[syslinux] extlinux from syslinux 4.04 failed hard after choosing bzImage.

Gene Cumm gene.cumm at gmail.com
Sun Nov 20 05:08:08 PST 2011

On Sun, Nov 20, 2011 at 05:25, Piotr Karbowski
<piotr.karbowski at gmail.com> wrote:
> On 20.11.2011 01:57, Tyler Durden wrote:
>> Yep, that fixes it properly, just tested. Thanks Gene!
>> So, it seems this is a Debian bug after all. It's extlinux-update script
>> populates linux.cfg file with "kernel" directives, but it should use
>> "linux" instead.
> I doubt that you can consider it as a debian bug, following the syslinux.txt
> the LINUX is default, so I presume that KERNEL = LINUX, but if there is .0
> on the end then it acting like PXE /bzImage-3.1.0

No, KERNEL != LINUX.  KERNEL first guesses based on file extension and
the fall back if no matches found is to load it like a Linux kernel,
then similar to LINUX (except LINUX forces the behavior).

I feel the complication is fivefold.  The builder script of Debian's
linux.cfg should use LINUX.  linux.c32 should somehow be included also
to aid usage from the "boot:" prompt (probably with a LABEL linux to
call it easily).  The naming of the kernel should perhaps be examined
(perhaps bzImage-3.1.0-0).  Attempting to use extensions that are
applicable only to certain variants should probably return a error
(but there may be circumstances for loading an NBP from SYSLINUX if
it's on a MEMDISK).  Should there be a way to execute commands from
the boot prompt (rather than always loading files) that does not
interfere with (at least common) file naming?

> Anyway I am happy that I can use syslinux, as it does support gpt and
> booting from raid1 with metadata-1 (for some reason I can't assemble 0.9
> arrays).
> -- Piotr,

SYSLINUX only works for RAID-1 MDs where the MD data is at the end of
the disk (1.0 specifically).


More information about the Syslinux mailing list