Re: [PATCH] xen/x86/pvh: Use HVM's flush_tlb_others op

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

On 12/15/2015 10:03 AM, Jan Beulich wrote:
>>>> On 15.12.15 at 15:36, <boris.ostrovsky@oracle.com> wrote:
>> On 12/14/2015 10:27 AM, Konrad Rzeszutek Wilk wrote:
>>> On Sat, Dec 12, 2015 at 07:25:55PM -0500, Boris Ostrovsky wrote:
>>>> Using MMUEXT_TLB_FLUSH_MULTI doesn't buy us much since the hypervisor
>>>> will likely perform same IPIs as would have the guest.
>>>>
>>> But if the VCPU is asleep, doing it via the hypervisor will save us waking
>>> up the guest VCPU, sending an IPI - just to do an TLB flush
>>> of that CPU. Which is pointless as the CPU hadn't been running the
>>> guest in the first place.
>>>
>>>> More importantly, using MMUEXT_INVLPG_MULTI may not to invalidate the
>>>> guest's address on remote CPU (when, for example, VCPU from another
>>>> guest
>>>> is running there).
>>> Right, so the hypervisor won't even send an IPI there.
>>>
>>> But if you do it via the normal guest IPI mechanism (which are opaque
>>> to the hypervisor) you and up scheduling the guest VCPU to do
>>> send an hypervisor callback. And the callback will go the IPI routine
>>> which will do an TLB flush. Not necessary.
>>>
>>> This is all in case of oversubscription of course. In the case where
>>> we are fine on vCPU resources it does not matter.
>>
>> So then should we keep these two operations (MMUEXT_INVLPG_MULTI and
>> MMUEXT_TLB_FLUSH_MULT) available to HVM/PVH guests? If the guest's VCPU
>> is not running then TLBs must have been flushed.
> While I followed the discussion, it didn't become clear to me what
> uses these are for HVM guests considering the separate address
> spaces.

To avoid unnecessary IPIs to VCPUs that are not currently scheduled (my 
mistake was that I didn't realize that IPIs to those pCPUs will be 
filtered out by the hypervisor).

> As long as they're useless if called, I'd still favor making
> them inaccessible.


VCPUs that are scheduled will receive the required flush requests.

-boris