Re: Bug#636123: linux-image-2.6.39 not booting due to older package (not in list of dependencies!)
On Tue, Aug 02, 2011 at 02:05:05PM +0100, Luke Kenneth Casson Leighton wrote:
> there's a rather complex set of interdependencies between
> linux-image-2.6.39, libdevmapper, initramfs-tools and LVM that has
> resulted in a system which last had 2.6.32-trunk-amd64 installed (and
> associated packages) failing to boot when the latest 2.6.39 kernel was
> installed. the root filesystem is on an LVM partition.
> the error that occurred will be familiar to anyone who has done kernel
> development especially embedded linux development: unable to find root
> filesystem, after only about 5-10 lines of kernel messages. it's
> incredibly early on in the boot process, and is probably in the
> initial ramdisk stuff, when looking for (duh) the root filesystem.
> the *only* thing that was installed, which resulted in the boot
> failure, was the linux 2.6.39 kernel. this, obviously, triggered an
> initrd build (but not of the 2.6.32 one, i don't recall seeing that:
> could be wrong though).
> now, i've discussed this on the bugtracker and there clearly isn't -
> and really shouldn't be - a listed debian dependency between
> linux-image-2.6.39 kernel and a userspace library. however, there
> clearly *is* a dependency because "It Don't Wurk (tm)".
> so the issue is: how the bloody hell should this clear dependency be
> expressed in "Debian Dependency" terms, such that nobody else runs
> smack into this same issue?
> ben's suggested that there may be something in initramfs-tools that's
> up the creek. however, i'm having difficulty finding a link (depends,
> rdepends) between libdevmapper and initramfs-tools (which may in fact
> be the issue).
> are there any other ideas on how this may be resolved?
the kernel is racy and before you were just lucky, it scsi is involved use
if not use bootparam
known stuff, no need to cry, documented