Re: [PATCH 3/3] clk: split of_clk_get_parent_name() into two functions

[Masahiro Yamada]

2015-12-01 9:49 GMT+09:00 Stephen Boyd <sboyd@xxxxxxxxxxxxxx>:
> On 11/24, Masahiro Yamada wrote:
>> Hi Stephen,
>>
>>
>> 2015-11-22 14:44 GMT+09:00 Masahiro Yamada <yamada.masahiro@xxxxxxxxxxxxx>:
>> > Hi Stephen,
>> >
>> >
>> > 2015-11-21 9:37 GMT+09:00 Stephen Boyd <sboyd@xxxxxxxxxxxxxx>:
>> >> On 11/20, Masahiro Yamada wrote:
>> >>> Currently, there is no function to get the clock name of the given
>> >>> node.  Create a new helper function, of_clk_get_name().  This is
>> >>> useful to get the clock name where "clock-indices" property is used.
>> >>>
>> >>>   of_clk_get_name(): get the clock name in the given node
>> >>>   of_clk_get_parent_name(): get the name of the parent clock
>> >>>
>> >>> Signed-off-by: Masahiro Yamada <yamada.masahiro@xxxxxxxxxxxxx>
>> >>> ---
>> >>>
>> >>> I want to use of_clk_get_name() for my clk drivers for my SoCs,
>> >>> which I will submit later.
>> >>>
>> >>> I found this helper function is useful.
>> >>
>> >> I don't see how this is useful. Is the new driver so generic it
>> >> doesn't know what clocks it's outputting? We've been trying to
>> >> move people away from using clock-output-names, so most likely
>> >> this sort of information should be conveyed from DT via the
>> >> compatible string and a table in the driver that matches up the
>> >> compatible string with the list of clock names.
>> >
>> >
>> > What I implemented in my driver was:
>> >
>> > [1] Use the clock names built in the driver, if "clock-output-names"
>> >     is missing in the DT.
>> >
>> > [2] If "clock-output-names" is specified, the driver overrides
>> >     the clock names, respecting the "clock-output-names".
>> >
>> >
>> > The following is a snippet from my driver code.
>> >
>> >
>> >     /*
>> >      * update the clock name with the one provided by
>> >      * clock-output-names property, if exists
>> >      */
>> >     new_name = of_clk_get_name(np, index);
>> >     if (new_name)
>> >             init_data[i].name = new_name;
>> >     else
>> >             pr_info("DT does not specify output name for %s[%d]\n",
>> >                     np->name, index);
>> >
>> >
>> > I read the Documentation/devicetree/bindings/clock-bindings.txt
>> > before I stared my driver development.
>> >
>> > I did not know we are deprecating the "clock-output-names".
>> > (Should it be mentioned in the clock-bindings.txt?)
>>
>>
>>
>>
>> Please let me clarify the driver implementation guideline.
>>
>> [1] Do not add "clock-output-names" in new device trees.
>> [2] New clock drivers should not rely on "clock-output-names" at all.
>>
>>
>> Is this correct?
>>
>
> Seems a little extreme, but yes, we would like to see clock
> provider drivers stop using clock-output-names. The binding isn't
> deprecated, so we shouldn't say that in the binding document, but
> perhaps we could add a comment that it's strongly suggested that
> another way be found.
>
> The only use I can think of is for something like fixed rate
> clocks or other pure DT described clocks that want to tell us
> their name. But the end goal is to make that irrelevant by using
> something besides strings (or at least, user definded strings) to
> describe the linkages between clocks in the clock tree. When this
> is done, the clock-output-names property will be unused.

Right.

What is weird about clk_register() is that
it takes the parent name(s), not pointers of the parent(s).

The current way allows us to add clock providers in any order,
but I am not sure if this flexibility is really needed.




-- 
Best Regards
Masahiro Yamada
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/