1bd6a2efe1
Xe KMD originally put the exec queue snapshot in the wrong topic, XE_TOPIC_GUC_CT. Add it to the right one while still keeping a fallback to the previous place so the new version of the tool is able to parse both before and after the kernel change. Based on previous version by José Roberto de Souza <jose.souza@intel.com> at https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32617 Cc: stable Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com> Reviewed-by: José Roberto de Souza <jose.souza@intel.com> Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33177> |
||
|---|---|---|
| .. | ||
| imgui | ||
| aub_mem.c | ||
| aub_mem.h | ||
| aub_read.c | ||
| aub_read.h | ||
| aub_write.c | ||
| aub_write.h | ||
| aubinator.c | ||
| aubinator_error_decode.c | ||
| aubinator_error_decode_lib.c | ||
| aubinator_error_decode_lib.h | ||
| aubinator_error_decode_xe.c | ||
| aubinator_error_decode_xe.h | ||
| aubinator_viewer.cpp | ||
| aubinator_viewer.h | ||
| aubinator_viewer_decoder.cpp | ||
| aubinator_viewer_urb.h | ||
| error2aub.c | ||
| error2hangdump.c | ||
| error2hangdump_lib.c | ||
| error2hangdump_lib.h | ||
| error2hangdump_xe.c | ||
| error2hangdump_xe.h | ||
| error_decode_lib.c | ||
| error_decode_lib.h | ||
| error_decode_xe_lib.c | ||
| error_decode_xe_lib.h | ||
| gfx8_context.h | ||
| gfx10_context.h | ||
| intel_aub.h | ||
| intel_context.h | ||
| intel_dev_info.c | ||
| intel_dump_gpu.c | ||
| intel_dump_gpu.in | ||
| intel_hang_replay.c | ||
| intel_hang_viewer.cpp | ||
| intel_noop_drm_shim.c | ||
| intel_sanitize_gpu.c | ||
| intel_sanitize_gpu.in | ||
| intel_stub_gpu.in | ||
| meson.build | ||
| README.md | ||
aubinator_error_decode
This tool allows decode an error state dump generated by the kernel driver.
Make a copy of the error state file (i915) :
$ sudo cp /sys/devices/pci0000:00/0000:00:02.0/drm/card0/error /tmp/error
$ sudo chmod go+rw /tmp/error
Make a copy of the error state file (Xe) :
$ sudo cp /sys/class/drm/card0/device/devcoredump/data /tmp/error
$ sudo chmod go+rw /tmp/error
Then visualize the error state (from the mesa build directory) :
$ ./src/intel/tools/aubinator_error_decode /tmp/error
intel_stub_gpu
This tool allows faking a particular device at the ioctl level. This is useful to test compiler changes (using shader-db/fossils-db replays) across a number of device you don't necessarily have access too.
Faking a DG2 device :
$ intel_stub_gpu -p dg2 fossilize-replay /path/to/fossilize.foz --enable-pipeline-stats /tmp/dg2.csv
Faking a ICL device :
$ intel_stub_gpu -p icl fossilize-replay /path/to/fossilize.foz --enable-pipeline-stats /tmp/icl.csv
intel_error2hangdump
This tool converts an error state dump into an
uncompressed/easy-to-modify binary file (format described in
src/intel/common/intel_hang_dump.h).
Capture the error state (i915) :
$ sudo cp /sys/devices/pci0000:00/0000:00:02.0/drm/card0/error /tmp/error
$ sudo chmod go+rw /tmp/error
Capture the error state (Xe) :
$ sudo cp /sys/class/drm/card0/device/devcoredump/data /tmp/error
$ sudo chmod go+rw /tmp/error
Convert an error state (from the mesa build directory) :
$ ./src/intel/tools/intel_error2hang /tmp/error
A resulting error.dmp file will be created
intel_hang_replay
This tool replays a binary file generated by intel_error2hangdump.
From the mesa build directory :
$ ./src/intel/tools/intel_hang_replay -d /tmp/error.dmp
intel_hang_viewer
This tool is a editor/visualizer for the file generated by
intel_error2hangdump. When replaying on simulation a few tweaks
might be needed in order for the simulator to accept the input
commands (this is due to divergences between the simulation & HW
implementations).
To visualize a file generated from a DG2 error state (from the mesa build directory) :
./src/intel/tools/intel_hang_viewer -p dg2 /tmp/error.dmp
To visualize and edit a file generated from a DG2 error state (from the mesa build directory) :
./src/intel/tools/intel_hang_viewer -e -p dg2 /tmp/error.dmp