Re: [PATCH v4 01/29] Revert "treewide: Fix probing of devices in DT overlays"

[Geert Uytterhoeven]

Hi Hervé,

Revisiting his old thread (sorry for the delay)...

On Thu, 11 Dec 2025 at 16:19, Herve Codina <herve.codina@xxxxxxxxxxx> wrote:
> On Thu, 11 Dec 2025 15:52:28 +0200
> Matti Vaittinen <mazziesaccount@xxxxxxxxx> wrote:
> > On 11/12/2025 14:20, Herve Codina wrote:
> > > On Thu, 11 Dec 2025 10:34:46 +0200
> > > Matti Vaittinen <mazziesaccount@xxxxxxxxx> wrote:
> > > Do you see the same trace with:
> > > - "pinctrl-0 = <&i2c1_pins>;" in your overlay
> > > - fragment0 removed from the overlay (i2c1_pins definition removed from
> > >    the overlay.
> > > - i2c1_pins node defined in your base DT.
> >
> > Just tested. The i2c1 appears and the test-overlay probe gets called,
> > when the i2c1_pins is in the base-dt and not in the overlay.
>
> Geert, do you expirement same results?

Yes, after moving the pin control subnode from the overlay[1] to the
base DTS, the SPI bus and FLASH work after loading the overlay for
the first time.

> > > In other word, is the issues related to adding a pinctrl sub-node (pinctrl
> > > pins definition) in the overlay or is it something else?
> >
> > Seems to be related to the pinctrl.
>
> I don't think that the issue is related to pinctrl itself.
>
> IMHO, I think the issue is related to overlays and fw_devlink.
> The distinction between "a new node is going to lead to a device" vs "a new
> node is just data and will never been attached to a new device" when an
> overlay is applied is broken.
>
> This is broken with the upstream "treewide: Fix probing of devices in DT
> overlays" commit I've tried to revert. Indeed, on the LAN966x PCI device
> use case devlinks created are not correct with this commit applied.
>
> I am not sure also that devlinks created with a more complex overlay will be
> correct. For instance, Matti, with your overlay not sure that a phandle from
> the oscillator node referencing the pmic node will lead to a correct
> provider/consumer devlink between the pmic device and the oscillator device.
>
> On the other hand, this is broken with "of: dynamic: Fix overlayed devices
> not probing because of fw_devlink" works for the LAN966x PCI device use case
> an lead to correct devlinks but breaks your use cases.

Loading my overlay[1] causes the following changes under
/sys/class/devlink/:

    + genpd_provider:ca53-cpu0--platform:e6e90000.spi ->
../../devices/virtual/devlink/genpd_provider:ca53-cpu0--platform:e6e90000.spi
    + platform:e6055000.gpio--platform:e6e90000.spi ->
../../devices/virtual/devlink/platform:e6055000.gpio--platform:e6e90000.spi
    + platform:e6060000.pinctrl--platform:e6e90000.spi ->
../../devices/virtual/devlink/platform:e6060000.pinctrl--platform:e6e90000.spi
    - platform:e6060000.pinctrl--platform:keys ->
../../devices/virtual/devlink/platform:e6060000.pinctrl--platform:keys
    + platform:e6150000.clock-controller--platform:e6e90000.spi ->
../../devices/virtual/devlink/platform:e6150000.clock-controller--platform:e6e90000.spi
    + platform:soc--platform:e6e90000.spi ->
../../devices/virtual/devlink/platform:soc--platform:e6e90000.spi

Note that these changes are exactly the same in the working and the
non-working case.

Removing the overlay again removes all added links, but does not
restore the keys link:

    - genpd_provider:ca53-cpu0--platform:e6e90000.spi ->
../../devices/virtual/devlink/genpd_provider:ca53-cpu0--platform:e6e90000.spi
    - platform:e6055000.gpio--platform:e6e90000.spi ->
../../devices/virtual/devlink/platform:e6055000.gpio--platform:e6e90000.spi
    - platform:e6060000.pinctrl--platform:e6e90000.spi ->
../../devices/virtual/devlink/platform:e6060000.pinctrl--platform:e6e90000.spi
    - platform:e6150000.clock-controller--platform:e6e90000.spi ->
../../devices/virtual/devlink/platform:e6150000.clock-controller--platform:e6e90000.spi
    - platform:soc--platform:e6e90000.spi ->
../../devices/virtual/devlink/platform:soc--platform:e6e90000.spi

Loading the overlay again causes no changes in the links, but the SPI
bus and FLASH always work.

[1] https://web.git.kernel.org/pub/scm/linux/kernel/git/geert/renesas-drivers.git/commit/?h=topic/renesas-overlays&id=383285b905a20d6734bfcbf7bcf715c1c2b45395

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

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