Re: [PATCH v2] gpio: Add driver for TI TPIC2810

[Linus Walleij]

On Wed, Feb 10, 2016 at 3:29 PM, Andy Shevchenko
<andy.shevchenko@xxxxxxxxx> wrote:
> On Wed, Feb 10, 2016 at 4:21 PM, Linus Walleij <linus.walleij@xxxxxxxxxx> wrote:
>> On Sun, Jan 31, 2016 at 11:52 PM, Andy Shevchenko
>> <andy.shevchenko@xxxxxxxxx> wrote:
>>
>>> It reminds me how 12 channel PWM chip is used on Intel Galileo Gen 2.
>>> Half pins are PWM, the other half is GPIO used for discrete based pin
>>> muxing and control. Nevertheless I think it's a userspace issue for
>>> now, otherwise we have to provide some 'semi-virtual' way of
>>> presenting pins as GPIO lines.
>>
>> That sounds like an MFD spawning a GPIO and a PWM cell.
>> That it is called "a PWM chip" is no big deal, it should be
>> modeled according to what it is, not what it claims to be.
>
> Although I agree with model I barely imagine how in this case drivers
> should access PWM chip registers in non-race way (take into account
> that PWM itself is connected to i2c bus).

There is a pattern for that. You add a set of accessor functions that
performs the I2C traffic in the MFD layer.

The accessor functions take a mutex. Since this is all slowpath,
waiting/preempting in a mutex is perfectly fine for all subdrivers.

Look at this:

/**
 * stmpe_reg_write() - write a single STMPE register
 * @stmpe:      Device to write to
 * @reg:        Register to write
 * @val:        Value to write
 */
int stmpe_reg_write(struct stmpe *stmpe, u8 reg, u8 val)
{
        int ret;

        mutex_lock(&stmpe->lock);
        ret = __stmpe_reg_write(stmpe, reg, val);
        mutex_unlock(&stmpe->lock);

        return ret;
}
EXPORT_SYMBOL_GPL(stmpe_reg_write);

Yours,
Linus Walleij