Bug 12274 - pixmap cache management problem on server startup/reset
Summary: pixmap cache management problem on server startup/reset
Status: RESOLVED INVALID
Alias: None
Product: xorg
Classification: Unclassified
Component: Driver/Radeon (show other bugs)
Version: 7.1 (2006.05)
Hardware: PowerPC Linux (All)
: medium normal
Assignee: xf86-video-ati maintainers
QA Contact: Xorg Project Team
URL:
Whiteboard:
Keywords: NEEDINFO
Depends on:
Blocks:
 
Reported: 2007-09-04 00:00 UTC by Branden Robinson
Modified: 2008-12-03 00:14 UTC (History)
0 users

See Also:
i915 platform:
i915 features:


Attachments
xorg.conf (1.75 KB, application/octet-stream)
2007-09-04 11:26 UTC, Branden Robinson
no flags Details
Xorg.0.log (70.42 KB, text/plain)
2007-09-04 11:31 UTC, Branden Robinson
no flags Details
xorg.conf (1.75 KB, text/plain)
2007-09-04 11:41 UTC, Branden Robinson
no flags Details

Description Branden Robinson 2007-09-04 00:00:48 UTC
Scenario: PowerPC architecture, running Debian lenny

Note that my architecture is big-endian; this might matter.

Video Card:
0000:00:10.0 0300: 1002:4966 (rev 01)
0000:00:10.0 VGA compatible controller: ATI Technologies Inc Radeon RV250 If [Radeon 9000] (rev 01)

Monitor: Samsung SyncMaster 213T connected via DVI port

I'm running a bleeding-edge ATI driver.

The problem is that when kdm draws its greeter screen on X server startup
or reset, it seems to be pulling from stale pixmap cache memory.

The following URLs should illustrate this.  The first is kdm coming up
after an X server start; the second is after a reset.  As you may be able
to guess, there is content from windows unmapped at the end of the previous
X session, and image data that the web browser had used recently.

http://redwald.deadbeast.net/tmp/branden_grief_5.jpeg
http://redwald.deadbeast.net/tmp/branden_grief_6.jpeg

Note that there is also some interference in the video signal that
manifests as pinkness and moire-pattern-like banding.  That's a separate
issue and Alex Deucher is working with me on it.
Comment 1 Michel Dänzer 2007-09-04 04:35:09 UTC
Please attach the full log file, preferably the config file as well.

Does it also happen with

        Option "RenderAccel" "off"

or

        Option "AccelMethod" "EXA"

?

P.S. in the future please attach files here instead of referencing them, so they can't get out of sync with the bug report.
Comment 2 Branden Robinson 2007-09-04 11:26:46 UTC
Created attachment 11415 [details]
xorg.conf
Comment 3 Branden Robinson 2007-09-04 11:31:54 UTC
Created attachment 11416 [details]
Xorg.0.log
Comment 4 Branden Robinson 2007-09-04 11:41:44 UTC
Created attachment 11418 [details]
xorg.conf
Comment 5 Branden Robinson 2007-09-04 12:34:22 UTC
Michel,

The image files are too large to attach.

The only reason I didn't attach my config and log files when submitting the
bug is that I couldn't find where to do so.  It's obvious *after* the bug
submitted, though.
Comment 6 Branden Robinson 2007-09-06 23:33:18 UTC
Hi Michel,

It appears to be a problem specifically with XAA acceleration.

Again, I'm linking to the images because they're 1.3 to 1.9MB each, and
fd.o Bugzilla doesn't like that for attachments.

I've also tried to give them more useful filenames.

First, the status quo.  This is a server regen after shutting down a
several days' long X session.  As you can see, the video memory being used
is chock full of stale pixmaps.

http://redwald.deadbeast.net/tmp/freedesktop.org_bug_12274_server_regen_with_default_RenderAccel_and_AccelMethod.jpeg

Next, I tried:
	Option	"RenderAccel"	"false"

While the result looks different, I think the root window was just painted
from a different wrong location in video memory.  It looks like we started
to traverse into the region we actually wanted at 0,0 partway through the
framebuffer.

http://redwald.deadbeast.net/tmp/freedesktop.org_bug_12274_server_restart_with_RenderAccel_off_and_default_AccelMethod.jpeg

Next, I tried:
	Option	"AccelMethod"	"EXA"

Bang.  Appears to fix the problem.

http://redwald.deadbeast.net/tmp/freedesktop.org_bug_12274_server_restart_with_default_RenderAccel_and_EXA_AccelMethod.jpeg

To ensure that it wasn't a fluke, I went ahead and started a session, fired
up IceWeasel, loaded up a few image-heavy webpages, and quit the session.

Still looks good.

http://redwald.deadbeast.net/tmp/freedesktop.org_bug_12274_server_regen_with_default_RenderAccel_and_EXA_AccelMethod.jpeg

You know better than I, but this smells like an XAA problem to me.

Please let me know if you need further information.
Comment 7 Michel Dänzer 2007-09-07 01:01:22 UTC
(In reply to comment #6)
> 
> Again, I'm linking to the images because they're 1.3 to 1.9MB each, and
> fd.o Bugzilla doesn't like that for attachments.

Okay, though it should be pretty easy to make them smaller using something like the GIMP.


> You know better than I, but this smells like an XAA problem to me.

Yeah, the question is whether it's in XAA itself or the radeon driver's XAA code. Does limiting the maximum virtual size with something like

        Virtual 1600 1200

make any difference?
Comment 8 Benjamin Close 2008-01-11 02:39:14 UTC
Bugzilla Upgrade Mass Bug Change

NEEDSINFO state was removed in Bugzilla 3.x, reopening any bugs previously listed as NEEDSINFO.

  - benjsc
    fd.o Wrangler
Comment 9 Alex Deucher 2008-12-03 00:14:55 UTC
closing due to lack of feedback.


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.