Re: [PATCHv4 1/2] x86/ept: invalidate guest physical mappings on VMENTER

[George Dunlap <george.dunlap@citrix.com>]

On 14/12/15 15:09, Andrew Cooper wrote:
> On 14/12/15 15:00, David Vrabel wrote:
>> On 14/12/15 14:52, Andrew Cooper wrote:
>>> On 14/12/15 14:39, David Vrabel wrote:
>>>> diff --git a/xen/arch/x86/mm/p2m-ept.c b/xen/arch/x86/mm/p2m-ept.c
>>>> index eef0372..6e0cf89 100644
>>>> --- a/xen/arch/x86/mm/p2m-ept.c
>>>> +++ b/xen/arch/x86/mm/p2m-ept.c
>> [...]
>>>> +    on_selected_cpus(d->domain_dirty_cpumask,
>>>>                       __ept_sync_domain, p2m, 1);
>>> You can drop __ept_sync_domain() entirely by using
>>> smp_send_event_check_mask() instead, which is a no-op IPI (and slightly
>>> less overhead while holding the IPI lock).
>> We need to wait until the IPI has been handled on the remote PCPUs since
>> we may immediately free a page table page.  If a VCPU was still running
>> it may use paging-structure-cache entries referring to that freed page.
> 
> Ah yes.  Better not do that.
> 
> As some future cleanup it would be nice to be able to do this without
> specifying a function, but that is a minor detail and not relevant to
> this patch.

This is exactly the conclusion I came to when reviewing v3 of this
series.  Actually calling __ept_sync_domain() is unnecessary, but at
that point you might as well.

 -George