[PATCH v2 00/15] Add Paravirtual RDMA Driver

[PATCH v2 00/15] Add Paravirtual RDMA Driver

[Adit Ranadive]

Hi Doug, others,

This patch series adds a driver for a paravirtual RDMA device. The device
is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs on
ESX. We recently did a presentation in the OFA Workshop regarding this
device. See - https://goo.gl/pHOXJ8 for our workshop presentation. Note,
that this patch series should be applied all together. I split out the
commits so that it may be easier to review.

Description and RDMA Support
-------------------------------
The virtual device is exposed as a dual function PCIe device. One part is
a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.

These patches add support and the all required infrastructure for letting
applications use such a device. We support the mandatory Verbs API as well
as the base memory management extensions (Local Inv, Send with Inv and Fast
Register Work Requests). We currently support both Reliable Connected and
Unreliable Datagram QPs but do not support Shared Receive Queues (SRQs).
Also, we support the following types of Work Requests:
 o Send/Receive (with or without Immediate Data)
 o RDMA Write (with or without Immediate Data)
 o RDMA Read
 o Local Invalidate
 o Send with Invalidate
 o Fast Register Work Requests

This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based and
IP-based GIDs. 

I have also created a git tree for our user-level driver here:
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary

Testing
--------
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several runs
of the performance tests (included with OFED), Intel MPI PingPong benchmark
on OpenMPI, krping for FRWRs. Mellanox has been kind enough to test the
backported version of the driver internally on their hardware using a
VMware provided ESX build. I have also applied and tested this with Doug's
master, k.o/for-4.7, k.o/for-4.8 branches.

Changes v1 -> v2: 
 Patch [07/15] - Addressed Yuval Shaia's comments and 32-bit build errors.

---

Adit Ranadive (15):
  IB/pvrdma: Add paravirtual rdma device
  IB/pvrdma: Add device command support
  IB/pvrdma: Add support for Completion Queues
  IB/pvrdma: Add the paravirtual RDMA device specification
  IB/pvrdma: Add UAR support
  IB/pvrdma: Add virtual device RDMA structures
  IB/pvrdma: Add the main driver module for PVRDMA
  IB/pvrdma: Add helper functions
  IB/pvrdma: Add support for memory regions
  IB/pvrdma: Add Queue Pair support
  IB/pvrdma: Add user-level shared functions
  IB/pvrdma: Add functions for Verbs support
  IB/pvrdma: Add Kconfig and Makefile
  IB: Add PVRDMA driver
  MAINTAINERS: Update for PVRDMA driver

 MAINTAINERS                                    |    7 +
 drivers/infiniband/Kconfig                     |    1 +
 drivers/infiniband/hw/Makefile                 |    1 +
 drivers/infiniband/hw/pvrdma/Kconfig           |    8 +
 drivers/infiniband/hw/pvrdma/Makefile          |    3 +
 drivers/infiniband/hw/pvrdma/pvrdma.h          |  459 +++++++++
 drivers/infiniband/hw/pvrdma/pvrdma_cmd.c      |  104 +++
 drivers/infiniband/hw/pvrdma/pvrdma_cq.c       |  436 +++++++++
 drivers/infiniband/hw/pvrdma/pvrdma_defs.h     |  300 ++++++
 drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h  |  342 +++++++
 drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c |  128 +++
 drivers/infiniband/hw/pvrdma/pvrdma_ib_verbs.h |  450 +++++++++
 drivers/infiniband/hw/pvrdma/pvrdma_main.c     | 1189 ++++++++++++++++++++++++
 drivers/infiniband/hw/pvrdma/pvrdma_misc.c     |  309 ++++++
 drivers/infiniband/hw/pvrdma/pvrdma_mr.c       |  333 +++++++
 drivers/infiniband/hw/pvrdma/pvrdma_qp.c       |  976 +++++++++++++++++++
 drivers/infiniband/hw/pvrdma/pvrdma_uapi.h     |  247 +++++
 drivers/infiniband/hw/pvrdma/pvrdma_user.h     |   99 ++
 drivers/infiniband/hw/pvrdma/pvrdma_verbs.c    |  593 ++++++++++++
 drivers/infiniband/hw/pvrdma/pvrdma_verbs.h    |  108 +++
 20 files changed, 6093 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/Kconfig
 create mode 100644 drivers/infiniband/hw/pvrdma/Makefile
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma.h
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_cmd.c
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_cq.c
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_defs.h
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_ib_verbs.h
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_main.c
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_misc.c
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_mr.c
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_qp.c
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_uapi.h
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_user.h
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_verbs.c
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_verbs.h

-- 
2.7.4

--
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 01/15] IB/pvrdma: Add paravirtual rdma device

[Adit Ranadive]

This patch adds the main device-level structures and functions to be used
to provide RDMA functionality. Also, we define conversion functions from
the IB core stack structures to the device-specific ones.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/pvrdma/pvrdma.h | 459 ++++++++++++++++++++++++++++++++++
 1 file changed, 459 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma.h

diff --git a/drivers/infiniband/hw/pvrdma/pvrdma.h b/drivers/infiniband/hw/pvrdma/pvrdma.h
new file mode 100644
index 0000000..0dc5bd2
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma.h
@@ -0,0 +1,459 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef __PVRDMA_H__
+#define __PVRDMA_H__
+
+#include <linux/compiler.h>
+#include <linux/interrupt.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/semaphore.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_verbs.h>
+
+#include "pvrdma_defs.h"
+#include "pvrdma_dev_api.h"
+#include "pvrdma_verbs.h"
+
+/* NOT the same as BIT_MASK(). */
+#define PVRDMA_MASK(n) ((n << 1) - 1)
+
+/*
+ * VMware PVRDMA vendor id and PCI device id.
+ */
+#ifndef PCI_VENDOR_ID_VMWARE
+#define PCI_VENDOR_ID_VMWARE		0x15AD
+#endif
+#define PCI_DEVICE_ID_VMWARE_PVRDMA	0x0820
+
+struct pvrdma_dev;
+
+struct pvrdma_page_dir {
+	dma_addr_t dir_dma;
+	u64 *dir;
+	int ntables;
+	u64 **tables;
+	u64 npages;
+	void **pages;
+};
+
+struct pvrdma_cq {
+	struct ib_cq ibcq;
+	int offset;
+	spinlock_t cq_lock; /* Poll lock. */
+	struct pvrdma_uar_map *uar;
+	struct ib_umem *umem;
+	struct pvrdma_ring_state *ring_state;
+	struct pvrdma_page_dir pdir;
+	u32 cq_handle;
+	bool is_kernel;
+	atomic_t refcnt;
+	wait_queue_head_t wait;
+};
+
+struct pvrdma_id_table {
+	u32 last;
+	u32 top;
+	u32 max;
+	u32 mask;
+	spinlock_t lock; /* Table lock. */
+	unsigned long *table;
+};
+
+struct pvrdma_uar_map {
+	unsigned long pfn;
+	void __iomem *map;
+	int index;
+};
+
+struct pvrdma_uar_table {
+	struct pvrdma_id_table tbl;
+	int size;
+};
+
+struct pvrdma_ucontext {
+	struct ib_ucontext ibucontext;
+	struct pvrdma_dev *dev;
+	struct pvrdma_uar_map uar;
+	u64 ctx_handle;
+};
+
+struct pvrdma_pd {
+	struct ib_pd ibpd;
+	u32 pdn;
+	u32 pd_handle;
+	int priviledged;
+};
+
+struct pvrdma_mr {
+	u32 mr_handle;
+	u64 iova;
+	u64 size;
+};
+
+struct pvrdma_user_mr {
+	struct ib_mr ibmr;
+	struct ib_umem *umem;
+	struct pvrdma_mr mmr;
+	struct pvrdma_page_dir pdir;
+	u64 *pages;
+	u32 npages;
+	u32 max_pages;
+	u32 page_shift;
+};
+
+struct pvrdma_wq {
+	struct pvrdma_ring *ring;
+	spinlock_t lock; /* Work queue lock. */
+	int wqe_cnt;
+	int wqe_size;
+	int max_gs;
+	int offset;
+};
+
+struct pvrdma_ah {
+	struct ib_ah ibah;
+	struct pvrdma_av av;
+};
+
+struct pvrdma_qp {
+	struct ib_qp ibqp;
+	u32 qp_handle;
+	u32 qkey;
+	struct pvrdma_wq sq;
+	struct pvrdma_wq rq;
+	struct ib_umem *rumem;
+	struct ib_umem *sumem;
+	struct pvrdma_page_dir pdir;
+	int npages;
+	int npages_send;
+	int npages_recv;
+	u32 flags;
+	u8 port;
+	u8 state;
+	bool is_kernel;
+	struct mutex mutex; /* QP state mutex. */
+	atomic_t refcnt;
+	wait_queue_head_t wait;
+};
+
+struct pvrdma_dev {
+	/* PCI device-related information. */
+	struct ib_device ib_dev;
+	struct pci_dev *pdev;
+	void __iomem *regs;
+	struct pvrdma_device_shared_region *dsr; /* Shared region pointer */
+	dma_addr_t dsrbase; /* Shared region base address */
+	void *cmd_slot;
+	void *resp_slot;
+	unsigned long flags;
+	struct list_head device_link;
+
+	/* Locking and interrupt information. */
+	spinlock_t cmd_lock; /* Command lock. */
+	struct semaphore cmd_sema;
+	struct completion cmd_done;
+	struct {
+		enum pvrdma_intr_type type; /* Intr type */
+		struct msix_entry msix_entry[PVRDMA_MAX_INTERRUPTS];
+		irq_handler_t handler[PVRDMA_MAX_INTERRUPTS];
+		u8 enabled[PVRDMA_MAX_INTERRUPTS];
+		u8 size;
+	} intr;
+
+	/* RDMA-related device information. */
+	union ib_gid *sgid_tbl;
+	struct pvrdma_ring_state *async_ring_state;
+	struct pvrdma_page_dir async_pdir;
+	struct pvrdma_ring_state *cq_ring_state;
+	struct pvrdma_page_dir cq_pdir;
+	struct pvrdma_cq **cq_tbl;
+	spinlock_t cq_tbl_lock;
+	struct pvrdma_qp **qp_tbl;
+	spinlock_t qp_tbl_lock;
+	struct pvrdma_uar_table uar_table;
+	struct pvrdma_uar_map driver_uar;
+	__be64 sys_image_guid;
+	spinlock_t desc_lock; /* Device modification lock. */
+	u32 port_cap_mask;
+	struct mutex port_mutex; /* Port modification mutex. */
+	bool ib_active;
+	atomic_t num_qps;
+	atomic_t num_cqs;
+	atomic_t num_pds;
+	atomic_t num_ahs;
+
+	/* Network device information. */
+	struct net_device *netdev;
+	struct notifier_block nb_netdev;
+};
+
+static inline struct pvrdma_dev *to_vdev(struct ib_device *ibdev)
+{
+	return container_of(ibdev, struct pvrdma_dev, ib_dev);
+}
+
+static inline struct
+pvrdma_ucontext *to_vucontext(struct ib_ucontext *ibucontext)
+{
+	return container_of(ibucontext, struct pvrdma_ucontext, ibucontext);
+}
+
+static inline struct pvrdma_pd *to_vpd(struct ib_pd *ibpd)
+{
+	return container_of(ibpd, struct pvrdma_pd, ibpd);
+}
+
+static inline struct pvrdma_cq *to_vcq(struct ib_cq *ibcq)
+{
+	return container_of(ibcq, struct pvrdma_cq, ibcq);
+}
+
+static inline struct pvrdma_user_mr *to_vmr(struct ib_mr *ibmr)
+{
+	return container_of(ibmr, struct pvrdma_user_mr, ibmr);
+}
+
+static inline struct pvrdma_qp *to_vqp(struct ib_qp *ibqp)
+{
+	return container_of(ibqp, struct pvrdma_qp, ibqp);
+}
+
+static inline struct pvrdma_ah *to_vah(struct ib_ah *ibah)
+{
+	return container_of(ibah, struct pvrdma_ah, ibah);
+}
+
+static inline void pvrdma_write_reg(struct pvrdma_dev *dev, u32 reg, u32 val)
+{
+	writel(cpu_to_le32(val), dev->regs + reg);
+}
+
+static inline u32 pvrdma_read_reg(struct pvrdma_dev *dev, u32 reg)
+{
+	return le32_to_cpu(readl(dev->regs + reg));
+}
+
+static inline void *pvrdma_page_dir_get_ptr(struct pvrdma_page_dir *pdir,
+					    u64 offset)
+{
+	return pdir->pages[offset / PAGE_SIZE] + (offset % PAGE_SIZE);
+}
+
+static inline enum pvrdma_mtu ib_mtu_to_pvrdma(enum ib_mtu mtu)
+{
+	return (enum pvrdma_mtu)mtu;
+}
+
+static inline enum ib_mtu pvrdma_mtu_to_ib(enum pvrdma_mtu mtu)
+{
+	return (enum ib_mtu)mtu;
+}
+
+static inline enum pvrdma_port_state ib_port_state_to_pvrdma(
+					enum ib_port_state state)
+{
+	return (enum pvrdma_port_state)state;
+}
+
+static inline enum ib_port_state pvrdma_port_state_to_ib(
+					enum pvrdma_port_state state)
+{
+	return (enum ib_port_state)state;
+}
+
+static inline int ib_port_cap_flags_to_pvrdma(int flags)
+{
+	return flags & PVRDMA_MASK(PVRDMA_PORT_CAP_FLAGS_MAX);
+}
+
+static inline int pvrdma_port_cap_flags_to_ib(int flags)
+{
+	return flags;
+}
+
+static inline enum pvrdma_port_width ib_port_width_to_pvrdma(
+					enum ib_port_width width)
+{
+	return (enum pvrdma_port_width)width;
+}
+
+static inline enum ib_port_width pvrdma_port_width_to_ib(
+					enum pvrdma_port_width width)
+{
+	return (enum ib_port_width)width;
+}
+
+static inline enum pvrdma_port_speed ib_port_speed_to_pvrdma(
+					enum ib_port_speed speed)
+{
+	return (enum pvrdma_port_speed)speed;
+}
+
+static inline enum ib_port_speed pvrdma_port_speed_to_ib(
+					enum pvrdma_port_speed speed)
+{
+	return (enum ib_port_speed)speed;
+}
+
+static inline int pvrdma_qp_attr_mask_to_ib(int attr_mask)
+{
+	return attr_mask;
+}
+
+static inline int ib_qp_attr_mask_to_pvrdma(int attr_mask)
+{
+	return attr_mask & PVRDMA_MASK(PVRDMA_QP_ATTR_MASK_MAX);
+}
+
+static inline enum pvrdma_mig_state ib_mig_state_to_pvrdma(
+					enum ib_mig_state state)
+{
+	return (enum pvrdma_mig_state)state;
+}
+
+static inline enum ib_mig_state pvrdma_mig_state_to_ib(
+					enum pvrdma_mig_state state)
+{
+	return (enum ib_mig_state)state;
+}
+
+static inline int ib_access_flags_to_pvrdma(int flags)
+{
+	return flags;
+}
+
+static inline int pvrdma_access_flags_to_ib(int flags)
+{
+	return flags & PVRDMA_MASK(PVRDMA_ACCESS_FLAGS_MAX);
+}
+
+static inline enum pvrdma_qp_type ib_qp_type_to_pvrdma(enum ib_qp_type type)
+{
+	return (enum pvrdma_qp_type)type;
+}
+
+static inline enum ib_qp_type pvrdma_qp_type_to_ib(enum pvrdma_qp_type type)
+{
+	return (enum ib_qp_type)type;
+}
+
+static inline enum pvrdma_qp_state ib_qp_state_to_pvrdma(enum ib_qp_state state)
+{
+	return (enum pvrdma_qp_state)state;
+}
+
+static inline enum ib_qp_state pvrdma_qp_state_to_ib(enum pvrdma_qp_state state)
+{
+	return (enum ib_qp_state)state;
+}
+
+static inline enum pvrdma_wr_opcode ib_wr_opcode_to_pvrdma(enum ib_wr_opcode op)
+{
+	return (enum pvrdma_wr_opcode)op;
+}
+
+static inline enum ib_wc_status pvrdma_wc_status_to_ib(
+					enum pvrdma_wc_status status)
+{
+	return (enum ib_wc_status)status;
+}
+
+static inline int pvrdma_wc_opcode_to_ib(int opcode)
+{
+	return opcode;
+}
+
+static inline int pvrdma_wc_flags_to_ib(int flags)
+{
+	return flags;
+}
+
+static inline int ib_send_flags_to_pvrdma(int flags)
+{
+	return flags & PVRDMA_MASK(PVRDMA_SEND_FLAGS_MAX);
+}
+
+void pvrdma_qp_cap_to_ib(const struct pvrdma_qp_cap *src,
+			 struct ib_qp_cap *dst);
+void ib_qp_cap_to_pvrdma(const struct ib_qp_cap *src,
+			 struct pvrdma_qp_cap *dst);
+void pvrdma_gid_to_ib(const union pvrdma_gid *src, union ib_gid *dst);
+void ib_gid_to_pvrdma(const union ib_gid *src, union pvrdma_gid *dst);
+void pvrdma_global_route_to_ib(const struct pvrdma_global_route *src,
+			       struct ib_global_route *dst);
+void ib_global_route_to_pvrdma(const struct ib_global_route *src,
+			       struct pvrdma_global_route *dst);
+void pvrdma_ah_attr_to_ib(const struct pvrdma_ah_attr *src,
+			  struct ib_ah_attr *dst);
+void ib_ah_attr_to_pvrdma(const struct ib_ah_attr *src,
+			  struct pvrdma_ah_attr *dst);
+
+int pvrdma_uar_table_init(struct pvrdma_dev *dev);
+void pvrdma_uar_table_cleanup(struct pvrdma_dev *dev);
+
+int pvrdma_uar_alloc(struct pvrdma_dev *dev, struct pvrdma_uar_map *uar);
+void pvrdma_uar_free(struct pvrdma_dev *dev, struct pvrdma_uar_map *uar);
+
+void pvrdma_flush_cqe(struct pvrdma_qp *qp, struct pvrdma_cq *cq);
+
+int pvrdma_page_dir_init(struct pvrdma_dev *dev, struct pvrdma_page_dir *pdir,
+			 u64 npages, bool alloc_pages);
+void pvrdma_page_dir_cleanup(struct pvrdma_dev *dev,
+			     struct pvrdma_page_dir *pdir);
+int pvrdma_page_dir_insert_dma(struct pvrdma_page_dir *pdir, u64 idx,
+			       dma_addr_t daddr);
+int pvrdma_page_dir_insert_umem(struct pvrdma_page_dir *pdir,
+				struct ib_umem *umem, u64 offset);
+dma_addr_t pvrdma_page_dir_get_dma(struct pvrdma_page_dir *pdir, u64 idx);
+int pvrdma_page_dir_insert_page_list(struct pvrdma_page_dir *pdir,
+				     u64 *page_list, int num_pages);
+
+int pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
+		    bool expect, union pvrdma_cmd_resp *rsp);
+
+#endif /* __PVRDMA_H__ */
-- 
2.7.4

--
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 02/15] IB/pvrdma: Add device command support

[Adit Ranadive]

This patch enables posting Verb requests and receiving responses to/from
the backend PVRDMA emulation layer.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/pvrdma/pvrdma_cmd.c | 104 ++++++++++++++++++++++++++++++
 1 file changed, 104 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_cmd.c

diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_cmd.c b/drivers/infiniband/hw/pvrdma/pvrdma_cmd.c
new file mode 100644
index 0000000..c22debf
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_cmd.c
@@ -0,0 +1,104 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <linux/list.h>
+
+#include "pvrdma.h"
+
+#define PVRDMA_CMD_TIMEOUT	10000 /* ms */
+
+static int pvrdma_cmd_recv(struct pvrdma_dev *dev, union pvrdma_cmd_resp *resp)
+{
+	dev_dbg(&dev->pdev->dev, "receive response from device\n");
+
+	spin_lock(&dev->cmd_lock);
+	memcpy(resp, dev->resp_slot, sizeof(*resp));
+	spin_unlock(&dev->cmd_lock);
+
+	return 0;
+}
+
+int
+pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
+		bool expect_resp, union pvrdma_cmd_resp *resp)
+{
+	int err;
+
+	dev_dbg(&dev->pdev->dev, "post request to device\n");
+
+	/* Serializiation */
+	down(&dev->cmd_sema);
+
+	spin_lock(&dev->cmd_lock);
+	memcpy(dev->cmd_slot, req, sizeof(*req));
+	spin_unlock(&dev->cmd_lock);
+
+	init_completion(&dev->cmd_done);
+	pvrdma_write_reg(dev, PVRDMA_REG_REQUEST, 0);
+
+	/* Make sure the request is written before reading status. */
+	mb();
+	err = pvrdma_read_reg(dev, PVRDMA_REG_ERR);
+	if (err == 0) {
+		if (expect_resp) {
+			err = wait_for_completion_interruptible_timeout(
+				&dev->cmd_done,
+				msecs_to_jiffies(PVRDMA_CMD_TIMEOUT));
+			if (err == 0) {
+				dev_warn(&dev->pdev->dev,
+					 "completion timeout\n");
+				err = -ETIMEDOUT;
+			} else {
+				err = pvrdma_cmd_recv(dev, resp);
+			}
+		}
+	} else {
+		dev_warn(&dev->pdev->dev, "failed to write request %d\n", err);
+	}
+
+	up(&dev->cmd_sema);
+
+	return err;
+}
-- 
2.7.4

--
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 03/15] IB/pvrdma: Add support for Completion Queues

[Adit Ranadive]

This patch adds the support for creating and destroying completion queues
on the paravirtual RDMA device.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/pvrdma/pvrdma_cq.c | 436 +++++++++++++++++++++++++++++++
 1 file changed, 436 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_cq.c

diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_cq.c b/drivers/infiniband/hw/pvrdma/pvrdma_cq.c
new file mode 100644
index 0000000..9a4b42c
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_cq.c
@@ -0,0 +1,436 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <asm/page.h>
+#include <linux/io.h>
+#include <linux/wait.h>
+#include <rdma/ib_addr.h>
+#include <rdma/ib_smi.h>
+#include <rdma/ib_user_verbs.h>
+
+#include "pvrdma.h"
+#include "pvrdma_user.h"
+
+/**
+ * pvrdma_req_notify_cq - request notification for a completion queue
+ * @ibcq: the completion queue
+ * @notify_flags: notification flags
+ *
+ * @return: 0 for success.
+ */
+int pvrdma_req_notify_cq(struct ib_cq *ibcq,
+			 enum ib_cq_notify_flags notify_flags)
+{
+	struct pvrdma_dev *dev = to_vdev(ibcq->device);
+	struct pvrdma_cq *cq = to_vcq(ibcq);
+	u32 val = cq->cq_handle;
+
+	val |= (notify_flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED ?
+		PVRDMA_UAR_CQ_ARM_SOL : PVRDMA_UAR_CQ_ARM;
+
+	writel(cpu_to_le32(val), dev->driver_uar.map + PVRDMA_UAR_CQ_OFFSET);
+
+	return 0;
+}
+
+/**
+ * pvrdma_create_cq - create completion queue
+ * @ibdev: the device
+ * @attr: completion queue attributes
+ * @context: user context
+ * @udata: user data
+ *
+ * @return: ib_cq completion queue pointer on success,
+ *          otherwise returns negative errno.
+ */
+struct ib_cq *pvrdma_create_cq(struct ib_device *ibdev,
+			       const struct ib_cq_init_attr *attr,
+			       struct ib_ucontext *context,
+			       struct ib_udata *udata)
+{
+	int entries = attr->cqe;
+	struct pvrdma_dev *dev = to_vdev(ibdev);
+	struct pvrdma_cq *cq;
+	int ret;
+	int npages;
+	unsigned long flags;
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_create_cq *cmd = &req.create_cq;
+	struct pvrdma_cmd_create_cq_resp *resp = &rsp.create_cq_resp;
+	struct pvrdma_create_cq ucmd;
+
+	BUILD_BUG_ON(sizeof(struct pvrdma_cqe) != 64);
+
+	entries = roundup_pow_of_two(entries);
+	if (entries < 1 || entries > dev->dsr->caps.max_cqe)
+		return ERR_PTR(-EINVAL);
+
+	if (!atomic_add_unless(&dev->num_cqs, 1, dev->dsr->caps.max_cq))
+		return ERR_PTR(-EINVAL);
+
+	cq = kzalloc(sizeof(*cq), GFP_KERNEL);
+	if (!cq) {
+		atomic_dec(&dev->num_cqs);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	cq->ibcq.cqe = entries;
+
+	if (context) {
+		if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) {
+			ret = -EFAULT;
+			goto err_cq;
+		}
+
+		cq->umem = ib_umem_get(context, ucmd.buf_addr, ucmd.buf_size,
+				       IB_ACCESS_LOCAL_WRITE, 1);
+		if (IS_ERR(cq->umem)) {
+			ret = PTR_ERR(cq->umem);
+			goto err_cq;
+		}
+
+		npages = ib_umem_page_count(cq->umem);
+	} else {
+		cq->is_kernel = true;
+
+		/* One extra page for shared ring state */
+		npages = 1 + (entries * sizeof(struct pvrdma_cqe) +
+			      PAGE_SIZE - 1) / PAGE_SIZE;
+
+		/* Skip header page. */
+		cq->offset = PAGE_SIZE;
+	}
+
+	if (npages < 0 || npages > PVRDMA_PAGE_DIR_MAX_PAGES) {
+		dev_warn(&dev->pdev->dev,
+			 "overflow pages in completion queue\n");
+		ret = -EINVAL;
+		goto err_umem;
+	}
+
+	ret = pvrdma_page_dir_init(dev, &cq->pdir, npages, cq->is_kernel);
+	if (ret) {
+		dev_warn(&dev->pdev->dev,
+			 "could not allocate page directory\n");
+		goto err_umem;
+	}
+
+	if (cq->is_kernel) {
+		/* Ring state is always the first page. */
+		cq->ring_state = cq->pdir.pages[0];
+	} else {
+		pvrdma_page_dir_insert_umem(&cq->pdir, cq->umem, 0);
+	}
+
+	atomic_set(&cq->refcnt, 1);
+	init_waitqueue_head(&cq->wait);
+	spin_lock_init(&cq->cq_lock);
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_CREATE_CQ;
+	cmd->nchunks = npages;
+	cmd->ctx_handle = (context) ?
+		(u64)to_vucontext(context)->ctx_handle : 0;
+	cmd->cqe = entries;
+	cmd->pdir_dma = cq->pdir.dir_dma;
+	ret = pvrdma_cmd_post(dev, &req, true, &rsp);
+
+	if (ret < 0 || rsp.hdr.ack != PVRDMA_CMD_CREATE_CQ_RESP) {
+		dev_warn(&dev->pdev->dev,
+			 "could not create completion queue\n");
+		goto err_page_dir;
+	}
+
+	cq->ibcq.cqe = resp->cqe;
+	cq->cq_handle = resp->cq_handle;
+	spin_lock_irqsave(&dev->cq_tbl_lock, flags);
+	dev->cq_tbl[cq->cq_handle % dev->dsr->caps.max_cq] = cq;
+	spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
+
+	if (context) {
+		cq->uar = &(to_vucontext(context)->uar);
+
+		/* Copy udata back. */
+		if (ib_copy_to_udata(udata, &cq->cq_handle, sizeof(__u32))) {
+			dev_warn(&dev->pdev->dev,
+				 "failed to copy back udata\n");
+			ret = -EINVAL;
+			goto err_page_dir;
+		}
+	}
+
+	return &cq->ibcq;
+
+err_page_dir:
+	pvrdma_page_dir_cleanup(dev, &cq->pdir);
+err_umem:
+	if (context)
+		ib_umem_release(cq->umem);
+err_cq:
+	atomic_dec(&dev->num_cqs);
+	kfree(cq);
+
+	return ERR_PTR(ret);
+}
+
+static void pvrdma_free_cq(struct pvrdma_dev *dev, struct pvrdma_cq *cq)
+{
+	atomic_dec(&cq->refcnt);
+	wait_event(cq->wait, !atomic_read(&cq->refcnt));
+
+	if (!cq->is_kernel)
+		ib_umem_release(cq->umem);
+
+	pvrdma_page_dir_cleanup(dev, &cq->pdir);
+	kfree(cq);
+}
+
+/**
+ * pvrdma_destroy_cq - destroy completion queue
+ * @cq: the completion queue to destroy.
+ *
+ * @return: 0 for success.
+ */
+int pvrdma_destroy_cq(struct ib_cq *cq)
+{
+	struct pvrdma_cq *vcq = to_vcq(cq);
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_destroy_cq *cmd = &req.destroy_cq;
+	struct pvrdma_dev *dev = to_vdev(cq->device);
+	unsigned long flags;
+	int ret;
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_DESTROY_CQ;
+	cmd->cq_handle = vcq->cq_handle;
+
+	ret = pvrdma_cmd_post(dev, &req, false, &rsp);
+	if (ret < 0)
+		dev_warn(&dev->pdev->dev,
+			 "could not destroy completion queue\n");
+
+	/* free cq's resources */
+	spin_lock_irqsave(&dev->cq_tbl_lock, flags);
+	dev->cq_tbl[vcq->cq_handle] = NULL;
+	spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
+
+	pvrdma_free_cq(dev, vcq);
+	atomic_dec(&dev->num_cqs);
+
+	return ret;
+}
+
+/**
+ * pvrdma_modify_cq - modify the CQ moderation parameters
+ * @ibcq: the CQ to modify
+ * @cq_count: number of CQEs that will trigger an event
+ * @cq_period: max period of time in usec before triggering an event
+ *
+ * @return: -EOPNOTSUPP as CQ resize is not supported.
+ */
+int pvrdma_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
+{
+	return -EOPNOTSUPP;
+}
+
+static inline struct pvrdma_cqe *get_cqe(struct pvrdma_cq *cq, int i)
+{
+	return (struct pvrdma_cqe *)pvrdma_page_dir_get_ptr(
+					&cq->pdir,
+					cq->offset +
+					sizeof(struct pvrdma_cqe) * i);
+}
+
+void pvrdma_flush_cqe(struct pvrdma_qp *qp, struct pvrdma_cq *cq)
+{
+	int head;
+	int has_data;
+
+	if (!cq->is_kernel)
+		return;
+
+	/* Lock held */
+	has_data = pvrdma_idx_ring_has_data(&cq->ring_state->rx,
+					    cq->ibcq.cqe, &head);
+	if (unlikely(has_data > 0)) {
+		int items;
+		int curr;
+		int tail = pvrdma_idx(&cq->ring_state->rx.prod_tail,
+				      cq->ibcq.cqe);
+		struct pvrdma_cqe *cqe;
+		struct pvrdma_cqe *curr_cqe;
+
+		items = (tail > head) ? (tail - head) :
+			(cq->ibcq.cqe - head + tail);
+		curr = --tail;
+		while (items-- > 0) {
+			if (curr < 0)
+				curr = cq->ibcq.cqe - 1;
+			if (tail < 0)
+				tail = cq->ibcq.cqe - 1;
+			curr_cqe = get_cqe(cq, curr);
+			if ((curr_cqe->qp & 0xFFFF) != qp->qp_handle) {
+				if (curr != tail) {
+					cqe = get_cqe(cq, tail);
+					*cqe = *curr_cqe;
+				}
+				tail--;
+			} else {
+				pvrdma_idx_ring_inc(
+					&cq->ring_state->rx.cons_head,
+					cq->ibcq.cqe);
+			}
+			curr--;
+		}
+	}
+}
+
+static int pvrdma_poll_one(struct pvrdma_cq *cq, struct pvrdma_qp **cur_qp,
+			   struct ib_wc *wc)
+{
+	struct pvrdma_dev *dev = to_vdev(cq->ibcq.device);
+	int has_data;
+	unsigned int head;
+	bool tried = false;
+	struct pvrdma_cqe *cqe;
+
+retry:
+	has_data = pvrdma_idx_ring_has_data(&cq->ring_state->rx,
+					    cq->ibcq.cqe, &head);
+	if (has_data == 0) {
+		u32 val;
+
+		if (tried)
+			return -EAGAIN;
+
+		/* Pass down POLL to give physical HCA a chance to poll. */
+		val = cq->cq_handle | PVRDMA_UAR_CQ_POLL;
+		writel(cpu_to_le32(val),
+		       dev->driver_uar.map + PVRDMA_UAR_CQ_OFFSET);
+
+		tried = true;
+		goto retry;
+	} else if (has_data == -1) {
+		return -EINVAL;
+	}
+
+	cqe = get_cqe(cq, head);
+
+	/* Ensure cqe is valid. */
+	rmb();
+	if (dev->qp_tbl[cqe->qp & 0xffff])
+		*cur_qp = (struct pvrdma_qp *)dev->qp_tbl[cqe->qp & 0xffff];
+	else
+		return -EINVAL;
+
+	wc->opcode = pvrdma_wc_opcode_to_ib(cqe->opcode);
+	wc->status = pvrdma_wc_status_to_ib(cqe->status);
+	wc->wr_id = cqe->wr_id;
+	wc->qp = &(*cur_qp)->ibqp;
+	wc->byte_len = cqe->byte_len;
+	wc->ex.imm_data = cqe->imm_data;
+	wc->src_qp = cqe->src_qp;
+	wc->wc_flags = pvrdma_wc_flags_to_ib(cqe->wc_flags);
+	wc->pkey_index = cqe->pkey_index;
+	wc->slid = cqe->slid;
+	wc->sl = cqe->sl;
+	wc->dlid_path_bits = cqe->dlid_path_bits;
+	wc->port_num = cqe->port_num;
+	wc->vendor_err = 0;
+
+	/* Update shared ring state */
+	pvrdma_idx_ring_inc(&cq->ring_state->rx.cons_head, cq->ibcq.cqe);
+
+	return 0;
+}
+
+/**
+ * pvrdma_poll_cq - poll for work completion queue entries
+ * @ibcq: completion queue
+ * @num_entries: the maximum number of entries
+ * @entry: pointer to work completion array
+ *
+ * @return: number of polled completion entries
+ */
+int pvrdma_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
+{
+	struct pvrdma_cq *cq = to_vcq(ibcq);
+	struct pvrdma_qp *cur_qp = NULL;
+	unsigned long flags;
+	int npolled;
+	int ret;
+
+	if (num_entries < 1)
+		return -EINVAL;
+
+	spin_lock_irqsave(&cq->cq_lock, flags);
+	for (npolled = 0; npolled < num_entries; ++npolled) {
+		ret = pvrdma_poll_one(cq, &cur_qp, wc + npolled);
+		if (ret)
+			break;
+	}
+
+	spin_unlock_irqrestore(&cq->cq_lock, flags);
+
+	if ((ret == 0) || (ret == -EAGAIN))
+		return npolled;
+	else
+		return ret;
+}
+
+/**
+ * pvrdma_resize_cq - resize CQ
+ * @ibcq: the completion queue
+ * @entries: CQ entries
+ * @udata: user data
+ *
+ * @return: -EOPNOTSUPP as CQ resize is not supported.
+ */
+int pvrdma_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
+{
+	return -EOPNOTSUPP;
+}
-- 
2.7.4

--
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 04/15] IB/pvrdma: Add the paravirtual RDMA device specification

[Adit Ranadive]

This patch describes the main specification of the underlying virtual RDMA
device. The pvrdma_dev_api header file defines the Verbs commands and
their parameters that can be issued to the device backend.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/pvrdma/pvrdma_defs.h    | 300 ++++++++++++++++++++++
 drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h | 342 ++++++++++++++++++++++++++
 2 files changed, 642 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_defs.h
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h

diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_defs.h b/drivers/infiniband/hw/pvrdma/pvrdma_defs.h
new file mode 100644
index 0000000..225cba4
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_defs.h
@@ -0,0 +1,300 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef __PVRDMA_DEFS_H__
+#define __PVRDMA_DEFS_H__
+
+#include <linux/types.h>
+#include "pvrdma_ib_verbs.h"
+#include "pvrdma_uapi.h"
+
+/*
+ * Masks and accessors for page directory, which is a two-level lookup:
+ * page directory -> page table -> page. Only one directory for now, but we
+ * could expand that easily. 9 bits for tables, 9 bits for pages, gives one
+ * gigabyte for memory regions and so forth.
+ */
+
+#define PVRDMA_PAGE_DIR_DIR(x)		(((x) >> 18) & 0x1)
+#define PVRDMA_PAGE_DIR_TABLE(x)	(((x) >> 9) & 0x1ff)
+#define PVRDMA_PAGE_DIR_PAGE(x)		((x) & 0x1ff)
+#define PVRDMA_PAGE_DIR_MAX_PAGES	(1 * 512 * 512)
+#define PVRDMA_MAX_FAST_REG_PAGES	128
+
+/*
+ * Max MSI-X vectors.
+ */
+
+#define PVRDMA_MAX_INTERRUPTS	3
+
+/* Register offsets within PCI resource on BAR1. */
+#define PVRDMA_REG_VERSION	0x00	/* R: Version of device. */
+#define PVRDMA_REG_DSRLOW	0x04	/* W: Device shared region low PA. */
+#define PVRDMA_REG_DSRHIGH	0x08	/* W: Device shared region high PA. */
+#define PVRDMA_REG_CTL		0x0c	/* W: PVRDMA_DEVICE_CTL */
+#define PVRDMA_REG_REQUEST	0x10	/* W: Indicate device request. */
+#define PVRDMA_REG_ERR		0x14	/* R: Device error. */
+#define PVRDMA_REG_ICR		0x18	/* R: Interrupt cause. */
+#define PVRDMA_REG_IMR		0x1c	/* R/W: Interrupt mask. */
+#define PVRDMA_REG_MACL		0x20	/* R/W: MAC address low. */
+#define PVRDMA_REG_MACH		0x24	/* R/W: MAC address high. */
+
+/* Object flags. */
+#define PVRDMA_CQ_FLAG_ARMED_SOL	BIT(0)	/* Armed for solicited-only. */
+#define PVRDMA_CQ_FLAG_ARMED		BIT(1)	/* Armed. */
+#define PVRDMA_MR_FLAG_DMA		BIT(0)	/* DMA region. */
+#define PVRDMA_MR_FLAG_FRMR		BIT(1)	/* Fast reg memory region. */
+
+/*
+ * Atomic operation capability (masked versions are extended atomic
+ * operations.
+ */
+
+#define PVRDMA_ATOMIC_OP_COMP_SWAP	BIT(0) /* Compare and swap. */
+#define PVRDMA_ATOMIC_OP_FETCH_ADD	BIT(1) /* Fetch and add. */
+#define PVRDMA_ATOMIC_OP_MASK_COMP_SWAP	BIT(2) /* Masked compare and swap. */
+#define PVRDMA_ATOMIC_OP_MASK_FETCH_ADD	BIT(3) /* Masked fetch and add. */
+
+/*
+ * Base Memory Management Extension flags to support Fast Reg Memory Regions
+ * and Fast Reg Work Requests. Each flag represents a verb operation and we
+ * must support all of them to qualify for the BMME device cap.
+ */
+
+#define PVRDMA_BMME_FLAG_LOCAL_INV	BIT(0) /* Local Invalidate. */
+#define PVRDMA_BMME_FLAG_REMOTE_INV	BIT(1) /* Remote Invalidate. */
+#define PVRDMA_BMME_FLAG_FAST_REG_WR	BIT(2) /* Fast Reg Work Request. */
+
+/*
+ * GID types. The interpretation of the gid_types bit field in the device
+ * capabilities will depend on the device mode. For now, the device only
+ * supports RoCE as mode, so only the different GID types for RoCE are
+ * defined.
+ */
+
+#define PVRDMA_GID_TYPE_FLAG_ROCE_V1 BIT(0)
+#define PVRDMA_GID_TYPE_FLAG_ROCE_V2 BIT(1)
+
+enum pvrdma_pci_resource {
+	PVRDMA_PCI_RESOURCE_MSIX = 0,	/* BAR0: MSI-X, MMIO. */
+	PVRDMA_PCI_RESOURCE_REG,	/* BAR1: Registers, MMIO. */
+	PVRDMA_PCI_RESOURCE_UAR,	/* BAR2: UAR pages, MMIO, 64-bit. */
+	PVRDMA_PCI_RESOURCE_LAST,	/* Last. */
+};
+
+enum pvrdma_device_ctl {
+	PVRDMA_DEVICE_CTL_ACTIVATE = 0,	/* Activate device. */
+	PVRDMA_DEVICE_CTL_QUIESCE,	/* Quiesce device. */
+	PVRDMA_DEVICE_CTL_RESET,	/* Reset device. */
+};
+
+enum pvrdma_intr_vector {
+	PVRDMA_INTR_VECTOR_RESPONSE	= 0,	/* Command response. */
+	PVRDMA_INTR_VECTOR_ASYNC	= 1,	/* Async events. */
+	PVRDMA_INTR_VECTOR_CQ		= 2,	/* CQ notification. */
+	/* Additional CQ notification vectors. */
+};
+
+enum pvrdma_intr_cause {
+	PVRDMA_INTR_CAUSE_RESPONSE	= (1 << PVRDMA_INTR_VECTOR_RESPONSE),
+	PVRDMA_INTR_CAUSE_ASYNC		= (1 << PVRDMA_INTR_VECTOR_ASYNC),
+	PVRDMA_INTR_CAUSE_CQ		= (1 << PVRDMA_INTR_VECTOR_CQ),
+};
+
+enum pvrdma_intr_type {
+	PVRDMA_INTR_TYPE_INTX = 0,	/* Legacy. */
+	PVRDMA_INTR_TYPE_MSI,		/* MSI. */
+	PVRDMA_INTR_TYPE_MSIX,		/* MSI-X. */
+};
+
+enum pvrdma_gos_bits {
+	PVRDMA_GOS_BITS_UNK = 0,	/* Unknown. */
+	PVRDMA_GOS_BITS_32,		/* 32-bit. */
+	PVRDMA_GOS_BITS_64,		/* 64-bit. */
+};
+
+enum pvrdma_gos_type {
+	PVRDMA_GOS_TYPE_UNK = 0,	/* Unknown. */
+	PVRDMA_GOS_TYPE_LINUX,		/* Linux. */
+};
+
+enum pvrdma_device_mode {
+	PVRDMA_DEVICE_MODE_ROCE = 0,	/* RoCE. */
+	PVRDMA_DEVICE_MODE_IWARP,	/* iWarp. */
+	PVRDMA_DEVICE_MODE_IB		/* InfiniBand. */
+};
+
+struct pvrdma_gos_info {
+	__u32 gos_bits:2;	/* W: PVRDMA_GOS_BITS_ */
+	__u32 gos_type:4;	/* W: PVRDMA_GOS_TYPE_ */
+	__u32 gos_ver:16;	/* W: Guest OS version. */
+	__u32 gos_misc:10;	/* W: Other. */
+	__u32 pad;		/*    Pad to 8-byte alignment. */
+};
+
+struct pvrdma_device_caps {
+	__u64 fw_ver;				/* R: Query device. */
+	__be64 node_guid;
+	__be64 sys_image_guid;
+	__u64 max_mr_size;
+	__u64 page_size_cap;
+	__u64 atomic_arg_sizes;			/* EXP verbs. */
+	__u32 exp_comp_mask;			/* EXP verbs. */
+	__u32 device_cap_flags2;		/* EXP verbs. */
+	__u32 max_fa_bit_boundary;		/* EXP verbs. */
+	__u32 log_max_atomic_inline_arg;	/* EXP verbs. */
+	__u32 vendor_id;
+	__u32 vendor_part_id;
+	__u32 hw_ver;
+	__u32 max_qp;
+	__u32 max_qp_wr;
+	__u32 device_cap_flags;
+	__u32 max_sge;
+	__u32 max_sge_rd;
+	__u32 max_cq;
+	__u32 max_cqe;
+	__u32 max_mr;
+	__u32 max_pd;
+	__u32 max_qp_rd_atom;
+	__u32 max_ee_rd_atom;
+	__u32 max_res_rd_atom;
+	__u32 max_qp_init_rd_atom;
+	__u32 max_ee_init_rd_atom;
+	__u32 max_ee;
+	__u32 max_rdd;
+	__u32 max_mw;
+	__u32 max_raw_ipv6_qp;
+	__u32 max_raw_ethy_qp;
+	__u32 max_mcast_grp;
+	__u32 max_mcast_qp_attach;
+	__u32 max_total_mcast_qp_attach;
+	__u32 max_ah;
+	__u32 max_fmr;
+	__u32 max_map_per_fmr;
+	__u32 max_srq;
+	__u32 max_srq_wr;
+	__u32 max_srq_sge;
+	__u32 max_uar;
+	__u32 gid_tbl_len;
+	__u16 max_pkeys;
+	__u8  local_ca_ack_delay;
+	__u8  phys_port_cnt;
+	__u8  mode;		/* PVRDMA_DEVICE_MODE_ */
+	__u8  atomic_ops;	/* PVRDMA_ATOMIC_OP_* bits */
+	__u8  bmme_flags;	/* Memory Mgmt Extensions for FRMR/FRWR */
+	__u8  gid_types;	/* PVRDMA_GID_TYPE_FLAG_ */
+	__u8  reserved[4];
+};
+
+struct pvrdma_ring_page_info {
+	__u32 num_pages;			/* Num pages incl. header. */
+	__u32 reserved;				/* Reserved. */
+	__u64 pdir_dma;				/* Page directory PA. */
+};
+
+#pragma pack(push, 1)
+
+struct pvrdma_device_shared_region {
+	__u32 driver_version;			/* W: Driver version. */
+	__u32 pad;				/*    Pad to 8-byte align. */
+	struct pvrdma_gos_info gos_info;	/* W: Guest OS information. */
+	__u64 cmd_slot_dma;			/* W: Command slot address. */
+	__u64 resp_slot_dma;			/* W: Response slot address. */
+	struct pvrdma_ring_page_info async_ring_pages;
+						/* W: Async ring page info. */
+	struct pvrdma_ring_page_info cq_ring_pages;
+						/* W: CQ ring page info. */
+	__u32 uar_pfn;				/* W: UAR pageframe. */
+	__u32 pad2;				/*    Pad to 8-byte align. */
+	struct pvrdma_device_caps caps;		/* R: Device capabilities. */
+};
+
+#pragma pack(pop)
+
+
+/* Event types. Currently a 1:1 mapping with enum ib_event. */
+enum pvrdma_eqe_type {
+	PVRDMA_EVENT_CQ_ERR = 0,
+	PVRDMA_EVENT_QP_FATAL,
+	PVRDMA_EVENT_QP_REQ_ERR,
+	PVRDMA_EVENT_QP_ACCESS_ERR,
+	PVRDMA_EVENT_COMM_EST,
+	PVRDMA_EVENT_SQ_DRAINED,
+	PVRDMA_EVENT_PATH_MIG,
+	PVRDMA_EVENT_PATH_MIG_ERR,
+	PVRDMA_EVENT_DEVICE_FATAL,
+	PVRDMA_EVENT_PORT_ACTIVE,
+	PVRDMA_EVENT_PORT_ERR,
+	PVRDMA_EVENT_LID_CHANGE,
+	PVRDMA_EVENT_PKEY_CHANGE,
+	PVRDMA_EVENT_SM_CHANGE,
+	PVRDMA_EVENT_SRQ_ERR,
+	PVRDMA_EVENT_SRQ_LIMIT_REACHED,
+	PVRDMA_EVENT_QP_LAST_WQE_REACHED,
+	PVRDMA_EVENT_CLIENT_REREGISTER,
+	PVRDMA_EVENT_GID_CHANGE,
+};
+
+/* Event queue element. */
+struct pvrdma_eqe {
+	__u32 type;	/* Event type. */
+	__u32 info;	/* Handle, other. */
+};
+
+/* CQ notification queue element. */
+struct pvrdma_cqne {
+	__u32 info;	/* Handle */
+};
+
+static inline void pvrdma_init_cqe(struct pvrdma_cqe *cqe, __u64 wr_id,
+				   __u64 qp)
+{
+	memset(cqe, 0, sizeof(*cqe));
+	cqe->status = PVRDMA_WC_GENERAL_ERR;
+	cqe->wr_id = wr_id;
+	cqe->qp = qp;
+}
+
+#endif /* __PVRDMA_DEFS_H__ */
diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h b/drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h
new file mode 100644
index 0000000..4b4d8e5
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h
@@ -0,0 +1,342 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef __PVRDMA_DEV_API_H__
+#define __PVRDMA_DEV_API_H__
+
+#include <linux/types.h>
+#include "pvrdma_ib_verbs.h"
+
+enum {
+	PVRDMA_CMD_FIRST = 0,
+	PVRDMA_CMD_QUERY_PORT = PVRDMA_CMD_FIRST,
+	PVRDMA_CMD_QUERY_PKEY,
+	PVRDMA_CMD_CREATE_PD,
+	PVRDMA_CMD_DESTROY_PD,
+	PVRDMA_CMD_CREATE_MR,
+	PVRDMA_CMD_DESTROY_MR,
+	PVRDMA_CMD_CREATE_CQ,
+	PVRDMA_CMD_RESIZE_CQ,
+	PVRDMA_CMD_DESTROY_CQ,
+	PVRDMA_CMD_CREATE_QP,
+	PVRDMA_CMD_MODIFY_QP,
+	PVRDMA_CMD_QUERY_QP,
+	PVRDMA_CMD_DESTROY_QP,
+	PVRDMA_CMD_CREATE_UC,
+	PVRDMA_CMD_DESTROY_UC,
+	PVRDMA_CMD_CREATE_BIND,
+	PVRDMA_CMD_DESTROY_BIND,
+	PVRDMA_CMD_MAX,
+};
+
+enum {
+	PVRDMA_CMD_FIRST_RESP = (1 << 31),
+	PVRDMA_CMD_QUERY_PORT_RESP = PVRDMA_CMD_FIRST_RESP,
+	PVRDMA_CMD_QUERY_PKEY_RESP,
+	PVRDMA_CMD_CREATE_PD_RESP,
+	PVRDMA_CMD_DESTROY_PD_RESP_NOOP,
+	PVRDMA_CMD_CREATE_MR_RESP,
+	PVRDMA_CMD_DESTROY_MR_RESP_NOOP,
+	PVRDMA_CMD_CREATE_CQ_RESP,
+	PVRDMA_CMD_RESIZE_CQ_RESP,
+	PVRDMA_CMD_DESTROY_CQ_RESP_NOOP,
+	PVRDMA_CMD_CREATE_QP_RESP,
+	PVRDMA_CMD_MODIFY_QP_RESP,
+	PVRDMA_CMD_QUERY_QP_RESP,
+	PVRDMA_CMD_DESTROY_QP_RESP,
+	PVRDMA_CMD_CREATE_UC_RESP,
+	PVRDMA_CMD_DESTROY_UC_RESP_NOOP,
+	PVRDMA_CMD_CREATE_BIND_RESP_NOOP,
+	PVRDMA_CMD_DESTROY_BIND_RESP_NOOP,
+	PVRDMA_CMD_MAX_RESP,
+};
+
+struct pvrdma_cmd_hdr {
+	__u64 response;	/* Key for response lookup. */
+	__u32 cmd;	/* PVRDMA_CMD_ */
+	__u32 reserved;	/* Reserved. */
+};
+
+struct pvrdma_cmd_resp_hdr {
+	__u64 response;		/* From cmd hdr. */
+	__u32 ack;		/* PVRDMA_CMD_XXX_RESP */
+	__u8 err;		/* Error. */
+	__u8 reserved[3];	/* Reserved. */
+};
+
+struct pvrdma_cmd_query_port {
+	struct pvrdma_cmd_hdr hdr;
+	__u8 port_num;
+	__u8 reserved[7];
+};
+
+struct pvrdma_cmd_query_port_resp {
+	struct pvrdma_cmd_resp_hdr hdr;
+	struct pvrdma_port_attr attrs;
+};
+
+struct pvrdma_cmd_query_pkey {
+	struct pvrdma_cmd_hdr hdr;
+	__u8 port_num;
+	__u8 index;
+	__u8 reserved[6];
+};
+
+struct pvrdma_cmd_query_pkey_resp {
+	struct pvrdma_cmd_resp_hdr hdr;
+	__u16 pkey;
+	__u8 reserved[6];
+};
+
+struct pvrdma_cmd_create_uc {
+	struct pvrdma_cmd_hdr hdr;
+	__u32 pfn; /* UAR page frame number */
+	__u8 reserved[4];
+};
+
+struct pvrdma_cmd_create_uc_resp {
+	struct pvrdma_cmd_resp_hdr hdr;
+	__u32 ctx_handle;
+	__u8 reserved[4];
+};
+
+struct pvrdma_cmd_destroy_uc {
+	struct pvrdma_cmd_hdr hdr;
+	__u32 ctx_handle;
+	__u8 reserved[4];
+};
+
+struct pvrdma_cmd_create_pd {
+	struct pvrdma_cmd_hdr hdr;
+	__u32 ctx_handle;
+	__u8 reserved[4];
+};
+
+struct pvrdma_cmd_create_pd_resp {
+	struct pvrdma_cmd_resp_hdr hdr;
+	__u32 pd_handle;
+	__u8 reserved[4];
+};
+
+struct pvrdma_cmd_destroy_pd {
+	struct pvrdma_cmd_hdr hdr;
+	__u32 pd_handle;
+	__u8 reserved[4];
+};
+
+struct pvrdma_cmd_create_mr {
+	struct pvrdma_cmd_hdr hdr;
+	__u64 start;
+	__u64 length;
+	__u64 pdir_dma;
+	__u32 pd_handle;
+	__u32 access_flags;
+	__u32 flags;
+	__u32 nchunks;
+};
+
+struct pvrdma_cmd_create_mr_resp {
+	struct pvrdma_cmd_resp_hdr hdr;
+	__u32 mr_handle;
+	__u32 lkey;
+	__u32 rkey;
+	__u8 reserved[4];
+};
+
+struct pvrdma_cmd_destroy_mr {
+	struct pvrdma_cmd_hdr hdr;
+	__u32 mr_handle;
+	__u8 reserved[4];
+};
+
+struct pvrdma_cmd_create_cq {
+	struct pvrdma_cmd_hdr hdr;
+	__u64 pdir_dma;
+	__u32 ctx_handle;
+	__u32 cqe;
+	__u32 nchunks;
+	__u8 reserved[4];
+};
+
+struct pvrdma_cmd_create_cq_resp {
+	struct pvrdma_cmd_resp_hdr hdr;
+	__u32 cq_handle;
+	__u32 cqe;
+};
+
+struct pvrdma_cmd_resize_cq {
+	struct pvrdma_cmd_hdr hdr;
+	__u32 cq_handle;
+	__u32 cqe;
+};
+
+struct pvrdma_cmd_resize_cq_resp {
+	struct pvrdma_cmd_resp_hdr hdr;
+	__u32 cqe;
+	__u8 reserved[4];
+};
+
+struct pvrdma_cmd_destroy_cq {
+	struct pvrdma_cmd_hdr hdr;
+	__u32 cq_handle;
+	__u8 reserved[4];
+};
+
+struct pvrdma_cmd_create_qp {
+	struct pvrdma_cmd_hdr hdr;
+	__u64 pdir_dma;
+	__u32 pd_handle;
+	__u32 send_cq_handle;
+	__u32 recv_cq_handle;
+	__u32 srq_handle;
+	__u32 max_send_wr;
+	__u32 max_recv_wr;
+	__u32 max_send_sge;
+	__u32 max_recv_sge;
+	__u32 max_inline_data;
+	__u32 lkey;
+	__u32 access_flags;
+	__u16 total_chunks;
+	__u16 send_chunks;
+	__u16 max_atomic_arg;
+	__u8 sq_sig_all;
+	__u8 qp_type;
+	__u8 is_srq;
+	__u8 reserved[3];
+};
+
+struct pvrdma_cmd_create_qp_resp {
+	struct pvrdma_cmd_resp_hdr hdr;
+	__u32 qpn;
+	__u32 max_send_wr;
+	__u32 max_recv_wr;
+	__u32 max_send_sge;
+	__u32 max_recv_sge;
+	__u32 max_inline_data;
+};
+
+struct pvrdma_cmd_modify_qp {
+	struct pvrdma_cmd_hdr hdr;
+	__u32 qp_handle;
+	__u32 attr_mask;
+	struct pvrdma_qp_attr attrs;
+};
+
+struct pvrdma_cmd_query_qp {
+	struct pvrdma_cmd_hdr hdr;
+	__u32 qp_handle;
+	__u32 attr_mask;
+};
+
+struct pvrdma_cmd_query_qp_resp {
+	struct pvrdma_cmd_resp_hdr hdr;
+	struct pvrdma_qp_attr attrs;
+};
+
+struct pvrdma_cmd_destroy_qp {
+	struct pvrdma_cmd_hdr hdr;
+	__u32 qp_handle;
+	__u8 reserved[4];
+};
+
+struct pvrdma_cmd_destroy_qp_resp {
+	struct pvrdma_cmd_resp_hdr hdr;
+	__u32 events_reported;
+	__u8 reserved[4];
+};
+
+struct pvrdma_cmd_create_bind {
+	struct pvrdma_cmd_hdr hdr;
+	__u32 mtu;
+	__u32 vlan;
+	__u32 index;
+	__u8 new_gid[16];
+	__u8 gid_type;
+	__u8 reserved[3];
+};
+
+struct pvrdma_cmd_destroy_bind {
+	struct pvrdma_cmd_hdr hdr;
+	__u32 index;
+	__u8 dest_gid[16];
+	__u8 reserved[4];
+};
+
+union pvrdma_cmd_req {
+	struct pvrdma_cmd_hdr hdr;
+	struct pvrdma_cmd_query_port query_port;
+	struct pvrdma_cmd_query_pkey query_pkey;
+	struct pvrdma_cmd_create_uc create_uc;
+	struct pvrdma_cmd_destroy_uc destroy_uc;
+	struct pvrdma_cmd_create_pd create_pd;
+	struct pvrdma_cmd_destroy_pd destroy_pd;
+	struct pvrdma_cmd_create_mr create_mr;
+	struct pvrdma_cmd_destroy_mr destroy_mr;
+	struct pvrdma_cmd_create_cq create_cq;
+	struct pvrdma_cmd_resize_cq resize_cq;
+	struct pvrdma_cmd_destroy_cq destroy_cq;
+	struct pvrdma_cmd_create_qp create_qp;
+	struct pvrdma_cmd_modify_qp modify_qp;
+	struct pvrdma_cmd_query_qp query_qp;
+	struct pvrdma_cmd_destroy_qp destroy_qp;
+	struct pvrdma_cmd_create_bind create_bind;
+	struct pvrdma_cmd_destroy_bind destroy_bind;
+};
+
+union pvrdma_cmd_resp {
+	struct pvrdma_cmd_resp_hdr hdr;
+	struct pvrdma_cmd_query_port_resp query_port_resp;
+	struct pvrdma_cmd_query_pkey_resp query_pkey_resp;
+	struct pvrdma_cmd_create_uc_resp create_uc_resp;
+	struct pvrdma_cmd_create_pd_resp create_pd_resp;
+	struct pvrdma_cmd_create_mr_resp create_mr_resp;
+	struct pvrdma_cmd_create_cq_resp create_cq_resp;
+	struct pvrdma_cmd_resize_cq_resp resize_cq_resp;
+	struct pvrdma_cmd_create_qp_resp create_qp_resp;
+	struct pvrdma_cmd_query_qp_resp query_qp_resp;
+	struct pvrdma_cmd_destroy_qp_resp destroy_qp_resp;
+};
+
+#endif /* __PVRDMA_DEV_API_H__ */
-- 
2.7.4

--
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 05/15] IB/pvrdma: Add UAR support

[Adit Ranadive]

This patch adds the UAR support for the paravirtual RDMA device. The UAR
pages are MMIO pages from the virtual PCI space. We define offsets within
this page to provide the fast data-path operations.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c | 128 +++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c

diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c b/drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c
new file mode 100644
index 0000000..d854f6d
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c
@@ -0,0 +1,128 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <linux/bitmap.h>
+#include <linux/errno.h>
+#include <linux/slab.h>
+
+#include "pvrdma.h"
+
+int pvrdma_uar_table_init(struct pvrdma_dev *dev)
+{
+	u32 num = dev->dsr->caps.max_uar;
+	u32 mask = num - 1;
+	struct pvrdma_id_table *tbl = &dev->uar_table.tbl;
+
+	/* Power of 2 */
+	if (num != 1 << (ffs(num) - 1))
+		return -EINVAL;
+
+	tbl->last = 0;
+	tbl->top = 0;
+	tbl->max = num;
+	tbl->mask = mask;
+	spin_lock_init(&tbl->lock);
+	tbl->table = kcalloc(BITS_TO_LONGS(num), sizeof(long), GFP_KERNEL);
+	if (!tbl->table)
+		return -ENOMEM;
+
+	/* 0th UAR is taken by the device. */
+	set_bit(0, tbl->table);
+
+	return 0;
+}
+
+void pvrdma_uar_table_cleanup(struct pvrdma_dev *dev)
+{
+	struct pvrdma_id_table *tbl = &dev->uar_table.tbl;
+
+	kfree(tbl->table);
+}
+
+int pvrdma_uar_alloc(struct pvrdma_dev *dev, struct pvrdma_uar_map *uar)
+{
+	struct pvrdma_id_table *tbl;
+	unsigned long flags;
+	u32 obj;
+
+	tbl = &dev->uar_table.tbl;
+
+	spin_lock_irqsave(&tbl->lock, flags);
+	obj = find_next_zero_bit(tbl->table, tbl->max, tbl->last);
+	if (obj >= tbl->max) {
+		tbl->top = (tbl->top + tbl->max) & tbl->mask;
+		obj = find_first_zero_bit(tbl->table, tbl->max);
+	}
+
+	if (obj < tbl->max) {
+		set_bit(obj, tbl->table);
+		obj |= tbl->top;
+	} else {
+		spin_unlock_irqrestore(&tbl->lock, flags);
+		return -ENOMEM;
+	}
+
+	spin_unlock_irqrestore(&tbl->lock, flags);
+
+	uar->index = obj;
+	uar->pfn = (pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_UAR) >>
+		    PAGE_SHIFT) + uar->index;
+
+	return 0;
+}
+
+void pvrdma_uar_free(struct pvrdma_dev *dev, struct pvrdma_uar_map *uar)
+{
+	struct pvrdma_id_table *tbl = &dev->uar_table.tbl;
+	unsigned long flags;
+	u32 obj;
+
+	obj = uar->index & (tbl->max - 1);
+	spin_lock_irqsave(&tbl->lock, flags);
+	clear_bit(obj, tbl->table);
+	tbl->last = min(tbl->last, obj);
+	tbl->top = (tbl->top + tbl->max) & tbl->mask;
+	spin_unlock_irqrestore(&tbl->lock, flags);
+}
-- 
2.7.4

--
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 06/15] IB/pvrdma: Add virtual device RDMA structures

[Adit Ranadive]

This patch adds the various Verbs structures that we support in the
virtual RDMA device. We have re-mapped the ones from the RDMA core stack
to make sure we can maintain compatibility with our backend.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/pvrdma/pvrdma_ib_verbs.h | 450 +++++++++++++++++++++++++
 1 file changed, 450 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_ib_verbs.h

diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_ib_verbs.h b/drivers/infiniband/hw/pvrdma/pvrdma_ib_verbs.h
new file mode 100644
index 0000000..f97d43d
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_ib_verbs.h
@@ -0,0 +1,450 @@
+/*
+ * [PLEASE NOTE:  VMWARE, INC. ELECTS TO USE AND DISTRIBUTE THIS COMPONENT
+ * UNDER THE TERMS OF THE OpenIB.org BSD license.  THE ORIGINAL LICENSE TERMS
+ * ARE REPRODUCED BELOW ONLY AS A REFERENCE.]
+ *
+ * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
+ * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
+ * Copyright (c) 2004 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
+ * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
+ * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
+ * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
+ * Copyright (c) 2015-2016 VMware, 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 __PVRDMA_IB_VERBS_H__
+#define __PVRDMA_IB_VERBS_H__
+
+#include <linux/types.h>
+
+union pvrdma_gid {
+	__u8	raw[16];
+	struct {
+		__be64	subnet_prefix;
+		__be64	interface_id;
+	} global;
+};
+
+enum pvrdma_link_layer {
+	PVRDMA_LINK_LAYER_UNSPECIFIED,
+	PVRDMA_LINK_LAYER_INFINIBAND,
+	PVRDMA_LINK_LAYER_ETHERNET,
+};
+
+enum pvrdma_mtu {
+	PVRDMA_MTU_256  = 1,
+	PVRDMA_MTU_512  = 2,
+	PVRDMA_MTU_1024 = 3,
+	PVRDMA_MTU_2048 = 4,
+	PVRDMA_MTU_4096 = 5
+};
+
+static inline int pvrdma_mtu_enum_to_int(enum pvrdma_mtu mtu)
+{
+	switch (mtu) {
+	case PVRDMA_MTU_256:	return  256;
+	case PVRDMA_MTU_512:	return  512;
+	case PVRDMA_MTU_1024:	return 1024;
+	case PVRDMA_MTU_2048:	return 2048;
+	case PVRDMA_MTU_4096:	return 4096;
+	default:		return   -1;
+	}
+}
+
+static inline enum pvrdma_mtu pvrdma_mtu_int_to_enum(int mtu)
+{
+	switch (mtu) {
+	case 256:	return PVRDMA_MTU_256;
+	case 512:	return PVRDMA_MTU_512;
+	case 1024:	return PVRDMA_MTU_1024;
+	case 2048:	return PVRDMA_MTU_2048;
+	case 4096:
+	default:	return PVRDMA_MTU_4096;
+	}
+}
+
+enum pvrdma_port_state {
+	PVRDMA_PORT_NOP			= 0,
+	PVRDMA_PORT_DOWN		= 1,
+	PVRDMA_PORT_INIT		= 2,
+	PVRDMA_PORT_ARMED		= 3,
+	PVRDMA_PORT_ACTIVE		= 4,
+	PVRDMA_PORT_ACTIVE_DEFER	= 5
+};
+
+enum pvrdma_port_cap_flags {
+	PVRDMA_PORT_SM				= 1 <<  1,
+	PVRDMA_PORT_NOTICE_SUP			= 1 <<  2,
+	PVRDMA_PORT_TRAP_SUP			= 1 <<  3,
+	PVRDMA_PORT_OPT_IPD_SUP			= 1 <<  4,
+	PVRDMA_PORT_AUTO_MIGR_SUP		= 1 <<  5,
+	PVRDMA_PORT_SL_MAP_SUP			= 1 <<  6,
+	PVRDMA_PORT_MKEY_NVRAM			= 1 <<  7,
+	PVRDMA_PORT_PKEY_NVRAM			= 1 <<  8,
+	PVRDMA_PORT_LED_INFO_SUP		= 1 <<  9,
+	PVRDMA_PORT_SM_DISABLED			= 1 << 10,
+	PVRDMA_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
+	PVRDMA_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
+	PVRDMA_PORT_EXTENDED_SPEEDS_SUP		= 1 << 14,
+	PVRDMA_PORT_CM_SUP			= 1 << 16,
+	PVRDMA_PORT_SNMP_TUNNEL_SUP		= 1 << 17,
+	PVRDMA_PORT_REINIT_SUP			= 1 << 18,
+	PVRDMA_PORT_DEVICE_MGMT_SUP		= 1 << 19,
+	PVRDMA_PORT_VENDOR_CLASS_SUP		= 1 << 20,
+	PVRDMA_PORT_DR_NOTICE_SUP		= 1 << 21,
+	PVRDMA_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
+	PVRDMA_PORT_BOOT_MGMT_SUP		= 1 << 23,
+	PVRDMA_PORT_LINK_LATENCY_SUP		= 1 << 24,
+	PVRDMA_PORT_CLIENT_REG_SUP		= 1 << 25,
+	PVRDMA_PORT_IP_BASED_GIDS		= 1 << 26,
+	PVRDMA_PORT_CAP_FLAGS_MAX		= PVRDMA_PORT_IP_BASED_GIDS,
+};
+
+enum pvrdma_port_width {
+	PVRDMA_WIDTH_1X		= 1,
+	PVRDMA_WIDTH_4X		= 2,
+	PVRDMA_WIDTH_8X		= 4,
+	PVRDMA_WIDTH_12X	= 8
+};
+
+static inline int pvrdma_width_enum_to_int(enum pvrdma_port_width width)
+{
+	switch (width) {
+	case PVRDMA_WIDTH_1X:	return  1;
+	case PVRDMA_WIDTH_4X:	return  4;
+	case PVRDMA_WIDTH_8X:	return  8;
+	case PVRDMA_WIDTH_12X:	return 12;
+	default:		return -1;
+	}
+}
+
+enum pvrdma_port_speed {
+	PVRDMA_SPEED_SDR	= 1,
+	PVRDMA_SPEED_DDR	= 2,
+	PVRDMA_SPEED_QDR	= 4,
+	PVRDMA_SPEED_FDR10	= 8,
+	PVRDMA_SPEED_FDR	= 16,
+	PVRDMA_SPEED_EDR	= 32
+};
+
+struct pvrdma_port_attr {
+	enum pvrdma_port_state	state;
+	enum pvrdma_mtu		max_mtu;
+	enum pvrdma_mtu		active_mtu;
+	__u32			gid_tbl_len;
+	__u32			port_cap_flags;
+	__u32			max_msg_sz;
+	__u32			bad_pkey_cntr;
+	__u32			qkey_viol_cntr;
+	__u16			pkey_tbl_len;
+	__u16			lid;
+	__u16			sm_lid;
+	__u8			lmc;
+	__u8			max_vl_num;
+	__u8			sm_sl;
+	__u8			subnet_timeout;
+	__u8			init_type_reply;
+	__u8			active_width;
+	__u8			active_speed;
+	__u8			phys_state;
+	__u8			reserved[2];
+};
+
+struct pvrdma_global_route {
+	union pvrdma_gid	dgid;
+	__u32			flow_label;
+	__u8			sgid_index;
+	__u8			hop_limit;
+	__u8			traffic_class;
+	__u8			reserved;
+};
+
+struct pvrdma_grh {
+	__be32			version_tclass_flow;
+	__be16			paylen;
+	__u8			next_hdr;
+	__u8			hop_limit;
+	union pvrdma_gid	sgid;
+	union pvrdma_gid	dgid;
+};
+
+enum pvrdma_ah_flags {
+	PVRDMA_AH_GRH = 1
+};
+
+enum pvrdma_rate {
+	PVRDMA_RATE_PORT_CURRENT = 0,
+	PVRDMA_RATE_2_5_GBPS = 2,
+	PVRDMA_RATE_5_GBPS   = 5,
+	PVRDMA_RATE_10_GBPS  = 3,
+	PVRDMA_RATE_20_GBPS  = 6,
+	PVRDMA_RATE_30_GBPS  = 4,
+	PVRDMA_RATE_40_GBPS  = 7,
+	PVRDMA_RATE_60_GBPS  = 8,
+	PVRDMA_RATE_80_GBPS  = 9,
+	PVRDMA_RATE_120_GBPS = 10,
+	PVRDMA_RATE_14_GBPS  = 11,
+	PVRDMA_RATE_56_GBPS  = 12,
+	PVRDMA_RATE_112_GBPS = 13,
+	PVRDMA_RATE_168_GBPS = 14,
+	PVRDMA_RATE_25_GBPS  = 15,
+	PVRDMA_RATE_100_GBPS = 16,
+	PVRDMA_RATE_200_GBPS = 17,
+	PVRDMA_RATE_300_GBPS = 18
+};
+
+struct pvrdma_ah_attr {
+	struct pvrdma_global_route	grh;
+	__u16				dlid;
+	__u16				vlan_id;
+	__u8				sl;
+	__u8				src_path_bits;
+	__u8				static_rate;
+	__u8				ah_flags;
+	__u8				port_num;
+	__u8				dmac[6];
+	__u8				reserved;
+};
+
+enum pvrdma_wc_status {
+	PVRDMA_WC_SUCCESS,
+	PVRDMA_WC_LOC_LEN_ERR,
+	PVRDMA_WC_LOC_QP_OP_ERR,
+	PVRDMA_WC_LOC_EEC_OP_ERR,
+	PVRDMA_WC_LOC_PROT_ERR,
+	PVRDMA_WC_WR_FLUSH_ERR,
+	PVRDMA_WC_MW_BIND_ERR,
+	PVRDMA_WC_BAD_RESP_ERR,
+	PVRDMA_WC_LOC_ACCESS_ERR,
+	PVRDMA_WC_REM_INV_REQ_ERR,
+	PVRDMA_WC_REM_ACCESS_ERR,
+	PVRDMA_WC_REM_OP_ERR,
+	PVRDMA_WC_RETRY_EXC_ERR,
+	PVRDMA_WC_RNR_RETRY_EXC_ERR,
+	PVRDMA_WC_LOC_RDD_VIOL_ERR,
+	PVRDMA_WC_REM_INV_RD_REQ_ERR,
+	PVRDMA_WC_REM_ABORT_ERR,
+	PVRDMA_WC_INV_EECN_ERR,
+	PVRDMA_WC_INV_EEC_STATE_ERR,
+	PVRDMA_WC_FATAL_ERR,
+	PVRDMA_WC_RESP_TIMEOUT_ERR,
+	PVRDMA_WC_GENERAL_ERR
+};
+
+enum pvrdma_wc_opcode {
+	PVRDMA_WC_SEND,
+	PVRDMA_WC_RDMA_WRITE,
+	PVRDMA_WC_RDMA_READ,
+	PVRDMA_WC_COMP_SWAP,
+	PVRDMA_WC_FETCH_ADD,
+	PVRDMA_WC_BIND_MW,
+	PVRDMA_WC_LSO,
+	PVRDMA_WC_LOCAL_INV,
+	PVRDMA_WC_FAST_REG_MR,
+	PVRDMA_WC_MASKED_COMP_SWAP,
+	PVRDMA_WC_MASKED_FETCH_ADD,
+	PVRDMA_WC_RECV = 1 << 7,
+	PVRDMA_WC_RECV_RDMA_WITH_IMM
+};
+
+enum pvrdma_wc_flags {
+	PVRDMA_WC_GRH			= 1 << 0,
+	PVRDMA_WC_WITH_IMM		= 1 << 1,
+	PVRDMA_WC_WITH_INVALIDATE	= 1 << 2,
+	PVRDMA_WC_IP_CSUM_OK		= 1 << 3,
+	PVRDMA_WC_WITH_SMAC		= 1 << 4,
+	PVRDMA_WC_WITH_VLAN		= 1 << 5,
+	PVRDMA_WC_FLAGS_MAX		= PVRDMA_WC_WITH_VLAN,
+};
+
+enum pvrdma_cq_notify_flags {
+	PVRDMA_CQ_SOLICITED		= 1 << 0,
+	PVRDMA_CQ_NEXT_COMP		= 1 << 1,
+	PVRDMA_CQ_SOLICITED_MASK	= PVRDMA_CQ_SOLICITED |
+					  PVRDMA_CQ_NEXT_COMP,
+	PVRDMA_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
+};
+
+struct pvrdma_qp_cap {
+	__u32	max_send_wr;
+	__u32	max_recv_wr;
+	__u32	max_send_sge;
+	__u32	max_recv_sge;
+	__u32	max_inline_data;
+	__u32	reserved;
+};
+
+enum pvrdma_sig_type {
+	PVRDMA_SIGNAL_ALL_WR,
+	PVRDMA_SIGNAL_REQ_WR
+};
+
+enum pvrdma_qp_type {
+	PVRDMA_QPT_SMI,
+	PVRDMA_QPT_GSI,
+	PVRDMA_QPT_RC,
+	PVRDMA_QPT_UC,
+	PVRDMA_QPT_UD,
+	PVRDMA_QPT_RAW_IPV6,
+	PVRDMA_QPT_RAW_ETHERTYPE,
+	PVRDMA_QPT_RAW_PACKET = 8,
+	PVRDMA_QPT_XRC_INI = 9,
+	PVRDMA_QPT_XRC_TGT,
+	PVRDMA_QPT_MAX,
+};
+
+enum pvrdma_qp_create_flags {
+	PVRDMA_QP_CREATE_IPOPVRDMA_UD_LSO		= 1 << 0,
+	PVRDMA_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
+};
+
+enum pvrdma_qp_attr_mask {
+	PVRDMA_QP_STATE			= 1 << 0,
+	PVRDMA_QP_CUR_STATE		= 1 << 1,
+	PVRDMA_QP_EN_SQD_ASYNC_NOTIFY	= 1 << 2,
+	PVRDMA_QP_ACCESS_FLAGS		= 1 << 3,
+	PVRDMA_QP_PKEY_INDEX		= 1 << 4,
+	PVRDMA_QP_PORT			= 1 << 5,
+	PVRDMA_QP_QKEY			= 1 << 6,
+	PVRDMA_QP_AV			= 1 << 7,
+	PVRDMA_QP_PATH_MTU		= 1 << 8,
+	PVRDMA_QP_TIMEOUT		= 1 << 9,
+	PVRDMA_QP_RETRY_CNT		= 1 << 10,
+	PVRDMA_QP_RNR_RETRY		= 1 << 11,
+	PVRDMA_QP_RQ_PSN		= 1 << 12,
+	PVRDMA_QP_MAX_QP_RD_ATOMIC	= 1 << 13,
+	PVRDMA_QP_ALT_PATH		= 1 << 14,
+	PVRDMA_QP_MIN_RNR_TIMER		= 1 << 15,
+	PVRDMA_QP_SQ_PSN		= 1 << 16,
+	PVRDMA_QP_MAX_DEST_RD_ATOMIC	= 1 << 17,
+	PVRDMA_QP_PATH_MIG_STATE	= 1 << 18,
+	PVRDMA_QP_CAP			= 1 << 19,
+	PVRDMA_QP_DEST_QPN		= 1 << 20,
+	PVRDMA_QP_ATTR_MASK_MAX		= PVRDMA_QP_DEST_QPN,
+};
+
+enum pvrdma_qp_state {
+	PVRDMA_QPS_RESET,
+	PVRDMA_QPS_INIT,
+	PVRDMA_QPS_RTR,
+	PVRDMA_QPS_RTS,
+	PVRDMA_QPS_SQD,
+	PVRDMA_QPS_SQE,
+	PVRDMA_QPS_ERR
+};
+
+enum pvrdma_mig_state {
+	PVRDMA_MIG_MIGRATED,
+	PVRDMA_MIG_REARM,
+	PVRDMA_MIG_ARMED
+};
+
+enum pvrdma_mw_type {
+	PVRDMA_MW_TYPE_1 = 1,
+	PVRDMA_MW_TYPE_2 = 2
+};
+
+struct pvrdma_qp_attr {
+	enum pvrdma_qp_state	qp_state;
+	enum pvrdma_qp_state	cur_qp_state;
+	enum pvrdma_mtu		path_mtu;
+	enum pvrdma_mig_state	path_mig_state;
+	__u32			qkey;
+	__u32			rq_psn;
+	__u32			sq_psn;
+	__u32			dest_qp_num;
+	__u32			qp_access_flags;
+	__u16			pkey_index;
+	__u16			alt_pkey_index;
+	__u8			en_sqd_async_notify;
+	__u8			sq_draining;
+	__u8			max_rd_atomic;
+	__u8			max_dest_rd_atomic;
+	__u8			min_rnr_timer;
+	__u8			port_num;
+	__u8			timeout;
+	__u8			retry_cnt;
+	__u8			rnr_retry;
+	__u8			alt_port_num;
+	__u8			alt_timeout;
+	__u8			reserved[5];
+	struct pvrdma_qp_cap	cap;
+	struct pvrdma_ah_attr	ah_attr;
+	struct pvrdma_ah_attr	alt_ah_attr;
+};
+
+enum pvrdma_wr_opcode {
+	PVRDMA_WR_RDMA_WRITE,
+	PVRDMA_WR_RDMA_WRITE_WITH_IMM,
+	PVRDMA_WR_SEND,
+	PVRDMA_WR_SEND_WITH_IMM,
+	PVRDMA_WR_RDMA_READ,
+	PVRDMA_WR_ATOMIC_CMP_AND_SWP,
+	PVRDMA_WR_ATOMIC_FETCH_AND_ADD,
+	PVRDMA_WR_LSO,
+	PVRDMA_WR_SEND_WITH_INV,
+	PVRDMA_WR_RDMA_READ_WITH_INV,
+	PVRDMA_WR_LOCAL_INV,
+	PVRDMA_WR_FAST_REG_MR,
+	PVRDMA_WR_MASKED_ATOMIC_CMP_AND_SWP,
+	PVRDMA_WR_MASKED_ATOMIC_FETCH_AND_ADD,
+	PVRDMA_WR_BIND_MW,
+	PVRDMA_WR_REG_SIG_MR,
+};
+
+enum pvrdma_send_flags {
+	PVRDMA_SEND_FENCE	= 1 << 0,
+	PVRDMA_SEND_SIGNALED	= 1 << 1,
+	PVRDMA_SEND_SOLICITED	= 1 << 2,
+	PVRDMA_SEND_INLINE	= 1 << 3,
+	PVRDMA_SEND_IP_CSUM	= 1 << 4,
+	PVRDMA_SEND_FLAGS_MAX	= PVRDMA_SEND_IP_CSUM,
+};
+
+struct pvrdma_sge {
+	__u64	addr;
+	__u32	length;
+	__u32	lkey;
+};
+
+enum pvrdma_access_flags {
+	PVRDMA_ACCESS_LOCAL_WRITE	= 1 << 0,
+	PVRDMA_ACCESS_REMOTE_WRITE	= 1 << 1,
+	PVRDMA_ACCESS_REMOTE_READ	= 1 << 2,
+	PVRDMA_ACCESS_REMOTE_ATOMIC	= 1 << 3,
+	PVRDMA_ACCESS_MW_BIND		= 1 << 4,
+	PVRDMA_ZERO_BASED		= 1 << 5,
+	PVRDMA_ACCESS_ON_DEMAND		= 1 << 6,
+	PVRDMA_ACCESS_FLAGS_MAX		= PVRDMA_ACCESS_ON_DEMAND,
+};
+
+#endif /* __PVRDMA_IB_VERBS_H__ */
-- 
2.7.4

--
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 07/15] IB/pvrdma: Add the main driver module for PVRDMA

[Adit Ranadive]

This patch adds the support to register a RDMA device with the kernel RDMA
stack as well as a kernel module. This also initializes the underlying
virtual PCI device.

Changes from v1: Addressed 32-bit build errors. Addressed comments by
Yuval Shaia.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/pvrdma/pvrdma_main.c | 1189 ++++++++++++++++++++++++++++
 1 file changed, 1189 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_main.c

diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_main.c b/drivers/infiniband/hw/pvrdma/pvrdma_main.c
new file mode 100644
index 0000000..20073b7
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_main.c
@@ -0,0 +1,1189 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <linux/errno.h>
+#include <linux/inetdevice.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <rdma/ib_addr.h>
+#include <rdma/ib_smi.h>
+#include <rdma/ib_user_verbs.h>
+#include <net/addrconf.h>
+
+#include "pvrdma.h"
+#include "pvrdma_user.h"
+
+#define PCI_DEVICE_ID_VMWARE_VMXNET3	0x07B0
+
+#define DRV_NAME	"pvrdma"
+#define DRV_VERSION	"1.0"
+#define DRV_RELDATE	"January 1, 2013"
+
+static const char pvrdma_version[] =
+	DRV_NAME ": PVRDMA InfiniBand driver v"
+	DRV_VERSION " (" DRV_RELDATE ")\n";
+
+static DEFINE_MUTEX(pvrdma_device_list_lock);
+static LIST_HEAD(pvrdma_device_list);
+
+static int pvrdma_add_gid(struct ib_device *ibdev,
+			  u8 port_num,
+			  unsigned int index,
+			  const union ib_gid *gid,
+			  const struct ib_gid_attr *attr,
+			  void **context);
+static int pvrdma_del_gid(struct ib_device *ibdev,
+			  u8 port_num,
+			  unsigned int index,
+			  void **context);
+
+
+static ssize_t show_hca(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	return sprintf(buf, "PVRDMA%s\n", DRV_VERSION);
+}
+
+static ssize_t show_fw_ver(struct device *device,
+			   struct device_attribute *attr, char *buf)
+{
+	struct pvrdma_dev *dev =
+		container_of(device, struct pvrdma_dev, ib_dev.dev);
+
+	return sprintf(buf, "%d.%d.%d\n",
+		       (int)(dev->dsr->caps.fw_ver >> 32),
+		       (int)(dev->dsr->caps.fw_ver >> 16) & 0xffff,
+		       (int)dev->dsr->caps.fw_ver & 0xffff);
+}
+
+static ssize_t show_rev(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	return sprintf(buf, "%d\n", PVRDMA_REV_ID);
+}
+
+static ssize_t show_board(struct device *device, struct device_attribute *attr,
+			  char *buf)
+{
+	return sprintf(buf, "%d\n", PVRDMA_BOARD_ID);
+}
+
+static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,	   NULL);
+static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
+static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,	   NULL);
+static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
+
+static struct device_attribute *pvrdma_class_attributes[] = {
+	&dev_attr_hw_rev,
+	&dev_attr_fw_ver,
+	&dev_attr_hca_type,
+	&dev_attr_board_id
+};
+
+static int pvrdma_init_device(struct pvrdma_dev *dev)
+{
+	/*  Initialize some device related stuff */
+	spin_lock_init(&dev->cmd_lock);
+	sema_init(&dev->cmd_sema, 1);
+	atomic_set(&dev->num_qps, 0);
+	atomic_set(&dev->num_cqs, 0);
+	atomic_set(&dev->num_pds, 0);
+	atomic_set(&dev->num_ahs, 0);
+
+	return 0;
+}
+
+static int pvrdma_port_immutable(struct ib_device *ibdev, u8 port_num,
+				 struct ib_port_immutable *immutable)
+{
+	struct ib_port_attr attr;
+	int err;
+
+	err = pvrdma_query_port(ibdev, port_num, &attr);
+	if (err)
+		return err;
+
+	immutable->pkey_tbl_len = attr.pkey_tbl_len;
+	immutable->gid_tbl_len = attr.gid_tbl_len;
+	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
+	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
+	return 0;
+}
+
+static struct net_device *pvrdma_get_netdev(struct ib_device *ibdev,
+					    u8 port_num)
+{
+	struct net_device *netdev;
+	struct pvrdma_dev *dev = to_vdev(ibdev);
+
+	if (port_num != 1)
+		return NULL;
+
+	rcu_read_lock();
+	netdev = dev->netdev;
+	if (netdev)
+		dev_hold(netdev);
+	rcu_read_unlock();
+
+	return netdev;
+}
+
+static int pvrdma_register_device(struct pvrdma_dev *dev)
+{
+	int ret = -1;
+	int i = 0;
+
+	strlcpy(dev->ib_dev.name, "pvrdma%d", IB_DEVICE_NAME_MAX);
+	dev->ib_dev.node_guid = dev->dsr->caps.node_guid;
+	dev->sys_image_guid = dev->dsr->caps.sys_image_guid;
+	dev->flags = 0;
+	dev->ib_dev.owner = THIS_MODULE;
+	dev->ib_dev.num_comp_vectors = 1;
+	dev->ib_dev.dma_device = &dev->pdev->dev;
+	dev->ib_dev.uverbs_abi_ver = PVRDMA_UVERBS_ABI_VERSION;
+	dev->ib_dev.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_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_POLL_CQ)		|
+		(1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ)	|
+		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
+		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
+		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
+		(1ull << IB_USER_VERBS_CMD_QUERY_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_CREATE_AH)		|
+		(1ull << IB_USER_VERBS_CMD_DESTROY_AH);
+
+	dev->ib_dev.node_type = RDMA_NODE_IB_CA;
+	dev->ib_dev.phys_port_cnt = dev->dsr->caps.phys_port_cnt;
+
+	dev->ib_dev.query_device = pvrdma_query_device;
+	dev->ib_dev.query_port = pvrdma_query_port;
+	dev->ib_dev.query_gid = pvrdma_query_gid;
+	dev->ib_dev.query_pkey = pvrdma_query_pkey;
+	dev->ib_dev.modify_port	= pvrdma_modify_port;
+	dev->ib_dev.alloc_ucontext = pvrdma_alloc_ucontext;
+	dev->ib_dev.dealloc_ucontext = pvrdma_dealloc_ucontext;
+	dev->ib_dev.mmap = pvrdma_mmap;
+	dev->ib_dev.alloc_pd = pvrdma_alloc_pd;
+	dev->ib_dev.dealloc_pd = pvrdma_dealloc_pd;
+	dev->ib_dev.create_ah = pvrdma_create_ah;
+	dev->ib_dev.destroy_ah = pvrdma_destroy_ah;
+	dev->ib_dev.create_qp = pvrdma_create_qp;
+	dev->ib_dev.modify_qp = pvrdma_modify_qp;
+	dev->ib_dev.query_qp = pvrdma_query_qp;
+	dev->ib_dev.destroy_qp = pvrdma_destroy_qp;
+	dev->ib_dev.post_send = pvrdma_post_send;
+	dev->ib_dev.post_recv = pvrdma_post_recv;
+	dev->ib_dev.create_cq = pvrdma_create_cq;
+	dev->ib_dev.modify_cq = pvrdma_modify_cq;
+	dev->ib_dev.resize_cq = pvrdma_resize_cq;
+	dev->ib_dev.destroy_cq = pvrdma_destroy_cq;
+	dev->ib_dev.poll_cq = pvrdma_poll_cq;
+	dev->ib_dev.req_notify_cq = pvrdma_req_notify_cq;
+	dev->ib_dev.get_dma_mr = pvrdma_get_dma_mr;
+	dev->ib_dev.reg_user_mr	= pvrdma_reg_user_mr;
+	dev->ib_dev.dereg_mr = pvrdma_dereg_mr;
+	dev->ib_dev.alloc_mr = pvrdma_alloc_mr;
+	dev->ib_dev.map_mr_sg = pvrdma_map_mr_sg;
+	dev->ib_dev.add_gid = pvrdma_add_gid;
+	dev->ib_dev.del_gid = pvrdma_del_gid;
+	dev->ib_dev.get_netdev = pvrdma_get_netdev;
+	dev->ib_dev.get_port_immutable = pvrdma_port_immutable;
+	dev->ib_dev.get_link_layer = pvrdma_port_link_layer;
+
+	mutex_init(&dev->port_mutex);
+	spin_lock_init(&dev->desc_lock);
+
+	dev->cq_tbl = kcalloc(dev->dsr->caps.max_cq, sizeof(void *),
+			      GFP_KERNEL);
+	if (!dev->cq_tbl)
+		return ret;
+	spin_lock_init(&dev->cq_tbl_lock);
+
+	dev->qp_tbl = kcalloc(dev->dsr->caps.max_qp, sizeof(void *),
+			      GFP_KERNEL);
+	if (!dev->qp_tbl)
+		goto err_cq_free;
+	spin_lock_init(&dev->qp_tbl_lock);
+
+	ret = ib_register_device(&dev->ib_dev, NULL);
+	if (ret)
+		goto err_qp_free;
+
+	for (i = 0; i < ARRAY_SIZE(pvrdma_class_attributes); ++i) {
+		ret = device_create_file(&dev->ib_dev.dev,
+					 pvrdma_class_attributes[i]);
+		if (ret)
+			goto err_class;
+	}
+
+	dev->ib_active = true;
+
+	return 0;
+
+err_class:
+	ib_unregister_device(&dev->ib_dev);
+err_qp_free:
+	kfree(dev->qp_tbl);
+err_cq_free:
+	kfree(dev->cq_tbl);
+
+	return ret;
+}
+
+static irqreturn_t pvrdma_intr0_handler(int irq, void *dev_id)
+{
+	u32 icr = PVRDMA_INTR_CAUSE_RESPONSE;
+	struct pvrdma_dev *dev = dev_id;
+
+	dev_dbg(&dev->pdev->dev, "interrupt 0 (response) handler\n");
+
+	if (dev->intr.type != PVRDMA_INTR_TYPE_MSIX) {
+		/* Legacy intr */
+		icr = pvrdma_read_reg(dev, PVRDMA_REG_ICR);
+		if (icr == 0)
+			return IRQ_NONE;
+	}
+
+	if (icr == PVRDMA_INTR_CAUSE_RESPONSE)
+		complete(&dev->cmd_done);
+
+	return IRQ_HANDLED;
+}
+
+static void pvrdma_qp_event(struct pvrdma_dev *dev, u32 qpn, int type)
+{
+	struct pvrdma_qp *qp;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dev->qp_tbl_lock, flags);
+	qp = dev->qp_tbl[qpn % dev->dsr->caps.max_qp];
+	if (qp)
+		atomic_inc(&qp->refcnt);
+	spin_unlock_irqrestore(&dev->qp_tbl_lock, flags);
+
+	if (qp && qp->ibqp.event_handler) {
+		struct ib_qp *ibqp = &qp->ibqp;
+		struct ib_event e;
+
+		e.device = ibqp->device;
+		e.element.qp = ibqp;
+		e.event = type; /* 1:1 mapping for now. */
+		ibqp->event_handler(&e, ibqp->qp_context);
+	}
+	if (qp) {
+		atomic_dec(&qp->refcnt);
+		if (atomic_read(&qp->refcnt) == 0)
+			wake_up(&qp->wait);
+	}
+}
+
+static void pvrdma_cq_event(struct pvrdma_dev *dev, u32 cqn, int type)
+{
+	struct pvrdma_cq *cq;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dev->cq_tbl_lock, flags);
+	cq = dev->cq_tbl[cqn % dev->dsr->caps.max_cq];
+	if (cq)
+		atomic_inc(&cq->refcnt);
+	spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
+
+	if (cq && cq->ibcq.event_handler) {
+		struct ib_cq *ibcq = &cq->ibcq;
+		struct ib_event e;
+
+		e.device = ibcq->device;
+		e.element.cq = ibcq;
+		e.event = type; /* 1:1 mapping for now. */
+		ibcq->event_handler(&e, ibcq->cq_context);
+	}
+	if (cq) {
+		atomic_dec(&cq->refcnt);
+		if (atomic_read(&cq->refcnt) == 0)
+			wake_up(&cq->wait);
+	}
+}
+
+static void pvrdma_dispatch_event(struct pvrdma_dev *dev, int port,
+				  enum ib_event_type event)
+{
+	struct ib_event ib_event;
+
+	memset(&ib_event, 0, sizeof(ib_event));
+	ib_event.device = &dev->ib_dev;
+	ib_event.element.port_num = port;
+	ib_event.event = event;
+	ib_dispatch_event(&ib_event);
+}
+
+static void pvrdma_dev_event(struct pvrdma_dev *dev, u8 port, int type)
+{
+	if (port < 1 || port > dev->dsr->caps.phys_port_cnt) {
+		dev_warn(&dev->pdev->dev, "event on port %d\n", port);
+		return;
+	}
+
+	pvrdma_dispatch_event(dev, port, type);
+}
+
+static inline struct pvrdma_eqe *get_eqe(struct pvrdma_dev *dev, unsigned int i)
+{
+	return (struct pvrdma_eqe *)pvrdma_page_dir_get_ptr(
+					&dev->async_pdir,
+					PAGE_SIZE +
+					sizeof(struct pvrdma_eqe) * i);
+}
+
+static irqreturn_t pvrdma_intr1_handler(int irq, void *dev_id)
+{
+	struct pvrdma_dev *dev = dev_id;
+	struct pvrdma_ring *ring = &dev->async_ring_state->rx;
+	int ring_slots = (dev->dsr->async_ring_pages.num_pages - 1) *
+			 PAGE_SIZE / sizeof(struct pvrdma_eqe);
+	unsigned int head;
+
+	dev_dbg(&dev->pdev->dev, "interrupt 1 (async event) handler\n");
+
+	/*
+	 * Don't process events until the IB device is registered. Otherwise
+	 * we'll try to ib_dispatch_event() on an invalid device.
+	 */
+	if (!dev->ib_active)
+		return IRQ_HANDLED;
+
+	while (pvrdma_idx_ring_has_data(ring, ring_slots, &head) > 0) {
+		struct pvrdma_eqe *eqe;
+
+		eqe = get_eqe(dev, head);
+
+		switch (eqe->type) {
+		case PVRDMA_EVENT_QP_FATAL:
+		case PVRDMA_EVENT_QP_REQ_ERR:
+		case PVRDMA_EVENT_QP_ACCESS_ERR:
+		case PVRDMA_EVENT_COMM_EST:
+		case PVRDMA_EVENT_SQ_DRAINED:
+		case PVRDMA_EVENT_PATH_MIG:
+		case PVRDMA_EVENT_PATH_MIG_ERR:
+		case PVRDMA_EVENT_QP_LAST_WQE_REACHED:
+			pvrdma_qp_event(dev, eqe->info, eqe->type);
+			break;
+
+		case PVRDMA_EVENT_CQ_ERR:
+			pvrdma_cq_event(dev, eqe->info, eqe->type);
+			break;
+
+		case PVRDMA_EVENT_SRQ_ERR:
+		case PVRDMA_EVENT_SRQ_LIMIT_REACHED:
+			break;
+
+		case PVRDMA_EVENT_PORT_ACTIVE:
+		case PVRDMA_EVENT_PORT_ERR:
+		case PVRDMA_EVENT_LID_CHANGE:
+		case PVRDMA_EVENT_PKEY_CHANGE:
+		case PVRDMA_EVENT_SM_CHANGE:
+		case PVRDMA_EVENT_CLIENT_REREGISTER:
+		case PVRDMA_EVENT_GID_CHANGE:
+			pvrdma_dev_event(dev, eqe->info, eqe->type);
+			break;
+
+		case PVRDMA_EVENT_DEVICE_FATAL:
+			pvrdma_dev_event(dev, 1, eqe->type);
+			break;
+
+		default:
+			break;
+		}
+
+		pvrdma_idx_ring_inc(&ring->cons_head, ring_slots);
+	}
+
+	return IRQ_HANDLED;
+}
+
+static inline struct pvrdma_cqne *get_cqne(struct pvrdma_dev *dev,
+					  unsigned int i)
+{
+	return (struct pvrdma_cqne *)pvrdma_page_dir_get_ptr(
+					&dev->cq_pdir,
+					PAGE_SIZE +
+					sizeof(struct pvrdma_cqne) * i);
+}
+
+static irqreturn_t pvrdma_intrx_handler(int irq, void *dev_id)
+{
+	struct pvrdma_dev *dev = dev_id;
+	struct pvrdma_ring *ring = &dev->cq_ring_state->rx;
+	int ring_slots = (dev->dsr->cq_ring_pages.num_pages - 1) * PAGE_SIZE /
+			 sizeof(struct pvrdma_cqne);
+	unsigned int head;
+	unsigned long flags;
+
+	dev_dbg(&dev->pdev->dev, "interrupt x (completion) handler\n");
+
+	while (pvrdma_idx_ring_has_data(ring, ring_slots, &head) > 0) {
+		struct pvrdma_cqne *cqne;
+		struct pvrdma_cq *cq;
+
+		cqne = get_cqne(dev, head);
+		spin_lock_irqsave(&dev->cq_tbl_lock, flags);
+		cq = dev->cq_tbl[cqne->info % dev->dsr->caps.max_cq];
+		if (cq)
+			atomic_inc(&cq->refcnt);
+		spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
+
+		if (cq && cq->ibcq.comp_handler)
+			cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
+		if (cq) {
+			atomic_dec(&cq->refcnt);
+			if (atomic_read(&cq->refcnt))
+				wake_up(&cq->wait);
+		}
+		pvrdma_idx_ring_inc(&ring->cons_head, ring_slots);
+	}
+
+	return IRQ_HANDLED;
+}
+
+static void pvrdma_disable_msi_all(struct pvrdma_dev *dev)
+{
+	if (dev->intr.type == PVRDMA_INTR_TYPE_MSIX)
+		pci_disable_msix(dev->pdev);
+	else if (dev->intr.type == PVRDMA_INTR_TYPE_MSI)
+		pci_disable_msi(dev->pdev);
+}
+
+static void pvrdma_free_irq(struct pvrdma_dev *dev)
+{
+	int i;
+
+	dev_dbg(&dev->pdev->dev, "freeing interrupts\n");
+
+	if (dev->intr.type == PVRDMA_INTR_TYPE_MSIX) {
+		for (i = 0; i < dev->intr.size; i++) {
+			if (dev->intr.enabled[i]) {
+				free_irq(dev->intr.msix_entry[i].vector, dev);
+				dev->intr.enabled[i] = 0;
+			}
+		}
+	} else if (dev->intr.type == PVRDMA_INTR_TYPE_INTX ||
+		   dev->intr.type == PVRDMA_INTR_TYPE_MSI) {
+		free_irq(dev->pdev->irq, dev);
+	}
+}
+
+static void pvrdma_enable_intrs(struct pvrdma_dev *dev)
+{
+	dev_dbg(&dev->pdev->dev, "enable interrupts\n");
+	pvrdma_write_reg(dev, PVRDMA_REG_IMR, 0);
+}
+
+static void pvrdma_disable_intrs(struct pvrdma_dev *dev)
+{
+	dev_dbg(&dev->pdev->dev, "disable interrupts\n");
+	pvrdma_write_reg(dev, PVRDMA_REG_IMR, ~0);
+}
+
+static int pvrdma_enable_msix(struct pci_dev *pdev, struct pvrdma_dev *dev)
+{
+	int i;
+	int ret;
+
+	for (i = 0; i < PVRDMA_MAX_INTERRUPTS; i++) {
+		dev->intr.msix_entry[i].entry = i;
+		dev->intr.msix_entry[i].vector = i;
+
+		switch (i) {
+		case 0:
+			/* CMD ring handler */
+			dev->intr.handler[i] = pvrdma_intr0_handler;
+			break;
+		case 1:
+			/* Async event ring handler */
+			dev->intr.handler[i] = pvrdma_intr1_handler;
+			break;
+		default:
+			/* Completion queue handler */
+			dev->intr.handler[i] = pvrdma_intrx_handler;
+			break;
+		}
+	}
+
+	ret = pci_enable_msix(pdev, dev->intr.msix_entry,
+			      PVRDMA_MAX_INTERRUPTS);
+	if (!ret) {
+		dev->intr.type = PVRDMA_INTR_TYPE_MSIX;
+		dev->intr.size = PVRDMA_MAX_INTERRUPTS;
+	} else if (ret > 0) {
+		ret = pci_enable_msix(pdev, dev->intr.msix_entry, ret);
+		if (!ret) {
+			dev->intr.type = PVRDMA_INTR_TYPE_MSIX;
+			dev->intr.size = ret;
+		} else {
+			dev->intr.size = 0;
+		}
+	}
+
+	dev_dbg(&pdev->dev, "using interrupt type %d, size %d\n",
+		dev->intr.type, dev->intr.size);
+
+	return ret;
+}
+
+static int pvrdma_alloc_intrs(struct pvrdma_dev *dev)
+{
+	int ret = 0;
+	int i;
+
+	if (pci_find_capability(dev->pdev, PCI_CAP_ID_MSIX) &&
+	    pvrdma_enable_msix(dev->pdev, dev)) {
+		/* Try MSI */
+		ret = pci_enable_msi(dev->pdev);
+		if (!ret) {
+			dev->intr.type = PVRDMA_INTR_TYPE_MSI;
+		} else {
+			/* Legacy INTR */
+			dev->intr.type = PVRDMA_INTR_TYPE_INTX;
+		}
+	}
+
+	/* Request First IRQ */
+	switch (dev->intr.type) {
+	case PVRDMA_INTR_TYPE_INTX:
+	case PVRDMA_INTR_TYPE_MSI:
+		ret = request_irq(dev->pdev->irq, pvrdma_intr0_handler,
+				  IRQF_SHARED, DRV_NAME, dev);
+		if (ret) {
+			dev_err(&dev->pdev->dev,
+				"failed to request interrupt\n");
+			goto disable_msi;
+		}
+		break;
+	case PVRDMA_INTR_TYPE_MSIX:
+		ret = request_irq(dev->intr.msix_entry[0].vector,
+				  pvrdma_intr0_handler, 0, DRV_NAME, dev);
+		if (ret) {
+			dev_err(&dev->pdev->dev,
+				"failed to request interrupt 0\n");
+			goto disable_msi;
+		}
+		dev->intr.enabled[0] = 1;
+		break;
+	default:
+		/* Not reached */
+		break;
+	}
+
+	/* For MSIX: request intr for each vector */
+	if (dev->intr.size > 1) {
+		ret = request_irq(dev->intr.msix_entry[1].vector,
+				  pvrdma_intr1_handler, 0, DRV_NAME, dev);
+		if (ret) {
+			dev_err(&dev->pdev->dev,
+				"failed to request interrupt 1\n");
+			goto free_irq;
+		}
+		dev->intr.enabled[1] = 1;
+
+		for (i = 2; i < dev->intr.size; i++) {
+			ret = request_irq(dev->intr.msix_entry[i].vector,
+					  pvrdma_intrx_handler, 0,
+					  DRV_NAME, dev);
+			if (ret) {
+				dev_err(&dev->pdev->dev,
+					"failed to request interrupt %d\n", i);
+				goto free_irq;
+			}
+			dev->intr.enabled[i] = 1;
+		}
+	}
+
+	return 0;
+
+free_irq:
+	pvrdma_free_irq(dev);
+disable_msi:
+	pvrdma_disable_msi_all(dev);
+	return ret;
+}
+
+static void pvrdma_free_slots(struct pvrdma_dev *dev)
+{
+	struct pci_dev *pdev = dev->pdev;
+
+	if (dev->resp_slot)
+		dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->resp_slot,
+				  dev->dsr->resp_slot_dma);
+	if (dev->cmd_slot)
+		dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->cmd_slot,
+				  dev->dsr->cmd_slot_dma);
+}
+
+static int pvrdma_add_gid_at_index(struct pvrdma_dev *dev,
+				   const union ib_gid *gid,
+				   int index)
+{
+	int ret;
+	union pvrdma_cmd_req req;
+	struct pvrdma_cmd_create_bind *cmd_bind = &req.create_bind;
+
+	if (!dev->sgid_tbl) {
+		dev_warn(&dev->pdev->dev, "sgid table not initialized\n");
+		return -EINVAL;
+	}
+
+	memset(cmd_bind, 0, sizeof(*cmd_bind));
+	cmd_bind->hdr.cmd = PVRDMA_CMD_CREATE_BIND;
+	memcpy(cmd_bind->new_gid, gid->raw, 16);
+	cmd_bind->mtu = ib_mtu_enum_to_int(IB_MTU_1024);
+	cmd_bind->vlan = 0xfff;
+	cmd_bind->index = index;
+	cmd_bind->gid_type = PVRDMA_GID_TYPE_FLAG_ROCE_V1;
+
+	ret = pvrdma_cmd_post(dev, &req, false, NULL);
+	if (ret < 0) {
+		dev_warn(&dev->pdev->dev, "could not create binding\n");
+		return -EINVAL;
+	}
+	memcpy(&dev->sgid_tbl[index], gid, sizeof(*gid));
+	return 0;
+}
+
+static int pvrdma_add_gid(struct ib_device *ibdev,
+			  u8 port_num,
+			  unsigned int index,
+			  const union ib_gid *gid,
+			  const struct ib_gid_attr *attr,
+			  void **context)
+{
+	struct pvrdma_dev *dev = to_vdev(ibdev);
+
+	return pvrdma_add_gid_at_index(dev, gid, index);
+}
+
+static int pvrdma_del_gid_at_index(struct pvrdma_dev *dev, int index)
+{
+	int ret;
+	union pvrdma_cmd_req req;
+	struct pvrdma_cmd_destroy_bind *cmd_dest = &req.destroy_bind;
+
+	/* Update sgid table. */
+	if (!dev->sgid_tbl) {
+		dev_warn(&dev->pdev->dev, "sgid table not initialized\n");
+		return -EINVAL;
+	}
+
+	memset(cmd_dest, 0, sizeof(*cmd_dest));
+	cmd_dest->hdr.cmd = PVRDMA_CMD_DESTROY_BIND;
+	memcpy(cmd_dest->dest_gid, &dev->sgid_tbl[index], 16);
+	cmd_dest->index = index;
+
+	ret = pvrdma_cmd_post(dev, &req, false, NULL);
+	if (ret < 0) {
+		dev_warn(&dev->pdev->dev,
+			 "could not destroy binding\n");
+		return ret;
+	}
+	memset(&dev->sgid_tbl[index], 0, 16);
+	return 0;
+}
+
+static int pvrdma_del_gid(struct ib_device *ibdev,
+			  u8 port_num,
+			  unsigned int index,
+			  void **context)
+{
+	struct pvrdma_dev *dev = to_vdev(ibdev);
+
+	dev_dbg(&dev->pdev->dev, "removing gid at index %u from %s",
+		index, dev->netdev->name);
+
+	return pvrdma_del_gid_at_index(dev, index);
+}
+
+static void pvrdma_netdevice_event_handle(struct pvrdma_dev *dev,
+					  unsigned long event)
+{
+	struct net_device *netdev;
+
+	netdev = dev->netdev;
+	switch (event) {
+	case NETDEV_REBOOT:
+	case NETDEV_DOWN:
+		pvrdma_dispatch_event(dev, 1, IB_EVENT_PORT_ERR);
+		break;
+	case NETDEV_UP:
+		pvrdma_dispatch_event(dev, 1, IB_EVENT_PORT_ACTIVE);
+		break;
+	default:
+		dev_dbg(&dev->pdev->dev, "ignore netdevice event %ld on %s\n",
+			event, dev->ib_dev.name);
+		break;
+	}
+}
+
+static int pvrdma_netdevice_event(struct notifier_block *this,
+				  unsigned long event, void *ptr)
+{
+	struct net_device *event_netdev = netdev_notifier_info_to_dev(ptr);
+	struct pvrdma_dev *dev;
+
+	mutex_lock(&pvrdma_device_list_lock);
+	list_for_each_entry(dev, &pvrdma_device_list, device_link) {
+		if (dev->netdev == event_netdev) {
+			pvrdma_netdevice_event_handle(dev, event);
+			break;
+		}
+	}
+	mutex_unlock(&pvrdma_device_list_lock);
+
+	return NOTIFY_DONE;
+}
+
+static int pvrdma_pci_probe(struct pci_dev *pdev,
+			    const struct pci_device_id *id)
+{
+	struct pci_dev *pdev_net;
+	struct pvrdma_dev *dev;
+	int ret;
+	unsigned long start;
+	unsigned long len;
+	unsigned int version;
+	dma_addr_t slot_dma = 0;
+
+	dev_dbg(&pdev->dev, "initializing driver %s\n", pci_name(pdev));
+
+	/* Allocate zero-out device */
+	dev = (struct pvrdma_dev *)ib_alloc_device(sizeof(*dev));
+	if (!dev) {
+		dev_err(&pdev->dev, "failed to allocate IB device\n");
+		return -ENOMEM;
+	}
+
+	mutex_lock(&pvrdma_device_list_lock);
+	list_add(&dev->device_link, &pvrdma_device_list);
+	mutex_unlock(&pvrdma_device_list_lock);
+
+	ret = pvrdma_init_device(dev);
+	if (ret)
+		goto err_free_device;
+
+	dev->pdev = pdev;
+	pci_set_drvdata(pdev, dev);
+
+	ret = pci_enable_device(pdev);
+	if (ret) {
+		dev_err(&pdev->dev, "cannot enable PCI device\n");
+		goto err_free_device;
+	}
+
+	/*
+	 * DEBUG: Checking PCI: 0x20200; _MEM; len=0x2000
+	 */
+	dev_dbg(&pdev->dev, "PCI resource flags BAR0 %#lx\n",
+		pci_resource_flags(pdev, 0));
+	dev_dbg(&pdev->dev, "PCI resource len %#llx\n",
+		(unsigned long long)pci_resource_len(pdev, 0));
+	dev_dbg(&pdev->dev, "PCI resource start %#llx\n",
+		(unsigned long long)pci_resource_start(pdev, 0));
+	dev_dbg(&pdev->dev, "PCI resource flags BAR1 %#lx\n",
+		pci_resource_flags(pdev, 1));
+	dev_dbg(&pdev->dev, "PCI resource len %#llx\n",
+		(unsigned long long)pci_resource_len(pdev, 1));
+	dev_dbg(&pdev->dev, "PCI resource start %#llx\n",
+		(unsigned long long)pci_resource_start(pdev, 1));
+
+	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
+	    !(pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
+		dev_err(&pdev->dev, "PCI BAR region not MMIO\n");
+		ret = -ENOMEM;
+		goto err_free_device;
+	}
+
+	ret = pci_request_regions(pdev, DRV_NAME);
+	if (ret) {
+		dev_err(&pdev->dev, "cannot request PCI resources\n");
+		goto err_disable_pdev;
+	}
+
+	/* Enable 64-Bit DMA */
+	if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
+		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
+		if (ret != 0) {
+			dev_err(&pdev->dev,
+				"pci_set_consistent_dma_mask failed\n");
+			goto err_free_resource;
+		}
+	} else {
+		ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
+		if (ret != 0) {
+			dev_err(&pdev->dev,
+				"pci_set_dma_mask failed\n");
+			goto err_free_resource;
+		}
+	}
+
+	pci_set_master(pdev);
+
+	/* Map register space */
+	start = pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_REG);
+	len = pci_resource_len(dev->pdev, PVRDMA_PCI_RESOURCE_REG);
+	dev->regs = ioremap(start, len);
+	if (!dev->regs) {
+		dev_err(&pdev->dev, "register mapping failed\n");
+		ret = -ENOMEM;
+		goto err_free_resource;
+	}
+
+	/* Setup per-device UAR. */
+	dev->driver_uar.index = 0;
+	dev->driver_uar.pfn =
+		pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_UAR) >>
+		PAGE_SHIFT;
+	dev->driver_uar.map =
+		ioremap(dev->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
+	if (!dev->driver_uar.map) {
+		dev_err(&pdev->dev, "failed to remap UAR pages\n");
+		ret = -ENOMEM;
+		goto err_unmap_regs;
+	}
+
+	version = pvrdma_read_reg(dev, PVRDMA_REG_VERSION);
+	dev_info(&pdev->dev, "device version %d, driver version %d\n",
+		 version, PVRDMA_VERSION);
+	if (version < PVRDMA_VERSION) {
+		dev_err(&pdev->dev, "incompatible device version\n");
+		goto err_uar_unmap;
+	}
+
+	dev->dsr = dma_alloc_coherent(&pdev->dev, sizeof(*dev->dsr),
+				      &dev->dsrbase, GFP_KERNEL);
+	if (!dev->dsr) {
+		dev_err(&pdev->dev, "failed to allocate shared region\n");
+		ret = -ENOMEM;
+		goto err_uar_unmap;
+	}
+
+	/* Setup the shared region */
+	memset(dev->dsr, 0, sizeof(*dev->dsr));
+	dev->dsr->driver_version = PVRDMA_VERSION;
+	dev->dsr->gos_info.gos_bits = sizeof(void *) == 4 ?
+		PVRDMA_GOS_BITS_32 :
+		PVRDMA_GOS_BITS_64;
+	dev->dsr->gos_info.gos_type = PVRDMA_GOS_TYPE_LINUX;
+	dev->dsr->gos_info.gos_ver = 1;
+	dev->dsr->uar_pfn = dev->driver_uar.pfn;
+
+	/* Command slot. */
+	dev->cmd_slot = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
+					   &slot_dma, GFP_KERNEL);
+	if (!dev->cmd_slot) {
+		ret = -ENOMEM;
+		goto err_free_dsr;
+	}
+
+	dev->dsr->cmd_slot_dma = (u64)slot_dma;
+
+	/* Response slot. */
+	dev->resp_slot = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
+					    &slot_dma, GFP_KERNEL);
+	if (!dev->resp_slot) {
+		ret = -ENOMEM;
+		goto err_free_slots;
+	}
+
+	dev->dsr->resp_slot_dma = (u64)slot_dma;
+
+	/* Async event ring */
+	dev->dsr->async_ring_pages.num_pages = 4;
+	ret = pvrdma_page_dir_init(dev, &dev->async_pdir,
+				   dev->dsr->async_ring_pages.num_pages, true);
+	if (ret)
+		goto err_free_slots;
+	dev->async_ring_state = dev->async_pdir.pages[0];
+	dev->dsr->async_ring_pages.pdir_dma = dev->async_pdir.dir_dma;
+
+	/* CQ notification ring */
+	dev->dsr->cq_ring_pages.num_pages = 4;
+	ret = pvrdma_page_dir_init(dev, &dev->cq_pdir,
+				   dev->dsr->cq_ring_pages.num_pages, true);
+	if (ret)
+		goto err_free_async_ring;
+	dev->cq_ring_state = dev->cq_pdir.pages[0];
+	dev->dsr->cq_ring_pages.pdir_dma = dev->cq_pdir.dir_dma;
+
+	/*
+	 * Write the PA of the shared region to the device. The writes must be
+	 * ordered such that the high bits are written last. When the writes
+	 * complete, the device will have filled out the capabilities.
+	 */
+
+	pvrdma_write_reg(dev, PVRDMA_REG_DSRLOW, (u32)dev->dsrbase);
+	pvrdma_write_reg(dev, PVRDMA_REG_DSRHIGH,
+			 (u32)((u64)(dev->dsrbase) >> 32));
+
+	/* Make sure the write is complete before reading status. */
+	mb();
+
+	/* Currently, the driver only supports RoCE mode. */
+	if (dev->dsr->caps.mode != PVRDMA_DEVICE_MODE_ROCE) {
+		dev_err(&pdev->dev, "unsupported transport %d\n",
+			dev->dsr->caps.mode);
+		ret = -EINVAL;
+		goto err_free_cq_ring;
+	}
+
+	/* Currently, the driver only supports RoCE V1. */
+	if (!(dev->dsr->caps.gid_types & PVRDMA_GID_TYPE_FLAG_ROCE_V1)) {
+		dev_err(&pdev->dev, "driver needs RoCE v1 support\n");
+		ret = -EINVAL;
+		goto err_free_cq_ring;
+	}
+
+	/* Paired vmxnet3 will have same bus, slot. But func will be 0 */
+	pdev_net = pci_get_slot(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 0));
+	if (!pdev_net) {
+		dev_err(&pdev->dev, "failed to find paired net device\n");
+		ret = -ENODEV;
+		goto err_free_cq_ring;
+	}
+
+	if (pdev_net->vendor != PCI_VENDOR_ID_VMWARE ||
+	    pdev_net->device != PCI_DEVICE_ID_VMWARE_VMXNET3) {
+		dev_err(&pdev->dev, "failed to find paired vmxnet3 device\n");
+		pci_dev_put(pdev_net);
+		ret = -ENODEV;
+		goto err_free_cq_ring;
+	}
+
+	dev->netdev = pci_get_drvdata(pdev_net);
+	pci_dev_put(pdev_net);
+	if (!dev->netdev) {
+		dev_err(&pdev->dev, "failed to get vmxnet3 device\n");
+		ret = -ENODEV;
+		goto err_free_cq_ring;
+	}
+
+	dev_info(&pdev->dev, "paired device to %s\n", dev->netdev->name);
+
+	dev->nb_netdev.notifier_call = pvrdma_netdevice_event;
+	ret = register_netdevice_notifier(&dev->nb_netdev);
+	if (ret) {
+		dev_err(&pdev->dev, "failed to register netdevice events\n");
+		goto err_free_cq_ring;
+	}
+
+	/* Interrupt setup */
+	ret = pvrdma_alloc_intrs(dev);
+	if (ret) {
+		dev_err(&pdev->dev, "failed to allocate interrupts\n");
+		ret = -ENOMEM;
+		goto err_netdevice;
+	}
+
+	/* Allocate UAR table. */
+	ret = pvrdma_uar_table_init(dev);
+	if (ret) {
+		dev_err(&pdev->dev, "failed to allocate UAR table\n");
+		ret = -ENOMEM;
+		goto err_free_intrs;
+	}
+
+	/* Allocate GID table */
+	dev->sgid_tbl = kcalloc(dev->dsr->caps.gid_tbl_len,
+				sizeof(union ib_gid), GFP_KERNEL);
+	if (!dev->sgid_tbl) {
+		ret = -ENOMEM;
+		goto err_free_uar_table;
+	}
+	dev_dbg(&pdev->dev, "gid table len %d\n", dev->dsr->caps.gid_tbl_len);
+
+	pvrdma_enable_intrs(dev);
+
+	/* Activate pvrdma device */
+	pvrdma_write_reg(dev, PVRDMA_REG_CTL, PVRDMA_DEVICE_CTL_ACTIVATE);
+
+	/* Make sure the write is complete before reading status. */
+	mb();
+
+	/* Check if device was successfully activated */
+	ret = pvrdma_read_reg(dev, PVRDMA_REG_ERR);
+	if (ret != 0) {
+		dev_err(&pdev->dev, "failed to activate device\n");
+		ret = -EINVAL;
+		goto err_disable_intr;
+	}
+
+	/* Register IB device */
+	ret = pvrdma_register_device(dev);
+	if (ret) {
+		dev_err(&pdev->dev, "failed to register IB device\n");
+		goto err_disable_intr;
+	}
+
+	dev_info(&pdev->dev, "attached to device\n");
+	return 0;
+
+err_disable_intr:
+	pvrdma_disable_intrs(dev);
+	kfree(dev->sgid_tbl);
+err_free_uar_table:
+	pvrdma_uar_table_cleanup(dev);
+err_free_intrs:
+	pvrdma_free_irq(dev);
+	pvrdma_disable_msi_all(dev);
+err_netdevice:
+	unregister_netdevice_notifier(&dev->nb_netdev);
+err_free_cq_ring:
+	pvrdma_page_dir_cleanup(dev, &dev->cq_pdir);
+err_free_async_ring:
+	pvrdma_page_dir_cleanup(dev, &dev->async_pdir);
+err_free_slots:
+	pvrdma_free_slots(dev);
+err_free_dsr:
+	dma_free_coherent(&pdev->dev, sizeof(*dev->dsr), dev->dsr,
+			  dev->dsrbase);
+err_uar_unmap:
+	iounmap(dev->driver_uar.map);
+err_unmap_regs:
+	iounmap(dev->regs);
+err_free_resource:
+	pci_release_regions(pdev);
+err_disable_pdev:
+	pci_disable_device(pdev);
+	pci_set_drvdata(pdev, NULL);
+err_free_device:
+	mutex_lock(&pvrdma_device_list_lock);
+	list_del(&dev->device_link);
+	mutex_unlock(&pvrdma_device_list_lock);
+	ib_dealloc_device(&dev->ib_dev);
+	return ret;
+}
+
+static void pvrdma_pci_remove(struct pci_dev *pdev)
+{
+	struct pvrdma_dev *dev = pci_get_drvdata(pdev);
+
+	if (!dev)
+		return;
+
+	dev_info(&pdev->dev, "detaching from device\n");
+
+	mutex_lock(&pvrdma_device_list_lock);
+	list_del(&dev->device_link);
+	mutex_unlock(&pvrdma_device_list_lock);
+
+	/* Unregister ib device */
+	ib_unregister_device(&dev->ib_dev);
+
+	pvrdma_disable_intrs(dev);
+	pvrdma_free_irq(dev);
+	pvrdma_disable_msi_all(dev);
+
+	unregister_netdevice_notifier(&dev->nb_netdev);
+
+	/* Deactivate pvrdma device */
+	pvrdma_write_reg(dev, PVRDMA_REG_CTL, PVRDMA_DEVICE_CTL_RESET);
+	pvrdma_page_dir_cleanup(dev, &dev->cq_pdir);
+	pvrdma_page_dir_cleanup(dev, &dev->async_pdir);
+	pvrdma_free_slots(dev);
+
+	iounmap(dev->regs);
+	kfree(dev->sgid_tbl);
+	kfree(dev->cq_tbl);
+	kfree(dev->qp_tbl);
+	pvrdma_uar_table_cleanup(dev);
+	iounmap(dev->driver_uar.map);
+
+	ib_dealloc_device(&dev->ib_dev);
+
+	/* Free pci resources */
+	pci_release_regions(pdev);
+	pci_disable_device(pdev);
+	pci_set_drvdata(pdev, NULL);
+}
+
+static struct pci_device_id pvrdma_pci_table[] = {
+	{ PCI_DEVICE(PCI_VENDOR_ID_VMWARE, PCI_DEVICE_ID_VMWARE_PVRDMA), },
+	{ 0 },
+};
+
+MODULE_DEVICE_TABLE(pci, pvrdma_pci_table);
+
+static struct pci_driver pvrdma_driver = {
+	.name		= DRV_NAME,
+	.id_table	= pvrdma_pci_table,
+	.probe		= pvrdma_pci_probe,
+	.remove		= pvrdma_pci_remove,
+};
+
+static int __init pvrdma_init(void)
+{
+	return pci_register_driver(&pvrdma_driver);
+}
+
+static void __exit pvrdma_cleanup(void)
+{
+	pci_unregister_driver(&pvrdma_driver);
+}
+
+module_init(pvrdma_init);
+module_exit(pvrdma_cleanup);
+
+MODULE_AUTHOR("VMware, Inc");
+MODULE_DESCRIPTION("VMware Virtual RDMA/InfiniBand driver");
+MODULE_VERSION(DRV_VERSION);
+MODULE_LICENSE("Dual BSD/GPL");
-- 
2.7.4

--
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 08/15] IB/pvrdma: Add helper functions

[Adit Ranadive]

This patch adds helper functions to store guest page addresses in a page
directory structure. The page directory pointer is passed down to the
backend which then maps the entire memory for the RDMA object by
traversing the directory. We add some more helper functions for converting
to/from RDMA stack address handles from/to PVRDMA ones.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/pvrdma/pvrdma_misc.c | 309 +++++++++++++++++++++++++++++
 1 file changed, 309 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_misc.c

diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_misc.c b/drivers/infiniband/hw/pvrdma/pvrdma_misc.c
new file mode 100644
index 0000000..97a1fcd
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_misc.c
@@ -0,0 +1,309 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <linux/errno.h>
+#include <linux/slab.h>
+#include <linux/bitmap.h>
+
+#include "pvrdma.h"
+
+int pvrdma_page_dir_init(struct pvrdma_dev *dev, struct pvrdma_page_dir *pdir,
+			 u64 npages, bool alloc_pages)
+{
+	u64 i;
+
+	if (npages > PVRDMA_PAGE_DIR_MAX_PAGES)
+		return -EINVAL;
+
+	memset(pdir, 0, sizeof(*pdir));
+
+	pdir->dir = dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE,
+				       &pdir->dir_dma, GFP_KERNEL);
+	if (!pdir->dir)
+		goto err;
+
+	pdir->ntables = PVRDMA_PAGE_DIR_TABLE(npages - 1) + 1;
+	pdir->tables = kcalloc(pdir->ntables, sizeof(*pdir->tables),
+			       GFP_KERNEL);
+	if (!pdir->tables)
+		goto err;
+
+	for (i = 0; i < pdir->ntables; i++) {
+		u64 *table;
+		dma_addr_t table_base;
+
+		table = dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE,
+					   &table_base, GFP_KERNEL);
+		if (!table)
+			goto err;
+
+		pdir->tables[i] = table;
+		pdir->dir[i] = table_base;
+	}
+
+	pdir->npages = npages;
+
+	if (alloc_pages) {
+		pdir->pages = kcalloc(npages, sizeof(*pdir->pages),
+				      GFP_KERNEL);
+		if (!pdir->pages)
+			goto err;
+
+		for (i = 0; i < pdir->npages; i++) {
+			void *page;
+			dma_addr_t page_dma;
+
+			page = dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE,
+						  &page_dma, GFP_KERNEL);
+			if (!page)
+				goto err;
+
+			pdir->pages[i] = page;
+			pvrdma_page_dir_insert_dma(pdir, i, page_dma);
+		}
+	}
+
+	return 0;
+
+err:
+	pvrdma_page_dir_cleanup(dev, pdir);
+
+	return -ENOMEM;
+}
+
+static u64 *pvrdma_page_dir_table(struct pvrdma_page_dir *pdir, u64 idx)
+{
+	return pdir->tables[PVRDMA_PAGE_DIR_TABLE(idx)];
+}
+
+dma_addr_t pvrdma_page_dir_get_dma(struct pvrdma_page_dir *pdir, u64 idx)
+{
+	return pvrdma_page_dir_table(pdir, idx)[PVRDMA_PAGE_DIR_PAGE(idx)];
+}
+
+static void pvrdma_page_dir_cleanup_pages(struct pvrdma_dev *dev,
+					  struct pvrdma_page_dir *pdir)
+{
+	if (pdir->pages) {
+		u64 i;
+
+		for (i = 0; i < pdir->npages && pdir->pages[i]; i++) {
+			dma_addr_t page_dma = pvrdma_page_dir_get_dma(pdir, i);
+
+			dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
+					  pdir->pages[i], page_dma);
+		}
+
+		kfree(pdir->pages);
+	}
+}
+
+static void pvrdma_page_dir_cleanup_tables(struct pvrdma_dev *dev,
+					   struct pvrdma_page_dir *pdir)
+{
+	if (pdir->tables) {
+		int i;
+
+		pvrdma_page_dir_cleanup_pages(dev, pdir);
+
+		for (i = 0; i < pdir->ntables; i++) {
+			u64 *table = pdir->tables[i];
+
+			if (table)
+				dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
+						  table, pdir->dir[i]);
+		}
+
+		kfree(pdir->tables);
+	}
+}
+
+void pvrdma_page_dir_cleanup(struct pvrdma_dev *dev,
+			     struct pvrdma_page_dir *pdir)
+{
+	if (pdir->dir) {
+		pvrdma_page_dir_cleanup_tables(dev, pdir);
+		dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
+				  pdir->dir, pdir->dir_dma);
+	}
+}
+
+int pvrdma_page_dir_insert_dma(struct pvrdma_page_dir *pdir, u64 idx,
+			       dma_addr_t daddr)
+{
+	u64 *table;
+
+	if (idx >= pdir->npages)
+		return -EINVAL;
+
+	table = pvrdma_page_dir_table(pdir, idx);
+	table[PVRDMA_PAGE_DIR_PAGE(idx)] = daddr;
+
+	return 0;
+}
+
+int pvrdma_page_dir_insert_umem(struct pvrdma_page_dir *pdir,
+				struct ib_umem *umem, u64 offset)
+{
+	u64 i = offset;
+	int j, entry;
+	int ret = 0, len = 0;
+	struct scatterlist *sg;
+
+	if (offset >= pdir->npages)
+		return -EINVAL;
+
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+		len = sg_dma_len(sg) >> PAGE_SHIFT;
+		for (j = 0; j < len; j++) {
+			dma_addr_t addr = sg_dma_address(sg) +
+					  umem->page_size * j;
+
+			ret = pvrdma_page_dir_insert_dma(pdir, i, addr);
+			if (ret)
+				goto exit;
+
+			i++;
+		}
+	}
+
+exit:
+	return ret;
+}
+
+int pvrdma_page_dir_insert_page_list(struct pvrdma_page_dir *pdir,
+				     u64 *page_list,
+				     int num_pages)
+{
+	int i;
+	int ret;
+
+	if (num_pages > pdir->npages)
+		return -EINVAL;
+
+	for (i = 0; i < num_pages; i++) {
+		ret = pvrdma_page_dir_insert_dma(pdir, i, page_list[i]);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+void ib_qp_cap_to_pvrdma(const struct ib_qp_cap *src, struct pvrdma_qp_cap *dst)
+{
+	dst->max_send_wr = src->max_send_wr;
+	dst->max_recv_wr = src->max_recv_wr;
+	dst->max_send_sge = src->max_send_sge;
+	dst->max_recv_sge = src->max_recv_sge;
+	dst->max_inline_data = src->max_inline_data;
+}
+
+void pvrdma_qp_cap_to_ib(const struct pvrdma_qp_cap *src, struct ib_qp_cap *dst)
+{
+	dst->max_send_wr = src->max_send_wr;
+	dst->max_recv_wr = src->max_recv_wr;
+	dst->max_send_sge = src->max_send_sge;
+	dst->max_recv_sge = src->max_recv_sge;
+	dst->max_inline_data = src->max_inline_data;
+}
+
+void pvrdma_gid_to_ib(const union pvrdma_gid *src, union ib_gid *dst)
+{
+	BUILD_BUG_ON(sizeof(union pvrdma_gid) != sizeof(union ib_gid));
+	memcpy(dst, src, sizeof(*src));
+}
+
+void ib_gid_to_pvrdma(const union ib_gid *src, union pvrdma_gid *dst)
+{
+	BUILD_BUG_ON(sizeof(union pvrdma_gid) != sizeof(union ib_gid));
+	memcpy(dst, src, sizeof(*src));
+}
+
+void ib_global_route_to_pvrdma(const struct ib_global_route *src,
+			       struct pvrdma_global_route *dst)
+{
+	ib_gid_to_pvrdma(&src->dgid, &dst->dgid);
+	dst->flow_label = src->flow_label;
+	dst->sgid_index = src->sgid_index;
+	dst->hop_limit = src->hop_limit;
+	dst->traffic_class = src->traffic_class;
+}
+
+void pvrdma_global_route_to_ib(const struct pvrdma_global_route *src,
+			       struct ib_global_route *dst)
+{
+	pvrdma_gid_to_ib(&src->dgid, &dst->dgid);
+	dst->flow_label = src->flow_label;
+	dst->sgid_index = src->sgid_index;
+	dst->hop_limit = src->hop_limit;
+	dst->traffic_class = src->traffic_class;
+}
+
+void ib_ah_attr_to_pvrdma(const struct ib_ah_attr *src,
+			  struct pvrdma_ah_attr *dst)
+{
+	ib_global_route_to_pvrdma(&src->grh, &dst->grh);
+	dst->dlid = src->dlid;
+	dst->sl = src->sl;
+	dst->src_path_bits = src->src_path_bits;
+	dst->static_rate = src->static_rate;
+	dst->ah_flags = src->ah_flags;
+	dst->port_num = src->port_num;
+	memcpy(&dst->dmac, &src->dmac, sizeof(dst->dmac));
+}
+
+void pvrdma_ah_attr_to_ib(const struct pvrdma_ah_attr *src,
+			  struct ib_ah_attr *dst)
+{
+	pvrdma_global_route_to_ib(&src->grh, &dst->grh);
+	dst->dlid = src->dlid;
+	dst->sl = src->sl;
+	dst->src_path_bits = src->src_path_bits;
+	dst->static_rate = src->static_rate;
+	dst->ah_flags = src->ah_flags;
+	dst->port_num = src->port_num;
+	memcpy(&dst->dmac, &src->dmac, sizeof(dst->dmac));
+}
-- 
2.7.4

--
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 09/15] IB/pvrdma: Add support for memory regions

[Adit Ranadive]

This patch adds support for creating and destroying memory regions. The
PVRDMA device supports User MRs, DMA MRs (no Remote Read/Write support),
Fast Register MRs.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/pvrdma/pvrdma_mr.c | 333 +++++++++++++++++++++++++++++++
 1 file changed, 333 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_mr.c

diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_mr.c b/drivers/infiniband/hw/pvrdma/pvrdma_mr.c
new file mode 100644
index 0000000..ea5edd5
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_mr.c
@@ -0,0 +1,333 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <linux/list.h>
+#include <linux/slab.h>
+
+#include "pvrdma.h"
+
+/**
+ * pvrdma_get_dma_mr - get a DMA memory region
+ * @pd: protection domain
+ * @acc: access flags
+ *
+ * @return: ib_mr pointer on success, otherwise returns an errno.
+ */
+struct ib_mr *pvrdma_get_dma_mr(struct ib_pd *pd, int acc)
+{
+	struct pvrdma_dev *dev = to_vdev(pd->device);
+	struct pvrdma_user_mr *mr;
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_create_mr *cmd = &req.create_mr;
+	struct pvrdma_cmd_create_mr_resp *resp = &rsp.create_mr_resp;
+	int ret;
+
+	if (acc != IB_ACCESS_LOCAL_WRITE) {
+		dev_warn(&dev->pdev->dev, "unsupported dma mr access flags\n");
+		return ERR_PTR(-EOPNOTSUPP);
+	}
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_CREATE_MR;
+	cmd->pd_handle = to_vpd(pd)->pd_handle;
+	cmd->access_flags = acc;
+	cmd->flags = PVRDMA_MR_FLAG_DMA;
+
+	ret = pvrdma_cmd_post(dev, &req, true, &rsp);
+	if (ret < 0) {
+		dev_warn(&dev->pdev->dev, "could not get DMA mem region\n");
+		kfree(mr);
+		return ERR_PTR(ret);
+	}
+
+	mr->mmr.mr_handle = resp->mr_handle;
+	mr->ibmr.lkey = resp->lkey;
+	mr->ibmr.rkey = resp->rkey;
+
+	return &mr->ibmr;
+}
+
+
+/**
+ * pvrdma_reg_user_mr - register a userspace memory region
+ * @pd: protection domain
+ * @start: starting address
+ * @length: length of region
+ * @virt_addr: I/O virtual address
+ * @access_flags: access flags for memory region
+ * @udata: user data
+ *
+ * @return: ib_mr pointer on success, otherwise returns an errno.
+ */
+struct ib_mr *pvrdma_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+				 u64 virt_addr, int access_flags,
+				 struct ib_udata *udata)
+{
+	struct pvrdma_dev *dev = to_vdev(pd->device);
+	struct pvrdma_user_mr *mr = NULL;
+	struct ib_umem *umem;
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_create_mr *cmd = &req.create_mr;
+	struct pvrdma_cmd_create_mr_resp *resp = &rsp.create_mr_resp;
+	int nchunks;
+	int ret;
+	int entry;
+	struct scatterlist *sg;
+
+	if (length == 0 || length > dev->dsr->caps.max_mr_size) {
+		dev_warn(&dev->pdev->dev, "invalid mem region length\n");
+		return ERR_PTR(-EINVAL);
+	}
+
+	umem = ib_umem_get(pd->uobject->context, start,
+			   length, access_flags, 0);
+	if (IS_ERR(umem)) {
+		dev_warn(&dev->pdev->dev,
+			 "could not get umem for mem region\n");
+		return ERR_CAST(umem);
+	}
+
+	nchunks = 0;
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry)
+		nchunks += sg_dma_len(sg) >> PAGE_SHIFT;
+
+	if (nchunks < 0 || nchunks > PVRDMA_PAGE_DIR_MAX_PAGES) {
+		dev_warn(&dev->pdev->dev, "overflow %d pages in mem region\n",
+			 nchunks);
+		ret = -EINVAL;
+		goto err_umem;
+	}
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr) {
+		ret = -ENOMEM;
+		goto err_umem;
+	}
+
+	mr->mmr.iova = virt_addr;
+	mr->mmr.size = length;
+	mr->umem = umem;
+
+	ret = pvrdma_page_dir_init(dev, &mr->pdir, nchunks, false);
+	if (ret) {
+		dev_warn(&dev->pdev->dev,
+			 "could not allocate page directory\n");
+		goto err_umem;
+	}
+
+	ret = pvrdma_page_dir_insert_umem(&mr->pdir, mr->umem, 0);
+	if (ret)
+		goto err_pdir;
+
+	/* Send to host */
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_CREATE_MR;
+	cmd->start = start;
+	cmd->length = length;
+	cmd->pd_handle = to_vpd(pd)->pd_handle;
+	cmd->access_flags = access_flags;
+	cmd->nchunks = nchunks;
+	cmd->pdir_dma = mr->pdir.dir_dma;
+
+	ret = pvrdma_cmd_post(dev, &req, true, &rsp);
+
+	if (ret < 0 || rsp.hdr.ack != PVRDMA_CMD_CREATE_MR_RESP) {
+		dev_warn(&dev->pdev->dev, "could not register mem region\n");
+		goto err_pdir;
+	}
+
+	mr->mmr.mr_handle = resp->mr_handle;
+	mr->ibmr.lkey = resp->lkey;
+	mr->ibmr.rkey = resp->rkey;
+
+	return &mr->ibmr;
+
+err_pdir:
+	pvrdma_page_dir_cleanup(dev, &mr->pdir);
+err_umem:
+	ib_umem_release(umem);
+	kfree(mr);
+
+	return ERR_PTR(ret);
+}
+
+/**
+ * pvrdma_alloc_mr - allocate a memory region
+ * @pd: protection domain
+ * @mr_type: type of memory region
+ * @max_num_sg: maximum number of pages
+ *
+ * @return: ib_mr pointer on success, otherwise returns an errno.
+ */
+struct ib_mr *pvrdma_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
+			      u32 max_num_sg)
+{
+	struct pvrdma_dev *dev = to_vdev(pd->device);
+	struct pvrdma_user_mr *mr;
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_create_mr *cmd = &req.create_mr;
+	struct pvrdma_cmd_create_mr_resp *resp = &rsp.create_mr_resp;
+	int size = max_num_sg * sizeof(u64);
+	int ret;
+
+	if (mr_type != IB_MR_TYPE_MEM_REG ||
+	    max_num_sg > PVRDMA_MAX_FAST_REG_PAGES)
+		return ERR_PTR(-EINVAL);
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	mr->pages = kzalloc(size, GFP_KERNEL);
+	if (!mr->pages) {
+		ret = -ENOMEM;
+		goto freemr;
+	}
+
+	ret = pvrdma_page_dir_init(dev, &mr->pdir, max_num_sg, false);
+	if (ret) {
+		dev_warn(&dev->pdev->dev,
+			 "failed to allocate page dir for mr\n");
+		ret = -ENOMEM;
+		goto freepages;
+	}
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_CREATE_MR;
+	cmd->pd_handle = to_vpd(pd)->pd_handle;
+	cmd->access_flags = 0;
+	cmd->flags = PVRDMA_MR_FLAG_FRMR;
+	cmd->nchunks = max_num_sg;
+
+	ret = pvrdma_cmd_post(dev, &req, true, &rsp);
+	if (ret < 0) {
+		dev_warn(&dev->pdev->dev, "could not create FR mem region\n");
+		ret = -EINVAL;
+		goto freepdir;
+	}
+
+	mr->max_pages = max_num_sg;
+	mr->mmr.mr_handle = resp->mr_handle;
+	mr->ibmr.lkey = resp->lkey;
+	mr->ibmr.rkey = resp->rkey;
+	mr->page_shift = PAGE_SHIFT;
+	mr->umem = NULL;
+
+	return &mr->ibmr;
+
+freepdir:
+	pvrdma_page_dir_cleanup(dev, &mr->pdir);
+freepages:
+	kfree(mr->pages);
+freemr:
+	kfree(mr);
+	return ERR_PTR(ret);
+}
+
+/**
+ * pvrdma_dereg_mr - deregister a memory region
+ * @ibmr: memory region
+ *
+ * @return: 0 on success.
+ */
+int pvrdma_dereg_mr(struct ib_mr *ibmr)
+{
+	struct pvrdma_user_mr *mr = to_vmr(ibmr);
+	struct pvrdma_dev *dev = to_vdev(ibmr->device);
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_destroy_mr *cmd = &req.destroy_mr;
+	int ret;
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_DESTROY_MR;
+	cmd->mr_handle = mr->mmr.mr_handle;
+	ret = pvrdma_cmd_post(dev, &req, false, &rsp);
+	if (ret < 0)
+		dev_warn(&dev->pdev->dev, "could not deregister mem region\n");
+
+	pvrdma_page_dir_cleanup(dev, &mr->pdir);
+	if (mr->umem)
+		ib_umem_release(mr->umem);
+
+	kfree(mr->pages);
+	kfree(mr);
+
+	return 0;
+}
+
+static int pvrdma_set_page(struct ib_mr *ibmr, u64 addr)
+{
+	struct pvrdma_user_mr *mr = to_vmr(ibmr);
+
+	if (mr->npages == mr->max_pages)
+		return -ENOMEM;
+
+	mr->pages[mr->npages++] = addr;
+	return 0;
+}
+
+int pvrdma_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents,
+		     unsigned int *sg_offset)
+{
+	struct pvrdma_user_mr *mr = to_vmr(ibmr);
+	struct pvrdma_dev *dev = to_vdev(ibmr->device);
+	int ret;
+
+	mr->npages = 0;
+
+	ret = ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, pvrdma_set_page);
+	if (ret < 0)
+		dev_warn(&dev->pdev->dev, "could not map sg to pages\n");
+
+	return ret;
+}
-- 
2.7.4

--
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 10/15] IB/pvrdma: Add Queue Pair support

[Adit Ranadive]

This patch adds the ability to create, modify, query and destroy QPs. The
PVRDMA device supports RC, UD and GSI QPs.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/pvrdma/pvrdma_qp.c | 976 +++++++++++++++++++++++++++++++
 1 file changed, 976 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_qp.c

diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_qp.c b/drivers/infiniband/hw/pvrdma/pvrdma_qp.c
new file mode 100644
index 0000000..18fd2c8
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_qp.c
@@ -0,0 +1,976 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <asm/page.h>
+#include <linux/io.h>
+#include <linux/wait.h>
+#include <rdma/ib_addr.h>
+#include <rdma/ib_smi.h>
+#include <rdma/ib_user_verbs.h>
+
+#include "pvrdma.h"
+#include "pvrdma_user.h"
+
+static inline void get_cqs(struct pvrdma_qp *qp, struct pvrdma_cq **send_cq,
+			   struct pvrdma_cq **recv_cq)
+{
+	*send_cq = to_vcq(qp->ibqp.send_cq);
+	*recv_cq = to_vcq(qp->ibqp.recv_cq);
+}
+
+static void pvrdma_reset_qp(struct pvrdma_qp *qp)
+{
+	struct pvrdma_cq *scq, *rcq;
+	unsigned long flags;
+
+	/* Clean up cqes */
+	get_cqs(qp, &scq, &rcq);
+	spin_lock_irqsave(&scq->cq_lock, flags);
+	pvrdma_flush_cqe(qp, scq);
+	spin_unlock_irqrestore(&scq->cq_lock, flags);
+	if (rcq != scq) {
+		spin_lock_irqsave(&rcq->cq_lock, flags);
+		pvrdma_flush_cqe(qp, rcq);
+		spin_unlock_irqrestore(&rcq->cq_lock, flags);
+	}
+
+	/*
+	 * Reset queuepair. The checks are because usermode queuepairs won't
+	 * have kernel ringstates.
+	 */
+	if (qp->rq.ring) {
+		atomic_set(&qp->rq.ring->cons_head, 0);
+		atomic_set(&qp->rq.ring->prod_tail, 0);
+	}
+	if (qp->sq.ring) {
+		atomic_set(&qp->sq.ring->cons_head, 0);
+		atomic_set(&qp->sq.ring->prod_tail, 0);
+	}
+}
+
+static int pvrdma_set_rq_size(struct pvrdma_dev *dev,
+			      struct ib_qp_cap *cap,
+			      struct pvrdma_qp *qp)
+{
+	if (cap->max_recv_wr > dev->dsr->caps.max_qp_wr ||
+	    cap->max_recv_sge > dev->dsr->caps.max_sge) {
+		dev_warn(&dev->pdev->dev, "recv queue size invalid\n");
+		return -EINVAL;
+	}
+
+	qp->rq.wqe_cnt = roundup_pow_of_two(max(1U, cap->max_recv_wr));
+	qp->rq.max_gs = roundup_pow_of_two(max(1U, cap->max_recv_sge));
+
+	/* Write back */
+	cap->max_recv_wr = qp->rq.wqe_cnt;
+	cap->max_recv_sge = qp->rq.max_gs;
+
+	qp->rq.wqe_size = roundup_pow_of_two(sizeof(struct pvrdma_rq_wqe_hdr) +
+					     sizeof(struct ib_sge) *
+					     qp->rq.max_gs);
+	qp->npages_recv = (qp->rq.wqe_cnt * qp->rq.wqe_size + PAGE_SIZE - 1) /
+			  PAGE_SIZE;
+
+	return 0;
+}
+
+static int pvrdma_set_sq_size(struct pvrdma_dev *dev, struct ib_qp_cap *cap,
+			      enum ib_qp_type type, struct pvrdma_qp *qp)
+{
+	if (cap->max_send_wr > dev->dsr->caps.max_qp_wr ||
+	    cap->max_send_sge > dev->dsr->caps.max_sge) {
+		dev_warn(&dev->pdev->dev, "send queue size invalid\n");
+		return -EINVAL;
+	}
+
+	qp->sq.wqe_cnt = roundup_pow_of_two(max(1U, cap->max_send_wr));
+	qp->sq.max_gs = roundup_pow_of_two(max(1U, cap->max_send_sge));
+
+	/* Write back */
+	cap->max_send_wr = qp->sq.wqe_cnt;
+	cap->max_send_sge = qp->sq.max_gs;
+
+	qp->sq.wqe_size = roundup_pow_of_two(sizeof(struct pvrdma_sq_wqe_hdr) +
+					     sizeof(struct ib_sge) *
+					     qp->sq.max_gs);
+	/* Note: one extra page for the header. */
+	qp->npages_send = 1 + (qp->sq.wqe_cnt * qp->sq.wqe_size +
+			       PAGE_SIZE - 1) / PAGE_SIZE;
+
+	return 0;
+}
+
+/**
+ * pvrdma_create_qp - create queue pair
+ * @pd: protection domain
+ * @init_attr: queue pair attributes
+ * @udata: user data
+ *
+ * @return: the ib_qp pointer on success, otherwise returns an errno.
+ */
+struct ib_qp *pvrdma_create_qp(struct ib_pd *pd,
+			       struct ib_qp_init_attr *init_attr,
+			       struct ib_udata *udata)
+{
+	struct pvrdma_qp *qp;
+	struct pvrdma_dev *dev = to_vdev(pd->device);
+	struct pvrdma_cq *send_cq, *recv_cq;
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_create_qp *cmd = &req.create_qp;
+	struct pvrdma_cmd_create_qp_resp *resp = &rsp.create_qp_resp;
+	struct pvrdma_create_qp ucmd;
+	unsigned long flags;
+	int ret;
+
+	if (!atomic_add_unless(&dev->num_qps, 1, dev->dsr->caps.max_qp))
+		return ERR_PTR(-EINVAL);
+
+	if (init_attr->create_flags) {
+		dev_warn(&dev->pdev->dev,
+			 "invalid create queuepair flags %#x\n",
+			 init_attr->create_flags);
+		atomic_dec(&dev->num_qps);
+		return ERR_PTR(-EINVAL);
+	}
+
+	if (init_attr->qp_type != IB_QPT_RC &&
+	    init_attr->qp_type != IB_QPT_UD &&
+	    init_attr->qp_type != IB_QPT_GSI) {
+		dev_warn(&dev->pdev->dev, "queuepair type %d not supported\n",
+			 init_attr->qp_type);
+		atomic_dec(&dev->num_qps);
+		return ERR_PTR(-EINVAL);
+	}
+
+	switch (init_attr->qp_type) {
+	case IB_QPT_GSI:
+		if (init_attr->port_num == 0 ||
+		    init_attr->port_num > pd->device->phys_port_cnt ||
+		    udata) {
+			dev_warn(&dev->pdev->dev, "invalid queuepair attrs\n");
+			atomic_dec(&dev->num_qps);
+			return ERR_PTR(-EINVAL);
+		}
+		/* fall through */
+	case IB_QPT_RC:
+	case IB_QPT_UD:
+		qp = kzalloc(sizeof(*qp), GFP_KERNEL);
+		if (!qp) {
+			atomic_dec(&dev->num_qps);
+			return ERR_PTR(-ENOMEM);
+		}
+
+		spin_lock_init(&qp->sq.lock);
+		spin_lock_init(&qp->rq.lock);
+		mutex_init(&qp->mutex);
+		atomic_set(&qp->refcnt, 1);
+		init_waitqueue_head(&qp->wait);
+
+		qp->state = IB_QPS_RESET;
+
+		if (pd->uobject && udata) {
+			dev_dbg(&dev->pdev->dev,
+				"create queuepair from user space\n");
+
+			if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) {
+				ret = -EFAULT;
+				goto err_qp;
+			}
+
+			/* set qp->sq.wqe_cnt, shift, buf_size.. */
+			qp->rumem = ib_umem_get(pd->uobject->context,
+						ucmd.rbuf_addr,
+						ucmd.rbuf_size, 0, 0);
+			if (IS_ERR(qp->rumem)) {
+				ret = PTR_ERR(qp->rumem);
+				goto err_qp;
+			}
+
+			qp->sumem = ib_umem_get(pd->uobject->context,
+						ucmd.sbuf_addr,
+						ucmd.sbuf_size, 0, 0);
+			if (IS_ERR(qp->sumem)) {
+				ib_umem_release(qp->rumem);
+				ret = PTR_ERR(qp->sumem);
+				goto err_qp;
+			}
+
+			qp->npages_send = ib_umem_page_count(qp->sumem);
+			qp->npages_recv = ib_umem_page_count(qp->rumem);
+			qp->npages = qp->npages_send + qp->npages_recv;
+		} else {
+			qp->is_kernel = true;
+
+			send_cq = to_vcq(init_attr->send_cq);
+			recv_cq = to_vcq(init_attr->recv_cq);
+
+			ret = pvrdma_set_sq_size(to_vdev(pd->device),
+						 &init_attr->cap,
+						 init_attr->qp_type, qp);
+			if (ret)
+				goto err_qp;
+
+			ret = pvrdma_set_rq_size(to_vdev(pd->device),
+						 &init_attr->cap, qp);
+			if (ret)
+				goto err_qp;
+
+			qp->npages = qp->npages_send + qp->npages_recv;
+
+			/* Skip header page. */
+			qp->sq.offset = PAGE_SIZE;
+
+			/* Recv queue pages are after send pages. */
+			qp->rq.offset = qp->npages_send * PAGE_SIZE;
+		}
+
+		if (qp->npages < 0 || qp->npages > PVRDMA_PAGE_DIR_MAX_PAGES) {
+			dev_warn(&dev->pdev->dev,
+				 "overflow pages in queuepair\n");
+			goto err_umem;
+		}
+
+		ret = pvrdma_page_dir_init(dev, &qp->pdir, qp->npages,
+					   qp->is_kernel);
+		if (ret) {
+			dev_warn(&dev->pdev->dev,
+				 "could not allocate page directory\n");
+			goto err_umem;
+		}
+
+		if (!qp->is_kernel) {
+			pvrdma_page_dir_insert_umem(&qp->pdir, qp->sumem, 0);
+			pvrdma_page_dir_insert_umem(&qp->pdir, qp->rumem,
+						    qp->npages_send);
+		} else {
+			/* Ring state is always the first page. */
+			qp->sq.ring = qp->pdir.pages[0];
+			qp->rq.ring = &qp->sq.ring[1];
+		}
+		break;
+	default:
+		atomic_dec(&dev->num_qps);
+		return ERR_PTR(-EINVAL);
+	}
+
+	/* Not supported */
+	init_attr->cap.max_inline_data = 0;
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_CREATE_QP;
+	cmd->pd_handle = to_vpd(pd)->pd_handle;
+	cmd->send_cq_handle = to_vcq(init_attr->send_cq)->cq_handle;
+	cmd->recv_cq_handle = to_vcq(init_attr->recv_cq)->cq_handle;
+	cmd->srq_handle = 0;
+	cmd->max_send_wr = init_attr->cap.max_send_wr;
+	cmd->max_recv_wr = init_attr->cap.max_recv_wr;
+	cmd->max_send_sge = init_attr->cap.max_send_sge;
+	cmd->max_recv_sge = init_attr->cap.max_recv_sge;
+	cmd->max_inline_data = init_attr->cap.max_inline_data;
+	cmd->sq_sig_all = (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR) ? 1 : 0;
+	cmd->qp_type = ib_qp_type_to_pvrdma(init_attr->qp_type);
+	cmd->is_srq = 0; /* No support */
+	cmd->lkey = 0;
+	cmd->access_flags = IB_ACCESS_LOCAL_WRITE;
+	cmd->total_chunks = qp->npages;
+	cmd->send_chunks = qp->npages_send - 1;
+	cmd->pdir_dma = qp->pdir.dir_dma;
+
+	dev_dbg(&dev->pdev->dev, "create queuepair with %d, %d, %d, %d\n",
+		cmd->max_send_wr, cmd->max_recv_wr, cmd->max_send_sge,
+		cmd->max_recv_sge);
+
+	ret = pvrdma_cmd_post(dev, &req, true, &rsp);
+
+	if (ret < 0 || resp->hdr.ack != PVRDMA_CMD_CREATE_QP_RESP) {
+		dev_warn(&dev->pdev->dev, "could not create queuepair\n");
+		goto err_pdir;
+	}
+
+	/* max_send_wr/_recv_wr/_send_sge/_recv_sge/_inline_data */
+	qp->qp_handle = resp->qpn;
+	qp->port = init_attr->port_num;
+	qp->ibqp.qp_num = resp->qpn;
+	spin_lock_irqsave(&dev->qp_tbl_lock, flags);
+	dev->qp_tbl[qp->qp_handle % dev->dsr->caps.max_qp] = qp;
+	spin_unlock_irqrestore(&dev->qp_tbl_lock, flags);
+
+	return &qp->ibqp;
+
+err_pdir:
+	pvrdma_page_dir_cleanup(dev, &qp->pdir);
+err_umem:
+	if (pd->uobject && udata) {
+		if (qp->rumem)
+			ib_umem_release(qp->rumem);
+		if (qp->sumem)
+			ib_umem_release(qp->sumem);
+	}
+err_qp:
+	kfree(qp);
+	atomic_dec(&dev->num_qps);
+
+	return ERR_PTR(ret);
+}
+
+static void pvrdma_free_qp(struct pvrdma_qp *qp)
+{
+	struct pvrdma_dev *dev = to_vdev(qp->ibqp.device);
+	struct pvrdma_cq *scq;
+	struct pvrdma_cq *rcq;
+	unsigned long flags, flags1, flags2;
+
+	/* In case cq is polling */
+	get_cqs(qp, &scq, &rcq);
+	if ((uintptr_t)scq <= (uintptr_t)rcq) {
+		spin_lock_irqsave(&scq->cq_lock, flags1);
+		pvrdma_flush_cqe(qp, scq);
+		if (rcq != scq) {
+			spin_lock_irqsave(&rcq->cq_lock, flags2);
+			pvrdma_flush_cqe(qp, rcq);
+		}
+	} else {
+		spin_lock_irqsave(&rcq->cq_lock, flags2);
+		pvrdma_flush_cqe(qp, rcq);
+		spin_lock_irqsave(&scq->cq_lock, flags1);
+		pvrdma_flush_cqe(qp, scq);
+	}
+
+	spin_lock_irqsave(&dev->qp_tbl_lock, flags);
+	dev->qp_tbl[qp->qp_handle] = NULL;
+	spin_unlock_irqrestore(&dev->qp_tbl_lock, flags);
+
+	if ((uintptr_t)scq <= (uintptr_t)rcq) {
+		if (rcq != scq)
+			spin_unlock_irqrestore(&rcq->cq_lock, flags2);
+		spin_unlock_irqrestore(&scq->cq_lock, flags1);
+	} else {
+		spin_unlock_irqrestore(&scq->cq_lock, flags1);
+		spin_unlock_irqrestore(&rcq->cq_lock, flags2);
+	}
+
+	atomic_dec(&qp->refcnt);
+	wait_event(qp->wait, !atomic_read(&qp->refcnt));
+
+	pvrdma_page_dir_cleanup(dev, &qp->pdir);
+
+	kfree(qp);
+
+	atomic_dec(&dev->num_qps);
+}
+
+/**
+ * pvrdma_destroy_qp - destroy a queue pair
+ * @qp: the queue pair to destroy
+ *
+ * @return: 0 on success.
+ */
+int pvrdma_destroy_qp(struct ib_qp *qp)
+{
+	struct pvrdma_qp *vqp = to_vqp(qp);
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_destroy_qp *cmd = &req.destroy_qp;
+	int ret;
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_DESTROY_QP;
+	cmd->qp_handle = vqp->qp_handle;
+
+	ret = pvrdma_cmd_post(to_vdev(qp->device), &req, false, &rsp);
+	if (ret < 0) {
+		struct pvrdma_dev *dev = to_vdev(qp->device);
+
+		dev_warn(&dev->pdev->dev, "destroy queuepair failed\n");
+	}
+
+	pvrdma_free_qp(vqp);
+
+	return 0;
+}
+
+/**
+ * pvrdma_modify_qp - modify queue pair attributes
+ * @ibqp: the queue pair
+ * @attr: the new queue pair's attributes
+ * @attr_mask: attributes mask
+ * @udata: user data
+ *
+ * @returns 0 on success, otherwise returns an errno.
+ */
+int pvrdma_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		     int attr_mask, struct ib_udata *udata)
+{
+	struct pvrdma_dev *dev = to_vdev(ibqp->device);
+	struct pvrdma_qp *qp = to_vqp(ibqp);
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_modify_qp *cmd = &req.modify_qp;
+	int cur_state, next_state;
+	int ret;
+
+	/* Sanity checking. Should need lock here */
+	mutex_lock(&qp->mutex);
+	cur_state = (attr_mask & IB_QP_CUR_STATE) ? attr->cur_qp_state :
+		qp->state;
+	next_state = (attr_mask & IB_QP_STATE) ? attr->qp_state : cur_state;
+
+	if (!ib_modify_qp_is_ok(cur_state, next_state, ibqp->qp_type,
+				attr_mask, IB_LINK_LAYER_ETHERNET)) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (attr_mask & IB_QP_PORT) {
+		if (attr->port_num == 0 ||
+		    attr->port_num > ibqp->device->phys_port_cnt) {
+			ret = -EINVAL;
+			goto out;
+		}
+	}
+
+	if (attr_mask & IB_QP_MIN_RNR_TIMER) {
+		if (attr->min_rnr_timer > 31) {
+			ret = -EINVAL;
+			goto out;
+		}
+	}
+
+	if (attr_mask & IB_QP_PKEY_INDEX) {
+		if (attr->pkey_index >= dev->dsr->caps.max_pkeys) {
+			ret = -EINVAL;
+			goto out;
+		}
+	}
+
+	if (attr_mask & IB_QP_QKEY)
+		qp->qkey = attr->qkey;
+
+	if (cur_state == next_state && cur_state == IB_QPS_RESET) {
+		ret = 0;
+		goto out;
+	}
+
+	switch (next_state) {
+	case IB_QPS_RESET:
+	case IB_QPS_INIT:
+	case IB_QPS_RTR:
+	case IB_QPS_RTS:
+	case IB_QPS_SQD:
+	case IB_QPS_SQE:
+	case IB_QPS_ERR:
+	default:
+		qp->state = next_state;
+		memset(cmd, 0, sizeof(*cmd));
+		cmd->hdr.cmd = PVRDMA_CMD_MODIFY_QP;
+		cmd->qp_handle = qp->qp_handle;
+		cmd->attr_mask = ib_qp_attr_mask_to_pvrdma(attr_mask);
+		cmd->attrs.qp_state = ib_qp_state_to_pvrdma(attr->qp_state);
+		cmd->attrs.cur_qp_state =
+			ib_qp_state_to_pvrdma(attr->cur_qp_state);
+		cmd->attrs.path_mtu = ib_mtu_to_pvrdma(attr->path_mtu);
+		cmd->attrs.path_mig_state =
+			ib_mig_state_to_pvrdma(attr->path_mig_state);
+		cmd->attrs.qkey = attr->qkey;
+		cmd->attrs.rq_psn = attr->rq_psn;
+		cmd->attrs.sq_psn = attr->sq_psn;
+		cmd->attrs.dest_qp_num = attr->dest_qp_num;
+		cmd->attrs.qp_access_flags =
+			ib_access_flags_to_pvrdma(attr->qp_access_flags);
+		cmd->attrs.pkey_index = attr->pkey_index;
+		cmd->attrs.alt_pkey_index = attr->alt_pkey_index;
+		cmd->attrs.en_sqd_async_notify = attr->en_sqd_async_notify;
+		cmd->attrs.sq_draining = attr->sq_draining;
+		cmd->attrs.max_rd_atomic = attr->max_rd_atomic;
+		cmd->attrs.max_dest_rd_atomic = attr->max_dest_rd_atomic;
+		cmd->attrs.min_rnr_timer = attr->min_rnr_timer;
+		cmd->attrs.port_num = attr->port_num;
+		cmd->attrs.timeout = attr->timeout;
+		cmd->attrs.retry_cnt = attr->retry_cnt;
+		cmd->attrs.rnr_retry = attr->rnr_retry;
+		cmd->attrs.alt_port_num = attr->alt_port_num;
+		cmd->attrs.alt_timeout = attr->alt_timeout;
+		ib_qp_cap_to_pvrdma(&attr->cap, &cmd->attrs.cap);
+		ib_ah_attr_to_pvrdma(&attr->ah_attr, &cmd->attrs.ah_attr);
+		ib_ah_attr_to_pvrdma(&attr->alt_ah_attr,
+				     &cmd->attrs.alt_ah_attr);
+		ret = pvrdma_cmd_post(dev, &req, true, &rsp);
+		if (ret < 0 || rsp.hdr.ack != PVRDMA_CMD_MODIFY_QP_RESP ||
+		    rsp.hdr.err > 0) {
+			dev_warn(&dev->pdev->dev,
+				 "could not modify queuepair\n");
+			if (ret == 0) {
+				if (rsp.hdr.ack != PVRDMA_CMD_MODIFY_QP_RESP)
+					ret = -EINVAL;
+				else
+					ret = rsp.hdr.err;
+			}
+		}
+
+		/* Fall through */
+		break;
+	}
+
+	if (ret == 0 && next_state == IB_QPS_RESET)
+		pvrdma_reset_qp(qp);
+
+out:
+	mutex_unlock(&qp->mutex);
+
+	return ret;
+}
+
+static inline void *get_sq_wqe(struct pvrdma_qp *qp, int n)
+{
+	return pvrdma_page_dir_get_ptr(&qp->pdir,
+				       qp->sq.offset + n * qp->sq.wqe_size);
+}
+
+static inline void *get_rq_wqe(struct pvrdma_qp *qp, int n)
+{
+	return pvrdma_page_dir_get_ptr(&qp->pdir,
+				       qp->rq.offset + n * qp->rq.wqe_size);
+}
+
+static int set_reg_seg(struct pvrdma_sq_wqe_hdr *wqe_hdr, struct ib_reg_wr *wr)
+{
+	struct pvrdma_user_mr *mr = to_vmr(wr->mr);
+
+	wqe_hdr->wr.fast_reg.iova_start = mr->ibmr.iova;
+	wqe_hdr->wr.fast_reg.pl_pdir_dma = mr->pdir.dir_dma;
+	wqe_hdr->wr.fast_reg.page_shift = mr->page_shift;
+	wqe_hdr->wr.fast_reg.page_list_len = mr->npages;
+	wqe_hdr->wr.fast_reg.length = mr->ibmr.length;
+	wqe_hdr->wr.fast_reg.access_flags = wr->access;
+	wqe_hdr->wr.fast_reg.rkey = wr->key;
+
+	return pvrdma_page_dir_insert_page_list(&mr->pdir, mr->pages,
+						mr->npages);
+}
+
+/**
+ * pvrdma_post_send - post send work request entries on a QP
+ * @ibqp: the QP
+ * @wr: work request list to post
+ * @bad_wr: the first bad WR returned
+ *
+ * @return: 0 on success, otherwise errno returned.
+ */
+int pvrdma_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+		     struct ib_send_wr **bad_wr)
+{
+	struct pvrdma_qp *qp = to_vqp(ibqp);
+	struct pvrdma_dev *dev = to_vdev(ibqp->device);
+	unsigned long flags;
+	struct pvrdma_sq_wqe_hdr *wqe_hdr;
+	struct ib_sge *sge;
+	int i, index;
+	int nreq;
+	int ret;
+
+	/*
+	 * In states lower than RTS, we can fail immediately. In other states,
+	 * just post and let the device figure it out.
+	 */
+	if (qp->state < IB_QPS_RTS) {
+		*bad_wr = wr;
+		return -EINVAL;
+	}
+
+	spin_lock_irqsave(&qp->sq.lock, flags);
+
+	index = pvrdma_idx(&qp->sq.ring->prod_tail, qp->sq.wqe_cnt);
+	for (nreq = 0; wr; nreq++, wr = wr->next) {
+		unsigned int tail;
+
+		if (unlikely(!pvrdma_idx_ring_has_space(
+				qp->sq.ring, qp->sq.wqe_cnt, &tail))) {
+			dev_warn(&dev->pdev->dev, "send queue is full\n");
+			*bad_wr = wr;
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		if (unlikely(wr->num_sge > qp->sq.max_gs || wr->num_sge < 0)) {
+			dev_warn(&dev->pdev->dev, "send SGE overflow\n");
+			*bad_wr = wr;
+			ret = -EINVAL;
+			goto out;
+		}
+
+		if (unlikely(wr->opcode < 0)) {
+			dev_warn(&dev->pdev->dev, "invalid send opcode\n");
+			*bad_wr = wr;
+			ret = -EINVAL;
+			goto out;
+		}
+
+		/*
+		 * Only support UD, RC.
+		 * Need to check opcode table for thorough checking.
+		 * opcode		_UD	_UC	_RC
+		 * _SEND		x	x	x
+		 * _SEND_WITH_IMM	x	x	x
+		 * _RDMA_WRITE			x	x
+		 * _RDMA_WRITE_WITH_IMM		x	x
+		 * _LOCAL_INV			x	x
+		 * _SEND_WITH_INV		x	x
+		 * _RDMA_READ				x
+		 * _ATOMIC_CMP_AND_SWP			x
+		 * _ATOMIC_FETCH_AND_ADD		x
+		 * _MASK_ATOMIC_CMP_AND_SWP		x
+		 * _MASK_ATOMIC_FETCH_AND_ADD		x
+		 * _REG_MR				x
+		 *
+		 */
+		if (qp->ibqp.qp_type != IB_QPT_UD &&
+		    qp->ibqp.qp_type != IB_QPT_RC &&
+			wr->opcode != IB_WR_SEND) {
+			dev_warn(&dev->pdev->dev,
+				 "unsupported queuepair type\n");
+			*bad_wr = wr;
+			ret = -EINVAL;
+			goto out;
+		} else if (qp->ibqp.qp_type == IB_QPT_UD ||
+			   qp->ibqp.qp_type == IB_QPT_GSI) {
+			if (wr->opcode != IB_WR_SEND &&
+			    wr->opcode != IB_WR_SEND_WITH_IMM) {
+				dev_warn(&dev->pdev->dev,
+					 "invalid send opcode\n");
+				*bad_wr = wr;
+				ret = -EINVAL;
+				goto out;
+			}
+		}
+
+		wqe_hdr = (struct pvrdma_sq_wqe_hdr *)get_sq_wqe(qp, index);
+		memset(wqe_hdr, 0, sizeof(*wqe_hdr));
+		wqe_hdr->wr_id = wr->wr_id;
+		wqe_hdr->num_sge = wr->num_sge;
+		wqe_hdr->opcode = ib_wr_opcode_to_pvrdma(wr->opcode);
+		wqe_hdr->send_flags = ib_send_flags_to_pvrdma(wr->send_flags);
+		if (wr->opcode == IB_WR_SEND_WITH_IMM ||
+		    wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM)
+			wqe_hdr->ex.imm_data = wr->ex.imm_data;
+
+		switch (qp->ibqp.qp_type) {
+		case IB_QPT_GSI:
+		case IB_QPT_UD:
+			if (unlikely(!ud_wr(wr)->ah)) {
+				dev_warn(&dev->pdev->dev,
+					 "invalid address handle\n");
+				*bad_wr = wr;
+				ret = -EINVAL;
+				goto out;
+			}
+
+			/*
+			 * Use qkey from qp context if high order bit set,
+			 * otherwise from work request.
+			 */
+			wqe_hdr->wr.ud.remote_qpn = ud_wr(wr)->remote_qpn;
+			wqe_hdr->wr.ud.remote_qkey =
+				ud_wr(wr)->remote_qkey & 0x80000000 ?
+				qp->qkey : ud_wr(wr)->remote_qkey;
+			wqe_hdr->wr.ud.av = to_vah(ud_wr(wr)->ah)->av;
+
+			break;
+		case IB_QPT_RC:
+			switch (wr->opcode) {
+			case IB_WR_RDMA_READ:
+			case IB_WR_RDMA_WRITE:
+			case IB_WR_RDMA_WRITE_WITH_IMM:
+				wqe_hdr->wr.rdma.remote_addr =
+					rdma_wr(wr)->remote_addr;
+				wqe_hdr->wr.rdma.rkey = rdma_wr(wr)->rkey;
+				break;
+			case IB_WR_LOCAL_INV:
+			case IB_WR_SEND_WITH_INV:
+				wqe_hdr->ex.invalidate_rkey =
+					wr->ex.invalidate_rkey;
+				break;
+			case IB_WR_ATOMIC_CMP_AND_SWP:
+			case IB_WR_ATOMIC_FETCH_AND_ADD:
+				wqe_hdr->wr.atomic.remote_addr =
+					atomic_wr(wr)->remote_addr;
+				wqe_hdr->wr.atomic.rkey = atomic_wr(wr)->rkey;
+				wqe_hdr->wr.atomic.compare_add =
+					atomic_wr(wr)->compare_add;
+				if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP)
+					wqe_hdr->wr.atomic.swap =
+						atomic_wr(wr)->swap;
+				break;
+			case IB_WR_REG_MR:
+				ret = set_reg_seg(wqe_hdr, reg_wr(wr));
+				if (ret < 0) {
+					dev_warn(&dev->pdev->dev,
+						 "Failed to set fast register work request\n");
+					*bad_wr = wr;
+					goto out;
+				}
+				break;
+			default:
+				break;
+			}
+
+			break;
+		default:
+			dev_warn(&dev->pdev->dev, "invalid queuepair type\n");
+			ret = -EINVAL;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		sge = (struct ib_sge *)(wqe_hdr + 1);
+		for (i = 0; i < wr->num_sge; i++) {
+			/* Need to check wqe_size 0 or max size */
+			sge->addr = wr->sg_list[i].addr;
+			sge->length = wr->sg_list[i].length;
+			sge->lkey = wr->sg_list[i].lkey;
+			sge++;
+		}
+
+		/* Make sure wqe is written before index update */
+		smp_wmb();
+
+		index++;
+		if (unlikely(index >= qp->sq.wqe_cnt))
+			index = 0;
+		/* Update shared sq ring */
+		pvrdma_idx_ring_inc(&qp->sq.ring->prod_tail,
+				    qp->sq.wqe_cnt);
+	}
+
+	ret = 0;
+
+out:
+	spin_unlock_irqrestore(&qp->sq.lock, flags);
+
+	if (!ret)
+		writel(cpu_to_le32(PVRDMA_UAR_QP_SEND | qp->qp_handle),
+		       dev->driver_uar.map + PVRDMA_UAR_QP_OFFSET);
+
+	return ret;
+}
+
+/**
+ * pvrdma_post_receive - post receive work request entries on a QP
+ * @ibqp: the QP
+ * @wr: the work request list to post
+ * @bad_wr: the first bad WR returned
+ *
+ * @return: 0 on success, otherwise errno returned.
+ */
+int pvrdma_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+		     struct ib_recv_wr **bad_wr)
+{
+	struct pvrdma_dev *dev = to_vdev(ibqp->device);
+	unsigned long flags;
+	struct pvrdma_qp *qp = to_vqp(ibqp);
+	struct pvrdma_rq_wqe_hdr *wqe_hdr;
+	struct ib_sge *sge;
+	int index, nreq;
+	int ret = 0;
+	int i;
+
+	/*
+	 * In the RESET state, we can fail immediately. For other states,
+	 * just post and let the device figure it out.
+	 */
+	if (qp->state == IB_QPS_RESET) {
+		*bad_wr = wr;
+		return -EINVAL;
+	}
+
+	spin_lock_irqsave(&qp->rq.lock, flags);
+
+	index = pvrdma_idx(&qp->rq.ring->prod_tail, qp->rq.wqe_cnt);
+	for (nreq = 0; wr; nreq++, wr = wr->next) {
+		unsigned int tail;
+
+		if (unlikely(wr->num_sge > qp->rq.max_gs ||
+			     wr->num_sge < 0)) {
+			ret = -EINVAL;
+			*bad_wr = wr;
+			dev_warn(&dev->pdev->dev, "recv SGE overflow\n");
+			goto out;
+		}
+
+		if (unlikely(!pvrdma_idx_ring_has_space(
+				qp->rq.ring, qp->rq.wqe_cnt, &tail))) {
+			ret = -ENOMEM;
+			*bad_wr = wr;
+			dev_warn(&dev->pdev->dev, "recv queue full\n");
+			goto out;
+		}
+
+		wqe_hdr = (struct pvrdma_rq_wqe_hdr *)get_rq_wqe(qp, index);
+		wqe_hdr->wr_id = wr->wr_id;
+		wqe_hdr->num_sge = wr->num_sge;
+		wqe_hdr->total_len = 0;
+
+		sge = (struct ib_sge *)(wqe_hdr + 1);
+		for (i = 0; i < wr->num_sge; i++) {
+			sge->addr = wr->sg_list[i].addr;
+			sge->length = wr->sg_list[i].length;
+			sge->lkey = wr->sg_list[i].lkey;
+			sge++;
+		}
+
+		/* Make sure wqe is written before index update */
+		smp_wmb();
+
+		index++;
+		if (unlikely(index >= qp->rq.wqe_cnt))
+			index = 0;
+		/* Update shared rq ring */
+		pvrdma_idx_ring_inc(&qp->rq.ring->prod_tail,
+				    qp->rq.wqe_cnt);
+	}
+
+	spin_unlock_irqrestore(&qp->rq.lock, flags);
+
+	writel(cpu_to_le32(PVRDMA_UAR_QP_RECV | qp->qp_handle),
+	       dev->driver_uar.map + PVRDMA_UAR_QP_OFFSET);
+
+	return ret;
+
+out:
+	spin_unlock_irqrestore(&qp->rq.lock, flags);
+
+	return ret;
+}
+
+/**
+ * pvrdma_query_qp - query a queue pair's attributes
+ * @ibqp: the queue pair to query
+ * @attr: the queue pair's attributes
+ * @attr_mask: attributes mask
+ * @init_attr: initial queue pair attributes
+ *
+ * @returns 0 on success, otherwise returns an errno.
+ */
+int pvrdma_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		    int attr_mask, struct ib_qp_init_attr *init_attr)
+{
+	struct pvrdma_dev *dev = to_vdev(ibqp->device);
+	struct pvrdma_qp *qp = to_vqp(ibqp);
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_query_qp *cmd = &req.query_qp;
+	struct pvrdma_cmd_query_qp_resp *resp = &rsp.query_qp_resp;
+	int ret = 0;
+
+	mutex_lock(&qp->mutex);
+
+	if (qp->state == IB_QPS_RESET) {
+		attr->qp_state = IB_QPS_RESET;
+		goto out;
+	}
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_QUERY_QP;
+	cmd->qp_handle = qp->qp_handle;
+	cmd->attr_mask = ib_qp_attr_mask_to_pvrdma(attr_mask);
+
+	ret = pvrdma_cmd_post(dev, &req, true, &rsp);
+	if (ret < 0 || resp->hdr.ack != PVRDMA_CMD_QUERY_QP_RESP) {
+		dev_warn(&dev->pdev->dev, "could not query queuepair\n");
+		ret = -EINVAL;
+		goto out;
+	}
+
+	attr->qp_state = pvrdma_qp_state_to_ib(resp->attrs.qp_state);
+	attr->cur_qp_state =
+		pvrdma_qp_state_to_ib(resp->attrs.cur_qp_state);
+	attr->path_mtu = pvrdma_mtu_to_ib(resp->attrs.path_mtu);
+	attr->path_mig_state =
+		pvrdma_mig_state_to_ib(resp->attrs.path_mig_state);
+	attr->qkey = resp->attrs.qkey;
+	attr->rq_psn = resp->attrs.rq_psn;
+	attr->sq_psn = resp->attrs.sq_psn;
+	attr->dest_qp_num = resp->attrs.dest_qp_num;
+	attr->qp_access_flags =
+		pvrdma_access_flags_to_ib(resp->attrs.qp_access_flags);
+	attr->pkey_index = resp->attrs.pkey_index;
+	attr->alt_pkey_index = resp->attrs.alt_pkey_index;
+	attr->en_sqd_async_notify = resp->attrs.en_sqd_async_notify;
+	attr->sq_draining = resp->attrs.sq_draining;
+	attr->max_rd_atomic = resp->attrs.max_rd_atomic;
+	attr->max_dest_rd_atomic = resp->attrs.max_dest_rd_atomic;
+	attr->min_rnr_timer = resp->attrs.min_rnr_timer;
+	attr->port_num = resp->attrs.port_num;
+	attr->timeout = resp->attrs.timeout;
+	attr->retry_cnt = resp->attrs.retry_cnt;
+	attr->rnr_retry = resp->attrs.rnr_retry;
+	attr->alt_port_num = resp->attrs.alt_port_num;
+	attr->alt_timeout = resp->attrs.alt_timeout;
+	pvrdma_qp_cap_to_ib(&resp->attrs.cap, &attr->cap);
+	pvrdma_ah_attr_to_ib(&resp->attrs.ah_attr, &attr->ah_attr);
+	pvrdma_ah_attr_to_ib(&resp->attrs.alt_ah_attr, &attr->alt_ah_attr);
+
+	qp->state = attr->qp_state;
+
+	ret = 0;
+
+out:
+	attr->cur_qp_state = attr->qp_state;
+
+	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->xrcd = NULL;
+	init_attr->cap = attr->cap;
+	init_attr->sq_sig_type = 0;
+	init_attr->qp_type = qp->ibqp.qp_type;
+	init_attr->create_flags = 0;
+	init_attr->port_num = qp->port;
+
+	mutex_unlock(&qp->mutex);
+	return ret;
+}
-- 
2.7.4

--
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 11/15] IB/pvrdma: Add user-level shared functions

[Adit Ranadive]

We share some common structures with the user-level driver. This patch
adds those structures and shared functions to traverse the QP/CQ rings.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/pvrdma/pvrdma_uapi.h | 247 +++++++++++++++++++++++++++++
 drivers/infiniband/hw/pvrdma/pvrdma_user.h |  99 ++++++++++++
 2 files changed, 346 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_uapi.h
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_user.h

diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_uapi.h b/drivers/infiniband/hw/pvrdma/pvrdma_uapi.h
new file mode 100644
index 0000000..274328c
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_uapi.h
@@ -0,0 +1,247 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef __PVRDMA_UAPI_H__
+#define __PVRDMA_UAPI_H__
+
+#include <linux/types.h>
+
+#define PVRDMA_VERSION 17
+
+#define PVRDMA_UAR_HANDLE_MASK	0x00FFFFFF	/* Bottom 24 bits. */
+#define PVRDMA_UAR_QP_OFFSET	0		/* Offset of QP doorbell. */
+#define PVRDMA_UAR_QP_SEND	BIT(30)		/* Send bit. */
+#define PVRDMA_UAR_QP_RECV	BIT(31)		/* Recv bit. */
+#define PVRDMA_UAR_CQ_OFFSET	4		/* Offset of CQ doorbell. */
+#define PVRDMA_UAR_CQ_ARM_SOL	BIT(29)		/* Arm solicited bit. */
+#define PVRDMA_UAR_CQ_ARM	BIT(30)		/* Arm bit. */
+#define PVRDMA_UAR_CQ_POLL	BIT(31)		/* Poll bit. */
+
+/* PVRDMA atomic compare and swap */
+struct pvrdma_exp_cmp_swap {
+	__u64 swap_val;
+	__u64 compare_val;
+	__u64 swap_mask;
+	__u64 compare_mask;
+};
+
+/* PVRDMA atomic fetch and add */
+struct pvrdma_exp_fetch_add {
+	__u64 add_val;
+	__u64 field_boundary;
+};
+
+/* PVRDMA address vector. */
+struct pvrdma_av {
+	__u32 port_pd;
+	__u32 sl_tclass_flowlabel;
+	__u8 dgid[16];
+	__u8 src_path_bits;
+	__u8 gid_index;
+	__u8 stat_rate;
+	__u8 hop_limit;
+	__u8 dmac[6];
+	__u8 reserved[6];
+};
+
+/* PVRDMA receive queue work request */
+struct pvrdma_rq_wqe_hdr {
+	__u64 wr_id;		/* wr id */
+	__u32 num_sge;		/* size of s/g array */
+	__u32 total_len;	/* reserved */
+};
+/* Use pvrdma_sge (ib_sge) for receieve queue s/g array elements. */
+
+/* PVRDMA send queue work request */
+struct pvrdma_sq_wqe_hdr {
+	__u64 wr_id;		/* wr id */
+	__u32 num_sge;		/* size of s/g array */
+	__u32 total_len;	/* reserved */
+	__u32 opcode;		/* operation type */
+	__u32 send_flags;	/* wr flags */
+	union {
+		__u32 imm_data;
+		__u32 invalidate_rkey;
+	} ex;
+	__u32 reserved;
+	union {
+		struct {
+			__u64 remote_addr;
+			__u32 rkey;
+			__u8 reserved[4];
+		} rdma;
+		struct {
+			__u64 remote_addr;
+			__u64 compare_add;
+			__u64 swap;
+			__u32 rkey;
+			__u32 reserved;
+		} atomic;
+		struct {
+			__u64 remote_addr;
+			__u32 log_arg_sz;
+			__u32 rkey;
+			union {
+				struct pvrdma_exp_cmp_swap  cmp_swap;
+				struct pvrdma_exp_fetch_add fetch_add;
+			} wr_data;
+		} masked_atomics;
+		struct {
+			__u64 iova_start;
+			__u64 pl_pdir_dma;
+			__u32 page_shift;
+			__u32 page_list_len;
+			__u32 length;
+			__u32 access_flags;
+			__u32 rkey;
+		} fast_reg;
+		struct {
+			__u32 remote_qpn;
+			__u32 remote_qkey;
+			struct pvrdma_av av;
+		} ud;
+	} wr;
+};
+/* Use pvrdma_sge (ib_sge) for send queue s/g array elements. */
+
+/* Completion queue element. */
+struct pvrdma_cqe {
+	__u64 wr_id;
+	__u64 qp;
+	__u32 opcode;
+	__u32 status;
+	__u32 byte_len;
+	__u32 imm_data;
+	__u32 src_qp;
+	__u32 wc_flags;
+	__u32 vendor_err;
+	__u16 pkey_index;
+	__u16 slid;
+	__u8 sl;
+	__u8 dlid_path_bits;
+	__u8 port_num;
+	__u8 smac[6];
+	__u8 reserved2[7]; /* Pad to next power of 2 (64). */
+};
+
+struct pvrdma_ring {
+	atomic_t prod_tail;	/* Producer tail. */
+	atomic_t cons_head;	/* Consumer head. */
+};
+
+struct pvrdma_ring_state {
+	struct pvrdma_ring tx;	/* Tx ring. */
+	struct pvrdma_ring rx;	/* Rx ring. */
+};
+
+static inline int pvrdma_idx_valid(__u32 idx, __u32 max_elems)
+{
+	/* Generates fewer instructions than a less-than. */
+	return (idx & ~((max_elems << 1) - 1)) == 0;
+}
+
+static inline __s32 pvrdma_idx(atomic_t *var, __u32 max_elems)
+{
+	const unsigned int idx = atomic_read(var);
+
+	if (pvrdma_idx_valid(idx, max_elems))
+		return idx & (max_elems - 1);
+	return -1;
+}
+
+static inline void pvrdma_idx_ring_inc(atomic_t *var, __u32 max_elems)
+{
+	__u32 idx = atomic_read(var) + 1;	/* Increment. */
+
+	idx &= (max_elems << 1) - 1;		/* Modulo size, flip gen. */
+	atomic_set(var, idx);
+}
+
+static inline __s32 pvrdma_idx_ring_has_space(const struct pvrdma_ring *r,
+					      __u32 max_elems, __u32 *out_tail)
+{
+	const __u32 tail = atomic_read(&r->prod_tail);
+	const __u32 head = atomic_read(&r->cons_head);
+
+	if (pvrdma_idx_valid(tail, max_elems) &&
+	    pvrdma_idx_valid(head, max_elems)) {
+		*out_tail = tail & (max_elems - 1);
+		return tail != (head ^ max_elems);
+	}
+	return -1;
+}
+
+static inline __s32 pvrdma_idx_ring_has_data(const struct pvrdma_ring *r,
+					     __u32 max_elems, __u32 *out_head)
+{
+	const __u32 tail = atomic_read(&r->prod_tail);
+	const __u32 head = atomic_read(&r->cons_head);
+
+	if (pvrdma_idx_valid(tail, max_elems) &&
+	    pvrdma_idx_valid(head, max_elems)) {
+		*out_head = head & (max_elems - 1);
+		return tail != head;
+	}
+	return -1;
+}
+
+static inline __s32 pvrdma_idx_ring_is_valid_idx(const struct pvrdma_ring *r,
+						 __u32 max_elems, __u32 *idx)
+{
+	const __u32 tail = atomic_read(&r->prod_tail);
+	const __u32 head = atomic_read(&r->cons_head);
+
+	if (pvrdma_idx_valid(tail, max_elems) &&
+	    pvrdma_idx_valid(head, max_elems) &&
+	    pvrdma_idx_valid(*idx, max_elems)) {
+		if (tail > head && (*idx < tail && *idx >= head))
+			return 1;
+		else if (head > tail && (*idx >= head || *idx < tail))
+			return 1;
+	}
+	return 0;
+}
+
+#endif /* __PVRDMA_UAPI_H__ */
diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_user.h b/drivers/infiniband/hw/pvrdma/pvrdma_user.h
new file mode 100644
index 0000000..35f8c1e
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_user.h
@@ -0,0 +1,99 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef __PVRDMA_USER_H__
+#define __PVRDMA_USER_H__
+
+#include <linux/types.h>
+
+#define PVRDMA_UVERBS_ABI_VERSION	3
+#define PVRDMA_BOARD_ID			1
+#define PVRDMA_REV_ID			1
+
+struct pvrdma_alloc_ucontext_resp {
+	u32 qp_tab_size;
+	u32 reserved;
+};
+
+struct pvrdma_alloc_pd_resp {
+	u32 pdn;
+	u32 reserved;
+};
+
+struct pvrdma_create_cq {
+	u64 buf_addr;
+	u32 buf_size;
+	u32 reserved;
+};
+
+struct pvrdma_create_cq_resp {
+	u32 cqn;
+	u32 reserved;
+};
+
+struct pvrdma_resize_cq {
+	u64 buf_addr;
+	u32 buf_size;
+	u32 reserved;
+};
+
+struct pvrdma_create_srq {
+	u64 buf_addr;
+};
+
+struct pvrdma_create_srq_resp {
+	u32 srqn;
+	u32 reserved;
+};
+
+struct pvrdma_create_qp {
+	u64 rbuf_addr;
+	u64 sbuf_addr;
+	u32 rbuf_size;
+	u32 sbuf_size;
+	u64 qp_addr;
+};
+
+#endif /* __PVRDMA_USER_H__ */
-- 
2.7.4

--
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 12/15] IB/pvrdma: Add functions for Verbs support

[Adit Ranadive]

This patch implements the remaining Verbs functions registered with the
core RDMA stack.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/pvrdma/pvrdma_verbs.c | 593 ++++++++++++++++++++++++++++
 drivers/infiniband/hw/pvrdma/pvrdma_verbs.h | 108 +++++
 2 files changed, 701 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_verbs.c
 create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_verbs.h

diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_verbs.c b/drivers/infiniband/hw/pvrdma/pvrdma_verbs.c
new file mode 100644
index 0000000..bfc5f89
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_verbs.c
@@ -0,0 +1,593 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <asm/page.h>
+#include <linux/inet.h>
+#include <linux/io.h>
+#include <rdma/ib_addr.h>
+#include <rdma/ib_smi.h>
+#include <rdma/ib_user_verbs.h>
+
+#include "pvrdma.h"
+#include "pvrdma_user.h"
+
+/**
+ * pvrdma_query_device - query device
+ * @ibdev: the device to query
+ * @props: the device properties
+ * @uhw: user data
+ *
+ * @return: 0 on success, otherwise negative errno
+ */
+int pvrdma_query_device(struct ib_device *ibdev,
+			struct ib_device_attr *props,
+			struct ib_udata *uhw)
+{
+	struct pvrdma_dev *dev = to_vdev(ibdev);
+
+	if (uhw->inlen || uhw->outlen)
+		return -EINVAL;
+
+	memset(props, 0, sizeof(*props));
+
+	props->fw_ver = dev->dsr->caps.fw_ver;
+	props->sys_image_guid = dev->dsr->caps.sys_image_guid;
+	props->max_mr_size = dev->dsr->caps.max_mr_size;
+	props->page_size_cap = dev->dsr->caps.page_size_cap;
+	props->vendor_id = dev->dsr->caps.vendor_id;
+	props->vendor_part_id = dev->pdev->device;
+	props->hw_ver = dev->dsr->caps.hw_ver;
+	props->max_qp = dev->dsr->caps.max_qp;
+	props->max_qp_wr = dev->dsr->caps.max_qp_wr;
+	props->device_cap_flags = dev->dsr->caps.device_cap_flags;
+	props->max_sge = dev->dsr->caps.max_sge;
+	props->max_sge_rd = dev->dsr->caps.max_sge_rd;
+	props->max_cq = dev->dsr->caps.max_cq;
+	props->max_cqe = dev->dsr->caps.max_cqe;
+	props->max_mr = dev->dsr->caps.max_mr;
+	props->max_pd = dev->dsr->caps.max_pd;
+	props->max_qp_rd_atom = dev->dsr->caps.max_qp_rd_atom;
+	props->max_ee_rd_atom = dev->dsr->caps.max_ee_rd_atom;
+	props->max_res_rd_atom = dev->dsr->caps.max_res_rd_atom;
+	props->max_qp_init_rd_atom = dev->dsr->caps.max_qp_init_rd_atom;
+	props->max_ee_init_rd_atom = dev->dsr->caps.max_ee_init_rd_atom;
+	props->atomic_cap =
+		dev->dsr->caps.atomic_ops &
+		(PVRDMA_ATOMIC_OP_COMP_SWAP | PVRDMA_ATOMIC_OP_FETCH_ADD) ?
+		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
+	props->masked_atomic_cap = props->atomic_cap;
+	props->max_ee = dev->dsr->caps.max_ee;
+	props->max_rdd = dev->dsr->caps.max_rdd;
+	props->max_mw = dev->dsr->caps.max_mw;
+	props->max_raw_ipv6_qp = dev->dsr->caps.max_raw_ipv6_qp;
+	props->max_raw_ethy_qp = dev->dsr->caps.max_raw_ethy_qp;
+	props->max_mcast_grp = dev->dsr->caps.max_mcast_grp;
+	props->max_mcast_qp_attach = dev->dsr->caps.max_mcast_qp_attach;
+	props->max_total_mcast_qp_attach =
+		dev->dsr->caps.max_total_mcast_qp_attach;
+	props->max_ah = dev->dsr->caps.max_ah;
+	props->max_fmr = dev->dsr->caps.max_fmr;
+	props->max_map_per_fmr = dev->dsr->caps.max_map_per_fmr;
+	props->max_srq = dev->dsr->caps.max_srq;
+	props->max_srq_wr = dev->dsr->caps.max_srq_wr;
+	props->max_srq_sge = dev->dsr->caps.max_srq_sge;
+	props->max_fast_reg_page_list_len = 0;
+	props->max_pkeys = dev->dsr->caps.max_pkeys;
+	props->local_ca_ack_delay = dev->dsr->caps.local_ca_ack_delay;
+	if ((dev->dsr->caps.bmme_flags & PVRDMA_BMME_FLAG_LOCAL_INV) &&
+	    (dev->dsr->caps.bmme_flags & PVRDMA_BMME_FLAG_REMOTE_INV) &&
+	    (dev->dsr->caps.bmme_flags & PVRDMA_BMME_FLAG_FAST_REG_WR)) {
+		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
+	}
+
+	return 0;
+}
+
+/**
+ * pvrdma_query_port - query device port attributes
+ * @ibdev: the device to query
+ * @port: the port number
+ * @props: the device properties
+ *
+ * @return: 0 on success, otherwise negative errno
+ */
+int pvrdma_query_port(struct ib_device *ibdev, u8 port,
+		      struct ib_port_attr *props)
+{
+	struct pvrdma_dev *dev = to_vdev(ibdev);
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_query_port *cmd = &req.query_port;
+	struct pvrdma_cmd_query_port_resp *resp = &rsp.query_port_resp;
+	int err;
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_QUERY_PORT;
+	cmd->port_num = port;
+
+	err = pvrdma_cmd_post(dev, &req, true, &rsp);
+	if (err < 0 || rsp.hdr.ack != PVRDMA_CMD_QUERY_PORT_RESP) {
+		dev_warn(&dev->pdev->dev, "could not query port\n");
+		return -EFAULT;
+	}
+
+	memset(props, 0, sizeof(*props));
+
+	props->state = pvrdma_port_state_to_ib(resp->attrs.state);
+	props->max_mtu = pvrdma_mtu_to_ib(resp->attrs.max_mtu);
+	props->active_mtu = pvrdma_mtu_to_ib(resp->attrs.active_mtu);
+	props->gid_tbl_len = resp->attrs.gid_tbl_len;
+	props->port_cap_flags =
+		pvrdma_port_cap_flags_to_ib(resp->attrs.port_cap_flags);
+	props->max_msg_sz = resp->attrs.max_msg_sz;
+	props->bad_pkey_cntr = resp->attrs.bad_pkey_cntr;
+	props->qkey_viol_cntr = resp->attrs.qkey_viol_cntr;
+	props->pkey_tbl_len = resp->attrs.pkey_tbl_len;
+	props->lid = resp->attrs.lid;
+	props->sm_lid = resp->attrs.sm_lid;
+	props->lmc = resp->attrs.lmc;
+	props->max_vl_num = resp->attrs.max_vl_num;
+	props->sm_sl = resp->attrs.sm_sl;
+	props->subnet_timeout = resp->attrs.subnet_timeout;
+	props->init_type_reply = resp->attrs.init_type_reply;
+	props->active_width = pvrdma_port_width_to_ib(resp->attrs.active_width);
+	props->active_speed = pvrdma_port_speed_to_ib(resp->attrs.active_speed);
+	props->phys_state = resp->attrs.phys_state;
+
+	return 0;
+}
+
+/**
+ * pvrdma_query_gid - query device gid
+ * @ibdev: the device to query
+ * @port: the port number
+ * @index: the index
+ * @gid: the device gid value
+ *
+ * @return: 0 on success, otherwise negative errno
+ */
+int pvrdma_query_gid(struct ib_device *ibdev, u8 port, int index,
+		     union ib_gid *gid)
+{
+	struct pvrdma_dev *dev = to_vdev(ibdev);
+
+	if (index >= dev->dsr->caps.gid_tbl_len)
+		return -EINVAL;
+
+	memcpy(gid, &dev->sgid_tbl[index], sizeof(union ib_gid));
+
+	return 0;
+}
+
+/**
+ * pvrdma_query_pkey - query device port's P_Key table
+ * @ibdev: the device to query
+ * @port: the port number
+ * @index: the index
+ * @pkey: the device P_Key value
+ *
+ * @return: 0 on success, otherwise negative errno
+ */
+int pvrdma_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+		      u16 *pkey)
+{
+	int err = 0;
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_query_pkey *cmd = &req.query_pkey;
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_QUERY_PKEY;
+	cmd->port_num = port;
+	cmd->index = index;
+
+	err = pvrdma_cmd_post(to_vdev(ibdev), &req, true, &rsp);
+	if (err < 0 || rsp.hdr.ack != PVRDMA_CMD_QUERY_PKEY_RESP) {
+		struct pvrdma_dev *dev = to_vdev(ibdev);
+
+		dev_warn(&dev->pdev->dev, "could not query device pkey\n");
+		return -EFAULT;
+	}
+
+	*pkey = rsp.query_pkey_resp.pkey;
+
+	return 0;
+}
+
+enum rdma_link_layer pvrdma_port_link_layer(struct ib_device *ibdev,
+					    u8 port)
+{
+	return IB_LINK_LAYER_ETHERNET;
+}
+
+int pvrdma_modify_device(struct ib_device *ibdev, int mask,
+			 struct ib_device_modify *props)
+{
+	unsigned long flags;
+	int ret;
+
+	if (mask & ~(IB_DEVICE_MODIFY_SYS_IMAGE_GUID |
+		     IB_DEVICE_MODIFY_NODE_DESC)) {
+		struct pvrdma_dev *dev = to_vdev(ibdev);
+
+		dev_warn(&dev->pdev->dev, "unsupported device modify mask\n");
+		ret = -EOPNOTSUPP;
+		goto err_out;
+	}
+
+	if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
+		spin_lock_irqsave(&to_vdev(ibdev)->desc_lock, flags);
+		memcpy(ibdev->node_desc, props->node_desc, 64);
+		spin_unlock_irqrestore(&to_vdev(ibdev)->desc_lock, flags);
+	}
+
+	if (mask & IB_DEVICE_MODIFY_SYS_IMAGE_GUID) {
+		mutex_lock(&to_vdev(ibdev)->port_mutex);
+		to_vdev(ibdev)->sys_image_guid =
+			cpu_to_be64(props->sys_image_guid);
+		mutex_unlock(&to_vdev(ibdev)->port_mutex);
+	}
+
+	return 0;
+
+err_out:
+	return ret;
+}
+
+/**
+ * pvrdma_modify_port - modify device port attributes
+ * @ibdev: the device to modify
+ * @port: the port number
+ * @mask: attributes to modify
+ * @props: the device properties
+ *
+ * @return: 0 on success, otherwise negative errno
+ */
+int pvrdma_modify_port(struct ib_device *ibdev, u8 port, int mask,
+		       struct ib_port_modify *props)
+{
+	struct ib_port_attr attr;
+	struct pvrdma_dev *vdev = to_vdev(ibdev);
+	int ret;
+
+	mutex_lock(&vdev->port_mutex);
+	ret = pvrdma_query_port(ibdev, port, &attr);
+	if (ret)
+		goto err_out;
+
+	vdev->port_cap_mask |= props->set_port_cap_mask;
+	vdev->port_cap_mask &= ~props->clr_port_cap_mask;
+
+	if (mask & IB_PORT_SHUTDOWN)
+		vdev->ib_active = false;
+
+	if (mask & (IB_PORT_INIT_TYPE | IB_PORT_RESET_QKEY_CNTR)) {
+		ret = -EOPNOTSUPP;
+		goto err_out;
+	}
+
+	ret = 0;
+
+err_out:
+	mutex_unlock(&vdev->port_mutex);
+	return ret;
+}
+
+/**
+ * pvrdma_alloc_ucontext - allocate ucontext
+ * @ibdev: the IB device
+ * @udata: user data
+ *
+ * @return: the ib_ucontext pointer on success, otherwise errno.
+ */
+struct ib_ucontext *pvrdma_alloc_ucontext(struct ib_device *ibdev,
+					  struct ib_udata *udata)
+{
+	struct pvrdma_dev *vdev = to_vdev(ibdev);
+	struct pvrdma_ucontext *context;
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_create_uc *cmd = &req.create_uc;
+	struct pvrdma_cmd_create_uc_resp *resp = &rsp.create_uc_resp;
+	struct pvrdma_alloc_ucontext_resp uresp;
+	int ret;
+
+	if (!vdev->ib_active)
+		return ERR_PTR(-EAGAIN);
+
+	context = kmalloc(sizeof(*context), GFP_KERNEL);
+	if (!context)
+		return ERR_PTR(-ENOMEM);
+
+	context->dev = vdev;
+	ret = pvrdma_uar_alloc(vdev, &context->uar);
+	if (ret) {
+		kfree(context);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	/* get ctx_handle from host */
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->pfn = context->uar.pfn;
+	cmd->hdr.cmd = PVRDMA_CMD_CREATE_UC;
+	ret = pvrdma_cmd_post(vdev, &req, true, &rsp);
+	if (ret < 0 || rsp.hdr.ack != PVRDMA_CMD_CREATE_UC_RESP) {
+		struct pvrdma_dev *dev = to_vdev(ibdev);
+
+		dev_warn(&dev->pdev->dev, "could not create ucontext\n");
+		pvrdma_uar_free(vdev, &context->uar);
+		kfree(context);
+		return ERR_PTR(-EFAULT);
+	}
+
+	context->ctx_handle = resp->ctx_handle;
+
+	/* copy back to user */
+	uresp.qp_tab_size = vdev->dsr->caps.max_qp;
+	ret = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+	if (ret) {
+		pvrdma_uar_free(vdev, &context->uar);
+		context->ibucontext.device = ibdev;
+		pvrdma_dealloc_ucontext(&context->ibucontext);
+		return ERR_PTR(-EFAULT);
+	}
+
+	return &context->ibucontext;
+}
+
+/**
+ * pvrdma_dealloc_ucontext - deallocate ucontext
+ * @ibcontext: the ucontext
+ *
+ * @return: 0 on success, otherwise errno.
+ */
+int pvrdma_dealloc_ucontext(struct ib_ucontext *ibcontext)
+{
+	struct pvrdma_ucontext *context = to_vucontext(ibcontext);
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_destroy_uc *cmd = &req.destroy_uc;
+	int ret;
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_DESTROY_UC;
+	cmd->ctx_handle = context->ctx_handle;
+
+	ret = pvrdma_cmd_post(context->dev, &req, false, &rsp);
+	if (ret < 0) {
+		dev_warn(&context->dev->pdev->dev,
+			 "destroy ucontext failed\n");
+		/* Just do it */
+	}
+	pvrdma_uar_free(to_vdev(ibcontext->device), &context->uar);
+	kfree(context);
+
+	return ret;
+}
+
+/**
+ * pvrdma_mmap - create mmap region
+ * @ibcontext: the user context
+ * @vma: the VMA
+ *
+ * @return: 0 on success, otherwise errno.
+ */
+int pvrdma_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vma)
+{
+	struct pvrdma_ucontext *context = to_vucontext(ibcontext);
+	unsigned long start = vma->vm_start;
+	unsigned long size = vma->vm_end - vma->vm_start;
+	unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
+
+	dev_dbg(&context->dev->pdev->dev, "create mmap region\n");
+
+	if ((size != PAGE_SIZE) || (offset & ~PAGE_MASK)) {
+		dev_warn(&context->dev->pdev->dev,
+			 "invalid params for mmap region\n");
+		return -EINVAL;
+	}
+
+	/* Map UAR to kernel space, VM_LOCKED? */
+	vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
+	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+	if (io_remap_pfn_range(vma, start, context->uar.pfn, size,
+			       vma->vm_page_prot))
+		return -EAGAIN;
+
+	return 0;
+}
+
+/**
+ * pvrdma_alloc_pd - allocate protection domain
+ * @ibdev: the IB device
+ * @context: user context
+ * @udata: user data
+ *
+ * @return: the ib_pd protection domain pointer on success, otherwise errno.
+ */
+struct ib_pd *pvrdma_alloc_pd(struct ib_device *ibdev,
+			      struct ib_ucontext *context,
+			      struct ib_udata *udata)
+{
+	struct pvrdma_pd *pd;
+	struct pvrdma_dev *dev = to_vdev(ibdev);
+	union pvrdma_cmd_req req;
+	union pvrdma_cmd_resp rsp;
+	struct pvrdma_cmd_create_pd *cmd = &req.create_pd;
+	struct pvrdma_cmd_create_pd_resp *resp = &rsp.create_pd_resp;
+	int ret;
+
+	/* Check allowed max pds */
+	if (!atomic_add_unless(&dev->num_pds, 1, dev->dsr->caps.max_pd))
+		return ERR_PTR(-EINVAL);
+
+	pd = kmalloc(sizeof(*pd), GFP_KERNEL);
+	if (!pd) {
+		atomic_dec(&dev->num_pds);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	pd->priviledged = !context;
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_CREATE_PD;
+	cmd->ctx_handle = (context) ? to_vucontext(context)->ctx_handle : 0;
+	ret = pvrdma_cmd_post(dev, &req, true, &rsp);
+	if (ret < 0 || resp->hdr.ack != PVRDMA_CMD_CREATE_PD_RESP) {
+		dev_warn(&dev->pdev->dev,
+			 "failed to allocate protection domain\n");
+		kfree(pd);
+		atomic_dec(&dev->num_pds);
+		return ERR_PTR(-EFAULT);
+	}
+
+	pd->pd_handle = resp->pd_handle;
+	pd->pdn = resp->pd_handle;
+
+	if (context) {
+		if (ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) {
+			dev_warn(&dev->pdev->dev,
+				 "failed to copy back protection domain\n");
+			pvrdma_dealloc_pd(&pd->ibpd);
+			atomic_dec(&dev->num_pds);
+			return ERR_PTR(-EFAULT);
+		}
+	}
+
+	/* u32 pd handle */
+	return  &pd->ibpd;
+}
+
+/**
+ * pvrdma_dealloc_pd - deallocate protection domain
+ * @pd: the protection domain to be released
+ *
+ * @return: 0 on success, otherwise errno.
+ */
+int pvrdma_dealloc_pd(struct ib_pd *pd)
+{
+	struct pvrdma_dev *dev = to_vdev(pd->device);
+	union pvrdma_cmd_req req;
+	struct pvrdma_cmd_destroy_pd *cmd = &req.destroy_pd;
+	int ret;
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->hdr.cmd = PVRDMA_CMD_DESTROY_PD;
+	cmd->pd_handle = to_vpd(pd)->pd_handle;
+
+	ret = pvrdma_cmd_post(dev, &req, false, NULL);
+	if (ret)
+		dev_warn(&dev->pdev->dev,
+			 "could not dealloc protection domain\n");
+
+	kfree(to_vpd(pd));
+	atomic_dec(&dev->num_pds);
+
+	return 0;
+}
+
+/**
+ * pvrdma_create_ah - create an address handle
+ * @pd: the protection domain
+ * @ah_attr: the attributes of the AH
+ *
+ * @return: the ib_ah pointer on success, oterwise errno.
+ */
+struct ib_ah *pvrdma_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr)
+{
+	struct pvrdma_dev *dev = to_vdev(pd->device);
+	struct pvrdma_ah *ah;
+	enum rdma_link_layer ll;
+
+	if (!(ah_attr->ah_flags & IB_AH_GRH))
+		return ERR_PTR(-EINVAL);
+
+	ll = rdma_port_get_link_layer(pd->device, ah_attr->port_num);
+
+	if (ll != IB_LINK_LAYER_ETHERNET ||
+	    rdma_is_multicast_addr((struct in6_addr *)ah_attr->grh.dgid.raw))
+		return ERR_PTR(-EINVAL);
+
+	if (!atomic_add_unless(&dev->num_ahs, 1, dev->dsr->caps.max_ah))
+		return ERR_PTR(-EINVAL);
+
+	ah = kzalloc(sizeof(*ah), GFP_KERNEL);
+	if (!ah) {
+		atomic_dec(&dev->num_ahs);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	ah->av.port_pd = to_vpd(pd)->pd_handle | (ah_attr->port_num << 24);
+	ah->av.src_path_bits = ah_attr->src_path_bits;
+	ah->av.src_path_bits |= 0x80;
+	ah->av.gid_index = ah_attr->grh.sgid_index;
+	ah->av.hop_limit = ah_attr->grh.hop_limit;
+	ah->av.sl_tclass_flowlabel = (ah_attr->grh.traffic_class << 20) |
+				      ah_attr->grh.flow_label;
+	memcpy(ah->av.dgid, ah_attr->grh.dgid.raw, 16);
+	memcpy(ah->av.dmac, ah_attr->dmac, 6);
+
+	ah->ibah.device = pd->device;
+	ah->ibah.pd = pd;
+	ah->ibah.uobject = NULL;
+
+	return &ah->ibah;
+}
+
+/**
+ * pvrdma_destroy_ah - destroy an address handle
+ * @ah: the address handle to destroyed
+ *
+ * @return: 0 on success.
+ */
+int pvrdma_destroy_ah(struct ib_ah *ah)
+{
+	struct pvrdma_dev *dev = to_vdev(ah->device);
+
+	kfree(to_vah(ah));
+	atomic_dec(&dev->num_ahs);
+
+	return 0;
+}
diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_verbs.h b/drivers/infiniband/hw/pvrdma/pvrdma_verbs.h
new file mode 100644
index 0000000..e52fbe5
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/pvrdma_verbs.h
@@ -0,0 +1,108 @@
+/*
+ * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of EITHER the GNU General Public License
+ * version 2 as published by the Free Software Foundation or the BSD
+ * 2-Clause License. This program is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License version 2 for more details at
+ * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program available in the file COPYING in the main
+ * directory of this source tree.
+ *
+ * The BSD 2-Clause License
+ *
+ *     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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef __PVRDMA_VERBS_H__
+#define __PVRDMA_VERBS_H__
+
+int pvrdma_query_device(struct ib_device *ibdev,
+			struct ib_device_attr *props,
+			struct ib_udata *udata);
+int pvrdma_query_port(struct ib_device *ibdev, u8 port,
+		      struct ib_port_attr *props);
+int pvrdma_query_gid(struct ib_device *ibdev, u8 port,
+		     int index, union ib_gid *gid);
+int pvrdma_query_pkey(struct ib_device *ibdev, u8 port,
+		      u16 index, u16 *pkey);
+enum rdma_link_layer pvrdma_port_link_layer(struct ib_device *ibdev,
+					    u8 port);
+int pvrdma_modify_device(struct ib_device *ibdev, int mask,
+			 struct ib_device_modify *props);
+int pvrdma_modify_port(struct ib_device *ibdev, u8 port,
+		       int mask, struct ib_port_modify *props);
+int pvrdma_mmap(struct ib_ucontext *context, struct vm_area_struct *vma);
+struct ib_ucontext *pvrdma_alloc_ucontext(struct ib_device *ibdev,
+					  struct ib_udata *udata);
+int pvrdma_dealloc_ucontext(struct ib_ucontext *context);
+struct ib_pd *pvrdma_alloc_pd(struct ib_device *ibdev,
+			      struct ib_ucontext *context,
+			      struct ib_udata *udata);
+int pvrdma_dealloc_pd(struct ib_pd *ibpd);
+struct ib_mr *pvrdma_get_dma_mr(struct ib_pd *pd, int acc);
+struct ib_mr *pvrdma_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+				 u64 virt_addr, int access_flags,
+				 struct ib_udata *udata);
+int pvrdma_dereg_mr(struct ib_mr *mr);
+struct ib_mr *pvrdma_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
+			      u32 max_num_sg);
+int pvrdma_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
+		     int sg_nents, unsigned int *sg_offset);
+int pvrdma_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
+int pvrdma_resize_cq(struct ib_cq *ibcq, int entries,
+		     struct ib_udata *udata);
+struct ib_cq *pvrdma_create_cq(struct ib_device *ibdev,
+			       const struct ib_cq_init_attr *attr,
+			       struct ib_ucontext *context,
+			       struct ib_udata *udata);
+int pvrdma_resize_cq(struct ib_cq *ibcq, int entries,
+		     struct ib_udata *udata);
+int pvrdma_destroy_cq(struct ib_cq *cq);
+int pvrdma_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
+int pvrdma_req_notify_cq(struct ib_cq *cq, enum ib_cq_notify_flags flags);
+struct ib_ah *pvrdma_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
+int pvrdma_destroy_ah(struct ib_ah *ah);
+struct ib_qp *pvrdma_create_qp(struct ib_pd *pd,
+			       struct ib_qp_init_attr *init_attr,
+			       struct ib_udata *udata);
+int pvrdma_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		     int attr_mask, struct ib_udata *udata);
+int pvrdma_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
+		    int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
+int pvrdma_destroy_qp(struct ib_qp *qp);
+int pvrdma_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+		     struct ib_send_wr **bad_wr);
+int pvrdma_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+		     struct ib_recv_wr **bad_wr);
+
+#endif /* __PVRDMA_VERBS_H__ */
-- 
2.7.4

