On Tue, 5 Oct 2021 at 14:03, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Tue, Oct 05, 2021 at 12:58:58PM +0200, Marco Elver wrote:
> > @@ -59,6 +60,7 @@ atomic_add(int i, atomic_t *v)
> > static __always_inline int
> > atomic_add_return(int i, atomic_t *v)
> > {
> > + kcsan_mb();
> > instrument_atomic_read_write(v, sizeof(*v));
> > return arch_atomic_add_return(i, v);
> > }
>
> This and others,.. is this actually correct? Should that not be
> something like:
>
> kscan_mb();
> instrument_atomic_read_write(...);
> ret = arch_atomic_add_return(i, v);
> kcsan_mb();
> return ret;
>
> ?
In theory, yes, but right now it's redundant.
Because right now KCSAN only models "buffering", and no "prefetching".
So there's no way that a later instruction would be reordered before
this point. And atomic accesses are never considered for reordering,
so it's also impossible that it would be reordered later.
Each kcsan_mb() is a call, so right now it makes sense to just have 1
call to be a bit more efficient.
