Bug 4177 - composite + radeon + glxinfo = 1h 90%CPU
Summary: composite + radeon + glxinfo = 1h 90%CPU
Status: RESOLVED INVALID
Alias: None
Product: Mesa
Classification: Unclassified
Component: Demos (show other bugs)
Version: unspecified
Hardware: x86 (IA32) Linux (All)
: high normal
Assignee: Xorg Project Team
QA Contact:
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2005-08-21 23:05 UTC by Dirk Salewski
Modified: 2011-10-15 16:04 UTC (History)
1 user (show)

See Also:
i915 platform:
i915 features:


Attachments
xorg.conf (4.69 KB, text/plain)
2005-08-21 23:07 UTC, Dirk Salewski
Details
Xorg.0.log (working, without Composite) (45.04 KB, text/plain)
2005-08-21 23:08 UTC, Dirk Salewski
Details
Xorg.0.log (with Composite, before glxinfo) (44.71 KB, text/plain)
2005-08-21 23:17 UTC, Dirk Salewski
Details
Xorg.0.log (with Composite, after glxinfo, CPU=100%) (44.71 KB, text/plain)
2005-08-21 23:19 UTC, Dirk Salewski
Details
Output of glxinfo with Composite enabled (4.45 KB, text/plain)
2005-08-21 23:19 UTC, Dirk Salewski
Details
Output of glxinfo with Composite disabled (4.38 KB, text/plain)
2005-08-21 23:21 UTC, Dirk Salewski
Details

Description Dirk Salewski 2005-08-21 23:05:50 UTC
After starting X with Composite extension enabled everything looks fine (very
fine, due to transparency). However, when running "glxinfo" CPU-usage jumps up
to 100% for some seconds, then I can see the glxinfo-info and after that
CPU-usage remains high (70-100%) for one hour or more (depends on something I
wasn't able to find out yet). When disabling the Composite extension, everything
is fine, running glxinfo or not.

My system:
AMD Athlon64 (Gentoo Linux, 64bit-Version)
Sapphire ATI Radeon 9200
radeon-driver from kernel 2.6.9

Programs running: 
xfce4 (the whole desktop)
gkrellm2
psi
root-tail
terminal (xfce4-terminal)

Somebody mentioned a similar problem when running gkrellm2, but stopping it
didn't have the desired effect on my machine.

Attached You will find my xorg.conf and Xorg.0.log (with and without Composite
enabled).

Greetings,

Dirk
Comment 1 Dirk Salewski 2005-08-21 23:07:04 UTC
Created attachment 2976 [details]
xorg.conf
Comment 2 Dirk Salewski 2005-08-21 23:08:07 UTC
Created attachment 2977 [details]
Xorg.0.log (working, without Composite)
Comment 3 Dirk Salewski 2005-08-21 23:15:05 UTC
What I forgot to mention: glxinfo is just a bulletproof way to trigger the bug -
You might think "just don't use it". Unfortunately there are other programs that
do it, too. I wasn't able to compile a list, though, because there's no user
interaction required. Sometimes I come back to my desk and CPU=80%. So I can't
just stop using glxinfo and expect everything to be alright.
Comment 4 Dirk Salewski 2005-08-21 23:17:59 UTC
Created attachment 2978 [details]
Xorg.0.log (with Composite, before glxinfo)
Comment 5 Dirk Salewski 2005-08-21 23:19:05 UTC
Created attachment 2979 [details]
Xorg.0.log (with Composite, after glxinfo, CPU=100%)
Comment 6 Dirk Salewski 2005-08-21 23:19:59 UTC
Created attachment 2980 [details]
Output of glxinfo with Composite enabled
Comment 7 Dirk Salewski 2005-08-21 23:21:55 UTC
Created attachment 2981 [details]
Output of glxinfo with Composite disabled
Comment 8 Dirk Salewski 2005-08-30 07:31:51 UTC
I tested this with external radeon modules (not in-kernel). Same effect. Very
strange, though: When CPU is near 100% after glxinfo it drops to 3%-10% when
running glxgears (can be invisible, hidden, on other workspace). So I'm now able
to work when I run glxgears...
Comment 9 Octoploid 2005-12-18 20:06:04 UTC
Just to chime in. This bug is still present in X11R6.9-RC4.
As soon as the dri driver gets loaded (triggered by running glxinfo),
the whole system becomes extremly sluggish and virtually unuseable.
I'm using the latest cvs radeon_dri.so driver (Radeon 7500 in my case).

Comment 10 Michel Dänzer 2006-01-06 02:22:30 UTC
Does this also happen if you don't enable backing store?
Comment 11 Dirk Salewski 2006-01-09 07:55:28 UTC
(In reply to comment #10)
> Does this also happen if you don't enable backing store?

Short answer, probably misleading: No!

Long answer: I got so frustrated with all this eyecandystuff that I started to
use icewm again - icewm does not have this sort of problem (the problems occured
in xfce4 with transparency, window-shadows and all possible wizardry enabled).
Now, when I read Your post I tried it all again - and it seems that I cannot
trigger the bug anymore (both with and without backingstore).

There's one thing to note, however: I run "conky" and "root-tail", both of which
draw directly to the root window. Conky is put into "single buffer" mode,
because otherwise it makes root-tail invisible. When I ENABLE backingstore and
issue "glxinfo" conky starts to update itself r ea ll y  s l o oo o w. So I
guess it has to do something with backingstore and improved a bit since my last
post. 

Greetings,

Dirk
Comment 12 Erik Andren 2006-04-22 23:58:04 UTC
Dirk, are you willing to go back to XFCE to do some more debugging or should we
close this bug?
Comment 13 Dirk Salewski 2006-04-23 22:57:29 UTC
Hey Erik,

I'd prefer to stay with icewm for now. I have to get some serious work done, and
I simply don't have the time for debugging eyecandy at the moment. 

(Sometimes I think by myself: "Hey, isn't all of this going into the wrong
direction, and shouldn't everybody be using wmii by now?")

If I ever get into the lucky situation to have the time to tinker with these
configfiles again, I will reopen the bug (if it persists).

Greetings,

Dirk
Comment 14 Erik Andren 2006-04-23 23:15:45 UTC
Closing. 


Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.