Hi all,
I've come to the conclusion that in order to stay backwards
compatible while moving root-mounting stuff to userspace, in the
initial patch everything in prepare_namespace() and south needs to be
fully supported in userspace. This looks perfectly doable, but is a
fair bit of work.
The one piece of ugliness I've encountered has to do with the
load_ramdisk= option; this causes a ramdisk to be loaded from an
external device, usually a floppy. The ugliness has to do with the
fact that it requires the kernel itself to deduce the size of the
ramdisk, which is filesystem-specific. Although this code is
currently run for initrds as well, it doesn't need to, since the
kernel knows the size of an initrd.
This code isn't complex by any means, but it's ugly and complex, and
I'm trying to make something a bit cleaner than just copying the
existing in-kernel code to userspace.
So, in short:
a) Does anyone use the load_ramdisk= option anymore, or is it
legitimate to drop?
b) If it is necessary to retain, does anyone care if this option would
only support gzip format in the future, i.e. NOT support uncompressed
filesystem images? Since a gzip stream is self-terminating, this
takes care of the problem of finding the end, but adds a sizable chunk
of code to the kinit binary.
-hpa
On Tue, 26 Oct 2004 [email protected] wrote:
> Date: Tue, 26 Oct 2004 03:48:40 +0000 (UTC)
> From: [email protected]
> To: [email protected]
> Subject: Is anyone using the load_ramdisk= option in the kernel still?
>
> Hi all,
>
> I've come to the conclusion that in order to stay backwards
> compatible while moving root-mounting stuff to userspace, in the
> initial patch everything in prepare_namespace() and south needs to be
> fully supported in userspace. This looks perfectly doable, but is a
> fair bit of work.
>
> The one piece of ugliness I've encountered has to do with the
> load_ramdisk= option; this causes a ramdisk to be loaded from an
> external device, usually a floppy. The ugliness has to do with the
> fact that it requires the kernel itself to deduce the size of the
> ramdisk, which is filesystem-specific. Although this code is
> currently run for initrds as well, it doesn't need to, since the
> kernel knows the size of an initrd.
>
> This code isn't complex by any means, but it's ugly and complex, and
> I'm trying to make something a bit cleaner than just copying the
> existing in-kernel code to userspace.
>
> So, in short:
>
> a) Does anyone use the load_ramdisk= option anymore, or is it
> legitimate to drop?
>
I know Slackware Linux uses load_ramdisk= when you use floppies to start
the install in the situations where booting from CD for some reason
doesn't work. Slackware uses one boot disk and then 2 root disks in this
case.
--
Jesper Juhl
hpa,
On Tuesday 26 October 2004 05:48, H. Peter Anvin wrote:
> So, in short:
>
> a) Does anyone use the load_ramdisk= option anymore, or is it
> legitimate to drop?
I'm pretty sure it is used by the Debian installer when bootstrapping from a
floppy, but...
> b) If it is necessary to retain, does anyone care if this option would
> only support gzip format in the future, i.e. NOT support uncompressed
> filesystem images? Since a gzip stream is self-terminating, this
> takes care of the problem of finding the end, but adds a sizable chunk
> of code to the kinit binary.
... the image is read from a raw device (/dev/fd0) and it is gzipped.
Booting the installer is probably the only large scale use case for the
load_ramdisk parameter.
Cheers,
Alessandro
Alessandro Amici <[email protected]> wrote:
> On Tuesday 26 October 2004 05:48, H. Peter Anvin wrote:
>> a) Does anyone use the load_ramdisk= option anymore, or is it
>> legitimate to drop?
>
> I'm pretty sure it is used by the Debian installer when bootstrapping from a
> floppy, but...
I believe that this is no longer the case with the new installer.
--
Matthew Garrett | [email protected]
Matthew,
On Tuesday 26 October 2004 16:09, Matthew Garrett wrote:
> Alessandro Amici <[email protected]> wrote:
> > On Tuesday 26 October 2004 05:48, H. Peter Anvin wrote:
> >> a) Does anyone use the load_ramdisk= option anymore, or is it
> >> legitimate to drop?
> >
> > I'm pretty sure it is used by the Debian installer when bootstrapping
> > from a floppy, but...
>
> I believe that this is no longer the case with the new installer.
If you have the root filesystem on a floppy you still need load_ramdisk
(checked a few days ago ;)
cheers,
alessandro