[Bug ld/22397] BFD internal error when message locale isn't C

[amodra at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=22397

Alan Modra  changed:

   What|Removed |Added

 Status|UNCONFIRMED |ASSIGNED
   Last reconfirmed||2017-11-05
 CC||amodra at gmail dot com
   Assignee|unassigned at sourceware dot org   |amodra at gmail dot com
 Ever confirmed|0   |1

--- Comment #1 from Alan Modra  ---
The relevant lines from fr.po are these:

# Les %s%s au début de %s%s« %s » peuvent être « non défini » ou rien pour le
premier et
# un des fragments ci-dessus pour le second ce qui donne par exemple « symbole
interne non défini» en inversant les deux %s.
# Le dernier %s collé au dernier mot peut être rien ou un des fragments
traduits ci-dessus et commençant par un point-virgule.
#: elf64-x86-64.c:1947
#, c-format
msgid "%B: relocation %s against %s%s`%s' can not be used when making a shared
object%s"
msgstr "%1$B: réadressage %2$s vers %4$s%3$s« %5$s » ne peut pas être utilisé
en créant un objet partagé%6$s"

The internal error is due to the use of 1$ and other parameter number
specifiers.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
___
bug-binutils mailing list
bug-binutils@gnu.org
https://lists.gnu.org/mailman/listinfo/bug-binutils

[Bug ld/22394] hppa-linux-ld fails to emit dynamic relocations on function pointer initialization

[amodra at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=22394

Alan Modra  changed:

   What|Removed |Added

 Status|ASSIGNED|RESOLVED
 Resolution|--- |FIXED

--- Comment #5 from Alan Modra  ---
Dave, thanks for the commentary on ABI issues.  Fixed, I think in line with
your comments.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
___
bug-binutils mailing list
bug-binutils@gnu.org
https://lists.gnu.org/mailman/listinfo/bug-binutils

[Bug ld/22394] hppa-linux-ld fails to emit dynamic relocations on function pointer initialization

[cvs-commit at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=22394

--- Comment #4 from cvs-commit at gcc dot gnu.org  ---
The master branch has been updated by Alan Modra :

https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=127e8e9f62ed0e7145dfa1172c6253275fc150dd

commit 127e8e9f62ed0e7145dfa1172c6253275fc150dd
Author: Alan Modra 
Date:   Tue Oct 31 18:13:03 2017 +1030

PR22394, hppa-linux-ld fails to emit dynamic relocations

gcc -mfast-indirect-calls emits a function pointer initialization
without a P% (plabel) modifier.  ld does not create the necessary
dynamic relocations for this to work.  It turns out that the problem
is caused by the non_got_ref symbol flag.  This flag is set for
non-pic by check_relocs to indicate that the symbol might need copy
relocations or dynamic relocations.  Later, the backend
adjust_dynamic_symbol clears the flag to indicate dynamic relocations
are needed, but leaves it set when copy relocations were created.  The
inversion in meaning is insane, but it's that way because the backend
adjust_dynamic_symbol function doesn't get to look at all symbols..
Anyway, the insanity works for non-function symbols.  However, the
flag is left set on any function symbol with a dynamic relocation.

This patch fixes the non_got_ref handling for function symbols, adds
-z nocopyreloc for hppa-elf, reports where textrel occurs, and expands
comments.  The check_relocs change just stops creation of dyn_relocs
we always threw away later.

PR 22394
* elf32-hppa.c (elf32_hppa_check_relocs): Don't create dyn_relocs
for plabels when non-pic.
(maybe_set_textrel): New function.
(readonly_dynrelocs): Move and rewrite.
(elf32_hppa_adjust_dynamic_symbol): Use it.  Don't create copy
relocs when def_regular or -z nocopyreloc.  Handle non_got_ref
for functions.  Expand non_got_ref comments.
(elf32_hppa_size_dynamic_sections): Use maybe_set_textrel.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
___
bug-binutils mailing list
bug-binutils@gnu.org
https://lists.gnu.org/mailman/listinfo/bug-binutils

[Bug ld/22394] hppa-linux-ld fails to emit dynamic relocations on function pointer initialization

[danglin at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=22394

--- Comment #3 from John David Anglin  ---
On HP-UX, the 64-bit HP linker treats the DIR64 and FPTR64 relocations the same
when "symbol" is a function pointer.  Both point to a function descriptor.

