Re: per-cpu blk_plug_list

[Andrew Morton <akpm@osdl.org>]

Andrew Morton <akpm@osdl.org> wrote:
>
> @@ -1251,6 +1251,9 @@ void blk_run_queues(void)
>  {
>  	LIST_HEAD(local_plug_list);
>  
> +	if (list_empty(&blk_plug_list))
> +		return;
> +

hmm, no, that won't help.  There will always be at least one plugged disk.

Perhaps what we need here is a per-task container of "queues which I have
plugged".  Or, more accurately, "queues which need to be unplugged so that
I can complete".  It'll get tricky because in some situations a task can
need unplugging of a queue which it did not plug, and even a queue to which
it did not submit any I/O.

You should try just nuking the whole plugging scheme, btw.  Stick a bunch
of `return' statements at the top of all the plugging functions.  I bet
that would run nicely...