib_sd7220_ib_img,
+		sizeof(qib_sd7220_ib_img), 0);
+}
+
+int qib_sd7220_ib_vfy(struct qib_devdata *dd)
+{
+	return qib_sd7220_prog_vfy(dd, IB_7220_SERDES, qib_sd7220_ib_img,
+		sizeof(qib_sd7220_ib_img), 0);
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 36/51] IB/qib: Add qib_sdma.c

[Ralph Campbell]

creates the qib_sdma.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_sdma.c | 1037 ++++++++++++++++++++++++++++++++++
 1 files changed, 1037 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_sdma.c

diff --git a/drivers/infiniband/hw/qib/qib_sdma.c b/drivers/infiniband/hw/qib/qib_sdma.c
new file mode 100644
index 0000000..4b58f26
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_sdma.c
@@ -0,0 +1,1037 @@
+/*
+ * Copyright (c) 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/spinlock.h>
+#include <linux/netdevice.h>
+
+#include "qib.h"
+#include "qib_common.h"
+
+/* default pio off, sdma on */
+static ushort sdma_descq_cnt = 256;
+module_param_named(sdma_descq_cnt, sdma_descq_cnt, ushort, S_IRUGO);
+MODULE_PARM_DESC(sdma_descq_cnt, "Number of SDMA descq entries");
+
+/*
+ * Bits defined in the send DMA descriptor.
+ */
+#define SDMA_DESC_LAST          (1ULL << 11)
+#define SDMA_DESC_FIRST         (1ULL << 12)
+#define SDMA_DESC_DMA_HEAD      (1ULL << 13)
+#define SDMA_DESC_USE_LARGE_BUF (1ULL << 14)
+#define SDMA_DESC_INTR          (1ULL << 15)
+#define SDMA_DESC_COUNT_LSB     16
+#define SDMA_DESC_GEN_LSB       30
+
+char *qib_sdma_state_names[] = {
+	[qib_sdma_state_s00_hw_down]          = "s00_HwDown",
+	[qib_sdma_state_s10_hw_start_up_wait] = "s10_HwStartUpWait",
+	[qib_sdma_state_s20_idle]             = "s20_Idle",
+	[qib_sdma_state_s30_sw_clean_up_wait] = "s30_SwCleanUpWait",
+	[qib_sdma_state_s40_hw_clean_up_wait] = "s40_HwCleanUpWait",
+	[qib_sdma_state_s50_hw_halt_wait]     = "s50_HwHaltWait",
+	[qib_sdma_state_s99_running]          = "s99_Running",
+};
+
+char *qib_sdma_event_names[] = {
+	[qib_sdma_event_e00_go_hw_down]   = "e00_GoHwDown",
+	[qib_sdma_event_e10_go_hw_start]  = "e10_GoHwStart",
+	[qib_sdma_event_e20_hw_started]   = "e20_HwStarted",
+	[qib_sdma_event_e30_go_running]   = "e30_GoRunning",
+	[qib_sdma_event_e40_sw_cleaned]   = "e40_SwCleaned",
+	[qib_sdma_event_e50_hw_cleaned]   = "e50_HwCleaned",
+	[qib_sdma_event_e60_hw_halted]    = "e60_HwHalted",
+	[qib_sdma_event_e70_go_idle]      = "e70_GoIdle",
+	[qib_sdma_event_e7220_err_halted] = "e7220_ErrHalted",
+	[qib_sdma_event_e7322_err_halted] = "e7322_ErrHalted",
+	[qib_sdma_event_e90_timer_tick]   = "e90_TimerTick",
+};
+
+/* declare all statics here rather than keep sorting */
+static int alloc_sdma(struct qib_pportdata *);
+static void sdma_complete(struct kref *);
+static void sdma_finalput(struct qib_sdma_state *);
+static void sdma_get(struct qib_sdma_state *);
+static void sdma_put(struct qib_sdma_state *);
+static void sdma_set_state(struct qib_pportdata *, enum qib_sdma_states);
+static void sdma_start_sw_clean_up(struct qib_pportdata *);
+static void sdma_sw_clean_up_task(unsigned long);
+static void unmap_desc(struct qib_pportdata *, unsigned);
+
+static void sdma_get(struct qib_sdma_state *ss)
+{
+	kref_get(&ss->kref);
+}
+
+static void sdma_complete(struct kref *kref)
+{
+	struct qib_sdma_state *ss =
+		container_of(kref, struct qib_sdma_state, kref);
+
+	complete(&ss->comp);
+}
+
+static void sdma_put(struct qib_sdma_state *ss)
+{
+	kref_put(&ss->kref, sdma_complete);
+}
+
+static void sdma_finalput(struct qib_sdma_state *ss)
+{
+	sdma_put(ss);
+	wait_for_completion(&ss->comp);
+}
+
+/*
+ * Complete all the sdma requests on the active list, in the correct
+ * order, and with appropriate processing.   Called when cleaning up
+ * after sdma shutdown, and when new sdma requests are submitted for
+ * a link that is down.   This matches what is done for requests
+ * that complete normally, it's just the full list.
+ *
+ * Must be called with sdma_lock held
+ */
+static void clear_sdma_activelist(struct qib_pportdata *ppd)
+{
+	struct qib_sdma_txreq *txp, *txp_next;
+
+	list_for_each_entry_safe(txp, txp_next, &ppd->sdma_activelist, list) {
+		list_del_init(&txp->list);
+		if (txp->flags & QIB_SDMA_TXREQ_F_FREEDESC) {
+			unsigned idx;
+
+			idx = txp->start_idx;
+			while (idx != txp->next_descq_idx) {
+				unmap_desc(ppd, idx);
+				if (++idx == ppd->sdma_descq_cnt)
+					idx = 0;
+			}
+		}
+		if (txp->callback)
+			(*txp->callback)(txp, QIB_SDMA_TXREQ_S_ABORTED);
+	}
+}
+
+static void sdma_sw_clean_up_task(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *) opaque;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	/*
+	 * At this point, the following should always be true:
+	 * - We are halted, so no more descriptors are getting retired.
+	 * - We are not running, so no one is submitting new work.
+	 * - Only we can send the e40_sw_cleaned, so we can't start
+	 *   running again until we say so.  So, the active list and
+	 *   descq are ours to play with.
+	 */
+
+	/* Process all retired requests. */
+	qib_sdma_make_progress(ppd);
+
+	clear_sdma_activelist(ppd);
+
+	/*
+	 * Resync count of added and removed.  It is VERY important that
+	 * sdma_descq_removed NEVER decrement - user_sdma depends on it.
+	 */
+	ppd->sdma_descq_removed = ppd->sdma_descq_added;
+
+	/*
+	 * Reset our notion of head and tail.
+	 * Note that the HW registers will be reset when switching states
+	 * due to calling __qib_sdma_process_event() below.
+	 */
+	ppd->sdma_descq_tail = 0;
+	ppd->sdma_descq_head = 0;
+	ppd->sdma_head_dma[0] = 0;
+	ppd->sdma_generation = 0;
+
+	__qib_sdma_process_event(ppd, qib_sdma_event_e40_sw_cleaned);
+
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+}
+
+/*
+ * This is called when changing to state qib_sdma_state_s10_hw_start_up_wait
+ * as a result of send buffer errors or send DMA descriptor errors.
+ * We want to disarm the buffers in these cases.
+ */
+static void sdma_hw_start_up(struct qib_pportdata *ppd)
+{
+	struct qib_sdma_state *ss = &ppd->sdma_state;
+	unsigned bufno;
+
+	for (bufno = ss->first_sendbuf; bufno < ss->last_sendbuf; ++bufno)
+		ppd->dd->f_sendctrl(ppd, QIB_SENDCTRL_DISARM_BUF(bufno));
+
+	ppd->dd->f_sdma_hw_start_up(ppd);
+}
+
+static void sdma_sw_tear_down(struct qib_pportdata *ppd)
+{
+	struct qib_sdma_state *ss = &ppd->sdma_state;
+
+	/* Releasing this reference means the state machine has stopped. */
+	sdma_put(ss);
+}
+
+static void sdma_start_sw_clean_up(struct qib_pportdata *ppd)
+{
+	tasklet_hi_schedule(&ppd->sdma_sw_clean_up_task);
+}
+
+static void sdma_set_state(struct qib_pportdata *ppd,
+	enum qib_sdma_states next_state)
+{
+	struct qib_sdma_state *ss = &ppd->sdma_state;
+	struct sdma_set_state_action *action = ss->set_state_action;
+	unsigned op = 0;
+
+	/* debugging bookkeeping */
+	ss->previous_state = ss->current_state;
+	ss->previous_op = ss->current_op;
+
+	ss->current_state = next_state;
+
+	if (action[next_state].op_enable)
+		op |= QIB_SDMA_SENDCTRL_OP_ENABLE;
+
+	if (action[next_state].op_intenable)
+		op |= QIB_SDMA_SENDCTRL_OP_INTENABLE;
+
+	if (action[next_state].op_halt)
+		op |= QIB_SDMA_SENDCTRL_OP_HALT;
+
+	if (action[next_state].op_drain)
+		op |= QIB_SDMA_SENDCTRL_OP_DRAIN;
+
+	if (action[next_state].go_s99_running_tofalse)
+		ss->go_s99_running = 0;
+
+	if (action[next_state].go_s99_running_totrue)
+		ss->go_s99_running = 1;
+
+	ss->current_op = op;
+
+	ppd->dd->f_sdma_sendctrl(ppd, ss->current_op);
+}
+
+static void unmap_desc(struct qib_pportdata *ppd, unsigned head)
+{
+	__le64 *descqp = &ppd->sdma_descq[head].qw[0];
+	u64 desc[2];
+	dma_addr_t addr;
+	size_t len;
+
+	desc[0] = le64_to_cpu(descqp[0]);
+	desc[1] = le64_to_cpu(descqp[1]);
+
+	addr = (desc[1] << 32) | (desc[0] >> 32);
+	len = (desc[0] >> 14) & (0x7ffULL << 2);
+	dma_unmap_single(&ppd->dd->pcidev->dev, addr, len, DMA_TO_DEVICE);
+}
+
+static int alloc_sdma(struct qib_pportdata *ppd)
+{
+	ppd->sdma_descq_cnt = sdma_descq_cnt;
+	if (!ppd->sdma_descq_cnt)
+		ppd->sdma_descq_cnt = 256;
+
+	/* Allocate memory for SendDMA descriptor FIFO */
+	qib_cdbg(SDMA, "IB%u:%u allocating %u SDMA descq entries\n",
+		 ppd->dd->unit, ppd->port, (unsigned)ppd->sdma_descq_cnt);
+	ppd->sdma_descq = dma_alloc_coherent(&ppd->dd->pcidev->dev,
+		ppd->sdma_descq_cnt * sizeof(u64[2]), &ppd->sdma_descq_phys,
+		GFP_KERNEL);
+
+	if (!ppd->sdma_descq) {
+		qib_dev_err(ppd->dd, "failed to allocate SendDMA descriptor "
+			    "FIFO memory\n");
+		goto bail;
+	}
+
+	/* Allocate memory for DMA of head register to memory */
+	ppd->sdma_head_dma = dma_alloc_coherent(&ppd->dd->pcidev->dev,
+		PAGE_SIZE, &ppd->sdma_head_phys, GFP_KERNEL);
+	if (!ppd->sdma_head_dma) {
+		qib_dev_err(ppd->dd, "failed to allocate SendDMA "
+			    "head memory\n");
+		goto cleanup_descq;
+	}
+	ppd->sdma_head_dma[0] = 0;
+	return 0;
+
+cleanup_descq:
+	dma_free_coherent(&ppd->dd->pcidev->dev,
+		ppd->sdma_descq_cnt * sizeof(u64[2]), (void *)ppd->sdma_descq,
+		ppd->sdma_descq_phys);
+	ppd->sdma_descq = NULL;
+	ppd->sdma_descq_phys = 0;
+bail:
+	ppd->sdma_descq_cnt = 0;
+	return -ENOMEM;
+}
+
+static void free_sdma(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+
+	if (ppd->sdma_head_dma) {
+		dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
+				  (void *)ppd->sdma_head_dma,
+				  ppd->sdma_head_phys);
+		ppd->sdma_head_dma = NULL;
+		ppd->sdma_head_phys = 0;
+	}
+
+	if (ppd->sdma_descq) {
+		dma_free_coherent(&dd->pcidev->dev,
+				  ppd->sdma_descq_cnt * sizeof(u64[2]),
+				  ppd->sdma_descq, ppd->sdma_descq_phys);
+		ppd->sdma_descq = NULL;
+		ppd->sdma_descq_phys = 0;
+	}
+}
+
+static inline void make_sdma_desc(struct qib_pportdata *ppd,
+				  u64 *sdmadesc, u64 addr, u64 dwlen,
+				  u64 dwoffset)
+{
+
+	WARN_ON(addr & 3);
+	/* SDmaPhyAddr[47:32] */
+	sdmadesc[1] = addr >> 32;
+	/* SDmaPhyAddr[31:0] */
+	sdmadesc[0] = (addr & 0xfffffffcULL) << 32;
+	/* SDmaGeneration[1:0] */
+	sdmadesc[0] |= (ppd->sdma_generation & 3ULL) <<
+		SDMA_DESC_GEN_LSB;
+	/* SDmaDwordCount[10:0] */
+	sdmadesc[0] |= (dwlen & 0x7ffULL) << SDMA_DESC_COUNT_LSB;
+	/* SDmaBufOffset[12:2] */
+	sdmadesc[0] |= dwoffset & 0x7ffULL;
+}
+
+/* sdma_lock must be held */
+int qib_sdma_make_progress(struct qib_pportdata *ppd)
+{
+	struct list_head *lp = NULL;
+	struct qib_sdma_txreq *txp = NULL;
+	struct qib_devdata *dd = ppd->dd;
+	int progress = 0;
+	u16 hwhead;
+	u16 idx = 0;
+
+	hwhead = dd->f_sdma_gethead(ppd);
+
+	/* The reason for some of the complexity of this code is that
+	 * not all descriptors have corresponding txps.  So, we have to
+	 * be able to skip over descs until we wander into the range of
+	 * the next txp on the list.
+	 */
+
+	if (!list_empty(&ppd->sdma_activelist)) {
+		lp = ppd->sdma_activelist.next;
+		txp = list_entry(lp, struct qib_sdma_txreq, list);
+		idx = txp->start_idx;
+	}
+
+	while (ppd->sdma_descq_head != hwhead) {
+		/* if desc is part of this txp, unmap if needed */
+		if (txp && (txp->flags & QIB_SDMA_TXREQ_F_FREEDESC) &&
+		    (idx == ppd->sdma_descq_head)) {
+			unmap_desc(ppd, ppd->sdma_descq_head);
+			if (++idx == ppd->sdma_descq_cnt)
+				idx = 0;
+		}
+
+		/* increment dequed desc count */
+		ppd->sdma_descq_removed++;
+
+		/* advance head, wrap if needed */
+		if (++ppd->sdma_descq_head == ppd->sdma_descq_cnt)
+			ppd->sdma_descq_head = 0;
+
+		/* if now past this txp's descs, do the callback */
+		if (txp && txp->next_descq_idx == ppd->sdma_descq_head) {
+			/* remove from active list */
+			list_del_init(&txp->list);
+			if (txp->callback)
+				(*txp->callback)(txp, QIB_SDMA_TXREQ_S_OK);
+			/* see if there is another txp */
+			if (list_empty(&ppd->sdma_activelist))
+				txp = NULL;
+			else {
+				lp = ppd->sdma_activelist.next;
+				txp = list_entry(lp, struct qib_sdma_txreq,
+					list);
+				idx = txp->start_idx;
+			}
+		}
+		progress = 1;
+	}
+	if (progress)
+		qib_verbs_sdma_desc_avail(ppd, qib_sdma_descq_freecnt(ppd));
+	return progress;
+}
+
+/*
+ * This is called from interrupt context.
+ */
+void qib_sdma_intr(struct qib_pportdata *ppd)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	__qib_sdma_intr(ppd);
+
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+}
+
+void __qib_sdma_intr(struct qib_pportdata *ppd)
+{
+	if (__qib_sdma_running(ppd))
+		qib_sdma_make_progress(ppd);
+}
+
+int qib_setup_sdma(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+	int ret = 0;
+
+	ret = alloc_sdma(ppd);
+	if (ret)
+		goto bail;
+
+	/* set consistent sdma state */
+	ppd->dd->f_sdma_init_early(ppd);
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+	sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+	/* set up reference counting */
+	kref_init(&ppd->sdma_state.kref);
+	init_completion(&ppd->sdma_state.comp);
+
+	ppd->sdma_generation = 0;
+	ppd->sdma_descq_head = 0;
+	ppd->sdma_descq_removed = 0;
+	ppd->sdma_descq_added = 0;
+
+	INIT_LIST_HEAD(&ppd->sdma_activelist);
+
+	tasklet_init(&ppd->sdma_sw_clean_up_task, sdma_sw_clean_up_task,
+		(unsigned long)ppd);
+
+	ret = dd->f_init_sdma_regs(ppd);
+	if (ret)
+		goto bail_alloc;
+
+	qib_sdma_process_event(ppd, qib_sdma_event_e10_go_hw_start);
+
+	return 0;
+
+bail_alloc:
+	qib_teardown_sdma(ppd);
+bail:
+	return ret;
+}
+
+void qib_teardown_sdma(struct qib_pportdata *ppd)
+{
+	qib_sdma_process_event(ppd, qib_sdma_event_e00_go_hw_down);
+
+	/*
+	 * This waits for the state machine to exit so it is not
+	 * necessary to kill the sdma_sw_clean_up_task to make sure
+	 * it is not running.
+	 */
+	sdma_finalput(&ppd->sdma_state);
+
+	free_sdma(ppd);
+}
+
+int qib_sdma_running(struct qib_pportdata *ppd)
+{
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+	ret = __qib_sdma_running(ppd);
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+	return ret;
+}
+
+/*
+ * Complete a request when sdma not running; likely only request
+ * but to simplify the code, always queue it, then process the full
+ * activelist.  We process the entire list to ensure that this particular
+ * request does get it's callback, but in the correct order.
+ * Must be called with sdma_lock held
+ */
+static void complete_sdma_err_req(struct qib_pportdata *ppd,
+				  struct qib_verbs_txreq *tx)
+{
+	atomic_inc(&tx->qp->s_dma_busy);
+	/* no sdma descriptors, so no unmap_desc */
+	tx->txreq.start_idx = 0;
+	tx->txreq.next_descq_idx = 0;
+	list_add_tail(&tx->txreq.list, &ppd->sdma_activelist);
+	clear_sdma_activelist(ppd);
+}
+
+/*
+ * This function queues one IB packet onto the send DMA queue per call.
+ * The caller is responsible for checking:
+ * 1) The number of send DMA descriptor entries is less than the size of
+ *    the descriptor queue.
+ * 2) The IB SGE addresses and lengths are 32-bit aligned
+ *    (except possibly the last SGE's length)
+ * 3) The SGE addresses are suitable for passing to dma_map_single().
+ */
+int qib_sdma_verbs_send(struct qib_pportdata *ppd,
+			struct qib_sge_state *ss, u32 dwords,
+			struct qib_verbs_txreq *tx)
+{
+	unsigned long flags;
+	struct qib_sge *sge;
+	struct qib_qp *qp;
+	int ret = 0;
+	u16 tail;
+	__le64 *descqp;
+	u64 sdmadesc[2];
+	u32 dwoffset;
+	dma_addr_t addr;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+retry:
+	if (unlikely(!__qib_sdma_running(ppd))) {
+		complete_sdma_err_req(ppd, tx);
+		goto unlock;
+	}
+
+	if (tx->txreq.sg_count > qib_sdma_descq_freecnt(ppd)) {
+		if (qib_sdma_make_progress(ppd))
+			goto retry;
+		if (ppd->dd->flags & QIB_HAS_SDMA_TIMEOUT)
+			ppd->dd->f_sdma_set_desc_cnt(ppd,
+					ppd->sdma_descq_cnt / 2);
+		goto busy;
+	}
+
+	dwoffset = tx->hdr_dwords;
+	make_sdma_desc(ppd, sdmadesc, (u64) tx->txreq.addr, dwoffset, 0);
+
+	sdmadesc[0] |= SDMA_DESC_FIRST;
+	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_USELARGEBUF)
+		sdmadesc[0] |= SDMA_DESC_USE_LARGE_BUF;
+
+	/* write to the descq */
+	tail = ppd->sdma_descq_tail;
+	descqp = &ppd->sdma_descq[tail].qw[0];
+	*descqp++ = cpu_to_le64(sdmadesc[0]);
+	*descqp++ = cpu_to_le64(sdmadesc[1]);
+
+	/* increment the tail */
+	if (++tail == ppd->sdma_descq_cnt) {
+		tail = 0;
+		descqp = &ppd->sdma_descq[0].qw[0];
+		++ppd->sdma_generation;
+	}
+
+	tx->txreq.start_idx = tail;
+
+	sge = &ss->sge;
+	while (dwords) {
+		u32 dw;
+		u32 len;
+
+		len = dwords << 2;
+		if (len > sge->length)
+			len = sge->length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		dw = (len + 3) >> 2;
+		addr = dma_map_single(&ppd->dd->pcidev->dev, sge->vaddr,
+				      dw << 2, DMA_TO_DEVICE);
+		if (dma_mapping_error(&ppd->dd->pcidev->dev, addr))
+			goto unmap;
+		sdmadesc[0] = 0;
+		make_sdma_desc(ppd, sdmadesc, (u64) addr, dw, dwoffset);
+		/* SDmaUseLargeBuf has to be set in every descriptor */
+		if (tx->txreq.flags & QIB_SDMA_TXREQ_F_USELARGEBUF)
+			sdmadesc[0] |= SDMA_DESC_USE_LARGE_BUF;
+		/* write to the descq */
+		*descqp++ = cpu_to_le64(sdmadesc[0]);
+		*descqp++ = cpu_to_le64(sdmadesc[1]);
+
+		/* increment the tail */
+		if (++tail == ppd->sdma_descq_cnt) {
+			tail = 0;
+			descqp = &ppd->sdma_descq[0].qw[0];
+			++ppd->sdma_generation;
+		}
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (--ss->num_sge)
+				*sge = *ss->sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+
+		dwoffset += dw;
+		dwords -= dw;
+	}
+
+	if (!tail)
+		descqp = &ppd->sdma_descq[ppd->sdma_descq_cnt].qw[0];
+	descqp -= 2;
+	descqp[0] |= cpu_to_le64(SDMA_DESC_LAST);
+	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_HEADTOHOST)
+		descqp[0] |= cpu_to_le64(SDMA_DESC_DMA_HEAD);
+	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_INTREQ)
+		descqp[0] |= cpu_to_le64(SDMA_DESC_INTR);
+
+	atomic_inc(&tx->qp->s_dma_busy);
+	tx->txreq.next_descq_idx = tail;
+	ppd->dd->f_sdma_update_tail(ppd, tail);
+	ppd->sdma_descq_added += tx->txreq.sg_count;
+	list_add_tail(&tx->txreq.list, &ppd->sdma_activelist);
+	goto unlock;
+
+unmap:
+	for (;;) {
+		if (!tail)
+			tail = ppd->sdma_descq_cnt - 1;
+		else
+			tail--;
+		if (tail == ppd->sdma_descq_tail)
+			break;
+		unmap_desc(ppd, tail);
+	}
+	qp = tx->qp;
+	qib_dbg("QP%d I/O error\n", qp->ibqp.qp_num);
+	qib_put_txreq(tx);
+	spin_lock(&qp->s_lock);
+	if (qp->ibqp.qp_type == IB_QPT_RC) {
+		/* XXX what about error sending RDMA read responses? */
+		if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)
+			qib_error_qp(qp, IB_WC_GENERAL_ERR);
+	} else if (qp->s_wqe)
+		qib_send_complete(qp, qp->s_wqe, IB_WC_GENERAL_ERR);
+	spin_unlock(&qp->s_lock);
+	/* return zero to process the next send work request */
+	goto unlock;
+
+busy:
+	qp = tx->qp;
+	spin_lock(&qp->s_lock);
+	if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK) {
+		struct qib_ibdev *dev;
+
+		/*
+		 * If we couldn't queue the DMA request, save the info
+		 * and try again later rather than destroying the
+		 * buffer and undoing the side effects of the copy.
+		 */
+		tx->ss = ss;
+		tx->dwords = dwords;
+		qp->s_tx = tx;
+		dev = &ppd->dd->verbs_dev;
+		spin_lock(&dev->pending_lock);
+		if (list_empty(&qp->iowait)) {
+			struct qib_ibport *ibp;
+
+			ibp = &ppd->ibport_data;
+			ibp->n_dmawait++;
+			qp->s_flags |= QIB_S_WAIT_DMA_DESC;
+			list_add_tail(&qp->iowait, &dev->dmawait);
+		}
+		spin_unlock(&dev->pending_lock);
+		qp->s_flags &= ~QIB_S_BUSY;
+		spin_unlock(&qp->s_lock);
+		ret = -EBUSY;
+	} else {
+		spin_unlock(&qp->s_lock);
+		qib_put_txreq(tx);
+	}
+unlock:
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+	return ret;
+}
+
+/*
+ * sdma_lock should be acquired before calling this routine
+ */
+void dump_sdma_state(struct qib_pportdata *ppd)
+{
+	struct qib_sdma_desc *descq;
+	struct qib_sdma_txreq *txp, *txpnext;
+	__le64 *descqp;
+	u64 desc[2];
+	dma_addr_t addr;
+	u16 gen, dwlen, dwoffset;
+	u16 head, tail, cnt;
+
+	head = ppd->sdma_descq_head;
+	tail = ppd->sdma_descq_tail;
+	cnt = qib_sdma_descq_freecnt(ppd);
+	descq = ppd->sdma_descq;
+
+	qib_cdbg(SDMA, "IB%u: ppd->sdma_descq_head: %u\n", ppd->port, head);
+	qib_cdbg(SDMA, "IB%u: ppd->sdma_descq_tail: %u\n", ppd->port, tail);
+	qib_cdbg(SDMA, "IB%u: sdma_descq_freecnt: %u\n", ppd->port, cnt);
+
+	/* print info for each entry in the descriptor queue */
+	while (head != tail) {
+		char flags[6] = { 'x', 'x', 'x', 'x', 'x', 0 };
+
+		descqp = &descq[head].qw[0];
+		desc[0] = le64_to_cpu(descqp[0]);
+		desc[1] = le64_to_cpu(descqp[1]);
+		flags[0] = (desc[0] & 1<<15) ? 'I' : '-';
+		flags[1] = (desc[0] & 1<<14) ? 'L' : 'S';
+		flags[2] = (desc[0] & 1<<13) ? 'H' : '-';
+		flags[3] = (desc[0] & 1<<12) ? 'F' : '-';
+		flags[4] = (desc[0] & 1<<11) ? 'L' : '-';
+		addr = (desc[1] << 32) | ((desc[0] >> 32) & 0xfffffffcULL);
+		gen = (desc[0] >> 30) & 3ULL;
+		dwlen = (desc[0] >> 14) & (0x7ffULL << 2);
+		dwoffset = (desc[0] & 0x7ffULL) << 2;
+		qib_cdbg(SDMA, "IB%u: sdmadesc[%u]: flags:%s addr:0x%016llx "
+			 "gen:%u len:%u bytes offset:%u bytes\n", ppd->port,
+			 head, flags, addr, gen, dwlen, dwoffset);
+		if (++head == ppd->sdma_descq_cnt)
+			head = 0;
+	}
+
+	if (qib_debug & __QIB_VERBDBG)
+		/* print dma descriptor indices from the TX requests */
+		list_for_each_entry_safe(txp, txpnext, &ppd->sdma_activelist,
+					 list)
+			qib_cdbg(SDMA, "IB%u: txp->start_idx: %u "
+				"txp->next_descq_idx: %u\n", ppd->port,
+				txp->start_idx, txp->next_descq_idx);
+}
+
+void qib_sdma_process_event(struct qib_pportdata *ppd,
+	enum qib_sdma_events event)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	__qib_sdma_process_event(ppd, event);
+
+	if (ppd->sdma_state.current_state == qib_sdma_state_s99_running)
+		qib_verbs_sdma_desc_avail(ppd, qib_sdma_descq_freecnt(ppd));
+
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+}
+
+void __qib_sdma_process_event(struct qib_pportdata *ppd,
+	enum qib_sdma_events event)
+{
+	struct qib_sdma_state *ss = &ppd->sdma_state;
+
+	qib_cdbg(SDMA, "IB%u:%u [%s] %s\n", ppd->dd->unit, ppd->port,
+		 qib_sdma_state_names[ss->current_state],
+		 qib_sdma_event_names[event]);
+	switch (ss->current_state) {
+	case qib_sdma_state_s00_hw_down:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			/* This reference means the state machine is started */
+			sdma_get(&ppd->sdma_state);
+			sdma_set_state(ppd,
+				       qib_sdma_state_s10_hw_start_up_wait);
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			sdma_sw_tear_down(ppd);
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s10_hw_start_up_wait:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_sw_tear_down(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			sdma_set_state(ppd, ss->go_s99_running ?
+				       qib_sdma_state_s99_running :
+				       qib_sdma_state_s20_idle);
+			break;
+		case qib_sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s20_idle:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_sw_tear_down(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			sdma_set_state(ppd, qib_sdma_state_s99_running);
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			/*
+			 * Some chips need sends flushed after we go idle
+			 * in some conditions, in case there are stuck sends
+			 * after hw cleanup
+			 */
+			if (ppd->dd->f_sdma_flush_sends(ppd))
+				sdma_set_state(ppd,
+					qib_sdma_state_s40_hw_clean_up_wait);
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s30_sw_clean_up_wait:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s10_hw_start_up_wait);
+			sdma_hw_start_up(ppd);
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s40_hw_clean_up_wait:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s30_sw_clean_up_wait);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s50_hw_halt_wait:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s40_hw_clean_up_wait);
+			ppd->dd->f_sdma_hw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s99_running:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s30_sw_clean_up_wait);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e70_go_idle:
+			sdma_set_state(ppd, qib_sdma_state_s50_hw_halt_wait);
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s30_sw_clean_up_wait);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			sdma_set_state(ppd, qib_sdma_state_s50_hw_halt_wait);
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+	}
+
+	ss->last_event = event;
+
+	qib_cdbg(SDMA, "IB%u:%u [%s]\n", ppd->dd->unit, ppd->port,
+		 qib_sdma_state_names[ss->current_state]);
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 37/51] IB/qib: Add qib_srq.c

[Ralph Campbell]

creates the qib_srq.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_srq.c |  374 +++++++++++++++++++++++++++++++++++
 1 files changed, 374 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_srq.c

diff --git a/drivers/infiniband/hw/qib/qib_srq.c b/drivers/infiniband/hw/qib/qib_srq.c
new file mode 100644
index 0000000..d79ae33
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_srq.c
@@ -0,0 +1,374 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/err.h>
+#include <linux/vmalloc.h>
+
+#include "qib_verbs.h"
+
+/**
+ * qib_post_srq_receive - post a receive on a shared receive queue
+ * @ibsrq: the SRQ to post the receive on
+ * @wr: the list of work requests to post
+ * @bad_wr: A pointer to the first WR to cause a problem is put here
+ *
+ * This may be called from interrupt context.
+ */
+int qib_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
+			 struct ib_recv_wr **bad_wr)
+{
+	struct qib_srq *srq = to_isrq(ibsrq);
+	struct qib_rwq *wq;
+	unsigned long flags;
+	int ret;
+
+	for (; wr; wr = wr->next) {
+		struct qib_rwqe *wqe;
+		u32 next;
+		int i;
+
+		if ((unsigned) wr->num_sge > srq->rq.max_sge) {
+			*bad_wr = wr;
+			ret = -EINVAL;
+			goto bail;
+		}
+
+		spin_lock_irqsave(&srq->rq.lock, flags);
+		wq = srq->rq.wq;
+		next = wq->head + 1;
+		if (next >= srq->rq.size)
+			next = 0;
+		if (next == wq->tail) {
+			spin_unlock_irqrestore(&srq->rq.lock, flags);
+			*bad_wr = wr;
+			ret = -ENOMEM;
+			goto bail;
+		}
+
+		wqe = get_rwqe_ptr(&srq->rq, wq->head);
+		wqe->wr_id = wr->wr_id;
+		wqe->num_sge = wr->num_sge;
+		for (i = 0; i < wr->num_sge; i++)
+			wqe->sg_list[i] = wr->sg_list[i];
+		/* Make sure queue entry is written before the head index. */
+		smp_wmb();
+		wq->head = next;
+		spin_unlock_irqrestore(&srq->rq.lock, flags);
+	}
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_create_srq - create a shared receive queue
+ * @ibpd: the protection domain of the SRQ to create
+ * @srq_init_attr: the attributes of the SRQ
+ * @udata: data from libibverbs when creating a user SRQ
+ */
+struct ib_srq *qib_create_srq(struct ib_pd *ibpd,
+			      struct ib_srq_init_attr *srq_init_attr,
+			      struct ib_udata *udata)
+{
+	struct qib_ibdev *dev = to_idev(ibpd->device);
+	struct qib_srq *srq;
+	u32 sz;
+	struct ib_srq *ret;
+
+	if (srq_init_attr->attr.max_sge == 0 ||
+	    srq_init_attr->attr.max_sge > ib_qib_max_srq_sges ||
+	    srq_init_attr->attr.max_wr == 0 ||
+	    srq_init_attr->attr.max_wr > ib_qib_max_srq_wrs) {
+		ret = ERR_PTR(-EINVAL);
+		goto done;
+	}
+
+	srq = kmalloc(sizeof(*srq), GFP_KERNEL);
+	if (!srq) {
+		ret = ERR_PTR(-ENOMEM);
+		goto done;
+	}
+
+	/*
+	 * Need to use vmalloc() if we want to support large #s of entries.
+	 */
+	srq->rq.size = srq_init_attr->attr.max_wr + 1;
+	srq->rq.max_sge = srq_init_attr->attr.max_sge;
+	sz = sizeof(struct ib_sge) * srq->rq.max_sge +
+		sizeof(struct qib_rwqe);
+	srq->rq.wq = vmalloc_user(sizeof(struct qib_rwq) + srq->rq.size * sz);
+	if (!srq->rq.wq) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_srq;
+	}
+
+	/*
+	 * Return the address of the RWQ as the offset to mmap.
+	 * See qib_mmap() for details.
+	 */
+	if (udata && udata->outlen >= sizeof(__u64)) {
+		int err;
+		u32 s = sizeof(struct qib_rwq) + srq->rq.size * sz;
+
+		srq->ip =
+		    qib_create_mmap_info(dev, s, ibpd->uobject->context,
+					 srq->rq.wq);
+		if (!srq->ip) {
+			ret = ERR_PTR(-ENOMEM);
+			goto bail_wq;
+		}
+
+		err = ib_copy_to_udata(udata, &srq->ip->offset,
+				       sizeof(srq->ip->offset));
+		if (err) {
+			ret = ERR_PTR(err);
+			goto bail_ip;
+		}
+	} else
+		srq->ip = NULL;
+
+	/*
+	 * ib_create_srq() will initialize srq->ibsrq.
+	 */
+	spin_lock_init(&srq->rq.lock);
+	srq->rq.wq->head = 0;
+	srq->rq.wq->tail = 0;
+	srq->limit = srq_init_attr->attr.srq_limit;
+
+	spin_lock(&dev->n_srqs_lock);
+	if (dev->n_srqs_allocated == ib_qib_max_srqs) {
+		spin_unlock(&dev->n_srqs_lock);
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_ip;
+	}
+
+	dev->n_srqs_allocated++;
+	spin_unlock(&dev->n_srqs_lock);
+
+	if (srq->ip) {
+		spin_lock_irq(&dev->pending_lock);
+		list_add(&srq->ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
+	ret = &srq->ibsrq;
+	goto done;
+
+bail_ip:
+	kfree(srq->ip);
+bail_wq:
+	vfree(srq->rq.wq);
+bail_srq:
+	kfree(srq);
+done:
+	return ret;
+}
+
+/**
+ * qib_modify_srq - modify a shared receive queue
+ * @ibsrq: the SRQ to modify
+ * @attr: the new attributes of the SRQ
+ * @attr_mask: indicates which attributes to modify
+ * @udata: user data for libibverbs.so
+ */
+int qib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
+		   enum ib_srq_attr_mask attr_mask,
+		   struct ib_udata *udata)
+{
+	struct qib_srq *srq = to_isrq(ibsrq);
+	struct qib_rwq *wq;
+	int ret = 0;
+
+	if (attr_mask & IB_SRQ_MAX_WR) {
+		struct qib_rwq *owq;
+		struct qib_rwqe *p;
+		u32 sz, size, n, head, tail;
+
+		/* Check that the requested sizes are below the limits. */
+		if ((attr->max_wr > ib_qib_max_srq_wrs) ||
+		    ((attr_mask & IB_SRQ_LIMIT) ?
+		     attr->srq_limit : srq->limit) > attr->max_wr) {
+			ret = -EINVAL;
+			goto bail;
+		}
+
+		sz = sizeof(struct qib_rwqe) +
+			srq->rq.max_sge * sizeof(struct ib_sge);
+		size = attr->max_wr + 1;
+		wq = vmalloc_user(sizeof(struct qib_rwq) + size * sz);
+		if (!wq) {
+			ret = -ENOMEM;
+			goto bail;
+		}
+
+		/* Check that we can write the offset to mmap. */
+		if (udata && udata->inlen >= sizeof(__u64)) {
+			__u64 offset_addr;
+			__u64 offset = 0;
+
+			ret = ib_copy_from_udata(&offset_addr, udata,
+						 sizeof(offset_addr));
+			if (ret)
+				goto bail_free;
+			udata->outbuf =
+				(void __user *) (unsigned long) offset_addr;
+			ret = ib_copy_to_udata(udata, &offset,
+					       sizeof(offset));
+			if (ret)
+				goto bail_free;
+		}
+
+		spin_lock_irq(&srq->rq.lock);
+		/*
+		 * validate head and tail pointer values and compute
+		 * the number of remaining WQEs.
+		 */
+		owq = srq->rq.wq;
+		head = owq->head;
+		tail = owq->tail;
+		if (head >= srq->rq.size || tail >= srq->rq.size) {
+			ret = -EINVAL;
+			goto bail_unlock;
+		}
+		n = head;
+		if (n < tail)
+			n += srq->rq.size - tail;
+		else
+			n -= tail;
+		if (size <= n) {
+			ret = -EINVAL;
+			goto bail_unlock;
+		}
+		n = 0;
+		p = wq->wq;
+		while (tail != head) {
+			struct qib_rwqe *wqe;
+			int i;
+
+			wqe = get_rwqe_ptr(&srq->rq, tail);
+			p->wr_id = wqe->wr_id;
+			p->num_sge = wqe->num_sge;
+			for (i = 0; i < wqe->num_sge; i++)
+				p->sg_list[i] = wqe->sg_list[i];
+			n++;
+			p = (struct qib_rwqe *)((char *) p + sz);
+			if (++tail >= srq->rq.size)
+				tail = 0;
+		}
+		srq->rq.wq = wq;
+		srq->rq.size = size;
+		wq->head = n;
+		wq->tail = 0;
+		if (attr_mask & IB_SRQ_LIMIT)
+			srq->limit = attr->srq_limit;
+		spin_unlock_irq(&srq->rq.lock);
+
+		vfree(owq);
+
+		if (srq->ip) {
+			struct qib_mmap_info *ip = srq->ip;
+			struct qib_ibdev *dev = to_idev(srq->ibsrq.device);
+			u32 s = sizeof(struct qib_rwq) + size * sz;
+
+			qib_update_mmap_info(dev, ip, s, wq);
+
+			/*
+			 * Return the offset to mmap.
+			 * See qib_mmap() for details.
+			 */
+			if (udata && udata->inlen >= sizeof(__u64)) {
+				ret = ib_copy_to_udata(udata, &ip->offset,
+						       sizeof(ip->offset));
+				if (ret)
+					goto bail;
+			}
+
+			/*
+			 * Put user mapping info onto the pending list
+			 * unless it already is on the list.
+			 */
+			spin_lock_irq(&dev->pending_lock);
+			if (list_empty(&ip->pending_mmaps))
+				list_add(&ip->pending_mmaps,
+					 &dev->pending_mmaps);
+			spin_unlock_irq(&dev->pending_lock);
+		}
+	} else if (attr_mask & IB_SRQ_LIMIT) {
+		spin_lock_irq(&srq->rq.lock);
+		if (attr->srq_limit >= srq->rq.size)
+			ret = -EINVAL;
+		else
+			srq->limit = attr->srq_limit;
+		spin_unlock_irq(&srq->rq.lock);
+	}
+	goto bail;
+
+bail_unlock:
+	spin_unlock_irq(&srq->rq.lock);
+bail_free:
+	vfree(wq);
+bail:
+	return ret;
+}
+
+int qib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr)
+{
+	struct qib_srq *srq = to_isrq(ibsrq);
+
+	attr->max_wr = srq->rq.size - 1;
+	attr->max_sge = srq->rq.max_sge;
+	attr->srq_limit = srq->limit;
+	return 0;
+}
+
+/**
+ * qib_destroy_srq - destroy a shared receive queue
+ * @ibsrq: the SRQ to destroy
+ */
+int qib_destroy_srq(struct ib_srq *ibsrq)
+{
+	struct qib_srq *srq = to_isrq(ibsrq);
+	struct qib_ibdev *dev = to_idev(ibsrq->device);
+
+	spin_lock(&dev->n_srqs_lock);
+	dev->n_srqs_allocated--;
+	spin_unlock(&dev->n_srqs_lock);
+	if (srq->ip)
+		kref_put(&srq->ip->ref, qib_release_mmap_info);
+	else
+		vfree(srq->rq.wq);
+	kfree(srq);
+
+	return 0;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Ralph Campbell]

creates the qib_sysfs.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_sysfs.c |  736 +++++++++++++++++++++++++++++++++
 1 files changed, 736 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_sysfs.c

