| Message ID | 20250121190935.1984508-1-alan.previn.teres.alexis@intel.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Maintenence of devcoredump <-> GuC-Err-Capture plumbing | expand |
On Tue, Jan 21, 2025 at 11:09:34AM -0800, Alan Previn wrote: > The GuC-Error-Capture is currently reaching into xe_devcoredump > structure to store its own place-holder snaphot to workaround > the race between G2H-Error-Capture-Notification vs Drm-Scheduler > triggering GuC-Submission-exec-queue-timeout/kill. > > Part of that race workaround design included GuC-Error-Capture taking > on some of the front-end functions for xe_hw_engine_snapshot > generation because of the orthogonal debugfs for raw dumps of engine > registers without any job association. We want this to also be handled, > even if indirectly, by GuC-Error-Capture since there is a lot to manage > when it comes to reading and printing the register lists. > > However, logically speaking, GuC-Error-Capture node management, > despite being the majority of an engine-snapshot work, is still > a subset of xe_hw_engine_snapshot. > > This series intends to re-design the plumbing for future A 'series' of 1 patch is not a series. Cover letter is not needed. However, this patch is the size of a series and it should be split. I'm really surprised that someone went over and *really* reviewed it. Even the subject of the patch doesn't make a lot of sense to me. I don't even know what to write in the pull-request with a patch like this. Please break this in small patches. > maintenence and scalability, rearranging the layering > back to what its should be (xe_devcoredump_snapshot owns > xe_hw_engine_snapshot owns xe_guc_capture_snapshot).. > > Alan Previn (1): > drm/xe/guc/capture: Maintenence of devcoredump <-> GuC-Err-Capture > plumbing > > drivers/gpu/drm/xe/xe_devcoredump.c | 3 - > drivers/gpu/drm/xe/xe_devcoredump_types.h | 6 - > drivers/gpu/drm/xe/xe_guc_capture.c | 406 ++++++++---------- > drivers/gpu/drm/xe/xe_guc_capture.h | 10 +- > .../drm/xe/xe_guc_capture_snapshot_types.h | 68 +++ > drivers/gpu/drm/xe/xe_guc_submit.c | 21 +- > drivers/gpu/drm/xe/xe_hw_engine.c | 117 +++-- > drivers/gpu/drm/xe/xe_hw_engine.h | 4 +- > drivers/gpu/drm/xe/xe_hw_engine_types.h | 13 +- > 9 files changed, 359 insertions(+), 289 deletions(-) > create mode 100644 drivers/gpu/drm/xe/xe_guc_capture_snapshot_types.h > > > base-commit: cfa9d40db8c30d894171010fe765d96e9bc6a47e > -- > 2.34.1 >
The GuC-Error-Capture is currently reaching into xe_devcoredump structure to store its own place-holder snaphot to workaround the race between G2H-Error-Capture-Notification vs Drm-Scheduler triggering GuC-Submission-exec-queue-timeout/kill. Part of that race workaround design included GuC-Error-Capture taking on some of the front-end functions for xe_hw_engine_snapshot generation because of the orthogonal debugfs for raw dumps of engine registers without any job association. We want this to also be handled, even if indirectly, by GuC-Error-Capture since there is a lot to manage when it comes to reading and printing the register lists. However, logically speaking, GuC-Error-Capture node management, despite being the majority of an engine-snapshot work, is still a subset of xe_hw_engine_snapshot. This series intends to re-design the plumbing for future maintenence and scalability, rearranging the layering back to what its should be (xe_devcoredump_snapshot owns xe_hw_engine_snapshot owns xe_guc_capture_snapshot).. Alan Previn (1): drm/xe/guc/capture: Maintenence of devcoredump <-> GuC-Err-Capture plumbing drivers/gpu/drm/xe/xe_devcoredump.c | 3 - drivers/gpu/drm/xe/xe_devcoredump_types.h | 6 - drivers/gpu/drm/xe/xe_guc_capture.c | 406 ++++++++---------- drivers/gpu/drm/xe/xe_guc_capture.h | 10 +- .../drm/xe/xe_guc_capture_snapshot_types.h | 68 +++ drivers/gpu/drm/xe/xe_guc_submit.c | 21 +- drivers/gpu/drm/xe/xe_hw_engine.c | 117 +++-- drivers/gpu/drm/xe/xe_hw_engine.h | 4 +- drivers/gpu/drm/xe/xe_hw_engine_types.h | 13 +- 9 files changed, 359 insertions(+), 289 deletions(-) create mode 100644 drivers/gpu/drm/xe/xe_guc_capture_snapshot_types.h base-commit: cfa9d40db8c30d894171010fe765d96e9bc6a47e