Ingo Weinhold <bonefish@xxxxxxxxxxxxxxx> wrote: > Not quite. If the boot platform (BIOS, Open Firmware, EFI) presents > the USB > disk as it would present any other disk (I believe the BIOS does just > that, > and OF is saner anyway), then our boot loader already supports > booting from > those disks. Exactly - it would just need to tell the kernel that it was loaded from USB, and make those modules available. > > I agree that having an initramfs isn't simpler, but I would say > > that it is a > > bit more flexible than a monolithic bootloader: initramfs for the standard boot is a bad solution, as it would force you to update it for every new supported hardware or hardware changes (just like Linux). > In case of networking you're right, though, since the boot loader > would > probably need to implement all the "high level" protocols we want to > support. That's true. > BTW, we already have a method similar to initramfs, which is used on > a boot > floppy (or CD). The kernel and required modules live in a tgz archive > that > the boot loader reads straight from a fixed position on the disk. I > guess it > wouldn't be all that hard to support retrieving that archive from > another > channel (e.g. TFTP). > > I wouldn't use the initramfs method by default (i.e. for booting from > standard disks), but for network boot, it might indeed be the best > choice. Not only for network boot - also for coping with buggy or incapable firmware. For example, I wanted to use the tarfs to boot the PPC version on the Pegasos later - or at least until I have found ways to work around that firmware in a sane way :-) I would think it should be possible to link the tarfs into the boot loader, such that the network and firmware loader would automatically load everything we need in a single go - and that would also allow us to remove the networking mini stack from the boot loader (even if optionally). Bye, Axel.