[PATCH 00/21] IB/hfi1: First round of fixes for 4.7

[PATCH 00/21] IB/hfi1: First round of fixes for 4.7

[Dennis Dalessandro]

These patches do some clean ups and fix issues which may not be appropriate for
an RC so we are targeting these for 4.7. The current patch series applies on top
of the "Important bug fixes for 4.6 RC" patch set submitted just prior to this.

Patches can be viewed in my GitHub repo at:
https://github.com/ddalessa/kernel/tree/for-4.7.

---

Dean Luick (13):
      IB/hfi1: Fix sysfs file offset usage
      IB/hfi1: Fix i2c resource reservation checks
      IB/hfi1: Fix QOS num_vl bit width
      IB/hfi1: Remove invalid QOS check
      IB/hfi1: Fix QOS rule mappings
      IB/hfi1: Correctly obtain the full service class
      IB/hfi1: Simplify init_qpmap_table()
      IB/hfi1: Guard against concurrent I2C access across all chains
      IB/hfi1: Fix double QSFP resource acquire on cache refresh
      IB/hfi1: Extract RSM map table init from QOS
      IB/hfi1: Move QOS decision logic into its own function
      IB/hfi1: Create a routine to set a receive side mapping rule
      IB/hfi1: Add RSM rule for user FECN handling

Easwar Hariharan (2):
      IB/hfi1: Always turn on CDRs for low power QSFP modules
      IB/hfi1: Remove module presence check outside pre-LNI checks

Jianxin Xiong (1):
      IB/hfi1: Reduce kernel context pio buffer allocation

Jubin John (1):
      IB/hfi1: Change default number of user contexts

Mike Marciniszyn (2):
      IB/hfi1: Remove unreachable code
      IB/hfi1: Use global defines for upper bits in opcode

Sebastian Sanchez (2):
      IB/hfi1: Adjust default MTU to be 10KB
      IB/hfi1: Check P_KEY for all sent packets from user mode


 drivers/infiniband/sw/rdmavt/qp.c          |    6 
 drivers/staging/rdma/hfi1/affinity.c       |   93 +++--
 drivers/staging/rdma/hfi1/affinity.h       |   19 +
 drivers/staging/rdma/hfi1/chip.c           |  485 +++++++++++++++++++---------
 drivers/staging/rdma/hfi1/chip.h           |    1 
 drivers/staging/rdma/hfi1/chip_registers.h |    1 
 drivers/staging/rdma/hfi1/diag.c           |    3 
 drivers/staging/rdma/hfi1/driver.c         |    3 
 drivers/staging/rdma/hfi1/firmware.c       |    9 -
 drivers/staging/rdma/hfi1/hfi.h            |   11 -
 drivers/staging/rdma/hfi1/init.c           |    1 
 drivers/staging/rdma/hfi1/pio.c            |   52 ++-
 drivers/staging/rdma/hfi1/pio.h            |    4 
 drivers/staging/rdma/hfi1/platform.c       |   99 +++---
 drivers/staging/rdma/hfi1/qp.c             |    6 
 drivers/staging/rdma/hfi1/qsfp.c           |   58 ++-
 drivers/staging/rdma/hfi1/qsfp.h           |   15 -
 drivers/staging/rdma/hfi1/rc.c             |    2 
 drivers/staging/rdma/hfi1/sysfs.c          |    4 
 drivers/staging/rdma/hfi1/user_sdma.c      |    7 
 drivers/staging/rdma/hfi1/verbs.c          |  108 ++++--
 drivers/staging/rdma/hfi1/verbs.h          |    3 
 22 files changed, 639 insertions(+), 351 deletions(-)

-- 
-Denny
--
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 01/21] IB/hfi1: Fix sysfs file offset usage

[Dennis Dalessandro]

From: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Two sysfs files do not pay attention to the file offset when
reading data. Fix that.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Jubin John <jubin.john-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/sysfs.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/sysfs.c b/drivers/staging/rdma/hfi1/sysfs.c
index c7f1271..8cd6df8 100644
--- a/drivers/staging/rdma/hfi1/sysfs.c
+++ b/drivers/staging/rdma/hfi1/sysfs.c
@@ -84,7 +84,7 @@ static ssize_t read_cc_table_bin(struct file *filp, struct kobject *kobj,
 		rcu_read_unlock();
 		return -EINVAL;
 	}
-	memcpy(buf, &cc_state->cct, count);
+	memcpy(buf, (void *)&cc_state->cct + pos, count);
 	rcu_read_unlock();
 
 	return count;
@@ -131,7 +131,7 @@ static ssize_t read_cc_setting_bin(struct file *filp, struct kobject *kobj,
 		rcu_read_unlock();
 		return -EINVAL;
 	}
-	memcpy(buf, &cc_state->cong_setting, count);
+	memcpy(buf, (void *)&cc_state->cong_setting + pos, count);
 	rcu_read_unlock();
 
 	return count;

--
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 02/21] IB/hfi1: Fix i2c resource reservation checks

[Dennis Dalessandro]

From: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The i2c and qsfp read/write routines should check for the resource
reservation of the incoming argument target rather than the implicit
target of the hardware HFI.

Reviewed-by: Easwar Hariharan <easwar.hariharan-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Jubin John <jubin.john-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/qsfp.c |    8 ++++----
 1 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/qsfp.c b/drivers/staging/rdma/hfi1/qsfp.c
index 9ed1963..ac03d80 100644
--- a/drivers/staging/rdma/hfi1/qsfp.c
+++ b/drivers/staging/rdma/hfi1/qsfp.c
@@ -96,7 +96,7 @@ int i2c_write(struct hfi1_pportdata *ppd, u32 target, int i2c_addr, int offset,
 {
 	int ret;
 
-	if (!check_chip_resource(ppd->dd, qsfp_resource(ppd->dd), __func__))
+	if (!check_chip_resource(ppd->dd, i2c_target(target), __func__))
 		return -EACCES;
 
 	/* make sure the TWSI bus is in a sane state */
@@ -162,7 +162,7 @@ int i2c_read(struct hfi1_pportdata *ppd, u32 target, int i2c_addr, int offset,
 {
 	int ret;
 
-	if (!check_chip_resource(ppd->dd, qsfp_resource(ppd->dd), __func__))
+	if (!check_chip_resource(ppd->dd, i2c_target(target), __func__))
 		return -EACCES;
 
 	/* make sure the TWSI bus is in a sane state */
@@ -192,7 +192,7 @@ int qsfp_write(struct hfi1_pportdata *ppd, u32 target, int addr, void *bp,
 	int ret;
 	u8 page;
 
-	if (!check_chip_resource(ppd->dd, qsfp_resource(ppd->dd), __func__))
+	if (!check_chip_resource(ppd->dd, i2c_target(target), __func__))
 		return -EACCES;
 
 	/* make sure the TWSI bus is in a sane state */
@@ -276,7 +276,7 @@ int qsfp_read(struct hfi1_pportdata *ppd, u32 target, int addr, void *bp,
 	int ret;
 	u8 page;
 
-	if (!check_chip_resource(ppd->dd, qsfp_resource(ppd->dd), __func__))
+	if (!check_chip_resource(ppd->dd, i2c_target(target), __func__))
 		return -EACCES;
 
 	/* make sure the TWSI bus is in a sane state */

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

[PATCH 03/21] IB/hfi1: Fix QOS num_vl bit width

[Dennis Dalessandro]

From: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The bit width for num_vls, n, needs to be calculated based on
the pow2 rounded up of the number of vls.  Otherwise num_vls of 3,
5, 6, and 7 will have misplaced QOS RSM map entries.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/chip.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index 16eb653..f1efbe5 100644
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -13508,7 +13508,7 @@ static void init_qos(struct hfi1_devdata *dd, u32 first_ctxt)
 		goto bail;
 	qpns_per_vl = __roundup_pow_of_two(max_by_vl);
 	/* determine bits vl */
-	n = ilog2(num_vls);
+	n = ilog2(__roundup_pow_of_two(num_vls));
 	/* determine bits for qpn */
 	m = ilog2(qpns_per_vl);
 	if ((m + n) > 7)

--
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 04/21] IB/hfi1: Remove invalid QOS check

[Dennis Dalessandro]

From: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Remove an invalid compare of the number of QOS RSM map table entries
against the number of physical receive contexts.  The RSM map table
has its own size and has no relation to the number of physical receive
contexts.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/chip.c |    2 --
 1 files changed, 0 insertions(+), 2 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index f1efbe5..8b54d66 100644
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -13513,8 +13513,6 @@ static void init_qos(struct hfi1_devdata *dd, u32 first_ctxt)
 	m = ilog2(qpns_per_vl);
 	if ((m + n) > 7)
 		goto bail;
-	if (num_vls * qpns_per_vl > dd->chip_rcv_contexts)
-		goto bail;
 	rsmmap = kmalloc_array(NUM_MAP_REGS, sizeof(u64), GFP_KERNEL);
 	if (!rsmmap)
 		goto bail;

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

[PATCH 05/21] IB/hfi1: Fix QOS rule mappings

[Dennis Dalessandro]

From: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The QOS RSM rule mappings are off by one, referencing a kernel receive
context that does not exist.

Correctly start the QOS RSM map entries at FIRST_KERNEL_CONTEXT rather
than MIN_KERNEL_KCTXTS.  Remove the cruft that hid this.

Change the QP map table so all traffic not caught by QOS RSM goes to
the control context rather than the first QOS context.

Correct comments to match the actual code operation and intent.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/chip.c |   48 +++++++++++++++++---------------------
 1 files changed, 21 insertions(+), 27 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index 8b54d66..79dc29a 100644
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -12675,20 +12675,20 @@ static int set_up_context_variables(struct hfi1_devdata *dd)
 	unsigned ngroups;
 
 	/*
-	 * Kernel contexts: (to be fixed later):
-	 * - min or 2 or 1 context/numa
+	 * Kernel receive contexts:
+	 * - min of 2 or 1 context/numa (excluding control context)
 	 * - Context 0 - control context (VL15/multicast/error)
-	 * - Context 1 - default context
+	 * - Context 1 - first kernel context
+	 * - Context 2 - second kernel context
+	 * ...
 	 */
 	if (n_krcvqs)
 		/*
-		 * Don't count context 0 in n_krcvqs since
-		 * is isn't used for normal verbs traffic.
-		 *
-		 * krcvqs will reflect number of kernel
-		 * receive contexts above 0.
+		 * n_krcvqs is the sum of module parameter kernel receive
+		 * contexts, krcvqs[].  It does not include the control
+		 * context, so add that.
 		 */
-		num_kernel_contexts = n_krcvqs + MIN_KERNEL_KCTXTS - 1;
+		num_kernel_contexts = n_krcvqs + 1;
 	else
 		num_kernel_contexts = num_online_nodes() + 1;
 	num_kernel_contexts =
@@ -13473,22 +13473,17 @@ static void init_qpmap_table(struct hfi1_devdata *dd,
 /**
  * init_qos - init RX qos
  * @dd - device data
- * @first_context
- *
- * This routine initializes Rule 0 and the
- * RSM map table to implement qos.
  *
- * If all of the limit tests succeed,
- * qos is applied based on the array
- * interpretation of krcvqs where
- * entry 0 is VL0.
+ * This routine initializes Rule 0 and the RSM map table to implement
+ * quality of service (qos).
  *
- * The number of vl bits (n) and the number of qpn
- * bits (m) are computed to feed both the RSM map table
- * and the single rule.
+ * If all of the limit tests succeed, qos is applied based on the array
+ * interpretation of krcvqs where entry 0 is VL0.
  *
+ * The number of vl bits (n) and the number of qpn bits (m) are computed to
+ * feed both the RSM map table and the single rule.
  */
-static void init_qos(struct hfi1_devdata *dd, u32 first_ctxt)
+static void init_qos(struct hfi1_devdata *dd)
 {
 	u8 max_by_vl = 0;
 	unsigned qpns_per_vl, ctxt, i, qpn, n = 1, m;
@@ -13518,7 +13513,7 @@ static void init_qos(struct hfi1_devdata *dd, u32 first_ctxt)
 		goto bail;
 	memset(rsmmap, rxcontext, NUM_MAP_REGS * sizeof(u64));
 	/* init the local copy of the table */
-	for (i = 0, ctxt = first_ctxt; i < num_vls; i++) {
+	for (i = 0, ctxt = FIRST_KERNEL_KCTXT; i < num_vls; i++) {
 		unsigned tctxt;
 
 		for (qpn = 0, tctxt = ctxt;
@@ -13546,7 +13541,7 @@ static void init_qos(struct hfi1_devdata *dd, u32 first_ctxt)
 	/* add rule0 */
 	write_csr(dd, RCV_RSM_CFG /* + (8 * 0) */,
 		  RCV_RSM_CFG_ENABLE_OR_CHAIN_RSM0_MASK <<
-		  RCV_RSM_CFG_ENABLE_OR_CHAIN_RSM0_SHIFT |
+			RCV_RSM_CFG_ENABLE_OR_CHAIN_RSM0_SHIFT |
 		  2ull << RCV_RSM_CFG_PACKET_TYPE_SHIFT);
 	write_csr(dd, RCV_RSM_SELECT /* + (8 * 0) */,
 		  LRH_BTH_MATCH_OFFSET << RCV_RSM_SELECT_FIELD1_OFFSET_SHIFT |
@@ -13563,8 +13558,8 @@ static void init_qos(struct hfi1_devdata *dd, u32 first_ctxt)
 	/* Enable RSM */
 	add_rcvctrl(dd, RCV_CTRL_RCV_RSM_ENABLE_SMASK);
 	kfree(rsmmap);
-	/* map everything else to first context */
-	init_qpmap_table(dd, FIRST_KERNEL_KCTXT, MIN_KERNEL_KCTXTS - 1);
+	/* map everything else to the mcast/err/vl15 context */
+	init_qpmap_table(dd, HFI1_CTRL_CTXT, HFI1_CTRL_CTXT);
 	dd->qos_shift = n + 1;
 	return;
 bail:
@@ -13577,8 +13572,7 @@ static void init_rxe(struct hfi1_devdata *dd)
 	/* enable all receive errors */
 	write_csr(dd, RCV_ERR_MASK, ~0ull);
 	/* setup QPN map table - start where VL15 context leaves off */
-	init_qos(dd, dd->n_krcv_queues > MIN_KERNEL_KCTXTS ?
-		 MIN_KERNEL_KCTXTS : 0);
+	init_qos(dd);
 	/*
 	 * make sure RcvCtrl.RcvWcb <= PCIe Device Control
 	 * Register Max_Payload_Size (PCI_EXP_DEVCTL in Linux PCIe config

--
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 06/21] IB/hfi1: Correctly obtain the full service class

[Dennis Dalessandro]

From: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The function hdr2sc was using an unshifted mask to obtain
the 5th bit of the service class.  Correct the issue by using
the shifted mask.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/hfi.h |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/hfi.h b/drivers/staging/rdma/hfi1/hfi.h
index 16cbdc4..ac553f1 100644
--- a/drivers/staging/rdma/hfi1/hfi.h
+++ b/drivers/staging/rdma/hfi1/hfi.h
@@ -1258,7 +1258,7 @@ void receive_interrupt_work(struct work_struct *work);
 static inline int hdr2sc(struct hfi1_message_header *hdr, u64 rhf)
 {
 	return ((be16_to_cpu(hdr->lrh[0]) >> 12) & 0xf) |
-	       ((!!(rhf & RHF_DC_INFO_MASK)) << 4);
+	       ((!!(rhf & RHF_DC_INFO_SMASK)) << 4);
 }
 
 static inline u16 generate_jkey(kuid_t uid)

--
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 07/21] IB/hfi1: Simplify init_qpmap_table()

[Dennis Dalessandro]

From: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Make init_qpmap_table() easier to understand by simplifying
the loop indexing and writing each register when it is "full",
removing the need for a follow-on register write.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/chip.c |    7 ++-----
 1 files changed, 2 insertions(+), 5 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index 79dc29a..56753c6 100644
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -13451,20 +13451,17 @@ static void init_qpmap_table(struct hfi1_devdata *dd,
 	int i;
 	u64 ctxt = first_ctxt;
 
-	for (i = 0; i < 256;) {
+	for (i = 0; i < 256; i++) {
 		reg |= ctxt << (8 * (i % 8));
-		i++;
 		ctxt++;
 		if (ctxt > last_ctxt)
 			ctxt = first_ctxt;
-		if (i % 8 == 0) {
+		if (i % 8 == 7) {
 			write_csr(dd, regno, reg);
 			reg = 0;
 			regno += 8;
 		}
 	}
-	if (i % 8)
-		write_csr(dd, regno, reg);
 
 	add_rcvctrl(dd, RCV_CTRL_RCV_QP_MAP_ENABLE_SMASK
 			| RCV_CTRL_RCV_BYPASS_ENABLE_SMASK);

--
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] [PATCH 08/21] IB/hfi1: Adjust default MTU to be 10KB

[Dennis Dalessandro]

From: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Increasing the default MTU size to 10KB improves performance
for PSM. Change the default MTU to 10KB but constrain
Verbs MTU to 8KB. Also update default MTU module parameter
description to be HFI1_DEFAULT_MAX_MTU.

Reviewed-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Mitko Haralanov <mitko.haralanov-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Jubin John <jubin.john-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/driver.c |    3 ++-
 drivers/staging/rdma/hfi1/hfi.h    |    4 ++--
 drivers/staging/rdma/hfi1/qp.c     |    6 +++++-
 3 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/driver.c b/drivers/staging/rdma/hfi1/driver.c
index 34511e5..700c6fa 100644
--- a/drivers/staging/rdma/hfi1/driver.c
+++ b/drivers/staging/rdma/hfi1/driver.c
@@ -75,7 +75,8 @@ DEFINE_MUTEX(hfi1_mutex);	/* general driver use */
 
 unsigned int hfi1_max_mtu = HFI1_DEFAULT_MAX_MTU;
 module_param_named(max_mtu, hfi1_max_mtu, uint, S_IRUGO);
-MODULE_PARM_DESC(max_mtu, "Set max MTU bytes, default is 8192");
+MODULE_PARM_DESC(max_mtu, "Set max MTU bytes, default is " __stringify(
+		 HFI1_DEFAULT_MAX_MTU));
 
 unsigned int hfi1_cu = 1;
 module_param_named(cu, hfi1_cu, uint, S_IRUGO);
diff --git a/drivers/staging/rdma/hfi1/hfi.h b/drivers/staging/rdma/hfi1/hfi.h
index ac553f1..ff04593 100644
--- a/drivers/staging/rdma/hfi1/hfi.h
+++ b/drivers/staging/rdma/hfi1/hfi.h
@@ -455,9 +455,9 @@ struct rvt_sge_state;
 #define HLS_UP (HLS_UP_INIT | HLS_UP_ARMED | HLS_UP_ACTIVE)
 
 /* use this MTU size if none other is given */
-#define HFI1_DEFAULT_ACTIVE_MTU 8192
+#define HFI1_DEFAULT_ACTIVE_MTU 10240
 /* use this MTU size as the default maximum */
-#define HFI1_DEFAULT_MAX_MTU 8192
+#define HFI1_DEFAULT_MAX_MTU 10240
 /* default partition key */
 #define DEFAULT_PKEY 0xffff
 
diff --git a/drivers/staging/rdma/hfi1/qp.c b/drivers/staging/rdma/hfi1/qp.c
index 29a5ad2..e68d08a 100644
--- a/drivers/staging/rdma/hfi1/qp.c
+++ b/drivers/staging/rdma/hfi1/qp.c
@@ -167,8 +167,12 @@ static inline int opa_mtu_enum_to_int(int mtu)
  */
 static inline int verbs_mtu_enum_to_int(struct ib_device *dev, enum ib_mtu mtu)
 {
-	int val = opa_mtu_enum_to_int((int)mtu);
+	int val;
 
+	/* Constraining 10KB packets to 8KB packets */
+	if (mtu == (enum ib_mtu)OPA_MTU_10240)
+		mtu = OPA_MTU_8192;
+	val = opa_mtu_enum_to_int((int)mtu);
 	if (val > 0)
 		return val;
 	return ib_mtu_enum_to_int(mtu);

--
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 09/21] IB/hfi1: Check P_KEY for all sent packets from user mode

[Dennis Dalessandro]

From: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Add the P_KEY check for user-context mechanism for
both PIO and SDMA. For PIO, the
SendCtxtCheckEnable.DisallowKDETHPackets is set by
default. When the P_KEY is set,
SendCtxtCheckEnable.DisallowKDETHPackets is cleared.
For SDMA, a software check was included. This change
requires user processes to set the P_KEY before sending
any packets, otherwise, the sent packet will fail. The
original submission didn't have this check but it's
required.

Reviewed-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Mikto Haralanov <mitko.haralanov-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/chip.c      |    1 
 drivers/staging/rdma/hfi1/hfi.h       |    4 +
 drivers/staging/rdma/hfi1/user_sdma.c |    7 ++
 drivers/staging/rdma/hfi1/verbs.c     |  106 ++++++++++++++++++++-------------
 4 files changed, 77 insertions(+), 41 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index 56753c6..b0a0a0d 100644
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -13751,6 +13751,7 @@ int hfi1_set_ctxt_pkey(struct hfi1_devdata *dd, unsigned ctxt, u16 pkey)
 	write_kctxt_csr(dd, sctxt, SEND_CTXT_CHECK_PARTITION_KEY, reg);
 	reg = read_kctxt_csr(dd, sctxt, SEND_CTXT_CHECK_ENABLE);
 	reg |= SEND_CTXT_CHECK_ENABLE_CHECK_PARTITION_KEY_SMASK;
+	reg &= ~SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK;
 	write_kctxt_csr(dd, sctxt, SEND_CTXT_CHECK_ENABLE, reg);
 done:
 	return ret;
diff --git a/drivers/staging/rdma/hfi1/hfi.h b/drivers/staging/rdma/hfi1/hfi.h
index ff04593..b1d4f60 100644
--- a/drivers/staging/rdma/hfi1/hfi.h
+++ b/drivers/staging/rdma/hfi1/hfi.h
@@ -1333,6 +1333,9 @@ void process_becn(struct hfi1_pportdata *ppd, u8 sl,  u16 rlid, u32 lqpn,
 void return_cnp(struct hfi1_ibport *ibp, struct rvt_qp *qp, u32 remote_qpn,
 		u32 pkey, u32 slid, u32 dlid, u8 sc5,
 		const struct ib_grh *old_grh);
+#define PKEY_CHECK_INVALID -1
+int egress_pkey_check(struct hfi1_pportdata *ppd, __be16 *lrh, __be32 *bth,
+		      u8 sc5, int8_t s_pkey_index);
 
 #define PACKET_EGRESS_TIMEOUT 350
 static inline void pause_for_credit_return(struct hfi1_devdata *dd)
@@ -1776,6 +1779,7 @@ extern struct mutex hfi1_mutex;
 
 #define HFI1_PKT_USER_SC_INTEGRITY					    \
 	(SEND_CTXT_CHECK_ENABLE_DISALLOW_NON_KDETH_PACKETS_SMASK	    \
+	| SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK		\
 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_SMASK		    \
 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_GRH_SMASK)
 
diff --git a/drivers/staging/rdma/hfi1/user_sdma.c b/drivers/staging/rdma/hfi1/user_sdma.c
index 635ddf8..0014c9c 100644
--- a/drivers/staging/rdma/hfi1/user_sdma.c
+++ b/drivers/staging/rdma/hfi1/user_sdma.c
@@ -600,6 +600,13 @@ int hfi1_user_sdma_process_request(struct file *fp, struct iovec *iovec,
 		goto free_req;
 	}
 
+	/* Checking P_KEY for requests from user-space */
+	if (egress_pkey_check(dd->pport, req->hdr.lrh, req->hdr.bth, sc,
+			      PKEY_CHECK_INVALID)) {
+		ret = -EINVAL;
+		goto free_req;
+	}
+
 	/*
 	 * Also should check the BTH.lnh. If it says the next header is GRH then
 	 * the RXE parsing will be off and will land in the middle of the KDETH
diff --git a/drivers/staging/rdma/hfi1/verbs.c b/drivers/staging/rdma/hfi1/verbs.c
index 89f2aad..c56c0cb 100644
--- a/drivers/staging/rdma/hfi1/verbs.c
+++ b/drivers/staging/rdma/hfi1/verbs.c
@@ -1089,16 +1089,16 @@ bail:
 
 /*
  * egress_pkey_matches_entry - return 1 if the pkey matches ent (ent
- * being an entry from the ingress partition key table), return 0
+ * being an entry from the partition key table), return 0
  * otherwise. Use the matching criteria for egress partition keys
  * specified in the OPAv1 spec., section 9.1l.7.
  */
 static inline int egress_pkey_matches_entry(u16 pkey, u16 ent)
 {
 	u16 mkey = pkey & PKEY_LOW_15_MASK;
-	u16 ment = ent & PKEY_LOW_15_MASK;
+	u16 mentry = ent & PKEY_LOW_15_MASK;
 
-	if (mkey == ment) {
+	if (mkey == mentry) {
 		/*
 		 * If pkey[15] is set (full partition member),
 		 * is bit 15 in the corresponding table element
@@ -1111,32 +1111,32 @@ static inline int egress_pkey_matches_entry(u16 pkey, u16 ent)
 	return 0;
 }
 
-/*
- * egress_pkey_check - return 0 if hdr's pkey matches according to the
- * criteria in the OPAv1 spec., section 9.11.7.
+/**
+ * egress_pkey_check - check P_KEY of a packet
+ * @ppd:    Physical IB port data
+ * @lrh: Local route header
+ * @bth: Base transport header
+ * @sc5:    SC for packet
+ * @s_pkey_index: It will be used for look up optimization for kernel contexts
+ * only. If it is negative value, then it means user contexts is calling this
+ * function.
+ *
+ * It checks if hdr's pkey is valid.
+ *
+ * Return: 0 on success, otherwise, 1
  */
-static inline int egress_pkey_check(struct hfi1_pportdata *ppd,
-				    struct hfi1_ib_header *hdr,
-				    struct rvt_qp *qp)
+int egress_pkey_check(struct hfi1_pportdata *ppd, __be16 *lrh, __be32 *bth,
+		      u8 sc5, int8_t s_pkey_index)
 {
-	struct hfi1_qp_priv *priv = qp->priv;
-	struct hfi1_other_headers *ohdr;
 	struct hfi1_devdata *dd;
-	int i = 0;
+	int i;
 	u16 pkey;
-	u8 lnh, sc5 = priv->s_sc;
+	int is_user_ctxt_mechanism = (s_pkey_index < 0);
 
 	if (!(ppd->part_enforce & HFI1_PART_ENFORCE_OUT))
 		return 0;
 
-	/* locate the pkey within the headers */
-	lnh = be16_to_cpu(hdr->lrh[0]) & 3;
-	if (lnh == HFI1_LRH_GRH)
-		ohdr = &hdr->u.l.oth;
-	else
-		ohdr = &hdr->u.oth;
-
-	pkey = (u16)be32_to_cpu(ohdr->bth[0]);
+	pkey = (u16)be32_to_cpu(bth[0]);
 
 	/* If SC15, pkey[0:14] must be 0x7fff */
 	if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
@@ -1146,28 +1146,37 @@ static inline int egress_pkey_check(struct hfi1_pportdata *ppd,
 	if ((pkey & PKEY_LOW_15_MASK) == 0)
 		goto bad;
 
-	/* The most likely matching pkey has index qp->s_pkey_index */
-	if (unlikely(!egress_pkey_matches_entry(pkey,
-						ppd->pkeys
-						[qp->s_pkey_index]))) {
-		/* no match - try the entire table */
-		for (; i < MAX_PKEY_VALUES; i++) {
-			if (egress_pkey_matches_entry(pkey, ppd->pkeys[i]))
-				break;
-		}
+	/*
+	 * For the kernel contexts only, if a qp is passed into the function,
+	 * the most likely matching pkey has index qp->s_pkey_index
+	 */
+	if (!is_user_ctxt_mechanism &&
+	    egress_pkey_matches_entry(pkey, ppd->pkeys[s_pkey_index])) {
+		return 0;
 	}
 
-	if (i < MAX_PKEY_VALUES)
-		return 0;
+	for (i = 0; i < MAX_PKEY_VALUES; i++) {
+		if (egress_pkey_matches_entry(pkey, ppd->pkeys[i]))
+			return 0;
+	}
 bad:
-	incr_cntr64(&ppd->port_xmit_constraint_errors);
-	dd = ppd->dd;
-	if (!(dd->err_info_xmit_constraint.status & OPA_EI_STATUS_SMASK)) {
-		u16 slid = be16_to_cpu(hdr->lrh[3]);
-
-		dd->err_info_xmit_constraint.status |= OPA_EI_STATUS_SMASK;
-		dd->err_info_xmit_constraint.slid = slid;
-		dd->err_info_xmit_constraint.pkey = pkey;
+	/*
+	 * For the user-context mechanism, the P_KEY check would only happen
+	 * once per SDMA request, not once per packet.  Therefore, there's no
+	 * need to increment the counter for the user-context mechanism.
+	 */
+	if (!is_user_ctxt_mechanism) {
+		incr_cntr64(&ppd->port_xmit_constraint_errors);
+		dd = ppd->dd;
+		if (!(dd->err_info_xmit_constraint.status &
+		      OPA_EI_STATUS_SMASK)) {
+			u16 slid = be16_to_cpu(lrh[3]);
+
+			dd->err_info_xmit_constraint.status |=
+				OPA_EI_STATUS_SMASK;
+			dd->err_info_xmit_constraint.slid = slid;
+			dd->err_info_xmit_constraint.pkey = pkey;
+		}
 	}
 	return 1;
 }
@@ -1227,11 +1236,26 @@ int hfi1_verbs_send(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
 {
 	struct hfi1_devdata *dd = dd_from_ibdev(qp->ibqp.device);
 	struct hfi1_qp_priv *priv = qp->priv;
+	struct hfi1_other_headers *ohdr;
+	struct hfi1_ib_header *hdr;
 	send_routine sr;
 	int ret;
+	u8 lnh;
+
+	hdr = &ps->s_txreq->phdr.hdr;
+	/* locate the pkey within the headers */
+	lnh = be16_to_cpu(hdr->lrh[0]) & 3;
+	if (lnh == HFI1_LRH_GRH)
+		ohdr = &hdr->u.l.oth;
+	else
+		ohdr = &hdr->u.oth;
 
 	sr = get_send_routine(qp, ps->s_txreq);
-	ret = egress_pkey_check(dd->pport, &ps->s_txreq->phdr.hdr, qp);
+	ret = egress_pkey_check(dd->pport,
+				hdr->lrh,
+				ohdr->bth,
+				priv->s_sc,
+				qp->s_pkey_index);
 	if (unlikely(ret)) {
 		/*
 		 * The value we are returning here does not get propagated to

--
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 10/21] IB/hfi1: Always turn on CDRs for low power QSFP modules

[Dennis Dalessandro]

From: Easwar Hariharan <easwar.hariharan-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Clock and data recovery mechanisms (CDRs) in active QSFP modules
can be turned on or off to improve the bit error rate observed on
the channel. Signal integrity and bit error rate requirements require
us to always turn on any CDRs present in low power cables (power
dissipation 2.5W or lower). However, we adhere to the platform
designer's settings (provided in the platform configuration) for
higher power cables (dissipation 3.5W or higher) if the platform
designer has determined that the platform requires the CDRs to be
turned on (or off) and is capable of supplying and cooling the higher
power modules.

This patch also introduces the get_qsfp_power_class function to
centralize the bit twiddling required to determine the QSFP power class
across the code. Reusing this function improves the readability of code
that depends on knowing the power class of the cable, such as the
active and optical channel tuning algorithm.

