Re: [PATCH v3 0/7] init: discoverable root partitions

From: Vincent Mailhol

Date: Wed Nov 11 2026 - 10:53:28 EST


On 10/11/2026 at 17:03, Christian Brauner wrote:
> On Mon, Nov 02, 2026 at 10:29:07PM +0100, Vincent Mailhol wrote:
>> This series adds root discovery directly into the kernel so that people
>> who are not using an initrd can still benefit from it.
>
> Nothing has changed since
> https://lore.kernel.org/r/20260617-irritation-rollen-wirst-7d636cbfec92@brauner

I replied to your earlier message to explain my position but did not
hear back.

Since v1, I have rewritten the cover letter to better explain why I
think implementing a subset of DPS in the kernel makes a
difference. The relevant paragraph is:

This is useful for booting a kernel manually, for example from the edk2
file browser, or recovering from a lost or corrupted bootloader
configuration without having to supply the correct root= argument or
load an initrd from the bootloader shell. Finally, it allows systems
that otherwise need no initrd to discover their root partition through
DPS without introducing an early userspace environment solely for that
lookup.

Similarly, I clarified the scope to address your concern about the
complexity of DPS:

The scope is deliberately limited to block devices that the kernel can
already mount through the root= parameter. Other DPS features, such as
setting up LUKS-encrypted partitions, mounting partitions other than
root or container setup, remain userspace responsibilities, with no
future plan to implement them in the kernel. The intended usage is
thus a hybrid approach in which the kernel DPS only mounts the root
partition and leaves the remaining setup to userspace (either with DPS
or with a classic fstab), just as it does today when root= specifies
the root device.

I would appreciate hearing whether these use cases and the limited
scope address your concerns.


Yours sincerely,
Vincent Mailhol