Merge "Defend against -fstack-protector in libc startup."

We had these symbols incorrectly versioned as LIBC_PRIVATE
in M release. This change moves __aeabi* symbols from LIBC
to LIBC_N and adds __gnu_Unwind_Find_exidx to the list

Bug: https://github.com/android-ndk/ndk/issues/1
Change-Id: I0b353012adeacb00ae29ea10c63b9d1cf1cadbe7

.clang-format

@@ -0,0 +1,15 @@
+BasedOnStyle: Google
+AllowShortBlocksOnASingleLine: false
+AllowShortFunctionsOnASingleLine: false
+
+ColumnLimit: 100
+CommentPragmas: NOLINT:.*
+DerivePointerAlignment: false
+IndentWidth: 2
+ContinuationIndentWidth: 2
+PointerAlignment: Left
+TabWidth: 2
+UseTab: Never
+PenaltyExcessCharacter: 32
+
+Cpp11BracedListStyle: false

ABI-bugs.txt

@@ -1,15 +0,0 @@
-KNOWN ABI BUGS
---------------
-
-  time_t is 32-bit. http://b/5819737
-
-  off_t is 32-bit. There is off64_t, but no _FILE_OFFSET_BITS support.
-
-  sigset_t is too small on ARM and x86 (but correct on MIPS), so support
-  for real-time signals is broken. http://b/5828899
-
-  Too few TLS slots mean we can't allocate 128 pthread_key_t instances,
-  which POSIX says should be the minimum.
-
-  atexit(3) handlers registered by a shared library aren't called on
-  dlclose(3); this only affects ARM. http://b/4998315

Android.bp

@@ -0,0 +1 @@
+subdirs = ["*"]

CPPLINT.cfg

@@ -0,0 +1,2 @@
+set noparent
+filter=-build/header_guard,-runtime/int,-readability/function

CleanSpec.mk

@@ -44,6 +44,10 @@
 #$(call add-clean-step, find $(OUT_DIR) -type f -name "IGTalkSession*" -print0 | xargs -0 rm -f)
 #$(call add-clean-step, rm -rf $(PRODUCT_OUT)/data/*)
 
+# Switching to jemalloc requires deleting these files.
+$(call add-clean-step, rm -rf $(PRODUCT_OUT)/obj/STATIC_LIBRARIES/libc_*)
+$(call add-clean-step, rm -rf $(PRODUCT_OUT)/obj/SHARED_LIBRARIES/libc_*)
+
 # ************************************************
 # NEWER CLEAN STEPS MUST BE AT THE END OF THE LIST
 # ************************************************

README.md

@@ -0,0 +1,290 @@
+Working on bionic
+=================
+
+What are the big pieces of bionic?
+----------------------------------
+
+#### libc/ --- libc.so, libc.a
+
+The C library. Stuff like `fopen(3)` and `kill(2)`.
+
+#### libm/ --- libm.so, libm.a
+
+The math library. Traditionally Unix systems kept stuff like `sin(3)` and
+`cos(3)` in a separate library to save space in the days before shared
+libraries.
+
+#### libdl/ --- libdl.so
+
+The dynamic linker interface library. This is actually just a bunch of stubs
+that the dynamic linker replaces with pointers to its own implementation at
+runtime. This is where stuff like `dlopen(3)` lives.
+
+#### libstdc++/ --- libstdc++.so
+
+The C++ ABI support functions. The C++ compiler doesn't know how to implement
+thread-safe static initialization and the like, so it just calls functions that
+are supplied by the system. Stuff like `__cxa_guard_acquire` and
+`__cxa_pure_virtual` live here.
+
+#### linker/ --- /system/bin/linker and /system/bin/linker64
+
+The dynamic linker. When you run a dynamically-linked executable, its ELF file
+has a `DT_INTERP` entry that says "use the following program to start me".  On
+Android, that's either `linker` or `linker64` (depending on whether it's a
+32-bit or 64-bit executable). It's responsible for loading the ELF executable
+into memory and resolving references to symbols (so that when your code tries to
+jump to `fopen(3)`, say, it lands in the right place).
+
+#### tests/ --- unit tests
+
+The `tests/` directory contains unit tests. Roughly arranged as one file per
+publicly-exported header file.
+
+#### benchmarks/ --- benchmarks
+
+The `benchmarks/` directory contains benchmarks.
+
+
+What's in libc/?
+----------------
+
+<pre>
+libc/
+  arch-arm/
+  arch-arm64/
+  arch-common/
+  arch-mips/
+  arch-mips64/
+  arch-x86/
+  arch-x86_64/
+    # Each architecture has its own subdirectory for stuff that isn't shared
+    # because it's architecture-specific. There will be a .mk file in here that
+    # drags in all the architecture-specific files.
+    bionic/
+      # Every architecture needs a handful of machine-specific assembler files.
+      # They live here.
+    include/
+      machine/
+        # The majority of header files are actually in libc/include/, but many
+        # of them pull in a <machine/something.h> for things like limits,
+        # endianness, and how floating point numbers are represented. Those
+        # headers live here.
+    string/
+      # Most architectures have a handful of optional assembler files
+      # implementing optimized versions of various routines. The <string.h>
+      # functions are particular favorites.
+    syscalls/
+      # The syscalls directories contain script-generated assembler files.
+      # See 'Adding system calls' later.
+
+  include/
+    # The public header files on everyone's include path. These are a mixture of
+    # files written by us and files taken from BSD.
+
+  kernel/
+    # The kernel uapi header files. These are scrubbed copies of the originals
+    # in external/kernel-headers/. These files must not be edited directly. The
+    # generate_uapi_headers.sh script should be used to go from a kernel tree to
+    # external/kernel-headers/ --- this takes care of the architecture-specific
+    # details. The update_all.py script should be used to regenerate bionic's
+    # scrubbed headers from external/kernel-headers/.
+
+  private/
+    # These are private header files meant for use within bionic itself.
+
+  dns/
+    # Contains the DNS resolver (originates from NetBSD code).
+
+  upstream-dlmalloc/
+  upstream-freebsd/
+  upstream-netbsd/
+  upstream-openbsd/
+    # These directories contain unmolested upstream source. Any time we can
+    # just use a BSD implementation of something unmodified, we should.
+    # The structure under these directories mimics the upstream tree,
+    # but there's also...
+    android/
+      include/
+        # This is where we keep the hacks necessary to build BSD source
+        # in our world. The *-compat.h files are automatically included
+        # using -include, but we also provide equivalents for missing
+        # header/source files needed by the BSD implementation.
+
+  bionic/
+    # This is the biggest mess. The C++ files are files we own, typically
+    # because the Linux kernel interface is sufficiently different that we
+    # can't use any of the BSD implementations. The C files are usually
+    # legacy mess that needs to be sorted out, either by replacing it with
+    # current upstream source in one of the upstream directories or by
+    # switching the file to C++ and cleaning it up.
+
+  stdio/
+    # These are legacy files of dubious provenance. We're working to clean
+    # this mess up, and this directory should disappear.
+
+  tools/
+    # Various tools used to maintain bionic.
+
+  tzcode/
+    # A modified superset of the IANA tzcode. Most of the modifications relate
+    # to Android's use of a single file (with corresponding index) to contain
+    # time zone data.
+  zoneinfo/
+    # Android-format time zone data.
+    # See 'Updating tzdata' later.
+</pre>
+
+
+Adding system calls
+-------------------
+
+Adding a system call usually involves:
+
+  1. Add entries to SYSCALLS.TXT.
+     See SYSCALLS.TXT itself for documentation on the format.
+  2. Run the gensyscalls.py script.
+  3. Add constants (and perhaps types) to the appropriate header file.
+     Note that you should check to see whether the constants are already in
+     kernel uapi header files, in which case you just need to make sure that
+     the appropriate POSIX header file in libc/include/ includes the
+     relevant file or files.
+  4. Add function declarations to the appropriate header file.
+  5. Add at least basic tests. Even a test that deliberately supplies
+     an invalid argument helps check that we're generating the right symbol
+     and have the right declaration in the header file. (And strace(1) can
+     confirm that the correct system call is being made.)
+
+
+Updating kernel header files
+----------------------------
+
+As mentioned above, this is currently a two-step process:
+
+  1. Use generate_uapi_headers.sh to go from a Linux source tree to appropriate
+     contents for external/kernel-headers/.
+  2. Run update_all.py to scrub those headers and import them into bionic.
+
+
+Updating tzdata
+---------------
+
+This is fully automated (and these days handled by the libcore team, because
+they own icu, and that needs to be updated in sync with bionic):
+
+  1. Run update-tzdata.py in external/icu/tools/.
+
+
+Verifying changes
+-----------------
+
+If you make a change that is likely to have a wide effect on the tree (such as a
+libc header change), you should run `make checkbuild`. A regular `make` will
+_not_ build the entire tree; just the minimum number of projects that are
+required for the device. Tests, additional developer tools, and various other
+modules will not be built. Note that `make checkbuild` will not be complete
+either, as `make tests` covers a few additional modules, but generally speaking
+`make checkbuild` is enough.
+
+
+Running the tests
+-----------------
+
+The tests are all built from the tests/ directory.
+
+### Device tests
+
+    $ mma
+    $ adb remount
+    $ adb sync
+    $ adb shell /data/nativetest/bionic-unit-tests/bionic-unit-tests32
+    $ adb shell \
+        /data/nativetest/bionic-unit-tests-static/bionic-unit-tests-static32
+    # Only for 64-bit targets
+    $ adb shell /data/nativetest64/bionic-unit-tests/bionic-unit-tests64
+    $ adb shell \
+        /data/nativetest64/bionic-unit-tests-static/bionic-unit-tests-static64
+
+### Host tests
+
+The host tests require that you have `lunch`ed either an x86 or x86_64 target.
+
+    $ mma
+    $ mm bionic-unit-tests-run-on-host32
+    $ mm bionic-unit-tests-run-on-host64  # For 64-bit *targets* only.
+
+### Against glibc
+
+As a way to check that our tests do in fact test the correct behavior (and not
+just the behavior we think is correct), it is possible to run the tests against
+the host's glibc. The executables are already in your path.
+
+    $ mma
+    $ bionic-unit-tests-glibc32
+    $ bionic-unit-tests-glibc64
+
+
+Gathering test coverage
+-----------------------
+
+For either host or target coverage, you must first:
+
+ * `$ export NATIVE_COVERAGE=true`
+     * Note that the build system is ignorant to this flag being toggled, i.e. if
+       you change this flag, you will have to manually rebuild bionic.
+ * Set `bionic_coverage=true` in `libc/Android.mk` and `libm/Android.mk`.
+
+### Coverage from device tests
+
+    $ mma
+    $ adb sync
+    $ adb shell \
+        GCOV_PREFIX=/data/local/tmp/gcov \
+        GCOV_PREFIX_STRIP=`echo $ANDROID_BUILD_TOP | grep -o / | wc -l` \
+        /data/nativetest/bionic-unit-tests/bionic-unit-tests32
+    $ acov
+
+`acov` will pull all coverage information from the device, push it to the right
+directories, run `lcov`, and open the coverage report in your browser.
+
+### Coverage from host tests
+
+First, build and run the host tests as usual (see above).
+
+    $ croot
+    $ lcov -c -d $ANDROID_PRODUCT_OUT -o coverage.info
+    $ genhtml -o covreport coverage.info # or lcov --list coverage.info
+
+The coverage report is now available at `covreport/index.html`.
+
+
+Attaching GDB to the tests
+--------------------------
+
+Bionic's test runner will run each test in its own process by default to prevent
+tests failures from impacting other tests. This also has the added benefit of
+running them in parallel, so they are much faster.
+
+However, this also makes it difficult to run the tests under GDB. To prevent
+each test from being forked, run the tests with the flag `--no-isolate`.
+
+
+32-bit ABI bugs
+---------------
+
+This probably belongs in the NDK documentation rather than here, but these
+are the known ABI bugs in the 32-bit ABI:
+
+ * `time_t` is 32-bit. <http://b/5819737>. In the 64-bit ABI, time_t is
+   64-bit.
+
+ * `off_t` is 32-bit. There is `off64_t`, and in newer releases there is
+   almost-complete support for `_FILE_OFFSET_BITS`. Unfortunately our stdio
+   implementation uses 32-bit offsets and -- worse -- function pointers to
+   functions that use 32-bit offsets, so there's no good way to implement
+   the last few pieces <http://b/24807045>. In the 64-bit ABI, off_t is
+   off64_t.
+
+ * `sigset_t` is too small on ARM and x86 (but correct on MIPS), so support
+   for real-time signals is broken. <http://b/5828899> In the 64-bit ABI,
+   `sigset_t` is the correct size for every architecture.

benchmarks/Android.mk

@@ -0,0 +1,130 @@
+#
+# Copyright (C) 2013 The Android Open Source Project
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+LOCAL_PATH := $(call my-dir)
+
+# -----------------------------------------------------------------------------
+# Benchmarks library, usable by projects outside this directory.
+# -----------------------------------------------------------------------------
+
+benchmark_cflags := \
+    -O2 \
+    -fno-builtin \
+    -Wall \
+    -Wextra \
+    -Werror \
+    -Wunused \
+
+benchmark_cppflags := \
+
+benchmarklib_src_files := \
+    Benchmark.cpp \
+    utils.cpp \
+    main.cpp \
+
+include $(CLEAR_VARS)
+LOCAL_MODULE := libbenchmark
+LOCAL_CFLAGS := $(benchmark_cflags)
+LOCAL_CPPFLAGS := $(benchmark_cppflags)
+LOCAL_SRC_FILES := $(benchmarklib_src_files)
+LOCAL_C_INCLUDES := $(benchmark_c_includes)
+LOCAL_STATIC_LIBRARIES := libbase
+include $(BUILD_STATIC_LIBRARY)
+
+# Only supported on linux systems.
+ifeq ($(HOST_OS),linux)
+
+include $(CLEAR_VARS)
+LOCAL_MODULE := libbenchmark
+LOCAL_CFLAGS := $(benchmark_cflags)
+LOCAL_CPPFLAGS := $(benchmark_cppflags)
+LOCAL_SRC_FILES := $(benchmarklib_src_files)
+LOCAL_C_INCLUDES := $(benchmark_c_includes)
+LOCAL_MULTILIB := both
+LOCAL_STATIC_LIBRARIES := libbase
+include $(BUILD_HOST_STATIC_LIBRARY)
+
+endif
+
+# -----------------------------------------------------------------------------
+# Benchmarks.
+# -----------------------------------------------------------------------------
+benchmark_src_files := \
+    math_benchmark.cpp \
+    property_benchmark.cpp \
+    pthread_benchmark.cpp \
+    semaphore_benchmark.cpp \
+    stdio_benchmark.cpp \
+    string_benchmark.cpp \
+    time_benchmark.cpp \
+    unistd_benchmark.cpp \
+
+# Build benchmarks for the device (with bionic's .so). Run with:
+#   adb shell bionic-benchmarks32
+#   adb shell bionic-benchmarks64
+include $(CLEAR_VARS)
+LOCAL_MODULE := bionic-benchmarks
+LOCAL_MODULE_STEM_32 := bionic-benchmarks32
+LOCAL_MODULE_STEM_64 := bionic-benchmarks64
+LOCAL_MULTILIB := both
+LOCAL_CFLAGS := $(benchmark_cflags)
+LOCAL_CPPFLAGS := $(benchmark_cppflags)
+LOCAL_SRC_FILES := $(benchmark_src_files)
+LOCAL_STATIC_LIBRARIES := libbenchmark libbase
+include $(BUILD_EXECUTABLE)
+
+# We don't build a static benchmark executable because it's not usually
+# useful. If you're trying to run the current benchmarks on an older
+# release, it's (so far at least) always because you want to measure the
+# performance of the old release's libc, and a static benchmark isn't
+# going to let you do that.
+
+# Only supported on linux systems.
+ifeq ($(HOST_OS),linux)
+
+# Build benchmarks for the host (against glibc!). Run with:
+include $(CLEAR_VARS)
+LOCAL_MODULE := bionic-benchmarks-glibc
+LOCAL_MODULE_STEM_32 := bionic-benchmarks-glibc32
+LOCAL_MODULE_STEM_64 := bionic-benchmarks-glibc64
+LOCAL_MULTILIB := both
+LOCAL_CFLAGS := $(benchmark_cflags)
+LOCAL_CPPFLAGS := $(benchmark_cppflags)
+LOCAL_LDFLAGS := -lrt
+LOCAL_SRC_FILES := $(benchmark_src_files)
+LOCAL_STATIC_LIBRARIES := libbenchmark libbase
+include $(BUILD_HOST_EXECUTABLE)
+
+endif
+
+ifeq ($(HOST_OS)-$(HOST_ARCH),$(filter $(HOST_OS)-$(HOST_ARCH),linux-x86 linux-x86_64))
+include $(LOCAL_PATH)/../build/run-on-host.mk
+
+ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),x86 x86_64))
+bionic-benchmarks-run-on-host32: bionic-benchmarks bionic-prepare-run-on-host
+	ANDROID_DATA=$(TARGET_OUT_DATA) \
+	ANDROID_ROOT=$(TARGET_OUT) \
+		$(TARGET_OUT_EXECUTABLES)/bionic-benchmarks32 $(BIONIC_BENCHMARKS_FLAGS)
+endif
+
+ifeq ($(TARGET_IS_64_BIT),true)
+bionic-benchmarks-run-on-host64: bionic-benchmarks bionic-prepare-run-on-host
+	ANDROID_DATA=$(TARGET_OUT_DATA) \
+	ANDROID_ROOT=$(TARGET_OUT) \
+		$(TARGET_OUT_EXECUTABLES)/bionic-benchmarks64 $(BIONIC_BENCHMARKS_FLAGS)
+endif
+
+endif

benchmarks/Benchmark.cpp

@@ -0,0 +1,159 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <inttypes.h>
+#include <regex.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <time.h>
+
+#include <string>
+#include <vector>
+
+#include <android-base/stringprintf.h>
+
+#include <benchmark/Benchmark.h>
+
+#include "utils.h"
+
+namespace testing {
+
+static uint64_t NanoTime() {
+  struct timespec t;
+  t.tv_sec = t.tv_nsec = 0;
+  clock_gettime(CLOCK_MONOTONIC, &t);
+  return static_cast<uint64_t>(t.tv_sec) * 1000000000LL + t.tv_nsec;
+}
+
+bool Benchmark::header_printed_;
+
+std::vector<Benchmark*>& Benchmark::List() {
+  static std::vector<Benchmark*> list;
+  return list;
+}
+
+int Benchmark::MaxNameColumnWidth() {
+  size_t max = 20;
+  for (auto& benchmark : List()) {
+    max = std::max(max, benchmark->NameColumnWidth());
+  }
+  return static_cast<int>(max);
+}
+
+size_t Benchmark::RunAll(std::vector<regex_t*>& regs) {
+  size_t benchmarks_run = 0;
+  header_printed_ = false;
+  for (auto& benchmark : List()) {
+    benchmarks_run += benchmark->RunAllArgs(regs);
+  }
+  return benchmarks_run;
+}
+
+void Benchmark::PrintHeader() {
+  if (!header_printed_) {
+    printf("%-*s %10s %10s\n", MaxNameColumnWidth(), "", "iterations", "ns/op");
+    header_printed_ = true;
+  }
+}
+
+template <typename T>
+bool BenchmarkT<T>::ShouldRun(std::vector<regex_t*>& regs, T arg) {
+  if (regs.empty()) {
+    return true;
+  }
+
+  for (const auto& re : regs) {
+    if (regexec(re, GetNameStr(arg).c_str(), 0, NULL, 0) != REG_NOMATCH) {
+      return true;
+    }
+  }
+  return false;
+}
+
+void Benchmark::StopBenchmarkTiming() {
+  if (start_time_ns_ != 0) {
+    total_time_ns_ += NanoTime() - start_time_ns_;
+  }
+  start_time_ns_ = 0;
+}
+
+void Benchmark::StartBenchmarkTiming() {
+  if (start_time_ns_ == 0) {
+    start_time_ns_ = NanoTime();
+  }
+}
+
+std::string BenchmarkWithoutArg::GetNameStr(void*) {
+  return Name();
+}
+
+template <>
+std::string BenchmarkWithArg<int>::GetNameStr(int arg) {
+  return Name() + "/" + PrettyInt(arg, 2);
+}
+
+template <>
+std::string BenchmarkWithArg<double>::GetNameStr(double arg) {
+  return Name() + "/" + android::base::StringPrintf("%0.6f", arg);
+}
+
+template<typename T>
+void BenchmarkT<T>::RunWithArg(T arg) {
+  int new_iterations = 1;
+  int iterations;
+  while (new_iterations < 1e8) {
+    bytes_processed_ = 0;
+    total_time_ns_ = 0;
+    start_time_ns_ = 0;
+
+    iterations = new_iterations;
+    RunIterations(iterations, arg);
+    if (total_time_ns_ >= 1e9) {
+      break;
+    }
+
+    if (total_time_ns_/iterations == 0) {
+      new_iterations = 1e9;
+    } else {
+      new_iterations = 1e9/ (total_time_ns_/iterations);
+    }
+    new_iterations = std::max(iterations + 1,
+                          std::min(new_iterations + new_iterations/2, 100*iterations));
+
+    new_iterations = Round(new_iterations);
+  }
+
+  printf("%-*s %10s %10" PRId64, MaxNameColumnWidth(), GetNameStr(arg).c_str(),
+         PrettyInt(iterations, 10).c_str(), total_time_ns_/iterations);
+
+  if (total_time_ns_ > 0 && bytes_processed_ > 0) {
+    double gib_processed = static_cast<double>(bytes_processed_)/1e9;
+    double seconds = static_cast<double>(total_time_ns_)/1e9;
+    printf(" %8.3f GiB/s", gib_processed/seconds);
+  }
+  printf("\n");
+  fflush(stdout);
+}
+
+template class BenchmarkT<int>;
+template class BenchmarkT<double>;
+template class BenchmarkT<void*>;
+
+template class BenchmarkWithArg<int>;
+template class BenchmarkWithArg<double>;
+
+}  // namespace testing

benchmarks/benchmark/Benchmark.h

@@ -0,0 +1,169 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef BENCHMARKS_BENCHMARK_H_
+#define BENCHMARKS_BENCHMARK_H_
+
+#include <regex.h>
+#include <stdint.h>
+
+#include <string>
+#include <vector>
+
+namespace testing {
+
+class Benchmark {
+public:
+  Benchmark() {
+    List().push_back(this);
+  }
+  virtual ~Benchmark() {}
+
+  virtual std::string Name() = 0;
+
+  virtual size_t RunAllArgs(std::vector<regex_t*>&) = 0;
+
+  void SetBenchmarkBytesProcessed(uint64_t bytes) { bytes_processed_ += bytes; }
+  void StopBenchmarkTiming();
+  void StartBenchmarkTiming();
+
+  // Run all of the benchmarks that have registered.
+  static size_t RunAll(std::vector<regex_t*>&);
+
+  static std::vector<Benchmark*>& List();
+
+  static int MaxNameColumnWidth();
+
+protected:
+  virtual size_t NameColumnWidth() = 0;
+
+  uint64_t bytes_processed_;
+  uint64_t total_time_ns_;
+  uint64_t start_time_ns_;
+
+  static bool header_printed_;
+
+  static void PrintHeader();
+};
+
+template <typename T>
+class BenchmarkT : public Benchmark {
+public:
+  BenchmarkT() {}
+  virtual ~BenchmarkT() {}
+
+protected:
+  bool ShouldRun(std::vector<regex_t*>&, T arg);
+  void RunWithArg(T arg);
+  virtual void RunIterations(int, T) = 0;
+  virtual std::string GetNameStr(T) = 0;
+};
+
+class BenchmarkWithoutArg : public BenchmarkT<void*> {
+public:
+  BenchmarkWithoutArg() {}
+  virtual ~BenchmarkWithoutArg() {}
+
+protected:
+  virtual size_t RunAllArgs(std::vector<regex_t*>& regs) override {
+    size_t benchmarks_run = 0;
+    if (BenchmarkT<void*>::ShouldRun(regs, nullptr)) {
+      PrintHeader();
+      RunWithArg(nullptr);
+      benchmarks_run++;
+    }
+    return benchmarks_run;
+  }
+
+  virtual void RunIterations(int iters, void*) override {
+    Run(iters);
+  }
+
+  virtual void Run(int) = 0;
+
+  virtual size_t NameColumnWidth() override {
+    return Name().size();
+  }
+
+  virtual std::string GetNameStr(void *) override;
+};
+
+template<typename T>
+class BenchmarkWithArg : public BenchmarkT<T> {
+public:
+  BenchmarkWithArg() {}
+  virtual ~BenchmarkWithArg() {}
+
+  BenchmarkWithArg* Arg(T arg) {
+    args_.push_back(arg);
+    return this;
+  }
+
+protected:
+  virtual size_t NameColumnWidth() override {
+    size_t max = 0;
+    for (const auto& arg : args_) {
+      max = std::max(max, GetNameStr(arg).size());
+    }
+    return max;
+  }
+
+  std::string GetNameStr(T arg) override;
+
+  virtual size_t RunAllArgs(std::vector<regex_t*>& regs) override {
+    size_t benchmarks_run = 0;
+    for (T& arg : args_) {
+      if (BenchmarkT<T>::ShouldRun(regs, arg)) {
+        Benchmark::PrintHeader();
+        BenchmarkT<T>::RunWithArg(arg);
+        benchmarks_run++;
+      }
+    }
+    return benchmarks_run;
+  }
+
+  virtual void RunIterations(int iters, T arg) override {
+    Run(iters, arg);
+  }
+
+  virtual void Run(int iters, T arg) = 0;
+
+private:
+  std::vector<T> args_;
+};
+
+}  // namespace testing
+
+#define BENCHMARK_START(f, super_class) \
+  class f : public super_class { \
+  public: \
+    f() {} \
+    virtual ~f() {} \
+    virtual std::string Name() override { return #f; } \
+
+#define BENCHMARK_NO_ARG(f) \
+  BENCHMARK_START(f, ::testing::BenchmarkWithoutArg) \
+    virtual void Run(int) override; \
+  }; \
+  static ::testing::Benchmark* __benchmark_##f = new f()
+
+#define BENCHMARK_WITH_ARG(f, arg_type) \
+  BENCHMARK_START(f, ::testing::BenchmarkWithArg<arg_type>) \
+    virtual void Run(int, arg_type) override; \
+  }; \
+  static ::testing::BenchmarkWithArg<arg_type>* __benchmark_##f = (new f())
+
+#endif  // BENCHMARKS_BENCHMARK_H_

benchmarks/main.cpp

@@ -0,0 +1,61 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+
+#include <regex.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <vector>
+
+#include <benchmark/Benchmark.h>
+
+int main(int argc, char* argv[]) {
+  if (::testing::Benchmark::List().empty()) {
+    fprintf(stderr, "No benchmarks registered!\n");
+    exit(EXIT_FAILURE);
+  }
+
+  std::vector<regex_t*> regs;
+  for (int i = 1; i < argc; i++) {
+    regex_t* re = new regex_t;
+    int errcode = regcomp(re, argv[i], 0);
+    if (errcode != 0) {
+      size_t errbuf_size = regerror(errcode, re, NULL, 0);
+      if (errbuf_size > 0) {
+        char* errbuf = new char[errbuf_size];
+        regerror(errcode, re, errbuf, errbuf_size);
+        fprintf(stderr, "Couldn't compile \"%s\" as a regular expression: %s\n",
+                argv[i], errbuf);
+      } else {
+        fprintf(stderr, "Unknown compile error for \"%s\" as a regular expression!\n", argv[i]);
+      }
+      exit(EXIT_FAILURE);
+    }
+    regs.push_back(re);
+  }
+
+  if (::testing::Benchmark::RunAll(regs) == 0) {
+    fprintf(stderr, "No matching benchmarks!\n");
+    fprintf(stderr, "Available benchmarks:\n");
+    for (const auto& benchmark : ::testing::Benchmark::List()) {
+      fprintf(stderr, "  %s\n", benchmark->Name().c_str());
+    }
+    exit(EXIT_FAILURE);
+  }
+
+  return 0;
+}

benchmarks/math_benchmark.cpp

@@ -0,0 +1,286 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <fenv.h>
+#include <math.h>
+
+#include <benchmark/Benchmark.h>
+
+#define AT_COMMON_VALS \
+    Arg(1234.0)->Arg(nan(""))->Arg(HUGE_VAL)->Arg(0.0)
+
+// Avoid optimization.
+volatile double d;
+volatile double v;
+
+BENCHMARK_NO_ARG(BM_math_sqrt);
+void BM_math_sqrt::Run(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = 2.0;
+  for (int i = 0; i < iters; ++i) {
+    d += sqrt(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_math_log10);
+void BM_math_log10::Run(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = 1234.0;
+  for (int i = 0; i < iters; ++i) {
+    d += log10(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_math_logb);
+void BM_math_logb::Run(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = 1234.0;
+  for (int i = 0; i < iters; ++i) {
+    d += logb(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_WITH_ARG(BM_math_isfinite_macro, double)->AT_COMMON_VALS;
+void BM_math_isfinite_macro::Run(int iters, double value) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = value;
+  for (int i = 0; i < iters; ++i) {
+    d += isfinite(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+#if defined(__BIONIC__)
+#define test_isfinite __isfinite
+#else
+#define test_isfinite __finite
+#endif
+BENCHMARK_WITH_ARG(BM_math_isfinite, double)->AT_COMMON_VALS;
+void BM_math_isfinite::Run(int iters, double value) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = value;
+  for (int i = 0; i < iters; ++i) {
+    d += test_isfinite(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_WITH_ARG(BM_math_isinf_macro, double)->AT_COMMON_VALS;
+void BM_math_isinf_macro::Run(int iters, double value) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = value;
+  for (int i = 0; i < iters; ++i) {
+    d += isinf(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_WITH_ARG(BM_math_isinf, double)->AT_COMMON_VALS;
+void BM_math_isinf::Run(int iters, double value) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = value;
+  for (int i = 0; i < iters; ++i) {
+    d += (isinf)(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_WITH_ARG(BM_math_isnan_macro, double)->AT_COMMON_VALS;
+void BM_math_isnan_macro::Run(int iters, double value) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = value;
+  for (int i = 0; i < iters; ++i) {
+    d += isnan(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_WITH_ARG(BM_math_isnan, double)->AT_COMMON_VALS;
+void BM_math_isnan::Run(int iters, double value) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = value;
+  for (int i = 0; i < iters; ++i) {
+    d += (isnan)(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_WITH_ARG(BM_math_isnormal_macro, double)->AT_COMMON_VALS;
+void BM_math_isnormal_macro::Run(int iters, double value) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = value;
+  for (int i = 0; i < iters; ++i) {
+    d += isnormal(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+#if defined(__BIONIC__)
+BENCHMARK_WITH_ARG(BM_math_isnormal, double)->AT_COMMON_VALS;
+void BM_math_isnormal::Run(int iters, double value) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = value;
+  for (int i = 0; i < iters; ++i) {
+    d += (__isnormal)(v);
+  }
+
+  StopBenchmarkTiming();
+}
+#endif
+
+BENCHMARK_NO_ARG(BM_math_sin_fast);
+void BM_math_sin_fast::Run(int iters) {
+  StartBenchmarkTiming();
+
+  d = 1.0;
+  for (int i = 0; i < iters; ++i) {
+    d += sin(d);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_math_sin_feupdateenv);
+void BM_math_sin_feupdateenv::Run(int iters) {
+  StartBenchmarkTiming();
+
+  d = 1.0;
+  for (int i = 0; i < iters; ++i) {
+    fenv_t __libc_save_rm;
+    feholdexcept(&__libc_save_rm);
+    fesetround(FE_TONEAREST);
+    d += sin(d);
+    feupdateenv(&__libc_save_rm);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_math_sin_fesetenv);
+void BM_math_sin_fesetenv::Run(int iters) {
+  StartBenchmarkTiming();
+
+  d = 1.0;
+  for (int i = 0; i < iters; ++i) {
+    fenv_t __libc_save_rm;
+    feholdexcept(&__libc_save_rm);
+    fesetround(FE_TONEAREST);
+    d += sin(d);
+    fesetenv(&__libc_save_rm);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_WITH_ARG(BM_math_fpclassify, double)->AT_COMMON_VALS;
+void BM_math_fpclassify::Run(int iters, double value) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = value;
+  for (int i = 0; i < iters; ++i) {
+    d += fpclassify(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_WITH_ARG(BM_math_signbit_macro, double)->AT_COMMON_VALS;
+void BM_math_signbit_macro::Run(int iters, double value) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = value;
+  for (int i = 0; i < iters; ++i) {
+    d += signbit(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_WITH_ARG(BM_math_signbit, double)->AT_COMMON_VALS;
+void BM_math_signbit::Run(int iters, double value) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = value;
+  for (int i = 0; i < iters; ++i) {
+    d += (__signbit)(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_WITH_ARG(BM_math_fabs_macro, double)->AT_COMMON_VALS;
+void BM_math_fabs_macro::Run(int iters, double value) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = value;
+  for (int i = 0; i < iters; ++i) {
+    d += fabs(v);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_WITH_ARG(BM_math_fabs, double)->AT_COMMON_VALS;
+void BM_math_fabs::Run(int iters, double value) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = value;
+  for (int i = 0; i < iters; ++i) {
+    d += (fabs)(v);
+  }
+
+  StopBenchmarkTiming();
+}

benchmarks/property_benchmark.cpp

@@ -0,0 +1,233 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <errno.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+
+#include <string>
+
+#if defined(__BIONIC__)
+
+#define _REALLY_INCLUDE_SYS__SYSTEM_PROPERTIES_H_
+#include <sys/_system_properties.h>
+
+#include <benchmark/Benchmark.h>
+
+extern void* __system_property_area__;
+
+// Do not exceed 512, that is about the largest number of properties
+// that can be created with the current property area size.
+#define TEST_NUM_PROPS \
+    Arg(1)->Arg(4)->Arg(16)->Arg(64)->Arg(128)->Arg(256)->Arg(512)
+
+struct LocalPropertyTestState {
+  LocalPropertyTestState(int nprops) : nprops(nprops), valid(false) {
+    static const char prop_name_chars[] = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ-_.";
+
+    const char* android_data = getenv("ANDROID_DATA");
+    if (android_data == NULL) {
+      printf("ANDROID_DATA environment variable not set\n");
+      return;
+    }
+    char dir_template[PATH_MAX];
+    snprintf(dir_template, sizeof(dir_template), "%s/local/tmp/prop-XXXXXX", android_data);
+    char* dirname = mkdtemp(dir_template);
+    if (!dirname) {
+      printf("making temp file for test state failed (is %s/local/tmp writable?): %s\n",
+             android_data, strerror(errno));
+      return;
+    }
+
+    old_pa = __system_property_area__;
+    __system_property_area__ = NULL;
+
+    pa_dirname = dirname;
+    pa_filename = pa_dirname + "/__properties__";
+
+    __system_property_set_filename(pa_filename.c_str());
+    __system_property_area_init();
+
+    names = new char* [nprops];
+    name_lens = new int[nprops];
+    values = new char* [nprops];
+    value_lens = new int[nprops];
+
+    srandom(nprops);
+
+    for (int i = 0; i < nprops; i++) {
+      // Make sure the name has at least 10 characters to make
+      // it very unlikely to generate the same random name.
+      name_lens[i] = (random() % (PROP_NAME_MAX - 10)) + 10;
+      names[i] = new char[PROP_NAME_MAX + 1];
+      size_t prop_name_len = sizeof(prop_name_chars) - 1;
+      for (int j = 0; j < name_lens[i]; j++) {
+        if (j == 0 || names[i][j-1] == '.' || j == name_lens[i] - 1) {
+          // Certain values are not allowed:
+          // - Don't start name with '.'
+          // - Don't allow '.' to appear twice in a row
+          // - Don't allow the name to end with '.'
+          // This assumes that '.' is the last character in the
+          // array so that decrementing the length by one removes
+          // the value from the possible values.
+          prop_name_len--;
+        }
+        names[i][j] = prop_name_chars[random() % prop_name_len];
+      }
+      names[i][name_lens[i]] = 0;
+
+      // Make sure the value contains at least 1 character.
+      value_lens[i] = (random() % (PROP_VALUE_MAX - 1)) + 1;
+      values[i] = new char[PROP_VALUE_MAX];
+      for (int j = 0; j < value_lens[i]; j++) {
+        values[i][j] = prop_name_chars[random() % (sizeof(prop_name_chars) - 1)];
+      }
+
+      if (__system_property_add(names[i], name_lens[i], values[i], value_lens[i]) < 0) {
+        printf("Failed to add a property, terminating...\n");
+        printf("%s = %.*s\n", names[i], value_lens[i], values[i]);
+        exit(1);
+      }
+    }
+
+    valid = true;
+  }
+
+  ~LocalPropertyTestState() {
+    if (!valid)
+      return;
+
+    __system_property_area__ = old_pa;
+
+    __system_property_set_filename(PROP_FILENAME);
+    unlink(pa_filename.c_str());
+    rmdir(pa_dirname.c_str());
+
+    for (int i = 0; i < nprops; i++) {
+      delete names[i];
+      delete values[i];
+    }
+    delete[] names;
+    delete[] name_lens;
+    delete[] values;
+    delete[] value_lens;
+  }
+public:
+  const int nprops;
+  char** names;
+  int* name_lens;
+  char** values;
+  int* value_lens;
+  bool valid;
+
+private:
+  std::string pa_dirname;
+  std::string pa_filename;
+  void* old_pa;
+};
+
+BENCHMARK_WITH_ARG(BM_property_get, int)->TEST_NUM_PROPS;
+void BM_property_get::Run(int iters, int nprops) {
+  StopBenchmarkTiming();
+
+  LocalPropertyTestState pa(nprops);
+  char value[PROP_VALUE_MAX];
+
+  if (!pa.valid)
+    return;
+
+  srandom(iters * nprops);
+
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; i++) {
+    __system_property_get(pa.names[random() % nprops], value);
+  }
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_WITH_ARG(BM_property_find, int)->TEST_NUM_PROPS;
+void BM_property_find::Run(int iters, int nprops) {
+  StopBenchmarkTiming();
+
+  LocalPropertyTestState pa(nprops);
+
+  if (!pa.valid)
+    return;
+
+  srandom(iters * nprops);
+
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; i++) {
+    __system_property_find(pa.names[random() % nprops]);
+  }
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_WITH_ARG(BM_property_read, int)->TEST_NUM_PROPS;
+void BM_property_read::Run(int iters, int nprops) {
+  StopBenchmarkTiming();
+
+  LocalPropertyTestState pa(nprops);
+
+  if (!pa.valid)
+    return;
+
+  srandom(iters * nprops);
+  const prop_info** pinfo = new const prop_info*[iters];
+  char propvalue[PROP_VALUE_MAX];
+
+  for (int i = 0; i < iters; i++) {
+    pinfo[i] = __system_property_find(pa.names[random() % nprops]);
+  }
+
+  StartBenchmarkTiming();
+  for (int i = 0; i < iters; i++) {
+    __system_property_read(pinfo[i], 0, propvalue);
+  }
+  StopBenchmarkTiming();
+
+  delete[] pinfo;
+}
+
+BENCHMARK_WITH_ARG(BM_property_serial, int)->TEST_NUM_PROPS;
+void BM_property_serial::Run(int iters, int nprops) {
+  StopBenchmarkTiming();
+
+  LocalPropertyTestState pa(nprops);
+
+  if (!pa.valid)
+    return;
+
+  srandom(iters * nprops);
+  const prop_info** pinfo = new const prop_info*[iters];
+
+  for (int i = 0; i < iters; i++) {
+    pinfo[i] = __system_property_find(pa.names[random() % nprops]);
+  }
+
+  StartBenchmarkTiming();
+  for (int i = 0; i < iters; i++) {
+    __system_property_serial(pinfo[i]);
+  }
+  StopBenchmarkTiming();
+
+  delete[] pinfo;
+}
+
+#endif  // __BIONIC__

benchmarks/pthread_benchmark.cpp

@@ -0,0 +1,233 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <pthread.h>
+
+#include <benchmark/Benchmark.h>
+
+// Stop GCC optimizing out our pure function.
+/* Must not be static! */ pthread_t (*pthread_self_fp)() = pthread_self;
+
+BENCHMARK_NO_ARG(BM_pthread_self);
+void BM_pthread_self::Run(int iters) {
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_self_fp();
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_pthread_getspecific);
+void BM_pthread_getspecific::Run(int iters) {
+  StopBenchmarkTiming();
+  pthread_key_t key;
+  pthread_key_create(&key, NULL);
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_getspecific(key);
+  }
+
+  StopBenchmarkTiming();
+  pthread_key_delete(key);
+}
+
+BENCHMARK_NO_ARG(BM_pthread_setspecific);
+void BM_pthread_setspecific::Run(int iters) {
+  StopBenchmarkTiming();
+  pthread_key_t key;
+  pthread_key_create(&key, NULL);
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_setspecific(key, NULL);
+  }
+
+  StopBenchmarkTiming();
+  pthread_key_delete(key);
+}
+
+static void DummyPthreadOnceInitFunction() {
+}
+
+BENCHMARK_NO_ARG(BM_pthread_once);
+void BM_pthread_once::Run(int iters) {
+  StopBenchmarkTiming();
+  pthread_once_t once = PTHREAD_ONCE_INIT;
+  pthread_once(&once, DummyPthreadOnceInitFunction);
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_once(&once, DummyPthreadOnceInitFunction);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_pthread_mutex_lock);
+void BM_pthread_mutex_lock::Run(int iters) {
+  StopBenchmarkTiming();
+  pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_mutex_lock(&mutex);
+    pthread_mutex_unlock(&mutex);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_pthread_mutex_lock_ERRORCHECK);
+void BM_pthread_mutex_lock_ERRORCHECK::Run(int iters) {
+  StopBenchmarkTiming();
+  pthread_mutex_t mutex = PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP;
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_mutex_lock(&mutex);
+    pthread_mutex_unlock(&mutex);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_pthread_mutex_lock_RECURSIVE);
+void BM_pthread_mutex_lock_RECURSIVE::Run(int iters) {
+  StopBenchmarkTiming();
+  pthread_mutex_t mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_mutex_lock(&mutex);
+    pthread_mutex_unlock(&mutex);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_pthread_rwlock_read);
+void BM_pthread_rwlock_read::Run(int iters) {
+  StopBenchmarkTiming();
+  pthread_rwlock_t lock;
+  pthread_rwlock_init(&lock, NULL);
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_rwlock_rdlock(&lock);
+    pthread_rwlock_unlock(&lock);
+  }
+
+  StopBenchmarkTiming();
+  pthread_rwlock_destroy(&lock);
+}
+
+BENCHMARK_NO_ARG(BM_pthread_rwlock_write);
+void BM_pthread_rwlock_write::Run(int iters) {
+  StopBenchmarkTiming();
+  pthread_rwlock_t lock;
+  pthread_rwlock_init(&lock, NULL);
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_rwlock_wrlock(&lock);
+    pthread_rwlock_unlock(&lock);
+  }
+
+  StopBenchmarkTiming();
+  pthread_rwlock_destroy(&lock);
+}
+
+static void* IdleThread(void*) {
+  return NULL;
+}
+
+BENCHMARK_NO_ARG(BM_pthread_create);
+void BM_pthread_create::Run(int iters) {
+  StopBenchmarkTiming();
+  pthread_t thread;
+
+  for (int i = 0; i < iters; ++i) {
+    StartBenchmarkTiming();
+    pthread_create(&thread, NULL, IdleThread, NULL);
+    StopBenchmarkTiming();
+    pthread_join(thread, NULL);
+  }
+}
+
+static void* RunThread(void* arg) {
+  ::testing::Benchmark* benchmark = reinterpret_cast<::testing::Benchmark*>(arg);
+  benchmark->StopBenchmarkTiming();
+  return NULL;
+}
+
+BENCHMARK_NO_ARG(BM_pthread_create_and_run);
+void BM_pthread_create_and_run::Run(int iters) {
+  StopBenchmarkTiming();
+  pthread_t thread;
+
+  for (int i = 0; i < iters; ++i) {
+    StartBenchmarkTiming();
+    pthread_create(&thread, NULL, RunThread, this);
+    pthread_join(thread, NULL);
+  }
+}
+
+static void* ExitThread(void* arg) {
+  ::testing::Benchmark* benchmark = reinterpret_cast<::testing::Benchmark*>(arg);
+  benchmark->StartBenchmarkTiming();
+  pthread_exit(NULL);
+}
+
+BENCHMARK_NO_ARG(BM_pthread_exit_and_join);
+void BM_pthread_exit_and_join::Run(int iters) {
+  StopBenchmarkTiming();
+  pthread_t thread;
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_create(&thread, NULL, ExitThread, this);
+    pthread_join(thread, NULL);
+    StopBenchmarkTiming();
+  }
+}
+
+BENCHMARK_NO_ARG(BM_pthread_key_create);
+void BM_pthread_key_create::Run(int iters) {
+  StopBenchmarkTiming();
+  pthread_key_t key;
+
+  for (int i = 0; i < iters; ++i) {
+    StartBenchmarkTiming();
+    pthread_key_create(&key, NULL);
+    StopBenchmarkTiming();
+    pthread_key_delete(key);
+  }
+}
+
+BENCHMARK_NO_ARG(BM_pthread_key_delete);
+void BM_pthread_key_delete::Run(int iters) {
+  StopBenchmarkTiming();
+  pthread_key_t key;
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_key_create(&key, NULL);
+    StartBenchmarkTiming();
+    pthread_key_delete(key);
+    StopBenchmarkTiming();
+  }
+}

benchmarks/semaphore_benchmark.cpp

@@ -0,0 +1,143 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <pthread.h>
+#include <semaphore.h>
+#include <stdatomic.h>
+#include <stdio.h>
+
+#include <benchmark/Benchmark.h>
+
+BENCHMARK_NO_ARG(BM_semaphore_sem_getvalue);
+void BM_semaphore_sem_getvalue::Run(int iters) {
+  StopBenchmarkTiming();
+  sem_t semaphore;
+  sem_init(&semaphore, 1, 1);
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    int dummy;
+    sem_getvalue(&semaphore, &dummy);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_semaphore_sem_wait_sem_post);
+void BM_semaphore_sem_wait_sem_post::Run(int iters) {
+  StopBenchmarkTiming();
+  sem_t semaphore;
+  sem_init(&semaphore, 1, 1);
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    sem_wait(&semaphore);
+    sem_post(&semaphore);
+  }
+
+  StopBenchmarkTiming();
+}
+
+/*
+ *    This test reports the overhead of the underlying futex wake syscall on
+ * the producer. It does not report the overhead from issuing the wake to the
+ * point where the posted consumer thread wakes up. It suffers from
+ * clock_gettime syscall overhead. Lock the CPU speed for consistent results
+ * as we may not reach >50% cpu utilization.
+ *
+ *    We will run a background thread that catches the sem_post wakeup and
+ * loops immediately returning back to sleep in sem_wait for the next one. This
+ * thread is run with policy SCHED_OTHER (normal policy), a middle policy.
+ *
+ *    The primary thread will run at SCHED_IDLE (lowest priority policy) when
+ * monitoring the background thread to detect when it hits sem_wait sleep. It
+ * will do so with no clock running. Once we are ready, we will switch to
+ * SCHED_FIFO (highest priority policy) to time the act of running sem_post
+ * with the benchmark clock running. This ensures nothing else in the system
+ * can preempt our timed activity, including the background thread. We are
+ * also protected with the scheduling policy of letting a process hit a
+ * resource limit rather than get hit with a context switch.
+ *
+ *    The background thread will start executing either on another CPU, or
+ * after we back down from SCHED_FIFO, but certainly not in the context of
+ * the timing of the sem_post.
+ */
+static atomic_int BM_semaphore_sem_post_running;
+
+static void *BM_semaphore_sem_post_start_thread(void *obj) {
+    sem_t *semaphore = reinterpret_cast<sem_t *>(obj);
+
+    while ((BM_semaphore_sem_post_running > 0) && !sem_wait(semaphore)) {
+        ;
+    }
+    BM_semaphore_sem_post_running = -1;
+    return NULL;
+}
+
+BENCHMARK_NO_ARG(BM_semaphore_sem_post);
+void BM_semaphore_sem_post::Run(int iters) {
+  StopBenchmarkTiming();
+
+  sem_t semaphore;
+  sem_init(&semaphore, 0, 0);
+
+  pthread_attr_t attr;
+  pthread_attr_init(&attr);
+  BM_semaphore_sem_post_running = 1;
+  struct sched_param param = { 0, };
+  pthread_attr_setschedparam(&attr, &param);
+  pthread_attr_setschedpolicy(&attr, SCHED_OTHER);
+  pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
+  pthread_t pthread;
+  pthread_create(&pthread, &attr, BM_semaphore_sem_post_start_thread, &semaphore);
+  pthread_attr_destroy(&attr);
+
+  sched_setscheduler((pid_t)0, SCHED_IDLE, &param);
+  for (int i = 0; i < iters; ++i) {
+    int trys = 3, dummy = 0;
+    do {
+      if (BM_semaphore_sem_post_running < 0) {
+        sched_setscheduler((pid_t)0, SCHED_OTHER, &param);
+        fprintf(stderr, "BM_semaphore_sem_post: start_thread died unexpectedly\n");
+        return;
+      }
+      sched_yield();
+      sem_getvalue(&semaphore, &dummy);
+      if (dummy < 0) {  // POSIX.1-2001 possibility 1
+        break;
+      }
+      if (dummy == 0) { // POSIX.1-2001 possibility 2
+        --trys;
+      }
+    } while (trys);
+    param.sched_priority = 1;
+    sched_setscheduler((pid_t)0, SCHED_FIFO, &param);
+    StartBenchmarkTiming();
+    sem_post(&semaphore);
+    StopBenchmarkTiming(); // Remember to subtract clock syscall overhead
+    param.sched_priority = 0;
+    sched_setscheduler((pid_t)0, SCHED_IDLE, &param);
+  }
+  sched_setscheduler((pid_t)0, SCHED_OTHER, &param);
+
+  if (BM_semaphore_sem_post_running > 0) {
+    BM_semaphore_sem_post_running = 0;
+  }
+  do {
+    sem_post(&semaphore);
+    sched_yield();
+  } while (!BM_semaphore_sem_post_running);
+}

benchmarks/stdio_benchmark.cpp

@@ -0,0 +1,94 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <stdio.h>
+#include <stdio_ext.h>
+#include <stdlib.h>
+
+#include <benchmark/Benchmark.h>
+
+#define KB 1024
+#define MB 1024*KB
+
+#define AT_COMMON_SIZES \
+    Arg(1)->Arg(2)->Arg(3)->Arg(4)->Arg(8)->Arg(16)->Arg(32)->Arg(64)->Arg(512)-> \
+    Arg(1*KB)->Arg(4*KB)->Arg(8*KB)->Arg(16*KB)->Arg(64*KB)
+
+template <typename Fn>
+void ReadWriteTest(::testing::Benchmark* benchmark, int iters, int chunk_size, Fn f, bool buffered) {
+  benchmark->StopBenchmarkTiming();
+  FILE* fp = fopen("/dev/zero", "rw");
+  __fsetlocking(fp, FSETLOCKING_BYCALLER);
+  char* buf = new char[chunk_size];
+  benchmark->StartBenchmarkTiming();
+
+  if (!buffered) {
+    setvbuf(fp, 0, _IONBF, 0);
+  }
+
+  for (int i = 0; i < iters; ++i) {
+    f(buf, chunk_size, 1, fp);
+  }
+
+  benchmark->StopBenchmarkTiming();
+  benchmark->SetBenchmarkBytesProcessed(int64_t(iters) * int64_t(chunk_size));
+  delete[] buf;
+  fclose(fp);
+}
+
+BENCHMARK_WITH_ARG(BM_stdio_fread, int)->AT_COMMON_SIZES;
+void BM_stdio_fread::Run(int iters, int chunk_size) {
+  ReadWriteTest(this, iters, chunk_size, fread, true);
+}
+
+BENCHMARK_WITH_ARG(BM_stdio_fwrite, int)->AT_COMMON_SIZES;
+void BM_stdio_fwrite::Run(int iters, int chunk_size) {
+  ReadWriteTest(this, iters, chunk_size, fwrite, true);
+}
+
+BENCHMARK_WITH_ARG(BM_stdio_fread_unbuffered, int)->AT_COMMON_SIZES;
+void BM_stdio_fread_unbuffered::Run(int iters, int chunk_size) {
+  ReadWriteTest(this, iters, chunk_size, fread, false);
+}
+
+BENCHMARK_WITH_ARG(BM_stdio_fwrite_unbuffered, int)->AT_COMMON_SIZES;
+void BM_stdio_fwrite_unbuffered::Run(int iters, int chunk_size) {
+  ReadWriteTest(this, iters, chunk_size, fwrite, false);
+}
+
+static void FopenFgetsFclose(int iters, bool no_locking) {
+  char buf[1024];
+  for (int i = 0; i < iters; ++i) {
+    FILE* fp = fopen("/proc/version", "re");
+    if (no_locking) __fsetlocking(fp, FSETLOCKING_BYCALLER);
+    if (fgets(buf, sizeof(buf), fp) == nullptr) abort();
+    fclose(fp);
+  }
+}
+
+BENCHMARK_NO_ARG(BM_stdio_fopen_fgets_fclose_locking);
+void BM_stdio_fopen_fgets_fclose_locking::Run(int iters) {
+  StartBenchmarkTiming();
+  FopenFgetsFclose(iters, false);
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_stdio_fopen_fgets_fclose_no_locking);
+void BM_stdio_fopen_fgets_fclose_no_locking::Run(int iters) {
+  StartBenchmarkTiming();
+  FopenFgetsFclose(iters, true);
+  StopBenchmarkTiming();
+}

benchmarks/string_benchmark.cpp

@@ -0,0 +1,113 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <stdint.h>
+#include <string.h>
+
+#include <benchmark/Benchmark.h>
+
+#define KB 1024
+#define MB 1024*KB
+
+#define AT_COMMON_SIZES \
+    Arg(8)->Arg(64)->Arg(512)->Arg(1*KB)->Arg(8*KB)->Arg(16*KB)->Arg(32*KB)->Arg(64*KB)
+
+// TODO: test unaligned operation too? (currently everything will be 8-byte aligned by malloc.)
+
+BENCHMARK_WITH_ARG(BM_string_memcmp, int)->AT_COMMON_SIZES;
+void BM_string_memcmp::Run(int iters, int nbytes) {
+  StopBenchmarkTiming();
+  char* src = new char[nbytes]; char* dst = new char[nbytes];
+  memset(src, 'x', nbytes);
+  memset(dst, 'x', nbytes);
+  StartBenchmarkTiming();
+
+  volatile int c __attribute__((unused)) = 0;
+  for (int i = 0; i < iters; ++i) {
+    c += memcmp(dst, src, nbytes);
+  }
+
+  StopBenchmarkTiming();
+  SetBenchmarkBytesProcessed(uint64_t(iters) * uint64_t(nbytes));
+  delete[] src;
+  delete[] dst;
+}
+
+BENCHMARK_WITH_ARG(BM_string_memcpy, int)->AT_COMMON_SIZES;
+void BM_string_memcpy::Run(int iters, int nbytes) {
+  StopBenchmarkTiming();
+  char* src = new char[nbytes]; char* dst = new char[nbytes];
+  memset(src, 'x', nbytes);
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    memcpy(dst, src, nbytes);
+  }
+
+  StopBenchmarkTiming();
+  SetBenchmarkBytesProcessed(uint64_t(iters) * uint64_t(nbytes));
+  delete[] src;
+  delete[] dst;
+}
+
+BENCHMARK_WITH_ARG(BM_string_memmove, int)->AT_COMMON_SIZES;
+void BM_string_memmove::Run(int iters, int nbytes) {
+  StopBenchmarkTiming();
+  char* buf = new char[nbytes + 64];
+  memset(buf, 'x', nbytes + 64);
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    memmove(buf, buf + 1, nbytes); // Worst-case overlap.
+  }
+
+  StopBenchmarkTiming();
+  SetBenchmarkBytesProcessed(uint64_t(iters) * uint64_t(nbytes));
+  delete[] buf;
+}
+
+BENCHMARK_WITH_ARG(BM_string_memset, int)->AT_COMMON_SIZES;
+void BM_string_memset::Run(int iters, int nbytes) {
+  StopBenchmarkTiming();
+  char* dst = new char[nbytes];
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    memset(dst, 0, nbytes);
+  }
+
+  StopBenchmarkTiming();
+  SetBenchmarkBytesProcessed(uint64_t(iters) * uint64_t(nbytes));
+  delete[] dst;
+}
+
+BENCHMARK_WITH_ARG(BM_string_strlen, int)->AT_COMMON_SIZES;
+void BM_string_strlen::Run(int iters, int nbytes) {
+  StopBenchmarkTiming();
+  char* s = new char[nbytes];
+  memset(s, 'x', nbytes);
+  s[nbytes - 1] = 0;
+  StartBenchmarkTiming();
+
+  volatile int c __attribute__((unused)) = 0;
+  for (int i = 0; i < iters; ++i) {
+    c += strlen(s);
+  }
+
+  StopBenchmarkTiming();
+  SetBenchmarkBytesProcessed(uint64_t(iters) * uint64_t(nbytes));
+  delete[] s;
+}

benchmarks/time_benchmark.cpp

@@ -0,0 +1,81 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <sys/syscall.h>
+#include <sys/time.h>
+#include <time.h>
+#include <unistd.h>
+
+#include <benchmark/Benchmark.h>
+
+BENCHMARK_NO_ARG(BM_time_clock_gettime);
+void BM_time_clock_gettime::Run(int iters) {
+  StartBenchmarkTiming();
+
+  timespec t;
+  for (int i = 0; i < iters; ++i) {
+    clock_gettime(CLOCK_MONOTONIC, &t);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_time_clock_gettime_syscall);
+void BM_time_clock_gettime_syscall::Run(int iters) {
+  StartBenchmarkTiming();
+
+  timespec t;
+  for (int i = 0; i < iters; ++i) {
+    syscall(__NR_clock_gettime, CLOCK_MONOTONIC, &t);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_time_gettimeofday);
+void BM_time_gettimeofday::Run(int iters) {
+  StartBenchmarkTiming();
+
+  timeval tv;
+  for (int i = 0; i < iters; ++i) {
+    gettimeofday(&tv, NULL);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_time_gettimeofday_syscall);
+void BM_time_gettimeofday_syscall::Run(int iters) {
+  StartBenchmarkTiming();
+
+  timeval tv;
+  for (int i = 0; i < iters; ++i) {
+    syscall(__NR_gettimeofday, &tv, NULL);
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_time_time);
+void BM_time_time::Run(int iters) {
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    time(NULL);
+  }
+
+  StopBenchmarkTiming();
+}

benchmarks/unistd_benchmark.cpp

@@ -0,0 +1,71 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <sys/syscall.h>
+#include <unistd.h>
+
+#include <benchmark/Benchmark.h>
+
+BENCHMARK_NO_ARG(BM_unistd_getpid);
+void BM_unistd_getpid::Run(int iters) {
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    getpid();
+  }
+
+  StopBenchmarkTiming();
+}
+
+BENCHMARK_NO_ARG(BM_unistd_getpid_syscall);
+void BM_unistd_getpid_syscall::Run(int iters) {
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    syscall(__NR_getpid);
+  }
+
+  StopBenchmarkTiming();
+}
+
+#if defined(__BIONIC__)
+
+// Stop GCC optimizing out our pure function.
+/* Must not be static! */ pid_t (*gettid_fp)() = gettid;
+
+BENCHMARK_NO_ARG(BM_unistd_gettid);
+void BM_unistd_gettid::Run(int iters) {
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    gettid_fp();
+  }
+
+  StopBenchmarkTiming();
+}
+
+#endif
+
+BENCHMARK_NO_ARG(BM_unistd_gettid_syscall);
+void BM_unistd_gettid_syscall::Run(int iters) {
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    syscall(__NR_gettid);
+  }
+
+  StopBenchmarkTiming();
+}

benchmarks/utils.cpp

@@ -0,0 +1,80 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "utils.h"
+
+#include <inttypes.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <stdlib.h>
+
+#include <string>
+
+#include <android-base/stringprintf.h>
+
+int Round(int n) {
+  int base = 1;
+  while (base*10 < n) {
+    base *= 10;
+  }
+  if (n < 2*base) {
+    return 2*base;
+  }
+  if (n < 5*base) {
+    return 5*base;
+  }
+  return 10*base;
+}
+
+// Similar to the code in art, but supporting both binary and decimal prefixes.
+std::string PrettyInt(long value, size_t base) {
+  if (base != 2 && base != 10) abort();
+
+  uint64_t count = static_cast<uint64_t>(value);
+  bool negative_number = false;
+  if (value < 0) {
+    negative_number = true;
+    count = static_cast<uint64_t>(-value);
+  }
+
+  // The byte thresholds at which we display amounts. A count is displayed
+  // in unit U when kUnitThresholds[U] <= bytes < kUnitThresholds[U+1].
+  static const uint64_t kUnitThresholds2[] = {
+    1024*1024*1024 /* Gi */, 2*1024*1024 /* Mi */, 3*1024 /* Ki */, 0,
+  };
+  static const uint64_t kUnitThresholds10[] = {
+    1000*1000*1000 /* G */, 2*1000*1000 /* M */, 3*1000 /* k */, 0,
+  };
+  static const uint64_t kAmountPerUnit2[] = { 1024*1024*1024, 1024*1024, 1024, 1 };
+  static const uint64_t kAmountPerUnit10[] = { 1000*1000*1000, 1000*1000, 1000, 1 };
+  static const char* const kUnitStrings2[] = { "Gi", "Mi", "Ki", "" };
+  static const char* const kUnitStrings10[] = { "G", "M", "k", "" };
+
+  // Which set are we using?
+  const uint64_t* kUnitThresholds = ((base == 2) ? kUnitThresholds2 : kUnitThresholds10);
+  const uint64_t* kAmountPerUnit = ((base == 2) ? kAmountPerUnit2 : kAmountPerUnit10);
+  const char* const* kUnitStrings = ((base == 2) ? kUnitStrings2 : kUnitStrings10);
+
+  size_t i = 0;
+  for (; kUnitThresholds[i] != 0; ++i) {
+    if (count >= kUnitThresholds[i]) {
+      break;
+    }
+  }
+  return android::base::StringPrintf("%s%" PRId64 "%s",
+                                     negative_number ? "-" : "",
+                                     count / kAmountPerUnit[i], kUnitStrings[i]);
+}

benchmarks/utils.h

@@ -0,0 +1,26 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef BENCHMARKS_UTILS_H
+#define BENCHMARKS_UTILS_H
+
+#include <stddef.h>
+#include <string>
+
+int Round(int n);
+std::string PrettyInt(long value, size_t base);
+
+#endif  // BENCHMARKS_UTILS_H

build/run-on-host.mk

@@ -0,0 +1,44 @@
+#
+# Copyright (C) 2014 The Android Open Source Project
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+# Include once
+ifneq ($(bionic_run_on_host_mk_included),true)
+bionic_run_on_host_mk_included:=true
+
+ifneq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),arm mips x86))
+LINKER = linker64
+else
+LINKER = linker
+endif
+
+ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),x86 x86_64))
+# gtest needs ANDROID_DATA/local/tmp for death test output.
+# Make sure to create ANDROID_DATA/local/tmp if doesn't exist.
+# bionic itself should always work relative to ANDROID_DATA or ANDROID_ROOT.
+bionic-prepare-run-on-host: $(TARGET_OUT_EXECUTABLES)/$(LINKER) $(TARGET_OUT)/etc/hosts $(TARGET_OUT_EXECUTABLES)/sh
+	if [ ! -d /system ]; then \
+	  echo "Attempting to create /system"; \
+	  sudo mkdir -p -m 0777 /system; \
+	fi
+	mkdir -p $(TARGET_OUT_DATA)/local/tmp
+	ln -fs `realpath $(TARGET_OUT)/bin` /system/
+	ln -fs `realpath $(TARGET_OUT)/etc` /system/
+	ln -fs `realpath $(TARGET_OUT)/lib` /system/
+	if [ -d "$(TARGET_OUT)/lib64" ]; then \
+	  ln -fs `realpath $(TARGET_OUT)/lib64` /system/; \
+	fi
+endif
+endif

libc/CAVEATS

@@ -1,26 +0,0 @@
-Bionic is a very small C library because we have decided to *not* implement various features
-of the POSIX standard. we only add functions on a as-needed basis, and there are a few things
-we wish we'll never put in there.
-
-this file is here to document explicitely what we don't want to support in Bionic:
-
-- C++ exceptions are not supported. on embedded systems, they lead to extremely larger and
-  slower code for no good reason (even when so-called zero-cost exception schemes are
-  implemented, they enforce very large numbers of registers spills to the stack, even
-  in functions that do not throw an exception themselves).
-
-- pthread cancellation is *not* supported. this seemingly simple "feature" is the source
-  of much bloat and complexity in a C library. Besides, you'd better write correct
-  multi-threaded code instead of relying on this stuff.
-
-- pthread_once() doesn't support C++ exceptions thrown from the init function, or the init
-  function doing a fork().
-
-- locales and wide characters are not supported. we use ICU for all this i18n stuff, which
-  is much better than the ill-designed related C libraries functions.
-
-- at the moment, several user-account-related functions like getpwd are stubbed and return
-  the values corresponding to root. this will be fixed when we'll be able to have distinct
-  users on the Android filesystem. :-(
-
-  see bionic/stubs.c for the details

libc/SYSCALLS.TXT

@@ -1,326 +1,348 @@
-# This file is used to automatically generate bionic's the system calls stubs.
+# This file is used to automatically generate bionic's system call stubs.
 #
-# Each non comment line has the following format:
+# Each non-blank, non-comment line has the following format:
 #
-# return_type    func_name[:syscall_name[:call_id]]([parameter_list])  (1|-1|"stub")
+# return_type func_name[|alias_list][:syscall_name[:socketcall_id]]([parameter_list]) arch_list
 #
-# Note that:
+# where:
+#       arch_list ::= "all" | arch+
+#       arch      ::= "arm" | "arm64" | "mips" | "mips64" | "x86" | "x86_64"
+#
+# Note:
 #      - syscall_name corresponds to the name of the syscall, which may differ from
 #        the exported function name (example: the exit syscall is implemented by the _exit()
 #        function, which is not the same as the standard C exit() function which calls it)
-#        The call_id parameter, given that func_name and syscall_name have
+#
+#      - alias_list is optional comma separated list of function aliases.
+#
+#      - The call_id parameter, given that func_name and syscall_name have
 #        been provided, allows the user to specify dispatch style syscalls.
 #        For example, socket() syscall on i386 actually becomes:
 #          socketcall(__NR_socket, 1, *(rest of args on stack)).
 #
-#      - each parameter type is assumed to be stored on 32 bits, there is no plan to support
-#        64-bit architectures at the moment
-#
-#      - the final field can be "1", meaning: generate a stub for each architecture,
-#        taking the constants from the kernel header files.
-#
-#      - the final field can be "stub" meaning: do not generate any stubs ---
-#        in this case, a hand-written custom stub must be provided.
-#        TODO: replace this with something like "custom" or "none", or remove
-#        it entirely.
-#
-#      - the final field can be a three-element list of 1s and -1 meaning:
-#        this system call is only available on some of the architectures (1),
-#        and no stub should be generated for those architectures marked with -1.
-#        the order is arm,x86,mips.
-#        TODO: replace this with something more readable like "-arm,-mips" (meaning x86 only).
+#      - Each parameter type is assumed to be stored in 32 bits.
 #
 # This file is processed by a python script named gensyscalls.py.
-#
 
-# process management
-void    _exit:exit_group (int)      1
-void    _exit_thread:exit (int)     1
-pid_t   __fork:fork (void)           1
-pid_t   _waitpid:waitpid (pid_t, int*, int, struct rusage*)   -1,1,1
-int     __waitid:waitid(int, pid_t, struct siginfo_t*, int,void*)          1
-pid_t   wait4(pid_t pid, int *status, int options, struct rusage *rusage)   1
+int     execve(const char*, char* const*, char* const*)  all
 
-# NOTE: this system call is never called directly, but we list it there
-#       to have __NR_clone properly defined.
-#
-pid_t   __sys_clone:clone (int, void*, int*, void*, int*) 1
+uid_t   getuid:getuid32()         arm,x86
+uid_t   getuid:getuid()           arm64,mips,mips64,x86_64
+gid_t   getgid:getgid32()         arm,x86
+gid_t   getgid:getgid()           arm64,mips,mips64,x86_64
+uid_t   geteuid:geteuid32()       arm,x86
+uid_t   geteuid:geteuid()         arm64,mips,mips64,x86_64
+gid_t   getegid:getegid32()       arm,x86
+gid_t   getegid:getegid()         arm64,mips,mips64,x86_64
+uid_t   getresuid:getresuid32(uid_t* ruid, uid_t* euid, uid_t* suid)   arm,x86
+uid_t   getresuid:getresuid(uid_t* ruid, uid_t* euid, uid_t* suid)     arm64,mips,mips64,x86_64
+gid_t   getresgid:getresgid32(gid_t* rgid, gid_t* egid, gid_t* sgid)   arm,x86
+gid_t   getresgid:getresgid(gid_t* rgid, gid_t* egid, gid_t* sgid)     arm64,mips,mips64,x86_64
+ssize_t readahead(int, off64_t, size_t)     all
+int     getgroups:getgroups32(int, gid_t*)    arm,x86
+int     getgroups:getgroups(int, gid_t*)      arm64,mips,mips64,x86_64
+pid_t   getpgid(pid_t)             all
+pid_t   getppid()                  all
+pid_t   getsid(pid_t)              all
+pid_t   setsid()                   all
+int     setgid:setgid32(gid_t)     arm,x86
+int     setgid:setgid(gid_t)       arm64,mips,mips64,x86_64
+int     setuid:setuid32(uid_t)    arm,x86
+int     setuid:setuid(uid_t)      arm64,mips,mips64,x86_64
+int     setreuid:setreuid32(uid_t, uid_t)   arm,x86
+int     setreuid:setreuid(uid_t, uid_t)     arm64,mips,mips64,x86_64
+int     setresuid:setresuid32(uid_t, uid_t, uid_t)   arm,x86
+int     setresuid:setresuid(uid_t, uid_t, uid_t)     arm64,mips,mips64,x86_64
+int     setresgid:setresgid32(gid_t, gid_t, gid_t)   arm,x86
+int     setresgid:setresgid(gid_t, gid_t, gid_t)     arm64,mips,mips64,x86_64
+void*   __brk:brk(void*)           all
+int     kill(pid_t, int)           all
+int     tgkill(pid_t tgid, pid_t tid, int sig)  all
+int     __ptrace:ptrace(int request, int pid, void* addr, void* data)  all
 
-int     execve (const char*, char* const*, char* const*)  1
+# <sys/resource.h>
+int getrusage(int, struct rusage*)  all
+int __getpriority:getpriority(int, int)  all
+int setpriority(int, int, int)   all
+# On LP64, rlimit and rlimit64 are the same.
+# On 32-bit systems we use prlimit64 to implement the rlimit64 functions.
+int getrlimit:ugetrlimit(int, struct rlimit*)  arm,x86
+int getrlimit(int, struct rlimit*)  mips
+int getrlimit|getrlimit64(int, struct rlimit*)  arm64,mips64,x86_64
+int setrlimit(int, const struct rlimit*)  arm,mips,x86
+int setrlimit|setrlimit64(int, const struct rlimit*)  arm64,mips64,x86_64
+int prlimit64|prlimit(pid_t, int, struct rlimit64*, const struct rlimit64*)  arm64,mips64,x86_64
+int prlimit64(pid_t, int, struct rlimit64*, const struct rlimit64*)  arm,mips,x86
 
-int     __setuid:setuid32 (uid_t)    1,1,-1
-int     __setuid:setuid (uid_t)   -1,-1,1
-uid_t   getuid:getuid32 ()         1,1,-1
-uid_t   getuid:getuid ()           -1,-1,1
-gid_t   getgid:getgid32 ()         1,1,-1
-gid_t   getgid:getgid ()           -1,-1,1
-uid_t   geteuid:geteuid32 ()       1,1,-1
-uid_t   geteuid:geteuid ()         -1,-1,1
-gid_t   getegid:getegid32 ()       1,1,-1
-gid_t   getegid:getegid ()         -1,-1,1
-uid_t   getresuid:getresuid32 (uid_t *ruid, uid_t *euid, uid_t *suid)   1,1,-1
-uid_t   getresuid:getresuid (uid_t *ruid, uid_t *euid, uid_t *suid)     -1,-1,1
-gid_t   getresgid:getresgid32 (gid_t *rgid, gid_t *egid, gid_t *sgid)   1,1,-1
-gid_t   getresgid:getresgid (gid_t *rgid, gid_t *egid, gid_t *sgid)     -1,-1,1
-pid_t   gettid()                   1
-ssize_t readahead(int, off64_t, size_t)     1
-int     getgroups:getgroups32(int, gid_t *)    1,1,-1
-int     getgroups:getgroups(int, gid_t *)      -1,-1,1
-pid_t   getpgid(pid_t)             1
-pid_t   getppid()                  1
-pid_t   getsid(pid_t)              1
-pid_t   setsid()                   1
-int     setgid:setgid32(gid_t)     1,1,-1
-int     setgid:setgid(gid_t)       -1,-1,1
-int     seteuid:seteuid32(uid_t)   stub
-int     __setreuid:setreuid32(uid_t, uid_t)   1,1,-1
-int     __setreuid:setreuid(uid_t, uid_t)     -1,-1,1
-int     __setresuid:setresuid32(uid_t, uid_t, uid_t)   1,1,-1
-int     __setresuid:setresuid(uid_t, uid_t, uid_t)     -1,-1,1
-int     setresgid:setresgid32(gid_t, gid_t, gid_t)   1,1,-1
-int     setresgid:setresgid(gid_t, gid_t, gid_t)     -1,-1,1
-void*   __brk:brk(void*)           1
-# see comments in arch-arm/bionic/kill.S to understand why we don't generate an ARM stub for kill/tkill
-int     kill(pid_t, int)           -1,1,1
-int     tkill(pid_t tid, int sig)  -1,1,1
-int     tgkill(pid_t tgid, pid_t tid, int sig)  -1,1,1
-int     __ptrace:ptrace(int request, int pid, void* addr, void* data)  1
-int     __set_thread_area:set_thread_area(void*  user_desc)  -1,1,1
-int     __getpriority:getpriority(int, int)  1
-int     setpriority(int, int, int)   1
-int     setrlimit(int resource, const struct rlimit *rlp)  1
-int     getrlimit:ugetrlimit(int resource, struct rlimit *rlp)  1,1,-1
-int     getrlimit:getrlimit(int resource, struct rlimit *rlp)  -1,-1,1
-int     getrusage(int who, struct rusage*  r_usage)  1
-int     setgroups:setgroups32(int, const gid_t *)   1,1,-1
-int     setgroups:setgroups(int, const gid_t *)     -1,-1,1
-pid_t   getpgrp(void)  stub
-int     setpgid(pid_t, pid_t)  1
-pid_t   vfork(void)  1,-1,-1
-int     setregid:setregid32(gid_t, gid_t)  1,1,-1
-int     setregid:setregid(gid_t, gid_t)    -1,-1,1
-int     chroot(const char *)  1
-# IMPORTANT: Even though <sys/prctl.h> declares prctl(int,...), the syscall stub must take 6 arguments
-#            to match the kernel implementation.
-int     prctl(int option, unsigned int arg2, unsigned int arg3, unsigned int arg4, unsigned int arg5) 1
-int     capget(cap_user_header_t header, cap_user_data_t data) 1
-int     capset(cap_user_header_t header, const cap_user_data_t data) 1
-int     sigaltstack(const stack_t*, stack_t*) 1
-int     acct(const char*  filepath)  1
+int     setgroups:setgroups32(int, const gid_t*)   arm,x86
+int     setgroups:setgroups(int, const gid_t*)     arm64,mips,mips64,x86_64
+int     setpgid(pid_t, pid_t)  all
+int     setregid:setregid32(gid_t, gid_t)  arm,x86
+int     setregid:setregid(gid_t, gid_t)    arm64,mips,mips64,x86_64
+int     chroot(const char*)  all
+int     prctl(int, unsigned long, unsigned long, unsigned long, unsigned long) all
+long    __arch_prctl:arch_prctl(int, unsigned long) x86_64
+int     capget(cap_user_header_t header, cap_user_data_t data) all
+int     capset(cap_user_header_t header, const cap_user_data_t data) all
+int     sigaltstack(const stack_t*, stack_t*) all
+int     acct(const char*  filepath)  all
 
 # file descriptors
-ssize_t     read (int, void*, size_t)        1
-ssize_t     write (int, const void*, size_t)       1
-ssize_t     pread64 (int, void *, size_t, off64_t) 1
-ssize_t     pwrite64 (int, void *, size_t, off64_t) 1
-int         __open:open (const char*, int, mode_t)  1
-int         __openat:openat (int, const char*, int, mode_t) 1
-int         close (int)                      1
-int         creat(const char*, mode_t)       stub
-off_t       lseek(int, off_t, int)           1
-int         __llseek:_llseek (int, unsigned long, unsigned long, loff_t*, int)  1
-pid_t       getpid ()    1
-void *      mmap(void *, size_t, int, int, int, long)  stub
-void *      __mmap2:mmap2(void*, size_t, int, int, int, long)   1
-int         munmap(void *, size_t)  1
-void *      mremap(void *, size_t, size_t, unsigned long)  1
-int         msync(const void *, size_t, int)    1
-int         mprotect(const void *, size_t, int)  1
-int         madvise(const void *, size_t, int)  1
-int         mlock(const void *addr, size_t len)    1
-int         munlock(const void *addr, size_t len)   1
-int         mlockall(int flags)   1
-int         munlockall()   1
-int         mincore(void*  start, size_t  length, unsigned char*  vec)   1
-int         __ioctl:ioctl(int, int, void *)  1
-int         readv(int, const struct iovec *, int)   1
-int         writev(int, const struct iovec *, int)  1
-int         __fcntl:fcntl(int, int, void*)  1
-int         flock(int, int)   1
-int         fchmod(int, mode_t)  1
-int         dup(int)  1
-int         pipe(int *)  1,1,-1
-int         pipe2(int *, int) 1
-int         dup2(int, int)   1
-int         select:_newselect(int, struct fd_set *, struct fd_set *, struct fd_set *, struct timeval *)  1
-int         ftruncate(int, off_t)  1
-int         ftruncate64(int, off64_t) 1
-int         getdents:getdents64(unsigned int, struct dirent *, unsigned int)   1
-int         fsync(int)  1
-int         fdatasync(int) 1
-int         fchown:fchown32(int, uid_t, gid_t)  1,1,-1
-int         fchown:fchown(int, uid_t, gid_t)    -1,-1,1
-void        sync(void)  1
-int         __fcntl64:fcntl64(int, int, void *)  1
-int         __fstatfs64:fstatfs64(int, size_t, struct statfs *)  1
-ssize_t     sendfile(int out_fd, int in_fd, off_t *offset, size_t count)  1
-int         fstatat:fstatat64(int dirfd, const char *path, struct stat *buf, int flags)   1
-int         mkdirat(int dirfd, const char *pathname, mode_t mode)  1
-int         fchownat(int dirfd, const char *path, uid_t owner, gid_t group, int flags)  1
-int         fchmodat(int dirfd, const char *path, mode_t mode, int flags)  1
-int         renameat(int olddirfd, const char *oldpath, int newdirfd, const char *newpath)  1
-int         fsetxattr(int, const char *, const void *, size_t, int) 1
-ssize_t     fgetxattr(int, const char *, void *, size_t) 1
-ssize_t     flistxattr(int, char *, size_t) 1
-int         fremovexattr(int, const char *) 1
+ssize_t     read(int, void*, size_t)        all
+ssize_t     write(int, const void*, size_t)       all
+ssize_t     pread64(int, void*, size_t, off64_t) arm,mips,x86
+ssize_t     pread64|pread(int, void*, size_t, off_t) arm64,mips64,x86_64
+ssize_t     pwrite64(int, void*, size_t, off64_t) arm,mips,x86
+ssize_t     pwrite64|pwrite(int, void*, size_t, off_t) arm64,mips64,x86_64
+
+# On LP32, preadv/pwritev don't use off64_t --- they use pairs of 32-bit
+# arguments to avoid problems on architectures like ARM where 64-bit arguments
+# must be in a register pair starting with an even-numbered register.
+# See linux/fs/read_write.c and https://lwn.net/Articles/311630/.
+ssize_t     __preadv64:preadv(int, const struct iovec*, int, long, long) arm,mips,x86
+ssize_t     preadv|preadv64(int, const struct iovec*, int, off_t) arm64,mips64,x86_64
+ssize_t     __pwritev64:pwritev(int, const struct iovec*, int, long, long) arm,mips,x86
+ssize_t     pwritev|pwritev64(int, const struct iovec*, int, off_t) arm64,mips64,x86_64
+
+int         ___close:close(int)  all
+pid_t       __getpid:getpid()  all
+int         munmap(void*, size_t)  all
+void*       ___mremap:mremap(void*, size_t, size_t, int, void*)  all
+int         msync(const void*, size_t, int)    all
+int         mprotect(const void*, size_t, int)  all
+int         madvise(void*, size_t, int)  all
+int         mlock(const void* addr, size_t len)    all
+int         munlock(const void* addr, size_t len)   all
+int         mlockall(int flags)   all
+int         munlockall()   all
+int         mincore(void*  start, size_t  length, unsigned char*  vec)   all
+int         __ioctl:ioctl(int, int, void*)  all
+int         readv(int, const struct iovec*, int)   all
+int         writev(int, const struct iovec*, int)  all
+int         __fcntl64:fcntl64(int, int, void*)  arm,mips,x86
+int         fcntl(int, int, void*)  arm64,mips64,x86_64
+int         flock(int, int)   all
+int         ___fchmod:fchmod(int, mode_t)  all
+int         dup(int)  all
+int         pipe2(int*, int) all
+int         dup3(int, int, int)   all
+int         fsync(int)  all
+int         fdatasync(int) all
+int         fchown:fchown32(int, uid_t, gid_t)  arm,x86
+int         fchown:fchown(int, uid_t, gid_t)    arm64,mips,mips64,x86_64
+void        sync(void)  all
+int         ___fsetxattr:fsetxattr(int, const char*, const void*, size_t, int) all
+ssize_t     ___fgetxattr:fgetxattr(int, const char*, void*, size_t) all
+ssize_t     ___flistxattr:flistxattr(int, char*, size_t) all
+int         fremovexattr(int, const char*) all
+
+int __getdents64:getdents64(unsigned int, struct dirent*, unsigned int)   arm,arm64,mips,mips64,x86,x86_64
+
+int __openat:openat(int, const char*, int, mode_t) all
+int ___faccessat:faccessat(int, const char*, int)  all
+int ___fchmodat:fchmodat(int, const char*, mode_t)  all
+int fchownat(int, const char*, uid_t, gid_t, int)  all
+int fstatat64|fstatat:fstatat64(int, const char*, struct stat*, int)   arm,mips,x86
+int fstatat64|fstatat:newfstatat(int, const char*, struct stat*, int)  arm64,x86_64
+int linkat(int, const char*, int, const char*, int)  all
+int mkdirat(int, const char*, mode_t)  all
+int mknodat(int, const char*, mode_t, dev_t)  all
+int readlinkat(int, const char*, char*, size_t)  all
+int renameat(int, const char*, int, const char*)  all
+int symlinkat(const char*, int, const char*)  all
+int unlinkat(int, const char*, int)   all
+int utimensat(int, const char*, const struct timespec times[2], int)  all
+
+# Paired off_t/off64_t system calls. On 64-bit systems,
+# sizeof(off_t) == sizeof(off64_t), so there we emit two symbols that are
+# aliases. On 32-bit systems, we have two different system calls.
+# That means that every system call in this section should take three lines.
+off_t lseek(int, off_t, int) arm,mips,x86
+int __llseek:_llseek(int, unsigned long, unsigned long, off64_t*, int) arm,mips,x86
+off_t lseek|lseek64(int, off_t, int) arm64,mips64,x86_64
+int ftruncate64(int, off64_t) arm,mips,x86
+int ftruncate|ftruncate64(int, off_t) arm64,mips64,x86_64
+ssize_t sendfile(int out_fd, int in_fd, off_t* offset, size_t count) arm,mips,x86
+ssize_t sendfile64(int out_fd, int in_fd, off64_t* offset, size_t count) arm,mips,x86
+ssize_t sendfile|sendfile64(int out_fd, int in_fd, off_t* offset, size_t count) arm64,mips64,x86_64
+int truncate(const char*, off_t) arm,mips,x86
+int truncate64(const char*, off64_t) arm,mips,x86
+int truncate|truncate64(const char*, off_t) arm64,mips64,x86_64
+# (mmap only gets two lines because we only used the 64-bit variant on 32-bit systems.)
+void* __mmap2:mmap2(void*, size_t, int, int, int, long)   arm,mips,x86
+void* mmap|mmap64(void*, size_t, int, int, int, off_t)  arm64,mips64,x86_64
+# (fallocate only gets two lines because there is no 32-bit variant.)
+int fallocate64:fallocate(int, int, off64_t, off64_t) arm,mips,x86
+int fallocate|fallocate64(int, int, off_t, off_t) arm64,mips64,x86_64
+
+# posix_fadvise64 is awkward: arm has shuffled arguments,
+# the POSIX functions don't set errno, and no architecture has posix_fadvise.
+int __arm_fadvise64_64:arm_fadvise64_64(int, int, off64_t, off64_t) arm
+int __fadvise64:fadvise64_64(int, off64_t, off64_t, int) x86
+int __fadvise64:fadvise64(int, off64_t, off64_t, int) arm64,mips,mips64,x86_64
+
+int __fstatfs64:fstatfs64(int, size_t, struct statfs*)  arm,mips,x86
+int __fstatfs:fstatfs(int, struct statfs*)  arm64,mips64,x86_64
+int __statfs64:statfs64(const char*, size_t, struct statfs*)  arm,mips,x86
+int __statfs:statfs(const char*, struct statfs*)  arm64,mips64,x86_64
+
+int     fstat64|fstat:fstat64(int, struct stat*)    arm,mips,x86
+int     fstat64|fstat:fstat(int, struct stat*)    arm64,x86_64
 
 # file system
-int     link (const char*, const char*)  1
-int     unlink (const char*)             1
-int     unlinkat (int, const char *, int)   1
-int     chdir (const char*)              1
-int     mknod (const char*, mode_t, dev_t)  1
-int     chmod (const char*,mode_t)          1
-int     chown:chown32(const char *, uid_t, gid_t)  1,1,-1
-int     chown:chown(const char *, uid_t, gid_t)    -1,-1,1
-int     lchown:lchown32 (const char*, uid_t, gid_t)  1,1,-1
-int     lchown:lchown (const char*, uid_t, gid_t)  -1,-1,1
-int     mount (const char*, const char*, const char*, unsigned long, const void*)  1
-int     umount(const char*)  stub
-int     umount2 (const char*, int)  1
-int     fstat:fstat64(int, struct stat*)    1
-int     stat:stat64(const char *, struct stat *)  1
-int     lstat:lstat64(const char *, struct stat *)  1
-int     mkdir(const char *, mode_t) 1
-int     readlink(const char *, char *, size_t)  1
-int     rmdir(const char *)  1
-int     rename(const char *, const char *)  1
-int     __getcwd:getcwd(char * buf, size_t size)  1
-int     access(const char *, int)  1
-int     faccessat(int, const char *, int, int)  1
-int     symlink(const char *, const char *)  1
-int     fchdir(int)    1
-int     truncate(const char*, off_t)    1
-int     setxattr(const char *, const char *, const void *, size_t, int) 1
-int     lsetxattr(const char *, const char *, const void *, size_t, int) 1
-ssize_t getxattr(const char *, const char *, void *, size_t) 1
-ssize_t lgetxattr(const char *, const char *, void *, size_t) 1
-ssize_t listxattr(const char *, char *, size_t) 1
-ssize_t llistxattr(const char *, char *, size_t) 1
-int     removexattr(const char *, const char *) 1
-int     lremovexattr(const char *, const char *) 1
-int     __statfs64:statfs64(const char *, size_t, struct statfs *)  1
-long    unshare(unsigned long)  1
-int     swapon(const char *, int) 1
-int     swapoff(const char *) 1
+int     chdir(const char*)              all
+int     mount(const char*, const char*, const char*, unsigned long, const void*)  all
+int     umount2(const char*, int)  all
+int     __getcwd:getcwd(char* buf, size_t size)  all
+int     fchdir(int)    all
+int     setxattr(const char*, const char*, const void*, size_t, int) all
+int     lsetxattr(const char*, const char*, const void*, size_t, int) all
+ssize_t getxattr(const char*, const char*, void*, size_t) all
+ssize_t lgetxattr(const char*, const char*, void*, size_t) all
+ssize_t listxattr(const char*, char*, size_t) all
+ssize_t llistxattr(const char*, char*, size_t) all
+int     removexattr(const char*, const char*) all
+int     lremovexattr(const char*, const char*) all
+int     swapon(const char*, int) all
+int     swapoff(const char*) all
 
 # time
-int           pause ()                       1
-int           gettimeofday(struct timeval*, struct timezone*)       1
-int           settimeofday(const struct timeval*, const struct timezone*)   1
-clock_t       times(struct tms *)       1
-int           nanosleep(const struct timespec *, struct timespec *)   1
-int           clock_gettime(clockid_t clk_id, struct timespec *tp)    1
-int           clock_settime(clockid_t clk_id, const struct timespec *tp)  1
-int           clock_getres(clockid_t clk_id, struct timespec *res)   1
-int           clock_nanosleep(clockid_t clock_id, int flags, const struct timespec *req, struct timespec *rem)  1
-int           getitimer(int, const struct itimerval *)   1
-int           setitimer(int, const struct itimerval *, struct itimerval *)  1
-int           __timer_create:timer_create(clockid_t clockid, struct sigevent *evp, timer_t *timerid)    1
-int           __timer_settime:timer_settime(timer_t, int, const struct itimerspec*, struct itimerspec*) 1
-int           __timer_gettime:timer_gettime(timer_t, struct itimerspec*)                                1
-int           __timer_getoverrun:timer_getoverrun(timer_t)                                              1
-int           __timer_delete:timer_delete(timer_t)                                                      1
-int           utimes(const char*, const struct timeval tvp[2])                          1
-int           utimensat(int, const char *, const struct timespec times[2], int)         1
-int           timerfd_create(clockid_t, int)   1
-int           timerfd_settime(int, int, const struct itimerspec *, struct itimerspec *)   1
-int           timerfd_gettime(int, struct itimerspec *)   1
+int           settimeofday(const struct timeval*, const struct timezone*)   all
+clock_t       times(struct tms*)       all
+int           nanosleep(const struct timespec*, struct timespec*)   all
+int           clock_settime(clockid_t, const struct timespec*)  all
+int           clock_getres(clockid_t, struct timespec*)   all
+int           ___clock_nanosleep:clock_nanosleep(clockid_t, int, const struct timespec*, struct timespec*)  all
+int           getitimer(int, const struct itimerval*)   all
+int           setitimer(int, const struct itimerval*, struct itimerval*)  all
+int           __timer_create:timer_create(clockid_t clockid, struct sigevent* evp, __kernel_timer_t* timerid)    all
+int           __timer_settime:timer_settime(__kernel_timer_t, int, const struct itimerspec*, struct itimerspec*) all
+int           __timer_gettime:timer_gettime(__kernel_timer_t, struct itimerspec*)                                all
+int           __timer_getoverrun:timer_getoverrun(__kernel_timer_t)                                              all
+int           __timer_delete:timer_delete(__kernel_timer_t)                                                      all
+int           timerfd_create(clockid_t, int)   all
+int           timerfd_settime(int, int, const struct itimerspec*, struct itimerspec*)   all
+int           timerfd_gettime(int, struct itimerspec*)   all
 
 # signals
-int     sigaction(int, const struct sigaction *, struct sigaction *)  1
-int     sigprocmask(int, const sigset_t *, sigset_t *)  1
-int     __sigsuspend:sigsuspend(int unused1, int unused2, unsigned mask)  1,1,-1
-int     __sigsuspend:sigsuspend(const sigset_t *mask)  -1,-1,1
-int     __rt_sigaction:rt_sigaction (int sig, const struct sigaction *act, struct sigaction *oact, size_t sigsetsize)  1
-int     __rt_sigprocmask:rt_sigprocmask (int  how, const sigset_t *set, sigset_t *oset, size_t sigsetsize)  1
-int     __rt_sigtimedwait:rt_sigtimedwait(const sigset_t *set, struct siginfo_t  *info, struct timespec_t  *timeout, size_t  sigset_size)  1
-int     sigpending(sigset_t *)  1
-int     signalfd4(int fd, const sigset_t *mask, size_t sizemask, int flags)  1
+int     __sigaction:sigaction(int, const struct sigaction*, struct sigaction*)  arm,mips,x86
+int     __rt_sigaction:rt_sigaction(int, const struct sigaction*, struct sigaction*, size_t)  all
+int     __rt_sigpending:rt_sigpending(sigset_t*, size_t)  all
+int     __rt_sigprocmask:rt_sigprocmask(int, const sigset_t*, sigset_t*, size_t)  all
+int     __rt_sigsuspend:rt_sigsuspend(const sigset_t*, size_t)  all
+int     __rt_sigtimedwait:rt_sigtimedwait(const sigset_t*, struct siginfo_t*, struct timespec_t*, size_t)  all
+int     ___rt_sigqueueinfo:rt_sigqueueinfo(pid_t, int, siginfo_t*)  all
+int     __signalfd4:signalfd4(int, const sigset_t*, size_t, int)  all
 
 # sockets
-int           socket(int, int, int)              1,-1,1
-int           socketpair(int, int, int, int*)    1,-1,1
-int           bind(int, struct sockaddr *, int)  1,-1,1
-int           connect(int, struct sockaddr *, socklen_t)   1,-1,1
-int           listen(int, int)                   1,-1,1
-int           accept(int, struct sockaddr *, socklen_t *)  1,-1,1
-int           getsockname(int, struct sockaddr *, socklen_t *)  1,-1,1
-int           getpeername(int, struct sockaddr *, socklen_t *)  1,-1,1
-int           sendto(int, const void *, size_t, int, const struct sockaddr *, socklen_t)  1,-1,1
-int           recvfrom(int, void *, size_t, unsigned int, struct sockaddr *, socklen_t *)  1,-1,1
-int           shutdown(int, int)  1,-1,1
-int           setsockopt(int, int, int, const void *, socklen_t)  1,-1,1
-int           getsockopt(int, int, int, void *, socklen_t *)    1,-1,1
-int           sendmsg(int, const struct msghdr *, unsigned int)  1,-1,1
-int           recvmsg(int, struct msghdr *, unsigned int)   1,-1,1
+int           __socket:socket(int, int, int)              arm,arm64,mips,mips64,x86_64
+int           socketpair(int, int, int, int*)    arm,arm64,mips,mips64,x86_64
+int           bind(int, struct sockaddr*, int)  arm,arm64,mips,mips64,x86_64
+int           __connect:connect(int, struct sockaddr*, socklen_t)   arm,arm64,mips,mips64,x86_64
+int           listen(int, int)                   arm,arm64,mips,mips64,x86_64
+int           __accept4:accept4(int, struct sockaddr*, socklen_t*, int)  arm,arm64,mips,mips64,x86_64
+int           getsockname(int, struct sockaddr*, socklen_t*)  arm,arm64,mips,mips64,x86_64
+int           getpeername(int, struct sockaddr*, socklen_t*)  arm,arm64,mips,mips64,x86_64
+int           sendto(int, const void*, size_t, int, const struct sockaddr*, socklen_t)  arm,arm64,mips,mips64,x86_64
+int           recvfrom(int, void*, size_t, unsigned int, struct sockaddr*, socklen_t*)  arm,arm64,mips,mips64,x86_64
+int           shutdown(int, int)  arm,arm64,mips,mips64,x86_64
+int           setsockopt(int, int, int, const void*, socklen_t)  arm,arm64,mips,mips64,x86_64
+int           getsockopt(int, int, int, void*, socklen_t*)    arm,arm64,mips,mips64,x86_64
+int           sendmsg(int, const struct msghdr*, unsigned int)  arm,arm64,mips,mips64,x86_64
+int           recvmsg(int, struct msghdr*, unsigned int)   arm,arm64,mips,mips64,x86_64
+int           recvmmsg(int, struct mmsghdr*, unsigned int, int, const struct timespec*)   arm,arm64,mips,mips64,x86_64
+int           sendmmsg(int, struct mmsghdr*, unsigned int, int)   arm,arm64,mips,mips64,x86_64
 
 # sockets for x86. These are done as an "indexed" call to socketcall syscall.
-int           socket:socketcall:1 (int, int, int) -1,1,-1
-int           bind:socketcall:2 (int, struct sockaddr *, int)  -1,1,-1
-int           connect:socketcall:3(int, struct sockaddr *, socklen_t)   -1,1,-1
-int           listen:socketcall:4(int, int)                   -1,1,-1
-int           accept:socketcall:5(int, struct sockaddr *, socklen_t *)  -1,1,-1
-int           getsockname:socketcall:6(int, struct sockaddr *, socklen_t *)  -1,1,-1
-int           getpeername:socketcall:7(int, struct sockaddr *, socklen_t *)  -1,1,-1
-int           socketpair:socketcall:8(int, int, int, int*)    -1,1,-1
-int           sendto:socketcall:11(int, const void *, size_t, int, const struct sockaddr *, socklen_t)  -1,1,-1
-int           recvfrom:socketcall:12(int, void *, size_t, unsigned int, struct sockaddr *, socklen_t *)  -1,1,-1
-int           shutdown:socketcall:13(int, int)  -1,1,-1
-int           setsockopt:socketcall:14(int, int, int, const void *, socklen_t)  -1,1,-1
-int           getsockopt:socketcall:15(int, int, int, void *, socklen_t *)    -1,1,-1
-int           sendmsg:socketcall:16(int, const struct msghdr *, unsigned int)  -1,1,-1
-int           recvmsg:socketcall:17(int, struct msghdr *, unsigned int)   -1,1,-1
+int           __socket:socketcall:1(int, int, int) x86
+int           bind:socketcall:2(int, struct sockaddr*, int)  x86
+int           __connect:socketcall:3(int, struct sockaddr*, socklen_t)   x86
+int           listen:socketcall:4(int, int)                   x86
+int           getsockname:socketcall:6(int, struct sockaddr*, socklen_t*)  x86
+int           getpeername:socketcall:7(int, struct sockaddr*, socklen_t*)  x86
+int           socketpair:socketcall:8(int, int, int, int*)    x86
+int           sendto:socketcall:11(int, const void*, size_t, int, const struct sockaddr*, socklen_t)  x86
+int           recvfrom:socketcall:12(int, void*, size_t, unsigned int, struct sockaddr*, socklen_t*)  x86
+int           shutdown:socketcall:13(int, int)  x86
+int           setsockopt:socketcall:14(int, int, int, const void*, socklen_t)  x86
+int           getsockopt:socketcall:15(int, int, int, void*, socklen_t*)    x86
+int           sendmsg:socketcall:16(int, const struct msghdr*, unsigned int)  x86
+int           recvmsg:socketcall:17(int, struct msghdr*, unsigned int)   x86
+int           __accept4:socketcall:18(int, struct sockaddr*, socklen_t*, int)  x86
+int           recvmmsg:socketcall:19(int, struct mmsghdr*, unsigned int, int, const struct timespec*)   x86
+int           sendmmsg:socketcall:20(int, struct mmsghdr*, unsigned int, int)   x86
 
 # scheduler & real-time
-int sched_setscheduler(pid_t pid, int policy, const struct sched_param *param)  1
-int sched_getscheduler(pid_t pid)  1
-int sched_yield(void)  1
-int sched_setparam(pid_t pid, const struct sched_param *param)  1
-int sched_getparam(pid_t pid, struct sched_param *param)  1
-int sched_get_priority_max(int policy)  1
-int sched_get_priority_min(int policy)  1
-int sched_rr_get_interval(pid_t pid, struct timespec *interval)  1
-int sched_setaffinity(pid_t pid, size_t setsize, const cpu_set_t* set) 1
-int __sched_getaffinity:sched_getaffinity(pid_t pid, size_t setsize, cpu_set_t* set)  1
-int __getcpu:getcpu(unsigned *cpu, unsigned *node, void *unused) 1
-
-# io priorities
-int ioprio_set(int which, int who, int ioprio) 1
-int ioprio_get(int which, int who) 1
+int sched_setscheduler(pid_t pid, int policy, const struct sched_param* param)  all
+int sched_getscheduler(pid_t pid)  all
+int sched_yield(void)  all
+int sched_setparam(pid_t pid, const struct sched_param* param)  all
+int sched_getparam(pid_t pid, struct sched_param* param)  all
+int sched_get_priority_max(int policy)  all
+int sched_get_priority_min(int policy)  all
+int sched_rr_get_interval(pid_t pid, struct timespec* interval)  all
+int sched_setaffinity(pid_t pid, size_t setsize, const cpu_set_t* set) all
+int setns(int, int) all
+int unshare(int) all
+int __sched_getaffinity:sched_getaffinity(pid_t pid, size_t setsize, cpu_set_t* set)  all
+int __getcpu:getcpu(unsigned*, unsigned*, void*) all
 
 # other
-int     uname(struct utsname *)  1
-mode_t  umask(mode_t)  1
-int      __reboot:reboot(int, int, int, void *)  1
-int     __syslog:syslog(int, char *, int)  1
-int     init_module(void *, unsigned long, const char *)  1
-int     delete_module(const char*, unsigned int)   1
-int     klogctl:syslog(int, char *, int)   1
-int     sysinfo(struct sysinfo *)  1
-int     personality(unsigned long)  1
-long    perf_event_open(struct perf_event_attr *attr_uptr, pid_t pid, int cpu, int group_fd, unsigned long flags) 1
+int     uname(struct utsname*)  all
+mode_t  umask(mode_t)  all
+int     __reboot:reboot(int, int, int, void*)  all
+int     init_module(void*, unsigned long, const char*)  all
+int     delete_module(const char*, unsigned int)   all
+int     klogctl:syslog(int, char*, int)   all
+int     sysinfo(struct sysinfo*)  all
+int     personality(unsigned long)  all
 
-# futex
-int	futex(void *, int, int, void *, void *, int) 1
+ssize_t tee(int, int, size_t, unsigned int)  all
+ssize_t splice(int, off64_t*, int, off64_t*, size_t, unsigned int)  all
+ssize_t vmsplice(int, const struct iovec*, size_t, unsigned int)  all
 
-# epoll
-int     epoll_create(int size)     1
-int     epoll_ctl(int epfd, int op, int fd, struct epoll_event *event)    1
-int     epoll_wait(int epfd, struct epoll_event *events, int max, int timeout)   1
+int epoll_create1(int)  all
+int epoll_ctl(int, int op, int, struct epoll_event*)  all
+int __epoll_pwait:epoll_pwait(int, struct epoll_event*, int, int, const sigset_t*, size_t)  all
 
-int     inotify_init(void)      1
-int     inotify_add_watch(int, const char *, unsigned int)  1
-int     inotify_rm_watch(int, unsigned int)  1
+int eventfd:eventfd2(unsigned int, int)  all
 
-int     poll(struct pollfd *, unsigned int, long)  1
+void _exit|_Exit:exit_group(int)  all
+void __exit:exit(int)  all
 
-int     eventfd:eventfd2(unsigned int, int)  1
+int inotify_init1(int)  all
+int inotify_add_watch(int, const char*, unsigned int)  all
+int inotify_rm_watch(int, unsigned int)  all
 
-# ARM-specific ARM_NR_BASE == 0x0f0000 == 983040
-int     __set_tls:__ARM_NR_set_tls(void*)                                 1,-1,-1
-int     cacheflush:__ARM_NR_cacheflush(long start, long end, long flags)  1,-1,-1
+int __pselect6:pselect6(int, fd_set*, fd_set*, fd_set*, timespec*, void*)  all
+int __ppoll:ppoll(pollfd*, unsigned int, timespec*, const sigset_t*, size_t)  all
+
+ssize_t process_vm_readv(pid_t, const struct iovec*, unsigned long, const struct iovec*, unsigned long, unsigned long)  all
+ssize_t process_vm_writev(pid_t, const struct iovec*, unsigned long, const struct iovec*, unsigned long, unsigned long)  all
+
+int __set_tid_address:set_tid_address(int*)  all
+
+int setfsgid(gid_t)  all
+int setfsuid(uid_t)  all
+
+int sethostname(const char*, size_t)  all
+
+pid_t wait4(pid_t, int*, int, struct rusage*)  all
+int __waitid:waitid(int, pid_t, struct siginfo_t*, int, void*)  all
+
+# ARM-specific
+int     __set_tls:__ARM_NR_set_tls(void*)                                 arm
+int     cacheflush:__ARM_NR_cacheflush(long start, long end, long flags)  arm
 
 # MIPS-specific
-int     _flush_cache:cacheflush(char *addr, const int nbytes, const int op) -1,-1,1
-int     syscall(int number,...) -1,-1,1
+int     _flush_cache:cacheflush(char* addr, const int nbytes, const int op) mips
+int     __set_tls:set_thread_area(void*) mips,mips64
+
+# x86-specific
+int     __set_thread_area:set_thread_area(void*) x86
+
+# vdso stuff.
+int clock_gettime(clockid_t, timespec*)                 arm,mips,mips64
+int __clock_gettime:clock_gettime(clockid_t, timespec*) arm64,x86,x86_64
+int gettimeofday(timeval*, timezone*)                   arm,mips,mips64
+int __gettimeofday:gettimeofday(timeval*, timezone*)    arm64,x86,x86_64

libc/arch-arm/arm.mk

@@ -1,44 +1,63 @@
-_LIBC_ARCH_COMMON_SRC_FILES := \
-    arch-arm/bionic/abort_arm.S \
-    arch-arm/bionic/atomics_arm.c \
-    arch-arm/bionic/clone.S \
-    arch-arm/bionic/eabi.c \
-    arch-arm/bionic/_exit_with_stack_teardown.S \
-    arch-arm/bionic/ffs.S \
-    arch-arm/bionic/futex_arm.S \
-    arch-arm/bionic/__get_sp.S \
-    arch-arm/bionic/kill.S \
-    arch-arm/bionic/libgcc_compat.c \
-    arch-arm/bionic/memcmp16.S \
-    arch-arm/bionic/memcmp.S \
-    arch-arm/bionic/_setjmp.S \
-    arch-arm/bionic/setjmp.S \
-    arch-arm/bionic/sigsetjmp.S \
-    arch-arm/bionic/syscall.S \
-    arch-arm/bionic/tgkill.S \
-    arch-arm/bionic/tkill.S \
+# 32-bit arm.
 
-# These are used by the static and dynamic versions of the libc
-# respectively.
-_LIBC_ARCH_STATIC_SRC_FILES := \
-    arch-arm/bionic/exidx_static.c
+libc_bionic_src_files_arm += \
+    arch-arm/generic/bionic/memcmp.S \
+    arch-arm/generic/bionic/memcpy.S \
+    arch-arm/generic/bionic/memset.S \
+    arch-arm/generic/bionic/strcmp.S \
+    arch-arm/generic/bionic/strcpy.S \
+    arch-arm/generic/bionic/strlen.c \
 
-_LIBC_ARCH_DYNAMIC_SRC_FILES := \
-    arch-arm/bionic/exidx_dynamic.c
-
-# Remove the C++ fortify function implementations for which there is an
-# arm assembler version.
-_LIBC_FORTIFY_FILES_TO_REMOVE := \
+libc_bionic_src_files_exclude_arm += \
     bionic/__memcpy_chk.cpp \
     bionic/__memset_chk.cpp \
-    bionic/__strcpy_chk.cpp \
-    bionic/__strcat_chk.cpp \
 
-libc_common_src_files := \
-    $(filter-out $(_LIBC_FORTIFY_FILES_TO_REMOVE),$(libc_common_src_files))
+libc_openbsd_src_files_exclude_arm += \
+    upstream-openbsd/lib/libc/string/strcpy.c \
 
-ifeq ($(strip $(wildcard bionic/libc/arch-arm/$(TARGET_CPU_VARIANT)/$(TARGET_CPU_VARIANT).mk)),)
-$(error "TARGET_CPU_VARIANT not set or set to an unknown value. Possible values are cortex-a7, cortex-a8, cortex-a9, cortex-a15, krait. Use generic for devices that do not have a CPU similar to any of the supported cpu variants.")
+#
+# Inherently architecture-specific code.
+#
+
+libc_bionic_src_files_arm += \
+    arch-arm/bionic/abort_arm.S \
+    arch-arm/bionic/atomics_arm.c \
+    arch-arm/bionic/__bionic_clone.S \
+    arch-arm/bionic/_exit_with_stack_teardown.S \
+    arch-arm/bionic/libgcc_compat.c \
+    arch-arm/bionic/popcount_tab.c \
+    arch-arm/bionic/__restore.S \
+    arch-arm/bionic/setjmp.S \
+    arch-arm/bionic/syscall.S \
+    arch-arm/bionic/vfork.S \
+
+libc_arch_static_src_files_arm := arch-arm/bionic/exidx_static.c
+libc_arch_dynamic_src_files_arm := arch-arm/bionic/exidx_dynamic.c
+
+## CPU variant specific source files
+ifeq ($(strip $(TARGET_$(my_2nd_arch_prefix)CPU_VARIANT)),)
+  $(warning TARGET_$(my_2nd_arch_prefix)ARCH is arm, but TARGET_$(my_2nd_arch_prefix)CPU_VARIANT is not defined)
+endif
+ifneq ($(TARGET_$(my_2nd_arch_prefix)CPU_VARIANT),generic)
+cpu_variant_mk := $(LOCAL_PATH)/arch-arm/$(TARGET_$(my_2nd_arch_prefix)CPU_VARIANT)/$(TARGET_$(my_2nd_arch_prefix)CPU_VARIANT).mk
+ifeq ($(wildcard $(cpu_variant_mk)),)
+$(error "TARGET_$(my_2nd_arch_prefix)CPU_VARIANT not set or set to an unknown value. Possible values are cortex-a7, cortex-a8, cortex-a9, cortex-a15, krait, denver. Use generic for devices that do not have a CPU similar to any of the supported cpu variants.")
+endif
+include $(cpu_variant_mk)
+libc_common_additional_dependencies += $(cpu_variant_mk)
+
+cpu_variant_mk :=
 endif
 
-include bionic/libc/arch-arm/$(TARGET_CPU_VARIANT)/$(TARGET_CPU_VARIANT).mk
+
+libc_crt_target_cflags_arm := \
+    -I$(LOCAL_PATH)/arch-arm/include \
+    -mthumb-interwork
+
+libc_crt_target_so_cflags_arm :=
+
+libc_crt_target_crtbegin_file_arm := \
+    $(LOCAL_PATH)/arch-common/bionic/crtbegin.c
+
+libc_crt_target_crtbegin_so_file_arm := \
+    $(LOCAL_PATH)/arch-common/bionic/crtbegin_so.c

libc/arch-arm/bionic/__aeabi.c

@@ -0,0 +1,185 @@
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#if defined(__clang__)
+// clang interprets -fno-builtin more loosely than you might expect,
+// and thinks it's okay to still substitute builtins as long as they're
+// named __aeabi_* rather than __builtin_*, which causes infinite
+// recursion if we have the fortified memcpy visible in this file.
+#undef _FORTIFY_SOURCE
+#endif
+
+#include <stddef.h>
+#include <string.h>
+
+extern int __cxa_atexit(void (*)(void*), void*, void*);
+
+// All of these are weak symbols to avoid multiple definition errors when
+// linking with libstdc++-v3 or compiler-rt.
+
+/* The "C++ ABI for ARM" document states that static C++ constructors,
+ * which are called from the .init_array, should manually call
+ * __aeabi_atexit() to register static destructors explicitly.
+ *
+ * Note that 'dso_handle' is the address of a magic linker-generate
+ * variable from the shared object that contains the constructor/destructor
+ */
+
+int __attribute__((weak))
+__aeabi_atexit_impl(void *object, void (*destructor) (void *), void *dso_handle) {
+    return __cxa_atexit(destructor, object, dso_handle);
+}
+
+int __attribute__((weak))
+__aeabi_atexit_impl2(void *object, void (*destructor) (void *), void *dso_handle) {
+    return __cxa_atexit(destructor, object, dso_handle);
+}
+
+
+void __attribute__((weak)) __aeabi_memcpy8_impl(void *dest, const void *src, size_t n) {
+    memcpy(dest, src, n);
+}
+
+void __attribute__((weak)) __aeabi_memcpy4_impl(void *dest, const void *src, size_t n) {
+    memcpy(dest, src, n);
+}
+
+void __attribute__((weak)) __aeabi_memcpy_impl(void *dest, const void *src, size_t n) {
+    memcpy(dest, src, n);
+}
+
+void __attribute__((weak)) __aeabi_memcpy8_impl2(void *dest, const void *src, size_t n) {
+    memcpy(dest, src, n);
+}
+
+void __attribute__((weak)) __aeabi_memcpy4_impl2(void *dest, const void *src, size_t n) {
+    memcpy(dest, src, n);
+}
+
+void __attribute__((weak)) __aeabi_memcpy_impl2(void *dest, const void *src, size_t n) {
+    memcpy(dest, src, n);
+}
+
+
+void __attribute__((weak)) __aeabi_memmove8_impl(void *dest, const void *src, size_t n) {
+    memmove(dest, src, n);
+}
+
+void __attribute__((weak)) __aeabi_memmove4_impl(void *dest, const void *src, size_t n) {
+    memmove(dest, src, n);
+}
+
+void __attribute__((weak)) __aeabi_memmove_impl(void *dest, const void *src, size_t n) {
+    memmove(dest, src, n);
+}
+
+void __attribute__((weak)) __aeabi_memmove8_impl2(void *dest, const void *src, size_t n) {
+    memmove(dest, src, n);
+}
+
+void __attribute__((weak)) __aeabi_memmove4_impl2(void *dest, const void *src, size_t n) {
+    memmove(dest, src, n);
+}
+
+void __attribute__((weak)) __aeabi_memmove_impl2(void *dest, const void *src, size_t n) {
+    memmove(dest, src, n);
+}
+
+/*
+ * __aeabi_memset has the order of its second and third arguments reversed.
+ *  This allows __aeabi_memclr to tail-call __aeabi_memset
+ */
+
+void __attribute__((weak)) __aeabi_memset8_impl(void *dest, size_t n, int c) {
+    memset(dest, c, n);
+}
+
+void __attribute__((weak)) __aeabi_memset4_impl(void *dest, size_t n, int c) {
+    memset(dest, c, n);
+}
+
+void __attribute__((weak)) __aeabi_memset_impl(void *dest, size_t n, int c) {
+    memset(dest, c, n);
+}
+
+void __attribute__((weak)) __aeabi_memset8_impl2(void *dest, size_t n, int c) {
+    memset(dest, c, n);
+}
+
+void __attribute__((weak)) __aeabi_memset4_impl2(void *dest, size_t n, int c) {
+    memset(dest, c, n);
+}
+
+void __attribute__((weak)) __aeabi_memset_impl2(void *dest, size_t n, int c) {
+    memset(dest, c, n);
+}
+
+
+void __attribute__((weak)) __aeabi_memclr8_impl(void *dest, size_t n) {
+    __aeabi_memset8_impl(dest, n, 0);
+}
+
+void __attribute__((weak)) __aeabi_memclr4_impl(void *dest, size_t n) {
+    __aeabi_memset4_impl(dest, n, 0);
+}
+
+void __attribute__((weak)) __aeabi_memclr_impl(void *dest, size_t n) {
+    __aeabi_memset_impl(dest, n, 0);
+}
+
+void __attribute__((weak)) __aeabi_memclr8_impl2(void *dest, size_t n) {
+    __aeabi_memset8_impl(dest, n, 0);
+}
+
+void __attribute__((weak)) __aeabi_memclr4_impl2(void *dest, size_t n) {
+    __aeabi_memset4_impl(dest, n, 0);
+}
+
+void __attribute__((weak)) __aeabi_memclr_impl2(void *dest, size_t n) {
+    __aeabi_memset_impl(dest, n, 0);
+}
+
+#define __AEABI_SYMVERS(fn_name) \
+__asm__(".symver " #fn_name "_impl, " #fn_name "@@LIBC_N"); \
+__asm__(".symver " #fn_name "_impl2, " #fn_name "@LIBC_PRIVATE")
+
+__AEABI_SYMVERS(__aeabi_atexit);
+__AEABI_SYMVERS(__aeabi_memcpy8);
+__AEABI_SYMVERS(__aeabi_memcpy4);
+__AEABI_SYMVERS(__aeabi_memcpy);
+__AEABI_SYMVERS(__aeabi_memmove8);
+__AEABI_SYMVERS(__aeabi_memmove4);
+__AEABI_SYMVERS(__aeabi_memmove);
+__AEABI_SYMVERS(__aeabi_memset8);
+__AEABI_SYMVERS(__aeabi_memset4);
+__AEABI_SYMVERS(__aeabi_memset);
+__AEABI_SYMVERS(__aeabi_memclr8);
+__AEABI_SYMVERS(__aeabi_memclr4);
+__AEABI_SYMVERS(__aeabi_memclr);
+
+#undef __AEABI_SYMVERS

libc/arch-arm/bionic/__bionic_clone.S

@@ -0,0 +1,70 @@
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <private/bionic_asm.h>
+
+// pid_t __bionic_clone(int flags, void* child_stack, pid_t* parent_tid, void* tls, pid_t* child_tid, int (*fn)(void*), void* arg);
+ENTRY(__bionic_clone)
+    mov     ip, sp
+    # save registers to parent stack
+    stmfd   sp!, {r4, r5, r6, r7}
+    .cfi_def_cfa_offset 16
+    .cfi_rel_offset r4, 0
+    .cfi_rel_offset r5, 4
+    .cfi_rel_offset r6, 8
+    .cfi_rel_offset r7, 12
+
+    # load extra parameters
+    ldmfd   ip, {r4, r5, r6}
+
+    # Push 'fn' and 'arg' onto the child stack.
+    stmdb   r1!, {r5, r6}
+
+    # Make the system call.
+    ldr     r7, =__NR_clone
+    swi     #0
+
+    # Are we the child?
+    movs    r0, r0
+    beq     1f
+
+    # In the parent, reload saved registers then either return or set errno.
+    ldmfd   sp!, {r4, r5, r6, r7}
+    cmn     r0, #(MAX_ERRNO + 1)
+    bxls    lr
+    neg     r0, r0
+    b       __set_errno_internal
+
+1:  # The child.
+    # Setting lr to 0 will make the unwinder stop at __start_thread
+    mov    lr, #0
+    # Call __start_thread with the 'fn' and 'arg' we stored on the child stack.
+    pop    {r0, r1}
+    b      __start_thread
+END(__bionic_clone)
+.hidden __bionic_clone

libc/arch-arm/bionic/__get_sp.S

@@ -1,34 +0,0 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include <machine/asm.h>
-
-ENTRY(__get_sp)
-  mov r0, sp
-  bx lr
-END(__get_sp)

libc/arch-arm/bionic/__restore.S

@@ -0,0 +1,63 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <private/bionic_asm.h>
+
+// gdb is smart enough to unwind through signal frames with just the regular
+// CFI information but libgcc and libunwind both need extra help. We do this
+// by using .fnstart/.fnend and inserting a nop before both __restore and
+// __restore_rt (but covered by the .fnstart/.fnend) so that although they're
+// not inside the functions from objdump's point of view, an unwinder that
+// blindly looks at the previous instruction (but is then smart enough to check
+// the unwind information to find out where it landed) gets the right answer.
+// Make sure not to have both DWARF and ARM unwind information, so only
+// use the ARM unwind information.
+
+// We need to place .fnstart ourselves (but we may as well keep the free .fnend).
+#undef __bionic_asm_custom_entry
+#define __bionic_asm_custom_entry(f)
+
+  .fnstart
+  .save {r0-r15}
+  .pad #32
+  nop
+ENTRY_PRIVATE_NO_DWARF(__restore)
+  // This function must have exactly this instruction sequence.
+  mov r7, #__NR_sigreturn
+  swi #0
+END_NO_DWARF(__restore)
+
+  .fnstart
+  .save {r0-r15}
+  .pad #160
+  nop
+ENTRY_PRIVATE_NO_DWARF(__restore_rt)
+  // This function must have exactly this instruction sequence.
+  mov r7, #__NR_rt_sigreturn
+  swi #0
+END_NO_DWARF(__restore_rt)

libc/arch-arm/bionic/_exit_with_stack_teardown.S

@@ -26,19 +26,16 @@
  * SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
-#include <asm/unistd.h>
+#include <private/bionic_asm.h>
 
-// void _exit_with_stack_teardown(void* stackBase, int stackSize, int retCode)
-ENTRY(_exit_with_stack_teardown)
-    mov     lr, r2
-    ldr     r7, =__NR_munmap
-    swi     #0              // the stack is destroyed by this call
-    mov     r0, lr
-    ldr     r7, =__NR_exit
-    swi     #0
+// void _exit_with_stack_teardown(void* stackBase, size_t stackSize)
+ENTRY_PRIVATE(_exit_with_stack_teardown)
+  ldr r7, =__NR_munmap
+  swi #0
+  // If munmap failed, we ignore the failure and exit anyway.
 
-    // exit() should never return, cause a crash if it does
-    mov     r0, #0
-    ldr     r0, [r0]
+  mov r0, #0
+  ldr r7, =__NR_exit
+  swi #0
+  // The exit syscall does not return.
 END(_exit_with_stack_teardown)

libc/arch-arm/bionic/_setjmp.S

@@ -1,113 +0,0 @@
-/*	$OpenBSD: _setjmp.S,v 1.2 2004/02/01 05:40:52 drahn Exp $	*/
-/*	$NetBSD: _setjmp.S,v 1.5 2003/04/05 23:08:51 bjh21 Exp $	*/
-
-/*
- * Copyright (c) 1997 Mark Brinicombe
- * Copyright (c) 2010 Android Open Source Project.
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- * 3. All advertising materials mentioning features or use of this software
- *    must display the following acknowledgement:
- *	This product includes software developed by Mark Brinicombe
- * 4. Neither the name of the University nor the names of its contributors
- *    may be used to endorse or promote products derived from this software
- *    without specific prior written permission.
- *
- * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
- * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
- * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
- * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
- * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
- * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
- * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
- * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
- * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
- * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include <machine/asm.h>
-#include <machine/setjmp.h>
-#include <machine/cpu-features.h>
-
-/*
- * C library -- _setjmp, _longjmp
- *
- *	_longjmp(a,v)
- * will generate a "return(v)" from the last call to
- *	_setjmp(a)
- * by restoring registers from the stack.
- * The previous signal state is NOT restored.
- *
- * Note: r0 is the return value
- *       r1-r3 are scratch registers in functions
- */
-
-ENTRY(_setjmp)
-	ldr	r1, .L_setjmp_magic
-	str	r1, [r0, #(_JB_MAGIC * 4)]
-
-	/* Store core registers */
-	add     r1, r0, #(_JB_CORE_BASE * 4)
-	stmia   r1, {r4-r14}
-
-#ifdef __ARM_HAVE_VFP
-	/* Store floating-point registers */
-	add     r1, r0, #(_JB_FLOAT_BASE * 4)
-	vstmia  r1, {d8-d15}
-	/* Store floating-point state */
-	fmrx    r1, fpscr
-	str     r1, [r0, #(_JB_FLOAT_STATE * 4)]
-#endif  /* __ARM_HAVE_VFP */
-
-        mov	r0, #0x00000000
-        bx      lr
-END(_setjmp)
-
-.L_setjmp_magic:
-	.word	_JB_MAGIC__SETJMP
-
-ENTRY(_longjmp)
-	ldr	r2, .L_setjmp_magic
-	ldr	r3, [r0, #(_JB_MAGIC * 4)]
-	teq	r2, r3
-	bne	botch
-
-#ifdef __ARM_HAVE_VFP
-	/* Restore floating-point registers */
-	add     r2, r0, #(_JB_FLOAT_BASE * 4)
-	vldmia  r2, {d8-d15}
-	/* Restore floating-point state */
-	ldr     r2, [r0, #(_JB_FLOAT_STATE * 4)]
-	fmxr    fpscr, r2
-#endif /* __ARM_HAVE_VFP */
-
-	/* Restore core registers */
-	add     r2, r0, #(_JB_CORE_BASE * 4)
-	ldmia   r2, {r4-r14}
-
-	/* Validate sp and r14 */
-	teq	sp, #0
-	teqne	r14, #0
-	beq	botch
-
-	/* Set return value */
-	mov	r0, r1
-	teq	r0, #0x00000000
-	moveq	r0, #0x00000001
-	bx      lr
-
-	/* validation failed, die die die. */
-botch:
-	bl	PIC_SYM(_C_LABEL(longjmperror), PLT)
-	bl	PIC_SYM(_C_LABEL(abort), PLT)
-	b	. - 8		/* Cannot get here */
-END(_longjmp)

libc/arch-arm/bionic/abort_arm.S

@@ -26,7 +26,7 @@
  * SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
+#include <private/bionic_asm.h>
 
 /*
  * Coding the abort function in assembly so that registers are guaranteed to
@@ -36,7 +36,9 @@
  * sequence when the crash happens.
  */
 ENTRY(abort)
-    .save   {r3, r14}
     stmfd   sp!, {r3, r14}
-    bl      PIC_SYM(_C_LABEL(__libc_android_abort), PLT)
+    .cfi_def_cfa_offset 8
+    .cfi_rel_offset r3, 0
+    .cfi_rel_offset r14, 4
+    bl      __libc_android_abort
 END(abort)

libc/arch-arm/bionic/atexit.h

@@ -1,35 +0,0 @@
-/*
- * Copyright (C) 2012 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-extern void *__dso_handle;
-
-__attribute__ ((visibility ("hidden")))
-int atexit(void (*func)(void))
-{
-  return (__cxa_atexit((void (*)(void *))func, (void *)0, &__dso_handle));
-}

libc/arch-arm/bionic/atexit_legacy.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2012 The Android Open Source Project
+ * Copyright (C) 2012 The Android Open Source Project
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -27,9 +27,10 @@
  */
 
