Confirming this on **gfx1152 (GC 11.5.2, "Krackan Point")**, which I don't think is represented in this bug yet - previous reports here are Phoenix/Strix class. The reproducer in this bug (a page with 32 `<video>` elements) triggers it reliably for me, and one of my two faults landed on the same `0x3f800000` address noted earlier in this report.
**Hardware** - Lenovo ThinkPad P14s Gen 6 AMD (21QL0017US), BIOS R2XET39W 1.19 (2026-03-19) - AMD Ryzen AI 7 PRO 350 w/ Radeon 860M - GPU: `1002:1114` rev d2, subsystem `17aa:512f` — GC 11.5.2 / gfx1152 - amdgpu firmware `gc_11_5_2_*` dated 2026-08-18 (shipped in the pc-kernel snap) **Software** - Ubuntu 26.04.1 LTS - Kernel 7.0.0-31-generic (Canonical snap `pc-kernel` 7.0.0-31.31, rev 3699) - Mesa 26.0.8-1ubuntu0.3 - GNOME Shell 50.1, Wayland - Brave 1.94.121 (Chromium), `--ozone-platform=wayland` **OpenGL path, not Vulkan.** The faulting GPU process maps `libgallium-26.0.8-1ubuntu0.3.so`, `libEGL_mesa.so`, `dri_gbm.so` and `libdrm_amdgpu.so`, and does **not** map `libvulkan_radeon.so` — so this is radeonsi, consistent with this being filed against mesa. **Fault 1 — organic, during normal browsing** ``` amdgpu 0000:c4:00.0: [gfxhub] page fault (src_id:0 ring:24 vmid:7 pasid:125) amdgpu 0000:c4:00.0: Process brave pid 99451 thread brave:cs0 pid 99481 amdgpu 0000:c4:00.0: in page starting at address 0x000000003a19e000 from client 10 amdgpu 0000:c4:00.0: GCVM_L2_PROTECTION_FAULT_STATUS:0x00701430 amdgpu 0000:c4:00.0: Faulty UTCL2 client ID: SQC (data) (0xa) amdgpu 0000:c4:00.0: MORE_FAULTS: 0x0 amdgpu 0000:c4:00.0: WALKER_ERROR: 0x0 amdgpu 0000:c4:00.0: PERMISSION_FAULTS: 0x3 amdgpu 0000:c4:00.0: MAPPING_ERROR: 0x0 amdgpu 0000:c4:00.0: RW: 0x0 amdgpu 0000:c4:00.0: ring gfx_0.0.0 timeout, signaled seq=337002, emitted seq=337004 amdgpu 0000:c4:00.0: Starting gfx_0.0.0 ring reset amdgpu 0000:c4:00.0: Ring gfx_0.0.0 reset succeeded amdgpu 0000:c4:00.0: [drm] device wedged, but recovered through reset ``` **Fault 2 — from this bug's 32-`<video>` reproducer, at `0x3f800000`** ``` amdgpu 0000:c4:00.0: [gfxhub] page fault (src_id:0 ring:24 vmid:5 pasid:143) amdgpu 0000:c4:00.0: Process brave pid 286647 thread brave:cs0 pid 286654 amdgpu 0000:c4:00.0: in page starting at address 0x000000003f800000 from client 10 amdgpu 0000:c4:00.0: GCVM_L2_PROTECTION_FAULT_STATUS:0x00501430 amdgpu 0000:c4:00.0: Faulty UTCL2 client ID: SQC (data) (0xa) amdgpu 0000:c4:00.0: PERMISSION_FAULTS: 0x3 amdgpu 0000:c4:00.0: ring gfx_0.0.0 timeout, signaled seq=384501, emitted seq=384503 amdgpu 0000:c4:00.0: Starting gfx_0.0.0 ring reset amdgpu 0000:c4:00.0: Ring gfx_0.0.0 reset succeeded amdgpu 0000:c4:00.0: [drm] device wedged, but recovered through reset ``` Note `GCVM_L2_PROTECTION_FAULT_STATUS` differs only in the VMID field (`0x00701430` vs `0x00501430`); the client (SQC data, 0xa) and `PERMISSION_FAULTS: 0x3` are identical. **Outcome:** the ring reset succeeds both times and the GPU recovers — the browser just restarts its GPU process. This differs from LP #2147367 (gfx1103), where recovery failed. I have an `amdgpu` devcoredump (6 MB) from the first fault and can attach it or run further tests on gfx1152 if that is useful. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2125697 Title: amdgpu crashes with page fault To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/mesa/+bug/2125697/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
