Hi, Just wanted to quickly report another issue with EXA acceleration and the radeon driver. JProfiler is next to unusable because of massive font corruption when scrolling. * Compiz or not does not matter. * Disabling EXA-acceleration fixes the problem. * ATI X1400 radeon mobile GPU. * radeon driver 6.9.0+git20081003.f9826a56-0ubuntu2.1 Please see the attached screenshot. lspci here (from other bug report): http://launchpadlibrarian.net/19073276/lspci-vvnn.txt Xorg log here (from other bug report): https://bugs.freedesktop.org/attachment.cgi?id=20090 Regards, Øyvind
Created attachment 20399 [details] Screenshot showing the corruption
What sets JProfiler apart from most other apps I use, is that it's a Swing-based Java GUI-application. Maybe they do font rendering slightly differently, which triggers the problems.
The Java's X11 pipeline does font-rendering in software, so I guess its not something font-related at all. However they are using ShmPut/Get a lot for such operations, maybe there is a race somewhere. By the way I've written an RENDER based backend for Java: http://78.31.67.79:8080/jxrender/ This backend uses hw-accaleration almost everywhere, would be great if you could give it a try :)
Does the problem also persist if you start it with -Dsun.java2d.pmoffscreen=false ?
Have you tried disabling the usual suspects like AccelDFS or RenderAccel to see if that works around the problem?
(In reply to comment #4) > Does the problem also persist if you start it with > -Dsun.java2d.pmoffscreen=false ? > This fixes the problem completely.
(In reply to comment #5) > Have you tried disabling the usual suspects like AccelDFS or RenderAccel to see > if that works around the problem? > Will do ..
(In reply to comment #7) > (In reply to comment #5) > > Have you tried disabling the usual suspects like AccelDFS or RenderAccel to see > > if that works around the problem? > > > > Will do .. > Hi again, Some more data points: AccelDFS==false => does not help, corruption the same. RenderAccel==false => does not help, corruption the same. AccelDFS==false && RenderAccel==false => does not help, corruption the same. When using old XAA, there is no corruption. Disabling use of offscreen-pixmaps in Java2D solves the problem (like I've stated in earlier comment). But this workaround most be done on a per-application basis. Regards, Øyvind
I've donated a fix last week which automatically disables pmoffscreen if SHMPixmaps are not available - it will be integrated into JDK6u12 (next update). This should also improve performance when EXA drivers are used. After all, there is still a problem somewhere in the driver, even if is hidden by the fix.
(In reply to comment #9) > I've donated a fix last week which automatically disables pmoffscreen if > SHMPixmaps are not available - it will be integrated into JDK6u12 (next > update). > This should also improve performance when EXA drivers are used. > > After all, there is still a problem somewhere in the driver, even if is hidden > by the fix. > That sounds nice :) You are right, there must be a problem in the driver somewhere as well, since when using old XAA, there is no corruption at all.
(In reply to comment #9) > I've donated a fix last week which automatically disables pmoffscreen if > SHMPixmaps are not available - it will be integrated into JDK6u12 (next > update). Is it possible that pmoffscreen is actually doing something wrong when SHMPixmaps aren't available? Is there a testcase simpler than JProfiler, or at least freely available?
> Is it possible that pmoffscreen is actually doing something wrong when > SHMPixmaps aren't available? I don't think so. However when SHMPixmaps aren't available the code agressivly does SHMGet/ShmPut. > Is there a testcase simpler than JProfiler, or at least freely available? In general it should happen with any Swing application, sadly I don't have an ATI card. Øyvind: Could you try Jedit to see wether the problem persists also there?
(In reply to comment #12) > > Is it possible that pmoffscreen is actually doing something wrong when > > SHMPixmaps aren't available? > I don't think so. However when SHMPixmaps aren't available the code agressivly > does SHMGet/ShmPut. > > > Is there a testcase simpler than JProfiler, or at least freely available? > In general it should happen with any Swing application, sadly I don't have an > ATI card. > Øyvind: Could you try Jedit to see wether the problem persists also there? > Tested two more Java Swing apps with radeon+EXA: * JEdit is OK, no problems. * OxygenXML is OK, no problems. Might be JProfiler that's doing something unusual which triggers problems. I should also note, JProfiler is the only app that defaults to native look-and-feel, the others use the classic Swing theme it seems. Btw. when starting up Opera, I get this message: opera: X Shared memory extension is not available. ZPixmap not supported So, shared memory pixmaps are disabled, just like recent nvidia drivers do by default ?
This problem does not seem to occur anymore with JProfiler5 + latest Java 6 update 11. :)
(In reply to comment #14) > This problem does not seem to occur anymore with JProfiler5 + latest Java 6 > update 11. :) > And incidentally, that might also be the reason why there is no corruption in OxygenXML or Jedit, either.
7.4.99+git20140806.fbf575cb-0ubuntu0sarvatt3less /u
ubuntu-gnome 14.10 with : 7.4.99+git20140806.fbf575cb-0ubuntu0sarvatt3 The corruption is still there, rendering matlab editor (java based) useless, unless -Dsun.java2d.pmoffscreen=false is used.
-- GitLab Migration Automatic Message -- This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity. You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/xorg/driver/xf86-video-ati/issues/4.
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.