Re: [PATCH/RFC] asm-generic/mutex-dec.h: add SMP support

[Eric Dumazet]

Le lundi 06 juin 2011 Ã 14:23 -0700, Andrew Morton a Ãcrit :
> On Sun, 29 May 2011 23:19:28 -0400
> Mike Frysinger <vapier@xxxxxxxxxx> wrote:
> 
> > To make these guys work on SMP systems, we just need to sprinkle a few
> > barriers around.
> > 
> > Signed-off-by: Mike Frysinger <vapier@xxxxxxxxxx>
> > ---
> > note: this is what the Blackfin SMP port is using, but it doesn't seem
> > like other SMP ports are ... so I wonder if we're just trying too hard
> > and these barriers aren't actually necessary ?
> > 
> >  include/asm-generic/mutex-dec.h |    8 +++++++-
> >  1 files changed, 7 insertions(+), 1 deletions(-)
> > 
> > diff --git a/include/asm-generic/mutex-dec.h b/include/asm-generic/mutex-dec.h
> > index f104af7..e746c3c 100644
> > --- a/include/asm-generic/mutex-dec.h
> > +++ b/include/asm-generic/mutex-dec.h
> > @@ -22,6 +22,8 @@ __mutex_fastpath_lock(atomic_t *count, void (*fail_fn)(atomic_t *))
> >  {
> >  	if (unlikely(atomic_dec_return(count) < 0))
> >  		fail_fn(count);

Check Documentation/memory-barriers.txt, around line 1688

atomic_dec_return() implies a full memory barrier on each side of the
operation.

smp_mb() is therefore not needed here


> > +	else
> > +		smp_mb();
> >  }
> >  
> >  /**
> > @@ -39,6 +41,7 @@ __mutex_fastpath_lock_retval(atomic_t *count, int (*fail_fn)(atomic_t *))
> >  {
> >  	if (unlikely(atomic_dec_return(count) < 0))
> >  		return fail_fn(count);
> > +	smp_mb();

atomic_dec_return() implies a full memory barrier.
	smp_mb() is therefore not needed here

> >  	return 0;
> >  }
> >  
> > @@ -58,6 +61,7 @@ __mutex_fastpath_lock_retval(atomic_t *count, int (*fail_fn)(atomic_t *))
> >  static inline void
> >  __mutex_fastpath_unlock(atomic_t *count, void (*fail_fn)(atomic_t *))
> >  {
> > +	smp_mb();

atomic_inc_return() implies a full memory barrier.
	smp_mb() is therefore not needed here


> >  	if (unlikely(atomic_inc_return(count) <= 0))
> >  		fail_fn(count);
> >  }
> > @@ -82,8 +86,10 @@ __mutex_fastpath_unlock(atomic_t *count, void (*fail_fn)(atomic_t *))
> >  static inline int
> >  __mutex_fastpath_trylock(atomic_t *count, int (*fail_fn)(atomic_t *))
> >  {
> > -	if (likely(atomic_cmpxchg(count, 1, 0) == 1))
> > +	if (likely(atomic_cmpxchg(count, 1, 0) == 1)) {
> > +		smp_mb();

atomic_cmpxchg() implies a full memory barrier.
	smp_mb() is therefore not needed here


> >  		return 1;
> > +	}
> >  	return 0;
> >  }
> 
> This patch basically reverts Nick's a8ddac7e53e89cb ("mutex: speed up
> generic mutex implementations").  What's up with that?
> 
> I could try to review this patch but I'm pathetic with barriers.  Help.

Well, I really dont understand this patch, it makes no sense.



--
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/