--
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 13/15] IB/pvrdma: Add Kconfig and Makefile

[Adit Ranadive]

This patch adds a Kconfig and Makefile for the PVRDMA driver.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/pvrdma/Kconfig  | 8 ++++++++
 drivers/infiniband/hw/pvrdma/Makefile | 3 +++
 2 files changed, 11 insertions(+)
 create mode 100644 drivers/infiniband/hw/pvrdma/Kconfig
 create mode 100644 drivers/infiniband/hw/pvrdma/Makefile

diff --git a/drivers/infiniband/hw/pvrdma/Kconfig b/drivers/infiniband/hw/pvrdma/Kconfig
new file mode 100644
index 0000000..cefa441
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/Kconfig
@@ -0,0 +1,8 @@
+config INFINIBAND_PVRDMA
+	tristate "VMware Paravirtualized RDMA Driver"
+	depends on NETDEVICES && ETHERNET && PCI && INET
+	select VMXNET3
+	---help---
+	  This driver provides low-level support for VMware Paravirtual
+	  RDMA adapter. It interacts with the VMXNet3 driver to provide
+	  Ethernet capabilities.
diff --git a/drivers/infiniband/hw/pvrdma/Makefile b/drivers/infiniband/hw/pvrdma/Makefile
new file mode 100644
index 0000000..e6f078b
--- /dev/null
+++ b/drivers/infiniband/hw/pvrdma/Makefile
@@ -0,0 +1,3 @@
+obj-$(CONFIG_INFINIBAND_PVRDMA) += pvrdma.o
+
+pvrdma-y := pvrdma_cmd.o pvrdma_cq.o pvrdma_doorbell.o pvrdma_main.o pvrdma_misc.o pvrdma_mr.o pvrdma_qp.o pvrdma_verbs.o
-- 
2.7.4

--
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 14/15] IB: Add PVRDMA driver

[Adit Ranadive]

This patch updates the InfiniBand subsystem to build the PVRDMA driver.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/Kconfig     | 1 +
 drivers/infiniband/hw/Makefile | 1 +
 2 files changed, 2 insertions(+)

diff --git a/drivers/infiniband/Kconfig b/drivers/infiniband/Kconfig
index 2137adf..d677d8f 100644
--- a/drivers/infiniband/Kconfig
+++ b/drivers/infiniband/Kconfig
@@ -86,5 +86,6 @@ source "drivers/infiniband/ulp/isert/Kconfig"
 source "drivers/infiniband/sw/rdmavt/Kconfig"
 
 source "drivers/infiniband/hw/hfi1/Kconfig"
+source "drivers/infiniband/hw/pvrdma/Kconfig"
 
 endif # INFINIBAND
diff --git a/drivers/infiniband/hw/Makefile b/drivers/infiniband/hw/Makefile
index c0c7cf8..eb24449 100644
--- a/drivers/infiniband/hw/Makefile
+++ b/drivers/infiniband/hw/Makefile
@@ -9,3 +9,4 @@ obj-$(CONFIG_INFINIBAND_NES)		+= nes/
 obj-$(CONFIG_INFINIBAND_OCRDMA)		+= ocrdma/
 obj-$(CONFIG_INFINIBAND_USNIC)		+= usnic/
 obj-$(CONFIG_INFINIBAND_HFI1)		+= hfi1/
+obj-$(CONFIG_INFINIBAND_PVRDMA)		+= pvrdma/
-- 
2.7.4

--
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 15/15] MAINTAINERS: Update for PVRDMA driver

[Adit Ranadive]

Add maintainer info for the PVRDMA driver.

Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
---
 MAINTAINERS | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/MAINTAINERS b/MAINTAINERS
index 1209323..abf62e7 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -12302,6 +12302,13 @@ S:	Maintained
 F:	drivers/scsi/vmw_pvscsi.c
 F:	drivers/scsi/vmw_pvscsi.h
 
+VMWARE PVRDMA DRIVER
+M:	Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
+M:	VMware PV-Drivers <pv-drivers-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
+L:	linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
+S:	Maintained
+F:	drivers/infiniband/hw/pvrdma
+
 VOLTAGE AND CURRENT REGULATOR FRAMEWORK
 M:	Liam Girdwood <lgirdwood-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
 M:	Mark Brown <broonie-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
-- 
2.7.4

--
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

Re: [PATCH v2 00/15] Add Paravirtual RDMA Driver

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 245 bytes --]

On Tue, Jul 12, 2016 at 12:36:30PM -0700, Adit Ranadive wrote:
> Changes v1 -> v2: 
>  Patch [07/15] - Addressed Yuval Shaia's comments and 32-bit build errors.

Do I understand correctly that you started from v1 and not from v0?

Thanks

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 00/15] Add Paravirtual RDMA Driver

[Adit Ranadive]

On 7/12/16, 12:50 PM, "Leon Romanovsky" <leon@kernel.org> wrote:

>On Tue, Jul 12, 2016 at 12:36:30PM -0700, Adit Ranadive wrote:
>> Changes v1 -> v2: 
>>  Patch [07/15] - Addressed Yuval Shaia's comments and 32-bit build errors.
>
>Do I understand correctly that you started from v1 and not from v0?
>
>Thanks

Yes, sorry about that.

Thanks,
Adit

Re: [PATCH v2 01/15] IB/pvrdma: Add paravirtual rdma device

[Yuval Shaia]

On Tue, Jul 12, 2016 at 12:36:31PM -0700, Adit Ranadive wrote:
> This patch adds the main device-level structures and functions to be used
> to provide RDMA functionality. Also, we define conversion functions from
> the IB core stack structures to the device-specific ones.
> 
> Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> ---
>  drivers/infiniband/hw/pvrdma/pvrdma.h | 459 ++++++++++++++++++++++++++++++++++
>  1 file changed, 459 insertions(+)
>  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma.h
> 
> diff --git a/drivers/infiniband/hw/pvrdma/pvrdma.h b/drivers/infiniband/hw/pvrdma/pvrdma.h
> new file mode 100644
> index 0000000..0dc5bd2
> --- /dev/null
> +++ b/drivers/infiniband/hw/pvrdma/pvrdma.h
> @@ -0,0 +1,459 @@
> +/*
> + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of EITHER the GNU General Public License
> + * version 2 as published by the Free Software Foundation or the BSD
> + * 2-Clause License. This program is distributed in the hope that it
> + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> + * See the GNU General Public License version 2 for more details at
> + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program available in the file COPYING in the main
> + * directory of this source tree.
> + *
> + * The BSD 2-Clause License
> + *
> + *     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.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> + * OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#ifndef __PVRDMA_H__
> +#define __PVRDMA_H__
> +
> +#include <linux/compiler.h>
> +#include <linux/interrupt.h>
> +#include <linux/list.h>
> +#include <linux/mutex.h>
> +#include <linux/pci.h>
> +#include <linux/semaphore.h>
> +#include <rdma/ib_umem.h>
> +#include <rdma/ib_verbs.h>
> +
> +#include "pvrdma_defs.h"
> +#include "pvrdma_dev_api.h"
> +#include "pvrdma_verbs.h"
> +
> +/* NOT the same as BIT_MASK(). */
> +#define PVRDMA_MASK(n) ((n << 1) - 1)
> +
> +/*
> + * VMware PVRDMA vendor id and PCI device id.
> + */
> +#ifndef PCI_VENDOR_ID_VMWARE
> +#define PCI_VENDOR_ID_VMWARE		0x15AD
> +#endif
> +#define PCI_DEVICE_ID_VMWARE_PVRDMA	0x0820
> +
> +struct pvrdma_dev;
> +
> +struct pvrdma_page_dir {
> +	dma_addr_t dir_dma;
> +	u64 *dir;
> +	int ntables;
> +	u64 **tables;
> +	u64 npages;
> +	void **pages;
> +};
> +
> +struct pvrdma_cq {
> +	struct ib_cq ibcq;
> +	int offset;
> +	spinlock_t cq_lock; /* Poll lock. */
> +	struct pvrdma_uar_map *uar;
> +	struct ib_umem *umem;
> +	struct pvrdma_ring_state *ring_state;
> +	struct pvrdma_page_dir pdir;
> +	u32 cq_handle;
> +	bool is_kernel;
> +	atomic_t refcnt;
> +	wait_queue_head_t wait;
> +};
> +
> +struct pvrdma_id_table {
> +	u32 last;
> +	u32 top;
> +	u32 max;
> +	u32 mask;
> +	spinlock_t lock; /* Table lock. */
> +	unsigned long *table;
> +};
> +
> +struct pvrdma_uar_map {
> +	unsigned long pfn;
> +	void __iomem *map;
> +	int index;
> +};
> +
> +struct pvrdma_uar_table {
> +	struct pvrdma_id_table tbl;
> +	int size;
> +};
> +
> +struct pvrdma_ucontext {
> +	struct ib_ucontext ibucontext;
> +	struct pvrdma_dev *dev;
> +	struct pvrdma_uar_map uar;
> +	u64 ctx_handle;
> +};
> +
> +struct pvrdma_pd {
> +	struct ib_pd ibpd;
> +	u32 pdn;
> +	u32 pd_handle;
> +	int priviledged;
> +};
> +
> +struct pvrdma_mr {
> +	u32 mr_handle;
> +	u64 iova;
> +	u64 size;
> +};
> +
> +struct pvrdma_user_mr {
> +	struct ib_mr ibmr;
> +	struct ib_umem *umem;
> +	struct pvrdma_mr mmr;
> +	struct pvrdma_page_dir pdir;
> +	u64 *pages;
> +	u32 npages;
> +	u32 max_pages;
> +	u32 page_shift;
> +};
> +
> +struct pvrdma_wq {
> +	struct pvrdma_ring *ring;
> +	spinlock_t lock; /* Work queue lock. */
> +	int wqe_cnt;
> +	int wqe_size;
> +	int max_gs;
> +	int offset;
> +};
> +
> +struct pvrdma_ah {
> +	struct ib_ah ibah;
> +	struct pvrdma_av av;
> +};
> +
> +struct pvrdma_qp {
> +	struct ib_qp ibqp;
> +	u32 qp_handle;
> +	u32 qkey;
> +	struct pvrdma_wq sq;
> +	struct pvrdma_wq rq;
> +	struct ib_umem *rumem;
> +	struct ib_umem *sumem;
> +	struct pvrdma_page_dir pdir;
> +	int npages;
> +	int npages_send;
> +	int npages_recv;
> +	u32 flags;
> +	u8 port;
> +	u8 state;
> +	bool is_kernel;
> +	struct mutex mutex; /* QP state mutex. */
> +	atomic_t refcnt;
> +	wait_queue_head_t wait;
> +};
> +
> +struct pvrdma_dev {
> +	/* PCI device-related information. */
> +	struct ib_device ib_dev;
> +	struct pci_dev *pdev;
> +	void __iomem *regs;
> +	struct pvrdma_device_shared_region *dsr; /* Shared region pointer */
> +	dma_addr_t dsrbase; /* Shared region base address */
> +	void *cmd_slot;
> +	void *resp_slot;
> +	unsigned long flags;
> +	struct list_head device_link;
> +
> +	/* Locking and interrupt information. */
> +	spinlock_t cmd_lock; /* Command lock. */
> +	struct semaphore cmd_sema;
> +	struct completion cmd_done;
> +	struct {
> +		enum pvrdma_intr_type type; /* Intr type */
> +		struct msix_entry msix_entry[PVRDMA_MAX_INTERRUPTS];
> +		irq_handler_t handler[PVRDMA_MAX_INTERRUPTS];
> +		u8 enabled[PVRDMA_MAX_INTERRUPTS];
> +		u8 size;
> +	} intr;
> +
> +	/* RDMA-related device information. */
> +	union ib_gid *sgid_tbl;
> +	struct pvrdma_ring_state *async_ring_state;
> +	struct pvrdma_page_dir async_pdir;
> +	struct pvrdma_ring_state *cq_ring_state;
> +	struct pvrdma_page_dir cq_pdir;
> +	struct pvrdma_cq **cq_tbl;
> +	spinlock_t cq_tbl_lock;
> +	struct pvrdma_qp **qp_tbl;
> +	spinlock_t qp_tbl_lock;
> +	struct pvrdma_uar_table uar_table;
> +	struct pvrdma_uar_map driver_uar;
> +	__be64 sys_image_guid;
> +	spinlock_t desc_lock; /* Device modification lock. */
> +	u32 port_cap_mask;
> +	struct mutex port_mutex; /* Port modification mutex. */
> +	bool ib_active;
> +	atomic_t num_qps;
> +	atomic_t num_cqs;
> +	atomic_t num_pds;
> +	atomic_t num_ahs;
> +
> +	/* Network device information. */
> +	struct net_device *netdev;
> +	struct notifier_block nb_netdev;
> +};
> +
> +static inline struct pvrdma_dev *to_vdev(struct ib_device *ibdev)
> +{
> +	return container_of(ibdev, struct pvrdma_dev, ib_dev);
> +}
> +
> +static inline struct
> +pvrdma_ucontext *to_vucontext(struct ib_ucontext *ibucontext)
> +{
> +	return container_of(ibucontext, struct pvrdma_ucontext, ibucontext);
> +}
> +
> +static inline struct pvrdma_pd *to_vpd(struct ib_pd *ibpd)
> +{
> +	return container_of(ibpd, struct pvrdma_pd, ibpd);
> +}
> +
> +static inline struct pvrdma_cq *to_vcq(struct ib_cq *ibcq)
> +{
> +	return container_of(ibcq, struct pvrdma_cq, ibcq);
> +}
> +
> +static inline struct pvrdma_user_mr *to_vmr(struct ib_mr *ibmr)
> +{
> +	return container_of(ibmr, struct pvrdma_user_mr, ibmr);
> +}
> +
> +static inline struct pvrdma_qp *to_vqp(struct ib_qp *ibqp)
> +{
> +	return container_of(ibqp, struct pvrdma_qp, ibqp);
> +}
> +
> +static inline struct pvrdma_ah *to_vah(struct ib_ah *ibah)
> +{
> +	return container_of(ibah, struct pvrdma_ah, ibah);
> +}
> +
> +static inline void pvrdma_write_reg(struct pvrdma_dev *dev, u32 reg, u32 val)
> +{
> +	writel(cpu_to_le32(val), dev->regs + reg);
> +}
> +
> +static inline u32 pvrdma_read_reg(struct pvrdma_dev *dev, u32 reg)
> +{
> +	return le32_to_cpu(readl(dev->regs + reg));
> +}
> +
> +static inline void *pvrdma_page_dir_get_ptr(struct pvrdma_page_dir *pdir,
> +					    u64 offset)
> +{
> +	return pdir->pages[offset / PAGE_SIZE] + (offset % PAGE_SIZE);
> +}
> +
> +static inline enum pvrdma_mtu ib_mtu_to_pvrdma(enum ib_mtu mtu)
> +{
> +	return (enum pvrdma_mtu)mtu;
> +}
> +
> +static inline enum ib_mtu pvrdma_mtu_to_ib(enum pvrdma_mtu mtu)
> +{
> +	return (enum ib_mtu)mtu;
> +}
> +
> +static inline enum pvrdma_port_state ib_port_state_to_pvrdma(
> +					enum ib_port_state state)
> +{
> +	return (enum pvrdma_port_state)state;
> +}
> +
> +static inline enum ib_port_state pvrdma_port_state_to_ib(
> +					enum pvrdma_port_state state)
> +{
> +	return (enum ib_port_state)state;
> +}
> +
> +static inline int ib_port_cap_flags_to_pvrdma(int flags)
> +{
> +	return flags & PVRDMA_MASK(PVRDMA_PORT_CAP_FLAGS_MAX);
> +}
> +
> +static inline int pvrdma_port_cap_flags_to_ib(int flags)
> +{
> +	return flags;
> +}
> +
> +static inline enum pvrdma_port_width ib_port_width_to_pvrdma(
> +					enum ib_port_width width)
> +{
> +	return (enum pvrdma_port_width)width;
> +}
> +
> +static inline enum ib_port_width pvrdma_port_width_to_ib(
> +					enum pvrdma_port_width width)
> +{
> +	return (enum ib_port_width)width;
> +}
> +
> +static inline enum pvrdma_port_speed ib_port_speed_to_pvrdma(
> +					enum ib_port_speed speed)
> +{
> +	return (enum pvrdma_port_speed)speed;
> +}
> +
> +static inline enum ib_port_speed pvrdma_port_speed_to_ib(
> +					enum pvrdma_port_speed speed)
> +{
> +	return (enum ib_port_speed)speed;
> +}
> +
> +static inline int pvrdma_qp_attr_mask_to_ib(int attr_mask)
> +{
> +	return attr_mask;
> +}
> +
> +static inline int ib_qp_attr_mask_to_pvrdma(int attr_mask)
> +{
> +	return attr_mask & PVRDMA_MASK(PVRDMA_QP_ATTR_MASK_MAX);
> +}
> +
> +static inline enum pvrdma_mig_state ib_mig_state_to_pvrdma(
> +					enum ib_mig_state state)
> +{
> +	return (enum pvrdma_mig_state)state;
> +}
> +
> +static inline enum ib_mig_state pvrdma_mig_state_to_ib(
> +					enum pvrdma_mig_state state)
> +{
> +	return (enum ib_mig_state)state;
> +}
> +
> +static inline int ib_access_flags_to_pvrdma(int flags)
> +{
> +	return flags;
> +}
> +
> +static inline int pvrdma_access_flags_to_ib(int flags)
> +{
> +	return flags & PVRDMA_MASK(PVRDMA_ACCESS_FLAGS_MAX);
> +}
> +
> +static inline enum pvrdma_qp_type ib_qp_type_to_pvrdma(enum ib_qp_type type)
> +{
> +	return (enum pvrdma_qp_type)type;
> +}
> +
> +static inline enum ib_qp_type pvrdma_qp_type_to_ib(enum pvrdma_qp_type type)
> +{
> +	return (enum ib_qp_type)type;
> +}
> +
> +static inline enum pvrdma_qp_state ib_qp_state_to_pvrdma(enum ib_qp_state state)
> +{
> +	return (enum pvrdma_qp_state)state;
> +}
> +
> +static inline enum ib_qp_state pvrdma_qp_state_to_ib(enum pvrdma_qp_state state)
> +{
> +	return (enum ib_qp_state)state;
> +}
> +
> +static inline enum pvrdma_wr_opcode ib_wr_opcode_to_pvrdma(enum ib_wr_opcode op)
> +{
> +	return (enum pvrdma_wr_opcode)op;
> +}
> +
> +static inline enum ib_wc_status pvrdma_wc_status_to_ib(
> +					enum pvrdma_wc_status status)
> +{
> +	return (enum ib_wc_status)status;
> +}
> +
> +static inline int pvrdma_wc_opcode_to_ib(int opcode)
> +{
> +	return opcode;
> +}
> +
> +static inline int pvrdma_wc_flags_to_ib(int flags)
> +{
> +	return flags;
> +}
> +
> +static inline int ib_send_flags_to_pvrdma(int flags)
> +{
> +	return flags & PVRDMA_MASK(PVRDMA_SEND_FLAGS_MAX);
> +}
> +
> +void pvrdma_qp_cap_to_ib(const struct pvrdma_qp_cap *src,
> +			 struct ib_qp_cap *dst);
> +void ib_qp_cap_to_pvrdma(const struct ib_qp_cap *src,
> +			 struct pvrdma_qp_cap *dst);
> +void pvrdma_gid_to_ib(const union pvrdma_gid *src, union ib_gid *dst);
> +void ib_gid_to_pvrdma(const union ib_gid *src, union pvrdma_gid *dst);
> +void pvrdma_global_route_to_ib(const struct pvrdma_global_route *src,
> +			       struct ib_global_route *dst);
> +void ib_global_route_to_pvrdma(const struct ib_global_route *src,
> +			       struct pvrdma_global_route *dst);
> +void pvrdma_ah_attr_to_ib(const struct pvrdma_ah_attr *src,
> +			  struct ib_ah_attr *dst);
> +void ib_ah_attr_to_pvrdma(const struct ib_ah_attr *src,
> +			  struct pvrdma_ah_attr *dst);
> +
> +int pvrdma_uar_table_init(struct pvrdma_dev *dev);
> +void pvrdma_uar_table_cleanup(struct pvrdma_dev *dev);
> +
> +int pvrdma_uar_alloc(struct pvrdma_dev *dev, struct pvrdma_uar_map *uar);
> +void pvrdma_uar_free(struct pvrdma_dev *dev, struct pvrdma_uar_map *uar);
> +
> +void pvrdma_flush_cqe(struct pvrdma_qp *qp, struct pvrdma_cq *cq);
> +
> +int pvrdma_page_dir_init(struct pvrdma_dev *dev, struct pvrdma_page_dir *pdir,
> +			 u64 npages, bool alloc_pages);
> +void pvrdma_page_dir_cleanup(struct pvrdma_dev *dev,
> +			     struct pvrdma_page_dir *pdir);
> +int pvrdma_page_dir_insert_dma(struct pvrdma_page_dir *pdir, u64 idx,
> +			       dma_addr_t daddr);
> +int pvrdma_page_dir_insert_umem(struct pvrdma_page_dir *pdir,
> +				struct ib_umem *umem, u64 offset);
> +dma_addr_t pvrdma_page_dir_get_dma(struct pvrdma_page_dir *pdir, u64 idx);
> +int pvrdma_page_dir_insert_page_list(struct pvrdma_page_dir *pdir,
> +				     u64 *page_list, int num_pages);
> +
> +int pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
> +		    bool expect, union pvrdma_cmd_resp *rsp);

Patch #1 uses unions pvrdma_cmd_req & pvrdma_cmd_resp that added in patch
#4.
Suggesting to change the order.

> +
> +#endif /* __PVRDMA_H__ */
> -- 
> 2.7.4
> 
> --
> 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
--
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 02/15] IB/pvrdma: Add device command support

[Yuval Shaia]

On Tue, Jul 12, 2016 at 12:36:32PM -0700, Adit Ranadive wrote:
> This patch enables posting Verb requests and receiving responses to/from
> the backend PVRDMA emulation layer.
> 
> Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> ---
>  drivers/infiniband/hw/pvrdma/pvrdma_cmd.c | 104 ++++++++++++++++++++++++++++++
>  1 file changed, 104 insertions(+)
>  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_cmd.c
> 
> diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_cmd.c b/drivers/infiniband/hw/pvrdma/pvrdma_cmd.c
> new file mode 100644
> index 0000000..c22debf
> --- /dev/null
> +++ b/drivers/infiniband/hw/pvrdma/pvrdma_cmd.c
> @@ -0,0 +1,104 @@
> +/*
> + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of EITHER the GNU General Public License
> + * version 2 as published by the Free Software Foundation or the BSD
> + * 2-Clause License. This program is distributed in the hope that it
> + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> + * See the GNU General Public License version 2 for more details at
> + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program available in the file COPYING in the main
> + * directory of this source tree.
> + *
> + * The BSD 2-Clause License
> + *
> + *     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.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> + * OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#include <linux/list.h>
> +
> +#include "pvrdma.h"
> +
> +#define PVRDMA_CMD_TIMEOUT	10000 /* ms */
> +
> +static int pvrdma_cmd_recv(struct pvrdma_dev *dev, union pvrdma_cmd_resp *resp)
> +{
> +	dev_dbg(&dev->pdev->dev, "receive response from device\n");
> +
> +	spin_lock(&dev->cmd_lock);
> +	memcpy(resp, dev->resp_slot, sizeof(*resp));
> +	spin_unlock(&dev->cmd_lock);
> +
> +	return 0;
> +}
> +
> +int
> +pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
> +		bool expect_resp, union pvrdma_cmd_resp *resp)
> +{
> +	int err;
> +
> +	dev_dbg(&dev->pdev->dev, "post request to device\n");
> +
> +	/* Serializiation */
> +	down(&dev->cmd_sema);
> +
> +	spin_lock(&dev->cmd_lock);
> +	memcpy(dev->cmd_slot, req, sizeof(*req));

Just to protect against memory corruption suggesting to do:
	memcpy(dev->cmd_slot, req, min(PAGE_SIZE, sizeof(*req)));

> +	spin_unlock(&dev->cmd_lock);
> +
> +	init_completion(&dev->cmd_done);
> +	pvrdma_write_reg(dev, PVRDMA_REG_REQUEST, 0);
> +
> +	/* Make sure the request is written before reading status. */
> +	mb();
> +	err = pvrdma_read_reg(dev, PVRDMA_REG_ERR);
> +	if (err == 0) {
> +		if (expect_resp) {

How about utilizing resp instead of extra argument.
This will save you for example from the local variable rsp in function
pvrdma_destroy_qp (patch #10)

> +			err = wait_for_completion_interruptible_timeout(
> +				&dev->cmd_done,
> +				msecs_to_jiffies(PVRDMA_CMD_TIMEOUT));
> +			if (err == 0) {
> +				dev_warn(&dev->pdev->dev,
> +					 "completion timeout\n");

Is this really a warning only? i.e. i assume some high level flow is
break now.

> +				err = -ETIMEDOUT;
> +			} else {
> +				err = pvrdma_cmd_recv(dev, resp);
> +			}
> +		}
> +	} else {
> +		dev_warn(&dev->pdev->dev, "failed to write request %d\n", err);
> +	}
> +
> +	up(&dev->cmd_sema);
> +
> +	return err;
> +}
> -- 
> 2.7.4
> 
> --
> 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
--
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 02/15] IB/pvrdma: Add device command support

[Yuval Shaia]

On Tue, Jul 12, 2016 at 12:36:32PM -0700, Adit Ranadive wrote:
> This patch enables posting Verb requests and receiving responses to/from
> the backend PVRDMA emulation layer.
> 
> Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> ---
>  drivers/infiniband/hw/pvrdma/pvrdma_cmd.c | 104 ++++++++++++++++++++++++++++++
>  1 file changed, 104 insertions(+)
>  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_cmd.c
> 
> diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_cmd.c b/drivers/infiniband/hw/pvrdma/pvrdma_cmd.c
> new file mode 100644
> index 0000000..c22debf
> --- /dev/null
> +++ b/drivers/infiniband/hw/pvrdma/pvrdma_cmd.c
> @@ -0,0 +1,104 @@
> +/*
> + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of EITHER the GNU General Public License
> + * version 2 as published by the Free Software Foundation or the BSD
> + * 2-Clause License. This program is distributed in the hope that it
> + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> + * See the GNU General Public License version 2 for more details at
> + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program available in the file COPYING in the main
> + * directory of this source tree.
> + *
> + * The BSD 2-Clause License
> + *
> + *     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.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> + * OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#include <linux/list.h>
> +
> +#include "pvrdma.h"
> +
> +#define PVRDMA_CMD_TIMEOUT	10000 /* ms */
> +
> +static int pvrdma_cmd_recv(struct pvrdma_dev *dev, union pvrdma_cmd_resp *resp)
> +{
> +	dev_dbg(&dev->pdev->dev, "receive response from device\n");
> +
> +	spin_lock(&dev->cmd_lock);
> +	memcpy(resp, dev->resp_slot, sizeof(*resp));
> +	spin_unlock(&dev->cmd_lock);

In case your virtual HW supports more than one concurrent commands, if
cmd_lock protects cmd_slot i would expect to have extra resp_lock to
protect resp_slot.
btw, otherwise why not just lock the entire flow:
	memcpy(dev->cmd_slot, req, sizeof(*req));
	pvrdma_write_reg(dev, PVRDMA_REG_REQUEST, 0);
	pvrdma_read_reg(dev, PVRDMA_REG_ERR);
	memcpy(resp, dev->resp_slot, sizeof(*resp));

This might improve performances.

> +
> +	return 0;
> +}
> +
> +int
> +pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
> +		bool expect_resp, union pvrdma_cmd_resp *resp)
> +{
> +	int err;
> +
> +	dev_dbg(&dev->pdev->dev, "post request to device\n");
> +
> +	/* Serializiation */
> +	down(&dev->cmd_sema);
> +
> +	spin_lock(&dev->cmd_lock);
> +	memcpy(dev->cmd_slot, req, sizeof(*req));
> +	spin_unlock(&dev->cmd_lock);
> +
> +	init_completion(&dev->cmd_done);
> +	pvrdma_write_reg(dev, PVRDMA_REG_REQUEST, 0);
> +
> +	/* Make sure the request is written before reading status. */
> +	mb();
> +	err = pvrdma_read_reg(dev, PVRDMA_REG_ERR);
> +	if (err == 0) {
> +		if (expect_resp) {
> +			err = wait_for_completion_interruptible_timeout(
> +				&dev->cmd_done,
> +				msecs_to_jiffies(PVRDMA_CMD_TIMEOUT));
> +			if (err == 0) {
> +				dev_warn(&dev->pdev->dev,
> +					 "completion timeout\n");
> +				err = -ETIMEDOUT;
> +			} else {
> +				err = pvrdma_cmd_recv(dev, resp);

Maybe it is a silly question but in case of two commands that issued almost
simultaneously, it possible that fast execution of second command will
reach this point first and the reasponse here is of the first one?

> +			}
> +		}
> +	} else {
> +		dev_warn(&dev->pdev->dev, "failed to write request %d\n", err);
> +	}
> +
> +	up(&dev->cmd_sema);
> +
> +	return err;
> +}
> -- 
> 2.7.4
> 
> --
> 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
--
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 05/15] IB/pvrdma: Add UAR support

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 3969 bytes --]

On Tue, Jul 12, 2016 at 12:36:35PM -0700, Adit Ranadive wrote:
> This patch adds the UAR support for the paravirtual RDMA device. The UAR
> pages are MMIO pages from the virtual PCI space. We define offsets within
> this page to provide the fast data-path operations.
> 
> Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> ---
>  drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c | 128 +++++++++++++++++++++++++
>  1 file changed, 128 insertions(+)
>  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c
> 
> diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c b/drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c
> new file mode 100644
> index 0000000..d854f6d
> --- /dev/null
> +++ b/drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c
> @@ -0,0 +1,128 @@
> +/*
> + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of EITHER the GNU General Public License
> + * version 2 as published by the Free Software Foundation or the BSD
> + * 2-Clause License. This program is distributed in the hope that it
> + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> + * See the GNU General Public License version 2 for more details at
> + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program available in the file COPYING in the main
> + * directory of this source tree.
> + *
> + * The BSD 2-Clause License
> + *
> + *     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.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> + * OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#include <linux/bitmap.h>
> +#include <linux/errno.h>
> +#include <linux/slab.h>
> +
> +#include "pvrdma.h"
> +
> +int pvrdma_uar_table_init(struct pvrdma_dev *dev)
> +{
> +	u32 num = dev->dsr->caps.max_uar;
> +	u32 mask = num - 1;
> +	struct pvrdma_id_table *tbl = &dev->uar_table.tbl;
> +
> +	/* Power of 2 */
> +	if (num != 1 << (ffs(num) - 1))
> +		return -EINVAL;

Can functions/macros defined in include/linux/log2.h replace the code
above?

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 11/15] IB/pvrdma: Add user-level shared functions

[Yuval Shaia]

On Tue, Jul 12, 2016 at 12:36:41PM -0700, Adit Ranadive wrote:
> We share some common structures with the user-level driver. This patch
> adds those structures and shared functions to traverse the QP/CQ rings.
> 
> Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> ---
>  drivers/infiniband/hw/pvrdma/pvrdma_uapi.h | 247 +++++++++++++++++++++++++++++
>  drivers/infiniband/hw/pvrdma/pvrdma_user.h |  99 ++++++++++++
>  2 files changed, 346 insertions(+)
>  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_uapi.h
>  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_user.h
> 
> diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_uapi.h b/drivers/infiniband/hw/pvrdma/pvrdma_uapi.h
> new file mode 100644
> index 0000000..274328c
> --- /dev/null
> +++ b/drivers/infiniband/hw/pvrdma/pvrdma_uapi.h
> @@ -0,0 +1,247 @@
> +/*
> + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of EITHER the GNU General Public License
> + * version 2 as published by the Free Software Foundation or the BSD
> + * 2-Clause License. This program is distributed in the hope that it
> + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> + * See the GNU General Public License version 2 for more details at
> + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program available in the file COPYING in the main
> + * directory of this source tree.
> + *
> + * The BSD 2-Clause License
> + *
> + *     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.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> + * OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#ifndef __PVRDMA_UAPI_H__
> +#define __PVRDMA_UAPI_H__
> +
> +#include <linux/types.h>
> +
> +#define PVRDMA_VERSION 17
> +
> +#define PVRDMA_UAR_HANDLE_MASK	0x00FFFFFF	/* Bottom 24 bits. */
> +#define PVRDMA_UAR_QP_OFFSET	0		/* Offset of QP doorbell. */
> +#define PVRDMA_UAR_QP_SEND	BIT(30)		/* Send bit. */
> +#define PVRDMA_UAR_QP_RECV	BIT(31)		/* Recv bit. */
> +#define PVRDMA_UAR_CQ_OFFSET	4		/* Offset of CQ doorbell. */
> +#define PVRDMA_UAR_CQ_ARM_SOL	BIT(29)		/* Arm solicited bit. */
> +#define PVRDMA_UAR_CQ_ARM	BIT(30)		/* Arm bit. */
> +#define PVRDMA_UAR_CQ_POLL	BIT(31)		/* Poll bit. */
> +
> +/* PVRDMA atomic compare and swap */
> +struct pvrdma_exp_cmp_swap {
> +	__u64 swap_val;
> +	__u64 compare_val;
> +	__u64 swap_mask;
> +	__u64 compare_mask;
> +};
> +
> +/* PVRDMA atomic fetch and add */
> +struct pvrdma_exp_fetch_add {
> +	__u64 add_val;
> +	__u64 field_boundary;
> +};
> +
> +/* PVRDMA address vector. */
> +struct pvrdma_av {
> +	__u32 port_pd;
> +	__u32 sl_tclass_flowlabel;
> +	__u8 dgid[16];
> +	__u8 src_path_bits;
> +	__u8 gid_index;
> +	__u8 stat_rate;
> +	__u8 hop_limit;
> +	__u8 dmac[6];
> +	__u8 reserved[6];
> +};
> +
> +/* PVRDMA receive queue work request */
> +struct pvrdma_rq_wqe_hdr {
> +	__u64 wr_id;		/* wr id */
> +	__u32 num_sge;		/* size of s/g array */
> +	__u32 total_len;	/* reserved */
> +};
> +/* Use pvrdma_sge (ib_sge) for receieve queue s/g array elements. */
> +
> +/* PVRDMA send queue work request */
> +struct pvrdma_sq_wqe_hdr {
> +	__u64 wr_id;		/* wr id */
> +	__u32 num_sge;		/* size of s/g array */
> +	__u32 total_len;	/* reserved */
> +	__u32 opcode;		/* operation type */
> +	__u32 send_flags;	/* wr flags */
> +	union {
> +		__u32 imm_data;
> +		__u32 invalidate_rkey;
> +	} ex;
> +	__u32 reserved;
> +	union {
> +		struct {
> +			__u64 remote_addr;
> +			__u32 rkey;
> +			__u8 reserved[4];
> +		} rdma;
> +		struct {
> +			__u64 remote_addr;
> +			__u64 compare_add;
> +			__u64 swap;
> +			__u32 rkey;
> +			__u32 reserved;
> +		} atomic;
> +		struct {
> +			__u64 remote_addr;
> +			__u32 log_arg_sz;
> +			__u32 rkey;
> +			union {
> +				struct pvrdma_exp_cmp_swap  cmp_swap;
> +				struct pvrdma_exp_fetch_add fetch_add;
> +			} wr_data;
> +		} masked_atomics;
> +		struct {
> +			__u64 iova_start;
> +			__u64 pl_pdir_dma;
> +			__u32 page_shift;
> +			__u32 page_list_len;
> +			__u32 length;
> +			__u32 access_flags;
> +			__u32 rkey;
> +		} fast_reg;
> +		struct {
> +			__u32 remote_qpn;
> +			__u32 remote_qkey;
> +			struct pvrdma_av av;
> +		} ud;
> +	} wr;
> +};
> +/* Use pvrdma_sge (ib_sge) for send queue s/g array elements. */
> +
> +/* Completion queue element. */
> +struct pvrdma_cqe {
> +	__u64 wr_id;
> +	__u64 qp;
> +	__u32 opcode;
> +	__u32 status;
> +	__u32 byte_len;
> +	__u32 imm_data;
> +	__u32 src_qp;
> +	__u32 wc_flags;
> +	__u32 vendor_err;
> +	__u16 pkey_index;
> +	__u16 slid;
> +	__u8 sl;
> +	__u8 dlid_path_bits;
> +	__u8 port_num;
> +	__u8 smac[6];
> +	__u8 reserved2[7]; /* Pad to next power of 2 (64). */
> +};
> +
> +struct pvrdma_ring {
> +	atomic_t prod_tail;	/* Producer tail. */
> +	atomic_t cons_head;	/* Consumer head. */
> +};
> +
> +struct pvrdma_ring_state {
> +	struct pvrdma_ring tx;	/* Tx ring. */
> +	struct pvrdma_ring rx;	/* Rx ring. */
> +};
> +
> +static inline int pvrdma_idx_valid(__u32 idx, __u32 max_elems)
> +{
> +	/* Generates fewer instructions than a less-than. */
> +	return (idx & ~((max_elems << 1) - 1)) == 0;
> +}
> +
> +static inline __s32 pvrdma_idx(atomic_t *var, __u32 max_elems)
> +{
> +	const unsigned int idx = atomic_read(var);
> +
> +	if (pvrdma_idx_valid(idx, max_elems))
> +		return idx & (max_elems - 1);
> +	return -1;
> +}
> +
> +static inline void pvrdma_idx_ring_inc(atomic_t *var, __u32 max_elems)
> +{
> +	__u32 idx = atomic_read(var) + 1;	/* Increment. */
> +
> +	idx &= (max_elems << 1) - 1;		/* Modulo size, flip gen. */
> +	atomic_set(var, idx);
> +}
> +
> +static inline __s32 pvrdma_idx_ring_has_space(const struct pvrdma_ring *r,
> +					      __u32 max_elems, __u32 *out_tail)
> +{
> +	const __u32 tail = atomic_read(&r->prod_tail);
> +	const __u32 head = atomic_read(&r->cons_head);
> +
> +	if (pvrdma_idx_valid(tail, max_elems) &&
> +	    pvrdma_idx_valid(head, max_elems)) {
> +		*out_tail = tail & (max_elems - 1);
> +		return tail != (head ^ max_elems);
> +	}
> +	return -1;
> +}
> +
> +static inline __s32 pvrdma_idx_ring_has_data(const struct pvrdma_ring *r,
> +					     __u32 max_elems, __u32 *out_head)
> +{
> +	const __u32 tail = atomic_read(&r->prod_tail);
> +	const __u32 head = atomic_read(&r->cons_head);
> +
> +	if (pvrdma_idx_valid(tail, max_elems) &&
> +	    pvrdma_idx_valid(head, max_elems)) {
> +		*out_head = head & (max_elems - 1);
> +		return tail != head;
> +	}
> +	return -1;

What is the meaning of -1?

