On Tue, Mar 12, 2013 at 02:34:07AM -0400, CAI Qian wrote:
> ----- Original Message -----
> > From: "Dave Chinner" <david@xxxxxxxxxxxxx>
> > To: "CAI Qian" <caiqian@xxxxxxxxxx>
> > Cc: xfs@xxxxxxxxxxx
> > Sent: Tuesday, March 12, 2013 2:07:01 PM
> > Subject: Re: 3.9-rc2 xfs panic
> > On Tue, Mar 12, 2013 at 12:32:28AM -0400, CAI Qian wrote:
> > > Just came across when running xfstests using 3.9-rc2 kernel on a
> > > power7
> > > box with addition of this patch which fixed a known issue,
> > > http://people.redhat.com/qcai/stable/01-fix-double-fetch-hlist.patch
> > >
> > > The log shows it was happened around test case 370 with
> > > TEST_PARAM_BLKSIZE = 2048
> > That doesn't sound like xfstests. it only has 305 tests, and no
> > parameters like TEST_PARAM_BLKSIZE....
> Sorry, it is a typo, test case 270 not 370. TEST_PARAM_BLKSIZE was
> from an internal wrapper to be used to create new filessytem not from the
> original xfstests.
OK, so that means you're testing 2k filesystem block size on a 64k
page size machine? Are you running with CONFIG_XFS_DEBUG=y?
> > So, looks like memory corruption - a corrupted slab, perhaps? Can
> > you turn on memory poisoning, debugging, etc?
Does this turn anything up?