[PATCH v8 14/16] ARM: dts: Introduce STM32F429 MCU

[daniel.thompson@linaro.org (Daniel Thompson)]

On 22/05/15 13:32, Maxime Coquelin wrote:
> 2015-05-22 12:07 GMT+02:00 Philipp Zabel <p.zabel@pengutronix.de>:
>> And that's how I'd expect it to be described by the device tree:
>>
>>      rcc: rcc at 40023800 {
>>          compatible = "st,stm32-rcc";
>>          reg = <0x40023800 0xc0>;
>>      };
>>
>
> Doing that, since this register bank contains both reset and clock
> registers, the reset device cannot get the IO resource at probe time
> because the clock driver has already reserved it.
> Daniel, who has started to work on the clock driver is facing this issue.
> This is why I proposed this binding for "reg" property.

Temporarily I've kept things working for me by avoiding 
of_io_request_and_map() in the clock driver (I am using raw of_iomap() 
instead).

I view this as a hack rather than a solution!

Note that, with or without the hack, I do hope to be able to post the 
clock driver this weekend. I am acutely aware that at the moment we are 
discussing code I haven't posted yet.


> We could think of creating a MFD driver, but the problem is that clock
> need to be intialized before a MFD device can be probed.
>
> Maybe there is a way to have you binding working properly, but I
> haven't found one for now.

I guess a driver doesn't *have* to reserve all the register space 
described in DT. The driver could replace of_io_request_and_map() with 
lower level calls.

Thus if we wanted to keep this detail out of DT then I guess the two 
drivers could simply make an agreement not to request registers they do 
not use.