Cross compiling MIPS kernel under x86

Hello

I have an octeon board. I'm trying to use a custom kernel from kernel.org
instead of the Cavium one.

Under x86, I installed the CrossTools from kegel:
http://www.kegel.com/crosstool/

I tried to cross compile:
octane@darkstar:/opt/linux-2.6.34$ make ARCH=mips
CROSS_COMPILE=/opt/crosstool/gcc-3.4.5-glibc-2.3.6/mips-unknown-linux-gnu/bin/mips-unknown-linux-gnu-
  CHK     include/linux/version.h
  CHK     include/generated/utsrelease.h
  Checking missing-syscalls for N32
  CALL    scripts/checksyscalls.sh
  Checking missing-syscalls for O32
  CALL    scripts/checksyscalls.sh
  CALL    scripts/checksyscalls.sh
  CC      scripts/mod/empty.o
Assembler messages:
Error: Bad value (octeon) for -march
make[2]: *** [scripts/mod/empty.o] Error 1
make[1]: *** [scripts/mod] Error 2
make: *** [scripts] Error 2

So, I'm obviously missing a thing, but what?

Here is other information:
octane@darkstar:/opt$ cat world.c 
#include <stdio.h>
int main()
{
    printf("Hello world!\n");
    return 0;
}
octane@darkstar:/opt$
/opt/crosstool/gcc-3.4.5-glibc-2.3.6/mips-unknown-linux-gnu/bin/mips-unknown-linux-gnu-gcc
-march=octeon -o hello world.c 
world.c:1: error: bad value (octeon) for -march

31 more lines

octane indice wrote:

Hello

I have an octeon board. I'm trying to use a custom kernel from kernel.org
instead of the Cavium one.

Under x86, I installed the CrossTools from kegel:
http://www.kegel.com/crosstool/

I tried to cross compile:
octane@darkstar:/opt/linux-2.6.34$ make ARCH=mips
CROSS_COMPILE=/opt/crosstool/gcc-3.4.5-glibc-2.3.6/mips-unknown-linux-gnu/bin/mips-unknown-linux-gnu-
  CHK     include/linux/version.h
  CHK     include/generated/utsrelease.h
  Checking missing-syscalls for N32
  CALL    scripts/checksyscalls.sh
  Checking missing-syscalls for O32
  CALL    scripts/checksyscalls.sh
  CALL    scripts/checksyscalls.sh
  CC      scripts/mod/empty.o
Assembler messages:
Error: Bad value (octeon) for -march
make[2]: *** [scripts/mod/empty.o] Error 1
make[1]: *** [scripts/mod] Error 2
make: *** [scripts] Error 2

So, I'm obviously missing a thing, but what?

Here is other information:
octane@darkstar:/opt$ cat world.c #include <stdio.h>
int main()
{
    printf("Hello world!\n");
    return 0;
}
octane@darkstar:/opt$
/opt/crosstool/gcc-3.4.5-glibc-2.3.6/mips-unknown-linux-gnu/bin/mips-unknown-linux-gnu-gcc
-march=octeon -o hello world.c world.c:1: error: bad value (octeon) for -march
octane@darkstar:/opt$

octane@darkstar:/opt$ ls

32 more quoted lines

gcc-3.4.5-glibc-2.3.6


The recognition of "octeon" as an option does not show up in GCC until mainstream version 4.4.
looks like your gcc is too old gcc-3.4.5-glibc-2.3.6


Jan

On Mon, May 24, 2010 at 5:24 PM, octane indice [off-list ref] wrote:

Hello

I have an octeon board. I'm trying to use a custom kernel from kernel.org
instead of the Cavium one.

[skipped]

So, I'm obviously missing a thing, but what?

It looks like your toolchain is quite old. I just tried building a
Cavium Octeon defconfig using my custom toolchain based on GCC 4.3.1
and binutils 2.19.51.20090304, and the build was successfull. Before
you ask: yes, GCC did receive `-march=octeon' :)

Dmitri

On Mon, May 24, 2010 at 05:33:01PM +0300, Dmitri Vorobiev wrote:

It looks like your toolchain is quite old. I just tried building a
Cavium Octeon defconfig using my custom toolchain based on GCC 4.3.1
and binutils 2.19.51.20090304, and the build was successfull. Before
you ask: yes, GCC did receive `-march=octeon' :)

Tools requirements to build a kernel have become a little bit confusing.
I'm sure there are more restrictions that I've forgot.

 * The Lemote 2F defconfig requires binutils 2.20 to build.
 * GCC 3.2 is a lost cause for building 64-bit kernels
 * GCC 3.3 is broken but can just about be kludged to build a 64-bit kernel.
 * GCC 4.4 or a patched older version is required to build a kernel O2 or
   Indigo² with R10000 processors.
 * GCC 3.2 used to work for the rest but it's a very long time since I
   tested this for a modern kernel.
 * Linux 2.6.29 and older need a GCC older than 4.4.0 to compile. 

  Ralf

On Tue, May 25, 2010 at 09:13:41AM -0400, Ralf Baechle wrote:

On Mon, May 24, 2010 at 05:33:01PM +0300, Dmitri Vorobiev wrote:

quoted

It looks like your toolchain is quite old. I just tried building a
Cavium Octeon defconfig using my custom toolchain based on GCC 4.3.1
and binutils 2.19.51.20090304, and the build was successfull. Before
you ask: yes, GCC did receive `-march=octeon' :)

Tools requirements to build a kernel have become a little bit confusing.
I'm sure there are more restrictions that I've forgot.

 * The Lemote 2F defconfig requires binutils 2.20 to build.
 * GCC 3.2 is a lost cause for building 64-bit kernels
 * GCC 3.3 is broken but can just about be kludged to build a 64-bit kernel.
 * GCC 4.4 or a patched older version is required to build a kernel O2 or
   Indigo² with R10000 processors.
 * GCC 3.2 used to work for the rest but it's a very long time since I
   tested this for a modern kernel.
 * Linux 2.6.29 and older need a GCC older than 4.4.0 to compile. 

The toolchain from emdebian seems to work fine for me, though it is at times
a bit difficult to set up due to bad dependencies.

Guenter

En réponse à Ralf Baechle [off-list ref] :

On Mon, May 24, 2010 at 05:33:01PM +0300, Dmitri Vorobiev
wrote:

quoted

It looks like your toolchain is quite old. I just tried building a
Cavium Octeon defconfig using my custom toolchain based on
GCC 4.3.1 and binutils 2.19.51.20090304, and the build was
successfull. Before you ask: yes, GCC did receive
`-march=octeon' :)

Ok, thanks :)

Tools requirements to build a kernel have become a little bit
confusing. I'm sure there are more restrictions that I've forgot.

 * The Lemote 2F defconfig requires binutils 2.20 to build.
 * GCC 3.2 is a lost cause for building 64-bit kernels
 * GCC 3.3 is broken but can just about be kludged to build a
64-bit kernel.
 * GCC 4.4 or a patched older version is required to build a
kernel O2 or
   Indigo² with R10000 processors.
 * GCC 3.2 used to work for the rest but it's a very long time
since I tested this for a modern kernel.
 * Linux 2.6.29 and older need a GCC older than 4.4.0 to
compile. 

Ok. I'm following the info there:
http://www.linux-mips.org/wiki/Toolchains
but infos seems a bit old, so I followed the main idea
with newer binutils and gcc-4.4.4

It is still compiling gcc. When it will finish, I will ping back
and say if it compiles an octeon kernel 2.6.34

------------------- Fin du message d'origine ---------------------

Envoyé avec Inmano, ma messagerie renversante et gratuite : http://www.inmano.com

On Tue, May 25, 2010 at 03:58:25PM +0200, octane indice wrote:

Ok. I'm following the info there:
http://www.linux-mips.org/wiki/Toolchains
but infos seems a bit old, so I followed the main idea
with newer binutils and gcc-4.4.4

That page just documents available toolchains and is suffering from too
many details.  Somebody should do a bit weed killing on that page.

The build dependencies for the kernel are documented on

  http://www.linux-mips.org/wiki/Kernel_Build

which I've updated just before my previous posting.

It is still compiling gcc. When it will finish, I will ping back
and say if it compiles an octeon kernel 2.6.34

It will :-)

  Ralf

Sorry to bother you another time, but it's a FAIL :-(

I'm exactly in that situation:
http://gcc.gnu.org/ml/gcc-help/2009-08/msg00198.html

I read some docs, relaunched the configure process another time, read that I
should build with
make all-gcc 
But error is still present.

thanks

En réponse à Ralf Baechle [off-list ref] :

------------------ Début du message d'origine --------------------

On Tue, May 25, 2010 at 03:58:25PM +0200, octane indice wrote:

quoted

Ok. I'm following the info there:
http://www.linux-mips.org/wiki/Toolchains
but infos seems a bit old, so I followed the main idea
with newer binutils and gcc-4.4.4

That page just documents available toolchains and is suffering
from too
many details.  Somebody should do a bit weed killing on that
page.

The build dependencies for the kernel are documented on

  http://www.linux-mips.org/wiki/Kernel_Build

which I've updated just before my previous posting.

quoted

It is still compiling gcc. When it will finish, I will ping

back

quoted

and say if it compiles an octeon kernel 2.6.34

It will :-)

  Ralf

------------------- Fin du message d'origine ---------------------

Envoyé avec Inmano, ma messagerie renversante et gratuite : http://www.inmano.com

On Wed, May 26, 2010 at 4:11 PM, octane indice [off-list ref] wrote:

Sorry to bother you another time, but it's a FAIL :-(

Just checked that the following steps result in a successful build of
a vanilla 2.6.34 vmlinux:

cd /work/tmp/zoo/src
wget http://ftp.gnu.org/gnu/binutils/binutils-2.20.1.tar.bz2
cd ../build
../src/binutils-2.20.1/configure --prefix=/work/tmp/zoo
--target=mips64-unknown-linux-gnu
make
export PATH=/work/tmp/zoo/bin:$PATH
rm -rf *
cd ../src/
wget http://ftp.gnu.org/gnu/gcc/gcc-4.4.4/gcc-core-4.4.4.tar.bz2
tar jxf gcc-core-4.4.4.tar.bz2
cd ../build
../src/gcc-4.4.4/configure --target=mips64-unknown-linux-gnu
--prefix=/work/tmp/zoo --disable-threads --disable-shared
--disable-multilib --disable-libgcc --disable-libmudflap
--disable-libssp --disable-libgomp
make
make install
cd ../src
wget http://www.kernel.org/pub/linux/kernel/v2.6/linux-2.6.34.tar.bz2
tar jxf linux-2.6.34.tar.bz2
cd linux-2.6.34/
make ARCH=mips cavium-octeon_defconfig
make ARCH=mips CROSS_COMPILE=/work/tmp/zoo/bin/mips64-unknown-linux-gnu- vmlinux

Hope that helps.

Dmitri

Response to Dmitri Vorobiev [off-list ref] :

Just checked that the following steps result in a successful
build of a vanilla 2.6.34 vmlinux:

Thanks for taking the time to do it.

http://ftp.gnu.org/gnu/gcc/gcc-4.4.4/gcc-core-4.4.4.tar.bz2
tar jxf gcc-core-4.4.4.tar.bz2
cd ../build
../src/gcc-4.4.4/configure --target=mips64-unknown-linux-gnu
--prefix=/work/tmp/zoo --disable-threads --disable-shared
--disable-multilib --disable-libgcc --disable-libmudflap
--disable-libssp --disable-libgomp
make

It fails here with something related to stdc++. With adding a
--enable-language=c it works. So the configure line I used is:
../gcc-4.4.4/configure --target=mips64-unknown-linux-gnu
--prefix=/var/samba/mips --disable-threads --disable-shared
--disable-multilib --disable-libgcc --disable-libmudflap --disable-libssp
--disable-libgomp --enable-languages=c

make ARCH=mips cavium-octeon_defconfig
make ARCH=mips
CROSS_COMPILE=/work/tmp/zoo/bin/mips64-unknown-linux-gnu- vmlinux

Hope that helps.

It helped me a lot, thank you. The kernel compiles fine. 
The kernel is very huge: 39MBytes (!).
After a mips64-unknown-linux-gnu-strip, it downsized to 3.3MBytes.
But that kernel doesn't boot.

The system in the board I have uses in U-boot:
ext2load ide 0 4000000 vmlinux
bootoctlinux 4000000 (other args..)

But when I replace the vmlinux file with mine called 'mips', it says:
RSEC-K1# ext2load ide 0 4000000 mips

3362840 bytes read
WARNING: Data loaded outside of the reserved load area, memory corruption
may occur.
WARNING: Please refer to the bootloader memory map documentation for more
information.
RSEC-K1# bootoctlinux 4000000
ELF file is 64 bit
Attempting to allocate memory for ELF segment: addr: 0xffffffff81100000
(adjusted to: 0x0000000001100000), size 0x355c00
Allocated memory for ELF segment: addr: 0xffffffff81100000, size 0x355c00
Attempting to allocate memory for ELF segment: addr: 0xffffffff81343da0
(adjusted to: 0x0000000001343da0), size 0x24
Error allocating memory for elf image!
## ERROR loading File!
RSEC-K1#

I tried a lot of different numbers like 2000000 or greater, depending of
what I find in mailing lists, but I don't find the right value. Is there a
way to compute it, or it's kernel I made that is not right?

Thanks

-------------------  ---------------------

Envoyé avec Inmano, ma messagerie renversante et gratuite : http://www.inmano.com

On Thu, May 27, 2010 at 7:29 PM, octane indice [off-list ref] wrote:

Response to Dmitri Vorobiev [off-list ref] :

I tried a lot of different numbers like 2000000 or greater, depending of
what I find in mailing lists, but I don't find the right value. Is there a
way to compute it, or it's kernel I made that is not right?

Well, I can't say offhand, and I don't have hardware to check that.
One thing that attracts attention is that the size of 0x24 is
apparently too small.

Hard to say what's going on without knowing the hardware and seeing
the docs for it.

Dmitri

On 05/27/2010 09:29 AM, octane indice wrote:

Response to Dmitri Vorobiev[off-list ref]  :

quoted

Just checked that the following steps result in a successful
build of a vanilla 2.6.34 vmlinux:

Thanks for taking the time to do it.

quoted

http://ftp.gnu.org/gnu/gcc/gcc-4.4.4/gcc-core-4.4.4.tar.bz2
tar jxf gcc-core-4.4.4.tar.bz2
cd ../build
../src/gcc-4.4.4/configure --target=mips64-unknown-linux-gnu
--prefix=/work/tmp/zoo --disable-threads --disable-shared
--disable-multilib --disable-libgcc --disable-libmudflap
--disable-libssp --disable-libgomp
make

It fails here with something related to stdc++. With adding a
--enable-language=c it works. So the configure line I used is:
../gcc-4.4.4/configure --target=mips64-unknown-linux-gnu
--prefix=/var/samba/mips --disable-threads --disable-shared
--disable-multilib --disable-libgcc --disable-libmudflap --disable-libssp
--disable-libgomp --enable-languages=c

quoted

make ARCH=mips cavium-octeon_defconfig
make ARCH=mips
CROSS_COMPILE=/work/tmp/zoo/bin/mips64-unknown-linux-gnu- vmlinux

Hope that helps.

It helped me a lot, thank you. The kernel compiles fine.
The kernel is very huge: 39MBytes (!).
After a mips64-unknown-linux-gnu-strip, it downsized to 3.3MBytes.
But that kernel doesn't boot.

