IRQ deadlocks

[Doug Ledford]

Matthew Jacob wrote:
> 
> You should always note that this is with Linux- not FreeBSD- and
> the answer is, yes, on occasion I certainly do.
> 
> And you shouldn't be putting Linux SMP into a production environment
> unless you know what you're doing.
> 
> On Sun, 20 Sep 1998, System Administrator wrote:
> 
> > anybody get IRQ deadlocks on their motherboard? If so, have you found them
> > related to the SCSI drivers? My servers doesn't tell me crap about them?
> > Just IRQ DEADLOCK DETECTED BY CPU 0 or CPU 12 (I run SMP) and the syslog
> > doesn't say a dang thing.
> >
> > running caldera linux, kernel 2.0.35 with pre7 drivers on an intel
> > providence motherboard, adaptec 7880 chipset.
> >
> > sorry if this may be a bit off topic, it's just tickin me off since this
> > server's in a production environment, i'm trying to get all hardware
> > views.

Disable SMP and live a happy life.  That, or downgrade the kernel to version
2.0.33 and patch in the latest driver.  The thing is, I have information
from a kernel deadlock where in I gathered the EIP information using the
print EIP patch, and the deadlock isn't related to the aic7xxx driver at all
(however, a faster disk subsystem will trigger the problem more often).  It
appears to be a generic vm bug in the 34 and 35 kernels under SMP.  With SMP
disabled, all is fine.  With 2.0.33, all was fine.  FWIW, the bug probably
existed in 2.0.33 as well, but for 2.0.34 thye made a change to the vm
subsystem that allowed swapping of shared COW pages (which 2.0.33 didn't do)
and this appears to have made it much easier to trigger the actual bug,
which is somewhere in the fork() code.  That's my take on it anyway.

-- 

 Doug Ledford  <dledford at dialnet.net>
  Opinions expressed are my own, but
     they should be everybody's.

To Unsubscribe: send mail to majordomo at FreeBSD.org
with "unsubscribe freebsd-aic7xxx" in the body of the message