drm/xe: Use ordered WQ for G2H handler

System work queues are shared, use a dedicated work queue for G2H
processing to avoid G2H processing getting block behind system tasks.

Fixes: dd08ebf ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Cc: <stable@vger.kernel.org>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Francois Dugast <francois.dugast@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240506034758.3697397-1-matthew.brost@intel.com
(cherry picked from commit 50aec9665e0babd62b9eee4e613d9a1ef8d2b7de)
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drivers/gpu/drm/xe/xe_guc_ct.c

@@ -120,6 +120,7 @@ static void guc_ct_fini(struct drm_device *drm, void *arg)
 {
 	struct xe_guc_ct *ct = arg;
 
+	destroy_workqueue(ct->g2h_wq);
 	xa_destroy(&ct->fence_lookup);
 }
 
@@ -145,6 +146,10 @@ int xe_guc_ct_init(struct xe_guc_ct *ct)
 
 	xe_assert(xe, !(guc_ct_size() % PAGE_SIZE));
 
+	ct->g2h_wq = alloc_ordered_workqueue("xe-g2h-wq", 0);
+	if (!ct->g2h_wq)
+		return -ENOMEM;
+
 	spin_lock_init(&ct->fast_lock);
 	xa_init(&ct->fence_lookup);
 	INIT_WORK(&ct->g2h_worker, g2h_worker_func);

drivers/gpu/drm/xe/xe_guc_ct.h

@@ -34,7 +34,7 @@ static inline void xe_guc_ct_irq_handler(struct xe_guc_ct *ct)
 		return;
 
 	wake_up_all(&ct->wq);
-	queue_work(system_unbound_wq, &ct->g2h_worker);
+	queue_work(ct->g2h_wq, &ct->g2h_worker);
 	xe_guc_ct_fast_path(ct);
 }
 

drivers/gpu/drm/xe/xe_guc_ct_types.h

@@ -120,6 +120,8 @@ struct xe_guc_ct {
 	wait_queue_head_t wq;
 	/** @g2h_fence_wq: wait queue used for G2H fencing */
 	wait_queue_head_t g2h_fence_wq;
+	/** @g2h_wq: used to process G2H */
+	struct workqueue_struct *g2h_wq;
 	/** @msg: Message buffer */
 	u32 msg[GUC_CTB_MSG_MAX_LEN];
 	/** @fast_msg: Message buffer */