Hey Phil,

(CCing Jean for some additional expertise)

quoted

quoted

In preparation to adding of_i2c_setup_smbus_alert() move
i2c_setup_smbus_alert() to core module. of_i2c_setup_smbus_alert()
will call i2c_setup_smbus_alert() and this avoid module dependecy issues.

I am not very happy with this but don't want to cause another delay. So,
I hope we can discuss and fix it incrementally.

From what it does, I really think both setup_alert functions belong
to i2c-smbus.c. Three possibilities come to my mind (untested,
though):

a) use try_then_request_module somehow

b) add to CONFIG_I2C something like: select I2C_SMBUS if OF

c) get rid of i2c-smbus.c entirely and move it all into the core

Dunno if a) or b) have been tried in the course of this series
already?

Nope hadn't tried either a) or b). a) is not something I was aware of.
Had a brief play and not sure how to make this work. This gives me the
same linking problem.

Which one? I2C core is built-in and the driver is a module? And
i2c-smbus is a module then, too?

A variant of b) was tried by selecting I2C_SMBUS from driver modules
having smbalert support, but that caused problems if the i2c core was
builtin and the driver was a module. But this looks to work for me.

It should work(tm). Yet, it will cause overhead, because OF is big these
days and I2C_SMBUS is still rare.

A variant of c) with option to exclude the i2c-smbus code still.
Add i2c-smbus to the i2c-core module by:
Change Kconfig I2C_SMBUS to a bool
add i2c-smbus to i2c-core object.
fixup duplicate init_module calls etc.

I see. From my side, why not. But I'd really like Jean's opinion here on
why i2c-smbus.c was a seperate module and not included in the core.
Because I split the I2C core into multiple source files recently, we
could have a seperate source file for i2c-smbus-alert without ifdeffery
but with some Makefile magic.

Possibly also rename i2c-smbus to i2c-smbus-alert?

Why the rename? I thought we put all the code into the core?

Let me know your preference and I'll put something together.

Thanks for the assistance. I am afraid we need a bit more discussion
first, though.

All the best,

   Wolfram