The patch below does not apply to the 4.9-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to stable@vger.kernel.org.
thanks,
greg k-h
------------------ original commit in Linus's tree ------------------
From cf131a81967583ae737df6383a0893b9fee75b4e Mon Sep 17 00:00:00 2001
From: Mike Marciniszyn mike.marciniszyn@intel.com Date: Fri, 14 Jun 2019 12:32:26 -0400 Subject: [PATCH] IB/hfi1: Avoid hardlockup with flushlist_lock
Heavy contention of the sde flushlist_lock can cause hard lockups at extreme scale when the flushing logic is under stress.
Mitigate by replacing the item at a time copy to the local list with an O(1) list_splice_init() and using the high priority work queue to do the flushes.
Fixes: 7724105686e7 ("IB/hfi1: add driver files") Cc: stable@vger.kernel.org Reviewed-by: Dennis Dalessandro dennis.dalessandro@intel.com Signed-off-by: Mike Marciniszyn mike.marciniszyn@intel.com Signed-off-by: Dennis Dalessandro dennis.dalessandro@intel.com Signed-off-by: Doug Ledford dledford@redhat.com
diff --git a/drivers/infiniband/hw/hfi1/sdma.c b/drivers/infiniband/hw/hfi1/sdma.c index b0110728f541..70828de7436b 100644 --- a/drivers/infiniband/hw/hfi1/sdma.c +++ b/drivers/infiniband/hw/hfi1/sdma.c @@ -410,10 +410,7 @@ static void sdma_flush(struct sdma_engine *sde) sdma_flush_descq(sde); spin_lock_irqsave(&sde->flushlist_lock, flags); /* copy flush list */ - list_for_each_entry_safe(txp, txp_next, &sde->flushlist, list) { - list_del_init(&txp->list); - list_add_tail(&txp->list, &flushlist); - } + list_splice_init(&sde->flushlist, &flushlist); spin_unlock_irqrestore(&sde->flushlist_lock, flags); /* flush from flush list */ list_for_each_entry_safe(txp, txp_next, &flushlist, list) @@ -2413,7 +2410,7 @@ int sdma_send_txreq(struct sdma_engine *sde, list_add_tail(&tx->list, &sde->flushlist); spin_unlock(&sde->flushlist_lock); iowait_inc_wait_count(wait, tx->num_desc); - schedule_work(&sde->flush_worker); + queue_work_on(sde->cpu, system_highpri_wq, &sde->flush_worker); ret = -ECOMM; goto unlock; nodesc: @@ -2511,7 +2508,7 @@ int sdma_send_txlist(struct sdma_engine *sde, struct iowait_work *wait, iowait_inc_wait_count(wait, tx->num_desc); } spin_unlock(&sde->flushlist_lock); - schedule_work(&sde->flush_worker); + queue_work_on(sde->cpu, system_highpri_wq, &sde->flush_worker); ret = -ECOMM; goto update_tail; nodesc: