Bug: 23396112
Switch from "-fsanitize-undefined-trap-on-error" to
"-fsanitize-trap=all". The former ends up accidentally leaving
unresolved calls to __ubsan* helper functions in the object file with
clang 3.8. The latter is used when we don't include address sanitizer,
and replaces any misbehavior with a direct call to abort().
Change-Id: I8a67461b45f5f1dd9f2d179b6b64a4ca905e999f
Force gcc and clang to always output color diagnostics. Ninja will
strip the ANSI color codes if it is not running in a terminal.
Bug: 24273983
Change-Id: Ic9252549bfacf3bbb3f4b9d20f63010fc24ac21d
This doesn't seem to be a problem any more, and mac builds
that differ from linux builds is causing other problems.
This reverts commit 7c40fb95dc.
Bug: 24339250
Change-Id: I8a0da11df42d58ba9c3ded4571f1db870f400a21
Instead of using recursive make to change the HOST_OS when building the
windows SDK under linux, add the concept of cross-building to another
host os.
Bug: 23566667
Change-Id: I6dc525b601b6251d458d197c30bf4660d7485502
We have deferred the CC_WRAPPER prepending to binary.mk per module.
Bug: 22612634
Change-Id: If2c0549503989d17ff7be619fdd49c26fd78f0a5
(cherry-pick from commit 66c8e8c756)
Another change in bionic/linker adds linker_asan/linker_asan64 that
know where to find ASan shared libraries.
Also, include linker_asan to the required packages list when building
for ASan.
Change-Id: I8ebe7c0091bbeb0c135708a891d33d9844373d37
Upstream clang r239152 (http://reviews.llvm.org/D10239) caused a pretty
significant change in behavior. Passing an FPU feature via -mfpu
disables any feature not supported by those flags (in addition to the
old behavior of enabling features supported by the flags). For e.g.,
-mfpu=neon used to just pass +neon, +vfp3 as target features to the
backend. Now, -mfpu=neon also passes -vfp4, -fp16, -fp-armv8.
The backend has always disabled implied feature bits if a feature is
disabled. Upon seeing the target feature -vfp4, it will disable any
processor/feature that implies vfp4, including the bit that the
processor is a Krait. Since Krait has both Neon and vfp4, it is safe to
pass '-mfpu=neon-vfpv4'.
Change-Id: Ibbb992e80b8cbc8dc36d5df556885d0912baea22
Sanitized RPATH now mentions /system/vendor/lib to preserve overlay
in the case when a sanitized version of a vendor library can not be
built.
Bug: 22199458
Change-Id: I3222d2e1d6c08fdd1e0404fcb7db347aa4a92bb7
A fully (or even mostly) asan-instrumented device will have 2 copies of each
shared library, which might not fit on system partition. Moving instrumented
libraries to /data.
Bug: 21785137
Change-Id: I64184261da2eb24a1382c67e4931c34a5a38b3c0
This also does a bit of cleanup in config_sanitizers.mk. The result is
that `LOCAL_SANITIZE := <any arbitrary ubsan group>` should function
fine for both host and target.
This is a superset of LOCAL_DETECT_INTEGER_OVERFLOWS, so remove that.
This also checks integer division by zero. It's supposed to cover
shifting undefined behaviors as well, but apparently it does not
(though `LOCAL_SANITIZE := shift` works fine).
Change-Id: I4ac99eafa6920a3f8cb82af37ce56ff0fdb95223
GDB does not yet have support for compact branches, and is unable to
set a breakpoint on them. Turn compact branch generation off, until
GDB is fixed.
Change-Id: Ie7d6fb891e9934f8fc645fe9cf8f706be15a5f77
Clang++ for Mips and Mips64 generates read-only exception unwind
tables that trigger DT_TEXTREL warnings at link time. Until Clang
is fixed, ignore those performance warnings instead of failing the build.
With this patch, Mips clang++ can be (optionally) used when building
libdeqp.
NDK's Mips llvm has been using an alternate temporary fix, marking the
.gcc_exception_table section as read-write for Mips only:
https://android-review.googlesource.com/#/c/119660/
A permanent fix using a read-only exception table is pending upstream:
http://reviews.llvm.org/D9669
Change-Id: Ie0cd7da398acbe45dbe39adc251e7fd5b5ca1445
ASan runtime library (when using dynamic linking) must be the first
dependency of the main executable to achieve correct symbol
interposition. This matches how the clang driver works.
In multilib setup, ASan-RT name depends on the target arch:
/system/lib/libclang_rt.asan-arm-android.so
/system/lib64/libclang_rt.asan-arm64-android.so
We also set RPATH to /system/lib/asan or /system/lib64/asan
to have a place for ASan-only versions of system libraries.
Change-Id: I937d202077b6e433ba476c075d31be818b662d53
Also filter out gcc-only flags -Wno-clobbered and -fno-devirtualize
when compiled with clang/llvm.
BUG: 19872411
Change-Id: I6de57583be04da607f569df65e93531787dbb789
(cherry picked from AOSP commit 90036610dd)
Also filter out gcc-only flags -Wno-clobbered and -fno-devirtualize
when compiled with clang/llvm.
