Re: [PATCH RFC 1/1] gpio: mcp23s08: convert driver to DT

[Grant Likely <grant.likely@secretlab.ca>]

On Wed, 6 Feb 2013 10:31:04 +0100, Lars Poeschel <poeschel@lemonage.de> wrote:
> On Tuesday 05 February 2013 at 15:29:09, Grant Likely wrote:
> > On Thu, 31 Jan 2013 21:51:36 +0100, Linus Walleij 
> <linus.walleij@linaro.org> wrote:
> > > On Thu, Jan 31, 2013 at 4:58 PM, Lars Poeschel <larsi@wh2.tu-dresden.de> 
> wrote:
> > > > --- /dev/null
> > > > +++ b/Documentation/devicetree/bindings/gpio/gpio-mcp23s08.txt
> > > > @@ -0,0 +1,27 @@
> > > > +Microchip MCP2308/MCP23S08/MCP23017/MCP23S17 driver for
> > > > +8-/16-bit I/O expander with serial interface (I2C/SPI)
> > > > +
> > > > +Required properties:
> > > > +- compatible : Should be "mcp,mcp23s08-gpio", "mcp,mcp23s17-gpio",
> > > > +                       "mcp,mcp23008-gpio" or "mcp,mcp23017-gpio"
> > > > +- base : The first gpio number that should be assigned by this chip.
> > > 
> > > No. We do not tie the global GPIO numbers into the device tree.
> > > 
> > > In the DT GPIOs are referenced by ampersand <&gpio0 1 2>
> > > notation referring to the instance, so as you realize DT itself
> > > has no need for that number.
> > > 
> > > Further it is not OS-neutral.
> > > 
> > > You have to find another way to handle this in the driver code.
> > > In worst case: use AUXDATA.
> > 
> > Hi Lars,
> > 
> > The trick is to declare the io expander to be a "gpio-controller" and
> > use the #gpio-cells property to declare how many cells (32-bit numbers)
> > are need to specify a single gpio line. Most gpio controllers use
> > "gpio-cells=<2>"; The first cell is the *controller local* gpio
> > number, and the second cell is used for flags. That way your gpio
> > controller can be referenced by other nodes in the tree with a "gpios"
> > property.
> > 
> > You can find lots of examples of this in the tree.
> 
> Linus, Grant, thanks for the explanations. I think I have catched where it 
> should go.
> The thing that confused me was, that the platform_data for the chip has a 
> mandatory "base" member, that sets the linux global gpio number at which the 
> chip should appear. A value of -1 for automatic assigning gpio number is not 
> allowed, the chip will not probe.
> I have to change the driver to allow at least this -1 as an additional value.
> As Linus pointed out, it is not desirable to set the global gpio base number 
> from device tree, right ? If I have 3 instances of this chips then, how can 
> userspace sw distinguish then to which one it is talking ?

You look in sysfs to find the chip you are interested in. That is the
place to find out how dynamic numbers have been assigned.

g.