Reviewed-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Easwar Hariharan <easwar.hariharan-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/chip.c     |   49 +++------------------------------
 drivers/staging/rdma/hfi1/chip.h     |    1 -
 drivers/staging/rdma/hfi1/hfi.h      |    1 -
 drivers/staging/rdma/hfi1/init.c     |    1 -
 drivers/staging/rdma/hfi1/platform.c |   51 +++++++++++++++++++---------------
 drivers/staging/rdma/hfi1/qsfp.c     |   38 +++++++++++++++++++++++--
 drivers/staging/rdma/hfi1/qsfp.h     |   15 +++++-----
 7 files changed, 74 insertions(+), 82 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index b0a0a0d..962cca0 100644
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -6199,18 +6199,13 @@ static void hreq_response(struct hfi1_devdata *dd, u8 return_code, u16 rsp_data)
 
 /*
  * Handle host requests from the 8051.
- *
- * This is a work-queue function outside of the interrupt.
  */
-void handle_8051_request(struct work_struct *work)
+static void handle_8051_request(struct hfi1_pportdata *ppd)
 {
-	struct hfi1_pportdata *ppd = container_of(work, struct hfi1_pportdata,
-							dc_host_req_work);
 	struct hfi1_devdata *dd = ppd->dd;
 	u64 reg;
 	u16 data = 0;
-	u8 type, i, lanes, *cache = ppd->qsfp_info.cache;
-	u8 cdr_ctrl_byte = cache[QSFP_CDR_CTRL_BYTE_OFFS];
+	u8 type;
 
 	reg = read_csr(dd, DC_DC8051_CFG_EXT_DEV_1);
 	if ((reg & DC_DC8051_CFG_EXT_DEV_1_REQ_NEW_SMASK) == 0)
@@ -6231,46 +6226,11 @@ void handle_8051_request(struct work_struct *work)
 	case HREQ_READ_CONFIG:
 	case HREQ_SET_TX_EQ_ABS:
 	case HREQ_SET_TX_EQ_REL:
+	case HREQ_ENABLE:
 		dd_dev_info(dd, "8051 request: request 0x%x not supported\n",
 			    type);
 		hreq_response(dd, HREQ_NOT_SUPPORTED, 0);
 		break;
-
-	case HREQ_ENABLE:
-		lanes = data & 0xF;
-		for (i = 0; lanes; lanes >>= 1, i++) {
-			if (!(lanes & 1))
-				continue;
-			if (data & 0x200) {
-				/* enable TX CDR */
-				if (cache[QSFP_MOD_PWR_OFFS] & 0x8 &&
-				    cache[QSFP_CDR_INFO_OFFS] & 0x80)
-					cdr_ctrl_byte |= (1 << (i + 4));
-			} else {
-				/* disable TX CDR */
-				if (cache[QSFP_MOD_PWR_OFFS] & 0x8 &&
-				    cache[QSFP_CDR_INFO_OFFS] & 0x80)
-					cdr_ctrl_byte &= ~(1 << (i + 4));
-			}
-
-			if (data & 0x800) {
-				/* enable RX CDR */
-				if (cache[QSFP_MOD_PWR_OFFS] & 0x4 &&
-				    cache[QSFP_CDR_INFO_OFFS] & 0x40)
-					cdr_ctrl_byte |= (1 << i);
-			} else {
-				/* disable RX CDR */
-				if (cache[QSFP_MOD_PWR_OFFS] & 0x4 &&
-				    cache[QSFP_CDR_INFO_OFFS] & 0x40)
-					cdr_ctrl_byte &= ~(1 << i);
-			}
-		}
-		one_qsfp_write(ppd, dd->hfi1_id, QSFP_CDR_CTRL_BYTE_OFFS,
-			       &cdr_ctrl_byte, 1);
-		hreq_response(dd, HREQ_SUCCESS, data);
-		refresh_qsfp_cache(ppd, &ppd->qsfp_info);
-		break;
-
 	case HREQ_CONFIG_DONE:
 		hreq_response(dd, HREQ_SUCCESS, 0);
 		break;
@@ -6278,7 +6238,6 @@ void handle_8051_request(struct work_struct *work)
 	case HREQ_INTERFACE_TEST:
 		hreq_response(dd, HREQ_SUCCESS, data);
 		break;
-
 	default:
 		dd_dev_err(dd, "8051 request: unknown request 0x%x\n", type);
 		hreq_response(dd, HREQ_NOT_SUPPORTED, 0);
@@ -7534,7 +7493,7 @@ static void handle_8051_interrupt(struct hfi1_devdata *dd, u32 unused, u64 reg)
 			host_msg &= ~(u64)LINKUP_ACHIEVED;
 		}
 		if (host_msg & EXT_DEVICE_CFG_REQ) {
-			queue_work(ppd->hfi1_wq, &ppd->dc_host_req_work);
+			handle_8051_request(ppd);
 			host_msg &= ~(u64)EXT_DEVICE_CFG_REQ;
 		}
 		if (host_msg & VERIFY_CAP_FRAME) {
diff --git a/drivers/staging/rdma/hfi1/chip.h b/drivers/staging/rdma/hfi1/chip.h
index 4f3b878..e02e006 100644
--- a/drivers/staging/rdma/hfi1/chip.h
+++ b/drivers/staging/rdma/hfi1/chip.h
@@ -691,7 +691,6 @@ void handle_verify_cap(struct work_struct *work);
 void handle_freeze(struct work_struct *work);
 void handle_link_up(struct work_struct *work);
 void handle_link_down(struct work_struct *work);
-void handle_8051_request(struct work_struct *work);
 void handle_link_downgrade(struct work_struct *work);
 void handle_link_bounce(struct work_struct *work);
 void handle_sma_message(struct work_struct *work);
diff --git a/drivers/staging/rdma/hfi1/hfi.h b/drivers/staging/rdma/hfi1/hfi.h
index b1d4f60..7b78d56 100644
--- a/drivers/staging/rdma/hfi1/hfi.h
+++ b/drivers/staging/rdma/hfi1/hfi.h
@@ -606,7 +606,6 @@ struct hfi1_pportdata {
 	struct work_struct link_vc_work;
 	struct work_struct link_up_work;
 	struct work_struct link_down_work;
-	struct work_struct dc_host_req_work;
 	struct work_struct sma_message_work;
 	struct work_struct freeze_work;
 	struct work_struct link_downgrade_work;
diff --git a/drivers/staging/rdma/hfi1/init.c b/drivers/staging/rdma/hfi1/init.c
index cfcdc16..991a49d 100644
--- a/drivers/staging/rdma/hfi1/init.c
+++ b/drivers/staging/rdma/hfi1/init.c
@@ -496,7 +496,6 @@ void hfi1_init_pportdata(struct pci_dev *pdev, struct hfi1_pportdata *ppd,
 	INIT_WORK(&ppd->link_vc_work, handle_verify_cap);
 	INIT_WORK(&ppd->link_up_work, handle_link_up);
 	INIT_WORK(&ppd->link_down_work, handle_link_down);
-	INIT_WORK(&ppd->dc_host_req_work, handle_8051_request);
 	INIT_WORK(&ppd->freeze_work, handle_freeze);
 	INIT_WORK(&ppd->link_downgrade_work, handle_link_downgrade);
 	INIT_WORK(&ppd->sma_message_work, handle_sma_message);
diff --git a/drivers/staging/rdma/hfi1/platform.c b/drivers/staging/rdma/hfi1/platform.c
index 0a1d074..19d0223 100644
--- a/drivers/staging/rdma/hfi1/platform.c
+++ b/drivers/staging/rdma/hfi1/platform.c
@@ -114,21 +114,11 @@ static int qual_power(struct hfi1_pportdata *ppd)
 	if (ret)
 		return ret;
 
-	if (QSFP_HIGH_PWR(cache[QSFP_MOD_PWR_OFFS]) != 4)
-		cable_power_class = QSFP_HIGH_PWR(cache[QSFP_MOD_PWR_OFFS]);
-	else
-		cable_power_class = QSFP_PWR(cache[QSFP_MOD_PWR_OFFS]);
+	cable_power_class = get_qsfp_power_class(cache[QSFP_MOD_PWR_OFFS]);
 
-	if (cable_power_class <= 3 && cable_power_class > (power_class_max - 1))
-		ppd->offline_disabled_reason =
-			HFI1_ODR_MASK(OPA_LINKDOWN_REASON_POWER_POLICY);
-	else if (cable_power_class > 4 && cable_power_class > (power_class_max))
+	if (cable_power_class > power_class_max)
 		ppd->offline_disabled_reason =
 			HFI1_ODR_MASK(OPA_LINKDOWN_REASON_POWER_POLICY);
-	/*
-	 * cable_power_class will never have value 4 as this simply
-	 * means the high power settings are unused
-	 */
 
 	if (ppd->offline_disabled_reason ==
 			HFI1_ODR_MASK(OPA_LINKDOWN_REASON_POWER_POLICY)) {
@@ -173,12 +163,9 @@ static int set_qsfp_high_power(struct hfi1_pportdata *ppd)
 	u8 *cache = ppd->qsfp_info.cache;
 	int ret;
 
-	if (QSFP_HIGH_PWR(cache[QSFP_MOD_PWR_OFFS]) != 4)
-		cable_power_class = QSFP_HIGH_PWR(cache[QSFP_MOD_PWR_OFFS]);
-	else
-		cable_power_class = QSFP_PWR(cache[QSFP_MOD_PWR_OFFS]);
+	cable_power_class = get_qsfp_power_class(cache[QSFP_MOD_PWR_OFFS]);
 
-	if (cable_power_class) {
+	if (cable_power_class > QSFP_POWER_CLASS_1) {
 		power_ctrl_byte = cache[QSFP_PWR_CTRL_BYTE_OFFS];
 
 		power_ctrl_byte |= 1;
@@ -190,8 +177,7 @@ static int set_qsfp_high_power(struct hfi1_pportdata *ppd)
 		if (ret != 1)
 			return -EIO;
 
-		if (cable_power_class > 3) {
-			/* > power class 4*/
+		if (cable_power_class > QSFP_POWER_CLASS_4) {
 			power_ctrl_byte |= (1 << 2);
 			ret = qsfp_write(ppd, ppd->dd->hfi1_id,
 					 QSFP_PWR_CTRL_BYTE_OFFS,
@@ -212,12 +198,21 @@ static void apply_rx_cdr(struct hfi1_pportdata *ppd,
 {
 	u32 rx_preset;
 	u8 *cache = ppd->qsfp_info.cache;
+	int cable_power_class;
 
 	if (!((cache[QSFP_MOD_PWR_OFFS] & 0x4) &&
 	      (cache[QSFP_CDR_INFO_OFFS] & 0x40)))
 		return;
 
-	/* rx_preset preset to zero to catch error */
+	/* RX CDR present, bypass supported */
+	cable_power_class = get_qsfp_power_class(cache[QSFP_MOD_PWR_OFFS]);
+
+	if (cable_power_class <= QSFP_POWER_CLASS_3) {
+		/* Power class <= 3, ignore config & turn RX CDR on */
+		*cdr_ctrl_byte |= 0xF;
+		return;
+	}
+
 	get_platform_config_field(
 		ppd->dd, PLATFORM_CONFIG_RX_PRESET_TABLE,
 		rx_preset_index, RX_PRESET_TABLE_QSFP_RX_CDR_APPLY,
@@ -250,15 +245,25 @@ static void apply_rx_cdr(struct hfi1_pportdata *ppd,
 
 static void apply_tx_cdr(struct hfi1_pportdata *ppd,
 			 u32 tx_preset_index,
-			 u8 *ctr_ctrl_byte)
+			 u8 *cdr_ctrl_byte)
 {
 	u32 tx_preset;
 	u8 *cache = ppd->qsfp_info.cache;
+	int cable_power_class;
 
 	if (!((cache[QSFP_MOD_PWR_OFFS] & 0x8) &&
 	      (cache[QSFP_CDR_INFO_OFFS] & 0x80)))
 		return;
 
+	/* TX CDR present, bypass supported */
+	cable_power_class = get_qsfp_power_class(cache[QSFP_MOD_PWR_OFFS]);
+
+	if (cable_power_class <= QSFP_POWER_CLASS_3) {
+		/* Power class <= 3, ignore config & turn TX CDR on */
+		*cdr_ctrl_byte |= 0xF0;
+		return;
+	}
+
 	get_platform_config_field(
 		ppd->dd,
 		PLATFORM_CONFIG_TX_PRESET_TABLE, tx_preset_index,
@@ -282,10 +287,10 @@ static void apply_tx_cdr(struct hfi1_pportdata *ppd,
 			(tx_preset << 2) | (tx_preset << 3));
 
 	if (tx_preset)
-		*ctr_ctrl_byte |= (tx_preset << 4);
+		*cdr_ctrl_byte |= (tx_preset << 4);
 	else
 		/* Preserve current/determined RX CDR status */
-		*ctr_ctrl_byte &= ((tx_preset << 4) | 0xF);
+		*cdr_ctrl_byte &= ((tx_preset << 4) | 0xF);
 }
 
 static void apply_cdr_settings(
diff --git a/drivers/staging/rdma/hfi1/qsfp.c b/drivers/staging/rdma/hfi1/qsfp.c
index ac03d80..dc5d186 100644
--- a/drivers/staging/rdma/hfi1/qsfp.c
+++ b/drivers/staging/rdma/hfi1/qsfp.c
@@ -466,7 +466,28 @@ const char * const hfi1_qsfp_devtech[16] = {
 #define QSFP_DUMP_CHUNK 16 /* Holds longest string */
 #define QSFP_DEFAULT_HDR_CNT 224
 
-static const char *pwr_codes = "1.5W2.0W2.5W3.5W";
+#define QSFP_PWR(pbyte) (((pbyte) >> 6) & 3)
+#define QSFP_HIGH_PWR(pbyte) ((pbyte) & 3)
+/* For use with QSFP_HIGH_PWR macro */
+#define QSFP_HIGH_PWR_UNUSED	0 /* Bits [1:0] = 00 implies low power module */
+
+/*
+ * Takes power class byte [Page 00 Byte 129] in SFF 8636
+ * Returns power class as integer (1 through 7, per SFF 8636 rev 2.4)
+ */
+int get_qsfp_power_class(u8 power_byte)
+{
+	if (QSFP_HIGH_PWR(power_byte) == QSFP_HIGH_PWR_UNUSED)
+		/* power classes count from 1, their bit encodings from 0 */
+		return (QSFP_PWR(power_byte) + 1);
+	/*
+	 * 00 in the high power classes stands for unused, bringing
+	 * balance to the off-by-1 offset above, we add 4 here to
+	 * account for the difference between the low and high power
+	 * groups
+	 */
+	return (QSFP_HIGH_PWR(power_byte) + 4);
+}
 
 int qsfp_mod_present(struct hfi1_pportdata *ppd)
 {
@@ -537,6 +558,16 @@ set_zeroes:
 	return ret;
 }
 
+static const char *pwr_codes[8] = {"N/AW",
+				  "1.5W",
+				  "2.0W",
+				  "2.5W",
+				  "3.5W",
+				  "4.0W",
+				  "4.5W",
+				  "5.0W"
+				 };
+
 int qsfp_dump(struct hfi1_pportdata *ppd, char *buf, int len)
 {
 	u8 *cache = &ppd->qsfp_info.cache[0];
@@ -546,6 +577,7 @@ int qsfp_dump(struct hfi1_pportdata *ppd, char *buf, int len)
 	int bidx = 0;
 	u8 *atten = &cache[QSFP_ATTEN_OFFS];
 	u8 *vendor_oui = &cache[QSFP_VOUI_OFFS];
+	u8 power_byte = 0;
 
 	sofar = 0;
 	lenstr[0] = ' ';
@@ -555,9 +587,9 @@ int qsfp_dump(struct hfi1_pportdata *ppd, char *buf, int len)
 		if (QSFP_IS_CU(cache[QSFP_MOD_TECH_OFFS]))
 			sprintf(lenstr, "%dM ", cache[QSFP_MOD_LEN_OFFS]);
 
+		power_byte = cache[QSFP_MOD_PWR_OFFS];
 		sofar += scnprintf(buf + sofar, len - sofar, "PWR:%.3sW\n",
-				pwr_codes +
-				(QSFP_PWR(cache[QSFP_MOD_PWR_OFFS]) * 4));
+				pwr_codes[get_qsfp_power_class(power_byte)]);
 
 		sofar += scnprintf(buf + sofar, len - sofar, "TECH:%s%s\n",
 				lenstr,
diff --git a/drivers/staging/rdma/hfi1/qsfp.h b/drivers/staging/rdma/hfi1/qsfp.h
index 831fe4c..dadc66c 100644
--- a/drivers/staging/rdma/hfi1/qsfp.h
+++ b/drivers/staging/rdma/hfi1/qsfp.h
@@ -82,8 +82,9 @@
 /* Byte 128 is Identifier: must be 0x0c for QSFP, or 0x0d for QSFP+ */
 #define QSFP_MOD_ID_OFFS 128
 /*
- * Byte 129 is "Extended Identifier". We only care about D7,D6: Power class
- *  0:1.5W, 1:2.0W, 2:2.5W, 3:3.5W
+ * Byte 129 is "Extended Identifier".
+ * For bits [7:6]: 0:1.5W, 1:2.0W, 2:2.5W, 3:3.5W
+ * For bits [1:0]: 0:Unused, 1:4W, 2:4.5W, 3:5W
  */
 #define QSFP_MOD_PWR_OFFS 129
 /* Byte 130 is Connector type. Not Intel req'd */
@@ -190,6 +191,9 @@ extern const char *const hfi1_qsfp_devtech[16];
 #define QSFP_HIGH_BIAS_WARNING		0x22
 #define QSFP_LOW_BIAS_WARNING		0x11
 
+#define QSFP_ATTEN_SDR(attenarray) (attenarray[0])
+#define QSFP_ATTEN_DDR(attenarray) (attenarray[1])
+
 /*
  * struct qsfp_data encapsulates state of QSFP device for one port.
  * it will be part of port-specific data if a board supports QSFP.
@@ -201,12 +205,6 @@ extern const char *const hfi1_qsfp_devtech[16];
  * and let the qsfp_lock arbitrate access to common resources.
  *
  */
-
-#define QSFP_PWR(pbyte) (((pbyte) >> 6) & 3)
-#define QSFP_HIGH_PWR(pbyte) (((pbyte) & 3) | 4)
-#define QSFP_ATTEN_SDR(attenarray) (attenarray[0])
-#define QSFP_ATTEN_DDR(attenarray) (attenarray[1])
-
 struct qsfp_data {
 	/* Helps to find our way */
 	struct hfi1_pportdata *ppd;
@@ -223,6 +221,7 @@ struct qsfp_data {
 
 int refresh_qsfp_cache(struct hfi1_pportdata *ppd,
 		       struct qsfp_data *cp);
+int get_qsfp_power_class(u8 power_byte);
 int qsfp_mod_present(struct hfi1_pportdata *ppd);
 int get_cable_info(struct hfi1_devdata *dd, u32 port_num, u32 addr,
 		   u32 len, u8 *data);

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

[PATCH 11/21] IB/hfi1: Remove module presence check outside pre-LNI checks

[Dennis Dalessandro]

From: Easwar Hariharan <easwar.hariharan-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The pre-LNI SerDes and channel tuning algorithm already checks for
module presence assertion for the relevant port types. The extraneous
check removed in this patch blocks link up for port types for which
the module presence assertion is not relevant.

