2.35.50 breaks ld -no-pie

[Christian Ehrhardt ]

The qemu build reaches (and always did) a step where it tries to link some
img files. That is done via the command:
  $ ld -m elf_i386 -T /<<PKGBUILDDIR>>/pc-bios/optionrom//flat.lds -no-pie -s -o multiboot.img multiboot.o

Recently that still works in Debian [1] but no more in Ubuntu [2].

I think that the new binutils broke me.
In hirsute proposed those are at 2.35.50.20201210-0ubuntu1

The issue is easily isolated, and by copying the two files around I found the following:

Hirsute: 2.35.50.20201210-0ubuntu1 - bad
Hirsute: 2.35.50.20201207-0ubuntu1 - bad
Sid: 2.35.1-4 - good
Groovy: 2.35.1-1ubuntu1 - good
Focal: 2.34-6ubuntu1 - good

I'll attach these two files to the bug, just thro them into a directory and
run the command:
 $ ld -m elf_i386 -T ./flat.lds -no-pie -s -o linuxboot.img linuxboot.o

If that is an intentional change please guide how this is now supposed to work.

[1]: https://buildd.debian.org/status/fetch.php?pkg=qemu&arch=amd64&ver=1%3A5.2%2Bdfsg-2&stamp=1607598738&raw=1
[2]: https://launchpadlibrarian.net/510801929/buildlog_ubuntu-hirsute-amd64.qemu_1%3A5.2+dfsg-2ubuntu1~ppa2_BUILD

[Matthias Klose (doko)]

works with the 2.35 branch, fails with the trunk:

$ ld.gold -pie
ld.gold: fatal error: no input files

$ ld.gold -no-pie
ld.gold: fatal error: no input files

$ ld.bfd -pie
ld.bfd: no input files

$ ld.bfd -no-pie
ld.bfd: Error: unable to disambiguate: -no-pie (did you mean --no-pie ?)

[Hjl-tools (hjl-tools)]

It is caused by

commit 983d925db6a09ac90f6bed90be16eb69267b58e0
Author: Nick Clifton <email address hidden>
Date: Mon Oct 5 13:53:59 2020 +0100

    Update the BFD linker so that it deprecates grouped short options.

            * lexsup.c (parse_args): Generate an error or warning message when
            multiple short options are used together.

[Hjl-tools (hjl-tools)]

$ valgrind ./ld/ld-new -no-pie
==290969== Memcheck, a memory error detector
==290969== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==290969== Using Valgrind-3.16.1 and LibVEX; rerun with -h for copyright info
==290969== Command: ./ld/ld-new -no-pie
==290969==
==290969== Conditional jump or move depends on uninitialised value(s)
==290969== at 0x409A96: parse_args (lexsup.c:749)
==290969== by 0x403DC6: main (ldmain.c:370)
==290969==
./ld/ld-new: Error: unable to disambiguate: -no-pie (did you mean --no-pie ?)
==290969==
==290969== HEAP SUMMARY:
==290969== in use at exit: 21,161 bytes in 10 blocks
==290969== total heap usage: 72 allocs, 62 frees, 43,232 bytes allocated
==290969==
==290969== LEAK SUMMARY:
==290969== definitely lost: 0 bytes in 0 blocks
==290969== indirectly lost: 0 bytes in 0 blocks
==290969== possibly lost: 0 bytes in 0 blocks
==290969== still reachable: 21,161 bytes in 10 blocks
==290969== suppressed: 0 bytes in 0 blocks
==290969== Rerun with --leak-check=full to see details of leaked memory
==290969==
==290969== Use --track-origins=yes to see where uninitialised values come from
==290969== For lists of detected and suppressed errors, rerun with: -s
==290969== ERROR SUMMARY: 2 errors from 1 contexts (suppressed: 0 from 0)

[Hjl-tools (hjl-tools)]

longind may not be set by getopt_long_only:

diff --git a/ld/lexsup.c b/ld/lexsup.c
index 0d10bc6fba..60e3f7d8b5 100644
--- a/ld/lexsup.c
+++ b/ld/lexsup.c
@@ -735,6 +735,7 @@ parse_args (unsigned argc, char **argv)
       /* getopt_long_only is like getopt_long, but '-' as well as '--'
    can indicate a long option. */
       opterr = 0;
+ longind = -1;
       last_optind = optind;
       optc = getopt_long_only (argc, argv, shortopts, longopts, &longind);
       if (optc == '?')

[Hjl-tools (hjl-tools)]

   LONGIND returns the index in LONGOPT of the long-named option found.
   It is only valid when a long-named option has been found by the most
   recent call.

Since the long option isn't found, LONGIND is undefined.

[Nickc (nickc)]

    Not to be facetious but did we ever really support -no-pie ?

    It does not appear to be documented, and it looks like it is treated internally as two options: -n and -o-pie.

    I have a patch that fixes the uninitialised memory problem detected by valgrind, but I am wondering whether it is better to add explicit support for -no-pie as an option to turn off -pie, or else keep the linker's error message, and maybe add a --no-pie option instead.

[Hjl-tools (hjl-tools)]

(In reply to Nick Clifton from comment #5)
> Not to be facetious but did we ever really support -no-pie ?
>

ld/testsuite/config/default.exp: set NOPIE_LDFLAGS "-no-pie"

[Matthias Klose (doko)]

that was reported for qemu:

./configure:# Check we support --no-pie first; we will need this for building ROMs.
./configure:if compile_prog "-Werror -fno-pie" "-no-pie"; then
./configure: LDFLAGS_NOPIE="-no-pie"

[Andreas Schwab (schwab-linux-m68k)]

That's the *compiler* flag.

