Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 21 Feb 2018 11:21:38 +0000
From:   Will Deacon <will.deacon@arm.com>
To:     Andrea Parri <parri.andrea@gmail.com>
Cc:     Ingo Molnar <mingo@kernel.org>,
        "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>,
        Alan Stern <stern@rowland.harvard.edu>,
        Richard Henderson <rth@twiddle.net>,
        Ivan Kokshaysky <ink@jurassic.park.msu.ru>,
        Matt Turner <mattst88@gmail.com>, linux-alpha@vger.kernel.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] xchg/alpha: Add unconditional memory barrier to cmpxchg
Message-ID: <20180221112137.GA6165@arm.com>
References: <1519152356-4804-1-git-send-email-parri.andrea@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1519152356-4804-1-git-send-email-parri.andrea@gmail.com>
User-Agent: Mutt/1.5.23 (2014-03-12)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Andrea,

On Tue, Feb 20, 2018 at 07:45:56PM +0100, Andrea Parri wrote:
> Continuing along with the fight against smp_read_barrier_depends() [1]
> (or rather, against its improper use), add an unconditional barrier to
> cmpxchg.  This guarantees that dependency ordering is preserved when a
> dependency is headed by an unsuccessful cmpxchg.  As it turns out, the
> change could enable further simplification of LKMM as proposed in [2].
> 
> [1] https://marc.info/?l=linux-kernel&m=150884953419377&w=2
>     https://marc.info/?l=linux-kernel&m=150884946319353&w=2
>     https://marc.info/?l=linux-kernel&m=151215810824468&w=2
>     https://marc.info/?l=linux-kernel&m=151215816324484&w=2
> 
> [2] https://marc.info/?l=linux-kernel&m=151881978314872&w=2
> 
> Signed-off-by: Andrea Parri <parri.andrea@gmail.com>
> Acked-by: Peter Zijlstra <peterz@infradead.org>
> Cc: Will Deacon <will.deacon@arm.com>
> Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
> Cc: Alan Stern <stern@rowland.harvard.edu>
> Cc: Richard Henderson <rth@twiddle.net>
> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
> Cc: Matt Turner <mattst88@gmail.com>
> Cc: linux-alpha@vger.kernel.org
> Cc: linux-kernel@vger.kernel.org
> ---
>  arch/alpha/include/asm/xchg.h | 15 +++++++--------
>  1 file changed, 7 insertions(+), 8 deletions(-)
> 
> diff --git a/arch/alpha/include/asm/xchg.h b/arch/alpha/include/asm/xchg.h
> index 68dfb3cb71454..e2660866ce972 100644
> --- a/arch/alpha/include/asm/xchg.h
> +++ b/arch/alpha/include/asm/xchg.h
> @@ -128,10 +128,9 @@ ____xchg(, volatile void *ptr, unsigned long x, int size)
>   * store NEW in MEM.  Return the initial value in MEM.  Success is
>   * indicated by comparing RETURN with OLD.
>   *
> - * The memory barrier should be placed in SMP only when we actually
> - * make the change. If we don't change anything (so if the returned
> - * prev is equal to old) then we aren't acquiring anything new and
> - * we don't need any memory barrier as far I can tell.
> + * The memory barrier is placed in SMP unconditionally, in order to
> + * guarantee that dependency ordering is preserved when a dependency
> + * is headed by an unsuccessful operation.
>   */
>  
>  static inline unsigned long
> @@ -150,8 +149,8 @@ ____cmpxchg(_u8, volatile char *m, unsigned char old, unsigned char new)
>  	"	or	%1,%2,%2\n"
>  	"	stq_c	%2,0(%4)\n"
>  	"	beq	%2,3f\n"
> -		__ASM__MB
>  	"2:\n"
> +		__ASM__MB
>  	".subsection 2\n"
>  	"3:	br	1b\n"
>  	".previous"

It might be better just to add smp_read_barrier_depends() into the cmpxchg
macro, then remove all of the __ASM__MB stuff.

That said, I don't actually understand how the Alpha cmpxchg or xchg
implementations satisfy the memory model, since they only appear to have
a barrier after the operation.

So MP using xchg:

WRITE_ONCE(x, 1)
xchg(y, 1)

smp_load_acquire(y) == 1
READ_ONCE(x) == 0

would be allowed. What am I missing?

Since I'm in the mood for dumb questions, do we need to care about
this_cpu_cmpxchg? I'm sure I've seen code that allows concurrent access to
per-cpu variables, but the asm-generic implementation of this_cpu_cmpxchg
doesn't use READ_ONCE.

Will