[ECOS] Using a different toolchain

[ECOS] Using a different toolchain

[John R. Dunning]

Hi all.  I'm trying to find hints on how to get ecos to build using a
non-stock toolchain.  It's gcc, but not one of the canned ones that
comes with ecos.

First, is there a way to convince ecosconfig and friends to generate a
different tool prefix in the makefiles?  It seemed like it was
designed for that, but I couldn't find it.  If not, it's no big deal,
I can tweak the makefiles after generation.

Second, is there documentation on what sections and other
constructions are required in the linker script?  I tried using a
linker script from another (non-ecos) project, but that failed
miserably.  I have the impression that the script I'm using just
doesn't have enough in it, but I don't speak ld well enough to know
what I'm missing.  Do I just need to go read up on that stuff in order
to get my linker script right?

Any other hints or things I should look at?  TIA...


-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

Re: [ECOS] Using a different toolchain

[Andrew Lunn]

On Tue, Apr 04, 2006 at 09:05:44AM -0400, John R. Dunning wrote:
> Hi all.  I'm trying to find hints on how to get ecos to build using a
> non-stock toolchain.  It's gcc, but not one of the canned ones that
> comes with ecos.
> 
> First, is there a way to convince ecosconfig and friends to generate a
> different tool prefix in the makefiles?  It seemed like it was
> designed for that, but I couldn't find it.  If not, it's no big deal,
> I can tweak the makefiles after generation.

You want to change CYGBLD_GLOBAL_COMMAND_PREFIX

> Second, is there documentation on what sections and other
> constructions are required in the linker script?  I tried using a
> linker script from another (non-ecos) project, but that failed
> miserably.  I have the impression that the script I'm using just
> doesn't have enough in it, but I don't speak ld well enough to know
> what I'm missing.  Do I just need to go read up on that stuff in order
> to get my linker script right?

What is wrong with the linker script eCos generates? Since you are
still using gcc and i assume you are still using binutils, i would of
thought target.ld would work.

        Andrew

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

[ECOS] Using a different toolchain

[John R. Dunning]

    From: Andrew Lunn <andrew@lunn.ch>
    Date: Tue, 4 Apr 2006 15:16:47 +0200
    
    
    You want to change CYGBLD_GLOBAL_COMMAND_PREFIX

Ok, tnx.
    
    
    What is wrong with the linker script eCos generates? Since you are
    still using gcc and i assume you are still using binutils, 

Yes.
							       i would of
    thought target.ld would work.
    
Well, I dunno; I may just be not looking at the right thing.  I'm
trying to use what amounts to a 64-bit mips toolchain (it's a mips
instruction set, with some extra hacks added to it for custom
hardware) so I started with the 32-bit stuff, got that to build ok,
then tried swapping in the 64-bit stuff.  The 64-bit linker barfed on
the references to elf32-littlemips in target.ld, so I tried changing
them to what I thought were equivalent 64-bit things, but that didn't
work either.  I tried using the linker script that I use for other
(non-ecos) binaries on this platform, and that said it built an
executable, but when I objdump it, there's nothing in .text.

Perhaps I need to go off and read up on linker scripts, because my
non-ecos one bears approximately zero resemblance to the ecos one, and
I can intuit only vaguely what either one of them are doing :-{


-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

[ECOS] Re: Using a different toolchain

[Grant Edwards]

In gmane.os.ecos.general, you wrote:

> Hi all.  I'm trying to find hints on how to get ecos to build using a
> non-stock toolchain.

There's no trick to it: you just do it.

> It's gcc, but not one of the canned ones that comes with ecos.

I've never used a "canned" toolkit.

> First, is there a way to convince ecosconfig and friends to
> generate a different tool prefix in the makefiles?

You mean different filenames for gcc et al?

It's probably easiest to just create symlinks so that the
default filenames work.

> It seemed like it was designed for that, but I couldn't find
> it.  If not, it's no big deal, I can tweak the makefiles after
> generation.
>
> Second, is there documentation on what sections and other
> constructions are required in the linker script?  I tried
> using a linker script from another (non-ecos) project,

Again, why?

> but that failed miserably.

In what way?

> I have the impression that the script I'm using just doesn't
> have enough in it, but I don't speak ld well enough to know
> what I'm missing.  Do I just need to go read up on that stuff
> in order to get my linker script right?

Either that or tell us how it failed.

