workqueue: re-add lockdep dependencies for flushing

In flush_work(), we need to create a lockdep dependency so that
the following scenario is appropriately tagged as a problem:

  work_function()
  {
    mutex_lock(&mutex);
    ...
  }

  other_function()
  {
    mutex_lock(&mutex);
    flush_work(&work); // or cancel_work_sync(&work);
  }

This is a problem since the work might be running and be blocked
on trying to acquire the mutex.

Similarly, in flush_workqueue().

These were removed after cross-release partially caught these
problems, but now cross-release was reverted anyway. IMHO the
removal was erroneous anyway though, since lockdep should be
able to catch potential problems, not just actual ones, and
cross-release would only have caught the problem when actually
invoking wait_for_completion().

Fixes: fd1a5b04df ("workqueue: Remove now redundant lock acquisitions wrt. workqueue flushes")
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
This commit is contained in:
Johannes Berg 2018-08-22 11:49:04 +02:00 committed by Tejun Heo
parent d6e89786be
commit 87915adc3f

View File

@ -2652,6 +2652,9 @@ void flush_workqueue(struct workqueue_struct *wq)
if (WARN_ON(!wq_online))
return;
lock_map_acquire(&wq->lockdep_map);
lock_map_release(&wq->lockdep_map);
mutex_lock(&wq->mutex);
/*
@ -2905,6 +2908,11 @@ static bool __flush_work(struct work_struct *work, bool from_cancel)
if (WARN_ON(!wq_online))
return false;
if (!from_cancel) {
lock_map_acquire(&work->lockdep_map);
lock_map_release(&work->lockdep_map);
}
if (start_flush_work(work, &barr, from_cancel)) {
wait_for_completion(&barr.done);
destroy_work_on_stack(&barr.work);