Re: [PATCH 00/18] locking/atomic: atomic64 type cleanup

[Andrea Parri]

On Tue, May 28, 2019 at 12:47:19PM +0200, Peter Zijlstra wrote:
> On Sat, May 25, 2019 at 12:43:40AM +0200, Andrea Parri wrote:
> > > ---
> > > Subject: Documentation/atomic_t.txt: Clarify pure non-rmw usage
> > > 
> > > Clarify that pure non-RMW usage of atomic_t is pointless, there is
> > > nothing 'magical' about atomic_set() / atomic_read().
> > > 
> > > This is something that seems to confuse people, because I happen upon it
> > > semi-regularly.
> > > 
> > > Acked-by: Will Deacon <will.deacon@xxxxxxx>
> > > Reviewed-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> > > Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> > > ---
> > >  Documentation/atomic_t.txt | 6 ++++--
> > >  1 file changed, 4 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/Documentation/atomic_t.txt b/Documentation/atomic_t.txt
> > > index dca3fb0554db..89eae7f6b360 100644
> > > --- a/Documentation/atomic_t.txt
> > > +++ b/Documentation/atomic_t.txt
> > > @@ -81,9 +81,11 @@ SEMANTICS
> > >  
> > >  The non-RMW ops are (typically) regular LOADs and STOREs and are canonically
> > >  implemented using READ_ONCE(), WRITE_ONCE(), smp_load_acquire() and
> > > -smp_store_release() respectively.
> > > +smp_store_release() respectively. Therefore, if you find yourself only using
> > > +the Non-RMW operations of atomic_t, you do not in fact need atomic_t at all
> > > +and are doing it wrong.
> > 
> > The counterargument (not so theoretic, just look around in the kernel!) is:
> > we all 'forget' to use READ_ONCE() and WRITE_ONCE(), it should be difficult
> > or more difficult to forget to use atomic_read() and atomic_set()...   IAC,
> > I wouldn't call any of them 'wrong'.
> 
> I'm thinking you mean that the type system isn't helping us with
> READ/WRITE_ONCE() like it does with atomic_t ?

Yep.


> And while I agree that
> there is room for improvement there, that doesn't mean we should start
> using atomic*_t all over the place for that.

Agreed.  But this still doesn't explain that "and are doing it wrong",
AFAICT; maybe just remove that part?

  Andrea


> 
> Part of the problem with READ/WRITE_ONCE() is that it serves a dual
> purpose; we've tried to untangle that at some point, but Linus wasn't
> having it.