-- 
Grant Edwards                   grante             Yow!  I feel better about
                                  at               world problems now!
                               visi.com            

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

[ECOS] Using a different toolchain

[John R. Dunning]

    From: "John R. Dunning" <jrd@jrd.org>
    Date: Tue, 4 Apr 2006 10:11:53 -0400
    
        From: Andrew Lunn <andrew@lunn.ch>
        Date: Tue, 4 Apr 2006 15:16:47 +0200
    							       i would of
        thought target.ld would work.

Further info, I went back and started over from scratch.  When I use
an out-of-the box config, using one of the mips targets, everything
builds correctly.  I then changed toolchains, by fixing the tool
prefix so as to invoke my 64bit-plus-special-sauce toolchain.

Here's some basic info on the toolchain:

    jrd@test ~/sw/ecos/devo/hack $ xxgcc --version
    mips64el-unknown-linux-gnu-gcc (GCC) 3.4.5 (Gentoo 3.4.5-r1, ssp-3.4.5-1.0, pie-8.7.9)
    Copyright (C) 2004 Free Software Foundation, Inc.
    This is free software; see the source for copying conditions.  There is NO
    warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

Cleaned, and compiled everything.  That works ok, but when I get to
the link step, trying to build one of the test executables, I get the
following:

    xxgcc -c  -I/net/home/jrd/sw/ecos/devo/hack/install/include -I/home/jrd/sw/ecos/dist/ecos-2.0/packages/hal/common/v2_0 -I/home/jrd/sw/ecos/dist/ecos-2.0/packages/hal/common/v2_0/src -I/home/jrd/sw/ecos/dist/ecos-2.0/packages/hal/common/v2_0/tests -I. -I/home/jrd/sw/ecos/dist/ecos-2.0/packages/hal/common/v2_0/tests/ -mips64 -EL -msoft-float -Wall -Wpointer-arith -Wstrict-prototypes -Winline -Wundef -Woverloaded-virtual -g -O2 -ffunction-sections -fdata-sections -fno-rtti -fno-exceptions -fvtable-gc -G0 -Wp,-MD,tests/context.tmp -o tests/context.o /home/jrd/sw/ecos/dist/ecos-2.0/packages/hal/common/v2_0/tests/context.c
    cc1: warning: command line option "-Woverloaded-virtual" is valid for C++/ObjC++ but not for C
    cc1: warning: command line option "-fno-rtti" is valid for C++/ObjC++ but not for C
    cc1: warning: command line option "-fvtable-gc" is valid for C++/ObjC++ but not for C
    xxgcc -EL -msoft-float -g -nostdlib -Wl,--gc-sections -Wl,-static -L/net/home/jrd/sw/ecos/devo/hack/install/lib -Ttarget.ld -o /net/home/jrd/sw/ecos/devo/hack/install/tests/hal/common/v2_0/tests/context tests/context.o
    /usr/lib/gcc/mips64el-unknown-linux-gnu/3.4.5/../../../../mips64el-unknown-linux-gnu/bin/ld: target elf32-littlemips not found
    collect2: ld returned 1 exit status
    make[1]: *** [/net/home/jrd/sw/ecos/devo/hack/install/tests/hal/common/v2_0/tests/context] Error 1
    make[1]: Leaving directory `/net/home/jrd/sw/ecos/devo/hack/hal/common/v2_0'
    make: *** [tests] Error 2

I tried a few tweaks to target.ld, primarily changing the
OUTPUT_FORMAT spec to some 64bit values (elf64-tradbigmips,
elf64-tradlittlemips) which I cribbed from another (much more
complicated, but working) linker script that's used by gcc compiling
stuff for this target.  Using that modified target.ld, I got:

    xxgcc -EL -msoft-float -g -nostdlib -Wl,--gc-sections -Wl,-static -L/net/home/jrd/sw/ecos/devo/hack/install/lib -Ttarget.ld -o /net/home/jrd/sw/ecos/devo/hack/install/tests/hal/common/v2_0/tests/context tests/context.o
    /usr/lib/gcc/mips64el-unknown-linux-gnu/3.4.5/../../../../mips64el-unknown-linux-gnu/bin/ld: /net/home/jrd/sw/ecos/devo/hack/install/lib/vectors.o: ABI is incompatible with that of the selected emulation

... followed by lots more of that error.

