Re: [MEH PATCH] fs/pipe: write to ->poll_usage only once

[Jan Kara]

On Sat 16-05-26 04:18:52, Mateusz Guzik wrote:
> Both GNU and BSD makes share a "token pipe" between their instances, as
> a result a -j $BIGNUM invocation results in multicore perf problems in
> the poll handler.
> 
> Avoiding the store will reduce it a little bit. However, the crux of the
> problem is the locked queuing up in poll_wait().
> 
> Signed-off-by: Mateusz Guzik <mjguzik@xxxxxxxxx>

I guess doesn't harm to reduce cacheline invalidations. Feel free to add:

Reviewed-by: Jan Kara <jack@xxxxxxx>

								Honza

> ---
> 
> disclaimer: could not be bothered to bench this
> 
>  fs/pipe.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/pipe.c b/fs/pipe.c
> index 9841648c9cf3..e37c79935ecb 100644
> --- a/fs/pipe.c
> +++ b/fs/pipe.c
> @@ -664,7 +664,8 @@ pipe_poll(struct file *filp, poll_table *wait)
>  	union pipe_index idx;
>  
>  	/* Epoll has some historical nasty semantics, this enables them */
> -	WRITE_ONCE(pipe->poll_usage, true);
> +	if (unlikely(!READ_ONCE(pipe->poll_usage)))
> +		WRITE_ONCE(pipe->poll_usage, true);
>  
>  	/*
>  	 * Reading pipe state only -- no need for acquiring the semaphore.
> -- 
> 2.48.1
> 
-- 
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR