Cary Coutant [Sat, 2 Dec 2017 06:03:09 +0000 (22:03 -0800)]
Fix incremental linking failure with GCC 7+.
With the new compiler, we're running out of patch space for the .eh_frame
section. To workaround that issue, we compile the before and after versions
both with no unwind tables.
gold/
PR gold/22309
* testsuite/Makefile.am (two_file_test_1_v1_ndebug.o): Compile with
no EH information.
(two_file_test_1_ndebug.o): Likewise.
* testsuite/Makefile.in: Regenerate.
* testsuite/two_file_test_1.cc: Touch to force recompilation with new
flags.
* testsuite/two_file_test_1_v1.cc: Likewise.
Cary Coutant [Thu, 30 Nov 2017 21:44:28 +0000 (13:44 -0800)]
Fix problem where erratum stubs are not always applied.
I checked over the results of applying --fix-cortex-a53-843419 to
a very large program (gitit) with two stub tables and thousands
of erratum fixes. I noticed that all the erratum_stubs were being
created but about 1/3 of them were being skipped over by
fix_errata_and_relocate_erratum_stubs(). By skipped over I mean
no branch relocation or adrp -> adr transformation was applied to
the erratum address, leaving the erratum_stub unreachable, and
with a branch with a 0 immediate.
The root cause of the skipped over erratum_stubs is
Erratum_stub::invalidate_erratum_stub() that is used to set
relobj_ to NULL when an erratum_stub has been processed.
Unfortunately relobj_ is used in operator<() so altering relobj
makes the results from erratum_stubs_.lower_bound() as used in
find_erratum_stubs_for_input_section() unreliable.
2017-11-30 Peter Smith <peter.smith@linaro.org>
Cary Coutant <ccoutant@gmail.com>
gold/
PR gold/20765
* aarch64.cc (Erratum_stub::invalidate_erratum_stub): Use erratum_insn_
instead of relobj_ to invalidate the stub.
(Erratum_stub::is_invalidated_erratum_stub): Likewise.
Peter Smith [Thu, 30 Nov 2017 21:30:58 +0000 (13:30 -0800)]
Fix segfault in relocate_erratum_stub on aarch64.
The fix for PR21868 (an internal error when --fix-cortex-a53-843419
is applied) has a small mistake in it. When the stub_owner section
needs an erratum fix an incorrect address for the stubs for the section
is given to relocate_erratum_stub. If we are lucky we will get a segfault;
if we aren't, an incorrect patch or data corruption is possible.
The error is visible in PR21868, but the side-effects aren't fatal.
Alan Modra [Tue, 14 Nov 2017 08:17:34 +0000 (18:47 +1030)]
PR22431, powerpc64 ld segfault when .plt discarded
The fix for the PR is to not use input_section->output_section->owner
to get to the output bfd, but use the output bfd directly since it is
available nowadays in struct bfd_link_info.
I thought it worth warning when non-empty dynamic sections are
discarded too, which meant a tweak to one of the ld tests to avoid the
warning.
bfd/
PR 22431
* elf64-ppc.c (ppc64_elf_size_dynamic_sections): Warn on discarding
non-empty dynamic section.
(ppc_build_one_stub): Take elf_gp from output bfd, not output
section owner.
(ppc_size_one_stub, ppc64_elf_next_toc_section): Likewise.
ld/
* testsuite/ld-elf/note-3.t: Don't discard .got.
Alan Modra [Sun, 5 Nov 2017 05:52:55 +0000 (16:22 +1030)]
PR22397, BFD internal error when message locale isn't C
This adds positional parameter support to the bfd error handler,
something that was lost 2017-04-13 when _doprnt was added with commit c08bb8dd. The number of format args is now limited to 9, which is
sufficient for current _bfd_error_handler messages. If someone
exceeds 9 args they get the joy of modifying this code to support more
args (shouldn't be too difficult).
PR 22397
* bfd.c (union _bfd_doprnt_args): New.
(PRINT_TYPE): Add FIELD arg. Take value from args.
(_bfd_doprnt): Replace ap parameter with args. Adjust all
PRINT_TYPE invocations and reading of format args to suit.
Move "%%" handling out of switch handling args. Support
positional parameters.
(_bfd_doprnt_scan): New function.
(error_handler_internal): Call _bfd_doprnt_scan and read args.
