[Intel-gfx] Updated patch series for removing EXA
stavallo at unina.it
Wed Feb 25 19:22:04 CET 2009
On Wednesday 25 February 2009 19:08:55 Eric Anholt wrote:
> On Wed, 2009-02-25 at 10:00 +0100, Johannes Engel wrote:
> > Eric Anholt wrote:
> > > Here's another cut at removing EXA.
> > > I've fixed up UXA to run on server 1.4 (though the lack of glyph cache
> > > hurts pretty badly), and fixed a bunch of failure that I noticed with
> > > UXA and !DRI2, and a bit of failure with UXA and !GEM. The !GEM case
> > > is still incredibly painful, due to too much moving of object data in
> > > and out. We could probably fix up the fake bufmgr to handle this
> > > better with a little effort.
> > >
> > > If people like the looks of this, I'm thinking of pushing all but the
> > > last patch pretty soon.
> > Hi Eric,
> > are you aware of the memory leak using UXA on at least some 945GM
> > configurations? This makes UXA with Composite like kwin4 pretty much
> > unusable at the moment...
> > This happens for me also with supposedly GEM enabled kernels
> > (2.6.29-rc*), so according to what you say seems not to be related to
> > the fake bufmgr.
> The 2.6.2 release I just put out included a fix for a major memory
> "leak" (actually not a leak, just choosing a massive working set). We
> need information on how to reproduce any further failure if you have a
> way to.
As I wrote in a post to xorg@, the current git code of everything solved most
of the problems with memory usage increase of Xorg. However, I just found a
way to reproducibly increase the memory usage of Xorg (as reported by top): VT
switching. The sequence Ctrl+Alt+F1 and Alt+F7 brought, for instance, the
memory usage from 4.7% to 6.9%. I am using DRI2 on GM965.
More information about the Intel-gfx