On Tue, Apr 09, 2024 at 08:33:22AM +0200, Krzysztof Kozlowski wrote:
> On 08/04/2024 17:36, Frank Li wrote:
> > On Fri, Apr 05, 2024 at 08:21:18PM +0200, Krzysztof Kozlowski wrote:
> >> On 05/04/2024 16:46, Frank Li wrote:
> >>> On Fri, Apr 05, 2024 at 08:41:59AM +0200, Krzysztof Kozlowski wrote:
> >>>> On 04/04/2024 18:19, Frank Li wrote:
> >>>>> imx8qxp-mek use two kind audio codec, wm8960 and wm8962. Using dummy gpio
> >>>>> i2c bus mux to connect both i2c devices. One will probe failure and other
> >>>>> will probe success when devices driver check whoami. So one dtb can cover
> >>>>> both board configuration.
> >>>>
> >>>> I don't understand it. Either you add real device or not. If one board
> >>>> has two devices, then why do you need to check for failures?
> >>>>
> >>>> Anyway, don't add fake stuff to DTS.
> >>>
> >>> NAK can't resolve the problem. It should be common problem for long time
> >>> cycle boards. Some chipes will be out life cycle. such as some sensor. So
> >>> chips on boards have been replace by some pin to pin compatible sensor. For
> >>> example:
> >>> old boards: use sensor A with address 0x1a
> >>> new bench: use sensor B with address 0x1b.
> >>>
> >>> You can treat it as two kind boards, RevA or RevB. But most user want to
> >>> use one dtb to handle such small differences. For this case, it should be
> >>> simple. Just add a super set.
> >>> i2c
> >>> {
> >>> sensorA@1a
> >>> {
> >>> }
> >>> sensorB@1b
> >>> {
> >>> }
> >>> }
> >>>
> >>> It also depend on whoami check by i2c devices. Only A or B will probe.
> >>>
> >>> wm8960 and wm8962 are more complex example. wm8960 is out of life. But
> >>> wm8962 and wm8960 have the same i2c address. The current i2c frame can't
> >>> allow the same i2c address in one i2c bus.
> >>>
> >>> You are feel to NAK my method, but I hope you also provide constructive
> >>> solution to help resolve the problem.
> >>
> >> Yes, we resolved it long time ago. Your bootloader can (usually easily)
> >> detect revision of the board and load appropriate DTS or DTS+DTSO.
> >
> > I knewn it. But the problem is one development boards A have many options,
> > so create many child dts for files, A1, A2, ... An which base on A
>
> So use DTSO, what's the problem? Other vendors, liek Rpi does not have
> problem with it and it works well. No confusion.
>
> >
> > If there are difference happen at A, create new B. then create all child
> > dtb, B1, B2, ... Bn. DTB number will increase exponent.
> >
> > If change is quite bit, we have to do that. But if change is quite small,
> > One dtb can cover it by driver auto detect, which will work like some
> > adaptor card have not plug into boards, or some sensor or NOR-flash have
> > not installed because reduce cost.
>
> You have two boards, not 20 here!
Actually, it is around ~20 derived boards. It is not upstream just because
we have not time to do that yet. After some clean up, I estimate about 7 -
- 10.
>
> >
> > Although boot loader can update dts or choose difference dts, It also cause
> > many confusition, such as layerscape, uboot update many kernel dtb's
> > information, which actually increase dependence between uboot and kernel.
> > Also it confuse people, for example, when try to debug kernel dtb, why
> > change have not token affect when change dts because not realized uboot
> > over write it.
> >
> > What's I dide is that trying to reduce unnecessary dts.
>
> There is no confusion. That's normal process, so if someone is confused
> by having variants of board, this someone will be even more confused by
> seeing non-existing hardware in his DTS.
How about existed dummy_clk and dummy regulator in dts?
>
> This problem was solved long time ago and you do not bring any
> reasonable new arguments. All vendors solved it (just look at ongoing
> discussions on board id) but only you have problem with their solution.
I never said solution was not work. Just not friend for user enough for
this case. It likes USB TypeC vs USB A port. Both works, just TypeC can't
care direction.
>
> NAK
>
> Best regards,
> Krzysztof
>
