Re: [RFC] I2C: fsl-i2c: make device probing configurable via FDT

On 7/16/08, Grant Likely [off-list ref] wrote:

On Wed, Jul 16, 2008 at 10:24:22AM -0400, Jon Smirl wrote:
 > On 7/16/08, Wolfgang Grandegger [off-list ref] wrote:
 > > Jon Smirl wrote:
 > >
 > > > On 7/16/08, Wolfgang Grandegger [off-list ref] wrote:
 > > >
 > > > > Currently, the I2C buses are probed for HWMON I2C devices, which might
 > > > >  not be acceptable in same cases. This patch makes device probing
 > > > >  configurable through the property "probe" of the FDT I2C device node:
 > > > >
 > > >
 > > > All this patch seems to be doing is removing class I2C_CLASS_HWMON via
 > > > a device tree flag, PROBE.
 >
 > If this is implemented it shouldn't be a device tree option, it should
 > be a compile time option in the Kconfig system. We don't want to
 > pollute a platform independent device tree with Linux clutter.


I'm not sure about this.  It is somewhat describing the hardware.  It
 sounds like it is saying "hey, I've got an i2c bus here, but I don't
 know what is on it; so instead of providing a set of child nodes, you
 should probe this bus to find out what it has".

Probing an i2c bus does not necessarily come to a good conclusion. The
probes for some chips can alter the states in others. People have
accidentally changed voltage settings and fried CPU chips. The process
is not well defined.

I'm not saying remove i2c probing altogether, some things require
probing like EDID data from a monitor. Since these are embedded system
I just think it is pretty unlikely that stuff like that is going to
get hooked to an i2c-mpc bus.  That was behind the idea to push it
back to a compile time option.

 I don't see any problem specifying it in the device tree.

 Making it a Kconfig option forces *all* i2c busses on supported
 platforms into either probe or no-probe mode.  Not a multiplatform
 friendly solution.  Heck, we've got multiple i2c busses on some chips and
 it is conceivable that only one would need to be probed.

 Cheers,

g.

-- 
Jon Smirl
jonsmirl@gmail.com