 #include <sys/types.h>
-#include <private/libc_logging.h>
 #include <stdio.h>
 
+#include "private/libc_logging.h"
+
 /*
  * This source file should only be included by libc.so, its purpose is
  * to support legacy ARM binaries by exporting a publicly visible

libc/arch-arm/bionic/clone.S

@@ -1,111 +0,0 @@
-/*
- * Copyright (C) 2008-2010 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include <linux/err.h>
-#include <machine/asm.h>
-#include <asm/unistd.h>
-
-// int  __pthread_clone(void* (*fn)(void*), void* child_stack, int flags, void* arg);
-ENTRY(__pthread_clone)
-    # Push 'fn' and 'arg' onto 'child_stack'.
-    stmdb   r1!, {r0, r3}
-
-    # The sys_clone system call only takes two arguments: 'flags' and 'child_stack'.
-    # 'child_stack' is already in r1, but we need to move 'flags' into position.
-    mov     r0, r2
-
-    # System call.
-    mov     ip, r7
-    ldr     r7, =__NR_clone
-    swi     #0
-
-    # Child?
-    movs    r0, r0
-    beq     1f
-
-    # Parent.
-    mov     r7, ip
-    cmn     r0, #(MAX_ERRNO + 1)
-    bxls    lr
-    neg     r0, r0
-    b       __set_errno
-
-1:  # Child.
-    # Pop 'fn' and 'arg' back off the stack and call __thread_entry.
-    pop     {r0, r1}
-    # __thread_entry also needs our stack pointer.
-    mov     r2, sp
-    b       __thread_entry
-END(__pthread_clone)
-
-
-    #
-    # This function is defined as:
-    #
-    #   pid_t  __bionic_clone( int  flags, void *child_stack,
-    #                          pid_t *pid, void *tls, pid_t *ctid,
-    #                          int  (*fn)(void *), void* arg );
-    #
-    # NOTE: This is not the same signature as the glibc
-    #       __clone function. Placing 'fn' and 'arg'
-    #       at the end of the parameter list makes the
-    #       implementation much simpler.
-    #
-
-ENTRY(__bionic_clone)
-    mov     ip, sp
-    .save   {r4, r5, r6, r7}
-
-    # save registers to parent stack
-    stmfd   sp!, {r4, r5, r6, r7}
-
-    # load extra parameters
-    ldmfd   ip, {r4, r5, r6}
-
-    # store 'fn' and 'arg' to the child stack
-    str     r5, [r1, #-4]
-    str     r6, [r1, #-8]
-
-    # System call
-    ldr     r7, =__NR_clone
-    swi     #0
-    movs    r0, r0
-    beq     1f
-
-    # In the parent, reload saved registers then either return or set errno.
-    ldmfd   sp!, {r4, r5, r6, r7}
-    cmn     r0, #(MAX_ERRNO + 1)
-    bxls    lr
-    neg     r0, r0
-    b       __set_errno
-
-1:  # The child.
-    ldr    r0, [sp, #-4]
-    ldr    r1, [sp, #-8]
-    b      __bionic_clone_entry
-END(__bionic_clone)

libc/arch-arm/bionic/crtbegin.c

@@ -1,53 +0,0 @@
-/*
- * Copyright (C) 2012 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include "../../bionic/libc_init_common.h"
-#include <stddef.h>
-#include <stdint.h>
-
-__attribute__ ((section (".preinit_array")))
-void (*__PREINIT_ARRAY__)(void) = (void (*)(void)) -1;
-
-__attribute__ ((section (".init_array")))
-void (*__INIT_ARRAY__)(void) = (void (*)(void)) -1;
-
-__attribute__ ((section (".fini_array")))
-void (*__FINI_ARRAY__)(void) = (void (*)(void)) -1;
-
-__LIBC_HIDDEN__ void _start() {
-  structors_array_t array;
-  array.preinit_array = &__PREINIT_ARRAY__;
-  array.init_array = &__INIT_ARRAY__;
-  array.fini_array = &__FINI_ARRAY__;
-
-  void* raw_args = (void*) ((uintptr_t) __builtin_frame_address(0) + sizeof(void*));
-  __libc_init(raw_args, NULL, &main, &array);
-}
-
-#include "__dso_handle.h"
-#include "atexit.h"

libc/arch-arm/bionic/crtbegin_so.c

@@ -1,58 +0,0 @@
-/*
- * Copyright (C) 2012 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-extern void __cxa_finalize(void *);
-extern void *__dso_handle;
-
-__attribute__((visibility("hidden"),destructor))
-void __on_dlclose() {
-  __cxa_finalize(&__dso_handle);
-}
-
-/* CRT_LEGACY_WORKAROUND should only be defined when building
- * this file as part of the platform's C library.
- *
- * The C library already defines a function named 'atexit()'
- * for backwards compatibility with older NDK-generated binaries.
- *
- * For newer ones, 'atexit' is actually embedded in the C
- * runtime objects that are linked into the final ELF
- * binary (shared library or executable), and will call
- * __cxa_atexit() in order to un-register any atexit()
- * handler when a library is unloaded.
- *
- * This function must be global *and* hidden. Only the
- * code inside the same ELF binary should be able to access it.
- */
-
-#ifdef CRT_LEGACY_WORKAROUND
-#include "__dso_handle.h"
-#else
-#include "__dso_handle_so.h"
-#include "atexit.h"
-#endif

libc/arch-arm/bionic/crtend.S

@@ -1,40 +0,0 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-	
-	.section .preinit_array, "aw"
-	.long 0
-
-	.section .init_array, "aw"
-	.long 0
-
-	.section .fini_array, "aw"
-	.long 0
-
-#if defined(__linux__) && defined(__ELF__)
-	.section .note.GNU-stack,"",%progbits
-#endif

libc/arch-arm/bionic/crtend_so.S

@@ -1,31 +0,0 @@
-/*
- * Copyright (C) 2010 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#if defined(__linux__) && defined(__ELF__)
-	.section .note.GNU-stack,"",%progbits
-#endif

libc/arch-arm/bionic/eabi.c

@@ -1,103 +0,0 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-#include <stddef.h>
-#include <string.h>
-
-extern int  __cxa_atexit(void (*)(void*), void*, void* );
-
-/* The "C++ ABI for ARM" document states that static C++ constructors,
- * which are called from the .init_array, should manually call
- * __aeabi_atexit() to register static destructors explicitely.
- *
- * Note that 'dso_handle' is the address of a magic linker-generate
- * variable from the shared object that contains the constructor/destructor
- */
-
-/* Make this a weak symbol to avoid a multiple definition error when linking
- * with libstdc++-v3.  */
-int __attribute__((weak))
-__aeabi_atexit (void *object, void (*destructor) (void *), void *dso_handle)
-{
-    return __cxa_atexit(destructor, object, dso_handle);
-}
-
-
-void __aeabi_memcpy8(void *dest, const void *src, size_t n) {
-    memcpy(dest, src, n);
-}
-
-void __aeabi_memcpy4(void *dest, const void *src, size_t n) {
-    memcpy(dest, src, n);
-}
-
-void __aeabi_memcpy(void *dest, const void *src, size_t n) {
-    memcpy(dest, src, n);
-}
-
-
-void __aeabi_memmove8(void *dest, const void *src, size_t n) {
-    memmove(dest, src, n);
-}
-
-void __aeabi_memmove4(void *dest, const void *src, size_t n) {
-    memmove(dest, src, n);
-}
-
-void __aeabi_memmove(void *dest, const void *src, size_t n) {
-    memmove(dest, src, n);
-}
-
-/*
- * __aeabi_memset has the order of its second and third arguments reversed. 
- *  This allows __aeabi_memclr to tail-call __aeabi_memset
- */
- 
-void __aeabi_memset8(void *dest, size_t n, int c) {
-    memset(dest, c, n);
-}
-
-void __aeabi_memset4(void *dest, size_t n, int c) {
-    memset(dest, c, n);
-}
-
-void __aeabi_memset(void *dest, size_t n, int c) {
-    memset(dest, c, n);
-}
-
-
-void __aeabi_memclr8(void *dest, size_t n) {
-    __aeabi_memset8(dest, n, 0);
-}
-
-void __aeabi_memclr4(void *dest, size_t n) {
-    __aeabi_memset4(dest, n, 0);
-}
-
-void __aeabi_memclr(void *dest, size_t n) {
-    __aeabi_memset(dest, n, 0);
-}

libc/arch-arm/bionic/exidx_dynamic.c

@@ -37,7 +37,13 @@
  * the expectation that libc will define it and call through to
  * a differently-named function in the dynamic linker.
  */
-_Unwind_Ptr __gnu_Unwind_Find_exidx(_Unwind_Ptr pc, int *pcount)
-{
+_Unwind_Ptr __gnu_Unwind_Find_exidx_impl(_Unwind_Ptr pc, int *pcount) {
     return dl_unwind_find_exidx(pc, pcount);
 }
+
+_Unwind_Ptr __gnu_Unwind_Find_exidx_impl2(_Unwind_Ptr pc, int *pcount) {
+    return dl_unwind_find_exidx(pc, pcount);
+}
+
+__asm__(".symver __gnu_Unwind_Find_exidx_impl,__gnu_Unwind_Find_exidx@LIBC_PRIVATE");
+__asm__(".symver __gnu_Unwind_Find_exidx_impl2,__gnu_Unwind_Find_exidx@@LIBC_N");

libc/arch-arm/bionic/ffs.S

@@ -1,41 +0,0 @@
-/*	$NetBSD: ffs.S,v 1.5 2003/04/05 23:08:52 bjh21 Exp $	*/
-/*
- * Copyright (c) 2001 Christopher Gilbert
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- * 3. The name of the company nor the name of the author may be used to
- *    endorse or promote products derived from this software without specific
- *    prior written permission.
- *
- * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
- * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
- * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
- * IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
- * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
- * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
- * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
- * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
- * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
- * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include <machine/asm.h>
-#include <machine/cpu-features.h>
-
-ENTRY(ffs)
-	/* Standard trick to isolate bottom bit in r0 or 0 if r0 = 0 on entry */
- 	rsb     r1, r0, #0
- 	ands    r0, r0, r1
-	clzne	r0, r0
-	rsbne	r0, r0, #32
-	bx		lr
-END(ffs)

libc/arch-arm/bionic/futex_arm.S

@@ -1,70 +0,0 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include <asm/unistd.h>
-#include <machine/asm.h>
-
-#define FUTEX_WAIT 0
-#define FUTEX_WAKE 1
-
-// __futex_syscall3(*ftx, op, val)
-ENTRY(__futex_syscall3)
-    mov     ip, r7
-    ldr     r7, =__NR_futex
-    swi     #0
-    mov     r7, ip
-    bx      lr
-END(__futex_syscall3)
-
-// __futex_syscall4(*ftx, op, val, *timespec)
-ENTRY(__futex_syscall4)
-    b __futex_syscall3
-END(__futex_syscall4)
-
-// __futex_wait(*ftx, val, *timespec)
-ENTRY(__futex_wait)
-    mov     ip, r7
-    mov     r3, r2
-    mov     r2, r1
-    mov     r1, #FUTEX_WAIT
-    ldr     r7, =__NR_futex
-    swi     #0
-    mov     r7, ip
-    bx      lr
-END(__futex_wait)
-
-// __futex_wake(*ftx, counter)
-ENTRY(__futex_wake)
-    mov     ip, r7
-    mov     r2, r1
-    mov     r1, #FUTEX_WAKE
-    ldr     r7, =__NR_futex
-    swi     #0
-    mov     r7, ip
-    bx      lr
-END(__futex_wake)

libc/arch-arm/bionic/kill.S

@@ -1,50 +0,0 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include <linux/err.h>
-#include <asm/unistd.h>
-#include <machine/asm.h>
-
-/* unlike our auto-generated syscall stubs, this code saves lr
-   on the stack, as well as a few other registers. this makes
-   our stack unwinder happy, when we generate debug stack
-   traces after the C library or other parts of the system
-   abort due to a fatal runtime error (e.g. detection
-   of a corrupted malloc heap).
-*/
-
-ENTRY(kill)
-    stmfd   sp!, {r4-r7, ip, lr}
-    ldr     r7, =__NR_kill
-    swi     #0
-    ldmfd   sp!, {r4-r7, ip, lr}
-    cmn     r0, #(MAX_ERRNO + 1)
-    bxls    lr
-    neg     r0, r0
-    b       __set_errno
-END(kill)

libc/arch-arm/bionic/libgcc_compat.c

@@ -1,170 +1,159 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
+/* Generated by genlibgcc_compat.py */
 
-/* This file contains dummy references to libgcc.a functions to force the
- * dynamic linker to copy their definition into the final libc.so binary.
- *
- * They are required to ensure backwards binary compatibility with
- * libc.so provided by the platform and binaries built with the NDK or
- * different versions/configurations of toolchains.
- *
- * Now, for a more elaborate description of the issue:
- *
- * libgcc.a is a compiler-specific library containing various helper
- * functions used to implement certain operations that are not necessarily
- * supported by the target CPU. For example, integer division doesn't have a
- * corresponding CPU instruction on ARMv5, and is instead implemented in the
- * compiler-generated machine code as a call to an __idiv helper function.
- *
- * Normally, one has to place libgcc.a in the link command used to generate
- * target binaries (shared libraries and executables) after all objects and
- * static libraries, but before dependent shared libraries, i.e. something
- * like:
- *         gcc <options> -o libfoo.so  foo.a libgcc.a -lc -lm
- *
- * This ensures that any helper function needed by the code in foo.a is copied
- * into the final libfoo.so. However, doing so will link a bunch of other __cxa
- * functions from libgcc.a into each .so and executable, causing 4k+ increase
- * in every binary. Therefore the Android platform build system has been
- * using this instead:
- *
- *         gcc <options> -o libfoo.so foo.a -lc -lm libgcc.a
- *
- * The problem with this is that if one helper function needed by foo.a has
- * already been copied into libc.so or libm.so, then nothing will be copied
- * into libfoo.so. Instead, a symbol import definition will be added to it
- * so libfoo.so can directly call the one in libc.so at runtime.
- *
- * When refreshing toolchains for new versions or using different architecture
- * flags, the set of helper functions copied to libc.so may change, which
- * resulted in some native shared libraries generated with the NDK or prebuilts
- * from vendors to fail to load properly.
- *
- * The NDK has been fixed after 1.6_r1 to use the correct link command, so
- * any native shared library generated with it should now be safe from that
- * problem. On the other hand, existing shared libraries distributed with
- * applications that were generated with a previous version of the NDK
- * still need all 1.5/1.6 helper functions in libc.so and libm.so
- *
- * After 3.2, the toolchain was updated again, adding __aeabi_f2uiz to the
- * list of requirements. Technically, this is due to mis-linked NDK libraries
- * but it is easier to add a single function here than asking several app
- * developers to fix their build.
- *
- * The __aeabi_idiv function is added to the list since cortex-a15 supports
- * HW idiv instructions so the system libc.so doesn't pull in the reference to
- * __aeabi_idiv but legacy libraries built against cortex-a9 targets still need
- * it.
- *
- * Final note: some of the functions below should really be in libm.so to
- *             completely reflect the state of 1.5/1.6 system images. However,
- *             since libm.so depends on libc.so, it's easier to put all of
- *             these in libc.so instead, since the dynamic linker will always
- *             search in libc.so before libm.so for dependencies.
- */
+extern char __adddf3;
+extern char __addsf3;
+extern char __aeabi_cdcmpeq;
+extern char __aeabi_cdcmple;
+extern char __aeabi_cdrcmple;
+extern char __aeabi_d2f;
+extern char __aeabi_d2iz;
+extern char __aeabi_dadd;
+extern char __aeabi_dcmpeq;
+extern char __aeabi_dcmpge;
+extern char __aeabi_dcmpgt;
+extern char __aeabi_dcmple;
+extern char __aeabi_dcmplt;
+extern char __aeabi_dcmpun;
+extern char __aeabi_ddiv;
+extern char __aeabi_dmul;
+extern char __aeabi_drsub;
+extern char __aeabi_dsub;
+extern char __aeabi_f2d;
+extern char __aeabi_f2iz;
+extern char __aeabi_f2uiz;
+extern char __aeabi_fadd;
+extern char __aeabi_fcmpun;
+extern char __aeabi_fdiv;
+extern char __aeabi_fmul;
+extern char __aeabi_frsub;
+extern char __aeabi_fsub;
+extern char __aeabi_i2d;
+extern char __aeabi_i2f;
+extern char __aeabi_idiv;
+extern char __aeabi_idivmod;
+extern char __aeabi_l2d;
+extern char __aeabi_l2f;
+extern char __aeabi_lasr;
+extern char __aeabi_ldivmod;
+extern char __aeabi_llsl;
+extern char __aeabi_llsr;
+extern char __aeabi_lmul;
+extern char __aeabi_ui2d;
+extern char __aeabi_ui2f;
+extern char __aeabi_uidiv;
+extern char __aeabi_uidivmod;
+extern char __aeabi_ul2d;
+extern char __aeabi_ul2f;
+extern char __aeabi_uldivmod;
+extern char __aeabi_unwind_cpp_pr0;
+extern char __aeabi_unwind_cpp_pr1;
+extern char __cmpdf2;
+extern char __divdf3;
+extern char __divsf3;
+extern char __eqdf2;
+extern char __extendsfdf2;
+extern char __fixdfsi;
+extern char __fixsfsi;
+extern char __floatdidf;
+extern char __floatdisf;
+extern char __floatsidf;
+extern char __floatsisf;
+extern char __floatundidf;
+extern char __floatundisf;
+extern char __floatunsidf;
+extern char __floatunsisf;
+extern char __gedf2;
+extern char __gtdf2;
+extern char __ledf2;
+extern char __ltdf2;
+extern char __muldf3;
+extern char __muldi3;
+extern char __mulsf3;
+extern char __nedf2;
+extern char __popcount_tab;
+extern char __popcountsi2;
+extern char __subdf3;
+extern char __subsf3;
+extern char __truncdfsf2;
+extern char __unorddf2;
+extern char __unordsf2;
 
-#define   COMPAT_FUNCTIONS_LIST \
-    XX(__adddf3)             \
-    XX(__addsf3)             \
-    XX(__aeabi_cdcmpeq)      \
-    XX(__aeabi_cdcmple)      \
-    XX(__aeabi_cdrcmple)     \
-    XX(__aeabi_d2f)          \
-    XX(__aeabi_d2iz)         \
-    XX(__aeabi_dadd)         \
-    XX(__aeabi_dcmpeq)       \
-    XX(__aeabi_dcmpge)       \
-    XX(__aeabi_dcmpgt)       \
-    XX(__aeabi_dcmple)       \
-    XX(__aeabi_dcmplt)       \
-    XX(__aeabi_dcmpun)       \
-    XX(__aeabi_ddiv)         \
-    XX(__aeabi_dmul)         \
-    XX(__aeabi_drsub)        \
-    XX(__aeabi_dsub)         \
-    XX(__aeabi_f2d)          \
-    XX(__aeabi_f2iz)         \
-    XX(__aeabi_f2uiz)        \
-    XX(__aeabi_fadd)         \
-    XX(__aeabi_fcmpun)       \
-    XX(__aeabi_fdiv)         \
-    XX(__aeabi_fmul)         \
-    XX(__aeabi_frsub)        \
-    XX(__aeabi_fsub)         \
-    XX(__aeabi_i2d)          \
-    XX(__aeabi_i2f)          \
-    XX(__aeabi_idiv)         \
-    XX(__aeabi_l2d)          \
-    XX(__aeabi_l2f)          \
-    XX(__aeabi_lasr)         \
-    XX(__aeabi_lmul)         \
-    XX(__aeabi_llsl)         \
-    XX(__aeabi_llsr)         \
-    XX(__aeabi_ui2d)         \
-    XX(__aeabi_ui2f)         \
-    XX(__aeabi_ul2d)         \
-    XX(__aeabi_ul2f)         \
-    XX(__cmpdf2)             \
-    XX(__divdf3)             \
-    XX(__divsf3)             \
-    XX(__eqdf2)              \
-    XX(__extendsfdf2)        \
-    XX(__fixdfsi)            \
-    XX(__fixsfsi)            \
-    XX(__floatdidf)          \
-    XX(__floatdisf)          \
-    XX(__floatsidf)          \
-    XX(__floatsisf)          \
-    XX(__floatundidf)        \
-    XX(__floatundisf)        \
-    XX(__floatunsidf)        \
-    XX(__floatunsisf)        \
-    XX(__gedf2)              \
-    XX(__gtdf2)              \
-    XX(__ledf2)              \
-    XX(__ltdf2)              \
-    XX(__muldf3)             \
-    XX(__muldi3)             \
-    XX(__mulsf3)             \
-    XX(__nedf2)              \
-    XX(__subdf3)             \
-    XX(__subsf3)             \
-    XX(__truncdfsf2)         \
-    XX(__unorddf2)           \
-    XX(__unordsf2)           \
-
-#define  XX(f)    extern void f(void);
-COMPAT_FUNCTIONS_LIST
-#undef XX
-
-void  __bionic_libgcc_compat_hooks(void)
-{
-#define XX(f)    f();
-COMPAT_FUNCTIONS_LIST
-#undef XX
-}
+void* __bionic_libgcc_compat_symbols[] = {
+    &__adddf3,
+    &__addsf3,
+    &__aeabi_cdcmpeq,
+    &__aeabi_cdcmple,
+    &__aeabi_cdrcmple,
+    &__aeabi_d2f,
+    &__aeabi_d2iz,
+    &__aeabi_dadd,
+    &__aeabi_dcmpeq,
+    &__aeabi_dcmpge,
+    &__aeabi_dcmpgt,
+    &__aeabi_dcmple,
+    &__aeabi_dcmplt,
+    &__aeabi_dcmpun,
+    &__aeabi_ddiv,
+    &__aeabi_dmul,
+    &__aeabi_drsub,
+    &__aeabi_dsub,
+    &__aeabi_f2d,
+    &__aeabi_f2iz,
+    &__aeabi_f2uiz,
+    &__aeabi_fadd,
+    &__aeabi_fcmpun,
+    &__aeabi_fdiv,
+    &__aeabi_fmul,
+    &__aeabi_frsub,
+    &__aeabi_fsub,
+    &__aeabi_i2d,
+    &__aeabi_i2f,
+    &__aeabi_idiv,
+    &__aeabi_idivmod,
+    &__aeabi_l2d,
+    &__aeabi_l2f,
+    &__aeabi_lasr,
+    &__aeabi_ldivmod,
+    &__aeabi_llsl,
+    &__aeabi_llsr,
+    &__aeabi_lmul,
+    &__aeabi_ui2d,
+    &__aeabi_ui2f,
+    &__aeabi_uidiv,
+    &__aeabi_uidivmod,
+    &__aeabi_ul2d,
+    &__aeabi_ul2f,
+    &__aeabi_uldivmod,
+    &__aeabi_unwind_cpp_pr0,
+    &__aeabi_unwind_cpp_pr1,
+    &__cmpdf2,
+    &__divdf3,
+    &__divsf3,
+    &__eqdf2,
+    &__extendsfdf2,
+    &__fixdfsi,
+    &__fixsfsi,
+    &__floatdidf,
+    &__floatdisf,
+    &__floatsidf,
+    &__floatsisf,
+    &__floatundidf,
+    &__floatundisf,
+    &__floatunsidf,
+    &__floatunsisf,
+    &__gedf2,
+    &__gtdf2,
+    &__ledf2,
+    &__ltdf2,
+    &__muldf3,
+    &__muldi3,
+    &__mulsf3,
+    &__nedf2,
+    &__popcount_tab,
+    &__popcountsi2,
+    &__subdf3,
+    &__subsf3,
+    &__truncdfsf2,
+    &__unorddf2,
+    &__unordsf2,
+};

libc/arch-arm/bionic/memcmp.S

@@ -1,341 +0,0 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include <machine/cpu-features.h>
-#include <machine/asm.h>
-
-
-#ifdef HAVE_32_BYTE_CACHE_LINE
-#define CACHE_LINE_SIZE     32
-#else
-#define CACHE_LINE_SIZE     64
-#endif
-
-/*
- * Optimized memcmp() for Cortex-A9.
- */
-
-ENTRY(memcmp)
-        pld         [r0, #(CACHE_LINE_SIZE * 0)]
-        pld         [r0, #(CACHE_LINE_SIZE * 1)]
-
-        /* take of the case where length is 0 or the buffers are the same */
-        cmp         r0, r1
-        moveq       r0, #0
-        bxeq        lr
-
-        pld         [r1, #(CACHE_LINE_SIZE * 0)]
-        pld         [r1, #(CACHE_LINE_SIZE * 1)]
-
-        /* make sure we have at least 8+4 bytes, this simplify things below
-         * and avoid some overhead for small blocks
-         */
-        cmp        r2, #(8+4)
-        bmi        10f
-/*
- * Neon optimization
- * Comparing 32 bytes at a time
- */
-#if defined(__ARM_NEON__) && defined(NEON_UNALIGNED_ACCESS)
-        subs        r2, r2, #32
-        blo         3f
-
-        /* preload all the cache lines we need. */
-        pld         [r0, #(CACHE_LINE_SIZE * 2)]
-        pld         [r1, #(CACHE_LINE_SIZE * 2)]
-
-1:      /* The main loop compares 32 bytes at a time */
-        vld1.8      {d0 - d3}, [r0]!
-        pld         [r0, #(CACHE_LINE_SIZE * 2)]
-        vld1.8      {d4 - d7}, [r1]!
-        pld         [r1, #(CACHE_LINE_SIZE * 2)]
-
-        /* Start subtracting the values and merge results */
-        vsub.i8     q0, q2
-        vsub.i8     q1, q3
-        vorr        q2, q0, q1
-        vorr        d4, d5
-        vmov        r3, ip, d4
-        /* Check if there are any differences among the 32 bytes */
-        orrs        r3, ip
-        bne         2f
-        subs        r2, r2, #32
-        bhs         1b
-        b           3f
-2:
-        /* Check if the difference was in the first or last 16 bytes */
-        sub         r0, #32
-        vorr        d0, d1
-        sub         r1, #32
-        vmov        r3, ip, d0
-        orrs        r3, ip
-        /* if the first 16 bytes are equal, we only have to rewind 16 bytes */
-        ittt        eq
-        subeq       r2, #16
-        addeq       r0, #16
-        addeq       r1, #16
-
-3:      /* fix-up the remaining count */
-        add         r2, r2, #32
-
-        cmp        r2, #(8+4)
-        bmi        10f
-#endif
-
-        .save {r4, lr}
-        /* save registers */
-        stmfd       sp!, {r4, lr}
-
-        /* since r0 hold the result, move the first source
-         * pointer somewhere else
-         */
-         mov        r4, r0
-
-        /* align first pointer to word boundary
-         * offset = -src & 3
-         */
-        rsb         r3, r4, #0
-        ands        r3, r3, #3
-        beq         0f
-
-        /* align first pointer  */
-        sub         r2, r2, r3
-1:      ldrb        r0, [r4], #1
-        ldrb        ip, [r1], #1
-        subs        r0, r0, ip
-        bne         9f
-        subs        r3, r3, #1
-        bne         1b
-
-
-0:      /* here the first pointer is aligned, and we have at least 4 bytes
-         * to process.
-         */
-
-        /* see if the pointers are congruent */
-        eor         r0, r4, r1
-        ands        r0, r0, #3
-        bne         5f
-
-        /* congruent case, 32 bytes per iteration
-         * We need to make sure there are at least 32+4 bytes left
-         * because we effectively read ahead one word, and we could
-         * read past the buffer (and segfault) if we're not careful.
-         */
-
-        ldr         ip, [r1]
-        subs        r2, r2, #(32 + 4)
-        bmi         1f
-
-0:      pld         [r4, #(CACHE_LINE_SIZE * 2)]
-        pld         [r1, #(CACHE_LINE_SIZE * 2)]
-        ldr         r0, [r4], #4
-        ldr         lr, [r1, #4]!
-        eors        r0, r0, ip
-        ldreq       r0, [r4], #4
-        ldreq       ip, [r1, #4]!
-        eoreqs      r0, r0, lr
-        ldreq       r0, [r4], #4
-        ldreq       lr, [r1, #4]!
-        eoreqs      r0, r0, ip
-        ldreq       r0, [r4], #4
-        ldreq       ip, [r1, #4]!
-        eoreqs      r0, r0, lr
-        ldreq       r0, [r4], #4
-        ldreq       lr, [r1, #4]!
-        eoreqs      r0, r0, ip
-        ldreq       r0, [r4], #4
-        ldreq       ip, [r1, #4]!
-        eoreqs      r0, r0, lr
-        ldreq       r0, [r4], #4
-        ldreq       lr, [r1, #4]!
-        eoreqs      r0, r0, ip
-        ldreq       r0, [r4], #4
-        ldreq       ip, [r1, #4]!
-        eoreqs      r0, r0, lr
-        bne         2f
-        subs        r2, r2, #32
-        bhs         0b
-
-        /* do we have at least 4 bytes left? */
-1:      adds        r2, r2, #(32 - 4 + 4)
-        bmi         4f
-
-        /* finish off 4 bytes at a time */
-3:      ldr         r0, [r4], #4
-        ldr         ip, [r1], #4
-        eors        r0, r0, ip
-        bne         2f
-        subs        r2, r2, #4
-        bhs         3b
-
-        /* are we done? */
-4:      adds        r2, r2, #4
-        moveq       r0, #0
-        beq         9f
-
-        /* finish off the remaining bytes */
-        b           8f
-
-2:      /* the last 4 bytes are different, restart them */
-        sub         r4, r4, #4
-        sub         r1, r1, #4
-        mov         r2, #4
-
-        /* process the last few bytes */
-8:      ldrb        r0, [r4], #1
-        ldrb        ip, [r1], #1
-        // stall
-        subs        r0, r0, ip
-        bne         9f
-        subs        r2, r2, #1
-        bne         8b
-
-9:      /* restore registers and return */
-        ldmfd       sp!, {r4, lr}
-        bx          lr
-
-10:     /* process less than 12 bytes */
-        cmp         r2, #0
-        moveq       r0, #0
-        bxeq        lr
-        mov         r3, r0
-11:
-        ldrb        r0, [r3], #1
-        ldrb        ip, [r1], #1
-        subs        r0, ip
-        bxne        lr
-        subs        r2, r2, #1
-        bne         11b
-        bx          lr
-
-5:      /*************** non-congruent case ***************/
-        and         r0, r1, #3
-        cmp         r0, #2
-        bne         4f
-
-        /* here, offset is 2 (16-bits aligned, special cased) */
-
-        /* make sure we have at least 16 bytes to process */
-        subs        r2, r2, #16
-        addmi       r2, r2, #16
-        bmi         8b
-
-        /* align the unaligned pointer */
-        bic         r1, r1, #3
-        ldr         lr, [r1], #4
-
-6:      pld         [r1, #(CACHE_LINE_SIZE * 2)]
-        pld         [r4, #(CACHE_LINE_SIZE * 2)]
-        mov         ip, lr, lsr #16
-        ldr         lr, [r1], #4
-        ldr         r0, [r4], #4
-        orr         ip, ip, lr, lsl #16
-        eors        r0, r0, ip
-        moveq       ip, lr, lsr #16
-        ldreq       lr, [r1], #4
-        ldreq       r0, [r4], #4
-        orreq       ip, ip, lr, lsl #16
-        eoreqs      r0, r0, ip
-        moveq       ip, lr, lsr #16
-        ldreq       lr, [r1], #4
-        ldreq       r0, [r4], #4
-        orreq       ip, ip, lr, lsl #16
-        eoreqs      r0, r0, ip
-        moveq       ip, lr, lsr #16
-        ldreq       lr, [r1], #4
-        ldreq       r0, [r4], #4
-        orreq       ip, ip, lr, lsl #16
-        eoreqs      r0, r0, ip
-        bne         7f
-        subs        r2, r2, #16
-        bhs         6b
-        sub         r1, r1, #2
-        /* are we done? */
-        adds        r2, r2, #16
-        moveq       r0, #0
-        beq         9b
-        /* finish off the remaining bytes */
-        b           8b
-
-7:      /* fix up the 2 pointers and fallthrough... */
-        sub         r1, r1, #(4+2)
-        sub         r4, r4, #4
-        mov         r2, #4
-        b           8b
-
-
-4:      /*************** offset is 1 or 3 (less optimized) ***************/
-
-		stmfd		sp!, {r5, r6, r7}
-
-        // r5 = rhs
-        // r6 = lhs
-        // r7 = scratch
-
-        mov         r5, r0, lsl #3		/* r5 = right shift */
-        rsb         r6, r5, #32         /* r6 = left shift */
-
-        /* align the unaligned pointer */
-        bic         r1, r1, #3
-        ldr         r7, [r1], #4
-        sub         r2, r2, #8
-
-6:      mov         ip, r7, lsr r5
-        ldr         r7, [r1], #4
-        ldr         r0, [r4], #4
-        orr         ip, ip, r7, lsl r6
-        eors        r0, r0, ip
-        moveq       ip, r7, lsr r5
-        ldreq       r7, [r1], #4
-        ldreq       r0, [r4], #4
-        orreq       ip, ip, r7, lsl r6
-        eoreqs      r0, r0, ip
-        bne         7f
-        subs        r2, r2, #8
-        bhs         6b
-
-        sub         r1, r1, r6, lsr #3
-		ldmfd       sp!, {r5, r6, r7}
-
-        /* are we done? */
-        adds        r2, r2, #8
-        moveq       r0, #0
-        beq         9b
-
-        /* finish off the remaining bytes */
-        b           8b
-
-7:      /* fix up the 2 pointers and fallthrough... */
-        sub         r1, r1, #4
-        sub         r1, r1, r6, lsr #3
-        sub         r4, r4, #4
-        mov         r2, #4
-		ldmfd		sp!, {r5, r6, r7}
-        b           8b
-END(memcmp)

