Re: [PATCH 6/7] arm: mediatek: enable gpt6 on boot up to make arch timer working

[Matthias Brugger <matthias.bgg@gmail.com>]

2014-08-13 10:47 GMT+02:00 Linus Walleij <linus.walleij@linaro.org>:
> On Tue, Aug 12, 2014 at 11:02 AM, Matthias Brugger
> <matthias.bgg@gmail.com> wrote:
>> 2014-08-11 9:15 GMT+02:00 Linus Walleij <linus.walleij@linaro.org>:
>>> On Thu, Jul 31, 2014 at 6:42 PM, Matthias Brugger
>>> <matthias.bgg@gmail.com> wrote:
>>>
>>>> We enable GTP6 which ungates the arch timer clock. Apart we write the
>>>> frequency with which the timer is running in the CNTFREQ register.
>>>> In the future this should be done in the bootloader.
>>>>
>>>> Signed-off-by: Matthias Brugger <matthias.bgg@gmail.com>
>>> (...)
>>>
>>>> +       if (of_machine_is_compatible("mediatek,mt6589")) {
>>>> +               /* set cntfreq register which is not done in bootloader */
>>>> +               asm volatile("mcr p15, 0, %0, c14, c0, 0" : : "r" (13000000));
>>>
>>> I guess this is for something like the ARM arch timer in
>>> drivers/clocksource/arm_arch_timer.c
>>>
>>> Instead of doing this, use the DT property "clock-frequency" on
>>> the ARM arch timer node, as that overrides the CP15 setting.
>>>
>>>> +
>>>> +               /* turn on GPT6 which ungates arch timer clocks */
>>>> +               gpt_base = ioremap(GPT6_CON_MT65xx, 0x04);
>>>> +       }
>>>> +
>>>> +       /* enabel clock and set to free-run */
>>>> +       if (gpt_base)
>>>> +               writel(0x31, gpt_base);
>>>
>>> Why is this not done properly in the GPT driver (I guess
>>> in drivers/clocksource/mtk_timer.c) instead of remapping
>>> it and fiddling around in the machine?
>>
>> I didn't put it in the GPT driver, because that would mean, that you
>> need to have the mtk_timer to be able to use the ARM arch timer.
>> From my understanding this are two independent blocks in the SoC.
>> Apart from that, as stated in the commit message, this patch is a
>> workaround until we have a bootloader which does this for us.
>> Therefore I thought the init_time function would be the better place to do this.
>
> OK I get the point...
>
> What about the ARM arch timer fixup then, will that also be
> fixed in the boot loader eventually?

Do you refer to clock frequency in the cntfreq register? I think that
this should go in the DTS as you said.
Same holds for the magic 0x31, which should be done through defines.

Cheers,
Matthias

>
> Yours,
> Linus Walleij



-- 
motzblog.wordpress.com