Re: [PATCH v4 0/4] Documentation/litmus-tests: Add litmus tests for atomic APIs

[Joel Fernandes]

On Tue, Mar 31, 2020 at 09:40:37AM +0800, Boqun Feng wrote:
> On Fri, Mar 27, 2020 at 06:18:43PM -0400, Joel Fernandes wrote:
> > On Thu, Mar 26, 2020 at 10:40:18AM +0800, Boqun Feng wrote:
> > > A recent discussion raises up the requirement for having test cases for
> > > atomic APIs:
> > > 
> > > 	https://lore.kernel.org/lkml/20200213085849.GL14897@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
> > > 
> > > , and since we already have a way to generate a test module from a
> > > litmus test with klitmus[1]. It makes sense that we add more litmus
> > > tests for atomic APIs. And based on the previous discussion, I create a
> > > new directory Documentation/atomic-tests and put these litmus tests
> > > here.
> > > 
> > > This patchset starts the work by adding the litmus tests which are
> > > already used in atomic_t.txt, and also improve the atomic_t.txt to make
> > > it consistent with the litmus tests.
> > > 
> > > Previous version:
> > > v1: https://lore.kernel.org/linux-doc/20200214040132.91934-1-boqun.feng@xxxxxxxxx/
> > > v2: https://lore.kernel.org/lkml/20200219062627.104736-1-boqun.feng@xxxxxxxxx/
> > > v3: https://lore.kernel.org/linux-doc/20200227004049.6853-1-boqun.feng@xxxxxxxxx/
> > 
> > For full series:
> > 
> > Reviewed-by: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
> > 
> > One question I had was in the existing atomic_set() documentation, it talks
> > about atomic_add_unless() implementation based on locking could have issues.
> > It says the way to fix such cases is:
> > 
> > Quote:
> >     the typical solution is to then implement atomic_set{}() with
> >     atomic_xchg().
> > 
> > I didn't get how using atomic_xchg() fixes it. Is the assumption there that
> > atomic_xchg() would be implemented using locking to avoid atomic_set() having
> 
> Right, I think that's the intent of the sentence.
> 
> > issues? If so, we could clarify that in the document.
> > 
> 
> Patches are welcome ;-)


---8<-----------------------

Like this? I'll add it to my tree and send it to Paul during my next
series, unless you disagree ;-)

Subject: [PATCH] doc: atomic_t: Document better about the locking within
 atomic_xchg()

It is not fully clear how the atomic_set() would not cause an issue with
preservation of the atomicity of RMW in this example. Make it clear that
locking within atomic_xchg() would save the day.

Suggested-by: Boqun Feng <boqun.feng@xxxxxxxxx>
Signed-off-by: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
---
 Documentation/atomic_t.txt | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/Documentation/atomic_t.txt b/Documentation/atomic_t.txt
index 0f1fdedf36bbb..1d9c307c73a7c 100644
--- a/Documentation/atomic_t.txt
+++ b/Documentation/atomic_t.txt
@@ -129,6 +129,8 @@ with a lock:
     unlock();
 
 the typical solution is to then implement atomic_set{}() with atomic_xchg().
+The locking within the atomic_xchg() in CPU1 would ensure that the value read
+in CPU0 would not be overwritten.
 
 
 RMW ops:
-- 
2.26.0.292.g33ef6b2f38-goog