iw_cxgb4: Atomically flush per QP HW CQEs

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@vger.kernel.org
Signed-off-by: Potnuri Bharat Teja <bharat@chelsio.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

3 files changed, 13 insertions, 4 deletions

diff --git a/drivers/infiniband/hw/cxgb4/cq.c b/drivers/infiniband/hw/cxgb4/cq.c
index 6f2b26126c64..2be2e1ac1b5f 100644
--- a/drivers/infiniband/hw/cxgb4/cq.c
+++ b/drivers/infiniband/hw/cxgb4/cq.c
@@ -315,7 +315,7 @@ static void advance_oldest_read(struct t4_wq *wq)
  * 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 *chp)
                 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 *chp)
 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);
         }
 }
 
diff --git a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
index a60def23e9ef..831027717121 100644
--- 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 *rdev, int size);
 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);
diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c
index de77b6027d69..ae167b686608 100644
--- 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 *qhp, struct c4iw_cq *rchp,
         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);