So, back to the original question: Should I be able to swap in a
not-shipped-with-ecos toolchain, and come up with a way to get it to
compile and link?  The compilation part appears to be under control,
but I'm still not at all clear on what I need to do make my 64-bit
linker sufficiently happy to produce an elf for me.

As before, any info appreciated.


-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

[ECOS] Re: Using a different toolchain

[Grant Edwards]

On 2006-04-06, John R. Dunning <jrd@jrd.org> wrote:

> So, back to the original question: Should I be able to swap in a
> not-shipped-with-ecos toolchain, and come up with a way to get it to
> compile and link?

Certainly. I've never used a "shipped-with-ecos-toolchain".

> The compilation part appears to be under control, but I'm
> still not at all clear on what I need to do make my 64-bit
> linker sufficiently happy to produce an elf for me.

The linker scripts generated by eCos have always worked for my
(32-bit) "home-built" toolchains.

-- 
Grant Edwards                   grante             Yow!  Your CHEEKS sit like
                                  at               twin NECTARINES above
                               visi.com            a MOUTH that knows no
                                                   BOUNDS --


-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

Re: [ECOS] Using a different toolchain

[Andrew Lunn]

On Wed, Apr 05, 2006 at 11:11:17PM -0400, John R. Dunning wrote:
>     From: "John R. Dunning" <jrd@jrd.org>
>     Date: Tue, 4 Apr 2006 10:11:53 -0400
>     
>         From: Andrew Lunn <andrew@lunn.ch>
>         Date: Tue, 4 Apr 2006 15:16:47 +0200
>     							       i would of
>         thought target.ld would work.
> 
> Further info, I went back and started over from scratch.  When I use
> an out-of-the box config, using one of the mips targets, everything
> builds correctly.  I then changed toolchains, by fixing the tool
> prefix so as to invoke my 64bit-plus-special-sauce toolchain.
> 
> Here's some basic info on the toolchain:
> 
>     jrd@test ~/sw/ecos/devo/hack $ xxgcc --version
>     mips64el-unknown-linux-gnu-gcc (GCC) 3.4.5 (Gentoo 3.4.5-r1, ssp-3.4.5-1.0, pie-8.7.9)
>     Copyright (C) 2004 Free Software Foundation, Inc.
>     This is free software; see the source for copying conditions.  There is NO
>     warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
> 
> Cleaned, and compiled everything.  That works ok, but when I get to
> the link step, trying to build one of the test executables, I get the
> following:
> 
>     xxgcc -c  -I/net/home/jrd/sw/ecos/devo/hack/install/include -I/home/jrd/sw/ecos/dist/ecos-2.0/packages/hal/common/v2_0 -I/home/jrd/sw/ecos/dist/ecos-2.0/packages/hal/common/v2_0/src -I/home/jrd/sw/ecos/dist/ecos-2.0/packages/hal/common/v2_0/tests -I. -I/home/jrd/sw/ecos/dist/ecos-2.0/packages/hal/common/v2_0/tests/ -mips64 -EL -msoft-float -Wall -Wpointer-arith -Wstrict-prototypes -Winline -Wundef -Woverloaded-virtual -g -O2 -ffunction-sections -fdata-sections -fno-rtti -fno-exceptions -fvtable-gc -G0 -Wp,-MD,tests/context.tmp -o tests/context.o /home/jrd/sw/ecos/dist/ecos-2.0/packages/hal/common/v2_0/tests/context.c
>     cc1: warning: command line option "-Woverloaded-virtual" is valid for C++/ObjC++ but not for C
>     cc1: warning: command line option "-fno-rtti" is valid for C++/ObjC++ but not for C
>     cc1: warning: command line option "-fvtable-gc" is valid for C++/ObjC++ but not for C
>     xxgcc -EL -msoft-float -g -nostdlib -Wl,--gc-sections -Wl,-static -L/net/home/jrd/sw/ecos/devo/hack/install/lib -Ttarget.ld -o /net/home/jrd/sw/ecos/devo/hack/install/tests/hal/common/v2_0/tests/context tests/context.o
>     /usr/lib/gcc/mips64el-unknown-linux-gnu/3.4.5/../../../../mips64el-unknown-linux-gnu/bin/ld: target elf32-littlemips not found

Try -V so we can see what is being passed from gcc to ld. 

Have you been able to compile and link anything with the toolchain?
Are there any build in self tests? 

        Andrew

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss