Re: news about CD booting
Paul Slootman writes:
> On Thu 04 Mar 1999, Loic Prylli wrote:
>
> > To implement such a mechanism, I have modified the ramdisk
> > driver, to be able to load a ramdisk not only from a floppy but from
> > any device with a iso9660 image, I store the offset of the root image
>
> Cool!
> I wonder if this could also be of use to other arch's...
Sure there is nothing Alpha specific in the patch.
> > in an application_data field of the iso volume descriptor (aka
> > super-block). This field is written with a modified isomarkboot.
> > Hope it makes sense (at least I have got the first test working).
>
> Maybe at some later stage mkisofs / mkhybrid could be adapted to
> support this directly? In a way similar to the El Torito boot file
> image...
Yes there is probably something to do with the eltorito spec, however:
- it requires to put more complex stuff in the kernel (the super-block
point to a boot catalog that needs to be interpreted to find the
right boot image), so to simplify testing for slink, I choosed to
minimize the implementation complexity.
- eltorito specifies a boot image for i386, Debian/i386 use it with
initrd to provide both _kernel_ and ramdisk image. In any case for
Alpha, the kernel(and eventually milo) cannot be booted in this way,
and what we want is only to specify the root filesystem (so after we
already are runing Linux). As this cannot be done in a standard way
with the Eltorito spec (AFAIK), if we cannot follow any standard,
it is better to at least keep things simple, and it avoids confusion
between what is a ramdisk image and what is a boot image. Maybe the right
thing to do is to modify Linux to be able to pass to it an optional
full path to a ramdisk image on the command-line, and let it load it,
that would work for any filesystem and would provide maximum
flexibility without needing to define any other convention.
> > Before I generate a new set of kernels with this patch, is the last
> > floppy patch posted by Paul the right one to include?
>
> It was accepted by Alan Cox for 2.0.37-pre5, so I think it's
acceptable :-)
Cool :-)
> Doesn't the VGA/TGA display slow down to the speed of the serial port if
> serial console support is enabled? That would not be nice; it would help
> if that can be disabled by some boot parameter (or maybe enabled instead?).
>
> I can build a kernel with serial support and see how that works on my
> XLT tonight.
I was thinking of the serial-console patch from Miquel van Smoorenburg
(standard in 2.2) which allows to choose the console on the
command-line (not the existing serial echoing), the serial port is not
used at all if not specified on cmd-line (but it will still be
initialized in the normal serial driver init routine).
Loic
Reply to: