[PATCH] nvme-tcp: fix 'quiesce' imbalance for the admin queue

[PATCH] nvme-tcp: fix 'quiesce' imbalance for the admin queue

[Hannes Reinecke]

From: Hannes Reinecke <hare@suse.de>

nvme_tcp_teardown_ctrl() calls 'nvme_quiesce_admin_queue()' before
calling nvme_tcp_teardown_admin_queue() which also calls that function.
So move nvme_quiesce_admin_queue() out and require the callers to
do it.

Signed-off-by: Hannes Reinecke <hare@kernel.org>
---
 drivers/nvme/host/tcp.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c
index 9289bba54c82..8aabccb782f3 100644
--- a/drivers/nvme/host/tcp.c
+++ b/drivers/nvme/host/tcp.c
@@ -2318,7 +2318,6 @@ static int nvme_tcp_configure_admin_queue(struct nvme_ctrl *ctrl, bool new)
 static void nvme_tcp_teardown_admin_queue(struct nvme_ctrl *ctrl,
 		bool remove)
 {
-	nvme_quiesce_admin_queue(ctrl);
 	blk_sync_queue(ctrl->admin_q);
 	nvme_tcp_stop_queue(ctrl, 0);
 	nvme_cancel_admin_tagset(ctrl);
@@ -2407,6 +2406,7 @@ static int nvme_tcp_setup_ctrl(struct nvme_ctrl *ctrl, bool new)
 		/* See comments for nvme_tcp_key_revoke_needed() */
 		dev_dbg(ctrl->device, "restart admin queue for secure concatenation\n");
 		nvme_stop_keep_alive(ctrl);
+		nvme_quiesce_admin_queue(ctrl);
 		nvme_tcp_teardown_admin_queue(ctrl, false);
 		ret = nvme_tcp_configure_admin_queue(ctrl, false);
 		if (ret)
@@ -2473,6 +2473,7 @@ static int nvme_tcp_setup_ctrl(struct nvme_ctrl *ctrl, bool new)
 	}
 destroy_admin:
 	nvme_stop_keep_alive(ctrl);
+	nvme_quiesce_admin_queue(ctrl);
 	nvme_tcp_teardown_admin_queue(ctrl, new);
 	return ret;
 }
@@ -2516,6 +2517,7 @@ static void nvme_tcp_error_recovery_work(struct work_struct *work)
 	nvme_tcp_teardown_io_queues(ctrl, false);
 	/* unquiesce to fail fast pending requests */
 	nvme_unquiesce_io_queues(ctrl);
+	nvme_quiesce_admin_queue(ctrl);
 	nvme_tcp_teardown_admin_queue(ctrl, false);
 	nvme_unquiesce_admin_queue(ctrl);
 	nvme_auth_stop(ctrl);
-- 
2.35.3

Re: [PATCH] nvme-tcp: fix 'quiesce' imbalance for the admin queue

[Shinichiro Kawasaki]

On May 19, 2025 / 08:49, Hannes Reinecke wrote:
> From: Hannes Reinecke <hare@suse.de>
> 
> nvme_tcp_teardown_ctrl() calls 'nvme_quiesce_admin_queue()' before
> calling nvme_tcp_teardown_admin_queue() which also calls that function.
> So move nvme_quiesce_admin_queue() out and require the callers to
> do it.
> 
> Signed-off-by: Hannes Reinecke <hare@kernel.org>

Thanks for the action. I applied this patch on top of v6.15-rc7 kernel and ran
the test case nvme/063. Unfortunately, I still observe the failures that I
reported [1], both the WARN in blk_mq_unquiesce_queue and the KASAN sauf in
blk_mq_queue_tag_busy_iter...

[1] https://lore.kernel.org/linux-nvme/6mhxskdlbo6fk6hotsffvwriauurqky33dfb3s44mqtr5dsxmf@gywwmnyh3twm/