[PATCH 3/4] clk: samsung: Add driver to control CLKOUT line on Exynos SoCs

[Tomasz Figa]

On 22.05.2014 14:25, Sylwester Nawrocki wrote:
> On 22/05/14 14:01, Tomasz Figa wrote:
>>>>>>>> [ ... ]
>>>>>>>>
>>>>>>>>>> +    clkout->clk_table[0] = clk_register_composite(NULL, "clkout",
>>>>>>>>>> +                            parent_names, parent_count, &clkout->mux.hw,
>>>>>>>>>> +                            &clk_mux_ops, NULL, NULL, &clkout->gate.hw,
>>>>>>>>>> +                            &clk_gate_ops, CLK_SET_RATE_PARENT
>>>>>>>>>> +                            | CLK_SET_RATE_NO_REPARENT);
>>>>>>>>
>>>>>>>> Would you please remove CLK_SET_RATE_NO_REPARENT flag from here? Let me
>>>>>>>> know if you have reservations against this.
>>>>>>
>>>>>> The problem with clock reparenting is that there are certain parent
>>>>>> clocks of CLKOUT, rate of which changes at runtime, e.g. clocks derived
>>>>>> from APLL or bus clocks, which can be reconfigured by cpufreq or devfreq.
>>>>>>
>>>>
>>>> +CC: Sylwester Nawrocki
>>>>
>>>> Okay. But in cases where there is only 1 valid parent clock provided
>>>> through DT (at the moment for Exynos5250/Exynos5420), would it be safe
>>>> to set that clock as the parent of CLKOUT?
>>
>> This is not something to rely on. I have simply omitted remaining CLKOUT
>> parents on Exynos 5 SoCs, as I don't have any board with them on which I
>> could test this. Eventually they will be added.
>>
>>>> Otherwise, this clock is
>>>> not usable ATM.
>>
>> On many boards it is already configured properly by the bootloader.
>> Although I don't see any reason why you couldn't reparent it in
>> (board-specific) sound card driver right now.
> 
> This would require passing the parent's clock specifier in 'clocks'
> property of the sound card device node, which I assume is not something
> we're generally expected to do in mainline. Although some drivers
> happen to be doing it already I think that's a bad example. It sounds
> like an abuse of the current clock bindings.

Well, an already board-specific driver might have some knowledge of the
SoC, so it might not be a huge problem, but I agree that this is not an
optimal solution.

Best regards,
Tomasz