[Buildroot] building ARM toolchain with nofpu

[Buildroot] building ARM toolchain with nofpu

[Michel Benoit]

Hi,

I'm trying to build a toolchain with soft floating point enabled for
ARM arm26t target platform.  I'm doing this in order to build u-boot
for my target.

I downloaded the latest verison of buildroot and selected "Use
software floating point by defaul" in buildroot menuconfig and
unselected "Target CPU has a floating point unit (FPU) " in
uclib-menuconfig.

The build stops due to missing symbols while linking uclibc:

make[1]: Entering directory
`/home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/toolchain_build_arm_nofpu/uClibc'
  LD libuClibc-0.9.29.so
libc/libc_so.a(difftime.os): In function `difftime':
difftime.c:(.text+0x8): undefined reference to `__floatsidf'
difftime.c:(.text+0x2c): undefined reference to `__subdf3'
libc/libc_so.a(_fpmaxtostr.os): In function `_fpmaxtostr':
_fpmaxtostr.c:(.text+0xd4): undefined reference to `__nedf2'
_fpmaxtostr.c:(.text+0xf8): undefined reference to `__eqdf2'
_fpmaxtostr.c:(.text+0x114): undefined reference to `__divdf3'
_fpmaxtostr.c:(.text+0x120): undefined reference to `__ltdf2'
_fpmaxtostr.c:(.text+0x1c4): undefined reference to `__muldf3'
_fpmaxtostr.c:(.text+0x388): undefined reference to `__gedf2'
_fpmaxtostr.c:(.text+0x430): undefined reference to `__adddf3'
libc/libc_so.a(__psfs_do_numeric.os): In function `__psfs_do_numeric':
__psfs_do_numeric.c:(.text+0x520): undefined reference to `__truncdfsf2'
libc/libc_so.a(strtof.os): In function `strtof':
strtof.c:(.text+0x1c): undefined reference to `__extendsfdf2'
/home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/build_arm_nofpu/staging_dir/lib/gcc/arm-linux-uclibc/4.1.1/libgcc.a(_fixunsdfsi.o):
In function `__fixunsdfsi':
libgcc2.c:(.text+0x34): undefined reference to `__fixdfsi'
make[2]: *** [lib/libc.so] Error 1
make[1]: *** [lib/libc.so.0] Error 2
make[1]: Leaving directory
`/home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/toolchain_build_arm_nofpu/uClibc'
make: *** [/home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/toolchain_build_arm_nofpu/uClibc/lib/libc.a]
Error 2


Are there any other config options that need to be selected/unselected
in order to get this to work?

Michel

[Buildroot] building ARM toolchain with nofpu

[Yann E. MORIN]

Hello all!
Michel, Carl.

On Thursday 01 February 2007 010, Michel Benoit wrote:
> I'm trying to build a toolchain with soft floating point enabled for
> ARM arm26t target platform.  I'm doing this in order to build u-boot
> for my target.
[--SNIP--]
> The build stops due to missing symbols while linking uclibc:
> 
> make[1]: Entering directory
> `/home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/toolchain_build_arm_nofpu/uClibc'
>   LD libuClibc-0.9.29.so
> libc/libc_so.a(difftime.os): In function `difftime':
> difftime.c:(.text+0x8): undefined reference to `__floatsidf'
> difftime.c:(.text+0x2c): undefined reference to `__subdf3'
> libc/libc_so.a(_fpmaxtostr.os): In function `_fpmaxtostr':
> _fpmaxtostr.c:(.text+0xd4): undefined reference to `__nedf2'
> _fpmaxtostr.c:(.text+0xf8): undefined reference to `__eqdf2'
> _fpmaxtostr.c:(.text+0x114): undefined reference to `__divdf3'
> _fpmaxtostr.c:(.text+0x120): undefined reference to `__ltdf2'
> _fpmaxtostr.c:(.text+0x1c4): undefined reference to `__muldf3'
> _fpmaxtostr.c:(.text+0x388): undefined reference to `__gedf2'
> _fpmaxtostr.c:(.text+0x430): undefined reference to `__adddf3'
> libc/libc_so.a(__psfs_do_numeric.os): In function `__psfs_do_numeric':
> __psfs_do_numeric.c:(.text+0x520): undefined reference to `__truncdfsf2'
> libc/libc_so.a(strtof.os): In function `strtof':
> strtof.c:(.text+0x1c): undefined reference to `__extendsfdf2'
> /home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/build_arm_nofpu/staging_dir/lib/gcc/arm-linux-uclibc/4.1.1/libgcc.a(_fixunsdfsi.o):
> In function `__fixunsdfsi':
> libgcc2.c:(.text+0x34): undefined reference to `__fixdfsi'
> make[2]: *** [lib/libc.so] Error 1
> make[1]: *** [lib/libc.so.0] Error 2
> make[1]: Leaving directory
> `/home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/toolchain_build_arm_nofpu/uClibc'
> make: *** [/home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/toolchain_build_arm_nofpu/uClibc/lib/libc.a]
> Error 2

I'm tracking this down right now, and I've found that those functions are not
included in libgcc.a because gcc-4.1.1/gcc/config.gcc does not validate
arm/t-arm-elf for arm*-linux* targets, but arm/t-arm, which does not describe
those functions as to be included in libgcc.a.

I've tried a quick fix (replacing t-arm with t-arm-elf), but it still does
not work. I haven't had time right to investigate more. Later tongiht.

> Are there any other config options that need to be selected/unselected
> in order to get this to work?

You could try with a known working gcc (gcc-3.4.4 is working real fine here),
unless you really can't for very good reasons, but compiling u-boot should
Just Work (TM).

Regards,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +0/33 662376056 | Software  Designer | \ / CAMPAIGN     |   ^                |
| --==< ?_? >==-- ?------------.-------:  X  AGAINST      |  /e\  There is no  |
| http://ymorin.is-a-geek.org/ | (*_*) | / \ HTML MAIL    |  """  conspiracy.  |
?------------------------------?-------?------------------?--------------------?

[Buildroot] building ARM toolchain with nofpu

[Michel Benoit]

Thanks that worked!

On 2/1/07, Yann E. MORIN <yann.morin.1998@anciens.enib.fr> wrote:
> Hello all!
> Michel, Carl.
>
> On Thursday 01 February 2007 010, Michel Benoit wrote:
> > I'm trying to build a toolchain with soft floating point enabled for
> > ARM arm26t target platform.  I'm doing this in order to build u-boot
> > for my target.
> [--SNIP--]
> > The build stops due to missing symbols while linking uclibc:
> >
> > make[1]: Entering directory
> > `/home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/toolchain_build_arm_nofpu/uClibc'
> >   LD libuClibc-0.9.29.so
> > libc/libc_so.a(difftime.os): In function `difftime':
> > difftime.c:(.text+0x8): undefined reference to `__floatsidf'
> > difftime.c:(.text+0x2c): undefined reference to `__subdf3'
> > libc/libc_so.a(_fpmaxtostr.os): In function `_fpmaxtostr':
> > _fpmaxtostr.c:(.text+0xd4): undefined reference to `__nedf2'
> > _fpmaxtostr.c:(.text+0xf8): undefined reference to `__eqdf2'
> > _fpmaxtostr.c:(.text+0x114): undefined reference to `__divdf3'
> > _fpmaxtostr.c:(.text+0x120): undefined reference to `__ltdf2'
> > _fpmaxtostr.c:(.text+0x1c4): undefined reference to `__muldf3'
> > _fpmaxtostr.c:(.text+0x388): undefined reference to `__gedf2'
> > _fpmaxtostr.c:(.text+0x430): undefined reference to `__adddf3'
> > libc/libc_so.a(__psfs_do_numeric.os): In function `__psfs_do_numeric':
> > __psfs_do_numeric.c:(.text+0x520): undefined reference to `__truncdfsf2'
> > libc/libc_so.a(strtof.os): In function `strtof':
> > strtof.c:(.text+0x1c): undefined reference to `__extendsfdf2'
> > /home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/build_arm_nofpu/staging_dir/lib/gcc/arm-linux-uclibc/4.1.1/libgcc.a(_fixunsdfsi.o):
> > In function `__fixunsdfsi':
> > libgcc2.c:(.text+0x34): undefined reference to `__fixdfsi'
> > make[2]: *** [lib/libc.so] Error 1
> > make[1]: *** [lib/libc.so.0] Error 2
> > make[1]: Leaving directory
> > `/home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/toolchain_build_arm_nofpu/uClibc'
> > make: *** [/home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/toolchain_build_arm_nofpu/uClibc/lib/libc.a]
> > Error 2
>
> I'm tracking this down right now, and I've found that those functions are not
> included in libgcc.a because gcc-4.1.1/gcc/config.gcc does not validate
> arm/t-arm-elf for arm*-linux* targets, but arm/t-arm, which does not describe
> those functions as to be included in libgcc.a.
>
> I've tried a quick fix (replacing t-arm with t-arm-elf), but it still does
> not work. I haven't had time right to investigate more. Later tongiht.
>
> > Are there any other config options that need to be selected/unselected
> > in order to get this to work?
>
> You could try with a known working gcc (gcc-3.4.4 is working real fine here),
> unless you really can't for very good reasons, but compiling u-boot should
> Just Work (TM).
>
> Regards,
> Yann E. MORIN.
>
> --
> .-----------------.--------------------.------------------.--------------------.
> |  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
> | +0/33 662376056 | Software  Designer | \ / CAMPAIGN     |   ^                |
> | --==< ?_? >==-- ?------------.-------:  X  AGAINST      |  /e\  There is no  |
> | http://ymorin.is-a-geek.org/ | (*_*) | / \ HTML MAIL    |  """  conspiracy.  |
> ?------------------------------?-------?------------------?--------------------?
>
>

[Buildroot] building ARM toolchain with nofpu

[Ulf Samuelsson]

Michel Benoit wrote:
> Thanks that worked!
>
> On 2/1/07, Yann E. MORIN <yann.morin.1998@anciens.enib.fr> wrote:
>> Hello all!
>> Michel, Carl.
>>
>> On Thursday 01 February 2007 010, Michel Benoit wrote:
>>> I'm trying to build a toolchain with soft floating point enabled for
>>> ARM arm26t target platform.  I'm doing this in order to build u-boot
>>> for my target.
>> [--SNIP--]

You can build u-boot with a NWFPE compiler
if you just remove the "-msoft-float" from cpu/arm926ejs/config.mk and
from cpu/arm926ejs/at91sam926x/config.mk
That is what I do in my buildroot (That part yet to be submitted).


