Re: [PATCH 6/6] KVM: MMU: Avoid handling same rmap_pde in kvm_handle_hva_range()

[Avi Kivity <avi@redhat.com>]

On 07/01/2012 04:18 PM, Takuya Yoshikawa wrote:
> On Sun, 01 Jul 2012 10:41:05 +0300
> Avi Kivity <avi@redhat.com> wrote:
> 
>> >> > Note: in the new code we could not use trace_kvm_age_page(), so we just
>> >> > dropped the point from kvm_handle_hva_range().
>> >> > 
>> >> 
>> >> Can't it be pushed to handler()?
>> > 
>> > Yes, but it will be changed to print rmap, not hva and gfn.
>> > I will do in the next version.
>> 
>> We should try to avoid changing the tracepoint ABI.  If we pass the
>> memslot to handler(), we should be able to reconstruct the gfn and hva.
> 
> That's not enough because we only know rmapp in the handler: we also need
> level or something.  I agree that passing memslot for keeping the ABI is
> good, but how about two more arguments?

I don't see a problem with that.  But it would be good to move all the
address transformations into tp_fast_assign so we only perform them if
the tracepoint is active.

> BTW, trace_kvm_age_page() is tracing every handler, not just kvm_age_rmapp().
> Is this intentional?

No, it's a bug.  Pushing the trace into the handler will fix it though.

> If we trace in the caller, in kvm_age_hva(), we know hva, but not gfn.

gfn is useful to correlate to later events with the same gfn.

-- 
error compiling committee.c: too many arguments to function