Commit 3d98a716 authored by Matthew Brost's avatar Matthew Brost Committed by Thomas Hellström
Browse files

drm/xe/vf: Start re-emission from first unsignaled job during VF migration 



The LRC software ring tail is reset to the first unsignaled pending
job's head.

Fix the re-emission logic to begin submitting from the first unsignaled
job detected, rather than scanning all pending jobs, which can cause
imbalance.

v2:
 - Include missing local changes
v3:
 - s/skip_replay/restore_replay (Tomasz)

Fixes: c25c1010 ("drm/xe/vf: Replay GuC submission state on pause / unpause")
Signed-off-by: default avatarMatthew Brost <matthew.brost@intel.com>
Reviewed-by: default avatarTomasz Lis <tomasz.lis@intel.com>
Link: https://patch.msgid.link/20251121152750.240557-1-matthew.brost@intel.com


(cherry picked from commit 00937fe1)
Signed-off-by: default avatarThomas Hellström <thomas.hellstrom@linux.intel.com>
parent 14a8d83c

drivers/gpu/drm/xe/xe_gpu_scheduler.h

+3 −2
Original line number Diff line number Diff line
@@ -54,13 +54,14 @@ static inline void xe_sched_tdr_queue_imm(struct xe_gpu_scheduler *sched) 
static inline void xe_sched_resubmit_jobs(struct xe_gpu_scheduler *sched)

{

	struct drm_sched_job *s_job;

	bool restore_replay = false;



	list_for_each_entry(s_job, &sched->base.pending_list, list) {

		struct drm_sched_fence *s_fence = s_job->s_fence;

		struct dma_fence *hw_fence = s_fence->parent;



		if (to_xe_sched_job(s_job)->skip_emit ||

		    (hw_fence && !dma_fence_is_signaled(hw_fence)))

		restore_replay |= to_xe_sched_job(s_job)->restore_replay;

		if (restore_replay || (hw_fence && !dma_fence_is_signaled(hw_fence)))

			sched->base.ops->run_job(s_job);

	}

}

drivers/gpu/drm/xe/xe_guc_submit.c

+14 −11
Original line number Diff line number Diff line
@@ -822,7 +822,7 @@ static void submit_exec_queue(struct xe_exec_queue *q, struct xe_sched_job *job) 


	xe_gt_assert(guc_to_gt(guc), exec_queue_registered(q));



	if (!job->skip_emit || job->last_replay) {

	if (!job->restore_replay || job->last_replay) {

		if (xe_exec_queue_is_parallel(q))

			wq_item_append(q);

		else
@@ -881,10 +881,10 @@ guc_exec_queue_run_job(struct drm_sched_job *drm_job) 
	if (!killed_or_banned_or_wedged && !xe_sched_job_is_error(job)) {

		if (!exec_queue_registered(q))

			register_exec_queue(q, GUC_CONTEXT_NORMAL);

		if (!job->skip_emit)

		if (!job->restore_replay)

			q->ring_ops->emit_job(job);

		submit_exec_queue(q, job);

		job->skip_emit = false;

		job->restore_replay = false;

	}



	/*
@@ -2152,6 +2152,8 @@ static void guc_exec_queue_pause(struct xe_guc *guc, struct xe_exec_queue *q) 


	job = xe_sched_first_pending_job(sched);

	if (job) {

		job->restore_replay = true;



		/*

		 * Adjust software tail so jobs submitted overwrite previous

		 * position in ring buffer with new GGTT addresses.
@@ -2241,17 +2243,18 @@ static void guc_exec_queue_unpause_prepare(struct xe_guc *guc, 
					   struct xe_exec_queue *q)

{

	struct xe_gpu_scheduler *sched = &q->guc->sched;

	struct drm_sched_job *s_job;

	struct xe_sched_job *job = NULL;

	bool restore_replay = false;



	list_for_each_entry(s_job, &sched->base.pending_list, list) {

		job = to_xe_sched_job(s_job);



	list_for_each_entry(job, &sched->base.pending_list, drm.list) {

		restore_replay |= job->restore_replay;

		if (restore_replay) {

			xe_gt_dbg(guc_to_gt(guc), "Replay JOB - guc_id=%d, seqno=%d",

				  q->guc->id, xe_sched_job_seqno(job));



			q->ring_ops->emit_job(job);

		job->skip_emit = true;

			job->restore_replay = true;

		}

	}



	if (job)

drivers/gpu/drm/xe/xe_sched_job_types.h

+2 −2
Original line number Diff line number Diff line
@@ -63,8 +63,8 @@ struct xe_sched_job { 
	bool ring_ops_flush_tlb;

	/** @ggtt: mapped in ggtt. */

	bool ggtt;

	/** @skip_emit: skip emitting the job */

	bool skip_emit;

	/** @restore_replay: job being replayed for restore */

	bool restore_replay;

	/** @last_replay: last job being replayed */

	bool last_replay;

	/** @ptrs: per instance pointers. */