diff --git a/drivers/infiniband/hw/qib/qib_sysfs.c b/drivers/infiniband/hw/qib/qib_sysfs.c
new file mode 100644
index 0000000..747005a
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_sysfs.c
@@ -0,0 +1,736 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/ctype.h>
+
+#include "qib.h"
+
+/**
+ * qib_parse_ushort - parse an unsigned short value in an arbitrary base
+ * @str: the string containing the number
+ * @valp: where to put the result
+ *
+ * Returns the number of bytes consumed, or negative value on error.
+ */
+static int qib_parse_ushort(const char *str, unsigned short *valp)
+{
+	unsigned long val;
+	char *end;
+	int ret;
+
+	if (!isdigit(str[0])) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	val = simple_strtoul(str, &end, 0);
+
+	if (val > 0xffff) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	*valp = val;
+
+	ret = end + 1 - str;
+	if (ret == 0)
+		ret = -EINVAL;
+
+bail:
+	return ret;
+}
+
+/* start of per-port functions */
+/*
+ * Get/Set heartbeat enable. OR of 1=enabled, 2=auto
+ */
+static ssize_t show_hrtbt_enb(struct qib_pportdata *ppd, char *buf)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret;
+
+	ret = dd->f_get_ib_cfg(ppd, QIB_IB_CFG_HRTBT);
+	ret = scnprintf(buf, PAGE_SIZE, "%d\n", ret);
+	return ret;
+}
+
+static ssize_t store_hrtbt_enb(struct qib_pportdata *ppd, const char *buf,
+			       size_t count)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret;
+	u16 val;
+
+	ret = qib_parse_ushort(buf, &val);
+
+	/*
+	 * Set the "intentional" heartbeat enable per either of
+	 * "Enable" and "Auto", as these are normally set together.
+	 * This bit is consulted when leaving loopback mode,
+	 * because entering loopback mode overrides it and automatically
+	 * disables heartbeat.
+	 */
+	if (ret >= 0)
+		ret = dd->f_set_ib_cfg(ppd, QIB_IB_CFG_HRTBT, val);
+	if (ret < 0)
+		qib_dev_err(dd, "attempt to set invalid Heartbeat enable\n");
+	return ret < 0 ? ret : count;
+}
+
+static ssize_t store_loopback(struct qib_pportdata *ppd, const char *buf,
+			      size_t count)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret = count, r;
+
+	r = dd->f_set_ib_loopback(ppd, buf);
+	if (r < 0)
+		ret = r;
+
+	return ret;
+}
+
+static ssize_t store_led_override(struct qib_pportdata *ppd, const char *buf,
+				  size_t count)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret;
+	u16 val;
+
+	ret = qib_parse_ushort(buf, &val);
+	if (ret > 0)
+		qib_set_led_override(ppd, val);
+	else
+		qib_dev_err(dd, "attempt to set invalid LED override\n");
+	return ret < 0 ? ret : count;
+}
+
+static ssize_t show_qsfp(struct qib_pportdata *ppd, char *buf)
+{
+	ssize_t ret;
+
+	ret = qib_qsfp_dump(ppd, buf, PAGE_SIZE);
+	return ret;
+}
+
+static ssize_t show_status(struct qib_pportdata *ppd, char *buf)
+{
+	ssize_t ret;
+
+	if (!ppd->statusp)
+		ret = -EINVAL;
+	else
+		ret = scnprintf(buf, PAGE_SIZE, "0x%llx\n",
+				(unsigned long long) *(ppd->statusp));
+	return ret;
+}
+
+/*
+ * For userland compatibility, these offsets must remain fixed.
+ * They are strings for QIB_STATUS_*
+ */
+static const char *qib_status_str[] = {
+	"Initted",
+	"",
+	"",
+	"",
+	"",
+	"Present",
+	"IB_link_up",
+	"IB_configured",
+	"",
+	"Fatal_Hardware_Error",
+	NULL,
+};
+
+static ssize_t show_status_str(struct qib_pportdata *ppd, char *buf)
+{
+	int i, any;
+	u64 s;
+	ssize_t ret;
+
+	if (!ppd->statusp) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	s = *(ppd->statusp);
+	*buf = '\0';
+	for (any = i = 0; s && qib_status_str[i]; i++) {
+		if (s & 1) {
+			/* if overflow */
+			if (any && strlcat(buf, " ", PAGE_SIZE) >= PAGE_SIZE)
+				break;
+			if (strlcat(buf, qib_status_str[i], PAGE_SIZE) >=
+					PAGE_SIZE)
+				break;
+			any = 1;
+		}
+		s >>= 1;
+	}
+	if (any)
+		strlcat(buf, "\n", PAGE_SIZE);
+
+	ret = strlen(buf);
+
+bail:
+	return ret;
+}
+
+/* end of per-port functions */
+
+/*
+ * Start of per-port file structures and support code
+ * Because we are fitting into other infrastructure, we have to supply the
+ * full set of kobject/sysfs_ops structures and routines.
+ */
+#define QIB_PORT_ATTR(name, mode, show, store) \
+	static struct qib_port_attr qib_port_attr_##name = \
+		__ATTR(name, mode, show, store)
+
+struct qib_port_attr {
+	struct attribute attr;
+	ssize_t (*show)(struct qib_pportdata *, char *);
+	ssize_t (*store)(struct qib_pportdata *, const char *, size_t);
+};
+
+QIB_PORT_ATTR(loopback, S_IWUSR, NULL, store_loopback);
+QIB_PORT_ATTR(led_override, S_IWUSR, NULL, store_led_override);
+QIB_PORT_ATTR(hrtbt_enable, S_IWUSR | S_IRUGO, show_hrtbt_enb,
+	      store_hrtbt_enb);
+QIB_PORT_ATTR(qsfp, S_IRUGO, show_qsfp, NULL);
+QIB_PORT_ATTR(status, S_IRUGO, show_status, NULL);
+QIB_PORT_ATTR(status_str, S_IRUGO, show_status_str, NULL);
+
+static struct attribute *port_default_attributes[] = {
+	&qib_port_attr_loopback.attr,
+	&qib_port_attr_led_override.attr,
+	&qib_port_attr_hrtbt_enable.attr,
+	&qib_port_attr_status.attr,
+	&qib_port_attr_status_str.attr,
+	NULL
+};
+
+static ssize_t qib_portattr_show(struct kobject *kobj,
+	struct attribute *attr, char *buf)
+{
+	struct qib_port_attr *pattr =
+		container_of(attr, struct qib_port_attr, attr);
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, pport_kobj);
+
+	return pattr->show(ppd, buf);
+}
+
+static ssize_t qib_portattr_store(struct kobject *kobj,
+	struct attribute *attr, const char *buf, size_t len)
+{
+	struct qib_port_attr *pattr =
+		container_of(attr, struct qib_port_attr, attr);
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, pport_kobj);
+
+	return pattr->store(ppd, buf, len);
+}
+
+static void qib_port_release(struct kobject *kobj)
+{
+	/* nothing to do since memory is freed by qib_free_devdata() */
+}
+
+static struct sysfs_ops qib_port_ops = {
+	.show = qib_portattr_show,
+	.store = qib_portattr_store,
+};
+
+static struct kobj_type qib_port_ktype = {
+	.release = qib_port_release,
+	.sysfs_ops = &qib_port_ops,
+	.default_attrs = port_default_attributes
+};
+
+/* Start sl2vl */
+
+#define QIB_SL2VL_ATTR(N) \
+	static struct qib_sl2vl_attr qib_sl2vl_attr_##N = { \
+		.attr = { .name = __stringify(N), .mode = 0444 }, \
+		.sl = N \
+	}
+
+struct qib_sl2vl_attr {
+	struct attribute attr;
+	int sl;
+};
+
+QIB_SL2VL_ATTR(0);
+QIB_SL2VL_ATTR(1);
+QIB_SL2VL_ATTR(2);
+QIB_SL2VL_ATTR(3);
+QIB_SL2VL_ATTR(4);
+QIB_SL2VL_ATTR(5);
+QIB_SL2VL_ATTR(6);
+QIB_SL2VL_ATTR(7);
+QIB_SL2VL_ATTR(8);
+QIB_SL2VL_ATTR(9);
+QIB_SL2VL_ATTR(10);
+QIB_SL2VL_ATTR(11);
+QIB_SL2VL_ATTR(12);
+QIB_SL2VL_ATTR(13);
+QIB_SL2VL_ATTR(14);
+QIB_SL2VL_ATTR(15);
+
+static struct attribute *sl2vl_default_attributes[] = {
+	&qib_sl2vl_attr_0.attr,
+	&qib_sl2vl_attr_1.attr,
+	&qib_sl2vl_attr_2.attr,
+	&qib_sl2vl_attr_3.attr,
+	&qib_sl2vl_attr_4.attr,
+	&qib_sl2vl_attr_5.attr,
+	&qib_sl2vl_attr_6.attr,
+	&qib_sl2vl_attr_7.attr,
+	&qib_sl2vl_attr_8.attr,
+	&qib_sl2vl_attr_9.attr,
+	&qib_sl2vl_attr_10.attr,
+	&qib_sl2vl_attr_11.attr,
+	&qib_sl2vl_attr_12.attr,
+	&qib_sl2vl_attr_13.attr,
+	&qib_sl2vl_attr_14.attr,
+	&qib_sl2vl_attr_15.attr,
+	NULL
+};
+
+static ssize_t sl2vl_attr_show(struct kobject *kobj, struct attribute *attr,
+			       char *buf)
+{
+	struct qib_sl2vl_attr *sattr =
+		container_of(attr, struct qib_sl2vl_attr, attr);
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, sl2vl_kobj);
+	struct qib_ibport *qibp = &ppd->ibport_data;
+
+	return sprintf(buf, "%u\n", qibp->sl_to_vl[sattr->sl]);
+}
+
+static struct sysfs_ops qib_sl2vl_ops = {
+	.show = sl2vl_attr_show,
+};
+
+static struct kobj_type qib_sl2vl_ktype = {
+	.release = qib_port_release,
+	.sysfs_ops = &qib_sl2vl_ops,
+	.default_attrs = sl2vl_default_attributes
+};
+
+/* End sl2vl */
+
+/* end of per-port file structures and support code */
+
+/*
+ * Start of per-unit (or driver, in some cases, but replicated
+ * per unit) functions (these get a device *)
+ */
+static ssize_t show_rev(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+
+	return sprintf(buf, "%x\n", dd_from_dev(dev)->minrev);
+}
+
+static ssize_t show_hca(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	int ret;
+
+	if (!dd->boardname)
+		ret = -EINVAL;
+	else
+		ret = scnprintf(buf, PAGE_SIZE, "%s\n", dd->boardname);
+	return ret;
+}
+
+static const char *qp_type_str[] = {
+	"SMI", "GSI", "RC", "UC", "UD",
+};
+
+static ssize_t show_stats(struct device *device, struct device_attribute *attr,
+			  char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	unsigned pidx;
+	unsigned i;
+	int len = 0;
+	unsigned long flags;
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		struct qib_ibport *ibp = &dd->pport[pidx].ibport_data;
+
+		len += sprintf(buf + len,
+			       "Port %d:\n"
+			       "RC timeouts %d\n"
+			       "RC resends  %d\n"
+			       "RC QACKs    %d\n"
+			       "RC SEQ NAKs %d\n"
+			       "RC RDMA seq %d\n"
+			       "RC RNR NAKs %d\n"
+			       "RC OTH NAKs %d\n"
+			       "RC DComp    %d\n"
+			       "RCr dup req %d\n"
+			       "RCr SEQ NAK %d\n"
+			       "wait piobuf %d\n"
+			       "wait DMA    %d\n"
+			       "wait TX     %d\n"
+			       "unaligned   %d\n"
+			       "loop pkts   %d\n"
+			       "PKT drops   %d\n",
+			       dd->pport[pidx].port,
+			       ibp->n_timeouts, ibp->n_rc_resends,
+			       ibp->n_rc_qacks, ibp->n_seq_naks,
+			       ibp->n_rdma_seq, ibp->n_rnr_naks,
+			       ibp->n_other_naks, ibp->n_rc_delayed_comp,
+			       ibp->n_rc_dupreq, ibp->n_rc_seqnak,
+			       dev->n_piowait, ibp->n_dmawait, dev->n_txwait,
+			       ibp->n_unaligned, ibp->n_loop_pkts,
+			       ibp->n_pkt_drops);
+		for (i = 0; i < ARRAY_SIZE(ibp->opstats); i++) {
+			const struct qib_opcode_stats *si = &ibp->opstats[i];
+
+			if (!si->n_packets && !si->n_bytes)
+				continue;
+			len += sprintf(buf + len, "%02x %llu/%llu\n", i,
+				       (unsigned long long) si->n_packets,
+				       (unsigned long long) si->n_bytes);
+		}
+	}
+	len += sprintf(buf + len, "Ctx:npkts");
+	for (i = 0; i < dd->first_user_ctxt; i++) {
+		if (!dd->rcd[i])
+			continue;
+		len += sprintf(buf + len, " %u:%u", i,
+			       dd->rcd[i]->pkt_count);
+	}
+	len += sprintf(buf + len, "\n");
+	spin_lock_irqsave(&dev->qpt_lock, flags);
+	for (i = 0; i < dev->qp_table_size; i++) {
+		struct qib_qp *qp;
+		for (qp = dev->qp_table[i]; qp != NULL; qp = qp->next) {
+			struct qib_swqe *wqe;
+
+			if (qp->s_last == qp->s_acked &&
+			    qp->s_acked == qp->s_cur &&
+			    qp->s_cur == qp->s_tail &&
+			    qp->s_tail == qp->s_head)
+				continue;
+			if (len + 128 >= PAGE_SIZE)
+				break;
+			wqe = get_swqe_ptr(qp, qp->s_last);
+			len += sprintf(buf + len,
+				       "QP%u %s %u %u %u f=%x %u %u %u %u %u "
+				       "PSN %x %x %x %x %x "
+				       "(%u %u %u %u %u %u) QP%u LID %x\n",
+				       qp->ibqp.qp_num,
+				       qp_type_str[qp->ibqp.qp_type],
+				       qp->state,
+				       wqe->wr.opcode,
+				       qp->s_hdrwords,
+				       qp->s_flags,
+				       atomic_read(&qp->s_dma_busy),
+				       !list_empty(&qp->iowait),
+				       qp->timeout,
+				       wqe->ssn,
+				       qp->s_lsn,
+				       qp->s_last_psn,
+				       qp->s_psn, qp->s_next_psn,
+				       qp->s_sending_psn, qp->s_sending_hpsn,
+				       qp->s_last, qp->s_acked, qp->s_cur,
+				       qp->s_tail, qp->s_head, qp->s_size,
+				       qp->remote_qpn,
+				       qp->remote_ah_attr.dlid);
+		}
+	}
+	spin_unlock_irqrestore(&dev->qpt_lock, flags);
+	return len;
+}
+
+static ssize_t show_version(struct device *device,
+			    struct device_attribute *attr, char *buf)
+{
+	/* The string printed here is already newline-terminated. */
+	return scnprintf(buf, PAGE_SIZE, "%s", (char *)ib_qib_version);
+}
+
+static ssize_t show_boardversion(struct device *device,
+				 struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+
+	/* The string printed here is already newline-terminated. */
+	return scnprintf(buf, PAGE_SIZE, "%s", dd->boardversion);
+}
+
+
+static ssize_t show_localbus_info(struct device *device,
+				  struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+
+	/* The string printed here is already newline-terminated. */
+	return scnprintf(buf, PAGE_SIZE, "%s", dd->lbus_info);
+}
+
+
+static ssize_t show_nctxts(struct device *device,
+			   struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+
+	/* Return the number of user ports (contexts) available. */
+	return scnprintf(buf, PAGE_SIZE, "%u\n", dd->cfgctxts -
+		dd->first_user_ctxt);
+}
+
+static ssize_t show_serial(struct device *device,
+			   struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+
+	buf[sizeof dd->serial] = '\0';
+	memcpy(buf, dd->serial, sizeof dd->serial);
+	strcat(buf, "\n");
+	return strlen(buf);
+}
+
+static ssize_t store_chip_reset(struct device *device,
+				struct device_attribute *attr, const char *buf,
+				size_t count)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	int ret;
+
+	if (count < 5 || memcmp(buf, "reset", 5) || !dd->diag_client) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	ret = qib_reset_device(dd->unit);
+bail:
+	return ret < 0 ? ret : count;
+}
+
+static ssize_t show_logged_errs(struct device *device,
+				struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	int idx, count;
+
+	/* force consistency with actual EEPROM */
+	if (qib_update_eeprom_log(dd) != 0)
+		return -ENXIO;
+
+	count = 0;
+	for (idx = 0; idx < QIB_EEP_LOG_CNT; ++idx) {
+		count += scnprintf(buf + count, PAGE_SIZE - count, "%d%c",
+				   dd->eep_st_errs[idx],
+				   idx == (QIB_EEP_LOG_CNT - 1) ? '\n' : ' ');
+	}
+
+	return count;
+}
+
+/*
+ * Dump tempsense regs. in decimal, to ease shell-scripts.
+ */
+static ssize_t show_tempsense(struct device *device,
+			      struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	int ret;
+	int idx;
+	u8 regvals[8];
+
+	ret = -ENXIO;
+	for (idx = 0; idx < 8; ++idx) {
+		if (idx == 6)
+			continue;
+		ret = dd->f_tempsense_rd(dd, idx);
+		if (ret < 0)
+			break;
+		regvals[idx] = ret;
+	}
+	if (idx == 8)
+		ret = scnprintf(buf, PAGE_SIZE, "%d %d %02X %02X %d %d\n",
+				*(signed char *)(regvals),
+				*(signed char *)(regvals + 1),
+				regvals[2], regvals[3],
+				*(signed char *)(regvals + 5),
+				*(signed char *)(regvals + 7));
+	return ret;
+}
+
+/*
+ * end of per-unit (or driver, in some cases, but replicated
+ * per unit) functions
+ */
+
+/* start of per-unit file structures and support code */
+static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
+static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
+static DEVICE_ATTR(board_id, S_IRUGO, show_hca, NULL);
+static DEVICE_ATTR(stats, S_IRUGO, show_stats, NULL);
+static DEVICE_ATTR(version, S_IRUGO, show_version, NULL);
+static DEVICE_ATTR(nctxts, S_IRUGO, show_nctxts, NULL);
+static DEVICE_ATTR(serial, S_IRUGO, show_serial, NULL);
+static DEVICE_ATTR(boardversion, S_IRUGO, show_boardversion, NULL);
+static DEVICE_ATTR(logged_errors, S_IRUGO, show_logged_errs, NULL);
+static DEVICE_ATTR(tempsense, S_IRUGO, show_tempsense, NULL);
+static DEVICE_ATTR(localbus_info, S_IRUGO, show_localbus_info, NULL);
+static DEVICE_ATTR(chip_reset, S_IWUSR, NULL, store_chip_reset);
+
+static struct device_attribute *qib_attributes[] = {
+	&dev_attr_hw_rev,
+	&dev_attr_hca_type,
+	&dev_attr_board_id,
+	&dev_attr_stats,
+	&dev_attr_version,
+	&dev_attr_nctxts,
+	&dev_attr_serial,
+	&dev_attr_boardversion,
+	&dev_attr_logged_errors,
+	&dev_attr_tempsense,
+	&dev_attr_localbus_info,
+	&dev_attr_chip_reset,
+};
+
+static int create_port_files(struct ib_device *ibdev, u8 port_num,
+			     struct kobject *kobj)
+{
+	struct qib_pportdata *ppd;
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	int ret;
+
+	if (!port_num || port_num > dd->num_pports) {
+		qib_dev_err(dd, "Skipping infiniband class with "
+			    "invalid port %u\n", port_num);
+		ret = -ENODEV;
+		goto bail;
+	}
+	ppd = &dd->pport[port_num - 1];
+
+	ret = kobject_init_and_add(&ppd->pport_kobj, &qib_port_ktype, kobj,
+				   "linkcontrol");
+	if (ret) {
+		qib_dev_err(dd, "Skipping linkcontrol sysfs info, "
+			    "(err %d) port %u\n", ret, port_num);
+		goto bail;
+	}
+
+	if (dd->flags & QIB_HAS_QSFP) {
+		ret = sysfs_create_file(&ppd->pport_kobj,
+					&qib_port_attr_qsfp.attr);
+		if (ret) {
+			qib_dev_err(dd, "Skipping qsfp sysfs info, "
+				    "(err %d) port %u\n", ret, port_num);
+			goto bail;
+		}
+	}
+	kobject_uevent(&ppd->pport_kobj, KOBJ_ADD);
+
+	ret = kobject_init_and_add(&ppd->sl2vl_kobj, &qib_sl2vl_ktype, kobj,
+				   "sl2vl");
+	if (ret) {
+		qib_dev_err(dd, "Skipping sl2vl sysfs info, "
+			    "(err %d) port %u\n", ret, port_num);
+		goto bail_sl;
+	}
+	kobject_uevent(&ppd->sl2vl_kobj, KOBJ_ADD);
+
+	return 0;
+
+bail_sl:
+	kobject_put(&ppd->pport_kobj);
+bail:
+	return ret;
+}
+
+/*
+ * Register and create our files in /sys/class/infiniband.
+ */
+int qib_verbs_register_sysfs(struct qib_devdata *dd)
+{
+	struct ib_device *dev = &dd->verbs_dev.ibdev;
+	int i, ret;
+
+	for (i = 0; i < ARRAY_SIZE(qib_attributes); ++i)
+		if (device_create_file(&dev->dev, qib_attributes[i])) {
+			ret = 1;
+			goto bail;
+		}
+
+	ret = ib_sysfs_create_port_files(dev, create_port_files);
+bail:
+	return ret;
+}
+
+/*
+ * Unregister and remove our files in /sys/class/infiniband.
+ */
+void qib_verbs_unregister_sysfs(struct qib_devdata *dd)
+{
+	struct qib_pportdata *ppd;
+	int i;
+
+	for (i = 0; i < dd->num_pports; i++) {
+		ppd = &dd->pport[i];
+		kobject_put(&ppd->pport_kobj);
+		kobject_put(&ppd->sl2vl_kobj);
+	}
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Roland Dreier]

I didn't audit everything (so there may be more problems here too), but
certainly this violates the "one value per sysfs file" rule rather
egregiously:

 > +static ssize_t show_stats(struct device *device, struct device_attribute *attr,
 > +			  char *buf)
 > +{
 > +	struct qib_ibdev *dev =
 > +		container_of(device, struct qib_ibdev, ibdev.dev);
 > +	struct qib_devdata *dd = dd_from_dev(dev);
 > +	unsigned pidx;
 > +	unsigned i;
 > +	int len = 0;
 > +	unsigned long flags;
 > +
 > +	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
 > +		struct qib_ibport *ibp = &dd->pport[pidx].ibport_data;
 > +
 > +		len += sprintf(buf + len,
 > +			       "Port %d:\n"
 > +			       "RC timeouts %d\n"
 > +			       "RC resends  %d\n"
 > +			       "RC QACKs    %d\n"
 > +			       "RC SEQ NAKs %d\n"
 > +			       "RC RDMA seq %d\n"
 > +			       "RC RNR NAKs %d\n"
 > +			       "RC OTH NAKs %d\n"
 > +			       "RC DComp    %d\n"
 > +			       "RCr dup req %d\n"
 > +			       "RCr SEQ NAK %d\n"
 > +			       "wait piobuf %d\n"
 > +			       "wait DMA    %d\n"
 > +			       "wait TX     %d\n"
 > +			       "unaligned   %d\n"
 > +			       "loop pkts   %d\n"
 > +			       "PKT drops   %d\n",
 > +			       dd->pport[pidx].port,
 > +			       ibp->n_timeouts, ibp->n_rc_resends,
 > +			       ibp->n_rc_qacks, ibp->n_seq_naks,
 > +			       ibp->n_rdma_seq, ibp->n_rnr_naks,
 > +			       ibp->n_other_naks, ibp->n_rc_delayed_comp,
 > +			       ibp->n_rc_dupreq, ibp->n_rc_seqnak,
 > +			       dev->n_piowait, ibp->n_dmawait, dev->n_txwait,
 > +			       ibp->n_unaligned, ibp->n_loop_pkts,
 > +			       ibp->n_pkt_drops);
 > +		for (i = 0; i < ARRAY_SIZE(ibp->opstats); i++) {
 > +			const struct qib_opcode_stats *si = &ibp->opstats[i];
 > +
 > +			if (!si->n_packets && !si->n_bytes)
 > +				continue;
 > +			len += sprintf(buf + len, "%02x %llu/%llu\n", i,
 > +				       (unsigned long long) si->n_packets,
 > +				       (unsigned long long) si->n_bytes);
 > +		}
 > +	}
 > +	len += sprintf(buf + len, "Ctx:npkts");
 > +	for (i = 0; i < dd->first_user_ctxt; i++) {
 > +		if (!dd->rcd[i])
 > +			continue;
 > +		len += sprintf(buf + len, " %u:%u", i,
 > +			       dd->rcd[i]->pkt_count);
 > +	}
 > +	len += sprintf(buf + len, "\n");
 > +	spin_lock_irqsave(&dev->qpt_lock, flags);
 > +	for (i = 0; i < dev->qp_table_size; i++) {
 > +		struct qib_qp *qp;
 > +		for (qp = dev->qp_table[i]; qp != NULL; qp = qp->next) {
 > +			struct qib_swqe *wqe;
 > +
 > +			if (qp->s_last == qp->s_acked &&
 > +			    qp->s_acked == qp->s_cur &&
 > +			    qp->s_cur == qp->s_tail &&
 > +			    qp->s_tail == qp->s_head)
 > +				continue;
 > +			if (len + 128 >= PAGE_SIZE)
 > +				break;
 > +			wqe = get_swqe_ptr(qp, qp->s_last);
 > +			len += sprintf(buf + len,
 > +				       "QP%u %s %u %u %u f=%x %u %u %u %u %u "
 > +				       "PSN %x %x %x %x %x "
 > +				       "(%u %u %u %u %u %u) QP%u LID %x\n",
 > +				       qp->ibqp.qp_num,
 > +				       qp_type_str[qp->ibqp.qp_type],
 > +				       qp->state,
 > +				       wqe->wr.opcode,
 > +				       qp->s_hdrwords,
 > +				       qp->s_flags,
 > +				       atomic_read(&qp->s_dma_busy),
 > +				       !list_empty(&qp->iowait),
 > +				       qp->timeout,
 > +				       wqe->ssn,
 > +				       qp->s_lsn,
 > +				       qp->s_last_psn,
 > +				       qp->s_psn, qp->s_next_psn,
 > +				       qp->s_sending_psn, qp->s_sending_hpsn,
 > +				       qp->s_last, qp->s_acked, qp->s_cur,
 > +				       qp->s_tail, qp->s_head, qp->s_size,
 > +				       qp->remote_qpn,
 > +				       qp->remote_ah_attr.dlid);
 > +		}
 > +	}
 > +	spin_unlock_irqrestore(&dev->qpt_lock, flags);
 > +	return len;
 > +}

-- 
Roland Dreier <rolandd-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> || For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Ralph Campbell]

This was for debugging and clearly could use a better method.
Some stats are definitely useful and having files per value
would make scripting easier too.
I could add /sys/class/infiniband/qib0/ports/1/diag_counters/*

One other place that has multiple values is dumping the QSFP
data from the IB cable.

On Thu, 2010-04-08 at 13:50 -0700, Roland Dreier wrote:
> I didn't audit everything (so there may be more problems here too), but
> certainly this violates the "one value per sysfs file" rule rather
> egregiously:
> 
>  > +static ssize_t show_stats(struct device *device, struct device_attribute *attr,
>  > +			  char *buf)
>  > +{
>  > +	struct qib_ibdev *dev =
>  > +		container_of(device, struct qib_ibdev, ibdev.dev);
>  > +	struct qib_devdata *dd = dd_from_dev(dev);
>  > +	unsigned pidx;
>  > +	unsigned i;
>  > +	int len = 0;
>  > +	unsigned long flags;
>  > +
>  > +	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
>  > +		struct qib_ibport *ibp = &dd->pport[pidx].ibport_data;
>  > +
>  > +		len += sprintf(buf + len,
>  > +			       "Port %d:\n"
>  > +			       "RC timeouts %d\n"
>  > +			       "RC resends  %d\n"
>  > +			       "RC QACKs    %d\n"
>  > +			       "RC SEQ NAKs %d\n"
>  > +			       "RC RDMA seq %d\n"
>  > +			       "RC RNR NAKs %d\n"
>  > +			       "RC OTH NAKs %d\n"
>  > +			       "RC DComp    %d\n"
>  > +			       "RCr dup req %d\n"
>  > +			       "RCr SEQ NAK %d\n"
>  > +			       "wait piobuf %d\n"
>  > +			       "wait DMA    %d\n"
>  > +			       "wait TX     %d\n"
>  > +			       "unaligned   %d\n"
>  > +			       "loop pkts   %d\n"
>  > +			       "PKT drops   %d\n",
>  > +			       dd->pport[pidx].port,
>  > +			       ibp->n_timeouts, ibp->n_rc_resends,
>  > +			       ibp->n_rc_qacks, ibp->n_seq_naks,
>  > +			       ibp->n_rdma_seq, ibp->n_rnr_naks,
>  > +			       ibp->n_other_naks, ibp->n_rc_delayed_comp,
>  > +			       ibp->n_rc_dupreq, ibp->n_rc_seqnak,
>  > +			       dev->n_piowait, ibp->n_dmawait, dev->n_txwait,
>  > +			       ibp->n_unaligned, ibp->n_loop_pkts,
>  > +			       ibp->n_pkt_drops);
>  > +		for (i = 0; i < ARRAY_SIZE(ibp->opstats); i++) {
>  > +			const struct qib_opcode_stats *si = &ibp->opstats[i];
>  > +
>  > +			if (!si->n_packets && !si->n_bytes)
>  > +				continue;
>  > +			len += sprintf(buf + len, "%02x %llu/%llu\n", i,
>  > +				       (unsigned long long) si->n_packets,
>  > +				       (unsigned long long) si->n_bytes);
>  > +		}
>  > +	}
>  > +	len += sprintf(buf + len, "Ctx:npkts");
>  > +	for (i = 0; i < dd->first_user_ctxt; i++) {
>  > +		if (!dd->rcd[i])
>  > +			continue;
>  > +		len += sprintf(buf + len, " %u:%u", i,
>  > +			       dd->rcd[i]->pkt_count);
>  > +	}
>  > +	len += sprintf(buf + len, "\n");
>  > +	spin_lock_irqsave(&dev->qpt_lock, flags);
>  > +	for (i = 0; i < dev->qp_table_size; i++) {
>  > +		struct qib_qp *qp;
>  > +		for (qp = dev->qp_table[i]; qp != NULL; qp = qp->next) {
>  > +			struct qib_swqe *wqe;
>  > +
>  > +			if (qp->s_last == qp->s_acked &&
>  > +			    qp->s_acked == qp->s_cur &&
>  > +			    qp->s_cur == qp->s_tail &&
>  > +			    qp->s_tail == qp->s_head)
>  > +				continue;
>  > +			if (len + 128 >= PAGE_SIZE)
>  > +				break;
>  > +			wqe = get_swqe_ptr(qp, qp->s_last);
>  > +			len += sprintf(buf + len,
>  > +				       "QP%u %s %u %u %u f=%x %u %u %u %u %u "
>  > +				       "PSN %x %x %x %x %x "
>  > +				       "(%u %u %u %u %u %u) QP%u LID %x\n",
>  > +				       qp->ibqp.qp_num,
>  > +				       qp_type_str[qp->ibqp.qp_type],
>  > +				       qp->state,
>  > +				       wqe->wr.opcode,
>  > +				       qp->s_hdrwords,
>  > +				       qp->s_flags,
>  > +				       atomic_read(&qp->s_dma_busy),
>  > +				       !list_empty(&qp->iowait),
>  > +				       qp->timeout,
>  > +				       wqe->ssn,
>  > +				       qp->s_lsn,
>  > +				       qp->s_last_psn,
>  > +				       qp->s_psn, qp->s_next_psn,
>  > +				       qp->s_sending_psn, qp->s_sending_hpsn,
>  > +				       qp->s_last, qp->s_acked, qp->s_cur,
>  > +				       qp->s_tail, qp->s_head, qp->s_size,
>  > +				       qp->remote_qpn,
>  > +				       qp->remote_ah_attr.dlid);
>  > +		}
>  > +	}
>  > +	spin_unlock_irqrestore(&dev->qpt_lock, flags);
>  > +	return len;
>  > +}
> 


--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Roland Dreier]

 > This was for debugging and clearly could use a better method.
 > Some stats are definitely useful and having files per value
 > would make scripting easier too.
 > I could add /sys/class/infiniband/qib0/ports/1/diag_counters/*
 > 
 > One other place that has multiple values is dumping the QSFP
 > data from the IB cable.

You can stick debugging data under debugfs.  Or just kill it for now and
figure out how to add it back later.
-- 
Roland Dreier <rolandd-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> || For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Ira Weiny]

On Thu, 08 Apr 2010 14:33:59 -0700
Roland Dreier <rdreier-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> wrote:

>  > This was for debugging and clearly could use a better method.
>  > Some stats are definitely useful and having files per value
>  > would make scripting easier too.
>  > I could add /sys/class/infiniband/qib0/ports/1/diag_counters/*
>  > 
>  > One other place that has multiple values is dumping the QSFP
>  > data from the IB cable.
> 
> You can stick debugging data under debugfs.  Or just kill it for now and
> figure out how to add it back later.

I for one would rather not see this die.  We have debugged some critical
issues using this data.  The sysfs entries above are what Mellanox uses.
Should those also be changed?

Ira

> -- 
> Roland Dreier <rolandd-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> || For corporate legal information go to:
> http://*www.*cisco.com/web/about/doing_business/legal/cri/index.html
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://*vger.kernel.org/majordomo-info.html
> 


-- 
Ira Weiny
Math Programmer/Computer Scientist
Lawrence Livermore National Lab
925-423-8008
weiny2-i2BcT+NCU+M@public.gmane.org
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Roland Dreier]

 > I for one would rather not see this die.  We have debugged some critical
 > issues using this data.  The sysfs entries above are what Mellanox uses.
 > Should those also be changed?

Which sysfs entries?  I don't see any likely looking code in either of
the Mellanox drivers.

 - R.

-- 
Roland Dreier <rolandd-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> || For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Ralph Campbell]

On Thu, 2010-04-08 at 16:29 -0700, Roland Dreier wrote:
> > I for one would rather not see this die.  We have debugged some critical
>  > issues using this data.  The sysfs entries above are what Mellanox uses.
>  > Should those also be changed?
> 
> Which sysfs entries?  I don't see any likely looking code in either of
> the Mellanox drivers.
> 
>  - R.

I think Ira means:

# ls /sys/class/infiniband/mlx4_0/diag_counters/
clear_diag    rq_num_lqpoe    rq_num_udsdprd  sq_num_lqpoe   sq_num_rree
num_baddb     rq_num_mce      rq_num_wrfe     sq_num_mwbe    sq_num_rsync
num_cqovf     rq_num_oos      sq_num_bre      sq_num_oos     sq_num_tree
num_eqovf     rq_num_rae      sq_num_ieecne   sq_num_rabrte  sq_num_wrfe
rq_num_lae    rq_num_rire     sq_num_ieecse   sq_num_rae
rq_num_leeoe  rq_num_rnr      sq_num_leeoe    sq_num_rire
rq_num_lle    rq_num_rsync    sq_num_lle      sq_num_rnr
rq_num_lpe    rq_num_ucsdprd  sq_num_lpe      sq_num_roe


I couldn't find where these entries are added in the kernel.org code.
It is in the OFED tree though.

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Ira Weiny]

On Thu, 08 Apr 2010 16:29:06 -0700
Roland Dreier <rdreier-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> wrote:

>  > I for one would rather not see this die.  We have debugged some critical
>  > issues using this data.  The sysfs entries above are what Mellanox uses.
>  > Should those also be changed?
> 
> Which sysfs entries?  I don't see any likely looking code in either of
> the Mellanox drivers.

My apologies.  It appears this an OFED thing.[*]

16:31:04 > ls -la /sys/class/infiniband/mlx4_0/diag_counters/
total 0
drwxr-xr-x 2 root root    0 Apr  8 16:31 ./
drwxr-xr-x 4 root root    0 Apr  8 16:31 ../
--w--w--w- 1 root root 4096 Apr  8 16:31 clear_diag
-r--r--r-- 1 root root 4096 Apr  8 16:31 num_baddb
-r--r--r-- 1 root root 4096 Apr  8 16:31 num_cqovf
...

This is with the current RHEL5.4 drivers.

<sigh>

So is there an equivalent for mlx4 in the upstream kernel?  I don't see them.

Do you feel it is appropriate for the port counters to be in sysfs?

16:46:19 > ls -la /sys/class/infiniband/mlx4_0/ports/1/counters/
total 0
drwxr-xr-x 2 root root    0 Apr  8 16:46 ./
drwxr-xr-x 5 root root    0 Apr  6 09:33 ../
-r--r--r-- 1 root root 4096 Apr  8 16:46 VL15_dropped
-r--r--r-- 1 root root 4096 Apr  8 16:46 excessive_buffer_overrun_errors
...

Although these "diag_counters" are internal to the cards they do track counts
which are part of the IB spec like RNR NAK's.  So I am not sure they are
really "debug" in the strictest sense.

I will rephrase my statement, I would like to see these counters included as
they give valuable information about what is going on in a cluster.

Ira

* OFED MUST die!

> 
>  - R.
> 
> -- 
> Roland Dreier <rolandd-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org> || For corporate legal information go to:
> http://*www.*cisco.com/web/about/doing_business/legal/cri/index.html
> 


-- 
Ira Weiny
Math Programmer/Computer Scientist
Lawrence Livermore National Lab
925-423-8008
weiny2-i2BcT+NCU+M@public.gmane.org
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Ralph Campbell]

On Thu, 2010-04-08 at 14:33 -0700, Roland Dreier wrote:
> > This was for debugging and clearly could use a better method.
>  > Some stats are definitely useful and having files per value
>  > would make scripting easier too.
>  > I could add /sys/class/infiniband/qib0/ports/1/diag_counters/*
>  > 
>  > One other place that has multiple values is dumping the QSFP
>  > data from the IB cable.
> 
> You can stick debugging data under debugfs.  Or just kill it for now and
> figure out how to add it back later.

OK. I replaced the /sys/class/infiniband/qib0/stats file with

% ls /sys/class/infiniband/qib0/ports/1/diag_counters/
dmawait     pkt_drops        rc_dupreq   rc_seqnak    rnr_naks
loop_pkts   rc_acks          rc_qacks    rc_timeouts  seq_naks
other_naks  rc_delayed_comp  rc_resends  rdma_seq     unaligned

These are all simple counter values.

For the QSFP data, I hope I can leave it as is since it is
related to the link state that the other files contain.
It is a read-only file so no issue with trying to set a value.

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Jason Gunthorpe]

On Fri, Apr 09, 2010 at 05:13:24PM -0700, Ralph Campbell wrote:

> For the QSFP data, I hope I can leave it as is since it is
> related to the link state that the other files contain.
> It is a read-only file so no issue with trying to set a value.

There was some flack for other stuff like this a while back.

IMHO, it would be appropriate to have a hex dump of the entire QFSP
EEPROM and leave parsing to userspace, or put the parsed version in
debugfs..

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Ralph Campbell]

On Fri, 2010-04-09 at 17:27 -0700, Jason Gunthorpe wrote:
> On Fri, Apr 09, 2010 at 05:13:24PM -0700, Ralph Campbell wrote:
> 
> > For the QSFP data, I hope I can leave it as is since it is
> > related to the link state that the other files contain.
> > It is a read-only file so no issue with trying to set a value.
> 
> There was some flack for other stuff like this a while back.
> 
> IMHO, it would be appropriate to have a hex dump of the entire QFSP
> EEPROM and leave parsing to userspace, or put the parsed version in
> debugfs..
> 
> Jason

OK. I will move it to our file system which is used
to export binary data.

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Jason Gunthorpe]

On Thu, Apr 08, 2010 at 02:29:53PM -0700, Ralph Campbell wrote:

> One other place that has multiple values is dumping the QSFP
> data from the IB cable.

I was going to comment that building your own I2C subsystem is kinda
strange, can't the in-kernel stuff be used?

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Ralph Campbell]

On Thu, 2010-04-08 at 15:08 -0700, Jason Gunthorpe wrote:
> On Thu, Apr 08, 2010 at 02:29:53PM -0700, Ralph Campbell wrote:
> 
> > One other place that has multiple values is dumping the QSFP
> > data from the IB cable.
> 
> I was going to comment that building your own I2C subsystem is kinda
> strange, can't the in-kernel stuff be used?
> 
> Jason

It is a bit strange but we have had this discussion before with
the ib_ipath driver. Basically, the devices connected to the bus
weren't really I2C compliant. Since we had to support the
non-standard parts, it was easier to have only one set of code
that could handle both.

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Jason Gunthorpe]

On Thu, Apr 08, 2010 at 03:26:41PM -0700, Ralph Campbell wrote:
> On Thu, 2010-04-08 at 15:08 -0700, Jason Gunthorpe wrote:
> > On Thu, Apr 08, 2010 at 02:29:53PM -0700, Ralph Campbell wrote:
> > 
> > > One other place that has multiple values is dumping the QSFP
> > > data from the IB cable.
> > 
> > I was going to comment that building your own I2C subsystem is kinda
> > strange, can't the in-kernel stuff be used?

> It is a bit strange but we have had this discussion before with
> the ib_ipath driver. Basically, the devices connected to the bus
> weren't really I2C compliant. Since we had to support the
> non-standard parts, it was easier to have only one set of code
> that could handle both.

Is this still true on qib? I didn't notice anything too obviously
weird, and QSFPs are clearly standard i2c..

The i2c layer has a pretty flexible definition of i2c these days..

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Ralph Campbell]

On Thu, 2010-04-08 at 15:50 -0700, Jason Gunthorpe wrote:
> On Thu, Apr 08, 2010 at 03:26:41PM -0700, Ralph Campbell wrote:
> > On Thu, 2010-04-08 at 15:08 -0700, Jason Gunthorpe wrote:
> > > On Thu, Apr 08, 2010 at 02:29:53PM -0700, Ralph Campbell wrote:
> > > 
> > > > One other place that has multiple values is dumping the QSFP
> > > > data from the IB cable.
> > > 
> > > I was going to comment that building your own I2C subsystem is kinda
> > > strange, can't the in-kernel stuff be used?
> 
> > It is a bit strange but we have had this discussion before with
> > the ib_ipath driver. Basically, the devices connected to the bus
> > weren't really I2C compliant. Since we had to support the
> > non-standard parts, it was easier to have only one set of code
> > that could handle both.
> 
> Is this still true on qib? I didn't notice anything too obviously
> weird, and QSFPs are clearly standard i2c..
> 
> The i2c layer has a pretty flexible definition of i2c these days..
> 
> Jason

I will take another look but don't be surprised if it takes me
awhile since I'm also working on addressing the other comments
too.

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Ralph Campbell]

On Thu, 2010-04-08 at 15:50 -0700, Jason Gunthorpe wrote:
> On Thu, Apr 08, 2010 at 03:26:41PM -0700, Ralph Campbell wrote:
> > On Thu, 2010-04-08 at 15:08 -0700, Jason Gunthorpe wrote:
> > > On Thu, Apr 08, 2010 at 02:29:53PM -0700, Ralph Campbell wrote:
> > > 
> > > > One other place that has multiple values is dumping the QSFP
> > > > data from the IB cable.
> > > 
> > > I was going to comment that building your own I2C subsystem is kinda
> > > strange, can't the in-kernel stuff be used?
> 
> > It is a bit strange but we have had this discussion before with
> > the ib_ipath driver. Basically, the devices connected to the bus
> > weren't really I2C compliant. Since we had to support the
> > non-standard parts, it was easier to have only one set of code
> > that could handle both.
> 
> Is this still true on qib? I didn't notice anything too obviously
> weird, and QSFPs are clearly standard i2c..
> 
> The i2c layer has a pretty flexible definition of i2c these days..
> 
> Jason

I would be very reluctant to change this code.
One device we have doesn't have an i2c address, the "address" is the
address of memory within the chip, not the address of the chip on the
bus.

Reading some of the QSFP cable's EEPROM caused another EEPROM
on the bus to be erased. We had to put in defensive code to
check for this case.

I'm not the expert who wrote this code (he retired) so I would have
to do a lot of work making sure the older parts and cables with
these problems worked after a major rewrite of the code.

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 38/51] IB/qib: Add qib_sysfs.c

[Jason Gunthorpe]

On Fri, Apr 09, 2010 at 04:31:26PM -0700, Ralph Campbell wrote:

> I would be very reluctant to change this code.
> One device we have doesn't have an i2c address, the "address" is the
> address of memory within the chip, not the address of the chip on the
> bus.

Fair enough.. But mixing not-i2c + QSFP real i2c on the same bus =
too scary for me :)

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 39/51] IB/qib: Add qib_twsi.c

[Ralph Campbell]

creates the qib_twsi.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_twsi.c |  521 ++++++++++++++++++++++++++++++++++
 1 files changed, 521 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_twsi.c

diff --git a/drivers/infiniband/hw/qib/qib_twsi.c b/drivers/infiniband/hw/qib/qib_twsi.c
new file mode 100644
index 0000000..a9a1c92
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_twsi.c
@@ -0,0 +1,521 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/delay.h>
+#include <linux/pci.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+
+/*
+ * QLogic_IB "Two Wire Serial Interface" driver.
+ * Originally written for a not-quite-i2c serial eeprom, which is
+ * still used on some supported boards. Later boards have added a
+ * variety of other uses, most board-specific, so teh bit-boffing
+ * part has been split off to this file, while the other parts
+ * have been moved to chip-specific files.
+ *
+ * We have also dropped all pretense of fully generic (e.g. pretend
+ * we don't know whether '1' is the higher voltage) interface, as
+ * the restrictions of the generic i2c interface (e.g. no access from
+ * driver itself) make it unsuitable for this use.
+ */
+
+#define READ_CMD 1
+#define WRITE_CMD 0
+
+/**
+ * i2c_wait_for_writes - wait for a write
+ * @dd: the qlogic_ib device
+ *
+ * We use this instead of udelay directly, so we can make sure
+ * that previous register writes have been flushed all the way
+ * to the chip.  Since we are delaying anyway, the cost doesn't
+ * hurt, and makes the bit twiddling more regular
+ */
+static void i2c_wait_for_writes(struct qib_devdata *dd)
+{
+	/*
+	 * implicit read of EXTStatus is as good as explicit
+	 * read of scratch, if all we want to do is flush
+	 * writes.
+	 */
+	dd->f_gpio_mod(dd, 0, 0, 0);
+	rmb(); /* inlined, so prevent compiler reordering */
+}
+
+/*
+ * QSFP modules are allowed to hold SCL low for 500uSec. Allow twice that
+ * for "almost compliant" modules
+ */
+#define SCL_WAIT_USEC 1000
+
+/* BUF_WAIT is time bus must be free between STOP or ACK and to next START.
+ * Should be 20, but some chips need more.
+ */
+#define TWSI_BUF_WAIT_USEC 60
+
+static void scl_out(struct qib_devdata *dd, u8 bit)
+{
+	u32 mask;
+
+	udelay(1);
+
+	mask = 1UL << dd->gpio_scl_num;
+
+	/* SCL is meant to be bare-drain, so never set "OUT", just DIR */
+	dd->f_gpio_mod(dd, 0, bit ? 0 : mask, mask);
+
+	/*
+	 * Allow for slow slaves by simple
+	 * delay for falling edge, sampling on rise.
+	 */
+	if (!bit)
+		udelay(2);
+	else {
+		int rise_usec;
+		for (rise_usec = SCL_WAIT_USEC; rise_usec > 0; rise_usec -= 2) {
+			if (mask & dd->f_gpio_mod(dd, 0, 0, 0))
+				break;
+			udelay(2);
+		}
+		if (rise_usec <= 0)
+			qib_dev_err(dd, "TWSI SCL stuck low > %d uSec\n",
+				    SCL_WAIT_USEC);
+	}
+	i2c_wait_for_writes(dd);
+}
+
+static void sda_out(struct qib_devdata *dd, u8 bit)
+{
+	u32 mask;
+
+	mask = 1UL << dd->gpio_sda_num;
+
+	/* SDA is meant to be bare-drain, so never set "OUT", just DIR */
+	dd->f_gpio_mod(dd, 0, bit ? 0 : mask, mask);
+
+	i2c_wait_for_writes(dd);
+	udelay(2);
+}
+
+static u8 sda_in(struct qib_devdata *dd, int wait)
+{
+	int bnum;
+	u32 read_val, mask;
+
+	bnum = dd->gpio_sda_num;
+	mask = (1UL << bnum);
+	/* SDA is meant to be bare-drain, so never set "OUT", just DIR */
+	dd->f_gpio_mod(dd, 0, 0, mask);
+	read_val = dd->f_gpio_mod(dd, 0, 0, 0);
+	if (wait)
+		i2c_wait_for_writes(dd);
+	return (read_val & mask) >> bnum;
+}
+
+/**
+ * i2c_ackrcv - see if ack following write is true
+ * @dd: the qlogic_ib device
+ */
+static int i2c_ackrcv(struct qib_devdata *dd)
+{
+	u8 ack_received;
+
+	/* AT ENTRY SCL = LOW */
+	/* change direction, ignore data */
+	ack_received = sda_in(dd, 1);
+	scl_out(dd, 1);
+	ack_received = sda_in(dd, 1) == 0;
+	scl_out(dd, 0);
+	return ack_received;
+}
+
+static void stop_cmd(struct qib_devdata *dd);
+
+/**
+ * rd_byte - read a byte, sending STOP on last, else ACK
+ * @dd: the qlogic_ib device
+ *
+ * Returns byte shifted out of device
+ */
+static int rd_byte(struct qib_devdata *dd, int last)
+{
+	int bit_cntr, data;
+
+	data = 0;
+
+	for (bit_cntr = 7; bit_cntr >= 0; --bit_cntr) {
+		data <<= 1;
+		scl_out(dd, 1);
+		data |= sda_in(dd, 0);
+		scl_out(dd, 0);
+	}
+	if (last) {
+		scl_out(dd, 1);
+		stop_cmd(dd);
+	} else {
+		sda_out(dd, 0);
+		scl_out(dd, 1);
+		scl_out(dd, 0);
+		sda_out(dd, 1);
+	}
+	return data;
+}
+
+/**
+ * wr_byte - write a byte, one bit at a time
+ * @dd: the qlogic_ib device
+ * @data: the byte to write
+ *
+ * Returns 0 if we got the following ack, otherwise 1
+ */
+static int wr_byte(struct qib_devdata *dd, u8 data)
+{
+	int bit_cntr;
+	u8 bit;
+
+	for (bit_cntr = 7; bit_cntr >= 0; bit_cntr--) {
+		bit = (data >> bit_cntr) & 1;
+		sda_out(dd, bit);
+		scl_out(dd, 1);
+		scl_out(dd, 0);
+	}
+	return (!i2c_ackrcv(dd)) ? 1 : 0;
+}
+
+/*
+ * issue TWSI start sequence:
+ * (both clock/data high, clock high, data low while clock is high)
+ */
+static void start_seq(struct qib_devdata *dd)
+{
+	sda_out(dd, 1);
+	scl_out(dd, 1);
+	sda_out(dd, 0);
+	udelay(1);
+	scl_out(dd, 0);
+}
+
+/**
+ * stop_seq - transmit the stop sequence
+ * @dd: the qlogic_ib device
+ *
+ * (both clock/data low, clock high, data high while clock is high)
+ */
+static void stop_seq(struct qib_devdata *dd)
+{
+	scl_out(dd, 0);
+	sda_out(dd, 0);
+	scl_out(dd, 1);
+	sda_out(dd, 1);
+}
+
+/**
+ * stop_cmd - transmit the stop condition
+ * @dd: the qlogic_ib device
+ *
+ * (both clock/data low, clock high, data high while clock is high)
+ */
+static void stop_cmd(struct qib_devdata *dd)
+{
+	stop_seq(dd);
+	udelay(TWSI_BUF_WAIT_USEC);
+}
+
+/**
+ * qib_twsi_reset - reset I2C communication
+ * @dd: the qlogic_ib device
+ */
+
+int qib_twsi_reset(struct qib_devdata *dd)
+{
+	int clock_cycles_left = 9;
+	int was_high = 0;
+	u32 pins, mask;
+
+	/* Both SCL and SDA should be high. If not, there
+	 * is something wrong.
+	 */
+	mask = (1UL << dd->gpio_scl_num) | (1UL << dd->gpio_sda_num);
+
+	/*
+	 * Force pins to desired innocuous state.
+	 * This is the default power-on state with out=0 and dir=0,
+	 * So tri-stated and should be floating high (barring HW problems)
+	 */
+	dd->f_gpio_mod(dd, 0, 0, mask);
+
+	/*
+	 * Clock nine times to get all listeners into a sane state.
+	 * If SDA does not go high at any point, we are wedged.
+	 * One vendor recommends then issuing START followed by STOP.
+	 * we cannot use our "normal" functions to do that, because
+	 * if SCL drops between them, another vendor's part will
+	 * wedge, dropping SDA and keeping it low forever, at the end of
+	 * the next transaction (even if it was not the device addressed).
+	 * So our START and STOP take place with SCL held high.
+	 */
+	while (clock_cycles_left--) {
+		scl_out(dd, 0);
+		scl_out(dd, 1);
+		/* Note if SDA is high, but keep clocking to sync slave */
+		was_high |= sda_in(dd, 0);
+	}
+
+	if (was_high) {
+		/*
+		 * We saw a high, which we hope means the slave is sync'd.
+		 * Issue START, STOP, pause for T_BUF.
+		 */
+
+		pins = dd->f_gpio_mod(dd, 0, 0, 0);
+		if ((pins & mask) != mask)
+			qib_dev_err(dd, "TWSI pins not at rest: %d\n",
+				    pins & mask);
+		/* Drop SDA to issue START */
+		udelay(1); /* Guarantee .6 uSec setup */
+		sda_out(dd, 0);
+		udelay(1); /* Guarantee .6 uSec hold */
+		/* At this point, SCL is high, SDA low. Raise SDA for STOP */
+		sda_out(dd, 1);
+		udelay(TWSI_BUF_WAIT_USEC);
+	}
+
+	return !was_high;
+}
+
+#define QIB_TWSI_START 0x100
+#define QIB_TWSI_STOP 0x200
+
+/* Write byte to TWSI, optionally prefixed with START or suffixed with
+ * STOP.
+ * returns 0 if OK (ACK received), else != 0
+ */
+static int qib_twsi_wr(struct qib_devdata *dd, int data, int flags)
+{
+	int ret = 1;
+	if (flags & QIB_TWSI_START)
+		start_seq(dd);
+
+	ret = wr_byte(dd, data); /* Leaves SCL low (from i2c_ackrcv()) */
+
+	if (flags & QIB_TWSI_STOP)
+		stop_cmd(dd);
+	if (ret)
+		qib_cdbg(VERBOSE, "qib%d Failed TWSI wr %s 0x%02X %s\n",
+			dd->unit, (flags & QIB_TWSI_START) ? "START" : "", data,
+			(flags & QIB_TWSI_STOP) ? "STOP" : "");
+	return ret;
+}
+
+/* Added functionality for IBA7220-based cards */
+#define QIB_TEMP_DEV 0x98
+
+/*
+ * qib_twsi_blk_rd
+ * Formerly called qib_eeprom_internal_read, and only used for eeprom,
+ * but now the general interface for data transfer from twsi devices.
+ * One vestige of its former role is that it recognizes a device
+ * QIB_TWSI_NO_DEV and does the correct operation for the legacy part,
+ * which responded to all TWSI device codes, interpreting them as
+ * address within device. On all other devices found on board handled by
+ * this driver, the device is followed by a one-byte "address" which selects
+ * the "register" or "offset" within the device from which data should
+ * be read.
+ */
+int qib_twsi_blk_rd(struct qib_devdata *dd, int dev, int addr,
+		    void *buffer, int len)
+{
+	int ret;
+	u8 *bp = buffer;
+
+	ret = 1;
+
+	if (dev == QIB_TWSI_NO_DEV) {
+		/* legacy not-really-I2C */
+		qib_cdbg(VERBOSE, "qib%d Start command only address\n",
+			 dd->unit);
+		addr = (addr << 1) | READ_CMD;
+		ret = qib_twsi_wr(dd, addr, QIB_TWSI_START);
+	} else {
+		/* Actual I2C */
+		qib_cdbg(VERBOSE, "qib%d Start command uses devaddr %02X\n",
+			 dd->unit, dev);
+		ret = qib_twsi_wr(dd, dev | WRITE_CMD, QIB_TWSI_START);
+		if (ret) {
+			qib_dbg("qib%d Failed TWSI startcmd\n", dd->unit);
+			stop_cmd(dd);
+			ret = 1;
+			goto bail;
+		}
+		/*
+		 * SFF spec claims we do _not_ stop after the addr
+		 * but simply issue a start with the "read" dev-addr.
+		 * Since we are implicitely waiting for ACK here,
+		 * we need t_buf (nominally 20uSec) before that start,
+		 * and cannot rely on the delay built in to the STOP
+		 */
+		ret = qib_twsi_wr(dd, addr, 0);
+		udelay(TWSI_BUF_WAIT_USEC);
+
+		if (ret) {
+			qib_dev_err(dd, "Failed to write TWSI addr %02X\n",
+				    addr);
+			ret = 1;
+			goto bail;
+		}
+		ret = qib_twsi_wr(dd, dev | READ_CMD, QIB_TWSI_START);
+	}
+	if (ret) {
+		qib_dbg("qib%d Failed startcmd for dev %02X\n", dd->unit, dev);
+		stop_cmd(dd);
+		ret = 1;
+		goto bail;
+	}
+
+	/*
+	 * block devices keeps clocking data out as long as we ack,
+	 * automatically incrementing the address. Some have "pages"
+	 * whose boundaries will not be crossed, but the handling
+	 * of these is left to the caller, who is in a better
+	 * position to know.
+	 */
+	while (len-- > 0) {
+		/*
+		 * Get and store data, sending ACK if length remaining,
+		 * else STOP
+		 */
+		*bp++ = rd_byte(dd, !len);
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/*
+ * qib_twsi_blk_wr
+ * Formerly called qib_eeprom_internal_write, and only used for eeprom,
+ * but now the general interface for data transfer to twsi devices.
+ * One vestige of its former role is that it recognizes a device
+ * QIB_TWSI_NO_DEV and does the correct operation for the legacy part,
+ * which responded to all TWSI device codes, interpreting them as
+ * address within device. On all other devices found on board handled by
+ * this driver, the device is followed by a one-byte "address" which selects
+ * the "register" or "offset" within the device to which data should
+ * be written.
+ */
+int qib_twsi_blk_wr(struct qib_devdata *dd, int dev, int addr,
+		    const void *buffer, int len)
+{
+	int sub_len;
+	const u8 *bp = buffer;
+	int max_wait_time, i;
+	int ret;
+	ret = 1;
+
+	while (len > 0) {
+		if (dev == QIB_TWSI_NO_DEV) {
+			if (qib_twsi_wr(dd, (addr << 1) | WRITE_CMD,
+					QIB_TWSI_START)) {
+				qib_dbg("qib%d Failed to start cmd offset %u\n",
+					dd->unit, addr);
+				goto failed_write;
+			}
+		} else {
+			/* Real I2C */
+			qib_cdbg(VERBOSE, "qib%d write start command devaddr"
+				" %02X\n", dd->unit, dev);
+			if (qib_twsi_wr(dd, dev | WRITE_CMD, QIB_TWSI_START)) {
+				qib_dbg("qib%d Failed TWSI startcmd\n",
+					dd->unit);
+				goto failed_write;
+			}
+			ret = qib_twsi_wr(dd, addr, 0);
+			if (ret) {
+				qib_dev_err(dd, "Failed to write TWSI addr\n");
+				goto failed_write;
+			}
+		}
+
+		sub_len = min(len, 4);
+		addr += sub_len;
+		len -= sub_len;
+
+		for (i = 0; i < sub_len; i++) {
+			if (qib_twsi_wr(dd, *bp++, 0)) {
+				qib_dbg("qib%d no ack after byte %u/%u (%u "
+					"total remain)\n", dd->unit, i, sub_len,
+					len + sub_len - i);
+				goto failed_write;
+			}
+		}
+
+		stop_cmd(dd);
+
+		/*
+		 * Wait for write complete by waiting for a successful
+		 * read (the chip replies with a zero after the write
+		 * cmd completes, and before it writes to the eeprom.
+		 * The startcmd for the read will fail the ack until
+		 * the writes have completed.   We do this inline to avoid
+		 * the debug prints that are in the real read routine
+		 * if the startcmd fails.
+		 * We also use the proper device address, so it doesn't matter
+		 * whether we have real eeprom_dev. Legacy likes any address.
+		 */
+		max_wait_time = 100;
+		while (qib_twsi_wr(dd, dev | READ_CMD, QIB_TWSI_START)) {
+			stop_cmd(dd);
+			if (!--max_wait_time) {
+				qib_dbg("qib%d Did not get successful read to "
+					"complete write\n", dd->unit);
+				goto failed_write;
+			}
+		}
+		/* now read (and ignore) the resulting byte */
+		rd_byte(dd, 1);
+	}
+
+	ret = 0;
+	goto bail;
+
+failed_write:
+	stop_cmd(dd);
+	ret = 1;
+
+bail:
+	return ret;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 40/51] IB/qib: Add qib_tx.c

[Ralph Campbell]

creates the qib_tx.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_tx.c |  585 ++++++++++++++++++++++++++++++++++++
 1 files changed, 585 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_tx.c

diff --git a/drivers/infiniband/hw/qib/qib_tx.c b/drivers/infiniband/hw/qib/qib_tx.c
new file mode 100644
index 0000000..7d77f29
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_tx.c
@@ -0,0 +1,585 @@
+/*
+ * Copyright (c) 2008, 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/spinlock.h>
+#include <linux/pci.h>
+#include <linux/io.h>
+#include <linux/delay.h>
+#include <linux/netdevice.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+
+static unsigned qib_hol_timeout_ms = 3000;
+module_param_named(hol_timeout_ms, qib_hol_timeout_ms, uint, S_IRUGO);
+MODULE_PARM_DESC(hol_timeout_ms,
+		 "duration of user app suspension after link failure");
+
+unsigned qib_sdma_fetch_arb = 1;
+module_param_named(fetch_arb, qib_sdma_fetch_arb, uint, S_IRUGO);
+MODULE_PARM_DESC(fetch_arb, "IBA7220: change SDMA descriptor arbitration");
+
+/**
+ * qib_disarm_piobufs - cancel a range of PIO buffers
+ * @dd: the qlogic_ib device
+ * @first: the first PIO buffer to cancel
+ * @cnt: the number of PIO buffers to cancel
+ *
+ * Cancel a range of PIO buffers. Used at user process close,
+ * in case it died while writing to a PIO buffer.
+ */
+void qib_disarm_piobufs(struct qib_devdata *dd, unsigned first, unsigned cnt)
+{
+	unsigned long flags;
+	unsigned i;
+	unsigned last;
+
+	qib_cdbg(ERRPKT, "disarm %u PIObufs first=%u\n", cnt, first);
+	last = first + cnt;
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	for (i = first; i < last; i++) {
+		__clear_bit(i, dd->pio_need_disarm);
+		dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(i));
+	}
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+}
+
+/*
+ * This is called by a user process when it sees the DISARM_BUFS event
+ * bit is set.
+ */
+int qib_disarm_piobufs_ifneeded(struct qib_ctxtdata *rcd)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned i;
+	unsigned last;
+	unsigned n = 0;
+
+	last = rcd->pio_base + rcd->piocnt;
+	/*
+	 * Don't need uctxt_lock here, since user has called in to us.
+	 * Clear at start in case more interrupts set bits while we
+	 * are disarming
+	 */
+	if (rcd->user_event_mask) {
+		/*
+		 * subctxt_cnt is 0 if not shared, so do base
+		 * separately, first, then remaining subctxt, if any
+		 */
+		clear_bit(_QIB_EVENT_DISARM_BUFS_BIT, &rcd->user_event_mask[0]);
+		for (i = 1; i < rcd->subctxt_cnt; i++)
+			clear_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+				  &rcd->user_event_mask[i]);
+	}
+	spin_lock_irq(&dd->pioavail_lock);
+	for (i = rcd->pio_base; i < last; i++) {
+		if (__test_and_clear_bit(i, dd->pio_need_disarm)) {
+			n++;
+			dd->f_sendctrl(rcd->ppd, QIB_SENDCTRL_DISARM_BUF(i));
+		}
+	}
+	spin_unlock_irq(&dd->pioavail_lock);
+	qib_cdbg(ERRPKT, "Ctxt%u, User disarm (%u bufs (%u-%u)) done\n",
+		rcd->ctxt, n, rcd->pio_base, last);
+	return 0;
+}
+
+static struct qib_pportdata *is_sdma_buf(struct qib_devdata *dd, unsigned i)
+{
+	struct qib_pportdata *ppd;
+	unsigned pidx;
+
+	for (pidx = 0; pidx < dd->num_pports; pidx++) {
+		ppd = dd->pport + pidx;
+		if (i >= ppd->sdma_state.first_sendbuf &&
+		    i < ppd->sdma_state.last_sendbuf)
+			return ppd;
+	}
+	return NULL;
+}
+
+/*
+ * Return true if send buffer is being used by a user context.
+ * Sets  _QIB_EVENT_DISARM_BUFS_BIT in user_event_mask as a side effect
+ */
+static int find_ctxt(struct qib_devdata *dd, unsigned bufn)
+{
+	struct qib_ctxtdata *rcd;
+	unsigned ctxt;
+	int ret = 0;
+
+	spin_lock(&dd->uctxt_lock);
+	for (ctxt = dd->first_user_ctxt; ctxt < dd->cfgctxts; ctxt++) {
+		rcd = dd->rcd[ctxt];
+		if (!rcd || bufn < rcd->pio_base ||
+		    bufn >= rcd->pio_base + rcd->piocnt)
+			continue;
+		if (rcd->user_event_mask) {
+			int i;
+			/*
+			 * subctxt_cnt is 0 if not shared, so do base
+			 * separately, first, then remaining subctxt, if any
+			 */
+			set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+				&rcd->user_event_mask[0]);
+			for (i = 1; i < rcd->subctxt_cnt; i++)
+				set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+					&rcd->user_event_mask[i]);
+		}
+		ret = 1;
+		break;
+	}
+	spin_unlock(&dd->uctxt_lock);
+
+	return ret;
+}
+
+/*
+ * Disarm a set of send buffers.  If the buffer might be actively being
+ * written to, mark the buffer to be disarmed later when it is not being
+ * written to.
+ *
+ * This should only be called from the IRQ error handler.
+ */
+void qib_disarm_piobufs_set(struct qib_devdata *dd, unsigned long *mask,
+			    unsigned cnt)
+{
+	struct qib_pportdata *ppd;
+	unsigned i;
+	unsigned long flags;
+
+	for (i = 0; i < cnt; i++) {
+		int which;
+		if (!test_bit(i, mask))
+			continue;
+		/*
+		 * If the buffer is owned by the DMA hardware,
+		 * reset the DMA engine.
+		 */
+		ppd = is_sdma_buf(dd, i);
+		if (ppd) {
+			/* should do only once per port */
+			qib_cdbg(ERRPKT, "buf %u sdma; do cancel_sends\n", i);
+			qib_cancel_sends(ppd);
+			continue;
+		}
+		/*
+		 * If the kernel is writing the buffer or the buffer is
+		 * owned by a user process, we can't clear it yet.
+		 */
+		spin_lock_irqsave(&dd->pioavail_lock, flags);
+		if (test_bit(i, dd->pio_writing) ||
+		    (!test_bit(i << 1, dd->pioavailkernel) &&
+		     find_ctxt(dd, i))) {
+			__set_bit(i, dd->pio_need_disarm);
+			which = 0;
+		} else {
+			which = 1;
+			dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(i));
+		}
+		spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+		qib_cdbg(ERRPKT, "disarm buf %u %s\n", i, which ?
+			 "now" : "later");
+	}
+}
+
+/**
+ * update_send_bufs - update shadow copy of the PIO availability map
+ * @dd: the qlogic_ib device
+ *
+ * called whenever our local copy indicates we have run out of send buffers
+ */
+static void update_send_bufs(struct qib_devdata *dd)
+{
+	unsigned long flags;
+	unsigned i;
+	const unsigned piobregs = dd->pioavregs;
+
+	/*
+	 * If the generation (check) bits have changed, then we update the
+	 * busy bit for the corresponding PIO buffer.  This algorithm will
+	 * modify positions to the value they already have in some cases
+	 * (i.e., no change), but it's faster than changing only the bits
+	 * that have changed.
+	 *
+	 * We would like to do this atomicly, to avoid spinlocks in the
+	 * critical send path, but that's not really possible, given the
+	 * type of changes, and that this routine could be called on
+	 * multiple cpu's simultaneously, so we lock in this routine only,
+	 * to avoid conflicting updates; all we change is the shadow, and
+	 * it's a single 64 bit memory location, so by definition the update
+	 * is atomic in terms of what other cpu's can see in testing the
+	 * bits.  The spin_lock overhead isn't too bad, since it only
+	 * happens when all buffers are in use, so only cpu overhead, not
+	 * latency or bandwidth is affected.
+	 */
+	if (!dd->pioavailregs_dma) {
+		qib_dbg("Update shadow pioavail, but regs_dma NULL!\n");
+		return;
+	}
+	if (qib_debug & __QIB_VERBDBG) {
+		/* only if packet debug and verbose */
+		volatile __le64 *dma = dd->pioavailregs_dma; /* DMA'ed */
+		unsigned long *shadow = dd->pioavailshadow;
+
+		qib_dbg("Refill avail, ts%lx; dmacopy: "
+			"%llx %llx %llx %llx %llx %llx"
+			"; shadow: %lx %lx %lx %lx %lx %lx\n",
+			(unsigned long)get_cycles(),
+			(unsigned long long) le64_to_cpu(dma[0]),
+			(unsigned long long) le64_to_cpu(dma[1]),
+			(unsigned long long) le64_to_cpu(dma[2]),
+			(unsigned long long) le64_to_cpu(dma[3]),
+			(unsigned long long) le64_to_cpu(dma[4]),
+			(unsigned long long) le64_to_cpu(dma[5]),
+			shadow[0], shadow[1], shadow[2], shadow[3],
+			shadow[4], shadow[5]);
+	}
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	for (i = 0; i < piobregs; i++) {
+		u64 pchbusy, pchg, piov, pnew;
+
+		piov = le64_to_cpu(dd->pioavailregs_dma[i]);
+		pchg = dd->pioavailkernel[i] &
+			~(dd->pioavailshadow[i] ^ piov);
+		pchbusy = pchg << QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT;
+		if (pchg && (pchbusy & dd->pioavailshadow[i])) {
+			pnew = dd->pioavailshadow[i] & ~pchbusy;
+			pnew |= piov & pchbusy;
+			dd->pioavailshadow[i] = pnew;
+		}
+	}
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+}
+
+/*
+ * Debugging code and stats updates if no pio buffers available.
+ */
+static noinline void no_send_bufs(struct qib_devdata *dd)
+{
+	dd->upd_pio_shadow = 1;
+
+	/* not atomic, but if we lose a stat count in a while, that's OK */
+	qib_stats.sps_nopiobufs++;
+}
+
+/*
+ * Common code for normal driver send buffer allocation, and reserved
+ * allocation.
+ *
+ * Do appropriate marking as busy, etc.
+ * Returns buffer pointer if one is found, otherwise NULL.
+ */
+u32 __iomem *qib_getsendbuf_range(struct qib_devdata *dd, u32 *pbufnum,
+				  u32 first, u32 last)
+{
+	unsigned i, j, updated = 0;
+	unsigned nbufs;
+	unsigned long flags;
+	unsigned long *shadow = dd->pioavailshadow;
+	u32 __iomem *buf;
+
+	if (!(dd->flags & QIB_PRESENT))
+		return NULL;
+
+	nbufs = last - first + 1; /* number in range to check */
+	if (dd->upd_pio_shadow) {
+		/*
+		 * Minor optimization.  If we had no buffers on last call,
+		 * start out by doing the update; continue and do scan even
+		 * if no buffers were updated, to be paranoid.
+		 */
+		update_send_bufs(dd);
+		updated++;
+	}
+	i = first;
+rescan:
+	/*
+	 * While test_and_set_bit() is atomic, we do that and then the
+	 * change_bit(), and the pair is not.  See if this is the cause
+	 * of the remaining armlaunch errors.
+	 */
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	for (j = 0; j < nbufs; j++, i++) {
+		if (i > last)
+			i = first;
+		if (__test_and_set_bit((2 * i) + 1, shadow))
+			continue;
+		/* flip generation bit */
+		__change_bit(2 * i, shadow);
+		/* remember that the buffer can be written to now */
+		__set_bit(i, dd->pio_writing);
+		break;
+	}
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+
+	if (j == nbufs) {
+		if (!updated) {
+			/*
+			 * First time through; shadow exhausted, but may be
+			 * buffers available, try an update and then rescan.
+			 */
+			update_send_bufs(dd);
+			updated++;
+			i = first;
+			goto rescan;
+		}
+		no_send_bufs(dd);
+		buf = NULL;
+	} else {
+		if (i < dd->piobcnt2k)
+			buf = (u32 __iomem *)(dd->pio2kbase +
+				i * dd->palign);
+		else
+			buf = (u32 __iomem *)(dd->pio4kbase +
+				(i - dd->piobcnt2k) * dd->align4k);
+		if (pbufnum)
+			*pbufnum = i;
+		dd->upd_pio_shadow = 0;
+	}
+
+	return buf;
+}
+
+/*
+ * Record that the caller is finished writing to the buffer so we don't
+ * disarm it while it is being written and disarm it now if needed.
+ */
+void qib_sendbuf_done(struct qib_devdata *dd, unsigned n)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	__clear_bit(n, dd->pio_writing);
+	if (__test_and_clear_bit(n, dd->pio_need_disarm))
+		dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(n));
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+}
+
+/**
+ * qib_chg_pioavailkernel - change which send buffers are available for kernel
+ * @dd: the qlogic_ib device
+ * @start: the starting send buffer number
+ * @len: the number of send buffers
+ * @avail: true if the buffers are available for kernel use, false otherwise
+ */
+void qib_chg_pioavailkernel(struct qib_devdata *dd, unsigned start,
+	unsigned len, u32 avail, struct qib_ctxtdata *rcd)
+{
+	unsigned long flags;
+	unsigned end;
+	unsigned ostart = start;
+
+	if (rcd)
+		qib_cdbg(SDMA, "start %u len %u avail %d ctxt%u\n", start,
+			 len, avail, rcd->ctxt);
+	else
+		qib_cdbg(SDMA, "start %u len %u avail %d (nowait)\n",
+			 start, len, avail);
+
+	/* There are two bits per send buffer (busy and generation) */
+	start *= 2;
+	end = start + len * 2;
+
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	/* Set or clear the busy bit in the shadow. */
+	while (start < end) {
+		if (avail) {
+			unsigned long dma;
+			int i;
+
+			/*
+			 * The BUSY bit will never be set, because we disarm
+			 * the user buffers before we hand them back to the
+			 * kernel.  We do have to make sure the generation
+			 * bit is set correctly in shadow, since it could
+			 * have changed many times while allocated to user.
+			 * We can't use the bitmap functions on the full
+			 * dma array because it is always little-endian, so
+			 * we have to flip to host-order first.
+			 * BITS_PER_LONG is slightly wrong, since it's
+			 * always 64 bits per register in chip...
+			 * We only work on 64 bit kernels, so that's OK.
+			 */
+			i = start / BITS_PER_LONG;
+			__clear_bit(QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT + start,
+				    dd->pioavailshadow);
+			dma = (unsigned long)
+				le64_to_cpu(dd->pioavailregs_dma[i]);
+			if (test_bit((QLOGIC_IB_SENDPIOAVAIL_CHECK_SHIFT +
+				      start) % BITS_PER_LONG, &dma))
+				__set_bit(QLOGIC_IB_SENDPIOAVAIL_CHECK_SHIFT +
+					  start, dd->pioavailshadow);
+			else
+				__clear_bit(QLOGIC_IB_SENDPIOAVAIL_CHECK_SHIFT
+					    + start, dd->pioavailshadow);
+			__set_bit(start, dd->pioavailkernel);
+		} else {
+			__set_bit(start + QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT,
+				  dd->pioavailshadow);
+			__clear_bit(start, dd->pioavailkernel);
+		}
+		start += 2;
+	}
+
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+
+	dd->f_txchk_change(dd, ostart, len, avail, rcd);
+}
+
+/*
+ * Flush all sends that might be in the ready to send state, as well as any
+ * that are in the process of being sent.  Used whenever we need to be
+ * sure the send side is idle.  Cleans up all buffer state by canceling
+ * all pio buffers, and issuing an abort, which cleans up anything in the
+ * launch fifo.  The cancel is superfluous on some chip versions, but
+ * it's safer to always do it.
+ * PIOAvail bits are updated by the chip as if a normal send had happened.
+ */
+void qib_cancel_sends(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	struct qib_ctxtdata *rcd;
+	unsigned long flags;
+	unsigned ctxt;
+	unsigned i;
+	unsigned last;
+
+	/*
+	 * Tell PSM to disarm buffers again before trying to reuse them.
+	 * We need to be sure the rcd doesn't change out from under us
+	 * while we do so.  We hold the two locks sequentially.  We might
+	 * needlessly set some need_disarm bits as a result, if the
+	 * context is closed after we release the uctxt_lock, but that's
+	 * fairly benign, and safer than nesting the locks.
+	 */
+	for (ctxt = dd->first_user_ctxt; ctxt < dd->cfgctxts; ctxt++) {
+		spin_lock_irqsave(&dd->uctxt_lock, flags);
+		rcd = dd->rcd[ctxt];
+		if (rcd && rcd->ppd == ppd) {
+			last = rcd->pio_base + rcd->piocnt;
+			if (rcd->user_event_mask) {
+				/*
+				 * subctxt_cnt is 0 if not shared, so do base
+				 * separately, first, then remaining subctxt,
+				 * if any
+				 */
+				set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+					&rcd->user_event_mask[0]);
+				for (i = 1; i < rcd->subctxt_cnt; i++)
+					set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+						&rcd->user_event_mask[i]);
+			}
+			i = rcd->pio_base;
+			spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+			spin_lock_irqsave(&dd->pioavail_lock, flags);
+			for (; i < last; i++)
+				__set_bit(i, dd->pio_need_disarm);
+			spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+		} else
+			spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+	}
+
+	if (dd->flags & QIB_HAS_SEND_DMA)
+		qib_sdma_process_event(ppd, qib_sdma_event_e70_go_idle);
+	else
+		dd->f_sendctrl(ppd, QIB_SENDCTRL_DISARM_ALL |
+				    QIB_SENDCTRL_FLUSH);
+}
+
+/*
+ * Force an update of in-memory copy of the pioavail registers, when
+ * needed for any of a variety of reasons.
+ * If already off, this routine is a nop, on the assumption that the
+ * caller (or set of callers) will "do the right thing".
+ * This is a per-device operation, so just the first port.
+ */
+void qib_force_pio_avail_update(struct qib_devdata *dd)
+{
+	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_AVAIL_BLIP);
+}
+
+void qib_hol_down(struct qib_pportdata *ppd)
+{
+	/*
+	 * Cancel sends when the link goes DOWN so that we aren't doing it
+	 * at INIT when we might be trying to send SMI packets.
+	 */
+	if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG))
+		qib_cancel_sends(ppd);
+}
+
+/*
+ * Link is at INIT.
+ * We start the HoL timer so we can detect stuck packets blocking SMP replies.
+ * Timer may already be running, so use mod_timer, not add_timer.
+ */
+void qib_hol_init(struct qib_pportdata *ppd)
+{
+	if (ppd->hol_state != QIB_HOL_INIT) {
+		ppd->hol_state = QIB_HOL_INIT;
+		mod_timer(&ppd->hol_timer,
+			  jiffies + msecs_to_jiffies(qib_hol_timeout_ms));
+	}
+}
+
+/*
+ * Link is up, continue any user processes, and ensure timer
+ * is a nop, if running.  Let timer keep running, if set; it
+ * will nop when it sees the link is up.
+ */
+void qib_hol_up(struct qib_pportdata *ppd)
+{
+	ppd->hol_state = QIB_HOL_UP;
+}
+
+/*
+ * This is only called via the timer.
+ */
+void qib_hol_event(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *)opaque;
+
+	/* If hardware error, etc, skip. */
+	if (!(ppd->dd->flags & QIB_INITTED))
+		return;
+
+	if (ppd->hol_state != QIB_HOL_UP) {
+		/*
+		 * Try to flush sends in case a stuck packet is blocking
+		 * SMP replies.
+		 */
+		qib_hol_down(ppd);
+		mod_timer(&ppd->hol_timer,
+			  jiffies + msecs_to_jiffies(qib_hol_timeout_ms));
+	}
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 41/51] IB/qib: Add qib_uc.c

