Re: [Xen-devel] [PATCH v3 5/6] vpt: add support for level interrupts

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Fri, Jun 22, 2018 at 09:53:26AM -0600, Jan Beulich wrote:
> >>> On 22.06.18 at 17:24, <roger.pau@xxxxxxxxxx> wrote:
> > On Fri, Jun 22, 2018 at 08:23:02AM -0600, Jan Beulich wrote:
> >> >>> On 08.06.18 at 17:07, <roger.pau@xxxxxxxxxx> wrote:
> >> > @@ -316,7 +317,9 @@ int pt_update_irq(struct vcpu *v)
> >> >          if ( pt->pending_intr_nr )
> >> >          {
> >> >              /* RTC code takes care of disabling the timer itself. */
> >> > -            if ( (pt->irq != RTC_IRQ || !pt->priv) && pt_irq_masked(pt) 
> >> > )
> >> > +            if ( (pt->irq != RTC_IRQ || !pt->priv) && pt_irq_masked(pt) 
> >> > &&
> >> > +                 /* Level interrupts should be asserted even if masked. 
> >> > */
> >> > +                 !pt->level )
> >> >              {
> >> >                  /* suspend timer emulation */
> >> 
> >> Especially with this comment I'm not convinced this change is fully
> >> correct: Once a level triggered interrupt is latched in IRR, no
> >> further assertions are meaningful, and hence emulation could (and
> >> hence should) still be stopped to reduce resource consumption.
> > 
> > Hm, I can see your point, but that's going to make the implementation
> > of HPET level trigger interrupts more complex.
> > 
> > From my reading of the spec, when a level triggered HPET interrupt
> > fires the ISR bit must be set, regardless of whether the IRR of the
> > io-apic entry is set or not.
> > 
> > Maybe the right solution is to add a pt_irq_pending that checks the
> > IRR bit, and a new flag that the caller can set in order to requests
> > callbacks regardless of whether the interrupt has been injected or
> > not. Would you agree to that plan?
> 
> That sounds like a reasonable option. Another would be to add a
> parameter to the callback allowing the context to be identified to the
> callee.

Yes, but that seems less optimal since all callback will be executed
and current callers would just exit if the interrupt was masked. I
will add the new option an send a new version.

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel