Description of problem: If KMS is active, glyphs in fonts change their shape to solid rectangle which renders text unreadable. The change seems to be gradual and takes effect after several minutes of work in X. See attached screenshot. If KMS is NOT active (nomodeset kernel parameter is set), this does not happen. Version-Release number of selected component (if applicable): kernel-2.6.27.7-130.fc10.x86_64 xorg-x11-server-common-1.5.3-5.fc10.x86_64 xorg-x11-drv-ati-6.9.0-59.fc10.x86_64 $ lspci | grep VGA 01:05.0 VGA compatible controller: ATI Technologies Inc RS690 [Radeon X1200 Series] How reproducible: 100%, but I do not know exactly the action Steps to Reproduce: 1. boot with KMS active (no kernel parameter added) 2. Browse the internet or otherwise use X applications. 3. Some glyps start to be rendered as solid rectangle and situation is gradually getting worse, still more and more glyphs become unreadable. Also, some icons disappear or get ugly/broken. The font glyphs, pixmaps and icons should be rendered as intended.
Created attachment 325214 [details] screenshot, glyphs rendered as solid rectangles
The same bug shows up on my laptop, also with RS690. To reproduce it, I open Slashdot.org main page in Firefox and repeatedly zoom in, out and scroll. Still, it can take a few minutes. I cannot reproduce it with kernel 2.6.27.5-123. I've seen it with 2.6.27.7-128. From looking at the diff between the two versions, I started suspecting the GART sizing change in the radeon driver. Another hint is I'm getting 'Unable to use IGP GART table size 524288' in dmesg. I am now testing a workaround: Adding the parameter 'gartsize=32' to radeon module. So far looks good. I think I can see a bug in radeon_set_igpgart(). switch(dev_priv->gart_size) assumes the size is in megabytes, but it is in bytes. When it prints the error message, it attempts to fall back to 32 MB GART size, but I'm afraid the original size is still preserved in several variables. I'll test a patch and post it soon.
Created attachment 325233 [details] radeon: fix IGP GART size setting I am now running with this patch instead of the workaround. The error message is gone. The black rectangles have not appeared yet. I'll let you know should they reappear.
Scratch build of the kernel with the patch applied (x86_64 only): http://koji.fedoraproject.org/koji/taskinfo?taskID=967454
A bug I filed (bug 473815) seems to look very similiar!
*** Bug 473815 has been marked as a duplicate of this bug. ***
Thanks Michal, kernel-2.6.27.7-134.fc10 is in koji with this patch. Let me know if it works.
kernel-2.6.27.7-134.fc10 works for me.
ditto.
Replied too early... while the screen corruption is fixed, resume for suspend doesn't work - have a backlit blank screen; does not respond to ctrl-alt-f*. according to syslog, it looks like networkmanager restarts correctly, so it should be possible to lock in from outside (the lapop) to try something, if there is something to try. syslog itself says resume error: ---------- Dec 2 17:06:34 localhost kernel: [drm] Num pipes: 1 Dec 2 17:06:34 localhost kernel: [drm:radeon_resume] *ERROR* Dec 2 17:06:34 localhost kernel: [drm] Loading RS690/RS740 Microcode Dec 2 17:06:34 localhost kernel: [drm] Num pipes: 1 Dec 2 17:06:34 localhost kernel: [drm] writeback test succeeded in 1 usecs Dec 2 17:06:34 localhost kernel: [drm] LVDS-8: set mode 9 ----------- I also have these leftover from my fglrx days, just in case it matters. # more /etc/pm/config.d/video POST_VIDEO=false DISPLAY_QUIRK_S3_BIOS=true to be honest I have not tried suspend/resume in this configuration (EXA+modeset) before, but suspend/resume with XAA+nomodeset works.
Created attachment 325417 [details] screen corruption/tearing with nomodeset Since I am not having suspend/resume (it worked every other time so far), I thought I'll try putting nomodeset back, without xorg.conf (i.e. EXA) - and I am getting screen corruption/tearing as shown. xorg-x11-drv-ati-6.9.0-61.fc10 kernel-2.6.27.7-134.fc10 (this probably should be a separate bug, since this screenshot is about EXA+nomodeset - but I am just expressing my general frustration that switching between EXA/XAA, modeset/nomodeset, radeon/radeonhd, none really work happily. The best for general usage - for me - is still XAA+nomodeset)
kernel-2.6.27.7-134.fc10 has been pushed to the Fedora 10 testing repository. If problems still persist, please make note of it in this bug report. If you want to test the update, you can install it with su -c 'yum --enablerepo=updates-testing update kernel'. You can provide feedback for this update here: http://admin.fedoraproject.org/updates/F10/FEDORA-2008-10772
kernel-2.6.27.7-134.fc10 has been pushed to the Fedora 10 stable repository. If problems still persist, please make note of it in this bug report.
Please note that the kernel containing the fix, kernel-2.6.27.7-134.fc10, introduces a regression at least on some hardware: #477140 .