[Ralph Campbell]

creates the qib_uc.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_uc.c |  586 ++++++++++++++++++++++++++++++++++++
 1 files changed, 586 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_uc.c

diff --git a/drivers/infiniband/hw/qib/qib_uc.c b/drivers/infiniband/hw/qib/qib_uc.c
new file mode 100644
index 0000000..f077489
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_uc.c
@@ -0,0 +1,586 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "qib.h"
+
+/* cut down ridiculously long IB macro names */
+#define OP(x) IB_OPCODE_UC_##x
+
+/**
+ * qib_make_uc_req - construct a request packet (SEND, RDMA write)
+ * @qp: a pointer to the QP
+ *
+ * Return 1 if constructed; otherwise, return 0.
+ */
+int qib_make_uc_req(struct qib_qp *qp)
+{
+	struct qib_other_headers *ohdr;
+	struct qib_swqe *wqe;
+	unsigned long flags;
+	u32 hwords;
+	u32 bth0;
+	u32 len;
+	u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
+	int ret = 0;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_SEND_OK)) {
+		if (!(ib_qib_state_ops[qp->state] & QIB_FLUSH_SEND))
+			goto bail;
+		/* We are in the error state, flush the work request. */
+		if (qp->s_last == qp->s_head)
+			goto bail;
+		/* If DMAs are in progress, we can't flush immediately. */
+		if (atomic_read(&qp->s_dma_busy)) {
+			qp->s_flags |= QIB_S_WAIT_DMA;
+			goto bail;
+		}
+		wqe = get_swqe_ptr(qp, qp->s_last);
+		qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
+		goto done;
+	}
+
+	ohdr = &qp->s_hdr.u.oth;
+	if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
+		ohdr = &qp->s_hdr.u.l.oth;
+
+	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
+	hwords = 5;
+	bth0 = 1 << 22; /* Set M bit */
+
+	/* Get the next send request. */
+	wqe = get_swqe_ptr(qp, qp->s_cur);
+	qp->s_wqe = NULL;
+	switch (qp->s_state) {
+	default:
+		if (!(ib_qib_state_ops[qp->state] &
+		    QIB_PROCESS_NEXT_SEND_OK))
+			goto bail;
+		/* Check if send work queue is empty. */
+		if (qp->s_cur == qp->s_head)
+			goto bail;
+		/*
+		 * Start a new request.
+		 */
+		wqe->psn = qp->s_next_psn;
+		qp->s_psn = qp->s_next_psn;
+		qp->s_sge.sge = wqe->sg_list[0];
+		qp->s_sge.sg_list = wqe->sg_list + 1;
+		qp->s_sge.num_sge = wqe->wr.num_sge;
+		qp->s_sge.total_len = wqe->length;
+		len = wqe->length;
+		qp->s_len = len;
+		switch (wqe->wr.opcode) {
+		case IB_WR_SEND:
+		case IB_WR_SEND_WITH_IMM:
+			if (len > pmtu) {
+				qp->s_state = OP(SEND_FIRST);
+				len = pmtu;
+				break;
+			}
+			if (wqe->wr.opcode == IB_WR_SEND)
+				qp->s_state = OP(SEND_ONLY);
+			else {
+				qp->s_state =
+					OP(SEND_ONLY_WITH_IMMEDIATE);
+				/* Immediate data comes after the BTH */
+				ohdr->u.imm_data = wqe->wr.ex.imm_data;
+				hwords += 1;
+			}
+			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+				bth0 |= IB_BTH_SOLICITED;
+			qp->s_wqe = wqe;
+			if (++qp->s_cur >= qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		case IB_WR_RDMA_WRITE:
+		case IB_WR_RDMA_WRITE_WITH_IMM:
+			ohdr->u.rc.reth.vaddr =
+				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+			ohdr->u.rc.reth.rkey =
+				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+			ohdr->u.rc.reth.length = cpu_to_be32(len);
+			hwords += sizeof(struct ib_reth) / 4;
+			if (len > pmtu) {
+				qp->s_state = OP(RDMA_WRITE_FIRST);
+				len = pmtu;
+				break;
+			}
+			if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
+				qp->s_state = OP(RDMA_WRITE_ONLY);
+			else {
+				qp->s_state =
+					OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE);
+				/* Immediate data comes after the RETH */
+				ohdr->u.rc.imm_data = wqe->wr.ex.imm_data;
+				hwords += 1;
+				if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+					bth0 |= IB_BTH_SOLICITED;
+			}
+			qp->s_wqe = wqe;
+			if (++qp->s_cur >= qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		default:
+			goto bail;
+		}
+		break;
+
+	case OP(SEND_FIRST):
+		qp->s_state = OP(SEND_MIDDLE);
+		/* FALLTHROUGH */
+	case OP(SEND_MIDDLE):
+		len = qp->s_len;
+		if (len > pmtu) {
+			len = pmtu;
+			break;
+		}
+		if (wqe->wr.opcode == IB_WR_SEND)
+			qp->s_state = OP(SEND_LAST);
+		else {
+			qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE);
+			/* Immediate data comes after the BTH */
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
+			hwords += 1;
+		}
+		if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+			bth0 |= IB_BTH_SOLICITED;
+		qp->s_wqe = wqe;
+		if (++qp->s_cur >= qp->s_size)
+			qp->s_cur = 0;
+		break;
+
+	case OP(RDMA_WRITE_FIRST):
+		qp->s_state = OP(RDMA_WRITE_MIDDLE);
+		/* FALLTHROUGH */
+	case OP(RDMA_WRITE_MIDDLE):
+		len = qp->s_len;
+		if (len > pmtu) {
+			len = pmtu;
+			break;
+		}
+		if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
+			qp->s_state = OP(RDMA_WRITE_LAST);
+		else {
+			qp->s_state =
+				OP(RDMA_WRITE_LAST_WITH_IMMEDIATE);
+			/* Immediate data comes after the BTH */
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
+			hwords += 1;
+			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+				bth0 |= IB_BTH_SOLICITED;
+		}
+		qp->s_wqe = wqe;
+		if (++qp->s_cur >= qp->s_size)
+			qp->s_cur = 0;
+		break;
+	}
+	qp->s_len -= len;
+	qp->s_hdrwords = hwords;
+	qp->s_cur_sge = &qp->s_sge;
+	qp->s_cur_size = len;
+	qib_make_ruc_header(qp, ohdr, bth0 | (qp->s_state << 24),
+			    qp->s_next_psn++ & QIB_PSN_MASK);
+done:
+	ret = 1;
+	goto unlock;
+
+bail:
+	qp->s_flags &= ~QIB_S_BUSY;
+unlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+	return ret;
+}
+
+static void fix_mr_refcount(struct qib_qp *qp)
+{
+	unsigned i;
+
+	if (qp->r_sge.num_sge == qp->s_rdma_read_sge.num_sge)
+		return;
+	while (qp->r_sge.num_sge) {
+		atomic_dec(&qp->r_sge.sge.mr->refcount);
+		if (--qp->r_sge.num_sge)
+			qp->r_sge.sge = *qp->r_sge.sg_list++;
+	}
+	for (i = 0; i < qp->s_rdma_read_sge.num_sge; i++) {
+		struct qib_sge *sge = i ?
+			&qp->s_rdma_read_sge.sg_list[i - 1] :
+			&qp->s_rdma_read_sge.sge;
+
+		atomic_inc(&sge->mr->refcount);
+	}
+}
+
+/**
+ * qib_uc_rcv - handle an incoming UC packet
+ * @ibp: the port the packet came in on
+ * @hdr: the header of the packet
+ * @has_grh: true if the packet has a GRH
+ * @data: the packet data
+ * @tlen: the length of the packet
+ * @qp: the QP for this packet.
+ *
+ * This is called from qib_qp_rcv() to process an incoming UC packet
+ * for the given QP.
+ * Called at interrupt level.
+ */
+void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp)
+{
+	struct qib_other_headers *ohdr;
+	unsigned long flags;
+	u32 opcode;
+	u32 hdrsize;
+	u32 psn;
+	u32 pad;
+	struct ib_wc wc;
+	u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
+	struct ib_reth *reth;
+	int ret;
+
+	/* Check for GRH */
+	if (!has_grh) {
+		ohdr = &hdr->u.oth;
+		hdrsize = 8 + 12;       /* LRH + BTH */
+	} else {
+		ohdr = &hdr->u.l.oth;
+		hdrsize = 8 + 40 + 12;  /* LRH + GRH + BTH */
+	}
+
+	opcode = be32_to_cpu(ohdr->bth[0]);
+	spin_lock_irqsave(&qp->s_lock, flags);
+	if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode))
+		goto sunlock;
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+
+	psn = be32_to_cpu(ohdr->bth[2]);
+	opcode >>= 24;
+	memset(&wc, 0, sizeof wc);
+
+	/* Prevent simultaneous processing after APM on different CPUs */
+	spin_lock(&qp->r_lock);
+
+	/* Compare the PSN verses the expected PSN. */
+	if (unlikely(qib_cmp24(psn, qp->r_psn) != 0)) {
+		/*
+		 * Handle a sequence error.
+		 * Silently drop any current message.
+		 */
+		qp->r_psn = psn;
+inv:
+		while (qp->r_sge.num_sge) {
+			atomic_dec(&qp->r_sge.sge.mr->refcount);
+			if (--qp->r_sge.num_sge)
+				qp->r_sge.sge = *qp->r_sge.sg_list++;
+		}
+		qp->r_state = OP(SEND_LAST);
+		switch (opcode) {
+		case OP(SEND_FIRST):
+		case OP(SEND_ONLY):
+		case OP(SEND_ONLY_WITH_IMMEDIATE):
+			goto send_first;
+
+		case OP(RDMA_WRITE_FIRST):
+		case OP(RDMA_WRITE_ONLY):
+		case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE):
+			goto rdma_first;
+
+		default:
+			ibp->n_pkt_drops++;
+			goto runlock;
+		}
+	}
+
+	/* Check for opcode sequence errors. */
+	switch (qp->r_state) {
+	case OP(SEND_FIRST):
+	case OP(SEND_MIDDLE):
+		if (opcode == OP(SEND_MIDDLE) ||
+		    opcode == OP(SEND_LAST) ||
+		    opcode == OP(SEND_LAST_WITH_IMMEDIATE))
+			break;
+		goto inv;
+
+	case OP(RDMA_WRITE_FIRST):
+	case OP(RDMA_WRITE_MIDDLE):
+		if (opcode == OP(RDMA_WRITE_MIDDLE) ||
+		    opcode == OP(RDMA_WRITE_LAST) ||
+		    opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
+			break;
+		goto inv;
+
+	default:
+		if (opcode == OP(SEND_FIRST) ||
+		    opcode == OP(SEND_ONLY) ||
+		    opcode == OP(SEND_ONLY_WITH_IMMEDIATE) ||
+		    opcode == OP(RDMA_WRITE_FIRST) ||
+		    opcode == OP(RDMA_WRITE_ONLY) ||
+		    opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
+			break;
+		goto inv;
+	}
+
+	if (qp->state == IB_QPS_RTR && !(qp->r_flags & QIB_R_COMM_EST)) {
+		qp->r_flags |= QIB_R_COMM_EST;
+		if (qp->ibqp.event_handler) {
+			struct ib_event ev;
+
+			ev.device = qp->ibqp.device;
+			ev.element.qp = &qp->ibqp;
+			ev.event = IB_EVENT_COMM_EST;
+			qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+		}
+	}
+
+	/* OK, process the packet. */
+	switch (opcode) {
+	case OP(SEND_FIRST):
+	case OP(SEND_ONLY):
+	case OP(SEND_ONLY_WITH_IMMEDIATE):
+send_first:
+		if (qp->r_flags & QIB_R_REUSE_SGE) {
+			qp->r_flags &= ~QIB_R_REUSE_SGE;
+			fix_mr_refcount(qp);
+			qp->r_sge = qp->s_rdma_read_sge;
+		} else {
+			ret = qib_get_rwqe(qp, 0);
+			if (ret < 0)
+				goto op_err;
+			if (!ret) {
+				ibp->n_pkt_drops++;
+				goto runlock;
+			}
+			qp->s_rdma_read_sge = qp->r_sge;
+		}
+		qp->r_rcv_len = 0;
+		if (opcode == OP(SEND_ONLY))
+			goto send_last;
+		else if (opcode == OP(SEND_ONLY_WITH_IMMEDIATE))
+			goto send_last_imm;
+		/* FALLTHROUGH */
+	case OP(SEND_MIDDLE):
+		/* Check for invalid length PMTU or posted rwqe len. */
+		if (unlikely(tlen != (hdrsize + pmtu + 4))) {
+			qp->r_flags |= QIB_R_REUSE_SGE;
+			ibp->n_pkt_drops++;
+			goto runlock;
+		}
+		qp->r_rcv_len += pmtu;
+		if (unlikely(qp->r_rcv_len > qp->r_len)) {
+			qp->r_flags |= QIB_R_REUSE_SGE;
+			ibp->n_pkt_drops++;
+			goto runlock;
+		}
+		qib_copy_sge(&qp->r_sge, data, pmtu, 1);
+		break;
+
+	case OP(SEND_LAST_WITH_IMMEDIATE):
+send_last_imm:
+		wc.ex.imm_data = ohdr->u.imm_data;
+		hdrsize += 4;
+		wc.wc_flags = IB_WC_WITH_IMM;
+		/* FALLTHROUGH */
+	case OP(SEND_LAST):
+send_last:
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/* Check for invalid length. */
+		/* XXX LAST len should be >= 1 */
+		if (unlikely(tlen < (hdrsize + pad + 4))) {
+			qp->r_flags |= QIB_R_REUSE_SGE;
+			ibp->n_pkt_drops++;
+			goto runlock;
+		}
+		/* Don't count the CRC. */
+		tlen -= (hdrsize + pad + 4);
+		wc.byte_len = tlen + qp->r_rcv_len;
+		if (unlikely(wc.byte_len > qp->r_len)) {
+			qp->r_flags |= QIB_R_REUSE_SGE;
+			ibp->n_pkt_drops++;
+			goto runlock;
+		}
+		wc.opcode = IB_WC_RECV;
+last_imm:
+		qib_copy_sge(&qp->r_sge, data, tlen, 1);
+		while (qp->r_sge.num_sge) {
+			atomic_dec(&qp->r_sge.sge.mr->refcount);
+			if (--qp->r_sge.num_sge)
+				qp->r_sge.sge = *qp->r_sge.sg_list++;
+		}
+		wc.wr_id = qp->r_wr_id;
+		wc.status = IB_WC_SUCCESS;
+		wc.qp = &qp->ibqp;
+		wc.src_qp = qp->remote_qpn;
+		wc.slid = qp->remote_ah_attr.dlid;
+		wc.sl = qp->remote_ah_attr.sl;
+		/* Signal completion event if the solicited bit is set. */
+		qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+			     (ohdr->bth[0] &
+				cpu_to_be32(IB_BTH_SOLICITED)) != 0);
+		break;
+
+	case OP(RDMA_WRITE_FIRST):
+	case OP(RDMA_WRITE_ONLY):
+	case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE): /* consume RWQE */
+rdma_first:
+		if (unlikely(!(qp->qp_access_flags &
+			       IB_ACCESS_REMOTE_WRITE))) {
+			ibp->n_pkt_drops++;
+			goto runlock;
+		}
+		reth = &ohdr->u.rc.reth;
+		hdrsize += sizeof(*reth);
+		qp->r_len = be32_to_cpu(reth->length);
+		qp->r_rcv_len = 0;
+		qp->r_sge.sg_list = NULL;
+		if (qp->r_len != 0) {
+			u32 rkey = be32_to_cpu(reth->rkey);
+			u64 vaddr = be64_to_cpu(reth->vaddr);
+			int ok;
+
+			/* Check rkey */
+			ok = qib_rkey_ok(qp, &qp->r_sge.sge, qp->r_len,
+					 vaddr, rkey, IB_ACCESS_REMOTE_WRITE);
+			if (unlikely(!ok)) {
+				ibp->n_pkt_drops++;
+				goto runlock;
+			}
+			qp->r_sge.num_sge = 1;
+		} else {
+			qp->r_sge.num_sge = 0;
+			qp->r_sge.sge.mr = NULL;
+			qp->r_sge.sge.vaddr = NULL;
+			qp->r_sge.sge.length = 0;
+			qp->r_sge.sge.sge_length = 0;
+		}
+		if (opcode == OP(RDMA_WRITE_ONLY))
+			goto rdma_last;
+		else if (opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
+			goto rdma_last_imm;
+		/* FALLTHROUGH */
+	case OP(RDMA_WRITE_MIDDLE):
+		/* Check for invalid length PMTU or posted rwqe len. */
+		if (unlikely(tlen != (hdrsize + pmtu + 4))) {
+			ibp->n_pkt_drops++;
+			goto runlock;
+		}
+		qp->r_rcv_len += pmtu;
+		if (unlikely(qp->r_rcv_len > qp->r_len)) {
+			ibp->n_pkt_drops++;
+			goto runlock;
+		}
+		qib_copy_sge(&qp->r_sge, data, pmtu, 1);
+		break;
+
+	case OP(RDMA_WRITE_LAST_WITH_IMMEDIATE):
+rdma_last_imm:
+		wc.ex.imm_data = ohdr->u.imm_data;
+		hdrsize += 4;
+		wc.wc_flags = IB_WC_WITH_IMM;
+
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/* Check for invalid length. */
+		/* XXX LAST len should be >= 1 */
+		if (unlikely(tlen < (hdrsize + pad + 4))) {
+			ibp->n_pkt_drops++;
+			goto runlock;
+		}
+		/* Don't count the CRC. */
+		tlen -= (hdrsize + pad + 4);
+		if (unlikely(tlen + qp->r_rcv_len != qp->r_len)) {
+			ibp->n_pkt_drops++;
+			goto runlock;
+		}
+		if (qp->r_flags & QIB_R_REUSE_SGE)
+			qp->r_flags &= ~QIB_R_REUSE_SGE;
+		else {
+			ret = qib_get_rwqe(qp, 1);
+			if (ret < 0)
+				goto op_err;
+			if (!ret) {
+				ibp->n_pkt_drops++;
+				goto runlock;
+			}
+		}
+		wc.byte_len = qp->r_len;
+		wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
+		goto last_imm;
+
+	case OP(RDMA_WRITE_LAST):
+rdma_last:
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/* Check for invalid length. */
+		/* XXX LAST len should be >= 1 */
+		if (unlikely(tlen < (hdrsize + pad + 4))) {
+			ibp->n_pkt_drops++;
+			goto runlock;
+		}
+		/* Don't count the CRC. */
+		tlen -= (hdrsize + pad + 4);
+		if (unlikely(tlen + qp->r_rcv_len != qp->r_len)) {
+			ibp->n_pkt_drops++;
+			goto runlock;
+		}
+		qib_copy_sge(&qp->r_sge, data, tlen, 1);
+		while (qp->r_sge.num_sge) {
+			atomic_dec(&qp->r_sge.sge.mr->refcount);
+			if (--qp->r_sge.num_sge)
+				qp->r_sge.sge = *qp->r_sge.sg_list++;
+		}
+		break;
+
+	default:
+		/* Drop packet for unknown opcodes. */
+		ibp->n_pkt_drops++;
+		goto runlock;
+	}
+	qp->r_psn++;
+	qp->r_state = opcode;
+runlock:
+	spin_unlock(&qp->r_lock);
+	return;
+
+op_err:
+	qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
+	spin_unlock(&qp->r_lock);
+	return;
+
+sunlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 42/51] IB/qib: Add qib_ud.c

[Ralph Campbell]

creates the qib_ud.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_ud.c |  607 ++++++++++++++++++++++++++++++++++++
 1 files changed, 607 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_ud.c

diff --git a/drivers/infiniband/hw/qib/qib_ud.c b/drivers/infiniband/hw/qib/qib_ud.c
new file mode 100644
index 0000000..c838cda
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_ud.c
@@ -0,0 +1,607 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_smi.h>
+
+#include "qib.h"
+#include "qib_mad.h"
+
+/**
+ * qib_ud_loopback - handle send on loopback QPs
+ * @sqp: the sending QP
+ * @swqe: the send work request
+ *
+ * This is called from qib_make_ud_req() to forward a WQE addressed
+ * to the same HCA.
+ * Note that the receive interrupt handler may be calling qib_ud_rcv()
+ * while this is being called.
+ */
+static void qib_ud_loopback(struct qib_qp *sqp, struct qib_swqe *swqe)
+{
+	struct qib_ibport *ibp = to_iport(sqp->ibqp.device, sqp->port_num);
+	struct qib_pportdata *ppd;
+	struct qib_qp *qp;
+	struct ib_ah_attr *ah_attr;
+	unsigned long flags;
+	struct qib_sge_state ssge;
+	struct qib_sge *sge;
+	struct ib_wc wc;
+	u32 length;
+
+	qp = qib_lookup_qpn(ibp, swqe->wr.wr.ud.remote_qpn);
+	if (!qp) {
+		ibp->n_pkt_drops++;
+		return;
+	}
+	if (qp->ibqp.qp_type != sqp->ibqp.qp_type ||
+	    !(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)) {
+		ibp->n_pkt_drops++;
+		goto drop;
+	}
+
+	ah_attr = &to_iah(swqe->wr.wr.ud.ah)->attr;
+	ppd = ppd_from_ibp(ibp);
+
+	if (qp->ibqp.qp_num > 1) {
+		u16 pkey1;
+		u16 pkey2;
+		u16 lid;
+
+		pkey1 = qib_get_pkey(ibp, sqp->s_pkey_index);
+		pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
+		if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
+			lid = ppd->lid | (ah_attr->src_path_bits &
+					  ((1 << ppd->lmc) - 1));
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY, pkey1,
+				      ah_attr->sl,
+				      sqp->ibqp.qp_num, qp->ibqp.qp_num,
+				      cpu_to_be16(lid),
+				      cpu_to_be16(ah_attr->dlid));
+			goto drop;
+		}
+	}
+
+	/*
+	 * Check that the qkey matches (except for QP0, see 9.6.1.4.1).
+	 * Qkeys with the high order bit set mean use the
+	 * qkey from the QP context instead of the WR (see 10.2.5).
+	 */
+	if (qp->ibqp.qp_num) {
+		u32 qkey;
+
+		qkey = (int)swqe->wr.wr.ud.remote_qkey < 0 ?
+			sqp->qkey : swqe->wr.wr.ud.remote_qkey;
+		if (unlikely(qkey != qp->qkey)) {
+			u16 lid;
+
+			lid = ppd->lid | (ah_attr->src_path_bits &
+					  ((1 << ppd->lmc) - 1));
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_QKEY, qkey,
+				      ah_attr->sl,
+				      sqp->ibqp.qp_num, qp->ibqp.qp_num,
+				      cpu_to_be16(lid),
+				      cpu_to_be16(ah_attr->dlid));
+			goto drop;
+		}
+	}
+
+	/*
+	 * A GRH is expected to preceed the data even if not
+	 * present on the wire.
+	 */
+	length = swqe->length;
+	memset(&wc, 0, sizeof wc);
+	wc.byte_len = length + sizeof(struct ib_grh);
+
+	if (swqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
+		wc.wc_flags = IB_WC_WITH_IMM;
+		wc.ex.imm_data = swqe->wr.ex.imm_data;
+	}
+
+	spin_lock_irqsave(&qp->r_lock, flags);
+
+	/*
+	 * Get the next work request entry to find where to put the data.
+	 */
+	if (qp->r_flags & QIB_R_REUSE_SGE)
+		qp->r_flags &= ~QIB_R_REUSE_SGE;
+	else {
+		int ret;
+
+		ret = qib_get_rwqe(qp, 0);
+		if (ret < 0) {
+			qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
+			goto bail_unlock;
+		}
+		if (!ret) {
+			if (qp->ibqp.qp_num == 0)
+				ibp->n_vl15_dropped++;
+			goto bail_unlock;
+		}
+	}
+	/* Silently drop packets which are too big. */
+	if (unlikely(wc.byte_len > qp->r_len)) {
+		qp->r_flags |= QIB_R_REUSE_SGE;
+		ibp->n_pkt_drops++;
+		goto bail_unlock;
+	}
+
+	if (ah_attr->ah_flags & IB_AH_GRH) {
+		qib_copy_sge(&qp->r_sge, &ah_attr->grh,
+			     sizeof(struct ib_grh), 1);
+		wc.wc_flags |= IB_WC_GRH;
+	} else
+		qib_skip_sge(&qp->r_sge, sizeof(struct ib_grh), 1);
+	ssge.sg_list = swqe->sg_list + 1;
+	ssge.sge = *swqe->sg_list;
+	ssge.num_sge = swqe->wr.num_sge;
+	sge = &ssge.sge;
+	while (length) {
+		u32 len = sge->length;
+
+		if (len > length)
+			len = length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		qib_copy_sge(&qp->r_sge, sge->vaddr, len, 1);
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (--ssge.num_sge)
+				*sge = *ssge.sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+		length -= len;
+	}
+	while (qp->r_sge.num_sge) {
+		atomic_dec(&qp->r_sge.sge.mr->refcount);
+		if (--qp->r_sge.num_sge)
+			qp->r_sge.sge = *qp->r_sge.sg_list++;
+	}
+	if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
+		goto bail_unlock;
+	wc.wr_id = qp->r_wr_id;
+	wc.status = IB_WC_SUCCESS;
+	wc.opcode = IB_WC_RECV;
+	wc.qp = &qp->ibqp;
+	wc.src_qp = sqp->ibqp.qp_num;
+	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
+		swqe->wr.wr.ud.pkey_index : 0;
+	wc.slid = ppd->lid | (ah_attr->src_path_bits & ((1 << ppd->lmc) - 1));
+	wc.sl = ah_attr->sl;
+	wc.dlid_path_bits = ah_attr->dlid & ((1 << ppd->lmc) - 1);
+	wc.port_num = qp->port_num;
+	/* Signal completion event if the solicited bit is set. */
+	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+		     swqe->wr.send_flags & IB_SEND_SOLICITED);
+	ibp->n_loop_pkts++;
+bail_unlock:
+	spin_unlock_irqrestore(&qp->r_lock, flags);
+drop:
+	if (atomic_dec_and_test(&qp->refcount))
+		wake_up(&qp->wait);
+}
+
+/**
+ * qib_make_ud_req - construct a UD request packet
+ * @qp: the QP
+ *
+ * Return 1 if constructed; otherwise, return 0.
+ */
+int qib_make_ud_req(struct qib_qp *qp)
+{
+	struct qib_other_headers *ohdr;
+	struct ib_ah_attr *ah_attr;
+	struct qib_pportdata *ppd;
+	struct qib_ibport *ibp;
+	struct qib_swqe *wqe;
+	unsigned long flags;
+	u32 nwords;
+	u32 extra_bytes;
+	u32 bth0;
+	u16 lrh0;
+	u16 lid;
+	int ret = 0;
+	int next_cur;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_NEXT_SEND_OK)) {
+		if (!(ib_qib_state_ops[qp->state] & QIB_FLUSH_SEND))
+			goto bail;
+		/* We are in the error state, flush the work request. */
+		if (qp->s_last == qp->s_head)
+			goto bail;
+		/* If DMAs are in progress, we can't flush immediately. */
+		if (atomic_read(&qp->s_dma_busy)) {
+			qp->s_flags |= QIB_S_WAIT_DMA;
+			goto bail;
+		}
+		wqe = get_swqe_ptr(qp, qp->s_last);
+		qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
+		goto done;
+	}
+
+	if (qp->s_cur == qp->s_head)
+		goto bail;
+
+	wqe = get_swqe_ptr(qp, qp->s_cur);
+	next_cur = qp->s_cur + 1;
+	if (next_cur >= qp->s_size)
+		next_cur = 0;
+
+	/* Construct the header. */
+	ibp = to_iport(qp->ibqp.device, qp->port_num);
+	ppd = ppd_from_ibp(ibp);
+	ah_attr = &to_iah(wqe->wr.wr.ud.ah)->attr;
+	if (ah_attr->dlid >= QIB_MULTICAST_LID_BASE) {
+		if (ah_attr->dlid != QIB_PERMISSIVE_LID)
+			ibp->n_multicast_xmit++;
+		else
+			ibp->n_unicast_xmit++;
+	} else {
+		ibp->n_unicast_xmit++;
+		lid = ah_attr->dlid & ~((1 << ppd->lmc) - 1);
+		if (unlikely(lid == ppd->lid)) {
+			/*
+			 * If DMAs are in progress, we can't generate
+			 * a completion for the loopback packet since
+			 * it would be out of order.
+			 * XXX Instead of waiting, we could queue a
+			 * zero length descriptor so we get a callback.
+			 */
+			if (atomic_read(&qp->s_dma_busy)) {
+				qp->s_flags |= QIB_S_WAIT_DMA;
+				goto bail;
+			}
+			qp->s_cur = next_cur;
+			spin_unlock_irqrestore(&qp->s_lock, flags);
+			qib_ud_loopback(qp, wqe);
+			spin_lock_irqsave(&qp->s_lock, flags);
+			qib_send_complete(qp, wqe, IB_WC_SUCCESS);
+			goto done;
+		}
+	}
+
+	qp->s_cur = next_cur;
+	extra_bytes = -wqe->length & 3;
+	nwords = (wqe->length + extra_bytes) >> 2;
+
+	/* header size in 32-bit words LRH+BTH+DETH = (8+12+8)/4. */
+	qp->s_hdrwords = 7;
+	qp->s_cur_size = wqe->length;
+	qp->s_cur_sge = &qp->s_sge;
+	qp->s_srate = ah_attr->static_rate;
+	qp->s_wqe = wqe;
+	qp->s_sge.sge = wqe->sg_list[0];
+	qp->s_sge.sg_list = wqe->sg_list + 1;
+	qp->s_sge.num_sge = wqe->wr.num_sge;
+	qp->s_sge.total_len = wqe->length;
+
+	if (ah_attr->ah_flags & IB_AH_GRH) {
+		/* Header size in 32-bit words. */
+		qp->s_hdrwords += qib_make_grh(ibp, &qp->s_hdr.u.l.grh,
+					       &ah_attr->grh,
+					       qp->s_hdrwords, nwords);
+		lrh0 = QIB_LRH_GRH;
+		ohdr = &qp->s_hdr.u.l.oth;
+		/*
+		 * Don't worry about sending to locally attached multicast
+		 * QPs.  It is unspecified by the spec. what happens.
+		 */
+	} else {
+		/* Header size in 32-bit words. */
+		lrh0 = QIB_LRH_BTH;
+		ohdr = &qp->s_hdr.u.oth;
+	}
+	if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
+		qp->s_hdrwords++;
+		ohdr->u.ud.imm_data = wqe->wr.ex.imm_data;
+		bth0 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE << 24;
+	} else
+		bth0 = IB_OPCODE_UD_SEND_ONLY << 24;
+	lrh0 |= ah_attr->sl << 4;
+	if (qp->ibqp.qp_type == IB_QPT_SMI)
+		lrh0 |= 0xF000; /* Set VL (see ch. 13.5.3.1) */
+	else
+		lrh0 |= ibp->sl_to_vl[ah_attr->sl] << 12;
+	qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
+	qp->s_hdr.lrh[1] = cpu_to_be16(ah_attr->dlid);  /* DEST LID */
+	qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
+	lid = ppd->lid;
+	if (lid) {
+		lid |= ah_attr->src_path_bits & ((1 << ppd->lmc) - 1);
+		qp->s_hdr.lrh[3] = cpu_to_be16(lid);
+	} else
+		qp->s_hdr.lrh[3] = IB_LID_PERMISSIVE;
+	if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+		bth0 |= IB_BTH_SOLICITED;
+	bth0 |= extra_bytes << 20;
+	bth0 |= qp->ibqp.qp_type == IB_QPT_SMI ? QIB_DEFAULT_P_KEY :
+		qib_get_pkey(ibp, qp->ibqp.qp_type == IB_QPT_GSI ?
+			     wqe->wr.wr.ud.pkey_index : qp->s_pkey_index);
+	ohdr->bth[0] = cpu_to_be32(bth0);
+	/*
+	 * Use the multicast QP if the destination LID is a multicast LID.
+	 */
+	ohdr->bth[1] = ah_attr->dlid >= QIB_MULTICAST_LID_BASE &&
+		ah_attr->dlid != QIB_PERMISSIVE_LID ?
+		cpu_to_be32(QIB_MULTICAST_QPN) :
+		cpu_to_be32(wqe->wr.wr.ud.remote_qpn);
+	ohdr->bth[2] = cpu_to_be32(qp->s_next_psn++ & QIB_PSN_MASK);
+	/*
+	 * Qkeys with the high order bit set mean use the
+	 * qkey from the QP context instead of the WR (see 10.2.5).
+	 */
+	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->wr.wr.ud.remote_qkey < 0 ?
+					 qp->qkey : wqe->wr.wr.ud.remote_qkey);
+	ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);
+
+done:
+	ret = 1;
+	goto unlock;
+
+bail:
+	qp->s_flags &= ~QIB_S_BUSY;
+unlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+	return ret;
+}
+
+static unsigned qib_lookup_pkey(struct qib_ibport *ibp, u16 pkey)
+{
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct qib_devdata *dd = ppd->dd;
+	unsigned ctxt = ppd->hw_pidx;
+	unsigned i;
+
+	pkey &= 0x7fff;	/* remove limited/full membership bit */
+
+	for (i = 0; i < ARRAY_SIZE(dd->rcd[ctxt]->pkeys); ++i)
+		if ((dd->rcd[ctxt]->pkeys[i] & 0x7fff) == pkey)
+			return i;
+
+	/*
+	 * Should not get here, this means hardware failed to validate pkeys.
+	 * Punt and return index 0.
+	 */
+	return 0;
+}
+
+/**
+ * qib_ud_rcv - receive an incoming UD packet
+ * @ibp: the port the packet came in on
+ * @hdr: the packet header
+ * @has_grh: true if the packet has a GRH
+ * @data: the packet data
+ * @tlen: the packet length
+ * @qp: the QP the packet came on
+ *
+ * This is called from qib_qp_rcv() to process an incoming UD packet
+ * for the given QP.
+ * Called at interrupt level.
+ */
+void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp)
+{
+	struct qib_other_headers *ohdr;
+	int opcode;
+	u32 hdrsize;
+	u32 pad;
+	struct ib_wc wc;
+	u32 qkey;
+	u32 src_qp;
+	u16 dlid;
+
+	/* Check for GRH */
+	if (!has_grh) {
+		ohdr = &hdr->u.oth;
+		hdrsize = 8 + 12 + 8;   /* LRH + BTH + DETH */
+	} else {
+		ohdr = &hdr->u.l.oth;
+		hdrsize = 8 + 40 + 12 + 8; /* LRH + GRH + BTH + DETH */
+	}
+	qkey = be32_to_cpu(ohdr->u.ud.deth[0]);
+	src_qp = be32_to_cpu(ohdr->u.ud.deth[1]) & QIB_QPN_MASK;
+
+	/* Get the number of bytes the message was padded by. */
+	pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+	if (unlikely(tlen < (hdrsize + pad + 4))) {
+		/* Drop incomplete packets. */
+		ibp->n_pkt_drops++;
+		goto bail;
+	}
+	tlen -= hdrsize + pad + 4;
+
+	/*
+	 * Check that the permissive LID is only used on QP0
+	 * and the QKEY matches (see 9.6.1.4.1 and 9.6.1.5.1).
+	 */
+	if (qp->ibqp.qp_num) {
+		if (unlikely(hdr->lrh[1] == IB_LID_PERMISSIVE ||
+			     hdr->lrh[3] == IB_LID_PERMISSIVE)) {
+			ibp->n_pkt_drops++;
+			goto bail;
+		}
+		if (qp->ibqp.qp_num > 1) {
+			u16 pkey1, pkey2;
+
+			pkey1 = be32_to_cpu(ohdr->bth[0]);
+			pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
+			if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
+				qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY,
+					      pkey1,
+					      (be16_to_cpu(hdr->lrh[0]) >> 4) &
+						0xF,
+					      src_qp, qp->ibqp.qp_num,
+					      hdr->lrh[3], hdr->lrh[1]);
+				goto bail;
+			}
+		}
+		if (unlikely(qkey != qp->qkey)) {
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_QKEY, qkey,
+				      (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
+				      src_qp, qp->ibqp.qp_num,
+				      hdr->lrh[3], hdr->lrh[1]);
+			goto bail;
+		}
+		/* Drop invalid MAD packets (see 13.5.3.1). */
+		if (unlikely(qp->ibqp.qp_num == 1 &&
+			     (tlen != 256 ||
+			      (be16_to_cpu(hdr->lrh[0]) >> 12) == 15))) {
+			ibp->n_pkt_drops++;
+			goto bail;
+		}
+	} else {
+		struct ib_smp *smp;
+
+		/* Drop invalid MAD packets (see 13.5.3.1). */
+		if (tlen != 256 || (be16_to_cpu(hdr->lrh[0]) >> 12) != 15) {
+			ibp->n_pkt_drops++;
+			goto bail;
+		}
+		smp = (struct ib_smp *) data;
+		if ((hdr->lrh[1] == IB_LID_PERMISSIVE ||
+		     hdr->lrh[3] == IB_LID_PERMISSIVE) &&
+		    smp->mgmt_class != IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
+			ibp->n_pkt_drops++;
+			goto bail;
+		}
+	}
+
+	/*
+	 * The opcode is in the low byte when its in network order
+	 * (top byte when in host order).
+	 */
+	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
+	if (qp->ibqp.qp_num > 1 &&
+	    opcode == IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE) {
+		wc.ex.imm_data = ohdr->u.ud.imm_data;
+		wc.wc_flags = IB_WC_WITH_IMM;
+		hdrsize += sizeof(u32);
+	} else if (opcode == IB_OPCODE_UD_SEND_ONLY) {
+		wc.ex.imm_data = 0;
+		wc.wc_flags = 0;
+	} else {
+		ibp->n_pkt_drops++;
+		goto bail;
+	}
+
+	/*
+	 * A GRH is expected to preceed the data even if not
+	 * present on the wire.
+	 */
+	wc.byte_len = tlen + sizeof(struct ib_grh);
+
+	/*
+	 * We need to serialize getting a receive work queue entry and
+	 * generating a completion for it against QPs sending to this QP
+	 * locally.
+	 */
+	spin_lock(&qp->r_lock);
+
+	/*
+	 * Get the next work request entry to find where to put the data.
+	 */
+	if (qp->r_flags & QIB_R_REUSE_SGE)
+		qp->r_flags &= ~QIB_R_REUSE_SGE;
+	else {
+		int ret;
+
+		ret = qib_get_rwqe(qp, 0);
+		if (ret < 0) {
+			qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
+			goto bail_unlock;
+		}
+		if (!ret) {
+			if (qp->ibqp.qp_num == 0)
+				ibp->n_vl15_dropped++;
+			goto bail_unlock;
+		}
+	}
+	/* Silently drop packets which are too big. */
+	if (unlikely(wc.byte_len > qp->r_len)) {
+		qp->r_flags |= QIB_R_REUSE_SGE;
+		ibp->n_pkt_drops++;
+		goto bail_unlock;
+	}
+	if (has_grh) {
+		qib_copy_sge(&qp->r_sge, &hdr->u.l.grh,
+			     sizeof(struct ib_grh), 1);
+		wc.wc_flags |= IB_WC_GRH;
+	} else
+		qib_skip_sge(&qp->r_sge, sizeof(struct ib_grh), 1);
+	qib_copy_sge(&qp->r_sge, data, wc.byte_len - sizeof(struct ib_grh), 1);
+	while (qp->r_sge.num_sge) {
+		atomic_dec(&qp->r_sge.sge.mr->refcount);
+		if (--qp->r_sge.num_sge)
+			qp->r_sge.sge = *qp->r_sge.sg_list++;
+	}
+	if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
+		goto bail_unlock;
+	wc.wr_id = qp->r_wr_id;
+	wc.status = IB_WC_SUCCESS;
+	wc.opcode = IB_WC_RECV;
+	wc.vendor_err = 0;
+	wc.qp = &qp->ibqp;
+	wc.src_qp = src_qp;
+	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
+		qib_lookup_pkey(ibp, be32_to_cpu(ohdr->bth[0])) : 0;
+	wc.slid = be16_to_cpu(hdr->lrh[3]);
+	wc.sl = (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF;
+	dlid = be16_to_cpu(hdr->lrh[1]);
+	/*
+	 * Save the LMC lower bits if the destination LID is a unicast LID.
+	 */
+	wc.dlid_path_bits = dlid >= QIB_MULTICAST_LID_BASE ? 0 :
+		dlid & ((1 << ppd_from_ibp(ibp)->lmc) - 1);
+	wc.port_num = qp->port_num;
+	/* Signal completion event if the solicited bit is set. */
+	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+		     (ohdr->bth[0] &
+			cpu_to_be32(IB_BTH_SOLICITED)) != 0);
+bail_unlock:
+	spin_unlock(&qp->r_lock);
+bail:;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 43/51] IB/qib: Add qib_user_pages.c

[Ralph Campbell]

creates the qib_user_pages.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_user_pages.c |  163 ++++++++++++++++++++++++++++
 1 files changed, 163 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_user_pages.c

diff --git a/drivers/infiniband/hw/qib/qib_user_pages.c b/drivers/infiniband/hw/qib/qib_user_pages.c
new file mode 100644
index 0000000..a0406a2
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_user_pages.c
@@ -0,0 +1,163 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/mm.h>
+#include <linux/device.h>
+
+#include "qib.h"
+
+static void __qib_release_user_pages(struct page **p, size_t num_pages,
+				     int dirty)
+{
+	size_t i;
+
+	for (i = 0; i < num_pages; i++) {
+		qib_cdbg(MM, "%lu/%lu put_page %p\n", (unsigned long) i,
+			 (unsigned long) num_pages, p[i]);
+		if (dirty)
+			set_page_dirty_lock(p[i]);
+		put_page(p[i]);
+	}
+}
+
+/*
+ * Call with current->mm->mmap_sem held.
+ */
+static int __get_user_pages(unsigned long start_page, size_t num_pages,
+			    struct page **p, struct vm_area_struct **vma)
+{
+	unsigned long lock_limit;
+	size_t got;
+	int ret;
+
+	lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >>
+		PAGE_SHIFT;
+
+	if (num_pages > lock_limit) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	qib_cdbg(VERBOSE, "pin %lx pages from vaddr %lx\n",
+		 (unsigned long) num_pages, start_page);
+
+	for (got = 0; got < num_pages; got += ret) {
+		ret = get_user_pages(current, current->mm,
+				     start_page + got * PAGE_SIZE,
+				     num_pages - got, 1, 1,
+				     p + got, vma);
+		if (ret < 0)
+			goto bail_release;
+	}
+
+	current->mm->locked_vm += num_pages;
+
+	ret = 0;
+	goto bail;
+
+bail_release:
+	__qib_release_user_pages(p, got, 0);
+bail:
+	return ret;
+}
+
+/**
+ * qib_map_page - a safety wrapper around pci_map_page()
+ *
+ * A dma_addr of all 0's is interpreted by the chip as "disabled".
+ * Unfortunately, it can also be a valid dma_addr returned on some
+ * architectures.
+ *
+ * The powerpc iommu assigns dma_addrs in ascending order, so we don't
+ * have to bother with retries or mapping a dummy page to insure we
+ * don't just get the same mapping again.
+ *
+ * I'm sure we won't be so lucky with other iommu's, so FIXME.
+ */
+dma_addr_t qib_map_page(struct pci_dev *hwdev, struct page *page,
+			unsigned long offset, size_t size, int direction)
+{
+	dma_addr_t phys;
+
+	phys = pci_map_page(hwdev, page, offset, size, direction);
+
+	if (phys == 0) {
+		pci_unmap_page(hwdev, phys, size, direction);
+		phys = pci_map_page(hwdev, page, offset, size, direction);
+		/*
+		 * FIXME: If we get 0 again, we should keep this page,
+		 * map another, then free the 0 page.
+		 */
+	}
+
+	return phys;
+}
+
+/**
+ * qib_get_user_pages - lock user pages into memory
+ * @start_page: the start page
+ * @num_pages: the number of pages
+ * @p: the output page structures
+ *
+ * This function takes a given start page (page aligned user virtual
+ * address) and pins it and the following specified number of pages.  For
+ * now, num_pages is always 1, but that will probably change at some point
+ * (because caller is doing expected sends on a single virtually contiguous
+ * buffer, so we can do all pages at once).
+ */
+int qib_get_user_pages(unsigned long start_page, size_t num_pages,
+		       struct page **p)
+{
+	int ret;
+
+	down_write(&current->mm->mmap_sem);
+
+	ret = __get_user_pages(start_page, num_pages, p, NULL);
+
+	up_write(&current->mm->mmap_sem);
+
+	return ret;
+}
+
+void qib_release_user_pages(struct page **p, size_t num_pages)
+{
+	if (current->mm) /* during close after signal, mm can be NULL */
+		down_write(&current->mm->mmap_sem);
+
+	__qib_release_user_pages(p, num_pages, 1);
+
+	if (current->mm) {
+		current->mm->locked_vm -= num_pages;
+		up_write(&current->mm->mmap_sem);
+	}
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 44/51] IB/qib: Add qib_user_sdma.c

