Re: [Lse-tech] [RFC,PATCH] use rcu for fasync_lock

[Stephen Hemminger]

Manfred Spraul wrote:

> Hi,
>
> kill_fasync and fasync_helper were intended for mice and similar, rare 
> users, thus it uses a simple rwlock for the locking. This is not true 
> anymore: e.g. every pipe read and write operation calls kill_fasync, 
> which must acquire the rwlock before handling the fasync list.
> What about switching to rcu? I did a reaim run on a 4-way pIII with 
> STP, and it reduced the time within kill_fasync by 80%:
>
> diffprofile reaim_End_stock reaim_End_rcu              21166     1.2% 
> default_idle
>     18882     0.9% total
>       290    12.8% page_address
>       269    23.5% group_send_sig_info
>       259    41.1% do_brk
>       244     6.3% current_kernel_time
> [ delta < 200: skipped]
>      -205   -16.1% get_signal_to_deliver
>      -240    -3.7% page_add_rmap
>      -364    -4.7% __might_sleep
>      -369    -8.4% page_remove_rmap
>      -975   -81.2% kill_fasync
>
> What do you think? Patch against 2.6.0 is attached.
>
> --
>    Manfred
>
> ------------------------------------------------------------------------
>
> --- 2.6/fs/fcntl.c	2003-12-04 19:44:38.000000000 +0100
> +++ build-2.6/fs/fcntl.c	2003-12-20 10:56:23.344256035 +0100
> @@ -537,9 +537,19 @@
> 	return ret;
> }
>
> -static rwlock_t fasync_lock = RW_LOCK_UNLOCKED;
> +static spinlock_t fasync_lock = SPIN_LOCK_UNLOCKED;
> static kmem_cache_t *fasync_cache;
>
> +struct fasync_rcu_struct {
> +	struct fasync_struct data;
> +	struct rcu_head rcu;
> +};
>  
Why do needless wrapping of existing structure? Just add and rcu element 
to it!


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/