ivampiresp/Spdk
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

rdma: Correct WR type checking

Browse Source 
The previous patch ce6b8a1313
added a wrong assumption that every WC of RDMA_WR_TYPE_DATA
type must point to rdma_req with IBV_WC_RDMA_READ opcode since
RDMA_WRITE operations are non-signaled. However it is wrong
since in the case of error all WRs will have WCs. Revert part
of the problematic patch.

Change-Id: I8d270c5313ebfe1ec44a338820a62f085996eb8f
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1334
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
This commit is contained in:
Alexey Marchuk 2020-03-17 16:42:41 +03:00 committed by Jim Harris
parent 3509fc12df
commit 5e2101ceb2
1 changed files with 9 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
lib/nvmf/rdma.c
View File

@ -3915,13 +3915,12 @@ spdk_nvmf_rdma_poller_poll(struct spdk_nvmf_rdma_transport *rtransport,
rqpair = SPDK_CONTAINEROF(rdma_req->req.qpair, struct spdk_nvmf_rdma_qpair, qpair); rqpair = SPDK_CONTAINEROF(rdma_req->req.qpair, struct spdk_nvmf_rdma_qpair, qpair);
assert(rdma_req->num_outstanding_data_wr > 0); assert(rdma_req->num_outstanding_data_wr > 0);
assert(rdma_req->data.wr.opcode == IBV_WR_RDMA_READ);
rqpair->current_send_depth--; rqpair->current_send_depth--;
rqpair->current_read_depth--;
rdma_req->num_outstanding_data_wr--; rdma_req->num_outstanding_data_wr--;
if (!wc[i].status) { if (!wc[i].status) {
assert(wc[i].opcode == IBV_WC_RDMA_READ); assert(wc[i].opcode == IBV_WC_RDMA_READ);
rqpair->current_read_depth--;
/* wait for all outstanding reads associated with the same rdma_req to complete before proceeding. */ /* wait for all outstanding reads associated with the same rdma_req to complete before proceeding. */
if (rdma_req->num_outstanding_data_wr == 0) { if (rdma_req->num_outstanding_data_wr == 0) {
rdma_req->state = RDMA_REQUEST_STATE_READY_TO_EXECUTE; rdma_req->state = RDMA_REQUEST_STATE_READY_TO_EXECUTE;
@ -3929,9 +3928,14 @@ spdk_nvmf_rdma_poller_poll(struct spdk_nvmf_rdma_transport *rtransport,
} }
} else { } else {
/* If the data transfer fails still force the queue into the error state, /* If the data transfer fails still force the queue into the error state,
* in case of RDMA_READ, we need to force the request into a completed state */ * if we were performing an RDMA_READ, we need to force the request into a
if (rdma_req->num_outstanding_data_wr == 0) { * completed state since it wasn't linked to a send. However, in the RDMA_WRITE
rdma_req->state = RDMA_REQUEST_STATE_COMPLETED; * case, we should wait for the SEND to complete. */
if (rdma_req->data.wr.opcode == IBV_WR_RDMA_READ) {
rqpair->current_read_depth--;
if (rdma_req->num_outstanding_data_wr == 0) {
rdma_req->state = RDMA_REQUEST_STATE_COMPLETED;
}
} }
} }
break; break;