https://bugzilla.kernel.org/show_bug.cgi?id=194761
Bug ID: 194761 Summary: amdgpu driver breaks on Oland (SI) Product: Drivers Version: 2.5 Kernel Version: 4.10 Hardware: x86-64 OS: Linux Tree: Mainline Status: NEW Severity: normal Priority: P1 Component: Video(DRI - non Intel) Assignee: drivers_video-dri@kernel-bugs.osdl.org Reporter: jdelvare@suse.de Regression: No
Created attachment 255045 --> https://bugzilla.kernel.org/attachment.cgi?id=255045&action=edit Checkerboard effect on glxgears
After upgrading my kernel from v4.9.11 to v4.10.0, all 3D applications are broken. The most visible problem is a checkerboard effect, with blinking of some areas. See the attached screenshot of glxgears for an example. The bug only happens when I use the amdgpu driver (with CONFIG_DRM_AMDGPU_SI=y.) If I use the radeon driver instead then all is fine.
There are no error messages in the kernel log, nor in /var/log/Xorg.0.log. However I could spot differences between the kernel logs:
-Linux version 4.9.11-1-default (geeko@buildhost) (gcc version 6.3.1 20170202 [gcc-6-branch revision 245119] (SUSE Linux) ) #1 SMP PREEMPT Sat Feb 18 17:59:27 UTC 2017 (cf9c670) +Linux version 4.10.1-1-default (geeko@buildhost) (gcc version 6.3.1 20170202 [gcc-6-branch revision 245119] (SUSE Linux) ) #1 SMP PREEMPT Sun Feb 26 12:43:10 UTC 2017 (1ecd5af)
[drm] GPU post is not needed amdgpu 0000:01:00.0: VRAM: 1024M 0x0000000000000000 - 0x000000003FFFFFFF (1024M used) -amdgpu 0000:01:00.0: GTT: 3991M 0x0000000040000000 - 0x00000001397847FF +amdgpu 0000:01:00.0: GTT: 1024M 0x0000000040000000 - 0x000000007FFFFFFF [drm] Detected VRAM RAM=1024M, BAR=256M [drm] RAM width 64bits DDR3 [drm] amdgpu: 1024M of VRAM memory ready -[drm] amdgpu: 3991M of GTT memory ready. -[drm] GART: num cpu pages 1021828, num gpu pages 1021828 -amdgpu 0000:01:00.0: PCIE GART of 3991M enabled (table at 0x0000000000040000). +[drm] amdgpu: 1024M of GTT memory ready. +[drm] GART: num cpu pages 262144, num gpu pages 262144 +amdgpu 0000:01:00.0: PCIE GART of 1024M enabled (table at 0x0000000000040000).
-[drm] Initialized amdgpu 3.8.0 20150101 for 0000:01:00.0 on minor 0 +[drm] Initialized amdgpu 3.9.0 20150101 for 0000:01:00.0 on minor 0
So something changed in the detection of the available memory. I don't know if this is related to the bug. As a comparison point, the radeon driver disagrees with both the old amdgpu driver and the new amdgpu driver:
radeon 0000:01:00.0: VRAM: 1024M 0x0000000000000000 - 0x000000003FFFFFFF (1024M used) radeon 0000:01:00.0: GTT: 2048M 0x0000000040000000 - 0x00000000BFFFFFFF [drm] Detected VRAM RAM=1024M, BAR=256M [drm] RAM width 64bits DDR [drm] radeon: 1024M of VRAM memory ready [drm] radeon: 2048M of GTT memory ready.