Best Regards,
Ulf Samuelsson
ulf at atmel.com
GSM:  +46 (706) 22 44 57
Tel:     +46  (8) 441 54 22
Fax:     +46 (8) 441 54 29
Mail: Box 2033  174 02 Sundbyberg
Visit: Kavalleriv?gen 24
          174 58 Sundbyberg'
Sweden

[Buildroot] building ARM toolchain with nofpu

[Michel Benoit]

Hej Ulf

I don't think that the NWFPE stuff is linked into the u-boot code
automatically.  The build instructions in the u-boot tree are very
explicit about using -msoft-float.

When I build u-boot with a buildroot toolchain based on gcc 3.4.3 with
no fpu support the bootloader works as it should.

When I build u-boot with fpu support in the toolchain and I remove the
-msfot-float option according to your suggestion, the bootloader
crashes almost immediately.  It may be related to NAND access but
could also be due to the printouts (printf can use float ops).

On 2/1/07, Ulf Samuelsson <ulf@atmel.com> wrote:
> Michel Benoit wrote:
> > Thanks that worked!
> >
> > On 2/1/07, Yann E. MORIN <yann.morin.1998@anciens.enib.fr> wrote:
> >> Hello all!
> >> Michel, Carl.
> >>
> >> On Thursday 01 February 2007 010, Michel Benoit wrote:
> >>> I'm trying to build a toolchain with soft floating point enabled for
> >>> ARM arm26t target platform.  I'm doing this in order to build u-boot
> >>> for my target.
> >> [--SNIP--]
>
> You can build u-boot with a NWFPE compiler
> if you just remove the "-msoft-float" from cpu/arm926ejs/config.mk and
> from cpu/arm926ejs/at91sam926x/config.mk
> That is what I do in my buildroot (That part yet to be submitted).
>
>
> Best Regards,
> Ulf Samuelsson
> ulf at atmel.com
> GSM:  +46 (706) 22 44 57
> Tel:     +46  (8) 441 54 22
> Fax:     +46 (8) 441 54 29
> Mail: Box 2033  174 02 Sundbyberg
> Visit: Kavalleriv?gen 24
>           174 58 Sundbyberg'
> Sweden
>
>

[Buildroot] building ARM toolchain with nofpu

[James Newell]

Has any one had any progress with building uclibc default soft float
using gcc 4.1.1?

Thanks,
Jim

On Thu, 2007-02-01 at 20:15 +0100, Yann E. MORIN wrote:
> Hello all!
> Michel, Carl.
> 
> On Thursday 01 February 2007 010, Michel Benoit wrote:
> > I'm trying to build a toolchain with soft floating point enabled for
> > ARM arm26t target platform.  I'm doing this in order to build u-boot
> > for my target.
> [--SNIP--]
> > The build stops due to missing symbols while linking uclibc:
> > 
> > make[1]: Entering directory
> > `/home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/toolchain_build_arm_nofpu/uClibc'
> >   LD libuClibc-0.9.29.so
> > libc/libc_so.a(difftime.os): In function `difftime':
> > difftime.c:(.text+0x8): undefined reference to `__floatsidf'
> > difftime.c:(.text+0x2c): undefined reference to `__subdf3'
> > libc/libc_so.a(_fpmaxtostr.os): In function `_fpmaxtostr':
> > _fpmaxtostr.c:(.text+0xd4): undefined reference to `__nedf2'
> > _fpmaxtostr.c:(.text+0xf8): undefined reference to `__eqdf2'
> > _fpmaxtostr.c:(.text+0x114): undefined reference to `__divdf3'
> > _fpmaxtostr.c:(.text+0x120): undefined reference to `__ltdf2'
> > _fpmaxtostr.c:(.text+0x1c4): undefined reference to `__muldf3'
> > _fpmaxtostr.c:(.text+0x388): undefined reference to `__gedf2'
> > _fpmaxtostr.c:(.text+0x430): undefined reference to `__adddf3'
> > libc/libc_so.a(__psfs_do_numeric.os): In function `__psfs_do_numeric':
> > __psfs_do_numeric.c:(.text+0x520): undefined reference to `__truncdfsf2'
> > libc/libc_so.a(strtof.os): In function `strtof':
> > strtof.c:(.text+0x1c): undefined reference to `__extendsfdf2'
> > /home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/build_arm_nofpu/staging_dir/lib/gcc/arm-linux-uclibc/4.1.1/libgcc.a(_fixunsdfsi.o):
> > In function `__fixunsdfsi':
> > libgcc2.c:(.text+0x34): undefined reference to `__fixdfsi'
> > make[2]: *** [lib/libc.so] Error 1
> > make[1]: *** [lib/libc.so.0] Error 2
> > make[1]: Leaving directory
> > `/home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/toolchain_build_arm_nofpu/uClibc'
> > make: *** [/home/michel/Projects/sepab/exyplus/sw/bsp/build/buildroot/toolchain_build_arm_nofpu/uClibc/lib/libc.a]
> > Error 2
> 
> I'm tracking this down right now, and I've found that those functions are not
> included in libgcc.a because gcc-4.1.1/gcc/config.gcc does not validate
> arm/t-arm-elf for arm*-linux* targets, but arm/t-arm, which does not describe
> those functions as to be included in libgcc.a.
> 
> I've tried a quick fix (replacing t-arm with t-arm-elf), but it still does
> not work. I haven't had time right to investigate more. Later tongiht.
> 
> > Are there any other config options that need to be selected/unselected
> > in order to get this to work?
> 
> You could try with a known working gcc (gcc-3.4.4 is working real fine here),
> unless you really can't for very good reasons, but compiling u-boot should
> Just Work (TM).
> 
> Regards,
> Yann E. MORIN.
> 

[Buildroot] building ARM toolchain with nofpu

[Yann E. MORIN]

On Wednesday 07 February 2007 210, James Newell wrote:
> Has any one had any progress with building uclibc default soft float
> using gcc 4.1.1?

Yes.

Short answer: it does not work.
See: http://gcc.gnu.org/bugzilla/show_bug.cgi?id)329  [1]
(not directly related, but yet, useless on ARM).

Long answer: I've got a patch (attached) that allow building gcc-4.1.1 with
uClibc-0.9.28.1 (didn't try svn head). You have to add it to the collection
of patches provided by buildroot. Unfortunately, that doesn't give a useable
x-compiler because of [1].

Regards,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +0/33 662376056 | Software  Designer | \ / CAMPAIGN     |   ^                |
| --=< ?_? >=-- ?------------.-------:  X  AGAINST      |  /e\  There is no  |
| http://ymorin.is-a-geek.org/ | (*_*) | / \ HTML MAIL    |  """  conspiracy.  |
?------------------------------?-------?------------------?--------------------?
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 801-softfloat-fix.patch
Type: text/x-diff
Size: 2827 bytes
Desc: not available
Url : http://busybox.net/lists/buildroot/attachments/20070207/e19457c3/attachment.bin 

[Buildroot] building ARM toolchain with nofpu

[Yann E. MORIN]

On Wednesday 07 February 2007 212, Yann E. MORIN wrote:
> Short answer: it does not work.
> See: http://gcc.gnu.org/bugzilla/show_bug.cgi?id)329  [1]
> (not directly related, but yet, useless on ARM).

I realise I'm being vague. Some explanations below.

In fact, the patch I posted allows one to successfully build a x-compiler
for soft-float armeb (mine is armeb-xscale-linux-uclibc, tuned for Xscale).
Unfortunately, the resulting gcc barfs on the squashfs code in the kernel.
A smaller test-case is available in the bug report, suggesting this is quite
easy to hit.

That's why I've reverted to gcc-4.0.4, which still needs the attached patch,
but Just Work (TM).

Regards,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +0/33 662376056 | Software  Designer | \ / CAMPAIGN     |   ^                |
| --=< ?_? >=-- ?------------.-------:  X  AGAINST      |  /e\  There is no  |
| http://ymorin.is-a-geek.org/ | (*_*) | / \ HTML MAIL    |  """  conspiracy.  |
?------------------------------?-------?------------------?--------------------?
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 801-arm-softfloat.patch
Type: text/x-diff
Size: 1137 bytes
Desc: not available
Url : http://busybox.net/lists/buildroot/attachments/20070207/5fbe69cd/attachment.bin 

[Buildroot] building ARM toolchain with nofpu

[carl zhu]

Hi Yann, so you mean if use this patch, GCC 4.0.4 can work with uclibc? I
still feel a little strange, because I have googled several days before,
and  I have found someone fixed this issue with applied IIRC, Does it do
help for this issue?

I am new for build cross-compiler.

related link:http://tech.groups.yahoo.com/group/gnuarm/message/1325


On 2/8/07, Yann E. MORIN <yann.morin.1998@anciens.enib.fr> wrote:
>
> On Wednesday 07 February 2007 212, Yann E. MORIN wrote:
> > Short answer: it does not work.
> > See: http://gcc.gnu.org/bugzilla/show_bug.cgi?id)329  [1]
> > (not directly related, but yet, useless on ARM).
>
> I realise I'm being vague. Some explanations below.
>
> In fact, the patch I posted allows one to successfully build a x-compiler
> for soft-float armeb (mine is armeb-xscale-linux-uclibc, tuned for
> Xscale).
> Unfortunately, the resulting gcc barfs on the squashfs code in the kernel.
> A smaller test-case is available in the bug report, suggesting this is
> quite
> easy to hit.
>
> That's why I've reverted to gcc-4.0.4, which still needs the attached
> patch,
> but Just Work (TM).
>
> Regards,
> Yann E. MORIN.
>
> --
>
> .-----------------.--------------------.------------------.--------------------.
> |  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics'
> conspiracy: |
> | +0/33 662376056 | Software  Designer | \ / CAMPAIGN     |
> ^                |
> | --=< ?_? >=-- ?------------.-------:  X  AGAINST      |  /e\  There is
> no  |
> | http://ymorin.is-a-geek.org/ | (*_*) | / \ HTML
> MAIL    |  """  conspiracy.  |
>
> ?------------------------------?-------?------------------?--------------------?
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://busybox.net/lists/buildroot/attachments/20070208/f3754625/attachment.htm 

[Buildroot] building ARM toolchain with nofpu

[James Newell]

On Thu, 2007-02-08 at 15:50 +0800, carl zhu wrote:
> Hi Yann, so you mean if use this patch, GCC 4.0.4 can work with
> uclibc? I still feel a little strange, because I have googled several
> days before, and  I have found someone fixed this issue with applied
> IIRC, Does it do help for this issue?

I think I'm just confused in general.

At first thought I needed a gcc EABI toolchain
(http://wiki.debian.org/ArmEabiPort) for performance reasons but looking
at this table for ARM cores
(http://en.wikipedia.org/wiki/ARM_architecture), I am using the Davinci
which is a ARM926EJ-S, and doesn't appear to have vfp support.  I assume
then it would be better to use soft toolchain.  Is there any way
distinguish an arm elf binary that was built with vfp, fpa, and soft
using the 'file' command or readelf command?  I appreciate the
discussion.  I'm fairly new to arm, and buildroot, but enjoying the
learning curve.
Thanks,
Jim


>  
> I am new for build cross-compiler.
>  
> related link:http://tech.groups.yahoo.com/group/gnuarm/message/1325
> 
>  
> On 2/8/07, Yann E. MORIN <yann.morin.1998@anciens.enib.fr> wrote: 
>         On Wednesday 07 February 2007 212, Yann E. MORIN wrote:
>         > Short answer: it does not work.
>         > See: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29329  [1]
>         > (not directly related, but yet, useless on ARM).
>         
>         I realise I'm being vague. Some explanations below. 
>         
>         In fact, the patch I posted allows one to successfully build a
>         x-compiler
>         for soft-float armeb (mine is armeb-xscale-linux-uclibc, tuned
>         for Xscale).
>         Unfortunately, the resulting gcc barfs on the squashfs code in
>         the kernel. 
>         A smaller test-case is available in the bug report, suggesting
>         this is quite
>         easy to hit.
>         
>         That's why I've reverted to gcc-4.0.4, which still needs the
>         attached patch,
>         but Just Work (TM).
>         
>         Regards, 
>         Yann E. MORIN.
>         
>         --
>         .-----------------.--------------------.------------------.--------------------.
>         |  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON |
>         Erics' conspiracy: |
>         | +0/33 662376056 | Software  Designer | \ / CAMPAIGN     |
>         ^                | 
>         | --==< ?_? >==--
>         ?------------.-------:  X  AGAINST      |  /e\  There is no  |
>         | http://ymorin.is-a-geek.org/ | (*_*) | / \ HTML
>         MAIL    |  """  conspiracy.  | 
>         ?------------------------------?-------?------------------?--------------------?
>         
> 

[Buildroot] building ARM toolchain with nofpu

[Bernhard Fischer]

On Wed, Feb 07, 2007 at 09:24:29PM +0100, Yann E. MORIN wrote:
>On Wednesday 07 February 2007 210, James Newell wrote:
>> Has any one had any progress with building uclibc default soft float
>> using gcc 4.1.1?
>
>Yes.
>
>Short answer: it does not work.
>See: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29329  [1]
>(not directly related, but yet, useless on ARM).

It would be of tremendous help to have one non-float mmu-less config for
a kernel that i can check against qemu (i.e. run the resulting
filesystem and kernel via current qemu).

Does anybody have such a .config i could play with?

[Buildroot] building ARM toolchain with nofpu

[Yann E. MORIN]

Bernhard,
All,

On Thursday 08 February 2007 182, Bernhard Fischer wrote:
> It would be of tremendous help to have one non-float mmu-less config for
> a kernel that i can check against qemu (i.e. run the resulting
> filesystem and kernel via current qemu).
> Does anybody have such a .config i could play with?

I'm using a highly modified, soon-to-be-available version of crosstool to
build my toolchains, not buildroot (except for vampirising its patches! :-) ).

I can try with buildroot, and submit a patch if I succeed. More later (I'm
kind of busy right now with a BluePAD tablet PC and my crosstool).

The problem lies more with gcc than with uClibc or binutils or the kernel
headers.

Regards,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +0/33 662376056 | Software  Designer | \ / CAMPAIGN     |   ^                |
| --==< ?_? >==-- ?------------.-------:  X  AGAINST      |  /e\  There is no  |
| http://ymorin.is-a-geek.org/ | (*_*) | / \ HTML MAIL    |  """  conspiracy.  |
?------------------------------?-------?------------------?--------------------?

[Buildroot] building ARM toolchain with nofpu

[Bernhard Fischer]

On Thu, Feb 08, 2007 at 06:33:53PM +0100, Yann E. MORIN wrote:
>Bernhard,
>All,
>
>On Thursday 08 February 2007 182, Bernhard Fischer wrote:
>> It would be of tremendous help to have one non-float mmu-less config for
>> a kernel that i can check against qemu (i.e. run the resulting
>> filesystem and kernel via current qemu).
>> Does anybody have such a .config i could play with?
>
>I'm using a highly modified, soon-to-be-available version of crosstool to
>build my toolchains, not buildroot (except for vampirising its patches! :-) ).
>
>I can try with buildroot, and submit a patch if I succeed. More later (I'm
>kind of busy right now with a BluePAD tablet PC and my crosstool).
>
>The problem lies more with gcc than with uClibc or binutils or the kernel
>headers.

I was talking about a kernel .config and a hint on what arm variant qemu
supports (nick-name plus cpu name like "dominator" arm0815xyz).

[Buildroot] building ARM toolchain with nofpu

[James Newell]

The buildroot only seems to go to gcc 4.0.3? 

On Wed, 2007-02-07 at 21:36 +0100, Yann E. MORIN wrote:
> On Wednesday 07 February 2007 212, Yann E. MORIN wrote:
> > Short answer: it does not work.
> > See: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29329  [1]
> > (not directly related, but yet, useless on ARM).
> 
> I realise I'm being vague. Some explanations below.
> 
> In fact, the patch I posted allows one to successfully build a x-compiler
> for soft-float armeb (mine is armeb-xscale-linux-uclibc, tuned for Xscale).
> Unfortunately, the resulting gcc barfs on the squashfs code in the kernel.
> A smaller test-case is available in the bug report, suggesting this is quite
> easy to hit.
> 
> That's why I've reverted to gcc-4.0.4, which still needs the attached patch,
> but Just Work (TM).
> 
> Regards,
> Yann E. MORIN.
> 

[Buildroot] building ARM toolchain with nofpu

[Bernhard Fischer]

On Thu, Feb 08, 2007 at 08:41:53PM -0500, James Newell wrote:
>The buildroot only seems to go to gcc 4.0.3? 

I'll add 4.0.4 later today (provided that it builds and works), fwiw

[Buildroot] building ARM toolchain with nofpu

[carl zhu]

Hi Yann, I have tried this patch. it fail to patch the code, do you tried it
on GCC 4.0.4?

On 2/8/07, Yann E. MORIN <yann.morin.1998@anciens.enib.fr> wrote:
>
> On Wednesday 07 February 2007 212, Yann E. MORIN wrote:
> > Short answer: it does not work.
> > See: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29329  [1]
> > (not directly related, but yet, useless on ARM).
>
> I realise I'm being vague. Some explanations below.
>
> In fact, the patch I posted allows one to successfully build a x-compiler
> for soft-float armeb (mine is armeb-xscale-linux-uclibc, tuned for
> Xscale).
> Unfortunately, the resulting gcc barfs on the squashfs code in the kernel.
> A smaller test-case is available in the bug report, suggesting this is
> quite
> easy to hit.
>
> That's why I've reverted to gcc-4.0.4, which still needs the attached
> patch,
> but Just Work (TM).
>
> Regards,
> Yann E. MORIN.
>
> --
>
> .-----------------.--------------------.------------------.--------------------.
> |  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics'
> conspiracy: |
> | +0/33 662376056 | Software  Designer | \ / CAMPAIGN     |
> ^                |
> | --==< ?_? >==-- ?------------.-------:  X  AGAINST      |  /e\  There is
> no  |
> | http://ymorin.is-a-geek.org/ | (*_*) | / \ HTML
> MAIL    |  """  conspiracy.  |
>
> ?------------------------------?-------?------------------?--------------------?
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://busybox.net/lists/buildroot/attachments/20070301/e42c38f0/attachment.htm 

[Buildroot] building ARM toolchain with nofpu

[Yann E. MORIN]

Bernhard,
All,

On Friday 09 February 2007 192, Bernhard Fischer wrote:
> On Fri, Feb 09, 2007 at 06:57:13PM +0100, Yann E. MORIN wrote:
> >On Friday 09 February 2007 152, Bernhard Fischer wrote:
> >> I'm curious as to why you think we don't have -lfloat?
> >I din't say that either.
> Well, IIRC one of your patches (for gcc-4.0.4 in this thread) removed
> -lfloat which begs the above question :)

Right. Fact is that I also find a library that is years old is either:
  - deprecated,
  - very stable.

As a default, I consider it deprecated, fumble in the other sources I have
(eg. gcc), find possible alternatives, try to make them work, and only if
that fails, will I reconsider libfloat (which I'll do this WE!)

> I simply do not know when gcc grew it's own float-fallbacks (and i don't
> want to search for it, TBH).

Bahhh... :-)
3.4.4: has it
4.0.4: code there, not used, activation patch needed
4.1.1: code there, not used, activation patch needed
Others: unknown.

> >As I said in an earlier post, I don't use buildroot to build my toolchains
> >(because I need toolchains, not a rootfs, and for other technical reasons).
> Fair enough, but let me point out that recently a patch went in that
> allows to use an external toolchain, fwiw ?).

Yep. I've seen (and not quite had time to test). But you still need a way to
actually build that external toolchain, no? :-)))

One of the limiting design with buildroot is that it mixes the destination
rootfs directory with the sys-root of the toolchain, thus requiring to either:
  - save the rootfs directory before fr-building a completely new rootfs (or
    the libc.so and the likes is lost),
  - re-build the full toolchain when oine 'accidentally' remove his/her rootfs.

> Could very well be that an external libfloat isn't needed anymore.

Could very well be that we still do... :-(

> Still, i, personally since i'm in the fortunate position to have an mmu
> as well as an FPU natively, have no idea how i could simulate
> appropriate targets, thus my plea for a kernel .config for an mmu-less,
> softfloat target that i can emulate with e.g. qemu :-/

I have none either. qemu emulates an MMU processor (for x86 and ARM at least).
I've tried yesterday evening to build qemu-bootable kernel and busybox-based
rootfs, but I failed miserably... :-( May be the WE will see it...

> We currently have in buildroot (short of /me having any test-target for
> [+-]mmu and/or float variants):
[--SNIP--]
> Help on that option can be obtained from invoke.texi from gcc or the
> respective online-docs somewhere at http://gcc.gnu.org/

Yep, I've seen that. I'm using --with-float and -mfpu, but not -mfloat-abi.
Will try it also.

> ?) Sounds like marketing mumble, really, but heck :)

Hehe! :-) Every one's trying to promote his tool! :-)

> friendly,
> Bernhard

Yours,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +0/33 662376056 | Software  Designer | \ / CAMPAIGN     |   ^                |
| --==< ?_? >==-- ?------------.-------:  X  AGAINST      |  /e\  There is no  |
| http://ymorin.is-a-geek.org/ | (*_*) | / \ HTML MAIL    |  """  conspiracy.  |
?------------------------------?-------?------------------?--------------------?

[Buildroot] building ARM toolchain with nofpu

[Bernhard Fischer]

On Fri, Feb 09, 2007 at 06:57:13PM +0100, Yann E. MORIN wrote:
>Bernhard,
>All,
>
>On Friday 09 February 2007 152, Bernhard Fischer wrote:
>> I'm curious as to why you think we don't have -lfloat?
>
>I din't say that either.

Well, IIRC one of your patches (for gcc-4.0.4 in this thread) removed
-lfloat which begs the above question :)

I simply do not know when gcc grew it's own float-fallbacks (and i don't
want to search for it, TBH).

>As I said in an earlier post, I don't use buildroot to build my toolchains
>(because I need toolchains, not a rootfs, and for other technical reasons).

Fair enough, but let me point out that recently a patch went in that
allows to use an external toolchain, fwiw ?).
>
>For reference, I'm using a highly modified version of crosstool, please see:
>  http://sources.redhat.com/ml/crossgcc/2007-02/msg00009.html
>
>> Could somebody please follow up on
>> http://www.busybox.net/lists/buildroot/2007-January/001182.html
>
>Quoted from the message referenced above:
>> Please edit package/libfloat/libfloat.mk
>> and remove the check for the gcc version 3.3 and let me know if that
>> fixed it.
>
>I'm gonna give it a try in the WE. The fact that having to link with an
>external library when gcc has the adequate code seems odd to me... :-/

Could very well be that an external libfloat isn't needed anymore.

Still, i, personally since i'm in the fortunate position to have an mmu
as well as an FPU natively, have no idea how i could simulate
appropriate targets, thus my plea for a kernel .config for an mmu-less,
softfloat target that i can emulate with e.g. qemu :-/


We currently have in buildroot (short of /me having any test-target for
[+-]mmu and/or float variants):

[quote from toolchain/gcc/Makefile.in]
ifeq ($(findstring 4.,$(GCC_VERSION)),4.)
SOFT_FLOAT_CONFIG_OPTION:=--with-float=soft
ifeq ($(BR2_SOFT_FLOAT_FP),y)
TARGET_SOFT_FLOAT:=-mfloat-abi=softfp
else # no fp at all
TARGET_SOFT_FLOAT:=-mfloat-abi=soft
endif
else # not gcc-4.x
TARGET_SOFT_FLOAT:=-msoft-float
endif
ARCH_FPU_SUFFIX:=_nofpu
else # no softfloat support
SOFT_FLOAT_CONFIG_OPTION:=
TARGET_SOFT_FLOAT:=
ARCH_FPU_SUFFIX:=
endif
[quote/]

Where currently "BR2_SOFT_FLOAT_FP" isn't set anywhere.
Help on that option can be obtained from invoke.texi from gcc or the
respective online-docs somewhere at http://gcc.gnu.org/

>Regards,
>Yann E. MORIN.

?) Sounds like marketing mumble, really, but heck :)

friendly,
Bernhard

[Buildroot] building ARM toolchain with nofpu

[Yann E. MORIN]

Bernhard,
All,

On Friday 09 February 2007 152, Bernhard Fischer wrote:
> I'm curious as to why you think we don't have -lfloat?

I din't say that either.
As I said in an earlier post, I don't use buildroot to build my toolchains
(because I need toolchains, not a rootfs, and for other technical reasons).

For reference, I'm using a highly modified version of crosstool, please see:
  http://sources.redhat.com/ml/crossgcc/2007-02/msg00009.html

> Could somebody please follow up on
> http://www.busybox.net/lists/buildroot/2007-January/001182.html

Quoted from the message referenced above:
> Please edit package/libfloat/libfloat.mk
> and remove the check for the gcc version 3.3 and let me know if that
> fixed it.

I'm gonna give it a try in the WE. The fact that having to link with an
external library when gcc has the adequate code seems odd to me... :-/

Regards,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +0/33 662376056 | Software  Designer | \ / CAMPAIGN     |   ^                |
| --==< ?_? >==-- ?------------.-------:  X  AGAINST      |  /e\  There is no  |
| http://ymorin.is-a-geek.org/ | (*_*) | / \ HTML MAIL    |  """  conspiracy.  |
?------------------------------?-------?------------------?--------------------?

[Buildroot] building ARM toolchain with nofpu

[James Newell]

 

> -----Original Message-----
> From: Bernhard Fischer [mailto:rep.dot.nop at gmail.com] 
> Sent: Friday, February 09, 2007 9:30 AM
> To: James Newell
> Cc: Yann E. MORIN; buildroot at uclibc.org
> Subject: Re: [Buildroot] building ARM toolchain with nofpu
> 
> On Fri, Feb 09, 2007 at 09:10:03AM -0500, James Newell wrote:
> > 
> >
> >> -----Original Message-----
> >> From: Bernhard Fischer [mailto:rep.dot.nop at gmail.com]
> >> Sent: Friday, February 09, 2007 8:11 AM
> >> To: James Newell
> >> Cc: Yann E. MORIN; buildroot at uclibc.org
> >> Subject: Re: [Buildroot] building ARM toolchain with nofpu
> >> 
> >> On Thu, Feb 08, 2007 at 08:41:53PM -0500, James Newell wrote:
> >> >The buildroot only seems to go to gcc 4.0.3? 
> >> 
> >> I'll add 4.0.4 later today (provided that it builds and 
> works), fwiw
> >> 
> >> 
> >
> >I was able to add 4.0.4 myself, and it worked as described by Yann.
> 
> I'm curious as to why you think we don't have -lfloat?
I didn't say that.  The 4.1.1 compiler doesn't seem to build the uclibc
with default soft-float enabled.

Did I miss an email?

> 
> Could somebody please follow up on
> http://www.busybox.net/lists/buildroot/2007-January/001182.html
> 
> 
> 

[Buildroot] building ARM toolchain with nofpu

[Bernhard Fischer]

On Fri, Feb 09, 2007 at 09:10:03AM -0500, James Newell wrote:
> 
>
>> -----Original Message-----
>> From: Bernhard Fischer [mailto:rep.dot.nop at gmail.com] 
>> Sent: Friday, February 09, 2007 8:11 AM
>> To: James Newell
>> Cc: Yann E. MORIN; buildroot at uclibc.org
>> Subject: Re: [Buildroot] building ARM toolchain with nofpu
>> 
>> On Thu, Feb 08, 2007 at 08:41:53PM -0500, James Newell wrote:
>> >The buildroot only seems to go to gcc 4.0.3? 
>> 
>> I'll add 4.0.4 later today (provided that it builds and works), fwiw
>> 
>> 
>
>I was able to add 4.0.4 myself, and it worked as described by Yann.

I'm curious as to why you think we don't have -lfloat?

Could somebody please follow up on
http://www.busybox.net/lists/buildroot/2007-January/001182.html

[Buildroot] building ARM toolchain with nofpu

[James Newell]

 

> -----Original Message-----
> From: Bernhard Fischer [mailto:rep.dot.nop at gmail.com] 
> Sent: Friday, February 09, 2007 8:11 AM
> To: James Newell
> Cc: Yann E. MORIN; buildroot at uclibc.org
> Subject: Re: [Buildroot] building ARM toolchain with nofpu
> 
> On Thu, Feb 08, 2007 at 08:41:53PM -0500, James Newell wrote:
> >The buildroot only seems to go to gcc 4.0.3? 
> 
> I'll add 4.0.4 later today (provided that it builds and works), fwiw
> 
> 

I was able to add 4.0.4 myself, and it worked as described by Yann.

Thanks,

Jim