On Fri, 9 Sep 2005, evilninja wrote:
> Jan Derfinak schrieb:
> > Helo.
> >
> > I switched from older SGI CVS kernel (2.6.12) to current SGI CVS kernel on
> > two systems (i386 and x86_64). But after some time ( <12 hours) on both
> > systems appeared "Corruption of in-memory data detected." error. I attached
> > /var/log/messages log from i386 system. In both cases I was unable to
> > umount FS
> > nor did clear shutdown. Both systems are workstation with very little load.
> >
> > Sep 7 00:08:26 alien kernel: Filesystem "hda10": xfs_iflush: Bad inode
> > 34351857 magic number 0x2f6c, ptr 0xcc580100
> > Sep 7 00:08:26 alien kernel: xfs_force_shutdown(hda10,0x8) called from
> > line 3294 of file fs/xfs/xfs_inode.c. Return address = 0xc02350fc
> > Sep 7 00:08:26 alien kernel: Filesystem "hda10": Corruption of in-memory
> > data detected. Shutting down filesystem: hda10
>
> ...and when you switch back to 2.6.12 it does not happen again? did you
2.6.12 (SGI-XFS CVS-2005-06-14_05:00_UTC with ACLs, no debug enabled) is
fine. I returned back to this kernel because I had another incident on my
/home partition. 2.6.13 from SGI CVS seems to be dangerous for me.
> run memtest86+ overnight? i don't know about the xfs cvs repos of the
Do you realy think that memory break on two different machines just after
switch to 2.6.13 kernel? But ok I will do it this night.
> kernel, but perhaps you can narrow it down a little bit (if xfs' cvs tree
> is in heavy flux, then this could be difficult, especially when the error
> occurs after 12h...)
The kernel is SGI-XFS CVS-2005-09-07_05:00_UTC. Partition /dev/hda10 is
mounted with ihashsize=64433,noikeep,logbufs=8. /home with
noikeep,logbufs=8. I attached kernel config but it is similiar to
2.6.12 working config.
One more notice: I tried to use CONFIG_KEXEC on x86_64 and kernel could not
recognize XFS superblock. I didn't try it on i386.
jan
--
config.bz2
Description: BZip2 compressed data
|