Re: [PATCH 2/2] docs: memory-barriers: Add note on plain-accesses to address-dependency barriers

[Paul E. McKenney]

On Fri, Aug 04, 2023 at 04:27:45PM +0000, Joel Fernandes wrote:
> On Fri, Aug 04, 2023 at 06:52:32AM -0700, Paul E. McKenney wrote:
> > On Fri, Aug 04, 2023 at 05:11:27AM +0000, Joel Fernandes wrote:
> > > On Thu, Aug 03, 2023 at 11:52:06AM -0700, Paul E. McKenney wrote:
> > > > On Thu, Aug 03, 2023 at 03:24:07AM +0000, Joel Fernandes (Google) wrote:
> > > > > The compiler has the ability to cause misordering by destroying
> > > > > address-dependency barriers if comparison operations are used. Add a
> > > > > note about this to memory-barriers.txt and point to rcu-dereference.rst
> > > > > for more information.
> > > > > 
> > > > > Signed-off-by: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
> > > > > ---
> > > > >  Documentation/memory-barriers.txt | 5 +++++
> > > > >  1 file changed, 5 insertions(+)
> > > > > 
> > > > > diff --git a/Documentation/memory-barriers.txt b/Documentation/memory-barriers.txt
> > > > > index 06e14efd8662..acc8ec5ce563 100644
> > > > > --- a/Documentation/memory-barriers.txt
> > > > > +++ b/Documentation/memory-barriers.txt
> > > > > @@ -435,6 +435,11 @@ Memory barriers come in four basic varieties:
> > > > >       variables such as READ_ONCE() and rcu_dereference() provide implicit
> > > > >       address-dependency barriers.
> > > > >  
> > > > > +     [!] Note that address dependency barriers can be destroyed by comparison
> > > > > +     of a pointer obtained by a marked accessor such as READ_ONCE() or
> > > > > +     rcu_dereference() with some value.  For an example of this, see
> > > > > +     rcu_dereference.rst (part where the comparison of pointers is discussed).
> > > > 
> > > > Hmmm...
> > > > 
> > > > Given that this is in a section marked "historical" (for the old
> > > > smp_read_barrier_depends() API), why not instead add a pointer to
> > > > Documentation/RCU/rcu_dereference.rst to the beginning of the section,
> > > > noted as the updated material?
> > > 
> > > Sounds good. There's also another section in the same file on Address
> > > dependency barriers (also marked historical). So something like the
> > > following?
> > 
> > Given a Signed-off-by and so forth, I would be happy to take this one.
> 
> Thank you for helping me improve the docs, here it goes:
> 
> ---8<-----------------------
> 
> From: "Joel Fernandes (Google)" <joel@xxxxxxxxxxxxxxxxx>
> Subject: [PATCH] docs: memory-barriers: Add note on compiler transformation
>  and address deps
> 
> The compiler has the ability to cause misordering by destroying
> address-dependency barriers if comparison operations are used. Add a
> note about this to memory-barriers.txt in the beginning of both the
> historical address-dependency sections and point to rcu-dereference.rst
> for more information.
> 
> Signed-off-by: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>

Queued and pushed, thank you!

							Thanx, Paul

> ---
>  Documentation/memory-barriers.txt | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/Documentation/memory-barriers.txt b/Documentation/memory-barriers.txt
> index acc8ec5ce563..ba50220716ca 100644
> --- a/Documentation/memory-barriers.txt
> +++ b/Documentation/memory-barriers.txt
> @@ -396,6 +396,10 @@ Memory barriers come in four basic varieties:
>  
>  
>   (2) Address-dependency barriers (historical).
> +     [!] This section is marked as HISTORICAL: For more up-to-date
> +     information, including how compiler transformations related to pointer
> +     comparisons can sometimes cause problems, see
> +     Documentation/RCU/rcu_dereference.rst.
>  
>       An address-dependency barrier is a weaker form of read barrier.  In the
>       case where two loads are performed such that the second depends on the
> @@ -561,6 +565,9 @@ There are certain things that the Linux kernel memory barriers do not guarantee:
>  
>  ADDRESS-DEPENDENCY BARRIERS (HISTORICAL)
>  ----------------------------------------
> +[!] This section is marked as HISTORICAL: For more up-to-date information,
> +including how compiler transformations related to pointer comparisons can
> +sometimes cause problems, see Documentation/RCU/rcu_dereference.rst.
>  
>  As of v4.15 of the Linux kernel, an smp_mb() was added to READ_ONCE() for
>  DEC Alpha, which means that about the only people who need to pay attention
> -- 
> 2.41.0.585.gd2178a4bd4-goog
>