|
Description
|
While running the ZFS test suite on a sparc machine with the ZFS errgate bits,
> !cat /etc/motd
Sun Microsystems Inc. SunOS 5.11 zfs-errgate-0201 Feb. 01, 2006
SunOS Internal Development: bricker 2006-02-01 [zfs-errbld]
bfu'ed from /net/zday.sfbay/export/zfs-errgate-x86/zfs-errbld/archives/sparc/daily.0201 on 2006-02-01
Sun Microsystems Inc. SunOS 5.11 snv_30 October 2007
I came across this panic :
> ::status
debugging crash dump vmcore.1 (64-bit) from smog
operating system: 5.11 zfs-errgate-0201 (sun4u)
panic message:
assertion failed: (dmfe_ring_get32(descp, i, 0) & (1UL<<31)) == 0, file: ../../s
un4u/io/dmfe/dmfe_main.c, line: 1024
dump content: kernel pages only
> $c
vpanic(12a17c0, 7b25e7f0, 7b25e820, 400, 16, 14)
assfail+0x7c(7b25e7f0, 7b25e820, 400, 1853c00, 12a1400, 0)
dmfe_reclaim_tx_desc+0xc0(30004db29c0, 1f, 1, 1f, fc740004, 6)
dmfe_tick_stall_check+0xcc(30004db29c0, 27, fc740004, 30004db2b30, 7, 1)
dmfe_cyclic+0x74(30004db2b20, ffffff5e, 7b25a7d8, 30004db29c0, 5, 30004db29c0)
cyclic_softint+0x38c(c, 1880cd0, 0, 1880c28, 1853e48, 30003b24000)
cbe_level1+8(0, 0, 180c000, 3f70, 7d0, 100f674)
intr_thread+0x2c0(0, 0, 180c000, 0, 0, 180c000)
swtch+0x218(2a10010fcc0, 180c000, 184c3c0, 184c3c0, 99973f, 0)
cv_wait+0x60(30002ab9978, 30002ab9978, 4, 127c000, 30002ab9958, 0)
taskq_thread+0xd8(30002ab9958, 30002ab99b0, 2a10010faca, 2a10010fac8,
19030e92e77e, 19030e92c7d2)
thread_start+4(30002ab9958, 0, 0, 0, 0, 0)
I've put the crash dump in
/net/zion.eng/export/dumps/timf/<bugid>
Is this reproducible with stock bits? It's pretty much impossible
for the generic FS changes we've made to be responsible for a
panic in the dmfe driver. At least not directly - changing I/O
patterns may tweak some latent bug in the driver. Despite the
fact that this was seen on pre-integration bits, I would assert
that this is a generic problem with dmfe.
Just trying to reproduce it now on the latest snv nightly. I wasn't able to reproduce just
by running the test that was running at the time of the panic, so I'm now running the
entire suite. I'll update this bug if I find any more (and otherwise, will try to reproduce
with the latest err-gate again)
Nope, can't reproduce on either the latest nv bits or the latest err-gate. Despite that, I suspect this could be reassigned to the network driver guys to see if they want to investigate futher. (flapjack_lite/dmfe I think) Will update this bug if I'm able to reproduce in the future.
|