On Tue, Jun 08, 2021 at 01:51:21PM +0000, Jerome Kasper wrote: > Hello bugs@ > > Since i'm following -current i lately encountered drm issues with my graphic > card that doesn't attach anymore while it was working on 6.9-stable. > sysupgrade -s done today and still having the issue. > Non attachement occurs at each reboot leaving graphics to efifb > > Any method to track down faulty code or simple way to bisect the commit > that broke things would be greatly appreciated? Wondering if it didn't > came along with the amdgpu firmware upgrade?
Most likely related. You can find the old firmware at http://firmware.openbsd.org/firmware/6.9/amdgpu-firmware-20201218.tgz to confirm if that is the case. I would also be interested to know how a kernel with the following diff goes. The mapping is an optimisation, there is a fallback. Index: sys/dev/pci/drm/amd/amdgpu/amdgpu_ttm.c =================================================================== RCS file: /cvs/src/sys/dev/pci/drm/amd/amdgpu/amdgpu_ttm.c,v retrieving revision 1.5 diff -u -p -r1.5 amdgpu_ttm.c --- sys/dev/pci/drm/amd/amdgpu/amdgpu_ttm.c 21 Oct 2020 02:16:53 -0000 1.5 +++ sys/dev/pci/drm/amd/amdgpu/amdgpu_ttm.c 8 Jun 2021 14:56:39 -0000 @@ -1870,11 +1870,11 @@ int amdgpu_ttm_init(struct amdgpu_device adev->gmc.visible_vram_size, BUS_SPACE_MAP_LINEAR | BUS_SPACE_MAP_PREFETCHABLE, &adev->mman.aper_bsh)) { - DRM_ERROR("Failed to remap VRAM\n"); - return -ENOMEM; + adev->mman.aper_base_kaddr = NULL; + } else { + adev->mman.aper_base_kaddr = bus_space_vaddr(adev->memt, + adev->mman.aper_bsh); } - adev->mman.aper_base_kaddr = bus_space_vaddr(adev->memt, - adev->mman.aper_bsh); #endif #endif