[PATCH 4.16 29/52] iw_cxgb4: Atomically flush per QP HW CQEs

[Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>]

4.16-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Bharat Potnuri <bharat@xxxxxxxxxxx>

commit 2df19e19ae90d94fd8724083f161f368a2797537 upstream.

When a CQ is shared by multiple QPs, c4iw_flush_hw_cq() needs to acquire
corresponding QP lock before moving the CQEs into its corresponding SW
queue and accessing the SQ contents for completing a WR.
Ignore CQEs if corresponding QP is already flushed.

Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Potnuri Bharat Teja <bharat@xxxxxxxxxxx>
Reviewed-by: Steve Wise <swise@xxxxxxxxxxxxxxxxxxxxx>
Signed-off-by: Doug Ledford <dledford@xxxxxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>

---
 drivers/infiniband/hw/cxgb4/cq.c       |   11 ++++++++++-
 drivers/infiniband/hw/cxgb4/iw_cxgb4.h |    2 +-
 drivers/infiniband/hw/cxgb4/qp.c       |    4 ++--
 3 files changed, 13 insertions(+), 4 deletions(-)

--- a/drivers/infiniband/hw/cxgb4/cq.c
+++ b/drivers/infiniband/hw/cxgb4/cq.c
@@ -315,7 +315,7 @@ static void advance_oldest_read(struct t
  * Deal with out-of-order and/or completions that complete
  * prior unsignalled WRs.
  */
-void c4iw_flush_hw_cq(struct c4iw_cq *chp)
+void c4iw_flush_hw_cq(struct c4iw_cq *chp, struct c4iw_qp *flush_qhp)
 {
 	struct t4_cqe *hw_cqe, *swcqe, read_cqe;
 	struct c4iw_qp *qhp;
@@ -339,6 +339,13 @@ void c4iw_flush_hw_cq(struct c4iw_cq *ch
 		if (qhp == NULL)
 			goto next_cqe;
 
+		if (flush_qhp != qhp) {
+			spin_lock(&qhp->lock);
+
+			if (qhp->wq.flushed == 1)
+				goto next_cqe;
+		}
+
 		if (CQE_OPCODE(hw_cqe) == FW_RI_TERMINATE)
 			goto next_cqe;
 
@@ -390,6 +397,8 @@ void c4iw_flush_hw_cq(struct c4iw_cq *ch
 next_cqe:
 		t4_hwcq_consume(&chp->cq);
 		ret = t4_next_hw_cqe(&chp->cq, &hw_cqe);
+		if (qhp && flush_qhp != qhp)
+			spin_unlock(&qhp->lock);
 	}
 }
 
--- a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
+++ b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
@@ -1053,7 +1053,7 @@ u32 c4iw_pblpool_alloc(struct c4iw_rdev
 void c4iw_pblpool_free(struct c4iw_rdev *rdev, u32 addr, int size);
 u32 c4iw_ocqp_pool_alloc(struct c4iw_rdev *rdev, int size);
 void c4iw_ocqp_pool_free(struct c4iw_rdev *rdev, u32 addr, int size);
-void c4iw_flush_hw_cq(struct c4iw_cq *chp);
+void c4iw_flush_hw_cq(struct c4iw_cq *chp, struct c4iw_qp *flush_qhp);
 void c4iw_count_rcqes(struct t4_cq *cq, struct t4_wq *wq, int *count);
 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp);
 int c4iw_flush_rq(struct t4_wq *wq, struct t4_cq *cq, int count);
--- a/drivers/infiniband/hw/cxgb4/qp.c
+++ b/drivers/infiniband/hw/cxgb4/qp.c
@@ -1343,12 +1343,12 @@ static void __flush_qp(struct c4iw_qp *q
 	qhp->wq.flushed = 1;
 	t4_set_wq_in_error(&qhp->wq);
 
-	c4iw_flush_hw_cq(rchp);
+	c4iw_flush_hw_cq(rchp, qhp);
 	c4iw_count_rcqes(&rchp->cq, &qhp->wq, &count);
 	rq_flushed = c4iw_flush_rq(&qhp->wq, &rchp->cq, count);
 
 	if (schp != rchp)
-		c4iw_flush_hw_cq(schp);
+		c4iw_flush_hw_cq(schp, qhp);
 	sq_flushed = c4iw_flush_sq(qhp);
 
 	spin_unlock(&qhp->lock);