Re: [Xen-devel] [Patch] x86/HVM: Fix RTC interrupt modelling

[Tim Deegan <tim@xxxxxxx>]

At 13:31 +0000 on 11 Feb (1392121899), Jan Beulich wrote:
> >>> On 11.02.14 at 14:15, Tim Deegan <tim@xxxxxxx> wrote:
> > At 12:50 +0000 on 11 Feb (1392119457), Jan Beulich wrote:
> >> Right, except you do this be reverting other stuff rather than
> >> adding the missing functionality on top.
> > 
> > Absolutely -- because once we went back to having PF set only when the
> > interrupt was injected, it seemed better to reduce the amount of
> > special-case plumbing for RTC than to add yet more.
> > 
> > But for the case of an OS polling for PF with PIE clear, I guess we
> > might need to keep all the current special cases.  Was that a known
> > observed bug or a theoretical one?
> 
> A theoretical one. Along the lines of the general theme of all the
> patches around that time: Get the emulation closer to what real
> hardware does.

Righto.

> > I can't see a way of handling both that case and the w2k3 case.
> 
> Since hardware can, software certainly could as well.

Hardware doesn't have to do things like no-missed-ticks mode; it
certainly doesn't have to deal with no-ack mode. :(

> > Or do we have to treat 'masked in REG_B/IOAPIC' differently from
> > 'masked by ISR/TPR/RFLAGS.IF/...'?
> 
> ... this might be the right direction (albeit I think it would be REG_B
> on one side and collectively IOAPIC/ISR/TPR/EFLAGS.IF).

Yeah, maybe.

Something like, if !PIE then we set PF and consume the tick in the
early vpt callback, otherwise we do it in the late callback (and in
both cases, the early vpt callback doesn't actually assert the line)?

Or: we drop the early callback, and go back to setting PF|IRQF in the
pt_intr_post callback (much like the patch under discussion), and
disable the vpt when the guest clears PIE.  Then in the REG_C read, if
!PIE, we can set the PF bit if a tick should have happened since
the last read (but saving ourselves the bother of running the actual
timer).  Then the only case where things are wrong is an OS which
_both_ polls for the edge _and_ relies on the vpt logic for adding the
right number of ticks after being descheduled.  Or an OS which asks
for interrupts, masks them in the *APIC/RFLAGS and then polls for PF.
But that guest, I think, is indistinguishable from w2k3 in the stuck
state.

Actually that second patch doesn't sound too bad.  If that sounds OK
to you I can look into coding it up on Thursday.

Tim.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel