[Clfs-support] [CLFS-DEV] Embedded GMP builds 64-bit binaries when asked for a 32-bit build

[Andrew Bradford]

On Mon, Jul 1, 2013, at 09:07 AM, Kirk Terrell wrote:
> On 07/01/2013 04:32 AM, Andrew Bradford wrote:
> > On Sat, Jun 29, 2013, at 10:33 AM, Kirk Terrell wrote:
> >> While trying to build a system using the latest version of the embedded
> >> GIT I ran into an issue at the beginning with the GMP build - I'ver
> >> tried twice. One time GMP compiled at the problem didn't manifest until
> >> gcc-static, where the it would build with the 64bit versions of GMP, MPC
> >> and MPFR. Based on some of the research i did it looks like GMP does a
> >> poor job of guessing processor, which leads to the wrong ABI. I don't
> >> know what i did different but the build failed, the following is a
> >> partial listing of the output:
> >>
> >>
> >> checking build system type... atom-pc-linux-gnu
> >> checking host system type... atom-pc-linux-gnu
> >
> > Is 'atom-pc-linux-gnu' a valid triplet?
> > If nothing else, I would think an embedded book build should be using
> > '-uclibc' as the ending portion of the triplet.  Unless I'm
> > misunderstanding something.
> >
> > If you are using 'atom-pc-linux-gnu', can you reproduce the issue using
> > a triplet of 'i686-pc-linux-uclibc' and report back?
> >
> >> checking for a BSD-compatible install... /usr/bin/install -c
> >> checking whether build environment is sane... yes
> >> checking for a thread-safe mkdir -p... /bin/mkdir -p
> >> checking for gawk... gawk
> >> checking whether make sets $(MAKE)... yes
> >> checking whether to enable maintainer-specific portions of Makefiles...
> >> no
> >> checking ABI=64
> >> checking compiler gcc -O2 -pedantic -fomit-frame-pointer -m64
> >> -fexceptions... yes
> >> checking compiler gcc -O2 -pedantic -fomit-frame-pointer -m64
> >> -fexceptions -mtune=atom... yes
> >> checking compiler gcc -O2 -pedantic -fomit-frame-pointer -m64
> >> -mtune=atom -fexceptions -march=atom... yes
> >> checking for gcc... gcc
> >> -------------------------
> >> -------------------------
> >> ../fib_table.h:4:1: warning: data definition has no type or storage
> >> class [enabled by default]
> >> ../fib_table.h:4:1: warning: type defaults to 'int' in declaration of
> >> 'Error' [enabled by default]
> >> ../fib_table.h:4:8: warning: type defaults to 'int' in declaration of
> >> 'error' [enabled by default]
> >> ../fib_table.h:4:20: error: expected '=', ',', ';', 'asm' or
> >> '__attribute__' before 'data'
> >> make[4]: *** [amd64check.lo] Error 1
> >> make[4]: Leaving directory
> >> `/opt/elfs20130613B.arm.little/sources/gmp-5.1.2/tests'
> >> make[3]: *** [check-am] Error 2
> >> make[3]: Leaving directory
> >> `/opt/elfs20130613B.arm.little/sources/gmp-5.1.2/tests'
> >> make[2]: *** [check-recursive] Error 1
> >> make[2]: Leaving directory
> >> `/opt/elfs20130613B.arm.little/sources/gmp-5.1.2/tests'
> >> make[1]: *** [check-recursive] Error 1
> >> make[1]: Leaving directory
> >> `/opt/elfs20130613B.arm.little/sources/gmp-5.1.2'
> >> make: *** [check] Error 2
> >> ------------------------------
> >> ----------------------------
> >> /usr/bin/ld: i386 architecture of input file `rand/.libs/randbui.o' is
> >> incompatible with i386:x86-64 output
> >> /usr/bin/ld: i386 architecture of input file `rand/.libs/randmui.o' is
> >> incompatible with i386:x86-64 output
> >> mpn/.libs/toom_interpolate_6pts.o: In function
> >> `__gmpn_toom_interpolate_6pts':
> >> toom_interpolate_6pts.c:(.text+0x2bc): undefined reference to
> >> `__gmpn_sublsh2_n_ip1'
> >> mpn/.libs/toom_interpolate_8pts.o: In function
> >> `__gmpn_toom_interpolate_8pts':
> >> toom_interpolate_8pts.c:(.text+0x3ca): undefined reference to
> >> `__gmpn_sublsh2_n_ip1'
> >> collect2: ld returned 1 exit status
> >> make[2]: *** [libgmp.la] Error 1
> >> make[2]: Leaving directory
> >> `/opt/elfs20130613B.arm.little/sources/gmp-5.1.2'
> >> make[1]: *** [install-recursive] Error 1
> >> make[1]: Leaving directory
> >> `/opt/elfs20130613B.arm.little/sources/gmp-5.1.2'
> >> make: *** [install] Error 2
> >> exitCode=2
> >>
> >>
> >> ---------------------------
> >>
> >> Some references that might be useful:
> >>
> >> http://gmplib.org/manual/Build-Options.html
> >>
> >> http://lists.gnu.org/archive/html/bug-standards/2012-01/msg00001.html
> >>
> >> http://lists.cross-lfs.org/htdig.cgi/clfs-support-cross-lfs.org/2009-September/000746.html
> >>
> >> --------------------
> >>
> >> i've built succesfully by seting ABI=32, not sure if this is the best
> >> solution.
> >
> > Based on [1] it seems it is.  And, somewhat good timing to catch this,
> > as armv8 is 64 bit now and we'll start seeing those real systems out
> > there, too :)
> >
> > [1]:http://gmplib.org/manual/ABI-and-ISA.html#ABI-and-ISA
> >
> > I've opened a bug [2] for this.
> >
> > [2]:http://trac.cross-lfs.org/ticket/925
> >
> I've built a complete system for vexpress-a9 running on QEMU using the 
> book w/ the following three exceptions:
> 
> 1)      added the --build=${CLFS_HOST} option when configuring GMP. For 
> tracker 925 - I'm not setting the triplet, GMP is guessing the triplet 
> for the host/build machine. It looks like some newer atom processors are 
> 64 bit, mine is too old to be one of those.             

Ah, sorry, I should drink coffee *before* emailing :)
Now I see that GMP is picking build and host, not target.

Are you building on a multilib distribution?  I'm curious why, if you've
only got 32 bit abilities on your host, that GMP is even able to think
you have 64 bit.  Or am I still not understanding this issue?

> 2)      Did not execute the second stage for gcc - I think I only need
> the 
> second stage for building C++, or other, compilers?

You'll need gcc-final if you want dynamic linking, I believe, plus for
GCC to know to use uClibc vs just the built-in newlib.
You should always build gcc-final.

> 3)      Deselected networking/nslookup in busybox .
> 
> for the build I used the following parameters:
> 
> export CLFS_ARCH=arm
> export CLFS_ABI="aapcs-linux"
> export CLFS_HOST="i686-cross-linux-gnu"
> export CLFS_ENDIAN="little"
> export CLFS_ARM_MODE="arm"
> export CLFS_ARM_ARCH="armv7-a"
> export CLFS_FLOAT="soft"
> export CLFS_TARGET="armel-unknown-linux-uclibceabi"
> export CC="armel-unknown-linux-uclibceabi-gcc"
> 
> 
> I did not need the soft float patch ( Ticket #919)
> I did not have an issues w/ the .size directive ( Ticket #876/ Tracker 
> #917)

OK, good to know! :)

> Here is an excerpt from the busybox output for the issues w/ nslookup.
> 
> networking/lib.a(nslookup.o): In function `nslookup_main':
> nslookup.c:(.text.nslookup_main+0x2c): undefined reference to
> `__res_init'
> nslookup.c:(.text.nslookup_main+0x54): undefined reference to
> `__res_state'
> nslookup.c:(.text.nslookup_main+0x78): undefined reference to
> `__res_state'
> nslookup.c:(.text.nslookup_main+0x8c): undefined reference to
> `__res_state'
> collect2: error: ld returned 1 exit status
> make: *** [busybox_unstripped] Error 1
> 
> I have not had a chance to see if a tweak to the uClibc config file 
> would address this.

Is this without building gcc-final?  If so, I expect all sort of issues
without building a gcc-final.

-Andrew