regression on generic/351 in 6.11-rc5?

regression on generic/351 in 6.11-rc5?

[Darrick J. Wong]

Hi everyone,

Has anyone else noticed the following regression in generic/351 between
6.11-rc4 and -rc5?

--- /tmp/fstests/tests/generic/351.out	2024-02-28 16:20:24.224889046 -0800
+++ /var/tmp/fstests/generic/351.out.bad	2024-08-26 00:03:35.701439178 -0700
@@ -25,7 +25,7 @@ b83f9394092e15bdcda585cd8e776dc6  SCSI_D
 Destroy device
 Create w/o unmap or writesame and format
 Zero punch, no fallback available
-fallocate: Operation not supported
+fallocate: Remote I/O error
 Zero range, write fallback
 Check contents
 0fc6bc93cd0cd97e3cde5ea39ea1185d  SCSI_DEBUG_DEV

Just speculating here, but seeing as that test messes with lbpme in
scsi-debug, that this might be a result of this patch:

https://lore.kernel.org/all/20240817005325.3319384-1-martin.petersen@oracle.com/

Will bisect in the morning...

--D

Re: regression on generic/351 in 6.11-rc5?

[Christoph Hellwig]

On Mon, Aug 26, 2024 at 07:07:14PM -0700, Darrick J. Wong wrote:
> Has anyone else noticed the following regression in generic/351 between
> 6.11-rc4 and -rc5?

Yes, I'm seeing this with a fresh -rc5 build.

Re: regression on generic/351 in 6.11-rc5?

[Darrick J. Wong]

On Mon, Aug 26, 2024 at 09:50:55PM -0700, Christoph Hellwig wrote:
> On Mon, Aug 26, 2024 at 07:07:14PM -0700, Darrick J. Wong wrote:
> > Has anyone else noticed the following regression in generic/351 between
> > 6.11-rc4 and -rc5?
> 
> Yes, I'm seeing this with a fresh -rc5 build.

Aha, it wasn't the scsi lbpme patch like I thought; it was a different
patch in the block layer that got overly aggressive about not querying
write same queue limits.  Patch incoming.

--D

Re: regression on generic/351 in 6.11-rc5?

[Zorro Lang]

On Mon, Aug 26, 2024 at 07:07:14PM -0700, Darrick J. Wong wrote:
> Hi everyone,
> 
> Has anyone else noticed the following regression in generic/351 between
> 6.11-rc4 and -rc5?

Hi Darrick,

The g/351 isn't in auto (default) group. So I think most of fstests users
don't run it regularly. So I think most of them didn't hit that.
How about add this case and more "fast and stable" cases to auto group?
Include some xfs_scrub and online repair test cases.

Thanks,
Zorro

> 
> --- /tmp/fstests/tests/generic/351.out	2024-02-28 16:20:24.224889046 -0800
> +++ /var/tmp/fstests/generic/351.out.bad	2024-08-26 00:03:35.701439178 -0700
> @@ -25,7 +25,7 @@ b83f9394092e15bdcda585cd8e776dc6  SCSI_D
>  Destroy device
>  Create w/o unmap or writesame and format
>  Zero punch, no fallback available
> -fallocate: Operation not supported
> +fallocate: Remote I/O error
>  Zero range, write fallback
>  Check contents
>  0fc6bc93cd0cd97e3cde5ea39ea1185d  SCSI_DEBUG_DEV
> 
> Just speculating here, but seeing as that test messes with lbpme in
> scsi-debug, that this might be a result of this patch:
> 
> https://lore.kernel.org/all/20240817005325.3319384-1-martin.petersen@oracle.com/
> 
> Will bisect in the morning...
> 
> --D
>