qed: rdma - don't wait for resources under hw error recovery flow
authorShai Malin <smalin@marvell.com>
Wed, 22 Sep 2021 10:53:26 +0000 (13:53 +0300)
committerDavid S. Miller <davem@davemloft.net>
Wed, 22 Sep 2021 13:38:34 +0000 (14:38 +0100)
If the HW device is during recovery, the HW resources will never return,
hence we shouldn't wait for the CID (HW context ID) bitmaps to clear.
This fix speeds up the error recovery flow.

Fixes: 64515dc899df ("qed: Add infrastructure for error detection and recovery")
Signed-off-by: Michal Kalderon <mkalderon@marvell.com>
Signed-off-by: Ariel Elior <aelior@marvell.com>
Signed-off-by: Shai Malin <smalin@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/qlogic/qed/qed_iwarp.c
drivers/net/ethernet/qlogic/qed/qed_roce.c

index fc8b3e6..186d004 100644 (file)
@@ -1297,6 +1297,14 @@ qed_iwarp_wait_cid_map_cleared(struct qed_hwfn *p_hwfn, struct qed_bmap *bmap)
        prev_weight = weight;
 
        while (weight) {
+               /* If the HW device is during recovery, all resources are
+                * immediately reset without receiving a per-cid indication
+                * from HW. In this case we don't expect the cid_map to be
+                * cleared.
+                */
+               if (p_hwfn->cdev->recov_in_prog)
+                       return 0;
+
                msleep(QED_IWARP_MAX_CID_CLEAN_TIME);
 
                weight = bitmap_weight(bmap->bitmap, bmap->max_count);
index f16a157..cf5baa5 100644 (file)
@@ -77,6 +77,14 @@ void qed_roce_stop(struct qed_hwfn *p_hwfn)
         * Beyond the added delay we clear the bitmap anyway.
         */
        while (bitmap_weight(rcid_map->bitmap, rcid_map->max_count)) {
+               /* If the HW device is during recovery, all resources are
+                * immediately reset without receiving a per-cid indication
+                * from HW. In this case we don't expect the cid bitmap to be
+                * cleared.
+                */
+               if (p_hwfn->cdev->recov_in_prog)
+                       return;
+
                msleep(100);
                if (wait_count++ > 20) {
                        DP_NOTICE(p_hwfn, "cid bitmap wait timed out\n");