binutils_os216

Fork

opcodes: link against libbfd.la for rpath deps

The reason opcodes doesn't link against libbfd.la is to workaround a
libtool bug where it uses installed -L paths ahead of DESTDIR paths.
The downside is that the library itself lacks rpath tags to find the
right version of libbfd.so.

Since Gentoo has patched the libtool bug for a while, we don't need
the workaround. Use the .la file so we get the rpath tags.

URL: https://bugs.gentoo.org/563934

libiberty: install PIC version of libiberty.a

This will install a PIC version of libiberty.a by overwriting the non-PIC
version of libiberty.a while compiling. We do this because there is no
shared version of libiberty for random apps to link against which means if
someone wants to use this in a shared library or PIE, they're out of luck.
It's arguable whether people should be able to use this in a shared lib,
but usage in PIE should be fine. You could argue that this penalizes the
non-PIE users, but the counter point is that people using this library in
general are fairly low, and we'd rather have things work for all of them.

gold/ld: enable gnu hash by default

Glibc first added .gnu.hash support to glibc-2.5 (released 29 Sep 2006),
and gold was first released after that. Let's default the gnu hash style
to the new "gnu" rather than the classic sysv.

gold/:
2012-02-03 Mike Frysinger <vapier@gentoo.org>

* options.h (General_options): Change default to gnu for hash_style.

ld: enable new dtags by default for linux/gnu targets

The "new" dtags options have been around for 14+ years now, so for Linux
and GNU targets, enable them by default.

2012-01-21 Mike Frysinger <vapier@gentoo.org>

* emultempl/elf32.em (gld${EMULATION_NAME}_before_parse): Set
link_info.new_dtags to TRUE for linux/gnu targets.
* NEWS: Mention new dtags default.

2013-01-22 Roland McGrath <mcgrathr@google.com>

* emultempl/elf32.em (gld${EMULATION_NAME}_before_parse): Set
new_dtags to TRUE for *-*-nacl* targets.

gold/ld: add support for poisoned system directories

This is based on the old CodeSourcery patch written by Joseph Myers to add
support to the link for detecting & rejecting bad -L paths when using a
cross-compiler. The differences here:
* The command line flags are always available.
* We can turn on & off the warning via the command line.
* The configure option controls the default warning behavior.
* Add support for gold.

It is not currently upstream, nor has it been submitted at all. There are
no plans to do so currently either.

BUG=chromium:488360
TEST=`cbuildbot chromiumos-sdk` passes # tests arm/amd64/mipsel/x86
TEST=`cbuildbot panther_moblab-full whirlwind-release` pass
TEST=`cbuildbot {x32,arm64}-generic-full` has no new failures
TEST=x86_64-cros-linux-gnu-ld throws warnings when using -L/lib (gold & bfd)

Reviewed-on: https://chromium-review.googlesource.com/272083

ld: always warn about textrels in files

textrels are bad for forcing copy-on-write (this affects everyone), and for
security/runtime code generation, this affects security ppl. But in either
case, it doesn't matter who needs textrels, it's the very fact that they're
needed at all.

Release 2.27

bfd/
2016-08-03 Tristan Gingold <gingold@adacore.com>

* version.m4: Bump version to 2.27
* configure: Regenerate.

binutils/
2016-08-03 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

gas/
2016-08-03 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

gprof/
2016-08-03 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

ld/
2016-08-03 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

opcodes/
2016-08-03 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

Regenerate pot files.

PowerPC64 ld segfault with code in non-executable sections

PR ld/20428
* elf64-ppc.c (ppc_get_stub_entry): Don't segfault on NULL group.

Automatic date update in version.in

Automatic date update in version.in

Automatic date update in version.in

Automatic date update in version.in

Automatic date update in version.in

Automatic date update in version.in

Automatic date update in version.in

Fix seg fault in linker when performing garbage collection on COFF based targets.

PR ld/20401
bfd * coffgen.c (fini_reloc_cookie_rels): Check for the extistence
of the coff_section_data before using it.

Automatic date update in version.in

Support -pie for aarch64*-elf targets.

ld/
* emulparams/aarch64elf.sh (GENERATE_PIE_SCRIPT): Set to yes.
* emulparams/aarch64elf32.sh: Likewise.

(cherry picked from commit 3af09e8db3084f5e03b7f09768a2ab6415418e63)

Expect the 'objcopy without global symbols' test to fail for AArch64 and ARM targets.

* testsuite/binutils-all/objcopy.exp
(objcopy_test_without_global_symbol): Expect this test to fail on
the AArch64 and ARM targets, since they preserve their mapping
symbols.

Bump version to 2.26.90 (on branch 2.27)

bfd/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* version.m4: Bump version to 2.26.90
* configure: Regenerate.

binutils/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

gas/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

gprof/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

ld/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

opcodes/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

Bump version to 2.27.51

bfd/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* version.m4: Bump version to 2.27.51
* configure: Regenerate.

binutils/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

gas/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

gprof/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

ld/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

opcodes/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* configure: Regenerate.

Add marker for 2.27 branch.

binutils/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* NEWS: Add marker for 2.27.

gas/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* NEWS: Add marker for 2.27.

ld/
2016-07-01 Tristan Gingold <gingold@adacore.com>

* NEWS: Add marker for 2.27.

Fix mis-placement in binutils.texi

binutils/
* doc/binutils.texi (objdump): Fix mis-placement.

x86-64/MPX: bndmk, bndldx, and bndstx don't allow RIP-relative addressing

Additionally warn about scaling factors other than 1 for the latter
two, as those get ignored by the hardware.

x86/MPX: fix address size handling

While address overrides are ignored in 64-bit mode (and hence shouldn't
really result in an error, but upon v1 converting this to a warning I
was told otherwise), trying to use 16-bit addressing is documented to
result in #UD, and hence the assembler should reject the attempt. (The
added test case at once also checks that bndc{l,n,u} won't accept
16-bit register operands.)

x86/Intel: don't accept bogus instructions

... due to their last byte looking like a suffix, when after its
stripping a matching instruction can be found. Since memory operand
size specifiers in Intel mode get converted into suffix representation
internally, we need to keep track of the actual mnemonic suffix which
may have got trimmed off, and check its validity while looking for a
matching template. I tripper over this quite some time again after
support for AMD's SSE5 instructions got removed, as at that point some
of the SSE5 mnemonics, other than expected, didn't fail to assemble.
But the problem affects many more instructions, namely (almost) all
MMX, SSE, and AVX ones as it looks. I don't think it makes sense to
add a testcase covering all of them, nor do I think it makes sense to
pick out some random examples for a new test case.

x86: allow suffix-less movzw and 64-bit movzb

... just like is already the case for 16- and 32-bit movzb: I can't see
why omitting suffixes on this (and movs{b,w,l}) is not allowed, when it
is allowed for all other instructions where the suffix is redundant
with (one of) the operands.

x86: remove stray instruction attributes

- with Cpu64 Disp16 makes no sense for memory operands
- with CpuNo64 Disp32S makes no sense
- non-64-bit lgdt doesn't allow 10-byte operands

x86/Intel: fix operand checking for MOVSD

The dual purpose mnemonic (string move vs scalar double move) breaks
the assumption that the isstring flag would be set on both the first
and last entry in the current set of templates, which results in bogus
or missing diagnostics for the string move variant of the mnemonic.
Short of mostly rewriting i386_index_check() and its interaction with
the rest of the code, simply shrink the template set to just string
instructions when encountering the second memory operand, and run
i386_index_check() a second time for the first memory operand after
that reduction.