I've been trying various kernel command line parameters that have been suggested in this thread. None of these made any noticeable difference: · acpi=off · systemd.restore_state=0 · init=/lib/sysvinit/init · init=/bin/sh I also found a parameter called initcall_debug that seemed like it might provide more data, but that didn't lead anywhere either. Even with init=/bin/sh I still see the string "starting version 232" – which is the version number of SystemD – and "uninitialized urandom read" messages that mention systemd-udevd and udevadm. Fortunately the presence or absence of "quiet" has an obvious effect. Otherwise I would have started to think that Grub ignored my changes to the command line. Unpacking and examining initrd.img-4.9.0-9-686 I find that /sbin/init is actually Busybox. So is /bin/sh, so that might explain why they both do the same thing. /lib/sysvinit/init doesn't exist in the initrd, but apparently Busybox gets started anyway. (Perhaps Linux falls back on the default if the requested init program doesn't exist?) It seems that I'll have to hack the initrd somehow if I am to get at a shell and poke around in the initrd environment. Booting from USB has been suggested several times. I'll do that if someone can tell me the magic incantation that makes Combios boot a USB device. I doubt it's possible. PXE booting is supposed to work. Maybe some day in a few weeks or months I'll find the time to set up a boot server. Björn Persson
Attachment:
pgpSuRSBA1wfo.pgp
Description: OpenPGP digital signatur