lkml.org 
[lkml]   [2014]   [Oct]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH] futex: Ensure get_futex_key_refs() always implies a barrier
From
On Mon, Oct 20, 2014 at 3:49 AM, Catalin Marinas
<catalin.marinas@arm.com> wrote:
>
> Since you mention symmetry, something like below makes the barriers more
> explicit.

Borken, for two reasons:

> diff --git a/kernel/futex.c b/kernel/futex.c
> index f3a3a071283c..5b9d857d0816 100644
> --- a/kernel/futex.c
> +++ b/kernel/futex.c
> @@ -143,9 +143,7 @@
> static inline void futex_get_mm(union futex_key *key)
> {
> atomic_inc(&key->private.mm->mm_count);
> - /*
> - * Ensure futex_get_mm() implies a full barrier such that
> - * get_futex_key() implies a full barrier. This is relied upon
> - * as full barrier (B), see the ordering comment above.
> - */
> - smp_mb__after_atomic();
> }

So the thing is, this means that we can't take advantage of the fact
that "atomic_inc" is already an atomic. So this is just a performance
breakage. But:

>
> static inline int hb_waiters_pending(struct futex_hash_bucket *hb)
> {
> + /*
> + * Full barrier (B), see the ordering comment above.
> + */
> + smp_mb__before_atomic();
> #ifdef CONFIG_SMP
> return atomic_read(&hb->waiters);

This is just entirely broken.

"atomic_read()" isn't really an "atomic op" at all. despite the name,
it's just a read that is basically ACCESS_ONCE.

So smp_mb__before_atomic() doesn't work for atomic_read(), and the
code is nonsensical and doesn't work. It would need to be a full
memory barrier.

Linus


\
 
 \ /
  Last update: 2014-10-20 17:41    [W:0.055 / U:1.260 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site