[Ralph Campbell]

creates the qib_user_sdma.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_user_sdma.c |  909 +++++++++++++++++++++++++++++
 1 files changed, 909 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_user_sdma.c

diff --git a/drivers/infiniband/hw/qib/qib_user_sdma.c b/drivers/infiniband/hw/qib/qib_user_sdma.c
new file mode 100644
index 0000000..869244e
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_user_sdma.c
@@ -0,0 +1,909 @@
+/*
+ * Copyright (c) 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/mm.h>
+#include <linux/types.h>
+#include <linux/device.h>
+#include <linux/dmapool.h>
+#include <linux/slab.h>
+#include <linux/list.h>
+#include <linux/highmem.h>
+#include <linux/io.h>
+#include <linux/uio.h>
+#include <linux/rbtree.h>
+#include <linux/spinlock.h>
+#include <linux/delay.h>
+
+#include "qib.h"
+#include "qib_user_sdma.h"
+
+/* minimum size of header */
+#define QIB_USER_SDMA_MIN_HEADER_LENGTH 64
+/* expected size of headers (for dma_pool) */
+#define QIB_USER_SDMA_EXP_HEADER_LENGTH 64
+
+struct qib_user_sdma_pkt {
+	u8 naddr;               /* dimension of addr (1..3) ... */
+	u32 counter;            /* sdma pkts queued counter for this entry */
+	u64 added;              /* global descq number of entries */
+
+	struct {
+		u32 offset;                     /* offset for kvaddr, addr */
+		u32 length;                     /* length in page */
+		u8  put_page;                   /* should we put_page? */
+		u8  dma_mapped;                 /* is page dma_mapped? */
+		struct page *page;              /* may be NULL (coherent mem) */
+		void *kvaddr;                   /* FIXME: only for pio hack */
+		dma_addr_t addr;
+	} addr[4];   /* max pages, any more and we coalesce */
+	struct list_head list;  /* list element */
+};
+
+struct qib_user_sdma_queue {
+	/*
+	 * pkts sent to dma engine are queued on this
+	 * list head.  the type of the elements of this
+	 * list are struct qib_user_sdma_pkt...
+	 */
+	struct list_head sent;
+
+	/* headers with expected length are allocated from here... */
+	char header_cache_name[64];
+	struct dma_pool *header_cache;
+
+	/* packets are allocated from the slab cache... */
+	char pkt_slab_name[64];
+	struct kmem_cache *pkt_slab;
+
+	/* as packets go on the queued queue, they are counted... */
+	u32 counter;
+	u32 sent_counter;
+
+	/* dma page table */
+	struct rb_root dma_pages_root;
+
+	/* protect everything above... */
+	struct mutex lock;
+};
+
+struct qib_user_sdma_queue *
+qib_user_sdma_queue_create(struct device *dev, int unit, int ctxt, int sctxt)
+{
+	struct qib_user_sdma_queue *pq =
+		kmalloc(sizeof(struct qib_user_sdma_queue), GFP_KERNEL);
+
+	if (!pq)
+		goto done;
+
+	pq->counter = 0;
+	pq->sent_counter = 0;
+	INIT_LIST_HEAD(&pq->sent);
+
+	mutex_init(&pq->lock);
+
+	snprintf(pq->pkt_slab_name, sizeof(pq->pkt_slab_name),
+		 "qib-user-sdma-pkts-%u-%02u.%02u", unit, ctxt, sctxt);
+	pq->pkt_slab = kmem_cache_create(pq->pkt_slab_name,
+					 sizeof(struct qib_user_sdma_pkt),
+					 0, 0, NULL);
+
+	if (!pq->pkt_slab)
+		goto err_kfree;
+
+	snprintf(pq->header_cache_name, sizeof(pq->header_cache_name),
+		 "qib-user-sdma-headers-%u-%02u.%02u", unit, ctxt, sctxt);
+	pq->header_cache = dma_pool_create(pq->header_cache_name,
+					   dev,
+					   QIB_USER_SDMA_EXP_HEADER_LENGTH,
+					   4, 0);
+	if (!pq->header_cache)
+		goto err_slab;
+
+	pq->dma_pages_root = RB_ROOT;
+
+	goto done;
+
+err_slab:
+	kmem_cache_destroy(pq->pkt_slab);
+err_kfree:
+	kfree(pq);
+	pq = NULL;
+
+done:
+	return pq;
+}
+
+static void qib_user_sdma_init_frag(struct qib_user_sdma_pkt *pkt,
+				    int i, size_t offset, size_t len,
+				    int put_page, int dma_mapped,
+				    struct page *page,
+				    void *kvaddr, dma_addr_t dma_addr)
+{
+	pkt->addr[i].offset = offset;
+	pkt->addr[i].length = len;
+	pkt->addr[i].put_page = put_page;
+	pkt->addr[i].dma_mapped = dma_mapped;
+	pkt->addr[i].page = page;
+	pkt->addr[i].kvaddr = kvaddr;
+	pkt->addr[i].addr = dma_addr;
+}
+
+static void qib_user_sdma_init_header(struct qib_user_sdma_pkt *pkt,
+				      u32 counter, size_t offset,
+				      size_t len, int dma_mapped,
+				      struct page *page,
+				      void *kvaddr, dma_addr_t dma_addr)
+{
+	pkt->naddr = 1;
+	pkt->counter = counter;
+	qib_user_sdma_init_frag(pkt, 0, offset, len, 0, dma_mapped, page,
+				kvaddr, dma_addr);
+}
+
+/* we've too many pages in the iovec, coalesce to a single page */
+static int qib_user_sdma_coalesce(const struct qib_devdata *dd,
+				  struct qib_user_sdma_pkt *pkt,
+				  const struct iovec *iov,
+				  unsigned long niov)
+{
+	int ret = 0;
+	struct page *page = alloc_page(GFP_KERNEL);
+	void *mpage_save;
+	char *mpage;
+	int i;
+	int len = 0;
+	dma_addr_t dma_addr;
+
+	if (!page) {
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	mpage = kmap(page);
+	mpage_save = mpage;
+	for (i = 0; i < niov; i++) {
+		int cfur;
+
+		cfur = copy_from_user(mpage,
+				      iov[i].iov_base, iov[i].iov_len);
+		if (cfur) {
+			ret = -EFAULT;
+			goto free_unmap;
+		}
+
+		mpage += iov[i].iov_len;
+		len += iov[i].iov_len;
+	}
+
+	dma_addr = dma_map_page(&dd->pcidev->dev, page, 0, len,
+				DMA_TO_DEVICE);
+	if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
+		ret = -ENOMEM;
+		goto free_unmap;
+	}
+
+	qib_user_sdma_init_frag(pkt, 1, 0, len, 0, 1, page, mpage_save,
+				dma_addr);
+	pkt->naddr = 2;
+
+	goto done;
+
+free_unmap:
+	kunmap(page);
+	__free_page(page);
+done:
+	return ret;
+}
+
+/*
+ * How many pages in this iovec element?
+ */
+static int qib_user_sdma_num_pages(const struct iovec *iov)
+{
+	const unsigned long addr  = (unsigned long) iov->iov_base;
+	const unsigned long  len  = iov->iov_len;
+	const unsigned long spage = addr & PAGE_MASK;
+	const unsigned long epage = (addr + len - 1) & PAGE_MASK;
+
+	return 1 + ((epage - spage) >> PAGE_SHIFT);
+}
+
+/*
+ * Truncate length to page boundry.
+ */
+static int qib_user_sdma_page_length(unsigned long addr, unsigned long len)
+{
+	const unsigned long offset = addr & ~PAGE_MASK;
+
+	return ((offset + len) > PAGE_SIZE) ? (PAGE_SIZE - offset) : len;
+}
+
+static void qib_user_sdma_free_pkt_frag(struct device *dev,
+					struct qib_user_sdma_queue *pq,
+					struct qib_user_sdma_pkt *pkt,
+					int frag)
+{
+	const int i = frag;
+
+	if (pkt->addr[i].page) {
+		if (pkt->addr[i].dma_mapped)
+			dma_unmap_page(dev,
+				       pkt->addr[i].addr,
+				       pkt->addr[i].length,
+				       DMA_TO_DEVICE);
+
+		if (pkt->addr[i].kvaddr)
+			kunmap(pkt->addr[i].page);
+
+		if (pkt->addr[i].put_page)
+			put_page(pkt->addr[i].page);
+		else
+			__free_page(pkt->addr[i].page);
+	} else if (pkt->addr[i].kvaddr)
+		/* free coherent mem from cache... */
+		dma_pool_free(pq->header_cache,
+			      pkt->addr[i].kvaddr, pkt->addr[i].addr);
+}
+
+/* return number of pages pinned... */
+static int qib_user_sdma_pin_pages(const struct qib_devdata *dd,
+				   struct qib_user_sdma_pkt *pkt,
+				   unsigned long addr, int tlen, int npages)
+{
+	struct page *pages[2];
+	int j;
+	int ret;
+
+	ret = get_user_pages(current, current->mm, addr,
+			     npages, 0, 1, pages, NULL);
+
+	if (ret != npages) {
+		int i;
+
+		for (i = 0; i < ret; i++)
+			put_page(pages[i]);
+
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	for (j = 0; j < npages; j++) {
+		/* map the pages... */
+		const int flen = qib_user_sdma_page_length(addr, tlen);
+		dma_addr_t dma_addr =
+			dma_map_page(&dd->pcidev->dev,
+				     pages[j], 0, flen, DMA_TO_DEVICE);
+		unsigned long fofs = addr & ~PAGE_MASK;
+
+		if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
+			ret = -ENOMEM;
+			goto done;
+		}
+
+		qib_user_sdma_init_frag(pkt, pkt->naddr, fofs, flen, 1, 1,
+					pages[j], kmap(pages[j]), dma_addr);
+
+		pkt->naddr++;
+		addr += flen;
+		tlen -= flen;
+	}
+
+done:
+	return ret;
+}
+
+static int qib_user_sdma_pin_pkt(const struct qib_devdata *dd,
+				 struct qib_user_sdma_queue *pq,
+				 struct qib_user_sdma_pkt *pkt,
+				 const struct iovec *iov,
+				 unsigned long niov)
+{
+	int ret = 0;
+	unsigned long idx;
+
+	for (idx = 0; idx < niov; idx++) {
+		const int npages = qib_user_sdma_num_pages(iov + idx);
+		const unsigned long addr = (unsigned long) iov[idx].iov_base;
+
+		ret = qib_user_sdma_pin_pages(dd, pkt, addr,
+					      iov[idx].iov_len, npages);
+		if (ret < 0)
+			goto free_pkt;
+	}
+
+	goto done;
+
+free_pkt:
+	for (idx = 0; idx < pkt->naddr; idx++)
+		qib_user_sdma_free_pkt_frag(&dd->pcidev->dev, pq, pkt, idx);
+
+done:
+	return ret;
+}
+
+static int qib_user_sdma_init_payload(const struct qib_devdata *dd,
+				      struct qib_user_sdma_queue *pq,
+				      struct qib_user_sdma_pkt *pkt,
+				      const struct iovec *iov,
+				      unsigned long niov, int npages)
+{
+	int ret = 0;
+
+	if (npages >= ARRAY_SIZE(pkt->addr))
+		ret = qib_user_sdma_coalesce(dd, pkt, iov, niov);
+	else
+		ret = qib_user_sdma_pin_pkt(dd, pq, pkt, iov, niov);
+
+	return ret;
+}
+
+/* free a packet list -- return counter value of last packet */
+static void qib_user_sdma_free_pkt_list(struct device *dev,
+					struct qib_user_sdma_queue *pq,
+					struct list_head *list)
+{
+	struct qib_user_sdma_pkt *pkt, *pkt_next;
+
+	list_for_each_entry_safe(pkt, pkt_next, list, list) {
+		int i;
+
+		for (i = 0; i < pkt->naddr; i++)
+			qib_user_sdma_free_pkt_frag(dev, pq, pkt, i);
+
+		kmem_cache_free(pq->pkt_slab, pkt);
+	}
+}
+
+/*
+ * copy headers, coalesce etc -- pq->lock must be held
+ *
+ * we queue all the packets to list, returning the
+ * number of bytes total.  list must be empty initially,
+ * as, if there is an error we clean it...
+ */
+static int qib_user_sdma_queue_pkts(const struct qib_devdata *dd,
+				    struct qib_user_sdma_queue *pq,
+				    struct list_head *list,
+				    const struct iovec *iov,
+				    unsigned long niov,
+				    int maxpkts)
+{
+	unsigned long idx = 0;
+	int ret = 0;
+	int npkts = 0;
+	struct page *page = NULL;
+	__le32 *pbc;
+	dma_addr_t dma_addr;
+	struct qib_user_sdma_pkt *pkt = NULL;
+	size_t len;
+	size_t nw;
+	u32 counter = pq->counter;
+	int dma_mapped = 0;
+
+	while (idx < niov && npkts < maxpkts) {
+		const unsigned long addr = (unsigned long) iov[idx].iov_base;
+		const unsigned long idx_save = idx;
+		unsigned pktnw;
+		unsigned pktnwc;
+		int nfrags = 0;
+		int npages = 0;
+		int cfur;
+
+		dma_mapped = 0;
+		len = iov[idx].iov_len;
+		nw = len >> 2;
+		page = NULL;
+
+		pkt = kmem_cache_alloc(pq->pkt_slab, GFP_KERNEL);
+		if (!pkt) {
+			ret = -ENOMEM;
+			goto free_list;
+		}
+
+		if (len < QIB_USER_SDMA_MIN_HEADER_LENGTH ||
+		    len > PAGE_SIZE || len & 3 || addr & 3) {
+			qib_dbg("iov_len %lx invalid\n", (long)len);
+			ret = -EINVAL;
+			goto free_pkt;
+		}
+
+		if (len == QIB_USER_SDMA_EXP_HEADER_LENGTH)
+			pbc = dma_pool_alloc(pq->header_cache, GFP_KERNEL,
+					     &dma_addr);
+		else
+			pbc = NULL;
+
+		if (!pbc) {
+			page = alloc_page(GFP_KERNEL);
+			if (!page) {
+				ret = -ENOMEM;
+				goto free_pkt;
+			}
+			pbc = kmap(page);
+		}
+
+		cfur = copy_from_user(pbc, iov[idx].iov_base, len);
+		if (cfur) {
+			ret = -EFAULT;
+			goto free_pbc;
+		}
+
+		/*
+		 * This assignment is a bit strange.  it's because the
+		 * the pbc counts the number of 32 bit words in the full
+		 * packet _except_ the first word of the pbc itself...
+		 */
+		pktnwc = nw - 1;
+
+		/*
+		 * pktnw computation yields the number of 32 bit words
+		 * that the caller has indicated in the PBC.  note that
+		 * this is one less than the total number of words that
+		 * goes to the send DMA engine as the first 32 bit word
+		 * of the PBC itself is not counted.  Armed with this count,
+		 * we can verify that the packet is consistent with the
+		 * iovec lengths.
+		 */
+		pktnw = le32_to_cpu(*pbc) & QIB_PBC_LENGTH_MASK;
+		if (pktnw < pktnwc || pktnw > pktnwc + (PAGE_SIZE >> 2)) {
+			qib_dbg("pbc nwords %x invalid (pbc32 %x)\n", pktnw,
+				le32_to_cpu(*pbc));
+			ret = -EINVAL;
+			goto free_pbc;
+		}
+
+		idx++;
+		while (pktnwc < pktnw && idx < niov) {
+			const size_t slen = iov[idx].iov_len;
+			const unsigned long faddr =
+				(unsigned long) iov[idx].iov_base;
+
+			if (slen & 3 || faddr & 3 || !slen ||
+			    slen > PAGE_SIZE) {
+				qib_dbg("iov_len %lx or addr %lx idx=%lu bad\n",
+					(long)slen, faddr, idx);
+				ret = -EINVAL;
+				goto free_pbc;
+			}
+
+			npages++;
+			if ((faddr & PAGE_MASK) !=
+			    ((faddr + slen - 1) & PAGE_MASK))
+				npages++;
+
+			pktnwc += slen >> 2;
+			idx++;
+			nfrags++;
+		}
+
+		if (pktnwc != pktnw) {
+			qib_dbg("pktnwc %x != pktnw %x, pbc32 %x\n",
+				pktnwc, pktnw, le32_to_cpu(*pbc));
+			ret = -EINVAL;
+			goto free_pbc;
+		}
+
+		if (page) {
+			dma_addr = dma_map_page(&dd->pcidev->dev,
+						page, 0, len, DMA_TO_DEVICE);
+			if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
+				ret = -ENOMEM;
+				goto free_pbc;
+			}
+
+			dma_mapped = 1;
+		}
+
+		qib_user_sdma_init_header(pkt, counter, 0, len, dma_mapped,
+					  page, pbc, dma_addr);
+
+		if (nfrags) {
+			ret = qib_user_sdma_init_payload(dd, pq, pkt,
+							 iov + idx_save + 1,
+							 nfrags, npages);
+			if (ret < 0)
+				goto free_pbc_dma;
+		}
+
+		counter++;
+		npkts++;
+
+		list_add_tail(&pkt->list, list);
+	}
+
+	ret = idx;
+	goto done;
+
+free_pbc_dma:
+	if (dma_mapped)
+		dma_unmap_page(&dd->pcidev->dev, dma_addr, len, DMA_TO_DEVICE);
+free_pbc:
+	if (page) {
+		kunmap(page);
+		__free_page(page);
+	} else
+		dma_pool_free(pq->header_cache, pbc, dma_addr);
+free_pkt:
+	kmem_cache_free(pq->pkt_slab, pkt);
+free_list:
+	qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, list);
+done:
+	return ret;
+}
+
+static void qib_user_sdma_set_complete_counter(struct qib_user_sdma_queue *pq,
+					       u32 c)
+{
+	pq->sent_counter = c;
+}
+
+/* try to clean out queue -- needs pq->lock */
+static int qib_user_sdma_queue_clean(struct qib_pportdata *ppd,
+				     struct qib_user_sdma_queue *pq)
+{
+	struct qib_devdata *dd = ppd->dd;
+	struct list_head free_list;
+	struct qib_user_sdma_pkt *pkt;
+	struct qib_user_sdma_pkt *pkt_prev;
+	int ret = 0;
+
+	INIT_LIST_HEAD(&free_list);
+
+	list_for_each_entry_safe(pkt, pkt_prev, &pq->sent, list) {
+		s64 descd = ppd->sdma_descq_removed - pkt->added;
+
+		if (descd < 0)
+			break;
+
+		list_move_tail(&pkt->list, &free_list);
+
+		/* one more packet cleaned */
+		ret++;
+	}
+
+	if (!list_empty(&free_list)) {
+		u32 counter;
+
+		pkt = list_entry(free_list.prev,
+				 struct qib_user_sdma_pkt, list);
+		counter = pkt->counter;
+
+		qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &free_list);
+		qib_user_sdma_set_complete_counter(pq, counter);
+	}
+
+	return ret;
+}
+
+void qib_user_sdma_queue_destroy(struct qib_user_sdma_queue *pq)
+{
+	if (!pq)
+		return;
+
+	kmem_cache_destroy(pq->pkt_slab);
+	dma_pool_destroy(pq->header_cache);
+	kfree(pq);
+}
+
+/* clean descriptor queue, returns > 0 if some elements cleaned */
+static int qib_user_sdma_hwqueue_clean(struct qib_pportdata *ppd)
+{
+	int ret;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+	ret = qib_sdma_make_progress(ppd);
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+	return ret;
+}
+
+/* we're in close, drain packets so that we can cleanup successfully... */
+void qib_user_sdma_queue_drain(struct qib_pportdata *ppd,
+			       struct qib_user_sdma_queue *pq)
+{
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+	int i;
+
+	if (!pq)
+		return;
+
+	for (i = 0; i < 500; i++) {
+		mutex_lock(&pq->lock);
+		if (list_empty(&pq->sent)) {
+			mutex_unlock(&pq->lock);
+			break;
+		}
+		qib_user_sdma_hwqueue_clean(ppd);
+		qib_user_sdma_queue_clean(ppd, pq);
+		mutex_unlock(&pq->lock);
+		msleep(10);
+	}
+
+	if (!list_empty(&pq->sent)) {
+		struct list_head free_list;
+
+		spin_lock_irqsave(&ppd->sdma_lock, flags);
+		ppd->dd->f_dump_sdma_state(ppd);
+		spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+		qib_dev_err(dd, "user sdma lists not empty: forcing!\n");
+		INIT_LIST_HEAD(&free_list);
+		mutex_lock(&pq->lock);
+		list_splice_init(&pq->sent, &free_list);
+		qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &free_list);
+		mutex_unlock(&pq->lock);
+	}
+}
+
+static inline __le64 qib_sdma_make_desc0(struct qib_pportdata *ppd,
+					 u64 addr, u64 dwlen, u64 dwoffset)
+{
+	u8 tmpgen;
+
+	tmpgen = ppd->sdma_generation;
+
+	return cpu_to_le64(/* SDmaPhyAddr[31:0] */
+			   ((addr & 0xfffffffcULL) << 32) |
+			   /* SDmaGeneration[1:0] */
+			   ((tmpgen & 3ULL) << 30) |
+			   /* SDmaDwordCount[10:0] */
+			   ((dwlen & 0x7ffULL) << 16) |
+			   /* SDmaBufOffset[12:2] */
+			   (dwoffset & 0x7ffULL));
+}
+
+static inline __le64 qib_sdma_make_first_desc0(__le64 descq)
+{
+	return descq | cpu_to_le64(1ULL << 12);
+}
+
+static inline __le64 qib_sdma_make_last_desc0(__le64 descq)
+{
+					      /* last */  /* dma head */
+	return descq | cpu_to_le64(1ULL << 11 | 1ULL << 13);
+}
+
+static inline __le64 qib_sdma_make_desc1(u64 addr)
+{
+	/* SDmaPhyAddr[47:32] */
+	return cpu_to_le64(addr >> 32);
+}
+
+static void qib_user_sdma_send_frag(struct qib_pportdata *ppd,
+				    struct qib_user_sdma_pkt *pkt, int idx,
+				    unsigned ofs, u16 tail)
+{
+	const u64 addr = (u64) pkt->addr[idx].addr +
+		(u64) pkt->addr[idx].offset;
+	const u64 dwlen = (u64) pkt->addr[idx].length / 4;
+	__le64 *descqp;
+	__le64 descq0;
+
+	descqp = &ppd->sdma_descq[tail].qw[0];
+
+	descq0 = qib_sdma_make_desc0(ppd, addr, dwlen, ofs);
+	if (idx == 0)
+		descq0 = qib_sdma_make_first_desc0(descq0);
+	if (idx == pkt->naddr - 1)
+		descq0 = qib_sdma_make_last_desc0(descq0);
+
+	descqp[0] = descq0;
+	descqp[1] = qib_sdma_make_desc1(addr);
+}
+
+/* pq->lock must be held, get packets on the wire... */
+static int qib_user_sdma_push_pkts(struct qib_pportdata *ppd,
+				   struct qib_user_sdma_queue *pq,
+				   struct list_head *pktlist)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret = 0;
+	unsigned long flags;
+	u16 tail;
+	u8 generation;
+	u64 descq_added;
+
+	if (list_empty(pktlist))
+		return 0;
+
+	if (unlikely(!(ppd->lflags & QIBL_LINKACTIVE)))
+		return -ECOMM;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	/* keep a copy for restoring purposes in case of problems */
+	generation = ppd->sdma_generation;
+	descq_added = ppd->sdma_descq_added;
+
+	if (unlikely(!__qib_sdma_running(ppd))) {
+		ret = -ECOMM;
+		goto unlock;
+	}
+
+	tail = ppd->sdma_descq_tail;
+	while (!list_empty(pktlist)) {
+		struct qib_user_sdma_pkt *pkt =
+			list_entry(pktlist->next, struct qib_user_sdma_pkt,
+				   list);
+		int i;
+		unsigned ofs = 0;
+		u16 dtail = tail;
+
+		if (pkt->naddr > qib_sdma_descq_freecnt(ppd))
+			goto unlock_check_tail;
+
+		for (i = 0; i < pkt->naddr; i++) {
+			qib_user_sdma_send_frag(ppd, pkt, i, ofs, tail);
+			ofs += pkt->addr[i].length >> 2;
+
+			if (++tail == ppd->sdma_descq_cnt) {
+				tail = 0;
+				++ppd->sdma_generation;
+			}
+		}
+
+		if ((ofs << 2) > ppd->ibmaxlen) {
+			qib_dbg("packet size %X > ibmax %X, fail\n",
+				ofs << 2, ppd->ibmaxlen);
+			ret = -EMSGSIZE;
+			goto unlock;
+		}
+
+		/*
+		 * If the packet is >= 2KB mtu equivalent, we have to use
+		 * the large buffers, and have to mark each descriptor as
+		 * part of a large buffer packet.
+		 */
+		if (ofs > dd->piosize2kmax_dwords) {
+			for (i = 0; i < pkt->naddr; i++) {
+				ppd->sdma_descq[dtail].qw[0] |=
+					cpu_to_le64(1ULL << 14);
+				if (++dtail == ppd->sdma_descq_cnt)
+					dtail = 0;
+			}
+		}
+
+		ppd->sdma_descq_added += pkt->naddr;
+		pkt->added = ppd->sdma_descq_added;
+		list_move_tail(&pkt->list, &pq->sent);
+		ret++;
+	}
+
+unlock_check_tail:
+	/* advance the tail on the chip if necessary */
+	if (ppd->sdma_descq_tail != tail)
+		dd->f_sdma_update_tail(ppd, tail);
+
+unlock:
+	if (unlikely(ret < 0)) {
+		ppd->sdma_generation = generation;
+		ppd->sdma_descq_added = descq_added;
+	}
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+	return ret;
+}
+
+int qib_user_sdma_writev(struct qib_ctxtdata *rcd,
+			 struct qib_user_sdma_queue *pq,
+			 const struct iovec *iov,
+			 unsigned long dim)
+{
+	struct qib_devdata *dd = rcd->dd;
+	struct qib_pportdata *ppd = rcd->ppd;
+	int ret = 0;
+	struct list_head list;
+	int npkts = 0;
+
+	INIT_LIST_HEAD(&list);
+
+	mutex_lock(&pq->lock);
+
+	/* why not -ECOMM like qib_user_sdma_push_pkts() below? */
+	if (!qib_sdma_running(ppd))
+		goto done_unlock;
+
+	if (ppd->sdma_descq_added != ppd->sdma_descq_removed) {
+		qib_user_sdma_hwqueue_clean(ppd);
+		qib_user_sdma_queue_clean(ppd, pq);
+	}
+
+	while (dim) {
+		const int mxp = 8;
+
+		down_write(&current->mm->mmap_sem);
+		ret = qib_user_sdma_queue_pkts(dd, pq, &list, iov, dim, mxp);
+		up_write(&current->mm->mmap_sem);
+
+		if (ret <= 0)
+			goto done_unlock;
+		else {
+			dim -= ret;
+			iov += ret;
+		}
+
+		/* force packets onto the sdma hw queue... */
+		if (!list_empty(&list)) {
+			/*
+			 * Lazily clean hw queue.  the 4 is a guess of about
+			 * how many sdma descriptors a packet will take (it
+			 * doesn't have to be perfect).
+			 */
+			if (qib_sdma_descq_freecnt(ppd) < ret * 4) {
+				qib_user_sdma_hwqueue_clean(ppd);
+				qib_user_sdma_queue_clean(ppd, pq);
+			}
+
+			ret = qib_user_sdma_push_pkts(ppd, pq, &list);
+			if (ret < 0)
+				goto done_unlock;
+			else {
+				npkts += ret;
+				pq->counter += ret;
+
+				if (!list_empty(&list))
+					goto done_unlock;
+			}
+		}
+	}
+
+done_unlock:
+	if (!list_empty(&list))
+		qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &list);
+	mutex_unlock(&pq->lock);
+
+	return (ret < 0) ? ret : npkts;
+}
+
+int qib_user_sdma_make_progress(struct qib_pportdata *ppd,
+				struct qib_user_sdma_queue *pq)
+{
+	int ret = 0;
+
+	mutex_lock(&pq->lock);
+	qib_user_sdma_hwqueue_clean(ppd);
+	ret = qib_user_sdma_queue_clean(ppd, pq);
+	mutex_unlock(&pq->lock);
+
+	return ret;
+}
+
+u32 qib_user_sdma_complete_counter(const struct qib_user_sdma_queue *pq)
+{
+	return pq ? pq->sent_counter : 0;
+}
+
+u32 qib_user_sdma_inflight_counter(struct qib_user_sdma_queue *pq)
+{
+	return pq ? pq->counter : 0;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 45/51] IB/qib: Add qib_user_sdma.h

[Ralph Campbell]

creates the qib_user_sdma.h file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_user_sdma.h |   52 +++++++++++++++++++++++++++++
 1 files changed, 52 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_user_sdma.h

diff --git a/drivers/infiniband/hw/qib/qib_user_sdma.h b/drivers/infiniband/hw/qib/qib_user_sdma.h
new file mode 100644
index 0000000..ce8cbaf
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_user_sdma.h
@@ -0,0 +1,52 @@
+/*
+ * Copyright (c) 2007, 2008 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/device.h>
+
+struct qib_user_sdma_queue;
+
+struct qib_user_sdma_queue *
+qib_user_sdma_queue_create(struct device *dev, int unit, int port, int sport);
+void qib_user_sdma_queue_destroy(struct qib_user_sdma_queue *pq);
+
+int qib_user_sdma_writev(struct qib_ctxtdata *pd,
+			 struct qib_user_sdma_queue *pq,
+			 const struct iovec *iov,
+			 unsigned long dim);
+
+int qib_user_sdma_make_progress(struct qib_pportdata *ppd,
+				struct qib_user_sdma_queue *pq);
+
+void qib_user_sdma_queue_drain(struct qib_pportdata *ppd,
+			       struct qib_user_sdma_queue *pq);
+
+u32 qib_user_sdma_complete_counter(const struct qib_user_sdma_queue *pq);
+u32 qib_user_sdma_inflight_counter(struct qib_user_sdma_queue *pq);

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 46/51] IB/qib: Add qib_verbs.c

[Ralph Campbell]

creates the qib_verbs.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_verbs.c | 2383 +++++++++++++++++++++++++++++++++
 1 files changed, 2383 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_verbs.c

diff --git a/drivers/infiniband/hw/qib/qib_verbs.c b/drivers/infiniband/hw/qib/qib_verbs.c
new file mode 100644
index 0000000..fa1e74b
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_verbs.c
@@ -0,0 +1,2383 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_mad.h>
+#include <rdma/ib_user_verbs.h>
+#include <linux/io.h>
+#include <linux/utsname.h>
+#include <linux/rculist.h>
+#include <linux/mm.h>
+
+#include "qib.h"
+#include "qib_common.h"
+
+static unsigned int ib_qib_qp_table_size = 251;
+module_param_named(qp_table_size, ib_qib_qp_table_size, uint, S_IRUGO);
+MODULE_PARM_DESC(qp_table_size, "QP table size");
+
+unsigned int ib_qib_lkey_table_size = 16;
+module_param_named(lkey_table_size, ib_qib_lkey_table_size, uint,
+		   S_IRUGO);
+MODULE_PARM_DESC(lkey_table_size,
+		 "LKEY table size in bits (2^n, 1 <= n <= 23)");
+
+static unsigned int ib_qib_max_pds = 0xFFFF;
+module_param_named(max_pds, ib_qib_max_pds, uint, S_IRUGO);
+MODULE_PARM_DESC(max_pds,
+		 "Maximum number of protection domains to support");
+
+static unsigned int ib_qib_max_ahs = 0xFFFF;
+module_param_named(max_ahs, ib_qib_max_ahs, uint, S_IRUGO);
+MODULE_PARM_DESC(max_ahs, "Maximum number of address handles to support");
+
+unsigned int ib_qib_max_cqes = 0x2FFFF;
+module_param_named(max_cqes, ib_qib_max_cqes, uint, S_IRUGO);
+MODULE_PARM_DESC(max_cqes,
+		 "Maximum number of completion queue entries to support");
+
+unsigned int ib_qib_max_cqs = 0x1FFFF;
+module_param_named(max_cqs, ib_qib_max_cqs, uint, S_IRUGO);
+MODULE_PARM_DESC(max_cqs, "Maximum number of completion queues to support");
+
+unsigned int ib_qib_max_qp_wrs = 0x3FFF;
+module_param_named(max_qp_wrs, ib_qib_max_qp_wrs, uint, S_IRUGO);
+MODULE_PARM_DESC(max_qp_wrs, "Maximum number of QP WRs to support");
+
+unsigned int ib_qib_max_qps = 16384;
+module_param_named(max_qps, ib_qib_max_qps, uint, S_IRUGO);
+MODULE_PARM_DESC(max_qps, "Maximum number of QPs to support");
+
+unsigned int ib_qib_max_sges = 0x60;
+module_param_named(max_sges, ib_qib_max_sges, uint, S_IRUGO);
+MODULE_PARM_DESC(max_sges, "Maximum number of SGEs to support");
+
+unsigned int ib_qib_max_mcast_grps = 16384;
+module_param_named(max_mcast_grps, ib_qib_max_mcast_grps, uint, S_IRUGO);
+MODULE_PARM_DESC(max_mcast_grps,
+		 "Maximum number of multicast groups to support");
+
+unsigned int ib_qib_max_mcast_qp_attached = 16;
+module_param_named(max_mcast_qp_attached, ib_qib_max_mcast_qp_attached,
+		   uint, S_IRUGO);
+MODULE_PARM_DESC(max_mcast_qp_attached,
+		 "Maximum number of attached QPs to support");
+
+unsigned int ib_qib_max_srqs = 1024;
+module_param_named(max_srqs, ib_qib_max_srqs, uint, S_IRUGO);
+MODULE_PARM_DESC(max_srqs, "Maximum number of SRQs to support");
+
+unsigned int ib_qib_max_srq_sges = 128;
+module_param_named(max_srq_sges, ib_qib_max_srq_sges, uint, S_IRUGO);
+MODULE_PARM_DESC(max_srq_sges, "Maximum number of SRQ SGEs to support");
+
+unsigned int ib_qib_max_srq_wrs = 0x1FFFF;
+module_param_named(max_srq_wrs, ib_qib_max_srq_wrs, uint, S_IRUGO);
+MODULE_PARM_DESC(max_srq_wrs, "Maximum number of SRQ WRs support");
+
+static unsigned int ib_qib_disable_sma;
+module_param_named(disable_sma, ib_qib_disable_sma, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(disable_sma, "Disable the SMA");
+
+/*
+ * Note that it is OK to post send work requests in the SQE and ERR
+ * states; qib_do_send() will process them and generate error
+ * completions as per IB 1.2 C10-96.
+ */
+const int ib_qib_state_ops[IB_QPS_ERR + 1] = {
+	[IB_QPS_RESET] = 0,
+	[IB_QPS_INIT] = QIB_POST_RECV_OK,
+	[IB_QPS_RTR] = QIB_POST_RECV_OK | QIB_PROCESS_RECV_OK,
+	[IB_QPS_RTS] = QIB_POST_RECV_OK | QIB_PROCESS_RECV_OK |
+	    QIB_POST_SEND_OK | QIB_PROCESS_SEND_OK |
+	    QIB_PROCESS_NEXT_SEND_OK,
+	[IB_QPS_SQD] = QIB_POST_RECV_OK | QIB_PROCESS_RECV_OK |
+	    QIB_POST_SEND_OK | QIB_PROCESS_SEND_OK,
+	[IB_QPS_SQE] = QIB_POST_RECV_OK | QIB_PROCESS_RECV_OK |
+	    QIB_POST_SEND_OK | QIB_FLUSH_SEND,
+	[IB_QPS_ERR] = QIB_POST_RECV_OK | QIB_FLUSH_RECV |
+	    QIB_POST_SEND_OK | QIB_FLUSH_SEND,
+};
+
+struct qib_ucontext {
+	struct ib_ucontext ibucontext;
+};
+
+static inline struct qib_ucontext *to_iucontext(struct ib_ucontext
+						  *ibucontext)
+{
+	return container_of(ibucontext, struct qib_ucontext, ibucontext);
+}
+
+/*
+ * Translate ib_wr_opcode into ib_wc_opcode.
+ */
+const enum ib_wc_opcode ib_qib_wc_opcode[] = {
+	[IB_WR_RDMA_WRITE] = IB_WC_RDMA_WRITE,
+	[IB_WR_RDMA_WRITE_WITH_IMM] = IB_WC_RDMA_WRITE,
+	[IB_WR_SEND] = IB_WC_SEND,
+	[IB_WR_SEND_WITH_IMM] = IB_WC_SEND,
+	[IB_WR_RDMA_READ] = IB_WC_RDMA_READ,
+	[IB_WR_ATOMIC_CMP_AND_SWP] = IB_WC_COMP_SWAP,
+	[IB_WR_ATOMIC_FETCH_AND_ADD] = IB_WC_FETCH_ADD
+};
+
+/*
+ * System image GUID.
+ */
+__be64 ib_qib_sys_image_guid;
+
+/**
+ * qib_copy_sge - copy data to SGE memory
+ * @ss: the SGE state
+ * @data: the data to copy
+ * @length: the length of the data
+ */
+void qib_copy_sge(struct qib_sge_state *ss, void *data, u32 length, int release)
+{
+	struct qib_sge *sge = &ss->sge;
+
+	while (length) {
+		u32 len = sge->length;
+
+		if (len > length)
+			len = length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		memcpy(sge->vaddr, data, len);
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (release)
+				atomic_dec(&sge->mr->refcount);
+			if (--ss->num_sge)
+				*sge = *ss->sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+		data += len;
+		length -= len;
+	}
+}
+
+/**
+ * qib_skip_sge - skip over SGE memory - XXX almost dup of prev func
+ * @ss: the SGE state
+ * @length: the number of bytes to skip
+ */
+void qib_skip_sge(struct qib_sge_state *ss, u32 length, int release)
+{
+	struct qib_sge *sge = &ss->sge;
+
+	while (length) {
+		u32 len = sge->length;
+
+		if (len > length)
+			len = length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (release)
+				atomic_dec(&sge->mr->refcount);
+			if (--ss->num_sge)
+				*sge = *ss->sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+		length -= len;
+	}
+}
+
+/*
+ * Count the number of DMA descriptors needed to send length bytes of data.
+ * Don't modify the qib_sge_state to get the count.
+ * Return zero if any of the segments is not aligned.
+ */
+static u32 qib_count_sge(struct qib_sge_state *ss, u32 length)
+{
+	struct qib_sge *sg_list = ss->sg_list;
+	struct qib_sge sge = ss->sge;
+	u8 num_sge = ss->num_sge;
+	u32 ndesc = 1;  /* count the header */
+
+	while (length) {
+		u32 len = sge.length;
+
+		if (len > length)
+			len = length;
+		if (len > sge.sge_length)
+			len = sge.sge_length;
+		BUG_ON(len == 0);
+		if (((long) sge.vaddr & (sizeof(u32) - 1)) ||
+		    (len != length && (len & (sizeof(u32) - 1)))) {
+			ndesc = 0;
+			break;
+		}
+		ndesc++;
+		sge.vaddr += len;
+		sge.length -= len;
+		sge.sge_length -= len;
+		if (sge.sge_length == 0) {
+			if (--num_sge)
+				sge = *sg_list++;
+		} else if (sge.length == 0 && sge.mr->lkey) {
+			if (++sge.n >= QIB_SEGSZ) {
+				if (++sge.m >= sge.mr->mapsz)
+					break;
+				sge.n = 0;
+			}
+			sge.vaddr =
+				sge.mr->map[sge.m]->segs[sge.n].vaddr;
+			sge.length =
+				sge.mr->map[sge.m]->segs[sge.n].length;
+		}
+		length -= len;
+	}
+	return ndesc;
+}
+
+/*
+ * Copy from the SGEs to the data buffer.
+ */
+static void qib_copy_from_sge(void *data, struct qib_sge_state *ss, u32 length)
+{
+	struct qib_sge *sge = &ss->sge;
+
+	while (length) {
+		u32 len = sge->length;
+
+		if (len > length)
+			len = length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		memcpy(data, sge->vaddr, len);
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (--ss->num_sge)
+				*sge = *ss->sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+		data += len;
+		length -= len;
+	}
+}
+
+/**
+ * qib_post_one_send - post one RC, UC, or UD send work request
+ * @qp: the QP to post on
+ * @wr: the work request to send
+ */
+static int qib_post_one_send(struct qib_qp *qp, struct ib_send_wr *wr)
+{
+	struct qib_swqe *wqe;
+	u32 next;
+	int i;
+	int j;
+	int acc;
+	int ret;
+	unsigned long flags;
+	struct qib_lkey_table *rkt;
+	struct qib_pd *pd;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	/* Check that state is OK to post send. */
+	if (unlikely(!(ib_qib_state_ops[qp->state] & QIB_POST_SEND_OK)))
+		goto bail_inval;
+
+	/* IB spec says that num_sge == 0 is OK. */
+	if (wr->num_sge > qp->s_max_sge)
+		goto bail_inval;
+
+	/*
+	 * Don't allow RDMA reads or atomic operations on UC or
+	 * undefined operations.
+	 * Make sure buffer is large enough to hold the result for atomics.
+	 */
+	if (qp->ibqp.qp_type == IB_QPT_UC) {
+		if ((unsigned) wr->opcode >= IB_WR_RDMA_READ)
+			goto bail_inval;
+	} else if (qp->ibqp.qp_type != IB_QPT_RC) {
+		/* Check IB_QPT_SMI, IB_QPT_GSI, IB_QPT_UD opcode */
+		if (wr->opcode != IB_WR_SEND &&
+		    wr->opcode != IB_WR_SEND_WITH_IMM)
+			goto bail_inval;
+		/* Check UD destination address PD */
+		if (qp->ibqp.pd != wr->wr.ud.ah->pd)
+			goto bail_inval;
+	} else if ((unsigned) wr->opcode > IB_WR_ATOMIC_FETCH_AND_ADD)
+		goto bail_inval;
+	else if (wr->opcode >= IB_WR_ATOMIC_CMP_AND_SWP &&
+		   (wr->num_sge == 0 ||
+		    wr->sg_list[0].length < sizeof(u64) ||
+		    wr->sg_list[0].addr & (sizeof(u64) - 1)))
+		goto bail_inval;
+	else if (wr->opcode >= IB_WR_RDMA_READ && !qp->s_max_rd_atomic)
+		goto bail_inval;
+
+	next = qp->s_head + 1;
+	if (next >= qp->s_size)
+		next = 0;
+	if (next == qp->s_last) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	rkt = &to_idev(qp->ibqp.device)->lk_table;
+	pd = to_ipd(qp->ibqp.pd);
+	wqe = get_swqe_ptr(qp, qp->s_head);
+	wqe->wr = *wr;
+	wqe->length = 0;
+	j = 0;
+	if (wr->num_sge) {
+		acc = wr->opcode >= IB_WR_RDMA_READ ?
+			IB_ACCESS_LOCAL_WRITE : 0;
+		for (i = 0; i < wr->num_sge; i++) {
+			u32 length = wr->sg_list[i].length;
+			int ok;
+
+			if (length == 0)
+				continue;
+			ok = qib_lkey_ok(rkt, pd, &wqe->sg_list[j],
+					 &wr->sg_list[i], acc);
+			if (!ok)
+				goto bail_inval_free;
+			wqe->length += length;
+			j++;
+		}
+		wqe->wr.num_sge = j;
+	}
+	if (qp->ibqp.qp_type == IB_QPT_UC ||
+	    qp->ibqp.qp_type == IB_QPT_RC) {
+		if (wqe->length > 0x80000000U)
+			goto bail_inval_free;
+	} else if (wqe->length > (dd_from_ibdev(qp->ibqp.device)->pport +
+				  qp->port_num - 1)->ibmtu)
+		goto bail_inval_free;
+	else
+		atomic_inc(&to_iah(wr->wr.ud.ah)->refcount);
+	wqe->ssn = qp->s_ssn++;
+	qp->s_head = next;
+
+	ret = 0;
+	goto bail;
+
+bail_inval_free:
+	while (j) {
+		struct qib_sge *sge = &wqe->sg_list[--j];
+
+		atomic_dec(&sge->mr->refcount);
+	}
+bail_inval:
+	ret = -EINVAL;
+bail:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+	return ret;
+}
+
+/**
+ * qib_post_send - post a send on a QP
+ * @ibqp: the QP to post the send on
+ * @wr: the list of work requests to post
+ * @bad_wr: the first bad WR is put here
+ *
+ * This may be called from interrupt context.
+ */
+static int qib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+			 struct ib_send_wr **bad_wr)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+	int err = 0;
+
+	for (; wr; wr = wr->next) {
+		err = qib_post_one_send(qp, wr);
+		if (err) {
+			*bad_wr = wr;
+			goto bail;
+		}
+	}
+
+	/* Try to do the send work in the caller's context. */
+	qib_do_send(&qp->s_work);
+
+bail:
+	return err;
+}
+
+/**
+ * qib_post_receive - post a receive on a QP
+ * @ibqp: the QP to post the receive on
+ * @wr: the WR to post
+ * @bad_wr: the first bad WR is put here
+ *
+ * This may be called from interrupt context.
+ */
+static int qib_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+			    struct ib_recv_wr **bad_wr)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+	struct qib_rwq *wq = qp->r_rq.wq;
+	unsigned long flags;
+	int ret;
+
+	/* Check that state is OK to post receive. */
+	if (!(ib_qib_state_ops[qp->state] & QIB_POST_RECV_OK) || !wq) {
+		*bad_wr = wr;
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	for (; wr; wr = wr->next) {
+		struct qib_rwqe *wqe;
+		u32 next;
+		int i;
+
+		if ((unsigned) wr->num_sge > qp->r_rq.max_sge) {
+			*bad_wr = wr;
+			ret = -EINVAL;
+			goto bail;
+		}
+
+		spin_lock_irqsave(&qp->r_rq.lock, flags);
+		next = wq->head + 1;
+		if (next >= qp->r_rq.size)
+			next = 0;
+		if (next == wq->tail) {
+			spin_unlock_irqrestore(&qp->r_rq.lock, flags);
+			*bad_wr = wr;
+			ret = -ENOMEM;
+			goto bail;
+		}
+
+		wqe = get_rwqe_ptr(&qp->r_rq, wq->head);
+		wqe->wr_id = wr->wr_id;
+		wqe->num_sge = wr->num_sge;
+		for (i = 0; i < wr->num_sge; i++)
+			wqe->sg_list[i] = wr->sg_list[i];
+		/* Make sure queue entry is written before the head index. */
+		smp_wmb();
+		wq->head = next;
+		spin_unlock_irqrestore(&qp->r_rq.lock, flags);
+	}
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+static void qib_dump_hdr(u32 control, u32 plen, __be32 *hdr, unsigned hlen)
+{
+	qib_cdbg(VPKT, "PBC %08x %x HDR len %u\n", control, plen, hlen);
+	while (hlen >= 8) {
+		qib_cdbg(VPKT, "%08x %08x %08x %08x %08x %08x %08x %08x\n",
+			be32_to_cpu(hdr[0]),
+			be32_to_cpu(hdr[1]),
+			be32_to_cpu(hdr[2]),
+			be32_to_cpu(hdr[3]),
+			be32_to_cpu(hdr[4]),
+			be32_to_cpu(hdr[5]),
+			be32_to_cpu(hdr[6]),
+			be32_to_cpu(hdr[7]));
+		hlen -= 8;
+		hdr += 8;
+	}
+	switch (hlen) {
+	case 1:
+		qib_cdbg(VPKT, "%08x\n", be32_to_cpu(hdr[0]));
+		break;
+
+	case 2:
+		qib_cdbg(VPKT, "%08x %08x\n",
+			be32_to_cpu(hdr[0]),
+			be32_to_cpu(hdr[1]));
+		break;
+
+	case 3:
+		qib_cdbg(VPKT, "%08x %08x %08x\n",
+			be32_to_cpu(hdr[0]),
+			be32_to_cpu(hdr[1]),
+			be32_to_cpu(hdr[2]));
+		break;
+
+	case 4:
+		qib_cdbg(VPKT, "%08x %08x %08x %08x\n",
+			be32_to_cpu(hdr[0]),
+			be32_to_cpu(hdr[1]),
+			be32_to_cpu(hdr[2]),
+			be32_to_cpu(hdr[3]));
+		break;
+
+	case 5:
+		qib_cdbg(VPKT, "%08x %08x %08x %08x %08x\n",
+			be32_to_cpu(hdr[0]),
+			be32_to_cpu(hdr[1]),
+			be32_to_cpu(hdr[2]),
+			be32_to_cpu(hdr[3]),
+			be32_to_cpu(hdr[4]));
+		break;
+
+	case 6:
+		qib_cdbg(VPKT, "%08x %08x %08x %08x %08x %08x\n",
+			be32_to_cpu(hdr[0]),
+			be32_to_cpu(hdr[1]),
+			be32_to_cpu(hdr[2]),
+			be32_to_cpu(hdr[3]),
+			be32_to_cpu(hdr[4]),
+			be32_to_cpu(hdr[5]));
+		break;
+
+	case 7:
+		qib_cdbg(VPKT, "%08x %08x %08x %08x %08x %08x %08x\n",
+			be32_to_cpu(hdr[0]),
+			be32_to_cpu(hdr[1]),
+			be32_to_cpu(hdr[2]),
+			be32_to_cpu(hdr[3]),
+			be32_to_cpu(hdr[4]),
+			be32_to_cpu(hdr[5]),
+			be32_to_cpu(hdr[6]));
+		break;
+	}
+}
+
+static void qib_dump_data(char *msg, __be32 *dat, unsigned dlen)
+{
+	qib_cdbg(VPKT, "%s len %u\n", msg, dlen);
+	while (dlen >= 8) {
+		qib_cdbg(VPKT, "%08x %08x %08x %08x %08x %08x %08x %08x\n",
+			be32_to_cpu(dat[0]),
+			be32_to_cpu(dat[1]),
+			be32_to_cpu(dat[2]),
+			be32_to_cpu(dat[3]),
+			be32_to_cpu(dat[4]),
+			be32_to_cpu(dat[5]),
+			be32_to_cpu(dat[6]),
+			be32_to_cpu(dat[7]));
+		dlen -= 8;
+		dat += 8;
+	}
+	switch (dlen) {
+	case 1:
+		qib_cdbg(VPKT, "%08x\n", be32_to_cpu(dat[0]));
+		break;
+
+	case 2:
+		qib_cdbg(VPKT, "%08x %08x\n",
+			be32_to_cpu(dat[0]),
+			be32_to_cpu(dat[1]));
+		break;
+
+	case 3:
+		qib_cdbg(VPKT, "%08x %08x %08x\n",
+			be32_to_cpu(dat[0]),
+			be32_to_cpu(dat[1]),
+			be32_to_cpu(dat[2]));
+		break;
+
+	case 4:
+		qib_cdbg(VPKT, "%08x %08x %08x %08x\n",
+			be32_to_cpu(dat[0]),
+			be32_to_cpu(dat[1]),
+			be32_to_cpu(dat[2]),
+			be32_to_cpu(dat[3]));
+		break;
+
+	case 5:
+		qib_cdbg(VPKT, "%08x %08x %08x %08x %08x\n",
+			be32_to_cpu(dat[0]),
+			be32_to_cpu(dat[1]),
+			be32_to_cpu(dat[2]),
+			be32_to_cpu(dat[3]),
+			be32_to_cpu(dat[4]));
+		break;
+
+	case 6:
+		qib_cdbg(VPKT, "%08x %08x %08x %08x %08x %08x\n",
+			be32_to_cpu(dat[0]),
+			be32_to_cpu(dat[1]),
+			be32_to_cpu(dat[2]),
+			be32_to_cpu(dat[3]),
+			be32_to_cpu(dat[4]),
+			be32_to_cpu(dat[5]));
+		break;
+
+	case 7:
+		qib_cdbg(VPKT, "%08x %08x %08x %08x %08x %08x %08x\n",
+			be32_to_cpu(dat[0]),
+			be32_to_cpu(dat[1]),
+			be32_to_cpu(dat[2]),
+			be32_to_cpu(dat[3]),
+			be32_to_cpu(dat[4]),
+			be32_to_cpu(dat[5]),
+			be32_to_cpu(dat[6]));
+		break;
+	}
+}
+
+/**
+ * qib_qp_rcv - processing an incoming packet on a QP
+ * @rcd: the context pointer
+ * @hdr: the packet header
+ * @has_grh: true if the packet has a GRH
+ * @data: the packet data
+ * @tlen: the packet length
+ * @qp: the QP the packet came on
+ *
+ * This is called from qib_ib_rcv() to process an incoming packet
+ * for the given QP.
+ * Called at interrupt level.
+ */
+static void qib_qp_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
+		       int has_grh, void *data, u32 tlen, struct qib_qp *qp)
+{
+	struct qib_ibport *ibp = &rcd->ppd->ibport_data;
+
+	/* Check for valid receive state. */
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)) {
+		ibp->n_pkt_drops++;
+		return;
+	}
+
+	switch (qp->ibqp.qp_type) {
+	case IB_QPT_SMI:
+	case IB_QPT_GSI:
+		if (ib_qib_disable_sma) {
+			qib_cdbg(ERRPKT, "Dropped typ %x qp%u SMA disabled\n",
+				 qp->ibqp.qp_type, qp->ibqp.qp_num);
+			break;
+		}
+		/* FALLTHROUGH */
+	case IB_QPT_UD:
+		qib_ud_rcv(ibp, hdr, has_grh, data, tlen, qp);
+		break;
+
+	case IB_QPT_RC:
+		qib_rc_rcv(rcd, hdr, has_grh, data, tlen, qp);
+		break;
+
+	case IB_QPT_UC:
+		qib_uc_rcv(ibp, hdr, has_grh, data, tlen, qp);
+		break;
+
+	default:
+		break;
+	}
+}
+
+/**
+ * qib_ib_rcv - process an incoming packet
+ * @rcd: the context pointer
+ * @rhdr: the header of the packet
+ * @data: the packet payload
+ * @tlen: the packet length
+ *
+ * This is called from qib_kreceive() to process an incoming packet at
+ * interrupt level. Tlen is the length of the header + data + CRC in bytes.
+ */
+void qib_ib_rcv(struct qib_ctxtdata *rcd, void *rhdr, void *data, u32 tlen)
+{
+	struct qib_pportdata *ppd = rcd->ppd;
+	struct qib_ibport *ibp = &ppd->ibport_data;
+	struct qib_ib_header *hdr = rhdr;
+	struct qib_other_headers *ohdr;
+	struct qib_qp *qp;
+	u32 qp_num;
+	int lnh;
+	u8 opcode;
+	u16 lid;
+
+	/* 24 == LRH+BTH+CRC */
+	if (unlikely(tlen < 24))
+		goto drop;
+
+	/* Check for a valid destination LID (see ch. 7.11.1). */
+	lid = be16_to_cpu(hdr->lrh[1]);
+	if (lid < QIB_MULTICAST_LID_BASE) {
+		lid &= ~((1 << ppd->lmc) - 1);
+		if (unlikely(lid != ppd->lid))
+			goto drop;
+	}
+
+	/* Check for GRH */
+	lnh = be16_to_cpu(hdr->lrh[0]) & 3;
+	if (lnh == QIB_LRH_BTH)
+		ohdr = &hdr->u.oth;
+	else if (lnh == QIB_LRH_GRH) {
+		u32 vtf;
+
+		ohdr = &hdr->u.l.oth;
+		if (hdr->u.l.grh.next_hdr != IB_GRH_NEXT_HDR)
+			goto drop;
+		vtf = be32_to_cpu(hdr->u.l.grh.version_tclass_flow);
+		if ((vtf >> IB_GRH_VERSION_SHIFT) != IB_GRH_VERSION)
+			goto drop;
+	} else
+		goto drop;
+
+	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
+	ibp->opstats[opcode & 0x7f].n_bytes += tlen;
+	ibp->opstats[opcode & 0x7f].n_packets++;
+
+	/* Get the destination QP number. */
+	qp_num = be32_to_cpu(ohdr->bth[1]) & QIB_QPN_MASK;
+	if (qp_num <= 1) {
+		qib_dump_data("RHDR", rhdr, 7); /* LRH+BTH+DETH */
+		qib_dump_data("RDATA", data, 32); /* first 128 bytes */
+	}
+	if (qp_num == QIB_MULTICAST_QPN) {
+		struct qib_mcast *mcast;
+		struct qib_mcast_qp *p;
+
+		if (lnh != QIB_LRH_GRH)
+			goto drop;
+		mcast = qib_mcast_find(ibp, &hdr->u.l.grh.dgid);
+		if (mcast == NULL)
+			goto drop;
+		ibp->n_multicast_rcv++;
+		list_for_each_entry_rcu(p, &mcast->qp_list, list)
+			qib_qp_rcv(rcd, hdr, 1, data, tlen, p->qp);
+		/*
+		 * Notify qib_multicast_detach() if it is waiting for us
+		 * to finish.
+		 */
+		if (atomic_dec_return(&mcast->refcount) <= 1)
+			wake_up(&mcast->wait);
+	} else {
+		qp = qib_lookup_qpn(ibp, qp_num);
+		if (!qp)
+			goto drop;
+		ibp->n_unicast_rcv++;
+		qib_qp_rcv(rcd, hdr, lnh == QIB_LRH_GRH, data, tlen, qp);
+		/*
+		 * Notify qib_destroy_qp() if it is waiting
+		 * for us to finish.
+		 */
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	}
+	return;
+
+drop:
+	ibp->n_pkt_drops++;
+}
+
+/*
+ * This is called from a timer to check for QPs
+ * which need kernel memory in order to send a packet.
+ */
+static void mem_timer(unsigned long data)
+{
+	struct qib_ibdev *dev = (struct qib_ibdev *) data;
+	struct list_head *list = &dev->memwait;
+	struct qib_qp *qp = NULL;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dev->pending_lock, flags);
+	if (!list_empty(list)) {
+		qp = list_entry(list->next, struct qib_qp, iowait);
+		list_del_init(&qp->iowait);
+		atomic_inc(&qp->refcount);
+		if (!list_empty(list))
+			mod_timer(&dev->mem_timer, jiffies + 1);
+	}
+	spin_unlock_irqrestore(&dev->pending_lock, flags);
+
+	if (qp) {
+		spin_lock_irqsave(&qp->s_lock, flags);
+		if (qp->s_flags & QIB_S_WAIT_KMEM) {
+			qp->s_flags &= ~QIB_S_WAIT_KMEM;
+			qib_schedule_send(qp);
+		}
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	}
+}
+
+static void update_sge(struct qib_sge_state *ss, u32 length)
+{
+	struct qib_sge *sge = &ss->sge;
+
+	sge->vaddr += length;
+	sge->length -= length;
+	sge->sge_length -= length;
+	if (sge->sge_length == 0) {
+		if (--ss->num_sge)
+			*sge = *ss->sg_list++;
+	} else if (sge->length == 0 && sge->mr->lkey) {
+		if (++sge->n >= QIB_SEGSZ) {
+			if (++sge->m >= sge->mr->mapsz)
+				return;
+			sge->n = 0;
+		}
+		sge->vaddr = sge->mr->map[sge->m]->segs[sge->n].vaddr;
+		sge->length = sge->mr->map[sge->m]->segs[sge->n].length;
+	}
+}
+
+#ifdef __LITTLE_ENDIAN
+static inline u32 get_upper_bits(u32 data, u32 shift)
+{
+	return data >> shift;
+}
+
+static inline u32 set_upper_bits(u32 data, u32 shift)
+{
+	return data << shift;
+}
+
+static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off)
+{
+	data <<= ((sizeof(u32) - n) * BITS_PER_BYTE);
+	data >>= ((sizeof(u32) - n - off) * BITS_PER_BYTE);
+	return data;
+}
+#else
+static inline u32 get_upper_bits(u32 data, u32 shift)
+{
+	return data << shift;
+}
+
+static inline u32 set_upper_bits(u32 data, u32 shift)
+{
+	return data >> shift;
+}
+
+static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off)
+{
+	data >>= ((sizeof(u32) - n) * BITS_PER_BYTE);
+	data <<= ((sizeof(u32) - n - off) * BITS_PER_BYTE);
+	return data;
+}
+#endif
+
+static void copy_io(u32 __iomem *piobuf, struct qib_sge_state *ss,
+		    u32 length, unsigned flush_wc)
+{
+	u32 extra = 0;
+	u32 data = 0;
+	u32 last;
+
+	qib_cdbg(VPKT, "UDATA len %u (bytes) %u %lu\n", length, ss->num_sge,
+		(unsigned long)ss->sge.vaddr & (sizeof(u32) - 1));
+	while (1) {
+		u32 len = ss->sge.length;
+		u32 off;
+
+		if (len > length)
+			len = length;
+		if (len > ss->sge.sge_length)
+			len = ss->sge.sge_length;
+		BUG_ON(len == 0);
+		/* If the source address is not aligned, try to align it. */
+		off = (unsigned long)ss->sge.vaddr & (sizeof(u32) - 1);
+		if (off) {
+			u32 *addr = (u32 *)((unsigned long)ss->sge.vaddr &
+					    ~(sizeof(u32) - 1));
+			u32 v = get_upper_bits(*addr, off * BITS_PER_BYTE);
+			u32 y;
+
+			y = sizeof(u32) - off;
+			if (len > y)
+				len = y;
+			if (len + extra >= sizeof(u32)) {
+				data |= set_upper_bits(v, extra *
+						       BITS_PER_BYTE);
+				len = sizeof(u32) - extra;
+				if (len == length) {
+					last = data;
+					break;
+				}
+				__raw_writel(data, piobuf);
+				piobuf++;
+				extra = 0;
+				data = 0;
+			} else {
+				/* Clear unused upper bytes */
+				data |= clear_upper_bytes(v, len, extra);
+				if (len == length) {
+					last = data;
+					break;
+				}
+				extra += len;
+			}
+		} else if (extra) {
+			/* Source address is aligned. */
+			u32 *addr = (u32 *) ss->sge.vaddr;
+			int shift = extra * BITS_PER_BYTE;
+			int ushift = 32 - shift;
+			u32 l = len;
+
+			while (l >= sizeof(u32)) {
+				u32 v = *addr;
+
+				data |= set_upper_bits(v, shift);
+				__raw_writel(data, piobuf);
+				data = get_upper_bits(v, ushift);
+				piobuf++;
+				addr++;
+				l -= sizeof(u32);
+			}
+			/*
+			 * We still have 'extra' number of bytes leftover.
+			 */
+			if (l) {
+				u32 v = *addr;
+
+				if (l + extra >= sizeof(u32)) {
+					data |= set_upper_bits(v, shift);
+					len -= l + extra - sizeof(u32);
+					if (len == length) {
+						last = data;
+						break;
+					}
+					__raw_writel(data, piobuf);
+					piobuf++;
+					extra = 0;
+					data = 0;
+				} else {
+					/* Clear unused upper bytes */
+					data |= clear_upper_bytes(v, l, extra);
+					if (len == length) {
+						last = data;
+						break;
+					}
+					extra += l;
+				}
+			} else if (len == length) {
+				last = data;
+				break;
+			}
+		} else if (len == length) {
+			u32 w;
+
+			/*
+			 * Need to round up for the last dword in the
+			 * packet.
+			 */
+			w = (len + 3) >> 2;
+			qib_dump_data("UDATA", (__be32 *) ss->sge.vaddr, w);
+			qib_pio_copy(piobuf, ss->sge.vaddr, w - 1);
+			piobuf += w - 1;
+			last = ((u32 *) ss->sge.vaddr)[w - 1];
+			break;
+		} else {
+			u32 w = len >> 2;
+
+			qib_dump_data("UDATA", (__be32 *) ss->sge.vaddr, w);
+			qib_pio_copy(piobuf, ss->sge.vaddr, w);
+			piobuf += w;
+
+			extra = len & (sizeof(u32) - 1);
+			if (extra) {
+				u32 v = ((u32 *) ss->sge.vaddr)[w];
+
+				/* Clear unused upper bytes */
+				data = clear_upper_bytes(v, extra, 0);
+			}
+		}
+		update_sge(ss, len);
+		length -= len;
+	}
+	/* Update address before sending packet. */
+	update_sge(ss, length);
+	if (flush_wc) {
+		/* must flush early everything before trigger word */
+		qib_flush_wc();
+		__raw_writel(last, piobuf);
+		/* be sure trigger word is written */
+		qib_flush_wc();
+	} else
+		__raw_writel(last, piobuf);
+}
+
+static struct qib_verbs_txreq *get_txreq(struct qib_ibdev *dev,
+					 struct qib_qp *qp, int *retp)
+{
+	struct qib_verbs_txreq *tx;
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+	spin_lock(&dev->pending_lock);
+
+	if (!list_empty(&dev->txreq_free)) {
+		struct list_head *l = dev->txreq_free.next;
+
+		list_del(l);
+		tx = list_entry(l, struct qib_verbs_txreq, txreq.list);
+		*retp = 0;
+	} else {
+		if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK &&
+		    list_empty(&qp->iowait)) {
+			dev->n_txwait++;
+			qp->s_flags |= QIB_S_WAIT_TX;
+			list_add_tail(&qp->iowait, &dev->txwait);
+		}
+		tx = NULL;
+		qp->s_flags &= ~QIB_S_BUSY;
+		*retp = -EBUSY;
+	}
+
+	spin_unlock(&dev->pending_lock);
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+
+	return tx;
+}
+
+void qib_put_txreq(struct qib_verbs_txreq *tx)
+{
+	struct qib_ibdev *dev;
+	struct qib_qp *qp;
+	unsigned long flags;
+
+	qp = tx->qp;
+	dev = to_idev(qp->ibqp.device);
+
+	if (atomic_dec_and_test(&qp->refcount))
+		wake_up(&qp->wait);
+	if (tx->mr) {
+		atomic_dec(&tx->mr->refcount);
+		tx->mr = NULL;
+	}
+	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_FREEBUF) {
+		tx->txreq.flags &= ~QIB_SDMA_TXREQ_F_FREEBUF;
+		dma_unmap_single(&dd_from_dev(dev)->pcidev->dev,
+				 tx->txreq.addr, tx->hdr_dwords << 2,
+				 DMA_TO_DEVICE);
+		kfree(tx->align_buf);
+	}
+
+	spin_lock_irqsave(&dev->pending_lock, flags);
+
+	/* Put struct back on free list */
+	list_add(&tx->txreq.list, &dev->txreq_free);
+
+	if (!list_empty(&dev->txwait)) {
+		/* Wake up first QP wanting a free struct */
+		qp = list_entry(dev->txwait.next, struct qib_qp, iowait);
+		list_del_init(&qp->iowait);
+		atomic_inc(&qp->refcount);
+		spin_unlock_irqrestore(&dev->pending_lock, flags);
+
+		spin_lock_irqsave(&qp->s_lock, flags);
+		if (qp->s_flags & QIB_S_WAIT_TX) {
+			qp->s_flags &= ~QIB_S_WAIT_TX;
+			qib_schedule_send(qp);
+		}
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	} else
+		spin_unlock_irqrestore(&dev->pending_lock, flags);
+}
+
+/*
+ * This is called when there are send DMA descriptors that might be
+ * available.
+ *
+ * This is called with ppd->sdma_lock held.
+ */
+void qib_verbs_sdma_desc_avail(struct qib_pportdata *ppd, unsigned avail)
+{
+	struct qib_qp *qp, *nqp;
+	struct qib_qp *qps[20];
+	struct qib_ibdev *dev;
+	unsigned i, n;
+
+	n = 0;
+	dev = &ppd->dd->verbs_dev;
+	spin_lock(&dev->pending_lock);
+
+	/* Search wait list for first QP wanting DMA descriptors. */
+	list_for_each_entry_safe(qp, nqp, &dev->dmawait, iowait) {
+		if (qp->port_num != ppd->port)
+			continue;
+		if (n == ARRAY_SIZE(qps))
+			break;
+		if (qp->s_tx->txreq.sg_count > avail)
+			break;
+		avail -= qp->s_tx->txreq.sg_count;
+		list_del_init(&qp->iowait);
+		atomic_inc(&qp->refcount);
+		qps[n++] = qp;
+	}
+
+	spin_unlock(&dev->pending_lock);
+
+	for (i = 0; i < n; i++) {
+		qp = qps[i];
+		spin_lock(&qp->s_lock);
+		if (qp->s_flags & QIB_S_WAIT_DMA_DESC) {
+			qp->s_flags &= ~QIB_S_WAIT_DMA_DESC;
+			qib_schedule_send(qp);
+		}
+		spin_unlock(&qp->s_lock);
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	}
+}
+
+/*
+ * This is called with ppd->sdma_lock held.
+ */
+static void sdma_complete(struct qib_sdma_txreq *cookie, int status)
+{
+	struct qib_verbs_txreq *tx =
+		container_of(cookie, struct qib_verbs_txreq, txreq);
+	struct qib_qp *qp = tx->qp;
+
+	spin_lock(&qp->s_lock);
+	if (tx->wqe)
+		qib_send_complete(qp, tx->wqe, IB_WC_SUCCESS);
+	else if (qp->ibqp.qp_type == IB_QPT_RC) {
+		struct qib_ib_header *hdr;
+
+		if (tx->txreq.flags & QIB_SDMA_TXREQ_F_FREEBUF)
+			hdr = &tx->align_buf->hdr;
+		else {
+			struct qib_ibdev *dev = to_idev(qp->ibqp.device);
+
+			hdr = &dev->pio_hdrs[tx->hdr_inx].hdr;
+		}
+		qib_rc_send_complete(qp, hdr);
+	}
+	if (atomic_dec_and_test(&qp->s_dma_busy)) {
+		if (qp->state == IB_QPS_RESET)
+			wake_up(&qp->wait_dma);
+		else if (qp->s_flags & QIB_S_WAIT_DMA) {
+			qp->s_flags &= ~QIB_S_WAIT_DMA;
+			qib_schedule_send(qp);
+		}
+	}
+	spin_unlock(&qp->s_lock);
+
+	qib_put_txreq(tx);
+}
+
+static int wait_kmem(struct qib_ibdev *dev, struct qib_qp *qp)
+{
+	unsigned long flags;
+	int ret = 0;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+	if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK) {
+		spin_lock(&dev->pending_lock);
+		if (list_empty(&qp->iowait)) {
+			if (list_empty(&dev->memwait))
+				mod_timer(&dev->mem_timer, jiffies + 1);
+			qp->s_flags |= QIB_S_WAIT_KMEM;
+			list_add_tail(&qp->iowait, &dev->memwait);
+		}
+		spin_unlock(&dev->pending_lock);
+		qp->s_flags &= ~QIB_S_BUSY;
+		ret = -EBUSY;
+	}
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+
+	return ret;
+}
+
+static int qib_verbs_send_dma(struct qib_qp *qp, struct qib_ib_header *hdr,
+			      u32 hdrwords, struct qib_sge_state *ss, u32 len,
+			      u32 plen, u32 dwords)
+{
+	struct qib_ibdev *dev = to_idev(qp->ibqp.device);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct qib_verbs_txreq *tx;
+	struct qib_pio_header *phdr;
+	u32 control;
+	u32 ndesc;
+	int ret;
+
+	tx = qp->s_tx;
+	if (tx) {
+		qp->s_tx = NULL;
+		/* resend previously constructed packet */
+		ret = qib_sdma_verbs_send(ppd, tx->ss, tx->dwords, tx);
+		goto bail;
+	}
+
+	tx = get_txreq(dev, qp, &ret);
+	if (!tx)
+		goto bail;
+
+	control = dd->f_setpbc_control(ppd, plen, qp->s_srate,
+				       be16_to_cpu(hdr->lrh[0]) >> 12);
+	tx->qp = qp;
+	atomic_inc(&qp->refcount);
+	tx->wqe = qp->s_wqe;
+	tx->mr = qp->s_rdma_mr;
+	if (qp->s_rdma_mr)
+		qp->s_rdma_mr = NULL;
+	tx->txreq.callback = sdma_complete;
+	if (dd->flags & QIB_HAS_SDMA_TIMEOUT)
+		tx->txreq.flags = QIB_SDMA_TXREQ_F_HEADTOHOST;
+	else
+		tx->txreq.flags = QIB_SDMA_TXREQ_F_INTREQ;
+	if (plen + 1 > dd->piosize2kmax_dwords)
+		tx->txreq.flags |= QIB_SDMA_TXREQ_F_USELARGEBUF;
+
+	if (len) {
+		/*
+		 * Don't try to DMA if it takes more descriptors than
+		 * the queue holds.
+		 */
+		ndesc = qib_count_sge(ss, len);
+		if (ndesc >= ppd->sdma_descq_cnt)
+			ndesc = 0;
+	} else
+		ndesc = 1;
+	if (ndesc) {
+		phdr = &dev->pio_hdrs[tx->hdr_inx];
+		phdr->pbc[0] = cpu_to_le32(plen);
+		phdr->pbc[1] = cpu_to_le32(control);
+		memcpy(&phdr->hdr, hdr, hdrwords << 2);
+		tx->txreq.flags |= QIB_SDMA_TXREQ_F_FREEDESC;
+		tx->txreq.sg_count = ndesc;
+		tx->txreq.addr = dev->pio_hdrs_phys +
+			tx->hdr_inx * sizeof(struct qib_pio_header);
+		tx->hdr_dwords = hdrwords + 2; /* add PBC length */
