This implements the two missing CLKOUT clocks for the ux500
(well really U8500/DB8500) SoC.

The clocks are initialized using a specific parent and
divider and these are specified in the device tree, see
the separate binding patch.

The implementation is a bit different in that it will only
create the clock in the clock framework if a user appears
in the device tree, rather than it being registered upfront
like most of the other clocks. This is because the clock
needs parameters for source and divider from the consumer
phandle for the clock to be set up properly when the clock
is registered.

There could be more than one user of a CLKOUT clock, but
we have not seen this in practice. If this happens the
framework prints and info and returns the previously
registered clock.

Using the clocks requires also muxing the CLKOUT1 or
CLKOUT2 to the appropriate pad. In practice this is
achived in a pinctrl handle in the DTS node for the device
using the CLKOUT clock, so this muxing is done separately
from the clock itself. Example:

  haptic@49 {
    compatible = "immersion,isa1200";
    reg = <0x49>;
    (...)
    /* clkout1 from ACLK divided by 8 */
    clocks = <&clkout_clk DB8500_CLKOUT_1 DB8500_CLKOUT_SRC_ACLK 8>;
    pinctrl-names = "default";
    pinctrl-0 = <&isa1200_janice_default>;
  };

  isa1200_janice_default: isa1200_janice {
    /* Bring out clkout1 on pin GPIO227 pin AH7 */
    janice_mux {
      function = "clkout";
      groups = "clkout1_a_1";
    };
    janice_cfg1 {
      pins = "GPIO227_AH7";
      ste,config = <&out_lo>;
    };
  (...)

This was tested successfully with the Immersion ISA1200
haptic feedback unit on the Samsung Galaxy S Advance GT-I9070
(Janice) mobile phone.

As the CLKOUT clocks need some undefined fixed rate parent
clocks that are currently missing from the PRCMU clock
implementation, the three simplest are added in this patch:
clk38m_to_clkgen, aclk and sysclk. The only parent not yet
available in the implementation is clk009, which is a kind
of special muxed and divided clock which isn't even
implemented in the vendor clock driver.
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Link: https://lore.kernel.org/r/20220414221751.323525-6-linus.walleij@linaro.orgSigned-off-by: Stephen Boyd <sboyd@kernel.org>