Re: [PATCH] KVM: nVMX: Fix IRQs inject to L2 which belong to L1 since race

[Wanpeng Li <wanpeng.li@linux.intel.com>]

On Mon, Jul 07, 2014 at 01:38:37PM -0400, Bandan Das wrote:
>Paolo Bonzini <pbonzini@redhat.com> writes:
>
>> Il 07/07/2014 19:31, Bandan Das ha scritto:
>>>> >
>>>> > Which patch?  Yours or Bandan's?
>>> Why don't we hold off on Wanpeng's patch and instead apply the one I proposed
>>> to call check_nested_events() when checking for interrupt in inject_pending_event() ?
>>
>> Exactly, yours seemed better to apply as a quick regression fix.
>>
>> Can you post it as a toplevel patch, so that the commit message
>> explains what's happening?  Perhaps add a comment in the code as well.
>
>Ok, will do, thanks!

As Jan metioned in http://www.spinics.net/lists/kvm/msg105238.html, "In any case, 
unconditionally setting KVM_REQ_EVENT seems strange and should be changed." Your 
trick still keep the unconditionally setting KVM_REQ_EVENT which is the root cause 
of the race there, anyway, I focus on fix the hang currently and a patch will be 
submitted soon. 

Regards,
Wanpeng Li 

>
>> Paolo
>>
>>> I think that will take care of https://bugzilla.kernel.org/show_bug.cgi?id=72381
>>> too. Once, we figure out what's causing hangs under certain conditions with his
>>> patch, we can apply that and revert this change.
>>>
>>>