Bug#986741: Please enable CONFIG_IP_PNP_DHCP=y in cloud image
On Mon, Apr 12, 2021 at 11:18:43PM +0200, Ben Hutchings wrote:
> On Sun, 2021-04-11 at 18:44 -0700, Josh Triplett wrote:
> > On Sun, Apr 11, 2021 at 10:45:27PM +0200, Ben Hutchings wrote:
> > That wouldn't actually solve the issue as reported. I was hoping
> > specifically for the in-kernel support.
> >
> > I'd like to use this on a cloud instance that uses ext4 on NVMe, and
> > thus doesn't currently need an initramfs.
>
> I think I missed or forgot about that change when it happened. I'm not
> super happy about it, to be honest. It's been half-assed in that the
> cloud flavours still have a dependency on an initramfs builder, that
> should be a Recommends if we really want to support non-initramfs
> booting. There's no documentation that explains when that should work.
I didn't realize this would be an issue, and I *definitely* didn't
intend for this change to be a surprise. In this context, I was much
more concerned about boot speed than disk usage, and I was willing to
take what I could get.
I would be willing (after the release) to do more work to make this a
supported configuration, if you'd be willing to consider that. I'd love
to see initramfs-tools in Recommends, and I'd be happy to write
documentation for the Description that explains when it's required.
Something like this:
"""
This kernel package Recommends initramfs-tools. On cloud systems meeting
all of the following requirements, you can skip installing
initramfs-tools and boot without an initramfs:
- The root filesystem lives on an NVMe drive (no RAID, LVM, dm-crypt, or
similar).
- The root filesystem uses ext4.
- The root filesystem is identified by PARTUUID or by device name.
- Required system components such as /usr are part of the root
filesystem.
"""
Does that sound potentially reasonable?
> Have you tried using tiny-initramfs? Does that still significantly
> slow down boot relative to not using an initramfs image?
This is the first time I've heard of tiny-initramfs.
I'm currently counting the milliseconds, and loading any initramfs at
all does substantially add to boot time, compared to not requiring an
initramfs; the kernel has to extract it into memory and mount it, and
the bootloader has to load it from disk.
- Josh Triplett
Reply to: