[PATCH v2 1/1] nvme-pci: Hold cq_poll_lock while completing CQEs

[PATCH v2 1/1] nvme-pci: Hold cq_poll_lock while completing CQEs

[Bijan Mottahedeh]

Avoid scheduling a polling thread if it has cqe's to complete, otherwise
another thread may free enough tags out of order, allowing a queue wrap.

This is an interim solution. Longer term, the two-pass completion should
be fixed.  It was done that way when submit and complete processing were
using the same spinlock, so reducing the critical section was helpful,
but that's not needed anymore.

Fixes: dabcefab45d3 ("nvme: provide optimized poll function for separate poll queues")
Signed-off-by: Bijan Mottahedeh <bijan.mottahedeh@oracle.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
---
 drivers/nvme/host/pci.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index ace4dd9..d3f23d6 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -1078,9 +1078,9 @@ static int nvme_poll(struct blk_mq_hw_ctx *hctx)
 
 	spin_lock(&nvmeq->cq_poll_lock);
 	found = nvme_process_cq(nvmeq, &start, &end, -1);
+	nvme_complete_cqes(nvmeq, start, end);
 	spin_unlock(&nvmeq->cq_poll_lock);
 
-	nvme_complete_cqes(nvmeq, start, end);
 	return found;
 }
 
-- 
1.8.3.1


_______________________________________________
linux-nvme mailing list
linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

Re: [PATCH v2 1/1] nvme-pci: Hold cq_poll_lock while completing CQEs

[Keith Busch]

On Wed, Feb 26, 2020 at 06:53:43PM -0800, Bijan Mottahedeh wrote:
> Avoid scheduling a polling thread if it has cqe's to complete, otherwise
> another thread may free enough tags out of order, allowing a queue wrap.
> 
> This is an interim solution. Longer term, the two-pass completion should
> be fixed.  It was done that way when submit and complete processing were
> using the same spinlock, so reducing the critical section was helpful,
> but that's not needed anymore.
> 
> Fixes: dabcefab45d3 ("nvme: provide optimized poll function for separate poll queues")
> Signed-off-by: Bijan Mottahedeh <bijan.mottahedeh@oracle.com>
> Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
> Reviewed-by: Jens Axboe <axboe@kernel.dk>

Thanks, applied to a new nvme-5.6-rc4 branch.

_______________________________________________
linux-nvme mailing list
linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

Re: [PATCH v2 1/1] nvme-pci: Hold cq_poll_lock while completing CQEs

[Christoph Hellwig]

nvme_poll_irqdisable will need the same fix.

Also I think we can merge nvme_process_cq and nvme_complete_cqes now
as the split is becoming rather pointless.

_______________________________________________
linux-nvme mailing list
linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

Re: [PATCH v2 1/1] nvme-pci: Hold cq_poll_lock while completing CQEs

[Bijan Mottahedeh]

On 2/29/2020 10:57 AM, Christoph Hellwig wrote:
> nvme_poll_irqdisable will need the same fix.
>
> Also I think we can merge nvme_process_cq and nvme_complete_cqes now
> as the split is becoming rather pointless.
Ok, I'll work on a patch for both.

--bijan

_______________________________________________
linux-nvme mailing list
linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme