[PATCH, RFC] xfs: remove i_iolock and use i_rwsem in the VFS inode instead

[Peter Zijlstra]

On Sat, Aug 20, 2016 at 08:37:23AM +0200, Christoph Hellwig wrote:
> On Fri, Aug 19, 2016 at 03:27:36PM +0200, Peter Zijlstra wrote:
> > Hurm, if you're going to directly use that maybe we should pick a better
> > name ;-)
> 
> Fine with that.
> 
> > Also, be sure to check the debug_locks variable, if that's cleared the
> > result of _lockdep_is_held() isn't reliable -- we stop tracking lock
> > state when there's an error.
> 
> I already do.  But I'm wondering if we can't simply move the 
> debug_locks check into lockdep_is_held?  It's already used directly
> in a few places, and that would also solve the whole naming issue.

Reason I didn't do that initially was that I used lock_is_held() for
both positive and negative tests (ie. assert a lock is held and a lock
is not held). Given that, you cannot pick a right return value when
!debug_locks.

Not sure we still do that, but I distinctly remember running into that
when I did as you suggest now. But that was many years ago.