RE: [PATCH v4 1/3] mfd: Add new mfd device TPS68470

[Mani, Rajmohan]

Hi Lee, Andy,

> Subject: Re: [PATCH v4 1/3] mfd: Add new mfd device TPS68470
> 
> On Wed, Jul 26, 2017 at 11:23 AM, Lee Jones <lee.jones@xxxxxxxxxx> wrote:
> > On Tue, 25 Jul 2017, Andy Shevchenko wrote:
> >> On Tue, Jul 25, 2017 at 12:10 PM, Lee Jones <lee.jones@xxxxxxxxxx> wrote:
> 
> >> I briefly checked few ->read() and ->write() implementations and
> >> didn't find any evidence of positive numbers that can be returned.
> >> Documentation (kernel doc) doesn't shed a light on that. So, to me it
> >> sounds unspecified.
> >>
> >> So, for now (until documentation will be fixed) I would rely on if
> >> (ret < 0)
> >
> > It's not unspecified.  The regmap methods call into regcache_write(),
> > where the kerneldoc is clear.
> 

Since, we are interested in the regmap for the I2C bus here, I looked into the implementation of
 __devm_regmap_init()
	__regmap_init()
		regcache_init()
for I2C bus.

At the end of __devm_regmap_init() call from devm_regmap_init_i2c() inside tps68470_probe(), I see that both cache_bypass and defer_caching flags of i2c regmap struct are set. So, it looks regcache_write/read calls do not come into play here.

So, regmap_write()
	_regmap_write()
		map->reg_write (drivers/base/regmap/regmap.c:1665) translates to
		regmap_i2c_write(drivers/base/regmap/regmap-i2c.c:128)

These checks in regmap_i2c_write() ensure all return values from i2c_master_send() other than the requested number of bytes to write, are converted into negative values.

        if (ret == count)
                return 0;
        else if (ret < 0)
                return ret;
        else
                return -EIO;

Similar argument goes for regmap_read() as well.
With that, for regmap over I2C bus, it sounds like 'if (ret < 0)' looks to be a better choice. Please see if I missed anything here.
		
> drivers/base/regmap/regcache.c:266
> 
> " * Return a negative value on failure, 0 on success."
> 
> I can hardly find this any cleaner than "unspecified".
> 
> >  Perhaps we can also change the regmap kerneldoc headers too to match,
> > which might lessen the disparity.
> 
> I'm not familiar with the guts of regmap API, so, I would stick with if (ret < 0)
> for now until documentation specifies positive return values.
> 

Ack