i.MX35 Kernel without ARM EABI support cannot execute init

[H Hartley Sweeten]

On Friday, August 21, 2009 7:37 AM, hoefle marco wrote:
> Hello everybody on the new mailing list,
> I have an issue with the ARM EABI kernel option.
> right now we have two ARM boards and two embedded linux distributions
> (Emdebian and OSELAS). The CPUs are PXA270 and Freescale i.MX35.
> The aim is to have emdebian on both boards.
> When trying to boot the i.MX35 Kernel (with ARM EABI or without, the
> kernel panic remains exactly the same) using the emdebian rootfs I get
> this:
> 
> Kernel panic - not syncing: Attempted to kill init!
> Backtrace: 
> [<c0026430>] (dump_backtrace+0x0/0x10c) from [<c02515dc>] (dump_stack
> +0x18/0x1c)
>  r7:c7817f54 r6:c7813c40 r5:c7813c40 r4:00000004
> [<c02515c4>] (dump_stack+0x0/0x1c) from [<c025161c>] (panic+0x3c/0x110)
> [<c02515e0>] (panic+0x0/0x110) from [<c0034a14>] (do_exit+0x6c/0x5b8)
>  r3:c0301bd4 r2:c7813c40 r1:c7817d14 r0:c02b478a
> [<c00349a8>] (do_exit+0x0/0x5b8) from [<c0034fe0>] (do_group_exit
> +0x80/0xb4)
> [<c0034f60>] (do_group_exit+0x0/0xb4) from [<c003ea60>]
> (get_signal_to_deliver+0x2c0/0x2f0)
>  r4:00000004
> [<c003e7a0>] (get_signal_to_deliver+0x0/0x2f0) from [<c0024ecc>]
> (do_signal+0x58/0x4e4)
> [<c0024e74>] (do_signal+0x0/0x4e4) from [<c0025384>] (do_notify_resume
> +0x2c/0x30)
> [<c0025358>] (do_notify_resume+0x0/0x30) from [<c0022dac>] (work_pending
> +0x1c/0x20)
> 
> The PXA270 kernel (same settings except ARM system type and ARM EABI
> disabled) boots the emdebian root file-system successfully.
> 
> The i.MX35 Kernel boots with ARM EABI support the OSELAS file-system.
> Both Kernels are version 2.6.30.4 and were compiled with the same
> arm-linux-gnu-gcc compiler.
> 
> >From my understanding it should be possible to use the same root
> file-system on both CPUs.
> 
> Has anybody an idea why this is not working?

Curious.  I'm having the same issue on a ep93xx system.

I switched from an OABI toolchain to an EABI one a while back.  Everything
was working fine but I realized that my rootfs was still built with the
older OABI toolchain.  So I nuked the rootfs and re-built everything.

Now the kernel boots fine but it hangs with the same Kernel panic you
are seeing.

In my case I think it's an issue with the external toolchain support
in buildroot.  The toolchain I am using (CodeSourcery 2009q1-203) is
a multilib toolchain and needs the -march=armv4t option to use the
correct instructions for the architecture.  It also needs this option
to correctly identify the directory location for the libraries.

I have been able to staticlly compile a test init and replaced the
one in my rootfs.  Doing that I don't get the Kernel panic so I think
I'm on the right track to finding a fix.

Are you also using buildroot?  Could your problem be similar?

I have cc'ed the buildroot list and Thomas Petrazzoni just in case
your issue is the same.

Regards,
Hartley