Re: [PATCH v2 0/4] iio: adc: ad4080: add support for AD4880 dual-channel ADC

[David Lechner]

On 2/16/26 1:14 AM, Andy Shevchenko wrote:
> On Sun, Feb 15, 2026 at 05:16:47PM -0600, David Lechner wrote:
>> On 2/15/26 2:03 AM, Andy Shevchenko wrote:
>>> On Sat, Feb 14, 2026 at 12:31:12PM -0600, David Lechner wrote:
>>>> On 2/14/26 12:11 PM, Andy Shevchenko wrote:
>>>>> On Sat, Feb 14, 2026 at 04:08:52PM +0000, Jonathan Cameron wrote:
>>>>>> On Sun, 8 Feb 2026 14:50:23 +0200
>>>>>> Andy Shevchenko <andriy.shevchenko@xxxxxxxxx> wrote:
>>>>>>> On Fri, Feb 06, 2026 at 06:07:12PM +0200, Antoniu Miclaus wrote:
> 
> ...
> 
>>>>>>> I believe there is a better approach, what you need is rather a flag
>>>>>>> to SPI core to tell that this is the device with shared CS.
>>>>>>
>>>>>> Antoniu, this comment from Andy needs addressing before we move
>>>>>> on. It seems fairly fundamental and I'm not seeing a reply to it on list.
>>>>>>
>>>>>> I'm not entirely sure what Andy is suggesting will work but this
>>>>>> is perhaps a mismatch in really understanding what is going on here.
>>>>>> Andy, how would a flag work given they seem to be separately addressable
>>>>>> SPI buses. I think this isn't a shared SPI CS, but rather a device
>>>>>> with two entirely separate SPI buses. I think the only reason
>>>>>> we are bothering to implement it as a single device at all is the
>>>>>> shared backend.
>>>>>
>>>>> My understanding that there are two devices that for whatever reason share
>>>>
>>>> It is the opposite. It is a _single_ device with _two_ CS lines.
>>>
>>> Don't we have already support for that? This changes the picture even more towards
>>> NAKing this. See below why.
>>
>> Yes, spi_new_ancillary_device() was introduced exactly for this sort
>> of thing, which is why I think it makes sense to use it.
>>
>>>> adc@0 {
>>>> 	reg = <0>, <1>;
>>>> 	...
>>>> };
>>>>
>>>>> the same CS line. Yes, I probably misread the idea behind, but I meant
>>>>> some flag for SPI device that tells SPI core that the CS it wants is shared
>>>>> (maybe a high bit in the cs field or so), then CS core won't complain on
>>>>> validation about using the same cs number which is "already in use".
>>>>
>>>> There was one existing user in the kernel of spi_new_ancillary_device()
>>>> that looked like this, so it seemed the right way to approach it. However,
>>>> code was added later that caused the primary SPI device to "claim" both
>>>> CS lines for itself and probably broke the one existing user of
>>>> spi_new_ancillary_device() (hard to tell without hardware to test).
>>>>
>>>> The idea here was to unbreak that so we could use spi_new_ancillary_device()
>>>> just as in the existing use case.
>>>>
>>>> The patch for that could have been a bit more strict to only allow the
>>>> spi_new_ancillary_device() to take CS 1 and fail otherwise, but users
>>>> are going to notice if it isn't working right anyway, so I didn't ask
>>>> for more checking.
>>>
>>>>>> There is an argument that maybe we should be looking at how
>>>>>> to do data muxing backends to support the more general case of two
>>>>>> separate chips feeding into a single buffer, but that's a complex
>>>>>> beast and I'm not sure if it is something we actually need.
>>>>
>>>> I think it would actually be quite similar to what is done in this
>>>> series.
>>>
>>> TBH, the change sounds to me like a hack. It doesn't cover other potential ways
>>> of the multi-cs devices come into play. Given that SPI core supports multi-cs
>>> I don't see a good justification for this patch.
>>>
>>> What did I miss?
>>
>> As far as I can tell, other than the one existing user of
>> spi_new_ancillary_device(), other SPI multi-CS stuff is only used
>> by SPI flash memory devices, not general SPI devices. There code
>> that is being modified here was introduced to support the SPI
>> flash memory devices, so that use case is already covered by
>> existing code.
> 
> Right. And obvious question why can't we apply the same approach
> to any SPI device? Like extending existing code to cover generic
> cases.
> 

spi_new_ancillary_device() was already accepted in the kernel as the
solution for this sort of use case, so isn't it already the generic
approach?

I can see that it could possibly be nice if the SPI core saw that
there was more than one CS and called spi_new_ancillary_device()
automatically and somehow passed that along with the main SPI device
to the driver probe function. But since this is only the second user
of spi_new_ancillary_device(), I don't think we have enough data
points to be able to say if this is really what all peripheral drivers
would want.