[Intel-gfx] [PATCH 1/6] RFCish: write only mappings (aka non-blocking)

Daniel Vetter daniel at ffwll.ch
Wed Sep 21 09:02:24 CEST 2011


On Tue, Sep 20, 2011 at 10:16:39PM +0100, Chris Wilson wrote:
> On Tue, 20 Sep 2011 13:06:43 +0200, Daniel Vetter <daniel at ffwll.ch> wrote:
> >   Now non-blocking cpu mmaps make very much sense on llc/snooped buffer
> >   objects. So I think we actually need an ioctl to get obj->cache_level so
> >   userspace can decide whether it should use non-blocking gtt mmaps or cpu
> >   (non-blocking) cpu mmaps. We might as well go full-circle, make Chris
> >   happy and merge the corresponding set_cache_level ioclt to enable
> >   snooped buffers on machines with ilk-like coherency (i.e. that atom
> >   thing I'm hearing about ...). But imo that's material for non-blocking
> >   mmaps, step 2.
> 
> That's not enough to make me completely happy. I also have the use-case
> of wanting to efficiently handle compositing with client allocated
> memory i.e. persistent ShmPixmaps, where coherency either needs to be
> transparent to the client or (worse) synchronicity imposed by the server.

I know, just snoopable bo support only makes you half-happy and you need
fancy stuff in addition ;-) But if your sna branch can halfway easily
profit from simle snoopable bo support on !llc machines, I think that'll
give us an easy way to check api sanity for the new ioctl.
-Daniel
-- 
Daniel Vetter
Mail: daniel at ffwll.ch
Mobile: +41 (0)79 365 57 48



More information about the Intel-gfx mailing list