On Tue, Jan 9, 2018 at 9:23 PM, Bart Van Assche bart.vanassche@wdc.com wrote:
The following sequence:
- Change queue pair state into IB_QPS_ERR.
- Post a work request on the queue pair.
Triggers the following race condition in the rdma_rxe driver:
- rxe_qp_error() triggers an asynchronous call of rxe_completer(), the function that examines the QP send queue.
- rxe_post_send() posts a work request on the QP send queue.
Avoid that this race causes a work request to be ignored by scheduling an rxe_completer() call from rxe_post_send() for queues that are in the error state.
Signed-off-by: Bart Van Assche bart.vanassche@wdc.com Cc: Moni Shoua monis@mellanox.com Cc: stable@vger.kernel.org # v4.8
drivers/infiniband/sw/rxe/rxe_verbs.c | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/drivers/infiniband/sw/rxe/rxe_verbs.c b/drivers/infiniband/sw/rxe/rxe_verbs.c index a6fbed48db8a..8f631d64c192 100644 --- a/drivers/infiniband/sw/rxe/rxe_verbs.c +++ b/drivers/infiniband/sw/rxe/rxe_verbs.c @@ -814,6 +814,8 @@ static int rxe_post_send_kernel(struct rxe_qp *qp, struct ib_send_wr *wr, (queue_count(qp->sq.queue) > 1);
rxe_run_task(&qp->req.task, must_sched);
if (unlikely(qp->req.state == QP_STATE_ERROR))rxe_run_task(&qp->comp.task, 1); return err;}
2.15.1
Maybe I am missing something but I think that the race is when qp is in ERROR state and the following functions run in parallel * rxe_drain_req_pkts (called from rxe_requester after post_send) * rxe_drain_resp_pkts (called from rxe_completer after modify to ERROR)
Am I right?