Does Xorg server pin composition buffer's MFNs?

[Dave Airlie]

On Fri, Apr 16, 2010 at 6:47 PM, Rafal Wojtczuk
<rafal at invisiblethingslab.com> wrote:
> On Wed, Apr 14, 2010 at 03:59:39PM -0400, Adam Jackson wrote:
>> On Wed, 2010-04-14 at 20:07 +0200, Rafal Wojtczuk wrote:
>> > Hello,
>> > A bit of background: apparently under Linux, the mfn (say, physical address)
>> > of the frame, that holds an usermode page, can change, unless precautions
>> > are made. See http://lists.xensource.com/archives/html/xen-devel/2010-04/msg00514.html
>> > why it might be an issue in certain virtualized environments.
>> >
>> > Question 1: does X server (running under Linux), allocate the composition
>> > buffers in an usual way (using malloc() or similar), or does it do anything
>> > fancier in order to make sure the physical address of composition buffers is
>> > stable in time ?
>> > Question 2: Does ever X server schedule a DMA transaction (to graphics card
>> > presumably) from a composition buffer ? (in such case it would make sense to
>> > pin its physical pages).
>>
>> I'm assuming by "composition buffer" you mean the thing you're actually
>> scanning out on the display.
> No. I mean the per-window offscreen storage mechanism, activated by
> XCompositeRedirectSubwindows() function. Referred to as "windows backing
> pixmap" in http://ktown.kde.org/~fredrik/composite_howto.html. Apologies if
> I did not make it clear enough.
>
>>  I'll use the word "framebuffer" for that, since that's the usual X name for it.
>> In the absence of an accelerated driver, X doesn't care about the
>> ...
> Unfortunately, that is not the piece of information I would like to know. In
> fact if CompositeRedirectManual flag is specified, the actual framebuffer is
> not touched by xlib calls at all.
>
>> It's not really clear what you're asking though, or what you're trying
>> to accomplish.
> Briefly, the goal is to get the location of a composition buffer created by
> X server running in virtual machine A, and map it in the address space of
> virtual machine B. Such mapping has to be defined in terms of physical
> addresses; consequently, it is crucial to make sure that the frames backing a
> composition buffer do not change in time.
>
> So, once again: is it true that X server allocates a composition buffer (a
> window backing pixmap) by a normal malloc call, and does no additional
> effort to pin its physical pages so that they do not change in time ?
>

I could be anywhere, X doesn't care. The X acceleration architecture used for
instance EXA, can migrate the pixmap to from the driver from malloced memory
and the driver can migrate the pixmap to/from VRAM.

X has not access to the physical pages as its a userspace process.

So its a driver question really not a generic X server question.

Dave.