We (as in Debian) have been receiving lots of bug reports about display corruption with EXA, especially since Intel 2.1.99 arrived with EXA enabled by default, but also before. I thought it was related to #11402 but this one is marked as "worksforme" and our problems are more than font antialiasing, people at least have icon rendering problems too. The symptoms also seem similar to #13367 but nobody talked about suspend/resume in our reports. I am opening this new bug instead of messing the other ones in case they are unrelated in the end, feel free to merge... It started 3 months ago with Jos van Wolput with driver 2.1.1 and Xserver 1.3. He said "The background of gdm becomes blurred, the fonts unreadable. The icons and the fonts of my xfce desktop almost disappear making it unusable." and "After booting and login the screen always looks corrupted but sometimes it becomes clear after restarting X. This behavior seems unpredictable. I also started glxgears in a corrupted screen which suddenly became clear, first only the terminal running glxgears and then the whole screen! I compared the "good" and the "bad" logs. The only differences are found in some memory addresses." Mark Broadbent confirmed some font rendering problems with 2.1.99 and Xserver 1.4: "Font corruption occurs in gdm and on the Gnome desktop when the experimental intel driver is installed." Details available at http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=439210 Andrew Dingman reported for 2.2+1.4 on 82G965: "All fonts were completely illegible (black boxes) unless I set Gnome's font rendering preferences to 'monochrome'. At that point, *most* fonts showed up, but not all, including none in the gnome terminal. Xterm worked just fine. However, in addition to the font rendering problems, most icons were either rendered impropperly or surrounded by black rectangles. Text consoles failed to display at all, even with GDM stopped." http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=451791 Philip Frei with 2.2+1.4 on 82852/855GM: "Since I installed the new driver my xserver doesn't show any fonts. Not in input boxes, not in menus. I can type something (in GDM or in iceweasel) but the letters aren't there." Confirmed by Dann Frazier on GM965/GL960. Details on http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=451893 I can ask the reporters to post configs/logs here if the one from the URL above don't help. If you need screenshots, feel free to ask.
The mentioned display corruption with EXA disappears completely after running glxdemo or glxgears and restarting X, as long as the pc isn't shut down. Jos van Wolput
Does EXA ever work correctly on these machines? I mean does it work before or with current upstream master? It looks people are having problem with 855GM and 965 desktop and laptop, which all work fine here, although I use master branches.
How about try with Option "Tiling" "FALSE"?
(In reply to comment #4) > How about try with Option "Tiling" "FALSE"? > Doesn't make any difference!
(In reply to comment #3) > Does EXA ever work correctly on these machines? I mean does it work before or > with current upstream master? > > It looks people are having problem with 855GM and 965 desktop and laptop, which > all work fine here, although I use master branches. > EXA never worked correctly since I first tried it with Debian's v. 2.0.0 of xserver-xorg-video-intel half a year ago. It only works after running something like glxdemo or another app from /usr/share/mesademos/progs/xdemos. I am using Debian (unstable) on a 855GM laptop.
(In reply to comment #6) > EXA never worked correctly since I first tried it with Debian's v. 2.0.0 of > xserver-xorg-video-intel half a year ago. It only works after running > something like glxdemo or another app from /usr/share/mesademos/progs/xdemos. > I am using Debian (unstable) on a 855GM laptop. > What's your 855gm model? I just can't reproduce this on hp nx5000 as noted on bug 11402. So is it issue with current exa rendering state init on 855gm that may be fixed after some mesa apps run? Could you first try Option "EXANoComposite" "TRUE" to see if thing goes fine?
(In reply to comment #7) > > What's your 855gm model? > It 's a Jewel Topaz 9600 (dutch brand), made in China by Uniwill, type 223ii0 > > Could you first try Option "EXANoComposite" "TRUE" to see if thing goes fine? > No more problems with Option "EXANoComposite" "TRUE", only RENDER acceleration disabled.
I took a few screenshots of my system, available here: http://www.dannf.org/bugs/451791/ The first shows a gdm login screen where I've typed a few characters. The cursor moves forward, but I see no text appear. The next two show the screen right after login. The corruption in the center of the screen is where the GNOME splash screen should be. Also, the top panel normally contains the normal GNOME menu pulldowns (Applications, Places, System) and a few other icons, but with this issue I only see the battery charge monitor. I can click where the other components *should* be, and they still respond - though not correctly (no text in the menus, etc). I'm successfully working around the issue by using: Option "AccelMethod" "XAA"
(In reply to comment #8) > No more problems with Option "EXANoComposite" "TRUE", only RENDER acceleration > disabled. No, render should not be disabled, just composite operation fall back. You mean if you add EXANoComposite in xorg.conf, then there's no more rendering failure you've seen, right? So there must be some place which doesn't get initialized correctly..
(In reply to comment #9) > I took a few screenshots of my system, available here: > http://www.dannf.org/bugs/451791/ What's your chipset please? > I'm successfully working around the issue by using: > Option "AccelMethod" "XAA" > how about "EXANoComposite" "TRUE"?
(In reply to comment #10) > No, render should not be disabled, just composite operation fall back. render is not disabled, only render acceleration. From Xorg.0.log: (**) Extension "RENDER" is enabled (**) intel(0): Option "EXANoComposite" "TRUE" (II) intel(0): EXA: Disabling Composite operation (RENDER acceleration) > You mean if you add EXANoComposite in xorg.conf, then there's no more rendering > failure you've seen, right? So there must be some place which doesn't get > initialized correctly.. Correct, no more rendering failure after adding Option "EXANoComposite" "TRUE"
Created attachment 12922 [details] [review] try to emit invarient state every time Could you try this patch with your origin xorg.conf? I suspect some states should always to set, but this's just for debug.
(In reply to comment #13) > Could you try this patch with your origin xorg.conf? I tried the patch but it doesn't solve the problem, the rendering failure still appears when Option "EXANoComposite" "TRUE" is NOT set!
I've tried to reproduce this on a sony VGN-T26C, I'm looking into this.
(In reply to comment #15) > I've tried to reproduce this on a sony VGN-T26C, I'm looking into this. > I did some tests with EXA options. Setting Option "MigrationHeuristic" "greedy" (and removing "EXANoComposite") gives a clear screen,apps look good but the Terminal (an xfce4 package) is still corrupted. Setting to "always" or "smart" produces corrupted screen.
Created attachment 12993 [details] [review] i830 render fix This patch fixes exa issue on my sony 855gm. Pls test it to see if it also fixes your problems. i830 3d more states initialized according with mesa code.
(In reply to comment #17) > This patch fixes exa issue on my sony 855gm. Pls test it to see if it also > fixes your problems. Yes, that's it, also on my laptop no more exa render failure with your patch of i830_3d.c Thank you very much! 谢谢您！
Thanks to verify this! 不用谢!:) Brice, as this patch fixes for some 855gm users, I'm not sure about your copied issues on those i965s... Anyway I'll let our QA team double-check if we can reproduce or not.
I think we may close this, as it seems only affect 855gm. If you still can reproduce on some i965 boxes, pls reopen or open new bug. Thanks.
I am still waiting for some feedback of the bug reporters. So far, one i965 said he still had problems (missing fonts and icons). I'll open a new bug for this when I will have gathered enough info.
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.