github-mirror/linux
1
0
Fork 0
mirror of https://github.com/torvalds/linux.git synced 2026-05-30 10:04:04 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity Graph

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

Browse Source 
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: c25c1010df ("drm/xe/vf: Replay GuC submission state on pause / unpause")
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Tomasz Lis <tomasz.lis@intel.com>
Link: https://patch.msgid.link/20251121152750.240557-1-matthew.brost@intel.com
This commit is contained in:
Matthew Brost 2025-11-21 07:27:50 -08:00
parent 2e02254ef5
commit 00937fe192
3 changed files with 19 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
drivers/gpu/drm/xe/xe_gpu_scheduler.h
View File

@ -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) static inline void xe_sched_resubmit_jobs(struct xe_gpu_scheduler *sched)
{ {
struct drm_sched_job *s_job; struct drm_sched_job *s_job;
bool restore_replay = false;
list_for_each_entry(s_job, &sched->base.pending_list, list) { list_for_each_entry(s_job, &sched->base.pending_list, list) {
struct drm_sched_fence *s_fence = s_job->s_fence; struct drm_sched_fence *s_fence = s_job->s_fence;
struct dma_fence *hw_fence = s_fence->parent; struct dma_fence *hw_fence = s_fence->parent;
if (to_xe_sched_job(s_job)->skip_emit || restore_replay |= to_xe_sched_job(s_job)->restore_replay;
(hw_fence && !dma_fence_is_signaled(hw_fence))) if (restore_replay || (hw_fence && !dma_fence_is_signaled(hw_fence)))
sched->base.ops->run_job(s_job); sched->base.ops->run_job(s_job);
} }
} }

25
drivers/gpu/drm/xe/xe_guc_submit.c
View File

@ -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)); 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)) if (xe_exec_queue_is_parallel(q))
wq_item_append(q); wq_item_append(q);
else 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 (!killed_or_banned_or_wedged && !xe_sched_job_is_error(job)) {
if (!exec_queue_registered(q)) if (!exec_queue_registered(q))
register_exec_queue(q, GUC_CONTEXT_NORMAL); register_exec_queue(q, GUC_CONTEXT_NORMAL);
if (!job->skip_emit) if (!job->restore_replay)
q->ring_ops->emit_job(job); q->ring_ops->emit_job(job);
submit_exec_queue(q, job); submit_exec_queue(q, job);
job->skip_emit = false; job->restore_replay = false;
} }
/* /*
@ -2147,6 +2147,8 @@ static void guc_exec_queue_pause(struct xe_guc *guc, struct xe_exec_queue *q)
job = xe_sched_first_pending_job(sched); job = xe_sched_first_pending_job(sched);
if (job) { if (job) {
job->restore_replay = true;
/* /*
* Adjust software tail so jobs submitted overwrite previous * Adjust software tail so jobs submitted overwrite previous
* position in ring buffer with new GGTT addresses. * position in ring buffer with new GGTT addresses.
@ -2236,17 +2238,18 @@ static void guc_exec_queue_unpause_prepare(struct xe_guc *guc,
struct xe_exec_queue *q) struct xe_exec_queue *q)
{ {
struct xe_gpu_scheduler *sched = &q->guc->sched; struct xe_gpu_scheduler *sched = &q->guc->sched;
struct drm_sched_job *s_job;
struct xe_sched_job *job = NULL; struct xe_sched_job *job = NULL;
bool restore_replay = false;
list_for_each_entry(s_job, &sched->base.pending_list, list) { list_for_each_entry(job, &sched->base.pending_list, drm.list) {
job = to_xe_sched_job(s_job); 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));
xe_gt_dbg(guc_to_gt(guc), "Replay JOB - guc_id=%d, seqno=%d", q->ring_ops->emit_job(job);
q->guc->id, xe_sched_job_seqno(job)); job->restore_replay = true;
}
q->ring_ops->emit_job(job);
job->skip_emit = true;
} }
if (job) if (job)

4
drivers/gpu/drm/xe/xe_sched_job_types.h
View File

@ -63,8 +63,8 @@ struct xe_sched_job {
bool ring_ops_flush_tlb; bool ring_ops_flush_tlb;
/** @ggtt: mapped in ggtt. */ /** @ggtt: mapped in ggtt. */
bool ggtt; bool ggtt;
/** @skip_emit: skip emitting the job */ /** @restore_replay: job being replayed for restore */
bool skip_emit; bool restore_replay;
/** @last_replay: last job being replayed */ /** @last_replay: last job being replayed */
bool last_replay; bool last_replay;
/** @ptrs: per instance pointers. */ /** @ptrs: per instance pointers. */