[Kernel-packages] [Bug 1950205] [NEW] Some shaders can crash the amdgpu kernel module, tainting
Public bug reported: Hi there, I am experiencing, at random, the AMD-GPU "fences timed out!" bug reported previously as being either a kernel issue (https://bugzilla.kernel.org/show_bug.cgi?id=213145) or a mesa (https://gitlab.freedesktop.org/mesa/mesa/-/issues/4866) issue. The later was apparently fixed in 21.2.4 -- I have experienced it both on the stock mesa package versions current in play in 21.10 and the "bleeding edge" git-based builds. It is difficult to reproduce deterministically. For me, at present, roleplaying game Pathfinder: Wrath of the Righteous running under proton with Lutris (0.5.8.3/wine version lutris-fshack-6.14-3-x86_64 with DXVK v1.8.1L) triggers far more infrequently on the lowest graphics settings, but occasionally absolutely deterministically at certain points (I have a save game that can do this in the highly unlikely event that anyone wants to take me up on the issue and has the GOG game -- email me at jack.mil...@physics.org). This manifests as extreme screen corruption requiring a restart of at a minimum lightdm to fix, and more likely the whole computer. Memory issues can occur either on the card or on the host, namely: [ 1208.225974] [drm:amdgpu_dm_commit_planes [amdgpu]] *ERROR* Waiting for fences timed out! [ 1213.01] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, signaled seq=112591, emitted seq=112593 [ 1213.100119] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* Process information: process Wrath.exe pid 18793 thread Wrath.exe pid 18984 [ 1213.100224] amdgpu :08:00.0: amdgpu: GPU reset begin! [ 1213.349988] [drm:amdgpu_dm_commit_planes [amdgpu]] *ERROR* Waiting for fences timed out! [ 1213.350131] [drm:dce110_vblank_set [amdgpu]] *ERROR* Failed to get VBLANK! [ 1213.509424] amdgpu: cp is busy, skip halt cp [ 1213.95] amdgpu: rlc is busy, skip halt rlc [ 1213.778816] amdgpu :08:00.0: amdgpu: BACO reset [ 1214.425001] amdgpu :08:00.0: amdgpu: GPU reset succeeded, trying to resume [ 1214.427319] [drm] PCIE GART of 256M enabled (table at 0x00F4007E9000). [ 1214.427335] [drm] VRAM is lost due to GPU reset! [ 1214.600175] [ cut here ] [ 1214.600178] amdgpu :08:00.0: drm_WARN_ON(atomic_read(&vblank->refcount) == 0) [ 1214.600244] WARNING: CPU: 7 PID: 0 at drivers/gpu/drm/drm_vblank.c:1210 drm_vblank_put+0xef/0x100 [drm] [ 1214.600274] Modules linked in: xfrm_user xfrm_algo xt_addrtype br_netfilter vboxnetadp(OE) vboxnetflt(OE) vboxdrv(OE) rfcomm ashmem_linux(C) binder_linux iptable_mangle xt_comment iptable_nat iptable_filter bpfilter xt_CHECKSUM xt_MASQUERADE ipt_REJECT nf_reject_ipv4 nf_conntrack_netlink xt_conntrack nft_chain_nat nf_nat xt_NFQUEUE xt_tcpudp nf_conntrack nft_compat nf_defrag_ipv6 nf_defrag_ipv4 tcp_diag inet_diag nft_counter nfnetlink_queue nf_tables nfnetlink bridge stp llc cmac algif_hash algif_skcipher af_alg bnep overlay binfmt_misc nls_iso8859_1 zfs(PO) zunicode(PO) zzstd(O) zlua(O) zavl(PO) icp(PO) zcommon(PO) snd_hda_codec_realtek intel_rapl_msr znvpair(PO) snd_hda_codec_generic intel_rapl_common spl(O) ledtrig_audio snd_hda_codec_hdmi edac_mce_amd snd_hda_intel snd_intel_dspcfg snd_intel_sdw_acpi snd_hda_codec kvm_amd snd_hda_core btusb ath9k snd_hwdep btrtl kvm ath9k_common snd_pcm btbcm btintel ath9k_hw snd_seq_midi rapl bluetooth snd_seq_midi_event ath snd_rawmidi [ 1214.600304] ecdh_generic joydev mac80211 snd_seq ecc input_leds snd_seq_device apple_mfi_fastcharge snd_timer snd eeepc_wmi soundcore ccp efi_pstore wmi_bmof cfg80211 k10temp libarc4 mac_hid sch_fq_codel nct6775 hwmon_vid msr nfsd parport_pc ppdev auth_rpcgss lp nfs_acl parport lockd grace sunrpc ip_tables x_tables autofs4 btrfs blake2b_generic zstd_compress dm_crypt raid10 raid456 async_raid6_recov async_memcpy async_pq async_xor async_tx xor raid6_pq libcrc32c raid1 raid0 multipath linear hid_logitech_dj hid_apple hid_logitech_hidpp hid_generic usbhid hid mfd_aaeon asus_wmi sparse_keymap amdgpu video iommu_v2 gpu_sched i2c_algo_bit drm_ttm_helper ttm crct10dif_pclmul crc32_pclmul ghash_clmulni_intel drm_kms_helper syscopyarea aesni_intel sysfillrect sysimgblt fb_sys_fops cec crypto_simd rc_core cryptd drm r8169 ahci i2c_piix4 xhci_pci libahci xhci_pci_renesas realtek wmi [ 1214.600336] CPU: 7 PID: 0 Comm: swapper/7 Tainted: P C OE 5.13.0-21-generic #21-Ubuntu [ 1214.600337] Hardware name: System manufacturer System Product Name/TUF GAMING X570-PLUS, BIOS 1201 09/09/2019 [ 1214.600339] RIP: 0010:drm_vblank_put+0xef/0x100 [drm] [ 1214.600353] Code: 8b 7f 08 4c 8b 67 50 4d 85 e4 74 22 e8 ea 55 ef d7 48 c7 c1 d8 22 4e c0 4c 89 e2 48 c7 c7 75 d5 4d c0 48 89 c6 e8 c0 75 31 d8 <0f> 0b eb 8a 4c 8b 27 eb d9 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 [ 1214.600354] RSP: 0018:a654803a8d78 EFLAGS: 00010086 [ 1214.600356] RAX: RBX: 94289d0a RCX: 0027 [ 1214.600357] RDX: 942f7ebd89c8 RSI: 0001 RDI: 942f7ebd89c0 [ 1214.600358] RBP
[Kernel-packages] [Bug 1950205] Re: Some shaders can crash the amdgpu kernel module, tainting
Edit -- it transpires that, despite what I said, the magic <0 to <=0 change hasn't been made on the main branch of mesa's git repo yet. I've successfully compiled a local copy (with meson & ninja) and am running it in a separate root to avoid clobbering the system libraries using LD_LIBRARY_PATH and friends which can get glxgears working at least. Frankly I am in a little over my depth. I'd like to really be able to build the packages from source, apply a two character / one line diff as suggested upstream to see if it actually works, and then test further. Any advice very welcome ;-). -- You received this bug notification because you are a member of Kernel Packages, which is subscribed to linux in Ubuntu. https://bugs.launchpad.net/bugs/1950205 Title: Some shaders can crash the amdgpu kernel module, tainting Status in linux package in Ubuntu: Confirmed Bug description: Hi there, I am experiencing, at random, the AMD-GPU "fences timed out!" bug reported previously as being either a kernel issue (https://bugzilla.kernel.org/show_bug.cgi?id=213145) or a mesa (https://gitlab.freedesktop.org/mesa/mesa/-/issues/4866) issue. The later was apparently fixed in 21.2.4 -- I have experienced it both on the stock mesa package versions current in play in 21.10 and the "bleeding edge" git-based builds. It is difficult to reproduce deterministically. For me, at present, roleplaying game Pathfinder: Wrath of the Righteous running under proton with Lutris (0.5.8.3/wine version lutris-fshack-6.14-3-x86_64 with DXVK v1.8.1L) triggers far more infrequently on the lowest graphics settings, but occasionally absolutely deterministically at certain points (I have a save game that can do this in the highly unlikely event that anyone wants to take me up on the issue and has the GOG game -- email me at jack.mil...