Re: XFS + LVM + Software RAID5 on Debian testing

[Charles Steinkuehler <charles@xxxxxxxxxxxxxxxx>]

Charles Steinkuehler wrote:

> Christian Guggenberger wrote:
>
> > On Wed, 2004-06-23 at 19:20, Charles Steinkuehler wrote:
> > > Chris Wedgwood wrote:
> > >
> > > > I guess open a bug if you haven't done so already, nothing obvious
> > > > springs to mind and I don't know much about LVs (I just assumed they
> > > > were simple enough and would work as expected).
> > >
> > > I'm becoming less conviced this is an XFS problem, and am thinking there 
> > > might be some issues with the 'stacking' of the new SATA drivers, 
> > > software RAID5, and LVM2 in linux.  I'm still doing testing, and if 
> > > possible am going to try a 2.6 kernel, LVM1 (instead of LVM2), and the 
> > > kernel from debian-stable (if it will talk to my NForce2 chipset and 
> > > SiImage SATA controllers).
> >
> > are u talking about the siimage or the sata_sil driver?
> > I remember jgarzik posting about data corruption with sata_sil recently
> > on lkml.
> >
> > http://marc.theaimsgroup.com/?l=linux-kernel&m=108793683025643&w=4
>
> I'm running the siimage module, not the sata_sil, but it looks like I am 
> using the 'quirky' si3112 chipset (x2 - 1 PCI plug-in-card and 1 on the 
> motherboard).
>
> I'm currently trying to figure out if the problem is with the controller 
> or not.  The odd thing, is I wasn't having any problems prior to getting 
> the second controller (PCI card) and 2 more drives (after running out of 
> space on my ext3-on-LVM-on-RAID1 setup with 2x 160G drives).

OK, looks like a red herring (at least as far as XFS is concerned).

Further testing indicates my on-board Silicon Image 3112 controller 
works fine, but the new plug-in card with the same chip (different rev) 
causes data corruption.

Chip details for google, should anyone find it useful:
On-board controller (works):
  Silicon Image
  SATALink (tm)
  Sil3112ACT144
  Q221668.1A
  0242
  1.1

Plug-in card controller (fails):
  Silicon Image
  SATALink (tm)
  Sil3112ACT144
  Q30350.1
  0342
  1.21

If I tie two drives (Seagate ST31600023AS) to the on-board controller 
(plug-in-card removed), I have a stable system.

If I tie the two drives to the plug-in-card (on-board controller 
disabled via jumper), I get data corruption.

I suspect something is wrong with the newer version of the chip, or 
given the LKML traffic about this part, perhpas the driver 'fixes' for 
the old chip version don't work well with the new (possibly fixed?) version.

Anyway, I have a new controller (promise TX4) on the way, so I'll see if 
that fixes my corruption problems.

--
Charles Steinkuehler
charles@xxxxxxxxxxxxxxxx