I've come across a difference in how initrd is handled in 2.2.x and
2.4.x. This related directly to TILO (sparc TFTP image with ramdisk).
Basically what we have is a kernel image with ramdisk and initrd
enabled, and a root disk image slapped on the end that is loaded via
initrd.
On 2.2.x, this works without problems; the ramdisk is loaded, and
/sbin/init is executed. However, with 2.4.x, it's quite different.
It loads the initial ramdisk, mounts it fine, tries to execute /linuxrc
(same as in 2.2.x, but it isn't there, so it continues), and then
complains with this:
VFS: Mounted root (ext2 filesystem).
VFS: Cannot open root device "" or 02:00
For some reason it is trying to mount /dev/fd, and totally forgets
about /dev/ram. If I pass root=/dev/ram to the command line, it works
fine, but I don't want to have to do this :)
I can't seem to find the relevant place where this broke. Any ideas?
--
.----------=======-=-======-=========-----------=====------------=-=-----.
/ Ben Collins -- Debian GNU/Linux \
` [email protected] -- [email protected] -- [email protected] '
`---=========------=======-------------=-=-----=-===-======-------=--=---'
Ben Collins wrote:
> Basically what we have is a kernel image with ramdisk and initrd
> enabled, and a root disk image slapped on the end that is loaded via
> initrd.
>
> On 2.2.x, this works without problems; the ramdisk is loaded, and
> /sbin/init is executed. However, with 2.4.x, it's quite different.
>
> It loads the initial ramdisk, mounts it fine, tries to execute /linuxrc
> (same as in 2.2.x, but it isn't there, so it continues), and then
> complains with this:
>
> VFS: Mounted root (ext2 filesystem).
> VFS: Cannot open root device "" or 02:00
>
> For some reason it is trying to mount /dev/fd, and totally forgets
> about /dev/ram. If I pass root=/dev/ram to the command line, it works
> fine, but I don't want to have to do this :)
hrm, if your root filesystem is indeed in RAM, then root=/dev/ram seems
appropriate on both 2.2.x and 2.4.x. That's what 2.2.x and 2.4.x
Documentation/initrd.txt seem to indicate to me, anyway.
Jeff
--
Jeff Garzik | Only so many songs can be sung
Building 1024 | with two lips, two lungs, and one tongue.
MandrakeSoft | - nomeansno
On Tue, Nov 13, 2001 at 02:49:31PM -0500, Jeff Garzik wrote:
> Ben Collins wrote:
> > Basically what we have is a kernel image with ramdisk and initrd
> > enabled, and a root disk image slapped on the end that is loaded via
> > initrd.
> >
> > On 2.2.x, this works without problems; the ramdisk is loaded, and
> > /sbin/init is executed. However, with 2.4.x, it's quite different.
> >
> > It loads the initial ramdisk, mounts it fine, tries to execute /linuxrc
> > (same as in 2.2.x, but it isn't there, so it continues), and then
> > complains with this:
> >
> > VFS: Mounted root (ext2 filesystem).
> > VFS: Cannot open root device "" or 02:00
> >
> > For some reason it is trying to mount /dev/fd, and totally forgets
> > about /dev/ram. If I pass root=/dev/ram to the command line, it works
> > fine, but I don't want to have to do this :)
>
> hrm, if your root filesystem is indeed in RAM, then root=/dev/ram seems
> appropriate on both 2.2.x and 2.4.x. That's what 2.2.x and 2.4.x
> Documentation/initrd.txt seem to indicate to me, anyway.
Well, the point being that 2.2.x worked implicitly, and 2.4.x doesn't. I
don't want to have to tell people who have been using tilo forever and a
day that they now have to add additional command line to get it to work
with 2.4.x.
--
.----------=======-=-======-=========-----------=====------------=-=-----.
/ Ben Collins -- Debian GNU/Linux \
` [email protected] -- [email protected] -- [email protected] '
`---=========------=======-------------=-=-----=-===-======-------=--=---'
Ben Collins <[email protected]> wrote:
> Well, the point being that 2.2.x worked implicitly, and 2.4.x doesn't. I
> don't want to have to tell people who have been using tilo forever and a
> day that they now have to add additional command line to get it to work
> with 2.4.x.
You don't have to. Just setup linuxrc to echo the right stuff into
/proc/sys/kernel/real-root-dev
--
Debian GNU/Linux 2.2 is out! ( http://www.debian.org/ )
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
On Wed, Nov 14, 2001 at 07:50:00AM +1100, Herbert Xu wrote:
> Ben Collins <[email protected]> wrote:
>
> > Well, the point being that 2.2.x worked implicitly, and 2.4.x doesn't. I
> > don't want to have to tell people who have been using tilo forever and a
> > day that they now have to add additional command line to get it to work
> > with 2.4.x.
>
> You don't have to. Just setup linuxrc to echo the right stuff into
> /proc/sys/kernel/real-root-dev
Yeah, which is listed under the "Obsolete" section in
Documentation/initrd.txt. The assumption I'm making here is that if
/linuxrc fails to execute, it falls back to /sbin/init on the currently
mounted root filesystem. Assumptions are bad, but I don't see why it
can't work like this. If there is a filesystem already mounted, it
should be used.
Ben
--
.----------=======-=-======-=========-----------=====------------=-=-----.
/ Ben Collins -- Debian GNU/Linux \
` [email protected] -- [email protected] -- [email protected] '
`---=========------=======-------------=-=-----=-===-======-------=--=---'
Ben Collins wrote:
>
> On Wed, Nov 14, 2001 at 07:50:00AM +1100, Herbert Xu wrote:
> > Ben Collins <[email protected]> wrote:
> >
> > > Well, the point being that 2.2.x worked implicitly, and 2.4.x doesn't. I
> > > don't want to have to tell people who have been using tilo forever and a
> > > day that they now have to add additional command line to get it to work
> > > with 2.4.x.
> >
> > You don't have to. Just setup linuxrc to echo the right stuff into
> > /proc/sys/kernel/real-root-dev
>
> Yeah, which is listed under the "Obsolete" section in
> Documentation/initrd.txt. The assumption I'm making here is that if
> /linuxrc fails to execute, it falls back to /sbin/init on the currently
> mounted root filesystem. Assumptions are bad, but I don't see why it
> can't work like this. If there is a filesystem already mounted, it
> should be used.
Really? I always thought the standard behavior of initrd was "umount
ramdisk unless" not "umount ramdisk if", implying you need to do
something special (like root=/dev/ram) to keep the initrd around.
--
Jeff Garzik | Only so many songs can be sung
Building 1024 | with two lips, two lungs, and one tongue.
MandrakeSoft | - nomeansno
On Tue, Nov 13, 2001 at 05:48:57PM -0500, Jeff Garzik wrote:
> Ben Collins wrote:
> >
> > On Wed, Nov 14, 2001 at 07:50:00AM +1100, Herbert Xu wrote:
> > > Ben Collins <[email protected]> wrote:
> > >
> > > > Well, the point being that 2.2.x worked implicitly, and 2.4.x doesn't. I
> > > > don't want to have to tell people who have been using tilo forever and a
> > > > day that they now have to add additional command line to get it to work
> > > > with 2.4.x.
> > >
> > > You don't have to. Just setup linuxrc to echo the right stuff into
> > > /proc/sys/kernel/real-root-dev
> >
> > Yeah, which is listed under the "Obsolete" section in
> > Documentation/initrd.txt. The assumption I'm making here is that if
> > /linuxrc fails to execute, it falls back to /sbin/init on the currently
> > mounted root filesystem. Assumptions are bad, but I don't see why it
> > can't work like this. If there is a filesystem already mounted, it
> > should be used.
>
> Really? I always thought the standard behavior of initrd was "umount
> ramdisk unless" not "umount ramdisk if", implying you need to do
> something special (like root=/dev/ram) to keep the initrd around.
That's if /linuxrc remains the initial process (which should in turn
exec /sbin/init, atleast according to my reading of the same document):
When finished with its duties, linuxrc typically changes the
root device and proceeds with starting the Linux system on the
"real" root device.
After it changes root (pivot_root):
Now, the initrd can be unmounted and the memory allocated by the
RAM disk can be freed:
# umount /initrd
# blockdev --flushbufs /dev/ram0 # /dev/rd/0 if using devfs
Then there's this note:
Note: if linuxrc or any program exec'ed from it terminates for
some reason, the old change_root mechanism is invoked (see
section "Obsolete root change mechanism").
The old mechanism is described as using /proc/sys/kernel/real-root-dev,
rdev or root= mechanisms.
Now, on 2.2, if any of those were empty (which is the case), it just
used the already mounted filesystem (the ramdisk). In 2.4 it assumes
that if initrd is enabled, and the root device string is empty, then
/dev/fd is the default device. Why this is done, I'm not sure.
Ben
--
.----------=======-=-======-=========-----------=====------------=-=-----.
/ Ben Collins -- Debian GNU/Linux \
` [email protected] -- [email protected] -- [email protected] '
`---=========------=======-------------=-=-----=-===-======-------=--=---'
Silly TILO. Seems 2.4.x is correctly doing things. TILO was setting HdrS
in the kernel to 0x0200 (floppy of course). So, 2.2.x was ignoring it
(the wrong behavior), and 2.4.x respects it. I changed TILO to use
0x0100, and now it boots just fine.
Sorry for the hubub.
--
.----------=======-=-======-=========-----------=====------------=-=-----.
/ Ben Collins -- Debian GNU/Linux \
` [email protected] -- [email protected] -- [email protected] '
`---=========------=======-------------=-=-----=-===-======-------=--=---'