Re: [PATCH v2 2/2] i2c: add support for Socionext SynQuacer I2C controller

From: Andy Shevchenko
Date: Mon Feb 26 2018 - 12:06:06 EST


On Mon, Feb 26, 2018 at 4:58 PM, Ard Biesheuvel
<ard.biesheuvel@xxxxxxxxxx> wrote:
> On 26 February 2018 at 11:35, Andy Shevchenko <andy.shevchenko@xxxxxxxxx> wrote:
>> On Mon, Feb 26, 2018 at 11:59 AM, Ard Biesheuvel
>> <ard.biesheuvel@xxxxxxxxxx> wrote:
>>> On 23 February 2018 at 13:12, Andy Shevchenko <andy.shevchenko@xxxxxxxxx> wrote:
>>>> On Fri, Feb 23, 2018 at 2:40 PM, Ard Biesheuvel
>>>> <ard.biesheuvel@xxxxxxxxxx> wrote:

>>>> Replace 'baudclk' with 'pclk' and p->uartclk with i2c->clkrate in
>>>> above and you are almost done.

>>> I don't think this is better.
>>
>> It's a pattern over ACPI vs. clk cases at least for now.
>> But hold on. We have already an example of dealing with ACPI /
>> non-ACPI cases for I2C controllers â i2c-designware-platdrv.c.
>> Check how it's done there.
>>
>> I actually totally forgot about ACPI slaves described in the table. We
>> need to take into account the ones with lowest bus speed.
>>
>
> Wow, that code is absolutely terrible.

To some degree I may say yes it is.

> So even while _DSD device properties require vendor prefixes, which
> are lacking in this case,

What kind? clock-frequency? Does it require prefix?

> and the fact that the ACPI flavor of the
> Designware I2C controller now provides two different ways to get the
> timing parameters (using device properties or using SSCN/FMCN/etc ACPI
> methods), you think this is a shining example of how this should be
> implemented?

No, those methods because of windows driver and existed ACPI tables at
that time.
You are not supposed to uglify your case.

> Also, I still think implementing a clock device using rate X just to
> interrogate it for its rate (returning X) is absolutely pointless.

OTOH the deviation in the driver is what I absolutely against of.
Driver must not know the resource provider (ideally at all).

> So what I can do is invent an ACPI method that returns the PCLK rate.
> Would that work for you?

Again, looking into existing examples (UART, I2C, etc) we better to
create a generic helper in clock framework that would provide us a
clock based on property value.
But doing different paths for different resource providers is not what
we are looking for.

P.S. To move this somehow forward I may propose to submit an OF
driver, and discuss ACPI part after.

--
With Best Regards,
Andy Shevchenko