Re: "cell-index" vs. "index" vs. no index in I2C device nodes

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

On Wed, Jun 4, 2008 at 8:41 PM, David Gibson <dwg@au1.ibm.com> wrote:
> On Wed, Jun 04, 2008 at 09:19:42PM -0500, Josh Boyer wrote:
>> On Wed, 4 Jun 2008 10:43:51 -0500
>> Scott Wood <scottwood@freescale.com> wrote:
>>
>> > On Wed, Jun 04, 2008 at 10:24:15AM -0500, Timur Tabi wrote:
>> > > Stefan Roese wrote:
>> > > > I'm wondering what is currently recommended in the I2C device tree nodes? The
>> > > > current IBM I2C driver (i2c-ibm_iic.c) checks "index" and most FSL dts files
>> > > > use "cell-index". Some 4xx dts files implement "cell-index" some have no
>> > > > index at all.
>> > > >
>> > > > So what should be used here. Please advise and I'll prepare a patch for it.
>> > >
>> > > I just posted a patch for the FSL I2C driver to check for cell-index.  I'm under
>> > > the impression that cell-index is the standard for enumerating devices in the
>> > > device tree.
>> >
>> > No, it's the standard for correlating devices with portions of a shared
>> > register block elsewhere.  Your use in the I2C node is merely a hack to
>> > deal with Linux wanting to deal with indices rather than pointers,
>> > combined with a lack of a decent way to look up a device struct from the
>> > device node.
>>
>> So if possible, I'd like to eliminate the *index stuff all together
>> from the 4xx driver.  The private data structure contains an idx
>> parameter, but this can be populated based on probe order or something.
>>
>> >From a device tree perspective, index and cell-index are both
>> incorrect.  The IIC macros don't share register blocks with anything,
>> are enumerated as unique instances per macro in the device tree, and
>> should be able to be distinguished by "regs" and/or unit address.
>>
>> Does anyone disagree with that?
>
> Hear, hear.
>
> Aliases can also provide a reasonable way of enumerating devices, if
> "reg" isn't suitable on its own.  Though obviously, drivers will need
> some sort of fallback if suitable aliases don't exist in the tree.

The fallback is to just let the i2c layer auto-assign an ID.  The only
reason I can think of to want to assign a specific id to an i2c bus is
so that a userspace application can reference a specific bus.  The
drivers really shouldn't care.

Cheers,
g.

-- 
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.