Reviewed-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Easwar Hariharan <easwar.hariharan-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/chip.c     |   20 ++++++--------------
 drivers/staging/rdma/hfi1/platform.c |    8 ++++++--
 2 files changed, 12 insertions(+), 16 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index 962cca0..38c0e97 100644
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -6849,7 +6849,7 @@ void handle_link_down(struct work_struct *work)
 	 * If there is no cable attached, turn the DC off. Otherwise,
 	 * start the link bring up.
 	 */
-	if (!qsfp_mod_present(ppd)) {
+	if (ppd->port_type == PORT_TYPE_QSFP && !qsfp_mod_present(ppd)) {
 		dc_shutdown(ppd->dd);
 	} else {
 		tune_serdes(ppd);
@@ -9008,9 +9008,9 @@ set_local_link_attributes_fail:
 }
 
 /*
- * Call this to start the link.  Schedule a retry if the cable is not
- * present or if unable to start polling.  Do not do anything if the
- * link is disabled.  Returns 0 if link is disabled or moved to polling
+ * Call this to start the link.
+ * Do not do anything if the link is disabled.
+ * Returns 0 if link is disabled, moved to polling, or the driver is not ready.
  */
 int start_link(struct hfi1_pportdata *ppd)
 {
@@ -9027,15 +9027,7 @@ int start_link(struct hfi1_pportdata *ppd)
 		return 0;
 	}
 
-	if (qsfp_mod_present(ppd) || loopback == LOOPBACK_SERDES ||
-	    loopback == LOOPBACK_LCB ||
-	    ppd->dd->icode == ICODE_FUNCTIONAL_SIMULATOR)
-		return set_link_state(ppd, HLS_DN_POLL);
-
-	dd_dev_info(ppd->dd,
-		    "%s: stopping link start because no cable is present\n",
-		    __func__);
-	return -EAGAIN;
+	return set_link_state(ppd, HLS_DN_POLL);
 }
 
 static void wait_for_qsfp_init(struct hfi1_pportdata *ppd)
@@ -9206,7 +9198,7 @@ static int handle_qsfp_error_conditions(struct hfi1_pportdata *ppd,
 	return 0;
 }
 
-/* This routine will only be scheduled if the QSFP module is present */
+/* This routine will only be scheduled if the QSFP module present is asserted */
 void qsfp_event(struct work_struct *work)
 {
 	struct qsfp_data *qd;
diff --git a/drivers/staging/rdma/hfi1/platform.c b/drivers/staging/rdma/hfi1/platform.c
index 19d0223..b97027d 100644
--- a/drivers/staging/rdma/hfi1/platform.c
+++ b/drivers/staging/rdma/hfi1/platform.c
@@ -838,9 +838,11 @@ void tune_serdes(struct hfi1_pportdata *ppd)
 			total_atten = platform_atten + remote_atten;
 
 			tuning_method = OPA_PASSIVE_TUNING;
-		} else
+		} else {
 			ppd->offline_disabled_reason =
 			     HFI1_ODR_MASK(OPA_LINKDOWN_REASON_CHASSIS_CONFIG);
+			goto bail;
+		}
 		break;
 	case PORT_TYPE_QSFP:
 		if (qsfp_mod_present(ppd)) {
@@ -869,10 +871,12 @@ void tune_serdes(struct hfi1_pportdata *ppd)
 					   __func__);
 				goto bail;
 			}
-		} else
+		} else {
 			ppd->offline_disabled_reason =
 			   HFI1_ODR_MASK(
 				OPA_LINKDOWN_REASON_LOCAL_MEDIA_NOT_INSTALLED);
+			goto bail;
+		}
 		break;
 	default:
 		dd_dev_info(ppd->dd, "%s: Unknown port type\n", __func__);

--
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 12/21] IB/hfi1: Guard against concurrent I2C access across all chains

[Dennis Dalessandro]

From: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The discrete ASIC board design makes the two I2C chains not
independent of each other.  That is, only one chain can safely
be accessed at a time.  For discrete ASIC devices, adjust the
resource locking so that access to one I2C chain will lock both
of the chains.

Reviewed-by: Easwar Hariharan <easwar.hariharan-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/firmware.c |    9 ++++++++-
 1 files changed, 8 insertions(+), 1 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/firmware.c b/drivers/staging/rdma/hfi1/firmware.c
index 3040162..ed680fd 100644
--- a/drivers/staging/rdma/hfi1/firmware.c
+++ b/drivers/staging/rdma/hfi1/firmware.c
@@ -1413,8 +1413,15 @@ static int __acquire_chip_resource(struct hfi1_devdata *dd, u32 resource)
 
 	if (resource & CR_DYN_MASK) {
 		/* a dynamic resource is in use if either HFI has set the bit */
-		all_bits = resource_mask(0, resource) |
+		if (dd->pcidev->device == PCI_DEVICE_ID_INTEL0 &&
+		    (resource & (CR_I2C1 | CR_I2C2))) {
+			/* discrete devices must serialize across both chains */
+			all_bits = resource_mask(0, CR_I2C1 | CR_I2C2) |
+					resource_mask(1, CR_I2C1 | CR_I2C2);
+		} else {
+			all_bits = resource_mask(0, resource) |
 						resource_mask(1, resource);
+		}
 		my_bit = resource_mask(dd->hfi1_id, resource);
 	} else {
 		/* non-dynamic resources are not split between HFIs */

--
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 13/21] IB/hfi1: Fix double QSFP resource acquire on cache refresh

[Dennis Dalessandro]

From: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The function refresh_qsfp_cache() acquires the i2c chain resource,
but one caller already holds the resource.  Change the acquire so
all calls to refresh_qsfp_cache() are covered by the acquire and
remove the acquire within refresh_qsfp_cache().

Reviewed-by: Easwar Hariharan <easwar.hariharan-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/platform.c |   40 +++++++++++++++++-----------------
 drivers/staging/rdma/hfi1/qsfp.c     |   12 +++-------
 2 files changed, 23 insertions(+), 29 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/platform.c b/drivers/staging/rdma/hfi1/platform.c
index b97027d..8fe8a20 100644
--- a/drivers/staging/rdma/hfi1/platform.c
+++ b/drivers/staging/rdma/hfi1/platform.c
@@ -603,6 +603,7 @@ static void apply_tunings(
 		       "Applying TX settings");
 }
 
+/* Must be holding the QSFP i2c resource */
 static int tune_active_qsfp(struct hfi1_pportdata *ppd, u32 *ptr_tx_preset,
 			    u32 *ptr_rx_preset, u32 *ptr_total_atten)
 {
@@ -610,26 +611,19 @@ static int tune_active_qsfp(struct hfi1_pportdata *ppd, u32 *ptr_tx_preset,
 	u16 lss = ppd->link_speed_supported, lse = ppd->link_speed_enabled;
 	u8 *cache = ppd->qsfp_info.cache;
 
-	ret = acquire_chip_resource(ppd->dd, qsfp_resource(ppd->dd), QSFP_WAIT);
-	if (ret) {
-		dd_dev_err(ppd->dd, "%s: hfi%d: cannot lock i2c chain\n",
-			   __func__, (int)ppd->dd->hfi1_id);
-		return ret;
-	}
-
 	ppd->qsfp_info.limiting_active = 1;
 
 	ret = set_qsfp_tx(ppd, 0);
 	if (ret)
-		goto bail_unlock;
+		return ret;
 
 	ret = qual_power(ppd);
 	if (ret)
-		goto bail_unlock;
+		return ret;
 
 	ret = qual_bitrate(ppd);
 	if (ret)
-		goto bail_unlock;
+		return ret;
 
 	if (ppd->qsfp_info.reset_needed) {
 		reset_qsfp(ppd);
@@ -641,7 +635,7 @@ static int tune_active_qsfp(struct hfi1_pportdata *ppd, u32 *ptr_tx_preset,
 
 	ret = set_qsfp_high_power(ppd);
 	if (ret)
-		goto bail_unlock;
+		return ret;
 
 	if (cache[QSFP_EQ_INFO_OFFS] & 0x4) {
 		ret = get_platform_config_field(
@@ -651,7 +645,7 @@ static int tune_active_qsfp(struct hfi1_pportdata *ppd, u32 *ptr_tx_preset,
 			ptr_tx_preset, 4);
 		if (ret) {
 			*ptr_tx_preset = OPA_INVALID_INDEX;
-			goto bail_unlock;
+			return ret;
 		}
 	} else {
 		ret = get_platform_config_field(
@@ -661,7 +655,7 @@ static int tune_active_qsfp(struct hfi1_pportdata *ppd, u32 *ptr_tx_preset,
 			ptr_tx_preset, 4);
 		if (ret) {
 			*ptr_tx_preset = OPA_INVALID_INDEX;
-			goto bail_unlock;
+			return ret;
 		}
 	}
 
@@ -670,7 +664,7 @@ static int tune_active_qsfp(struct hfi1_pportdata *ppd, u32 *ptr_tx_preset,
 		PORT_TABLE_RX_PRESET_IDX, ptr_rx_preset, 4);
 	if (ret) {
 		*ptr_rx_preset = OPA_INVALID_INDEX;
-		goto bail_unlock;
+		return ret;
 	}
 
 	if ((lss & OPA_LINK_SPEED_25G) && (lse & OPA_LINK_SPEED_25G))
@@ -690,8 +684,6 @@ static int tune_active_qsfp(struct hfi1_pportdata *ppd, u32 *ptr_tx_preset,
 
 	ret = set_qsfp_tx(ppd, 1);
 
-bail_unlock:
-	release_chip_resource(ppd->dd, qsfp_resource(ppd->dd));
 	return ret;
 }
 
@@ -846,6 +838,14 @@ void tune_serdes(struct hfi1_pportdata *ppd)
 		break;
 	case PORT_TYPE_QSFP:
 		if (qsfp_mod_present(ppd)) {
+			ret = acquire_chip_resource(ppd->dd,
+						    qsfp_resource(ppd->dd),
+						    QSFP_WAIT);
+			if (ret) {
+				dd_dev_err(ppd->dd, "%s: hfi%d: cannot lock i2c chain\n",
+					   __func__, (int)ppd->dd->hfi1_id);
+				goto bail;
+			}
 			refresh_qsfp_cache(ppd, &ppd->qsfp_info);
 
 			if (ppd->qsfp_info.cache_valid) {
@@ -860,17 +860,17 @@ void tune_serdes(struct hfi1_pportdata *ppd)
 				 * update the cache to reflect the changes
 				 */
 				refresh_qsfp_cache(ppd, &ppd->qsfp_info);
-				if (ret)
-					goto bail;
-
 				limiting_active =
 						ppd->qsfp_info.limiting_active;
 			} else {
 				dd_dev_err(dd,
 					   "%s: Reading QSFP memory failed\n",
 					   __func__);
-				goto bail;
+				ret = -EINVAL; /* a fail indication */
 			}
+			release_chip_resource(ppd->dd, qsfp_resource(ppd->dd));
+			if (ret)
+				goto bail;
 		} else {
 			ppd->offline_disabled_reason =
 			   HFI1_ODR_MASK(
diff --git a/drivers/staging/rdma/hfi1/qsfp.c b/drivers/staging/rdma/hfi1/qsfp.c
index dc5d186..2441669 100644
--- a/drivers/staging/rdma/hfi1/qsfp.c
+++ b/drivers/staging/rdma/hfi1/qsfp.c
@@ -355,6 +355,8 @@ int one_qsfp_read(struct hfi1_pportdata *ppd, u32 target, int addr, void *bp,
  * The calls to qsfp_{read,write} in this function correctly handle the
  * address map difference between this mapping and the mapping implemented
  * by those functions
+ *
+ * The caller must be holding the QSFP i2c chain resource.
  */
 int refresh_qsfp_cache(struct hfi1_pportdata *ppd, struct qsfp_data *cp)
 {
@@ -371,13 +373,9 @@ int refresh_qsfp_cache(struct hfi1_pportdata *ppd, struct qsfp_data *cp)
 
 	if (!qsfp_mod_present(ppd)) {
 		ret = -ENODEV;
-		goto bail_no_release;
+		goto bail;
 	}
 
-	ret = acquire_chip_resource(ppd->dd, qsfp_resource(ppd->dd), QSFP_WAIT);
-	if (ret)
-		goto bail_no_release;
-
 	ret = qsfp_read(ppd, target, 0, cache, QSFP_PAGESIZE);
 	if (ret != QSFP_PAGESIZE) {
 		dd_dev_info(ppd->dd,
@@ -440,8 +438,6 @@ int refresh_qsfp_cache(struct hfi1_pportdata *ppd, struct qsfp_data *cp)
 		}
 	}
 
-	release_chip_resource(ppd->dd, qsfp_resource(ppd->dd));
-
 	spin_lock_irqsave(&ppd->qsfp_info.qsfp_lock, flags);
 	ppd->qsfp_info.cache_valid = 1;
 	ppd->qsfp_info.cache_refresh_required = 0;
@@ -450,8 +446,6 @@ int refresh_qsfp_cache(struct hfi1_pportdata *ppd, struct qsfp_data *cp)
 	return 0;
 
 bail:
-	release_chip_resource(ppd->dd, qsfp_resource(ppd->dd));
-bail_no_release:
 	memset(cache, 0, (QSFP_MAX_NUM_PAGES * 128));
 	return ret;
 }

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

[PATCH 14/21] IB/hfi1: Remove unreachable code

[Dennis Dalessandro]

From: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Remove unreachable code from RC ack handling to fix an
smatch error.

Fixes: 633d27399514 ("staging/rdma/hfi1: use mod_timer when appropriate")
Reported-by: Dan Carpenter <dan.carpenter-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/rc.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/rc.c b/drivers/staging/rdma/hfi1/rc.c
index 0d7e101..c614779 100644
--- a/drivers/staging/rdma/hfi1/rc.c
+++ b/drivers/staging/rdma/hfi1/rc.c
@@ -1497,7 +1497,7 @@ reserved:
 		/* Ignore reserved NAK codes. */
 		goto bail_stop;
 	}
-	return ret;
+	/* cannot be reached  */
 bail_stop:
 	hfi1_stop_rc_timers(qp);
 	return ret;

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

[PATCH 15/21] IB/hfi1: Use global defines for upper bits in opcode

[Dennis Dalessandro]

From: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The awkward coding for setting the allowed_ops field
was tripping an smatch warning.

This patch uses the more appropriate defines from include/rdma
to avoid the issue.

As part of the patch remove a mask that was duplicated
in rdmavt include files and use that mask as appropriate.

Fixes: 8bea6b1cfe6f ("IB/rdmavt: Add create queue pair functionality")
Reported-by: Dan Carpenter <dan.carpenter-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/sw/rdmavt/qp.c |    6 +++---
 drivers/staging/rdma/hfi1/verbs.c |    2 +-
 drivers/staging/rdma/hfi1/verbs.h |    3 ---
 3 files changed, 4 insertions(+), 7 deletions(-)

diff --git a/drivers/infiniband/sw/rdmavt/qp.c b/drivers/infiniband/sw/rdmavt/qp.c
index a9e3bcc..0f12c21 100644
--- a/drivers/infiniband/sw/rdmavt/qp.c
+++ b/drivers/infiniband/sw/rdmavt/qp.c
@@ -829,13 +829,13 @@ struct ib_qp *rvt_create_qp(struct ib_pd *ibpd,
 	case IB_QPT_SMI:
 	case IB_QPT_GSI:
 	case IB_QPT_UD:
-		qp->allowed_ops = IB_OPCODE_UD_SEND_ONLY & RVT_OPCODE_QP_MASK;
+		qp->allowed_ops = IB_OPCODE_UD;
 		break;
 	case IB_QPT_RC:
-		qp->allowed_ops = IB_OPCODE_RC_SEND_ONLY & RVT_OPCODE_QP_MASK;
+		qp->allowed_ops = IB_OPCODE_RC;
 		break;
 	case IB_QPT_UC:
-		qp->allowed_ops = IB_OPCODE_UC_SEND_ONLY & RVT_OPCODE_QP_MASK;
+		qp->allowed_ops = IB_OPCODE_UC;
 		break;
 	default:
 		ret = ERR_PTR(-EINVAL);
diff --git a/drivers/staging/rdma/hfi1/verbs.c b/drivers/staging/rdma/hfi1/verbs.c
index c56c0cb..9cdc85f 100644
--- a/drivers/staging/rdma/hfi1/verbs.c
+++ b/drivers/staging/rdma/hfi1/verbs.c
@@ -545,7 +545,7 @@ static inline int qp_ok(int opcode, struct hfi1_packet *packet)
 
 	if (!(ib_rvt_state_ops[packet->qp->state] & RVT_PROCESS_RECV_OK))
 		goto dropit;
-	if (((opcode & OPCODE_QP_MASK) == packet->qp->allowed_ops) ||
+	if (((opcode & RVT_OPCODE_QP_MASK) == packet->qp->allowed_ops) ||
 	    (opcode == IB_OPCODE_CNP))
 		return 1;
 dropit:
diff --git a/drivers/staging/rdma/hfi1/verbs.h b/drivers/staging/rdma/hfi1/verbs.h
index 2ba1373..3ee2239 100644
--- a/drivers/staging/rdma/hfi1/verbs.h
+++ b/drivers/staging/rdma/hfi1/verbs.h
@@ -335,9 +335,6 @@ int hfi1_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
 #endif
 #define PSN_MODIFY_MASK 0xFFFFFF
 
-/* Number of bits to pay attention to in the opcode for checking qp type */
-#define OPCODE_QP_MASK 0xE0
-
 /*
  * Compare the lower 24 bits of the msn values.
  * Returns an integer <, ==, or > than zero.

--
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 16/21] IB/hfi1: Change default number of user contexts

[Dennis Dalessandro]

From: Jubin John <jubin.john-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Change the default number of user contexts to the number of real
(non-HT) cpu cores in order to reduce the division of hfi1 hardware
contexts in the case of high core counts with hyper-threading enabled.

Reviewed-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Mitko Haralanov <mitko.haralanov-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Jubin John <jubin.john-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/affinity.c |   93 +++++++++++++++++-----------------
 drivers/staging/rdma/hfi1/affinity.h |   19 +++++++
 drivers/staging/rdma/hfi1/chip.c     |   26 +++++++---
 3 files changed, 84 insertions(+), 54 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/affinity.c b/drivers/staging/rdma/hfi1/affinity.c
index 2cb8ca7..6e7050a 100644
--- a/drivers/staging/rdma/hfi1/affinity.c
+++ b/drivers/staging/rdma/hfi1/affinity.c
@@ -53,20 +53,6 @@
 #include "sdma.h"
 #include "trace.h"
 
-struct cpu_mask_set {
-	struct cpumask mask;
-	struct cpumask used;
-	uint gen;
-};
-
-struct hfi1_affinity {
-	struct cpu_mask_set def_intr;
-	struct cpu_mask_set rcv_intr;
-	struct cpu_mask_set proc;
-	/* spin lock to protect affinity struct */
-	spinlock_t lock;
-};
-
 /* Name of IRQ types, indexed by enum irq_type */
 static const char * const irq_type_names[] = {
 	"SDMA",
@@ -82,6 +68,48 @@ static inline void init_cpu_mask_set(struct cpu_mask_set *set)
 	set->gen = 0;
 }
 
+/* Initialize non-HT cpu cores mask */
+int init_real_cpu_mask(struct hfi1_devdata *dd)
+{
+	struct hfi1_affinity *info;
+	int possible, curr_cpu, i, ht;
+
+	info = kzalloc(sizeof(*info), GFP_KERNEL);
+	if (!info)
+		return -ENOMEM;
+
+	cpumask_clear(&info->real_cpu_mask);
+
+	/* Start with cpu online mask as the real cpu mask */
+	cpumask_copy(&info->real_cpu_mask, cpu_online_mask);
+
+	/*
+	 * Remove HT cores from the real cpu mask.  Do this in two steps below.
+	 */
+	possible = cpumask_weight(&info->real_cpu_mask);
+	ht = cpumask_weight(topology_sibling_cpumask(
+					cpumask_first(&info->real_cpu_mask)));
+	/*
+	 * Step 1.  Skip over the first N HT siblings and use them as the
+	 * "real" cores.  Assumes that HT cores are not enumerated in
+	 * succession (except in the single core case).
+	 */
+	curr_cpu = cpumask_first(&info->real_cpu_mask);
+	for (i = 0; i < possible / ht; i++)
+		curr_cpu = cpumask_next(curr_cpu, &info->real_cpu_mask);
+	/*
+	 * Step 2.  Remove the remaining HT siblings.  Use cpumask_next() to
+	 * skip any gaps.
+	 */
+	for (; i < possible; i++) {
+		cpumask_clear_cpu(curr_cpu, &info->real_cpu_mask);
+		curr_cpu = cpumask_next(curr_cpu, &info->real_cpu_mask);
+	}
+
+	dd->affinity = info;
+	return 0;
+}
+
 /*
  * Interrupt affinity.
  *
@@ -93,20 +121,17 @@ static inline void init_cpu_mask_set(struct cpu_mask_set *set)
  * to the node relative 1 as necessary.
  *
  */
-int hfi1_dev_affinity_init(struct hfi1_devdata *dd)
+void hfi1_dev_affinity_init(struct hfi1_devdata *dd)
 {
 	int node = pcibus_to_node(dd->pcidev->bus);
-	struct hfi1_affinity *info;
+	struct hfi1_affinity *info = dd->affinity;
 	const struct cpumask *local_mask;
-	int curr_cpu, possible, i, ht;
+	int curr_cpu, possible, i;
 
 	if (node < 0)
 		node = numa_node_id();
 	dd->node = node;
 
-	info = kzalloc(sizeof(*info), GFP_KERNEL);
-	if (!info)
-		return -ENOMEM;
 	spin_lock_init(&info->lock);
 
 	init_cpu_mask_set(&info->def_intr);
@@ -116,30 +141,8 @@ int hfi1_dev_affinity_init(struct hfi1_devdata *dd)
 	local_mask = cpumask_of_node(dd->node);
 	if (cpumask_first(local_mask) >= nr_cpu_ids)
 		local_mask = topology_core_cpumask(0);
-	/* use local mask as default */
-	cpumask_copy(&info->def_intr.mask, local_mask);
-	/*
-	 * Remove HT cores from the default mask.  Do this in two steps below.
-	 */
-	possible = cpumask_weight(&info->def_intr.mask);
-	ht = cpumask_weight(topology_sibling_cpumask(
-					cpumask_first(&info->def_intr.mask)));
-	/*
-	 * Step 1.  Skip over the first N HT siblings and use them as the
-	 * "real" cores.  Assumes that HT cores are not enumerated in
-	 * succession (except in the single core case).
-	 */
-	curr_cpu = cpumask_first(&info->def_intr.mask);
-	for (i = 0; i < possible / ht; i++)
-		curr_cpu = cpumask_next(curr_cpu, &info->def_intr.mask);
-	/*
-	 * Step 2.  Remove the remaining HT siblings.  Use cpumask_next() to
-	 * skip any gaps.
-	 */
-	for (; i < possible; i++) {
-		cpumask_clear_cpu(curr_cpu, &info->def_intr.mask);
-		curr_cpu = cpumask_next(curr_cpu, &info->def_intr.mask);
-	}
+	/* Use the "real" cpu mask of this node as the default */
+	cpumask_and(&info->def_intr.mask, &info->real_cpu_mask, local_mask);
 
 	/*  fill in the receive list */
 	possible = cpumask_weight(&info->def_intr.mask);
@@ -167,8 +170,6 @@ int hfi1_dev_affinity_init(struct hfi1_devdata *dd)
 	}
 
 	cpumask_copy(&info->proc.mask, cpu_online_mask);
-	dd->affinity = info;
-	return 0;
 }
 
 void hfi1_dev_affinity_free(struct hfi1_devdata *dd)
diff --git a/drivers/staging/rdma/hfi1/affinity.h b/drivers/staging/rdma/hfi1/affinity.h
index b287e49..20f52fe 100644
--- a/drivers/staging/rdma/hfi1/affinity.h
+++ b/drivers/staging/rdma/hfi1/affinity.h
@@ -64,10 +64,27 @@ enum affinity_flags {
 	AFF_IRQ_LOCAL
 };
 
+struct cpu_mask_set {
+	struct cpumask mask;
+	struct cpumask used;
+	uint gen;
+};
+
+struct hfi1_affinity {
+	struct cpu_mask_set def_intr;
+	struct cpu_mask_set rcv_intr;
+	struct cpu_mask_set proc;
+	struct cpumask real_cpu_mask;
+	/* spin lock to protect affinity struct */
+	spinlock_t lock;
+};
+
 struct hfi1_msix_entry;
 
+/* Initialize non-HT cpu cores mask */
+int init_real_cpu_mask(struct hfi1_devdata *);
 /* Initialize driver affinity data */
-int hfi1_dev_affinity_init(struct hfi1_devdata *);
+void hfi1_dev_affinity_init(struct hfi1_devdata *);
 /* Free driver affinity data */
 void hfi1_dev_affinity_free(struct hfi1_devdata *);
 /*
diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index 38c0e97..1984454 100644
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -12656,12 +12656,13 @@ static int set_up_context_variables(struct hfi1_devdata *dd)
 		num_kernel_contexts = dd->chip_send_contexts - num_vls - 1;
 	}
 	/*
-	 * User contexts: (to be fixed later)
-	 *	- default to 1 user context per CPU if num_user_contexts is
-	 *	  negative
+	 * User contexts:
+	 *	- default to 1 user context per real (non-HT) CPU core if
+	 *	  num_user_contexts is negative
 	 */
 	if (num_user_contexts < 0)
-		num_user_contexts = num_online_cpus();
+		num_user_contexts =
+			cpumask_weight(&dd->affinity->real_cpu_mask);
 
 	total_contexts = num_kernel_contexts + num_user_contexts;
 
@@ -14089,6 +14090,19 @@ struct hfi1_devdata *hfi1_init_dd(struct pci_dev *pdev,
 		 (dd->revision >> CCE_REVISION_SW_SHIFT)
 		    & CCE_REVISION_SW_MASK);
 
+	/*
+	 * The real cpu mask is part of the affinity struct but has to be
+	 * initialized earlier than the rest of the affinity struct because it
+	 * is needed to calculate the number of user contexts in
+	 * set_up_context_variables(). However, hfi1_dev_affinity_init(),
+	 * which initializes the rest of the affinity struct members,
+	 * depends on set_up_context_variables() for the number of kernel
+	 * contexts, so it cannot be called before set_up_context_variables().
+	 */
+	ret = init_real_cpu_mask(dd);
+	if (ret)
+		goto bail_cleanup;
+
 	ret = set_up_context_variables(dd);
 	if (ret)
 		goto bail_cleanup;
@@ -14102,9 +14116,7 @@ struct hfi1_devdata *hfi1_init_dd(struct pci_dev *pdev,
 	/* set up KDETH QP prefix in both RX and TX CSRs */
 	init_kdeth_qp(dd);
 
-	ret = hfi1_dev_affinity_init(dd);
-	if (ret)
-		goto bail_cleanup;
+	hfi1_dev_affinity_init(dd);
 
 	/* send contexts must be set up before receive contexts */
 	ret = init_send_contexts(dd);

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

[PATCH 17/21] IB/hfi1: Reduce kernel context pio buffer allocation

[Dennis Dalessandro]

From: Jianxin Xiong <jianxin.xiong-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The pio buffers were pooled evenly among all kernel contexts and
user contexts. However, the demand from kernel contexts is much
lower than user contexts. This patch reduces the allocation for
kernel contexts and thus makes more credits available for PSM,
helping performance. This is especially useful on high core-count
systems where large numbers of contexts are used.

A new context type SC_VL15 is added to distinguish the context used
for VL15 from other kernel contexts. The reason is that VL15 needs
to support 2KB sized packet while other kernel contexts need only
support packets up to the size determined by "piothreshold", which
has a default value of 256.

The new allocation method allows triple buffering of largest pio
packets configured for these contexts. This is sufficient to maintain
verbs performance. The largest pio packet size is 2048B for VL15
and "piothreshold" for other kernel contexts. A cap is applied to
"piothreshold" to avoid excessive buffer allocation.

The special case that SDMA is disable is handled differently. In
that case, the original pooling allocation is used to better
support the much higher pio traffic.

Notice that if adaptive pio is disabled (piothreshold==0), the pio
buffer size doesn't matter for non-VL15 kernel send contexts when
SDMA is enabled because pio is not used at all on these contexts
and thus the new allocation is still valid. If SDMA is disabled then
pooling allocation is used as mentioned in previous paragraph.

Adjustment is also made to the calculation of the credit return
threshold for the kernel contexts. Instead of purely based on
the MTU size, a percentage based threshold is also considered and
the smaller one of the two is chosen. This is necessary to ensure
that with the reduced buffer allocation credits are returned in
time to avoid unnecessary stall in the send path.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Mark Debbage <mark.debbage-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Jubin John <jubin.john-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Jianxin Xiong <jianxin.xiong-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/chip.c |   27 +++++++++++---------
 drivers/staging/rdma/hfi1/diag.c |    3 +-
 drivers/staging/rdma/hfi1/pio.c  |   52 ++++++++++++++++++++++++++++++--------
 drivers/staging/rdma/hfi1/pio.h  |    4 ++-
 4 files changed, 61 insertions(+), 25 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index 1984454..96badb4 100644
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -5661,7 +5661,7 @@ static int sc_to_vl(struct hfi1_devdata *dd, int sw_index)
 	sci = &dd->send_contexts[sw_index];
 
 	/* there is no information for user (PSM) and ack contexts */
-	if (sci->type != SC_KERNEL)
+	if ((sci->type != SC_KERNEL) && (sci->type != SC_VL15))
 		return -1;
 
 	sc = sci->sc;
@@ -9627,6 +9627,7 @@ static void set_send_length(struct hfi1_pportdata *ppd)
 			      & SEND_LEN_CHECK1_LEN_VL15_MASK) <<
 		SEND_LEN_CHECK1_LEN_VL15_SHIFT;
 	int i;
+	u32 thres;
 
 	for (i = 0; i < ppd->vls_supported; i++) {
 		if (dd->vld[i].mtu > maxvlmtu)
@@ -9645,16 +9646,17 @@ static void set_send_length(struct hfi1_pportdata *ppd)
 	/* adjust kernel credit return thresholds based on new MTUs */
 	/* all kernel receive contexts have the same hdrqentsize */
 	for (i = 0; i < ppd->vls_supported; i++) {
-		sc_set_cr_threshold(dd->vld[i].sc,
-				    sc_mtu_to_threshold(dd->vld[i].sc,
-							dd->vld[i].mtu,
-							dd->rcd[0]->
-							rcvhdrqentsize));
-	}
-	sc_set_cr_threshold(dd->vld[15].sc,
-			    sc_mtu_to_threshold(dd->vld[15].sc,
-						dd->vld[15].mtu,
+		thres = min(sc_percent_to_threshold(dd->vld[i].sc, 50),
+			    sc_mtu_to_threshold(dd->vld[i].sc,
+						dd->vld[i].mtu,
 						dd->rcd[0]->rcvhdrqentsize));
+		sc_set_cr_threshold(dd->vld[i].sc, thres);
+	}
+	thres = min(sc_percent_to_threshold(dd->vld[15].sc, 50),
+		    sc_mtu_to_threshold(dd->vld[15].sc,
+					dd->vld[15].mtu,
+					dd->rcd[0]->rcvhdrqentsize));
+	sc_set_cr_threshold(dd->vld[15].sc, thres);
 
 	/* Adjust maximum MTU for the port in DC */
 	dcmtu = maxvlmtu == 10240 ? DCC_CFG_PORT_MTU_CAP_10240 :
@@ -12728,12 +12730,13 @@ static int set_up_context_variables(struct hfi1_devdata *dd)
 		dd->num_send_contexts = ret;
 		dd_dev_info(
 			dd,
-			"send contexts: chip %d, used %d (kernel %d, ack %d, user %d)\n",
+			"send contexts: chip %d, used %d (kernel %d, ack %d, user %d, vl15 %d)\n",
 			dd->chip_send_contexts,
 			dd->num_send_contexts,
 			dd->sc_sizes[SC_KERNEL].count,
 			dd->sc_sizes[SC_ACK].count,
-			dd->sc_sizes[SC_USER].count);
+			dd->sc_sizes[SC_USER].count,
+			dd->sc_sizes[SC_VL15].count);
 		ret = 0;	/* success */
 	}
 
diff --git a/drivers/staging/rdma/hfi1/diag.c b/drivers/staging/rdma/hfi1/diag.c
index c5b520b..bb2409a 100644
--- a/drivers/staging/rdma/hfi1/diag.c
+++ b/drivers/staging/rdma/hfi1/diag.c
@@ -413,7 +413,8 @@ static ssize_t diagpkt_send(struct diag_pkt *dp)
 		goto bail;
 	}
 	/* can only use kernel contexts */
-	if (dd->send_contexts[dp->sw_index].type != SC_KERNEL) {
+	if (dd->send_contexts[dp->sw_index].type != SC_KERNEL &&
+	    dd->send_contexts[dp->sw_index].type != SC_VL15) {
 		ret = -EINVAL;
 		goto bail;
 	}
diff --git a/drivers/staging/rdma/hfi1/pio.c b/drivers/staging/rdma/hfi1/pio.c
index c6849ce..c67b9ad 100644
--- a/drivers/staging/rdma/hfi1/pio.c
+++ b/drivers/staging/rdma/hfi1/pio.c
@@ -139,23 +139,30 @@ void pio_send_control(struct hfi1_devdata *dd, int op)
 /* Send Context Size (SCS) wildcards */
 #define SCS_POOL_0 -1
 #define SCS_POOL_1 -2
+
 /* Send Context Count (SCC) wildcards */
 #define SCC_PER_VL -1
 #define SCC_PER_CPU  -2
-
 #define SCC_PER_KRCVQ  -3
-#define SCC_ACK_CREDITS  32
+
+/* Send Context Size (SCS) constants */
+#define SCS_ACK_CREDITS  32
+#define SCS_VL15_CREDITS 102	/* 3 pkts of 2048B data + 128B header */
+
+#define PIO_THRESHOLD_CEILING 4096
 
 #define PIO_WAIT_BATCH_SIZE 5
 
 /* default send context sizes */
 static struct sc_config_sizes sc_config_sizes[SC_MAX] = {
 	[SC_KERNEL] = { .size  = SCS_POOL_0,	/* even divide, pool 0 */
-			.count = SCC_PER_VL },/* one per NUMA */
-	[SC_ACK]    = { .size  = SCC_ACK_CREDITS,
+			.count = SCC_PER_VL },	/* one per NUMA */
+	[SC_ACK]    = { .size  = SCS_ACK_CREDITS,
 			.count = SCC_PER_KRCVQ },
 	[SC_USER]   = { .size  = SCS_POOL_0,	/* even divide, pool 0 */
 			.count = SCC_PER_CPU },	/* one per CPU */
+	[SC_VL15]   = { .size  = SCS_VL15_CREDITS,
+			.count = 1 },
 
 };
 
@@ -202,7 +209,8 @@ static int wildcard_to_pool(int wc)
 static const char *sc_type_names[SC_MAX] = {
 	"kernel",
 	"ack",
-	"user"
+	"user",
+	"vl15"
 };
 
 static const char *sc_type_name(int index)
@@ -231,6 +239,22 @@ int init_sc_pools_and_sizes(struct hfi1_devdata *dd)
 	int i;
 
 	/*
+	 * When SDMA is enabled, kernel context pio packet size is capped by
+	 * "piothreshold". Reduce pio buffer allocation for kernel context by
+	 * setting it to a fixed size. The allocation allows 3-deep buffering
+	 * of the largest pio packets plus up to 128 bytes header, sufficient
+	 * to maintain verbs performance.
+	 *
+	 * When SDMA is disabled, keep the default pooling allocation.
+	 */
+	if (HFI1_CAP_IS_KSET(SDMA)) {
+		u16 max_pkt_size = (piothreshold < PIO_THRESHOLD_CEILING) ?
+					 piothreshold : PIO_THRESHOLD_CEILING;
+		sc_config_sizes[SC_KERNEL].size =
+			3 * (max_pkt_size + 128) / PIO_BLOCK_SIZE;
+	}
+
+	/*
 	 * Step 0:
 	 *	- copy the centipercents/absolute sizes from the pool config
 	 *	- sanity check these values
@@ -311,7 +335,7 @@ int init_sc_pools_and_sizes(struct hfi1_devdata *dd)
 		if (i == SC_ACK) {
 			count = dd->n_krcv_queues;
 		} else if (i == SC_KERNEL) {
-			count = (INIT_SC_PER_VL * num_vls) + 1 /* VL15 */;
+			count = INIT_SC_PER_VL * num_vls;
 		} else if (count == SCC_PER_CPU) {
 			count = dd->num_rcv_contexts - dd->n_krcv_queues;
 		} else if (count < 0) {
@@ -596,7 +620,7 @@ u32 sc_mtu_to_threshold(struct send_context *sc, u32 mtu, u32 hdrqentsize)
  * Return value is what to write into the CSR: trigger return when
  * unreturned credits pass this count.
  */
-static u32 sc_percent_to_threshold(struct send_context *sc, u32 percent)
+u32 sc_percent_to_threshold(struct send_context *sc, u32 percent)
 {
 	return (sc->credits * percent) / 100;
 }
@@ -790,7 +814,10 @@ struct send_context *sc_alloc(struct hfi1_devdata *dd, int type,
 	 * For Ack contexts, set a threshold for half the credits.
 	 * For User contexts use the given percentage.  This has been
 	 * sanitized on driver start-up.
-	 * For Kernel contexts, use the default MTU plus a header.
+	 * For Kernel contexts, use the default MTU plus a header
+	 * or half the credits, whichever is smaller. This should
+	 * work for both the 3-deep buffering allocation and the
+	 * pooling allocation.
 	 */
 	if (type == SC_ACK) {
 		thresh = sc_percent_to_threshold(sc, 50);
@@ -798,7 +825,9 @@ struct send_context *sc_alloc(struct hfi1_devdata *dd, int type,
 		thresh = sc_percent_to_threshold(sc,
 						 user_credit_return_threshold);
 	} else { /* kernel */
-		thresh = sc_mtu_to_threshold(sc, hfi1_max_mtu, hdrqentsize);
+		thresh = min(sc_percent_to_threshold(sc, 50),
+			     sc_mtu_to_threshold(sc, hfi1_max_mtu,
+						 hdrqentsize));
 	}
 	reg = thresh << SC(CREDIT_CTRL_THRESHOLD_SHIFT);
 	/* add in early return */
@@ -1531,7 +1560,8 @@ static void sc_piobufavail(struct send_context *sc)
 	unsigned long flags;
 	unsigned i, n = 0;
 
-	if (dd->send_contexts[sc->sw_index].type != SC_KERNEL)
+	if (dd->send_contexts[sc->sw_index].type != SC_KERNEL &&
+	    dd->send_contexts[sc->sw_index].type != SC_VL15)
 		return;
 	list = &sc->piowait;
 	/*
@@ -1900,7 +1930,7 @@ int init_pervl_scs(struct hfi1_devdata *dd)
 	u32 ctxt;
 	struct hfi1_pportdata *ppd = dd->pport;
 
-	dd->vld[15].sc = sc_alloc(dd, SC_KERNEL,
+	dd->vld[15].sc = sc_alloc(dd, SC_VL15,
 				  dd->rcd[0]->rcvhdrqentsize, dd->node);
 	if (!dd->vld[15].sc)
 		goto nomem;
diff --git a/drivers/staging/rdma/hfi1/pio.h b/drivers/staging/rdma/hfi1/pio.h
index 0026976..53a08ed 100644
--- a/drivers/staging/rdma/hfi1/pio.h
+++ b/drivers/staging/rdma/hfi1/pio.h
@@ -51,7 +51,8 @@
 #define SC_KERNEL 0
 #define SC_ACK    1
 #define SC_USER   2
-#define SC_MAX    3
+#define SC_VL15   3
+#define SC_MAX    4
 
 /* invalid send context index */
 #define INVALID_SCI 0xff
@@ -293,6 +294,7 @@ void sc_group_release_update(struct hfi1_devdata *dd, u32 hw_context);
 void sc_add_credit_return_intr(struct send_context *sc);
 void sc_del_credit_return_intr(struct send_context *sc);
 void sc_set_cr_threshold(struct send_context *sc, u32 new_threshold);
+u32 sc_percent_to_threshold(struct send_context *sc, u32 percent);
 u32 sc_mtu_to_threshold(struct send_context *sc, u32 mtu, u32 hdrqentsize);
 void hfi1_sc_wantpiobuf_intr(struct send_context *sc, u32 needint);
 void sc_wait(struct hfi1_devdata *dd);

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

[PATCH 18/21] IB/hfi1: Extract RSM map table init from QOS

[Dennis Dalessandro]

From: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Refactor the allocation, tracking, and writing of the RSM map table
into its own set of routines.  This will allow the map table to be
passed to multiple users to fill in as needed.  Start with the original
user, QOS.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/chip.c           |   91 ++++++++++++++++++++++------
 drivers/staging/rdma/hfi1/chip_registers.h |    1 
 2 files changed, 71 insertions(+), 21 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index 96badb4..234c078 100644
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -123,6 +123,8 @@ struct flag_table {
 
 #define MIN_KERNEL_KCTXTS         2
 #define FIRST_KERNEL_KCTXT        1
+/* sizes for both the QP and RSM map tables */
+#define NUM_MAP_ENTRIES		256
 #define NUM_MAP_REGS             32
 
 /* Bit offset into the GUID which carries HFI id information */
@@ -13422,9 +13424,52 @@ static void init_qpmap_table(struct hfi1_devdata *dd,
 			| RCV_CTRL_RCV_BYPASS_ENABLE_SMASK);
 }
 
+struct rsm_map_table {
+	u64 map[NUM_MAP_REGS];
+	unsigned int used;
+};
+
+/*
+ * Return an initialized RMT map table for users to fill in.  OK if it
+ * returns NULL, indicating no table.
+ */
+static struct rsm_map_table *alloc_rsm_map_table(struct hfi1_devdata *dd)
+{
+	struct rsm_map_table *rmt;
+	u8 rxcontext = is_ax(dd) ? 0 : 0xff;  /* 0 is default if a0 ver. */
+
+	rmt = kmalloc(sizeof(*rmt), GFP_KERNEL);
+	if (rmt) {
+		memset(rmt->map, rxcontext, sizeof(rmt->map));
+		rmt->used = 0;
+	}
+
+	return rmt;
+}
+
+/*
+ * Write the final RMT map table to the chip and free the table.  OK if
+ * table is NULL.
+ */
+static void complete_rsm_map_table(struct hfi1_devdata *dd,
+				   struct rsm_map_table *rmt)
+{
+	int i;
+
+	if (rmt) {
+		/* write table to chip */
+		for (i = 0; i < NUM_MAP_REGS; i++)
+			write_csr(dd, RCV_RSM_MAP_TABLE + (8 * i), rmt->map[i]);
+
+		/* enable RSM */
+		add_rcvctrl(dd, RCV_CTRL_RCV_RSM_ENABLE_SMASK);
+	}
+}
+
 /**
  * init_qos - init RX qos
  * @dd - device data
+ * @rmt - RSM map table
  *
  * This routine initializes Rule 0 and the RSM map table to implement
  * quality of service (qos).
@@ -13435,16 +13480,16 @@ static void init_qpmap_table(struct hfi1_devdata *dd,
  * The number of vl bits (n) and the number of qpn bits (m) are computed to
  * feed both the RSM map table and the single rule.
  */
-static void init_qos(struct hfi1_devdata *dd)
+static void init_qos(struct hfi1_devdata *dd, struct rsm_map_table *rmt)
 {
 	u8 max_by_vl = 0;
 	unsigned qpns_per_vl, ctxt, i, qpn, n = 1, m;
-	u64 *rsmmap;
+	unsigned int rmt_entries;
 	u64 reg;
-	u8  rxcontext = is_ax(dd) ? 0 : 0xff;  /* 0 is default if a0 ver. */
 
 	/* validate */
-	if (dd->n_krcv_queues <= MIN_KERNEL_KCTXTS ||
+	if (!rmt ||
+	    dd->n_krcv_queues <= MIN_KERNEL_KCTXTS ||
 	    num_vls == 1 ||
 	    krcvqsset <= 1)
 		goto bail;
@@ -13460,11 +13505,11 @@ static void init_qos(struct hfi1_devdata *dd)
 	m = ilog2(qpns_per_vl);
 	if ((m + n) > 7)
 		goto bail;
-	rsmmap = kmalloc_array(NUM_MAP_REGS, sizeof(u64), GFP_KERNEL);
-	if (!rsmmap)
+	/* enough room in the map table? */
+	rmt_entries = 1 << (m + n);
+	if (rmt->used + rmt_entries >= NUM_MAP_ENTRIES)
 		goto bail;
-	memset(rsmmap, rxcontext, NUM_MAP_REGS * sizeof(u64));
-	/* init the local copy of the table */
+	/* add qos entries to the the RSM map table */
 	for (i = 0, ctxt = FIRST_KERNEL_KCTXT; i < num_vls; i++) {
 		unsigned tctxt;
 
@@ -13472,26 +13517,24 @@ static void init_qos(struct hfi1_devdata *dd)
 		     krcvqs[i] && qpn < qpns_per_vl; qpn++) {
 			unsigned idx, regoff, regidx;
 
-			/* generate index <= 128 */
-			idx = (qpn << n) ^ i;
+			/* generate the index the hardware will produce */
+			idx = rmt->used + ((qpn << n) ^ i);
 			regoff = (idx % 8) * 8;
 			regidx = idx / 8;
-			reg = rsmmap[regidx];
-			/* replace 0xff with context number */
+			/* replace default with context number */
+			reg = rmt->map[regidx];
 			reg &= ~(RCV_RSM_MAP_TABLE_RCV_CONTEXT_A_MASK
 				<< regoff);
 			reg |= (u64)(tctxt++) << regoff;
-			rsmmap[regidx] = reg;
+			rmt->map[regidx] = reg;
 			if (tctxt == ctxt + krcvqs[i])
 				tctxt = ctxt;
 		}
 		ctxt += krcvqs[i];
 	}
-	/* flush cached copies to chip */
-	for (i = 0; i < NUM_MAP_REGS; i++)
-		write_csr(dd, RCV_RSM_MAP_TABLE + (8 * i), rsmmap[i]);
 	/* add rule0 */
 	write_csr(dd, RCV_RSM_CFG /* + (8 * 0) */,
+		  (u64)rmt->used << RCV_RSM_CFG_OFFSET_SHIFT |
 		  RCV_RSM_CFG_ENABLE_OR_CHAIN_RSM0_MASK <<
 			RCV_RSM_CFG_ENABLE_OR_CHAIN_RSM0_SHIFT |
 		  2ull << RCV_RSM_CFG_PACKET_TYPE_SHIFT);
@@ -13507,9 +13550,8 @@ static void init_qos(struct hfi1_devdata *dd)
 		  LRH_BTH_VALUE << RCV_RSM_MATCH_VALUE1_SHIFT |
 		  LRH_SC_MASK << RCV_RSM_MATCH_MASK2_SHIFT |
 		  LRH_SC_VALUE << RCV_RSM_MATCH_VALUE2_SHIFT);
-	/* Enable RSM */
-	add_rcvctrl(dd, RCV_CTRL_RCV_RSM_ENABLE_SMASK);
-	kfree(rsmmap);
+	/* mark RSM map entries as used */
+	rmt->used += rmt_entries;
 	/* map everything else to the mcast/err/vl15 context */
 	init_qpmap_table(dd, HFI1_CTRL_CTXT, HFI1_CTRL_CTXT);
 	dd->qos_shift = n + 1;
@@ -13521,10 +13563,17 @@ bail:
 
 static void init_rxe(struct hfi1_devdata *dd)
 {
+	struct rsm_map_table *rmt;
+
 	/* enable all receive errors */
 	write_csr(dd, RCV_ERR_MASK, ~0ull);
-	/* setup QPN map table - start where VL15 context leaves off */
-	init_qos(dd);
+
+	rmt = alloc_rsm_map_table(dd);
+	/* set up QOS, including the QPN map table */
+	init_qos(dd, rmt);
+	complete_rsm_map_table(dd, rmt);
+	kfree(rmt);
+
 	/*
 	 * make sure RcvCtrl.RcvWcb <= PCIe Device Control
 	 * Register Max_Payload_Size (PCI_EXP_DEVCTL in Linux PCIe config
diff --git a/drivers/staging/rdma/hfi1/chip_registers.h b/drivers/staging/rdma/hfi1/chip_registers.h
index 770f05c..8744de6 100644
--- a/drivers/staging/rdma/hfi1/chip_registers.h
+++ b/drivers/staging/rdma/hfi1/chip_registers.h
@@ -771,6 +771,7 @@
 #define RCV_RSM_CFG_ENABLE_OR_CHAIN_RSM0_MASK 0x1ull
 #define RCV_RSM_CFG_ENABLE_OR_CHAIN_RSM0_SHIFT 0
 #define RCV_RSM_CFG_PACKET_TYPE_SHIFT 60
+#define RCV_RSM_CFG_OFFSET_SHIFT 32
 #define RCV_RSM_MAP_TABLE (RXE + 0x000000000900)
 #define RCV_RSM_MAP_TABLE_RCV_CONTEXT_A_MASK 0xFFull
 #define RCV_RSM_MATCH (RXE + 0x000000000800)

--
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 19/21] IB/hfi1: Move QOS decision logic into its own function

[Dennis Dalessandro]

From: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The decision to use QOS affects other resource allocation.
Move the QOS decision logic into its own function so it can
be called by other interested parties.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/chip.c |   67 ++++++++++++++++++++++++++++----------
 1 files changed, 50 insertions(+), 17 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index 234c078..b2c037c 100644
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -13466,6 +13466,50 @@ static void complete_rsm_map_table(struct hfi1_devdata *dd,
 	}
 }
 
