[Xen-devel] [PATCH 1/3] arm_arch_timer: introduce arch_timer_stolen_ticks

[Stefano Stabellini]

On Fri, 3 May 2013, Marc Zyngier wrote:
> On 03/05/13 11:43, Stefano Stabellini wrote:
> > On Thu, 2 May 2013, Christopher Covington wrote:
> >>> So the virtual timer should appear to have been running even while time
> >>> is being stolen and therefore stolen time needs to be accounted via some
> >>> other means.
> >>
> >> Something that's not currently obvious to me is that given that the stolen
> >> cycle accounting should be done, what makes the architected timer interrupt
> >> handler the ideal place to do it?
> > 
> > That is a good question and I would appreciate suggestions to improve
> > the patch.
> > 
> > Given that Xen x86 and ia64 does stolen time accounting from the timer
> > interrupt handler:
> > 
> > arch/x86/xen/time.c:xen_timer_interrupt
> > arch/ia64/kernel/time.c:timer_interrupt
> > 
> > and given that the arch_timer is the only timer used by Xen on ARM and
> > that it includes a virt_timer that is made on purpose to be used by
> > virtual machines, I thought that it might be a good place for it.
> > 
> > I also thought that doing it this way, KVM should be able to reuse the
> > same hook.
> 
> Indeed. I just need to understand how time stealing works there ;-).
> 
> Now, KVM is not necessarily limited to arch_timers, and we've run KVM
> using a QEMU-provided timer in the past. Can you think of a more generic
> location for this hook? Possibly something that would satisfy the
> requirements of other architectures while we're at it?

Probably the best option would be to reuse

kernel/sched/cputime.c:steal_account_process_tick

but that also means introducing CONFIG_PARAVIRT on arm.
I am up for that, what do you think?