[i2c] Complex I2C device

[Jean Delvare]

Hi Rodolfo,

On Thu, 31 Jan 2008 10:16:19 +0100, Rodolfo Giometti wrote:
> On Wed, Jan 30, 2008 at 09:32:38PM +0100, Guennadi Liakhovetski wrote:
> > Hi Rodolfo,
> > 
> > On Wed, 30 Jan 2008, Rodolfo Giometti wrote:
> > 
> > > my custom board has a «complex» device... let me explain better: it's
> > > formed by a custom I2C device and an I/O extender PCA9539.
> > > 
> > > Ascii art:
> > > 
> > > 
> > >                 +---------+
> > >           --+---+         |
> > >             |   | PCA9539 |
> > >             |   +---------+
> > > Bus I2C ->> |      | | |
> > >             |      | | | <<--- GPIOs
> > >             |      | | |
> > >             |   +---------+
> > >             +---+         |
> > >                 |   CHIP  |
> > >                 +---------+  
> > > 
> > > Some input GPIOs of CHIP are managed by the PCA9539. So fisically I
> > > have two devices but logically they are merged together into one.
> > > 
> > > I can send commands to CHIP by both I2C bus and the GPIOs, which in
> > > turn are controlled by the PCA9539.
> > > 
> > > Can you please suggest me the best way to manage this problem? My
> > > solution was to provide PCA9539's driver of some exported symbols and
> > > using them into the CHIP's driver. Is that right? Or, can I "call" (in
> > > any way) PCA9539 driver's methods from CHIP's driver?
> > 
> > First, you want to use gpiolib, which in -mm tree now lives under 
> > drivers/gpio and includes a driver for pca9539. With it the pca9539 driver 
> > registers a gpio controller with the system, and then your driver for the 
> > other part can just request respective GPIOs as if they were normal CPU 
> > GPIOs or whatever else. And toggle them using the standard API. I will 
> > (hopefully tomorrow) post an example of exactly this to the 
> > video4linux-list at redhat.com list. In my case this other chip is a CMOS 
> > camera.
> 
> Great, this would be a good solution for this scenario... but I'm
> looking something more "general". A related problem of mine is on
> another custom board which has the following:

Guennadi's solution (based on David Brownell's work) is actually pretty
generic as far as GPIOs are concerned. I like it.

> 
> 
>                  +---------+
>            --+---+ Battery |
>              |   | Manager |
>              |   +---------+
>  Bus I2C ->> |
>              |
>              |
>              |   +---------+
>              +---+         |
>                  |  CHIP   |
>                  +---------+
> 
> A (complex) battery pack are managed by a "battery manager" and a
> custom chip connected by the I2C bus (my hardware designer _loves_ I2C
> bus :). Even these devices can be logically considered as only one
> (big) battery...
> 
> Looking at I2C code, which uses the new style driver model, I see the
> function i2c_register_board_info() where we can specify the bus number
> and the devices list attached on that I2C bus, so I suppose we can add
> a function as:
> 
> 	struct i2c_client *i2c_find_client(int busnum, unsigned short addr);
> 
> which can return the i2c_client struct associated to the given bus
> number and i2c address.
> 
> This should resolve any possible problem related to every I2C
> "complex" device. Shouldn't it?

This would be doing things the wrong way around IMHO. Assuming that
there is no clean interface between both physical components of your
complex logical device, that would make a GPIO-like solution possible,
what you really have is a single device with 2 I2C addresses. We
support this already thanks to the i2c_new_dummy() function, so I see
no need for an additional interface.

-- 
Jean Delvare