../../src/bfd/bfd.c:642:1: error: conflicting types for '_doprnt'
_doprnt (FILE *stream, const char *format, va_list ap)
^~~~~~~
In file included from ../../src/bfd/sysdep.h:37:0,
from ../../src/bfd/bfd.c:367:
/usr/lib64/gcc/i586-pc-msdosdjgpp/6.1.0/include-fixed/stdio.h:171:5: note: previous declaration of '_doprnt' was here
int _doprnt(const char *_fmt, va_list _args, FILE *_f);
^~~~~~~
Possible fixes the build on other systems too -- _doprnt was/is an
exported symbol on old Unix/BSD systems too.
bfd/ChangeLog:
2017-10-11 Pedro Alves <palves@redhat.com>
Alan Modra [Sun, 29 Oct 2017 23:50:29 +0000 (10:20 +1030)]
relocs_compatible test for gc-sections
I noticed when looking at pr22300 that before calling check_relocs we
have an elf_object_id test (added for pr11933) as well as the
relocs_compatible test. I believe backend gc_mark_hook and
gc_sweep_hook ought to be protected similarly from being confused by
unexpected relocations (for example, both elf64-ppc.c and elf32-ppc.c
use _bfd_elf_relocs_compatible, so I think it would be possible for
the ppc64 gc_mark_hook to be presented with a ppc32 relocatable
object).
* elflink.c (elf_gc_sweep): Test elf_object_id in addition to
relocs_compatible.
(bfd_elf_gc_sections): Likewise.
Alan Modra [Sat, 28 Oct 2017 03:40:55 +0000 (14:10 +1030)]
PR22300, Abort in elf32_hppa_relocate_section building polyml
polyml produces object files with the wrong OS/ABI for hppa-linux.
This, along with the fact that elf32-hppa.c is using the strictest
backend relocs_compatible, results in wrong merging of ELF symbols.
So, remove the relocs_compatible check in _bfd_elf_merge_symbol.
_bfd_elf_merge_symbol is only called nowadays from within blocks
protected by is_elf_hash_table, so "we are doing an ELF link" as the
removed comment says, is true.
Also relax relocs_compatible for hppa and powerpc. relocs_compatible
is used for more than just merging symbols, as the name suggests.
This allows objects that are in fact reasonably compatible to be
linked.
Alan Modra [Tue, 17 Oct 2017 14:03:05 +0000 (00:33 +1030)]
[GOLD] Fix powerpc64 optimization of TOC accesses
Fixes a thinko. Given code that puts variables into the TOC (a bad
idea, but some see the TOC as a small data section) this bug could
result in an attempt to optimize a sequence that should not be
optimized.
Passing enum as a first argument to variadic argument function
may lead to undefined behavior. The explanation on CERT site:
https://www.securecoding.cert.org/confluence/display/cplusplus/
EXP58-CPP.+Pass+an+object+of+the+correct+type+to+va_start
The bug was found by Kirill Nedostoev (nedostoev.ka@phystech.edu)
when he tried to build GNU binutils with Clang 7.
PR 22245
* bfd.c (bfd_set_error): Avoid UB on passing arg to va_start that
undergoes default promotion.
* bfd-in2.h: Regenerate.
Alan Modra [Thu, 28 Sep 2017 07:41:38 +0000 (17:11 +0930)]
PR22220, BFD linker wrongly marks symbols as PREVAILING_DEF_IRONLY
non_ir_ref_dynamic wasn't being set in the case where we have a
versioned dynamic symbol definition with a non-versioned matching IR
symbol.
bfd/
PR 22220
* elflink.c (_bfd_elf_merge_symbol): Set non_ir_ref_dynamic in
a case where plugin_notice isn't called.
ld/
* testsuite/ld-plugin/pr22220.h,
* testsuite/ld-plugin/pr22220lib.cc,
* testsuite/ld-plugin/pr22220lib.ver,
* testsuite/ld-plugin/pr22220main.cc: New test.
* testsuite/ld-plugin/lto.exp: Run it.
Alan Modra [Fri, 22 Sep 2017 01:34:41 +0000 (11:04 +0930)]
[GOLD] Set non-exec stack for ppc64
gcc doesn't emit stack notes for ELFv1, since ELFv1 never needs an
executable stack. Note that ELFv1 is usually big-endian and ELFv2
little-endian, but the ABI is really orthogonal to endiannes.
* powerpc.cc (Target_powerpc<64,*>::powerpc_info): Set
is_default_stack_executable false.
H.J. Lu [Fri, 22 Sep 2017 21:15:40 +0000 (14:15 -0700)]
x86: Guard against corrupted PLT
There should be only one entry in PLT for a given symbol. Set howto to
NULL after processing a PLT entry to guard against corrupted PLT so that
the duplicated PLT entries are skipped.
projects / thirdparty / binutils-gdb.git / log
