summaryrefslogtreecommitdiffstats
path: root/drivers/infiniband
diff options
context:
space:
mode:
authorEli Cohen <eli at mellanox.co.il>2008-01-24 06:38:06 -0800
committerRoland Dreier <rolandd@cisco.com>2008-02-04 20:20:44 -0800
commit1d368c546566e249da8181e933c53788093965cf (patch)
tree08ea2e0280f125f2981ab7ada0d335b5928cc670 /drivers/infiniband
parent1203c42e7be1aa0be641b701f42b6d38c2d94b39 (diff)
downloadlinux-stable-1d368c546566e249da8181e933c53788093965cf.tar.gz
linux-stable-1d368c546566e249da8181e933c53788093965cf.tar.bz2
linux-stable-1d368c546566e249da8181e933c53788093965cf.zip
IB/ib_mthca: Pre-link receive WQEs in Tavor mode
We have recently discovered that Tavor mode requires each WQE in a posted list of receive WQEs to have a valid NDA field at all times. This requirement holds true for regular QPs as well as for SRQs. This patch prelinks the receive queue in a regular QP and keeps the free list in SRQ always properly linked. Signed-off-by: Eli Cohen <eli@mellanox.co.il> Reviewed-by: Jack Morgenstein <jackm@mellanox.co.il> Signed-off-by: Roland Dreier <rolandd@cisco.com>
Diffstat (limited to 'drivers/infiniband')
-rw-r--r--drivers/infiniband/hw/mthca/mthca_qp.c13
-rw-r--r--drivers/infiniband/hw/mthca/mthca_srq.c23
2 files changed, 22 insertions, 14 deletions
diff --git a/drivers/infiniband/hw/mthca/mthca_qp.c b/drivers/infiniband/hw/mthca/mthca_qp.c
index 0e5461c65731..db5595bbf7f0 100644
--- a/drivers/infiniband/hw/mthca/mthca_qp.c
+++ b/drivers/infiniband/hw/mthca/mthca_qp.c
@@ -1175,6 +1175,7 @@ static int mthca_alloc_qp_common(struct mthca_dev *dev,
{
int ret;
int i;
+ struct mthca_next_seg *next;
qp->refcount = 1;
init_waitqueue_head(&qp->wait);
@@ -1217,7 +1218,6 @@ static int mthca_alloc_qp_common(struct mthca_dev *dev,
}
if (mthca_is_memfree(dev)) {
- struct mthca_next_seg *next;
struct mthca_data_seg *scatter;
int size = (sizeof (struct mthca_next_seg) +
qp->rq.max_gs * sizeof (struct mthca_data_seg)) / 16;
@@ -1240,6 +1240,13 @@ static int mthca_alloc_qp_common(struct mthca_dev *dev,
qp->sq.wqe_shift) +
qp->send_wqe_offset);
}
+ } else {
+ for (i = 0; i < qp->rq.max; ++i) {
+ next = get_recv_wqe(qp, i);
+ next->nda_op = htonl((((i + 1) % qp->rq.max) <<
+ qp->rq.wqe_shift) | 1);
+ }
+
}
qp->sq.last = get_send_wqe(qp, qp->sq.max - 1);
@@ -1863,7 +1870,6 @@ int mthca_tavor_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
prev_wqe = qp->rq.last;
qp->rq.last = wqe;
- ((struct mthca_next_seg *) wqe)->nda_op = 0;
((struct mthca_next_seg *) wqe)->ee_nds =
cpu_to_be32(MTHCA_NEXT_DBD);
((struct mthca_next_seg *) wqe)->flags = 0;
@@ -1885,9 +1891,6 @@ int mthca_tavor_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
qp->wrid[ind] = wr->wr_id;
- ((struct mthca_next_seg *) prev_wqe)->nda_op =
- cpu_to_be32((ind << qp->rq.wqe_shift) | 1);
- wmb();
((struct mthca_next_seg *) prev_wqe)->ee_nds =
cpu_to_be32(MTHCA_NEXT_DBD | size);
diff --git a/drivers/infiniband/hw/mthca/mthca_srq.c b/drivers/infiniband/hw/mthca/mthca_srq.c
index ec63adc1099c..a5ffff6e1026 100644
--- a/drivers/infiniband/hw/mthca/mthca_srq.c
+++ b/drivers/infiniband/hw/mthca/mthca_srq.c
@@ -175,9 +175,17 @@ static int mthca_alloc_srq_buf(struct mthca_dev *dev, struct mthca_pd *pd,
* scatter list L_Keys to the sentry value of 0x100.
*/
for (i = 0; i < srq->max; ++i) {
- wqe = get_wqe(srq, i);
+ struct mthca_next_seg *next;
- *wqe_to_link(wqe) = i < srq->max - 1 ? i + 1 : -1;
+ next = wqe = get_wqe(srq, i);
+
+ if (i < srq->max - 1) {
+ *wqe_to_link(wqe) = i + 1;
+ next->nda_op = htonl(((i + 1) << srq->wqe_shift) | 1);
+ } else {
+ *wqe_to_link(wqe) = -1;
+ next->nda_op = 0;
+ }
for (scatter = wqe + sizeof (struct mthca_next_seg);
(void *) scatter < wqe + (1 << srq->wqe_shift);
@@ -470,12 +478,15 @@ out:
void mthca_free_srq_wqe(struct mthca_srq *srq, u32 wqe_addr)
{
int ind;
+ struct mthca_next_seg *last_free;
ind = wqe_addr >> srq->wqe_shift;
spin_lock(&srq->lock);
- *wqe_to_link(get_wqe(srq, srq->last_free)) = ind;
+ last_free = get_wqe(srq, srq->last_free);
+ *wqe_to_link(last_free) = ind;
+ last_free->nda_op = htonl((ind << srq->wqe_shift) | 1);
*wqe_to_link(get_wqe(srq, ind)) = -1;
srq->last_free = ind;
@@ -516,7 +527,6 @@ int mthca_tavor_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
prev_wqe = srq->last;
srq->last = wqe;
- ((struct mthca_next_seg *) wqe)->nda_op = 0;
((struct mthca_next_seg *) wqe)->ee_nds = 0;
/* flags field will always remain 0 */
@@ -537,9 +547,6 @@ int mthca_tavor_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
if (i < srq->max_gs)
mthca_set_data_seg_inval(wqe);
- ((struct mthca_next_seg *) prev_wqe)->nda_op =
- cpu_to_be32((ind << srq->wqe_shift) | 1);
- wmb();
((struct mthca_next_seg *) prev_wqe)->ee_nds =
cpu_to_be32(MTHCA_NEXT_DBD);
@@ -613,8 +620,6 @@ int mthca_arbel_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
break;
}
- ((struct mthca_next_seg *) wqe)->nda_op =
- cpu_to_be32((next_ind << srq->wqe_shift) | 1);
((struct mthca_next_seg *) wqe)->ee_nds = 0;
/* flags field will always remain 0 */