Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/gi…

…t/dledford/rdma

Pull rdma fixes from Doug Ledford:
 "This is the second pull request for the rdma subsystem.  Most of the
  patches are small and obvious.  I took two patches in that are larger
  than I wanted this late in the cycle.

  The first is the hfi1 patch that implements a work queue to test the
  QSFP read state.  I originally rejected the first patch for this
  (which would have place up to 20 seconds worth of udelays in their
  probe routine).  They then rewrote it the way I wanted (use delayed
  work tasks to wait asynchronously up to 20 seconds for the QSFP to
  come alive), so I can't really complain about the size of getting what
  I asked for :-/.

  The second is large because it switches the rcu locking in the debugfs
  code.  Since a locking change like this is done all at once, the size
  it what it is.  It resolves a litany of debug messages from the
  kernel, so I pulled it in for -rc.

  The rest are all typical -rc worthy patches I think.

  There will still be a third -rc pull request from the rdma subsystem
  this release.  I hope to have that one ready to go by the end of this
  week or early next.

  Summary:

   - a smattering of small fixes across the core, ipoib, i40iw, isert,
     cxgb4, and mlx4

   - a slightly larger group of fixes to each of mlx5 and hfi1"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma:
  IB/hfi1: Rework debugfs to use SRCU
  IB/hfi1: Make n_krcvqs be an unsigned long integer
  IB/hfi1: Add QSFP sanity pre-check
  IB/hfi1: Fix AHG KDETH Intr shift
  IB/hfi1: Fix SGE length for misaligned PIO copy
  IB/mlx5: Don't return errors from poll_cq
  IB/mlx5: Use TIR number based on selector
  IB/mlx5: Simplify code by removing return variable
  IB/mlx5: Return EINVAL when caller specifies too many SGEs
  IB/mlx4: Don't return errors from poll_cq
  Revert "IB/mlx4: Return EAGAIN for any error in mlx4_ib_poll_one"
  IB/ipoib: Fix memory corruption in ipoib cm mode connect flow
  IB/core: Fix use after free in send_leave function
  IB/cxgb4: Make _free_qp static to silence build warning
  IB/isert: Properly release resources on DEVICE_REMOVAL
  IB/hfi1: Fix the size parameter to find_first_bit
  IB/mlx5: Fix the size parameter to find_first_bit
  IB/hfi1: Clean up type used and casting
  i40iw: Receive notification events correctly
  i40iw: Update hw_iwarp_state

drivers/infiniband/core/multicast.c

@@ -106,7 +106,6 @@ struct mcast_group {
 	atomic_t		refcount;
 	enum mcast_group_state	state;
 	struct ib_sa_query	*query;
-	int			query_id;
 	u16			pkey_index;
 	u8			leave_state;
 	int			retries;
@@ -340,11 +339,7 @@ static int send_join(struct mcast_group *group, struct mcast_member *member)
 				       member->multicast.comp_mask,
 				       3000, GFP_KERNEL, join_handler, group,
 				       &group->query);
-	if (ret >= 0) {
-		group->query_id = ret;
-		ret = 0;
-	}
-	return ret;
+	return (ret > 0) ? 0 : ret;
 }
 
 static int send_leave(struct mcast_group *group, u8 leave_state)
@@ -364,11 +359,7 @@ static int send_leave(struct mcast_group *group, u8 leave_state)
 				       IB_SA_MCMEMBER_REC_JOIN_STATE,
 				       3000, GFP_KERNEL, leave_handler,
 				       group, &group->query);
-	if (ret >= 0) {
-		group->query_id = ret;
-		ret = 0;
-	}
-	return ret;
+	return (ret > 0) ? 0 : ret;
 }
 
 static void join_group(struct mcast_group *group, struct mcast_member *member,

drivers/infiniband/hw/cxgb4/qp.c

@@ -683,7 +683,7 @@ static int build_inv_stag(union t4_wr *wqe, struct ib_send_wr *wr,
 	return 0;
 }
 
-void _free_qp(struct kref *kref)
+static void _free_qp(struct kref *kref)
 {
 	struct c4iw_qp *qhp;
 

drivers/infiniband/hw/hfi1/chip.c

@@ -9490,6 +9490,78 @@ static void init_lcb(struct hfi1_devdata *dd)
 	write_csr(dd, DC_LCB_CFG_TX_FIFOS_RESET, 0x00);
 }
 
