[U-Boot] [PATCH v4 03/33] pinctrl: Add the concept of peripheral IDs
Simon Glass
sjg at chromium.org
Wed Aug 26 15:20:44 CEST 2015
Hi Masahiro,
On 25 August 2015 at 21:51, Masahiro Yamada
<yamada.masahiro at socionext.com> wrote:
> 2015-08-26 13:38 GMT+09:00 Simon Glass <sjg at chromium.org>:
>> Hi Masahiro,
>>
>> On 25 August 2015 at 21:30, Masahiro Yamada
>> <yamada.masahiro at socionext.com> wrote:
>>> Hi Simon,
>>>
>>>
>>>
>>> 2015-08-25 0:12 GMT+09:00 Simon Glass <sjg at chromium.org>:
>>>> My original pinctrl patch operating using a peripheral ID enum. This was
>>>> shared between pinmux and clock and provides an easy way to specify a device
>>>> that needs to be controlled, even it is does not (yet) have a driver within
>>>> driver model.
>>>>
>>>> Masahiro's new simple pinctrl gets around this by providing a
>>>> set_state_simple() pinctrl method. By passing a device to that call the
>>>> peripheral ID becomes unnecessary. If the driver needs it, it can calculate
>>>> it itself and use it internally.
>>>>
>>>> However this does not solve the problem for peripheral clocks. The 'pure'
>>>> solution would be to pass a driver to the clock uclass also. But this
>>>> requires that all devices should have a driver, and a struct udevide. Also
>>>> a key optimisation of the clock uclass is allowing a peripheral clock to
>>>> be set even when there is no device for that clock.
>>>>
>>>> There may be a better way to achive the same goal, but for now it seems
>>>> expedient to add in peripheral ID to the pinctrl uclass. Two methods are
>>>> added - one to get the peripheral ID and one to select it. The existing
>>>> set_state_simple() is effectively the union of these.
>>>>
>>>> Signed-off-by: Simon Glass <sjg at chromium.org>
>>>> ---
>>>>
>>>> Changes in v4: None
>>>> Changes in v3: None
>>>> Changes in v2: None
>>>>
>>>> drivers/pinctrl/pinctrl-uclass.c | 33 ++++++++++++++++++++++
>>>> include/dm/pinctrl.h | 61 ++++++++++++++++++++++++++++++++++++++++
>>>> 2 files changed, 94 insertions(+)
>>>>
>>>> diff --git a/drivers/pinctrl/pinctrl-uclass.c b/drivers/pinctrl/pinctrl-uclass.c
>>>> index 01b0cf8..1d90fb0 100644
>>>> --- a/drivers/pinctrl/pinctrl-uclass.c
>>>> +++ b/drivers/pinctrl/pinctrl-uclass.c
>>>> @@ -9,6 +9,7 @@
>>>> #include <dm/device.h>
>>>> #include <dm/lists.h>
>>>> #include <dm/pinctrl.h>
>>>> +#include <dm/root.h>
>>>> #include <dm/uclass.h>
>>>> #include <linux/err.h>
>>>> #include <linux/list.h>
>>>> @@ -36,9 +37,41 @@ int pinctrl_select_state(struct udevice *dev, const char *ignored)
>>>> return ops->set_state_simple(pctldev, dev);
>>>> }
>>>>
>>>> +int pinctrl_request(struct udevice *dev, int func, int flags)
>>>> +{
>>>> + struct pinctrl_ops *ops = pinctrl_get_ops(dev);
>>>> +
>>>> + if (!ops->request)
>>>> + return -ENOSYS;
>>>> +
>>>> + return ops->request(dev, func, flags);
>>>> +}
>>>> +
>>>> +int pinctrl_request_noflags(struct udevice *dev, int func)
>>>> +{
>>>> + return pinctrl_request(dev, func, 0);
>>>> +}
>>>> +
>>>> +int pinctrl_get_periph_id(struct udevice *dev, struct udevice *periph)
>>>> +{
>>>> + struct pinctrl_ops *ops = pinctrl_get_ops(dev);
>>>> +
>>>> + if (!ops->get_periph_id)
>>>> + return -ENOSYS;
>>>> +
>>>> + return ops->get_periph_id(dev, periph);
>>>> +}
>>>
>>>
>>> If _get_periph_id() is shared between clk and pinctrl (that what I understood
>>> from your statement), where is it defined? Under arch/arm/mach-<soc>/ ?
>>
>> Yes.
>>
>>>
>>>
>>>
>>>
>>>> +static int pinctrl_post_bind(struct udevice *dev)
>>>> +{
>>>> + /* Scan the bus for devices */
>>>> + return dm_scan_fdt_node(dev, gd->fdt_blob, dev->of_offset, false);
>>>> +}
>>>
>>>
>>> I do not think this work.
>>>
>>> dm_scan_fdt_node() requires child nodes to be compatible to something
>>> for binding.
>>>
>>> Pin configuration nodes do not have '.compatible' properties.
>>
>> True in many cases, in which case it does not harm, but Rockchip has them:
>>
>> http://git.denx.de/?p=u-boot-dm.git;a=blob;f=arch/arm/dts/rk3288.dtsi;h=0f497099679470ea39078d6ca9e5b1ae550995b0;hb=refs/heads/rockchip-working
>
>
> My bad.
>
> It is possible a pinctrl device can have other devices under it (in
> this case, GPIO).
>
> So, it turned out my patch should be fixed.
> device_bind_driver_to_node() should be called for devices that have no
> .compatible.
>
OK.
>
>
>
>
>>>
>>>
>>>
>>>
>>>> UCLASS_DRIVER(pinctrl) = {
>>>> .id = UCLASS_PINCTRL,
>>>> .name = "pinctrl",
>>>> + .post_bind = pinctrl_post_bind,
>>>> };
>>>>
>>>> #else
>>>> diff --git a/include/dm/pinctrl.h b/include/dm/pinctrl.h
>>>> index b2ba0b4..2f9bf67 100644
>>>> --- a/include/dm/pinctrl.h
>>>> +++ b/include/dm/pinctrl.h
>>>> @@ -85,10 +85,37 @@ struct pinctrl_ops {
>>>> int (*pinconf_group_set)(struct udevice *dev, unsigned group_selector,
>>>> unsigned param, unsigned argument);
>>>> int (*set_state)(struct udevice *dev, struct udevice *config);
>>>> +
>>>> /* for pinctrl-simple */
>>>> int (*set_state_simple)(struct udevice *dev, struct udevice *periph);
>>>> + /**
>>>> + * request() - Request a particular pinctrl function
>>>> + *
>>>> + * This activates the selected function.
>>>> + *
>>>> + * @dev: Device to adjust (UCLASS_PINCTRL)
>>>> + * @func: Function number (driver-specific)
>>>> + * @return 0 if OK, -ve on error
>>>> + */
>>>> + int (*request)(struct udevice *dev, int func, int flags);
>>>> +
>>>> + /**
>>>> + * get_periph_id() - get the peripheral ID for a device
>>>> + *
>>>> + * This generally looks at the peripheral's device tree node to work
>>>> + * out the peripheral ID. The return value is normally interpreted as
>>>> + * enum periph_id. so long as this is defined by the platform (which it
>>>> + * should be).
>>>> + *
>>>> + * @dev: Pinctrl device to use for decoding
>>>> + * @periph: Device to check
>>>> + * @return peripheral ID of @periph, or -ENOENT on error
>>>> + */
>>>> + int (*get_periph_id)(struct udevice *dev, struct udevice *periph);
>>>> };
>>>>
>>>> +#define pinctrl_get_ops(dev) ((struct pinctrl_ops *)(dev)->driver->ops)
>>>> +
>>>> /**
>>>> * Generic pin configuration paramters
>>>> *
>>>> @@ -217,4 +244,38 @@ static inline int pinctrl_generic_set_state(struct udevice *pctldev,
>>>> }
>>>> #endif
>>>>
>>>> +/**
>>>> + * pinctrl_request() - Request a particular pinctrl function
>>>> + *
>>>> + * @dev: Device to check (UCLASS_PINCTRL)
>>>> + * @func: Function number (driver-specific)
>>>> + * @flags: Flags (driver-specific)
>>>> + * @return 0 if OK, -ve on error
>>>> + */
>>>> +int pinctrl_request(struct udevice *dev, int func, int flags);
>>>> +
>>>> +/**
>>>> + * pinctrl_request_noflags() - Request a particular pinctrl function
>>>> + *
>>>> + * This is similar to pinctrl_request() but uses 0 for @flags.
>>>> + *
>>>> + * @dev: Device to check (UCLASS_PINCTRL)
>>>> + * @func: Function number (driver-specific)
>>>> + * @return 0 if OK, -ve on error
>>>> + */
>>>> +int pinctrl_request_noflags(struct udevice *dev, int func);
>>>> +
>>>> +/**
>>>> + * pinctrl_get_periph_id() - get the peripheral ID for a device
>>>> + *
>>>> + * This generally looks at the peripheral's device tree node to work out the
>>>> + * peripheral ID. The return value is normally interpreted as enum periph_id.
>>>> + * so long as this is defined by the platform (which it should be).
>>>> + *
>>>> + * @dev: Pinctrl device to use for decoding
>>>> + * @periph: Device to check
>>>> + * @return peripheral ID of @periph, or -ENOENT on error
>>>> + */
>>>> +int pinctrl_get_periph_id(struct udevice *dev, struct udevice *periph);
>>>> +
>>>> #endif /* __PINCTRL_H */
>>>
>>>
>>>
>>> I am guessing these functions are only invoked from .set_state_simple(),
>>> If so, should they be exported to struct pinctrl_ops?
>>>
>>> I expect callbacks are invoked from the core framework (uclass).
>>>
>>
>> They can be involved from other places. In particular you need to call
>> get_periph_id() to get the peripheral ID for use with clocks.
>>
>
>
> In the first place, pinctrl_get_periph_id() should not be defined in
> a pinctrl driver.
>
> The first argument, pinctrl device, is not needed to calculate the
> peripheral ID,
> and the actual definition is placed under mach-<soc>, right?
So are you suggesting that each SoC provides a function (not in the
driver model framework) to get the peripheral ID in the simple pinctrl
case? This might be similar to clock_decode_periph_id(const void
*blob, int node) in tegra.
In fact if we implement a full clock controller binding (similar to
how you have implemented pinctrl) then we can avoid the need for
peripheral IDs once driver model is fully up. The need then exists
only in SPL, due to code size/memory constraints. It may perhaps also
be needed before relocation in U-Boot proper due to execution time
constraints but we can limit the number of devices brought up using
u-boot,dm-pre-reloc.
It would be good to limit the spread of the peripheral ID and only use
it when needed.
I'll take a look at creating a function outside of pinctrl which
returns a peripheral ID. That deals with the get_periph_id() method in
pinctrl.
However this still leaves the request() method - which takes a
peripheral ID. This is needed I think because:the device being set up
might not have a udevice in SPL. What do you think?
Regards,
Simon
More information about the U-Boot
mailing list