Created attachment 113333 [details] dmesg booted with drm.debug On Radeon Hawaii board the GPU acceleration is disabled. Since the new HSA driver is designed for 64-bit kernels only, the current Radeon accelaration should be fixed for 32-bit systems. [ 0.883020] [drm] Supports vblank timestamp caching Rev 2 (21.10.2013). [ 0.883020] [drm] Driver supports precise vblank timestamp query. [ 0.883036] radeon 0000:01:00.0: radeon: using MSI. [ 0.883053] [drm] radeon: irq initialized. [ 1.026624] [drm:cik_ring_test] *ERROR* radeon: ring 0 test failed (scratch(0x3010C)=0xCAFEDEAD) [ 1.026678] radeon 0000:01:00.0: disabling GPU acceleration
Created attachment 113334 [details] lspci -vvnn output
I think the problem is ttm uses an unsigned long to store the offsets which is only 32 bits on 32 bit kernels, but the GPU's address space is 64 bit so it gets cut off.
Created attachment 113971 [details] [review] possible fix Does this patch help?
Created attachment 114065 [details] [review] fix for nouveau_fbcon.c printk compilation warning
Your idea seems to be correct: [ 0.922337] [drm] Supports vblank timestamp caching Rev 2 (21.10.2013). [ 0.922383] [drm] Driver supports precise vblank timestamp query. [ 0.922444] radeon 0000:01:00.0: radeon: using MSI. [ 0.922505] [drm] radeon: irq initialized. [ 0.924622] [drm] ring test on 0 succeeded in 4 usecs [ 0.924743] [drm] ring test on 1 succeeded in 3 usecs [ 0.924799] [drm] ring test on 2 succeeded in 3 usecs [ 0.924945] [drm] ring test on 3 succeeded in 5 usecs [ 0.924997] [drm] ring test on 4 succeeded in 5 usecs [ 0.970689] [drm] ring test on 5 succeeded in 2 usecs [ 0.990579] [drm] UVD initialized successfully. [ 1.099778] [drm] ring test on 6 succeeded in 24 usecs [ 1.099831] [drm] ring test on 7 succeeded in 3 usecs [ 1.099876] [drm] VCE initialized successfully. [ 1.100006] [drm] ib test on ring 0 succeeded in 0 usecs [ 1.599633] [drm] ib test on ring 1 succeeded in 0 usecs [ 1.851505] tsc: Refined TSC clocksource calibration: 3092.973 MHz [ 2.099525] [drm] ib test on ring 2 succeeded in 0 usecs [ 2.099660] [drm] ib test on ring 3 succeeded in 0 usecs [ 2.099758] [drm] ib test on ring 4 succeeded in 0 usecs [ 2.619405] [drm] ib test on ring 5 succeeded [ 2.640200] [drm] ib test on ring 6 succeeded [ 2.640982] [drm] ib test on ring 7 succeeded glxgears shows much better numbers now. The side effect is a compilation warning for nouveau driver - patch attached.
Also did a sanity test on Nvidia 50 and Radon 635, no new problems noticed. Alex, if you think that glxgears test is a sufficient proof, we can close this bug.
http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=54c4cd68ed7abd9f245722bee39464d04ddb4cfd
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.