On 08/29/15 20:12, Bryan Kadzban wrote:
Tim Tassonis wrote:
On 29.08.2015 18:00, Bruce Dubbs wrote:
1. "Early system boot is the province of initrd." does not have to be
true. initrd is for commercial distros. We show in LFS that an initrd
is not needed and is quite a bit more complicated than booting without.
I would say that even a lot of distros could do away with the initrd
most of the time, if they would would provide two kernels:
- One fully modular, as is the case now
- One with the sata, ext4 and maybe md drivers compiled in.
I should note that that will only work if your rootfs is in a "normal"
partition (BIOS style partitions definitely work; I think UEFI might work now
too but am not 100% sure on that), with only an ext4 layer between the
partition table and the filesystem.
You mention md -- I'm not sure if that will autoassemble these days in the
kernel, or if it still requires userspace to run mdadm. Assuming it can
autoassemble, there's also LVM, which can't (it's self-describing but requires
that someone run "vgchange -a y"). There's also encrypted rootfs, which can't
(it requires someone run cryptsetup, either in LUKS mode asking for a
passphrase for one of the slots, or in raw mode, where you just have to give
it the raw key).
Yes, that's why I said 80 to 90 percent of cases, I'm aware there are
cases where an initrd is necessary. Now, I might be wrong about the
percentage of users that use lvm-based, or even encrypted root
partitions, I would have thought they are rather a minority.
GPT partitions definitely work as well, I'm writing this email on one
without initrd...
Not sure about uefi, though.
Kind regards
Tim
--
http://lists.linuxfromscratch.org/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page
