On Fri, May 24, 2019 at 12:37:31PM +0200, Peter Zijlstra wrote:
> On Thu, May 23, 2019 at 11:19:26AM +0100, Mark Rutland wrote:
>
> > [mark@lakrids:~/src/linux]% git grep '\(return\|=\)\s\+atomic\(64\)\?_set'
> > include/linux/vmw_vmci_defs.h: return atomic_set((atomic_t *)var, (u32)new_val);
> > include/linux/vmw_vmci_defs.h: return atomic64_set(var, new_val);
> >
>
> Oh boy, what a load of crap you just did find.
>
> How about something like the below? I've not read how that buffer is
> used, but the below preserves all broken without using atomic*_t.
Clarified by something along these lines?
---
Documentation/atomic_t.txt | 3 +++
1 file changed, 3 insertions(+)
diff --git a/Documentation/atomic_t.txt b/Documentation/atomic_t.txt
index dca3fb0554db..125c95ddbbc0 100644
--- a/Documentation/atomic_t.txt
+++ b/Documentation/atomic_t.txt
@@ -83,6 +83,9 @@ 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.
+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 one detail to this is that atomic_set{}() should be observable to the RMW
ops. That is:
