some guests hangs after migration (qemu-kvm-0.12)

[Michael Tokarev]

There's a quite good bugreport in Debian BTS about this, #580649: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=580649

This is not the same as lp#341682, since it's now 0.12.

Full initial message from #580649:

----
From: Apollon Oikonomopoulos <email address hidden>
To: Debian Bug Tracking System <email address hidden>
Subject: qemu-kvm: Guests hang after live migration
Date: Fri, 07 May 2010 16:29:10 +0300

Package: qemu-kvm
Version: 0.12.3+dfsg-4
Severity: important

Hi,

qemu-kvm 0.12.3 causes guests to hang after a live migration. The crash seems
to be related to the guest's virtio subsystem, as per the following backtrace
obtained _in the guest_:

May 7 14:17:32 kot kernel: [ 285.035681] irq 11: nobody cared (try booting with the "irqpoll" option)
May 7 14:17:32 kot kernel: [ 285.035681] Pid: 0, comm: swapper Not tainted 2.6.26-2-amd64 #1
May 7 14:17:32 kot kernel: [ 285.035681]
May 7 14:17:32 kot kernel: [ 285.035681] Call Trace:
May 7 14:17:32 kot kernel: [ 285.035681] <IRQ> [<ffffffffa008b06c>] :virtio_pci:vp_interrupt+0x27/0xb8
May 7 14:17:32 kot kernel: [ 285.035681] [<ffffffff8026c6d3>] __report_bad_irq+0x30/0x72
May 7 14:17:32 kot kernel: [ 285.035681] [<ffffffff8026c912>] note_interrupt+0x1fd/0x23b
May 7 14:17:32 kot kernel: [ 285.035681] [<ffffffff8026d19b>] handle_fasteoi_irq+0xa5/0xc8
May 7 14:17:32 kot kernel: [ 285.035681] [<ffffffff8020f5dc>] do_IRQ+0x6d/0xd9
May 7 14:17:32 kot kernel: [ 285.035681] [<ffffffff8020b093>] default_idle+0x0/0x49
May 7 14:17:32 kot kernel: [ 285.035681] [<ffffffff8020c45d>] ret_from_intr+0x0/0x19
May 7 14:17:32 kot kernel: [ 285.035681] <EOI> [<ffffffff8021a717>] lapic_next_event+0x0/0x13
May 7 14:17:32 kot kernel: [ 285.035681] [<ffffffff8021eb64>] native_safe_halt+0x2/0x3
May 7 14:17:32 kot kernel: [ 285.035681] [<ffffffff8021eb64>] native_safe_halt+0x2/0x3
May 7 14:17:32 kot kernel: [ 285.035681] [<ffffffff8020b0bd>] default_idle+0x2a/0x49
May 7 14:17:32 kot kernel: [ 285.035681] [<ffffffff8020ad04>] cpu_idle+0x8e/0xb8
May 7 14:17:32 kot kernel: [ 285.035681]
May 7 14:17:32 kot kernel: [ 285.035681] handlers:
May 7 14:17:32 kot kernel: [ 285.035681] [<ffffffff8038f27b>] (usb_hcd_irq+0x0/0x78)
May 7 14:17:32 kot kernel: [ 285.035681] [<ffffffffa008b045>] (vp_interrupt+0x0/0xb8 [virtio_pci])
May 7 14:17:32 kot kernel: [ 285.035681] Disabling IRQ #11

This happens in a reproducible fashion on a guest with the following setup:

  * Virtio net + block devices
  * 8 CPU SMP
  * 521 MB RAM
  * Debian Lenny with 2.6.26-2-amd64

Just migrating the guest a couple of times with a bit of I/O activity will
cause it to freeze completely.

It does *not* seem to impact the following:
  * Guests with no virtio devices
  * Guests running 2.6.32-4-amd64
  * Guests running under qemu-kvm-0.11.1

Thus, it seems to be virtio-specific and possibly related to the guest kernel,
but since qemu-kvm-0.11.1 works fine, I'm filing it as a qemu-kvm regression
and not a kernel bug.

Thank you