From: Wei Lin Guay <wei.lin.guay@oracle.com>
Date: Tue, 14 Jun 2016 12:33:07 +0000 (+0200)
Subject: sif: cq: Do not invalidate the CQ until completion of events
X-Git-Tag: v4.1.12-92~129^2~19
X-Git-Url: https://www.infradead.org/git/?a=commitdiff_plain;h=932e0ce2d9cafb3107130b311354f580b7f71410;p=users%2Fjedix%2Flinux-maple.git

sif: cq: Do not invalidate the CQ until completion of events

The CQ completion event might be performed while the CQ
has been invalidated. Besides, adding a check in the
sif_req_notify_cq for not rearming the CQ if the CQ
has been invalidated.

This fixes a scenario reported in Orabug: #23491094.

Signed-off-by: Wei Lin Guay <wei.lin.guay@oracle.com>
Reviewed-by: Knut Omang <knut.omang@oracle.com>
---

diff --git a/drivers/infiniband/hw/sif/sif_cq.c b/drivers/infiniband/hw/sif/sif_cq.c
index 08466881ceed9..7d0dc4b8fd6f5 100644
--- a/drivers/infiniband/hw/sif/sif_cq.c
+++ b/drivers/infiniband/hw/sif/sif_cq.c
@@ -325,6 +325,12 @@ int destroy_cq(struct sif_cq *cq)
 		atomic_add(miss_cnt, &sdev->cq_miss_cnt);
 		atomic_add(miss_occ, &sdev->cq_miss_occ);
 	}
+
+	/* Wait for any in-progress event queue entry for this CQ to be finished */
+	if (atomic_dec_and_test(&cq->refcnt))
+		complete(&cq->cleanup_ok);
+	wait_for_completion(&cq->cleanup_ok);
+
 	ret = sif_invalidate_cq_hw(sdev, index, PCM_WAIT);
 	if (ret) {
 		sif_log(sdev, SIF_INFO,
@@ -347,11 +353,6 @@ int sif_release_cq(struct sif_dev *sdev, int index)
 	struct sif_pd *pd = cq->pd;
 	struct sif_cq_sw *cq_sw = get_sif_cq_sw(sdev, index);
 
-	/* Wait for any in-progress event queue entry for this CQ to be finished */
-	if (atomic_dec_and_test(&cq->refcnt))
-		complete(&cq->cleanup_ok);
-	wait_for_completion(&cq->cleanup_ok);
-
 	/* Make sure any completions on the cq TLB invalidate
 	 * for priv.qp does arrive before the cq is destroyed..
 	 */
@@ -907,6 +908,10 @@ int sif_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
 	if (flags & IB_CQ_SOLICITED)
 		wr.se = 1;
 
+	/* If a CQ is not valid, do not rearm the CQ. */
+	if (!get_psif_cq_hw__valid(&cq->d))
+		return 0;
+
 	/* We should never miss events in psif so we have no need for a separate
 	 *  handling of IB_CQ_REPORT_MISSED_EVENTS - ignore it.
 	 */