Re: [PATCH 3/7] fs, xfs: convert xfs_buf_log_item.bli_refcount from atomic_t to refcount_t

[Peter Zijlstra]

On Tue, Feb 21, 2017 at 05:27:58PM +0100, Peter Zijlstra wrote:
> > True, but in some of this cases WARN might be ok, I think? As soon as
> > functionality is not changed and object is not reused (by doing
> > refcount_inc on it) anywhere later on. 
> 
> No, your conversion should not generate spurious WARN()s.
> 
> And also no, atomic_dec() must not hit 0. If you've been _that_ careful

refcount_dec() obviously, atomic_dec() doesn't care one way or the
other.

> with your reference counting and you absolutely _know_ this is the very
> last one, write something like:
> 
>   WARN_ON(!refcount_dec_if_one());
> 
> 
> Please, stop sending out conversions that haven't been tested. And take
> the time to actually look at your own patches. If I can spot fail just
> looking through them, so can you (or any of the other many people in
> your SoB chain).
> 
> Take a little more time and a little extra care.