diff --git a/techpack/display/msm/msm_atomic.c b/techpack/display/msm/msm_atomic.c
index eb92af6caea5..887c07ac3789 100644
--- a/techpack/display/msm/msm_atomic.c
+++ b/techpack/display/msm/msm_atomic.c
@@ -116,18 +116,6 @@ static inline bool _msm_seamless_for_conn(struct drm_connector *connector,
 /* clear specified crtcs (no longer pending update) */
 static void commit_destroy(struct msm_commit *c)
 {
-	struct msm_drm_private *priv = c->dev->dev_private;
-	uint32_t crtc_mask = c->crtc_mask;
-	uint32_t plane_mask = c->plane_mask;
-
-	/* End_atomic */
-	spin_lock(&priv->pending_crtcs_event.lock);
-	DBG("end: %08x", crtc_mask);
-	priv->pending_crtcs &= ~crtc_mask;
-	priv->pending_planes &= ~plane_mask;
-	wake_up_all_locked(&priv->pending_crtcs_event);
-	spin_unlock(&priv->pending_crtcs_event.lock);
-
 	if (c->nonblock)
 		kfree(c);
 }
@@ -469,6 +457,16 @@ static void msm_atomic_helper_commit_modeset_enables(struct drm_device *dev,
 	SDE_ATRACE_END("msm_enable");
 }
 
+static void complete_commit_cleanup(struct kthread_work *work)
+{
+	struct msm_commit *c = container_of(work, typeof(*c), commit_work);
+	struct drm_atomic_state *state = c->state;
+
+	drm_atomic_state_put(state);
+
+	commit_destroy(c);
+}
+
 int msm_atomic_prepare_fb(struct drm_plane *plane,
 			  struct drm_plane_state *new_state)
 {
@@ -505,6 +503,8 @@ static void complete_commit(struct msm_commit *c)
 	struct drm_device *dev = state->dev;
 	struct msm_drm_private *priv = dev->dev_private;
 	struct msm_kms *kms = priv->kms;
+	uint32_t crtc_mask = c->crtc_mask;
+	uint32_t plane_mask = c->plane_mask;
 
 	drm_atomic_helper_wait_for_fences(dev, state, false);
 
@@ -536,35 +536,43 @@ static void complete_commit(struct msm_commit *c)
 
 	kms->funcs->complete_commit(kms, state);
 
-	drm_atomic_state_put(state);
-
 	priv->complete_commit_time = ktime_get()/1000;
 
 	complete_time_generate_event(dev);
 
-	commit_destroy(c);
+	/* End_atomic */
+	spin_lock(&priv->pending_crtcs_event.lock);
+	DBG("end: %08x", crtc_mask);
+	priv->pending_crtcs &= ~crtc_mask;
+	priv->pending_planes &= ~plane_mask;
+	wake_up_all_locked(&priv->pending_crtcs_event);
+	spin_unlock(&priv->pending_crtcs_event.lock);
 }
 
 static void _msm_drm_commit_work_cb(struct kthread_work *work)
 {
-	struct msm_commit *commit = NULL;
+	struct msm_commit *c = container_of(work, typeof(*c), commit_work);
+	struct drm_atomic_state *state = c->state;
+	struct drm_device *dev = state->dev;
+	struct msm_drm_private *priv = dev->dev_private;
 	ktime_t start, end;
 	s64 duration;
 
-	if (!work) {
-		DRM_ERROR("%s: Invalid commit work data!\n", __func__);
-		return;
-	}
-
-	commit = container_of(work, struct msm_commit, commit_work);
-
 	start = ktime_get();
 	frame_stat_collector(0, COMMIT_START_TS);
 
 	SDE_ATRACE_BEGIN("complete_commit");
-	complete_commit(commit);
+	complete_commit(c);
 	SDE_ATRACE_END("complete_commit");
 
+	if (c->nonblock) {
+		/* Offload the cleanup onto little CPUs */
+		kthread_init_work(&c->commit_work, complete_commit_cleanup);
+		kthread_queue_work(&priv->clean_thread.worker, &c->commit_work);
+	} else {
+		complete_commit_cleanup(&c->commit_work);
+	}
+
 	end = ktime_get();
 	duration = ktime_to_ns(ktime_sub(end, start));
 	frame_stat_collector(duration, COMMIT_END_TS);
@@ -643,6 +651,7 @@ static void msm_atomic_commit_dispatch(struct drm_device *dev,
 		 * ensure that SW and HW state don't get out of sync.
 		 */
 		complete_commit(commit);
+		complete_commit_cleanup(&commit->commit_work);
 	} else if (!nonblock) {
 		kthread_flush_work(&commit->commit_work);
 	}
diff --git a/techpack/display/msm/msm_drv.c b/techpack/display/msm/msm_drv.c
index a9b644f2d55c..6aea0bf1152c 100644
--- a/techpack/display/msm/msm_drv.c
+++ b/techpack/display/msm/msm_drv.c
@@ -377,6 +377,8 @@ static int msm_drm_uninit(struct device *dev)
 		}
 	}
 
+	kthread_stop(priv->clean_thread.thread);
+
 	drm_kms_helper_poll_fini(ddev);
 
 	drm_mode_config_cleanup(ddev);
@@ -553,6 +555,11 @@ static int msm_drm_display_thread_create(struct sched_param param,
 {
 	int i, ret = 0;
 
+	kthread_init_worker(&priv->clean_thread.worker);
+	priv->clean_thread.thread = kthread_run_perf_critical(cpu_lp_mask,
+		kthread_worker_fn, &priv->clean_thread.worker, "drm_cleanup");
+	BUG_ON(IS_ERR(priv->clean_thread.thread));
+
 	/**
 	 * this priority was found during empiric testing to have appropriate
 	 * realtime scheduling to process display updates and interact with
diff --git a/techpack/display/msm/msm_drv.h b/techpack/display/msm/msm_drv.h
index ac106b0a19fc..22810e4fb0dc 100644
--- a/techpack/display/msm/msm_drv.h
+++ b/techpack/display/msm/msm_drv.h
@@ -667,6 +667,11 @@ struct msm_drm_private {
 	struct msm_drm_thread disp_thread[MAX_CRTCS];
 	struct msm_drm_thread event_thread[MAX_CRTCS];
 
+	struct {
+		struct task_struct *thread;
+		struct kthread_worker worker;
+	} clean_thread;
+
 	struct task_struct *pp_event_thread;
 	struct kthread_worker pp_event_worker;