> +}
> +
> +static inline __s32 pvrdma_idx_ring_is_valid_idx(const struct pvrdma_ring *r,
> +						 __u32 max_elems, __u32 *idx)
> +{
> +	const __u32 tail = atomic_read(&r->prod_tail);
> +	const __u32 head = atomic_read(&r->cons_head);
> +
> +	if (pvrdma_idx_valid(tail, max_elems) &&
> +	    pvrdma_idx_valid(head, max_elems) &&
> +	    pvrdma_idx_valid(*idx, max_elems)) {
> +		if (tail > head && (*idx < tail && *idx >= head))
> +			return 1;
> +		else if (head > tail && (*idx >= head || *idx < tail))
> +			return 1;
> +	}
> +	return 0;
> +}
> +
> +#endif /* __PVRDMA_UAPI_H__ */
> diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_user.h b/drivers/infiniband/hw/pvrdma/pvrdma_user.h
> new file mode 100644
> index 0000000..35f8c1e
> --- /dev/null
> +++ b/drivers/infiniband/hw/pvrdma/pvrdma_user.h
> @@ -0,0 +1,99 @@
> +/*
> + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of EITHER the GNU General Public License
> + * version 2 as published by the Free Software Foundation or the BSD
> + * 2-Clause License. This program is distributed in the hope that it
> + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> + * See the GNU General Public License version 2 for more details at
> + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program available in the file COPYING in the main
> + * directory of this source tree.
> + *
> + * The BSD 2-Clause License
> + *
> + *     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.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> + * OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#ifndef __PVRDMA_USER_H__
> +#define __PVRDMA_USER_H__
> +
> +#include <linux/types.h>
> +
> +#define PVRDMA_UVERBS_ABI_VERSION	3
> +#define PVRDMA_BOARD_ID			1
> +#define PVRDMA_REV_ID			1
> +
> +struct pvrdma_alloc_ucontext_resp {
> +	u32 qp_tab_size;
> +	u32 reserved;
> +};
> +
> +struct pvrdma_alloc_pd_resp {
> +	u32 pdn;
> +	u32 reserved;
> +};
> +
> +struct pvrdma_create_cq {
> +	u64 buf_addr;
> +	u32 buf_size;
> +	u32 reserved;
> +};
> +
> +struct pvrdma_create_cq_resp {
> +	u32 cqn;
> +	u32 reserved;
> +};
> +
> +struct pvrdma_resize_cq {
> +	u64 buf_addr;
> +	u32 buf_size;
> +	u32 reserved;
> +};
> +
> +struct pvrdma_create_srq {
> +	u64 buf_addr;
> +};
> +
> +struct pvrdma_create_srq_resp {
> +	u32 srqn;
> +	u32 reserved;
> +};
> +
> +struct pvrdma_create_qp {
> +	u64 rbuf_addr;
> +	u64 sbuf_addr;
> +	u32 rbuf_size;
> +	u32 sbuf_size;
> +	u64 qp_addr;
> +};
> +
> +#endif /* __PVRDMA_USER_H__ */
> -- 
> 2.7.4
> 
> --
> 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
--
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 03/15] IB/pvrdma: Add support for Completion Queues

[Yuval Shaia]

On Tue, Jul 12, 2016 at 12:36:33PM -0700, Adit Ranadive wrote:
> This patch adds the support for creating and destroying completion queues
> on the paravirtual RDMA device.
> 
> Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> ---
>  drivers/infiniband/hw/pvrdma/pvrdma_cq.c | 436 +++++++++++++++++++++++++++++++
>  1 file changed, 436 insertions(+)
>  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_cq.c
> 
> diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_cq.c b/drivers/infiniband/hw/pvrdma/pvrdma_cq.c
> new file mode 100644
> index 0000000..9a4b42c
> --- /dev/null
> +++ b/drivers/infiniband/hw/pvrdma/pvrdma_cq.c
> @@ -0,0 +1,436 @@
> +/*
> + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of EITHER the GNU General Public License
> + * version 2 as published by the Free Software Foundation or the BSD
> + * 2-Clause License. This program is distributed in the hope that it
> + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> + * See the GNU General Public License version 2 for more details at
> + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program available in the file COPYING in the main
> + * directory of this source tree.
> + *
> + * The BSD 2-Clause License
> + *
> + *     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.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> + * OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#include <asm/page.h>
> +#include <linux/io.h>
> +#include <linux/wait.h>
> +#include <rdma/ib_addr.h>
> +#include <rdma/ib_smi.h>
> +#include <rdma/ib_user_verbs.h>
> +
> +#include "pvrdma.h"
> +#include "pvrdma_user.h"
> +
> +/**
> + * pvrdma_req_notify_cq - request notification for a completion queue
> + * @ibcq: the completion queue
> + * @notify_flags: notification flags
> + *
> + * @return: 0 for success.
> + */
> +int pvrdma_req_notify_cq(struct ib_cq *ibcq,
> +			 enum ib_cq_notify_flags notify_flags)
> +{
> +	struct pvrdma_dev *dev = to_vdev(ibcq->device);
> +	struct pvrdma_cq *cq = to_vcq(ibcq);
> +	u32 val = cq->cq_handle;
> +
> +	val |= (notify_flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED ?
> +		PVRDMA_UAR_CQ_ARM_SOL : PVRDMA_UAR_CQ_ARM;
> +
> +	writel(cpu_to_le32(val), dev->driver_uar.map + PVRDMA_UAR_CQ_OFFSET);
> +
> +	return 0;
> +}
> +
> +/**
> + * pvrdma_create_cq - create completion queue
> + * @ibdev: the device
> + * @attr: completion queue attributes
> + * @context: user context
> + * @udata: user data
> + *
> + * @return: ib_cq completion queue pointer on success,
> + *          otherwise returns negative errno.
> + */
> +struct ib_cq *pvrdma_create_cq(struct ib_device *ibdev,
> +			       const struct ib_cq_init_attr *attr,
> +			       struct ib_ucontext *context,
> +			       struct ib_udata *udata)
> +{
> +	int entries = attr->cqe;
> +	struct pvrdma_dev *dev = to_vdev(ibdev);
> +	struct pvrdma_cq *cq;
> +	int ret;
> +	int npages;
> +	unsigned long flags;
> +	union pvrdma_cmd_req req;
> +	union pvrdma_cmd_resp rsp;
> +	struct pvrdma_cmd_create_cq *cmd = &req.create_cq;
> +	struct pvrdma_cmd_create_cq_resp *resp = &rsp.create_cq_resp;
> +	struct pvrdma_create_cq ucmd;
> +
> +	BUILD_BUG_ON(sizeof(struct pvrdma_cqe) != 64);
> +
> +	entries = roundup_pow_of_two(entries);
> +	if (entries < 1 || entries > dev->dsr->caps.max_cqe)
> +		return ERR_PTR(-EINVAL);
> +
> +	if (!atomic_add_unless(&dev->num_cqs, 1, dev->dsr->caps.max_cq))
> +		return ERR_PTR(-EINVAL);
> +
> +	cq = kzalloc(sizeof(*cq), GFP_KERNEL);
> +	if (!cq) {
> +		atomic_dec(&dev->num_cqs);
> +		return ERR_PTR(-ENOMEM);
> +	}
> +
> +	cq->ibcq.cqe = entries;
> +
> +	if (context) {
> +		if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) {
> +			ret = -EFAULT;
> +			goto err_cq;
> +		}
> +
> +		cq->umem = ib_umem_get(context, ucmd.buf_addr, ucmd.buf_size,
> +				       IB_ACCESS_LOCAL_WRITE, 1);
> +		if (IS_ERR(cq->umem)) {
> +			ret = PTR_ERR(cq->umem);
> +			goto err_cq;
> +		}
> +
> +		npages = ib_umem_page_count(cq->umem);
> +	} else {
> +		cq->is_kernel = true;
> +
> +		/* One extra page for shared ring state */
> +		npages = 1 + (entries * sizeof(struct pvrdma_cqe) +
> +			      PAGE_SIZE - 1) / PAGE_SIZE;
> +
> +		/* Skip header page. */
> +		cq->offset = PAGE_SIZE;
> +	}
> +
> +	if (npages < 0 || npages > PVRDMA_PAGE_DIR_MAX_PAGES) {
> +		dev_warn(&dev->pdev->dev,
> +			 "overflow pages in completion queue\n");
> +		ret = -EINVAL;
> +		goto err_umem;
> +	}
> +
> +	ret = pvrdma_page_dir_init(dev, &cq->pdir, npages, cq->is_kernel);
> +	if (ret) {
> +		dev_warn(&dev->pdev->dev,
> +			 "could not allocate page directory\n");
> +		goto err_umem;
> +	}
> +
> +	if (cq->is_kernel) {
> +		/* Ring state is always the first page. */
> +		cq->ring_state = cq->pdir.pages[0];
> +	} else {
> +		pvrdma_page_dir_insert_umem(&cq->pdir, cq->umem, 0);
> +	}
> +
> +	atomic_set(&cq->refcnt, 1);
> +	init_waitqueue_head(&cq->wait);
> +	spin_lock_init(&cq->cq_lock);
> +
> +	memset(cmd, 0, sizeof(*cmd));
> +	cmd->hdr.cmd = PVRDMA_CMD_CREATE_CQ;
> +	cmd->nchunks = npages;
> +	cmd->ctx_handle = (context) ?
> +		(u64)to_vucontext(context)->ctx_handle : 0;
> +	cmd->cqe = entries;
> +	cmd->pdir_dma = cq->pdir.dir_dma;
> +	ret = pvrdma_cmd_post(dev, &req, true, &rsp);
> +
> +	if (ret < 0 || rsp.hdr.ack != PVRDMA_CMD_CREATE_CQ_RESP) {
> +		dev_warn(&dev->pdev->dev,
> +			 "could not create completion queue\n");
> +		goto err_page_dir;
> +	}
> +
> +	cq->ibcq.cqe = resp->cqe;
> +	cq->cq_handle = resp->cq_handle;
> +	spin_lock_irqsave(&dev->cq_tbl_lock, flags);
> +	dev->cq_tbl[cq->cq_handle % dev->dsr->caps.max_cq] = cq;
> +	spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
> +
> +	if (context) {
> +		cq->uar = &(to_vucontext(context)->uar);
> +
> +		/* Copy udata back. */
> +		if (ib_copy_to_udata(udata, &cq->cq_handle, sizeof(__u32))) {
> +			dev_warn(&dev->pdev->dev,
> +				 "failed to copy back udata\n");
> +			ret = -EINVAL;
> +			goto err_page_dir;
> +		}
> +	}
> +
> +	return &cq->ibcq;
> +
> +err_page_dir:
> +	pvrdma_page_dir_cleanup(dev, &cq->pdir);
> +err_umem:
> +	if (context)
> +		ib_umem_release(cq->umem);
> +err_cq:
> +	atomic_dec(&dev->num_cqs);
> +	kfree(cq);
> +
> +	return ERR_PTR(ret);
> +}
> +
> +static void pvrdma_free_cq(struct pvrdma_dev *dev, struct pvrdma_cq *cq)
> +{
> +	atomic_dec(&cq->refcnt);
> +	wait_event(cq->wait, !atomic_read(&cq->refcnt));
> +
> +	if (!cq->is_kernel)
> +		ib_umem_release(cq->umem);
> +
> +	pvrdma_page_dir_cleanup(dev, &cq->pdir);
> +	kfree(cq);
> +}
> +
> +/**
> + * pvrdma_destroy_cq - destroy completion queue
> + * @cq: the completion queue to destroy.
> + *
> + * @return: 0 for success.
> + */
> +int pvrdma_destroy_cq(struct ib_cq *cq)
> +{
> +	struct pvrdma_cq *vcq = to_vcq(cq);
> +	union pvrdma_cmd_req req;
> +	union pvrdma_cmd_resp rsp;
> +	struct pvrdma_cmd_destroy_cq *cmd = &req.destroy_cq;
> +	struct pvrdma_dev *dev = to_vdev(cq->device);
> +	unsigned long flags;
> +	int ret;
> +
> +	memset(cmd, 0, sizeof(*cmd));
> +	cmd->hdr.cmd = PVRDMA_CMD_DESTROY_CQ;
> +	cmd->cq_handle = vcq->cq_handle;
> +
> +	ret = pvrdma_cmd_post(dev, &req, false, &rsp);
> +	if (ret < 0)
> +		dev_warn(&dev->pdev->dev,
> +			 "could not destroy completion queue\n");
> +
> +	/* free cq's resources */
> +	spin_lock_irqsave(&dev->cq_tbl_lock, flags);
> +	dev->cq_tbl[vcq->cq_handle] = NULL;
> +	spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
> +
> +	pvrdma_free_cq(dev, vcq);
> +	atomic_dec(&dev->num_cqs);
> +
> +	return ret;
> +}
> +
> +/**
> + * pvrdma_modify_cq - modify the CQ moderation parameters
> + * @ibcq: the CQ to modify
> + * @cq_count: number of CQEs that will trigger an event
> + * @cq_period: max period of time in usec before triggering an event
> + *
> + * @return: -EOPNOTSUPP as CQ resize is not supported.
> + */
> +int pvrdma_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
> +{
> +	return -EOPNOTSUPP;
> +}
> +
> +static inline struct pvrdma_cqe *get_cqe(struct pvrdma_cq *cq, int i)
> +{
> +	return (struct pvrdma_cqe *)pvrdma_page_dir_get_ptr(
> +					&cq->pdir,
> +					cq->offset +
> +					sizeof(struct pvrdma_cqe) * i);
> +}
> +
> +void pvrdma_flush_cqe(struct pvrdma_qp *qp, struct pvrdma_cq *cq)
> +{
> +	int head;
> +	int has_data;
> +
> +	if (!cq->is_kernel)
> +		return;
> +
> +	/* Lock held */
> +	has_data = pvrdma_idx_ring_has_data(&cq->ring_state->rx,
> +					    cq->ibcq.cqe, &head);

For easy review (and probably maintenance) it looks to me that we need to
preserve some sort of order where function/structures/union/enum etc will
be define in preceding patch in a patch-set.
This function implemented in patch #11 while we are in patch #3.

> +	if (unlikely(has_data > 0)) {
> +		int items;
> +		int curr;
> +		int tail = pvrdma_idx(&cq->ring_state->rx.prod_tail,
> +				      cq->ibcq.cqe);
> +		struct pvrdma_cqe *cqe;
> +		struct pvrdma_cqe *curr_cqe;
> +
> +		items = (tail > head) ? (tail - head) :
> +			(cq->ibcq.cqe - head + tail);
> +		curr = --tail;
> +		while (items-- > 0) {
> +			if (curr < 0)
> +				curr = cq->ibcq.cqe - 1;
> +			if (tail < 0)
> +				tail = cq->ibcq.cqe - 1;
> +			curr_cqe = get_cqe(cq, curr);
> +			if ((curr_cqe->qp & 0xFFFF) != qp->qp_handle) {
> +				if (curr != tail) {
> +					cqe = get_cqe(cq, tail);
> +					*cqe = *curr_cqe;
> +				}
> +				tail--;
> +			} else {
> +				pvrdma_idx_ring_inc(
> +					&cq->ring_state->rx.cons_head,
> +					cq->ibcq.cqe);
> +			}
> +			curr--;
> +		}
> +	}
> +}
> +
> +static int pvrdma_poll_one(struct pvrdma_cq *cq, struct pvrdma_qp **cur_qp,
> +			   struct ib_wc *wc)
> +{
> +	struct pvrdma_dev *dev = to_vdev(cq->ibcq.device);
> +	int has_data;
> +	unsigned int head;
> +	bool tried = false;
> +	struct pvrdma_cqe *cqe;
> +
> +retry:
> +	has_data = pvrdma_idx_ring_has_data(&cq->ring_state->rx,
> +					    cq->ibcq.cqe, &head);
> +	if (has_data == 0) {
> +		u32 val;
> +
> +		if (tried)
> +			return -EAGAIN;
> +
> +		/* Pass down POLL to give physical HCA a chance to poll. */
> +		val = cq->cq_handle | PVRDMA_UAR_CQ_POLL;
> +		writel(cpu_to_le32(val),
> +		       dev->driver_uar.map + PVRDMA_UAR_CQ_OFFSET);
> +
> +		tried = true;
> +		goto retry;
> +	} else if (has_data == -1) {

Is -1 represent fatal unrecoverable error or that next call to
pvrdma_idx_ring_has_data might succeeds?

> +		return -EINVAL;
> +	}
> +
> +	cqe = get_cqe(cq, head);
> +
> +	/* Ensure cqe is valid. */
> +	rmb();
> +	if (dev->qp_tbl[cqe->qp & 0xffff])
> +		*cur_qp = (struct pvrdma_qp *)dev->qp_tbl[cqe->qp & 0xffff];
> +	else
> +		return -EINVAL;

Is this can happen when QP was deleted while WR was processing?
Or that in first place it was pushed with invalid QP?

Can you explain why you choose EINVAL and not EAGAIN? i.e we polled the
"corrupted" CQE, next CEQ should be fine.
Reading pvrdma_poll_cq looks like that if it was requested to poll 10 CQE
for example and the 9th fails - the function will fail them all.

> +
> +	wc->opcode = pvrdma_wc_opcode_to_ib(cqe->opcode);
> +	wc->status = pvrdma_wc_status_to_ib(cqe->status);
> +	wc->wr_id = cqe->wr_id;
> +	wc->qp = &(*cur_qp)->ibqp;
> +	wc->byte_len = cqe->byte_len;
> +	wc->ex.imm_data = cqe->imm_data;
> +	wc->src_qp = cqe->src_qp;
> +	wc->wc_flags = pvrdma_wc_flags_to_ib(cqe->wc_flags);
> +	wc->pkey_index = cqe->pkey_index;
> +	wc->slid = cqe->slid;
> +	wc->sl = cqe->sl;
> +	wc->dlid_path_bits = cqe->dlid_path_bits;
> +	wc->port_num = cqe->port_num;
> +	wc->vendor_err = 0;
> +
> +	/* Update shared ring state */
> +	pvrdma_idx_ring_inc(&cq->ring_state->rx.cons_head, cq->ibcq.cqe);
> +
> +	return 0;
> +}
> +
> +/**
> + * pvrdma_poll_cq - poll for work completion queue entries
> + * @ibcq: completion queue
> + * @num_entries: the maximum number of entries
> + * @entry: pointer to work completion array
> + *
> + * @return: number of polled completion entries
> + */
> +int pvrdma_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
> +{
> +	struct pvrdma_cq *cq = to_vcq(ibcq);
> +	struct pvrdma_qp *cur_qp = NULL;
> +	unsigned long flags;
> +	int npolled;
> +	int ret;
> +
> +	if (num_entries < 1)
> +		return -EINVAL;
> +
> +	spin_lock_irqsave(&cq->cq_lock, flags);
> +	for (npolled = 0; npolled < num_entries; ++npolled) {
> +		ret = pvrdma_poll_one(cq, &cur_qp, wc + npolled);
> +		if (ret)
> +			break;
> +	}
> +
> +	spin_unlock_irqrestore(&cq->cq_lock, flags);
> +
> +	if ((ret == 0) || (ret == -EAGAIN))
> +		return npolled;
> +	else
> +		return ret;
> +}
> +
> +/**
> + * pvrdma_resize_cq - resize CQ
> + * @ibcq: the completion queue
> + * @entries: CQ entries
> + * @udata: user data
> + *
> + * @return: -EOPNOTSUPP as CQ resize is not supported.
> + */
> +int pvrdma_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
> +{
> +	return -EOPNOTSUPP;
> +}
> -- 
> 2.7.4
> 
> --
> 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
--
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 07/15] IB/pvrdma: Add the main driver module for PVRDMA

[Yuval Shaia]

Reviewed-by: Yuval Shaia <yuval.shaia-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org>

On Tue, Jul 12, 2016 at 12:36:37PM -0700, Adit Ranadive wrote:
> This patch adds the support to register a RDMA device with the kernel RDMA
> stack as well as a kernel module. This also initializes the underlying
> virtual PCI device.
> 
> Changes from v1: Addressed 32-bit build errors. Addressed comments by
> Yuval Shaia.
> 
> Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> ---
>  drivers/infiniband/hw/pvrdma/pvrdma_main.c | 1189 ++++++++++++++++++++++++++++
>  1 file changed, 1189 insertions(+)
>  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_main.c
> 
> diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_main.c b/drivers/infiniband/hw/pvrdma/pvrdma_main.c
> new file mode 100644
> index 0000000..20073b7
> --- /dev/null
> +++ b/drivers/infiniband/hw/pvrdma/pvrdma_main.c
> @@ -0,0 +1,1189 @@
> +/*
> + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of EITHER the GNU General Public License
> + * version 2 as published by the Free Software Foundation or the BSD
> + * 2-Clause License. This program is distributed in the hope that it
> + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> + * See the GNU General Public License version 2 for more details at
> + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program available in the file COPYING in the main
> + * directory of this source tree.
> + *
> + * The BSD 2-Clause License
> + *
> + *     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.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> + * OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#include <linux/errno.h>
> +#include <linux/inetdevice.h>
> +#include <linux/init.h>
> +#include <linux/module.h>
> +#include <linux/slab.h>
> +#include <rdma/ib_addr.h>
> +#include <rdma/ib_smi.h>
> +#include <rdma/ib_user_verbs.h>
> +#include <net/addrconf.h>
> +
> +#include "pvrdma.h"
> +#include "pvrdma_user.h"
> +
> +#define PCI_DEVICE_ID_VMWARE_VMXNET3	0x07B0
> +
> +#define DRV_NAME	"pvrdma"
> +#define DRV_VERSION	"1.0"
> +#define DRV_RELDATE	"January 1, 2013"
> +
> +static const char pvrdma_version[] =
> +	DRV_NAME ": PVRDMA InfiniBand driver v"
> +	DRV_VERSION " (" DRV_RELDATE ")\n";
> +
> +static DEFINE_MUTEX(pvrdma_device_list_lock);
> +static LIST_HEAD(pvrdma_device_list);
> +
> +static int pvrdma_add_gid(struct ib_device *ibdev,
> +			  u8 port_num,
> +			  unsigned int index,
> +			  const union ib_gid *gid,
> +			  const struct ib_gid_attr *attr,
> +			  void **context);
> +static int pvrdma_del_gid(struct ib_device *ibdev,
> +			  u8 port_num,
> +			  unsigned int index,
> +			  void **context);
> +
> +
> +static ssize_t show_hca(struct device *device, struct device_attribute *attr,
> +			char *buf)
> +{
> +	return sprintf(buf, "PVRDMA%s\n", DRV_VERSION);
> +}
> +
> +static ssize_t show_fw_ver(struct device *device,
> +			   struct device_attribute *attr, char *buf)
> +{
> +	struct pvrdma_dev *dev =
> +		container_of(device, struct pvrdma_dev, ib_dev.dev);
> +
> +	return sprintf(buf, "%d.%d.%d\n",
> +		       (int)(dev->dsr->caps.fw_ver >> 32),
> +		       (int)(dev->dsr->caps.fw_ver >> 16) & 0xffff,
> +		       (int)dev->dsr->caps.fw_ver & 0xffff);
> +}
> +
> +static ssize_t show_rev(struct device *device, struct device_attribute *attr,
> +			char *buf)
> +{
> +	return sprintf(buf, "%d\n", PVRDMA_REV_ID);
> +}
> +
> +static ssize_t show_board(struct device *device, struct device_attribute *attr,
> +			  char *buf)
> +{
> +	return sprintf(buf, "%d\n", PVRDMA_BOARD_ID);
> +}
> +
> +static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,	   NULL);
> +static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
> +static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,	   NULL);
> +static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
> +
> +static struct device_attribute *pvrdma_class_attributes[] = {
> +	&dev_attr_hw_rev,
> +	&dev_attr_fw_ver,
> +	&dev_attr_hca_type,
> +	&dev_attr_board_id
> +};
> +
> +static int pvrdma_init_device(struct pvrdma_dev *dev)
> +{
> +	/*  Initialize some device related stuff */
> +	spin_lock_init(&dev->cmd_lock);
> +	sema_init(&dev->cmd_sema, 1);
> +	atomic_set(&dev->num_qps, 0);
> +	atomic_set(&dev->num_cqs, 0);
> +	atomic_set(&dev->num_pds, 0);
> +	atomic_set(&dev->num_ahs, 0);
> +
> +	return 0;
> +}
> +
> +static int pvrdma_port_immutable(struct ib_device *ibdev, u8 port_num,
> +				 struct ib_port_immutable *immutable)
> +{
> +	struct ib_port_attr attr;
> +	int err;
> +
> +	err = pvrdma_query_port(ibdev, port_num, &attr);
> +	if (err)
> +		return err;
> +
> +	immutable->pkey_tbl_len = attr.pkey_tbl_len;
> +	immutable->gid_tbl_len = attr.gid_tbl_len;
> +	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
> +	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
> +	return 0;
> +}
> +
> +static struct net_device *pvrdma_get_netdev(struct ib_device *ibdev,
> +					    u8 port_num)
> +{
> +	struct net_device *netdev;
> +	struct pvrdma_dev *dev = to_vdev(ibdev);
> +
> +	if (port_num != 1)
> +		return NULL;
> +
> +	rcu_read_lock();
> +	netdev = dev->netdev;
> +	if (netdev)
> +		dev_hold(netdev);
> +	rcu_read_unlock();
> +
> +	return netdev;
> +}
> +
> +static int pvrdma_register_device(struct pvrdma_dev *dev)
> +{
> +	int ret = -1;
> +	int i = 0;
> +
> +	strlcpy(dev->ib_dev.name, "pvrdma%d", IB_DEVICE_NAME_MAX);
> +	dev->ib_dev.node_guid = dev->dsr->caps.node_guid;
> +	dev->sys_image_guid = dev->dsr->caps.sys_image_guid;
> +	dev->flags = 0;
> +	dev->ib_dev.owner = THIS_MODULE;
> +	dev->ib_dev.num_comp_vectors = 1;
> +	dev->ib_dev.dma_device = &dev->pdev->dev;
> +	dev->ib_dev.uverbs_abi_ver = PVRDMA_UVERBS_ABI_VERSION;
> +	dev->ib_dev.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_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_POLL_CQ)		|
> +		(1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ)	|
> +		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
> +		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
> +		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
> +		(1ull << IB_USER_VERBS_CMD_QUERY_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_CREATE_AH)		|
> +		(1ull << IB_USER_VERBS_CMD_DESTROY_AH);
> +
> +	dev->ib_dev.node_type = RDMA_NODE_IB_CA;
> +	dev->ib_dev.phys_port_cnt = dev->dsr->caps.phys_port_cnt;
> +
> +	dev->ib_dev.query_device = pvrdma_query_device;
> +	dev->ib_dev.query_port = pvrdma_query_port;
> +	dev->ib_dev.query_gid = pvrdma_query_gid;
> +	dev->ib_dev.query_pkey = pvrdma_query_pkey;
> +	dev->ib_dev.modify_port	= pvrdma_modify_port;
> +	dev->ib_dev.alloc_ucontext = pvrdma_alloc_ucontext;
> +	dev->ib_dev.dealloc_ucontext = pvrdma_dealloc_ucontext;
> +	dev->ib_dev.mmap = pvrdma_mmap;
> +	dev->ib_dev.alloc_pd = pvrdma_alloc_pd;
> +	dev->ib_dev.dealloc_pd = pvrdma_dealloc_pd;
> +	dev->ib_dev.create_ah = pvrdma_create_ah;
> +	dev->ib_dev.destroy_ah = pvrdma_destroy_ah;
> +	dev->ib_dev.create_qp = pvrdma_create_qp;
> +	dev->ib_dev.modify_qp = pvrdma_modify_qp;
> +	dev->ib_dev.query_qp = pvrdma_query_qp;
> +	dev->ib_dev.destroy_qp = pvrdma_destroy_qp;
> +	dev->ib_dev.post_send = pvrdma_post_send;
> +	dev->ib_dev.post_recv = pvrdma_post_recv;
> +	dev->ib_dev.create_cq = pvrdma_create_cq;
> +	dev->ib_dev.modify_cq = pvrdma_modify_cq;
> +	dev->ib_dev.resize_cq = pvrdma_resize_cq;
> +	dev->ib_dev.destroy_cq = pvrdma_destroy_cq;
> +	dev->ib_dev.poll_cq = pvrdma_poll_cq;
> +	dev->ib_dev.req_notify_cq = pvrdma_req_notify_cq;
> +	dev->ib_dev.get_dma_mr = pvrdma_get_dma_mr;
> +	dev->ib_dev.reg_user_mr	= pvrdma_reg_user_mr;
> +	dev->ib_dev.dereg_mr = pvrdma_dereg_mr;
> +	dev->ib_dev.alloc_mr = pvrdma_alloc_mr;
> +	dev->ib_dev.map_mr_sg = pvrdma_map_mr_sg;
> +	dev->ib_dev.add_gid = pvrdma_add_gid;
> +	dev->ib_dev.del_gid = pvrdma_del_gid;
> +	dev->ib_dev.get_netdev = pvrdma_get_netdev;
> +	dev->ib_dev.get_port_immutable = pvrdma_port_immutable;
> +	dev->ib_dev.get_link_layer = pvrdma_port_link_layer;
> +
> +	mutex_init(&dev->port_mutex);
> +	spin_lock_init(&dev->desc_lock);
> +
> +	dev->cq_tbl = kcalloc(dev->dsr->caps.max_cq, sizeof(void *),
> +			      GFP_KERNEL);
> +	if (!dev->cq_tbl)
> +		return ret;
> +	spin_lock_init(&dev->cq_tbl_lock);
> +
> +	dev->qp_tbl = kcalloc(dev->dsr->caps.max_qp, sizeof(void *),
> +			      GFP_KERNEL);
> +	if (!dev->qp_tbl)
> +		goto err_cq_free;
> +	spin_lock_init(&dev->qp_tbl_lock);
> +
> +	ret = ib_register_device(&dev->ib_dev, NULL);
> +	if (ret)
> +		goto err_qp_free;
> +
> +	for (i = 0; i < ARRAY_SIZE(pvrdma_class_attributes); ++i) {
> +		ret = device_create_file(&dev->ib_dev.dev,
> +					 pvrdma_class_attributes[i]);
> +		if (ret)
> +			goto err_class;
> +	}
> +
> +	dev->ib_active = true;
> +
> +	return 0;
> +
> +err_class:
> +	ib_unregister_device(&dev->ib_dev);
> +err_qp_free:
> +	kfree(dev->qp_tbl);
> +err_cq_free:
> +	kfree(dev->cq_tbl);
> +
> +	return ret;
> +}
> +
> +static irqreturn_t pvrdma_intr0_handler(int irq, void *dev_id)
> +{
> +	u32 icr = PVRDMA_INTR_CAUSE_RESPONSE;
> +	struct pvrdma_dev *dev = dev_id;
> +
> +	dev_dbg(&dev->pdev->dev, "interrupt 0 (response) handler\n");
> +
> +	if (dev->intr.type != PVRDMA_INTR_TYPE_MSIX) {
> +		/* Legacy intr */
> +		icr = pvrdma_read_reg(dev, PVRDMA_REG_ICR);
> +		if (icr == 0)
> +			return IRQ_NONE;
> +	}
> +
> +	if (icr == PVRDMA_INTR_CAUSE_RESPONSE)
> +		complete(&dev->cmd_done);
> +
> +	return IRQ_HANDLED;
> +}
> +
> +static void pvrdma_qp_event(struct pvrdma_dev *dev, u32 qpn, int type)
> +{
> +	struct pvrdma_qp *qp;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&dev->qp_tbl_lock, flags);
> +	qp = dev->qp_tbl[qpn % dev->dsr->caps.max_qp];
> +	if (qp)
> +		atomic_inc(&qp->refcnt);
> +	spin_unlock_irqrestore(&dev->qp_tbl_lock, flags);
> +
> +	if (qp && qp->ibqp.event_handler) {
> +		struct ib_qp *ibqp = &qp->ibqp;
> +		struct ib_event e;
> +
> +		e.device = ibqp->device;
> +		e.element.qp = ibqp;
> +		e.event = type; /* 1:1 mapping for now. */
> +		ibqp->event_handler(&e, ibqp->qp_context);
> +	}
> +	if (qp) {
> +		atomic_dec(&qp->refcnt);
> +		if (atomic_read(&qp->refcnt) == 0)
> +			wake_up(&qp->wait);
> +	}
> +}
> +
> +static void pvrdma_cq_event(struct pvrdma_dev *dev, u32 cqn, int type)
> +{
> +	struct pvrdma_cq *cq;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&dev->cq_tbl_lock, flags);
> +	cq = dev->cq_tbl[cqn % dev->dsr->caps.max_cq];
> +	if (cq)
> +		atomic_inc(&cq->refcnt);
> +	spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
> +
> +	if (cq && cq->ibcq.event_handler) {
> +		struct ib_cq *ibcq = &cq->ibcq;
> +		struct ib_event e;
> +
> +		e.device = ibcq->device;
> +		e.element.cq = ibcq;
> +		e.event = type; /* 1:1 mapping for now. */
> +		ibcq->event_handler(&e, ibcq->cq_context);
> +	}
> +	if (cq) {
> +		atomic_dec(&cq->refcnt);
> +		if (atomic_read(&cq->refcnt) == 0)
> +			wake_up(&cq->wait);
> +	}
> +}
> +
> +static void pvrdma_dispatch_event(struct pvrdma_dev *dev, int port,
> +				  enum ib_event_type event)
> +{
> +	struct ib_event ib_event;
> +
> +	memset(&ib_event, 0, sizeof(ib_event));
> +	ib_event.device = &dev->ib_dev;
> +	ib_event.element.port_num = port;
> +	ib_event.event = event;
> +	ib_dispatch_event(&ib_event);
> +}
> +
> +static void pvrdma_dev_event(struct pvrdma_dev *dev, u8 port, int type)
> +{
> +	if (port < 1 || port > dev->dsr->caps.phys_port_cnt) {
> +		dev_warn(&dev->pdev->dev, "event on port %d\n", port);
> +		return;
> +	}
> +
> +	pvrdma_dispatch_event(dev, port, type);
> +}
> +
> +static inline struct pvrdma_eqe *get_eqe(struct pvrdma_dev *dev, unsigned int i)
> +{
> +	return (struct pvrdma_eqe *)pvrdma_page_dir_get_ptr(
> +					&dev->async_pdir,
> +					PAGE_SIZE +
> +					sizeof(struct pvrdma_eqe) * i);
> +}
> +
> +static irqreturn_t pvrdma_intr1_handler(int irq, void *dev_id)
> +{
> +	struct pvrdma_dev *dev = dev_id;
> +	struct pvrdma_ring *ring = &dev->async_ring_state->rx;
> +	int ring_slots = (dev->dsr->async_ring_pages.num_pages - 1) *
> +			 PAGE_SIZE / sizeof(struct pvrdma_eqe);
> +	unsigned int head;
> +
> +	dev_dbg(&dev->pdev->dev, "interrupt 1 (async event) handler\n");
> +
> +	/*
> +	 * Don't process events until the IB device is registered. Otherwise
> +	 * we'll try to ib_dispatch_event() on an invalid device.
> +	 */
> +	if (!dev->ib_active)
> +		return IRQ_HANDLED;
> +
> +	while (pvrdma_idx_ring_has_data(ring, ring_slots, &head) > 0) {
> +		struct pvrdma_eqe *eqe;
> +
> +		eqe = get_eqe(dev, head);
> +
> +		switch (eqe->type) {
> +		case PVRDMA_EVENT_QP_FATAL:
> +		case PVRDMA_EVENT_QP_REQ_ERR:
> +		case PVRDMA_EVENT_QP_ACCESS_ERR:
> +		case PVRDMA_EVENT_COMM_EST:
> +		case PVRDMA_EVENT_SQ_DRAINED:
> +		case PVRDMA_EVENT_PATH_MIG:
> +		case PVRDMA_EVENT_PATH_MIG_ERR:
> +		case PVRDMA_EVENT_QP_LAST_WQE_REACHED:
> +			pvrdma_qp_event(dev, eqe->info, eqe->type);
> +			break;
> +
> +		case PVRDMA_EVENT_CQ_ERR:
> +			pvrdma_cq_event(dev, eqe->info, eqe->type);
> +			break;
> +
> +		case PVRDMA_EVENT_SRQ_ERR:
> +		case PVRDMA_EVENT_SRQ_LIMIT_REACHED:
> +			break;
> +
> +		case PVRDMA_EVENT_PORT_ACTIVE:
> +		case PVRDMA_EVENT_PORT_ERR:
> +		case PVRDMA_EVENT_LID_CHANGE:
> +		case PVRDMA_EVENT_PKEY_CHANGE:
> +		case PVRDMA_EVENT_SM_CHANGE:
> +		case PVRDMA_EVENT_CLIENT_REREGISTER:
> +		case PVRDMA_EVENT_GID_CHANGE:
> +			pvrdma_dev_event(dev, eqe->info, eqe->type);
> +			break;
> +
> +		case PVRDMA_EVENT_DEVICE_FATAL:
> +			pvrdma_dev_event(dev, 1, eqe->type);
> +			break;
> +
> +		default:
> +			break;
> +		}
> +
> +		pvrdma_idx_ring_inc(&ring->cons_head, ring_slots);
> +	}
> +
> +	return IRQ_HANDLED;
> +}
> +
> +static inline struct pvrdma_cqne *get_cqne(struct pvrdma_dev *dev,
> +					  unsigned int i)
> +{
> +	return (struct pvrdma_cqne *)pvrdma_page_dir_get_ptr(
> +					&dev->cq_pdir,
> +					PAGE_SIZE +
> +					sizeof(struct pvrdma_cqne) * i);
> +}
> +
> +static irqreturn_t pvrdma_intrx_handler(int irq, void *dev_id)
> +{
> +	struct pvrdma_dev *dev = dev_id;
> +	struct pvrdma_ring *ring = &dev->cq_ring_state->rx;
> +	int ring_slots = (dev->dsr->cq_ring_pages.num_pages - 1) * PAGE_SIZE /
> +			 sizeof(struct pvrdma_cqne);
> +	unsigned int head;
> +	unsigned long flags;
> +
> +	dev_dbg(&dev->pdev->dev, "interrupt x (completion) handler\n");
> +
> +	while (pvrdma_idx_ring_has_data(ring, ring_slots, &head) > 0) {
> +		struct pvrdma_cqne *cqne;
> +		struct pvrdma_cq *cq;
> +
> +		cqne = get_cqne(dev, head);
> +		spin_lock_irqsave(&dev->cq_tbl_lock, flags);
> +		cq = dev->cq_tbl[cqne->info % dev->dsr->caps.max_cq];
> +		if (cq)
> +			atomic_inc(&cq->refcnt);
> +		spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
> +
> +		if (cq && cq->ibcq.comp_handler)
> +			cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
> +		if (cq) {
> +			atomic_dec(&cq->refcnt);
> +			if (atomic_read(&cq->refcnt))
> +				wake_up(&cq->wait);
> +		}
> +		pvrdma_idx_ring_inc(&ring->cons_head, ring_slots);
> +	}
> +
> +	return IRQ_HANDLED;
> +}
> +
> +static void pvrdma_disable_msi_all(struct pvrdma_dev *dev)
> +{
> +	if (dev->intr.type == PVRDMA_INTR_TYPE_MSIX)
> +		pci_disable_msix(dev->pdev);
> +	else if (dev->intr.type == PVRDMA_INTR_TYPE_MSI)
> +		pci_disable_msi(dev->pdev);
> +}
> +
> +static void pvrdma_free_irq(struct pvrdma_dev *dev)
> +{
> +	int i;
> +
> +	dev_dbg(&dev->pdev->dev, "freeing interrupts\n");
> +
> +	if (dev->intr.type == PVRDMA_INTR_TYPE_MSIX) {
> +		for (i = 0; i < dev->intr.size; i++) {
> +			if (dev->intr.enabled[i]) {
> +				free_irq(dev->intr.msix_entry[i].vector, dev);
> +				dev->intr.enabled[i] = 0;
> +			}
> +		}
> +	} else if (dev->intr.type == PVRDMA_INTR_TYPE_INTX ||
> +		   dev->intr.type == PVRDMA_INTR_TYPE_MSI) {
> +		free_irq(dev->pdev->irq, dev);
> +	}
> +}
> +
> +static void pvrdma_enable_intrs(struct pvrdma_dev *dev)
> +{
> +	dev_dbg(&dev->pdev->dev, "enable interrupts\n");
> +	pvrdma_write_reg(dev, PVRDMA_REG_IMR, 0);
> +}
> +
> +static void pvrdma_disable_intrs(struct pvrdma_dev *dev)
> +{
> +	dev_dbg(&dev->pdev->dev, "disable interrupts\n");
> +	pvrdma_write_reg(dev, PVRDMA_REG_IMR, ~0);
> +}
> +
> +static int pvrdma_enable_msix(struct pci_dev *pdev, struct pvrdma_dev *dev)
> +{
> +	int i;
> +	int ret;
> +
> +	for (i = 0; i < PVRDMA_MAX_INTERRUPTS; i++) {
> +		dev->intr.msix_entry[i].entry = i;
> +		dev->intr.msix_entry[i].vector = i;
> +
> +		switch (i) {
> +		case 0:
> +			/* CMD ring handler */
> +			dev->intr.handler[i] = pvrdma_intr0_handler;
> +			break;
> +		case 1:
> +			/* Async event ring handler */
> +			dev->intr.handler[i] = pvrdma_intr1_handler;
> +			break;
> +		default:
> +			/* Completion queue handler */
> +			dev->intr.handler[i] = pvrdma_intrx_handler;
> +			break;
> +		}
> +	}
> +
> +	ret = pci_enable_msix(pdev, dev->intr.msix_entry,
> +			      PVRDMA_MAX_INTERRUPTS);
> +	if (!ret) {
> +		dev->intr.type = PVRDMA_INTR_TYPE_MSIX;
> +		dev->intr.size = PVRDMA_MAX_INTERRUPTS;
> +	} else if (ret > 0) {
> +		ret = pci_enable_msix(pdev, dev->intr.msix_entry, ret);
> +		if (!ret) {
> +			dev->intr.type = PVRDMA_INTR_TYPE_MSIX;
> +			dev->intr.size = ret;
> +		} else {
> +			dev->intr.size = 0;
> +		}
> +	}
> +
> +	dev_dbg(&pdev->dev, "using interrupt type %d, size %d\n",
> +		dev->intr.type, dev->intr.size);
> +
> +	return ret;
> +}
> +
> +static int pvrdma_alloc_intrs(struct pvrdma_dev *dev)
> +{
> +	int ret = 0;
> +	int i;
> +
> +	if (pci_find_capability(dev->pdev, PCI_CAP_ID_MSIX) &&
> +	    pvrdma_enable_msix(dev->pdev, dev)) {
> +		/* Try MSI */
> +		ret = pci_enable_msi(dev->pdev);
> +		if (!ret) {
> +			dev->intr.type = PVRDMA_INTR_TYPE_MSI;
> +		} else {
> +			/* Legacy INTR */
> +			dev->intr.type = PVRDMA_INTR_TYPE_INTX;
> +		}
> +	}
> +
> +	/* Request First IRQ */
> +	switch (dev->intr.type) {
> +	case PVRDMA_INTR_TYPE_INTX:
> +	case PVRDMA_INTR_TYPE_MSI:
> +		ret = request_irq(dev->pdev->irq, pvrdma_intr0_handler,
> +				  IRQF_SHARED, DRV_NAME, dev);
> +		if (ret) {
> +			dev_err(&dev->pdev->dev,
> +				"failed to request interrupt\n");
> +			goto disable_msi;
> +		}
> +		break;
> +	case PVRDMA_INTR_TYPE_MSIX:
> +		ret = request_irq(dev->intr.msix_entry[0].vector,
> +				  pvrdma_intr0_handler, 0, DRV_NAME, dev);
> +		if (ret) {
> +			dev_err(&dev->pdev->dev,
> +				"failed to request interrupt 0\n");
> +			goto disable_msi;
> +		}
> +		dev->intr.enabled[0] = 1;
> +		break;
> +	default:
> +		/* Not reached */
> +		break;
> +	}
> +
> +	/* For MSIX: request intr for each vector */
> +	if (dev->intr.size > 1) {
> +		ret = request_irq(dev->intr.msix_entry[1].vector,
> +				  pvrdma_intr1_handler, 0, DRV_NAME, dev);
> +		if (ret) {
> +			dev_err(&dev->pdev->dev,
> +				"failed to request interrupt 1\n");
> +			goto free_irq;
> +		}
> +		dev->intr.enabled[1] = 1;
> +
> +		for (i = 2; i < dev->intr.size; i++) {
> +			ret = request_irq(dev->intr.msix_entry[i].vector,
> +					  pvrdma_intrx_handler, 0,
> +					  DRV_NAME, dev);
> +			if (ret) {
> +				dev_err(&dev->pdev->dev,
> +					"failed to request interrupt %d\n", i);
> +				goto free_irq;
> +			}
> +			dev->intr.enabled[i] = 1;
> +		}
> +	}
> +
> +	return 0;
> +
> +free_irq:
> +	pvrdma_free_irq(dev);
> +disable_msi:
> +	pvrdma_disable_msi_all(dev);
> +	return ret;
> +}
> +
> +static void pvrdma_free_slots(struct pvrdma_dev *dev)
> +{
> +	struct pci_dev *pdev = dev->pdev;
> +
> +	if (dev->resp_slot)
> +		dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->resp_slot,
> +				  dev->dsr->resp_slot_dma);
> +	if (dev->cmd_slot)
> +		dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->cmd_slot,
> +				  dev->dsr->cmd_slot_dma);
> +}
> +
> +static int pvrdma_add_gid_at_index(struct pvrdma_dev *dev,
> +				   const union ib_gid *gid,
> +				   int index)
> +{
> +	int ret;
> +	union pvrdma_cmd_req req;
> +	struct pvrdma_cmd_create_bind *cmd_bind = &req.create_bind;
> +
> +	if (!dev->sgid_tbl) {
> +		dev_warn(&dev->pdev->dev, "sgid table not initialized\n");
> +		return -EINVAL;
> +	}
> +
> +	memset(cmd_bind, 0, sizeof(*cmd_bind));
> +	cmd_bind->hdr.cmd = PVRDMA_CMD_CREATE_BIND;
> +	memcpy(cmd_bind->new_gid, gid->raw, 16);
> +	cmd_bind->mtu = ib_mtu_enum_to_int(IB_MTU_1024);
> +	cmd_bind->vlan = 0xfff;
> +	cmd_bind->index = index;
> +	cmd_bind->gid_type = PVRDMA_GID_TYPE_FLAG_ROCE_V1;
> +
> +	ret = pvrdma_cmd_post(dev, &req, false, NULL);
> +	if (ret < 0) {
> +		dev_warn(&dev->pdev->dev, "could not create binding\n");
> +		return -EINVAL;
> +	}
> +	memcpy(&dev->sgid_tbl[index], gid, sizeof(*gid));
> +	return 0;
> +}
> +
> +static int pvrdma_add_gid(struct ib_device *ibdev,
> +			  u8 port_num,
> +			  unsigned int index,
> +			  const union ib_gid *gid,
> +			  const struct ib_gid_attr *attr,
> +			  void **context)
> +{
> +	struct pvrdma_dev *dev = to_vdev(ibdev);
> +
> +	return pvrdma_add_gid_at_index(dev, gid, index);
> +}
> +
> +static int pvrdma_del_gid_at_index(struct pvrdma_dev *dev, int index)
> +{
> +	int ret;
> +	union pvrdma_cmd_req req;
> +	struct pvrdma_cmd_destroy_bind *cmd_dest = &req.destroy_bind;
> +
> +	/* Update sgid table. */
> +	if (!dev->sgid_tbl) {
> +		dev_warn(&dev->pdev->dev, "sgid table not initialized\n");
> +		return -EINVAL;
> +	}
> +
> +	memset(cmd_dest, 0, sizeof(*cmd_dest));
> +	cmd_dest->hdr.cmd = PVRDMA_CMD_DESTROY_BIND;
> +	memcpy(cmd_dest->dest_gid, &dev->sgid_tbl[index], 16);
> +	cmd_dest->index = index;
> +
> +	ret = pvrdma_cmd_post(dev, &req, false, NULL);
> +	if (ret < 0) {
> +		dev_warn(&dev->pdev->dev,
> +			 "could not destroy binding\n");
> +		return ret;
> +	}
> +	memset(&dev->sgid_tbl[index], 0, 16);
> +	return 0;
> +}
> +
> +static int pvrdma_del_gid(struct ib_device *ibdev,
> +			  u8 port_num,
> +			  unsigned int index,
> +			  void **context)
> +{
> +	struct pvrdma_dev *dev = to_vdev(ibdev);
> +
> +	dev_dbg(&dev->pdev->dev, "removing gid at index %u from %s",
> +		index, dev->netdev->name);
> +
> +	return pvrdma_del_gid_at_index(dev, index);
> +}
> +
> +static void pvrdma_netdevice_event_handle(struct pvrdma_dev *dev,
> +					  unsigned long event)
> +{
> +	struct net_device *netdev;
> +
> +	netdev = dev->netdev;
> +	switch (event) {
> +	case NETDEV_REBOOT:
> +	case NETDEV_DOWN:
> +		pvrdma_dispatch_event(dev, 1, IB_EVENT_PORT_ERR);
> +		break;
> +	case NETDEV_UP:
> +		pvrdma_dispatch_event(dev, 1, IB_EVENT_PORT_ACTIVE);
> +		break;
> +	default:
> +		dev_dbg(&dev->pdev->dev, "ignore netdevice event %ld on %s\n",
> +			event, dev->ib_dev.name);
> +		break;
> +	}
> +}
> +
> +static int pvrdma_netdevice_event(struct notifier_block *this,
> +				  unsigned long event, void *ptr)
> +{
> +	struct net_device *event_netdev = netdev_notifier_info_to_dev(ptr);
> +	struct pvrdma_dev *dev;
> +
> +	mutex_lock(&pvrdma_device_list_lock);
> +	list_for_each_entry(dev, &pvrdma_device_list, device_link) {
> +		if (dev->netdev == event_netdev) {
> +			pvrdma_netdevice_event_handle(dev, event);
> +			break;
> +		}
> +	}
> +	mutex_unlock(&pvrdma_device_list_lock);
> +
> +	return NOTIFY_DONE;
> +}
> +
> +static int pvrdma_pci_probe(struct pci_dev *pdev,
> +			    const struct pci_device_id *id)
> +{
> +	struct pci_dev *pdev_net;
> +	struct pvrdma_dev *dev;
> +	int ret;
> +	unsigned long start;
> +	unsigned long len;
> +	unsigned int version;
> +	dma_addr_t slot_dma = 0;
> +
> +	dev_dbg(&pdev->dev, "initializing driver %s\n", pci_name(pdev));
> +
> +	/* Allocate zero-out device */
> +	dev = (struct pvrdma_dev *)ib_alloc_device(sizeof(*dev));
> +	if (!dev) {
> +		dev_err(&pdev->dev, "failed to allocate IB device\n");
> +		return -ENOMEM;
> +	}
> +
> +	mutex_lock(&pvrdma_device_list_lock);
> +	list_add(&dev->device_link, &pvrdma_device_list);
> +	mutex_unlock(&pvrdma_device_list_lock);
> +
> +	ret = pvrdma_init_device(dev);
> +	if (ret)
> +		goto err_free_device;
> +
> +	dev->pdev = pdev;
> +	pci_set_drvdata(pdev, dev);
> +
> +	ret = pci_enable_device(pdev);
> +	if (ret) {
> +		dev_err(&pdev->dev, "cannot enable PCI device\n");
> +		goto err_free_device;
> +	}
> +
> +	/*
> +	 * DEBUG: Checking PCI: 0x20200; _MEM; len=0x2000
> +	 */
> +	dev_dbg(&pdev->dev, "PCI resource flags BAR0 %#lx\n",
> +		pci_resource_flags(pdev, 0));
> +	dev_dbg(&pdev->dev, "PCI resource len %#llx\n",
> +		(unsigned long long)pci_resource_len(pdev, 0));
> +	dev_dbg(&pdev->dev, "PCI resource start %#llx\n",
> +		(unsigned long long)pci_resource_start(pdev, 0));
> +	dev_dbg(&pdev->dev, "PCI resource flags BAR1 %#lx\n",
> +		pci_resource_flags(pdev, 1));
> +	dev_dbg(&pdev->dev, "PCI resource len %#llx\n",
> +		(unsigned long long)pci_resource_len(pdev, 1));
> +	dev_dbg(&pdev->dev, "PCI resource start %#llx\n",
> +		(unsigned long long)pci_resource_start(pdev, 1));
> +
> +	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
> +	    !(pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
> +		dev_err(&pdev->dev, "PCI BAR region not MMIO\n");
> +		ret = -ENOMEM;
> +		goto err_free_device;
> +	}
> +
> +	ret = pci_request_regions(pdev, DRV_NAME);
> +	if (ret) {
> +		dev_err(&pdev->dev, "cannot request PCI resources\n");
> +		goto err_disable_pdev;
> +	}
> +
> +	/* Enable 64-Bit DMA */
> +	if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
> +		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
> +		if (ret != 0) {
> +			dev_err(&pdev->dev,
> +				"pci_set_consistent_dma_mask failed\n");
> +			goto err_free_resource;
> +		}
> +	} else {
> +		ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
> +		if (ret != 0) {
> +			dev_err(&pdev->dev,
> +				"pci_set_dma_mask failed\n");
> +			goto err_free_resource;
> +		}
> +	}
> +
> +	pci_set_master(pdev);
> +
> +	/* Map register space */
> +	start = pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_REG);
> +	len = pci_resource_len(dev->pdev, PVRDMA_PCI_RESOURCE_REG);
> +	dev->regs = ioremap(start, len);
> +	if (!dev->regs) {
> +		dev_err(&pdev->dev, "register mapping failed\n");
> +		ret = -ENOMEM;
> +		goto err_free_resource;
> +	}
> +
> +	/* Setup per-device UAR. */
> +	dev->driver_uar.index = 0;
> +	dev->driver_uar.pfn =
> +		pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_UAR) >>
> +		PAGE_SHIFT;
> +	dev->driver_uar.map =
> +		ioremap(dev->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
> +	if (!dev->driver_uar.map) {
> +		dev_err(&pdev->dev, "failed to remap UAR pages\n");
> +		ret = -ENOMEM;
> +		goto err_unmap_regs;
> +	}
> +
> +	version = pvrdma_read_reg(dev, PVRDMA_REG_VERSION);
> +	dev_info(&pdev->dev, "device version %d, driver version %d\n",
> +		 version, PVRDMA_VERSION);
> +	if (version < PVRDMA_VERSION) {
> +		dev_err(&pdev->dev, "incompatible device version\n");
> +		goto err_uar_unmap;
> +	}
> +
> +	dev->dsr = dma_alloc_coherent(&pdev->dev, sizeof(*dev->dsr),
> +				      &dev->dsrbase, GFP_KERNEL);
> +	if (!dev->dsr) {
> +		dev_err(&pdev->dev, "failed to allocate shared region\n");
> +		ret = -ENOMEM;
> +		goto err_uar_unmap;
> +	}
> +
> +	/* Setup the shared region */
> +	memset(dev->dsr, 0, sizeof(*dev->dsr));
> +	dev->dsr->driver_version = PVRDMA_VERSION;
> +	dev->dsr->gos_info.gos_bits = sizeof(void *) == 4 ?
> +		PVRDMA_GOS_BITS_32 :
> +		PVRDMA_GOS_BITS_64;
> +	dev->dsr->gos_info.gos_type = PVRDMA_GOS_TYPE_LINUX;
> +	dev->dsr->gos_info.gos_ver = 1;
> +	dev->dsr->uar_pfn = dev->driver_uar.pfn;
> +
> +	/* Command slot. */
> +	dev->cmd_slot = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
> +					   &slot_dma, GFP_KERNEL);
> +	if (!dev->cmd_slot) {
> +		ret = -ENOMEM;
> +		goto err_free_dsr;
> +	}
> +
> +	dev->dsr->cmd_slot_dma = (u64)slot_dma;
> +
> +	/* Response slot. */
> +	dev->resp_slot = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
> +					    &slot_dma, GFP_KERNEL);
> +	if (!dev->resp_slot) {
> +		ret = -ENOMEM;
> +		goto err_free_slots;
> +	}
> +
> +	dev->dsr->resp_slot_dma = (u64)slot_dma;
> +
> +	/* Async event ring */
> +	dev->dsr->async_ring_pages.num_pages = 4;
> +	ret = pvrdma_page_dir_init(dev, &dev->async_pdir,
> +				   dev->dsr->async_ring_pages.num_pages, true);
> +	if (ret)
> +		goto err_free_slots;
> +	dev->async_ring_state = dev->async_pdir.pages[0];
> +	dev->dsr->async_ring_pages.pdir_dma = dev->async_pdir.dir_dma;
> +
> +	/* CQ notification ring */
> +	dev->dsr->cq_ring_pages.num_pages = 4;
> +	ret = pvrdma_page_dir_init(dev, &dev->cq_pdir,
> +				   dev->dsr->cq_ring_pages.num_pages, true);
> +	if (ret)
> +		goto err_free_async_ring;
> +	dev->cq_ring_state = dev->cq_pdir.pages[0];
> +	dev->dsr->cq_ring_pages.pdir_dma = dev->cq_pdir.dir_dma;
> +
> +	/*
> +	 * Write the PA of the shared region to the device. The writes must be
> +	 * ordered such that the high bits are written last. When the writes
> +	 * complete, the device will have filled out the capabilities.
> +	 */
> +
> +	pvrdma_write_reg(dev, PVRDMA_REG_DSRLOW, (u32)dev->dsrbase);
> +	pvrdma_write_reg(dev, PVRDMA_REG_DSRHIGH,
> +			 (u32)((u64)(dev->dsrbase) >> 32));
> +
> +	/* Make sure the write is complete before reading status. */
> +	mb();
> +
> +	/* Currently, the driver only supports RoCE mode. */
> +	if (dev->dsr->caps.mode != PVRDMA_DEVICE_MODE_ROCE) {
> +		dev_err(&pdev->dev, "unsupported transport %d\n",
> +			dev->dsr->caps.mode);
> +		ret = -EINVAL;
> +		goto err_free_cq_ring;
> +	}
> +
> +	/* Currently, the driver only supports RoCE V1. */
> +	if (!(dev->dsr->caps.gid_types & PVRDMA_GID_TYPE_FLAG_ROCE_V1)) {
> +		dev_err(&pdev->dev, "driver needs RoCE v1 support\n");
> +		ret = -EINVAL;
> +		goto err_free_cq_ring;
> +	}
> +
> +	/* Paired vmxnet3 will have same bus, slot. But func will be 0 */
> +	pdev_net = pci_get_slot(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 0));
> +	if (!pdev_net) {
> +		dev_err(&pdev->dev, "failed to find paired net device\n");
> +		ret = -ENODEV;
> +		goto err_free_cq_ring;
> +	}
> +
> +	if (pdev_net->vendor != PCI_VENDOR_ID_VMWARE ||
> +	    pdev_net->device != PCI_DEVICE_ID_VMWARE_VMXNET3) {
> +		dev_err(&pdev->dev, "failed to find paired vmxnet3 device\n");
> +		pci_dev_put(pdev_net);
> +		ret = -ENODEV;
> +		goto err_free_cq_ring;
> +	}
> +
> +	dev->netdev = pci_get_drvdata(pdev_net);
> +	pci_dev_put(pdev_net);
> +	if (!dev->netdev) {
> +		dev_err(&pdev->dev, "failed to get vmxnet3 device\n");
> +		ret = -ENODEV;
> +		goto err_free_cq_ring;
> +	}
> +
> +	dev_info(&pdev->dev, "paired device to %s\n", dev->netdev->name);
> +
> +	dev->nb_netdev.notifier_call = pvrdma_netdevice_event;
> +	ret = register_netdevice_notifier(&dev->nb_netdev);
> +	if (ret) {
> +		dev_err(&pdev->dev, "failed to register netdevice events\n");
> +		goto err_free_cq_ring;
> +	}
> +
> +	/* Interrupt setup */
> +	ret = pvrdma_alloc_intrs(dev);
> +	if (ret) {
> +		dev_err(&pdev->dev, "failed to allocate interrupts\n");
> +		ret = -ENOMEM;
> +		goto err_netdevice;
> +	}
> +
> +	/* Allocate UAR table. */
> +	ret = pvrdma_uar_table_init(dev);
> +	if (ret) {
> +		dev_err(&pdev->dev, "failed to allocate UAR table\n");
> +		ret = -ENOMEM;
> +		goto err_free_intrs;
> +	}
> +
> +	/* Allocate GID table */
> +	dev->sgid_tbl = kcalloc(dev->dsr->caps.gid_tbl_len,
> +				sizeof(union ib_gid), GFP_KERNEL);
> +	if (!dev->sgid_tbl) {
> +		ret = -ENOMEM;
> +		goto err_free_uar_table;
> +	}
> +	dev_dbg(&pdev->dev, "gid table len %d\n", dev->dsr->caps.gid_tbl_len);
> +
> +	pvrdma_enable_intrs(dev);
> +
> +	/* Activate pvrdma device */
> +	pvrdma_write_reg(dev, PVRDMA_REG_CTL, PVRDMA_DEVICE_CTL_ACTIVATE);
> +
> +	/* Make sure the write is complete before reading status. */
> +	mb();
> +
> +	/* Check if device was successfully activated */
> +	ret = pvrdma_read_reg(dev, PVRDMA_REG_ERR);
> +	if (ret != 0) {
> +		dev_err(&pdev->dev, "failed to activate device\n");
> +		ret = -EINVAL;
> +		goto err_disable_intr;
> +	}
> +
> +	/* Register IB device */
> +	ret = pvrdma_register_device(dev);
> +	if (ret) {
> +		dev_err(&pdev->dev, "failed to register IB device\n");
> +		goto err_disable_intr;
> +	}
> +
> +	dev_info(&pdev->dev, "attached to device\n");
> +	return 0;
> +
> +err_disable_intr:
> +	pvrdma_disable_intrs(dev);
> +	kfree(dev->sgid_tbl);
> +err_free_uar_table:
> +	pvrdma_uar_table_cleanup(dev);
> +err_free_intrs:
> +	pvrdma_free_irq(dev);
> +	pvrdma_disable_msi_all(dev);
> +err_netdevice:
> +	unregister_netdevice_notifier(&dev->nb_netdev);
> +err_free_cq_ring:
> +	pvrdma_page_dir_cleanup(dev, &dev->cq_pdir);
> +err_free_async_ring:
> +	pvrdma_page_dir_cleanup(dev, &dev->async_pdir);
> +err_free_slots:
> +	pvrdma_free_slots(dev);
> +err_free_dsr:
> +	dma_free_coherent(&pdev->dev, sizeof(*dev->dsr), dev->dsr,
> +			  dev->dsrbase);
> +err_uar_unmap:
> +	iounmap(dev->driver_uar.map);
> +err_unmap_regs:
> +	iounmap(dev->regs);
> +err_free_resource:
> +	pci_release_regions(pdev);
> +err_disable_pdev:
> +	pci_disable_device(pdev);
> +	pci_set_drvdata(pdev, NULL);
> +err_free_device:
> +	mutex_lock(&pvrdma_device_list_lock);
> +	list_del(&dev->device_link);
> +	mutex_unlock(&pvrdma_device_list_lock);
> +	ib_dealloc_device(&dev->ib_dev);
> +	return ret;
> +}
> +
> +static void pvrdma_pci_remove(struct pci_dev *pdev)
> +{
> +	struct pvrdma_dev *dev = pci_get_drvdata(pdev);
> +
> +	if (!dev)
> +		return;
> +
> +	dev_info(&pdev->dev, "detaching from device\n");
> +
> +	mutex_lock(&pvrdma_device_list_lock);
> +	list_del(&dev->device_link);
> +	mutex_unlock(&pvrdma_device_list_lock);
> +
> +	/* Unregister ib device */
> +	ib_unregister_device(&dev->ib_dev);
> +
> +	pvrdma_disable_intrs(dev);
> +	pvrdma_free_irq(dev);
> +	pvrdma_disable_msi_all(dev);
> +
> +	unregister_netdevice_notifier(&dev->nb_netdev);
> +
> +	/* Deactivate pvrdma device */
> +	pvrdma_write_reg(dev, PVRDMA_REG_CTL, PVRDMA_DEVICE_CTL_RESET);
> +	pvrdma_page_dir_cleanup(dev, &dev->cq_pdir);
> +	pvrdma_page_dir_cleanup(dev, &dev->async_pdir);
> +	pvrdma_free_slots(dev);
> +
> +	iounmap(dev->regs);
> +	kfree(dev->sgid_tbl);
> +	kfree(dev->cq_tbl);
> +	kfree(dev->qp_tbl);
> +	pvrdma_uar_table_cleanup(dev);
> +	iounmap(dev->driver_uar.map);
> +
> +	ib_dealloc_device(&dev->ib_dev);
> +
> +	/* Free pci resources */
> +	pci_release_regions(pdev);
> +	pci_disable_device(pdev);
> +	pci_set_drvdata(pdev, NULL);
> +}
> +
> +static struct pci_device_id pvrdma_pci_table[] = {
> +	{ PCI_DEVICE(PCI_VENDOR_ID_VMWARE, PCI_DEVICE_ID_VMWARE_PVRDMA), },
> +	{ 0 },
> +};
> +
> +MODULE_DEVICE_TABLE(pci, pvrdma_pci_table);
> +
> +static struct pci_driver pvrdma_driver = {
> +	.name		= DRV_NAME,
> +	.id_table	= pvrdma_pci_table,
> +	.probe		= pvrdma_pci_probe,
> +	.remove		= pvrdma_pci_remove,
> +};
> +
> +static int __init pvrdma_init(void)
> +{
> +	return pci_register_driver(&pvrdma_driver);
> +}
> +
> +static void __exit pvrdma_cleanup(void)
> +{
> +	pci_unregister_driver(&pvrdma_driver);
> +}
> +
> +module_init(pvrdma_init);
> +module_exit(pvrdma_cleanup);
> +
> +MODULE_AUTHOR("VMware, Inc");
> +MODULE_DESCRIPTION("VMware Virtual RDMA/InfiniBand driver");
> +MODULE_VERSION(DRV_VERSION);
> +MODULE_LICENSE("Dual BSD/GPL");
> -- 
> 2.7.4
> 
> --
> 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
--
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 04/15] IB/pvrdma: Add the paravirtual RDMA device specification