+/* return the number of RSM map table entries that will be used for QOS */
+static int qos_rmt_entries(struct hfi1_devdata *dd, unsigned int *mp,
+			   unsigned int *np)
+{
+	int i;
+	unsigned int m, n;
+	u8 max_by_vl = 0;
+
+	/* is QOS active at all? */
+	if (dd->n_krcv_queues <= MIN_KERNEL_KCTXTS ||
+	    num_vls == 1 ||
+	    krcvqsset <= 1)
+		goto no_qos;
+
+	/* determine bits for qpn */
+	for (i = 0; i < min_t(unsigned int, num_vls, krcvqsset); i++)
+		if (krcvqs[i] > max_by_vl)
+			max_by_vl = krcvqs[i];
+	if (max_by_vl > 32)
+		goto no_qos;
+	m = ilog2(__roundup_pow_of_two(max_by_vl));
+
+	/* determine bits for vl */
+	n = ilog2(__roundup_pow_of_two(num_vls));
+
+	/* reject if too much is used */
+	if ((m + n) > 7)
+		goto no_qos;
+
+	if (mp)
+		*mp = m;
+	if (np)
+		*np = n;
+
+	return 1 << (m + n);
+
+no_qos:
+	if (mp)
+		*mp = 0;
+	if (np)
+		*np = 0;
+	return 0;
+}
+
 /**
  * init_qos - init RX qos
  * @dd - device data
@@ -13482,33 +13526,22 @@ static void complete_rsm_map_table(struct hfi1_devdata *dd,
  */
 static void init_qos(struct hfi1_devdata *dd, struct rsm_map_table *rmt)
 {
-	u8 max_by_vl = 0;
 	unsigned qpns_per_vl, ctxt, i, qpn, n = 1, m;
 	unsigned int rmt_entries;
 	u64 reg;
 
-	/* validate */
-	if (!rmt ||
-	    dd->n_krcv_queues <= MIN_KERNEL_KCTXTS ||
-	    num_vls == 1 ||
-	    krcvqsset <= 1)
-		goto bail;
-	for (i = 0; i < min_t(unsigned, num_vls, krcvqsset); i++)
-		if (krcvqs[i] > max_by_vl)
-			max_by_vl = krcvqs[i];
-	if (max_by_vl > 32)
+	if (!rmt)
 		goto bail;
-	qpns_per_vl = __roundup_pow_of_two(max_by_vl);
-	/* determine bits vl */
-	n = ilog2(__roundup_pow_of_two(num_vls));
-	/* determine bits for qpn */
-	m = ilog2(qpns_per_vl);
-	if ((m + n) > 7)
+	rmt_entries = qos_rmt_entries(dd, &m, &n);
+	if (rmt_entries == 0)
 		goto bail;
+	qpns_per_vl = 1 << m;
+
 	/* enough room in the map table? */
 	rmt_entries = 1 << (m + n);
 	if (rmt->used + rmt_entries >= NUM_MAP_ENTRIES)
 		goto bail;
+
 	/* add qos entries to the the RSM map table */
 	for (i = 0, ctxt = FIRST_KERNEL_KCTXT; i < num_vls; i++) {
 		unsigned tctxt;

--
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 20/21] IB/hfi1: Create a routine to set a receive side mapping rule

[Dennis Dalessandro]

From: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Move the rule setting code into its own routine for improved
searchability and reuse.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/chip.c |   75 +++++++++++++++++++++++++++++---------
 1 files changed, 57 insertions(+), 18 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index b2c037c..b5edc3a 100644
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -13429,6 +13429,21 @@ struct rsm_map_table {
 	unsigned int used;
 };
 
+struct rsm_rule_data {
+	u8 offset;
+	u8 pkt_type;
+	u32 field1_off;
+	u32 field2_off;
+	u32 index1_off;
+	u32 index1_width;
+	u32 index2_off;
+	u32 index2_width;
+	u32 mask1;
+	u32 value1;
+	u32 mask2;
+	u32 value2;
+};
+
 /*
  * Return an initialized RMT map table for users to fill in.  OK if it
  * returns NULL, indicating no table.
@@ -13466,6 +13481,30 @@ static void complete_rsm_map_table(struct hfi1_devdata *dd,
 	}
 }
 
+/*
+ * Add a receive side mapping rule.
+ */
+static void add_rsm_rule(struct hfi1_devdata *dd, u8 rule_index,
+			 struct rsm_rule_data *rrd)
+{
+	write_csr(dd, RCV_RSM_CFG + (8 * rule_index),
+		  (u64)rrd->offset << RCV_RSM_CFG_OFFSET_SHIFT |
+		  1ull << rule_index | /* enable bit */
+		  (u64)rrd->pkt_type << RCV_RSM_CFG_PACKET_TYPE_SHIFT);
+	write_csr(dd, RCV_RSM_SELECT + (8 * rule_index),
+		  (u64)rrd->field1_off << RCV_RSM_SELECT_FIELD1_OFFSET_SHIFT |
+		  (u64)rrd->field2_off << RCV_RSM_SELECT_FIELD2_OFFSET_SHIFT |
+		  (u64)rrd->index1_off << RCV_RSM_SELECT_INDEX1_OFFSET_SHIFT |
+		  (u64)rrd->index1_width << RCV_RSM_SELECT_INDEX1_WIDTH_SHIFT |
+		  (u64)rrd->index2_off << RCV_RSM_SELECT_INDEX2_OFFSET_SHIFT |
+		  (u64)rrd->index2_width << RCV_RSM_SELECT_INDEX2_WIDTH_SHIFT);
+	write_csr(dd, RCV_RSM_MATCH + (8 * rule_index),
+		  (u64)rrd->mask1 << RCV_RSM_MATCH_MASK1_SHIFT |
+		  (u64)rrd->value1 << RCV_RSM_MATCH_VALUE1_SHIFT |
+		  (u64)rrd->mask2 << RCV_RSM_MATCH_MASK2_SHIFT |
+		  (u64)rrd->value2 << RCV_RSM_MATCH_VALUE2_SHIFT);
+}
+
 /* return the number of RSM map table entries that will be used for QOS */
 static int qos_rmt_entries(struct hfi1_devdata *dd, unsigned int *mp,
 			   unsigned int *np)
@@ -13526,6 +13565,7 @@ no_qos:
  */
 static void init_qos(struct hfi1_devdata *dd, struct rsm_map_table *rmt)
 {
+	struct rsm_rule_data rrd;
 	unsigned qpns_per_vl, ctxt, i, qpn, n = 1, m;
 	unsigned int rmt_entries;
 	u64 reg;
@@ -13565,24 +13605,23 @@ static void init_qos(struct hfi1_devdata *dd, struct rsm_map_table *rmt)
 		}
 		ctxt += krcvqs[i];
 	}
-	/* add rule0 */
-	write_csr(dd, RCV_RSM_CFG /* + (8 * 0) */,
-		  (u64)rmt->used << RCV_RSM_CFG_OFFSET_SHIFT |
-		  RCV_RSM_CFG_ENABLE_OR_CHAIN_RSM0_MASK <<
-			RCV_RSM_CFG_ENABLE_OR_CHAIN_RSM0_SHIFT |
-		  2ull << RCV_RSM_CFG_PACKET_TYPE_SHIFT);
-	write_csr(dd, RCV_RSM_SELECT /* + (8 * 0) */,
-		  LRH_BTH_MATCH_OFFSET << RCV_RSM_SELECT_FIELD1_OFFSET_SHIFT |
-		  LRH_SC_MATCH_OFFSET << RCV_RSM_SELECT_FIELD2_OFFSET_SHIFT |
-		  LRH_SC_SELECT_OFFSET << RCV_RSM_SELECT_INDEX1_OFFSET_SHIFT |
-		  ((u64)n) << RCV_RSM_SELECT_INDEX1_WIDTH_SHIFT |
-		  QPN_SELECT_OFFSET << RCV_RSM_SELECT_INDEX2_OFFSET_SHIFT |
-		  ((u64)m + (u64)n) << RCV_RSM_SELECT_INDEX2_WIDTH_SHIFT);
-	write_csr(dd, RCV_RSM_MATCH /* + (8 * 0) */,
-		  LRH_BTH_MASK << RCV_RSM_MATCH_MASK1_SHIFT |
-		  LRH_BTH_VALUE << RCV_RSM_MATCH_VALUE1_SHIFT |
-		  LRH_SC_MASK << RCV_RSM_MATCH_MASK2_SHIFT |
-		  LRH_SC_VALUE << RCV_RSM_MATCH_VALUE2_SHIFT);
+
+	rrd.offset = rmt->used;
+	rrd.pkt_type = 2;
+	rrd.field1_off = LRH_BTH_MATCH_OFFSET;
+	rrd.field2_off = LRH_SC_MATCH_OFFSET;
+	rrd.index1_off = LRH_SC_SELECT_OFFSET;
+	rrd.index1_width = n;
+	rrd.index2_off = QPN_SELECT_OFFSET;
+	rrd.index2_width = m + n;
+	rrd.mask1 = LRH_BTH_MASK;
+	rrd.value1 = LRH_BTH_VALUE;
+	rrd.mask2 = LRH_SC_MASK;
+	rrd.value2 = LRH_SC_VALUE;
+
+	/* add rule 0 */
+	add_rsm_rule(dd, 0, &rrd);
+
 	/* mark RSM map entries as used */
 	rmt->used += rmt_entries;
 	/* map everything else to the mcast/err/vl15 context */

--
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 21/21] IB/hfi1: Add RSM rule for user FECN handling

[Dennis Dalessandro]

From: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Add a receive side mapping rule to extract expected user packets with
the FECN bit set and place them in an eager buffer.  This will allow
user libraries to recognize that a FECN was sent when using header
suppression and respond appropriately.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/staging/rdma/hfi1/chip.c |   84 ++++++++++++++++++++++++++++++++++++++
 1 files changed, 84 insertions(+), 0 deletions(-)

diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index b5edc3a..386309c 100644
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -1034,6 +1034,8 @@ static void read_planned_down_reason_code(struct hfi1_devdata *dd, u8 *pdrrc);
 static void handle_temp_err(struct hfi1_devdata *);
 static void dc_shutdown(struct hfi1_devdata *);
 static void dc_start(struct hfi1_devdata *);
+static int qos_rmt_entries(struct hfi1_devdata *dd, unsigned int *mp,
+			   unsigned int *np);
 
 /*
  * Error interrupt table entry.  This is used as input to the interrupt
@@ -12628,6 +12630,8 @@ static int set_up_context_variables(struct hfi1_devdata *dd)
 	int total_contexts;
 	int ret;
 	unsigned ngroups;
+	int qos_rmt_count;
+	int user_rmt_reduced;
 
 	/*
 	 * Kernel receive contexts:
@@ -12683,6 +12687,19 @@ static int set_up_context_variables(struct hfi1_devdata *dd)
 		total_contexts = num_kernel_contexts + num_user_contexts;
 	}
 
+	/* each user context requires an entry in the RMT */
+	qos_rmt_count = qos_rmt_entries(dd, NULL, NULL);
+	if (qos_rmt_count + num_user_contexts > NUM_MAP_ENTRIES) {
+		user_rmt_reduced = NUM_MAP_ENTRIES - qos_rmt_count;
+		dd_dev_err(dd,
+			   "RMT size is reducing the number of user receive contexts from %d to %d\n",
+			   (int)num_user_contexts,
+			   user_rmt_reduced);
+		/* recalculate */
+		num_user_contexts = user_rmt_reduced;
+		total_contexts = num_kernel_contexts + num_user_contexts;
+	}
+
 	/* the first N are kernel contexts, the rest are user contexts */
 	dd->num_rcv_contexts = total_contexts;
 	dd->n_krcv_queues = num_kernel_contexts;
@@ -13633,6 +13650,72 @@ bail:
 	init_qpmap_table(dd, FIRST_KERNEL_KCTXT, dd->n_krcv_queues - 1);
 }
 
+static void init_user_fecn_handling(struct hfi1_devdata *dd,
+				    struct rsm_map_table *rmt)
+{
+	struct rsm_rule_data rrd;
+	u64 reg;
+	int i, idx, regoff, regidx;
+	u8 offset;
+
+	/* there needs to be enough room in the map table */
+	if (rmt->used + dd->num_user_contexts >= NUM_MAP_ENTRIES) {
+		dd_dev_err(dd, "User FECN handling disabled - too many user contexts allocated\n");
+		return;
+	}
+
+	/*
+	 * RSM will extract the destination context as an index into the
+	 * map table.  The destination contexts are a sequential block
+	 * in the range first_user_ctxt...num_rcv_contexts-1 (inclusive).
+	 * Map entries are accessed as offset + extracted value.  Adjust
+	 * the added offset so this sequence can be placed anywhere in
+	 * the table - as long as the entries themselves do not wrap.
+	 * There are only enough bits in offset for the table size, so
+	 * start with that to allow for a "negative" offset.
+	 */
+	offset = (u8)(NUM_MAP_ENTRIES + (int)rmt->used -
+						(int)dd->first_user_ctxt);
+
+	for (i = dd->first_user_ctxt, idx = rmt->used;
+				i < dd->num_rcv_contexts; i++, idx++) {
+		/* replace with identity mapping */
+		regoff = (idx % 8) * 8;
+		regidx = idx / 8;
+		reg = rmt->map[regidx];
+		reg &= ~(RCV_RSM_MAP_TABLE_RCV_CONTEXT_A_MASK << regoff);
+		reg |= (u64)i << regoff;
+		rmt->map[regidx] = reg;
+	}
+
+	/*
+	 * For RSM intercept of Expected FECN packets:
+	 * o packet type 0 - expected
+	 * o match on F (bit 95), using select/match 1, and
+	 * o match on SH (bit 133), using select/match 2.
+	 *
+	 * Use index 1 to extract the 8-bit receive context from DestQP
+	 * (start at bit 64).  Use that as the RSM map table index.
+	 */
+	rrd.offset = offset;
+	rrd.pkt_type = 0;
+	rrd.field1_off = 95;
+	rrd.field2_off = 133;
+	rrd.index1_off = 64;
+	rrd.index1_width = 8;
+	rrd.index2_off = 0;
+	rrd.index2_width = 0;
+	rrd.mask1 = 1;
+	rrd.value1 = 1;
+	rrd.mask2 = 1;
+	rrd.value2 = 1;
+
+	/* add rule 1 */
+	add_rsm_rule(dd, 1, &rrd);
+
+	rmt->used += dd->num_user_contexts;
+}
+
 static void init_rxe(struct hfi1_devdata *dd)
 {
 	struct rsm_map_table *rmt;
@@ -13643,6 +13726,7 @@ static void init_rxe(struct hfi1_devdata *dd)
 	rmt = alloc_rsm_map_table(dd);
 	/* set up QOS, including the QPN map table */
 	init_qos(dd, rmt);
+	init_user_fecn_handling(dd, rmt);
 	complete_rsm_map_table(dd, rmt);
 	kfree(rmt);
 

--
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] [PATCH 08/21] IB/hfi1: Adjust default MTU to be 10KB

[Dennis Dalessandro]

Sorry about the messed up subject for this patch. stg mail went out to lunch 
and I tried to send by hand. Patchworks doesn't seem to care so I don't 
think I need to re-send anything.

-Denny
--
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] [PATCH 08/21] IB/hfi1: Adjust default MTU to be 10KB

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 3428 bytes --]

On Tue, Apr 12, 2016 at 11:17:09AM -0700, Dennis Dalessandro wrote:
> From: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> 
> Increasing the default MTU size to 10KB improves performance
> for PSM. Change the default MTU to 10KB but constrain
> Verbs MTU to 8KB. Also update default MTU module parameter
> description to be HFI1_DEFAULT_MAX_MTU.
> 
> Reviewed-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> Reviewed-by: Mitko Haralanov <mitko.haralanov-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> Reviewed-by: Jubin John <jubin.john-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> ---
>  drivers/staging/rdma/hfi1/driver.c |    3 ++-
>  drivers/staging/rdma/hfi1/hfi.h    |    4 ++--
>  drivers/staging/rdma/hfi1/qp.c     |    6 +++++-
>  3 files changed, 9 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/staging/rdma/hfi1/driver.c b/drivers/staging/rdma/hfi1/driver.c
> index 34511e5..700c6fa 100644
> --- a/drivers/staging/rdma/hfi1/driver.c
> +++ b/drivers/staging/rdma/hfi1/driver.c
> @@ -75,7 +75,8 @@ DEFINE_MUTEX(hfi1_mutex);	/* general driver use */
>  
>  unsigned int hfi1_max_mtu = HFI1_DEFAULT_MAX_MTU;
>  module_param_named(max_mtu, hfi1_max_mtu, uint, S_IRUGO);
> -MODULE_PARM_DESC(max_mtu, "Set max MTU bytes, default is 8192");
> +MODULE_PARM_DESC(max_mtu, "Set max MTU bytes, default is " __stringify(
> +		 HFI1_DEFAULT_MAX_MTU));
>  
>  unsigned int hfi1_cu = 1;
>  module_param_named(cu, hfi1_cu, uint, S_IRUGO);
> diff --git a/drivers/staging/rdma/hfi1/hfi.h b/drivers/staging/rdma/hfi1/hfi.h
> index ac553f1..ff04593 100644
> --- a/drivers/staging/rdma/hfi1/hfi.h
> +++ b/drivers/staging/rdma/hfi1/hfi.h
> @@ -455,9 +455,9 @@ struct rvt_sge_state;
>  #define HLS_UP (HLS_UP_INIT | HLS_UP_ARMED | HLS_UP_ACTIVE)
>  
>  /* use this MTU size if none other is given */
> -#define HFI1_DEFAULT_ACTIVE_MTU 8192
> +#define HFI1_DEFAULT_ACTIVE_MTU 10240
>  /* use this MTU size as the default maximum */
> -#define HFI1_DEFAULT_MAX_MTU 8192
> +#define HFI1_DEFAULT_MAX_MTU 10240
>  /* default partition key */
>  #define DEFAULT_PKEY 0xffff
>  
> diff --git a/drivers/staging/rdma/hfi1/qp.c b/drivers/staging/rdma/hfi1/qp.c
> index 29a5ad2..e68d08a 100644
> --- a/drivers/staging/rdma/hfi1/qp.c
> +++ b/drivers/staging/rdma/hfi1/qp.c
> @@ -167,8 +167,12 @@ static inline int opa_mtu_enum_to_int(int mtu)
>   */
>  static inline int verbs_mtu_enum_to_int(struct ib_device *dev, enum ib_mtu mtu)
>  {
> -	int val = opa_mtu_enum_to_int((int)mtu);
> +	int val;
>  
> +	/* Constraining 10KB packets to 8KB packets */
> +	if (mtu == (enum ib_mtu)OPA_MTU_10240)
> +		mtu = OPA_MTU_8192;
> +	val = opa_mtu_enum_to_int((int)mtu);

Just spotted my attention,
Curious names, castings and type conversions.

