Re: [PATCH] tools/memory-model: remove rb-dep, smp_read_barrier_depends, and lockless_dereference

[Andrea Parri]

On Fri, Feb 16, 2018 at 05:22:55PM -0500, Alan Stern wrote:
> Since commit 76ebbe78f739 ("locking/barriers: Add implicit
> smp_read_barrier_depends() to READ_ONCE()") was merged for the 4.15
> kernel, it has not been necessary to use smp_read_barrier_depends().
> Similarly, commit 59ecbbe7b31c ("locking/barriers: Kill
> lockless_dereference()") removed lockless_dereference() from the
> kernel.
> 
> Since these primitives are no longer part of the kernel, they do not
> belong in the Linux Kernel Memory Consistency Model.  This patch
> removes them, along with the internal rb-dep relation, and updates the
> revelant documentation.
> 
> Signed-off-by: Alan Stern <stern@xxxxxxxxxxxxxxxxxxx>
> 
> ---

[...]


> Index: usb-4.x/tools/memory-model/linux-kernel.def
> ===================================================================
> --- usb-4.x/tools/memory-model.orig/linux-kernel.def
> +++ usb-4.x/tools/memory-model/linux-kernel.def
> @@ -13,14 +13,12 @@ WRITE_ONCE(X,V) { __store{once}(X,V); }
>  smp_store_release(X,V) { __store{release}(*X,V); }
>  smp_load_acquire(X) __load{acquire}(*X)
>  rcu_assign_pointer(X,V) { __store{release}(X,V); }
> -lockless_dereference(X) __load{lderef}(X)
>  rcu_dereference(X) __load{deref}(X)

^^^ __load{once}


>  
>  // Fences
>  smp_mb() { __fence{mb} ; }
>  smp_rmb() { __fence{rmb} ; }
>  smp_wmb() { __fence{wmb} ; }
> -smp_read_barrier_depends() { __fence{rb_dep}; }
>  smp_mb__before_atomic() { __fence{before-atomic} ; }
>  smp_mb__after_atomic() { __fence{after-atomic} ; }
>  smp_mb__after_spinlock() { __fence{after-spinlock} ; }
> Index: usb-4.x/tools/memory-model/Documentation/cheatsheet.txt
> ===================================================================
> --- usb-4.x/tools/memory-model.orig/Documentation/cheatsheet.txt
> +++ usb-4.x/tools/memory-model/Documentation/cheatsheet.txt
> @@ -6,8 +6,7 @@
>  Store, e.g., WRITE_ONCE()            Y                                       Y
>  Load, e.g., READ_ONCE()              Y                              Y        Y
>  Unsuccessful RMW operation           Y                              Y        Y
> -smp_read_barrier_depends()              Y                       Y   Y
> -*_dereference()                      Y                          Y   Y        Y
> +rcu_dereference()                    Y                          Y   Y        Y
>  Successful *_acquire()               R                   Y  Y   Y   Y    Y   Y
>  Successful *_release()         C        Y  Y    Y     W                      Y
>  smp_rmb()                               Y       R        Y      Y        R

Akira's observation about READ_ONCE extends to all (annotated) loads.  In
fact, it also applies to loads corresponding to unsuccessful RMW operations;
consider, for example, the following variation of MP+onceassign+derefonce:

C T

{
y=z;
z=0;
}

P0(int *x, int **y)
{
	WRITE_ONCE(*x, 1);
	smp_store_release(y, x);
}

P1(int **y, int *z)
{
	int *r0;
	int r1;

	r0 = cmpxchg_relaxed(y, z, z);
	r1 = READ_ONCE(*r0);
}

exists (1:r0=x /\ 1:r1=0)

The final state is allowed w/o the patch, and forbidden w/ the patch.

This also reminds me of

   5a8897cc7631fa544d079c443800f4420d1b173f
   ("locking/atomics/alpha: Add smp_read_barrier_depends() to _release()/_relaxed() atomics")

(that we probably want to mention in the commit message).

  Andrea


> 
>