Disk I/O-related panics in 6.0-RELEASE?

[Brian Tao]

    Not sure if this is specifically an AMD64 issue or a 6.0 issue,
but I'm leaning towards the former, so I'll start here.  ;-)

FreeBSD 6.0-RELEASE-p4 #2: Tue Feb  7 18:04:40 EST 2006 taob at cabal.dm.egate.net:/depot/obj/depot/src/sys/LUXOKERNEL


    I'm seeing unpredictable kernel panics that appear to be related
to filesystem activity.  I have yet to be able to produce the panic on
demand.  I've seen panics after 30 seconds of running the postmark
benchmark, and other times it runs for hours without incident.
buildworlds will cause a panic maybe 1 out of 3 attempts.  The server
is very lightly loaded (it is not yet in full-scale production).

    Crash dumps show that a variety of processes can trigger the
panic.  I've seen pax, ar, fgrep, and even sync, bufdaemon and
fsck_ufs.  The instruction pointer in the kernel dump points to a
variety of places:

# fgrep -h "is in" vmcore*/kgdb.txt
0xffffffff802a01c9 is in cluster_wbuild (/depot/src/sys/kern/vfs_cluster.c:904).
0xffffffff8034bc96 is in softdep_disk_io_initiation (/depot/src/sys/ufs/ffs/ffs_softdep.c:3597).
0xffffffff8037e010 is in vm_page_io_start (/depot/src/sys/vm/vm_page.c:345).
0xffffffff8029d5c0 is in bufstrategy (vnode_if.h:928).
0xffffffff802a8f49 is in buf_vlist_remove (/depot/src/sys/kern/vfs_subr.c:1343).
0xffffffff8034bc96 is in softdep_disk_io_initiation (/depot/src/sys/ufs/ffs/ffs_softdep.c:3597).
0xffffffff8037e010 is in vm_page_io_start (/depot/src/sys/vm/vm_page.c:345).

    I've collected the output of kgdb from nine crash dumps, my kernel
config and the dmesg.boot here:

http://www.luxography.ca/tmp/amd64_panics.tgz

    The system has probably panicked about twenty times in the past
two months (it was turned up at the beginning of February).  I'm
hoping I can stick to the amd64 kernel, but I need better stability
than this.  Any help would be appreciated!
-- 
Brian Tao, Luxography
http://www.luxography.ca/ (main)
http://blog.luxography.ca/ (blog)
"The art of light"