Re: [PATCH 6/6] Documentation/memory-barriers.txt: Clarify cmpxchg()

[Manfred Spraul]

This is a multi-part message in MIME format.
Hello Peter,

On 10/14/19 3:03 PM, Peter Zijlstra wrote:
> On Sat, Oct 12, 2019 at 07:49:58AM +0200, Manfred Spraul wrote:
> > The documentation in memory-barriers.txt claims that
> > smp_mb__{before,after}_atomic() are for atomic ops that do not return a
> > value.
> >
> > This is misleading and doesn't match the example in atomic_t.txt,
> > and e.g. smp_mb__before_atomic() may and is used together with
> > cmpxchg_relaxed() in the wake_q code.
> >
> > The purpose of e.g. smp_mb__before_atomic() is to "upgrade" a following
> > RMW atomic operation to a full memory barrier.
> > The return code of the atomic operation has no impact, so all of the
> > following examples are valid:
> The value return of atomic ops is relevant in so far that
> (traditionally) all value returning atomic ops already implied full
> barriers. That of course changed when we added
> _release/_acquire/_relaxed variants.
I've updated the Change description accordingly
> > 1)
> > 	smp_mb__before_atomic();
> > 	atomic_add();
> >
> > 2)
> > 	smp_mb__before_atomic();
> > 	atomic_xchg_relaxed();
> >
> > 3)
> > 	smp_mb__before_atomic();
> > 	atomic_fetch_add_relaxed();
> >
> > Invalid would be:
> > 	smp_mb__before_atomic();
> > 	atomic_set();
> >
> > Signed-off-by: Manfred Spraul <manfred@xxxxxxxxxxxxxxxx>
> > Cc: Waiman Long <longman@xxxxxxxxxx>
> > Cc: Davidlohr Bueso <dave@xxxxxxxxxxxx>
> > Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> > ---
> >   Documentation/memory-barriers.txt | 11 ++++++-----
> >   1 file changed, 6 insertions(+), 5 deletions(-)
> >
> > diff --git a/Documentation/memory-barriers.txt b/Documentation/memory-barriers.txt
> > index 1adbb8a371c7..52076b057400 100644
> > --- a/Documentation/memory-barriers.txt
> > +++ b/Documentation/memory-barriers.txt
> > @@ -1873,12 +1873,13 @@ There are some more advanced barrier functions:
> >    (*) smp_mb__before_atomic();
> >    (*) smp_mb__after_atomic();
> >   
> > -     These are for use with atomic (such as add, subtract, increment and
> > -     decrement) functions that don't return a value, especially when used for
> > -     reference counting.  These functions do not imply memory barriers.
> > +     These are for use with atomic RMW functions (such as add, subtract,
> > +     increment, decrement, failed conditional operations, ...) that do
> > +     not imply memory barriers, but where the code needs a memory barrier,
> > +     for example when used for reference counting.
> >   
> > -     These are also used for atomic bitop functions that do not return a
> > -     value (such as set_bit and clear_bit).
> > +     These are also used for atomic RMW bitop functions that do imply a full
> s/do/do not/ ?
Sorry, yes, of course
> > +     memory barrier (such as set_bit and clear_bit).