Created attachment 26113 [details] Xorg.0.log Card is radeon HD3850 AGP using git driver, xserver is also git. Problem exists using drm from agd5f R6xx branch or mesa/drm R6xx. Does not happen with exa disabled. Also happens with radeon driver. Running x11perf -v1.3 -all stops rendering to screen during the calibration of certain tests. This is not just the output to the results xterm and test window, but the whole screen (had a counter running in a separate xterm and switching metacity desktops doesn't display - though it works) To restart rendering the mouse needs to be moved until the problematic calibration has finished - the mouse is jerky while doing this. 8 tests trigger this - tiled rectangle, both 1x1 and 10x10 with 4x4, 161x145 and 216x208. 500-pixel solid circle and 500-pixel filled ellipse. All other tests run OK.
(In reply to comment #0) More info - This is an untypical setup being based on LFS and running xorg and drivers from a git tree like - http://wiki.x.org/wiki/ModularDevelopersGuide. No desktop is installed and I am just running metacity as a window manager. AGP may be a factor but I am not sure about this. If I use lower aperture sizes exa fails to render properly, but in these cases I also get DRM errors like - [drm] Can't use AGP base @0xe0000000, won't fit [drm] writeback test failed 0 deadbeef
(In reply to comment #0) > Created an attachment (id=26113) [details] > Xorg.0.log > > Card is radeon HD3850 AGP using git driver, xserver is also git. Problem exists > using drm from agd5f R6xx branch or mesa/drm R6xx. > > Does not happen with exa disabled. > Also happens with radeon driver. > > Running x11perf -v1.3 -all stops rendering to screen during the calibration of > certain tests. This is not just the output to the results xterm and test > window, but the whole screen (had a counter running in a separate xterm and > switching metacity desktops doesn't display - though it works) > > To restart rendering the mouse needs to be moved until the problematic > calibration has finished - the mouse is jerky while doing this. > > 8 tests trigger this - > > tiled rectangle, both 1x1 and 10x10 with 4x4, 161x145 and 216x208. > > 500-pixel solid circle and 500-pixel filled ellipse. > > All other tests run OK. Reassigned to radeon as this is what I am testing with now and radeonhd bugzilla still has Luc as mail, so I don't even know if anyone at sees them! Have just found that Option BusType PCIE fixes this. It also fixes a similar issue I have just started to see with the mesa demos morph3d and geartrain. Interestingly using xserver/mesa/ddx/agd5f modules from todays' gits geartrain using AGP renders properly as long as nothing else writes to the screen, so moving mouse will cause artifacts. After 30sec however it will freeze and not start again until something else writes to screen. With BusType PCIE it has permanent artifacts but does not freeze.
> With BusType PCIE it has permanent artifacts but does not freeze. Just found that I get a hard lockup, when I quit X with PCIE.
(In reply to comment #3) > Just found that I get a hard lockup, when I quit X with PCIE. This only seems to happen if I ForceLowPowerMode + BusType PCIE.
(In reply to comment #2) > With BusType PCIE it has permanent artifacts but does not freeze. With recent changes, artifacts are now gone. I have found that the reason PCIE worked was that it disabled writebacks. If I modprobe radeon no_wb=1 then I no longer have the blocking behavior, so I guess this is just another AGP wierdness rather than a bug.
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.