drm/msm: Stop dropping struct_mutex in recover_worker()
authorKristian H. Kristensen <hoegsberg@gmail.com>
Wed, 20 Mar 2019 17:09:09 +0000 (10:09 -0700)
committerRob Clark <robdclark@chromium.org>
Fri, 19 Apr 2019 18:50:06 +0000 (11:50 -0700)
Now that we don't have the mmap_sem lock inversion, we don't need to
jump through this particular hoop anymore.

Signed-off-by: Kristian H. Kristensen <hoegsberg@chromium.org>
Signed-off-by: Rob Clark <robdclark@chromium.org>
drivers/gpu/drm/msm/msm_gpu.c

index 2b76657..bf4ee27 100644 (file)
@@ -450,21 +450,8 @@ static void recover_worker(struct work_struct *work)
                task = get_pid_task(submit->pid, PIDTYPE_PID);
                if (task) {
                        comm = kstrdup(task->comm, GFP_KERNEL);
-
-                       /*
-                        * So slightly annoying, in other paths like
-                        * mmap'ing gem buffers, mmap_sem is acquired
-                        * before struct_mutex, which means we can't
-                        * hold struct_mutex across the call to
-                        * get_cmdline().  But submits are retired
-                        * from the same in-order workqueue, so we can
-                        * safely drop the lock here without worrying
-                        * about the submit going away.
-                        */
-                       mutex_unlock(&dev->struct_mutex);
                        cmd = kstrdup_quotable_cmdline(task, GFP_KERNEL);
                        put_task_struct(task);
-                       mutex_lock(&dev->struct_mutex);
                }
 
                if (comm && cmd) {