The system in the board I have uses in U-boot:
ext2load ide 0 4000000 vmlinux
bootoctlinux 4000000 (other args..)

But when I replace the vmlinux file with mine called 'mips', it says:
RSEC-K1# ext2load ide 0 4000000 mips

16 more quoted lines

Early Octeon bootloaders cannot handle PT_NOTE program headers, I think that is what is biting you here.

If you can upgrade to an SDK-1.9 or later bootloader, I would recommend that.  Otherwise remove the PT_NOTE from your kernel image (the technique for doing this is left as an excise for the reader, but I have found that emacs hexl mode works well).

David Daney

On Fri, May 28, 2010 at 12:34 AM, David Daney [off-list ref] wrote:

On 05/27/2010 09:29 AM, octane indice wrote:

quoted

Otherwise remove the PT_NOTE from your kernel image (the technique
for doing this is left as an excise for the reader, but I have found that
emacs hexl mode works well).

I still believe using objcopy should be a bit more straightforward :)

Dmitri

On Thu, 27 May 2010 14:34:09 -0700, David Daney [off-list ref] wrote:

On 05/27/2010 09:29 AM, octane indice wrote:

quoted

Response to Dmitri Vorobiev[off-list ref]  :

quoted

Just checked that the following steps result in a successful
build of a vanilla 2.6.34 vmlinux:

Thanks for taking the time to do it.

quoted

http://ftp.gnu.org/gnu/gcc/gcc-4.4.4/gcc-core-4.4.4.tar.bz2
tar jxf gcc-core-4.4.4.tar.bz2
cd ../build
../src/gcc-4.4.4/configure --target=mips64-unknown-linux-gnu
--prefix=/work/tmp/zoo --disable-threads --disable-shared
--disable-multilib --disable-libgcc --disable-libmudflap
--disable-libssp --disable-libgomp
make

It fails here with something related to stdc++. With adding a
--enable-language=c it works. So the configure line I used is:
../gcc-4.4.4/configure --target=mips64-unknown-linux-gnu
--prefix=/var/samba/mips --disable-threads --disable-shared
--disable-multilib --disable-libgcc --disable-libmudflap --disable-libssp
--disable-libgomp --enable-languages=c

quoted

make ARCH=mips cavium-octeon_defconfig
make ARCH=mips
CROSS_COMPILE=/work/tmp/zoo/bin/mips64-unknown-linux-gnu- vmlinux

Hope that helps.

It helped me a lot, thank you. The kernel compiles fine.
The kernel is very huge: 39MBytes (!).
After a mips64-unknown-linux-gnu-strip, it downsized to 3.3MBytes.
But that kernel doesn't boot.

The system in the board I have uses in U-boot:
ext2load ide 0 4000000 vmlinux
bootoctlinux 4000000 (other args..)

But when I replace the vmlinux file with mine called 'mips', it says:
RSEC-K1# ext2load ide 0 4000000 mips

25 more quoted lines

In the top level Makefile, you'll find this:


# Use --build-id when available.
LDFLAGS_BUILD_ID = $(patsubst -Wl$(comma)%,%,\
			      $(call cc-ldoption, -Wl$(comma)--build-id,))

Comment that out and it should fix you up.

Phil

David Daney

-- 
Phil Staub, Senior Member of Technical Staff, Wind River
Direct: 702.290.0470 Fax: 702.982.0085

En réponse à Phil Staub [off-list ref] :

In the top level Makefile, you'll find this:

# Use --build-id when available.
LDFLAGS_BUILD_ID = $(patsubst -Wl$(comma)%,%,\
			      $(call cc-ldoption, -Wl$(comma)--build-id,))

Comment that out and it should fix you up.

Excellent. It boots. It fails next because it can't find
the root device, but I know how to manage that 

Phil

Thanks
 

-- 
Phil Staub, Senior Member of Technical Staff, Wind River
Direct: 702.290.0470 Fax: 702.982.0085

Envoyé avec Inmano, ma messagerie renversante et gratuite : http://www.inmano.com