Re: [tip:locking/core] locking/refcount: Create unchecked atomic_t implementation

[Peter Zijlstra]

On Mon, Sep 04, 2017 at 10:11:37AM -0700, Kees Cook wrote:
> On Mon, Sep 4, 2017 at 5:37 AM, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > On Wed, Jun 28, 2017 at 09:58:15AM -0700, tip-bot for Kees Cook wrote:
> >> locking/refcount: Create unchecked atomic_t implementation
> >
> > This seems to do only half the job. Here's the rest.
> >
> > ---
> > Subject: locking/refcount: Finish unchecked atomic_t implementation
> >
> > For some reason the unchecked atomic_t implementation stopped half-way
> > through, complete it it.
> 
> Hmm? The reason is that the implementation of the remaining functions
> is unchanged between full, unchecked, and x86.

But they're wasted code if !arch because the existing atomic functions
are adequate (and I would argue better in case of atomic_add_unless).

And arch implementations would certainly want to reimplement dec_not_one.

Plus, you completely failed mention any of this.