BUG: 19872411
Change-Id: I6de57583be04da607f569df65e93531787dbb789
ASan runtime library (when using dynamic linking) must be the first
dependency of the main executable to achieve correct symbol
interposition. This matches how the clang driver works.
In multilib setup, ASan-RT name depends on the target arch:
/system/lib/libclang_rt.asan-arm-android.so
/system/lib64/libclang_rt.asan-arm64-android.so
We also set RPATH to /system/lib/asan or /system/lib64/asan
to have a place for ASan-only versions of system libraries.
Change-Id: I5c0cdb89e5e08a1950eb276e406da9f31a6e52dd
Some of the ubsan checks expose a few pathological performance cases
in clang, and thus aren't suited to be used in SANITIZE_HOST.
This mode is also supported on the target despite not having the
target runtime libraries for ubsan by generating traps.
Change-Id: I0b0f0a08ca84d72e44e2174a66726b1c5e5cad7e
Bug: 20148343
The latest clang warns on "%p" being used with non-void pointer types, which
would require a great deal of cleanup in several sub-projects.
Change-Id: I4175d5cf8709504455067c906a2eb8ccc8e25432
Rather than adding LOCAL_UB_SANITIZER, LOCAL_THREAD_SANITIZER, etc for
each new sanitizer, deprecate LOCAL_ADDRESS_SANITIZER in favor of
LOCAL_SANITZE that mirrors the behavior of -fsanitize=<sanitizers>.
For example, the following will use both asan and ubsan:
LOCAL_SANITIZE := address undefined
We'll leave LOCAL_ADDRESS_SANITIZER around for compatibility until we
can clean up the tree.
Change-Id: I8a62315129d4753f8e992584ca6db1e5dfdd4d2a
Clang 3.6 (including the snapshot present in AOSP) added support for
-mcpu=cortex-a15 -- there's no need to fall back to generic armv7-a
anymore.
Change-Id: I8a582261ac35e32f8f1def1d244cf83f2096520d
Signed-off-by: Bernhard Rosenkränzer <Bernhard.Rosenkranzer@linaro.org>
Clang's tail call optimization for position independent code
only works when the callee was called (bound in GOT) before
the jump through GOT.
It does not work for some lazy bound Android tail callee.
We used special flag to disable all tail call optimization.
This change removes that global flag and depends on
clang compiler fix from upstream or the temporary patch in
https://android-review.googlesource.com/#/c/142241
See proposed llvm changes and pointers to llvm bugs in
https://android-review.googlesource.com/#/c/142792
Change-Id: Idb6e98d6b5f0e4c3676884b69eb3e767eba2491a
http://llvm.org/bugs/show_bug.cgi?id=15086,
llvm tail call optimization is wrong for x86.
For Android/x86 to use SSE* instructions safely, stack should be 16-byte
aligned before JNI function call, which isn't true for all x86 device,
so -mstackrealign should be the default.
BUG: 19234330
Change-Id: I4c6676366788772dbe64fd7f0dd33b3ed5c9b80e
Pass -fsanitize=address instead of manually specifying asan libraries
and other linker flags.
Note that we enable LOCAL_ALLOW_UNDEFINED_SYMBOLS by default for host
builds because ASAN only links symbols in the final executable, so
there will _always_ be undefined symbols in intermediate libraries.
Bug: 18208352
Change-Id: Ief55ab296e94974560eeb10507ec8d90f0025d5c
This change also defaults C compilations with clang to gnu99, and turns off
the -Winconsistent-missing-override flag, since we need further cleanup to
turn it on. -finline-functions is an unnecessary flag that only gcc needs.
Change-Id: I1cfcaedca32d9e3c248a6aa9460603205901c491
This will be necessary to support -std=gnu99 mode for clang 3.6, which
defaults to C11 mode (unlike prior releases that use C99).
Change-Id: Iea84582f9f12ba76b988463cbc0a20bd61042538
Bug: 18187181
Now that I checked in the latest prebuilts, clang will automatically use
color on terminals and no color on redirection to files or non-terminals.
Change-Id: I9be00c44947946cc18ce59c936b7f45d0ce2b6fc
Previously ccache is disabled when it fails calling clang's preprocessor with
unused arguments (such as '-Wa,--noexecstack') in the command line.
See http://petereisentraut.blogspot.com/2011/05/ccache-and-clang.html.
(-Qunused-arguments suppresses more than
-Wno-unused-command-line-argument does.)
Change-Id: I6cde307632c8395c053eb28063d7844d93070562
For mips target, gcc passes -KPIC to assembler by default,
but clang passes -KPIC only if -fPIC or -fpic is given.
BUG: 17895505
Change-Id: I7803adaf8ad7b40c6473872b67f4e0b4c502b38a
These aren't needed now that we only use the compiler/headers that exist in
the prebuilts/clang directory.
Change-Id: I9978efb10815e92577d45629db324e0a5094f880
This removes the hard-coded '3.5' in the various paths and makes switching to
a new toolchain easier from the command line (m LLVM_PREBUILTS_VERSION=3.6).
Change-Id: I46b10eb2fc177a03528de9c7b433f8647f632081
