On Thu, Jun 15, 2017 at 2:05 PM, Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote:
>
>
> On 15/06/2017 13:20, Ladi Prosek wrote:
>> @@ -4823,6 +4826,22 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu)
>> if (unlikely(svm->nested.exit_required))
>> return;
>>
>> + /*
>> + * Disable singlestep if we're injecting an interrupt/exception.
>> + * We don't want our modified rflags to be pushed on the stack where
>> + * we might not be able to easily reset them if we disabled NMI
>> + * singlestep later.
>> + */
>> + if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
>> + /*
>> + * We enabled NMI singlestepping because the NMI window was
>> + * closed. It's unlikely that injecting another event will make
>> + * it any better. Try again later, on next iret at the latest.
>> + */
>> + disable_nmi_singlestep(svm);
>> + set_intercept(svm, INTERCEPT_IRET);
>> + }
>> +
>> pre_svm_run(svm);
>>
>> sync_lapic_to_cr8(vcpu);
>>
>
> I wonder if we could just force an immediate vmexit instead of asking
> for one at the next IRET. Based on the AMD manual, event injection
> happens before external interrupts cause a vmexit.
>
> Interrupts here are disabled (through either IF or GIF) until VMRUN, so
> just a
>
> smp_send_reschedule(vcpu->cpu);
>
> should be enough after disable_nmi_singlestep. Or if you put the code
> in pre_svm_run you have a cpu local variable.
Yup, this seems to work. Thanks!
> Paolo
