Deadlock with Promise Driver, still in 2.6.9-rc1

Deadlock with Promise Driver, still in 2.6.9-rc1

[NM Lists]

This has already been reported I think, but here it is again. In some
bugzilla somewhere this has been incorrectly reported as a software Raid
vs. Promise driver incompatibility; in fact, it appears to be a problem
when both SATA channels are being used at the same time (which obviously
happens a lot with raid!)

How to trigger it:

# cat /dev/sda > /dev/null &
# hdparm -t /dev/sdb &
# hdparm -t /dev/sda &

Now any disk access to either disks is blocked, and in my case, incoming
mail (using qmail) starts piling up processes, and this ends up in lock
up.

Re: Deadlock with Promise Driver, still in 2.6.9-rc1

[Jeff Garzik]

NM Lists wrote:
> This has already been reported I think, but here it is again. In some
> bugzilla somewhere this has been incorrectly reported as a software Raid
> vs. Promise driver incompatibility; in fact, it appears to be a problem
> when both SATA channels are being used at the same time (which obviously
> happens a lot with raid!)
> 
> How to trigger it:
> 
> # cat /dev/sda > /dev/null &
> # hdparm -t /dev/sdb &
> # hdparm -t /dev/sda &


Well, I'm not familiar with this problem, at least.

Can you provide info as described in REPORTING-BUGS (in the kernel tree)?

Can you define ATA_DEBUG and ATA_VERBOSE_DEBUG in 
include/linux/libata.h, and provide output when it deadlocks?

Did you try 2.6.9-rc1 snapshot?

If nothing else, can you record the bug at http://bugzilla.kernel.org/ 
to make sure its not lost?

	Jeff