+		ret = qib_sdma_verbs_send(ppd, ss, dwords, tx);
+		goto bail;
+	}
+
+	/* Allocate a buffer and copy the header and payload to it. */
+	tx->hdr_dwords = plen + 1;
+	phdr = kmalloc(tx->hdr_dwords << 2, GFP_ATOMIC);
+	if (!phdr)
+		goto err_tx;
+	phdr->pbc[0] = cpu_to_le32(plen);
+	phdr->pbc[1] = cpu_to_le32(control);
+	memcpy(&phdr->hdr, hdr, hdrwords << 2);
+	qib_copy_from_sge((u32 *) &phdr->hdr + hdrwords, ss, len);
+
+	tx->txreq.addr = dma_map_single(&dd->pcidev->dev, phdr,
+					tx->hdr_dwords << 2, DMA_TO_DEVICE);
+	if (dma_mapping_error(&dd->pcidev->dev, tx->txreq.addr))
+		goto map_err;
+	tx->align_buf = phdr;
+	tx->txreq.flags |= QIB_SDMA_TXREQ_F_FREEBUF;
+	tx->txreq.sg_count = 1;
+	ret = qib_sdma_verbs_send(ppd, NULL, 0, tx);
+	goto unaligned;
+
+map_err:
+	kfree(phdr);
+err_tx:
+	qib_put_txreq(tx);
+	ret = wait_kmem(dev, qp);
+unaligned:
+	ibp->n_unaligned++;
+bail:
+	return ret;
+}
+
+/*
+ * If we are now in the error state, return zero to flush the
+ * send work request.
+ */
+static int no_bufs_available(struct qib_qp *qp)
+{
+	struct qib_ibdev *dev = to_idev(qp->ibqp.device);
+	struct qib_devdata *dd;
+	unsigned long flags;
+	int ret = 0;
+
+	/*
+	 * Note that as soon as want_buffer() is called and
+	 * possibly before it returns, qib_ib_piobufavail()
+	 * could be called. Therefore, put QP on the I/O wait list before
+	 * enabling the PIO avail interrupt.
+	 */
+	spin_lock_irqsave(&qp->s_lock, flags);
+	if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK) {
+		spin_lock(&dev->pending_lock);
+		if (list_empty(&qp->iowait)) {
+			dev->n_piowait++;
+			qp->s_flags |= QIB_S_WAIT_PIO;
+			list_add_tail(&qp->iowait, &dev->piowait);
+			dd = dd_from_dev(dev);
+			dd->f_wantpiobuf_intr(dd, 1);
+		}
+		spin_unlock(&dev->pending_lock);
+		qp->s_flags &= ~QIB_S_BUSY;
+		ret = -EBUSY;
+	}
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+	return ret;
+}
+
+static int qib_verbs_send_pio(struct qib_qp *qp, struct qib_ib_header *ibhdr,
+			      u32 hdrwords, struct qib_sge_state *ss, u32 len,
+			      u32 plen, u32 dwords)
+{
+	struct qib_devdata *dd = dd_from_ibdev(qp->ibqp.device);
+	struct qib_pportdata *ppd = dd->pport + qp->port_num - 1;
+	u32 *hdr = (u32 *) ibhdr;
+	u32 __iomem *piobuf_orig;
+	u32 __iomem *piobuf;
+	u64 pbc;
+	unsigned long flags;
+	unsigned flush_wc;
+	u32 control;
+	u32 pbufn;
+
+	control = dd->f_setpbc_control(ppd, plen, qp->s_srate,
+		be16_to_cpu(ibhdr->lrh[0]) >> 12);
+	pbc = ((u64) control << 32) | plen;
+	piobuf = dd->f_getsendbuf(ppd, pbc, &pbufn);
+	if (unlikely(piobuf == NULL))
+		return no_bufs_available(qp);
+
+	/*
+	 * Write the pbc.
+	 * We have to flush after the PBC for correctness on some cpus
+	 * or WC buffer can be written out of order.
+	 */
+	qib_dump_hdr(control, plen, (__be32 *)hdr, hdrwords);
+	writeq(pbc, piobuf);
+	piobuf_orig = piobuf;
+	piobuf += 2;
+
+	flush_wc = dd->flags & QIB_PIO_FLUSH_WC;
+	if (len == 0) {
+		/*
+		 * If there is just the header portion, must flush before
+		 * writing last word of header for correctness, and after
+		 * the last header word (trigger word).
+		 */
+		if (flush_wc) {
+			qib_flush_wc();
+			qib_pio_copy(piobuf, hdr, hdrwords - 1);
+			qib_flush_wc();
+			__raw_writel(hdr[hdrwords - 1], piobuf + hdrwords - 1);
+			qib_flush_wc();
+		} else
+			qib_pio_copy(piobuf, hdr, hdrwords);
+		goto done;
+	}
+
+	if (flush_wc)
+		qib_flush_wc();
+	qib_pio_copy(piobuf, hdr, hdrwords);
+	piobuf += hdrwords;
+
+	/* The common case is aligned and contained in one segment. */
+	if (likely(ss->num_sge == 1 && len <= ss->sge.length &&
+		   !((unsigned long)ss->sge.vaddr & (sizeof(u32) - 1)))) {
+		u32 *addr = (u32 *) ss->sge.vaddr;
+
+		qib_dump_data("DATA", ss->sge.vaddr, dwords);
+		/* Update address before sending packet. */
+		update_sge(ss, len);
+		if (flush_wc) {
+			qib_pio_copy(piobuf, addr, dwords - 1);
+			/* must flush early everything before trigger word */
+			qib_flush_wc();
+			__raw_writel(addr[dwords - 1], piobuf + dwords - 1);
+			/* be sure trigger word is written */
+			qib_flush_wc();
+		} else
+			qib_pio_copy(piobuf, addr, dwords);
+		goto done;
+	}
+	copy_io(piobuf, ss, len, flush_wc);
+done:
+	if (dd->flags & QIB_USE_SPCL_TRIG) {
+		u32 spcl_off = (pbufn >= dd->piobcnt2k) ? 2047 : 1023;
+		qib_flush_wc();
+		__raw_writel(0xaebecede, piobuf_orig + spcl_off);
+	}
+	qib_sendbuf_done(dd, pbufn);
+	if (qp->s_rdma_mr) {
+		atomic_dec(&qp->s_rdma_mr->refcount);
+		qp->s_rdma_mr = NULL;
+	}
+	if (qp->s_wqe) {
+		spin_lock_irqsave(&qp->s_lock, flags);
+		qib_send_complete(qp, qp->s_wqe, IB_WC_SUCCESS);
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+	} else if (qp->ibqp.qp_type == IB_QPT_RC) {
+		spin_lock_irqsave(&qp->s_lock, flags);
+		qib_rc_send_complete(qp, ibhdr);
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+	}
+	return 0;
+}
+
+/**
+ * qib_verbs_send - send a packet
+ * @qp: the QP to send on
+ * @hdr: the packet header
+ * @hdrwords: the number of 32-bit words in the header
+ * @ss: the SGE to send
+ * @len: the length of the packet in bytes
+ *
+ * Return zero if packet is sent or queued OK.
+ * Return non-zero and clear qp->s_flags QIB_S_BUSY otherwise.
+ */
+int qib_verbs_send(struct qib_qp *qp, struct qib_ib_header *hdr,
+		   u32 hdrwords, struct qib_sge_state *ss, u32 len)
+{
+	struct qib_devdata *dd = dd_from_ibdev(qp->ibqp.device);
+	u32 plen;
+	int ret;
+	u32 dwords = (len + 3) >> 2;
+
+	/*
+	 * Calculate the send buffer trigger address.
+	 * The +1 counts for the pbc control dword following the pbc length.
+	 */
+	plen = hdrwords + dwords + 1;
+
+	/*
+	 * VL15 packets (IB_QPT_SMI) will always use PIO, so we
+	 * can defer SDMA restart until link goes ACTIVE without
+	 * worrying about just how we got there.
+	 */
+	if (qp->ibqp.qp_type == IB_QPT_SMI ||
+	    !(dd->flags & QIB_HAS_SEND_DMA))
+		ret = qib_verbs_send_pio(qp, hdr, hdrwords, ss, len,
+					 plen, dwords);
+	else
+		ret = qib_verbs_send_dma(qp, hdr, hdrwords, ss, len,
+					 plen, dwords);
+
+	return ret;
+}
+
+int qib_snapshot_counters(struct qib_pportdata *ppd, u64 *swords,
+			  u64 *rwords, u64 *spkts, u64 *rpkts,
+			  u64 *xmit_wait)
+{
+	int ret;
+	struct qib_devdata *dd = ppd->dd;
+
+	if (!(dd->flags & QIB_PRESENT)) {
+		/* no hardware, freeze, etc. */
+		ret = -EINVAL;
+		goto bail;
+	}
+	*swords = dd->f_portcntr(ppd, QIBPORTCNTR_WORDSEND);
+	*rwords = dd->f_portcntr(ppd, QIBPORTCNTR_WORDRCV);
+	*spkts = dd->f_portcntr(ppd, QIBPORTCNTR_PKTSEND);
+	*rpkts = dd->f_portcntr(ppd, QIBPORTCNTR_PKTRCV);
+	*xmit_wait = dd->f_portcntr(ppd, QIBPORTCNTR_SENDSTALL);
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_get_counters - get various chip counters
+ * @dd: the qlogic_ib device
+ * @cntrs: counters are placed here
+ *
+ * Return the counters needed by recv_pma_get_portcounters().
+ */
+int qib_get_counters(struct qib_pportdata *ppd,
+		     struct qib_verbs_counters *cntrs)
+{
+	int ret;
+
+	if (!(ppd->dd->flags & QIB_PRESENT)) {
+		/* no hardware, freeze, etc. */
+		ret = -EINVAL;
+		goto bail;
+	}
+	cntrs->symbol_error_counter =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_IBSYMBOLERR);
+	cntrs->link_error_recovery_counter =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_IBLINKERRRECOV);
+	/*
+	 * The link downed counter counts when the other side downs the
+	 * connection.  We add in the number of times we downed the link
+	 * due to local link integrity errors to compensate.
+	 */
+	cntrs->link_downed_counter =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_IBLINKDOWN);
+	cntrs->port_rcv_errors =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_RXDROPPKT) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_RCVOVFL) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_ERR_RLEN) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_INVALIDRLEN) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_ERRLINK) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_ERRICRC) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_ERRVCRC) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_ERRLPCRC) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_BADFORMAT);
+	cntrs->port_rcv_errors +=
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_RXLOCALPHYERR);
+	cntrs->port_rcv_errors +=
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_RXVLERR);
+	cntrs->port_rcv_remphys_errors =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_RCVEBP);
+	cntrs->port_xmit_discards =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_UNSUPVL);
+	cntrs->port_xmit_data = ppd->dd->f_portcntr(ppd,
+			QIBPORTCNTR_WORDSEND);
+	cntrs->port_rcv_data = ppd->dd->f_portcntr(ppd,
+			QIBPORTCNTR_WORDRCV);
+	cntrs->port_xmit_packets = ppd->dd->f_portcntr(ppd,
+			QIBPORTCNTR_PKTSEND);
+	cntrs->port_rcv_packets = ppd->dd->f_portcntr(ppd,
+			QIBPORTCNTR_PKTRCV);
+	cntrs->local_link_integrity_errors =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_LLI);
+	cntrs->excessive_buffer_overrun_errors =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_EXCESSBUFOVFL);
+	cntrs->vl15_dropped =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_VL15PKTDROP);
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_ib_piobufavail - callback when a PIO buffer is available
+ * @dd: the device pointer
+ *
+ * This is called from qib_intr() at interrupt level when a PIO buffer is
+ * available after qib_verbs_send() returned an error that no buffers were
+ * available. Disable the interrupt if there are no more QPs waiting.
+ */
+void qib_ib_piobufavail(struct qib_devdata *dd)
+{
+	struct qib_ibdev *dev = &dd->verbs_dev;
+	struct list_head *list;
+	struct qib_qp *qps[5];
+	struct qib_qp *qp;
+	unsigned long flags;
+	unsigned i, n;
+
+	list = &dev->piowait;
+	n = 0;
+
+	/*
+	 * Note: checking that the piowait list is empty and clearing
+	 * the buffer available interrupt needs to be atomic or we
+	 * could end up with QPs on the wait list with the interrupt
+	 * disabled.
+	 */
+	spin_lock_irqsave(&dev->pending_lock, flags);
+	while (!list_empty(list)) {
+		if (n == ARRAY_SIZE(qps))
+			goto full;
+		qp = list_entry(list->next, struct qib_qp, iowait);
+		list_del_init(&qp->iowait);
+		atomic_inc(&qp->refcount);
+		qps[n++] = qp;
+	}
+	dd->f_wantpiobuf_intr(dd, 0);
+full:
+	spin_unlock_irqrestore(&dev->pending_lock, flags);
+
+	for (i = 0; i < n; i++) {
+		qp = qps[i];
+
+		spin_lock_irqsave(&qp->s_lock, flags);
+		if (qp->s_flags & QIB_S_WAIT_PIO) {
+			qp->s_flags &= ~QIB_S_WAIT_PIO;
+			qib_schedule_send(qp);
+		}
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+
+		/* Notify qib_destroy_qp() if it is waiting. */
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	}
+}
+
+static int qib_query_device(struct ib_device *ibdev,
+			    struct ib_device_attr *props)
+{
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	struct qib_ibdev *dev = to_idev(ibdev);
+
+	memset(props, 0, sizeof(*props));
+
+	props->device_cap_flags = IB_DEVICE_BAD_PKEY_CNTR |
+		IB_DEVICE_BAD_QKEY_CNTR | IB_DEVICE_SHUTDOWN_PORT |
+		IB_DEVICE_SYS_IMAGE_GUID | IB_DEVICE_RC_RNR_NAK_GEN |
+		IB_DEVICE_PORT_ACTIVE_EVENT | IB_DEVICE_SRQ_RESIZE;
+	props->page_size_cap = PAGE_SIZE;
+	props->vendor_id =
+		QIB_SRC_OUI_1 << 16 | QIB_SRC_OUI_2 << 8 | QIB_SRC_OUI_3;
+	props->vendor_part_id = dd->deviceid;
+	props->hw_ver = dd->minrev;
+	props->sys_image_guid = ib_qib_sys_image_guid;
+	props->max_mr_size = ~0ULL;
+	props->max_qp = ib_qib_max_qps;
+	props->max_qp_wr = ib_qib_max_qp_wrs;
+	props->max_sge = ib_qib_max_sges;
+	props->max_cq = ib_qib_max_cqs;
+	props->max_ah = ib_qib_max_ahs;
+	props->max_cqe = ib_qib_max_cqes;
+	props->max_mr = dev->lk_table.max;
+	props->max_fmr = dev->lk_table.max;
+	props->max_map_per_fmr = 32767;
+	props->max_pd = ib_qib_max_pds;
+	props->max_qp_rd_atom = QIB_MAX_RDMA_ATOMIC;
+	props->max_qp_init_rd_atom = 255;
+	/* props->max_res_rd_atom */
+	props->max_srq = ib_qib_max_srqs;
+	props->max_srq_wr = ib_qib_max_srq_wrs;
+	props->max_srq_sge = ib_qib_max_srq_sges;
+	/* props->local_ca_ack_delay */
+	props->atomic_cap = IB_ATOMIC_GLOB;
+	props->max_pkeys = qib_get_npkeys(dd);
+	props->max_mcast_grp = ib_qib_max_mcast_grps;
+	props->max_mcast_qp_attach = ib_qib_max_mcast_qp_attached;
+	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
+		props->max_mcast_grp;
+
+	return 0;
+}
+
+static int qib_query_port(struct ib_device *ibdev, u8 port,
+			  struct ib_port_attr *props)
+{
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	enum ib_mtu mtu;
+	u16 lid = ppd->lid;
+
+	memset(props, 0, sizeof(*props));
+	props->lid = lid ? lid : be16_to_cpu(IB_LID_PERMISSIVE);
+	props->lmc = ppd->lmc;
+	props->sm_lid = ibp->sm_lid;
+	props->sm_sl = ibp->sm_sl;
+	props->state = dd->f_iblink_state(ppd->lastibcstat);
+	props->phys_state = dd->f_ibphys_portstate(ppd->lastibcstat);
+	props->port_cap_flags = ibp->port_cap_flags;
+	props->gid_tbl_len = QIB_GUIDS_PER_PORT;
+	props->max_msg_sz = 0x80000000;
+	props->pkey_tbl_len = qib_get_npkeys(dd);
+	props->bad_pkey_cntr = ibp->pkey_violations;
+	props->qkey_viol_cntr = ibp->qkey_violations;
+	props->active_width = ppd->link_width_active;
+	/* See rate_show() */
+	props->active_speed = ppd->link_speed_active;
+	props->max_vl_num = qib_num_vls(ppd->vls_supported);
+	props->init_type_reply = 0;
+
+	props->max_mtu = qib_ibmtu ? qib_ibmtu : IB_MTU_4096;
+	switch (ppd->ibmtu) {
+	case 4096:
+		mtu = IB_MTU_4096;
+		break;
+	case 2048:
+		mtu = IB_MTU_2048;
+		break;
+	case 1024:
+		mtu = IB_MTU_1024;
+		break;
+	case 512:
+		mtu = IB_MTU_512;
+		break;
+	case 256:
+		mtu = IB_MTU_256;
+		break;
+	default:
+		mtu = IB_MTU_2048;
+	}
+	props->active_mtu = mtu;
+	props->subnet_timeout = ibp->subnet_timeout;
+
+	return 0;
+}
+
+static int qib_modify_device(struct ib_device *device,
+			     int device_modify_mask,
+			     struct ib_device_modify *device_modify)
+{
+	struct qib_devdata *dd = dd_from_ibdev(device);
+	unsigned i;
+	int ret;
+
+	if (device_modify_mask & ~(IB_DEVICE_MODIFY_SYS_IMAGE_GUID |
+				   IB_DEVICE_MODIFY_NODE_DESC)) {
+		ret = -EOPNOTSUPP;
+		goto bail;
+	}
+
+	if (device_modify_mask & IB_DEVICE_MODIFY_NODE_DESC) {
+		memcpy(device->node_desc, device_modify->node_desc, 64);
+		for (i = 0; i < dd->num_pports; i++) {
+			struct qib_ibport *ibp = &dd->pport[i].ibport_data;
+
+			qib_node_desc_chg(ibp);
+		}
+	}
+
+	if (device_modify_mask & IB_DEVICE_MODIFY_SYS_IMAGE_GUID) {
+		ib_qib_sys_image_guid =
+			cpu_to_be64(device_modify->sys_image_guid);
+		for (i = 0; i < dd->num_pports; i++) {
+			struct qib_ibport *ibp = &dd->pport[i].ibport_data;
+
+			qib_sys_guid_chg(ibp);
+		}
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+static int qib_modify_port(struct ib_device *ibdev, u8 port,
+			   int port_modify_mask, struct ib_port_modify *props)
+{
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+
+	ibp->port_cap_flags |= props->set_port_cap_mask;
+	ibp->port_cap_flags &= ~props->clr_port_cap_mask;
+	if (props->set_port_cap_mask || props->clr_port_cap_mask)
+		qib_cap_mask_chg(ibp);
+	if (port_modify_mask & IB_PORT_SHUTDOWN)
+		qib_set_linkstate(ppd, QIB_IB_LINKDOWN);
+	if (port_modify_mask & IB_PORT_RESET_QKEY_CNTR)
+		ibp->qkey_violations = 0;
+	return 0;
+}
+
+static int qib_query_gid(struct ib_device *ibdev, u8 port,
+			 int index, union ib_gid *gid)
+{
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	int ret = 0;
+
+	gid->global.subnet_prefix = ibp->gid_prefix;
+
+	if (index == 0)
+		gid->global.interface_id = ppd->guid;
+	else if (index < QIB_GUIDS_PER_PORT)
+		gid->global.interface_id = ibp->guids[index - 1];
+	else
+		ret = -EINVAL;
+
+	return ret;
+}
+
+static struct ib_pd *qib_alloc_pd(struct ib_device *ibdev,
+				  struct ib_ucontext *context,
+				  struct ib_udata *udata)
+{
+	struct qib_ibdev *dev = to_idev(ibdev);
+	struct qib_pd *pd;
+	struct ib_pd *ret;
+
+	/*
+	 * This is actually totally arbitrary.  Some correctness tests
+	 * assume there's a maximum number of PDs that can be allocated.
+	 * We don't actually have this limit, but we fail the test if
+	 * we allow allocations of more than we report for this value.
+	 */
+
+	pd = kmalloc(sizeof *pd, GFP_KERNEL);
+	if (!pd) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	spin_lock(&dev->n_pds_lock);
+	if (dev->n_pds_allocated == ib_qib_max_pds) {
+		spin_unlock(&dev->n_pds_lock);
+		kfree(pd);
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	dev->n_pds_allocated++;
+	spin_unlock(&dev->n_pds_lock);
+
+	/* ib_alloc_pd() will initialize pd->ibpd. */
+	pd->user = udata != NULL;
+
+	ret = &pd->ibpd;
+
+bail:
+	return ret;
+}
+
+static int qib_dealloc_pd(struct ib_pd *ibpd)
+{
+	struct qib_pd *pd = to_ipd(ibpd);
+	struct qib_ibdev *dev = to_idev(ibpd->device);
+
+	spin_lock(&dev->n_pds_lock);
+	dev->n_pds_allocated--;
+	spin_unlock(&dev->n_pds_lock);
+
+	kfree(pd);
+
+	return 0;
+}
+
+int qib_check_ah(struct ib_device *ibdev, struct ib_ah_attr *ah_attr)
+{
+	if ((ah_attr->ah_flags & IB_AH_GRH) &&
+	    ah_attr->grh.sgid_index >= QIB_GUIDS_PER_PORT)
+		goto bail;
+	if (ah_attr->dlid == 0)
+		goto bail;
+	if (ah_attr->port_num < 1 ||
+	    ah_attr->port_num > ibdev->phys_port_cnt)
+		goto bail;
+	if (ah_attr->static_rate != IB_RATE_PORT_CURRENT &&
+	    ib_rate_to_mult(ah_attr->static_rate) < 0)
+		goto bail;
+	if (ah_attr->sl > 15)
+		goto bail;
+	return 0;
+bail:
+	return -EINVAL;
+}
+
+/**
+ * qib_create_ah - create an address handle
+ * @pd: the protection domain
+ * @ah_attr: the attributes of the AH
+ *
+ * This may be called from interrupt context.
+ */
+static struct ib_ah *qib_create_ah(struct ib_pd *pd,
+				   struct ib_ah_attr *ah_attr)
+{
+	struct qib_ah *ah;
+	struct ib_ah *ret;
+	struct qib_ibdev *dev = to_idev(pd->device);
+
+	/* A multicast address requires a GRH (see ch. 8.4.1). */
+	if (ah_attr->dlid >= QIB_MULTICAST_LID_BASE &&
+	    ah_attr->dlid != QIB_PERMISSIVE_LID &&
+	    !(ah_attr->ah_flags & IB_AH_GRH)) {
+		ret = ERR_PTR(-EINVAL);
+		goto bail;
+	}
+	if (qib_check_ah(pd->device, ah_attr)) {
+		ret = ERR_PTR(-EINVAL);
+		goto bail;
+	}
+
+	ah = kmalloc(sizeof *ah, GFP_ATOMIC);
+	if (!ah) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	spin_lock(&dev->n_ahs_lock);
+	if (dev->n_ahs_allocated == ib_qib_max_ahs) {
+		spin_unlock(&dev->n_ahs_lock);
+		kfree(ah);
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	dev->n_ahs_allocated++;
+	spin_unlock(&dev->n_ahs_lock);
+
+	/* ib_create_ah() will initialize ah->ibah. */
+	ah->attr = *ah_attr;
+	atomic_set(&ah->refcount, 0);
+
+	ret = &ah->ibah;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_destroy_ah - destroy an address handle
+ * @ibah: the AH to destroy
+ *
+ * This may be called from interrupt context.
+ */
+static int qib_destroy_ah(struct ib_ah *ibah)
+{
+	struct qib_ibdev *dev = to_idev(ibah->device);
+	struct qib_ah *ah = to_iah(ibah);
+
+	if (atomic_read(&ah->refcount) != 0) {
+		qib_dbg("AH busy (cnt %u)\n", atomic_read(&ah->refcount));
+		return -EBUSY;
+	}
+
+	spin_lock(&dev->n_ahs_lock);
+	dev->n_ahs_allocated--;
+	spin_unlock(&dev->n_ahs_lock);
+
+	kfree(ah);
+
+	return 0;
+}
+
+static int qib_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr)
+{
+	struct qib_ah *ah = to_iah(ibah);
+
+	*ah_attr = ah->attr;
+
+	return 0;
+}
+
+/**
+ * qib_get_npkeys - return the size of the PKEY table for context 0
+ * @dd: the qlogic_ib device
+ */
+unsigned qib_get_npkeys(struct qib_devdata *dd)
+{
+	return ARRAY_SIZE(dd->rcd[0]->pkeys);
+}
+
+/*
+ * Return the indexed PKEY from the port PKEY table.
+ * No need to validate rcd[ctxt]; the port is setup if we are here.
+ */
+unsigned qib_get_pkey(struct qib_ibport *ibp, unsigned index)
+{
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct qib_devdata *dd = ppd->dd;
+	unsigned ctxt = ppd->hw_pidx;
+	unsigned ret;
+
+	/* dd->rcd null if mini_init or some init failures */
+	if (!dd->rcd || index >= ARRAY_SIZE(dd->rcd[ctxt]->pkeys))
+		ret = 0;
+	else
+		ret = dd->rcd[ctxt]->pkeys[index];
+
+	return ret;
+}
+
+static int qib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+			  u16 *pkey)
+{
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	int ret;
+
+	if (index >= qib_get_npkeys(dd)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	*pkey = qib_get_pkey(to_iport(ibdev, port), index);
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_alloc_ucontext - allocate a ucontest
+ * @ibdev: the infiniband device
+ * @udata: not used by the QLogic_IB driver
+ */
+
+static struct ib_ucontext *qib_alloc_ucontext(struct ib_device *ibdev,
+					      struct ib_udata *udata)
+{
+	struct qib_ucontext *context;
+	struct ib_ucontext *ret;
+
+	context = kmalloc(sizeof *context, GFP_KERNEL);
+	if (!context) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	ret = &context->ibucontext;
+
+bail:
+	return ret;
+}
+
+static int qib_dealloc_ucontext(struct ib_ucontext *context)
+{
+	kfree(to_iucontext(context));
+	return 0;
+}
+
+static void init_ibport(struct qib_pportdata *ppd)
+{
+	struct qib_verbs_counters cntrs;
+	struct qib_ibport *ibp = &ppd->ibport_data;
+
+	spin_lock_init(&ibp->lock);
+	/* Set the prefix to the default value (see ch. 4.1.1) */
+	ibp->gid_prefix = IB_DEFAULT_GID_PREFIX;
+	ibp->sm_lid = be16_to_cpu(IB_LID_PERMISSIVE);
+	ibp->port_cap_flags = IB_PORT_SYS_IMAGE_GUID_SUP |
+		IB_PORT_CLIENT_REG_SUP | IB_PORT_SL_MAP_SUP |
+		IB_PORT_TRAP_SUP | IB_PORT_AUTO_MIGR_SUP |
+		IB_PORT_DR_NOTICE_SUP | IB_PORT_CAP_MASK_NOTICE_SUP |
+		IB_PORT_OTHER_LOCAL_CHANGES_SUP;
+	if (ppd->dd->flags & QIB_HAS_LINK_LATENCY)
+		ibp->port_cap_flags |= IB_PORT_LINK_LATENCY_SUP;
+	ibp->pma_counter_select[0] = IB_PMA_PORT_XMIT_DATA;
+	ibp->pma_counter_select[1] = IB_PMA_PORT_RCV_DATA;
+	ibp->pma_counter_select[2] = IB_PMA_PORT_XMIT_PKTS;
+	ibp->pma_counter_select[3] = IB_PMA_PORT_RCV_PKTS;
+	ibp->pma_counter_select[4] = IB_PMA_PORT_XMIT_WAIT;
+
+	/* Snapshot current HW counters to "clear" them. */
+	qib_get_counters(ppd, &cntrs);
+	ibp->z_symbol_error_counter = cntrs.symbol_error_counter;
+	ibp->z_link_error_recovery_counter =
+		cntrs.link_error_recovery_counter;
+	ibp->z_link_downed_counter = cntrs.link_downed_counter;
+	ibp->z_port_rcv_errors = cntrs.port_rcv_errors;
+	ibp->z_port_rcv_remphys_errors = cntrs.port_rcv_remphys_errors;
+	ibp->z_port_xmit_discards = cntrs.port_xmit_discards;
+	ibp->z_port_xmit_data = cntrs.port_xmit_data;
+	ibp->z_port_rcv_data = cntrs.port_rcv_data;
+	ibp->z_port_xmit_packets = cntrs.port_xmit_packets;
+	ibp->z_port_rcv_packets = cntrs.port_rcv_packets;
+	ibp->z_local_link_integrity_errors =
+		cntrs.local_link_integrity_errors;
+	ibp->z_excessive_buffer_overrun_errors =
+		cntrs.excessive_buffer_overrun_errors;
+	ibp->z_vl15_dropped = cntrs.vl15_dropped;
+}
+
+/**
+ * qib_register_ib_device - register our device with the infiniband core
+ * @dd: the device data structure
+ * Return the allocated qib_ibdev pointer or NULL on error.
+ */
+int qib_register_ib_device(struct qib_devdata *dd)
+{
+	struct qib_ibdev *dev = &dd->verbs_dev;
+	struct ib_device *ibdev = &dev->ibdev;
+	struct qib_pportdata *ppd = dd->pport;
+	unsigned i, lk_tab_size;
+	int ret;
+
+	dev->qp_table_size = ib_qib_qp_table_size;
+	dev->qp_table = kzalloc(dev->qp_table_size * sizeof *dev->qp_table,
+				GFP_KERNEL);
+	if (!dev->qp_table) {
+		ret = -ENOMEM;
+		goto err_qpt;
+	}
+
+	for (i = 0; i < dd->num_pports; i++)
+		init_ibport(ppd + i);
+
+	/* Only need to initialize non-zero fields. */
+	spin_lock_init(&dev->qpt_lock);
+	spin_lock_init(&dev->n_pds_lock);
+	spin_lock_init(&dev->n_ahs_lock);
+	spin_lock_init(&dev->n_cqs_lock);
+	spin_lock_init(&dev->n_qps_lock);
+	spin_lock_init(&dev->n_srqs_lock);
+	spin_lock_init(&dev->n_mcast_grps_lock);
+	init_timer(&dev->mem_timer);
+	dev->mem_timer.function = mem_timer;
+	dev->mem_timer.data = (unsigned long) dev;
+
+	qib_init_qpn_table(dd, &dev->qpn_table);
+
+	/*
+	 * The top ib_qib_lkey_table_size bits are used to index the
+	 * table.  The lower 8 bits can be owned by the user (copied from
+	 * the LKEY).  The remaining bits act as a generation number or tag.
+	 */
+	spin_lock_init(&dev->lk_table.lock);
+	dev->lk_table.max = 1 << ib_qib_lkey_table_size;
+	lk_tab_size = dev->lk_table.max * sizeof(*dev->lk_table.table);
+	dev->lk_table.table = (struct qib_mregion **)
+		__get_free_pages(GFP_KERNEL, get_order(lk_tab_size));
+	if (dev->lk_table.table == NULL) {
+		ret = -ENOMEM;
+		goto err_lk;
+	}
+	memset(dev->lk_table.table, 0, lk_tab_size);
+	INIT_LIST_HEAD(&dev->pending_mmaps);
+	spin_lock_init(&dev->pending_lock);
+	dev->mmap_offset = PAGE_SIZE;
+	spin_lock_init(&dev->mmap_offset_lock);
+	INIT_LIST_HEAD(&dev->piowait);
+	INIT_LIST_HEAD(&dev->dmawait);
+	INIT_LIST_HEAD(&dev->txwait);
+	INIT_LIST_HEAD(&dev->memwait);
+	INIT_LIST_HEAD(&dev->rnrwait);
+	INIT_LIST_HEAD(&dev->txreq_free);
+
+	dev->pio_hdrs = dma_alloc_coherent(&dd->pcidev->dev,
+					   ppd->sdma_descq_cnt *
+					   sizeof(struct qib_pio_header),
+					   &dev->pio_hdrs_phys,
+					   GFP_KERNEL);
+	if (!dev->pio_hdrs) {
+		ret = -ENOMEM;
+		goto err_hdrs;
+	}
+
+	for (i = 0; i < ppd->sdma_descq_cnt; i++) {
+		struct qib_verbs_txreq *tx;
+
+		tx = kzalloc(sizeof *tx, GFP_KERNEL);
+		if (!tx) {
+			ret = -ENOMEM;
+			goto err_tx;
+		}
+		tx->hdr_inx = i;
+		list_add(&tx->txreq.list, &dev->txreq_free);
+	}
+
+	/*
+	 * The system image GUID is supposed to be the same for all
+	 * IB HCAs in a single system but since there can be other
+	 * device types in the system, we can't be sure this is unique.
+	 */
+	if (!ib_qib_sys_image_guid)
+		ib_qib_sys_image_guid = ppd->guid;
+
+	strlcpy(ibdev->name, "qib%d", IB_DEVICE_NAME_MAX);
+	ibdev->owner = THIS_MODULE;
+	ibdev->node_guid = ppd->guid;
+	ibdev->uverbs_abi_ver = QIB_UVERBS_ABI_VERSION;
+	ibdev->uverbs_cmd_mask =
+		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
+		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
+		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
+		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
+		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
+		(1ull << IB_USER_VERBS_CMD_CREATE_AH)           |
+		(1ull << IB_USER_VERBS_CMD_DESTROY_AH)          |
+		(1ull << IB_USER_VERBS_CMD_QUERY_AH)            |
+		(1ull << IB_USER_VERBS_CMD_REG_MR)              |
+		(1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
+		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
+		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
+		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
+		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
+		(1ull << IB_USER_VERBS_CMD_POLL_CQ)             |
+		(1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ)       |
+		(1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
+		(1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
+		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
+		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
+		(1ull << IB_USER_VERBS_CMD_POST_SEND)           |
+		(1ull << IB_USER_VERBS_CMD_POST_RECV)           |
+		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
+		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
+		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
+		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
+		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
+		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
+		(1ull << IB_USER_VERBS_CMD_POST_SRQ_RECV);
+	ibdev->node_type = RDMA_NODE_IB_CA;
+	ibdev->phys_port_cnt = dd->num_pports;
+	ibdev->num_comp_vectors = 1;
+	ibdev->dma_device = &dd->pcidev->dev;
+	ibdev->query_device = qib_query_device;
+	ibdev->modify_device = qib_modify_device;
+	ibdev->query_port = qib_query_port;
+	ibdev->modify_port = qib_modify_port;
+	ibdev->query_pkey = qib_query_pkey;
+	ibdev->query_gid = qib_query_gid;
+	ibdev->alloc_ucontext = qib_alloc_ucontext;
+	ibdev->dealloc_ucontext = qib_dealloc_ucontext;
+	ibdev->alloc_pd = qib_alloc_pd;
+	ibdev->dealloc_pd = qib_dealloc_pd;
+	ibdev->create_ah = qib_create_ah;
+	ibdev->destroy_ah = qib_destroy_ah;
+	ibdev->query_ah = qib_query_ah;
+	ibdev->create_srq = qib_create_srq;
+	ibdev->modify_srq = qib_modify_srq;
+	ibdev->query_srq = qib_query_srq;
+	ibdev->destroy_srq = qib_destroy_srq;
+	ibdev->create_qp = qib_create_qp;
+	ibdev->modify_qp = qib_modify_qp;
+	ibdev->query_qp = qib_query_qp;
+	ibdev->destroy_qp = qib_destroy_qp;
+	ibdev->post_send = qib_post_send;
+	ibdev->post_recv = qib_post_receive;
+	ibdev->post_srq_recv = qib_post_srq_receive;
+	ibdev->create_cq = qib_create_cq;
+	ibdev->destroy_cq = qib_destroy_cq;
+	ibdev->resize_cq = qib_resize_cq;
+	ibdev->poll_cq = qib_poll_cq;
+	ibdev->req_notify_cq = qib_req_notify_cq;
+	ibdev->get_dma_mr = qib_get_dma_mr;
+	ibdev->reg_phys_mr = qib_reg_phys_mr;
+	ibdev->reg_user_mr = qib_reg_user_mr;
+	ibdev->dereg_mr = qib_dereg_mr;
+	ibdev->alloc_fmr = qib_alloc_fmr;
+	ibdev->map_phys_fmr = qib_map_phys_fmr;
+	ibdev->unmap_fmr = qib_unmap_fmr;
+	ibdev->dealloc_fmr = qib_dealloc_fmr;
+	ibdev->attach_mcast = qib_multicast_attach;
+	ibdev->detach_mcast = qib_multicast_detach;
+	ibdev->process_mad = qib_process_mad;
+	ibdev->mmap = qib_mmap;
+	ibdev->dma_ops = &qib_dma_mapping_ops;
+
+	snprintf(ibdev->node_desc, sizeof(ibdev->node_desc),
+		 QIB_IDSTR " %s", init_utsname()->nodename);
+
+	ret = ib_register_device(ibdev);
+	if (ret)
+		goto err_reg;
+
+	ret = qib_create_agents(dev);
+	if (ret)
+		goto err_agents;
+
+	if (qib_verbs_register_sysfs(dd))
+		goto err_class;
+
+	goto bail;
+
+err_class:
+	qib_free_agents(dev);
+err_agents:
+	ib_unregister_device(ibdev);
+err_reg:
+err_tx:
+	while (!list_empty(&dev->txreq_free)) {
+		struct list_head *l = dev->txreq_free.next;
+		struct qib_verbs_txreq *tx;
+
+		list_del(l);
+		tx = list_entry(l, struct qib_verbs_txreq, txreq.list);
+		kfree(tx);
+	}
+	dma_free_coherent(&dd->pcidev->dev,
+			  ppd->sdma_descq_cnt * sizeof(struct qib_pio_header),
+			  dev->pio_hdrs, dev->pio_hdrs_phys);
+err_hdrs:
+	free_pages((unsigned long) dev->lk_table.table, get_order(lk_tab_size));
+err_lk:
+	kfree(dev->qp_table);
+err_qpt:
+	qib_dev_err(dd, "cannot register verbs: %d!\n", -ret);
+bail:
+	return ret;
+}
+
+void qib_unregister_ib_device(struct qib_devdata *dd)
+{
+	struct qib_ibdev *dev = &dd->verbs_dev;
+	struct ib_device *ibdev = &dev->ibdev;
+	u32 qps_inuse;
+	unsigned lk_tab_size;
+
+	qib_verbs_unregister_sysfs(dd);
+
+	qib_free_agents(dev);
+
+	ib_unregister_device(ibdev);
+
+	if (!list_empty(&dev->piowait))
+		qib_dev_err(dd, "piowait list not empty!\n");
+	if (!list_empty(&dev->dmawait))
+		qib_dev_err(dd, "dmawait list not empty!\n");
+	if (!list_empty(&dev->txwait))
+		qib_dev_err(dd, "txwait list not empty!\n");
+	if (!list_empty(&dev->memwait))
+		qib_dev_err(dd, "memwait list not empty!\n");
+	if (!list_empty(&dev->rnrwait))
+		qib_dev_err(dd, "rnrwait list not empty!\n");
+	if (dev->dma_mr)
+		qib_dev_err(dd, "DMA MR not NULL!\n");
+
+	qps_inuse = qib_free_all_qps(dd);
+	if (qps_inuse)
+		qib_dev_err(dd, "QP memory leak! %u still in use\n",
+			    qps_inuse);
+
+	del_timer_sync(&dev->mem_timer);
+	qib_free_qpn_table(&dev->qpn_table);
+	while (!list_empty(&dev->txreq_free)) {
+		struct list_head *l = dev->txreq_free.next;
+		struct qib_verbs_txreq *tx;
+
+		list_del(l);
+		tx = list_entry(l, struct qib_verbs_txreq, txreq.list);
+		kfree(tx);
+	}
+	dma_free_coherent(&dd->pcidev->dev,
+			  dd->pport->sdma_descq_cnt *
+				  sizeof(struct qib_pio_header),
+			  dev->pio_hdrs, dev->pio_hdrs_phys);
+	lk_tab_size = dev->lk_table.max * sizeof(*dev->lk_table.table);
+	free_pages((unsigned long) dev->lk_table.table,
+		   get_order(lk_tab_size));
+	kfree(dev->qp_table);
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 47/51] IB/qib: Add qib_verbs.h

