On 17/10/2016 14:54, Christian Borntraeger wrote:
> On 10/17/2016 02:50 PM, Paolo Bonzini wrote:
>>> Some testcase will trigger a guest panic state. For testing purposes
>>> it can be useful to exit QEMU anyway.
>>
>> I wonder if this should be done by default *unless* -no-shutdown is
>> provided. This would require some planning (and delay this to 2.9,
>> in all likelihood), but it probably would be pretty nice for general
>> usage.
>
> Yes, might also an option. There are basically two cases
> a: guest panic
> b: qemu panic (e.g. if KVM_RUN return EFAULT)
>
> I think for b, the current behaviour might be better.
(b) is not a guest panic, it's "INTERNAL_ERROR" right? It would be easy
to accomodate the difference. I tend to agree, since one may want to
play with the monitor in that case (e.g. x/10i $pc-20).
> In any
> case I want a tuneable and either -no-panic or the new -no-shutdown
> would allow that.
Let's change -no-shutdown then. Actually I think we might even change
it in 2.8, since for example Libvirt always uses -no-shutdown and
everyone else that doesn't use it would probably hang on panics.
>>> void qemu_system_guest_panicked(void)
>>> {
>>> + if (no_panic)
>>> + return qemu_system_shutdown_request();
>>> if (current_cpu) {
>>> current_cpu->crash_occurred = true;
>>> }
I think the "if (no_panic)" should go at the end so that the SHUTDOWN
event is sent after GUEST_PANICKED.
You would also have to add 'poweroff' to the GuestPanicAction enum too,
adjusting qemu_system_guest_panicked's call to
qapi_event_send_guest_panicked.
>>> @@ -3780,6 +3783,9 @@ int main(int argc, char **argv, char **envp)
>>> case QEMU_OPTION_no_shutdown:
>>> no_shutdown = 1;
>>> break;
>>> + case QEMU_OPTION_no_panic:
>>> + no_panic = 1;
>>> + break;
>>> case QEMU_OPTION_show_cursor:
>>> cursor_hide = 0;
>>> break;
>>> --
>>> 2.5.5
>>>
>>>
>>
>
>
>
