Re: [PATCH v8 0/6] of: teach overlay code to keep /aliases in sync

From: Rob Herring

Date: Thu Nov 12 2026 - 17:29:26 EST


On Fri, Nov 06, 2026 at 11:51:52AM -0800, Abdurrahman Hussain wrote:
> On Fri, Nov 06, 2026 at 07:24:32AM -0600, Rob Herring wrote:
> > What I would like is aliases are not supported in overlays (applied
> > after boot).
> >
> > Don't we already have consistent naming on hotpluggable devices such as
> > network adapters? All that is solved in userspace. Why do we need DT to
> > solve this here?
>
> Netdevs can be renamed, and that is why userspace can solve it for them.

I don't think that was always true, so if it can be added for network
devices, it can be added for anything. But netdev is a bit special...

> i2c adapters can't be renamed.

I didn't say rename. /dev/input/ has unstable indices *and* stable
paths. IIRC, udev creates all the stable symlinks. Disks have links by
uuid and label.

> I'm also not sure "after boot" is a clear boundary.

I mean bootloader applied before Linux is entered.

> Overlays applied by
> drivers such as lan966x_pci, or for an FPGA on an x86 host, are applied
> during probe. They describe fixed, soldered-down hardware in the same
> way a base DT would. The same /aliases node gives stable numbering when
> it is in the base DT and is silently ignored when it comes from an
> overlay, even though it is written to the live tree and visible in
> /proc/device-tree.

How aliases interact with an ACPI system is another aspect I have yet to
consider...

> If an alias is a request, as you said, then honoring it when possible
> and falling back with a warning when it isn't can't break a working
> system. Today such a request is simply never honored. That was the
> regression you were worried about, and the fallback covers it.

Right, but there is no fallback for some cases. Having that is the only
way this could work.

> If you still don't want overlays to set aliases, is there another way
> you'd accept to get stable bus numbers for hardware described by an
> overlay? I'm happy to rework the series in a different direction.

Stable bus numbers is a solution to your problem. What's the problem you
are trying to solve?

Rob