Re: [PATCH v6 06/12] Add PV MSR to enable asynchronous page faults delivery.

[Gleb Natapov <gleb@redhat.com>]

On Sun, Oct 10, 2010 at 02:47:43PM +0200, Avi Kivity wrote:
>  On 10/07/2010 07:53 PM, Gleb Natapov wrote:
> >On Thu, Oct 07, 2010 at 02:42:06PM +0200, Avi Kivity wrote:
> >>   On 10/04/2010 05:56 PM, Gleb Natapov wrote:
> >>  >Guest enables async PF vcpu functionality using this MSR.
> >>  >
> >>  >   			return NON_PRESENT;
> >>  >+
> >>  >+MSR_KVM_ASYNC_PF_EN: 0x4b564d02
> >>  >+	data: Bits 63-6 hold 64-byte aligned physical address of a 32bit memory
> >>
> >>  Given that it must be aligned anyway, we can require it to be a
> >>  64-byte region and also require that the guest zero it before
> >>  writing the MSR.  That will give us a little more flexibility in the
> >>  future.
> >>
> >No code change needed, so OK.
> 
> The guest needs to allocate a 64-byte per-cpu entry instead of a
> 4-byte entry.
> 
Yes, noticed that already :(

> 
> >>  >+
> >>  >+	kvm_async_pf_wakeup_all(vcpu);
> >>
> >>  Why is this needed?  If all apfs are flushed at disable time, what
> >>  do we need to wake up?
> >For migration. Destination will rewrite msr and all processes will be
> >waked up.
> 
> Ok. What happens to apf completions that happen after all vcpus are stopped?
> 
They will be cleaned by kvm_clear_async_pf_completion_queue() on vcpu
destroy.

--
			Gleb.