libc/arch-arm/bionic/memcmp16.S

@@ -1,232 +0,0 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include <machine/cpu-features.h>
-#include <machine/asm.h>
-
-/*
- * Optimized memcmp16() for ARM9.
- * This would not be optimal on XScale or ARM11, where more prefetching
- * and use of PLD will be needed.
- * The 2 major optimzations here are
- * (1) The main loop compares 16 bytes at a time
- * (2) The loads are scheduled in a way they won't stall
- */
-
-ENTRY(__memcmp16)
-        PLD         (r0, #0)
-        PLD         (r1, #0)
-
-        /* take of the case where length is nul or the buffers are the same */
-        cmp         r0, r1
-        cmpne       r2, #0
-        moveq       r0, #0
-        bxeq        lr
-
-        /* since r0 hold the result, move the first source
-         * pointer somewhere else
-         */
-
-        mov         r3, r0
-
-         /* make sure we have at least 12 words, this simplify things below
-          * and avoid some overhead for small blocks
-          */
-
-        cmp         r2, #12
-        bpl         0f
-
-        /* small blocks (less then 12 words) */
-        PLD         (r0, #32)
-        PLD         (r1, #32)
-
-1:      ldrh        r0, [r3], #2
-        ldrh        ip, [r1], #2
-        subs        r0, r0, ip
-        bxne        lr        
-        subs        r2, r2, #1
-        bne         1b
-        bx          lr
-
-
-        .save {r4, lr}
-        /* save registers */
-0:      stmfd       sp!, {r4, lr}
-        
-        /* align first pointer to word boundary */
-        tst         r3, #2
-        beq         0f
-        
-        ldrh        r0, [r3], #2
-        ldrh        ip, [r1], #2
-        sub         r2, r2, #1
-        subs        r0, r0, ip
-        /* restore registers and return */
-        ldmnefd     sp!, {r4, lr}
-        bxne        lr
-
-
-0:      /* here the first pointer is aligned, and we have at least 3 words
-         * to process.
-         */
-
-        /* see if the pointers are congruent */
-        eor         r0, r3, r1
-        ands        r0, r0, #2
-        bne         5f
-
-        /* congruent case, 16 half-words per iteration
-         * We need to make sure there are at least 16+2 words left
-         * because we effectively read ahead one long word, and we could
-         * read past the buffer (and segfault) if we're not careful.
-         */
-
-        ldr         ip, [r1]
-        subs        r2, r2, #(16 + 2)
-        bmi         1f
-        
-0:
-        PLD         (r3, #64)
-        PLD         (r1, #64)
-        ldr         r0, [r3], #4
-        ldr         lr, [r1, #4]!
-        eors        r0, r0, ip
-        ldreq       r0, [r3], #4
-        ldreq       ip, [r1, #4]!
-        eoreqs      r0, r0, lr
-        ldreq       r0, [r3], #4
-        ldreq       lr, [r1, #4]!
-        eoreqs      r0, r0, ip
-        ldreq       r0, [r3], #4
-        ldreq       ip, [r1, #4]!
-        eoreqs      r0, r0, lr
-        ldreq       r0, [r3], #4
-        ldreq       lr, [r1, #4]!
-        eoreqs      r0, r0, ip
-        ldreq       r0, [r3], #4
-        ldreq       ip, [r1, #4]!
-        eoreqs      r0, r0, lr
-        ldreq       r0, [r3], #4
-        ldreq       lr, [r1, #4]!
-        eoreqs      r0, r0, ip
-        ldreq       r0, [r3], #4
-        ldreq       ip, [r1, #4]!
-        eoreqs      r0, r0, lr
-        bne         2f        
-        subs        r2, r2, #16
-        bhs         0b
-
-        /* do we have at least 2 words left? */
-1:      adds        r2, r2, #(16 - 2 + 2)
-        bmi         4f
-        
-        /* finish off 2 words at a time */
-3:      ldr         r0, [r3], #4
-        ldr         ip, [r1], #4
-        eors        r0, r0, ip
-        bne         2f
-        subs        r2, r2, #2
-        bhs         3b
-
-        /* are we done? */
-4:      adds        r2, r2, #2
-        bne         8f
-        /* restore registers and return */
-        mov         r0, #0
-        ldmfd       sp!, {r4, lr}
-        bx          lr
-
-2:      /* the last 2 words are different, restart them */
-        ldrh        r0, [r3, #-4]
-        ldrh        ip, [r1, #-4]
-        subs        r0, r0, ip
-        ldreqh      r0, [r3, #-2]
-        ldreqh      ip, [r1, #-2]
-        subeqs      r0, r0, ip
-        /* restore registers and return */
-        ldmfd       sp!, {r4, lr}
-        bx          lr
-
-        /* process the last few words */
-8:      ldrh        r0, [r3], #2
-        ldrh        ip, [r1], #2
-        subs        r0, r0, ip
-        bne         9f
-        subs        r2, r2, #1
-        bne         8b
-
-9:      /* restore registers and return */
-        ldmfd       sp!, {r4, lr}
-        bx          lr
-
-
-5:      /*************** non-congruent case ***************/
-
-        /* align the unaligned pointer */
-        bic         r1, r1, #3
-        ldr         lr, [r1], #4
-        sub         r2, r2, #8
-
-6:
-        PLD         (r3, #64)
-        PLD         (r1, #64)
-        mov         ip, lr, lsr #16
-        ldr         lr, [r1], #4
-        ldr         r0, [r3], #4
-        orr         ip, ip, lr, lsl #16
-        eors        r0, r0, ip
-        moveq       ip, lr, lsr #16
-        ldreq       lr, [r1], #4
-        ldreq       r0, [r3], #4
-        orreq       ip, ip, lr, lsl #16
-        eoreqs      r0, r0, ip
-        moveq       ip, lr, lsr #16
-        ldreq       lr, [r1], #4
-        ldreq       r0, [r3], #4
-        orreq       ip, ip, lr, lsl #16
-        eoreqs      r0, r0, ip
-        moveq       ip, lr, lsr #16
-        ldreq       lr, [r1], #4
-        ldreq       r0, [r3], #4
-        orreq       ip, ip, lr, lsl #16
-        eoreqs      r0, r0, ip
-        bne         7f
-        subs        r2, r2, #8
-        bhs         6b
-        sub         r1, r1, #2
-        /* are we done? */
-        adds        r2, r2, #8
-        moveq       r0, #0
-        beq         9b
-        /* finish off the remaining bytes */
-        b           8b
-
-7:      /* fix up the 2 pointers and fallthrough... */
-        sub         r1, r1, #2
-        b           2b
-END(__memcmp16)

libc/arch-arm/bionic/memcpy.S

@@ -1,686 +0,0 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include <machine/cpu-features.h>
-#include <machine/asm.h>
-
-#if defined(__ARM_NEON__) && !defined(ARCH_ARM_USE_NON_NEON_MEMCPY)
-
-        .text
-        .fpu    neon
-
-#ifdef HAVE_32_BYTE_CACHE_LINE
-/* a prefetch distance of 2 cache-lines */
-#define CACHE_LINE_SIZE     32
-#else
-/* a prefetch distance of 4 cache-lines works best experimentally */
-#define CACHE_LINE_SIZE     64
-#endif
-
-ENTRY(memcpy)
-        .save       {r0, lr}
-        /* start preloading as early as possible */
-        pld         [r1, #(CACHE_LINE_SIZE * 0)]
-        stmfd       sp!, {r0, lr}
-        pld         [r1, #(CACHE_LINE_SIZE * 1)]
-
-/* If Neon supports unaligned access then remove the align code,
- * unless a size limit has been specified.
- */
-#ifndef NEON_UNALIGNED_ACCESS
-        /* do we have at least 16-bytes to copy (needed for alignment below) */
-        cmp         r2, #16
-        blo         5f
-
-        /* check if buffers are aligned. If so, run arm-only version */
-        eor         r3, r0, r1
-        ands        r3, r3, #0x3
-        beq         11f
-
-        /* align destination to cache-line for the write-buffer */
-        rsb         r3, r0, #0
-        ands        r3, r3, #0xF
-        beq         2f
-
-        /* copy up to 15-bytes (count in r3) */
-        sub         r2, r2, r3
-        movs        ip, r3, lsl #31
-        ldrmib      lr, [r1], #1
-        strmib      lr, [r0], #1
-        ldrcsb      ip, [r1], #1
-        ldrcsb      lr, [r1], #1
-        strcsb      ip, [r0], #1
-        strcsb      lr, [r0], #1
-        movs        ip, r3, lsl #29
-        bge         1f
-        // copies 4 bytes, destination 32-bits aligned
-        vld4.8      {d0[0], d1[0], d2[0], d3[0]}, [r1]!
-        vst4.8      {d0[0], d1[0], d2[0], d3[0]}, [r0, :32]!
-1:      bcc         2f
-        // copies 8 bytes, destination 64-bits aligned
-        vld1.8      {d0}, [r1]!
-        vst1.8      {d0}, [r0, :64]!
-2:
-        /* preload immediately the next cache line, which we may need */
-        pld         [r1, #(CACHE_LINE_SIZE * 0)]
-        pld         [r1, #(CACHE_LINE_SIZE * 1)]
-
-#ifdef HAVE_32_BYTE_CACHE_LINE
-        /* make sure we have at least 32 bytes to copy */
-        subs        r2, r2, #32
-        blo         4f
-
-        /* preload all the cache lines we need.
-         * NOTE: the number of pld below depends on PREFETCH_DISTANCE,
-         * ideally would would increase the distance in the main loop to
-         * avoid the goofy code below. In practice this doesn't seem to make
-         * a big difference.
-         */
-        pld         [r1, #(PREFETCH_DISTANCE)]
-
-1:      /* The main loop copies 32 bytes at a time */
-        vld1.8      {d0  - d3},   [r1]!
-        pld         [r1, #(PREFETCH_DISTANCE)]
-        subs        r2, r2, #32
-        vst1.8      {d0  - d3},   [r0, :128]!
-        bhs         1b
-#else
-        /* make sure we have at least 64 bytes to copy */
-        subs        r2, r2, #64
-        blo         2f
-
-        /* preload all the cache lines we need. */
-        pld         [r1, #(CACHE_LINE_SIZE * 2)]
-        pld         [r1, #(CACHE_LINE_SIZE * 3)]
-
-1:      /* The main loop copies 64 bytes at a time */
-        vld1.8      {d0 - d3}, [r1]!
-        vld1.8      {d4 - d7}, [r1]!
-#ifdef  HAVE_32_BYTE_CACHE_LINE
-        pld         [r1, #(CACHE_LINE_SIZE * 2)]
-        pld         [r1, #(CACHE_LINE_SIZE * 3)]
-#else
-        pld         [r1, #(CACHE_LINE_SIZE * 3)]
-#endif
-        subs        r2, r2, #64
-        vst1.8      {d0 - d3}, [r0, :128]!
-        vst1.8      {d4 - d7}, [r0, :128]!
-        bhs         1b
-
-2:      /* fix-up the remaining count and make sure we have >= 32 bytes left */
-        add         r2, r2, #64
-        subs        r2, r2, #32
-        blo         4f
-
-3:      /* 32 bytes at a time. These cache lines were already preloaded */
-        vld1.8      {d0 - d3}, [r1]!
-        subs        r2, r2, #32
-        vst1.8      {d0 - d3}, [r0, :128]!
-        bhs         3b
-#endif
-4:      /* less than 32 left */
-        add         r2, r2, #32
-        tst         r2, #0x10
-        beq         5f
-        // copies 16 bytes, 128-bits aligned
-        vld1.8      {d0, d1}, [r1]!
-        vst1.8      {d0, d1}, [r0, :128]!
-5:      /* copy up to 15-bytes (count in r2) */
-        movs        ip, r2, lsl #29
-        bcc         1f
-        vld1.8      {d0}, [r1]!
-        vst1.8      {d0}, [r0]!
-1:      bge         2f
-        vld4.8      {d0[0], d1[0], d2[0], d3[0]}, [r1]!
-        vst4.8      {d0[0], d1[0], d2[0], d3[0]}, [r0]!
-2:      movs        ip, r2, lsl #31
-        ldrmib      r3, [r1], #1
-        ldrcsb      ip, [r1], #1
-        ldrcsb      lr, [r1], #1
-        strmib      r3, [r0], #1
-        strcsb      ip, [r0], #1
-        strcsb      lr, [r0], #1
-
-        ldmfd       sp!, {r0, lr}
-        bx          lr
-
-#else   /* NEON_UNALIGNED_ACCESS */
-
-        // Check so divider is at least 16 bytes, needed for alignment code.
-        cmp         r2, #16
-        blo         5f
-
-#ifdef NEON_MEMCPY_ALIGNMENT_DIVIDER
-        /* Check the upper size limit for Neon unaligned memory access in memcpy */
-#if NEON_MEMCPY_ALIGNMENT_DIVIDER >= 16
-        cmp         r2, #NEON_MEMCPY_ALIGNMENT_DIVIDER
-        blo         3f
-#endif
-        /* check if buffers are aligned. If so, run arm-only version */
-        eor         r3, r0, r1
-        ands        r3, r3, #0x3
-        beq         11f
-
-        /* align destination to 16 bytes for the write-buffer */
-        rsb         r3, r0, #0
-        ands        r3, r3, #0xF
-        beq         3f
-
-        /* copy up to 15-bytes (count in r3) */
-        sub         r2, r2, r3
-        movs        ip, r3, lsl #31
-        ldrmib      lr, [r1], #1
-        strmib      lr, [r0], #1
-        ldrcsb      ip, [r1], #1
-        ldrcsb      lr, [r1], #1
-        strcsb      ip, [r0], #1
-        strcsb      lr, [r0], #1
-        movs        ip, r3, lsl #29
-        bge         1f
-        // copies 4 bytes, destination 32-bits aligned
-        vld1.32     {d0[0]}, [r1]!
-        vst1.32     {d0[0]}, [r0, :32]!
-1:      bcc         2f
-        // copies 8 bytes, destination 64-bits aligned
-        vld1.8      {d0}, [r1]!
-        vst1.8      {d0}, [r0, :64]!
-2:
-        /* preload immediately the next cache line, which we may need */
-        pld         [r1, #(CACHE_LINE_SIZE * 0)]
-        pld         [r1, #(CACHE_LINE_SIZE * 1)]
-3:
-#endif
-        /* make sure we have at least 64 bytes to copy */
-        subs        r2, r2, #64
-        blo         2f
-
-        /* preload all the cache lines we need */
-        pld         [r1, #(CACHE_LINE_SIZE * 2)]
-        pld         [r1, #(CACHE_LINE_SIZE * 3)]
-
-1:      /* The main loop copies 64 bytes at a time */
-        vld1.8      {d0 - d3}, [r1]!
-        vld1.8      {d4 - d7}, [r1]!
-#ifdef  HAVE_32_BYTE_CACHE_LINE
-        pld         [r1, #(CACHE_LINE_SIZE * 2)]
-        pld         [r1, #(CACHE_LINE_SIZE * 3)]
-#else
-        pld         [r1, #(CACHE_LINE_SIZE * 3)]
-#endif
-        subs        r2, r2, #64
-        vst1.8      {d0 - d3}, [r0]!
-        vst1.8      {d4 - d7}, [r0]!
-        bhs         1b
-
-2:      /* fix-up the remaining count and make sure we have >= 32 bytes left */
-        add         r2, r2, #64
-        subs        r2, r2, #32
-        blo         4f
-
-3:      /* 32 bytes at a time. These cache lines were already preloaded */
-        vld1.8      {d0 - d3}, [r1]!
-        subs        r2, r2, #32
-        vst1.8      {d0 - d3}, [r0]!
-        bhs         3b
-
-4:      /* less than 32 left */
-        add         r2, r2, #32
-        tst         r2, #0x10
-        beq         5f
-        // copies 16 bytes, 128-bits aligned
-        vld1.8      {d0, d1}, [r1]!
-        vst1.8      {d0, d1}, [r0]!
-5:      /* copy up to 15-bytes (count in r2) */
-        movs        ip, r2, lsl #29
-        bcc         1f
-        vld1.8      {d0}, [r1]!
-        vst1.8      {d0}, [r0]!
-1:      bge         2f
-        vld1.32     {d0[0]}, [r1]!
-        vst1.32     {d0[0]}, [r0]!
-2:      movs        ip, r2, lsl #31
-        ldrmib      r3, [r1], #1
-        ldrcsb      ip, [r1], #1
-        ldrcsb      lr, [r1], #1
-        strmib      r3, [r0], #1
-        strcsb      ip, [r0], #1
-        strcsb      lr, [r0], #1
-
-        ldmfd       sp!, {r0, lr}
-        bx          lr
-#endif  /* NEON_UNALIGNED_ACCESS */
-11:
-        /* Simple arm-only copy loop to handle aligned copy operations */
-        stmfd       sp!, {r4, r5, r6, r7, r8}
-        pld         [r1, #(CACHE_LINE_SIZE * 2)]
-
-        /* Check alignment */
-        rsb         r3, r1, #0
-        ands        r3, #3
-        beq         2f
-
-        /* align source to 32 bits. We need to insert 2 instructions between
-         * a ldr[b|h] and str[b|h] because byte and half-word instructions
-         * stall 2 cycles.
-         */
-        movs        r12, r3, lsl #31
-        sub         r2, r2, r3      /* we know that r3 <= r2 because r2 >= 4 */
-        ldrmib      r3, [r1], #1
-        ldrcsb      r4, [r1], #1
-        ldrcsb      r5, [r1], #1
-        strmib      r3, [r0], #1
-        strcsb      r4, [r0], #1
-        strcsb      r5, [r0], #1
-2:
-        subs        r2, #32
-        blt         5f
-        pld         [r1, #(CACHE_LINE_SIZE * 3)]
-3:      /* Main copy loop, copying 32 bytes at a time */
-        pld         [r1, #(CACHE_LINE_SIZE * 4)]
-        ldmia       r1!, {r3, r4, r5, r6, r7, r8, r12, lr}
-        subs        r2, r2, #32
-        stmia       r0!, {r3, r4, r5, r6, r7, r8, r12, lr}
-        bge         3b
-5:      /* Handle any remaining bytes */
-        adds        r2, #32
-        beq         6f
-
-        movs        r12, r2, lsl #28
-        ldmcsia     r1!, {r3, r4, r5, r6}   /* 16 bytes */
-        ldmmiia     r1!, {r7, r8}           /*  8 bytes */
-        stmcsia     r0!, {r3, r4, r5, r6}
-        stmmiia     r0!, {r7, r8}
-        movs        r12, r2, lsl #30
-        ldrcs       r3, [r1], #4            /*  4 bytes */
-        ldrmih      r4, [r1], #2            /*  2 bytes */
-        strcs       r3, [r0], #4
-        strmih      r4, [r0], #2
-        tst         r2, #0x1
-        ldrneb      r3, [r1]                /*  last byte  */
-        strneb      r3, [r0]
-6:
-        ldmfd       sp!, {r4, r5, r6, r7, r8}
-        ldmfd       sp!, {r0, pc}
-END(memcpy)
-
-
-#else   /* __ARM_ARCH__ < 7 */
-
-
-		/*
-		 * Optimized memcpy() for ARM.
-         *
-		 * note that memcpy() always returns the destination pointer,
-		 * so we have to preserve R0.
-		 */
-
-ENTRY(memcpy)
-		/* The stack must always be 64-bits aligned to be compliant with the
-		 * ARM ABI. Since we have to save R0, we might as well save R4
-		 * which we can use for better pipelining of the reads below
-		 */
-        .save       {r0, r4, lr}
-        stmfd       sp!, {r0, r4, lr}
-        /* Making room for r5-r11 which will be spilled later */
-        .pad        #28
-        sub         sp, sp, #28
-
-        // preload the destination because we'll align it to a cache line
-        // with small writes. Also start the source "pump".
-        PLD         (r0, #0)
-        PLD         (r1, #0)
-        PLD         (r1, #32)
-
-		/* it simplifies things to take care of len<4 early */
-		cmp			r2, #4
-		blo			copy_last_3_and_return
-
-		/* compute the offset to align the source
-		 * offset = (4-(src&3))&3 = -src & 3
-		 */
-		rsb			r3, r1, #0
-		ands		r3, r3, #3
-		beq			src_aligned
-
-		/* align source to 32 bits. We need to insert 2 instructions between
-		 * a ldr[b|h] and str[b|h] because byte and half-word instructions
-		 * stall 2 cycles.
-		 */
-		movs		r12, r3, lsl #31
-		sub			r2, r2, r3		/* we know that r3 <= r2 because r2 >= 4 */
-		ldrmib		r3, [r1], #1
-		ldrcsb		r4, [r1], #1
-		ldrcsb		r12,[r1], #1
-        strmib		r3, [r0], #1
-		strcsb		r4, [r0], #1
-		strcsb		r12,[r0], #1
-
-src_aligned:
-
-		/* see if src and dst are aligned together (congruent) */
-		eor			r12, r0, r1
-		tst			r12, #3
-		bne			non_congruent
-
-        /* Use post-incriment mode for stm to spill r5-r11 to reserved stack
-         * frame. Don't update sp.
-         */
-        stmea		sp, {r5-r11}
-
-		/* align the destination to a cache-line */
-		rsb         r3, r0, #0
-		ands		r3, r3, #0x1C
-		beq         congruent_aligned32
-		cmp         r3, r2
-		andhi		r3, r2, #0x1C
-
-		/* conditionnaly copies 0 to 7 words (length in r3) */
-		movs		r12, r3, lsl #28
-		ldmcsia		r1!, {r4, r5, r6, r7}	/* 16 bytes */
-		ldmmiia		r1!, {r8, r9}			/*  8 bytes */
-		stmcsia		r0!, {r4, r5, r6, r7}
-		stmmiia		r0!, {r8, r9}
-		tst         r3, #0x4
-		ldrne		r10,[r1], #4			/*  4 bytes */
-		strne		r10,[r0], #4
-		sub         r2, r2, r3
-
-congruent_aligned32:
-		/*
-		 * here source is aligned to 32 bytes.
-		 */
-
-cached_aligned32:
-        subs        r2, r2, #32
-        blo         less_than_32_left
-
-        /*
-         * We preload a cache-line up to 64 bytes ahead. On the 926, this will
-         * stall only until the requested world is fetched, but the linefill
-         * continues in the the background.
-         * While the linefill is going, we write our previous cache-line
-         * into the write-buffer (which should have some free space).
-         * When the linefill is done, the writebuffer will
-         * start dumping its content into memory
-         *
-         * While all this is going, we then load a full cache line into
-         * 8 registers, this cache line should be in the cache by now
-         * (or partly in the cache).
-         *
-         * This code should work well regardless of the source/dest alignment.
-         *
-         */
-
-        // Align the preload register to a cache-line because the cpu does
-        // "critical word first" (the first word requested is loaded first).
-        bic         r12, r1, #0x1F
-        add         r12, r12, #64
-
-1:      ldmia       r1!, { r4-r11 }
-        PLD         (r12, #64)
-        subs        r2, r2, #32
-
-        // NOTE: if r12 is more than 64 ahead of r1, the following ldrhi
-        // for ARM9 preload will not be safely guarded by the preceding subs.
-        // When it is safely guarded the only possibility to have SIGSEGV here
-        // is because the caller overstates the length.
-        ldrhi       r3, [r12], #32      /* cheap ARM9 preload */
-        stmia       r0!, { r4-r11 }
-		bhs         1b
-
-        add         r2, r2, #32
-
-
-
-
-less_than_32_left:
-		/*
-		 * less than 32 bytes left at this point (length in r2)
-		 */
-
-		/* skip all this if there is nothing to do, which should
-		 * be a common case (if not executed the code below takes
-		 * about 16 cycles)
-		 */
-		tst			r2, #0x1F
-		beq			1f
-
-		/* conditionnaly copies 0 to 31 bytes */
-		movs		r12, r2, lsl #28
-		ldmcsia		r1!, {r4, r5, r6, r7}	/* 16 bytes */
-		ldmmiia		r1!, {r8, r9}			/*  8 bytes */
-		stmcsia		r0!, {r4, r5, r6, r7}
-		stmmiia		r0!, {r8, r9}
-		movs		r12, r2, lsl #30
-		ldrcs		r3, [r1], #4			/*  4 bytes */
-		ldrmih		r4, [r1], #2			/*  2 bytes */
-		strcs		r3, [r0], #4
-		strmih		r4, [r0], #2
-		tst         r2, #0x1
-		ldrneb		r3, [r1]				/*  last byte  */
-		strneb		r3, [r0]
-
-		/* we're done! restore everything and return */
-1:		ldmfd		sp!, {r5-r11}
-		ldmfd		sp!, {r0, r4, lr}
-		bx			lr
-
-		/********************************************************************/
-
-non_congruent:
-		/*
-		 * here source is aligned to 4 bytes
-		 * but destination is not.
-		 *
-		 * in the code below r2 is the number of bytes read
-		 * (the number of bytes written is always smaller, because we have
-		 * partial words in the shift queue)
-		 */
-		cmp			r2, #4
-		blo			copy_last_3_and_return
-
-        /* Use post-incriment mode for stm to spill r5-r11 to reserved stack
-         * frame. Don't update sp.
-         */
-        stmea		sp, {r5-r11}
-
-		/* compute shifts needed to align src to dest */
-		rsb			r5, r0, #0
-		and			r5, r5, #3			/* r5 = # bytes in partial words */
-		mov			r12, r5, lsl #3		/* r12 = right */
-		rsb			lr, r12, #32		/* lr = left  */
-
-		/* read the first word */
-		ldr			r3, [r1], #4
-		sub			r2, r2, #4
-
-		/* write a partial word (0 to 3 bytes), such that destination
-		 * becomes aligned to 32 bits (r5 = nb of words to copy for alignment)
-		 */
-		movs		r5, r5, lsl #31
-		strmib		r3, [r0], #1
-		movmi		r3, r3, lsr #8
-		strcsb		r3, [r0], #1
-		movcs		r3, r3, lsr #8
-		strcsb		r3, [r0], #1
-		movcs		r3, r3, lsr #8
-
-		cmp			r2, #4
-		blo			partial_word_tail
-
-		/* Align destination to 32 bytes (cache line boundary) */
-1:		tst			r0, #0x1c
-		beq			2f
-		ldr			r5, [r1], #4
-		sub         r2, r2, #4
-		orr			r4, r3, r5,		lsl lr
-		mov			r3, r5,			lsr r12
-		str			r4, [r0], #4
-        cmp         r2, #4
-		bhs			1b
-		blo			partial_word_tail
-
-		/* copy 32 bytes at a time */
-2:		subs		r2, r2, #32
-		blo			less_than_thirtytwo
-
-		/* Use immediate mode for the shifts, because there is an extra cycle
-		 * for register shifts, which could account for up to 50% of
-		 * performance hit.
-		 */
-
-        cmp			r12, #24
-		beq			loop24
-		cmp			r12, #8
-		beq			loop8
-
-loop16:
-        ldr         r12, [r1], #4
-1:      mov         r4, r12
-		ldmia		r1!, {   r5,r6,r7,  r8,r9,r10,r11}
-        PLD         (r1, #64)
-        subs        r2, r2, #32
-        ldrhs       r12, [r1], #4
-		orr			r3, r3, r4,		lsl #16
-		mov			r4, r4,			lsr #16
-		orr			r4, r4, r5,		lsl #16
-		mov			r5, r5,			lsr #16
-		orr			r5, r5, r6,		lsl #16
-		mov			r6, r6,			lsr #16
-		orr			r6, r6, r7,		lsl #16
-		mov			r7, r7,			lsr #16
-		orr			r7, r7, r8,		lsl #16
-		mov			r8, r8,			lsr #16
-		orr			r8, r8, r9,		lsl #16
-		mov			r9, r9,			lsr #16
-		orr			r9, r9, r10,	lsl #16
-		mov			r10, r10,		lsr #16
-		orr			r10, r10, r11,	lsl #16
-		stmia		r0!, {r3,r4,r5,r6, r7,r8,r9,r10}
-		mov			r3, r11,		lsr #16
-		bhs			1b
-		b			less_than_thirtytwo
-
-loop8:
-        ldr         r12, [r1], #4
-1:      mov         r4, r12
-		ldmia		r1!, {   r5,r6,r7,  r8,r9,r10,r11}
-        PLD         (r1, #64)
-		subs		r2, r2, #32
-        ldrhs       r12, [r1], #4
-		orr			r3, r3, r4,		lsl #24
-		mov			r4, r4,			lsr #8
-		orr			r4, r4, r5,		lsl #24
-		mov			r5, r5,			lsr #8
-		orr			r5, r5, r6,		lsl #24
-		mov			r6, r6,			lsr #8
-		orr			r6, r6, r7,		lsl #24
-		mov			r7, r7,			lsr #8
-		orr			r7, r7, r8,		lsl #24
-		mov			r8, r8,			lsr #8
-		orr			r8, r8, r9,		lsl #24
-		mov			r9, r9,			lsr #8
-		orr			r9, r9, r10,	lsl #24
-		mov			r10, r10,		lsr #8
-		orr			r10, r10, r11,	lsl #24
-		stmia		r0!, {r3,r4,r5,r6, r7,r8,r9,r10}
-		mov			r3, r11,		lsr #8
-		bhs			1b
-		b			less_than_thirtytwo
-
-loop24:
-        ldr         r12, [r1], #4
-1:      mov         r4, r12
-		ldmia		r1!, {   r5,r6,r7,  r8,r9,r10,r11}
-        PLD         (r1, #64)
-		subs		r2, r2, #32
-        ldrhs       r12, [r1], #4
-		orr			r3, r3, r4,		lsl #8
-		mov			r4, r4,			lsr #24
-		orr			r4, r4, r5,		lsl #8
-		mov			r5, r5,			lsr #24
-		orr			r5, r5, r6,		lsl #8
-		mov			r6, r6,			lsr #24
-		orr			r6, r6, r7,		lsl #8
-		mov			r7, r7,			lsr #24
-		orr			r7, r7, r8,		lsl #8
-		mov			r8, r8,			lsr #24
-		orr			r8, r8, r9,		lsl #8
-		mov			r9, r9,			lsr #24
-		orr			r9, r9, r10,	lsl #8
-		mov			r10, r10,		lsr #24
-		orr			r10, r10, r11,	lsl #8
-		stmia		r0!, {r3,r4,r5,r6, r7,r8,r9,r10}
-		mov			r3, r11,		lsr #24
-		bhs			1b
-
-
-less_than_thirtytwo:
-		/* copy the last 0 to 31 bytes of the source */
-		rsb			r12, lr, #32		/* we corrupted r12, recompute it  */
-		add			r2, r2, #32
-		cmp			r2, #4
-		blo			partial_word_tail
-
-1:		ldr			r5, [r1], #4
-		sub         r2, r2, #4
-		orr			r4, r3, r5,		lsl lr
-		mov			r3,	r5,			lsr r12
-		str			r4, [r0], #4
-        cmp         r2, #4
-		bhs			1b
-
-partial_word_tail:
-		/* we have a partial word in the input buffer */
-		movs		r5, lr, lsl #(31-3)
-		strmib		r3, [r0], #1
-		movmi		r3, r3, lsr #8
-		strcsb		r3, [r0], #1
-		movcs		r3, r3, lsr #8
-		strcsb		r3, [r0], #1
-
-		/* Refill spilled registers from the stack. Don't update sp. */
-		ldmfd		sp, {r5-r11}
-
-copy_last_3_and_return:
-		movs		r2, r2, lsl #31	/* copy remaining 0, 1, 2 or 3 bytes */
-		ldrmib		r2, [r1], #1
-		ldrcsb		r3, [r1], #1
-		ldrcsb		r12,[r1]
-		strmib		r2, [r0], #1
-		strcsb		r3, [r0], #1
-		strcsb		r12,[r0]
-
-        /* we're done! restore sp and spilled registers and return */
-        add         sp,  sp, #28
-		ldmfd		sp!, {r0, r4, lr}
-		bx			lr
-END(memcpy)
-
-
-#endif    /* __ARM_ARCH__ < 7 */

