IB/hfi1: Close PSM sdma_progress sleep window
Subsystems (3)
PATCH 1/1 IB/hfi1: Close PSM sdma_progress sleep window
commit da9de5f8527f4b9efc82f967d29a583318c034c7 upstream.
The call to sdma_progress() is called outside the wait lock.
In this case, there is a race condition where sdma_progress() can return false and the sdma_engine can idle. If that happens, there will be no more sdma interrupts to cause the wakeup and the user_sdma xmit will hang.
Fix by moving the lock to enclose the sdma_progress() call.
Also, delete busycount. The need for this was removed by:
commit bcad29137a97 ("IB/hfi1: Serve the most starved iowait entry first")
Ported to linux-4.9.y.
--- drivers/infiniband/hw/hfi1/user_sdma.c | 13 ++++--------- 1 file changed, 4 insertions(+), 9 deletions(-)
drivers/infiniband/hw/hfi1/user_sdma.c +4 -9
@@ -260,7 +260,6 @@ struct user_sdma_txreq {struct list_head list;struct user_sdma_request *req;u16 flags; -unsigned busycount;u64 seqnum; };
@@ -323,25 +322,22 @@ static int defer_packet_queue(struct hfi1_user_sdma_pkt_q *pq = container_of(wait, struct hfi1_user_sdma_pkt_q, busy);struct hfi1_ibdev *dev = &pq->dd->verbs_dev; -struct user_sdma_txreq *tx = - container_of(txreq, struct user_sdma_txreq, txreq); -if (sdma_progress (sde, seq, txreq)) { -if (tx->busycount++ <MAX_DEFER_RETRY_COUNT ) - goto eagain; - } +write_seqlock (&dev->iowait_lock); +if (sdma_progress (sde, seq, txreq)) + goto eagain;/* * We are assuming that if the list is enqueued somewhere, it * is to the dmawait list since that is the only place where * it is supposed to be enqueued. */ xchg (&pq->state,SDMA_PKT_Q_DEFERRED ); -write_seqlock (&dev->iowait_lock);if (list_empty (&pq->busy.list))list_add_tail (&pq->busy.list, &sde->dmawait);write_sequnlock (&dev->iowait_lock);return -EBUSY ; eagain: +write_sequnlock (&dev->iowait_lock);return -EAGAIN ; }
@@ -925,7 +921,6 @@ static int user_sdma_send_pkts(struct user_sdma_request *req, unsigned maxpkts) tx->flags =0 ; tx->req = req; - tx->busycount =0 ;INIT_LIST_HEAD (&tx->list);if (req->seqnum == req->info.npkts -1 )
Quoted 8 lines · Mike Marciniszyn
commit
da9de5f8527f4b9efc82f967d29a583318c034c7upstream.The call to
sdma_progress()is called outside the wait lock.In this case, there is a race condition where
sdma_progress()can return false and thesdma_enginecan idle. If that happens, there will be no more sdma interrupts to cause the wakeup and theuser_sdmaxmit will hang.
Quoted 6 lines
Fix by moving the lock to enclose the
sdma_progress()call.Also, delete busycount. The need for this was removed by:
commitbcad29137a97("IB/hfi1: Serve the most starved iowait entry first")Ported to linux-4.9.y.
Now applied, thanks.
Note, this already is in 4.14.132 and 4.19.57 so I didn't need the backports for those kernels.
greg k-h