Re: Q: lockdep_assert_held_read() after downgrade_write()

[Peter Zijlstra]

On Mon, Jan 30, 2017 at 02:30:45PM -0700, Jens Axboe wrote:
> On 01/30/2017 02:25 PM, J. R. Okajima wrote:
> > Peter Zijlstra,
> > 
> > May I ask you a question?
> > v4.10-rc1 got a commit
> > 	f831948 2016-11-30 locking/lockdep: Provide a type check for lock_is_held
> > I've tested a little and lockdep splat a stack trace.
> > 
> > {
> > 	DECLARE_RWSEM(rw);
> > 	static struct lock_class_key key;
> > 	lockdep_set_class(&rw, &key);
> > 
> > 	down_read(&rw);
> > 	lockdep_assert_held_read(&rw);
> > 	up_read(&rw);
> > 
> > 	down_write(&rw);
> > 	lockdep_assert_held_exclusive(&rw);
> > 	up_write(&rw);
> > 
> > 	downgrade_write(&rw);
> > 	lockdep_assert_held_read(&rw);	<-- here
> > 	up_read(&rw);
> > }
> > 
> > I was expecting that lockdep_assert_held_read() splat nothing after
> > downgrade_write(). Is this warning an intentional behaviour?
> > 
> > Also the final up_read() gives me a warning too. It is produced at
> > 	lockdep.c:3514:lock_release(): DEBUG_LOCKS_WARN_ON(depth <= 0)
> 
> I don't think you understand how it works. downgrade_write() turns a write
> lock into read held. To make that last sequence valid, you'd need:

Correct, and I'm surprised that didn't explode in different ways.

> 
> 	down_write(&rw);
> 	downgrade_write(&rw);
> 	lockdep_assert_held_read(&rw)
> 	up_read(&rw);
> 
> or just not drop up_write() from the last section.

Right, but also, there seems to be a missing lockdep annotation to make
that work. That is, downgrade_write() doesn't have a lockdep annotation,
so it (lockdep) will still think its a write lock.


Let me try and fix both issues.