On Thu, 13 Jun 2024 15:09:40 -0500 David Lechner <dlechner@xxxxxxxxxxxx> wrote: > On 6/13/24 2:43 PM, Rob Herring wrote: > > On Thu, Jun 13, 2024 at 05:11:48PM +0200, Nuno Sá wrote: > >> On Thu, 2024-06-13 at 09:39 -0500, David Lechner wrote: > >>> On 6/13/24 9:18 AM, Krzysztof Kozlowski wrote: > >>>> On 13/06/2024 15:57, David Lechner wrote: > >>>>> > >>>>>> > >>>>>>> + - const: adi,ad4695 > >>>>>>> + - items: > >>>>>>> + - const: adi,ad4697-wlcsp > >>>>>>> + - const: adi,ad4697 > >>>>>>> + # same chips with higher max sample rate > >>>>> > >>>>> I suppose one could make the argument that the programming model is > >>>>> the same on these too, but the maximum sampling frequency does seem > >>>>> like an important bit of information so that you don't try to set > >>>>> the conversion trigger rate too high. > >>>>> > >>>> > >>>> which property is that? I don't see differences in the driver, so I > >>>> don't get how these wlcsp compatibles allow you to control value of > >>>> conversion trigger. > >>> > >>> This comment is unrelated to the package type (WLCSP or LFCSP). > >>> > >>> What I mean is that e.g. AD4695 and AD4696 are virtually identical > >>> other than the maximum allowable sample rate (500 kSPS or 1 MSPS). > >>> > >>> So my thinking was that it would make sense to have: > >>> > >>> compatible = "ad4695"; > >>> > >>> for the lower sample rate chip and > >>> > >>> compatible = "ad4696", "ad4695"; > >>> > >>> for the higher sample rate chip since ad4696 can do everything > >>> that ad4695 does plus a bit more. > >>> > >> > >> IMO, that would make sense yes. If the higher sample rate chip fallsback, it will > >> still work but not at full speed. The other way around is the one that we can't allow > >> naturally. > >> > >> But possibly dumb question now... since both devices will be supported at the same > >> time, do we actually care about having the fallback compatible? My understanding of > >> the fallback story is that we may load a DTS in an older kernel where chip A is > >> supported but chip B is not and it is ok for chip B to fallback to chip A. Since > >> these devices will be supported at the same time, do we need to care? Unless out of > >> tree stuff enters the equation? > > > > Yeah, it doesn't really matter much in that case. > > > >> Or is there another usecase that I'm not aware about (or maybe it just makes sense to > >> document properly...)? > > > > Somewhat I guess. Perhaps if there's a 3rd chip with higher rate, then > > it will be more obvious what to do and we don't have to have this > > discussion again for it. :) > > > > Rob > > It sounds like maybe the best thing to do here then is just keep it simple? > > Like this: > > compatible: > enum: > - adi,ad4695 > - adi,ad4696 > - adi,ad4697 > - adi,ad4698 > Quick comments late in discussion. Simply compatible list is fine as they are all coming together (though that may in theory not be true on some other OS!) I'm also fine with the fallback as you discussed so less capable chip is the one we fall back to. Definitely need to differentiate chips with different possible ranges of a parameter as that goes straight through to userspace as sampling_frequency_available and userspace is going to get confused if it gets told it can do things that silently don't work. This sort of thing is why I always want to see chip IDs listed even if we think they'll just work with another one. a) Documents hardware right and people expect to see the ID of the chip on the BoM b) Lets us deal with cases where we have missed subtle differences in capabilities because the driver didn't support them yet. So in conclusion either option you proposed is fine (as above, or fallback to the part we think is a subset of functionality.) Jonathan
