Re: [PATCH v2 3/3] gpio: pcf857x: Add OF support

[Tomasz Figa <tomasz.figa@gmail.com>]

On Saturday 24 of August 2013 02:54:07 Laurent Pinchart wrote:
> Hi Tomasz,
> 
> Thank you for the review.
> 
> On Saturday 24 August 2013 02:41:59 Tomasz Figa wrote:
> > On Tuesday 20 of August 2013 01:04:54 Laurent Pinchart wrote:
> > > Add DT bindings for the pcf857x-compatible chips and parse the
> > > device
> > > tree node in the driver.
> > > 
> > > Signed-off-by: Laurent Pinchart
> > > <laurent.pinchart+renesas@ideasonboard.com> ---
> > > 
> > >  .../devicetree/bindings/gpio/gpio-pcf857x.txt      | 71
> > >  +++++++++++++++++ drivers/gpio/gpio-pcf857x.c                     
> > >    | 57 ++++++++++++++--- 2 files changed, 119 insertions(+), 9
> > >  deletions(-)
> > >  
> > >  create mode 100644
> > > 
> > > Documentation/devicetree/bindings/gpio/gpio-pcf857x.txt
> > > 
> > > diff --git a/Documentation/devicetree/bindings/gpio/gpio-pcf857x.txt
> > > b/Documentation/devicetree/bindings/gpio/gpio-pcf857x.txt new file
> > > mode
> > > 100644
> > > index 0000000..df94462
> > > --- /dev/null
> > > +++ b/Documentation/devicetree/bindings/gpio/gpio-pcf857x.txt
> > > @@ -0,0 +1,71 @@
> 
> [snip]
> 
> > > +  - pins-initial-state: Bitmask that specifies the initial state of
> > > +    each pin. When a bit is set to zero, the corresponding pin will
> > > be
> > > +    initialized to the input (pulled-up) state. When the  bit is
> > > set to +    one, the pin will be initialized the the low-level
> > > output state. If +    the property is not specified all pins will
> > > be initialized to the +    input state.
> > 
> > Hmm, do you actually need to know whether those pins are outputs or
> > inputs before they get used for first time? I believe any driver
> > using GPIO will call gpio_direction_{in,out}put() before it starts
> > using the pin, which will initialize the pin to a known state.
> > 
> > What I'd suggest is making the driver handle this by having a bit mask
> > that marks states of pins as defined and flagging all pins as
> > undefined by default. Then any call to gpio_direction_output() or
> > _input() would mark it as defined and direction of the pin could be
> > stored in internal driver structures.
> 
> The problem is that all pins are controlled through a single I2C write.
> Setting the direction of a pin will set the direction of all other pins.
> I thus need to know what the initial settings are to avoid glitches.

Oh, that's a funny hardware, isn't it? :)

Well, I guess it can't be helped then. Sorry for the noise.

> > > +  The I/O expander can detect input state changes, and thus
> > > optionally
> > > +  act as an interrupt controller. When interrupts support is
> > > desired
> > 
> > I don't like this statement. Device tree should represent what the
> > device allows you to do, not what you want the device to do.
> > 
> > My opinion on this is that if the chip supports interrupts then it
> > should always be an interrupt-controller (unless its interrupt pin is
> > not wired on the board, but this still conforms to what I wrote
> > above).
> 
> I agree. What about the following text then ?
> 
> The I/O expander can detect input state changes, and thus optionally act
> as an interrupt controller. When the expander interrupt pin is
> connected all the following properties must be set. For more
> information please see the interrupt controller device tree bindings
> documentation available at
> Documentation/devicetree/bindings/interrupt-controller/interrupts.txt.

Sounds good.

> > > diff --git a/drivers/gpio/gpio-pcf857x.c
> > > b/drivers/gpio/gpio-pcf857x.c
> > > index 070e81f..50a90f1 100644
> > > --- a/drivers/gpio/gpio-pcf857x.c
> > > +++ b/drivers/gpio/gpio-pcf857x.c
> 
> [snip]
> 
> > > @@ -50,6 +52,27 @@ static const struct i2c_device_id pcf857x_id[] =
> > > {
> > > 
> > >  };
> > >  MODULE_DEVICE_TABLE(i2c, pcf857x_id);
> > > 
> > > +#ifdef CONFIG_OF
> > > +static const struct of_device_id pcf857x_of_table[] = {
> > > +	{ .compatible = "nxp,pcf8574", .data = (void *)8 },
> > > +	{ .compatible = "nxp,pcf8574a", .data = (void *)8 },
> > > +	{ .compatible = "nxp,pca8574", .data = (void *)8 },
> > > +	{ .compatible = "nxp,pca9670", .data = (void *)8 },
> > > +	{ .compatible = "nxp,pca9672", .data = (void *)8 },
> > > +	{ .compatible = "nxp,pca9674", .data = (void *)8 },
> > > +	{ .compatible = "nxp,pcf8575", .data = (void *)16 },
> > > +	{ .compatible = "nxp,pca8575", .data = (void *)16 },
> > > +	{ .compatible = "nxp,pca9671", .data = (void *)16 },
> > > +	{ .compatible = "nxp,pca9673", .data = (void *)16 },
> > > +	{ .compatible = "nxp,pca9675", .data = (void *)16 },
> > > +	{ .compatible = "maxim,max7328", .data = (void *)8 },
> > > +	{ .compatible = "maxim,max7329", .data = (void *)8 },
> > > +	{ .compatible = "ti,tca9554", .data = (void *)8 },
> > > +	{ }
> > > +};
> > > +MODULE_DEVICE_TABLE(of, pcf857x_of_table);
> > > +#endif
> > > +
> > > 
> > >  /*
> > >  
> > >   * The pcf857x, pca857x, and pca967x chips only expose one read and
> > >   one
> > >   * write register.  Writing a "one" bit (to match the reset state)
> > >   lets
> > > 
> > > @@ -257,14 +280,29 @@ fail:
> > >  static int pcf857x_probe(struct i2c_client *client,
> > >  
> > >  			 const struct i2c_device_id *id)
> > >  
> > >  {
> > > 
> > > -	struct pcf857x_platform_data	*pdata;
> > > +	struct pcf857x_platform_data	*pdata = client-
>dev.platform_data;
> > > +	struct device_node		*np = client->dev.of_node;
> > > 
> > >  	struct pcf857x			*gpio;
> > > 
> > > +	unsigned int			n_latch = 0;
> > > +	unsigned int			ngpio;
> > > 
> > >  	int				status;
> > > 
> > > -	pdata = client->dev.platform_data;
> > > -	if (!pdata) {
> > > +#ifdef CONFIG_OF
> > > +	if (np) {
> > 
> > Wouldn't if (IS_ENABLED(CONFIG_OF) && np) be sufficient here, without
> > the #ifdef? You would have to move the match table out of the #ifdef
> > in this case, though...
> 
> That's the exact reason why I've used #ifdef CONFIG_OF here, I didn't
> want to add the overhead of the pcf857x_of_table when CONFIG_OF isn't
> defined.

I'm not sure if I remember correctly, but I think there was something said 
in one of discussions some time ago, that we should be moving away from 
ifdef'ing such things, in favour of just having them compiled 
unconditionally.

[Adding DT maintainers on Cc for more opinions.]

Best regards,
Tomasz