[Ralph Campbell]

creates the qib_verbs.h file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_verbs.h | 1086 +++++++++++++++++++++++++++++++++
 1 files changed, 1086 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_verbs.h

diff --git a/drivers/infiniband/hw/qib/qib_verbs.h b/drivers/infiniband/hw/qib/qib_verbs.h
new file mode 100644
index 0000000..20c8ec1
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_verbs.h
@@ -0,0 +1,1086 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef QIB_VERBS_H
+#define QIB_VERBS_H
+
+#include <linux/types.h>
+#include <linux/spinlock.h>
+#include <linux/kernel.h>
+#include <linux/interrupt.h>
+#include <linux/kref.h>
+#include <linux/workqueue.h>
+#include <rdma/ib_pack.h>
+#include <rdma/ib_user_verbs.h>
+
+struct qib_ctxtdata;
+struct qib_pportdata;
+struct qib_devdata;
+struct qib_verbs_txreq;
+
+#define QIB_MAX_RDMA_ATOMIC     16
+#define QIB_GUIDS_PER_PORT	5
+
+#define QPN_MAX                 (1 << 24)
+#define QPNMAP_ENTRIES          (QPN_MAX / PAGE_SIZE / BITS_PER_BYTE)
+
+/*
+ * Increment this value if any changes that break userspace ABI
+ * compatibility are made.
+ */
+#define QIB_UVERBS_ABI_VERSION       2
+
+/*
+ * Define an ib_cq_notify value that is not valid so we know when CQ
+ * notifications are armed.
+ */
+#define IB_CQ_NONE      (IB_CQ_NEXT_COMP + 1)
+
+#define IB_SEQ_NAK	(3 << 29)
+
+/* AETH NAK opcode values */
+#define IB_RNR_NAK                      0x20
+#define IB_NAK_PSN_ERROR                0x60
+#define IB_NAK_INVALID_REQUEST          0x61
+#define IB_NAK_REMOTE_ACCESS_ERROR      0x62
+#define IB_NAK_REMOTE_OPERATIONAL_ERROR 0x63
+#define IB_NAK_INVALID_RD_REQUEST       0x64
+
+/* Flags for checking QP state (see ib_qib_state_ops[]) */
+#define QIB_POST_SEND_OK                0x01
+#define QIB_POST_RECV_OK                0x02
+#define QIB_PROCESS_RECV_OK             0x04
+#define QIB_PROCESS_SEND_OK             0x08
+#define QIB_PROCESS_NEXT_SEND_OK        0x10
+#define QIB_FLUSH_SEND			0x20
+#define QIB_FLUSH_RECV			0x40
+#define QIB_PROCESS_OR_FLUSH_SEND \
+	(QIB_PROCESS_SEND_OK | QIB_FLUSH_SEND)
+
+/* IB Performance Manager status values */
+#define IB_PMA_SAMPLE_STATUS_DONE       0x00
+#define IB_PMA_SAMPLE_STATUS_STARTED    0x01
+#define IB_PMA_SAMPLE_STATUS_RUNNING    0x02
+
+/* Mandatory IB performance counter select values. */
+#define IB_PMA_PORT_XMIT_DATA   cpu_to_be16(0x0001)
+#define IB_PMA_PORT_RCV_DATA    cpu_to_be16(0x0002)
+#define IB_PMA_PORT_XMIT_PKTS   cpu_to_be16(0x0003)
+#define IB_PMA_PORT_RCV_PKTS    cpu_to_be16(0x0004)
+#define IB_PMA_PORT_XMIT_WAIT   cpu_to_be16(0x0005)
+
+#define QIB_VENDOR_IPG		cpu_to_be16(0xFFA0)
+
+#define IB_BTH_REQ_ACK		(1 << 31)
+#define IB_BTH_SOLICITED	(1 << 23)
+#define IB_BTH_MIG_REQ		(1 << 22)
+
+/* XXX Should be defined in ib_verbs.h enum ib_port_cap_flags */
+#define IB_PORT_OTHER_LOCAL_CHANGES_SUP (1 << 26)
+
+#define IB_GRH_VERSION		6
+#define IB_GRH_VERSION_MASK	0xF
+#define IB_GRH_VERSION_SHIFT	28
+#define IB_GRH_TCLASS_MASK	0xFF
+#define IB_GRH_TCLASS_SHIFT	20
+#define IB_GRH_FLOW_MASK	0xFFFFF
+#define IB_GRH_FLOW_SHIFT	0
+#define IB_GRH_NEXT_HDR		0x1B
+
+#define IB_DEFAULT_GID_PREFIX	cpu_to_be64(0xfe80000000000000ULL)
+
+/* Values for set/get portinfo VLCap OperationalVLs */
+#define IB_VL_VL0       1
+#define IB_VL_VL0_1     2
+#define IB_VL_VL0_3     3
+#define IB_VL_VL0_7     4
+#define IB_VL_VL0_14    5
+
+static inline int qib_num_vls(int vls)
+{
+	switch (vls) {
+	default:
+	case IB_VL_VL0:
+		return 1;
+	case IB_VL_VL0_1:
+		return 2;
+	case IB_VL_VL0_3:
+		return 4;
+	case IB_VL_VL0_7:
+		return 8;
+	case IB_VL_VL0_14:
+		return 15;
+	}
+}
+
+struct ib_reth {
+	__be64 vaddr;
+	__be32 rkey;
+	__be32 length;
+} __attribute__ ((packed));
+
+struct ib_atomic_eth {
+	__be32 vaddr[2];        /* unaligned so access as 2 32-bit words */
+	__be32 rkey;
+	__be64 swap_data;
+	__be64 compare_data;
+} __attribute__ ((packed));
+
+struct qib_other_headers {
+	__be32 bth[3];
+	union {
+		struct {
+			__be32 deth[2];
+			__be32 imm_data;
+		} ud;
+		struct {
+			struct ib_reth reth;
+			__be32 imm_data;
+		} rc;
+		struct {
+			__be32 aeth;
+			__be32 atomic_ack_eth[2];
+		} at;
+		__be32 imm_data;
+		__be32 aeth;
+		struct ib_atomic_eth atomic_eth;
+	} u;
+} __attribute__ ((packed));
+
+/*
+ * Note that UD packets with a GRH header are 8+40+12+8 = 68 bytes
+ * long (72 w/ imm_data).  Only the first 56 bytes of the IB header
+ * will be in the eager header buffer.  The remaining 12 or 16 bytes
+ * are in the data buffer.
+ */
+struct qib_ib_header {
+	__be16 lrh[4];
+	union {
+		struct {
+			struct ib_grh grh;
+			struct qib_other_headers oth;
+		} l;
+		struct qib_other_headers oth;
+	} u;
+} __attribute__ ((packed));
+
+struct qib_pio_header {
+	__le32 pbc[2];
+	struct qib_ib_header hdr;
+} __attribute__ ((packed));
+
+/*
+ * There is one struct qib_mcast for each multicast GID.
+ * All attached QPs are then stored as a list of
+ * struct qib_mcast_qp.
+ */
+struct qib_mcast_qp {
+	struct list_head list;
+	struct qib_qp *qp;
+};
+
+struct qib_mcast {
+	struct rb_node rb_node;
+	union ib_gid mgid;
+	struct list_head qp_list;
+	wait_queue_head_t wait;
+	atomic_t refcount;
+	int n_attached;
+};
+
+/* Protection domain */
+struct qib_pd {
+	struct ib_pd ibpd;
+	int user;               /* non-zero if created from user space */
+};
+
+/* Address Handle */
+struct qib_ah {
+	struct ib_ah ibah;
+	struct ib_ah_attr attr;
+	atomic_t refcount;
+};
+
+/*
+ * This structure is used by qib_mmap() to validate an offset
+ * when an mmap() request is made.  The vm_area_struct then uses
+ * this as its vm_private_data.
+ */
+struct qib_mmap_info {
+	struct list_head pending_mmaps;
+	struct ib_ucontext *context;
+	void *obj;
+	__u64 offset;
+	struct kref ref;
+	unsigned size;
+};
+
+/*
+ * This structure is used to contain the head pointer, tail pointer,
+ * and completion queue entries as a single memory allocation so
+ * it can be mmap'ed into user space.
+ */
+struct qib_cq_wc {
+	u32 head;               /* index of next entry to fill */
+	u32 tail;               /* index of next ib_poll_cq() entry */
+	union {
+		/* these are actually size ibcq.cqe + 1 */
+		struct ib_uverbs_wc uqueue[0];
+		struct ib_wc kqueue[0];
+	};
+};
+
+/*
+ * The completion queue structure.
+ */
+struct qib_cq {
+	struct ib_cq ibcq;
+	struct tasklet_struct comptask;
+	spinlock_t lock; /* protect changes in this struct */
+	u8 notify;
+	u8 triggered;
+	struct qib_cq_wc *queue;
+	struct qib_mmap_info *ip;
+};
+
+/*
+ * A segment is a linear region of low physical memory.
+ * XXX Maybe we should use phys addr here and kmap()/kunmap().
+ * Used by the verbs layer.
+ */
+struct qib_seg {
+	void *vaddr;
+	size_t length;
+};
+
+/* The number of qib_segs that fit in a page. */
+#define QIB_SEGSZ     (PAGE_SIZE / sizeof(struct qib_seg))
+
+struct qib_segarray {
+	struct qib_seg segs[QIB_SEGSZ];
+};
+
+struct qib_mregion {
+	struct ib_pd *pd;       /* shares refcnt of ibmr.pd */
+	u64 user_base;          /* User's address for this region */
+	u64 iova;               /* IB start address of this region */
+	size_t length;
+	u32 lkey;
+	u32 offset;             /* offset (bytes) to start of region */
+	int access_flags;
+	u32 max_segs;           /* number of qib_segs in all the arrays */
+	u32 mapsz;              /* size of the map array */
+	atomic_t refcount;
+	struct qib_segarray *map[0];    /* the segments */
+};
+
+/*
+ * These keep track of the copy progress within a memory region.
+ * Used by the verbs layer.
+ */
+struct qib_sge {
+	struct qib_mregion *mr;
+	void *vaddr;            /* kernel virtual address of segment */
+	u32 sge_length;         /* length of the SGE */
+	u32 length;             /* remaining length of the segment */
+	u16 m;                  /* current index: mr->map[m] */
+	u16 n;                  /* current index: mr->map[m]->segs[n] */
+};
+
+/* Memory region */
+struct qib_mr {
+	struct ib_mr ibmr;
+	struct ib_umem *umem;
+	struct qib_mregion mr;  /* must be last */
+};
+
+/*
+ * Send work request queue entry.
+ * The size of the sg_list is determined when the QP is created and stored
+ * in qp->s_max_sge.
+ */
+struct qib_swqe {
+	struct ib_send_wr wr;   /* don't use wr.sg_list */
+	u32 psn;                /* first packet sequence number */
+	u32 lpsn;               /* last packet sequence number */
+	u32 ssn;                /* send sequence number */
+	u32 length;             /* total length of data in sg_list */
+	struct qib_sge sg_list[0];
+};
+
+/*
+ * Receive work request queue entry.
+ * The size of the sg_list is determined when the QP (or SRQ) is created
+ * and stored in qp->r_rq.max_sge (or srq->rq.max_sge).
+ */
+struct qib_rwqe {
+	u64 wr_id;
+	u8 num_sge;
+	struct ib_sge sg_list[0];
+};
+
+/*
+ * This structure is used to contain the head pointer, tail pointer,
+ * and receive work queue entries as a single memory allocation so
+ * it can be mmap'ed into user space.
+ * Note that the wq array elements are variable size so you can't
+ * just index into the array to get the N'th element;
+ * use get_rwqe_ptr() instead.
+ */
+struct qib_rwq {
+	u32 head;               /* new work requests posted to the head */
+	u32 tail;               /* receives pull requests from here. */
+	struct qib_rwqe wq[0];
+};
+
+struct qib_rq {
+	struct qib_rwq *wq;
+	spinlock_t lock; /* protect changes in this struct */
+	u32 size;               /* size of RWQE array */
+	u8 max_sge;
+};
+
+struct qib_srq {
+	struct ib_srq ibsrq;
+	struct qib_rq rq;
+	struct qib_mmap_info *ip;
+	/* send signal when number of RWQEs < limit */
+	u32 limit;
+};
+
+struct qib_sge_state {
+	struct qib_sge *sg_list;      /* next SGE to be used if any */
+	struct qib_sge sge;   /* progress state for the current SGE */
+	u32 total_len;
+	u8 num_sge;
+};
+
+/*
+ * This structure holds the information that the send tasklet needs
+ * to send a RDMA read response or atomic operation.
+ */
+struct qib_ack_entry {
+	u8 opcode;
+	u8 sent;
+	u32 psn;
+	u32 lpsn;
+	union {
+		struct qib_sge rdma_sge;
+		u64 atomic_data;
+	};
+};
+
+/*
+ * Variables prefixed with s_ are for the requester (sender).
+ * Variables prefixed with r_ are for the responder (receiver).
+ * Variables prefixed with ack_ are for responder replies.
+ *
+ * Common variables are protected by both r_rq.lock and s_lock in that order
+ * which only happens in modify_qp() or changing the QP 'state'.
+ */
+struct qib_qp {
+	struct ib_qp ibqp;
+	struct qib_qp *next;            /* link list for QPN hash table */
+	struct qib_qp *timer_next;      /* link list for qib_ib_timer() */
+	struct list_head iowait;        /* link for wait PIO buf */
+	struct list_head rspwait;       /* link for waititing to respond */
+	struct ib_ah_attr remote_ah_attr;
+	struct ib_ah_attr alt_ah_attr;
+	struct qib_ib_header s_hdr;     /* next packet header to send */
+	atomic_t refcount;
+	wait_queue_head_t wait;
+	wait_queue_head_t wait_dma;
+	struct timer_list s_timer;
+	struct work_struct s_work;
+	struct qib_mmap_info *ip;
+	struct qib_sge_state *s_cur_sge;
+	struct qib_verbs_txreq *s_tx;
+	struct qib_mregion *s_rdma_mr;
+	struct qib_sge_state s_sge;     /* current send request data */
+	struct qib_ack_entry s_ack_queue[QIB_MAX_RDMA_ATOMIC + 1];
+	struct qib_sge_state s_ack_rdma_sge;
+	struct qib_sge_state s_rdma_read_sge;
+	struct qib_sge_state r_sge;     /* current receive data */
+	spinlock_t r_lock;      /* used for APM */
+	spinlock_t s_lock;
+	atomic_t s_dma_busy;
+	unsigned processor_id;	/* Processor ID QP is bound to */
+	u32 s_flags;
+	u32 s_cur_size;         /* size of send packet in bytes */
+	u32 s_len;              /* total length of s_sge */
+	u32 s_rdma_read_len;    /* total length of s_rdma_read_sge */
+	u32 s_next_psn;         /* PSN for next request */
+	u32 s_last_psn;         /* last response PSN processed */
+	u32 s_sending_psn;      /* lowest PSN that is being sent */
+	u32 s_sending_hpsn;     /* highest PSN that is being sent */
+	u32 s_psn;              /* current packet sequence number */
+	u32 s_ack_rdma_psn;     /* PSN for sending RDMA read responses */
+	u32 s_ack_psn;          /* PSN for acking sends and RDMA writes */
+	u32 s_rnr_timeout;      /* number of milliseconds for RNR timeout */
+	u32 r_ack_psn;          /* PSN for next ACK or atomic ACK */
+	u64 r_wr_id;            /* ID for current receive WQE */
+	unsigned long r_aflags;
+	u32 r_len;              /* total length of r_sge */
+	u32 r_rcv_len;          /* receive data len processed */
+	u32 r_psn;              /* expected rcv packet sequence number */
+	u32 r_msn;              /* message sequence number */
+	u16 s_hdrwords;         /* size of s_hdr in 32 bit words */
+	u16 s_rdma_ack_cnt;
+	u8 state;               /* QP state */
+	u8 s_state;             /* opcode of last packet sent */
+	u8 s_ack_state;         /* opcode of packet to ACK */
+	u8 s_nak_state;         /* non-zero if NAK is pending */
+	u8 r_state;             /* opcode of last packet received */
+	u8 r_nak_state;         /* non-zero if NAK is pending */
+	u8 r_min_rnr_timer;     /* retry timeout value for RNR NAKs */
+	u8 r_flags;
+	u8 r_max_rd_atomic;     /* max number of RDMA read/atomic to receive */
+	u8 r_head_ack_queue;    /* index into s_ack_queue[] */
+	u8 qp_access_flags;
+	u8 s_max_sge;           /* size of s_wq->sg_list */
+	u8 s_retry_cnt;         /* number of times to retry */
+	u8 s_rnr_retry_cnt;
+	u8 s_retry;             /* requester retry counter */
+	u8 s_rnr_retry;         /* requester RNR retry counter */
+	u8 s_pkey_index;        /* PKEY index to use */
+	u8 s_alt_pkey_index;    /* Alternate path PKEY index to use */
+	u8 s_max_rd_atomic;     /* max number of RDMA read/atomic to send */
+	u8 s_num_rd_atomic;     /* number of RDMA read/atomic pending */
+	u8 s_tail_ack_queue;    /* index into s_ack_queue[] */
+	u8 s_srate;
+	u8 s_draining;
+	u8 s_mig_state;
+	u8 timeout;             /* Timeout for this QP */
+	u8 alt_timeout;         /* Alternate path timeout for this QP */
+	u8 port_num;
+	enum ib_mtu path_mtu;
+	u32 remote_qpn;
+	u32 qkey;               /* QKEY for this QP (for UD or RD) */
+	u32 s_size;             /* send work queue size */
+	u32 s_head;             /* new entries added here */
+	u32 s_tail;             /* next entry to process */
+	u32 s_cur;              /* current work queue entry */
+	u32 s_acked;            /* last un-ACK'ed entry */
+	u32 s_last;             /* last completed entry */
+	u32 s_ssn;              /* SSN of tail entry */
+	u32 s_lsn;              /* limit sequence number (credit) */
+	struct qib_swqe *s_wq;  /* send work queue */
+	struct qib_swqe *s_wqe;
+	struct qib_rq r_rq;             /* receive work queue */
+	struct qib_sge r_sg_list[0];    /* verified SGEs */
+};
+
+/*
+ * Atomic bit definitions for r_aflags.
+ */
+#define QIB_R_WRID_VALID        0
+
+/*
+ * Bit definitions for r_flags.
+ */
+#define QIB_R_REUSE_SGE 0x01
+#define QIB_R_RDMAR_SEQ 0x02
+#define QIB_R_RSP_NAK   0x04
+#define QIB_R_RSP_SEND  0x08
+#define QIB_R_COMM_EST  0x10
+
+/*
+ * Bit definitions for s_flags.
+ *
+ * QIB_S_SIGNAL_REQ_WR - set if QP send WRs contain completion signaled
+ * QIB_S_BUSY - send tasklet is processing the QP
+ * QIB_S_TIMER - the RC retry timer is active
+ * QIB_S_ACK_PENDING - an ACK is waiting to be sent after RDMA read/atomics
+ * QIB_S_WAIT_FENCE - waiting for all prior RDMA read or atomic SWQEs
+ *                         before processing the next SWQE
+ * QIB_S_WAIT_RDMAR - waiting for a RDMA read or atomic SWQE to complete
+ *                         before processing the next SWQE
+ * QIB_S_WAIT_RNR - waiting for RNR timeout
+ * QIB_S_WAIT_SSN_CREDIT - waiting for RC credits to process next SWQE
+ * QIB_S_WAIT_DMA - waiting for send DMA queue to drain before generating
+ *                  next send completion entry not via send DMA
+ * QIB_S_WAIT_PIO - waiting for a send buffer to be available
+ * QIB_S_WAIT_TX - waiting for a struct qib_verbs_txreq to be available
+ * QIB_S_WAIT_DMA_DESC - waiting for DMA descriptors to be available
+ * QIB_S_WAIT_KMEM - waiting for kernel memory to be available
+ * QIB_S_WAIT_PSN - waiting for a packet to exit the send DMA queue
+ * QIB_S_WAIT_ACK - waiting for an ACK packet before sending more requests
+ * QIB_S_SEND_ONE - send one packet, request ACK, then wait for ACK
+ */
+#define QIB_S_SIGNAL_REQ_WR	0x0001
+#define QIB_S_BUSY		0x0002
+#define QIB_S_TIMER		0x0004
+#define QIB_S_RESP_PENDING	0x0008
+#define QIB_S_ACK_PENDING	0x0010
+#define QIB_S_WAIT_FENCE	0x0020
+#define QIB_S_WAIT_RDMAR	0x0040
+#define QIB_S_WAIT_RNR		0x0080
+#define QIB_S_WAIT_SSN_CREDIT	0x0100
+#define QIB_S_WAIT_DMA		0x0200
+#define QIB_S_WAIT_PIO		0x0400
+#define QIB_S_WAIT_TX		0x0800
+#define QIB_S_WAIT_DMA_DESC	0x1000
+#define QIB_S_WAIT_KMEM		0x2000
+#define QIB_S_WAIT_PSN		0x4000
+#define QIB_S_WAIT_ACK		0x8000
+#define QIB_S_SEND_ONE		0x10000
+#define QIB_S_UNLIMITED_CREDIT	0x20000
+
+/*
+ * Wait flags that would prevent any packet type from being sent.
+ */
+#define QIB_S_ANY_WAIT_IO (QIB_S_WAIT_PIO | QIB_S_WAIT_TX | \
+	QIB_S_WAIT_DMA_DESC | QIB_S_WAIT_KMEM)
+
+/*
+ * Wait flags that would prevent send work requests from making progress.
+ */
+#define QIB_S_ANY_WAIT_SEND (QIB_S_WAIT_FENCE | QIB_S_WAIT_RDMAR | \
+	QIB_S_WAIT_RNR | QIB_S_WAIT_SSN_CREDIT | QIB_S_WAIT_DMA | \
+	QIB_S_WAIT_PSN | QIB_S_WAIT_ACK)
+
+#define QIB_S_ANY_WAIT (QIB_S_ANY_WAIT_IO | QIB_S_ANY_WAIT_SEND)
+
+#define QIB_PSN_CREDIT  16
+
+/*
+ * Since struct qib_swqe is not a fixed size, we can't simply index into
+ * struct qib_qp.s_wq.  This function does the array index computation.
+ */
+static inline struct qib_swqe *get_swqe_ptr(struct qib_qp *qp,
+					      unsigned n)
+{
+	return (struct qib_swqe *)((char *)qp->s_wq +
+				     (sizeof(struct qib_swqe) +
+				      qp->s_max_sge *
+				      sizeof(struct qib_sge)) * n);
+}
+
+/*
+ * Since struct qib_rwqe is not a fixed size, we can't simply index into
+ * struct qib_rwq.wq.  This function does the array index computation.
+ */
+static inline struct qib_rwqe *get_rwqe_ptr(struct qib_rq *rq, unsigned n)
+{
+	return (struct qib_rwqe *)
+		((char *) rq->wq->wq +
+		 (sizeof(struct qib_rwqe) +
+		  rq->max_sge * sizeof(struct ib_sge)) * n);
+}
+
+/*
+ * QPN-map pages start out as NULL, they get allocated upon
+ * first use and are never deallocated. This way,
+ * large bitmaps are not allocated unless large numbers of QPs are used.
+ */
+struct qpn_map {
+	void *page;
+};
+
+struct qib_qpn_table {
+	spinlock_t lock; /* protect changes in this struct */
+	unsigned flags;         /* flags for QP0/1 allocated for each port */
+	u32 last;               /* last QP number allocated */
+	u32 nmaps;              /* size of the map table */
+	u16 limit;
+	u16 mask;
+	/* bit map of free QP numbers other than 0/1 */
+	struct qpn_map map[QPNMAP_ENTRIES];
+};
+
+struct qib_lkey_table {
+	spinlock_t lock; /* protect changes in this struct */
+	u32 next;               /* next unused index (speeds search) */
+	u32 gen;                /* generation count */
+	u32 max;                /* size of the table */
+	struct qib_mregion **table;
+};
+
+struct qib_opcode_stats {
+	u64 n_packets;          /* number of packets */
+	u64 n_bytes;            /* total number of bytes */
+};
+
+struct qib_ibport {
+	struct qib_qp *qp0;
+	struct qib_qp *qp1;
+	struct ib_mad_agent *send_agent;	/* agent for SMI (traps) */
+	struct qib_ah *sm_ah;
+	struct qib_ah *smi_ah;
+	struct rb_root mcast_tree;
+	spinlock_t lock;		/* protect changes in this struct */
+
+	/* non-zero when timer is set */
+	unsigned long mkey_lease_timeout;
+	unsigned long trap_timeout;
+	__be64 gid_prefix;      /* in network order */
+	__be64 mkey;
+	__be64 guids[QIB_GUIDS_PER_PORT	- 1];	/* writable GUIDs */
+	u64 tid;		/* TID for traps */
+	u64 n_unicast_xmit;     /* total unicast packets sent */
+	u64 n_unicast_rcv;      /* total unicast packets received */
+	u64 n_multicast_xmit;   /* total multicast packets sent */
+	u64 n_multicast_rcv;    /* total multicast packets received */
+	u64 z_symbol_error_counter;             /* starting count for PMA */
+	u64 z_link_error_recovery_counter;      /* starting count for PMA */
+	u64 z_link_downed_counter;              /* starting count for PMA */
+	u64 z_port_rcv_errors;                  /* starting count for PMA */
+	u64 z_port_rcv_remphys_errors;          /* starting count for PMA */
+	u64 z_port_xmit_discards;               /* starting count for PMA */
+	u64 z_port_xmit_data;                   /* starting count for PMA */
+	u64 z_port_rcv_data;                    /* starting count for PMA */
+	u64 z_port_xmit_packets;                /* starting count for PMA */
+	u64 z_port_rcv_packets;                 /* starting count for PMA */
+	u32 z_local_link_integrity_errors;      /* starting count for PMA */
+	u32 z_excessive_buffer_overrun_errors;  /* starting count for PMA */
+	u32 z_vl15_dropped;                     /* starting count for PMA */
+	u32 n_rc_resends;
+	u32 n_rc_acks;
+	u32 n_rc_qacks;
+	u32 n_rc_delayed_comp;
+	u32 n_seq_naks;
+	u32 n_rdma_seq;
+	u32 n_rnr_naks;
+	u32 n_other_naks;
+	u32 n_loop_pkts;
+	u32 n_pkt_drops;
+	u32 n_vl15_dropped;
+	u32 n_timeouts;
+	u32 n_dmawait;
+	u32 n_unaligned;
+	u32 n_rc_dupreq;
+	u32 n_rc_seqnak;
+	u32 port_cap_flags;
+	u32 pma_sample_start;
+	u32 pma_sample_interval;
+	__be16 pma_counter_select[5];
+	u16 pma_tag;
+	u16 pkey_violations;
+	u16 qkey_violations;
+	u16 mkey_violations;
+	u16 mkey_lease_period;
+	u16 sm_lid;
+	u16 repress_traps;
+	u8 sm_sl;
+	u8 mkeyprot;
+	u8 subnet_timeout;
+	u8 vl_high_limit;
+	u8 sl_to_vl[16];
+
+	struct qib_opcode_stats opstats[128];
+};
+
+struct qib_ibdev {
+	struct ib_device ibdev;
+	struct list_head pending_mmaps;
+	spinlock_t mmap_offset_lock; /* protect mmap_offset */
+	u32 mmap_offset;
+	struct qib_mregion *dma_mr;
+
+	/* QP numbers are shared by all IB ports */
+	struct qib_qpn_table qpn_table;
+	struct qib_lkey_table lk_table;
+	struct list_head piowait;       /* list for wait PIO buf */
+	struct list_head dmawait;	/* list for wait DMA */
+	struct list_head txwait;        /* list for wait qib_verbs_txreq */
+	struct list_head memwait;       /* list for wait kernel memory */
+	struct list_head txreq_free;
+	struct timer_list mem_timer;
+	struct qib_qp **qp_table;
+	struct qib_pio_header *pio_hdrs;
+	dma_addr_t pio_hdrs_phys;
+	/* list of QPs waiting for RNR timer */
+	struct list_head rnrwait;
+	spinlock_t pending_lock; /* protect wait lists, PMA counters, etc. */
+	unsigned qp_table_size; /* size of the hash table */
+	spinlock_t qpt_lock;
+
+	u32 n_piowait;
+	u32 n_txwait;
+
+	u32 n_pds_allocated;    /* number of PDs allocated for device */
+	spinlock_t n_pds_lock;
+	u32 n_ahs_allocated;    /* number of AHs allocated for device */
+	spinlock_t n_ahs_lock;
+	u32 n_cqs_allocated;    /* number of CQs allocated for device */
+	spinlock_t n_cqs_lock;
+	u32 n_qps_allocated;    /* number of QPs allocated for device */
+	spinlock_t n_qps_lock;
+	u32 n_srqs_allocated;   /* number of SRQs allocated for device */
+	spinlock_t n_srqs_lock;
+	u32 n_mcast_grps_allocated; /* number of mcast groups allocated */
+	spinlock_t n_mcast_grps_lock;
+};
+
+struct qib_verbs_counters {
+	u64 symbol_error_counter;
+	u64 link_error_recovery_counter;
+	u64 link_downed_counter;
+	u64 port_rcv_errors;
+	u64 port_rcv_remphys_errors;
+	u64 port_xmit_discards;
+	u64 port_xmit_data;
+	u64 port_rcv_data;
+	u64 port_xmit_packets;
+	u64 port_rcv_packets;
+	u32 local_link_integrity_errors;
+	u32 excessive_buffer_overrun_errors;
+	u32 vl15_dropped;
+};
+
+static inline struct qib_mr *to_imr(struct ib_mr *ibmr)
+{
+	return container_of(ibmr, struct qib_mr, ibmr);
+}
+
+static inline struct qib_pd *to_ipd(struct ib_pd *ibpd)
+{
+	return container_of(ibpd, struct qib_pd, ibpd);
+}
+
+static inline struct qib_ah *to_iah(struct ib_ah *ibah)
+{
+	return container_of(ibah, struct qib_ah, ibah);
+}
+
+static inline struct qib_cq *to_icq(struct ib_cq *ibcq)
+{
+	return container_of(ibcq, struct qib_cq, ibcq);
+}
+
+static inline struct qib_srq *to_isrq(struct ib_srq *ibsrq)
+{
+	return container_of(ibsrq, struct qib_srq, ibsrq);
+}
+
+static inline struct qib_qp *to_iqp(struct ib_qp *ibqp)
+{
+	return container_of(ibqp, struct qib_qp, ibqp);
+}
+
+static inline struct qib_ibdev *to_idev(struct ib_device *ibdev)
+{
+	return container_of(ibdev, struct qib_ibdev, ibdev);
+}
+
+/*
+ * Send if not busy or waiting for I/O and either
+ * a RC response is pending or we can process send work requests.
+ */
+static inline int qib_send_ok(struct qib_qp *qp)
+{
+	return !(qp->s_flags & (QIB_S_BUSY | QIB_S_ANY_WAIT_IO)) &&
+		(qp->s_hdrwords || (qp->s_flags & QIB_S_RESP_PENDING) ||
+		 !(qp->s_flags & QIB_S_ANY_WAIT_SEND));
+}
+
+/*
+ * This must be called with s_lock held.
+ */
+static inline void qib_schedule_send(struct qib_qp *qp)
+{
+	if (qib_send_ok(qp)) {
+		if (qp->processor_id == smp_processor_id())
+			schedule_work(&qp->s_work);
+		else
+			schedule_work_on(qp->processor_id, &qp->s_work);
+	}
+}
+
+static inline int qib_pkey_ok(u16 pkey1, u16 pkey2)
+{
+	u16 p1 = pkey1 & 0x7FFF;
+	u16 p2 = pkey2 & 0x7FFF;
+
+	/*
+	 * Low 15 bits must be non-zero and match, and
+	 * one of the two must be a full member.
+	 */
+	return p1 && p1 == p2 && ((__s16)pkey1 < 0 || (__s16)pkey2 < 0);
+}
+
+void qib_bad_pqkey(struct qib_ibport *ibp, __be16 trap_num, u32 key, u32 sl,
+		   u32 qp1, u32 qp2, __be16 lid1, __be16 lid2);
+void qib_cap_mask_chg(struct qib_ibport *ibp);
+void qib_sys_guid_chg(struct qib_ibport *ibp);
+void qib_node_desc_chg(struct qib_ibport *ibp);
+int qib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
+		    struct ib_wc *in_wc, struct ib_grh *in_grh,
+		    struct ib_mad *in_mad, struct ib_mad *out_mad);
+int qib_create_agents(struct qib_ibdev *dev);
+void qib_free_agents(struct qib_ibdev *dev);
+
+/*
+ * Compare the lower 24 bits of the two values.
+ * Returns an integer <, ==, or > than zero.
+ */
+static inline int qib_cmp24(u32 a, u32 b)
+{
+	return (((int) a) - ((int) b)) << 8;
+}
+
+struct qib_mcast *qib_mcast_find(struct qib_ibport *ibp, union ib_gid *mgid);
+
+int qib_snapshot_counters(struct qib_pportdata *ppd, u64 *swords,
+			  u64 *rwords, u64 *spkts, u64 *rpkts,
+			  u64 *xmit_wait);
+
+int qib_get_counters(struct qib_pportdata *ppd,
+		     struct qib_verbs_counters *cntrs);
+
+int qib_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
+
+int qib_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
+
+int qib_mcast_tree_empty(struct qib_ibport *ibp);
+
+__be32 qib_compute_aeth(struct qib_qp *qp);
+
+struct qib_qp *qib_lookup_qpn(struct qib_ibport *ibp, u32 qpn);
+
+struct ib_qp *qib_create_qp(struct ib_pd *ibpd,
+			    struct ib_qp_init_attr *init_attr,
+			    struct ib_udata *udata);
+
+int qib_destroy_qp(struct ib_qp *ibqp);
+
+int qib_error_qp(struct qib_qp *qp, enum ib_wc_status err);
+
+int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		  int attr_mask, struct ib_udata *udata);
+
+int qib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		 int attr_mask, struct ib_qp_init_attr *init_attr);
+
+unsigned qib_free_all_qps(struct qib_devdata *dd);
+
+void qib_init_qpn_table(struct qib_devdata *dd, struct qib_qpn_table *qpt);
+
+void qib_free_qpn_table(struct qib_qpn_table *qpt);
+
+void qib_get_credit(struct qib_qp *qp, u32 aeth);
+
+unsigned qib_pkt_delay(u32 plen, u8 snd_mult, u8 rcv_mult);
+
+void qib_verbs_sdma_desc_avail(struct qib_pportdata *ppd, unsigned avail);
+
+void qib_put_txreq(struct qib_verbs_txreq *tx);
+
+int qib_verbs_send(struct qib_qp *qp, struct qib_ib_header *hdr,
+		   u32 hdrwords, struct qib_sge_state *ss, u32 len);
+
+void qib_copy_sge(struct qib_sge_state *ss, void *data, u32 length,
+		  int release);
+
+void qib_skip_sge(struct qib_sge_state *ss, u32 length, int release);
+
+void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp);
+
+void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp);
+
+int qib_check_ah(struct ib_device *ibdev, struct ib_ah_attr *ah_attr);
+
+void qib_rc_rnr_retry(unsigned long arg);
+
+void qib_rc_send_complete(struct qib_qp *qp, struct qib_ib_header *hdr);
+
+void qib_rc_error(struct qib_qp *qp, enum ib_wc_status err);
+
+int qib_post_ud_send(struct qib_qp *qp, struct ib_send_wr *wr);
+
+void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp);
+
+int qib_alloc_lkey(struct qib_lkey_table *rkt, struct qib_mregion *mr);
+
+int qib_free_lkey(struct qib_ibdev *dev, struct qib_mregion *mr);
+
+int qib_lkey_ok(struct qib_lkey_table *rkt, struct qib_pd *pd,
+		struct qib_sge *isge, struct ib_sge *sge, int acc);
+
+int qib_rkey_ok(struct qib_qp *qp, struct qib_sge *sge,
+		u32 len, u64 vaddr, u32 rkey, int acc);
+
+int qib_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
+			 struct ib_recv_wr **bad_wr);
+
+struct ib_srq *qib_create_srq(struct ib_pd *ibpd,
+			      struct ib_srq_init_attr *srq_init_attr,
+			      struct ib_udata *udata);
+
+int qib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
+		   enum ib_srq_attr_mask attr_mask,
+		   struct ib_udata *udata);
+
+int qib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr);
+
+int qib_destroy_srq(struct ib_srq *ibsrq);
+
+void qib_cq_enter(struct qib_cq *cq, struct ib_wc *entry, int sig);
+
+int qib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry);
+
+struct ib_cq *qib_create_cq(struct ib_device *ibdev, int entries,
+			    int comp_vector, struct ib_ucontext *context,
+			    struct ib_udata *udata);
+
+int qib_destroy_cq(struct ib_cq *ibcq);
+
+int qib_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags);
+
+int qib_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata);
+
+struct ib_mr *qib_get_dma_mr(struct ib_pd *pd, int acc);
+
+struct ib_mr *qib_reg_phys_mr(struct ib_pd *pd,
+			      struct ib_phys_buf *buffer_list,
+			      int num_phys_buf, int acc, u64 *iova_start);
+
+struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+			      u64 virt_addr, int mr_access_flags,
+			      struct ib_udata *udata);
+
+int qib_dereg_mr(struct ib_mr *ibmr);
+
+struct ib_fmr *qib_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
+			     struct ib_fmr_attr *fmr_attr);
+
+int qib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
+		     int list_len, u64 iova);
+
+int qib_unmap_fmr(struct list_head *fmr_list);
+
+int qib_dealloc_fmr(struct ib_fmr *ibfmr);
+
+void qib_release_mmap_info(struct kref *ref);
+
+struct qib_mmap_info *qib_create_mmap_info(struct qib_ibdev *dev, u32 size,
+					   struct ib_ucontext *context,
+					   void *obj);
+
+void qib_update_mmap_info(struct qib_ibdev *dev, struct qib_mmap_info *ip,
+			  u32 size, void *obj);
+
+int qib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma);
+
+int qib_get_rwqe(struct qib_qp *qp, int wr_id_only);
+
+void qib_migrate_qp(struct qib_qp *qp);
+
+int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		      int has_grh, struct qib_qp *qp, u32 bth0);
+
+u32 qib_make_grh(struct qib_ibport *ibp, struct ib_grh *hdr,
+		 struct ib_global_route *grh, u32 hwords, u32 nwords);
+
+void qib_make_ruc_header(struct qib_qp *qp, struct qib_other_headers *ohdr,
+			 u32 bth0, u32 bth2);
+
+void qib_do_send(struct work_struct *work);
+
+void qib_send_complete(struct qib_qp *qp, struct qib_swqe *wqe,
+		       enum ib_wc_status status);
+
+void qib_send_rc_ack(struct qib_qp *qp);
+
+int qib_make_rc_req(struct qib_qp *qp);
+
+int qib_make_uc_req(struct qib_qp *qp);
+
+int qib_make_ud_req(struct qib_qp *qp);
+
+int qib_register_ib_device(struct qib_devdata *);
+
+void qib_unregister_ib_device(struct qib_devdata *);
+
+void qib_ib_rcv(struct qib_ctxtdata *, void *, void *, u32);
+
+void qib_ib_piobufavail(struct qib_devdata *);
+
+unsigned qib_get_npkeys(struct qib_devdata *);
+
+unsigned qib_get_pkey(struct qib_ibport *, unsigned);
+
+extern const enum ib_wc_opcode ib_qib_wc_opcode[];
+
+/*
+ * Below  HCA-independent IB PhysPortState values, returned
+ * by the f_ibphys_portstate() routine.
+ */
+#define IB_PHYSPORTSTATE_SLEEP 1
+#define IB_PHYSPORTSTATE_POLL 2
+#define IB_PHYSPORTSTATE_DISABLED 3
+#define IB_PHYSPORTSTATE_CFG_TRAIN 4
+#define IB_PHYSPORTSTATE_LINKUP 5
+#define IB_PHYSPORTSTATE_LINK_ERR_RECOVER 6
+#define IB_PHYSPORTSTATE_CFG_DEBOUNCE 8
+#define IB_PHYSPORTSTATE_CFG_IDLE 0xB
+#define IB_PHYSPORTSTATE_RECOVERY_RETRAIN 0xC
+#define IB_PHYSPORTSTATE_RECOVERY_WAITRMT 0xE
+#define IB_PHYSPORTSTATE_RECOVERY_IDLE 0xF
+#define IB_PHYSPORTSTATE_CFG_ENH 0x10
+#define IB_PHYSPORTSTATE_CFG_WAIT_ENH 0x13
+
+extern const int ib_qib_state_ops[];
+
+extern __be64 ib_qib_sys_image_guid;    /* in network order */
+
+extern unsigned int ib_qib_lkey_table_size;
+
+extern unsigned int ib_qib_max_cqes;
+
+extern unsigned int ib_qib_max_cqs;
+
+extern unsigned int ib_qib_max_qp_wrs;
+
+extern unsigned int ib_qib_max_qps;
+
+extern unsigned int ib_qib_max_sges;
+
+extern unsigned int ib_qib_max_mcast_grps;
+
+extern unsigned int ib_qib_max_mcast_qp_attached;
+
+extern unsigned int ib_qib_max_srqs;
+
+extern unsigned int ib_qib_max_srq_sges;
+
+extern unsigned int ib_qib_max_srq_wrs;
+
+extern const u32 ib_qib_rnr_table[];
+
+extern struct ib_dma_mapping_ops qib_dma_mapping_ops;
+
+#endif                          /* QIB_VERBS_H */

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 48/51] IB/qib: Add qib_verbs_mcast.c

[Ralph Campbell]

creates the qib_verbs_mcast.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_verbs_mcast.c |  368 +++++++++++++++++++++++++++
 1 files changed, 368 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_verbs_mcast.c

diff --git a/drivers/infiniband/hw/qib/qib_verbs_mcast.c b/drivers/infiniband/hw/qib/qib_verbs_mcast.c
new file mode 100644
index 0000000..dabb697
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_verbs_mcast.c
@@ -0,0 +1,368 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/rculist.h>
+
+#include "qib.h"
+
+/**
+ * qib_mcast_qp_alloc - alloc a struct to link a QP to mcast GID struct
+ * @qp: the QP to link
+ */
+static struct qib_mcast_qp *qib_mcast_qp_alloc(struct qib_qp *qp)
+{
+	struct qib_mcast_qp *mqp;
+
+	mqp = kmalloc(sizeof *mqp, GFP_KERNEL);
+	if (!mqp)
+		goto bail;
+
+	mqp->qp = qp;
+	atomic_inc(&qp->refcount);
+
+bail:
+	return mqp;
+}
+
+static void qib_mcast_qp_free(struct qib_mcast_qp *mqp)
+{
+	struct qib_qp *qp = mqp->qp;
+
+	/* Notify qib_destroy_qp() if it is waiting. */
+	if (atomic_dec_and_test(&qp->refcount))
+		wake_up(&qp->wait);
+
+	kfree(mqp);
+}
+
+/**
+ * qib_mcast_alloc - allocate the multicast GID structure
+ * @mgid: the multicast GID
+ *
+ * A list of QPs will be attached to this structure.
+ */
+static struct qib_mcast *qib_mcast_alloc(union ib_gid *mgid)
+{
+	struct qib_mcast *mcast;
+
+	mcast = kmalloc(sizeof *mcast, GFP_KERNEL);
+	if (!mcast)
+		goto bail;
+
+	mcast->mgid = *mgid;
+	INIT_LIST_HEAD(&mcast->qp_list);
+	init_waitqueue_head(&mcast->wait);
+	atomic_set(&mcast->refcount, 0);
+	mcast->n_attached = 0;
+
+bail:
+	return mcast;
+}
+
+static void qib_mcast_free(struct qib_mcast *mcast)
+{
+	struct qib_mcast_qp *p, *tmp;
+
+	list_for_each_entry_safe(p, tmp, &mcast->qp_list, list)
+		qib_mcast_qp_free(p);
+
+	kfree(mcast);
+}
+
+/**
+ * qib_mcast_find - search the global table for the given multicast GID
+ * @ibp: the IB port structure
+ * @mgid: the multicast GID to search for
+ *
+ * Returns NULL if not found.
+ *
+ * The caller is responsible for decrementing the reference count if found.
+ */
+struct qib_mcast *qib_mcast_find(struct qib_ibport *ibp, union ib_gid *mgid)
+{
+	struct rb_node *n;
+	unsigned long flags;
+	struct qib_mcast *mcast;
+
+	spin_lock_irqsave(&ibp->lock, flags);
+	n = ibp->mcast_tree.rb_node;
+	while (n) {
+		int ret;
+
+		mcast = rb_entry(n, struct qib_mcast, rb_node);
+
+		ret = memcmp(mgid->raw, mcast->mgid.raw,
+			     sizeof(union ib_gid));
+		if (ret < 0)
+			n = n->rb_left;
+		else if (ret > 0)
+			n = n->rb_right;
+		else {
+			atomic_inc(&mcast->refcount);
+			spin_unlock_irqrestore(&ibp->lock, flags);
+			goto bail;
+		}
+	}
+	spin_unlock_irqrestore(&ibp->lock, flags);
+
+	mcast = NULL;
+
+bail:
+	return mcast;
+}
+
+/**
+ * qib_mcast_add - insert mcast GID into table and attach QP struct
+ * @mcast: the mcast GID table
+ * @mqp: the QP to attach
+ *
+ * Return zero if both were added.  Return EEXIST if the GID was already in
+ * the table but the QP was added.  Return ESRCH if the QP was already
+ * attached and neither structure was added.
+ */
+static int qib_mcast_add(struct qib_ibdev *dev, struct qib_ibport *ibp,
+			 struct qib_mcast *mcast, struct qib_mcast_qp *mqp)
+{
+	struct rb_node **n = &ibp->mcast_tree.rb_node;
+	struct rb_node *pn = NULL;
+	int ret;
+
+	spin_lock_irq(&ibp->lock);
+
+	while (*n) {
+		struct qib_mcast *tmcast;
+		struct qib_mcast_qp *p;
+
+		pn = *n;
+		tmcast = rb_entry(pn, struct qib_mcast, rb_node);
+
+		ret = memcmp(mcast->mgid.raw, tmcast->mgid.raw,
+			     sizeof(union ib_gid));
+		if (ret < 0) {
+			n = &pn->rb_left;
+			continue;
+		}
+		if (ret > 0) {
+			n = &pn->rb_right;
+			continue;
+		}
+
+		/* Search the QP list to see if this is already there. */
+		list_for_each_entry_rcu(p, &tmcast->qp_list, list) {
+			if (p->qp == mqp->qp) {
+				ret = ESRCH;
+				goto bail;
+			}
+		}
+		if (tmcast->n_attached == ib_qib_max_mcast_qp_attached) {
+			ret = ENOMEM;
+			goto bail;
+		}
+
+		tmcast->n_attached++;
+
+		list_add_tail_rcu(&mqp->list, &tmcast->qp_list);
+		ret = EEXIST;
+		goto bail;
+	}
+
+	spin_lock(&dev->n_mcast_grps_lock);
+	if (dev->n_mcast_grps_allocated == ib_qib_max_mcast_grps) {
+		spin_unlock(&dev->n_mcast_grps_lock);
+		ret = ENOMEM;
+		goto bail;
+	}
+
+	dev->n_mcast_grps_allocated++;
+	spin_unlock(&dev->n_mcast_grps_lock);
+
+	mcast->n_attached++;
+
+	list_add_tail_rcu(&mqp->list, &mcast->qp_list);
+
+	atomic_inc(&mcast->refcount);
+	rb_link_node(&mcast->rb_node, pn, n);
+	rb_insert_color(&mcast->rb_node, &ibp->mcast_tree);
+
+	ret = 0;
+
+bail:
+	spin_unlock_irq(&ibp->lock);
+
+	return ret;
+}
+
+int qib_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+	struct qib_ibdev *dev = to_idev(ibqp->device);
+	struct qib_ibport *ibp;
+	struct qib_mcast *mcast;
+	struct qib_mcast_qp *mqp;
+	int ret;
+
+	if (ibqp->qp_num <= 1 || qp->state == IB_QPS_RESET) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	/*
+	 * Allocate data structures since its better to do this outside of
+	 * spin locks and it will most likely be needed.
+	 */
+	mcast = qib_mcast_alloc(gid);
+	if (mcast == NULL) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+	mqp = qib_mcast_qp_alloc(qp);
+	if (mqp == NULL) {
+		qib_mcast_free(mcast);
+		ret = -ENOMEM;
+		goto bail;
+	}
+	ibp = to_iport(ibqp->device, qp->port_num);
+	switch (qib_mcast_add(dev, ibp, mcast, mqp)) {
+	case ESRCH:
+		/* Neither was used: OK to attach the same QP twice. */
+		qib_mcast_qp_free(mqp);
+		qib_mcast_free(mcast);
+		break;
+
+	case EEXIST:            /* The mcast wasn't used */
+		qib_mcast_free(mcast);
+		break;
+
+	case ENOMEM:
+		/* Exceeded the maximum number of mcast groups. */
+		qib_mcast_qp_free(mqp);
+		qib_mcast_free(mcast);
+		ret = -ENOMEM;
+		goto bail;
+
+	default:
+		break;
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+int qib_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+	struct qib_ibdev *dev = to_idev(ibqp->device);
+	struct qib_ibport *ibp = to_iport(ibqp->device, qp->port_num);
+	struct qib_mcast *mcast = NULL;
+	struct qib_mcast_qp *p, *tmp;
+	struct rb_node *n;
+	int last = 0;
+	int ret;
+
+	if (ibqp->qp_num <= 1 || qp->state == IB_QPS_RESET) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	spin_lock_irq(&ibp->lock);
+
+	/* Find the GID in the mcast table. */
+	n = ibp->mcast_tree.rb_node;
+	while (1) {
+		if (n == NULL) {
+			spin_unlock_irq(&ibp->lock);
+			ret = -EINVAL;
+			goto bail;
+		}
+
+		mcast = rb_entry(n, struct qib_mcast, rb_node);
+		ret = memcmp(gid->raw, mcast->mgid.raw,
+			     sizeof(union ib_gid));
+		if (ret < 0)
+			n = n->rb_left;
+		else if (ret > 0)
+			n = n->rb_right;
+		else
+			break;
+	}
+
+	/* Search the QP list. */
+	list_for_each_entry_safe(p, tmp, &mcast->qp_list, list) {
+		if (p->qp != qp)
+			continue;
+		/*
+		 * We found it, so remove it, but don't poison the forward
+		 * link until we are sure there are no list walkers.
+		 */
+		list_del_rcu(&p->list);
+		mcast->n_attached--;
+
+		/* If this was the last attached QP, remove the GID too. */
+		if (list_empty(&mcast->qp_list)) {
+			rb_erase(&mcast->rb_node, &ibp->mcast_tree);
+			last = 1;
+		}
+		break;
+	}
+
+	spin_unlock_irq(&ibp->lock);
+
+	if (p) {
+		/*
+		 * Wait for any list walkers to finish before freeing the
+		 * list element.
+		 */
+		wait_event(mcast->wait, atomic_read(&mcast->refcount) <= 1);
+		qib_mcast_qp_free(p);
+	}
+	if (last) {
+		atomic_dec(&mcast->refcount);
+		wait_event(mcast->wait, !atomic_read(&mcast->refcount));
+		qib_mcast_free(mcast);
+		spin_lock_irq(&dev->n_mcast_grps_lock);
+		dev->n_mcast_grps_allocated--;
+		spin_unlock_irq(&dev->n_mcast_grps_lock);
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+int qib_mcast_tree_empty(struct qib_ibport *ibp)
+{
+	return ibp->mcast_tree.rb_node == NULL;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 49/51] IB/qib: Add qib_wc_ppc64.c

[Ralph Campbell]

creates the qib_wc_ppc64.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_wc_ppc64.c |   62 ++++++++++++++++++++++++++++++
 1 files changed, 62 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_wc_ppc64.c

diff --git a/drivers/infiniband/hw/qib/qib_wc_ppc64.c b/drivers/infiniband/hw/qib/qib_wc_ppc64.c
new file mode 100644
index 0000000..673cf4c
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_wc_ppc64.c
@@ -0,0 +1,62 @@
+/*
+ * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This file is conditionally built on PowerPC only.  Otherwise weak symbol
+ * versions of the functions exported from here are used.
+ */
+
+#include "qib.h"
+
+/**
+ * qib_enable_wc - enable write combining for MMIO writes to the device
+ * @dd: qlogic_ib device
+ *
+ * Nothing to do on PowerPC, so just return without error.
+ */
+int qib_enable_wc(struct qib_devdata *dd)
+{
+	return 0;
+}
+
+/**
+ * qib_unordered_wc - indicate whether write combining is unordered
+ *
+ * Because our performance depends on our ability to do write
+ * combining mmio writes in the most efficient way, we need to
+ * know if we are on a processor that may reorder stores when
+ * write combining.
+ */
+int qib_unordered_wc(void)
+{
+	return 1;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 50/51] IB/qib: Add qib_wc_x86_64.c

[Ralph Campbell]

creates the qib_wc_x86_64.c file.

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/hw/qib/qib_wc_x86_64.c |  187 +++++++++++++++++++++++++++++
 1 files changed, 187 insertions(+), 0 deletions(-)
 create mode 100644 drivers/infiniband/hw/qib/qib_wc_x86_64.c

diff --git a/drivers/infiniband/hw/qib/qib_wc_x86_64.c b/drivers/infiniband/hw/qib/qib_wc_x86_64.c
new file mode 100644
index 0000000..594c198
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_wc_x86_64.c
@@ -0,0 +1,187 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This file is conditionally built on x86_64 only.  Otherwise weak symbol
+ * versions of the functions exported from here are used.
+ */
+
+#include <linux/pci.h>
+#include <asm/mtrr.h>
+#include <asm/processor.h>
+
+#include "qib.h"
+
+/**
+ * qib_enable_wc - enable write combining for MMIO writes to the device
+ * @dd: qlogic_ib device
+ *
+ * This routine is x86_64-specific; it twiddles the CPU's MTRRs to enable
+ * write combining.
+ */
+int qib_enable_wc(struct qib_devdata *dd)
+{
+	int ret = 0;
+	u64 pioaddr, piolen;
+	unsigned bits;
+	const unsigned long addr = pci_resource_start(dd->pcidev, 0);
+	const size_t len = pci_resource_len(dd->pcidev, 0);
+
+	/*
+	 * Set the PIO buffers to be WCCOMB, so we get HT bursts to the
+	 * chip.  Linux (possibly the hardware) requires it to be on a power
+	 * of 2 address matching the length (which has to be a power of 2).
+	 * For rev1, that means the base address, for rev2, it will be just
+	 * the PIO buffers themselves.
+	 * For chips with two sets of buffers, the calculations are
+	 * somewhat more complicated; we need to sum, and the piobufbase
+	 * register has both offsets, 2K in low 32 bits, 4K in high 32 bits.
+	 * The buffers are still packed, so a single range covers both.
+	 */
+	if (dd->piobcnt2k && dd->piobcnt4k) {
+		/* 2 sizes for chip */
+		unsigned long pio2kbase, pio4kbase;
+		pio2kbase = dd->piobufbase & 0xffffffffUL;
+		pio4kbase = (dd->piobufbase >> 32) & 0xffffffffUL;
+		if (pio2kbase < pio4kbase) {
+			/* all current chips */
+			pioaddr = addr + pio2kbase;
+			piolen = pio4kbase - pio2kbase +
+				dd->piobcnt4k * dd->align4k;
+		} else {
+			pioaddr = addr + pio4kbase;
+			piolen = pio2kbase - pio4kbase +
+				dd->piobcnt2k * dd->palign;
+		}
+	} else {  /* single buffer size (2K, currently) */
+		pioaddr = addr + dd->piobufbase;
+		piolen = dd->piobcnt2k * dd->palign +
+			dd->piobcnt4k * dd->align4k;
+	}
+
+	for (bits = 0; !(piolen & (1ULL << bits)); bits++)
+		/* do nothing */ ;
+
+	if (piolen != (1ULL << bits)) {
+		piolen >>= bits;
+		while (piolen >>= 1)
+			bits++;
+		piolen = 1ULL << (bits + 1);
+	}
+	if (pioaddr & (piolen - 1)) {
+		u64 atmp;
+		qib_dbg("pioaddr %llx not on right boundary for size "
+			  "%llx, fixing\n",
+			  (unsigned long long) pioaddr,
+			  (unsigned long long) piolen);
+		atmp = pioaddr & ~(piolen - 1);
+		if (atmp < addr || (atmp + piolen) > (addr + len)) {
+			qib_dev_err(dd, "No way to align address/size "
+				    "(%llx/%llx), no WC mtrr\n",
+				    (unsigned long long) atmp,
+				    (unsigned long long) piolen << 1);
+			ret = -ENODEV;
+		} else {
+			qib_dbg("changing WC base from %llx to %llx, "
+				"len from %llx to %llx\n",
+				(unsigned long long) pioaddr,
+				(unsigned long long) atmp,
+				(unsigned long long) piolen,
+				(unsigned long long) piolen << 1);
+			pioaddr = atmp;
+			piolen <<= 1;
+		}
+	}
+
+	if (!ret) {
+		int cookie;
+		qib_cdbg(VERBOSE, "Setting mtrr for chip to WC "
+			 "(addr %llx, len=0x%llx)\n",
+			 (unsigned long long) pioaddr,
+			 (unsigned long long) piolen);
+		cookie = mtrr_add(pioaddr, piolen, MTRR_TYPE_WRCOMB, 0);
+		if (cookie < 0) {
+			{
+				qib_devinfo(dd->pcidev,
+					 "mtrr_add()  WC for PIO bufs "
+					 "failed (%d)\n",
+					 cookie);
+				ret = -EINVAL;
+			}
+		} else {
+			qib_cdbg(VERBOSE, "Set mtrr for chip to WC, "
+				 "cookie is %d\n", cookie);
+			dd->wc_cookie = cookie;
+			dd->wc_base = (unsigned long) pioaddr;
+			dd->wc_len = (unsigned long) piolen;
+		}
+	}
+
+	return ret;
+}
+
+/**
+ * qib_disable_wc - disable write combining for MMIO writes to the device
+ * @dd: qlogic_ib device
+ */
+void qib_disable_wc(struct qib_devdata *dd)
+{
+	if (dd->wc_cookie) {
+		int r;
+
+		qib_cdbg(VERBOSE, "undoing WCCOMB on pio buffers\n");
+		r = mtrr_del(dd->wc_cookie, dd->wc_base,
+			     dd->wc_len);
+		if (r < 0)
+			qib_devinfo(dd->pcidev,
+				 "mtrr_del(%lx, %lx, %lx) failed: %d\n",
+				 dd->wc_cookie, dd->wc_base,
+				 dd->wc_len, r);
+		dd->wc_cookie = 0; /* even on failure */
+	}
+}
+
+/**
+ * qib_unordered_wc - indicate whether write combining is ordered
+ *
+ * Because our performance depends on our ability to do write combining mmio
+ * writes in the most efficient way, we need to know if we are on an Intel
+ * or AMD x86_64 processor.  AMD x86_64 processors flush WC buffers out in
+ * the order completed, and so no special flushing is required to get
+ * correct ordering.  Intel processors, however, will flush write buffers
+ * out in "random" orders, and so explicit ordering is needed at times.
+ */
+int qib_unordered_wc(void)
+{
+	return boot_cpu_data.x86_vendor != X86_VENDOR_AMD;
+}

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 51/51] IB/qib: Hooks for adding the QIB driver into the framework

[Ralph Campbell]

Signed-off-by: Ralph Campbell <ralph.campbell-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---

 drivers/infiniband/Kconfig  |    1 +
 drivers/infiniband/Makefile |    1 +
 2 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/drivers/infiniband/Kconfig b/drivers/infiniband/Kconfig
index dd0db67..c80f973 100644
--- a/drivers/infiniband/Kconfig
+++ b/drivers/infiniband/Kconfig
@@ -42,6 +42,7 @@ config INFINIBAND_ADDR_TRANS
 
 source "drivers/infiniband/hw/mthca/Kconfig"
 source "drivers/infiniband/hw/ipath/Kconfig"
+source "drivers/infiniband/hw/qib/Kconfig"
 source "drivers/infiniband/hw/ehca/Kconfig"
 source "drivers/infiniband/hw/amso1100/Kconfig"
 source "drivers/infiniband/hw/cxgb3/Kconfig"
diff --git a/drivers/infiniband/Makefile b/drivers/infiniband/Makefile
index ed35e44..27b66b1 100644
--- a/drivers/infiniband/Makefile
+++ b/drivers/infiniband/Makefile
@@ -1,6 +1,7 @@
 obj-$(CONFIG_INFINIBAND)		+= core/
 obj-$(CONFIG_INFINIBAND_MTHCA)		+= hw/mthca/
 obj-$(CONFIG_INFINIBAND_IPATH)		+= hw/ipath/
+obj-$(CONFIG_INFINIBAND_QIB)		+= hw/qib/
 obj-$(CONFIG_INFINIBAND_EHCA)		+= hw/ehca/
 obj-$(CONFIG_INFINIBAND_AMSO1100)	+= hw/amso1100/
 obj-$(CONFIG_INFINIBAND_CXGB3)		+= hw/cxgb3/

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 0/51] updated patches for adding QIB driver

[Roland Dreier]

have not had a chance to look at the new submission yet but I think you
need at least the following rolled in, to avoid calling __exit code that
might have been thrown away from the error path of your init function.

Building with CONFIG_DEBUG_SECTION_MISMATCH=y when you see that the
kernel build tells you that you introduced new section mismatches is
probably a good idea.

diff --git a/drivers/infiniband/hw/qib/qib_file_ops.c b/drivers/infiniband/hw/qib/qib_file_ops.c
index 781c87f..53c314f 100644
--- a/drivers/infiniband/hw/qib/qib_file_ops.c
+++ b/drivers/infiniband/hw/qib/qib_file_ops.c
@@ -2450,7 +2450,7 @@ done:
 	return ret;
 }
 
-void __exit qib_dev_cleanup(void)
+void qib_dev_cleanup(void)
 {
 	if (qib_class) {
 		class_destroy(qib_class);
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 0/51] updated patches for adding QIB driver

[Ralph Campbell]

OK.
Done.

On Wed, 2009-12-09 at 14:47 -0800, Roland Dreier wrote:
> have not had a chance to look at the new submission yet but I think you
> need at least the following rolled in, to avoid calling __exit code that
> might have been thrown away from the error path of your init function.
> 
> Building with CONFIG_DEBUG_SECTION_MISMATCH=y when you see that the
> kernel build tells you that you introduced new section mismatches is
> probably a good idea.
> 
> diff --git a/drivers/infiniband/hw/qib/qib_file_ops.c b/drivers/infiniband/hw/qib/qib_file_ops.c
> index 781c87f..53c314f 100644
> --- a/drivers/infiniband/hw/qib/qib_file_ops.c
> +++ b/drivers/infiniband/hw/qib/qib_file_ops.c
> @@ -2450,7 +2450,7 @@ done:
>  	return ret;
>  }
>  
> -void __exit qib_dev_cleanup(void)
> +void qib_dev_cleanup(void)
>  {
>  	if (qib_class) {
>  		class_destroy(qib_class);

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html