Re: [Xen-devel] HVM crash system on AMD APU A8-6600K

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 08.04.14 at 10:40, <highwaystar.ru@xxxxxxxxx> wrote:
> I can summirize and duplicate previous logs.
> 
> - Whole system crashed means system going to hard reset without any
> notice in log or on screen. screen goes black and next thing what I
> can see is POST of real hardware.
> - Without passing the GPU problem still here. Log provided made with
> empty hvm config:
> name = 'blank'
> builder = 'hvm'
> memory = 1024
> - logs with options sync_console and watchdog attached, both made with
> blank HVM config
> - lspci of my system attached
> - the way case my system doesnt crash is using debug xen version
> I can make new log on this system.

Okay, so
- the logs seem to indicate multiple VM starts ("AMD-Vi: Share p2m
  table with iommu: p2m table = 0x..."), but that's hard to tell
  without guest_loglvl=all
- the "APIC error on CPU..." messages seem to be issued at boot
  time, i.e. we can likely discount them being relevant for the reboot
  you observe

The system spontaneously rebooting is of course going to be harder
to debug than an ordinary crash. Since you can't see the issue with
a debug hypervisor, and since the "hvm_debug=" option by default
is effective only in debug hypervisors, you'll want to edit
xen/include/asm-x86/hvm/support.h to always enable this, and then
pass "hvm_debug=0xfff" on the Xen command line. This is going to
be very verbose, so you'll have to see whether you would better
turn off one or the other bit in that mask. All we want to see is
whether this allows us to see when in the process of starting the
guest the system dies, including whether this is always at the same
point. That may then allow us to hand you a debugging patch to
further isolate things.

Another thing to check is whether with the IOMMU turned off
entirely you also see the problem.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel