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: c25c1010df88 ("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
(cherry picked from commit 00937fe1921ab346b6f6a4beaa5c38e14733caa3)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

diff --git a/drivers/gpu/drm/xe/xe_gpu_scheduler.h b/drivers/gpu/drm/xe/xe_gpu_scheduler.h
index 9955397aaaa9a628eed4bc420506c1022dafa86f..c7a77a3a968193b32835e16923556387036b69ac 100644 (file)
--- a/drivers/gpu/drm/xe/xe_gpu_scheduler.h
+++ b/drivers/gpu/drm/xe/xe_gpu_scheduler.h
@@ -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);
        }
 }

diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c
index d4ffdb71ef3d71c405ebeaec745b85af2386deb1..c56fd44641f65533527e57da4389fcb78d22ae1b 100644 (file)
--- a/drivers/gpu/drm/xe/xe_guc_submit.c
+++ b/drivers/gpu/drm/xe/xe_guc_submit.c
@@ -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);
-
-               xe_gt_dbg(guc_to_gt(guc), "Replay JOB - guc_id=%d, seqno=%d",
-                         q->guc->id, xe_sched_job_seqno(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;
+                       q->ring_ops->emit_job(job);
+                       job->restore_replay = true;
+               }
        }
 
        if (job)

diff --git a/drivers/gpu/drm/xe/xe_sched_job_types.h b/drivers/gpu/drm/xe/xe_sched_job_types.h
index d26612abb4ca4f84e0a833bed2fedebda457c622..7c4c54fe920aec717ec15c099f89381320b82cc4 100644 (file)
--- a/drivers/gpu/drm/xe/xe_sched_job_types.h
+++ b/drivers/gpu/drm/xe/xe_sched_job_types.h
@@ -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. */