Re: [PATCH v5 05/14] clk: Add generic driver for Maxim PMIC clocks

[Javier Martinez Canillas <javier.martinez@collabora.co.uk>]

Hello Russell,

Thanks a lot for your suggestion.

On 06/30/2014 01:35 PM, Russell King - ARM Linux wrote:
> On Mon, Jun 30, 2014 at 12:58:57PM +0200, Javier Martinez Canillas wrote:
>> >> +       if (!max_gen->lookup)
>> >> +               return ERR_PTR(-ENOMEM);
>> >> +
>> >> +       max_gen->lookup->con_id = hw->init->name;
>> > 
>> > Also IMO,  init->name should be over-written if name is provided in DT,
>> > otherwise generic "clock-output-names" property will go futile,
>> > perhaps it should be done before clk_register.
>> > 
>> 
>> Even though Documentation/devicetree/bindings/clock/clock-bindings.txt says that
>> the "clock-output-names" property is optional I agree with you that will be
>> better to support it. So I'll add it on the next version as well.
> 
> However, remember that con_id is the _DEVICE_ specific connection name,
> not the _CLOCK_ name.  You will get a NAK from me if you violate this
> rule.
> 

Yes I know that con_id is the device specific connection name that is set by the
consumers using the "clock-names" property. But AFAIU the clk_lookup structs
added by clockdev_add() are only used by legacy non-DT drivers which don't set
their clock names and instead is the clock driver the one that set con_id to the
struct clk_init_data .name field.

I see that other drivers do the same, is that wrong as well? What value should
have con_id when registering clkdevs for legacy non-DT drivers?

DT enabled drivers whose devices define their "clock-names" are not affected by
this since the clock lookup is made by matching OF "clock-names" property + and
index.

Best regards,
Javier