>  	if (val > 0)
>  		return val;
>  	return ib_mtu_enum_to_int(mtu);
> 
> --
> 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

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

Re: [PATCH 00/21] IB/hfi1: First round of fixes for 4.7

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 502 bytes --]

On Tue, Apr 12, 2016 at 10:49:52AM -0700, Dennis Dalessandro wrote:
> These patches do some clean ups and fix issues which may not be appropriate for
> an RC so we are targeting these for 4.7. The current patch series applies on top
> of the "Important bug fixes for 4.6 RC" patch set submitted just prior to this.
> 
> Patches can be viewed in my GitHub repo at:
> https://github.com/ddalessa/kernel/tree/for-4.7.

Please follow SubmittingPatches guide - submitting fixes section.

Thanks.

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

Re: [PATCH] [PATCH 08/21] IB/hfi1: Adjust default MTU to be 10KB

[Dennis Dalessandro]

On Mon, Apr 18, 2016 at 10:07:08AM +0300, Leon Romanovsky wrote:
 
>> diff --git a/drivers/staging/rdma/hfi1/qp.c 
>> b/drivers/staging/rdma/hfi1/qp.c
>> index 29a5ad2..e68d08a 100644
>> --- a/drivers/staging/rdma/hfi1/qp.c
>> +++ b/drivers/staging/rdma/hfi1/qp.c
>> @@ -167,8 +167,12 @@ static inline int opa_mtu_enum_to_int(int mtu)
>>   */
>>  static inline int verbs_mtu_enum_to_int(struct ib_device *dev, enum ib_mtu mtu)
>>  {
>> -	int val = opa_mtu_enum_to_int((int)mtu);
>> +	int val;
>>  
>> +	/* Constraining 10KB packets to 8KB packets */
>> +	if (mtu == (enum ib_mtu)OPA_MTU_10240)
>> +		mtu = OPA_MTU_8192;
>> +	val = opa_mtu_enum_to_int((int)mtu);
>
>Just spotted my attention,
>Curious names, castings and type conversions.
>

Is there a question being implied here? I'm not clear on what kind of 
response you are looking for here.

-Denny
--
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 00/21] IB/hfi1: First round of fixes for 4.7

[Dennis Dalessandro]

On Mon, Apr 18, 2016 at 10:09:36AM +0300, Leon Romanovsky wrote:
>On Tue, Apr 12, 2016 at 10:49:52AM -0700, Dennis Dalessandro wrote:
>> These patches do some clean ups and fix issues which may not be appropriate for
>> an RC so we are targeting these for 4.7. The current patch series applies on top
>> of the "Important bug fixes for 4.6 RC" patch set submitted just prior to this.
>> 
>> Patches can be viewed in my GitHub repo at:
>> https://github.com/ddalessa/kernel/tree/for-4.7.
>
>Please follow SubmittingPatches guide - submitting fixes section.
>

Can you please be more specific with what your complaint is here? Which 
section are you referring to? There is no "submitting fixes" section in the 
document.

-Denny
--
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 00/21] IB/hfi1: First round of fixes for 4.7

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 1191 bytes --]

On Mon, Apr 18, 2016 at 08:07:55AM -0400, Dennis Dalessandro wrote:
> On Mon, Apr 18, 2016 at 10:09:36AM +0300, Leon Romanovsky wrote:
> >On Tue, Apr 12, 2016 at 10:49:52AM -0700, Dennis Dalessandro wrote:
> >>These patches do some clean ups and fix issues which may not be appropriate for
> >>an RC so we are targeting these for 4.7. The current patch series applies on top
> >>of the "Important bug fixes for 4.6 RC" patch set submitted just prior to this.
> >>
> >>Patches can be viewed in my GitHub repo at:
> >>https://github.com/ddalessa/kernel/tree/for-4.7.
> >
> >Please follow SubmittingPatches guide - submitting fixes section.
> >
> 
> Can you please be more specific with what your complaint is here? Which
> section are you referring to? There is no "submitting fixes" section in the
> document.


If your patch fixes a bug in a specific commit, e.g. you found an issue
using git-bisect, please use the 'Fixes:' tag with the first 12 characters of
the SHA-1 ID, and the one line summary.  For example:

  Fixes: e21d2170f366 ("video: remove unnecessary platform_set_drvdata()")

https://www.kernel.org/doc/Documentation/SubmittingPatches

>
> -Denny

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

Re: [PATCH] [PATCH 08/21] IB/hfi1: Adjust default MTU to be 10KB

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 1244 bytes --]

On Mon, Apr 18, 2016 at 08:03:36AM -0400, Dennis Dalessandro wrote:
> On Mon, Apr 18, 2016 at 10:07:08AM +0300, Leon Romanovsky wrote:
> 
> >>diff --git a/drivers/staging/rdma/hfi1/qp.c
> >>b/drivers/staging/rdma/hfi1/qp.c
> >>index 29a5ad2..e68d08a 100644
> >>--- a/drivers/staging/rdma/hfi1/qp.c
> >>+++ b/drivers/staging/rdma/hfi1/qp.c
> >>@@ -167,8 +167,12 @@ static inline int opa_mtu_enum_to_int(int mtu)
> >>  */
> >> static inline int verbs_mtu_enum_to_int(struct ib_device *dev, enum ib_mtu mtu)
> >> {
> >>-	int val = opa_mtu_enum_to_int((int)mtu);
> >>+	int val;
> >>+	/* Constraining 10KB packets to 8KB packets */
> >>+	if (mtu == (enum ib_mtu)OPA_MTU_10240)
> >>+		mtu = OPA_MTU_8192;
> >>+	val = opa_mtu_enum_to_int((int)mtu);
> >
> >Just spotted my attention,
> >Curious names, castings and type conversions.
> >
> 
> Is there a question being implied here? I'm not clear on what kind of
> response you are looking for here.

No, there is no question here.

mtu is type of "enum ib_mtu", which is converted to int and supplied to
function which has opa_mtu_ENUM_to_int name. From the name of that
function looks that the parameter supplied should be ENUM, but it is not
the case.

> 
> -Denny

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

Re: [PATCH 00/21] IB/hfi1: First round of fixes for 4.7

[Dennis Dalessandro]

On Mon, Apr 18, 2016 at 03:19:44PM +0300, Leon Romanovsky wrote:
>On Mon, Apr 18, 2016 at 08:07:55AM -0400, Dennis Dalessandro wrote:
>> On Mon, Apr 18, 2016 at 10:09:36AM +0300, Leon Romanovsky wrote:
>> >On Tue, Apr 12, 2016 at 10:49:52AM -0700, Dennis Dalessandro wrote:
>> >>These patches do some clean ups and fix issues which may not be appropriate for
>> >>an RC so we are targeting these for 4.7. The current patch series applies on top
>> >>of the "Important bug fixes for 4.6 RC" patch set submitted just prior to this.
>> >>
>> >>Patches can be viewed in my GitHub repo at:
>> >>https://github.com/ddalessa/kernel/tree/for-4.7.
>> >
>> >Please follow SubmittingPatches guide - submitting fixes section.
>> >
>> 
>> Can you please be more specific with what your complaint is here? Which
>> section are you referring to? There is no "submitting fixes" section in the
>> document.
>
>
>If your patch fixes a bug in a specific commit, e.g. you found an issue
>using git-bisect, please use the 'Fixes:' tag with the first 12 characters of
>the SHA-1 ID, and the one line summary.  For example:
>
>  Fixes: e21d2170f366 ("video: remove unnecessary platform_set_drvdata()")
>
>https://www.kernel.org/doc/Documentation/SubmittingPatches

When there is a specific commit that broke something, sure. I think folks 
usually add that when appropriate, it's not always the case. "Fix" is a 
general term in a lot of patches. Could be day-0 bug, or the result of 
multiple commits coming together, etc.

Regardless, is there a patch here in this series that you need more back 
story on?  Please let me know and I'll track it down.

-Denny
--
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 00/21] IB/hfi1: First round of fixes for 4.7

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 2129 bytes --]

On Mon, Apr 18, 2016 at 09:38:43AM -0400, Dennis Dalessandro wrote:
> On Mon, Apr 18, 2016 at 03:19:44PM +0300, Leon Romanovsky wrote:
> >On Mon, Apr 18, 2016 at 08:07:55AM -0400, Dennis Dalessandro wrote:
> >>On Mon, Apr 18, 2016 at 10:09:36AM +0300, Leon Romanovsky wrote:
> >>>On Tue, Apr 12, 2016 at 10:49:52AM -0700, Dennis Dalessandro wrote:
> >>>>These patches do some clean ups and fix issues which may not be appropriate for
> >>>>an RC so we are targeting these for 4.7. The current patch series applies on top
> >>>>of the "Important bug fixes for 4.6 RC" patch set submitted just prior to this.
> >>>>
> >>>>Patches can be viewed in my GitHub repo at:
> >>>>https://github.com/ddalessa/kernel/tree/for-4.7.
> >>>
> >>>Please follow SubmittingPatches guide - submitting fixes section.
> >>>
> >>
> >>Can you please be more specific with what your complaint is here? Which
> >>section are you referring to? There is no "submitting fixes" section in the
> >>document.
> >
> >
> >If your patch fixes a bug in a specific commit, e.g. you found an issue
> >using git-bisect, please use the 'Fixes:' tag with the first 12 characters of
> >the SHA-1 ID, and the one line summary.  For example:
> >
> > Fixes: e21d2170f366 ("video: remove unnecessary platform_set_drvdata()")
> >
> >https://www.kernel.org/doc/Documentation/SubmittingPatches
> 
> When there is a specific commit that broke something, sure. I think folks
> usually add that when appropriate, it's not always the case. "Fix" is a
> general term in a lot of patches. Could be day-0 bug, or the result of
> multiple commits coming together, etc.

Fixes line will help to the users and developers to get a sense if the
failure they see can be related to the code which they have (SHA-1).

Especially, it is valuable when it comes from the vendor who does
extensive testing and can perfectly explain what they are fixing.

> 
> Regardless, is there a patch here in this series that you need more back
> story on?  Please let me know and I'll track it down.

Every commit which suppose to fix something.


> 
> -Denny

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

Re: [PATCH] [PATCH 08/21] IB/hfi1: Adjust default MTU to be 10KB

[David Woodhouse]

[-- Attachment #1: Type: text/plain, Size: 243 bytes --]

On Mon, 2016-04-18 at 15:30 +0300, Leon Romanovsky wrote:
> From the name of that function looks that the parameter supplied
> should be ENUM, but it is not the case.

The implicit question might, perhaps, be "why?" :)

-- 
dwmw2



[-- Attachment #2: smime.p7s --]
[-- Type: application/x-pkcs7-signature, Size: 5691 bytes --]

Re: [PATCH] [PATCH 08/21] IB/hfi1: Adjust default MTU to be 10KB

[Ira Weiny]

On Mon, Apr 18, 2016 at 02:42:22PM -0400, David Woodhouse wrote:
> On Mon, 2016-04-18 at 15:30 +0300, Leon Romanovsky wrote:
> > From the name of that function looks that the parameter supplied
> > should be ENUM, but it is not the case.
> 
> The implicit question might, perhaps, be "why?" :)

OPA defines MTU values which are not "ib_mtu" enumerations.  This code is
processing 2 different "types" of enumerations.  The IB compatible ones as well
as new ones which are specific to OPA.

The OPA enums are not defined as a type but perhaps they should be?

From include/rdma/opa_port_info.h:

enum {  
        OPA_MTU_8192  = 6,
        OPA_MTU_10240 = 7,
};


Also I just noticed this comment for the function which is no longer
appropriate:

...
 * The actual flag used to determine "8k MTU" will change and is currently
 * unknown.
...

We should remove the comment.

Ira

--
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 00/21] IB/hfi1: First round of fixes for 4.7

[Doug Ledford]

[-- Attachment #1: Type: text/plain, Size: 3233 bytes --]

On 04/12/2016 01:49 PM, Dennis Dalessandro wrote:
> These patches do some clean ups and fix issues which may not be appropriate for
> an RC so we are targeting these for 4.7. The current patch series applies on top
> of the "Important bug fixes for 4.6 RC" patch set submitted just prior to this.
> 
> Patches can be viewed in my GitHub repo at:
> https://github.com/ddalessa/kernel/tree/for-4.7.
> 
> ---
> 
> Dean Luick (13):
>       IB/hfi1: Fix sysfs file offset usage
>       IB/hfi1: Fix i2c resource reservation checks
>       IB/hfi1: Fix QOS num_vl bit width
>       IB/hfi1: Remove invalid QOS check
>       IB/hfi1: Fix QOS rule mappings
>       IB/hfi1: Correctly obtain the full service class
>       IB/hfi1: Simplify init_qpmap_table()
>       IB/hfi1: Guard against concurrent I2C access across all chains
>       IB/hfi1: Fix double QSFP resource acquire on cache refresh
>       IB/hfi1: Extract RSM map table init from QOS
>       IB/hfi1: Move QOS decision logic into its own function
>       IB/hfi1: Create a routine to set a receive side mapping rule
>       IB/hfi1: Add RSM rule for user FECN handling
> 
> Easwar Hariharan (2):
>       IB/hfi1: Always turn on CDRs for low power QSFP modules
>       IB/hfi1: Remove module presence check outside pre-LNI checks
> 
> Jianxin Xiong (1):
>       IB/hfi1: Reduce kernel context pio buffer allocation
> 
> Jubin John (1):
>       IB/hfi1: Change default number of user contexts
> 
> Mike Marciniszyn (2):
>       IB/hfi1: Remove unreachable code
>       IB/hfi1: Use global defines for upper bits in opcode
> 
> Sebastian Sanchez (2):
>       IB/hfi1: Adjust default MTU to be 10KB
>       IB/hfi1: Check P_KEY for all sent packets from user mode
> 
> 
>  drivers/infiniband/sw/rdmavt/qp.c          |    6 
>  drivers/staging/rdma/hfi1/affinity.c       |   93 +++--
>  drivers/staging/rdma/hfi1/affinity.h       |   19 +
>  drivers/staging/rdma/hfi1/chip.c           |  485 +++++++++++++++++++---------
>  drivers/staging/rdma/hfi1/chip.h           |    1 
>  drivers/staging/rdma/hfi1/chip_registers.h |    1 
>  drivers/staging/rdma/hfi1/diag.c           |    3 
>  drivers/staging/rdma/hfi1/driver.c         |    3 
>  drivers/staging/rdma/hfi1/firmware.c       |    9 -
>  drivers/staging/rdma/hfi1/hfi.h            |   11 -
>  drivers/staging/rdma/hfi1/init.c           |    1 
>  drivers/staging/rdma/hfi1/pio.c            |   52 ++-
>  drivers/staging/rdma/hfi1/pio.h            |    4 
>  drivers/staging/rdma/hfi1/platform.c       |   99 +++---
>  drivers/staging/rdma/hfi1/qp.c             |    6 
>  drivers/staging/rdma/hfi1/qsfp.c           |   58 ++-
>  drivers/staging/rdma/hfi1/qsfp.h           |   15 -
>  drivers/staging/rdma/hfi1/rc.c             |    2 
>  drivers/staging/rdma/hfi1/sysfs.c          |    4 
>  drivers/staging/rdma/hfi1/user_sdma.c      |    7 
>  drivers/staging/rdma/hfi1/verbs.c          |  108 ++++--
>  drivers/staging/rdma/hfi1/verbs.h          |    3 
>  22 files changed, 639 insertions(+), 351 deletions(-)
> 

Series applied, thanks.

-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
              GPG KeyID: 0E572FDD



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 884 bytes --]