Re: [PATCH 2/2] Add the of_find_i2c_device_by_node function, V4

[Jean Delvare <khali@linux-fr.org>]

On Tue, 1 Jul 2008 13:27:57 -0400, Jon Smirl wrote:
> On 7/1/08, Jean Delvare <khali@linux-fr.org> wrote:
> > On Tue, 1 Jul 2008 13:00:08 -0400, Jon Smirl wrote:
> >  > On 7/1/08, Grant Likely <grant.likely@secretlab.ca> wrote:
> >
> > > > My preference is for things like of_spi and of_i2c to go with the
> >  > >  related busses; I think it makes more sense to keep all the I2C stuff
> >  > >  together, but I've already lost that battle once.
> >  >
> >  > This is a similar problem to adding aliases to the i2c driver drivers
> >  > for the device tree names of the i2c devices. Instead we have code in
> >  > drivers/of/of_i2c.c that tries to guess the translation from device
> >  > tree to linux names. Adding aliases to the drivers would eliminate the
> >  > need for of_find_i2c_driver().
> >  >
> >  > I've previously posted patches implementing device tree names in the
> >  > drivers that used ifdef to only instantiate on powerpc builds. For
> >  > example....
> >  >
> >  > diff --git a/drivers/i2c/chips/tps65010.c b/drivers/i2c/chips/tps65010.c
> >  > index e07274d..9cd1770 100644
> >  > --- a/drivers/i2c/chips/tps65010.c
> >  > +++ b/drivers/i2c/chips/tps65010.c
> >  > @@ -571,6 +571,10 @@ static const struct i2c_device_id tps65010_id[] = {
> >  >         { "tps65011", TPS65011 },
> >  >         { "tps65012", TPS65012 },
> >  >         { "tps65013", TPS65013 },
> >  > +       OF_ID("ti,tps65010", TPS65010)
> >  > +       OF_ID("ti,tps65011", TPS65011)
> >  > +       OF_ID("ti,tps65012", TPS65012)
> >  > +       OF_ID("ti,tps65013", TPS65013)
> >  >         { },
> >  >  };
> >  >  MODULE_DEVICE_TABLE(i2c, tps65010_id);
> >
> >
> > Yeah, yeah, you've been asking for this for months already, but it's
> >  just not going to happen, sorry. You want to abuse the standard Linux
> >  alias mechanism for your personal (i.e. openfirmware) use, but that's
> >  bad. Linux drivers shouldn't have to know whether they are used in
> >  openfirmware trees and what device names are used there. And device
> >  names as seen by user-space shouldn't vary depending on whether the
> >  device comes from an openfirmware tree or not - otherwise all
> >  user-space apps need to learn about both naming conversions.
> >
> >  Unsurprisingly, no other subsystem does what you propose.
> 
> Then what are all of the PCI aliases doing?
> 
> The only difference is that you are recognizing the PCI group as a
> naming authority and not recognizing the PowerPC device tree group.
> But on the PowerPC platform that is our naming authority. That's why I
> proposed adding the names on ifdefs so that they disappear on non
> PowerPC platforms.

You're comparing PCI devices those ID is built-in, with I2C devices
with no ID. This just can't compare.

> PS - adding an alias to a driver does not change the name of the
> driver. My PCI e1000 module has about 100 aliases but it is always
> e1000.

I said device names, not driver names. Linux I2C devices have a name
attribute which contains the (Linux) device name. That's something a
number of user-space applications are relying on.

-- 
Jean Delvare