[Intel-gfx] Huge memory leaks with GEM
Roger WANG
roger.wang at intel.com
Tue Jul 7 17:02:27 CEST 2009
On Tue, 2009-07-07 at 04:20 -0700 Johannes Engel wrote:
> Quoting Maxim Levitsky, On 07/06/2009 09:46 PM:
> > I have recently tried the latest versions of graphics stack for my G965
> > intel device.
> >
> >
> > First I must say thanks! The quality of everything did improve
> > significantly.
> >
> > I especially happy to see that GoogleEarth is finally usable. I remember
> > it being very slow back when I tried the intel drivers last time (I have
> > a nvidia card that I use most of the time on this system, and it is set
> > to hit the trash as soon as intel drivers became usable - I know that
> > performance will always be inferior, but as long as compiz, GE, and few
> > simple games work, I agree to that)
> >
> > However, it appears that current GEM implementation doesn't free buffers
> > at all.
> >
> > looking at /sys/kernel/debug/dri/0/gem_objects reveals that more a GB
> > of buffers are allocated, after few minutes of GE use. In same time
> > usual signs of swapping appear. top confirms that GE uses more that a GB
> > of memory (this system has just 1GB of memory)
> >
> > This is what I see there:
> >
> > 4614 objects
> > 1152802816 object bytes
> > 3 pinned
> > 13762560 pin bytes
> > 137949184 gtt bytes
> > 260308992 gtt total
> I can confirm the same behaviour using 945GM with KDE4 + compositing.
> Especially running a Skype video session using two XVideo ports makes
> the objects increase by approx. 9~10 per sec.
> Having turned of the swap partition the whole system gets eaten up by
> extensive hard disk activity after a while, then Xorg restarts and the
> game starts again.
I got the similar issue here and Jesse provided information[1] that
it's been fixed. I'll try it soon.
HTH
--Roger
[1] http://lkml.org/lkml/2009/7/6/238
>
> Cheers, Johannes
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx
More information about the Intel-gfx
mailing list