Re: [PATCH v7 4/4] arm64: dts: ti: hummingboard-t: add overlays for m.2 pci-e and usb-3

[Josua Mayer <josua@solid-run.com>]

Hi Geert,

Am 04.11.24 um 17:56 schrieb Geert Uytterhoeven:
> Hi Josua,
>
> On Wed, Oct 30, 2024 at 1:18 PM Josua Mayer <josua@solid-run.com> wrote:
>> Am 28.10.24 um 19:44 schrieb Vignesh Raghavendra:
>>> On 28/10/24 22:49, Josua Mayer wrote:
>>>> Am 28.10.24 um 16:31 schrieb Vignesh Raghavendra:
>>>>> On 25/10/24 19:27, Geert Uytterhoeven wrote:
>>>>>> On Mon, Feb 19, 2024 at 4:05 PM Josua Mayer <josua@solid-run.com> wrote:
>>>>>>> HummingBoard-T features two M.2 connectors labeled "M1" and "M2".
>>>>>>> The single SerDes lane of the SoC can be routed to either M1 pci-e
>>>>>>> signals, or M2 usb-3 signals by a gpio-controlled mux.
>>>>>>>
>>>>>>> Add overlays for each configuration.
>>>>>>>
>>>>>>> Signed-off-by: Josua Mayer <josua@solid-run.com>
>>>>>>> --- /dev/null
>>>>>>> +++ b/arch/arm64/boot/dts/ti/k3-am642-hummingboard-t-usb3.dtso
>>>>>>> +&usbss0 {
>>>>>>> +       /delete-property/ ti,usb2-only;
>>>>>> /delete-property/ (and /delete-node/) to delete something in the base DTS
>>>>>> does not work.
>>>> My understanding is that flags are equivalent to boolean, i.e:
>>>>
>>>> ti,usb2-only = <true>;
>>>> ti,usb2-only;
>>>>
>>>> are equivalent.
>>>>
>>>> If so, can we assign <false> within the overlay?
>> Is there any chance of reassigning <false> and making an argument
>> that this should be fixed?
> In theory, it can be done, if (1) all code that checks boolean
> properties would use of_property_read_bool() instead of
> of_property_present(), and (2) of_property_read_bool() would be changed
> to actually read the boolean value instead of just checking for the
> presence of the property.  And of course we have to do that in all
> software that uses DT (i.e. not just Linux).
> See [1][2][3] below for caveats.
Thank you for all those pointers!
>
> Using a similar solution for /delete-node/ is more complex, but still
> feasible, by setting its "status" property to "disabled" . I think
> that can be made to work if all DT core code that looks up nodes would
> just ignore any node that has a disabled status.
Agreed!
> I.e. callers would
> no longer see disabled nodes at all.
>
>> I find it frustrating that overlays can't override boolean properties,
>> and for consistency reasons I would otherwise change both
>> pcie and usb3 overlays to standalone dts.
> OK.

Setting a property = "false" would be quite preferential in my opinion,
over dropping the overlays.

Unfortunately I am preoccupied, and can't submit tested patches
within current merge window.

I will draft a minimal untested patch, and submit it to the list,
to get some comments.
For next window I could prepare a tested version.

>
> [1] The example in Documentation/devicetree/bindings/sound/rt5651.txt has:
>
>          realtek,dmic-en = "true";
>          realtek,in2-diff = "false";
>
>      Obviously the second line doesn't really work with the current
>      code, but fortunately there are no actual users of that (in
>      upstream DTS).
>      Note: "realtek,in2-diff" is a typo for "realtek,in2-differential".
>
> [2] The example in Documentation/devicetree/bindings/sound/pcm3060.txt has:
>
>          ti,out-single-ended = "true";
>
>      Again, this is an open invitation for replacing "true" by "false".
>      Fortunately there are no such users (in upstream DTS).
>
> [3] arch/arm/boot/dts/ti/omap/am335x-baltos.dtsi has:
>
>          gpmc,device-nand = "true";
>
>      But "gpmc,device-nand" does not exist. Oh, it is under removal:
>      https://lore.kernel.org/all/20241009-gpmc-omap-dtbx-v2-2-fc68124a090a@kernel.org/
>
>
>
> Gr{oetje,eeting}s,
>
>                          Geert
>
> --
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
>
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
>                                  -- Linus Torvalds