Re: [PATCH RFC 1/2] spmi: Add support for multi-master

[AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>]

Il 07/02/24 12:45, Dmitry Baryshkov ha scritto:
> On Wed, 7 Feb 2024 at 11:08, Abel Vesa <abel.vesa@linaro.org> wrote:
>>
>> On 24-02-07 09:23:09, Dmitry Baryshkov wrote:
>>> On Wed, 7 Feb 2024 at 09:19, Abel Vesa <abel.vesa@linaro.org> wrote:
>>>>
>>>> On 24-02-07 01:55:39, Dmitry Baryshkov wrote:
>>>>> On Wed, 7 Feb 2024 at 01:34, Abel Vesa <abel.vesa@linaro.org> wrote:
>>>>>>
>>>>>> Some newer SPMI controllers support multiple bus masters.
>>>>>> Such a master can control multiple slave devices. The generic
>>>>>> framework needs to be able to pass on the master id to the
>>>>>> controller-specific driver. So do that. The framework will
>>>>>> check if the devicetree child nodes are actually bus masters
>>>>>> and will register the devices for each master. The legacy
>>>>>> approach will still be supported for backwards compatibility.
>>>>>
>>>>> Please remind me, are those two actual bus musters driving a single
>>>>> bus in parallel or two SPMI buses being handled by a single device? In
>>>>> the latter case this implementation is incorrect. There should be
>>>>> multiple spmi_controller instances, one for each bus. Allocate them in
>>>>> a loop and set ctrl->dev.of_node after allocating.
>>>>
>>>> It's two SPMI buses (two sets of wires) handled by the same controller,
>>>> HW-wise.
>>>>
>>>> If we register two spmi controllers with the kernel framework, it will
>>>> be HW inaccurate, because there is just one controller which has
>>>> multiple masters.
>>>
>>> struct spmi_controller is a controller for a single bus. Inside your
>>> device you have two SPMI buses, each can be controlled by its own
>>> struct spmi_controller. Just like devices that control multiple I2C,
>>> SPI or USB busses register a separate instance of the bus controller.
>>
>> Well, this is what this patchset is trying to do in the generic part.
>> The SPMI controller supports multiple buses (HW-wise) and therefore SW
>> implementation shouldn't be tied to single bus requirement.
> 
> So, after the off-line discussion:
> - add new compatible string for sm8450+
> - register two spmi controller instances

Well, I don't know about the actual hardware that you're trying to implement
but, in my opinion, the "idea" of this series does actually make sense.

The SPMI specification says that SPMI supports up to 4 masters, and up to
16 slaves.

Just my two cents.

Cheers,
Angelo

> - drop the master-id from the SPMI interface
> - optionally: think about having a new separate driver for v7 SPMI.
> 
>>
>>>
>>>>
>>>> I'm not saying it might not work. But, to me, it looks more like a hack.
>>>>
>>>> Basically, we would be mapping HW bus masters to kernel controllers.
>>>
>>> Buses, not just masters.
>>>
>>>>
>>>>>
>>>>>>
>>>>>> Signed-off-by: Abel Vesa <abel.vesa@linaro.org>
>>>>>> ---
>>>>>>   drivers/spmi/spmi-mtk-pmif.c |  6 ++--
>>>>>>   drivers/spmi/spmi-pmic-arb.c | 10 +++---
>>>>>>   drivers/spmi/spmi.c          | 76 ++++++++++++++++++++++++++++++--------------
>>>>>>   include/linux/spmi.h         | 10 +++---
>>>>>>   4 files changed, 67 insertions(+), 35 deletions(-)
>>>>>
>>>>> --
>>>>> With best wishes
>>>>> Dmitry
>>>
>>>
>>>
>>> --
>>> With best wishes
>>> Dmitry
> 
> 
>