Re: [PATCH 0/8] btrfs: assemble a multi-device root filesystem without an initrd
From: Qu Wenruo
Date: Wed Nov 18 2026 - 15:33:52 EST
在 2026/11/18 22:37, Ismael Luceno 写道:
On Wed, 18 Nov 2026 06:30:31 +1030
Qu Wenruo <quwenruo.btrfs@xxxxxxx> wrote:
在 2026/11/18 04:06, Ismael Luceno 写道:
On Tue, 17 Nov 2026 10:11:46 +1030
Qu Wenruo <wqu@xxxxxxxx> wrote:
在 2026/11/17 09:13, Ismael Luceno 写道:
A multi-device btrfs root filesystem is normally assembled by the
device scanner that an initrd runs before mounting the root.
What is the problem here?
Is generating an initramfs really that complex?
Why depend on an initramfs?
Why not? One should not put everything into kernel.
E.g. what if the customer has some custom udev rules for devices
names? Now they can not use the custom devices names during boot,
which they can before.
Shouldn't this count as a regression?
I don't think that's a possibility since the initramfs takes precedence,
everything would follow the same sequence as before the patch.
If anything it makes the boot process more consistent and enables
migrating initramfs-less systems to a multi-device setup without
changes to the boot process, which was my main goal (unfortunately all
the generators produce bloated initramfs', and I didn't want to
introduce yet another implementation just for this, plus it would be
slower anyway).
I don't think it's even a good idea to put every workload in the kernel.
And personally speaking, I don't think the busy-box based mkinitcpio is that bloated nor slow, even with all LVM/cryptosetup included.
This is not a small change, but the benefit looks very limited, thus a strong argument is needed to push for this.