Re: [RESEND PATCH] documentation: memory-barriers: fix smp_mb__before_spinlock() semantics

[Paul E. McKenney]

On Tue, Mar 31, 2015 at 07:50:50PM +0200, Oleg Nesterov wrote:
> On 03/31, Will Deacon wrote:
> >
> > Could somebody pick this up please? I guess I could route it via the arm64
> > tree with an Ack, but I'd rather it went through Paul or -tip.
> 
> I think this would be the best route ;)
> 
> > --- a/Documentation/memory-barriers.txt
> > +++ b/Documentation/memory-barriers.txt
> > @@ -1768,10 +1768,9 @@ for each construct.  These operations all imply certain barriers:
> >  
> >       Memory operations issued before the ACQUIRE may be completed after
> >       the ACQUIRE operation has completed.  An smp_mb__before_spinlock(),
> > -     combined with a following ACQUIRE, orders prior loads against
> > -     subsequent loads and stores and also orders prior stores against
> > -     subsequent stores.  Note that this is weaker than smp_mb()!  The
> > -     smp_mb__before_spinlock() primitive is free on many architectures.
> > +     combined with a following ACQUIRE, orders prior stores against
> > +     subsequent loads and stores. Note that this is weaker than smp_mb()!
> > +     The smp_mb__before_spinlock() primitive is free on many architectures.
> 
> I agree, this description was always wrong.
> 
> But perhaps you can also update the comment above smp_mb__before_spinlock?
> It only documents the STORE - LOAD serialization, and this was on purpose.
> 
> But people started to use this helper assuming that it can also serialize
> the STOREs. Perhaps the changelog could also mention this fact, this is why
> we need to update this comment and fix memory-barriers.txt.

If Will agrees, like the following?

							Thanx, Paul

------------------------------------------------------------------------

    documentation: memory-barriers: Fix smp_mb__before_spinlock() semantics
    
    Our current documentation claims that, when followed by an ACQUIRE,
    smp_mb__before_spinlock() orders prior loads against subsequent loads
    and stores, which isn't the intent.  This commit therefore fixes the
    documentation to state that this sequence orders only prior stores
    against subsequent loads and stores.
    
    In addition, the original intent of smp_mb__before_spinlock() was to only
    order prior loads against subsequent stores, however, people have started
    using it as if it ordered prior loads against subsequent loads and stores.
    This commit therefore also updates smp_mb__before_spinlock()'s header
    comment to reflect this new reality.
    
    Cc: Oleg Nesterov <oleg@xxxxxxxxxx>
    Cc: "Paul E. McKenney" <paulmck@xxxxxxxxxxxxxxxxxx>
    Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
    Signed-off-by: Will Deacon <will.deacon@xxxxxxx>
    Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>

diff --git a/Documentation/memory-barriers.txt b/Documentation/memory-barriers.txt
index 6974f1c2b4e1..52c320e3f107 100644
--- a/Documentation/memory-barriers.txt
+++ b/Documentation/memory-barriers.txt
@@ -1784,10 +1784,9 @@ for each construct.  These operations all imply certain barriers:
 
      Memory operations issued before the ACQUIRE may be completed after
      the ACQUIRE operation has completed.  An smp_mb__before_spinlock(),
-     combined with a following ACQUIRE, orders prior loads against
-     subsequent loads and stores and also orders prior stores against
-     subsequent stores.  Note that this is weaker than smp_mb()!  The
-     smp_mb__before_spinlock() primitive is free on many architectures.
+     combined with a following ACQUIRE, orders prior stores against
+     subsequent loads and stores. Note that this is weaker than smp_mb()!
+     The smp_mb__before_spinlock() primitive is free on many architectures.
 
  (2) RELEASE operation implication:
 
diff --git a/include/linux/spinlock.h b/include/linux/spinlock.h
index 3e18379dfa6f..0063b24b4f36 100644
--- a/include/linux/spinlock.h
+++ b/include/linux/spinlock.h
@@ -120,7 +120,7 @@ do {								\
 /*
  * Despite its name it doesn't necessarily has to be a full barrier.
  * It should only guarantee that a STORE before the critical section
- * can not be reordered with a LOAD inside this section.
+ * can not be reordered with LOADs and STOREs inside this section.
  * spin_lock() is the one-way barrier, this LOAD can not escape out
  * of the region. So the default implementation simply ensures that
  * a STORE can not move into the critical section, smp_wmb() should

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/