The code was setting the bit 21 of the CPCCR register to use a divider of 2 for the "pll half" clock, and clearing the bit to use a divider of 1. This is the opposite of how this register field works: a cleared bit means that the /2 divider is used, and a set bit means that the divider is 1. Restore the correct behaviour using the newly introduced .div_table field. Signed-off-by: Paul Cercueil <paul@crapouillou.net> --- drivers/clk/ingenic/jz4740-cgu.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) diff --git a/drivers/clk/ingenic/jz4740-cgu.c b/drivers/clk/ingenic/jz4740-cgu.c index a7f8ce60c957..0957ba4a40a5 100644 --- a/drivers/clk/ingenic/jz4740-cgu.c +++ b/drivers/clk/ingenic/jz4740-cgu.c @@ -53,6 +53,10 @@ static const u8 jz4740_cgu_cpccr_div_table[] = { 1, 2, 3, 4, 6, 8, 12, 16, 24, 32, }; +static const u8 jz4740_cgu_pll_half_div_table[] = { + 2, 1, +}; + static const struct ingenic_cgu_clk_info jz4740_cgu_clocks[] = { /* External clocks */ @@ -86,7 +90,10 @@ static const struct ingenic_cgu_clk_info jz4740_cgu_clocks[] = { [JZ4740_CLK_PLL_HALF] = { "pll half", CGU_CLK_DIV, .parents = { JZ4740_CLK_PLL, -1, -1, -1 }, - .div = { CGU_REG_CPCCR, 21, 1, 1, -1, -1, -1 }, + .div = { + CGU_REG_CPCCR, 21, 1, 1, -1, -1, -1, + jz4740_cgu_pll_half_div_table, + }, }, [JZ4740_CLK_CCLK] = { -- 2.21.0.593.g511ec345e18
Quoting Paul Cercueil (2019-07-01 04:36:06)
> The code was setting the bit 21 of the CPCCR register to use a divider
> of 2 for the "pll half" clock, and clearing the bit to use a divider
> of 1.
>
> This is the opposite of how this register field works: a cleared bit
> means that the /2 divider is used, and a set bit means that the divider
> is 1.
>
> Restore the correct behaviour using the newly introduced .div_table
> field.
>
> Signed-off-by: Paul Cercueil <paul@crapouillou.net>
> ---
Applied to clk-next. Does this need a fixes tag?
Le mer. 7 août 2019 à 23:33, Stephen Boyd <sboyd@kernel.org> a écrit
:
> Quoting Paul Cercueil (2019-07-01 04:36:06)
>> The code was setting the bit 21 of the CPCCR register to use a
>> divider
>> of 2 for the "pll half" clock, and clearing the bit to use a divider
>> of 1.
>>
>> This is the opposite of how this register field works: a cleared bit
>> means that the /2 divider is used, and a set bit means that the
>> divider
>> is 1.
>>
>> Restore the correct behaviour using the newly introduced .div_table
>> field.
>>
>> Signed-off-by: Paul Cercueil <paul@crapouillou.net>
>> ---
>
> Applied to clk-next. Does this need a fixes tag?
It depends on commit a9fa2893fcc6 ("clk: ingenic: Add support for
divider tables") which was sent without a fixes tag, so it'd be
a bit difficult. Probably not worth the trouble.
-Paul
Quoting Paul Cercueil (2019-08-07 16:28:10)
>
>
> Le mer. 7 août 2019 à 23:33, Stephen Boyd <sboyd@kernel.org> a écrit
> :
> > Quoting Paul Cercueil (2019-07-01 04:36:06)
> >> The code was setting the bit 21 of the CPCCR register to use a
> >> divider
> >> of 2 for the "pll half" clock, and clearing the bit to use a divider
> >> of 1.
> >>
> >> This is the opposite of how this register field works: a cleared bit
> >> means that the /2 divider is used, and a set bit means that the
> >> divider
> >> is 1.
> >>
> >> Restore the correct behaviour using the newly introduced .div_table
> >> field.
> >>
> >> Signed-off-by: Paul Cercueil <paul@crapouillou.net>
> >> ---
> >
> > Applied to clk-next. Does this need a fixes tag?
>
> It depends on commit a9fa2893fcc6 ("clk: ingenic: Add support for
> divider tables") which was sent without a fixes tag, so it'd be
> a bit difficult. Probably not worth the trouble.
>
Does it need to go in as a fix for this -rc series then? Or is it not
causing issues for you so it's ok to wait until next merge window?
Le jeu. 8 août 2019 à 6:08, Stephen Boyd <sboyd@kernel.org> a écrit :
> Quoting Paul Cercueil (2019-08-07 16:28:10)
>>
>>
>> Le mer. 7 août 2019 à 23:33, Stephen Boyd <sboyd@kernel.org> a
>> écrit
>> :
>> > Quoting Paul Cercueil (2019-07-01 04:36:06)
>> >> The code was setting the bit 21 of the CPCCR register to use a
>> >> divider
>> >> of 2 for the "pll half" clock, and clearing the bit to use a
>> divider
>> >> of 1.
>> >>
>> >> This is the opposite of how this register field works: a
>> cleared bit
>> >> means that the /2 divider is used, and a set bit means that the
>> >> divider
>> >> is 1.
>> >>
>> >> Restore the correct behaviour using the newly introduced
>> .div_table
>> >> field.
>> >>
>> >> Signed-off-by: Paul Cercueil <paul@crapouillou.net>
>> >> ---
>> >
>> > Applied to clk-next. Does this need a fixes tag?
>>
>> It depends on commit a9fa2893fcc6 ("clk: ingenic: Add support for
>> divider tables") which was sent without a fixes tag, so it'd be
>> a bit difficult. Probably not worth the trouble.
>>
>
> Does it need to go in as a fix for this -rc series then? Or is it not
> causing issues for you so it's ok to wait until next merge window?
It can wait for the next merge window, yes.
-Paul