[linux-sunxi] Re: [PATCH v3 1/5] i2c: mv64xxx: Add reset deassert call

[Chen-Yu Tsai]

On Fri, Mar 7, 2014 at 6:34 PM, Russell King - ARM Linux
<linux at arm.linux.org.uk> wrote:
> On Fri, Mar 07, 2014 at 11:07:51AM +0100, Maxime Ripard wrote:
>> Hi Russell,
>>
>> On Fri, Mar 07, 2014 at 09:52:23AM +0000, Russell King - ARM Linux wrote:
>> > On Tue, Mar 04, 2014 at 05:28:37PM +0100, Maxime Ripard wrote:
>> > > The Allwinner A31 SoC using that IP has a reset controller maintaining
>> > > it reset unless told otherwise.
>> > >
>> > > Add some optional reset support to the driver.
>> > >
>> > > Signed-off-by: Maxime Ripard <maxime.ripard at free-electrons.com>
>> > > Reviewed-by: Gregory CLEMENT <gregory.clement at free-electrons.com>
>> > > Tested-by: Gregory CLEMENT <gregory.clement at free-electrons.com>
>> >
>> > This appears to be causing some build errors in Olof's next builder
>> > for many of the ARM platforms which make use of this:
>> >
>> > drivers/i2c/busses/i2c-mv64xxx.c:924: undefined reference to `reset_control_assert'
>> > drivers/i2c/busses/i2c-mv64xxx.c:904: undefined reference to `reset_control_assert'
>> > drivers/i2c/busses/i2c-mv64xxx.c:771: undefined reference to `devm_reset_control_get'
>> > drivers/i2c/busses/i2c-mv64xxx.c:778: undefined reference to `reset_control_deassert'
>>
>> The reset framework doesn't define its functions when its not
>> selected, and somehow I think it was not here. What's odd is that
>> there is an explicit select on RESET_CONTROLLER in the Kconfig. Maybe
>> it's the circular dependency issue that has been reported that cause
>> this and Wolfram sent a patch for: http://patchwork.ozlabs.org/patch/327573/
>
> If that patch has been taken, then yes, it will have caused the above -
> because now we have Dove and Kirkwood platforms trying to build this
> driver without RESET_CONTROLLER being set.
>
> The problem with depending on RESET_CONTROLLER is that then these
> platforms end up without their I2C controller - because there's nothing
> which enables RESET_CONTROLLER in their configuration.
>
> Since RESET_CONTROLLER is not required for those platforms, it really
> should be optional - and I think the real fix is for the reset controller
> support to provide stub functions.

Philipp Zabel suggested that adding a _optional variant that provides stubs
and doesn't depend on RESET_CONTROLLER is probably better. This keeps the
compile time checks for drivers requiring it.

See: https://lkml.org/lkml/2014/1/10/220

I ended up dropping my patch though.


Cheers,
ChenYu