RDMA/irdma: Return correct WC error for bind operation failure
authorSindhu-Devale <sindhu.devale@intel.com>
Tue, 6 Sep 2022 22:32:42 +0000 (17:32 -0500)
committerLeon Romanovsky <leon@kernel.org>
Wed, 7 Sep 2022 08:22:18 +0000 (11:22 +0300)
When a QP and a MR on a local host are in different PDs, the HW generates
an asynchronous event (AE). The same AE is generated when a QP and a MW
are in different PDs during a bind operation. Return the more appropriate
IBV_WC_MW_BIND_ERR for the latter case by checking the OP type from the
CQE in error.

Fixes: 551c46edc769 ("RDMA/irdma: Add user/kernel shared libraries")
Signed-off-by: Sindhu-Devale <sindhu.devale@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Link: https://lore.kernel.org/r/20220906223244.1119-4-shiraz.saleem@intel.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
drivers/infiniband/hw/irdma/uk.c

index daeab5d..d003ad8 100644 (file)
@@ -1005,6 +1005,7 @@ int irdma_uk_cq_poll_cmpl(struct irdma_cq_uk *cq,
        int ret_code;
        bool move_cq_head = true;
        u8 polarity;
+       u8 op_type;
        bool ext_valid;
        __le64 *ext_cqe;
 
@@ -1187,7 +1188,6 @@ int irdma_uk_cq_poll_cmpl(struct irdma_cq_uk *cq,
                        do {
                                __le64 *sw_wqe;
                                u64 wqe_qword;
-                               u8 op_type;
                                u32 tail;
 
                                tail = qp->sq_ring.tail;
@@ -1204,6 +1204,8 @@ int irdma_uk_cq_poll_cmpl(struct irdma_cq_uk *cq,
                                        break;
                                }
                        } while (1);
+                       if (op_type == IRDMA_OP_TYPE_BIND_MW && info->minor_err == FLUSH_PROT_ERR)
+                               info->minor_err = FLUSH_MW_BIND_ERR;
                        qp->sq_flush_seen = true;
                        if (!IRDMA_RING_MORE_WORK(qp->sq_ring))
                                qp->sq_flush_complete = true;