Test: None
Bug: None
Change-Id: Ibc80455fe8345e79b43807f83e7071369949a894

Test: None
Bug: None
Change-Id: I219d884c29a8d73f92a77ec930bed30a9e3a04e4

This was broken when I changed this API. ld can't be run from the
arch-specific bin directory because this is missing.

Test: Build, check that winpthread is copied
Bug: None
Change-Id: Ia01e88624b0e66edc5563c931bdfbea174d2a131

Commit 325ba6fb34 excluded degenerate zero length PT_LOAD segments,
but that only fixed part of the problem, which was that the load
segment limits were not calculated properly.

PR 22845
* elf.c (IS_TBSS): Define.
(_bfd_elf_map_sections_to_segments): Use IS_TBSS.
(assign_file_positions_for_non_load_sections): Revert last change.
Properly calculate load segment limits to compare against relro limits.

This change is backport of dbc88fc14992c556b94e77de563a8f7abcb0b653
  - https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=dbc88fc14992c556b94e77de563a8f7abcb0b653
to binutils-2.27.

Bug: http://b/135627985

Change-Id: I793d6b1c8082bef1b1c46662d194790d5e7d7577

Fixes a failure triggered by -z separate-code.  p_memsz is tested
rather than p_filesz for objcopy --only-keep-debug where p_filesz is
set to zero.

PR 22845
* elf.c (assign_file_positions_for_non_load_sections): Ignore
degenerate zero size PT_LOAD segments when finding one overlapping
the PT_GNU_RELRO segment.

This change is backport of commit 325ba6fb34be799c885fad9287d883e86b835c84
  - https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=f2731e0c374e5323ce4cdae2bcc7b7fe22da1a6f
to binutils-2.27.

It is needed to cleanly cherry-pick dbc88fc14992c556b94e77de563a8f7abcb0b653.

Bug: http://b/135627985

Change-Id: Ic860cd5fa2992214b45b3bd0c8c5f1a222973533

Gold linker by default turns on the --fix-arm1176 option. It has
missing cases for TAG_CPU_ARCH_V8 and various other arm v8
architectures. The result of this causes gold using v4t stubs for
arm v8 cpus, which results in measurable code-size and performance
impact. See https://sourceware.org/bugzilla/show_bug.cgi?id=24642
for more details on the bug.

This patches extend the condition to be any TAGs larger than v7,
instead of checking the tag individually. It enables any future
updates on the tags. All the tag are defined in include/elf/arm.h.

Bug: b/134709902
Test: ./checkbuild.by passes, after the ../prebuilt/ndk/binutils is updated with newly built binutils.

Change-Id: I66277f62db1b1377b19f0e6d056f3b2d0a6f129b

Test: ./build.py --arch arm64 --host windows64
Bug: None
Exempt-From-Owner-Approval: janitorial
Change-Id: I4403dd0e1d406b65577a3af0b96b9b4cb95c5c53

This function was removed since we're all Python 3 now.

Test: ./build.py --arch arm64 --host linux
Bug: None
Change-Id: I35a9f9afa9eed4b74c365a73b653bddd64a11c0e

Test: ./build.py --arch arm64 --host win
Bug: None
Exempt-From-Owner-Approval: trivial janitorial changes
Change-Id: I64b4a8caab872168e5a3a2dccb6a80abe9c839d9

Apparently their build scripts do not do this automatically.

Test: ./build.py --arch arm64 --host linux
Bug: https://issuetracker.google.com/118423828
Change-Id: I5bee38c76152b8b67360c61277b996b7ba3eee9e

The Python libraries we depend on from the NDK are now Python 3.

Test: ./build.py --arch arm64 --host linux
Bug: None
Change-Id: I7725069b37be531b43851f7bc37400493cff616d

Bug: http://b/117439724

Without specifying alignment, if __CTOR_LIST__ is not aligned to 4
bytes, we end up with garbage after the -1.  This causes access
violations in __do_global_ctors.

The 64-bit version aligns it to 8 bytes (ld/scrpittempl/pep.sc).

Test: With this fix, AOSP Windows tools built with Clang + MinGW +
libc++ no longer crash.

Change-Id: I76be11a265cdcf272d740fa8715024f1f6df3e1b

am: 49e1641c

Change-Id: I6d0d2c56e1a860de81c6f140b90d26728d4fc333

This option is intended to make gold not apply link-time values for
absolution relocations which have dynamic relocations emitted for
them, in order to workaround an android dynamic loader bug in old
versions of android.

Unfortunately, it also had the side-effect of breaking debug data,
because the dynamic relocations are not used for non-ALLOC sections,
but the flag was also suppressing the static relocation.

This fix was proposed by jyknight@ and it fixed the problem by filtering
out non-ALLOC sections with the --no-apply-dynamic-relocs option.

BUG: 70838247
TEST: Debug info appears with --no-apply-dynamic-relocs

am: fcb56510

Change-Id: I1123440e0e3eda6ac467eb0a45a8152c694bda38

This commit replaces `std::make_pair<int, std::string>(dirindex, path)`
with `std::make_pair(dirindex, path)` because C++11 replaces
`std::make_pair<T, V>(T t, V v)` with
`std::make_pair<T, V>(T &&t, V &&v)`.

This commit removes explicit template arguments because forward
references are only available during template arguments deduction.  If
the explicit template arguments are specified, the argument `t` and `v`
will become R-value references, which can not bind to an L-value.

Test: It builds with clang++ (or g++) with CXXFLAGS="-std=c++11"
Change-Id: I79eb94552445354f7e7bff02169879195c7f6283

am: 726e32dc

Change-Id: Iad6e3535a096c9b674fe72f3e135b980db353b9f

Based on upstream change 1e2926 and 15a769.

Bug: 110235326
Change-Id: Ieee5ad4cfb0aa21b46e918191215c4ce12248661

am: ee60eb7f

Change-Id: I15268f7baa80a3b2e40788ba56bec78b66451070

am: cce69c1b

Change-Id: Ie1dfbaeb7c0a3c0be167215e8bd190be16af3b01

lld lays out the relro segment differently to GNU ld, not bothering to
include the first few bytes of .got.plt and padding out to a page at
the end of the segment.  This patch teaches binutils to recognize the
different (and somewhat inferior) layout as valid.

    bfd/
        PR 22829
        * elf.c (assign_file_positions_for_non_load_sections): Rewrite
        PT_GNU_RELRO setup.
    ld/
        * testsuite/ld-x86-64/pr14207.d: Adjust relro p_filesz.

This change is backport of commit f2731e0c374e5323ce4cdae2bcc7b7fe22da1a6f
  - https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=f2731e0c374e5323ce4cdae2bcc7b7fe22da1a6f
to binutils-2.27.

Change-Id: I9d598adb1a0ef4778f0f29d0faf19c7c8112a74a

am: 49f78f5d

Change-Id: I0453330e4274ca698dfbaba0e63d004d4285cdf0

am: 2485f18b

Change-Id: Id8b65c16f617b6c4995a94a11002944c052b9dc2

Copy libwinpthread-1.dll alongside the binutils executables on Windows.

The install_winpthreads function is copied from llvm_android:
https://android.googlesource.com/toolchain/llvm_android/+/bd22d9779676661ae9571972dcd744c42c70ffd0/build.py#1063

Most C++ MinGW programs need libpthread, because even libstdc++'s
"operator new" pulls in EH code, which calls into libpthread. Currently,
the ld.gold.exe and dwp.exe binaries need libwinpthread-1.dll, which is
not generally in the PATH, so the binaries don't run.

Previously, the Android GCC build of binutils linked ld.gold.exe and
dwp.exe with -static, which selected the static libpthread.a. Packaging
libwinpthread-1.dll alongside the binaries ensures that ld.gold.exe and
the LLVMgold plugin use the same copy of the winpthreads runtime, which
might be important.

Test: ./toolchain/binutils/build.py --arch arm64 --host win64
Test: ./toolchain/binutils/build.py  --arch arm64 --host win
Test: ./toolchain/binutils/build.py --arch arm64  --host linux
Test: ./toolchain/binutils/build.py --arch x86 --host win64
Bug: none
Change-Id: I5fc2d38b67c72fa4760be1c2688ef88479759160

am: 19df9bb8

Change-Id: I7d0adc629592f7785b44ea20fbedac0d53f50ee5

am: b415c49e

Change-Id: I483b208c1052e9a7d33881bd4ceee90605fadc7f

Some cc files include both "gold.h" and "output.h", which makes the
compiler on mac complain about redefinition of __STDC_FORMAT_MACROS.

Change-Id: I11875d9b76d27449167485ae5f91f026acbda8e1

am: f1ff343f

Change-Id: I2bebcd2a3a6fc88f8865dce8db6051a1b4344407

am: 61006b81

Change-Id: I9d5b9c2f0f760e5833c66e9d2ec338350c3f9be8

am: b162eb11

Change-Id: Ifda56cebb333499e72c5692cced93b0bd425d559

am: b4a70cbe

Change-Id: Iee9ed64d9af676334c861e08485b79119247cf5a

If the relocation offset is odd, create the relocation in rela_dyn
section instead of the relr_dyn section.

Bug: None
Test: Verified relocations after linking test binary with odd offset.
Change-Id: If0d65176639bfb7401f7fb2dd43efc55c17670a4

Print a more informative error message instead of asserting out.
This prints errors for all the odd offsets before exiting gold.

Bug: None
Test: Verified error message on linking test binary with odd offset.
Change-Id: Ib481a4983cac2e637b1128ea0559d9606e671c35

am: 73fe135c

Change-Id: I721a8cab09b24a6de1bdee0f47de584367c2885b

am: 41d8fcbb

Change-Id: I5f394a221c1fa3e3efb92f7e9d8793545af37417

If a shared library contains an undefined symbol and LTO adds
a new reference to that same undefined symbol, the reference in the new
object added by the plugin would not trigger a rescan of the archive
containing the symbol.

2017-11-17  Stephen Crane  <sjc@immunant.com>

gold/
PR gold/22448
* symtab.cc (Symbol_table::add_from_object): Only rescan for
undefined symbols in regular, not dynamic, objects.

This change is backport of commit 71739b698d47945463f183619078ff680c277f0f
  - https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=71739b698d47945463f183619078ff680c277f0f
to binutils-2.27.

Change-Id: I9d926d8c56b02601785e18c88ef886ed2433c25b

am: 4c4cb9ac

Change-Id: If19e50676576b73072b86f75a92d9ed646b7e258

am: b01e31e3

Change-Id: I824423dc012b7ec47ac3a4a3866d4c1e6215cefe

The --disable-nls flag is also used to build gdb and gcc for Android.
Adding it here prevents 428 *.mo files from being added to the next NDK
release. (AFAICT, the SDK and NDK currently have no *.mo files.) It also
fixes a sporadic build failure that can happen when binutils needs its
bundled copy of libintl/gettext (see build 4635300 of ab/aosp-binutils or
the linked issue).

Test: ./toolchain/binutils/build.py --arch arm --host win
Test: ./toolchain/binutils/build.py --arch arm --host win64
Test: ./toolchain/binutils/build.py --arch arm --host linux
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=22941
Change-Id: I0810406dc52f366fac7491676c2d5cb000fcfcaa

am: 808f4aae

Change-Id: Ic9d4379eb08863289825d3d41be3836b3ca1846e