Re: [i2c] [PATCH 0/4] Series to add device tree naming to i2c

On Mon, Dec 10, 2007 at 08:38:46AM +1100, Benjamin Herrenschmidt wrote:

quoted

+static struct i2c_device_id rs5c372_id[] = {
+       {"rtc-rs5c372", rtc_rs5c372a},
+       {"rs5c372a", rtc_rs5c372a},
+       {"rs5c372b", rtc_rs5c372b},
+       {"rv5c386", rtc_rv5c386},
+       {"rv5c387a", rtc_rv5c387a},
+       DT_NAME({"ricoh,rs5c372a", rtc_rs5c372a},)
+       DT_NAME({"ricoh,rs5c372b", rtc_rs5c372b},)
+       DT_NAME({"ricoh,rv5c386", rtc_rv5c386},)
+       DT_NAME({"ricoh,rv5c387a", rtc_rv5c387a},)
+       {},

But what's the point in making these names specific to device trees?
They are perfectly valid names for the devices that could be used from
any platform.

The more I think about it, the more I tend to agree that tagging isn't
necessary and you are right. We should just match the name against the
"compatible" property of the OF nodes (which mean we need to support
multiple matches though since "compatible" is a list of strings).

Now, I have a question about your example: Why do you have both
"rs5c372a" and "ricoh,rs5c372a" ?

I would argue that we should keep only the later...

I think existing platforms register with the simpler name, so they would
need to be changed. Having both shouldn't do harm though, we tend to
sometimes do the same with of_platform drivers for legacy reasons.

Unless someone adds a ridicously simple match name (by mistake or
whatever), it shouldn't be a problem. And if someone does it to some
dts/firmware, then we'll just need to add device tree fixups to set the
"vendor,product" string instead, yet again similar to how we have to do
sometimes with other drivers/devices.


-Olof