[uClibc]Re: Strange busybox/uClibc problem

[Paul Komarek]

On Tue, 25 Jun 2002, Paul Komarek wrote:

>
>
> On Tue, 25 Jun 2002, Paul Komarek wrote:
>
> >
> > The message below was originally (and erroneously) posted by me to the
> > busybox mailing list.  At the end of the problem report, I mention that
> > things work fine when I replace uClibc with glib-2.1.3. Also, I'm now on
> > the uClibc and busybox mailing lists, so there is no need to forward to
> > me.
> >
> > On Sat, 22 Jun 2002, Paul Komarek wrote:
> >
> > >
> > > I am not (yet) on this mailing list -- the CMU mail admins claim my inbox
> > > is disturbing gravitational fields due to excessive mass.  Please cc me on
> > > any responses.
> > >
> > > I have a problem when booting a Soekris net4501 (AMD Elan SC520) using
> > > busybox (compiled against uClibc).  The problem occurs in init.c, but I
> > > hesitate to claim the problem is in init.c.
> > >
> > > The problem is that init stops at strange places:
> > >
> > >   VFS: Mounted root (nfs filesystem).
> > >   Freeing unused kernel memory: 220k freed
> > >   serial console detected.  Disabling virtual terminals.
> > >   init started:  BusyBox v0.60.3 (2002.06.22-05:44+0000) multi-call binary
> > >
> > > And that's it.  If I comment out init.c:311, then I get further (using
> > > message( CONSOLE, "Hello, world\n") to track my progress.
> > >
> > > init_main() calls check_memory() calls check_free_memory():
> > >
> > > static int check_free_memory(void)
> > > {
> > >     ...
> > >     u = info.mem_unit;
> > >     if (u==0) u=1;                   <------- Line 311, remove
> > >     ...
> > > }
> > >
> > > If I also comment out init.c:851, I make further progress.
> > >
> > > init_main() calls parse_inittab() calls new_init_action():
> > >
> > > static void new_init_action(int action, char *command, char *cons)
> > > {
> > >     ....
> > >     if (*cons == '\0')          <----------- Line 851, remove
> > >         cons = console;
> > >     ....
> > > }
> > >
> > > When repeating these tests to write this email, I didn't fiddle further
> > > with my inittab and such.  I did more tests earlier, but never got things
> > > working right (never got getty to work right, using my own and sample
> > > configs).
> > >
> > > I have this problem whether I do shared or static libs, and for kernels
> > > 2.4.17, 2.4.19-pre10, and 2.4.19-pre10-ac2 (you can see the desperation
> > > setting in... ;-).  The problem has occurred when netbooting or booting
> > > from a CF card.  I'm watching the output from a serial console, but I
> > > personally don't think that's part of the problem.  I have used a snapshot
> > > of uClibc from about June 19th, and another from June 22nd.  As you might
> > > guess, I haven't really taken the cross product of all these options.  If
> > > you want to know exactly what I've done, I'll be happy to enumerate the
> > > tests.  Of course, I'd be happy to provide most any information anyone
> > > things will help (kernel config, etc).
> > >
> > > I've made other net4501 boards work with a modified Red Hat 7.2
> > > distrobution, using a pared-down glibc and custom rh kernel 2.4.17-0.18,
> > > without any real troubles.
> > >
> > > Please let me know if it would be appropriate to post or cross-post to the
> > > uClibc list.
> > >
> > > -Paul Komarek
> >
> > I found a machine with gcc for i386 installed, and tested busybox with
> > static linking (and tinylogin with static linking) against glibc-2.1.3.
> > Everything works as expected (well, I haven't made getty work right for me
> > yet, but that' probably a PEBKAC situation =-).  I'm going to repost my
> > above message on the uClibc list now.
> >
> > -Paul Komarek
>
> I was suggest by one person that I clarify the cpu architecture involved:
>
> 1) the Soekris net4501 uses an AMD Elan SC520.  The Elan target in kernel
> the recent 2.4 kernels doesn't seem to work with it, and Configure.help
> states that the Elan target is for the SC400/SC410.
>
> 2) I have successfully used i386 libs with this board (using glibc instead
> of uClibc)
>
> 3) My uClibc toolchain-wrappers (the DEVEL directory) is name
> i386-linux-uclibc, which I'd hoped indicated that I was targeting i386.
> However, I didn't set CROSS to anything in Config, and my host machine
> uses gcc-i686.
>
> Since #3 looks like it could explain all my woes, I'm going to get the
> necessary cross-compiler built, and retry uClibc with that.
>
> -Paul Komarek

#3 was in fact the problem.  Once I added -mcpu=i386 for uClibc, and the
same for busybox and tinylogin, everything seems to be working properly.

Shouldn't -mcpu=i386 be part of Config.i386?  Rules.mak specifies
-march=i386 or the OPTIMIZE variable, but that's not enough to get an
i686-targeted gcc2 to compile properly for an i386 target.

-Paul Komarek