https://bugzilla.kernel.org/show_bug.cgi?id=202537 Bug ID: 202537 Summary: amdgpu/DC failed to reserve new abo buffer before flip Product: Drivers Version: 2.5 Kernel Version: 4.20 Hardware: All OS: Linux Tree: Mainline Status: NEW Severity: normal Priority: P1 Component: Video(DRI - non Intel) Assignee: drivers_video-dri@xxxxxxxxxxxxxxxxxxxx Reporter: linux@xxxxxxxxxxxxxxxxxxxx Regression: No I've been using amdgpu for a long time on my Kaveri (A7800) now and it works fine. In the recent kernel versions (I think since 4.15), I've been trying it with the DC activated and apart from some initial issue with the HDMI connection it works fine and on 4.19 it's rock-stable. However, when I tried 4.20 (all versions from 4.20.1 to 4.20.6), I'm experiencing a regression. Initially, everything works fine, but at some point especially video-related things stop working properly. vaapi seems more affected than vdpau, but at some point they both fail to setup the hw decoding. (btw, vdpau for some reason can only run for one video at the time, while vaapi can do multiple ones, but that's an unrelated issue. It used to be different a year ago or even earlier on) I think the problems start when I see a lot of messages like this: [drm:amdgpu_display_crtc_page_flip_target] *ERROR* failed to reserve new abo buffer before flip However, after that I can continue for a bit, possibly with the restriction of not being able to use vaapi but vdpau. At some point, the system will fail due to a memory leak, at least the OOM is starting to kill stuff until it ends up killing the window manager and X11. Before that I get these messages: [drm:amdgpu_cs_ioctl] *ERROR* amdgpu_vm_validate_pt_bos() failed. [drm:amdgpu_cs_ioctl] *ERROR* Not enough memory for command submission! [drm:amdgpu_cs_ioctl] *ERROR* amdgpu_cs_list_validate(validated) failed. [drm:amdgpu_cs_ioctl] *ERROR* Not enough memory for command submission! [drm:amdgpu_cs_ioctl] *ERROR* amdgpu_vm_validate_pt_bos() failed. [drm:amdgpu_cs_ioctl] *ERROR* Not enough memory for command submission! --- snip --- (lots of oom activity) and finally: [TTM] Out of kernel memory [TTM] Out of kernel memory [TTM] Out of kernel memory [TTM] Out of kernel memory [TTM] Out of kernel memory [TTM] Out of kernel memory [TTM] Out of kernel memory amdgpu 0000:00:01.0: (-12) failed to allocate kernel bo [drm:amdgpu_uvd_free_handles] *ERROR* Error destroying UVD -12! The latter one – I think – actually being the solution to the OOM problem, but I'm certainly not an expert. CPU/GPU is: vendor_id : AuthenticAMD cpu family : 21 model : 48 model name : AMD A10-7800 Radeon R7, 12 Compute Cores 4C+8G stepping : 1 microcode : 0x6003106 cpu MHz : 1592.730 cache size : 2048 KB Back to 4.19 for now since that runs beautifully. -- You are receiving this mail because: You are watching the assignee of the bug. _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/dri-devel
