(1) I compiled the new module-init-tools-3.1-pre5.tar.gz. It
claimed to be backward-compatible. After installing it, it
complained about something then seg-faulted. Nevertheless
`insmod` seemed to work so I proceeded.
(2) `make oldconfig` didn't work after copying over the
linux-2.4.26 .config file. This meant that I had to answer
hundreds of questions.
(3) `make bzImage` required that I install a new 'C' compiler.
This took several hours.
(4) Eventually "bzImage" got made. I tried `make modules`.
This took over 2 hours, went through everything several times.
This is a 2.8 GHz system. It usually takes about 6 minutes
to compile the kernel and all the modules. There is something
very wrong with the new compile method when it takes 120
times longer to compile than previously.
(5) `make modules_install` and `make install` seemed to work.
(6) The system would boot, but not find a file-system to mount.
(7) Tried to reboot using previous kernel. It failed to
load the required drivers for my SCSI disks so I have no
root file-system there.
(8) I am currently unable to use my main system. I will have
mount my main SCSI drive on this system, and replace the
module-init-tools with the previous modutils. This should
allow me to get "back" to my previous mounted root.
So much for "backwards compatibility".
The system that I use for development is not some "standard
distribution". Everything that runs on that system was built
on that system. That way, any embedded software developed
on that system is not likely to contain Trojan horses.
The latest module-init tools apparently assumes something
about the system, that it's some version of Red-Hat, perhaps?
Or some kernel functionality, necessary for a static version
of insmod to work, was removed!
In any event, it will not allow modules to be loaded from initrd.
The Red-Hat 'nash' that contains its own loader is not used.
Instead, the original ash.static plus a static version of
insmod is used.
As one of the `anonymous` developers who wrote the first module
code before taken over by Bas Laarhoven and Bjorn, I'm somewhat
pissed.
Richard B. Johnson
Project Engineer
Analogic Corporation
Penguin : Linux version 2.2.15 on an i586 machine (330.14 BogoMips).
On Fri, 24 Sep 2004 11:10:23 -0400 (EDT), Johnson, Richard
<[email protected]> wrote:
>
> (1) I compiled the new module-init-tools-3.1-pre5.tar.gz. It
> claimed to be backward-compatible. After installing it, it
> complained about something then seg-faulted. Nevertheless
> `insmod` seemed to work so I proceeded.
Error message would be nice to have.
> (2) `make oldconfig` didn't work after copying over the
> linux-2.4.26 .config file. This meant that I had to answer
> hundreds of questions.
This is unavoidable; 2.6 has lots and lots of new features.
> (3) `make bzImage` required that I install a new 'C' compiler.
> This took several hours.
gcc 2.95 should work for 2.6, although I have heard some mention on
the list of that not being true anymore.
> (4) Eventually "bzImage" got made. I tried `make modules`.
> This took over 2 hours, went through everything several times.
> This is a 2.8 GHz system. It usually takes about 6 minutes
> to compile the kernel and all the modules. There is something
> very wrong with the new compile method when it takes 120
> times longer to compile than previously.
Something is seriously wrong here, and it has nothing to do with the
"new compile method". Please post your 2.4 and 2.6 config files,
compiler version and platform info.
> (6) The system would boot, but not find a file-system to mount.
Details?
> (7) Tried to reboot using previous kernel. It failed to
> load the required drivers for my SCSI disks so I have no
> root file-system there.
>
> (8) I am currently unable to use my main system. I will have
> mount my main SCSI drive on this system, and replace the
> module-init-tools with the previous modutils. This should
> allow me to get "back" to my previous mounted root.
Related to (1), I assume.
Have you tried using a rescue disk to fix the problem? Knoppix (or a
distro rescue disk) should work fine.
Thanks,
Rahul
On Fri, 24 Sep 2004, Rahul Karnik wrote:
> On Fri, 24 Sep 2004 11:10:23 -0400 (EDT), Johnson, Richard
> <[email protected]> wrote:
> >
> > (1) I compiled the new module-init-tools-3.1-pre5.tar.gz. It
> > claimed to be backward-compatible. After installing it, it
> > complained about something then seg-faulted. Nevertheless
> > `insmod` seemed to work so I proceeded.
