linuxppc-dev.lists.ozlabs.org archive mirror
 help / color / mirror / Atom feed
* Re: [PATCH tip/core/rcu 48/55] powerpc: strengthen value-returning-atomics memory barriers
       [not found] ` <1315332049-2604-48-git-send-email-paulmck@linux.vnet.ibm.com>
@ 2011-09-09 17:23   ` Olof Johansson
  2011-09-09 17:34     ` Paul E. McKenney
  0 siblings, 1 reply; 3+ messages in thread
From: Olof Johansson @ 2011-09-09 17:23 UTC (permalink / raw)
  To: Paul E. McKenney
  Cc: mathieu.desnoyers, laijs, eric.dumazet, patches, peterz,
	linux-kernel, rostedt, josh, dhowells, darren, niv, linuxppc-dev,
	tglx, anton, Valdis.Kletnieks, mingo, akpm, paulus

[+linuxppc-dev]

On Tue, Sep 6, 2011 at 11:00 AM, Paul E. McKenney
<paulmck@linux.vnet.ibm.com> wrote:
> The trailing isync/lwsync in PowerPC value-returning atomics needs
> to be a sync in order to provide the required ordering properties.
> The leading lwsync/eieio can remain, as the remainder of the required
> ordering guarantees are provided by the atomic instructions: Any
> reordering will cause the stwcx to fail, which will result in a retry.

Admittedly, my powerpc barrier memory is starting to fade, but isn't
isync sufficient here? It will make sure all instructions before it
have retired, and will restart any speculative/issued instructions
beyond it.

lwsync not being sufficient makes sense since a load can overtake it.

> diff --git a/arch/powerpc/include/asm/synch.h b/arch/powerpc/include/asm/=
synch.h
> index d7cab44..4d97fbe 100644
> --- a/arch/powerpc/include/asm/synch.h
> +++ b/arch/powerpc/include/asm/synch.h
> @@ -37,11 +37,7 @@ static inline void isync(void)
> =A0#endif
>
> =A0#ifdef CONFIG_SMP
> -#define __PPC_ACQUIRE_BARRIER =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0\
> - =A0 =A0 =A0 START_LWSYNC_SECTION(97); =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 \
> - =A0 =A0 =A0 isync; =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0\
> - =A0 =A0 =A0 MAKE_LWSYNC_SECTION_ENTRY(97, __lwsync_fixup);
> -#define PPC_ACQUIRE_BARRIER =A0 =A0"\n" stringify_in_c(__PPC_ACQUIRE_BAR=
RIER)
> +#define PPC_ACQUIRE_BARRIER =A0 =A0"\n" stringify_in_c(sync;)

This can just be done as "\n\tsync\n" instead of the stringify stuff.


-Olof

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH tip/core/rcu 48/55] powerpc: strengthen value-returning-atomics memory barriers
  2011-09-09 17:23   ` [PATCH tip/core/rcu 48/55] powerpc: strengthen value-returning-atomics memory barriers Olof Johansson
@ 2011-09-09 17:34     ` Paul E. McKenney
  2011-09-09 18:43       ` Olof Johansson
  0 siblings, 1 reply; 3+ messages in thread
From: Paul E. McKenney @ 2011-09-09 17:34 UTC (permalink / raw)
  To: Olof Johansson
  Cc: mathieu.desnoyers, laijs, eric.dumazet, patches, peterz,
	linux-kernel, rostedt, josh, dhowells, darren, niv, linuxppc-dev,
	tglx, anton, Valdis.Kletnieks, mingo, akpm, paulus

On Fri, Sep 09, 2011 at 10:23:33AM -0700, Olof Johansson wrote:
> [+linuxppc-dev]
> 
> On Tue, Sep 6, 2011 at 11:00 AM, Paul E. McKenney
> <paulmck@linux.vnet.ibm.com> wrote:
> > The trailing isync/lwsync in PowerPC value-returning atomics needs
> > to be a sync in order to provide the required ordering properties.
> > The leading lwsync/eieio can remain, as the remainder of the required
> > ordering guarantees are provided by the atomic instructions: Any
> > reordering will cause the stwcx to fail, which will result in a retry.
> 
> Admittedly, my powerpc barrier memory is starting to fade, but isn't
> isync sufficient here? It will make sure all instructions before it
> have retired, and will restart any speculative/issued instructions
> beyond it.
> 
> lwsync not being sufficient makes sense since a load can overtake it.

As I understand it, although isync waits for the prior stwcx to execute,
it does not guarantee that the corresponding store is visible to all
processors before any following loads.

