"swap_pager" loop?

[Martin Schütte]

Kris Kennaway schrieb:
>> The machine has 4gb of ram, a 4gb swap partition on /dev/amrd0s1b, and
>> provides nfs/nis/samba/printing (cups) services.

I have the same problem with FreeBSD 6.1-RELEASE on my SMP-System 
(2xAthlon MP) with an asr-RAID (Adaptec 2110S).
The disks are OK (complete read with 'dd if=/dev/da...' showed no errors).

The deadlocks often occured during backups, so I suspect a race 
condition or too many interrupts to be the cause.

> You could try increasing the timeout in the msleep() in
> vm/swap_pager.c.  If your system is just so heavily loaded that the
> I/O command really is taking more than 20 seconds to complete, you may
> be getting false positives.

I set my timeout to 40 secs, without deadlocks since then.

But instead I got two Crashes with Trap 12: page fault.

lock order reversal: sleepable after non-sleepable
1st 0x... unp@/usr/src/sys/kern/uipc_usrreq.c:981
2nd 0x... usermap@/usr/src/sys/vm/vm_map.c:2997

Any suggestions are welcome.

-- 
Martin Schütte