[Yuval Shaia]

Two minor comments.
Besides that - Reviewed-by: Yuval Shaia <yuval.shaia-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org>

Yuval

On Tue, Jul 12, 2016 at 12:36:34PM -0700, Adit Ranadive wrote:
> This patch describes the main specification of the underlying virtual RDMA
> device. The pvrdma_dev_api header file defines the Verbs commands and
> their parameters that can be issued to the device backend.
> 
> Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> ---
>  drivers/infiniband/hw/pvrdma/pvrdma_defs.h    | 300 ++++++++++++++++++++++
>  drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h | 342 ++++++++++++++++++++++++++
>  2 files changed, 642 insertions(+)
>  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_defs.h
>  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h
> 
> diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_defs.h b/drivers/infiniband/hw/pvrdma/pvrdma_defs.h
> new file mode 100644
> index 0000000..225cba4
> --- /dev/null
> +++ b/drivers/infiniband/hw/pvrdma/pvrdma_defs.h
> @@ -0,0 +1,300 @@
> +/*
> + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of EITHER the GNU General Public License
> + * version 2 as published by the Free Software Foundation or the BSD
> + * 2-Clause License. This program is distributed in the hope that it
> + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> + * See the GNU General Public License version 2 for more details at
> + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program available in the file COPYING in the main
> + * directory of this source tree.
> + *
> + * The BSD 2-Clause License
> + *
> + *     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.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> + * OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#ifndef __PVRDMA_DEFS_H__
> +#define __PVRDMA_DEFS_H__
> +
> +#include <linux/types.h>
> +#include "pvrdma_ib_verbs.h"
> +#include "pvrdma_uapi.h"
> +
> +/*
> + * Masks and accessors for page directory, which is a two-level lookup:
> + * page directory -> page table -> page. Only one directory for now, but we
> + * could expand that easily. 9 bits for tables, 9 bits for pages, gives one
> + * gigabyte for memory regions and so forth.
> + */
> +
> +#define PVRDMA_PAGE_DIR_DIR(x)		(((x) >> 18) & 0x1)
> +#define PVRDMA_PAGE_DIR_TABLE(x)	(((x) >> 9) & 0x1ff)

Suggesting to add #define for the above 9 and 18, ex PVRDMA_PAGE_DIR_SHIFT
or something like that.

> +#define PVRDMA_PAGE_DIR_PAGE(x)		((x) & 0x1ff)
> +#define PVRDMA_PAGE_DIR_MAX_PAGES	(1 * 512 * 512)
> +#define PVRDMA_MAX_FAST_REG_PAGES	128
> +
> +/*
> + * Max MSI-X vectors.
> + */
> +
> +#define PVRDMA_MAX_INTERRUPTS	3
> +
> +/* Register offsets within PCI resource on BAR1. */
> +#define PVRDMA_REG_VERSION	0x00	/* R: Version of device. */
> +#define PVRDMA_REG_DSRLOW	0x04	/* W: Device shared region low PA. */
> +#define PVRDMA_REG_DSRHIGH	0x08	/* W: Device shared region high PA. */
> +#define PVRDMA_REG_CTL		0x0c	/* W: PVRDMA_DEVICE_CTL */
> +#define PVRDMA_REG_REQUEST	0x10	/* W: Indicate device request. */
> +#define PVRDMA_REG_ERR		0x14	/* R: Device error. */
> +#define PVRDMA_REG_ICR		0x18	/* R: Interrupt cause. */
> +#define PVRDMA_REG_IMR		0x1c	/* R/W: Interrupt mask. */
> +#define PVRDMA_REG_MACL		0x20	/* R/W: MAC address low. */
> +#define PVRDMA_REG_MACH		0x24	/* R/W: MAC address high. */
> +
> +/* Object flags. */
> +#define PVRDMA_CQ_FLAG_ARMED_SOL	BIT(0)	/* Armed for solicited-only. */
> +#define PVRDMA_CQ_FLAG_ARMED		BIT(1)	/* Armed. */
> +#define PVRDMA_MR_FLAG_DMA		BIT(0)	/* DMA region. */
> +#define PVRDMA_MR_FLAG_FRMR		BIT(1)	/* Fast reg memory region. */
> +
> +/*
> + * Atomic operation capability (masked versions are extended atomic
> + * operations.
> + */
> +
> +#define PVRDMA_ATOMIC_OP_COMP_SWAP	BIT(0) /* Compare and swap. */
> +#define PVRDMA_ATOMIC_OP_FETCH_ADD	BIT(1) /* Fetch and add. */
> +#define PVRDMA_ATOMIC_OP_MASK_COMP_SWAP	BIT(2) /* Masked compare and swap. */
> +#define PVRDMA_ATOMIC_OP_MASK_FETCH_ADD	BIT(3) /* Masked fetch and add. */
> +
> +/*
> + * Base Memory Management Extension flags to support Fast Reg Memory Regions
> + * and Fast Reg Work Requests. Each flag represents a verb operation and we
> + * must support all of them to qualify for the BMME device cap.
> + */
> +
> +#define PVRDMA_BMME_FLAG_LOCAL_INV	BIT(0) /* Local Invalidate. */
> +#define PVRDMA_BMME_FLAG_REMOTE_INV	BIT(1) /* Remote Invalidate. */
> +#define PVRDMA_BMME_FLAG_FAST_REG_WR	BIT(2) /* Fast Reg Work Request. */
> +
> +/*
> + * GID types. The interpretation of the gid_types bit field in the device
> + * capabilities will depend on the device mode. For now, the device only
> + * supports RoCE as mode, so only the different GID types for RoCE are
> + * defined.
> + */
> +
> +#define PVRDMA_GID_TYPE_FLAG_ROCE_V1 BIT(0)
> +#define PVRDMA_GID_TYPE_FLAG_ROCE_V2 BIT(1)
> +
> +enum pvrdma_pci_resource {
> +	PVRDMA_PCI_RESOURCE_MSIX = 0,	/* BAR0: MSI-X, MMIO. */
> +	PVRDMA_PCI_RESOURCE_REG,	/* BAR1: Registers, MMIO. */
> +	PVRDMA_PCI_RESOURCE_UAR,	/* BAR2: UAR pages, MMIO, 64-bit. */
> +	PVRDMA_PCI_RESOURCE_LAST,	/* Last. */
> +};
> +
> +enum pvrdma_device_ctl {
> +	PVRDMA_DEVICE_CTL_ACTIVATE = 0,	/* Activate device. */
> +	PVRDMA_DEVICE_CTL_QUIESCE,	/* Quiesce device. */
> +	PVRDMA_DEVICE_CTL_RESET,	/* Reset device. */
> +};
> +
> +enum pvrdma_intr_vector {
> +	PVRDMA_INTR_VECTOR_RESPONSE	= 0,	/* Command response. */
> +	PVRDMA_INTR_VECTOR_ASYNC	= 1,	/* Async events. */
> +	PVRDMA_INTR_VECTOR_CQ		= 2,	/* CQ notification. */
> +	/* Additional CQ notification vectors. */
> +};
> +
> +enum pvrdma_intr_cause {
> +	PVRDMA_INTR_CAUSE_RESPONSE	= (1 << PVRDMA_INTR_VECTOR_RESPONSE),
> +	PVRDMA_INTR_CAUSE_ASYNC		= (1 << PVRDMA_INTR_VECTOR_ASYNC),
> +	PVRDMA_INTR_CAUSE_CQ		= (1 << PVRDMA_INTR_VECTOR_CQ),
> +};
> +
> +enum pvrdma_intr_type {
> +	PVRDMA_INTR_TYPE_INTX = 0,	/* Legacy. */
> +	PVRDMA_INTR_TYPE_MSI,		/* MSI. */
> +	PVRDMA_INTR_TYPE_MSIX,		/* MSI-X. */
> +};
> +
> +enum pvrdma_gos_bits {
> +	PVRDMA_GOS_BITS_UNK = 0,	/* Unknown. */
> +	PVRDMA_GOS_BITS_32,		/* 32-bit. */
> +	PVRDMA_GOS_BITS_64,		/* 64-bit. */
> +};
> +
> +enum pvrdma_gos_type {
> +	PVRDMA_GOS_TYPE_UNK = 0,	/* Unknown. */
> +	PVRDMA_GOS_TYPE_LINUX,		/* Linux. */
> +};
> +
> +enum pvrdma_device_mode {
> +	PVRDMA_DEVICE_MODE_ROCE = 0,	/* RoCE. */
> +	PVRDMA_DEVICE_MODE_IWARP,	/* iWarp. */
> +	PVRDMA_DEVICE_MODE_IB		/* InfiniBand. */
> +};
> +
> +struct pvrdma_gos_info {
> +	__u32 gos_bits:2;	/* W: PVRDMA_GOS_BITS_ */
> +	__u32 gos_type:4;	/* W: PVRDMA_GOS_TYPE_ */
> +	__u32 gos_ver:16;	/* W: Guest OS version. */
> +	__u32 gos_misc:10;	/* W: Other. */
> +	__u32 pad;		/*    Pad to 8-byte alignment. */

Strip the spaces in comment.

> +};
> +
> +struct pvrdma_device_caps {
> +	__u64 fw_ver;				/* R: Query device. */
> +	__be64 node_guid;
> +	__be64 sys_image_guid;
> +	__u64 max_mr_size;
> +	__u64 page_size_cap;
> +	__u64 atomic_arg_sizes;			/* EXP verbs. */
> +	__u32 exp_comp_mask;			/* EXP verbs. */
> +	__u32 device_cap_flags2;		/* EXP verbs. */
> +	__u32 max_fa_bit_boundary;		/* EXP verbs. */
> +	__u32 log_max_atomic_inline_arg;	/* EXP verbs. */
> +	__u32 vendor_id;
> +	__u32 vendor_part_id;
> +	__u32 hw_ver;
> +	__u32 max_qp;
> +	__u32 max_qp_wr;
> +	__u32 device_cap_flags;
> +	__u32 max_sge;
> +	__u32 max_sge_rd;
> +	__u32 max_cq;
> +	__u32 max_cqe;
> +	__u32 max_mr;
> +	__u32 max_pd;
> +	__u32 max_qp_rd_atom;
> +	__u32 max_ee_rd_atom;
> +	__u32 max_res_rd_atom;
> +	__u32 max_qp_init_rd_atom;
> +	__u32 max_ee_init_rd_atom;
> +	__u32 max_ee;
> +	__u32 max_rdd;
> +	__u32 max_mw;
> +	__u32 max_raw_ipv6_qp;
> +	__u32 max_raw_ethy_qp;
> +	__u32 max_mcast_grp;
> +	__u32 max_mcast_qp_attach;
> +	__u32 max_total_mcast_qp_attach;
> +	__u32 max_ah;
> +	__u32 max_fmr;
> +	__u32 max_map_per_fmr;
> +	__u32 max_srq;
> +	__u32 max_srq_wr;
> +	__u32 max_srq_sge;
> +	__u32 max_uar;
> +	__u32 gid_tbl_len;
> +	__u16 max_pkeys;
> +	__u8  local_ca_ack_delay;
> +	__u8  phys_port_cnt;
> +	__u8  mode;		/* PVRDMA_DEVICE_MODE_ */
> +	__u8  atomic_ops;	/* PVRDMA_ATOMIC_OP_* bits */
> +	__u8  bmme_flags;	/* Memory Mgmt Extensions for FRMR/FRWR */
> +	__u8  gid_types;	/* PVRDMA_GID_TYPE_FLAG_ */
> +	__u8  reserved[4];
> +};
> +
> +struct pvrdma_ring_page_info {
> +	__u32 num_pages;			/* Num pages incl. header. */
> +	__u32 reserved;				/* Reserved. */
> +	__u64 pdir_dma;				/* Page directory PA. */
> +};
> +
> +#pragma pack(push, 1)
> +
> +struct pvrdma_device_shared_region {
> +	__u32 driver_version;			/* W: Driver version. */
> +	__u32 pad;				/*    Pad to 8-byte align. */
> +	struct pvrdma_gos_info gos_info;	/* W: Guest OS information. */
> +	__u64 cmd_slot_dma;			/* W: Command slot address. */
> +	__u64 resp_slot_dma;			/* W: Response slot address. */
> +	struct pvrdma_ring_page_info async_ring_pages;
> +						/* W: Async ring page info. */
> +	struct pvrdma_ring_page_info cq_ring_pages;
> +						/* W: CQ ring page info. */
> +	__u32 uar_pfn;				/* W: UAR pageframe. */
> +	__u32 pad2;				/*    Pad to 8-byte align. */
> +	struct pvrdma_device_caps caps;		/* R: Device capabilities. */
> +};
> +
> +#pragma pack(pop)
> +
> +
> +/* Event types. Currently a 1:1 mapping with enum ib_event. */
> +enum pvrdma_eqe_type {
> +	PVRDMA_EVENT_CQ_ERR = 0,
> +	PVRDMA_EVENT_QP_FATAL,
> +	PVRDMA_EVENT_QP_REQ_ERR,
> +	PVRDMA_EVENT_QP_ACCESS_ERR,
> +	PVRDMA_EVENT_COMM_EST,
> +	PVRDMA_EVENT_SQ_DRAINED,
> +	PVRDMA_EVENT_PATH_MIG,
> +	PVRDMA_EVENT_PATH_MIG_ERR,
> +	PVRDMA_EVENT_DEVICE_FATAL,
> +	PVRDMA_EVENT_PORT_ACTIVE,
> +	PVRDMA_EVENT_PORT_ERR,
> +	PVRDMA_EVENT_LID_CHANGE,
> +	PVRDMA_EVENT_PKEY_CHANGE,
> +	PVRDMA_EVENT_SM_CHANGE,
> +	PVRDMA_EVENT_SRQ_ERR,
> +	PVRDMA_EVENT_SRQ_LIMIT_REACHED,
> +	PVRDMA_EVENT_QP_LAST_WQE_REACHED,
> +	PVRDMA_EVENT_CLIENT_REREGISTER,
> +	PVRDMA_EVENT_GID_CHANGE,
> +};
> +
> +/* Event queue element. */
> +struct pvrdma_eqe {
> +	__u32 type;	/* Event type. */
> +	__u32 info;	/* Handle, other. */
> +};
> +
> +/* CQ notification queue element. */
> +struct pvrdma_cqne {
> +	__u32 info;	/* Handle */
> +};
> +
> +static inline void pvrdma_init_cqe(struct pvrdma_cqe *cqe, __u64 wr_id,
> +				   __u64 qp)
> +{
> +	memset(cqe, 0, sizeof(*cqe));
> +	cqe->status = PVRDMA_WC_GENERAL_ERR;
> +	cqe->wr_id = wr_id;
> +	cqe->qp = qp;
> +}
> +
> +#endif /* __PVRDMA_DEFS_H__ */
> diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h b/drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h
> new file mode 100644
> index 0000000..4b4d8e5
> --- /dev/null
> +++ b/drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h
> @@ -0,0 +1,342 @@
> +/*
> + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of EITHER the GNU General Public License
> + * version 2 as published by the Free Software Foundation or the BSD
> + * 2-Clause License. This program is distributed in the hope that it
> + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> + * See the GNU General Public License version 2 for more details at
> + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program available in the file COPYING in the main
> + * directory of this source tree.
> + *
> + * The BSD 2-Clause License
> + *
> + *     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.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> + * OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#ifndef __PVRDMA_DEV_API_H__
> +#define __PVRDMA_DEV_API_H__
> +
> +#include <linux/types.h>
> +#include "pvrdma_ib_verbs.h"
> +
> +enum {
> +	PVRDMA_CMD_FIRST = 0,
> +	PVRDMA_CMD_QUERY_PORT = PVRDMA_CMD_FIRST,
> +	PVRDMA_CMD_QUERY_PKEY,
> +	PVRDMA_CMD_CREATE_PD,
> +	PVRDMA_CMD_DESTROY_PD,
> +	PVRDMA_CMD_CREATE_MR,
> +	PVRDMA_CMD_DESTROY_MR,
> +	PVRDMA_CMD_CREATE_CQ,
> +	PVRDMA_CMD_RESIZE_CQ,
> +	PVRDMA_CMD_DESTROY_CQ,
> +	PVRDMA_CMD_CREATE_QP,
> +	PVRDMA_CMD_MODIFY_QP,
> +	PVRDMA_CMD_QUERY_QP,
> +	PVRDMA_CMD_DESTROY_QP,
> +	PVRDMA_CMD_CREATE_UC,
> +	PVRDMA_CMD_DESTROY_UC,
> +	PVRDMA_CMD_CREATE_BIND,
> +	PVRDMA_CMD_DESTROY_BIND,
> +	PVRDMA_CMD_MAX,
> +};
> +
> +enum {
> +	PVRDMA_CMD_FIRST_RESP = (1 << 31),
> +	PVRDMA_CMD_QUERY_PORT_RESP = PVRDMA_CMD_FIRST_RESP,
> +	PVRDMA_CMD_QUERY_PKEY_RESP,
> +	PVRDMA_CMD_CREATE_PD_RESP,
> +	PVRDMA_CMD_DESTROY_PD_RESP_NOOP,
> +	PVRDMA_CMD_CREATE_MR_RESP,
> +	PVRDMA_CMD_DESTROY_MR_RESP_NOOP,
> +	PVRDMA_CMD_CREATE_CQ_RESP,
> +	PVRDMA_CMD_RESIZE_CQ_RESP,
> +	PVRDMA_CMD_DESTROY_CQ_RESP_NOOP,
> +	PVRDMA_CMD_CREATE_QP_RESP,
> +	PVRDMA_CMD_MODIFY_QP_RESP,
> +	PVRDMA_CMD_QUERY_QP_RESP,
> +	PVRDMA_CMD_DESTROY_QP_RESP,
> +	PVRDMA_CMD_CREATE_UC_RESP,
> +	PVRDMA_CMD_DESTROY_UC_RESP_NOOP,
> +	PVRDMA_CMD_CREATE_BIND_RESP_NOOP,
> +	PVRDMA_CMD_DESTROY_BIND_RESP_NOOP,
> +	PVRDMA_CMD_MAX_RESP,
> +};
> +
> +struct pvrdma_cmd_hdr {
> +	__u64 response;	/* Key for response lookup. */
> +	__u32 cmd;	/* PVRDMA_CMD_ */
> +	__u32 reserved;	/* Reserved. */
> +};
> +
> +struct pvrdma_cmd_resp_hdr {
> +	__u64 response;		/* From cmd hdr. */
> +	__u32 ack;		/* PVRDMA_CMD_XXX_RESP */
> +	__u8 err;		/* Error. */
> +	__u8 reserved[3];	/* Reserved. */
> +};
> +
> +struct pvrdma_cmd_query_port {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u8 port_num;
> +	__u8 reserved[7];
> +};
> +
> +struct pvrdma_cmd_query_port_resp {
> +	struct pvrdma_cmd_resp_hdr hdr;
> +	struct pvrdma_port_attr attrs;
> +};
> +
> +struct pvrdma_cmd_query_pkey {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u8 port_num;
> +	__u8 index;
> +	__u8 reserved[6];
> +};
> +
> +struct pvrdma_cmd_query_pkey_resp {
> +	struct pvrdma_cmd_resp_hdr hdr;
> +	__u16 pkey;
> +	__u8 reserved[6];
> +};
> +
> +struct pvrdma_cmd_create_uc {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u32 pfn; /* UAR page frame number */
> +	__u8 reserved[4];
> +};
> +
> +struct pvrdma_cmd_create_uc_resp {
> +	struct pvrdma_cmd_resp_hdr hdr;
> +	__u32 ctx_handle;
> +	__u8 reserved[4];
> +};
> +
> +struct pvrdma_cmd_destroy_uc {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u32 ctx_handle;
> +	__u8 reserved[4];
> +};
> +
> +struct pvrdma_cmd_create_pd {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u32 ctx_handle;
> +	__u8 reserved[4];
> +};
> +
> +struct pvrdma_cmd_create_pd_resp {
> +	struct pvrdma_cmd_resp_hdr hdr;
> +	__u32 pd_handle;
> +	__u8 reserved[4];
> +};
> +
> +struct pvrdma_cmd_destroy_pd {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u32 pd_handle;
> +	__u8 reserved[4];
> +};
> +
> +struct pvrdma_cmd_create_mr {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u64 start;
> +	__u64 length;
> +	__u64 pdir_dma;
> +	__u32 pd_handle;
> +	__u32 access_flags;
> +	__u32 flags;
> +	__u32 nchunks;
> +};
> +
> +struct pvrdma_cmd_create_mr_resp {
> +	struct pvrdma_cmd_resp_hdr hdr;
> +	__u32 mr_handle;
> +	__u32 lkey;
> +	__u32 rkey;
> +	__u8 reserved[4];
> +};
> +
> +struct pvrdma_cmd_destroy_mr {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u32 mr_handle;
> +	__u8 reserved[4];
> +};
> +
> +struct pvrdma_cmd_create_cq {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u64 pdir_dma;
> +	__u32 ctx_handle;
> +	__u32 cqe;
> +	__u32 nchunks;
> +	__u8 reserved[4];
> +};
> +
> +struct pvrdma_cmd_create_cq_resp {
> +	struct pvrdma_cmd_resp_hdr hdr;
> +	__u32 cq_handle;
> +	__u32 cqe;
> +};
> +
> +struct pvrdma_cmd_resize_cq {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u32 cq_handle;
> +	__u32 cqe;
> +};
> +
> +struct pvrdma_cmd_resize_cq_resp {
> +	struct pvrdma_cmd_resp_hdr hdr;
> +	__u32 cqe;
> +	__u8 reserved[4];
> +};
> +
> +struct pvrdma_cmd_destroy_cq {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u32 cq_handle;
> +	__u8 reserved[4];
> +};
> +
> +struct pvrdma_cmd_create_qp {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u64 pdir_dma;
> +	__u32 pd_handle;
> +	__u32 send_cq_handle;
> +	__u32 recv_cq_handle;
> +	__u32 srq_handle;
> +	__u32 max_send_wr;
> +	__u32 max_recv_wr;
> +	__u32 max_send_sge;
> +	__u32 max_recv_sge;
> +	__u32 max_inline_data;
> +	__u32 lkey;
> +	__u32 access_flags;
> +	__u16 total_chunks;
> +	__u16 send_chunks;
> +	__u16 max_atomic_arg;
> +	__u8 sq_sig_all;
> +	__u8 qp_type;
> +	__u8 is_srq;
> +	__u8 reserved[3];
> +};
> +
> +struct pvrdma_cmd_create_qp_resp {
> +	struct pvrdma_cmd_resp_hdr hdr;
> +	__u32 qpn;
> +	__u32 max_send_wr;
> +	__u32 max_recv_wr;
> +	__u32 max_send_sge;
> +	__u32 max_recv_sge;
> +	__u32 max_inline_data;
> +};
> +
> +struct pvrdma_cmd_modify_qp {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u32 qp_handle;
> +	__u32 attr_mask;
> +	struct pvrdma_qp_attr attrs;
> +};
> +
> +struct pvrdma_cmd_query_qp {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u32 qp_handle;
> +	__u32 attr_mask;
> +};
> +
> +struct pvrdma_cmd_query_qp_resp {
> +	struct pvrdma_cmd_resp_hdr hdr;
> +	struct pvrdma_qp_attr attrs;
> +};
> +
> +struct pvrdma_cmd_destroy_qp {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u32 qp_handle;
> +	__u8 reserved[4];
> +};
> +
> +struct pvrdma_cmd_destroy_qp_resp {
> +	struct pvrdma_cmd_resp_hdr hdr;
> +	__u32 events_reported;
> +	__u8 reserved[4];
> +};
> +
> +struct pvrdma_cmd_create_bind {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u32 mtu;
> +	__u32 vlan;
> +	__u32 index;
> +	__u8 new_gid[16];
> +	__u8 gid_type;
> +	__u8 reserved[3];
> +};
> +
> +struct pvrdma_cmd_destroy_bind {
> +	struct pvrdma_cmd_hdr hdr;
> +	__u32 index;
> +	__u8 dest_gid[16];
> +	__u8 reserved[4];
> +};
> +
> +union pvrdma_cmd_req {
> +	struct pvrdma_cmd_hdr hdr;
> +	struct pvrdma_cmd_query_port query_port;
> +	struct pvrdma_cmd_query_pkey query_pkey;
> +	struct pvrdma_cmd_create_uc create_uc;
> +	struct pvrdma_cmd_destroy_uc destroy_uc;
> +	struct pvrdma_cmd_create_pd create_pd;
> +	struct pvrdma_cmd_destroy_pd destroy_pd;
> +	struct pvrdma_cmd_create_mr create_mr;
> +	struct pvrdma_cmd_destroy_mr destroy_mr;
> +	struct pvrdma_cmd_create_cq create_cq;
> +	struct pvrdma_cmd_resize_cq resize_cq;
> +	struct pvrdma_cmd_destroy_cq destroy_cq;
> +	struct pvrdma_cmd_create_qp create_qp;
> +	struct pvrdma_cmd_modify_qp modify_qp;
> +	struct pvrdma_cmd_query_qp query_qp;
> +	struct pvrdma_cmd_destroy_qp destroy_qp;
> +	struct pvrdma_cmd_create_bind create_bind;
> +	struct pvrdma_cmd_destroy_bind destroy_bind;
> +};
> +
> +union pvrdma_cmd_resp {
> +	struct pvrdma_cmd_resp_hdr hdr;
> +	struct pvrdma_cmd_query_port_resp query_port_resp;
> +	struct pvrdma_cmd_query_pkey_resp query_pkey_resp;
> +	struct pvrdma_cmd_create_uc_resp create_uc_resp;
> +	struct pvrdma_cmd_create_pd_resp create_pd_resp;
> +	struct pvrdma_cmd_create_mr_resp create_mr_resp;
> +	struct pvrdma_cmd_create_cq_resp create_cq_resp;
> +	struct pvrdma_cmd_resize_cq_resp resize_cq_resp;
> +	struct pvrdma_cmd_create_qp_resp create_qp_resp;
> +	struct pvrdma_cmd_query_qp_resp query_qp_resp;
> +	struct pvrdma_cmd_destroy_qp_resp destroy_qp_resp;
> +};
> +
> +#endif /* __PVRDMA_DEV_API_H__ */
> -- 
> 2.7.4
> 
> --
> 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
--
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 01/15] IB/pvrdma: Add paravirtual rdma device

[Adit Ranadive]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="utf-8", Size: 415 bytes --]

On 7/18/16, 5:27 AM, "Yuval Shaia" <yuval.shaia@oracle.com> wrote:



>Patch #1 uses unions pvrdma_cmd_req & pvrdma_cmd_resp that added in patch
>#4.
>Suggesting to change the order.

Sure. Will ensure the order is correct for the other patches too in v3.

Thanks,
Adit
N‹§²æìr¸›yúèšØb²X¬¶Ç§vØ^–)Þº{.nÇ+‰·¥Š{±­ÙšŠ{ayº\x1dʇڙë,j\a­¢f£¢·hš‹»öì\x17/oSc¾™Ú³9˜uÀ¦æå‰È&jw¨®\x03(­éšŽŠÝ¢j"ú\x1a¶^[m§ÿïêäz¹Þ–Šàþf£¢·hšˆ§~ˆmš

Re: [PATCH v2 01/15] IB/pvrdma: Add paravirtual rdma device

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 4241 bytes --]

On Tue, Jul 12, 2016 at 12:36:31PM -0700, Adit Ranadive wrote:
> This patch adds the main device-level structures and functions to be used
> to provide RDMA functionality. Also, we define conversion functions from
> the IB core stack structures to the device-specific ones.
> 
> Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> ---
>  drivers/infiniband/hw/pvrdma/pvrdma.h | 459 ++++++++++++++++++++++++++++++++++
>  1 file changed, 459 insertions(+)
>  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma.h
> 
> diff --git a/drivers/infiniband/hw/pvrdma/pvrdma.h b/drivers/infiniband/hw/pvrdma/pvrdma.h
> new file mode 100644
> index 0000000..0dc5bd2
> --- /dev/null
> +++ b/drivers/infiniband/hw/pvrdma/pvrdma.h
> @@ -0,0 +1,459 @@
> +/*
> + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of EITHER the GNU General Public License
> + * version 2 as published by the Free Software Foundation or the BSD
> + * 2-Clause License. This program is distributed in the hope that it
> + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> + * See the GNU General Public License version 2 for more details at
> + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program available in the file COPYING in the main
> + * directory of this source tree.
> + *
> + * The BSD 2-Clause License
> + *
> + *     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.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> + * OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#ifndef __PVRDMA_H__
> +#define __PVRDMA_H__
> +
> +#include <linux/compiler.h>
> +#include <linux/interrupt.h>
> +#include <linux/list.h>
> +#include <linux/mutex.h>
> +#include <linux/pci.h>
> +#include <linux/semaphore.h>
> +#include <rdma/ib_umem.h>
> +#include <rdma/ib_verbs.h>
> +
> +#include "pvrdma_defs.h"
> +#include "pvrdma_dev_api.h"
> +#include "pvrdma_verbs.h"
> +
> +/* NOT the same as BIT_MASK(). */
> +#define PVRDMA_MASK(n) ((n << 1) - 1)
> +
> +/*
> + * VMware PVRDMA vendor id and PCI device id.
> + */
> +#ifndef PCI_VENDOR_ID_VMWARE
> +#define PCI_VENDOR_ID_VMWARE		0x15AD
> +#endif

Any special reason to do it?
You have this define in include/linux/pci_ids.h file.

> +#define PCI_DEVICE_ID_VMWARE_PVRDMA	0x0820

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 06/15] IB/pvrdma: Add virtual device RDMA structures

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 1251 bytes --]

On Tue, Jul 12, 2016 at 12:36:36PM -0700, Adit Ranadive wrote:
> This patch adds the various Verbs structures that we support in the
> virtual RDMA device. We have re-mapped the ones from the RDMA core stack
> to make sure we can maintain compatibility with our backend.
> 
> Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> ---
>  drivers/infiniband/hw/pvrdma/pvrdma_ib_verbs.h | 450 +++++++++++++++++++++++++
>  1 file changed, 450 insertions(+)
>  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_ib_verbs.h
> 

...

> +enum pvrdma_wc_opcode {
> +	PVRDMA_WC_SEND,

...

> +	PVRDMA_WC_RECV_RDMA_WITH_IMM
> +};

....
> +
> +enum pvrdma_wc_flags {
> +	PVRDMA_WC_GRH			= 1 << 0,
....

> +	PVRDMA_WC_FLAGS_MAX		= PVRDMA_WC_WITH_VLAN,
> +};
> +

Sometimes you put "," at the last line of enums, sometimes you don't.
Can you please be more consistent?

Thanks.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 05/15] IB/pvrdma: Add UAR support

[Yuval Shaia]

On Tue, Jul 12, 2016 at 12:36:35PM -0700, Adit Ranadive wrote:
> This patch adds the UAR support for the paravirtual RDMA device. The UAR
> pages are MMIO pages from the virtual PCI space. We define offsets within
> this page to provide the fast data-path operations.
> 
> Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> ---
>  drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c | 128 +++++++++++++++++++++++++
>  1 file changed, 128 insertions(+)
>  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c
> 
> diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c b/drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c
> new file mode 100644
> index 0000000..d854f6d
> --- /dev/null
> +++ b/drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c
> @@ -0,0 +1,128 @@
> +/*
> + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of EITHER the GNU General Public License
> + * version 2 as published by the Free Software Foundation or the BSD
> + * 2-Clause License. This program is distributed in the hope that it
> + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> + * See the GNU General Public License version 2 for more details at
> + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program available in the file COPYING in the main
> + * directory of this source tree.
> + *
> + * The BSD 2-Clause License
> + *
> + *     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.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> + * OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#include <linux/bitmap.h>
> +#include <linux/errno.h>
> +#include <linux/slab.h>
> +
> +#include "pvrdma.h"
> +
> +int pvrdma_uar_table_init(struct pvrdma_dev *dev)
> +{
> +	u32 num = dev->dsr->caps.max_uar;
> +	u32 mask = num - 1;
> +	struct pvrdma_id_table *tbl = &dev->uar_table.tbl;
> +
> +	/* Power of 2 */
> +	if (num != 1 << (ffs(num) - 1))
> +		return -EINVAL;
> +
> +	tbl->last = 0;
> +	tbl->top = 0;
> +	tbl->max = num;
> +	tbl->mask = mask;
> +	spin_lock_init(&tbl->lock);
> +	tbl->table = kcalloc(BITS_TO_LONGS(num), sizeof(long), GFP_KERNEL);
> +	if (!tbl->table)
> +		return -ENOMEM;
> +
> +	/* 0th UAR is taken by the device. */
> +	set_bit(0, tbl->table);
> +
> +	return 0;
> +}
> +
> +void pvrdma_uar_table_cleanup(struct pvrdma_dev *dev)
> +{
> +	struct pvrdma_id_table *tbl = &dev->uar_table.tbl;
> +
> +	kfree(tbl->table);
> +}
> +
> +int pvrdma_uar_alloc(struct pvrdma_dev *dev, struct pvrdma_uar_map *uar)
> +{
> +	struct pvrdma_id_table *tbl;
> +	unsigned long flags;
> +	u32 obj;
> +
> +	tbl = &dev->uar_table.tbl;
> +
> +	spin_lock_irqsave(&tbl->lock, flags);
> +	obj = find_next_zero_bit(tbl->table, tbl->max, tbl->last);
> +	if (obj >= tbl->max) {
> +		tbl->top = (tbl->top + tbl->max) & tbl->mask;
> +		obj = find_first_zero_bit(tbl->table, tbl->max);
> +	}
> +
> +	if (obj < tbl->max) {
> +		set_bit(obj, tbl->table);
> +		obj |= tbl->top;
> +	} else {
> +		spin_unlock_irqrestore(&tbl->lock, flags);
> +		return -ENOMEM;
> +	}

Please consider restructuring the above block to something like this:
	if (obj >= tbl->max) {
		spin_unlock_irqrestore(&tbl->lock, flags);
		return -ENOMEM;
	}

	set_bit(obj, tbl->table);
	obj |= tbl->top;

> +
> +	spin_unlock_irqrestore(&tbl->lock, flags);
> +
> +	uar->index = obj;
> +	uar->pfn = (pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_UAR) >>
> +		    PAGE_SHIFT) + uar->index;
> +
> +	return 0;
> +}
> +
> +void pvrdma_uar_free(struct pvrdma_dev *dev, struct pvrdma_uar_map *uar)
> +{
> +	struct pvrdma_id_table *tbl = &dev->uar_table.tbl;
> +	unsigned long flags;
> +	u32 obj;
> +
> +	obj = uar->index & (tbl->max - 1);
> +	spin_lock_irqsave(&tbl->lock, flags);
> +	clear_bit(obj, tbl->table);
> +	tbl->last = min(tbl->last, obj);
> +	tbl->top = (tbl->top + tbl->max) & tbl->mask;
> +	spin_unlock_irqrestore(&tbl->lock, flags);
> +}
> -- 
> 2.7.4
> 
> --
> 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
--
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 05/15] IB/pvrdma: Add UAR support

[Adit Ranadive]

On 7/18/16 6:25 AM, Leon Romanovsky wrote:
>> +int pvrdma_uar_table_init(struct pvrdma_dev *dev)
>> +{
>> +	u32 num = dev->dsr->caps.max_uar;
>> +	u32 mask = num - 1;
>> +	struct pvrdma_id_table *tbl = &dev->uar_table.tbl;
>> +
>> +	/* Power of 2 */
>> +	if (num != 1 << (ffs(num) - 1))
>> +		return -EINVAL;
> 
> Can functions/macros defined in include/linux/log2.h replace the code
> above?
> 

I guess I can replace the ffs stuff with is_power_of_2 from log2.h.

Thanks,
Adit
--
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 02/15] IB/pvrdma: Add device command support

[Adit Ranadive]

On 7/18/16 6:13 AM, Yuval Shaia wrote:
>> +
>> +static int pvrdma_cmd_recv(struct pvrdma_dev *dev, union pvrdma_cmd_resp *resp)
>> +{
>> +	dev_dbg(&dev->pdev->dev, "receive response from device\n");
>> +
>> +	spin_lock(&dev->cmd_lock);
>> +	memcpy(resp, dev->resp_slot, sizeof(*resp));
>> +	spin_unlock(&dev->cmd_lock);
> 
> In case your virtual HW supports more than one concurrent commands, if
> cmd_lock protects cmd_slot i would expect to have extra resp_lock to
> protect resp_slot.

Thanks for the suggestion! Currently, our virtual hardware only supports
processing one command at a time. I'll skip adding the resp->lock since it
may not add much value here.

> btw, otherwise why not just lock the entire flow:
> 	memcpy(dev->cmd_slot, req, sizeof(*req));
> 	pvrdma_write_reg(dev, PVRDMA_REG_REQUEST, 0);
> 	pvrdma_read_reg(dev, PVRDMA_REG_ERR);
> 	memcpy(resp, dev->resp_slot, sizeof(*resp));
> 
> This might improve performances.
> 

Since this is in the slow control path and we execute 1 command at a time,
not sure if that would be beneficial.

>> +
>> +	return 0;
>> +}
>> +
>> +int
>> +pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
>> +		bool expect_resp, union pvrdma_cmd_resp *resp)
>> +{
>> +	int err;
>> +
>> +	dev_dbg(&dev->pdev->dev, "post request to device\n");
>> +
>> +	/* Serializiation */
>> +	down(&dev->cmd_sema);
>> +
>> +	spin_lock(&dev->cmd_lock);
>> +	memcpy(dev->cmd_slot, req, sizeof(*req));
>> +	spin_unlock(&dev->cmd_lock);
>> +
>> +	init_completion(&dev->cmd_done);
>> +	pvrdma_write_reg(dev, PVRDMA_REG_REQUEST, 0);
>> +
>> +	/* Make sure the request is written before reading status. */
>> +	mb();
>> +	err = pvrdma_read_reg(dev, PVRDMA_REG_ERR);
>> +	if (err == 0) {
>> +		if (expect_resp) {
>> +			err = wait_for_completion_interruptible_timeout(
>> +				&dev->cmd_done,
>> +				msecs_to_jiffies(PVRDMA_CMD_TIMEOUT));
>> +			if (err == 0) {
>> +				dev_warn(&dev->pdev->dev,
>> +					 "completion timeout\n");
>> +				err = -ETIMEDOUT;
>> +			} else {
>> +				err = pvrdma_cmd_recv(dev, resp);
> 
> Maybe it is a silly question but in case of two commands that issued almost
> simultaneously, it possible that fast execution of second command will
> reach this point first and the reasponse here is of the first one?
> 

I don't think that may happen since we do have a semaphore here to protect that.