The ABI description is here:
https://uclibc.org/docs/psABI-parisc.pdf

If DIR32 and PLABEL32 were to behave similarly on 32-bit ELF, this would
break the current fast indirect call implementation.  We would have one
additional layer for fast indirect calls.  This is reasonably efficient
when generating PA 2.0 code because there is the bve instruction.  However,
things are not efficient when generating PA 1.X code.  So, fast indirect
calls would no longer be useful.

The original HP SOM linker required the P' selector for procedure labels
for functions in shared libraries.  So, it wasn't possible to do a fast
indirect call to a function in a shared library.  When the function isn't
in a shared library, the call is direct and no function descriptor is used.

Thus, I think it best that the DIR32 relocation provide the actual function
address and not point to a descriptor.  This will enable calls into shared
libraries to functions that don't need the PIC register.  Having the ability
to directly load the address of a function is probably useful for other things.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
___
bug-binutils mailing list
bug-binutils@gnu.org
https://lists.gnu.org/mailman/listinfo/bug-binutils

[Bug ld/22397] New: BFD internal error when message locale isn't C

[nintendocatz at msn dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=22397

Bug ID: 22397
   Summary: BFD internal error when message locale isn't C
   Product: binutils
   Version: 2.29
Status: UNCONFIRMED
  Severity: normal
  Priority: P2
 Component: ld
  Assignee: unassigned at sourceware dot org
  Reporter: nintendocatz at msn dot com
  Target Milestone: ---

Hello,

Arch user here, I just hit this exact same bug with ld-2.29.1 without any
patches :
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=873485

I will repeat the information from this bug report here :

When ld tries to localize a you-need-fPIC error in fr_FR.UTF-8 (french),
it aborts ("abandon") with an internal error ("erreur interne"):

$ cat wtf.c++
#include 
void lol() { std::cout << "fail"; }

$ LC_ALL="fr_FR.UTF-8" g++ -shared wtf.c++
/usr/bin/ld: /usr/bin/ld: BFD erreur interne (GNU Binutils for Debian)
2.29, abandon à ../../bfd/bfd.c:843 dans _doprnt

/usr/bin/ld: Merci de rapporter cette anomalie.

collect2: error: ld returned 1 exit status

When switching to C locale, ld outputs an error as expected:

$ LC_ALL="C" g++ -shared wtf.c++
/usr/bin/ld: /tmp/ccomgT5o.o: relocation R_X86_64_PC32 against symbol
`_ZSt4cout@@GLIBCXX_3.4' can not be used when making a shared object;
recompile with -fPIC /usr/bin/ld: final link failed: Bad value
collect2: error: ld returned 1 exit status


I'm not sure if the bug can be triggered with any other locale.
I couldn't find any bug report looking similar to this one.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
___
bug-binutils mailing list
bug-binutils@gnu.org
https://lists.gnu.org/mailman/listinfo/bug-binutils

[Bug ld/22394] hppa-linux-ld fails to emit dynamic relocations on function pointer initialization

[danglin at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=22394

--- Comment #2 from John David Anglin  ---
The segmentation fault doesn't occur with trunk plus your proposed
elf32-hppa.c patch.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
___
bug-binutils mailing list
bug-binutils@gnu.org
https://lists.gnu.org/mailman/listinfo/bug-binutils

[Bug ld/22394] hppa-linux-ld fails to emit dynamic relocations on function pointer initialization

[danglin at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=22394

--- Comment #1 from John David Anglin  ---
Hi Alan,

Although your example is not technically supported, I think it should
work on linux because it doesn't use the PIC register and because of the
flat address space.

First, calls to functions in shared libraries are not supported using the
-mfast-indirect-calls gcc option:

  @item -mfast-indirect-calls
  @opindex mfast-indirect-calls
  Generate code that assumes calls never cross space boundaries.  This
  allows GCC to emit code that performs faster indirect calls.

  This option does not work in the presence of shared libraries or nested
  functions.

What works is:

$ gcc -o funcp funcp.c funcpdef.c -mfast-indirect-calls
$ ./funcp
$

Issues are the space register used for the call and the PIC register.

Looking at your example, it seems to me that R_PARISC_DIR32 relocations
are not handled correctly when they refer to a function.  This is the
assembly code:

.LEVEL 1.1
.globl fp
.section.data.rel,"aw",@progbits
.align 4
.type   fp, @object
.size   fp, 4
fp:
.word   foo

At runtime, fp is NULL causing the segmentation fault.  So, I agree
with your assessment.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
___
bug-binutils mailing list
bug-binutils@gnu.org
https://lists.gnu.org/mailman/listinfo/bug-binutils