Re: [RFT/PATCH] atkbd - speed up setting leds/repeat state

[Vojtech Pavlik <vojtech@suse.cz>]

On Mon, Oct 31, 2005 at 09:05:31AM -0500, Dmitry Torokhov wrote:
> On Monday 31 October 2005 07:47, Vojtech Pavlik wrote:
> > On Mon, Oct 31, 2005 at 02:24:02AM -0500, Dmitry Torokhov wrote:
> > > Input: atkbd - speed up setting leds/repeat state
> > > 
> > > Changing led state is pretty slow operation; when there are multiple
> > > requests coming at a high rate they may interfere with normal typing.
> > > Try optimize (skip) changing hardware state when multiple requests
> > > are coming back-to-back.
> > > 
> > > Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
> >  
> > It looks good - just two comments:
> > 
> > 	1) wmb() shouldn't be needed after set_bit()
> >
> 
> Judging by the comments in bitops only i386 implementation of set_bit
> implies memory barrier, other arches do not guarantee it. That's why
> I added wmb() there.

Hmm, OK. I always forget that atomicity doesn't imply a memory barrier.

> > 	2) maybe we want to enforce the delay before we send the 
> >            next SET_LED command.

> Well, with this patch "while true; do xset led 3; xset led -3; done"
> does not interfere with typing on my box and system load is staying
> low which means we don't have too many outstanding requests.

OK. I'll give it a spin if I find the time to.

-- 
Vojtech Pavlik
SuSE Labs, SuSE CR