[Cvs-commit (cvs-commit)]

The master branch has been updated by Nick Clifton <email address hidden>:

https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=3991c7acb29aa8d7d52150695eb3efa03a08dd50

commit 3991c7acb29aa8d7d52150695eb3efa03a08dd50
Author: Nick Clifton <email address hidden>
Date: Mon Dec 14 12:40:13 2020 +0000

    Fix a use of an uninitialised variable in the bfd linker.

            PR 27050
            * lexsup.c (parse_args): Ensure that the longind local variable is
            set.

[Nickc (nickc)]

Right - I have checked in the patch to fix the use of an unitialised variable.

Frankly I do not see any need to go any further. The linker does not have a -no-pie or a --no-pie command line option, so the error message is correct. Hence I am going to close this PR. If anyone has any strong objections they can reopen it and tell us what they think.

Cheers
  Nick

[Christian Ehrhardt  (paelzer)]

FYI - fix submitted to qemu upstream and for now added to the qemu package. If upstream eventually prefers a different solution I can refresh it accordingly.
=> https://lists.gnu.org/archive/html/qemu-devel/2020-12/msg03684.html

[Launchpad Janitor (janitor)]

This bug was fixed in the package binutils - 2.35.50.20201210-0ubuntu2

---------------
binutils (2.35.50.20201210-0ubuntu2) hirsute; urgency=medium

  * Revert linker options parsing patch. LP: #1907789.

 -- Matthias Klose <email address hidden> Fri, 11 Dec 2020 11:29:47 +0100

[Launchpad Janitor (janitor)]

This bug was fixed in the package qemu - 1:5.2+dfsg-2ubuntu1

---------------
qemu (1:5.2+dfsg-2ubuntu1) hirsute; urgency=medium

  * Merge with Debian unstable
    - includes fix for CVE-2020-17380
    - includes a fix for s390x PCI device reset (LP: #1907656)
    Remaining changes:
    - qemu-kvm to systemd unit
      - d/qemu-kvm-init: script for QEMU KVM preparation modules, ksm,
        hugepages and architecture specifics
      - d/qemu-system-common.qemu-kvm.service: systemd unit to call
        qemu-kvm-init
      - d/qemu-system-common.install: install helper script
      - d/qemu-system-common.qemu-kvm.default: defaults for
        /etc/default/qemu-kvm
      - d/rules: call dh_installinit and dh_installsystemd for qemu-kvm
    - Distribution specific machine type (LP: 1304107 1621042)
      - d/p/ubuntu/define-ubuntu-machine-types.patch: distro machine types
      - d/qemu-system-x86.NEWS Info on fixed machine type definitions
        for host-phys-bits=true (LP: 1776189)
      - add an info about -hpb machine type in debian/qemu-system-x86.NEWS
      - provide pseries-bionic-2.11-sxxm type as convenience with all
        meltdown/spectre workarounds enabled by default. (LP: 1761372).
      - ubuntu-q35 alias added to auto-select the most recent q35 ubuntu type
    - Enable nesting by default
      - d/p/ubuntu/enable-svm-by-default.patch: Enable nested svm by default
        in qemu64 on amd
        [ No more strictly needed, but required for backward compatibility ]
    - improved dependencies
      - Make qemu-system-common depend on qemu-block-extra
      - Make qemu-utils depend on qemu-block-extra
      - let qemu-utils recommend sharutils
    - tolerate ipxe size change on migrations to >=18.04 (LP: 1713490)
      - d/p/ubuntu/pre-bionic-256k-ipxe-efi-roms.patch: old machine types
        reference 256k path
      - d/control-in: depend on ipxe-qemu-256k-compat-efi-roms to be able to
        handle incoming migrations from former releases.
    - d/control-in: Disable capstone disassembler library support (universe)
    - d/qemu-system-x86.README.Debian: add info about updated nesting changes
    - d/control*, d/rules: disable xen by default, but provide universe
      package qemu-system-x86-xen as alternative
      [includes compat links changes of 5.0-5ubuntu4]
    - allow qemu to load old modules post upgrade (LP 1847361)
      - Drop d/qemu-block-extra.*.in, d/qemu-system-gui.*.in
      - d/rules: Drop generating package version into maintainer scripts
      - d/qemu-system-gui.prerm: add no-op prerm to overcome upgrade issues on
        the bad old prerm (LP 1906245 1905377)
  * Dropped Changes:
    - d/control, d/rules: build with gcc-9 on armhf as workaround until
      resolved in gcc-10 (LP: 1890435) [it is flaky still, but no more 100%
      fails]
  * Added Changes:
    - Refreshed ubuntu machine types for hirsute@5.2
    - d/control: regenerated from d/control-in
    - d/p/ubuntu/lp-1907789-build-no-pie-is-no-functional-liker-flag.patch: fix
      ld usage of -no-pie (LP: #1907789)

qemu (1:5.2+dfsg-2) unstable; urgency=medium

  * move ui-opengl.so module from qemu-system-gui to qemu-system-common,
    as other ...

[Lukas Märdian (slyon)]

Proposed to drop the -no-pie flag at upstream s390-tools:
https://github.com/ibm-s390-tools/s390-tools/pull/106

[Launchpad Janitor (janitor)]

This bug was fixed in the package s390-tools - 2.15.1-0ubuntu6

---------------
s390-tools (2.15.1-0ubuntu6) hirsute; urgency=medium

  * debian/patches/s390-tools-lp1903984-hirsute.patch
    zcryptstats: Fix handling of partial results with many domains
    Thanks to Ingo Franzki (LP: #1903984)

 -- Frank Heimes <email address hidden> Tue, 19 Jan 2021 18:58:26 +0100