by Enric Balletbo i Serra

[permalink] [raw]

Subject: Re: [PATCH v11 3/5] soc: mediatek: Move mt8173 MMSYS to platform driver

Hi all,

On 10/3/20 10:40, Matthias Brugger wrote:
>
>
> On 09/03/2020 23:50, Stephen Boyd wrote:
>> Quoting Enric Balletbo Serra (2020-03-06 14:09:50)
>>> Missatge de Stephen Boyd <[email protected]> del dia dv., 6 de mar
>>> 2020 a les 22:37:
>>>>
>>>> Quoting Enric Balletbo i Serra (2020-03-06 08:30:16)
>>>>> On 5/3/20 22:01, Stephen Boyd wrote:
>>>>>> Quoting Enric Balletbo i Serra (2020-03-02 03:01:26)
>>>>>>> diff --git a/drivers/soc/mediatek/mtk-mmsys.c b/drivers/soc/mediatek/mtk-mmsys.c
>>>>>>> new file mode 100644
>>>>>>> index 000000000000..473cdf732fb5
>>>>>>> --- /dev/null
>>>>>>> +++ b/drivers/soc/mediatek/mtk-mmsys.c
>>>>>>> @@ -0,0 +1,154 @@
>>>>>>> +// SPDX-License-Identifier: GPL-2.0-only
>>>>>>> +/*
>>>>>>> + * Copyright (c) 2014 MediaTek Inc.
>>>>>>> + * Author: James Liao <[email protected]>
>>>>>>> + */
>>>>>>> +
>>>>>>> +#include <linux/clk-provider.h>
>>>>>>> +#include <linux/of_device.h>
>>>>>>> +#include <linux/platform_device.h>
>>>>>>> +
>>>>>>> +#include "../../clk/mediatek/clk-gate.h"
>>>>>>> +#include "../../clk/mediatek/clk-mtk.h"
>>>>>>
>>>>>> Why not use include/linux/clk/?
>>>>>>
>>>>>
>>>>> I can move these files to include, this will impact a lot more of drivers but,
>>>>> yes, I think is the right way.
>>>>>
>>>>>> But I also don't understand why the clk driver is moved outside of
>>>>>> drivers/clk/ into drivers/soc/. Commit text saying that it has shared
>>>>>> registers doesn't mean it can't still keep the clk driver part in the
>>>>>> drivers/clk/ area.
>>>>>>
>>>>>
>>>>> Actually moving this to the soc directory has been requested by CK (mediatek) as
>>>>> a change in v8. You can see the discussion in [1]
>>>>>
>>>>
>>>> I can reply there in that thread if necessary, but we shouldn't need to
>>>> force simple-mfd into DT bindings to support this. Match the compatible
>>>> string in drivers/soc/ and register devices in software for the
>>>> different pieces of this overall hardware block. If necessary, pass down
>>>> the ioremapped addresss down through device data to each logical driver
>>>> in the respective subsystem.
>>>>
>>>> So yes, it looks like an MFD, but that doesn't mean we have to change
>>>> the DT binding or put it in drivers/mfd to support that. And we don't
>>>> have to fix any problems with allowing two drivers to probe the same
>>>> compatible string.
>>>>
>>>
>>> That thread maybe has too much information and things evolved since
>>> then. Note that the final solution is not an MFD neither we change the
>>> bindings. I pointed to that thread just because CK (CK please correct
>>> me if I'm wrong) thought that the driver is not a pure clock driver
>>> and he preferred to move to drivers/soc/mediatek (in that thread, he
>>> exposes his opinion on that). Sorry to introduce more confusion.
>>>
>>> You seem to be fine with the approach (just minor changes), so it
>>> looks to me that the only problem is if this should be in drivers/clk
>>> or drivers/soc. Honestly, this is not something I can't decide and
>>> I'll let you (the soc and clk maintainers) decide. I don't really have
>>> a strong opinion here. I don't mind move again to drivers/clk if that
>>> is what we want but let's come to an agreement.
>>>
>>
>> It's already in drivers/clk, so leave the clk part there and register
>> the clk device and any other devices by matching the compatible in
>> drivers/soc. That is my preferred solution. Can that be done?
>>
>
> I think we can once again create a platform device in drivers/soc which matches
> the drivers/clk and then do the routing in drivers/soc. Enric any thoughts?
>

Yes, working already on a new version.

Thanks,
Enric