[Qemu-devel] Rare linux boot failures ("divide error") with 0.9.1

[Vegard Nossum]

Hi,

I am writing because I've encountered a particular (but very rare)
error that sometimes happen with qemu. I am booting the linux kernel
and suddenly, I get a flood of messages like this:

[    2.335001] RIP  [<ffffffff815d44cb>] oops_end+0x6b/0xe0
[    2.335001]  RSP <ffff880001a49528>
[    2.335001] divide error: 0000 [#61] SMP
[    2.335001] last sysfs file:
[    2.335001] CPU 0
[    2.335001] Modules linked in:
[    2.335001] Pid: 1, comm: swapper Tainted: G      D    2.6.30-rc5 #401
[    2.335001] RIP: 0010:[<ffffffff815d44cb>]  [<ffffffff815d44cb>]
oops_end+0x6b/0xe0
[    2.335001] RSP: 0018:ffff880001a49328  EFLAGS: 00000292
[    2.335001] RAX: 0000000000000000 RBX: ffff880001a49478 RCX: 0000000000000000
[    2.335001] RDX: 000000000000d7d7 RSI: 0000000000000046 RDI: 0000000000000007
[    2.335001] RBP: ffff880001a49348 R08: 0000000000000000 R09: 0000000000000000
[    2.335001] R10: 0000000000000000 R11: 0000000000000002 R12: 0000000000000292
[    2.335001] R13: 000000000000000b R14: 0000000000000292 R15: 0000000000000008
[    2.335001] FS:  0000000000000000(0000) GS:ffff880001a4d000(0000)
knlGS:0000000000000000
[    2.335001] CS:  0010 DS: 0018 ES: 0018 CR0: 000000008005003b
[    2.335001] CR2: ffff8800078000d8 CR3: 0000000001001000 CR4: 00000000000006e0
[    2.335001] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[    2.335001] DR3: 0000000000000000 DR6: 0000000000004000 DR7: 0000000000000000
[    2.335001] Process swapper (pid: 1, threadinfo ffff8800078c2000,
task ffff8800078c8000)
[    2.335001] Stack:
[    2.335001]  ffff880001a49348 ffff880001a49478 0000000000000000
ffffffff817525b5
[    2.335001]  ffff880001a49378 ffffffff8100fb26 ffff8800078c8000
0000000000000000
[    2.335001]  ffff880001a49478 0000000000000000 ffff880001a493c8
ffffffff815d3ea6
[    2.335001] Call Trace:
[    2.335001] Code: c7 05 d6 86 23 00 ff ff ff ff e8 51 28 a7 ff 8b
05 5f 1c 3a 00 83 e8 01 85 c0 89 05 54 1c 3a 00 75 06 fe 05 44 1c 3a
00 41 54 9d <e8> f0 27 a7 ff 45 85 ed 74 43 65 48 8b 04 25 a8 b4 00 00
48 63

So apparently we got a divide error, but the instruction in question
(as reported by the CPU) is not a division:

Code starting with the faulting instruction
===========================================
   0:   e8 f0 27 a7 ff          callq  0xffffffffffa727f5
   5:   45 85 ed                test   %r13d,%r13d

I've seen this maybe 4-5 times out of hundreds of bootups, and the
next time I try to run it, everything works fine (this is also why I
can never capture the _first_ such message; I'm not expecting it and
therefore not logging it), with no change in image, disk, settings, or
command line.

I am using qemu version 0.9.1 on x86_64 (emulating x86_64 as well):

$ rpm -q -i qemu
Name        : qemu                         Relocations: (not relocatable)
Version     : 0.9.1                             Vendor: Fedora Project
Release     : 12.fc10                       Build Date: Sun 11 Jan
2009 06:48:00 PM CET
Install Date: Sun 26 Apr 2009 03:30:56 PM CEST      Build Host:
xenbuilder4.fedora.phx.redhat.com
Group       : Development/Tools             Source RPM:
qemu-0.9.1-12.fc10.src.rpm


Kind regards,

Vegard