Fix a potential thread dead lock issue

When bigo_worker_thread does kref_put on an instance, if the refcount
changes from 1 to 0, bigo_close will be invoked. If this is the last
instance, on_last_inst_close will be invoked and it will wait for
bigo_worker_thread to exit. It is a dead loop that bigo_worker_thread
is waiting for itself to exit. The fix is to make sure the worker thread
is stopped before closing the last instance.

Bug: 229580907

Signed-off-by: Ruofei Ma <ruofeim@google.com>
Change-Id: I0950a07e072c6d10ab76389bd10ff697c3d1fc79

1 files changed, 11 insertions, 11 deletions

diff --git a/bigo.c b/bigo.c
index 8895c81..5b613bf 100644
--- a/bigo.c
+++ b/bigo.c
@@ -108,16 +108,11 @@ exit:
 
 static inline void on_last_inst_close(struct bigo_core *core)
 {
-	int rc;
 #if IS_ENABLED(CONFIG_PM)
 	if (pm_runtime_put_sync_suspend(core->dev))
 		pr_warn("failed to suspend\n");
 #endif
 	bigo_pt_client_disable(core);
-
-	rc = kthread_stop(core->worker_thread);
-	if(rc)
-		pr_err("failed to stop worker thread rc = %d\n", rc);
 }
 
 static inline int bigo_count_inst(struct bigo_core *core)
@@ -199,13 +194,8 @@ static void bigo_close(struct kref *ref)
 		return;
 	}
 	bigo_unmap_all(inst);
-	mutex_lock(&core->lock);
-	list_del(&inst->list);
 	kfree(inst->job.regs);
 	kfree(inst);
-	if (list_empty(&core->instances))
-		on_last_inst_close(core);
-	mutex_unlock(&core->lock);
 	bigo_update_qos(core);
 	pr_info("closed instance\n");
 }
@@ -213,10 +203,20 @@ static void bigo_close(struct kref *ref)
 static int bigo_release(struct inode *inode, struct file *file)
 {
 	struct bigo_inst *inst = file->private_data;
+	struct bigo_core *core = inst->core;
 
-	if (!inst)
+	if (!inst || !core)
 		return -EINVAL;
 
+	mutex_lock(&core->lock);
+	list_del(&inst->list);
+	if (list_empty(&core->instances))
+	{
+		kthread_stop(core->worker_thread);
+		on_last_inst_close(core);
+	}
+	mutex_unlock(&core->lock);
+
 	kref_put(&inst->refcount, bigo_close);
 	return 0;
 }