On Wed, 13 Oct 2021 09:50:17 +0800
王贇 [off-list ref] wrote:

quoted

quoted

-	preempt_enable_notrace();
 	ftrace_test_recursion_unlock(bit);
 }  

I don't like this change much. We have preempt_disable there not because 
of ftrace_test_recursion, but because of RCU. ftrace_test_recursion was 
added later. Yes, it would work with the change, but it would also hide 
things which should not be hidden in my opinion.  

Not very sure about the backgroup stories, but just found this in
'Documentation/trace/ftrace-uses.rst':

  Note, on success,
  ftrace_test_recursion_trylock() will disable preemption, and the
  ftrace_test_recursion_unlock() will enable it again (if it was previously
  enabled).

Right that part is to be fixed by what you are adding here.

The point that Miroslav is complaining about is that the preemption
disabling is special in this case, and not just from the recursion
point of view, which is why the comment is still required.

-- Steve

Seems like this lock pair was supposed to take care the preemtion itself?