Thanks,
Adit
--
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 06/15] IB/pvrdma: Add virtual device RDMA structures

[Adit Ranadive]

On Tue, 26 Jul 2016 08:46:29 +0300
Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org> wrote:

> On Tue, Jul 12, 2016 at 12:36:36PM -0700, Adit Ranadive wrote:
> > This patch adds the various Verbs structures that we support in the
> > virtual RDMA device. We have re-mapped the ones from the RDMA core stack
> > to make sure we can maintain compatibility with our backend.
> > 
> > Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > ---
> >  drivers/infiniband/hw/pvrdma/pvrdma_ib_verbs.h | 450
> > +++++++++++++++++++++++++ 1 file changed, 450 insertions(+)
> >  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_ib_verbs.h
> >   
> 
> ...
> 
> > +enum pvrdma_wc_opcode {
> > +	PVRDMA_WC_SEND,  
> 
> ...
> 
> > +	PVRDMA_WC_RECV_RDMA_WITH_IMM
> > +};  
> 
> ....
> > +
> > +enum pvrdma_wc_flags {
> > +	PVRDMA_WC_GRH			= 1 << 0,  
> ....
> 
> > +	PVRDMA_WC_FLAGS_MAX		= PVRDMA_WC_WITH_VLAN,
> > +};
> > +  
> 
> Sometimes you put "," at the last line of enums, sometimes you don't.
> Can you please be more consistent?
> 
> Thanks.

Done. Added , to the enums missing it. Will send this out in v3.

Thanks,
Adit
--
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 11/15] IB/pvrdma: Add user-level shared functions

[Adit Ranadive]

On Mon, 18 Jul 2016 16:59:07 +0300
Yuval Shaia <yuval.shaia-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org> wrote:

> On Tue, Jul 12, 2016 at 12:36:41PM -0700, Adit Ranadive wrote:
> > We share some common structures with the user-level driver. This patch
> > adds those structures and shared functions to traverse the QP/CQ rings.
> > 
> > Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > ---
> >  drivers/infiniband/hw/pvrdma/pvrdma_uapi.h | 247
> > +++++++++++++++++++++++++++++ drivers/infiniband/hw/pvrdma/pvrdma_user.h |
> > 99 ++++++++++++ 2 files changed, 346 insertions(+)
> >  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_uapi.h
> >  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_user.h
> > 
> > diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_uapi.h
> > b/drivers/infiniband/hw/pvrdma/pvrdma_uapi.h new file mode 100644
> > index 0000000..274328c
> > --- /dev/null
> > +++ b/drivers/infiniband/hw/pvrdma/pvrdma_uapi.h
> > @@ -0,0 +1,247 @@
> > +/*
> > + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> > + *
> > + * This program is free software; you can redistribute it and/or
> > + * modify it under the terms of EITHER the GNU General Public License
> > + * version 2 as published by the Free Software Foundation or the BSD
> > + * 2-Clause License. This program is distributed in the hope that it
> > + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> > + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> > + * See the GNU General Public License version 2 for more details at
> > + *
> > https://urldefense.proofpoint.com/v2/url?u=http-3A__www.gnu.org_licenses_old-2Dlicenses_gpl-2D2.0.en.html&d=CwIBAg&c=Sqcl0Ez6M0X8aeM67LKIiDJAXVeAw-YihVMNtXt-uEs&r=uGpVYdpWLb65FRYUmM4EfQ&m=zWCNDfhYzlW7r0Soati7Ec2q9Vao5VNR2emCPrPrXLY&s=Tma3LbotbhNrOqLsPbVKn2noOZmFLnwQu2pOYlZqpfE&e= .
> > + *
> > + * You should have received a copy of the GNU General Public License
> > + * along with this program available in the file COPYING in the main
> > + * directory of this source tree.
> > + *
> > + * The BSD 2-Clause License
> > + *
> > + *     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.
> > + *
> > + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> > + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> > + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> > + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> > + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> > + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> > + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> > + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> > + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> > + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> > + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> > + * OF THE POSSIBILITY OF SUCH DAMAGE.
> > + */
> > +
> > +#ifndef __PVRDMA_UAPI_H__
> > +#define __PVRDMA_UAPI_H__
> > +
> > +#include <linux/types.h>
> > +
> > +#define PVRDMA_VERSION 17
> > +
> > +#define PVRDMA_UAR_HANDLE_MASK	0x00FFFFFF	/* Bottom 24 bits.
> > */ +#define PVRDMA_UAR_QP_OFFSET	0		/* Offset of QP
> > doorbell. */ +#define PVRDMA_UAR_QP_SEND	BIT(30)		/*
> > Send bit. */ +#define PVRDMA_UAR_QP_RECV	BIT(31)		/*
> > Recv bit. */ +#define PVRDMA_UAR_CQ_OFFSET	4		/*
> > Offset of CQ doorbell. */ +#define PVRDMA_UAR_CQ_ARM_SOL
> > BIT(29)		/* Arm solicited bit. */ +#define
> > PVRDMA_UAR_CQ_ARM	BIT(30)		/* Arm bit. */ +#define
> > PVRDMA_UAR_CQ_POLL	BIT(31)		/* Poll bit. */ +
> > +/* PVRDMA atomic compare and swap */
> > +struct pvrdma_exp_cmp_swap {
> > +	__u64 swap_val;
> > +	__u64 compare_val;
> > +	__u64 swap_mask;
> > +	__u64 compare_mask;
> > +};
> > +
> > +/* PVRDMA atomic fetch and add */
> > +struct pvrdma_exp_fetch_add {
> > +	__u64 add_val;
> > +	__u64 field_boundary;
> > +};
> > +
> > +/* PVRDMA address vector. */
> > +struct pvrdma_av {
> > +	__u32 port_pd;
> > +	__u32 sl_tclass_flowlabel;
> > +	__u8 dgid[16];
> > +	__u8 src_path_bits;
> > +	__u8 gid_index;
> > +	__u8 stat_rate;
> > +	__u8 hop_limit;
> > +	__u8 dmac[6];
> > +	__u8 reserved[6];
> > +};
> > +
> > +/* PVRDMA receive queue work request */
> > +struct pvrdma_rq_wqe_hdr {
> > +	__u64 wr_id;		/* wr id */
> > +	__u32 num_sge;		/* size of s/g array */
> > +	__u32 total_len;	/* reserved */
> > +};
> > +/* Use pvrdma_sge (ib_sge) for receieve queue s/g array elements. */
> > +
> > +/* PVRDMA send queue work request */
> > +struct pvrdma_sq_wqe_hdr {
> > +	__u64 wr_id;		/* wr id */
> > +	__u32 num_sge;		/* size of s/g array */
> > +	__u32 total_len;	/* reserved */
> > +	__u32 opcode;		/* operation type */
> > +	__u32 send_flags;	/* wr flags */
> > +	union {
> > +		__u32 imm_data;
> > +		__u32 invalidate_rkey;
> > +	} ex;
> > +	__u32 reserved;
> > +	union {
> > +		struct {
> > +			__u64 remote_addr;
> > +			__u32 rkey;
> > +			__u8 reserved[4];
> > +		} rdma;
> > +		struct {
> > +			__u64 remote_addr;
> > +			__u64 compare_add;
> > +			__u64 swap;
> > +			__u32 rkey;
> > +			__u32 reserved;
> > +		} atomic;
> > +		struct {
> > +			__u64 remote_addr;
> > +			__u32 log_arg_sz;
> > +			__u32 rkey;
> > +			union {
> > +				struct pvrdma_exp_cmp_swap  cmp_swap;
> > +				struct pvrdma_exp_fetch_add fetch_add;
> > +			} wr_data;
> > +		} masked_atomics;
> > +		struct {
> > +			__u64 iova_start;
> > +			__u64 pl_pdir_dma;
> > +			__u32 page_shift;
> > +			__u32 page_list_len;
> > +			__u32 length;
> > +			__u32 access_flags;
> > +			__u32 rkey;
> > +		} fast_reg;
> > +		struct {
> > +			__u32 remote_qpn;
> > +			__u32 remote_qkey;
> > +			struct pvrdma_av av;
> > +		} ud;
> > +	} wr;
> > +};
> > +/* Use pvrdma_sge (ib_sge) for send queue s/g array elements. */
> > +
> > +/* Completion queue element. */
> > +struct pvrdma_cqe {
> > +	__u64 wr_id;
> > +	__u64 qp;
> > +	__u32 opcode;
> > +	__u32 status;
> > +	__u32 byte_len;
> > +	__u32 imm_data;
> > +	__u32 src_qp;
> > +	__u32 wc_flags;
> > +	__u32 vendor_err;
> > +	__u16 pkey_index;
> > +	__u16 slid;
> > +	__u8 sl;
> > +	__u8 dlid_path_bits;
> > +	__u8 port_num;
> > +	__u8 smac[6];
> > +	__u8 reserved2[7]; /* Pad to next power of 2 (64). */
> > +};
> > +
> > +struct pvrdma_ring {
> > +	atomic_t prod_tail;	/* Producer tail. */
> > +	atomic_t cons_head;	/* Consumer head. */
> > +};
> > +
> > +struct pvrdma_ring_state {
> > +	struct pvrdma_ring tx;	/* Tx ring. */
> > +	struct pvrdma_ring rx;	/* Rx ring. */
> > +};
> > +
> > +static inline int pvrdma_idx_valid(__u32 idx, __u32 max_elems)
> > +{
> > +	/* Generates fewer instructions than a less-than. */
> > +	return (idx & ~((max_elems << 1) - 1)) == 0;
> > +}
> > +
> > +static inline __s32 pvrdma_idx(atomic_t *var, __u32 max_elems)
> > +{
> > +	const unsigned int idx = atomic_read(var);
> > +
> > +	if (pvrdma_idx_valid(idx, max_elems))
> > +		return idx & (max_elems - 1);
> > +	return -1;
> > +}
> > +
> > +static inline void pvrdma_idx_ring_inc(atomic_t *var, __u32 max_elems)
> > +{
> > +	__u32 idx = atomic_read(var) + 1;	/* Increment. */
> > +
> > +	idx &= (max_elems << 1) - 1;		/* Modulo size, flip
> > gen. */
> > +	atomic_set(var, idx);
> > +}
> > +
> > +static inline __s32 pvrdma_idx_ring_has_space(const struct pvrdma_ring *r,
> > +					      __u32 max_elems, __u32
> > *out_tail) +{
> > +	const __u32 tail = atomic_read(&r->prod_tail);
> > +	const __u32 head = atomic_read(&r->cons_head);
> > +
> > +	if (pvrdma_idx_valid(tail, max_elems) &&
> > +	    pvrdma_idx_valid(head, max_elems)) {
> > +		*out_tail = tail & (max_elems - 1);
> > +		return tail != (head ^ max_elems);
> > +	}
> > +	return -1;
> > +}
> > +
> > +static inline __s32 pvrdma_idx_ring_has_data(const struct pvrdma_ring *r,
> > +					     __u32 max_elems, __u32
> > *out_head) +{
> > +	const __u32 tail = atomic_read(&r->prod_tail);
> > +	const __u32 head = atomic_read(&r->cons_head);
> > +
> > +	if (pvrdma_idx_valid(tail, max_elems) &&
> > +	    pvrdma_idx_valid(head, max_elems)) {
> > +		*out_head = head & (max_elems - 1);
> > +		return tail != head;
> > +	}
> > +	return -1;  
> 
> What is the meaning of -1?

It just means its an invalid index. I'll make that more clear in v3.

Thanks,
Adit
--
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 05/15] IB/pvrdma: Add UAR support

[Adit Ranadive]

On Wed, 27 Jul 2016 17:06:12 +0300
Yuval Shaia <yuval.shaia-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org> wrote:

> On Tue, Jul 12, 2016 at 12:36:35PM -0700, Adit Ranadive wrote:
> > This patch adds the UAR support for the paravirtual RDMA device. The UAR
> > pages are MMIO pages from the virtual PCI space. We define offsets within
> > this page to provide the fast data-path operations.
> > 
> > Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > ---
> >  drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c | 128
> > +++++++++++++++++++++++++ 1 file changed, 128 insertions(+)
> >  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c
> > 
> > diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c
> > b/drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c new file mode 100644
> > index 0000000..d854f6d
> > --- /dev/null
> > +++ b/drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c
> > @@ -0,0 +1,128 @@
> > +/*
> > + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> > + *
> > + * This program is free software; you can redistribute it and/or
> > + * modify it under the terms of EITHER the GNU General Public License
> > + * version 2 as published by the Free Software Foundation or the BSD
> > + * 2-Clause License. This program is distributed in the hope that it
> > + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> > + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> > + * See the GNU General Public License version 2 for more details at
> > + *
> > https://urldefense.proofpoint.com/v2/url?u=http-3A__www.gnu.org_licenses_old-2Dlicenses_gpl-2D2.0.en.html&d=CwIBAg&c=Sqcl0Ez6M0X8aeM67LKIiDJAXVeAw-YihVMNtXt-uEs&r=uGpVYdpWLb65FRYUmM4EfQ&m=tNZWuWp3kE1s-gan-hE-5OXvDfvRDs9-8F-8KAmtUuA&s=sMzAhVyP3X8sxMREMqDO2OCcL6iGgSPFW_cbNPeGKUA&e= .
> > + *
> > + * You should have received a copy of the GNU General Public License
> > + * along with this program available in the file COPYING in the main
> > + * directory of this source tree.
> > + *
> > + * The BSD 2-Clause License
> > + *
> > + *     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.
> > + *
> > + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> > + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> > + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> > + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> > + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> > + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> > + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> > + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> > + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> > + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> > + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> > + * OF THE POSSIBILITY OF SUCH DAMAGE.
> > + */
> > +
> > +#include <linux/bitmap.h>
> > +#include <linux/errno.h>
> > +#include <linux/slab.h>
> > +
> > +#include "pvrdma.h"
> > +
> > +int pvrdma_uar_table_init(struct pvrdma_dev *dev)
> > +{
> > +	u32 num = dev->dsr->caps.max_uar;
> > +	u32 mask = num - 1;
> > +	struct pvrdma_id_table *tbl = &dev->uar_table.tbl;
> > +
> > +	/* Power of 2 */
> > +	if (num != 1 << (ffs(num) - 1))
> > +		return -EINVAL;
> > +
> > +	tbl->last = 0;
> > +	tbl->top = 0;
> > +	tbl->max = num;
> > +	tbl->mask = mask;
> > +	spin_lock_init(&tbl->lock);
> > +	tbl->table = kcalloc(BITS_TO_LONGS(num), sizeof(long), GFP_KERNEL);
> > +	if (!tbl->table)
> > +		return -ENOMEM;
> > +
> > +	/* 0th UAR is taken by the device. */
> > +	set_bit(0, tbl->table);
> > +
> > +	return 0;
> > +}
> > +
> > +void pvrdma_uar_table_cleanup(struct pvrdma_dev *dev)
> > +{
> > +	struct pvrdma_id_table *tbl = &dev->uar_table.tbl;
> > +
> > +	kfree(tbl->table);
> > +}
> > +
> > +int pvrdma_uar_alloc(struct pvrdma_dev *dev, struct pvrdma_uar_map *uar)
> > +{
> > +	struct pvrdma_id_table *tbl;
> > +	unsigned long flags;
> > +	u32 obj;
> > +
> > +	tbl = &dev->uar_table.tbl;
> > +
> > +	spin_lock_irqsave(&tbl->lock, flags);
> > +	obj = find_next_zero_bit(tbl->table, tbl->max, tbl->last);
> > +	if (obj >= tbl->max) {
> > +		tbl->top = (tbl->top + tbl->max) & tbl->mask;
> > +		obj = find_first_zero_bit(tbl->table, tbl->max);
> > +	}
> > +
> > +	if (obj < tbl->max) {
> > +		set_bit(obj, tbl->table);
> > +		obj |= tbl->top;
> > +	} else {
> > +		spin_unlock_irqrestore(&tbl->lock, flags);
> > +		return -ENOMEM;
> > +	}  
> 
> Please consider restructuring the above block to something like this:
> 	if (obj >= tbl->max) {
> 		spin_unlock_irqrestore(&tbl->lock, flags);
> 		return -ENOMEM;
> 	}
> 
> 	set_bit(obj, tbl->table);
> 	obj |= tbl->top;
> 

Done. Will add this in v3.

Thanks,
Adit
--
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 04/15] IB/pvrdma: Add the paravirtual RDMA device specification

[Adit Ranadive]

On Thu, 21 Jul 2016 11:45:09 +0300
Yuval Shaia <yuval.shaia-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org> wrote:

> Two minor comments.
> Besides that - Reviewed-by: Yuval Shaia <yuval.shaia-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org>
> 
> Yuval
> 
> On Tue, Jul 12, 2016 at 12:36:34PM -0700, Adit Ranadive wrote:
> > This patch describes the main specification of the underlying virtual RDMA
> > device. The pvrdma_dev_api header file defines the Verbs commands and
> > their parameters that can be issued to the device backend.
> > 
> > Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > ---
> >  drivers/infiniband/hw/pvrdma/pvrdma_defs.h    | 300 ++++++++++++++++++++++
> >  drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h | 342
> > ++++++++++++++++++++++++++ 2 files changed, 642 insertions(+)
> >  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_defs.h
> >  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h
> > 
> > diff --git a/drivers/infiniband/hw/pvrdma/pvrdma_defs.h
> > b/drivers/infiniband/hw/pvrdma/pvrdma_defs.h new file mode 100644
> > index 0000000..225cba4
> > --- /dev/null
> > +++ b/drivers/infiniband/hw/pvrdma/pvrdma_defs.h
> > @@ -0,0 +1,300 @@
> > +/*
> > + * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
> > + *
> > + * This program is free software; you can redistribute it and/or
> > + * modify it under the terms of EITHER the GNU General Public License
> > + * version 2 as published by the Free Software Foundation or the BSD
> > + * 2-Clause License. This program is distributed in the hope that it
> > + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
> > + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
> > + * See the GNU General Public License version 2 for more details at
> > + *
> > https://urldefense.proofpoint.com/v2/url?u=http-3A__www.gnu.org_licenses_old-2Dlicenses_gpl-2D2.0.en.html&d=CwIBAg&c=Sqcl0Ez6M0X8aeM67LKIiDJAXVeAw-YihVMNtXt-uEs&r=uGpVYdpWLb65FRYUmM4EfQ&m=czr8sWJhFSFR3J4meRKD6XraZLGcdnFbdCzUt9taPoM&s=w9TucKdjv27QS7PpvjqjVfU9uV35OBXlSCMXLdq9nQQ&e= .
> > + *
> > + * You should have received a copy of the GNU General Public License
> > + * along with this program available in the file COPYING in the main
> > + * directory of this source tree.
> > + *
> > + * The BSD 2-Clause License
> > + *
> > + *     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.
> > + *
> > + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> > + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> > + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
> > + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> > + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
> > + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
> > + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
> > + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
> > + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
> > + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> > + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
> > + * OF THE POSSIBILITY OF SUCH DAMAGE.
> > + */
> > +
> > +#ifndef __PVRDMA_DEFS_H__
> > +#define __PVRDMA_DEFS_H__
> > +
> > +#include <linux/types.h>
> > +#include "pvrdma_ib_verbs.h"
> > +#include "pvrdma_uapi.h"
> > +
> > +/*
> > + * Masks and accessors for page directory, which is a two-level lookup:
> > + * page directory -> page table -> page. Only one directory for now, but we
> > + * could expand that easily. 9 bits for tables, 9 bits for pages, gives one
> > + * gigabyte for memory regions and so forth.
> > + */
> > +
> > +#define PVRDMA_PAGE_DIR_DIR(x)		(((x) >> 18) & 0x1)
> > +#define PVRDMA_PAGE_DIR_TABLE(x)	(((x) >> 9) & 0x1ff)  
> 
> Suggesting to add #define for the above 9 and 18, ex PVRDMA_PAGE_DIR_SHIFT
> or something like that.
> 

Will add this in v3.

> > +#define PVRDMA_PAGE_DIR_PAGE(x)		((x) & 0x1ff)
> > +#define PVRDMA_PAGE_DIR_MAX_PAGES	(1 * 512 * 512)
> > +#define PVRDMA_MAX_FAST_REG_PAGES	128
> > +
> > +/*
> > + * Max MSI-X vectors.
> > + */
> > +
> > +#define PVRDMA_MAX_INTERRUPTS	3
> > +
> > +/* Register offsets within PCI resource on BAR1. */
> > +#define PVRDMA_REG_VERSION	0x00	/* R: Version of device. */
> > +#define PVRDMA_REG_DSRLOW	0x04	/* W: Device shared region
> > low PA. */ +#define PVRDMA_REG_DSRHIGH	0x08	/* W: Device
> > shared region high PA. */ +#define PVRDMA_REG_CTL
> > 0x0c	/* W: PVRDMA_DEVICE_CTL */ +#define PVRDMA_REG_REQUEST
> > 0x10	/* W: Indicate device request. */ +#define
> > PVRDMA_REG_ERR		0x14	/* R: Device error. */ +#define
> > PVRDMA_REG_ICR		0x18	/* R: Interrupt cause. */
> > +#define PVRDMA_REG_IMR		0x1c	/* R/W: Interrupt mask.
> > */ +#define PVRDMA_REG_MACL		0x20	/* R/W: MAC address
> > low. */ +#define PVRDMA_REG_MACH		0x24	/* R/W: MAC
> > address high. */ + +/* Object flags. */
> > +#define PVRDMA_CQ_FLAG_ARMED_SOL	BIT(0)	/* Armed for
> > solicited-only. */ +#define PVRDMA_CQ_FLAG_ARMED
> > BIT(1)	/* Armed. */ +#define PVRDMA_MR_FLAG_DMA
> > BIT(0)	/* DMA region. */ +#define PVRDMA_MR_FLAG_FRMR
> > BIT(1)	/* Fast reg memory region. */ +
> > +/*
> > + * Atomic operation capability (masked versions are extended atomic
> > + * operations.
> > + */
> > +
> > +#define PVRDMA_ATOMIC_OP_COMP_SWAP	BIT(0) /* Compare and swap. */
> > +#define PVRDMA_ATOMIC_OP_FETCH_ADD	BIT(1) /* Fetch and add. */
> > +#define PVRDMA_ATOMIC_OP_MASK_COMP_SWAP	BIT(2) /* Masked compare
> > and swap. */ +#define PVRDMA_ATOMIC_OP_MASK_FETCH_ADD	BIT(3) /*
> > Masked fetch and add. */ +
> > +/*
> > + * Base Memory Management Extension flags to support Fast Reg Memory
> > Regions
> > + * and Fast Reg Work Requests. Each flag represents a verb operation and we
> > + * must support all of them to qualify for the BMME device cap.
> > + */
> > +
> > +#define PVRDMA_BMME_FLAG_LOCAL_INV	BIT(0) /* Local Invalidate. */
> > +#define PVRDMA_BMME_FLAG_REMOTE_INV	BIT(1) /* Remote Invalidate. */
> > +#define PVRDMA_BMME_FLAG_FAST_REG_WR	BIT(2) /* Fast Reg Work
> > Request. */ +
> > +/*
> > + * GID types. The interpretation of the gid_types bit field in the device
> > + * capabilities will depend on the device mode. For now, the device only
> > + * supports RoCE as mode, so only the different GID types for RoCE are
> > + * defined.
> > + */
> > +
> > +#define PVRDMA_GID_TYPE_FLAG_ROCE_V1 BIT(0)
> > +#define PVRDMA_GID_TYPE_FLAG_ROCE_V2 BIT(1)
> > +
> > +enum pvrdma_pci_resource {
> > +	PVRDMA_PCI_RESOURCE_MSIX = 0,	/* BAR0: MSI-X, MMIO. */
> > +	PVRDMA_PCI_RESOURCE_REG,	/* BAR1: Registers, MMIO. */
> > +	PVRDMA_PCI_RESOURCE_UAR,	/* BAR2: UAR pages, MMIO, 64-bit.
> > */
> > +	PVRDMA_PCI_RESOURCE_LAST,	/* Last. */
> > +};
> > +
> > +enum pvrdma_device_ctl {
> > +	PVRDMA_DEVICE_CTL_ACTIVATE = 0,	/* Activate device. */
> > +	PVRDMA_DEVICE_CTL_QUIESCE,	/* Quiesce device. */
> > +	PVRDMA_DEVICE_CTL_RESET,	/* Reset device. */
> > +};
> > +
> > +enum pvrdma_intr_vector {
> > +	PVRDMA_INTR_VECTOR_RESPONSE	= 0,	/* Command
> > response. */
> > +	PVRDMA_INTR_VECTOR_ASYNC	= 1,	/* Async events. */
> > +	PVRDMA_INTR_VECTOR_CQ		= 2,	/* CQ
> > notification. */
> > +	/* Additional CQ notification vectors. */
> > +};
> > +
> > +enum pvrdma_intr_cause {
> > +	PVRDMA_INTR_CAUSE_RESPONSE	= (1 <<
> > PVRDMA_INTR_VECTOR_RESPONSE),
> > +	PVRDMA_INTR_CAUSE_ASYNC		= (1 <<
> > PVRDMA_INTR_VECTOR_ASYNC),
> > +	PVRDMA_INTR_CAUSE_CQ		= (1 << PVRDMA_INTR_VECTOR_CQ),
> > +};
> > +
> > +enum pvrdma_intr_type {
> > +	PVRDMA_INTR_TYPE_INTX = 0,	/* Legacy. */
> > +	PVRDMA_INTR_TYPE_MSI,		/* MSI. */
> > +	PVRDMA_INTR_TYPE_MSIX,		/* MSI-X. */
> > +};
> > +
> > +enum pvrdma_gos_bits {
> > +	PVRDMA_GOS_BITS_UNK = 0,	/* Unknown. */
> > +	PVRDMA_GOS_BITS_32,		/* 32-bit. */
> > +	PVRDMA_GOS_BITS_64,		/* 64-bit. */
> > +};
> > +
> > +enum pvrdma_gos_type {
> > +	PVRDMA_GOS_TYPE_UNK = 0,	/* Unknown. */
> > +	PVRDMA_GOS_TYPE_LINUX,		/* Linux. */
> > +};
> > +
> > +enum pvrdma_device_mode {
> > +	PVRDMA_DEVICE_MODE_ROCE = 0,	/* RoCE. */
> > +	PVRDMA_DEVICE_MODE_IWARP,	/* iWarp. */
> > +	PVRDMA_DEVICE_MODE_IB		/* InfiniBand. */
> > +};
> > +
> > +struct pvrdma_gos_info {
> > +	__u32 gos_bits:2;	/* W: PVRDMA_GOS_BITS_ */
> > +	__u32 gos_type:4;	/* W: PVRDMA_GOS_TYPE_ */
> > +	__u32 gos_ver:16;	/* W: Guest OS version. */
> > +	__u32 gos_misc:10;	/* W: Other. */
> > +	__u32 pad;		/*    Pad to 8-byte alignment. */  
> 
> Strip the spaces in comment.
> 

Ok. Did it here and other places below.

Thanks,
Adit
--
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 03/15] IB/pvrdma: Add support for Completion Queues

[Adit Ranadive]

On Mon, 18 Jul 2016 17:12:22 +0300 Yuval Shaia <yuval.shaia-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org> wrote:
> 
> On Tue, Jul 12, 2016 at 12:36:33PM -0700, Adit Ranadive wrote:
> > This patch adds the support for creating and destroying completion queues
> > on the paravirtual RDMA device.
> >
> > Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > ---
> >  drivers/infiniband/hw/pvrdma/pvrdma_cq.c | 436 +++++++++++++++++++++++++++++++
> >  1 file changed, 436 insertions(+)
> >  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_cq.c
> >

...

> > +void pvrdma_flush_cqe(struct pvrdma_qp *qp, struct pvrdma_cq *cq)
> > +{
> > +	int head;
> > +	int has_data;
> > +
> > +	if (!cq->is_kernel)
> > +		return;
> > +
> > +	/* Lock held */
> > +	has_data = pvrdma_idx_ring_has_data(&cq->ring_state->rx,
> > +					    cq->ibcq.cqe, &head);
> 
> For easy review (and probably maintenance) it looks to me that we need to
> preserve some sort of order where function/structures/union/enum etc will
> be define in preceding patch in a patch-set.
> This function implemented in patch #11 while we are in patch #3.

Ok will ensure that in v3.

> > +	if (unlikely(has_data > 0)) {
> > +		int items;
> > +		int curr;
> > +		int tail = pvrdma_idx(&cq->ring_state->rx.prod_tail,
> > +				      cq->ibcq.cqe);
> > +		struct pvrdma_cqe *cqe;
> > +		struct pvrdma_cqe *curr_cqe;
> > +
> > +		items = (tail > head) ? (tail - head) :
> > +			(cq->ibcq.cqe - head + tail);
> > +		curr = --tail;
> > +		while (items-- > 0) {
> > +			if (curr < 0)
> > +				curr = cq->ibcq.cqe - 1;
> > +			if (tail < 0)
> > +				tail = cq->ibcq.cqe - 1;
> > +			curr_cqe = get_cqe(cq, curr);
> > +			if ((curr_cqe->qp & 0xFFFF) != qp->qp_handle) {
> > +				if (curr != tail) {
> > +					cqe = get_cqe(cq, tail);
> > +					*cqe = *curr_cqe;
> > +				}
> > +				tail--;
> > +			} else {
> > +				pvrdma_idx_ring_inc(
> > +					&cq->ring_state->rx.cons_head,
> > +					cq->ibcq.cqe);
> > +			}
> > +			curr--;
> > +		}
> > +	}
> > +}
> > +
> > +static int pvrdma_poll_one(struct pvrdma_cq *cq, struct pvrdma_qp **cur_qp,
> > +			   struct ib_wc *wc)
> > +{
> > +	struct pvrdma_dev *dev = to_vdev(cq->ibcq.device);
> > +	int has_data;
> > +	unsigned int head;
> > +	bool tried = false;
> > +	struct pvrdma_cqe *cqe;
> > +
> > +retry:
> > +	has_data = pvrdma_idx_ring_has_data(&cq->ring_state->rx,
> > +					    cq->ibcq.cqe, &head);
> > +	if (has_data == 0) {
> > +		u32 val;
> > +
> > +		if (tried)
> > +			return -EAGAIN;
> > +
> > +		/* Pass down POLL to give physical HCA a chance to poll. */
> > +		val = cq->cq_handle | PVRDMA_UAR_CQ_POLL;
> > +		writel(cpu_to_le32(val),
> > +		       dev->driver_uar.map + PVRDMA_UAR_CQ_OFFSET);
> > +
> > +		tried = true;
> > +		goto retry;
> > +	} else if (has_data == -1) {
> 
> Is -1 represent fatal unrecoverable error or that next call to
> pvrdma_idx_ring_has_data might succeeds?

-1 does point to an invalid index on the ring. Its possible that the 
ring state is corrupted so treat it as unrecoverable for now.

> 
> > +		return -EINVAL;
> > +	}
> > +
> > +	cqe = get_cqe(cq, head);
> > +
> > +	/* Ensure cqe is valid. */
> > +	rmb();
> > +	if (dev->qp_tbl[cqe->qp & 0xffff])
> > +		*cur_qp = (struct pvrdma_qp *)dev->qp_tbl[cqe->qp & 0xffff];
> > +	else
> > +		return -EINVAL;
> 
> Is this can happen when QP was deleted while WR was processing?
> Or that in first place it was pushed with invalid QP?

If for some reason the QPN stored in the CQE was invalid we wanted to 
protect against that.

> Can you explain why you choose EINVAL and not EAGAIN? i.e we polled the
> "corrupted" CQE, next CEQ should be fine.
> Reading pvrdma_poll_cq looks like that if it was requested to poll 10 CQE
> for example and the 9th fails - the function will fail them all.

That makes sense. Ill replace this with EAGAIN. Its possible that the CQ is shared
with other QPs so other CQEs could still be valid.

Thanks,
Adit
--
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 01/15] IB/pvrdma: Add paravirtual rdma device

[Adit Ranadive]

On Tue, 26 Jul 2016 08:38:20 +0300 Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org> wrote:
> On Tue, Jul 12, 2016 at 12:36:31PM -0700, Adit Ranadive wrote:
> > This patch adds the main device-level structures and functions to be used
> > to provide RDMA functionality. Also, we define conversion functions from
> > the IB core stack structures to the device-specific ones.
> >
> > Reviewed-by: Jorgen Hansen <jhansen-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: George Zhang <georgezhang-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: Aditya Sarwade <asarwade-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Reviewed-by: Bryan Tan <bryantan-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > Signed-off-by: Adit Ranadive <aditr-pghWNbHTmq7QT0dZR+AlfA@public.gmane.org>
> > ---
> >  drivers/infiniband/hw/pvrdma/pvrdma.h | 459 ++++++++++++++++++++++++++++++++++
> >  1 file changed, 459 insertions(+)
> >  create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma.h

...

> > +/*
> > + * VMware PVRDMA vendor id and PCI device id.
> > + */
> > +#ifndef PCI_VENDOR_ID_VMWARE
> > +#define PCI_VENDOR_ID_VMWARE		0x15AD
> > +#endif
> 
> Any special reason to do it?
> You have this define in include/linux/pci_ids.h file.

Hmm, not necessarily. I believe it wasn't defined in some of the older
distros. We should be okay to remove it.

Thanks,
Adit

--
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 02/15] IB/pvrdma: Add device command support

[Adit Ranadive]

On Mon, 18 Jul 2016 15:46:53 +0300 Yuval Shaia <yuval.shaia-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org> wrote:
> On Tue, Jul 12, 2016 at 12:36:32PM -0700, Adit Ranadive wrote:
> > This patch enables posting Verb requests and receiving responses to/from
> > the backend PVRDMA emulation layer.
> >

...

> > +int
> > +pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
> > +		bool expect_resp, union pvrdma_cmd_resp *resp)
> > +{
> > +	int err;
> > +
> > +	dev_dbg(&dev->pdev->dev, "post request to device\n");
> > +
> > +	/* Serializiation */
> > +	down(&dev->cmd_sema);
> > +
> > +	spin_lock(&dev->cmd_lock);
> > +	memcpy(dev->cmd_slot, req, sizeof(*req));
> 
> Just to protect against memory corruption suggesting to do:
> 	memcpy(dev->cmd_slot, req, min(PAGE_SIZE, sizeof(*req)));
> 

Done.

> > +	spin_unlock(&dev->cmd_lock);
> > +
> > +	init_completion(&dev->cmd_done);
> > +	pvrdma_write_reg(dev, PVRDMA_REG_REQUEST, 0);
> > +
> > +	/* Make sure the request is written before reading status. */
> > +	mb();
> > +	err = pvrdma_read_reg(dev, PVRDMA_REG_ERR);
> > +	if (err == 0) {
> > +		if (expect_resp) {
> 
> How about utilizing resp instead of extra argument.
> This will save you for example from the local variable rsp in function
> pvrdma_destroy_qp (patch #10)
> 

Ok. Will remove the extra argument.

> > +			err = wait_for_completion_interruptible_timeout(
> > +				&dev->cmd_done,
> > +				msecs_to_jiffies(PVRDMA_CMD_TIMEOUT));
> > +			if (err == 0) {
> > +				dev_warn(&dev->pdev->dev,
> > +					 "completion timeout\n");
> 
> Is this really a warning only? i.e. i assume some high level flow is
> break now.
> 

I have changed this to a dev_err.

Thanks,
Adit
--
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 02/15] IB/pvrdma: Add device command support

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 1390 bytes --]

On Thu, Jul 28, 2016 at 10:51:39PM +0000, Adit Ranadive wrote:
> On Mon, 18 Jul 2016 15:46:53 +0300 Yuval Shaia <yuval.shaia-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org> wrote:
> > On Tue, Jul 12, 2016 at 12:36:32PM -0700, Adit Ranadive wrote:
> > > This patch enables posting Verb requests and receiving responses to/from
> > > the backend PVRDMA emulation layer.
> > >
> 
> ...
> 
> > > +int
> > > +pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
> > > +		bool expect_resp, union pvrdma_cmd_resp *resp)
> > > +{
> > > +	int err;
> > > +
> > > +	dev_dbg(&dev->pdev->dev, "post request to device\n");
> > > +
> > > +	/* Serializiation */
> > > +	down(&dev->cmd_sema);
> > > +
> > > +	spin_lock(&dev->cmd_lock);
> > > +	memcpy(dev->cmd_slot, req, sizeof(*req));
> > 
> > Just to protect against memory corruption suggesting to do:
> > 	memcpy(dev->cmd_slot, req, min(PAGE_SIZE, sizeof(*req)));
> > 
> 
> Done.

Hi Adit,
I didn't check how "dev->cmd_slot" was declared and if it is equal to
req in size. Additionally I didn't check if memcpy has any size
limitations, but wanted to warn you that implementation of this
suggestion as is will leave your system in much worse situation of
partially copied data.

if you really want to check sizes, please do it with BUILD_ON compilation
macro. I have serious doubts if you need it.

Thanks

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

RE: [PATCH v2 02/15] IB/pvrdma: Add device command support

[Adit Ranadive]

On Fri, 29 Jul 2016 12:37:29 +0000, Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org> wrote:
> On Thu, Jul 28, 2016 at 10:51:39PM +0000, Adit Ranadive wrote:
> > On Mon, 18 Jul 2016 15:46:53 +0300 Yuval Shaia <yuval.shaia-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org> wrote:
> > > On Tue, Jul 12, 2016 at 12:36:32PM -0700, Adit Ranadive wrote:
> > > > This patch enables posting Verb requests and receiving responses to/from
> > > > the backend PVRDMA emulation layer.
> > > >
> >
> > ...
> >
> > > > +int
> > > > +pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
> > > > +		bool expect_resp, union pvrdma_cmd_resp *resp)
> > > > +{
> > > > +	int err;
> > > > +
> > > > +	dev_dbg(&dev->pdev->dev, "post request to device\n");
> > > > +
> > > > +	/* Serializiation */
> > > > +	down(&dev->cmd_sema);
> > > > +
> > > > +	spin_lock(&dev->cmd_lock);
> > > > +	memcpy(dev->cmd_slot, req, sizeof(*req));
> > >
> > > Just to protect against memory corruption suggesting to do:
> > > 	memcpy(dev->cmd_slot, req, min(PAGE_SIZE, sizeof(*req)));
> > >
> >
> > Done.
> 
> Hi Adit,
> I didn't check how "dev->cmd_slot" was declared and if it is equal to
> req in size. Additionally I didn't check if memcpy has any size
> limitations, but wanted to warn you that implementation of this
> suggestion as is will leave your system in much worse situation of
> partially copied data.
> 
> if you really want to check sizes, please do it with BUILD_ON compilation
> macro. I have serious doubts if you need it.
> 
> Thanks

Hi Leon,

Thanks for taking a look!
I think this is okay to do was that we allocate a page for cmd_slot anyways
and the size of req will always be smaller than PAGE_SIZE. If the request is
malformed, our virtual hardware should throw an error for it.
I dont think we would run into partially copied data.

Having said that, I dont see a necessary advantage of using min here so am
willing to drop it if you feel strongly about it.

Can you clarify what you mean by memcpy size limitations?

Thanks,
Adit
--
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 02/15] IB/pvrdma: Add device command support

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 2448 bytes --]

On Fri, Jul 29, 2016 at 09:31:16PM +0000, Adit Ranadive wrote:
> On Fri, 29 Jul 2016 12:37:29 +0000, Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org> wrote:
> > On Thu, Jul 28, 2016 at 10:51:39PM +0000, Adit Ranadive wrote:
> > > On Mon, 18 Jul 2016 15:46:53 +0300 Yuval Shaia <yuval.shaia-QHcLZuEGTsthl2p70BpVqQ@public.gmane.orgm> wrote:
> > > > On Tue, Jul 12, 2016 at 12:36:32PM -0700, Adit Ranadive wrote:
> > > > > This patch enables posting Verb requests and receiving responses to/from
> > > > > the backend PVRDMA emulation layer.
> > > > >
> > >
> > > ...
> > >
> > > > > +int
> > > > > +pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
> > > > > +		bool expect_resp, union pvrdma_cmd_resp *resp)
> > > > > +{
> > > > > +	int err;
> > > > > +
> > > > > +	dev_dbg(&dev->pdev->dev, "post request to device\n");
> > > > > +
> > > > > +	/* Serializiation */
> > > > > +	down(&dev->cmd_sema);
> > > > > +
> > > > > +	spin_lock(&dev->cmd_lock);
> > > > > +	memcpy(dev->cmd_slot, req, sizeof(*req));
> > > >
> > > > Just to protect against memory corruption suggesting to do:
> > > > 	memcpy(dev->cmd_slot, req, min(PAGE_SIZE, sizeof(*req)));
> > > >
> > >
> > > Done.
> > 
> > Hi Adit,
> > I didn't check how "dev->cmd_slot" was declared and if it is equal to
> > req in size. Additionally I didn't check if memcpy has any size
> > limitations, but wanted to warn you that implementation of this
> > suggestion as is will leave your system in much worse situation of
> > partially copied data.
> > 
> > if you really want to check sizes, please do it with BUILD_ON compilation
> > macro. I have serious doubts if you need it.
> > 
> > Thanks
> 
> Hi Leon,
> 
> Thanks for taking a look!
> I think this is okay to do was that we allocate a page for cmd_slot anyways
> and the size of req will always be smaller than PAGE_SIZE. If the request is
> malformed, our virtual hardware should throw an error for it.
> I dont think we would run into partially copied data.
> 
> Having said that, I dont see a necessary advantage of using min here so am
> willing to drop it if you feel strongly about it.
> 
> Can you clarify what you mean by memcpy size limitations?

I didn't have time to look after reasons for Yuval's comment about
min(..), so I assumed that one of the possibilities will be limitation
of memcpy which I'm not aware.

> 
> Thanks,
> Adit

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 02/15] IB/pvrdma: Add device command support

[Yuval Shaia]

On Sun, Jul 31, 2016 at 09:15:47AM +0300, Leon Romanovsky wrote:
> On Fri, Jul 29, 2016 at 09:31:16PM +0000, Adit Ranadive wrote:
> > On Fri, 29 Jul 2016 12:37:29 +0000, Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org> wrote:
> > > On Thu, Jul 28, 2016 at 10:51:39PM +0000, Adit Ranadive wrote:
> > > > On Mon, 18 Jul 2016 15:46:53 +0300 Yuval Shaia <yuval.shaia-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org> wrote:
> > > > > On Tue, Jul 12, 2016 at 12:36:32PM -0700, Adit Ranadive wrote:
> > > > > > This patch enables posting Verb requests and receiving responses to/from
> > > > > > the backend PVRDMA emulation layer.
> > > > > >
> > > >
> > > > ...
> > > >
> > > > > > +int
> > > > > > +pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
> > > > > > +		bool expect_resp, union pvrdma_cmd_resp *resp)
> > > > > > +{
> > > > > > +	int err;
> > > > > > +
> > > > > > +	dev_dbg(&dev->pdev->dev, "post request to device\n");
> > > > > > +
> > > > > > +	/* Serializiation */
> > > > > > +	down(&dev->cmd_sema);
> > > > > > +
> > > > > > +	spin_lock(&dev->cmd_lock);
> > > > > > +	memcpy(dev->cmd_slot, req, sizeof(*req));
> > > > >
> > > > > Just to protect against memory corruption suggesting to do:
> > > > > 	memcpy(dev->cmd_slot, req, min(PAGE_SIZE, sizeof(*req)));
> > > > >
> > > >
> > > > Done.
> > > 
> > > Hi Adit,
> > > I didn't check how "dev->cmd_slot" was declared and if it is equal to
> > > req in size. Additionally I didn't check if memcpy has any size
> > > limitations, but wanted to warn you that implementation of this
> > > suggestion as is will leave your system in much worse situation of
> > > partially copied data.
> > > 
> > > if you really want to check sizes, please do it with BUILD_ON compilation
> > > macro. I have serious doubts if you need it.
> > > 
> > > Thanks
> > 
> > Hi Leon,
> > 
> > Thanks for taking a look!
> > I think this is okay to do was that we allocate a page for cmd_slot anyways
> > and the size of req will always be smaller than PAGE_SIZE. If the request is
> > malformed, our virtual hardware should throw an error for it.
> > I dont think we would run into partially copied data.
> > 
> > Having said that, I dont see a necessary advantage of using min here so am
> > willing to drop it if you feel strongly about it.
> > 
> > Can you clarify what you mean by memcpy size limitations?
> 
> I didn't have time to look after reasons for Yuval's comment about
> min(..), so I assumed that one of the possibilities will be limitation
> of memcpy which I'm not aware.

I understand that it is not such realistic that one of the structures in
union pvrdma_cmd_req will be more then PAGE_SIZE but any other alternative
to protect from such case?

> 
> > 
> > Thanks,
> > Adit


--
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 02/15] IB/pvrdma: Add device command support

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 3055 bytes --]

On Sun, Aug 14, 2016 at 02:25:36PM +0300, Yuval Shaia wrote:
> On Sun, Jul 31, 2016 at 09:15:47AM +0300, Leon Romanovsky wrote:
> > On Fri, Jul 29, 2016 at 09:31:16PM +0000, Adit Ranadive wrote:
> > > On Fri, 29 Jul 2016 12:37:29 +0000, Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org> wrote:
> > > > On Thu, Jul 28, 2016 at 10:51:39PM +0000, Adit Ranadive wrote:
> > > > > On Mon, 18 Jul 2016 15:46:53 +0300 Yuval Shaia <yuval.shaia@oracle.com> wrote:
> > > > > > On Tue, Jul 12, 2016 at 12:36:32PM -0700, Adit Ranadive wrote:
> > > > > > > This patch enables posting Verb requests and receiving responses to/from
> > > > > > > the backend PVRDMA emulation layer.
> > > > > > >
> > > > >
> > > > > ...
> > > > >
> > > > > > > +int
> > > > > > > +pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
> > > > > > > +		bool expect_resp, union pvrdma_cmd_resp *resp)
> > > > > > > +{
> > > > > > > +	int err;
> > > > > > > +
> > > > > > > +	dev_dbg(&dev->pdev->dev, "post request to device\n");
> > > > > > > +
> > > > > > > +	/* Serializiation */
> > > > > > > +	down(&dev->cmd_sema);
> > > > > > > +
> > > > > > > +	spin_lock(&dev->cmd_lock);
> > > > > > > +	memcpy(dev->cmd_slot, req, sizeof(*req));
> > > > > >
> > > > > > Just to protect against memory corruption suggesting to do:
> > > > > > 	memcpy(dev->cmd_slot, req, min(PAGE_SIZE, sizeof(*req)));
> > > > > >
> > > > >
> > > > > Done.
> > > > 
> > > > Hi Adit,
> > > > I didn't check how "dev->cmd_slot" was declared and if it is equal to
> > > > req in size. Additionally I didn't check if memcpy has any size
> > > > limitations, but wanted to warn you that implementation of this
> > > > suggestion as is will leave your system in much worse situation of
> > > > partially copied data.
> > > > 
> > > > if you really want to check sizes, please do it with BUILD_ON compilation
> > > > macro. I have serious doubts if you need it.
> > > > 
> > > > Thanks
> > > 
> > > Hi Leon,
> > > 
> > > Thanks for taking a look!
> > > I think this is okay to do was that we allocate a page for cmd_slot anyways
> > > and the size of req will always be smaller than PAGE_SIZE. If the request is
> > > malformed, our virtual hardware should throw an error for it.
> > > I dont think we would run into partially copied data.
> > > 
> > > Having said that, I dont see a necessary advantage of using min here so am
> > > willing to drop it if you feel strongly about it.
> > > 
> > > Can you clarify what you mean by memcpy size limitations?
> > 
> > I didn't have time to look after reasons for Yuval's comment about
> > min(..), so I assumed that one of the possibilities will be limitation
> > of memcpy which I'm not aware.
> 
> I understand that it is not such realistic that one of the structures in
> union pvrdma_cmd_req will be more then PAGE_SIZE but any other alternative
> to protect from such case?

BUILD_BUG_ON(...) is a perfect way to achieve it.

> 
> > 
> > > 
> > > Thanks,
> > > Adit
> 
> 

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

RE: [PATCH v2 02/15] IB/pvrdma: Add device command support

[Adit Ranadive]

On Sun, Aug 14, 2016 at 17:24:46 +0000, Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org> wrote:
> On Sun, Aug 14, 2016 at 02:25:36PM +0300, Yuval Shaia wrote:
> > On Sun, Jul 31, 2016 at 09:15:47AM +0300, Leon Romanovsky wrote:
> > > On Fri, Jul 29, 2016 at 09:31:16PM +0000, Adit Ranadive wrote:
> > > > On Fri, 29 Jul 2016 12:37:29 +0000, Leon Romanovsky <leon-DgEjT+Ai2yhQFI55V6+gNQ@public.gmane.orgg> wrote:
> > > > > On Thu, Jul 28, 2016 at 10:51:39PM +0000, Adit Ranadive wrote:
> > > > > > On Mon, 18 Jul 2016 15:46:53 +0300 Yuval Shaia <yuval.shaia@oracle.com> wrote:
> > > > > > > On Tue, Jul 12, 2016 at 12:36:32PM -0700, Adit Ranadive wrote:
> > > > > > > > This patch enables posting Verb requests and receiving responses to/from
> > > > > > > > the backend PVRDMA emulation layer.
> > > > > > > >
> > > > > >
> > > > > > ...
> > > > > >
> > > > > > > > +int
> > > > > > > > +pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
> > > > > > > > +		bool expect_resp, union pvrdma_cmd_resp *resp)
> > > > > > > > +{
> > > > > > > > +	int err;
> > > > > > > > +
> > > > > > > > +	dev_dbg(&dev->pdev->dev, "post request to device\n");
> > > > > > > > +
> > > > > > > > +	/* Serializiation */
> > > > > > > > +	down(&dev->cmd_sema);
> > > > > > > > +
> > > > > > > > +	spin_lock(&dev->cmd_lock);
> > > > > > > > +	memcpy(dev->cmd_slot, req, sizeof(*req));
> > > > > > >
> > > > > > > Just to protect against memory corruption suggesting to do:
> > > > > > > 	memcpy(dev->cmd_slot, req, min(PAGE_SIZE, sizeof(*req)));
> > > > > > >
> > > > > >
> > > > > > Done.
> > > > >
> > > > > Hi Adit,
> > > > > I didn't check how "dev->cmd_slot" was declared and if it is equal to
> > > > > req in size. Additionally I didn't check if memcpy has any size
> > > > > limitations, but wanted to warn you that implementation of this
> > > > > suggestion as is will leave your system in much worse situation of
> > > > > partially copied data.
> > > > >
> > > > > if you really want to check sizes, please do it with BUILD_ON compilation
> > > > > macro. I have serious doubts if you need it.
> > > > >
> > > > > Thanks
> > > >
> > > > Hi Leon,
> > > >
> > > > Thanks for taking a look!
> > > > I think this is okay to do was that we allocate a page for cmd_slot anyways
> > > > and the size of req will always be smaller than PAGE_SIZE. If the request is
> > > > malformed, our virtual hardware should throw an error for it.
> > > > I dont think we would run into partially copied data.
> > > >
> > > > Having said that, I dont see a necessary advantage of using min here so am
> > > > willing to drop it if you feel strongly about it.
> > > >
> > > > Can you clarify what you mean by memcpy size limitations?
> > >
> > > I didn't have time to look after reasons for Yuval's comment about
> > > min(..), so I assumed that one of the possibilities will be limitation
> > > of memcpy which I'm not aware.
> >
> > I understand that it is not such realistic that one of the structures in
> > union pvrdma_cmd_req will be more then PAGE_SIZE but any other alternative
> > to protect from such case?
> 
> BUILD_BUG_ON(...) is a perfect way to achieve it.

Thanks for your comments, Leon and Yuval! I will add that in v4 of the patch series.

--
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