@physics.org). This manifests as extreme screen corruption requiring a restart of at a minimum lightdm to fix, and more likely the whole computer. Memory issues can occur either on the card or on the host, namely: [ 1208.225974] [drm:amdgpu_dm_commit_planes [amdgpu]] *ERROR* Waiting for fences timed out! [ 1213.01] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, signaled seq=112591, emitted seq=112593 [ 1213.100119] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* Process information: process Wrath.exe pid 18793 thread Wrath.exe pid 18984 [ 1213.100224] amdgpu :08:00.0: amdgpu: GPU reset begin! [ 1213.349988] [drm:amdgpu_dm_commit_planes [amdgpu]] *ERROR* Waiting for fences timed out! [ 1213.350131] [drm:dce110_vblank_set [amdgpu]] *ERROR* Failed to get VBLANK! [ 1213.509424] amdgpu: cp is busy, skip halt cp [ 1213.95] amdgpu: rlc is busy, skip halt rlc [ 1213.778816] amdgpu :08:00.0: amdgpu: BACO reset [ 1214.425001] amdgpu :08:00.0: amdgpu: GPU reset succeeded, trying to resume [ 1214.427319] [drm] PCIE GART of 256M enabled (table at 0x00F4007E9000). [ 1214.427335] [drm] VRAM is lost due to GPU reset! [ 1214.600175] [ cut here ] [ 1214.600178] amdgpu :08:00.0: drm_WARN_ON(atomic_read(&vblank->refcount) == 0) [ 1214.600244] WARNING: CPU: 7 PID: 0 at drivers/gpu/drm/drm_vblank.c:1210 drm_vblank_put+0xef/0x100 [drm] [ 1214.600274] Modules linked in: xfrm_user xfrm_algo xt_addrtype br_netfilter vboxnetadp(OE) vboxnetflt(OE) vboxdrv(OE) rfcomm ashmem_linux(C) binder_linux iptable_mangle xt_comment iptable_nat iptable_filter bpfilter xt_CHECKSUM xt_MASQUERADE ipt_REJECT nf_reject_ipv4 nf_conntrack_netlink xt_conntrack nft_chain_nat nf_nat xt_NFQUEUE xt_tcpudp nf_conntrack nft_compat nf_defrag_ipv6 nf_defrag_ipv4 tcp_diag inet_diag nft_counter nfnetlink_queue nf_tables nfnetlink bridge stp llc cmac algif_hash algif_skcipher af_alg bnep overlay binfmt_misc nls_iso8859_1 zfs(PO) zunicode(PO) zzstd(O) zlua(O) zavl(PO) icp(PO) zcommon(PO) snd_hda_codec_realtek intel_rapl_msr znvpair(PO) snd_hda_codec_generic intel_rapl_common spl(O) ledtrig_audio snd_hda_codec_hdmi edac_mce_amd snd_hda_intel snd_intel_dspcfg snd_intel_sdw_acpi snd_hda_codec kvm_amd snd_hda_core btusb ath9k snd_hwdep btrtl kvm ath9k_common snd_pcm btbcm btintel ath9k_hw snd_seq_midi rapl bluetooth snd_seq_midi_event ath snd_rawmidi [ 1214.600304] ecdh_generic joydev mac80211 snd_seq ecc input_leds snd_seq_device apple_mfi_fastcharge snd_timer snd eeepc_wmi soundcore ccp efi_pstore wmi_bmof cfg80211 k10temp libarc4 mac_hid sch_fq_codel nct6775 hwmon_vid msr nfsd parport_pc ppdev auth_rpcgss lp nfs_acl parport lockd grace sunrpc ip_tables x_tables autofs4 btrfs blake2b_generic zstd_compress dm_crypt raid10 raid456 async_raid6_recov async_memcpy async_pq async_xor async_tx xor raid6_pq libcrc32c raid1 raid0 multipath linear hid_logitech_dj hid_apple hid_logitech_hidpp hid_generic usbhid hid mfd_aaeon asus_wmi sparse_keymap amdgpu video iommu_v2 gpu_sched i2c_algo_bit drm_ttm_helper ttm crct10dif_pclm
[Kernel-packages] [Bug 1950205] Re: Some shaders can crash the amdgpu kernel module, tainting
Edit: Please close this bug -- it is a duplicate of an upstream issue in dxvk (https://gitlab.freedesktop.org/mesa/mesa/-/issues/4866) and not the kernel per se (although that makes life worse by clearing out vram and hard-resetting the gpu once the crash occurs, which X and DMs have no way to recover from). ** Bug watch added: gitlab.freedesktop.org/mesa/mesa/-/issues #4866 https://gitlab.freedesktop.org/mesa/mesa/-/issues/4866 -- You received this bug notification because you are a member of Kernel Packages, which is subscribed to linux in Ubuntu. https://bugs.launchpad.net/bugs/1950205 Title: Some shaders can crash the amdgpu kernel module, tainting Status in linux package in Ubuntu: Confirmed Bug description: Hi there, I am experiencing, at random, the AMD-GPU "fences timed out!" bug reported previously as being either a kernel issue (https://bugzilla.kernel.org/show_bug.cgi?id=213145) or a mesa (https://gitlab.freedesktop.org/mesa/mesa/-/issues/4866) issue. The later was apparently fixed in 21.2.4 -- I have experienced it both on the stock mesa package versions current in play in 21.10 and the "bleeding edge" git-based builds. It is difficult to reproduce deterministically. For me, at present, roleplaying game Pathfinder: Wrath of the Righteous running under proton with Lutris (0.5.8.3/wine version lutris-fshack-6.14-3-x86_64 with DXVK v1.8.1L) triggers far more infrequently on the lowest graphics settings, but occasionally absolutely deterministically at certain points (I have a save game that can do this in the highly unlikely event that anyone wants to take me up on the issue and has the GOG game -- email me at jack.mil...@physics.org). This manifests as extreme screen corruption requiring a restart of at a minimum lightdm to fix, and more likely the whole computer. Memory issues can occur either on the card or on the host, namely: [ 1208.225974] [drm:amdgpu_dm_commit_planes [amdgpu]] *ERROR* Waiting for fences timed out! [ 1213.01] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, signaled seq=112591, emitted seq=112593 [ 1213.100119] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* Process information: process Wrath.exe pid 18793 thread Wrath.exe pid 18984 [ 1213.100224] amdgpu :08:00.0: amdgpu: GPU reset begin! [ 1213.349988] [drm:amdgpu_dm_commit_planes [amdgpu]] *ERROR* Waiting for fences timed out! [ 1213.350131] [drm:dce110_vblank_set [amdgpu]] *ERROR* Failed to get VBLANK! [ 1213.509424] amdgpu: cp is busy, skip halt cp [ 1213.95] amdgpu: rlc is busy, skip halt rlc [ 1213.778816] amdgpu :08:00.0: amdgpu: BACO reset [ 1214.425001] amdgpu :08:00.0: amdgpu: GPU reset succeeded, trying to resume [ 1214.427319] [drm] PCIE GART of 256M enabled (table at 0x00F4007E9000). [ 1214.427335] [drm] VRAM is lost due to GPU reset! [ 1214.600175] [ cut here ] [ 1214.600178] amdgpu :08:00.0: drm_WARN_ON(atomic_read(&vblank->refcount) == 0) [ 1214.600244] WARNING: CPU: 7 PID: 0 at drivers/gpu/drm/drm_vblank.c:1210 drm_vblank_put+0xef/0x100 [drm] [ 1214.600274] Modules linked in: xfrm_user xfrm_algo xt_addrtype br_netfilter vboxnetadp(OE) vboxnetflt(OE) vboxdrv(OE) rfcomm ashmem_linux(C) binder_linux iptable_mangle xt_comment iptable_nat iptable_filter bpfilter xt_CHECKSUM xt_MASQUERADE ipt_REJECT nf_reject_ipv4 nf_conntrack_netlink xt_conntrack nft_chain_nat nf_nat xt_NFQUEUE xt_tcpudp nf_conntrack nft_compat nf_defrag_ipv6 nf_defrag_ipv4 tcp_diag inet_diag nft_counter nfnetlink_queue nf_tables nfnetlink bridge stp llc cmac algif_hash algif_skcipher af_alg bnep overlay binfmt_misc nls_iso8859_1 zfs(PO) zunicode(PO) zzstd(O) zlua(O) zavl(PO) icp(PO) zcommon(PO) snd_hda_codec_realtek intel_rapl_msr znvpair(PO) snd_hda_codec_generic intel_rapl_common spl(O) ledtrig_audio snd_hda_codec_hdmi edac_mce_amd snd_hda_intel snd_intel_dspcfg snd_intel_sdw_acpi snd_hda_codec kvm_amd snd_hda_core btusb ath9k snd_hwdep btrtl kvm ath9k_common snd_pcm btbcm btintel ath9k_hw snd_seq_midi rapl bluetooth snd_seq_midi_event ath snd_rawmidi [ 1214.600304] ecdh_generic joydev mac80211 snd_seq ecc input_leds snd_seq_device apple_mfi_fastcharge snd_timer snd eeepc_wmi soundcore ccp efi_pstore wmi_bmof cfg80211 k10temp libarc4 mac_hid sch_fq_codel nct6775 hwmon_vid msr nfsd parport_pc ppdev auth_rpcgss lp nfs_acl parport lockd grace sunrpc ip_tables x_tables autofs4 btrfs blake2b_generic zstd_compress dm_crypt raid10 raid456 async_raid6_recov async_memcpy async_pq async_xor async_tx xor raid6_pq libcrc32c raid1 raid0 multipath linear hid_logitech_dj hid_apple hid_logitech_hidpp hid_generic usbhid hid mfd_aaeon asus_wmi sparse_keymap amdgpu video iommu_v2 gpu_sched i2c_algo_bit drm_ttm_helper ttm crct10dif_pclmul crc32_pclmul ghash_clmulni_intel drm_kms_helper syscopyarea aesni_intel sysfillrect sysimgblt fb_sys_fops cec crypto_simd rc_core cryptd drm r8169 ahci