Re: [PATCH -tip 2/2] x86: Issue at least one memory barrier in stop_machine_text_poke().

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

* Masami Hiramatsu (mhiramat@redhat.com) wrote:
> Fix stop_machine_text_poke() to issue smp_mb() before exiting waiting
> loop. Also, use ACCESS_ONCE() to check a flag according to Mathieu's
> comment.
> 
> Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
> Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
> Cc: Ingo Molnar <mingo@elte.hu>
> Cc: Jason Baron <jbaron@redhat.com>
> ---
> 
>  arch/x86/kernel/alternative.c |    6 +++---
>  1 files changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/x86/kernel/alternative.c b/arch/x86/kernel/alternative.c
> index 635e4f4..3236a11 100644
> --- a/arch/x86/kernel/alternative.c
> +++ b/arch/x86/kernel/alternative.c
> @@ -596,9 +596,9 @@ static int __kprobes stop_machine_text_poke(void *data)
>  		smp_wmb();	/* Make sure other cpus see that this has run */
>  		wrote_text = 1;
>  	} else {
> -		while (!wrote_text)
> -			smp_rmb();
> -		sync_core();
> +		while (!ACCESS_ONCE(wrote_text))

Well, cpu_relax() has a "memory" clobber, so the access once is not required to
ensure the variable is re-read. And, sorry to contradict my previous statement
somewhat, but given that we don't care if the compiler fetches wrote_text in
chunks or not, ACCESS_ONCE() is not required here. So rather than leaving people
wondering why we put an ACCESS_ONCE() here, it's probably better to leave it
out.

Thanks,

Mathieu

> +			cpu_relax();
> +		smp_mb();	/* Load wrote_text before following execution */
>  	}
>  
>  	flush_icache_range((unsigned long)tpp->addr,
> 
> 
> -- 
> Masami Hiramatsu
> 
> Software Engineer
> Hitachi Computer Products (America), Inc.
> Software Solutions Division
> 
> e-mail: mhiramat@redhat.com

-- 
Mathieu Desnoyers
Operating System Efficiency Consultant
EfficiOS Inc.
http://www.efficios.com