Re: [PATCH] Revert "scsi: scsi_debug: Address races following module load"

[Bob Pearson <rpearsonhpe@gmail.com>]

On 4/10/22 08:15, Yi Zhang wrote:
> Confirmed the blktests srp/ issue was fixed with this revert:
> 
> Tested-by: Yi Zhang <yi.zhang@redhat.com>
> 
> On Sat, Apr 9, 2022 at 12:37 PM Bart Van Assche <bvanassche@acm.org> wrote:
>>
>> Revert the patch mentioned in the subject since it blocks I/O after
>> module unload has started while this is a legitimate use case. For e.g.
>> blktests test case srp/001 that patch causes a command timeout to be
>> triggered for the following call stack:
>>

I also applied Bart's patch reverting the scsi_debug change. And find
that now blktest/check -q srp runs successfully and the trace does not
show the inconsistent lock state warning. So it looks to me that there
is not a current need to revert the remaining _bh locks in the rxe driver
to _irqsave locks.

But, as far as I know the root cause of those warnings, when valid, are due to
code that calls into the verbs APIs while holding _irqsave locks. If we want to
make that generally possible then perhaps we should just get rid of the _bh
locks in favor of _irqsave locks. I have a patch that does that if you think
it is needed.

Bob