Christoph, please excuse me for the long delay with this bugreport. I was busy, and later went to 2-week vacation, and now I'm busy again, you see... ;)
On 06.05.2012 18:15, Christoph Biedl wrote: > Package: qemu-kvm > Version: 1.0+dfsg-11 > Severity: important > > Dear Maintainer, > > after about ten days of searching and trying to reproduce the problem > I'm quite confident I've finally nailed down the problem to qemu-kvm. > > Given the situation: > > Two virtualisation hosts in the same network, designed for migration > (aktually using libvirt). The migration in general works quite fine > unless the guest is running a certain self-compiled 32bit kernel, > versions tested: 2.6.32.59, 3.0.30, 3.2.x, 3.4-rc5. In that case, the > guest will crash immediatly after migration causing a message like: > > divide error: 0000 [#1] > Pid: 0, comm: swapper Not tainted 3.4.0-rc5 #1 Bochs Bochs > EIP: 0060:[<c1006c42>] EFLAGS: 00000246 CPU: 0 > EIP is at default_idle+0x17/0x27 > EAX: 00000000 EBX: c160a044 ECX: 01000000 EDX: 00000000 > ESI: 00099800 EDI: c1591000 EBP: c158ffc8 ESP: c158ffc8 > DS: 007b ES: 007b FS: 0000 GS: 0000 SS: 0068 > CR0: 8005003b CR2: bf8bec3b CR3: 0f066000 CR4: 000006d0 > DR0: 00000000 DR1: 00000000 DR2: 00000000 DR3: 00000000 > DR6: ffff0ff0 DR7: 00000400 > Process swapper (pid: 0, ti=c158e000 task=c159b2e0 task.ti=c158e000) > Stack: > c158ffd0 c1006dcf c158ffd8 c1416096 c158ffe8 c15da876 c160a044 00000800 > c158fff8 c15da075 00000800 00099800 018c3003 00000000 > Call Trace: > [<c1006dcf>] cpu_idle+0x34/0x53 > [<c1416096>] rest_init+0x52/0x54 > [<c15da876>] start_kernel+0x2a9/0x2ae > [<c15da075>] i386_start_kernel+0x75/0x79 > Code: c8 8b 43 08 a8 08 75 08 31 c0 fb 0f 01 c9 eb 01 fb 5b 5d c3 55 89 > e0 89 e5 25 00 e0 ff ff 83 60 0c fb 8b 40 08 a8 08 75 04 fb f4 <eb> 01 fb 89 > e0 25 00 e0 ff ff 83 48 0c 04 5d c3 55 89 e5 0f b3 > EIP: [<c1006c42>] default_idle+0x17/0x27 SS:ESP 0068:c158ffc8 > ---[ end trace 27b93e2e7edb18aa ]--- > Kernel panic - not syncing: Attempted to kill the idle task! > > FWIW, this also happens while using qemu-kvm from Ubuntu precise > (1.0+noroms-0ubuntu13). > > This does not happen using qemu-kvm from squeeze > (0.12.5+dfsg-5+squeeze8) > > The interesting thing is an almost identical kernel configuration > compiled for x86_64 migrates the guest without any flaw, even if using > the very same (32bit) userland. > > This is not related to the host CPU, the AMD used below just has the > advantage of nested virtualisation, allowing me to create two > identical virtualisation hosts. The crash was seen on Core(TM)2 CPU > and Xeon(R) CPU E31220, too. > > Changing the guest CPU type to i686 did not solve the issue. > > Additionally, several vendor kernels tried (among them Debian and > grml) do not show this effect, but trying to somewhat manually bisect > the kernel konfiguration led to even more and even stranger [expletive > deleted] libvirt errors so I didn't continue there. This is the most important and interesting part -- which config option is the problem. But okay, since you provided the config, I guess it should be possible to reproduce the issue. But first, can you please try the new 1.1 version of qemu-kvm (currently in sid), whenever it shows the same issue too (please be aware about a new bug in 1.1, http://bugs.debian.org/679788, which should be easy to work around). If it still present, I'll try to work with your kernel config and see what's the problem (I don't have lots of hw power currently for kernel compiles, and for migration too -- so I'll try migrating on a single host between two kvm instances -- does this way shows the problem too -- provided libvirt supports it to start with?). Thank you for your patience! /mjt -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
