Bug 11939 - ld segfault in bfd_hash_allocate
Summary: ld segfault in bfd_hash_allocate
Alias: None
Product: binutils
Classification: Unclassified
Component: ld (show other bugs)
Version: 2.21
: P2 normal
Target Milestone: ---
Assignee: unassigned
Depends on:
Reported: 2010-08-23 06:37 UTC by Ralf Corsépius
Modified: 2010-08-28 04:08 UTC (History)
3 users (show)

See Also:
Host: x86_64-unknown-linux-gnu
Target: powerpc-rtems*, powerpc-elf*
Last reconfirmed:

data to reproduce the report (237.09 KB, application/octet-stream)
2010-08-23 06:42 UTC, Ralf Corsépius
A kludge (577 bytes, patch)
2010-08-25 18:29 UTC, H.J. Lu
Details | Diff

Note You need to log in before you can comment on or make changes to this bug.
Description Ralf Corsépius 2010-08-23 06:37:16 UTC
I am facing a segfault with powerpc-rtems*ld (powerpc-rtems is almost identical
to powerpc-*elf):

# powerpc-rtems4.11-ld -o minimum.ralf bootloader.o -b binary rtems.gz \
-T ppcboot.lds
Segmentation fault (core dumped)


#0  _int_malloc (av=0x3a2ab7ae80, bytes=<value optimized out>) at malloc.c:4439
#1  0x0000003a2a879a3d in __libc_malloc (bytes=4064) at malloc.c:3660
#2  0x00000000004e8a74 in _objalloc_alloc (o=0x2408390, len=72) at
#3  0x000000000043b2bf in bfd_hash_allocate (table=0x2408340, size=72) at
#4  0x000000000043bd32 in _bfd_generic_link_hash_newfunc (entry=0x0,
table=0x2408340, string=0x2425411 "em86_trap") at ../../bfd/linker.c:682
#5  0x000000000043aeb7 in bfd_hash_insert (table=0x2408340, string=0x2425411
"em86_trap", hash=127581402) at ../../bfd/hash.c:483
#6  0x000000000043ae87 in bfd_hash_lookup (table=0x2408340, string=0x2425411
"em86_trap", create=1, copy=0) at ../../bfd/hash.c:470
#7  0x000000000043b900 in bfd_link_hash_lookup (table=0x2408340,
string=0x2425411 "em86_trap", create=1, copy=0, follow=0) at ../../bfd/linker.c:501
#8  0x000000000043ce0a in _bfd_generic_link_add_one_symbol (info=0x75a6c0,
abfd=0x2412260, name=0x2425411 "em86_trap", flags=2, section=0x2414758,
value=46640, string=0x0, 
    copy=0, collect=0, hashp=0x2421c08) at ../../bfd/linker.c:1567
#9  0x000000000047c3f3 in elf_link_add_object_symbols (abfd=0x2412260,
info=0x75a6c0) at ../../bfd/elflink.c:4146
#10 0x000000000047e62e in bfd_elf_link_add_symbols (abfd=0x2412260,
info=0x75a6c0) at ../../bfd/elflink.c:5097
#11 0x000000000040f398 in load_symbols (entry=0x23fa0b0, place=0x7fff7881fe90)
at ../../ld/ldlang.c:2707
#12 0x000000000040fe95 in open_input_bfds (s=0x23fa0b0, force=0) at
#13 0x000000000041588b in lang_process () at ../../ld/ldlang.c:6287
#14 0x0000000000419726 in main (argc=9, argv=0x7fff78820058) at

Further info:

* I am able to deterministically reproduce it on fedora-13-x86_64, but 
am not able to reproduce it on fedora-13-i386 nor on fedora-12-{i386,x86_64}.

* I am able to reproduce the bug with ld from binutils-2.20.1,
binutils-2_20-branch and from binutils CVS-HEAD.

* The traceback above was produced by:
# powerpc-rtems4.11-ld --version
GNU ld (GNU Binutils)
Comment 1 Ralf Corsépius 2010-08-23 06:42:28 UTC
Created attachment 4947 [details]
data to reproduce the report

Extract the tarball and try

# powerpc-rtems4.11-ld -o minimum.ralf bootloader.o -b binary rtems.gz \
-T ppcboot.lds
Comment 2 H.J. Lu 2010-08-25 18:02:58 UTC
There is a fundamental problem in ld:

1. Different formats, ELF, COFF, XCOFF, ... have
different linker hash table and its entry.
2. It works as long as input and output have the 
same format.
3. When input format != output format, output linker
hash table and its entry have different formats from
the ones used by input.  We can't use output linker
hash table on input in this case.

We have been lucky most of time.  This bug isn't one of
Comment 3 H.J. Lu 2010-08-25 18:29:01 UTC
Created attachment 4950 [details]
A kludge

This is a kludge.
Comment 4 cvs-commit@gcc.gnu.org 2010-08-25 20:36:12 UTC
Subject: Bug 11939

CVSROOT:	/cvs/src
Module name:	src
Changes by:	hjl@sourceware.org	2010-08-25 20:35:59

Modified files:
	bfd            : ChangeLog elflink.c 

Log message:
	Don't set unique_global for non-ELF hash link table.
	2010-08-25  H.J. Lu  <hongjiu.lu@intel.com>
	PR ld/11939
	* elflink.c (elf_link_add_object_symbols): Don't set
	unique_global for non-ELF hash link table.


Comment 5 H.J. Lu 2010-08-25 20:38:17 UTC
Comment 6 Ralf Corsépius 2010-08-28 04:08:46 UTC
Thanks, your "kludge" seems to be working fine!