seabios version 0.6.1.2-0ubuntu1 failed to build on i386

[Matthias Klose]

seabios version 0.6.1.2-0ubuntu1 failed to build on i386
Link to failed build: https://launchpad.net/ubuntu/+archive/test-rebuild-20110329/+buildjob/2410113

Direct link to the build log: https://launchpad.net/ubuntu/+archive/test-rebuild-20110329/+buildjob/2410113/+files/buildlog_ubuntu-natty-i386.seabios_0.6.1.2-0ubuntu1_FAILEDTOBUILD.txt.gz

This log snippet might be of interest, since it triggered the matcher 'Purging chroot-autobuild'.
Excerpt 795 lines into the build log:

  Generating offset file out/asm-offsets.h
  Compiling (16bit) out/code16.o
  Compiling whole program out/ccode32flat.o
  Compiling whole program out/code32seg.o
  Building ld scripts (version "0.6.1.2-20110408_190953-actinium")
Fixed space: 0xe05b-0x10000 total: 8101 slack: 3 Percent slack: 0.0%
16bit size: 38896
32bit segmented size: 1636
32bit flat size: 64828
  Linking out/rom16.o
out/romlayout16.lds:695 cannot move location counter backwards (from 000000000000c9fa to 000000000000c9e0)
make[1]: *** [out/rom16.o] Error 1
make[1]: Leaving directory `/build/buildd/seabios-0.6.1.2'
make: *** [build-stamp] Error 2
dpkg-buildpackage: error: debian/rules build gave error exit status 2
******************************************************************************
Build finished at 20110408-1909
FAILED [dpkg-buildpackage died]
Purging chroot-autobuild/build/buildd/seabios-0.6.1.2

[Zero (wuwanli)]

Also failed on amd64 platform, the same error.

[Rguenth (rguenth)]

Created attachment 5702
testcase

While the 2.21 release links the attached testcase just fine, the 2.21 branch
taken at Apr 19th fails to link it with

> ld -T romlayout32seg.lds code32seg.o -o rom32seg.o
romlayout32seg.lds:66 cannot move location counter backwards (from 00000000000067b0 to 00000000000067a0)

If one changes the linker script to add 0x4 more room for the sections
(which corresponds to their alignment) like

*(.rodata.__func__.10274)
. = ( 0x6782 - code32seg_start ) ;
*(.rodata.__func__.8035)
. = ( 0x6796 - code32seg_start ) ;
*(.rodata.__func__.8048)
. = ( 0x67aa - code32seg_start ) ;
*(.rodata.__func__.8075)
. = ( 0x67c0 - code32seg_start ) ;
*(.rodata.__func__.8115)

then the link succeeds. This looks like a bogus alignment calculation
as with the size of 0xe for the section 0x10 should be enough space
to properly align the data.

[Rguenth (rguenth)]

Similar issue happens on 64bit.

[Rguenth (rguenth)]

Confirmed also with the current 2.21 branch head and also with

2010-12-04 Alan Modra <email address hidden>

        * ldlang.c (lang_size_sections_1): Align lma using same alignment
        as for vma.

reverted (that just looked maybe related).

[Alan Modra (amodra-gmail)]

Mine. Looks to be caused by the fix for PR12380. That delayed setting expld.phase = lang_final_phase_enum until just before the final lang_do_assignments, breaking the assignments done in lang_relax_sections.

[Cvs-commit (cvs-commit)]

CVSROOT: /cvs/src
Module name: src
Changes by: <email address hidden> 2011-05-03 14:56:14

Modified files:
 ld : ChangeLog ldexp.c ldexp.h ldlang.c ldlang.h
                  pe-dll.c

Log message:
 PR ld/12726
 * ldexp.h (lang_phase_type): Add lang_assigning_phase_enum.
 * ldexp.c (exp_fold_tree_1): Correct assign to dot comment. Don't
 assign to dot when lang_assigning_phase_enum.
 * ldlang.h (lang_do_assignments): Update prototype.
 * ldlang.c (lang_do_assignments): Add phase parameter. Update all
 callers.
 * pe-dll.c (pe_dll_fill_sections, pe_exe_fill_sections): Update
 lang_do_assignments calls.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/ChangeLog.diff?cvsroot=src&r1=1.2324&r2=1.2325
http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/ldexp.c.diff?cvsroot=src&r1=1.93&r2=1.94
http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/ldexp.h.diff?cvsroot=src&r1=1.28&r2=1.29
http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/ldlang.c.diff?cvsroot=src&r1=1.366&r2=1.367
http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/ldlang.h.diff?cvsroot=src&r1=1.96&r2=1.97
http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/pe-dll.c.diff?cvsroot=src&r1=1.136&r2=1.137

[Cvs-commit (cvs-commit)]

CVSROOT: /cvs/src
Module name: src
Branch: binutils-2_21-branch
Changes by: <email address hidden> 2011-05-03 15:16:40

Modified files:
 ld : ChangeLog ldexp.c ldexp.h ldlang.c ldlang.h
                  pe-dll.c

Log message:
 PR ld/12726
 * ldexp.h (lang_phase_type): Add lang_assigning_phase_enum.
 * ldexp.c (exp_fold_tree_1): Correct assign to dot comment. Don't
 assign to dot when lang_assigning_phase_enum.
 * ldlang.h (lang_do_assignments): Update prototype.
 * ldlang.c (lang_do_assignments): Add phase parameter. Update all
 callers.
 * pe-dll.c (pe_dll_fill_sections, pe_exe_fill_sections): Update
 lang_do_assignments calls.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/ChangeLog.diff?cvsroot=src&only_with_tag=binutils-2_21-branch&r1=1.2222.2.20&r2=1.2222.2.21
http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/ldexp.c.diff?cvsroot=src&only_with_tag=binutils-2_21-branch&r1=1.87.2.1&r2=1.87.2.2
http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/ldexp.h.diff?cvsroot=src&only_with_tag=binutils-2_21-branch&r1=1.23.2.1&r2=1.23.2.2
http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/ldlang.c.diff?cvsroot=src&only_with_tag=binutils-2_21-branch&r1=1.347.2.3&r2=1.347.2.4
http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/ldlang.h.diff?cvsroot=src&only_with_tag=binutils-2_21-branch&r1=1.91.2.2&r2=1.91.2.3
http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/pe-dll.c.diff?cvsroot=src&only_with_tag=binutils-2_21-branch&r1=1.133&r2=1.133.2.1

[Alan Modra (amodra-gmail)]

Fixed

[tolik2525 (tolik2525)]

It seems like this is binutils bug.
I reinstalled binutils from this snapshot ftp://sourceware.org/pub/binutils/snapshots/binutils-2.20.51.tar.bz2 and all went fine.

go@02-utest02:~/seabios-0.6.1.2$ ld -v
GNU ld (GNU Binutils) 2.20.51.20101105

go@02-utest02:~/seabios-0.6.1.2$ make
  Compiling whole program out/ccode.16.s
  Compiling to assembler out/asm-offsets.s
  Generating offset file out/asm-offsets.h
  Compiling (16bit) out/code16.o
  Compiling whole program out/ccode32flat.o
  Compiling whole program out/code32seg.o
  Building ld scripts (version "0.6.1.2-20110510_133622-02-utest02")
Fixed space: 0xe05b-0x10000 total: 8101 slack: 17 Percent slack: 0.2%
16bit size: 38928
32bit segmented size: 1464
32bit flat size: 66904
  Linking out/rom16.o
  Stripping out/rom16.strip.o
  Linking out/rom32seg.o
  Stripping out/rom32seg.strip.o
  Linking out/rom.o
  Prepping out/bios.bin
Total size: 107296 Free space: 23776 Percent used: 81.9% (128KiB rom)
go@02-utest02:~/seabios-0.6.1.2$

[tolik2525 (tolik2525)]

Forgot to mention I'm using amd64 version of Ubuntu 10.10

[Serge Hallyn (serge-hallyn)]

Reverting binutils to bzr commit 133 lets seabios compile. 135 does not. Have yet to test revision 134.

[Serge Hallyn (serge-hallyn)]

Revision 134 is also good. So something in revision 135 stops seabios from compiling.

[Matthias Klose (doko)]

r135 changes:
* Pass --enable-targets=all to configure in the multiarch pass;
  Debian #610745.

would it be possible to revert this change in a current binutils in natty/oneiric and recheck? I however cannot see any b-d on binutils-multiarch in seabios

[Loïc Minier (lool)]

Note that --enable-targets=all was later changed to use the list of dpkg linux triplets + a hardcoded list of triplets which was accumulated in binutils over time.

[Loïc Minier (lool)]

Looks like http://sourceware.org/bugzilla/show_bug.cgi?id=12066

[Loïc Minier (lool)]

There was another amd64-specific FTBFS which I uploaded a fix for (see Debian #625000); then I get the same issue as described here on amd64. We should give back the package when we get a new binutils snapshot with the fix.

[Loïc Minier (lool)]

Might also be http://sourceware.org/bugzilla/show_bug.cgi?id=12726

this is due to alignment changes which were now reverted

[Matthias Klose (doko)]

binutils fixed in oneiric, seabios built sucessfully