[PATCH] drivers: create a pinmux subsystem v2

[linus.walleij@linaro.org (Linus Walleij)]

2011/5/18 Stephen Warren <swarren@nvidia.com>:

>> +static struct foo_pmx_func myfuncs[] = {
>> + ? ? {
>> + ? ? ? ? ? ? .name = "spi0-0",
>> + ? ? ? ? ? ? .pins = spi0_0_pins,
>> + ? ? ? ? ? ? .num_pins = ARRAY_SIZE(spi0_1_pins),
>> + ? ? },
>> + ? ? {
>> + ? ? ? ? ? ? .name = "i2c0",
>> + ? ? ? ? ? ? .pins = i2c0_pins,
>> + ? ? ? ? ? ? .num_pins = ARRAY_SIZE(i2c0_pins),
>> + ? ? },
>> + ? ? {
>> + ? ? ? ? ? ? .name = "spi0-1",
>> + ? ? ? ? ? ? .pins = spi0_1_pins,
>> + ? ? ? ? ? ? .num_pins = ARRAY_SIZE(spi0_1_pins),
>> + ? ? },
>> +};
>
> Rather than defining a custom type (foo_pmx_func) for this array inside
> each driver, and then having to implement _list, _get_fname, _get_pins
> below, how about:
>
> * pinmux core defines a basic structure containing all the information
> ?that the core needs from the specific implementation.
>
> * This structure would need a field to point at the implementation-
> ?specific data.
>
> * We could get rid of _list, _get_fname, _get_pins completely from
> ?pinmux_ops.
>
> pinmux.h:
>
> struct pinmux_function {
> ? ? ? ?char *name;
> ? ? ? ?const unsigned int *pins;
> ? ? ? ?const unsigned num_pins;
> ? ? ?void *driver_data;
> };
>
> driver source:
>
> struct foo_pmx_func {
> ? ? ? ?int register;
> ? ? ? ?int mask;
> ? ? ? ?int value;
> };
>
> static struct foo_pmx_func spi0_0_func = {
> ? ? ? ?FOO_REG_PMX_A,
> ? ? ? ?0x30,
> ? ? ? ?0x10,
> };
> ...
> static struct pinmux_function myfuncs[] = {
> ? ? ? ?{
> ? ? ? ? ? ? ? ?.name = "spi0-0",
> ? ? ? ? ? ? ? ?.pins = spi0_0_pins,
> ? ? ? ? ? ? ? ?.num_pins = ARRAY_SIZE(spi0_1_pins),
> ? ? ? ? ? ? ? ?.driver_data = &spi0_0_func,
> ? ? ? ?},
> ? ? ? ?{
> ? ? ? ? ? ? ? ?.name = "i2c0",
> ? ? ? ? ? ? ? ?.pins = i2c0_pins,
> ? ? ? ? ? ? ? ?.num_pins = ARRAY_SIZE(i2c0_pins),
> ? ? ? ? ? ? ? ?.driver_data = &i2c0_func,
> ? ? ? ?},
> ? ? ? ?{
> ? ? ? ? ? ? ? ?.name = "spi0-1",
> ? ? ? ? ? ? ? ?.pins = spi0_1_pins,
> ? ? ? ? ? ? ? ?.num_pins = ARRAY_SIZE(spi0_1_pins),
> ? ? ? ? ? ? ? ?.driver_data = &spi0_1_func,
> ? ? ? ?},
> };
>
> This would remove some boiler-plate code from the SoC drivers,
> although it might be considered a bad breaking of abstraction barriers?

Yes it does, however I didn't want to make the initial submission
feature creepy. So I would like this to go in as is, then refactor
drivers to get help from the framework later on, if we see that
it is needed. (So the solution would evolve gradually rather
than being too much designed-in from the beginning.)

Do you think the driver support functions are needed from start?
I could do it I think...

Yours,
Linus Walleij