Re: [PATCH v1 01/11] arm64: dts: freescale: imx8mm-verdin: Add SPI alias

From: Frank Li

Date: Mon Nov 16 2026 - 11:05:34 EST


On Mon, Nov 16, 2026 at 12:45:09PM +0100, Francesco Dolcini wrote:
> Hello Frank,
>
>
> On Fri, Nov 13, 2026 at 03:35:02PM -0600, Frank Li wrote:
> > On Fri, Nov 13, 2026 at 06:19:27PM +0100, Francesco Dolcini wrote:
> > > On Fri, Nov 13, 2026 at 10:11:11AM -0600, Frank Li wrote:
> > > > On Fri, Nov 13, 2026 at 08:31:35AM +0100, Francesco Dolcini wrote:
> > > > > Hello Frank,
> > > > >
> > > > > On Thu, Nov 12, 2026 at 09:41:31AM -0600, Frank Li wrote:
> > > > > > On Thu, Nov 12, 2026 at 04:02:23PM +0100, Francesco Dolcini wrote:
> > > > > > > From: Francesco Dolcini <francesco.dolcini@xxxxxxxxxxx>
> > > > > > >
> > > > > > > The Verdin family pinout defines the SoM interfaces and their
> > > > > > > numbering, SPI_1 being the SPI bus available to the carrier board,
> > > > > > > implemented by ecspi2 on Verdin iMX8MM.
> > > > > > >
> > > > > > > imx8mm.dtsi numbers ecspi2 as spi1, so the Linux bus number in sysfs
> > > > > > > and the spidev name do not match the Verdin interface. Add a spi0
> > > > > > > alias in the SoM dtsi and delete the inherited spi1, so ecspi2 is
> > > > > > > bus 0, zero-based with SPI_1 being spi0, and carries a single alias.
> > > > > > > ecspi3, internal to the SoM with the CAN controller and the TPM, keeps
> > > > > > > the spi2 alias inherited from the SoC.
> > > > > > >
> > > > > > > Signed-off-by: Francesco Dolcini <francesco.dolcini@xxxxxxxxxxx>
> > > > > > > ---
> > > > > > > arch/arm64/boot/dts/freescale/imx8mm-verdin.dtsi | 2 ++
> > > > > > > 1 file changed, 2 insertions(+)
> > > > > > >
> > > > > > > diff --git a/arch/arm64/boot/dts/freescale/imx8mm-verdin.dtsi b/arch/arm64/boot/dts/freescale/imx8mm-verdin.dtsi
> > > > > > > index 1ac9d000d415..0369803a4d46 100644
> > > > > > > --- a/arch/arm64/boot/dts/freescale/imx8mm-verdin.dtsi
> > > > > > > +++ b/arch/arm64/boot/dts/freescale/imx8mm-verdin.dtsi
> > > > > > > @@ -16,6 +16,8 @@ chosen {
> > > > > > > aliases {
> > > > > > > rtc0 = &rtc_i2c;
> > > > > > > rtc1 = &snvs_rtc;
> > > > > > > + spi0 = &ecspi2;
> > > > > > > + /delete-property/ spi1;
> > > > > >
> > > > > > Can you avoid delete property for new platform?
> > > > >
> > > > > Unfortunately this is needed.
> > > > >
> > > > > The included imx8mm.dtsi has
> > > > >
> > > > > spi1 = &ecspi2;
> > > >
> > > > Create patch move aliase to each boards' dts from imx8mm.dtsi, which is
> > > > board feature, not soc features.
> > >
> > > Please see the linked thread [0][1]. It's not a viable option.
> >
> > where [0] and [1]
>
> Please see my previous email,
>
> https://lore.kernel.org/all/20261113073135.GA8515@francesco-nb/
>
> in which I tried to explain the issue.
>
>
> > > In addition, I can do this for my boards (Toradex), how do I know what I
> > > have to move for boards I do not know? I can't do that. The only way
> > > would be to duplicate the whole aliases on *every* board.
> >
> > at imx9, already move aliase to each boards, dupicate it and just make
> > sure build finial results the same.
> >
> > i.mx8 is still not too old yet.
>
> Do you want me to duplicate the whole aliases block on every board? I
> cannot do anything better then just duplicating everything, given I do
> not know what aliases make sense for each board. I can do better for my
> boards (Toradex), but for the others I can't.
>
> And this will likely break out-of-tree users (and to be clear, I think
> that out-of-tree users of Linux DT files is a fact, see U-Boot, barebox,
> ... and more), as I discussed in the previously linked discussion.

This is validate reason. We can try move one platform firstly later.

Frank

>
> To me this is not the correct thing to do, it will work for my boards,
> but I do not think this is the right decision for the whole i.MX
> ecosystem.
>
> And it's not about being old or not, it's about not introducing
> regressions and caring about the users.
>
> Francesco
>