On Thu, Jan 13, 2022 at 11:43:30AM +0000, Marc Zyngier wrote:
> On Thu, 13 Jan 2022 11:17:53 +0000,
> Mark Rutland <mark.rutland@xxxxxxx> wrote:
> >
> > On Tue, Jan 11, 2022 at 05:55:20PM +0000, Marc Zyngier wrote:
> > > On Tue, 11 Jan 2022 15:35:36 +0000,
> > > Mark Rutland <mark.rutland@xxxxxxx> wrote:
>
> [...]
>
> > > > @@ -891,26 +909,23 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
> > > > kvm_arch_vcpu_ctxsync_fp(vcpu);
> > > >
> > > > /*
> > > > - * We may have taken a host interrupt in HYP mode (ie
> > > > - * while executing the guest). This interrupt is still
> > > > - * pending, as we haven't serviced it yet!
> > > > + * We must ensure that any pending interrupts are taken before
> > > > + * we exit guest timing so that timer ticks are accounted as
> > > > + * guest time. Transiently unmask interrupts so that any
> > > > + * pending interrupts are taken.
> > > > *
> > > > - * We're now back in SVC mode, with interrupts
> > > > - * disabled. Enabling the interrupts now will have
> > > > - * the effect of taking the interrupt again, in SVC
> > > > - * mode this time.
> > > > + * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
> > > > + * context synchronization event) is necessary to ensure that
> > > > + * pending interrupts are taken.
> > > > */
> > > > local_irq_enable();
> > > > + isb();
> > > > + local_irq_disable();
> > >
> > > Small nit: we may be able to elide this enable/isb/disable dance if a
> > > read of ISR_EL1 returns 0.
> >
> > Wouldn't that be broken when using GIC priority masking, since that
> > can prevent IRQS being signalled ot the PE?
>
> You're right. But this can be made even simpler. We already know if
> we've exited the guest because of an IRQ (ret tells us that), and
> that's true whether we're using priority masking or not. It could be
> as simple as:
>
> if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
> // We exited because of an interrupt. Let's take
> // it now to account timer ticks to the guest.
> local_irq_enable();
> isb();
> local_irq_disable();
> }
>
> and that would avoid accounting the interrupt to the guest if it fired
> after the exit took place.
>
> > I'm happy to rework this, but I'll need to think a bit harder about
> > it. Would you be happy if we did that as a follow-up?
>
> Oh, absolutely. I want the flow to be correct before we make it
> fast(-ish).
Cool; I'll leave that for now on the assumption we'll address that with a
follow-up patch, though your suggestion above looks "obviously" correct to me.
Thanks,
Mark.
