[PATCH] nvme-pci: fix missed admin queue sq doorbell write

[PATCH] nvme-pci: fix missed admin queue sq doorbell write

[Keith Busch]

From: Keith Busch <kbusch@kernel.org>

We can batch admin commands, which means bd->last may be false, skipping
the doorbell write so that we aggregate multiple commands per write. If
a subsequent command can't be dispatched for whatever reason, we have to
provide the blk-mq ops' commit_rqs callback in order to ensure we
properly update the doorbell.

Fixes: 58e5bdeb9c2b ("nvme: enable uring-passthrough for admin commands")
Signed-off-by: Keith Busch <kbusch@kernel.org>
---
 drivers/nvme/host/pci.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index db5fc9bf66272..4c052ed18cb8d 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -2241,6 +2241,7 @@ static int nvme_create_queue(struct nvme_queue *nvmeq, int qid, bool polled)
 static const struct blk_mq_ops nvme_mq_admin_ops = {
 	.queue_rq	= nvme_queue_rq,
 	.complete	= nvme_pci_complete_rq,
+	.commit_rqs	= nvme_commit_rqs,
 	.init_hctx	= nvme_admin_init_hctx,
 	.init_request	= nvme_pci_init_request,
 	.timeout	= nvme_timeout,
-- 
2.52.0

Re: [PATCH] nvme-pci: fix missed admin queue sq doorbell write

[Christoph Hellwig]

On Tue, Apr 21, 2026 at 09:18:55AM -0700, Keith Busch wrote:
> From: Keith Busch <kbusch@kernel.org>
> 
> We can batch admin commands, which means bd->last may be false,

How do we end up batching them?  io_uring passthrough I guess?  Can
you expand the commit log a bit with that?

Otherwise looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH] nvme-pci: fix missed admin queue sq doorbell write

[Kanchan Joshi]

Thanks.

Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>

Re: [PATCH] nvme-pci: fix missed admin queue sq doorbell write

[Keith Busch]

On Wed, Apr 22, 2026 at 08:01:55AM +0200, Christoph Hellwig wrote:
> On Tue, Apr 21, 2026 at 09:18:55AM -0700, Keith Busch wrote:
> > From: Keith Busch <kbusch@kernel.org>
> > 
> > We can batch admin commands, which means bd->last may be false,
> 
> How do we end up batching them?  io_uring passthrough I guess?  Can
> you expand the commit log a bit with that?

Exactly, io_uring_cmd passthrough starts a plug. The "Fixes" tag is a
bit subtle, so I'll more explicitly state this in the commit message.