> > diff --git a/arch/powerpc/include/asm/synch.h b/arch/powerpc/include/asm/synch.h
> > index d7cab44..4d97fbe 100644
> > --- a/arch/powerpc/include/asm/synch.h
> > +++ b/arch/powerpc/include/asm/synch.h
> > @@ -37,11 +37,7 @@ static inline void isync(void)
> >  #endif
> >
> >  #ifdef CONFIG_SMP
> > -#define __PPC_ACQUIRE_BARRIER                          \
> > -       START_LWSYNC_SECTION(97);                       \
> > -       isync;                                          \
> > -       MAKE_LWSYNC_SECTION_ENTRY(97, __lwsync_fixup);
> > -#define PPC_ACQUIRE_BARRIER    "\n" stringify_in_c(__PPC_ACQUIRE_BARRIER)
> > +#define PPC_ACQUIRE_BARRIER    "\n" stringify_in_c(sync;)
> 
> This can just be done as "\n\tsync\n" instead of the stringify stuff.

That does sound a bit more straightforward, now that you mention it.  ;-)

							Thanx, Paul

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH tip/core/rcu 48/55] powerpc: strengthen value-returning-atomics memory barriers
  2011-09-09 17:34     ` Paul E. McKenney
@ 2011-09-09 18:43       ` Olof Johansson
  0 siblings, 0 replies; 3+ messages in thread
From: Olof Johansson @ 2011-09-09 18:43 UTC (permalink / raw)
  To: paulmck
  Cc: mathieu.desnoyers, laijs, eric.dumazet, patches, peterz,
	linux-kernel, rostedt, josh, dhowells, darren, niv, linuxppc-dev,
	tglx, anton, Valdis.Kletnieks, mingo, akpm, paulus

On Fri, Sep 9, 2011 at 10:34 AM, Paul E. McKenney
<paulmck@linux.vnet.ibm.com> wrote:
> On Fri, Sep 09, 2011 at 10:23:33AM -0700, Olof Johansson wrote:
>> [+linuxppc-dev]
>>
>> On Tue, Sep 6, 2011 at 11:00 AM, Paul E. McKenney
>> <paulmck@linux.vnet.ibm.com> wrote:
>> > The trailing isync/lwsync in PowerPC value-returning atomics needs
>> > to be a sync in order to provide the required ordering properties.
>> > The leading lwsync/eieio can remain, as the remainder of the required
>> > ordering guarantees are provided by the atomic instructions: Any
>> > reordering will cause the stwcx to fail, which will result in a retry.
>>
>> Admittedly, my powerpc barrier memory is starting to fade, but isn't
>> isync sufficient here? It will make sure all instructions before it
>> have retired, and will restart any speculative/issued instructions
>> beyond it.
>>
>> lwsync not being sufficient makes sense since a load can overtake it.
>
> As I understand it, although isync waits for the prior stwcx to execute,
> it does not guarantee that the corresponding store is visible to all
> processors before any following loads.

Ah yes, combined with brushing up on the semantics in
memory-barriers.txt, this sounds reasonable to me.

>> > diff --git a/arch/powerpc/include/asm/synch.h b/arch/powerpc/include/a=
sm/synch.h
>> > index d7cab44..4d97fbe 100644
>> > --- a/arch/powerpc/include/asm/synch.h
>> > +++ b/arch/powerpc/include/asm/synch.h
>> > @@ -37,11 +37,7 @@ static inline void isync(void)
>> > =A0#endif
>> >
>> > =A0#ifdef CONFIG_SMP
>> > -#define __PPC_ACQUIRE_BARRIER =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0 =A0\
>> > - =A0 =A0 =A0 START_LWSYNC_SECTION(97); =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 =A0 \
>> > - =A0 =A0 =A0 isync; =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0\
>> > - =A0 =A0 =A0 MAKE_LWSYNC_SECTION_ENTRY(97, __lwsync_fixup);
>> > -#define PPC_ACQUIRE_BARRIER =A0 =A0"\n" stringify_in_c(__PPC_ACQUIRE_=
BARRIER)
>> > +#define PPC_ACQUIRE_BARRIER =A0 =A0"\n" stringify_in_c(sync;)
>>
>> This can just be done as "\n\tsync\n" instead of the stringify stuff.
>
> That does sound a bit more straightforward, now that you mention it. =A0;=
-)

With that change, I'm:

Acked-by: Olof Johansson <olof@lixom.net>

But at least Ben or Anton should sign off on it too.


-Olof

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2011-09-09 18:43 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
     [not found] <20110906180015.GA2560@linux.vnet.ibm.com>
     [not found] ` <1315332049-2604-48-git-send-email-paulmck@linux.vnet.ibm.com>
2011-09-09 17:23   ` [PATCH tip/core/rcu 48/55] powerpc: strengthen value-returning-atomics memory barriers Olof Johansson
2011-09-09 17:34     ` Paul E. McKenney
2011-09-09 18:43       ` Olof Johansson

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).