Re: [PATCH] VMX: Corrections to XSA-60 fix

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 07/11/13 11:09, Jan Beulich wrote:
>>>> On 07.11.13 at 11:56, Andrew Cooper <andrew.cooper3@citrix.com> wrote:
>> This patch is somewhat RFC, partly because I have only compile-tested it, and
>> because of a few further questions.
>>
>> 1) Are there any other places other than unmap_domain_page() we might want to
>>    put hooks? Any well behaved access should use {map,unmap}_domain_page() as
>>    far as I am aware.
> Any writes done through map_domain_page_global() mappings.

Ooh yes.  I will hook unmap_domain_page_global() as well

>
>> 2) What is expected to happen with Qemu mappings from dom0?  As far as I am
>>    aware, they can't retoactivly have their pagetable caching type changed.
> Right, so the flag should also get set when completing ioreq-s.
>
> Jan
>

Ok, but that still doesn't help coherency of emulated device state which
changes without an ioreq.  Qemu scraping the framebuffer for VNC comes
to mind.

How bad would it be to have clean cacheline in RAM because of Qemu,
which is accessed as UC by the domain?  I suspect the answer is "there
be dragons", and thus best avoided.

~Andrew