Hacker News Reader

I'm relatively sure it worked without lilo (but that's so long ago I might be wrong).

This stackoverflow question suggests support for this was explicitly dropped from the linux kernel: https://superuser.com/questions/415429/how-to-boot-linux-ker...

View on HN

Yep, you had to modify the kernel to set some options so that it would know what device held the initrd (and where on the disk), and then you could dd that to the floppy and dd the initrd after it. At boot time the BIOS would load the kernel from the floopy, Linux would open the fd device, adjust the device to seek to the beginning of the initrd, mount the initrd, then load and execute init.

But the later convention was to create a loopback device tied to a disk image, mkfs.minix it, mount it, copy the kernel and an initrd, and run lilo on the loopback device. Then unmount and decouple the loopback from the disk image and dd that to the floppy. It didn't really take up any extra space and made it easier to update your initrd or kernel.

The BIOS only loads a 512 byte sector, never a whole kernel. So if you want to directly boot a zImage, the first 512 bytes of the zImage would need to implement floppy sector reading loops to load the rest of the kernel block by block.

I think you're right, it could be I used uncompressed kernels for the first example (it's been a few years...)

Reply on news.ycombinator.com