Re: 4.1 touchscreen regression on n900 -- pinpointed [was Re: linux-n900 v4.1-rc4]

[Pavel Machek <pavel@ucw.cz>]

On Fri 2015-05-29 14:36:49, Dmitry Torokhov wrote:
> On Fri, May 29, 2015 at 11:17:22PM +0200, Pavel Machek wrote:
> > On Fri 2015-05-29 21:57:40, Maxime Ripard wrote:
> > > On Fri, May 29, 2015 at 09:25:05PM +0200, Pavel Machek wrote:
> > > > On Fri 2015-05-29 21:08:16, Pavel Machek wrote:
> > > > > Hi!
> > > > > 
> > > > > > mh I remember having problems with tsc2005 before. It helped to
> > > > > > reset the controller (should actually happen automatically when it
> > > > > > hangs, but I'm not sure, that it actually works).
> > > > > 
> > > > > Ok, I did some more testing, and found out rather bogus values in
> > > > > evtest:
> > > > > 
> > > > > Input device name: "TSC2005 touchscreen"
> > > > > Supported events:
> > > > >   Event type 0 (EV_SYN)
> > > > >   Event type 1 (EV_KEY)
> > > > >   Event code 330 (BTN_TOUCH)
> > > > >   Event type 3 (EV_ABS)
> > > > >   Event code 0 (ABS_X)
> > > > >   Value   2514
> > > > >  Min        0
> > > > >  Max        0
> > > > >  Fuzz       4
> > > > > 
> > > > > Which made me go through the git logs, and these patches looked
> > > > > suspicious. After a revert... yes, touchscreen works as well as it
> > > > > worked before.
> > > > > 
> > > > > 0a363a380954e10fece7cd9931b66056eeb07d56
> > > > > 3eea8b5d68c801fec788b411582b803463834752
> > > > > 
> > > > > (It is impossible to revert just 3eea..)
> > > > 
> > > > Hmm, I see:
> > > > 
> > > >                 touchscreen-max-x = <4096>;
> > > >                 touchscreen-max-y = <4096>;
> > > > ...that's n900 dts.. this should be size-x/size-y... so we have a bug
> > > > in dts.
> > > > 
> > > > But the 3eea8b5d68c801fec788b411582b803463834752 is buggy, it should
> > > > not overwrite ->maximum for axis it has no devicetree data for.
> > > 
> > > What do you mean? touchscreen-max-* _is_ device tree data for an axis.
> > > 
> > > > Maybe replacing
> > > > 
> > > > +       if (maximum || fuzz)
> > > > 
> > > > in 3eea to (maximum && fuzz)... would help, but it is late in the
> > > > cycle now, so I'd suggest just reverting 3eea8b.
> > > 
> > > No, both maximum and fuzz are optional. You can perfectly have one
> > > without another.
> > 
> > Yes, which is something you got wrong.
> > 
> > Apply this on top of 3eea8b5d68c801fec788b411582b803463834752, and
> > you'll get simpler code, that works like it did before.
> > 
> > Maxime, since you caused the regression, can I ask you to put the
> > other patches on top of it, test it still works for you and submit it
> > properly?
> > 
> > [If you want to return in !test_bit case, and are sure it does not
> > break anything, feel free to do that.]
> > 
> > Thanks,
> > 								Pavel
> > 
> > Signed-off-by: Pavel Machek <pavel@ucw.cz>
> > 
> > diff --git a/drivers/input/touchscreen/of_touchscreen.c b/drivers/input/touchscreen/of_touchscreen.c
> > index b82b520..e626c8a 100644
> > --- a/drivers/input/touchscreen/of_touchscreen.c
> > +++ b/drivers/input/touchscreen/of_touchscreen.c
> > @@ -11,39 +11,23 @@
> >  
> >  #include <linux/of.h>
> >  #include <linux/input.h>
> > -#include <linux/input/mt.h>
> >  #include <linux/input/touchscreen.h>
> >  
> > -static u32 of_get_optional_u32(struct device_node *np,
> > -			       const char *property)
> > -{
> > -	u32 val = 0;
> > -
> > -	of_property_read_u32(np, property, &val);
> > -
> > -	return val;
> > -}
> > -
> >  static void touchscreen_set_params(struct input_dev *dev,
> >  				   unsigned long axis,
> > -				   int max, int fuzz)
> > +				   char *max, char *fuzz)
> >  {
> >  	struct input_absinfo *absinfo;
> > +	struct device_node *np = dev->dev.parent->of_node;
> >  
> >  	if (!test_bit(axis, dev->absbit)) {
> > -		/*
> > -		 * Emit a warning only if the axis is not a multitouch
> > -		 * axis, which might not be set by the driver.
> > -		 */
> > -		if (!input_is_mt_axis(axis))
> > -			dev_warn(&dev->dev,
> > -				 "DT specifies parameters but the axis is not set up\n");
> > -		return;
> > +		dev_warn(&dev->dev,
> > +			 "DT specifies parameters but the axis is not set up\n");
> 
> No, this is not right either. We do not want to bitch about axis not
> being set up if there is nothing in DT mentioning this axis.

Well, we should not bitch at all, that should have been followup patch.

> Plus the code structure won't work well if we decide to add more to it
> (such as min, or flat, or resolution).

Do you want to try to cook up something that works?
									Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html