[Buildroot] PXA270 and gcc-4.3.3 for iwMMXt

[Buildroot] PXA270 and gcc-4.3.3 for iwMMXt

[Laine Walker-Avina]

Hi,

I've been trying to build a busybox system with buildroot for a
PXA270-based system. I tried to use the latest gcc version in svn to
generate the kernel and rootfs, however when I try to run the system I
can't login and it doesn't run any of the init scripts in init.d. I
tried to chroot to the new rootfs from a previously built rootfs
compiled for a armv4 arch and it gave me a "Illegal Instruction" error
when trying to run ash. Most other commands from that busybox run
fine. And ash runs fine if i don't do the chroot and use the shared
libs from the other rootfs. Any thoughts?

-Laine Walker-Avina

[Buildroot] PXA270 and gcc-4.3.3 for iwMMXt

[Daniel Mack]

On Wed, Apr 22, 2009 at 09:08:28AM -0700, Laine Walker-Avina wrote:
> I've been trying to build a busybox system with buildroot for a
> PXA270-based system. I tried to use the latest gcc version in svn to
> generate the kernel and rootfs, however when I try to run the system I
> can't login and it doesn't run any of the init scripts in init.d. I
> tried to chroot to the new rootfs from a previously built rootfs
> compiled for a armv4 arch and it gave me a "Illegal Instruction" error
> when trying to run ash. Most other commands from that busybox run
> fine. And ash runs fine if i don't do the chroot and use the shared
> libs from the other rootfs. Any thoughts?

Check that your kernel and toolchain binaries are both built for EABI
and check for the following entries in your busybox .config:

BR2_iwmmxt=y
BR2_ARM_TYPE="ARM_IWMMXT"
# BR2_ARM_OABI is not set
BR2_ARM_EABI=y
BR2_ARCH="arm"
BR2_ENDIAN="LITTLE"
BR2_GCC_TARGET_TUNE="iwmmxt"
BR2_GCC_TARGET_ARCH="iwmmxt"
BR2_GCC_TARGET_ABI="aapcs-linux"

What you describe to see is very likely that kind of problem, we had
that many times already. The above settings should do well, though.

Daniel

[Buildroot] PXA270 and gcc-4.3.3 for iwMMXt

[Laine Walker-Avina]

On Wed, Apr 22, 2009 at 9:14 AM, Daniel Mack <daniel@caiaq.de> wrote:
> Check that your kernel and toolchain binaries are both built for EABI
> and check for the following entries in your busybox .config:
>
> BR2_iwmmxt=y
> BR2_ARM_TYPE="ARM_IWMMXT"
> # BR2_ARM_OABI is not set
> BR2_ARM_EABI=y
> BR2_ARCH="arm"
> BR2_ENDIAN="LITTLE"
> BR2_GCC_TARGET_TUNE="iwmmxt"
> BR2_GCC_TARGET_ARCH="iwmmxt"
> BR2_GCC_TARGET_ABI="aapcs-linux"
>
> What you describe to see is very likely that kind of problem, we had
> that many times already. The above settings should do well, though.
>
> Daniel

All of the above is the same as my config. I'm attaching the .config
for the kernel and buildroot; please let me know if there's anything
you see wrong.

-Laine
-------------- next part --------------
A non-text attachment was scrubbed...
Name: .config
Type: application/octet-stream
Size: 26928 bytes
Desc: not available
URL: <http://lists.busybox.net/pipermail/buildroot/attachments/20090422/ef6690e5/attachment-0001.obj>

[Buildroot] PXA270 and gcc-4.3.3 for iwMMXt

[Laine Walker-Avina]

> Is your kernel EABI enabled?
>
> Daniel
>

This is the kernel .config.

-Laine
-------------- next part --------------
A non-text attachment was scrubbed...
Name: .config
Type: application/octet-stream
Size: 30029 bytes
Desc: not available
URL: <http://lists.busybox.net/pipermail/buildroot/attachments/20090422/0db3ac82/attachment-0001.obj>

[Buildroot] PXA270 and gcc-4.3.3 for iwMMXt

[Daniel Mack]

On Wed, Apr 22, 2009 at 09:32:21AM -0700, Laine Walker-Avina wrote:
> > Daniel
> >
> 
> This is the kernel .config.

Hmm. Another common pitfall is a floating point mismatch between your
kernel and your toolchain, but that one seems to be set correctly in
your case as well. I'm afraid I can't provide a quick solution for you
at this point, but there are still some more things to consider.

- make sure you clean your buildroot tree properly after you changed
  low level settings like the ones discussed. Maybe you're just stuck on
  an old version of anything and br2 failed to rebuild the right bits.

- set CONFIG_DEBUG_USER in your kernel config and pass user_debug=255
  on startup. That will output more details when user space binaries
  crash.

- Consider using gcc 4.3.2 for your tests. I've never used a newer
  version for ARMs, so I can't prove that this is working.

- Try compiling a very simple program which just echos some text and
  cross-compile it for your target. Use the kernel's init= option to
  call it and see if it works.

If you manage to solve your problems, post your findings here for
others, please :)

Daniel