Message-ID: <20210306204219.6a7eae56@aktux>
Date:   Sat, 6 Mar 2021 20:42:19 +0100
From:   Andreas Kemnade <andreas@...nade.info>
To:     Jonathan Neuschäfer <j.neuschaefer@....net>
Cc:     lee.jones@...aro.org, linux-kernel@...r.kernel.org,
        Mark Brown <broonie@...nel.org>
Subject: Re: [PATCH -next] mfd: ntxec: Support for EC in Tolino Shine 2 HD

On Sat, 6 Mar 2021 20:14:46 +0100
Jonathan Neuschäfer <j.neuschaefer@....net> wrote:

> Hi,
> 
> (Cc'ing Mark Brown because of the regmap related questions)
> 
> On Sat, Mar 06, 2021 at 07:13:14PM +0100, Andreas Kemnade wrote:
> > Add the version of the EC in the Tolino Shine 2 HD
> > to the supported versions. It seems not to have an RTC
> > and does not ack data written to it.
> > The vendor kernel happily ignores write errors, using
> > I2C via userspace i2c-set also shows the error.
> > So add a quirk to ignore that error.
> > 
> > PWM can be successfully configured despite of that error.  
> 
> I'm curious, is this one of the variants with two PWM channels
> (for configurable color temperature)?
> 
No. Tolino Shine 3 and Kobo Clara HD have such things. There you have a
/sys/class/backlight/backlight_{cold,warm}.

> > 
> > Signed-off-by: Andreas Kemnade <andreas@...nade.info>
> > ---
> >  drivers/mfd/ntxec.c       | 57 ++++++++++++++++++++++++++++++++++++---
> >  include/linux/mfd/ntxec.h |  1 +
> >  2 files changed, 55 insertions(+), 3 deletions(-)
> > 
> > diff --git a/drivers/mfd/ntxec.c b/drivers/mfd/ntxec.c
> > index 957de2b03529..e7fe570127af 100644
> > --- a/drivers/mfd/ntxec.c
> > +++ b/drivers/mfd/ntxec.c
> > @@ -96,6 +96,36 @@ static struct notifier_block ntxec_restart_handler = {
> >  	.priority = 128,
> >  };
> >  
> > +static int regmap_ignore_write(void *context,
> > +			       unsigned int reg, unsigned int val)
> > +
> > +{
> > +	struct regmap *regmap = context;
> > +
> > +	regmap_write(regmap, reg, val);
> > +
> > +	return 0;
> > +}
> > +
> > +static int regmap_wrap_read(void *context, unsigned int reg,
> > +			    unsigned int *val)
> > +{
> > +	struct regmap *regmap = context;
> > +
> > +	return regmap_read(regmap, reg, val);
> > +}
> > +
> > +/* some firmware versions do not ack written data, add a wrapper */
> > +static const struct regmap_config regmap_config_noack = {
> > +	.name = "ntxec_noack",
> > +	.reg_bits = 8,
> > +	.val_bits = 16,
> > +	.cache_type = REGCACHE_NONE,
> > +	.val_format_endian = REGMAP_ENDIAN_BIG,
> > +	.reg_write = regmap_ignore_write,
> > +	.reg_read = regmap_wrap_read  
> 
> Is the read wrapper necessary? It seems to me from reading regmap.h
> that leaving .reg_read set to NULL should do the right thing, but I'm
> not sure.
>
well if we want to read from it, there need to be some function for it.
But if... I do not see anything worth to read besides of version.
I think we can leave ouf val_format_endian because a lot of stuff is
bypassed if no bus is set in regmap_init().
There is e.g. a goto skip_format_initialization.

> > +};
> > +
> >  static const struct regmap_config regmap_config = {
> >  	.name = "ntxec",
> >  	.reg_bits = 8,
> > @@ -109,10 +139,15 @@ static const struct mfd_cell ntxec_subdevices[] = {
> >  	{ .name = "ntxec-pwm" },
> >  };
> >  
> > +static const struct mfd_cell ntxec_subdev_pwm[] = {
> > +	{ .name = "ntxec-pwm" },
> > +};  
> 
> ntxec_subdevices vs. ntxec_subdev_pwm seems slightly inconsistent in
> naming. ntxec_subdevices_pwm would be a wrong plural, but IMHO slightly
> better because of consistency. Maybe rename ntxec_subdevices to
> ntxec_subdev?
> 
yes, I will change it.

> > +
> >  static int ntxec_probe(struct i2c_client *client)
> >  {
> >  	struct ntxec *ec;
> >  	unsigned int version;
> > +	bool has_rtc;
> >  	int res;
> >  
> >  	ec = devm_kmalloc(&client->dev, sizeof(*ec), GFP_KERNEL);
> > @@ -137,6 +172,15 @@ static int ntxec_probe(struct i2c_client *client)
> >  	/* Bail out if we encounter an unknown firmware version */
> >  	switch (version) {
> >  	case NTXEC_VERSION_KOBO_AURA:
> > +		has_rtc = true;
> > +		break;
> > +	case NTXEC_VERSION_TOLINO_SHINE2:
> > +		has_rtc = false;
> > +		ec->regmap = devm_regmap_init(ec->dev, NULL,
> > +					      ec->regmap,
> > +					      &regmap_config_noack);  
> 
> Ah— A custom regmap stacked on top of the old regmap… I think this
> deserves a comment.
> 
Yes, devm_regmap_init_i2c() sets a different set of callbacks depending
on circumstances. Seems to be the easiest way to avoid duplicating too
much code. I really hope that there are not so much devices requiring
such a dirty stuff that adding such thing to the generic regmap code
would be justified.

> > +		if (IS_ERR(ec->regmap))
> > +			return PTR_ERR(ec->regmap);
> >  		break;
> >  	default:
> >  		dev_err(ec->dev,
> > @@ -155,7 +199,6 @@ static int ntxec_probe(struct i2c_client *client)
> >  		 */
> >  		res = regmap_write(ec->regmap, NTXEC_REG_POWERKEEP,
> >  				   NTXEC_POWERKEEP_VALUE);
> > -		if (res < 0)
> >  			return res;  
> 
> This deletion looks like a mistake.
> 
Oops, sorry.

> >  
> >  		if (poweroff_restart_client)
> > @@ -181,8 +224,16 @@ static int ntxec_probe(struct i2c_client *client)
> >  
> >  	i2c_set_clientdata(client, ec);
> >  
> > -	res = devm_mfd_add_devices(ec->dev, PLATFORM_DEVID_NONE, ntxec_subdevices,
> > -				   ARRAY_SIZE(ntxec_subdevices), NULL, 0, NULL);
> > +	if (has_rtc)
> > +		res = devm_mfd_add_devices(ec->dev, PLATFORM_DEVID_NONE,
> > +					   ntxec_subdevices,
> > +					   ARRAY_SIZE(ntxec_subdevices),
> > +					   NULL, 0, NULL);
> > +	else
> > +		res = devm_mfd_add_devices(ec->dev, PLATFORM_DEVID_NONE,
> > +					   ntxec_subdev_pwm,
> > +					   ARRAY_SIZE(ntxec_subdev_pwm),
> > +					   NULL, 0, NULL);  
> 
> At some point, it will probably be simpler to have
> 
> 	struct mfd_cell *subdev = ntxec_subdevices;
> 	size_t subdev_size = ARRAY_SIZE(ntxec_subdevices);
> 
> on top of the probe function and override them in the switch statement,
> but at this point I think it doesn't matter, and either way is fine.
> 
Yes, that might be a good idea.

Regards,
Andreas

Powered by blists - more mailing lists