So, refreshing it all again with more correct information.
First, it appears that only 32bit version of qemu userspace is
affected, and only when using 32bit version of guest. So far,
either 64bit qemu userpace or 64bit windows7 guest is enough
for the issue to not show up. Second, both win7 and winXP
(32bit) are affected. Third, current qemu-kvm/master appears
to be unaffected too.
The problem is that on first and all subsequent guest REBOOT
(not first boot but really REBOOT) -- after the guest rebuilt
its HAL I guess, due to some changes in "hardware" it noticed, --
the guest BSODs during boot with this code:
STOP: 0x0000005C (0x000000000000010B, 0x0000000000000003,
0x000000000000000, 0x000000000000000)
After painful bisection between 1.0 and current git, the result
is this:
First bad commit:
commit 5b6fb069378e61c45c577bbec3d7ef60367f7e4c
Author: Avi Kivity <a...@redhat.com>
Date: Thu Mar 1 12:58:46 2012 +0200
pc-bios: update kvmvapic.bin
Signed-off-by: Avi Kivity <a...@redhat.com>
diff --git a/pc-bios/kvmvapic.bin b/pc-bios/kvmvapic.bin
new file mode 100755
index 0000000..045f5c2
Binary files /dev/null and b/pc-bios/kvmvapic.bin differ
Which, despite the name, actually INTRODUCES kvmvapic.bin,
not merely updates it. So apparently this introduction of
kvmvapic.bin resulted in autoloading implemented earlier to
starte entering the game, which appears to be this one:
commit e5ad936b0fd7dfd7fd7908be6f9f1ca88f63b96b
Author: Jan Kiszka <jan.kis...@siemens.com>
Date: Fri Feb 17 18:31:19 2012 +0100
kvmvapic: Introduce TPR access optimization for Windows guests
And indeed, we're talking about windows guests here...
Anything I can do further to debug/fix the issue?
Thanks,
/mjt