+/*
+ * Perform a test read on the QSFP.  Return 0 on success, -ERRNO
+ * on error.
+ */
+static int test_qsfp_read(struct hfi1_pportdata *ppd)
+{
+	int ret;
+	u8 status;
+
+	/* report success if not a QSFP */
+	if (ppd->port_type != PORT_TYPE_QSFP)
+		return 0;
+
+	/* read byte 2, the status byte */
+	ret = one_qsfp_read(ppd, ppd->dd->hfi1_id, 2, &status, 1);
+	if (ret < 0)
+		return ret;
+	if (ret != 1)
+		return -EIO;
+
+	return 0; /* success */
+}
+
+/*
+ * Values for QSFP retry.
+ *
+ * Give up after 10s (20 x 500ms).  The overall timeout was empirically
+ * arrived at from experience on a large cluster.
+ */
+#define MAX_QSFP_RETRIES 20
+#define QSFP_RETRY_WAIT 500 /* msec */
+
+/*
+ * Try a QSFP read.  If it fails, schedule a retry for later.
+ * Called on first link activation after driver load.
+ */
+static void try_start_link(struct hfi1_pportdata *ppd)
+{
+	if (test_qsfp_read(ppd)) {
+		/* read failed */
+		if (ppd->qsfp_retry_count >= MAX_QSFP_RETRIES) {
+			dd_dev_err(ppd->dd, "QSFP not responding, giving up\n");
+			return;
+		}
+		dd_dev_info(ppd->dd,
+			    "QSFP not responding, waiting and retrying %d\n",
+			    (int)ppd->qsfp_retry_count);
+		ppd->qsfp_retry_count++;
+		queue_delayed_work(ppd->hfi1_wq, &ppd->start_link_work,
+				   msecs_to_jiffies(QSFP_RETRY_WAIT));
+		return;
+	}
+	ppd->qsfp_retry_count = 0;
+
+	/*
+	 * Tune the SerDes to a ballpark setting for optimal signal and bit
+	 * error rate.  Needs to be done before starting the link.
+	 */
+	tune_serdes(ppd);
+	start_link(ppd);
+}
+
+/*
+ * Workqueue function to start the link after a delay.
+ */
+void handle_start_link(struct work_struct *work)
+{
+	struct hfi1_pportdata *ppd = container_of(work, struct hfi1_pportdata,
+						  start_link_work.work);
+	try_start_link(ppd);
+}
+
 int bringup_serdes(struct hfi1_pportdata *ppd)
 {
 	struct hfi1_devdata *dd = ppd->dd;
@@ -9525,14 +9597,8 @@ int bringup_serdes(struct hfi1_pportdata *ppd)
 		set_qsfp_int_n(ppd, 1);
 	}
 
-	/*
-	 * Tune the SerDes to a ballpark setting for
-	 * optimal signal and bit error rate
-	 * Needs to be done before starting the link
-	 */
-	tune_serdes(ppd);
-
-	return start_link(ppd);
+	try_start_link(ppd);
+	return 0;
 }
 
 void hfi1_quiet_serdes(struct hfi1_pportdata *ppd)
@@ -9549,6 +9615,10 @@ void hfi1_quiet_serdes(struct hfi1_pportdata *ppd)
 	ppd->driver_link_ready = 0;
 	ppd->link_enabled = 0;
 
+	ppd->qsfp_retry_count = MAX_QSFP_RETRIES; /* prevent more retries */
+	flush_delayed_work(&ppd->start_link_work);
+	cancel_delayed_work_sync(&ppd->start_link_work);
+
 	ppd->offline_disabled_reason =
 			HFI1_ODR_MASK(OPA_LINKDOWN_REASON_SMA_DISABLED);
 	set_link_down_reason(ppd, OPA_LINKDOWN_REASON_SMA_DISABLED, 0,
@@ -12865,7 +12935,7 @@ static int set_up_interrupts(struct hfi1_devdata *dd)
  */
 static int set_up_context_variables(struct hfi1_devdata *dd)
 {
-	int num_kernel_contexts;
+	unsigned long num_kernel_contexts;
 	int total_contexts;
 	int ret;
 	unsigned ngroups;
@@ -12894,9 +12964,9 @@ static int set_up_context_variables(struct hfi1_devdata *dd)
 	 */
 	if (num_kernel_contexts > (dd->chip_send_contexts - num_vls - 1)) {
 		dd_dev_err(dd,
-			   "Reducing # kernel rcv contexts to: %d, from %d\n",
+			   "Reducing # kernel rcv contexts to: %d, from %lu\n",
 			   (int)(dd->chip_send_contexts - num_vls - 1),
-			   (int)num_kernel_contexts);
+			   num_kernel_contexts);
 		num_kernel_contexts = dd->chip_send_contexts - num_vls - 1;
 	}
 	/*

drivers/infiniband/hw/hfi1/chip.h

@@ -706,6 +706,7 @@ void handle_link_up(struct work_struct *work);
 void handle_link_down(struct work_struct *work);
 void handle_link_downgrade(struct work_struct *work);
 void handle_link_bounce(struct work_struct *work);
+void handle_start_link(struct work_struct *work);
 void handle_sma_message(struct work_struct *work);
 void reset_qsfp(struct hfi1_pportdata *ppd);
 void qsfp_event(struct work_struct *work);