[PATCH] nvme-pci: silence a lockdep complaint

[Shinichiro Kawasaki]

On May 22, 2024 / 12:15, Sagi Grimberg wrote:
> lockdep complains about the timeout handler running concurrently with
> the reset work which is syncing the IO request queues (which in turn
> flushes the timeout work).
> 
> We know it cannot be the case because the ctrl state machine prevents
> the timeout handler from disabling the ctrl when the reset work is
> running (changing ctrl state to RESETTING will fail, and the state is not
> terminal). If this assumption happens to break in the future, we won't
> have lockdep to assist, but for the time being we are simply seeing
> false-positive complaints from it...
> 
> Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki at wdc.com>
> Suggested-by: Keith Busch <kbusch at kernel.org>
> Signed-off-by: Sagi Grimberg <sagi at grimberg.me>

Sagi, thank you very much. I confirmed this patch avoids the lockdep WARN
observed with the blktests test case nvme/050 [1]. When I repeated the test case
around 10 times on v6.9 kernel without this patch, the lockdep WARN was
recreated in stable manner. After applying this patch, the WARN was not
observed even when the test case is repeated 100 times.

Tested-by: Shin'ichiro Kawasaki <shinichiro.kawasaki at wdc.com>

[1] https://lore.kernel.org/linux-block/m6a437jvfwzq2jfytvvk62zpgu7e4bjvegr7x73pihhkp5me5c@sh6vs3s7w754/