Re: [Xen-devel] Time Skewing on Windows XP

[Tim Deegan <tim@xxxxxxx>]

At 16:43 +0000 on 14 Mar (1363279436), Jan Beulich wrote:
> >>> On 14.03.13 at 17:30, Tim Deegan <tim@xxxxxxx> wrote:
> > At 16:06 +0000 on 14 Mar (1363277205), Jan Beulich wrote:
> >> >>> On 14.03.13 at 16:16, Suravee Suthikulanit 
> >> >>> <suravee.suthikulpanit@xxxxxxx> 
> > wrote:
> >> > I finally traced the issue back to the patch that this first happened.  
> > This 
> >> > bug started in the patch :
> >> > 
> >> > H86/HVM: assorted RTC emulation adjustment (w/ git commit id 
> >> > 620d5dad54008e40798c4a0c4322aef274c36fa3)
> >> > 
> >> > I believe there are some issues with the changes in rtc_ioport_read in 
> >> > the 
> >> > arch/x86/hvm/rtc.c and in the pt_update_irq().
> >> 
> >> One thing you may want to try is remove the call from REG_C
> >> read to rtc_timer_update() again - on a second thought it may
> >> be wrong to do it here, as (other than check_update_timer()
> >> and alarm_timer_update()) the function doesn't change with
> >> RTC_PF getting cleared (i.e. I may have wrongly added the call
> >> in analogy to the other two).
> > 
> > The call is needed because of the new code in rtc_periodic_interrupt()
> > that disables the timer if a second passes with no RTC_C read to clear
> > the RTC_PF flag. 
> 
> Ah, right, of course. You see that it was many months back that I
> wrote that code.
> 
> So we could filter on this periodic timer not currently being active
> (which ought to be better than your calling into
> create_periodic_time() with just the time delta adjusted).

I tried adding a flag and gating on it but it didn't fix the drift. 
I may have got it wrong, or there may be something about piix4 RTCs that
Windows expects.  Reading the piix4 spec it's not at all clear that
leaving RTC_PF set suppresses interrupts (even if that's how an RTC
ought to behave).

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