[PATCH v3 3/5] drm/xe/guc: Add suspend refcount to exec queue ops

["Thomas Hellström" <thomas.hellstrom@linux.intel.com>]

Concurrent call sites that suspend exec queues (PM, mode switching,
preempt fences) each need to pair their own suspend/resume without
caring about other callers. Without coordination, a resume from one
path can prematurely re-enable a queue still suspended by another.

Add a reference count to exec queue suspend operations. A queue remains
suspended as long as any caller holds a suspend; it resumes only when
the last caller releases it. Each caller simply pairs its own suspend
with a resume without needing to know about others.

Assisted-by: GitHub_Copilot:claude-sonnet-4.6
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
---
 drivers/gpu/drm/xe/xe_guc_exec_queue_types.h |  7 ++++++
 drivers/gpu/drm/xe/xe_guc_submit.c           | 24 ++++++++++++--------
 2 files changed, 22 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h b/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h
index e5e53b421f29..1207d51cf770 100644
--- a/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h
+++ b/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h
@@ -49,6 +49,13 @@ struct xe_guc_exec_queue {
 	wait_queue_head_t suspend_wait;
 	/** @suspend_pending: a suspend of the exec_queue is pending */
 	bool suspend_pending;
+	/**
+	 * @suspend_count: Reference count of active suspend requests. The
+	 * exec_queue remains suspended while this is non-zero, allowing
+	 * multiple concurrent callers to independently hold a suspend without
+	 * prematurely re-enabling the queue. Protected by @sched.msg_lock.
+	 */
+	int suspend_count;
 	/**
 	 * @needs_cleanup: Needs a cleanup message during VF post migration
 	 * recovery.
diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c
index f1a6f13011b5..7800a9d1d377 100644
--- a/drivers/gpu/drm/xe/xe_guc_submit.c
+++ b/drivers/gpu/drm/xe/xe_guc_submit.c
@@ -2131,15 +2131,18 @@ static int guc_exec_queue_set_multi_queue_priority(struct xe_exec_queue *q,
 
 static int guc_exec_queue_suspend(struct xe_exec_queue *q)
 {
-	struct xe_gpu_scheduler *sched = &q->guc->sched;
-	struct xe_sched_msg *msg = q->guc->static_msgs + STATIC_MSG_SUSPEND;
+	struct xe_guc_exec_queue *ge = q->guc;
+	struct xe_gpu_scheduler *sched = &ge->sched;
+	struct xe_sched_msg *msg = ge->static_msgs + STATIC_MSG_SUSPEND;
 
 	if (exec_queue_killed_or_banned_or_wedged(q))
 		return -EINVAL;
 
 	xe_sched_msg_lock(sched);
-	if (guc_exec_queue_try_add_msg(q, msg, SUSPEND))
-		q->guc->suspend_pending = true;
+	if (++ge->suspend_count == 1) {
+		if (guc_exec_queue_try_add_msg(q, msg, SUSPEND))
+			ge->suspend_pending = true;
+	}
 	xe_sched_msg_unlock(sched);
 
 	return 0;
@@ -2190,14 +2193,17 @@ static int guc_exec_queue_suspend_wait(struct xe_exec_queue *q)
 
 static void guc_exec_queue_resume(struct xe_exec_queue *q)
 {
-	struct xe_gpu_scheduler *sched = &q->guc->sched;
-	struct xe_sched_msg *msg = q->guc->static_msgs + STATIC_MSG_RESUME;
+	struct xe_guc_exec_queue *ge = q->guc;
+	struct xe_gpu_scheduler *sched = &ge->sched;
+	struct xe_sched_msg *msg = ge->static_msgs + STATIC_MSG_RESUME;
 	struct xe_guc *guc = exec_queue_to_guc(q);
 
-	xe_gt_assert(guc_to_gt(guc), !q->guc->suspend_pending);
-
 	xe_sched_msg_lock(sched);
-	guc_exec_queue_try_add_msg(q, msg, RESUME);
+	xe_gt_assert(guc_to_gt(guc), ge->suspend_count > 0);
+	if (--ge->suspend_count == 0) {
+		xe_gt_assert(guc_to_gt(guc), !ge->suspend_pending);
+		guc_exec_queue_try_add_msg(q, msg, RESUME);
+	}
 	xe_sched_msg_unlock(sched);
 }
 
-- 
2.54.0