Re: lsi_scsi: error: Bad Status move errors with kvm-79

[Ryan Harper <ryanh@us.ibm.com>]

* Anssi Kolehmainen <anssi@aketzu.net> [2008-12-04 10:11]:
> On Thu, Dec 04, 2008 at 09:26:35AM -0600, Ryan Harper wrote:
> > * Anssi Kolehmainen <anssi@aketzu.net> [2008-12-04 08:51]:
> > > Hi,
> > > 
> > > I have kvm environment with linux-2.6.28-rc7 x86_64 (Xeon), kvm-79 host
> > > and bunch of Win2K3 guests. Sometimes I get 'lsi_scsi: error: Bad Status
> > > move' from kvm (qemu) and in Windows event log "The device,
> > > \Device\Scsi\sym_hi1, did not respond within the timeout period." These
> > > errors come somewhat at random, usually with 10-30 second intervals when
> > > there is enough disk usage in the guest (seems that installing Bea
> > > Weblogic or Oracle database is pretty nice for causing these errors).
> > > 
> > > Usually windows is able to recover from these but sometimes (=too often)
> > > I get random delays and hangups. Also I have gotten BSOD 0x77 (0x02,
> > > 0x00, 0x00, 0x5f4000) about once a day.
> > > 
> > > Any ideas how to debug / fix this problem?
> > 
> > Current KVM userspace has a bogus line in the scsi code relating to the
> > DBC register which looks like is what is tripping up the Bad Status, or
> > could be anyhow.  Try out with this patch applied to your qemu dir:
> > 
> > http://lists.gnu.org/archive/html/qemu-devel/2008-12/msg00043.html
> 
> That line is not in kvm-79... And with that line added Windows doesn't
> even seem to start.

Yeah, I was hoping it was something simple, but you're right, 79 was
released before that bogus line made it into qemu.  It's now been
removed from qemu cvs, and kvm-userspace just sync'ed.

> 
> > You can also enable debugging in qemu/hw/lsi53c895a.c and in
> > qemu/hw/scsi-disk.c   Sending that output here would be helpful if we're
> > still tracking it.
> 
> Nice 160mb log from just starting windows and running the installer
> until BSOD. Only 1.7mb with bzip2 compression:
> http://kelvin.aketzu.net/kvm-qemu.log.bz2

I scanned through that, but didn't see anything that jumped out at me.
Typically, I need a good run and a failing run to tell what's going on.

> > Are there free/downloadable copies of Bae or Oracle that I can use to
> > recreate?
> 
> Yeah, they are both available for free. You can get the Bea (now Oracle)
> Weblogic (application server) from:
> http://www.oracle.com/technology/software/products/ias/htdocs/wls_main.html
> The version we used is "Oracle WebLogic Server 10.0 MP1"
> 
> (I guess any big application which uncompresses to HD / installs lots of
> files might work.)

Using kvm-userspace.git (kvm-80 is equivalent w.r.t scsi level) I
installed win2k3 sp2 with scsi as the main device, qcow2 backed file.
Downloaded, unzipped and installed the Weblogic server.  All with no
issues at all.  I'll give it a run against raw devices, that seems to be
what you were using.  Any more details on how to reliably reproduce the
issue will help me track down the bug.


-- 
Ryan Harper
Software Engineer; Linux Technology Center
IBM Corp., Austin, Tx
ryanh@us.ibm.com