toolchain_gcc, branch master

Merge remote-tracking branch 'aosp/master' into replicant-6.0

2018-08-27T18:09:09+00:00

Merge "Add __emutls_unregister_key function"

2018-07-11T19:23:37+00:00

libgcc: work around old Bionic loader bug

2018-06-28T21:03:10+00:00

dl_iterate_phdr returns a 0 load_base for a PIE executable when it should
return the address where the executable was loaded (e.g. the load base or
load bias). Recalculate the load base when it is zero. This recalculation
should work on any ELF file with a PT_PHDR segment -- it will calculate 0
for a non-PIE executable.

The load base is added to an ELF virtual address to produce a run-time
address. Recalculate it by subtracting the PT_PHDR's virtual address from
its run-time address.

Bug: https://github.com/android-ndk/ndk/issues/505
Test: manual
Test: run NDK tests (./checkbuild.py && ./run_tests.py)
Change-Id: I7de46c07a8b04e794b59f07b4d554238cfd6d5d9

Add __emutls_unregister_key function

2018-06-06T04:10:07+00:00

Replace the existing __attribute__((destructor)) function,
unregister_emutls_key, with a function that something else must call
explicitly. We don't want the pthread key deleted at program exit, because
it's unnecessary and because it may delete the key before other tls-using
cleanup code runs. Exposing __emutls_unregister_key allows limiting the
cleanup to dlclose (e.g. by calling it only from crtbegin_so.c).

Reset emutls_key_created so multiple calls to __emutls_unregister_key are
safe.

Bug: b/80453944
Test: manual
Change-Id: I82d13614ae8042d0501fd2ca64f0ef6189669905

Delay emutls deallocation for one round

2018-05-22T06:03:29+00:00

With Android/Bionic, delay deallocation to round 2 of 4. It must run after
C++ thread_local destructors have been called, but before the final 2
rounds, because emutls calls free, and jemalloc then needs another 2
rounds to free its thread-specific data.

Bug: https://github.com/android-ndk/ndk/issues/687
Bug: b/78022094
Test: manual
Test: ./run_tests.py --rebuild --filter emutls-dealloc
Change-Id: I01bd634d97b7d22161b5cc8ca71b3cb94064a03e

Merge "gcc: backport patch to support vrnd<*>_f64 aarch64 intrinsics."

2018-04-26T21:06:26+00:00

gcc: backport patch to support vrnd<*>_f64 aarch64 intrinsics.

2018-04-25T05:15:37+00:00

This backports upstream patch to add support of vrdn<*>_f64 intrinisics
for aarch64. This is needed because glibc 2.27 uses some of them.

Author: mshawcroft 
Date:   Tue Apr 22 08:39:48 2014 +0000

    [AArch64] vrnd<*>_f64 patch

    This patch adds vrnd<*>_f64 aarch64 intrinsics.  A testcase for those
    intrinsics is added.  Run a complete LE and BE regression run with no
    regressions.

BUG=chromium:834385
TEST=./setup_board --board kevin --nousepkg
     cross-aarch64-cros-linux-gnu-glibc builds for glibc 2.27

Change-Id: Iaf9508a433e35d3304e44ac4b6c93071ff6c5ec0

gcc: backport an upstream patch to make it work with newer glibc.

2018-04-24T21:30:04+00:00

Newer glibc removes 'struct' tag from ucontext_t. Because of this, gcc
failed to bootstrap with newer glibc. This patch from upstream fixes
that.

BUG=chromium:834385
TEST=sudo emerge sys-devel/gcc passes with newer glibc installed.

commit 14c2f22a1877f6b60a2f7c2f83ffb032759456a6
Author: jsm28 
Date:   Tue Jul 4 10:22:56 2017 +0000

    Use ucontext_t not struct ucontext in linux-unwind.h files.

    Current glibc no longer gives the ucontext_t type the tag struct
    ucontext, to conform with POSIX namespace rules.  This requires
    various linux-unwind.h files in libgcc, that were previously using
    struct ucontext, to be fixed to use ucontext_t instead.  This is
    similar to the removal of the struct siginfo tag from siginfo_t some
    years ago.

    This patch changes those files to use ucontext_t instead.  As the
    standard name that should be unconditionally safe, so this is not
    restricted to architectures supported by glibc, or conditioned on the
    glibc version.

    Tested compilation together with current glibc with glibc's
    build-many-glibcs.py.

Change-Id: I433dd983b8d18c538d96486f7362b953ece3bfb8

Fix gcc 4.8.3 makeinfo build failure

2018-02-27T06:10:55+00:00

    /usr/local/google/home/rprichard/android/mingw/toolchain/gcc/gcc-4.8.3/gcc/doc/gcc.texi:88: warning: @tex should only appear at the beginning of a line
    /usr/local/google/home/rprichard/android/mingw/toolchain/gcc/gcc-4.8.3/gcc/doc/gcc.texi:208: no matching `@end tex'
    /usr/local/google/home/rprichard/android/mingw/toolchain/gcc/gcc-4.8.3/gcc/doc/gcc.texi:208: no matching `@end multitable'
    /usr/local/google/home/rprichard/android/mingw/toolchain/gcc/gcc-4.8.3/gcc/doc/gcc.texi:208: no matching `@end titlepage'
    Makefile:4353: recipe for target 'doc/gcc.info' failed

Even though it says "warning", the command exits unsuccessfully and fails
the build.

Adopt part of a patch from:

    https://gcc.gnu.org/ml/gcc-patches/2013-09/msg02100.html

This change is already in our gcc-4.9 directory.

For reference, here's the makeinfo version on my machine:

    $ makeinfo --version
    texi2any (GNU texinfo) 6.4.90

Test: ./toolchain/mingw/build.sh (in AOSP mingw branch)
Change-Id: Ib740c39d127ac8a127c929465fdf89477698a5e4

[GCC] Fix #endif problem in libgcc/emutls.c

2018-02-05T16:43:33+00:00

Currently in emutls.c 'emutls_key_created' is defined within part
of an ifdef, but is accessed outside the ifdef, which can cause
issues in builds where the ifdef is not taken.  This CL moves
the #endif to include the new function that accesses 'emutls_key_created'
so that all references to it are within the defining if-def.

Bug: b/72942688
Test: Fix tested in ChromeOS.
Change-Id: Ic1d6760d116b63abaef18f95e38de42af30952e4