>
> Error message would be nice to have.
>
>From memory:
Kernel requires old lsmod but couldn't run ./lsmod.old No such
file or directory.
I don't know how it 'knew' that I had proviously copied
all the modutil utilities to *.old before installing the
new ones. In any event, it would certainly never find
them at "./filename.old". They would have to be un
/sbin or /usr/sbin or /usr/local/something, never in
the current directory.
> > (2) `make oldconfig` didn't work after copying over the
> > linux-2.4.26 .config file. This meant that I had to answer
> > hundreds of questions.
>
> This is unavoidable; 2.6 has lots and lots of new features.
>
Actually not. It spends its time doing CRCs and linking in
anti-compatibility stuff (I'm actually being kind, stuff is
really not the best word).
> > (3) `make bzImage` required that I install a new 'C' compiler.
> > This took several hours.
>
> gcc 2.95 should work for 2.6, although I have heard some mention on
> the list of that not being true anymore.
>
Well I had to use gcc 3.2 which generates a lot of bloat.
egcs-2.91.66 worked best for about two years. The code was
much more compact. Unfortunately, people started using:
#define whatever(a,b,...)
... which the compiler didn't like, probably because a definition
isn't a function that can take a variable parameter list.
> > (4) Eventually "bzImage" got made. I tried `make modules`.
> > This took over 2 hours, went through everything several times.
> > This is a 2.8 GHz system. It usually takes about 6 minutes
> > to compile the kernel and all the modules. There is something
> > very wrong with the new compile method when it takes 120
> > times longer to compile than previously.
>
> Something is seriously wrong here, and it has nothing to do with the
> "new compile method". Please post your 2.4 and 2.6 config files,
> compiler version and platform info.
>
I can't get to them yet.
> > (6) The system would boot, but not find a file-system to mount.
>
> Details?
>
Like I said. `insmod` failed. It failed because somebody
removed kernel function call query_module, number 167.
It returns ENOSYS and that's the end of trying to install
modules. Therefore my root file-system (on a SCSI disk)
isn't available.
That `insmod` was the un-touched insmod.static that was
in the untouched initrd boot image.
Nobody should have been allowed to remove that functionality.
You are supposed to add new functionality to existing software,
not destroy backward compatibility. Last time I checked
there were 252 function call numbers. Since it's a 32-bit
int, there are quite a few more available before one needs to
delete previous ones to make room.
> > (7) Tried to reboot using previous kernel. It failed to
> > load the required drivers for my SCSI disks so I have no
> > root file-system there.
> >
> > (8) I am currently unable to use my main system. I will have
> > mount my main SCSI drive on this system, and replace the
> > module-init-tools with the previous modutils. This should
> > allow me to get "back" to my previous mounted root.
>
> Related to (1), I assume.
>
> Have you tried using a rescue disk to fix the problem? Knoppix (or a
> distro rescue disk) should work fine.
>
It's a SCSI disk. All I have to do is mount it on this system
can copy all the saved modutils stuff back.
> Thanks,
> Rahul
>
Also, further experiments shows that even my last-resort IDE
drive can't be used with the new kernel as a root file-system
because, even though IDE stuff is compiled in, the actual
drive number, i.e., /dev/hda, etc., is different and can't
be found, or it uses devfs or something else equally annoying.
Yep.
Doesn't have to be better, only different.
Richard B. Johnson
Project Engineer
Analogic Corporation
Penguin : Linux version 2.2.15 on an i586 machine (330.14 BogoMips).
> > On Fri, 24 Sep 2004 11:10:23 -0400 (EDT), Johnson, Richard
> I don't know how it 'knew' that I had proviously copied
> all the modutil utilities to *.old before installing the
> new ones. In any event, it would certainly never find
> them at "./filename.old". They would have to be un
> /sbin or /usr/sbin or /usr/local/something, never in
> the current directory.
Aha, there's the problem. module-init-tools does a copy of the old
tools itself and delegates to the old tools for 2.4 kernels. So to fix
your problem, install modutils and the install module-init-tools over
it. See the README file that comes with module-init-tools for further
info.
Thanks,
Rahul