mps0-troubles

[Kenneth D. Merry]

On Thu, Jan 13, 2011 at 01:14:50 +0100, Joachim Tingvold wrote:
> On Wed, Jan 12, 2011, at 23:29:53PM GMT+01:00, Joachim Tingvold wrote:
> >If I were copying from the AHCI-attached disk to the mps controller,  
> >and the AHCI-attached disk timeouts, wouldn't this cause the disks  
> >on the mps controller to timeout as well?
> 
> Now it happened again (while copying from 'zroot' to 'storage'). This  
> time only mps0 produced errors; 
> <http://home.komsys.org/~jocke/dmesg_mps0_freebsd-scsi_2.txt >. As the 
> timeout seem to be over quickly, I find it strange that  whatever process 
> that accessed the disks (in my case, 'mv'), doesn't  continue once the 
> disks are available -- or is this some kind of  safeguard against corrupted 
> data?

Did the system recover this time?

The 'out of chain frames' messages are somewhat worrysome.  From looking at
the logic in the driver (mpssas_action_scsiio() and mps_data_cb()), it
looks like if it runs out of chain frames, it won't cancel the timeout on
the command.  So you'll wind up getting timeouts.  But sending an abort for
a command that hasn't gone down to the chip is rather pointless.

Did you see any other messages before the 'out of chain frames' messages
popped up?

Try editing sys/dev/mps/mpsvar.h, and change MPS_CHAIN_FRAMES from 1024 to
2048 and see if that helps things any.

That won't fix the underlying problem, but it may help you avoid running
out of that resource.

Ken
-- 
Kenneth Merry
ken at FreeBSD.ORG