[regression] KVM: hangs and "irq timeout" booting HURD unless -no-kvm-irqchip passed

[Jonathan Nieder <jrnieder@gmail.com>]

Hi kvm-ers,

When I boot the HURD with

	kvm -m 768 -net nic,model=ne2k_pci -net user hurd-installed.qemu

it hangs and eventually produces two messages:

	hd0: unexpected_intr: status=0x58 { DriveReady SeekComplete DataRequest }
	hd0: irq timeout: status=0x58 { DriveReady Seek Complete DataRequest }

More details below[1].

Adding -no-kvm-irqchip to the kvm command line fixes it --- no
more hangs or confusing messages (thanks, Guillem!).

Bisects (thanks to Michael for the idea) to

	v2.6.37-rc1~142^2~39 (KVM: Check for pending events before
	attempting injection, 2010-07-27).

Bisection log and kernel configuration available upon request (but
probably not too relevant --- the 2.6.37 distro kernel from Debian
exhibits the same problem).  Reproducible with kvm/master (2d4b4d26,
2011-02-01).

CPU is a dual-core AMD Athlon II P360, family 16, model 6.

Any hints for tracking this down?  For those wanting to follow along
at home, you can find a HURD cd to try at [2].

Thanks for kvm --- now that I have hardware that can handle it, it's
become a favorite toy. :)

Jonathan

[1] Background: http://bugs.debian.org/612105
Result:

 1. Normal boot procedure.  Eventually the following message from HURD
    is on the console:

	start ext2fs: Hurd server bootstrap: ext2fs[device:hd0s1] exec

 2. Hang.  I hit ctrl+alt, "info registers", and so on in the hope of
    waking it up.  Eventually...

 3. A flood of repeated

	kb_setleds1: unexpected state (1)

    messages.

 4. Another message and another hang:

	hd0: unexpected_intr: status=0x58 { DriveReady SeekComplete DataRequest }
	hd0: irq timeout: status=0x58 { DriveReady Seek Complete DataRequest }

    (the two messages overlap sometimes).
[2] http://people.debian.org/~sthibault/hurd-i386/installer/cdimage/