IB/qib: Insure last cursor is updated prior to complete
This patch is a prerequisite for adding a separate lock
for post send.
The timing of updating s_last needs to be before returning
any send completion to avoid a race between a poll cq seeing
a completion and the post send checking for a full queue.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
diff --git a/drivers/infiniband/hw/qib/qib_rc.c b/drivers/infiniband/hw/qib/qib_rc.c
index 044525d9..ce886b2 100644
--- a/drivers/infiniband/hw/qib/qib_rc.c
+++ b/drivers/infiniband/hw/qib/qib_rc.c
@@ -1008,10 +1008,18 @@
start_timer(qp);
while (qp->s_last != qp->s_acked) {
+ u32 s_last;
+
wqe = rvt_get_swqe_ptr(qp, qp->s_last);
if (qib_cmp24(wqe->lpsn, qp->s_sending_psn) >= 0 &&
qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) <= 0)
break;
+ s_last = qp->s_last;
+ if (++s_last >= qp->s_size)
+ s_last = 0;
+ qp->s_last = s_last;
+ /* see post_send() */
+ barrier();
for (i = 0; i < wqe->wr.num_sge; i++) {
struct rvt_sge *sge = &wqe->sg_list[i];
@@ -1028,8 +1036,6 @@
wc.qp = &qp->ibqp;
rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.send_cq), &wc, 0);
}
- if (++qp->s_last >= qp->s_size)
- qp->s_last = 0;
}
/*
* If we were waiting for sends to complete before resending,
@@ -1068,11 +1074,19 @@
*/
if (qib_cmp24(wqe->lpsn, qp->s_sending_psn) < 0 ||
qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) > 0) {
+ u32 s_last;
+
for (i = 0; i < wqe->wr.num_sge; i++) {
struct rvt_sge *sge = &wqe->sg_list[i];
rvt_put_mr(sge->mr);
}
+ s_last = qp->s_last;
+ if (++s_last >= qp->s_size)
+ s_last = 0;
+ qp->s_last = s_last;
+ /* see post_send() */
+ barrier();
/* Post a send completion queue entry if requested. */
if (!(qp->s_flags & RVT_S_SIGNAL_REQ_WR) ||
(wqe->wr.send_flags & IB_SEND_SIGNALED)) {
@@ -1084,8 +1098,6 @@
wc.qp = &qp->ibqp;
rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.send_cq), &wc, 0);
}
- if (++qp->s_last >= qp->s_size)
- qp->s_last = 0;
} else
this_cpu_inc(*ibp->rvp.rc_delayed_comp);