1:orphan: 2 3.. _radv-debug-hang: 4 5Debugging GPU hangs with RADV 6============================= 7 8UMR (optional) 9-------------- 10 11UMR is needed for dumping a lot of useful information. Clone, build and install 12`UMR <https://gitlab.freedesktop.org/tomstdenis/umr>`__. Do not forget to run 13``chmod +s $(which umr)`` so RADV can actually access UMR. 14 15UMR needs to access some kernel debug interfaces: 16 17.. code-block:: sh 18 19 chmod 777 /sys/kernel/debug 20 chmod -R 777 /sys/kernel/debug/dri 21 22Secure boot has to be disabled as well. 23 24Generating and analyzing hang reports 25------------------------------------- 26 27With UMR installed, you can now set ``RADV_DEBUG=hang`` which makes RADV insert 28trace markers and synchronization and check for hangs. The hang report will be 29saved to ``~/radv_dumps_<pid>_<time>``. Inside the directory of the hang report, 30there are a couple of files: 31 32* ``*.spv``: SPIR-V binaries of the pipeline that was bound when the hang 33 occurred. 34* ``addr_binding_report.log``: VK_EXT_address_binding_report logs. 35* ``app_info.log``: ``VkApplicationInfo`` fields. 36* ``bo_history.log``: A list of every GPU memory allocation and deallocation. 37 If the GPU hang was caused by a page fault, you can use 38 `radv_check_va.py <https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/amd/vulkan/radv_check_va.py>`__ 39 to figure out if address is invalid or used after the memory was deallocated. 40* ``bo_ranges.log``: Address ranges that were valid at the time of submission. 41* ``dmesg.log``: Output of ``dmesg``, if available. 42* ``gpu_info.log``: Fields of ``radeon_info``. 43* ``pipeline.log``: IR of the shaders that were bound during the hang as well as 44 programm counters of waves executing said shaders and bound descriptors. 45* ``registers.log``: Various GPU state registers. 46* ``trace.log``: An annotated list of the command stream that caused the hang. 47 the commands that hung come after 48 ``!!!!! This is the last trace point that was reached by the CP !!!!!``. 49* ``umr_ring.log``: Similar to ``trace.log``. 50* ``umr_waves.log``: A list of waves that were active at the time of the hang, 51 including register values. 52* ``vm_fault.log``: The page fault address if a page fault occurred. 53 54Note: By default, the backend IR (ACO or LLVM) and the disassembly should be 55dumped to ``pipeline.log``. But due to shaders caching, you might need 56``RADV_DEBUG=hang,nocache`` to get SPIR-V and NIR in the GPU hang report. 57 58Debugging Steam games 59--------------------- 60 61Steam games require a bit more work so RADV can access UMR: In your Steam library, 62make sure **Tools** is checked and search for **Steam Linux Runtime**. 63Under **Properties** -> **Installed Files**, click **Browse**, open 64``_v2-entry-point`` and add 65 66.. code-block:: sh 67 68 shift 2 69 exec "$@" 70 71at the top of the file. Hang debugging can be enabled by selecting the faulting 72game and adding ``RADV_DEBUG=hang %command%`` under **Properties** -> **General** 73-> **LAUNCH OPTIONS**. 74 75Debugging hangs without RADV_DEBUG=hang 76--------------------------------------- 77 78In some situations, ``RADV_DEBUG=hang`` wouldn't be able to generate a GPU hang 79report, like for synchronization issues (because it enables 80``RADV_DEBUG=syncshaders`` behind the scene). An alternative solution is to 81disable GPU recovery by adding ``amdgpu.gpu_recovery=0`` to your kernel command 82line options. And then invoke UMR manually with 83``umr --by-pci <pci_id> -O bits,halt_waves -go 0 -wa <ring> -go 1 2>&1`` for 84dumping the waves and ``umr --by-pci <pci_id> -RS <ring> 2>&1`` for dumping the 85rings once the GPU hang occurred. 86