> From: Radim Krcmar
> Sent: Thursday, March 24, 2016 6:55 AM
>
> 2016-03-23 22:11+0100, Paolo Bonzini:
> > On 23/03/2016 22:06, Jan Kiszka wrote:
> >> > There is also at least one case where you could have one pending
> >> > (but not injected) and one latched NMI at instruction boundary, and
> >> > that is the special NMI shadow from the STI instruction.
> >>
> >> This sounds to me like we should try to address the issue Yuki is
> >> seeing without playing with the nmi_pending counter.
>
> I agree, the discussion was mostly offtopic, sorry.
>
> > Yes, it would be nice if my suggestion worked. :)
>
> And if not, we'll at least learn what 2.6.32 is doing. :)
I really appreciate your cooperation and valuable discussion. I understand that
changing nmi_pending count is not good because it is not guaranteed as safe.
Sorry my mistake.
I tested the logic suggested by Paolo in our environment, and it worked. I will
post v3, replacing second part by this logic.
��.n��������+%������w��{.n�����o�^n�r������&��z�ޗ�zf���h���~����������_��+v���)ߣ�
