Public bug reported: Suggested package: linux (Ubuntu), Ubuntu 26.04 / Resolute.
Two AMDGPU hangs on this laptop ended the GNOME Wayland login session while the kernel continued running. The second occurred during ordinary terminal use, despite a previously installed Ptyxis Cairo-renderer workaround. ## System - Lenovo IdeaPad Slim 5 16AHP9, product 83DD; BIOS NGCN34WW, dated 2026-01-16. - AMD HawkPoint1, PCI 1002:1900 (rev cc), subsystem 17aa:3807; amdgpu. - Ubuntu 26.04.1 LTS; kernel 7.0.0-31-generic, Ubuntu package 7.0.0-31.31 (base 7.0.14). - GNOME Shell 50.1-0ubuntu1.2; Mutter 50.1-0ubuntu2.2; Wayland. - Mesa / mesa-libgallium 26.0.8-1ubuntu0.3; Ptyxis 50.1-1ubuntu2. - AMD firmware: linux-firmware-amd-graphics 20260319.git217ca6e4-0ubuntu3.1. - Internal eDP panel only at inspection time; boot log reports PSR support 0. - No amdgpu overrides appear in the kernel command line. GPU performance policy is auto; ppfeaturemask is 0xfff7bfff. These versions match the released Ubuntu archive as checked on 2026-09-12. Package integrity verification reported no discrepancies for Mesa Gallium, Mesa DRI/Vulkan, libdrm-amdgpu, Mutter, Shell and AMD graphics firmware. ## What happened On 2026-09-12 at 10:27:02 UTC, the graphics ring timed out, naming GNOME Shell PID 10699 and its graphics-submission thread 10716. MES did not respond to RESET and then repeatedly failed REMOVE_QUEUE. A MODE2 GPU reset succeeded at 10:27:31. GNOME Shell reported loss of an innocent graphics context and aborted. At 10:27:39 systemd recorded status 6/ABRT for the Shell service; Ptyxis lost its Wayland compositor connection. The login session ended at 10:27:41, and a new login began at 10:28:22. The kernel did not reboot; its boot began on 2026-09-11 at 16:20:43 UTC. No preceding GPU page fault, DMCUB/flip timeout, kernel OOM kill or systemd-oomd kill was found in the inspected logs for this incident. A previous event on 2026-09-11 at 12:03:49 UTC, on the same kernel version, named Ptyxis in its gfx_0.0.0 timeout and also required a GPU reset. ## Symbolized Shell backtrace The crashing thread is LWP 10716, matching the submission thread named in the timeout. Matching Ubuntu Mesa debug symbols were selected by build ID b089ec2f62ded38e9327502246aa4ccfb7929b24: ``` abort() amdgpu_ctx_set_sw_reset_status() at amdgpu_cs.cpp:491 amdgpu_cs_submit_ib<(queue_type)0>() at amdgpu_cs.cpp:2113 util_queue_thread_func() at u_queue.c:294 impl_thrd_routine() at threads_posix.c:43 ``` The matching Mesa 26.0.8 source identifies line 2113 as the -ECANCELED / PIPE_INNOCENT_CONTEXT_RESET path. The subsequent abort explains the session loss. It does not establish what originally hung the graphics ring. ## Trigger and mitigation already tried The user was working with Codex CLI in Ptyxis when the second event occurred. There is no reliable reproduction command, and Codex is not established as the cause. These are two observed incidents, not a controlled reproduction. After the first incident, a user-local Ptyxis wrapper selected GSK_RENDERER=cairo and a D-Bus override routed desktop activation through it. Both were installed before the next boot. After the second incident and new login, the running Ptyxis process had GSK_RENDERER=cairo and no open /dev/dri/ handles. The closed terminal's environment cannot be checked retrospectively. The workaround's installation has therefore not established system stability. Enabled Shell extensions at inspection time: ``` [email protected] [email protected] [email protected] [email protected] [email protected] [email protected] [email protected] [email protected] ``` A comparison with extensions disabled has not been performed. Neither has a newer/mainline kernel test. The GPU device coredump expired before it was saved. Selected journals and the symbolized userspace backtrace are attached. ## Related reports and requested next step These reports have related symptoms but are not established duplicates: - https://gitlab.freedesktop.org/mesa/mesa/-/work_items/15668 — same GPU family and graphics-ring/MES recovery signature, with a different game trigger. - https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2158614 — innocent-context abort and session loss, on a different GPU with preceding Chrome page faults. Ubuntu Proposed currently has kernel 7.0.0-38.38, whose changelog includes AMD gfx11 queue-interrupt fixes. Is that the appropriate next comparison for this hardware, or should the next step be a specific Mesa/firmware test or additional diagnostics? No particular fix or initiating defect is claimed. This report was prepared with AI assistance from the user's local journals and existing crash dump. Raw process environments and memory dumps are not included in the attached text. ** Affects: linux (Ubuntu) Importance: Undecided Status: New ** Tags: amd64 amdgpu resolute wayland ** Attachment added: "Anonymized GPU reset journals and symbolized GNOME Shell backtrace" https://bugs.launchpad.net/bugs/2167160/+attachment/5999766/+files/evidence.txt -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2167160 Title: [HawkPoint1 1002:1900] gfx_0.0.0 timeout, failed MES ring reset, then GNOME Shell SIGABRT and session loss To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2167160/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
