Re: [RFT/PATCH] i8042: remove polling timer (v6)

[Vojtech Pavlik <vojtech@suse.cz>]

On Sun, Oct 29, 2006 at 10:34:00PM -0500, Dmitry Torokhov wrote:
> On Sunday 29 October 2006 18:20, Dave Neuer wrote:
> > On 8/23/06, Dmitry Torokhov <dtor@insightbb.com> wrote:
> > > Hi everyone,
> > >
> > > Here is another version of the patch removing polling timer from i8042
> > > which is needed if we want tickless kernel. Keyboards should now work
> > > on boxes that do not have mouse plugged in. PLease give it a test.
> > 
> >  What's the intent of this; just to allow tickless? Or is it also to
> > make the i8042 driver less racy? I ask because I've applied this over
> > (a modified) 2.6.18 on my Compaq Presario X1010us laptop which has
> > been driving me crazy w/ Synaptics problems and keyboard problems
> > (intermittent, but   frequent enough lately that I finally figured I
> > needed to do something about it).
> > 
> > If removing raciness is part of the goal, isn't the window in
> > i8042_aux_write still a problem?
> > 
> > 	if (port->mux == -1)
> > 		retval = i8042_command(&c, I8042_CMD_AUX_SEND);
> > 	else
> > 		retval = i8042_command(&c, I8042_CMD_MUX_SEND + port->mux);
> > 
> >         /* i8042_command has re-enabled interrupts;
> >            what happens if real interrupt happens here, before we call
> > the ISR ourselves? */
> > 
> > 	i8042_interrupt(0, NULL, NULL);
> > 	return retval;
> > }
> 
> Hi Dave,
> 
> i8042_interrupt() uses spinlock to serialize access to the KBC so if real
> interrupt happens before we call i8042_interrupt() manually (and it should
> normally happen) it will just process the response and second i8042_interrupt()
> will be just a no-op.

This would, however, create two reads of the i8042 controller
back-to-back, which has been a problem on old i8042's: IIRC IBM
documentation states that between the reads there should be a delay.

-- 
Vojtech Pavlik
Director SuSE Labs