Re: [BUG] log I/O completion GPF via xfs/006 and xfs/264 on 5.17.0-rc8

[Dave Chinner <david@fromorbit.com>]

On Fri, Mar 18, 2022 at 02:51:33PM -0700, Darrick J. Wong wrote:
> On Sat, Mar 19, 2022 at 08:48:31AM +1100, Dave Chinner wrote:
> > On Fri, Mar 18, 2022 at 09:46:53AM -0400, Brian Foster wrote:
> > > Hi,
> > > 
> > > I'm not sure if this is known and/or fixed already, but it didn't look
> > > familiar so here is a report. I hit a splat when testing Willy's
> > > prospective folio bookmark change and it turns out it replicates on
> > > Linus' current master (551acdc3c3d2). This initially reproduced on
> > > xfs/264 (mkfs defaults) and I saw a soft lockup warning variant via
> > > xfs/006, but when I attempted to reproduce the latter a second time I
> > > hit what looks like the same problem as xfs/264. Both tests seem to
> > > involve some form of error injection, so possibly the same underlying
> > > problem. The GPF splat from xfs/264 is below.
> > 
> > On a side note, I'm wondering if we should add xfs/006 and xfs/264
> > to the recoveryloop group - they do a shutdown under load and a
> > followup mount to ensure the filesystem gets recovered before
> > the test ends and the fs is checked, so while thy don't explicitly
> > test recovery, they do exercise it....
> > 
> > Thoughts?
> 
> Someone else asked about this the other day, and I proposed a 'recovery'
> group for tests that don't run in a loop.

That distinction is largely meaningless to me.

I tend to think of "recoveryloop" as the recovery tests I want to
run in a long running loop via iteration. e.g. isomething like 
'check -I 250 -g recoveryloop'. I don't really care if the tests
loop internally doing multiple recoveries - I'm wanting to run the
recovery tests that reproduce problems frequeently in a tight loop
repeatedly.

Hence I think we should just lump the shutdown+recovery tests all in
one group so that when we want to exercise shutdown/recovery we just
have one single group to run repeatedly in a loop. Whether that
group is named 'recovery' or 'recoveryloop' is largely irrelevant to
me.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com