[Intel-gfx] [RFC PATCH 00/11] i915 HW context support
eugeni at dodonov.net
Tue Feb 14 15:41:57 PST 2012
On Tue, Feb 14, 2012 at 21:17, Adam Jackson <ajax at redhat.com> wrote:
> On 2/14/12 4:09 PM, Ben Widawsky wrote:
> The IOCTLs defined here may also be used with ppgtt (real ppgtt) but the
>> interface may be insufficient for that. The contexts do provide a
> A... what?
> Series looks reasonable to me. The one thing that makes me nervous is
> 4/11 mentioning the BIOS, which honestly seems like an insane thing to let
> your BIOS have _any_ say in, but then Intel sure does like to give the BIOS
> a chance to make the OS less functional. The code in 7/11 effectively
> limits things to between 4k and 1M; does that actually make sense? I can't
> imagine wanting to allocate only 4k if it turns out we'd always need at
> least (say) 32k.
> I have admittedly not read any of the docs on how this is meant to work,
> apologies if that'd answer this.
The high-level overview of this feature is that it adds support for having
a per-context set of GPU items the processes have access to. This is mostly
important for GL_ARB_Robustness  and GL_EXT_Transform_feedback .
So what it does is adding a possibility of having a different context id
for each set of operations. This way, what processes in one context_id do
should not affect the processes in another set. For GL_ARB_Robustness, it
could prevent one WebGL applications from taking down all the other users
of the GPU for example; and for GL_EXT_Transform_feedback it would allow
each process to have a way to store its own feedback data for different
stages of the pipeline. And so on.
As far as I know, it is also needed for full PPGTT support; but I don't
know this part good enough to give more in-depth answer (Daniel?)
(I hope it clarified the things a bit :))
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Intel-gfx