Bug 807 - Minor display corruption with XAA offscreen pixmaps for mach64
Summary: Minor display corruption with XAA offscreen pixmaps for mach64
Status: RESOLVED FIXED
Alias: None
Product: xorg
Classification: Unclassified
Component: Driver/mach64 (show other bugs)
Version: 6.7.0
Hardware: x86 (IA32) Linux (All)
: high normal
Assignee: Xorg Project Team
QA Contact:
URL:
Whiteboard:
Keywords: patch
Depends on:
Blocks: xorg-7.2
  Show dependency treegraph
 
Reported: 2004-06-28 17:21 UTC by Sam Varshavchik
Modified: 2006-08-04 16:15 UTC (History)
5 users (show)

See Also:
i915 platform:
i915 features:


Attachments
Minor display corruption in Mozilla's URL text box. (33.38 KB, image/png)
2004-06-28 17:22 UTC, Sam Varshavchik
no flags Details
/var/log/XFree86.0.log (41.00 KB, text/plain)
2004-06-28 17:23 UTC, Sam Varshavchik
no flags Details
/var/log/Xorg.0.log (74.07 KB, text/plain)
2004-06-28 17:26 UTC, Sam Varshavchik
no flags Details
/etc/X11/xorg.conf (2.69 KB, text/plain)
2004-06-28 17:27 UTC, Sam Varshavchik
no flags Details
Output of lspci (11.24 KB, text/plain)
2004-06-28 17:29 UTC, Sam Varshavchik
no flags Details
Updated patch (7.70 KB, patch)
2006-04-04 08:45 UTC, Ivor Hewitt
no flags Details | Splinter Review

Description Sam Varshavchik 2004-06-28 17:21:46 UTC
Hello, I can reliably reproduce some minor display corruption on Fedora Core 2
using Mozilla 1.7.  Continuously typing and erasing text in the URL text box,
causing the URL history dropdown to pop up and down repeatedly, eventually
results in some minor display corruption in the URL text box.  I will attach a
short screen dump. The corruption appears to be minor, and it always gets erased
by the next keypress.

I believe that the corruption is caused by the driver/server, and not the
application, because if I set DISPLAY to a remote display, no corruption is
observed in the same situation.

Note: this is xorg 6.7.0 compiled for the x86_64 platform.  SMP kernel.
Comment 1 Sam Varshavchik 2004-06-28 17:22:58 UTC
Created attachment 423 [details]
Minor display corruption in Mozilla's URL text box.
Comment 2 Sam Varshavchik 2004-06-28 17:23:33 UTC
Created attachment 424 [details]
/var/log/XFree86.0.log
Comment 3 Sam Varshavchik 2004-06-28 17:26:12 UTC
Created attachment 425 [details]
/var/log/Xorg.0.log

Void previous attachment.  Correct log file attached.
Comment 4 Sam Varshavchik 2004-06-28 17:27:28 UTC
Created attachment 426 [details]
/etc/X11/xorg.conf
Comment 5 Sam Varshavchik 2004-06-28 17:29:08 UTC
Created attachment 427 [details]
Output of lspci
Comment 6 Adam Jackson 2004-07-06 15:37:18 UTC
does disabling XAA help?  if so, could you test the various "XaaNo..." options
(described in the xorg.conf manpage) to see which part of XAA is to blame?
Comment 7 Sam Varshavchik 2004-07-06 16:16:59 UTC
Option "XaaNoOffscreenPixmaps"

fixes the corruption :-)
Comment 8 Adam Jackson 2004-07-06 19:04:37 UTC
updating summary
Comment 9 Ivor Hewitt 2005-08-24 16:48:41 UTC
Just another datapoint, I see exactly the same corruption with a mach64 card.    
The problem shows itself when scaling a 1 pixel wide pixmap onto the screen.    
The effect is visually identical to that shown in this report although it   
persists when offscreen pixmaps are disabled too.  
  
Disabling ScreenToScreen copy solves the problem.   
   
Adding:   
  ATIMach64WaitForIdle(pATI);   
to the end of SubsequentScreenToScreenCopy also makes the problem disappear  
completely for me.  
 
Any ideas? 
Comment 10 Adam Jackson 2005-10-04 13:22:35 UTC
additional sync would be an acceptable workaround, but i'll see if i can do
better.  this should get fixed in 6.9.
Comment 11 Erik Andren 2006-04-04 05:31:17 UTC
Whats the status of this bug using a current version of xorg?
Comment 12 Sam Varshavchik 2006-04-04 08:29:16 UTC
I'm still on 6.8.2, but I don't seem to trigger this bug in 6.8.2.

I don't have any info on comment #9 -- I never experienced that variant of this bug.

Comment 13 Ivor Hewitt 2006-04-04 08:45:58 UTC
Created attachment 5190 [details] [review]
Updated patch

This patch provided by Marc Aurele La France completely resolved the problem
for me.
Comment 14 George - 2006-05-09 02:43:56 UTC
(In reply to comment #13)
> Created an attachment (id=5190) [edit]
> Updated patch
> 
> This patch provided by Marc Aurele La France completely resolved the problem
> for me.

The attached patch consists of two parts:
* read-back cache invalidation
* copy throttling

Both parts have been merged in the following repository:
git//git.freedesktop.org/~libv/xf86-video-mach64. xf86-video-mach64 is still in
development but we intend to to have it distribution ready in time for 7.2.

The bug with scaling 1x1 pixmaps affects RENDER on mach64 EXA. My guess is that
RENDER commonly uses 1x1R pixmaps, which often get reduced to solid operations.
When the destination pixmap is read back by the CPU, corruption appears.

With the "cache invalidation" fix, mach64 EXA passes rendercheck.

The "copy throttling" fix has been applied but should be reverted if a better
fix is found.

Thanks,
george.
Comment 15 Adam Jackson 2006-05-16 23:24:31 UTC
(In reply to comment #14)
> Both parts have been merged in the following repository:
> git//git.freedesktop.org/~libv/xf86-video-mach64. xf86-video-mach64 is still in
> development but we intend to to have it distribution ready in time for 7.2.

This looks pretty good.  Feel free to do an async mach64 release between now and
7.2.
Comment 16 Erik Andren 2006-06-05 12:26:28 UTC
Ok to close this bug?
Comment 17 George - 2006-06-05 13:48:37 UTC
(In reply to comment #16)
> Ok to close this bug?

Personally, I think it's better to close this bug when
git//git.freedesktop.org/~libv/xf86-video-mach64 makes its first release.
Comment 18 George - 2006-08-04 16:15:48 UTC
Commited to HEAD. Thanks!


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.