> On Jun 26, 2019, at 5:26 PM, Krish Sadhukhan <krish.sadhukhan@xxxxxxxxxx> wrote: > > > On 6/25/19 5:10 AM, Nadav Amit wrote: >> Do not assume that the local APIC is in a xAPIC mode after reset. >> Instead reset it first, since it might be in x2APIC mode, from which a >> transition in xAPIC is invalid. >> >> Note that we do not use the existing disable_apic() for the matter, >> since it also re-initializes apic_ops. > > > Is there any issue if apic_ops is reset ? So I checked again, and actually the problem was different. Beforehand, I used reset_apic(), which used apic_ops to write to SPIV. And the race with setting x2apic caused it to occasionally use the x2APIC MSR interface to set SPIV, which triggered an exception. I’ll send v2 that changes reset_apic() not to use apic_ops. Thanks.