libc/arch-arm/bionic/memcpy.a15.S

@@ -1,406 +0,0 @@
-/*
- * Copyright (c) 2013 ARM Ltd
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- * 3. The name of the company may not be used to endorse or promote
- *    products derived from this software without specific prior written
- *    permission.
- *
- * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
- * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
- * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
- * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
- * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
- * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
- * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
- * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
- * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- */
-
-    /* Prototype: void *memcpy (void *dst, const void *src, size_t count).  */
-
-        /* Use the version of memcpy implemented using LDRD and STRD.
-           This version is tuned for Cortex-A15.
-           This might not be the best for other ARMv7-A CPUs,
-           but there is no predefine to distinguish between
-           different CPUs in the same architecture,
-           and this version is better than the plain memcpy provided in newlib.
-
-           Therefore, we use this version for all ARMv7-A CPUS.  */
-
-        /* To make the same code compile for both ARM and Thumb instruction
-       sets, switch to unified syntax at the beginning of this function.
-           However, by using the same code, we may be missing optimization
-       opportunities.  For instance, in LDRD/STRD instructions, the first
-       destination register must be even and the second consecutive in
-       ARM state, but not in Thumb state.  */
-
-#include <machine/cpu-features.h>
-#include <machine/asm.h>
-
-        .syntax         unified
-
-ENTRY(memcpy)
-
-       /* Assumes that n >= 0, and dst, src are valid pointers.
-          If there is at least 8 bytes to copy, use LDRD/STRD.
-          If src and dst are misaligned with different offsets,
-          first copy byte by byte until dst is aligned,
-          and then copy using LDRD/STRD and shift if needed.
-          When less than 8 left, copy a word and then byte by byte.  */
-
-       /* Save registers (r0 holds the return value):
-          optimized push {r0, r4, r5, r6, r7, lr}.
-          To try and improve performance, stack layout changed,
-          i.e., not keeping the stack looking like users expect
-          (highest numbered register at highest address).  */
-        .save   {r0, lr}
-        push    {r0, lr}
-        .save   {r4, r5}
-        strd    r4, r5, [sp, #-8]!
-        .save   {r6, r7}
-        strd    r6, r7, [sp, #-8]!
-
-       /* TODO: Add debug frame directives.
-          We don't need exception unwind directives, because the code below
-      does not throw any exceptions and does not call any other functions.
-          Generally, newlib functions like this lack debug information for
-      assembler source.  */
-
-        /* Get copying of tiny blocks out of the way first.  */
-        /* Is there at least 4 bytes to copy?  */
-        subs    r2, r2, #4
-        blt     copy_less_than_4                 /* If n < 4.  */
-
-        /* Check word alignment.  */
-        ands    ip, r0, #3                       /* ip = last 2 bits of dst.  */
-        bne     dst_not_word_aligned             /* If dst is not word-aligned.  */
-
-        /* Get here if dst is word-aligned.  */
-        ands    ip, r1, #3                      /* ip = last 2 bits of src.  */
-        bne     src_not_word_aligned            /* If src is not word-aligned.  */
-word_aligned:
-        /* Get here if source and dst both are word-aligned.
-           The number of bytes remaining to copy is r2+4.  */
-
-        /* Is there is at least 64 bytes to copy?  */
-        subs    r2, r2, #60
-        blt     copy_less_than_64                /* If r2 + 4 < 64.  */
-
-        /* First, align the destination buffer to 8-bytes,
-           to make sure double loads and stores don't cross cache line boundary,
-           as they are then more expensive even if the data is in the cache
-           (require two load/store issue cycles instead of one).
-           If only one of the buffers is not 8-bytes aligned,
-           then it's more important to align dst than src,
-           because there is more penalty for stores
-           than loads that cross cacheline boundary.
-           This check and realignment are only worth doing
-           if there is a lot to copy.  */
-
-        /* Get here if dst is word aligned,
-           i.e., the 2 least significant bits are 0.
-           If dst is not 2w aligned (i.e., the 3rd bit is not set in dst),
-           then copy 1 word (4 bytes).  */
-        ands    r3, r0, #4
-        beq     11f                  /* If dst already two-word aligned.  */
-        ldr     r3, [r1], #4
-        str     r3, [r0], #4
-        subs    r2, r2, #4
-        blt     copy_less_than_64
-
-11:
-        /* TODO: Align to cacheline (useful for PLD optimization).  */
-
-        /* Every loop iteration copies 64 bytes.  */
-1:
-        .irp    offset, #0, #8, #16, #24, #32, #40, #48, #56
-        ldrd    r4, r5, [r1, \offset]
-        strd    r4, r5, [r0, \offset]
-        .endr
-
-        add     r0, r0, #64
-        add     r1, r1, #64
-        subs    r2, r2, #64
-        bge     1b                            /* If there is more to copy.  */
-
-copy_less_than_64:
-
-        /* Get here if less than 64 bytes to copy, -64 <= r2 < 0.
-           Restore the count if there is more than 7 bytes to copy.  */
-        adds    r2, r2, #56
-        blt     copy_less_than_8
-
-        /* Copy 8 bytes at a time.  */
-2:
-        ldrd    r4, r5, [r1], #8
-        strd    r4, r5, [r0], #8
-        subs    r2, r2, #8
-        bge     2b                            /* If there is more to copy.  */
-
-copy_less_than_8:
-
-        /* Get here if less than 8 bytes to copy, -8 <= r2 < 0.
-           Check if there is more to copy.  */
-        cmn     r2, #8
-        beq     return                          /* If r2 + 8 == 0.  */
-
-        /* Restore the count if there is more than 3 bytes to copy.  */
-        adds    r2, r2, #4
-        blt     copy_less_than_4
-
-        /* Copy 4 bytes.  */
-        ldr     r3, [r1], #4
-        str     r3, [r0], #4
-
-copy_less_than_4:
-        /* Get here if less than 4 bytes to copy, -4 <= r2 < 0.  */
-
-        /* Restore the count, check if there is more to copy.  */
-        adds    r2, r2, #4
-        beq     return                          /* If r2 == 0.  */
-
-        /* Get here with r2 is in {1,2,3}={01,10,11}.  */
-        /* Logical shift left r2, insert 0s, update flags.  */
-        lsls    r2, r2, #31
-
-        /* Copy byte by byte.
-           Condition ne means the last bit of r2 is 0.
-           Condition cs means the second to last bit of r2 is set,
-           i.e., r2 is 1 or 3.  */
-        itt     ne
-        ldrbne  r3, [r1], #1
-        strbne  r3, [r0], #1
-
-        itttt   cs
-        ldrbcs  r4, [r1], #1
-        ldrbcs  r5, [r1]
-        strbcs  r4, [r0], #1
-        strbcs  r5, [r0]
-
-return:
-        /* Restore registers: optimized pop {r0, r4, r5, r6, r7, pc}   */
-        /* This is the only return point of memcpy.  */
-        ldrd r6, r7, [sp], #8
-        ldrd r4, r5, [sp], #8
-        pop {r0, pc}
-
-#ifndef __ARM_FEATURE_UNALIGNED
-
-       /* The following assembly macro implements misaligned copy in software.
-          Assumes that dst is word aligned, src is at offset "pull" bits from
-      word, push = 32 - pull, and the number of bytes that remain to copy
-      is r2 + 4, r2 >= 0.  */
-
-       /* In the code below, r2 is the number of bytes that remain to be
-      written.  The number of bytes read is always larger, because we have
-      partial words in the shift queue.  */
-
-        .macro  miscopy pull push shiftleft shiftright
-
-        /* Align src to the previous word boundary.  */
-        bic     r1, r1, #3
-
-        /* Initialize the shift queue.  */
-        ldr     r5, [r1], #4                   /* Load a word from source.  */
-
-        subs    r2, r2, #4
-        blt     6f          /* Go to misaligned copy of less than 8 bytes.  */
-
-       /* Get here if there is more than 8 bytes to copy.
-          The number of bytes to copy is r2+8, r2 >= 0.  */
-
-       subs     r2, r2, #56
-       blt      4f         /* Go to misaligned copy of less than 64 bytes.  */
-
-3:
-       /* Get here if there is more than 64 bytes to copy.
-          The number of bytes to copy is r2+64, r2 >= 0.  */
-
-       /* Copy 64 bytes in every iteration.
-          Use a partial word from the shift queue.  */
-        .irp    offset, #0, #8, #16, #24, #32, #40, #48, #56
-        mov     r6, r5, \shiftleft #\pull
-        ldrd    r4, r5, [r1, \offset]
-        orr     r6, r6, r4, \shiftright #\push
-        mov     r7, r4, \shiftleft #\pull
-        orr     r7, r7, r5, \shiftright #\push
-        strd    r6, r7, [r0, \offset]
-        .endr
-
-        add     r1, r1, #64
-        add     r0, r0, #64
-        subs    r2, r2, #64
-        bge     3b
-
-4:
-       /* Get here if there is less than 64 bytes to copy (-64 <= r2 < 0)
-      and they are misaligned.  */
-
-       /* Restore the count if there is more than 7 bytes to copy.  */
-        adds    r2, r2, #56
-
-        blt     6f          /* Go to misaligned copy of less than 8 bytes.  */
-
-5:
-        /* Copy 8 bytes at a time.
-           Use a partial word from the shift queue.  */
-        mov     r6, r5, \shiftleft #\pull
-        ldrd    r4, r5, [r1], #8
-        orr     r6, r6, r4, \shiftright #\push
-        mov     r7, r4, \shiftleft #\pull
-        orr     r7, r7, r5, \shiftright #\push
-        strd    r6, r7, [r0], #8
-
-        subs    r2, r2, #8
-        bge     5b                        /* If there is more to copy.  */
-
-6:
-        /* Get here if there less than 8 bytes to copy (-8 <= r2 < 0)
-           and they are misaligned.  */
-
-        /* Check if there is more to copy.  */
-        cmn     r2, #8
-        beq     return
-
-        /* Check if there is less than 4 bytes to copy.  */
-        cmn     r2, #4
-
-        itt     lt
-        /* Restore src offset from word-align.  */
-        sublt   r1, r1, #(\push / 8)
-        blt     copy_less_than_4
-
-        /* Use a partial word from the shift queue.  */
-        mov     r3, r5, \shiftleft #\pull
-        /* Load a word from src, but without writeback
-           (this word is not fully written to dst).  */
-        ldr     r5, [r1]
-
-        /* Restore src offset from word-align.  */
-        add     r1, r1, #(\pull / 8)
-
-        /* Shift bytes to create one dst word and store it.  */
-        orr     r3, r3, r5, \shiftright #\push
-        str     r3, [r0], #4
-
-        /* Use single byte copying of the remaining bytes.  */
-        b       copy_less_than_4
-
-        .endm
-
-#endif /* not __ARM_FEATURE_UNALIGNED  */
-
-dst_not_word_aligned:
-
-       /* Get here when dst is not aligned and ip has the last 2 bits of dst,
-          i.e., ip is the offset of dst from word.
-          The number of bytes that remains to copy is r2 + 4,
-          i.e., there are at least 4 bytes to copy.
-          Write a partial word (0 to 3 bytes), such that dst becomes
-      word-aligned.  */
-
-       /* If dst is at ip bytes offset from a word (with 0 < ip < 4),
-          then there are (4 - ip) bytes to fill up to align dst to the next
-      word.  */
-        rsb     ip, ip, #4                        /* ip = #4 - ip.  */
-        cmp     ip, #2
-
-       /* Copy byte by byte with conditionals.  */
-        itt     gt
-        ldrbgt  r3, [r1], #1
-        strbgt  r3, [r0], #1
-
-        itt     ge
-        ldrbge  r4, [r1], #1
-        strbge  r4, [r0], #1
-
-        ldrb    lr, [r1], #1
-        strb    lr, [r0], #1
-
-       /* Update the count.
-          ip holds the number of bytes we have just copied.  */
-        subs    r2, r2, ip                        /* r2 = r2 - ip.  */
-        blt     copy_less_than_4                  /* If r2 < ip.  */
-
-       /* Get here if there are more than 4 bytes to copy.
-          Check if src is aligned.  If beforehand src and dst were not word
-      aligned but congruent (same offset), then now they are both
-      word-aligned, and we can copy the rest efficiently (without
-      shifting).  */
-        ands    ip, r1, #3                    /* ip = last 2 bits of src.  */
-        beq     word_aligned                  /* If r1 is word-aligned.  */
-
-src_not_word_aligned:
-       /* Get here when src is not word-aligned, but dst is word-aligned.
-          The number of bytes that remains to copy is r2+4.  */
-
-#ifdef __ARM_FEATURE_UNALIGNED
-       /* Copy word by word using LDR when alignment can be done in hardware,
-          i.e., SCTLR.A is set, supporting unaligned access in LDR and STR.  */
-        subs    r2, r2, #60
-        blt     8f
-
-7:
-        /* Copy 64 bytes in every loop iteration.  */
-        .irp    offset, #0, #4, #8, #12, #16, #20, #24, #28, #32, #36, #40, #44, #48, #52, #56, #60
-        ldr     r3, [r1, \offset]
-        str     r3, [r0, \offset]
-        .endr
-
-        add     r0, r0, #64
-        add     r1, r1, #64
-        subs    r2, r2, #64
-        bge     7b
-
-8:
-        /* Get here if less than 64 bytes to copy, -64 <= r2 < 0.
-           Check if there is more than 3 bytes to copy.  */
-        adds    r2, r2, #60
-        blt     copy_less_than_4
-
-9:
-       /* Get here if there is less than 64 but at least 4 bytes to copy,
-          where the number of bytes to copy is r2+4.  */
-        ldr     r3, [r1], #4
-        str     r3, [r0], #4
-        subs    r2, r2, #4
-        bge     9b
-
-        b       copy_less_than_4
-
-#else /* not __ARM_FEATURE_UNALIGNED  */
-
-       /* ip has last 2 bits of src,
-          i.e., ip is the offset of src from word, and ip > 0.
-          Compute shifts needed to copy from src to dst.  */
-        cmp     ip, #2
-        beq     miscopy_16_16             /* If ip == 2.  */
-        bge     miscopy_24_8              /* If ip == 3.  */
-
-        /* Get here if ip == 1.  */
-
-        /* Endian independent macros for shifting bytes within registers.  */
-
-#ifndef __ARMEB__
-miscopy_8_24:   miscopy pull=8 push=24 shiftleft=lsr shiftright=lsl
-miscopy_16_16:  miscopy pull=16 push=16 shiftleft=lsr shiftright=lsl
-miscopy_24_8:   miscopy pull=24 push=8 shiftleft=lsr shiftright=lsl
-#else  /* not __ARMEB__ */
-miscopy_8_24:   miscopy pull=8 push=24 shiftleft=lsl shiftright=lsr
-miscopy_16_16:  miscopy pull=16 push=16 shiftleft=lsl shiftright=lsr
-miscopy_24_8:   miscopy pull=24 push=8 shiftleft=lsl shiftright=lsr
-#endif  /* not __ARMEB__ */
-
-#endif  /* not __ARM_FEATURE_UNALIGNED  */
-
-END(memcpy)

libc/arch-arm/bionic/memcpy.a9.S

@@ -1,614 +0,0 @@
-/* Copyright (c) 2013, Linaro Limited
-   All rights reserved.
-
-   Redistribution and use in source and binary forms, with or without
-   modification, are permitted provided that the following conditions
-   are met:
-
-      * Redistributions of source code must retain the above copyright
-      notice, this list of conditions and the following disclaimer.
-
-      * Redistributions in binary form must reproduce the above copyright
-      notice, this list of conditions and the following disclaimer in the
-      documentation and/or other materials provided with the distribution.
-
-      * Neither the name of Linaro Limited nor the names of its
-      contributors may be used to endorse or promote products derived
-      from this software without specific prior written permission.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-   HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
- */
-
-/*
-   This memcpy routine is optimised for Cortex-A15 cores and takes advantage
-   of VFP or NEON when built with the appropriate flags.
-
-   Assumptions:
-
-    ARMv6 (ARMv7-a if using Neon)
-    ARM state
-    Unaligned accesses
-    LDRD/STRD support unaligned word accesses
-
- */
-
-#include <machine/cpu-features.h>
-#include <machine/asm.h>
-
-	.syntax unified
-	/* This implementation requires ARM state.  */
-	.arm
-
-#ifdef __ARM_NEON__
-
-	.fpu	neon
-	.arch	armv7-a
-# define FRAME_SIZE	4
-# define USE_VFP
-# define USE_NEON
-
-#elif !defined (__SOFTFP__)
-
-	.arch	armv6
-	.fpu	vfpv2
-# define FRAME_SIZE	32
-# define USE_VFP
-
-#else
-	.arch	armv6
-# define FRAME_SIZE    32
-
-#endif
-
-/* Old versions of GAS incorrectly implement the NEON align semantics.  */
-#ifdef BROKEN_ASM_NEON_ALIGN
-#define ALIGN(addr, align) addr,:align
-#else
-#define ALIGN(addr, align) addr:align
-#endif
-
-#define PC_OFFSET	8	/* PC pipeline compensation.  */
-#define INSN_SIZE	4
-
-/* Call parameters.  */
-#define dstin	r0
-#define src	r1
-#define count	r2
-
-/* Locals.  */
-#define tmp1	r3
-#define dst	ip
-#define tmp2	r10
-
-#ifndef USE_NEON
-/* For bulk copies using GP registers.  */
-#define	A_l	r2		/* Call-clobbered.  */
-#define	A_h	r3		/* Call-clobbered.  */
-#define	B_l	r4
-#define	B_h	r5
-#define	C_l	r6
-#define	C_h	r7
-#define	D_l	r8
-#define	D_h	r9
-#endif
-
-/* Number of lines ahead to pre-fetch data.  If you change this the code
-   below will need adjustment to compensate.  */
-
-#define prefetch_lines	5
-
-#ifdef USE_VFP
-	.macro	cpy_line_vfp vreg, base
-	vstr	\vreg, [dst, #\base]
-	vldr	\vreg, [src, #\base]
-	vstr	d0, [dst, #\base + 8]
-	vldr	d0, [src, #\base + 8]
-	vstr	d1, [dst, #\base + 16]
-	vldr	d1, [src, #\base + 16]
-	vstr	d2, [dst, #\base + 24]
-	vldr	d2, [src, #\base + 24]
-	vstr	\vreg, [dst, #\base + 32]
-	vldr	\vreg, [src, #\base + prefetch_lines * 64 - 32]
-	vstr	d0, [dst, #\base + 40]
-	vldr	d0, [src, #\base + 40]
-	vstr	d1, [dst, #\base + 48]
-	vldr	d1, [src, #\base + 48]
-	vstr	d2, [dst, #\base + 56]
-	vldr	d2, [src, #\base + 56]
-	.endm
-
-	.macro	cpy_tail_vfp vreg, base
-	vstr	\vreg, [dst, #\base]
-	vldr	\vreg, [src, #\base]
-	vstr	d0, [dst, #\base + 8]
-	vldr	d0, [src, #\base + 8]
-	vstr	d1, [dst, #\base + 16]
-	vldr	d1, [src, #\base + 16]
-	vstr	d2, [dst, #\base + 24]
-	vldr	d2, [src, #\base + 24]
-	vstr	\vreg, [dst, #\base + 32]
-	vstr	d0, [dst, #\base + 40]
-	vldr	d0, [src, #\base + 40]
-	vstr	d1, [dst, #\base + 48]
-	vldr	d1, [src, #\base + 48]
-	vstr	d2, [dst, #\base + 56]
-	vldr	d2, [src, #\base + 56]
-	.endm
-#endif
-
-	.p2align 6
-ENTRY(memcpy)
-
-	mov	dst, dstin	/* Preserve dstin, we need to return it.  */
-	cmp	count, #64
-	bge	.Lcpy_not_short
-	/* Deal with small copies quickly by dropping straight into the
-	   exit block.  */
-
-.Ltail63unaligned:
-#ifdef USE_NEON
-	and	tmp1, count, #0x38
-	rsb	tmp1, tmp1, #(56 - PC_OFFSET + INSN_SIZE)
-	add	pc, pc, tmp1
-	vld1.8	{d0}, [src]!	/* 14 words to go.  */
-	vst1.8	{d0}, [dst]!
-	vld1.8	{d0}, [src]!	/* 12 words to go.  */
-	vst1.8	{d0}, [dst]!
-	vld1.8	{d0}, [src]!	/* 10 words to go.  */
-	vst1.8	{d0}, [dst]!
-	vld1.8	{d0}, [src]!	/* 8 words to go.  */
-	vst1.8	{d0}, [dst]!
-	vld1.8	{d0}, [src]!	/* 6 words to go.  */
-	vst1.8	{d0}, [dst]!
-	vld1.8	{d0}, [src]!	/* 4 words to go.  */
-	vst1.8	{d0}, [dst]!
-	vld1.8	{d0}, [src]!	/* 2 words to go.  */
-	vst1.8	{d0}, [dst]!
-
-	tst	count, #4
-	ldrne	tmp1, [src], #4
-	strne	tmp1, [dst], #4
-#else
-	/* Copy up to 15 full words of data.  May not be aligned.  */
-	/* Cannot use VFP for unaligned data.  */
-	and	tmp1, count, #0x3c
-	add	dst, dst, tmp1
-	add	src, src, tmp1
-	rsb	tmp1, tmp1, #(60 - PC_OFFSET/2 + INSN_SIZE/2)
-	/* Jump directly into the sequence below at the correct offset.  */
-	add	pc, pc, tmp1, lsl #1
-
-	ldr	tmp1, [src, #-60]	/* 15 words to go.  */
-	str	tmp1, [dst, #-60]
-
-	ldr	tmp1, [src, #-56]	/* 14 words to go.  */
-	str	tmp1, [dst, #-56]
-	ldr	tmp1, [src, #-52]
-	str	tmp1, [dst, #-52]
-
-	ldr	tmp1, [src, #-48]	/* 12 words to go.  */
-	str	tmp1, [dst, #-48]
-	ldr	tmp1, [src, #-44]
-	str	tmp1, [dst, #-44]
-
-	ldr	tmp1, [src, #-40]	/* 10 words to go.  */
-	str	tmp1, [dst, #-40]
-	ldr	tmp1, [src, #-36]
-	str	tmp1, [dst, #-36]
-
-	ldr	tmp1, [src, #-32]	/* 8 words to go.  */
-	str	tmp1, [dst, #-32]
-	ldr	tmp1, [src, #-28]
-	str	tmp1, [dst, #-28]
-
-	ldr	tmp1, [src, #-24]	/* 6 words to go.  */
-	str	tmp1, [dst, #-24]
-	ldr	tmp1, [src, #-20]
-	str	tmp1, [dst, #-20]
-
-	ldr	tmp1, [src, #-16]	/* 4 words to go.  */
-	str	tmp1, [dst, #-16]
-	ldr	tmp1, [src, #-12]
-	str	tmp1, [dst, #-12]
-
-	ldr	tmp1, [src, #-8]	/* 2 words to go.  */
-	str	tmp1, [dst, #-8]
-	ldr	tmp1, [src, #-4]
-	str	tmp1, [dst, #-4]
-#endif
-
-	lsls	count, count, #31
-	ldrhcs	tmp1, [src], #2
-	ldrbne	src, [src]		/* Src is dead, use as a scratch.  */
-	strhcs	tmp1, [dst], #2
-	strbne	src, [dst]
-	bx	lr
-
-.Lcpy_not_short:
-	/* At least 64 bytes to copy, but don't know the alignment yet.  */
-	str	tmp2, [sp, #-FRAME_SIZE]!
-	and	tmp2, src, #7
-	and	tmp1, dst, #7
-	cmp	tmp1, tmp2
-	bne	.Lcpy_notaligned
-
-#ifdef USE_VFP
-	/* Magic dust alert!  Force VFP on Cortex-A9.  Experiments show
-	   that the FP pipeline is much better at streaming loads and
-	   stores.  This is outside the critical loop.  */
-	vmov.f32	s0, s0
-#endif
-
-	/* SRC and DST have the same mutual 32-bit alignment, but we may
-	   still need to pre-copy some bytes to get to natural alignment.
-	   We bring DST into full 64-bit alignment.  */
-	lsls	tmp2, dst, #29
-	beq	1f
-	rsbs	tmp2, tmp2, #0
-	sub	count, count, tmp2, lsr #29
-	ldrmi	tmp1, [src], #4
-	strmi	tmp1, [dst], #4
-	lsls	tmp2, tmp2, #2
-	ldrhcs	tmp1, [src], #2
-	ldrbne	tmp2, [src], #1
-	strhcs	tmp1, [dst], #2
-	strbne	tmp2, [dst], #1
-
-1:
-	subs	tmp2, count, #64	/* Use tmp2 for count.  */
-	blt	.Ltail63aligned
-
-	cmp	tmp2, #512
-	bge	.Lcpy_body_long
-
-.Lcpy_body_medium:			/* Count in tmp2.  */
-#ifdef USE_VFP
-1:
-	vldr	d0, [src, #0]
-	subs	tmp2, tmp2, #64
-	vldr	d1, [src, #8]
-	vstr	d0, [dst, #0]
-	vldr	d0, [src, #16]
-	vstr	d1, [dst, #8]
-	vldr	d1, [src, #24]
-	vstr	d0, [dst, #16]
-	vldr	d0, [src, #32]
-	vstr	d1, [dst, #24]
-	vldr	d1, [src, #40]
-	vstr	d0, [dst, #32]
-	vldr	d0, [src, #48]
-	vstr	d1, [dst, #40]
-	vldr	d1, [src, #56]
-	vstr	d0, [dst, #48]
-	add	src, src, #64
-	vstr	d1, [dst, #56]
-	add	dst, dst, #64
-	bge	1b
-	tst	tmp2, #0x3f
-	beq	.Ldone
-
-.Ltail63aligned:			/* Count in tmp2.  */
-	and	tmp1, tmp2, #0x38
-	add	dst, dst, tmp1
-	add	src, src, tmp1
-	rsb	tmp1, tmp1, #(56 - PC_OFFSET + INSN_SIZE)
-	add	pc, pc, tmp1
-
-	vldr	d0, [src, #-56]	/* 14 words to go.  */
-	vstr	d0, [dst, #-56]
-	vldr	d0, [src, #-48]	/* 12 words to go.  */
-	vstr	d0, [dst, #-48]
-	vldr	d0, [src, #-40]	/* 10 words to go.  */
-	vstr	d0, [dst, #-40]
-	vldr	d0, [src, #-32]	/* 8 words to go.  */
-	vstr	d0, [dst, #-32]
-	vldr	d0, [src, #-24]	/* 6 words to go.  */
-	vstr	d0, [dst, #-24]
-	vldr	d0, [src, #-16]	/* 4 words to go.  */
-	vstr	d0, [dst, #-16]
-	vldr	d0, [src, #-8]	/* 2 words to go.  */
-	vstr	d0, [dst, #-8]
-#else
-	sub	src, src, #8
-	sub	dst, dst, #8
-1:
-	ldrd	A_l, A_h, [src, #8]
-	strd	A_l, A_h, [dst, #8]
-	ldrd	A_l, A_h, [src, #16]
-	strd	A_l, A_h, [dst, #16]
-	ldrd	A_l, A_h, [src, #24]
-	strd	A_l, A_h, [dst, #24]
-	ldrd	A_l, A_h, [src, #32]
-	strd	A_l, A_h, [dst, #32]
-	ldrd	A_l, A_h, [src, #40]
-	strd	A_l, A_h, [dst, #40]
-	ldrd	A_l, A_h, [src, #48]
-	strd	A_l, A_h, [dst, #48]
-	ldrd	A_l, A_h, [src, #56]
-	strd	A_l, A_h, [dst, #56]
-	ldrd	A_l, A_h, [src, #64]!
-	strd	A_l, A_h, [dst, #64]!
-	subs	tmp2, tmp2, #64
-	bge	1b
-	tst	tmp2, #0x3f
-	bne	1f
-	ldr	tmp2,[sp], #FRAME_SIZE
-	bx	lr
-1:
-	add	src, src, #8
-	add	dst, dst, #8
-
-.Ltail63aligned:			/* Count in tmp2.  */
-	/* Copy up to 7 d-words of data.  Similar to Ltail63unaligned, but
-	   we know that the src and dest are 32-bit aligned so we can use
-	   LDRD/STRD to improve efficiency.  */
-	/* TMP2 is now negative, but we don't care about that.  The bottom
-	   six bits still tell us how many bytes are left to copy.  */
-
-	and	tmp1, tmp2, #0x38
-	add	dst, dst, tmp1
-	add	src, src, tmp1
-	rsb	tmp1, tmp1, #(56 - PC_OFFSET + INSN_SIZE)
-	add	pc, pc, tmp1
-	ldrd	A_l, A_h, [src, #-56]	/* 14 words to go.  */
-	strd	A_l, A_h, [dst, #-56]
-	ldrd	A_l, A_h, [src, #-48]	/* 12 words to go.  */
-	strd	A_l, A_h, [dst, #-48]
-	ldrd	A_l, A_h, [src, #-40]	/* 10 words to go.  */
-	strd	A_l, A_h, [dst, #-40]
-	ldrd	A_l, A_h, [src, #-32]	/* 8 words to go.  */
-	strd	A_l, A_h, [dst, #-32]
-	ldrd	A_l, A_h, [src, #-24]	/* 6 words to go.  */
-	strd	A_l, A_h, [dst, #-24]
-	ldrd	A_l, A_h, [src, #-16]	/* 4 words to go.  */
-	strd	A_l, A_h, [dst, #-16]
-	ldrd	A_l, A_h, [src, #-8]	/* 2 words to go.  */
-	strd	A_l, A_h, [dst, #-8]
-
-#endif
-	tst	tmp2, #4
-	ldrne	tmp1, [src], #4
-	strne	tmp1, [dst], #4
-	lsls	tmp2, tmp2, #31		/* Count (tmp2) now dead. */
-	ldrhcs	tmp1, [src], #2
-	ldrbne	tmp2, [src]
-	strhcs	tmp1, [dst], #2
-	strbne	tmp2, [dst]
-
-.Ldone:
-	ldr	tmp2, [sp], #FRAME_SIZE
-	bx	lr
-
-.Lcpy_body_long:			/* Count in tmp2.  */
-
-	/* Long copy.  We know that there's at least (prefetch_lines * 64)
-	   bytes to go.  */
-#ifdef USE_VFP
-	/* Don't use PLD.  Instead, read some data in advance of the current
-	   copy position into a register.  This should act like a PLD
-	   operation but we won't have to repeat the transfer.  */
-
-	vldr	d3, [src, #0]
-	vldr	d4, [src, #64]
-	vldr	d5, [src, #128]
-	vldr	d6, [src, #192]
-	vldr	d7, [src, #256]
-
-	vldr	d0, [src, #8]
-	vldr	d1, [src, #16]
-	vldr	d2, [src, #24]
-	add	src, src, #32
-
-	subs	tmp2, tmp2, #prefetch_lines * 64 * 2
-	blt	2f
-1:
-	cpy_line_vfp	d3, 0
-	cpy_line_vfp	d4, 64
-	cpy_line_vfp	d5, 128
-	add	dst, dst, #3 * 64
-	add	src, src, #3 * 64
-	cpy_line_vfp	d6, 0
-	cpy_line_vfp	d7, 64
-	add	dst, dst, #2 * 64
-	add	src, src, #2 * 64
-	subs	tmp2, tmp2, #prefetch_lines * 64
-	bge	1b
-
-2:
-	cpy_tail_vfp	d3, 0
-	cpy_tail_vfp	d4, 64
-	cpy_tail_vfp	d5, 128
-	add	src, src, #3 * 64
-	add	dst, dst, #3 * 64
-	cpy_tail_vfp	d6, 0
-	vstr	d7, [dst, #64]
-	vldr	d7, [src, #64]
-	vstr	d0, [dst, #64 + 8]
-	vldr	d0, [src, #64 + 8]
-	vstr	d1, [dst, #64 + 16]
-	vldr	d1, [src, #64 + 16]
-	vstr	d2, [dst, #64 + 24]
-	vldr	d2, [src, #64 + 24]
-	vstr	d7, [dst, #64 + 32]
-	add	src, src, #96
-	vstr	d0, [dst, #64 + 40]
-	vstr	d1, [dst, #64 + 48]
-	vstr	d2, [dst, #64 + 56]
-	add	dst, dst, #128
-	add	tmp2, tmp2, #prefetch_lines * 64
-	b	.Lcpy_body_medium
-#else
-	/* Long copy.  Use an SMS style loop to maximize the I/O
-	   bandwidth of the core.  We don't have enough spare registers
-	   to synthesise prefetching, so use PLD operations.  */
-	/* Pre-bias src and dst.  */
-	sub	src, src, #8
-	sub	dst, dst, #8
-	pld	[src, #8]
-	pld	[src, #72]
-	subs	tmp2, tmp2, #64
-	pld	[src, #136]
-	ldrd	A_l, A_h, [src, #8]
-	strd	B_l, B_h, [sp, #8]
-	ldrd	B_l, B_h, [src, #16]
-	strd	C_l, C_h, [sp, #16]
-	ldrd	C_l, C_h, [src, #24]
-	strd	D_l, D_h, [sp, #24]
-	pld	[src, #200]
-	ldrd	D_l, D_h, [src, #32]!
-	b	1f
-	.p2align	6
-2:
-	pld	[src, #232]
-	strd	A_l, A_h, [dst, #40]
-	ldrd	A_l, A_h, [src, #40]
-	strd	B_l, B_h, [dst, #48]
-	ldrd	B_l, B_h, [src, #48]
-	strd	C_l, C_h, [dst, #56]
-	ldrd	C_l, C_h, [src, #56]
-	strd	D_l, D_h, [dst, #64]!
-	ldrd	D_l, D_h, [src, #64]!
-	subs	tmp2, tmp2, #64
-1:
-	strd	A_l, A_h, [dst, #8]
-	ldrd	A_l, A_h, [src, #8]
-	strd	B_l, B_h, [dst, #16]
-	ldrd	B_l, B_h, [src, #16]
-	strd	C_l, C_h, [dst, #24]
-	ldrd	C_l, C_h, [src, #24]
-	strd	D_l, D_h, [dst, #32]
-	ldrd	D_l, D_h, [src, #32]
-	bcs	2b
-	/* Save the remaining bytes and restore the callee-saved regs.  */
-	strd	A_l, A_h, [dst, #40]
-	add	src, src, #40
-	strd	B_l, B_h, [dst, #48]
-	ldrd	B_l, B_h, [sp, #8]
-	strd	C_l, C_h, [dst, #56]
-	ldrd	C_l, C_h, [sp, #16]
-	strd	D_l, D_h, [dst, #64]
-	ldrd	D_l, D_h, [sp, #24]
-	add	dst, dst, #72
-	tst	tmp2, #0x3f
-	bne	.Ltail63aligned
-	ldr	tmp2, [sp], #FRAME_SIZE
-	bx	lr
-#endif
-
-.Lcpy_notaligned:
-	pld	[src]
-	pld	[src, #64]
-	/* There's at least 64 bytes to copy, but there is no mutual
-	   alignment.  */
-	/* Bring DST to 64-bit alignment.  */
-	lsls	tmp2, dst, #29
-	pld	[src, #(2 * 64)]
-	beq	1f
-	rsbs	tmp2, tmp2, #0
-	sub	count, count, tmp2, lsr #29
-	ldrmi	tmp1, [src], #4
-	strmi	tmp1, [dst], #4
-	lsls	tmp2, tmp2, #2
-	ldrbne	tmp1, [src], #1
-	ldrhcs	tmp2, [src], #2
-	strbne	tmp1, [dst], #1
-	strhcs	tmp2, [dst], #2
-1:
-	pld	[src, #(3 * 64)]
-	subs	count, count, #64
-	ldrmi	tmp2, [sp], #FRAME_SIZE
-	bmi	.Ltail63unaligned
-	pld	[src, #(4 * 64)]
-
-#ifdef USE_NEON
-	vld1.8	{d0-d3}, [src]!
-	vld1.8	{d4-d7}, [src]!
-	subs	count, count, #64
-	bmi	2f
-1:
-	pld	[src, #(4 * 64)]
-	vst1.8	{d0-d3}, [ALIGN (dst, 64)]!
-	vld1.8	{d0-d3}, [src]!
-	vst1.8	{d4-d7}, [ALIGN (dst, 64)]!
-	vld1.8	{d4-d7}, [src]!
-	subs	count, count, #64
-	bpl	1b
-2:
-	vst1.8	{d0-d3}, [ALIGN (dst, 64)]!
-	vst1.8	{d4-d7}, [ALIGN (dst, 64)]!
-	ands	count, count, #0x3f
-#else
-	/* Use an SMS style loop to maximize the I/O bandwidth.  */
-	sub	src, src, #4
-	sub	dst, dst, #8
-	subs	tmp2, count, #64	/* Use tmp2 for count.  */
-	ldr	A_l, [src, #4]
-	ldr	A_h, [src, #8]
-	strd	B_l, B_h, [sp, #8]
-	ldr	B_l, [src, #12]
-	ldr	B_h, [src, #16]
-	strd	C_l, C_h, [sp, #16]
-	ldr	C_l, [src, #20]
-	ldr	C_h, [src, #24]
-	strd	D_l, D_h, [sp, #24]
-	ldr	D_l, [src, #28]
-	ldr	D_h, [src, #32]!
-	b	1f
-	.p2align	6
-2:
-	pld	[src, #(5 * 64) - (32 - 4)]
-	strd	A_l, A_h, [dst, #40]
-	ldr	A_l, [src, #36]
-	ldr	A_h, [src, #40]
-	strd	B_l, B_h, [dst, #48]
-	ldr	B_l, [src, #44]
-	ldr	B_h, [src, #48]
-	strd	C_l, C_h, [dst, #56]
-	ldr	C_l, [src, #52]
-	ldr	C_h, [src, #56]
-	strd	D_l, D_h, [dst, #64]!
-	ldr	D_l, [src, #60]
-	ldr	D_h, [src, #64]!
-	subs	tmp2, tmp2, #64
-1:
-	strd	A_l, A_h, [dst, #8]
-	ldr	A_l, [src, #4]
-	ldr	A_h, [src, #8]
-	strd	B_l, B_h, [dst, #16]
-	ldr	B_l, [src, #12]
-	ldr	B_h, [src, #16]
-	strd	C_l, C_h, [dst, #24]
-	ldr	C_l, [src, #20]
-	ldr	C_h, [src, #24]
-	strd	D_l, D_h, [dst, #32]
-	ldr	D_l, [src, #28]
-	ldr	D_h, [src, #32]
-	bcs	2b
-
-	/* Save the remaining bytes and restore the callee-saved regs.  */
-	strd	A_l, A_h, [dst, #40]
-	add	src, src, #36
-	strd	B_l, B_h, [dst, #48]
-	ldrd	B_l, B_h, [sp, #8]
-	strd	C_l, C_h, [dst, #56]
-	ldrd	C_l, C_h, [sp, #16]
-	strd	D_l, D_h, [dst, #64]
-	ldrd	D_l, D_h, [sp, #24]
-	add	dst, dst, #72
-	ands	count, tmp2, #0x3f
-#endif
-	ldr	tmp2, [sp], #FRAME_SIZE
-	bne	.Ltail63unaligned
-	bx	lr
-END(memcpy)

libc/arch-arm/bionic/memset.S

@@ -1,200 +0,0 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include <machine/cpu-features.h>
-#include <machine/asm.h>
-
-		/*
-		 * Optimized memset() for ARM.
-         *
-         * memset() returns its first argument.
-		 */
-
-#if defined(__ARM_NEON__)
-    .fpu    neon
-#endif
-
-ENTRY(bzero)
-        mov     r2, r1
-        mov     r1, #0
-        // Fall through to memset...
-END(bzero)
-
-ENTRY(memset)
-#if defined(__ARM_NEON__)
-
-#ifdef  NEON_MEMSET_DIVIDER
-        cmp         r2, #NEON_MEMSET_DIVIDER
-        bhi         11f
-#endif
-        .save       {r0}
-        stmfd       sp!, {r0}
-
-        vdup.8      q0, r1
-
-#ifndef NEON_UNALIGNED_ACCESS
-        /* do we have at least 16-bytes to write (needed for alignment below) */
-        cmp         r2, #16
-        blo         3f
-
-        /* align destination to 16 bytes for the write-buffer */
-        rsb         r3, r0, #0
-        ands        r3, r3, #0xF
-        beq         2f
-
-        /* write up to 15-bytes (count in r3) */
-        sub         r2, r2, r3
-        movs        ip, r3, lsl #31
-        strmib      r1, [r0], #1
-        strcsb      r1, [r0], #1
-        strcsb      r1, [r0], #1
-        movs        ip, r3, lsl #29
-        bge         1f
-
-        // writes 4 bytes, 32-bits aligned
-        vst1.32     {d0[0]}, [r0, :32]!
-1:      bcc         2f
-
-        // writes 8 bytes, 64-bits aligned
-        vst1.8      {d0}, [r0, :64]!
-2:
-#endif
-        /* make sure we have at least 32 bytes to write */
-        subs        r2, r2, #32
-        blo         2f
-        vmov        q1, q0
-
-1:      /* The main loop writes 32 bytes at a time */
-        subs        r2, r2, #32
-#ifndef NEON_UNALIGNED_ACCESS
-        vst1.8      {d0 - d3}, [r0, :128]!
-#else
-        vst1.8      {d0 - d3}, [r0]!
-#endif
-        bhs         1b
-
-2:      /* less than 32 left */
-        add         r2, r2, #32
-        tst         r2, #0x10
-        beq         3f
-
-        // writes 16 bytes, 128-bits aligned
-#ifndef NEON_UNALIGNED_ACCESS
-        vst1.8      {d0, d1}, [r0, :128]!
-#else
-        vst1.8      {d0, d1}, [r0]!
-#endif
-3:      /* write up to 15-bytes (count in r2) */
-        movs        ip, r2, lsl #29
-        bcc         1f
-        vst1.8      {d0}, [r0]!
-1:      bge         2f
-        vst1.32     {d0[0]}, [r0]!
-2:      movs        ip, r2, lsl #31
-        strmib      r1, [r0], #1
-        strcsb      r1, [r0], #1
-        strcsb      r1, [r0], #1
-        ldmfd       sp!, {r0}
-        bx          lr
-11:
-#endif
-
-        /*
-         * Optimized memset() for ARM.
-         *
-         * memset() returns its first argument.
-         */
-
-		/* compute the offset to align the destination
-		 * offset = (4-(src&3))&3 = -src & 3
-		 */
-
-        .save       {r0, r4-r7, lr}
-		stmfd		sp!, {r0, r4-r7, lr}
-		rsb			r3, r0, #0
-		ands		r3, r3, #3
-        cmp         r3, r2
-        movhi       r3, r2
-
-        /* splat r1 */
-        mov         r1, r1, lsl #24
-        orr         r1, r1, r1, lsr #8
-        orr         r1, r1, r1, lsr #16
-
-		movs		r12, r3, lsl #31
-		strcsb		r1, [r0], #1    /* can't use strh (alignment unknown) */
-		strcsb		r1, [r0], #1
-		strmib		r1, [r0], #1
-		subs		r2, r2, r3
-        ldmlsfd     sp!, {r0, r4-r7, lr}   /* return */
-        bxls        lr
-
-		/* align the destination to a cache-line */
-        mov         r12, r1
-        mov         lr, r1
-        mov         r4, r1
-        mov         r5, r1
-        mov         r6, r1
-        mov         r7, r1
-
-		rsb         r3, r0, #0
-		ands		r3, r3, #0x1C
-		beq         3f
-		cmp         r3, r2
-		andhi		r3, r2, #0x1C
-		sub         r2, r2, r3
-
-		/* conditionally writes 0 to 7 words (length in r3) */
-		movs		r3, r3, lsl #28
-		stmcsia		r0!, {r1, lr}
-		stmcsia		r0!, {r1, lr}
-		stmmiia		r0!, {r1, lr}
-		movs		r3, r3, lsl #2
-        strcs       r1, [r0], #4
-
-3:
-        subs        r2, r2, #32
-        mov         r3, r1
-        bmi         2f
-1:      subs        r2, r2, #32
-        stmia		r0!, {r1,r3,r4,r5,r6,r7,r12,lr}
-        bhs         1b
-2:      add         r2, r2, #32
-
-		/* conditionally stores 0 to 31 bytes */
-		movs		r2, r2, lsl #28
-		stmcsia		r0!, {r1,r3,r12,lr}
-		stmmiia		r0!, {r1, lr}
-		movs		r2, r2, lsl #2
-        strcs       r1, [r0], #4
-		strmih		r1, [r0], #2
-		movs		r2, r2, lsl #2
-		strcsb		r1, [r0]
-        ldmfd		sp!, {r0, r4-r7, lr}
-        bx          lr
-END(memset)

libc/arch-arm/bionic/popcount_tab.c

@@ -0,0 +1,42 @@
+/*
+ * Copyright (C) 2015 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+// Export this to maintain ABI compatibilty with libgcc, since compiler-rt
+// doesn't use a table-driven implementation of __popcount.
+const unsigned char __popcount_tab[256] = {
+  0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4, 1, 2, 2, 3, 2, 3, 3, 4, 2, 3,
+  3, 4, 3, 4, 4, 5, 1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5, 2, 3, 3, 4,
+  3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, 1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4,
+  4, 5, 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, 2, 3, 3, 4, 3, 4, 4, 5,
+  3, 4, 4, 5, 4, 5, 5, 6, 3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7, 1, 2,
+  2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5, 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5,
+  4, 5, 5, 6, 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, 3, 4, 4, 5, 4, 5,
+  5, 6, 4, 5, 5, 6, 5, 6, 6, 7, 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
+  3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7, 3, 4, 4, 5, 4, 5, 5, 6, 4, 5,
+  5, 6, 5, 6, 6, 7, 4, 5, 5, 6, 5, 6, 6, 7, 5, 6, 6, 7, 6, 7, 7, 8
+};

libc/arch-arm/bionic/setjmp.S

@@ -1,6 +1,3 @@
-/*	$OpenBSD: setjmp.S,v 1.2 2004/02/01 05:40:52 drahn Exp $	*/
-/*	$NetBSD: setjmp.S,v 1.5 2003/04/05 23:08:51 bjh21 Exp $	*/
-
 /*
  * Copyright (c) 1997 Mark Brinicombe
  * Copyright (c) 2010 Android Open Source Project.
@@ -34,109 +31,204 @@
  * SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
-#include <machine/setjmp.h>
-#include <machine/cpu-features.h>
+#include <private/bionic_asm.h>
 
-/*
- * C library -- setjmp, longjmp
- *
- *	longjmp(a,v)
- * will generate a "return(v)" from the last call to
- *	setjmp(a)
- * by restoring registers from the stack.
- * The previous signal state is restored.
- */
+// According to the ARM AAPCS document, we only need to save
+// the following registers:
+//
+//  Core   r4-r14
+//
+//  VFP    d8-d15  (see section 5.1.2.1)
+//
+//      Registers s16-s31 (d8-d15, q4-q7) must be preserved across subroutine
+//      calls; registers s0-s15 (d0-d7, q0-q3) do not need to be preserved
+//      (and can be used for passing arguments or returning results in standard
+//      procedure-call variants). Registers d16-d31 (q8-q15), if present, do
+//      not need to be preserved.
+//
+//  FPSCR  saved because glibc does.
+
+// The internal structure of a jmp_buf is totally private.
+// Current layout (may change in the future):
+//
+// word   name            description
+// 0      sigflag/cookie  setjmp cookie in top 31 bits, signal mask flag in low bit
+// 1      sigmask         signal mask (not used with _setjmp / _longjmp)
+// 2      float_base      base of float registers (d8 to d15)
+// 18     float_state     floating-point status and control register
+// 19     core_base       base of core registers (r4 to r14)
+// 30     reserved        reserved entries (room to grow)
+// 64
+//
+// NOTE: float_base must be at an even word index, since the
+//       FP registers will be loaded/stored with instructions
+//       that expect 8-byte alignment.
+
+#define _JB_SIGFLAG     0
+#define _JB_SIGMASK     (_JB_SIGFLAG+1)
+#define _JB_FLOAT_BASE  (_JB_SIGMASK+1)
+#define _JB_FLOAT_STATE (_JB_FLOAT_BASE + (15-8+1)*2)
+#define _JB_CORE_BASE   (_JB_FLOAT_STATE+1)
 
 ENTRY(setjmp)
-	/* Block all signals and retrieve the old signal mask */
-	stmfd	sp!, {r0, r14}
-	mov	r0, #0x00000000
-
-	bl	PIC_SYM(_C_LABEL(sigblock), PLT)
-	mov	r1, r0
-
-	ldmfd	sp!, {r0, r14}
-
-	/* Store signal mask */
-	str	r1, [r0, #(_JB_SIGMASK * 4)]
-
-	ldr	r1, .Lsetjmp_magic
-	str	r1, [r0, #(_JB_MAGIC * 4)]
-
-	/* Store core registers */
-	add     r1, r0, #(_JB_CORE_BASE * 4)
-	stmia   r1, {r4-r14}
-
-#ifdef __ARM_HAVE_VFP
-	/* Store floating-point registers */
-	add     r1, r0, #(_JB_FLOAT_BASE * 4)
-	vstmia  r1, {d8-d15}
-	/* Store floating-point state */
-	fmrx    r1, fpscr
-	str     r1, [r0, #(_JB_FLOAT_STATE * 4)]
-#endif  /* __ARM_HAVE_VFP */
-
-	mov	r0, #0x00000000
-	bx      lr
+  mov r1, #1
+  b sigsetjmp
 END(setjmp)
 
-.Lsetjmp_magic:
-	.word	_JB_MAGIC_SETJMP
+ENTRY(_setjmp)
+  mov r1, #0
+  b sigsetjmp
+END(_setjmp)
 
-
-ENTRY(longjmp)
-	ldr	r2, .Lsetjmp_magic
-	ldr	r3, [r0, #(_JB_MAGIC * 4)]
-	teq	r2, r3
-	bne	botch
-
-	/* Fetch signal mask */
-	ldr	r2, [r0, #(_JB_SIGMASK * 4)]
-
-	/* Set signal mask */
-	stmfd	sp!, {r0, r1, r14}
-	sub	sp, sp, #4	/* align the stack */
-
-	mov	r0, r2
-	bl	PIC_SYM(_C_LABEL(sigsetmask), PLT)
-
-	add	sp, sp, #4	/* unalign the stack */
-	ldmfd	sp!, {r0, r1, r14} 
-
-#ifdef __ARM_HAVE_VFP
-	/* Restore floating-point registers */
-	add     r2, r0, #(_JB_FLOAT_BASE * 4)
-	vldmia  r2, {d8-d15}
-	/* Restore floating-point state */
-	ldr     r2, [r0, #(_JB_FLOAT_STATE * 4)]
-	fmxr    fpscr, r2
-#endif /* __ARM_HAVE_VFP */
-
-	/* Restore core registers */
-	add     r2, r0, #(_JB_CORE_BASE * 4)
-	ldmia   r2, {r4-r14}
-
-	/* Validate sp and r14 */
-	teq	sp, #0
-	teqne	r14, #0
-	beq	botch
-
-	/* Set return value */
-
-	mov	r0, r1
-	teq	r0, #0x00000000
-	moveq	r0, #0x00000001
-        bx      lr
-#ifdef __ARM_26__
-	mov	r15, r14
-#else
-	mov	r15, r14
+#define MANGLE_REGISTERS 1
+.macro m_mangle_registers reg
+#if MANGLE_REGISTERS
+  eor r4, r4, \reg
+  eor r5, r5, \reg
+  eor r6, r6, \reg
+  eor r7, r7, \reg
+  eor r8, r8, \reg
+  eor r9, r9, \reg
+  eor r10, r10, \reg
+  eor r11, r11, \reg
+  eor r12, r12, \reg
+  eor r13, r13, \reg
+  eor r14, r14, \reg
 #endif
+.endm
 
-	/* validation failed, die die die. */
-botch:
-	bl	PIC_SYM(_C_LABEL(longjmperror), PLT)
-	bl	PIC_SYM(_C_LABEL(abort), PLT)
-	b	. - 8		/* Cannot get here */
-END(longjmp)
+.macro m_unmangle_registers reg
+  m_mangle_registers \reg
+.endm
+
+// int sigsetjmp(sigjmp_buf env, int save_signal_mask);
+ENTRY(sigsetjmp)
+  stmfd sp!, {r0, lr}
+  .cfi_def_cfa_offset 8
+  .cfi_rel_offset r0, 0
+  .cfi_rel_offset lr, 4
+
+  mov r0, r1
+  bl __bionic_setjmp_cookie_get
+  mov r1, r0
+
+  ldmfd sp, {r0}
+
+  // Save the setjmp cookie for later.
+  bic r2, r1, #1
+  stmfd sp!, {r2}
+  .cfi_adjust_cfa_offset 4
+
+  // Record the setjmp cookie and whether or not we're saving the signal mask.
+  str r1, [r0, #(_JB_SIGFLAG * 4)]
+
+  // Do we need to save the signal mask?
+  tst r1, #1
+  beq 1f
+
+  // Align the stack.
+  sub sp, #4
+  .cfi_adjust_cfa_offset 4
+
+  // Save the current signal mask.
+  add r2, r0, #(_JB_SIGMASK * 4)
+  mov r0, #2 // SIG_SETMASK
+  mov r1, #0
+  bl sigprocmask
+
+  // Unalign the stack.
+  add sp, #4
+  .cfi_adjust_cfa_offset -4
+
+1:
+  ldmfd sp!, {r2}
+  .cfi_adjust_cfa_offset -4
+  ldmfd sp!, {r0, lr}
+  .cfi_adjust_cfa_offset -8
+  .cfi_restore r0
+  .cfi_restore lr
+
+  // Save core registers.
+  add r1, r0, #(_JB_CORE_BASE * 4)
+  m_mangle_registers r2
+
+  // ARM deprecates using sp in the register list for stmia.
+  stmia r1, {r4-r12, lr}
+  str sp, [r1, #(10 * 4)]
+  m_unmangle_registers r2
+
+  // Save floating-point registers.
+  add r1, r0, #(_JB_FLOAT_BASE * 4)
+  vstmia  r1, {d8-d15}
+
+  // Save floating-point state.
+  fmrx r1, fpscr
+  str r1, [r0, #(_JB_FLOAT_STATE * 4)]
+
+  mov r0, #0
+  bx lr
+END(sigsetjmp)
+
+// void siglongjmp(sigjmp_buf env, int value);
+ENTRY(siglongjmp)
+  stmfd sp!, {r0, r1, lr}
+  .cfi_def_cfa_offset 12
+  .cfi_rel_offset r0, 0
+  .cfi_rel_offset r1, 4
+  .cfi_rel_offset lr, 8
+
+  // Fetch the signal flag.
+  ldr r1, [r0, #(_JB_SIGFLAG * 4)]
+
+  // Do we need to restore the signal mask?
+  ands r1, r1, #1
+  beq 1f
+
+  // Restore the signal mask.
+  ldr r0, [r0, #(_JB_SIGMASK * 4)]
+  bl sigsetmask
+
+1:
+  ldmfd sp!, {r0, r1, lr}
+  .cfi_adjust_cfa_offset -12
+  .cfi_restore r0
+  .cfi_restore r1
+  .cfi_restore lr
+
+  // Restore floating-point registers.
+  add r2, r0, #(_JB_FLOAT_BASE * 4)
+  vldmia r2, {d8-d15}
+
+  // Restore floating-point state.
+  ldr r2, [r0, #(_JB_FLOAT_STATE * 4)]
+  fmxr fpscr, r2
+
+  // Restore core registers.
+  ldr r3, [r0, #(_JB_SIGFLAG * 4)]
+  bic r3, r3, #1
+  add r2, r0, #(_JB_CORE_BASE * 4)
+
+  // ARM deprecates using sp in the register list for ldmia.
+  ldmia r2, {r4-r12, lr}
+  ldr sp, [r2, #(10 * 4)]
+  m_unmangle_registers r3
+
+  // Save the return value/address and check the setjmp cookie.
+  stmfd sp!, {r1, lr}
+  .cfi_adjust_cfa_offset 8
+  .cfi_rel_offset lr, 4
+  mov r0, r3
+  bl __bionic_setjmp_cookie_check
+
+  // Restore return value/address.
+  ldmfd sp!, {r0, lr}
+  .cfi_adjust_cfa_offset -8
+  .cfi_restore lr
+
+  teq r0, #0
+  moveq r0, #1
+  bx lr
+END(siglongjmp)
+
+ALIAS_SYMBOL(longjmp, siglongjmp)
+ALIAS_SYMBOL(_longjmp, siglongjmp)

libc/arch-arm/bionic/sigsetjmp.S

@@ -1,66 +0,0 @@
-/*	$OpenBSD: sigsetjmp.S,v 1.2 2004/02/01 05:40:52 drahn Exp $	*/
-/*	$NetBSD: sigsetjmp.S,v 1.3 2002/08/17 19:54:30 thorpej Exp $	*/
-
-/*
- * Copyright (c) 1997 Mark Brinicombe
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- * 3. All advertising materials mentioning features or use of this software
- *    must display the following acknowledgement:
- *	This product includes software developed by Mark Brinicombe
- * 4. Neither the name of the University nor the names of its contributors
- *    may be used to endorse or promote products derived from this software
- *    without specific prior written permission.
- *
- * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
- * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
- * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
- * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
- * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
- * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
- * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
- * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
- * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
- * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#define _ALIGN_TEXT .align 0
-
-#include <machine/asm.h>
-#include <machine/setjmp.h>
-
-/*
- * C library -- sigsetjmp, siglongjmp
- *
- *	longjmp(a,v)
- * will generate a "return(v)" from the last call to
- *	setjmp(a, m)
- * by restoring registers from the stack.
- * The previous signal state is restored.
- */
-
-ENTRY(sigsetjmp)
-	teq	r1, #0
-	beq	PIC_SYM(_C_LABEL(_setjmp), PLT)
-	b	PIC_SYM(_C_LABEL(setjmp), PLT)
-END(sigsetjmp)
-
-.L_setjmp_magic:
-	.word	_JB_MAGIC__SETJMP
-
-ENTRY(siglongjmp)
-	ldr	r2, .L_setjmp_magic
-	ldr	r3, [r0]
-	teq	r2, r3
-	beq	PIC_SYM(_C_LABEL(_longjmp), PLT)
-	b	PIC_SYM(_C_LABEL(longjmp), PLT)
-END(siglongjmp)

libc/arch-arm/bionic/strcmp.S

@@ -1,317 +0,0 @@
-/*
- * Copyright (c) 2011 The Android Open Source Project
- * Copyright (c) 2008 ARM Ltd
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- * 3. The name of the company may not be used to endorse or promote
- *    products derived from this software without specific prior written
- *    permission.
- *
- * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
- * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
- * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
- * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
- * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
- * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
- * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
- * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
- * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- */
-
-#include <machine/cpu-features.h>
-#include <machine/asm.h>
-
-	.text
-
-#ifdef __ARMEB__
-#define SHFT2LSB lsl
-#define SHFT2LSBEQ lsleq
-#define SHFT2MSB lsr
-#define SHFT2MSBEQ lsreq
-#define MSB 0x000000ff
-#define LSB 0xff000000
-#else
-#define SHFT2LSB lsr
-#define SHFT2LSBEQ lsreq
-#define SHFT2MSB lsl
-#define SHFT2MSBEQ lsleq
-#define MSB 0xff000000
-#define LSB 0x000000ff
-#endif
-
-#define magic1(REG) REG
-#define magic2(REG) REG, lsl #7
-
-ENTRY(strcmp)
-	PLD(r0, #0)
-	PLD(r1, #0)
-	eor	r2, r0, r1
-	tst	r2, #3
-
-	/* Strings not at same byte offset from a word boundary.  */
-	bne	.Lstrcmp_unaligned
-	ands	r2, r0, #3
-	bic	r0, r0, #3
-	bic	r1, r1, #3
-	ldr	ip, [r0], #4
-	it	eq
-	ldreq	r3, [r1], #4
-	beq	1f
-
-	/* Although s1 and s2 have identical initial alignment, they are
-	 * not currently word aligned.  Rather than comparing bytes,
-	 * make sure that any bytes fetched from before the addressed
-	 * bytes are forced to 0xff.  Then they will always compare
-	 * equal.
-	 */
-	eor	r2, r2, #3
-	lsl	r2, r2, #3
-	mvn	r3, #MSB
-	SHFT2LSB	r2, r3, r2
-	ldr	r3, [r1], #4
-	orr	ip, ip, r2
-	orr	r3, r3, r2
-1:
-	/* Load the 'magic' constant 0x01010101. */
-	str	r4, [sp, #-4]!
-	mov	r4, #1
-	orr	r4, r4, r4, lsl #8
-	orr	r4, r4, r4, lsl #16
-	.p2align	2
-4:
-	PLD(r0, #8)
-	PLD(r1, #8)
-	sub	r2, ip, magic1(r4)
-	cmp	ip, r3
-	itttt	eq
-
-	/* check for any zero bytes in first word */
-	biceq	r2, r2, ip
-	tsteq	r2, magic2(r4)
-	ldreq	ip, [r0], #4
-	ldreq	r3, [r1], #4
-	beq	4b
-2:
-	/* There's a zero or a different byte in the word */
-	SHFT2MSB	r0, ip, #24
-	SHFT2LSB	ip, ip, #8
-	cmp	r0, #1
-	it	cs
-	cmpcs	r0, r3, SHFT2MSB #24
-	it	eq
-	SHFT2LSBEQ r3, r3, #8
-	beq	2b
-	/* On a big-endian machine, r0 contains the desired byte in bits
-	 * 0-7; on a little-endian machine they are in bits 24-31.  In
-	 * both cases the other bits in r0 are all zero.  For r3 the
-	 * interesting byte is at the other end of the word, but the
-	 * other bits are not necessarily zero.  We need a signed result
-	 * representing the differnece in the unsigned bytes, so for the
-	 * little-endian case we can't just shift the interesting bits up.
-	 */
-#ifdef __ARMEB__
-	sub	r0, r0, r3, lsr #24
-#else
-	and	r3, r3, #255
-	/* No RSB instruction in Thumb2 */
-#ifdef __thumb2__
-	lsr	r0, r0, #24
-	sub	r0, r0, r3
-#else
-	rsb	r0, r3, r0, lsr #24
-#endif
-#endif
-	ldr	r4, [sp], #4
-	bx	lr
-
-.Lstrcmp_unaligned:
-	wp1 .req r0
-	wp2 .req r1
-	b1  .req r2
-	w1  .req r4
-	w2  .req r5
-	t1  .req ip
-	@ r3 is scratch
-
-	/* First of all, compare bytes until wp1(sp1) is word-aligned. */
-1:
-	tst	wp1, #3
-	beq	2f
-	ldrb	r2, [wp1], #1
-	ldrb	r3, [wp2], #1
-	cmp	r2, #1
-	it	cs
-	cmpcs	r2, r3
-	beq	1b
-	sub	r0, r2, r3
-	bx	lr
-
-2:
-	str	r5, [sp, #-4]!
-	str	r4, [sp, #-4]!
-	mov	b1, #1
-	orr	b1, b1, b1, lsl #8
-	orr	b1, b1, b1, lsl #16
-
-	and	t1, wp2, #3
-	bic	wp2, wp2, #3
-	ldr	w1, [wp1], #4
-	ldr	w2, [wp2], #4
-	cmp	t1, #2
-	beq	2f
-	bhi	3f
-
-	/* Critical inner Loop: Block with 3 bytes initial overlap */
-	.p2align	2
-1:
-	bic	t1, w1, #MSB
-	cmp	t1, w2, SHFT2LSB #8
-	sub	r3, w1, b1
-	bic	r3, r3, w1
-	bne	4f
-	ands	r3, r3, b1, lsl #7
-	it	eq
-	ldreq	w2, [wp2], #4
-	bne	5f
-	eor	t1, t1, w1
-	cmp	t1, w2, SHFT2MSB #24
-	bne	6f
-	ldr	w1, [wp1], #4
-	b	1b
-4:
-	SHFT2LSB	w2, w2, #8
-	b	8f
-
-5:
-#ifdef __ARMEB__
-	/* The syndrome value may contain false ones if the string ends
-	 * with the bytes 0x01 0x00
-	 */
-	tst	w1, #0xff000000
-	itt	ne
-	tstne	w1, #0x00ff0000
-	tstne	w1, #0x0000ff00
-	beq	7f
-#else
-	bics	r3, r3, #0xff000000
-	bne	7f
-#endif
-	ldrb	w2, [wp2]
-	SHFT2LSB	t1, w1, #24
-#ifdef __ARMEB__
-	lsl	w2, w2, #24
-#endif
-	b	8f
-
-6:
-	SHFT2LSB	t1, w1, #24
-	and	w2, w2, #LSB
-	b	8f
-
-	/* Critical inner Loop: Block with 2 bytes initial overlap */
-	.p2align	2
-2:
-	SHFT2MSB	t1, w1, #16
-	sub	r3, w1, b1
-	SHFT2LSB	t1, t1, #16
-	bic	r3, r3, w1
-	cmp	t1, w2, SHFT2LSB #16
-	bne	4f
-	ands	r3, r3, b1, lsl #7
-	it	eq
-	ldreq	w2, [wp2], #4
-	bne	5f
-	eor	t1, t1, w1
-	cmp	t1, w2, SHFT2MSB #16
-	bne	6f
-	ldr	w1, [wp1], #4
-	b	2b
-
-5:
-#ifdef __ARMEB__
-	/* The syndrome value may contain false ones if the string ends
-	 * with the bytes 0x01 0x00
-	 */
-	tst	w1, #0xff000000
-	it	ne
-	tstne	w1, #0x00ff0000
-	beq	7f
-#else
-	lsls	r3, r3, #16
-	bne	7f
-#endif
-	ldrh	w2, [wp2]
-	SHFT2LSB	t1, w1, #16
-#ifdef __ARMEB__
-	lsl	w2, w2, #16
-#endif
-	b	8f
-
-6:
-	SHFT2MSB	w2, w2, #16
-	SHFT2LSB	t1, w1, #16
-4:
-	SHFT2LSB	w2, w2, #16
-	b	8f
-
-	/* Critical inner Loop: Block with 1 byte initial overlap */
-	.p2align	2
-3:
-	and	t1, w1, #LSB
-	cmp	t1, w2, SHFT2LSB #24
-	sub	r3, w1, b1
-	bic	r3, r3, w1
-	bne	4f
-	ands	r3, r3, b1, lsl #7
-	it	eq
-	ldreq	w2, [wp2], #4
-	bne	5f
-	eor	t1, t1, w1
-	cmp	t1, w2, SHFT2MSB #8
-	bne	6f
-	ldr	w1, [wp1], #4
-	b	3b
-4:
-	SHFT2LSB	w2, w2, #24
-	b	8f
-5:
-	/* The syndrome value may contain false ones if the string ends
-	 * with the bytes 0x01 0x00
-	 */
-	tst	w1, #LSB
-	beq	7f
-	ldr	w2, [wp2], #4
-6:
-	SHFT2LSB	t1, w1, #8
-	bic	w2, w2, #MSB
-	b	8f
-7:
-	mov	r0, #0
-	ldr	r4, [sp], #4
-	ldr	r5, [sp], #4
-	bx	lr
-
-8:
-	and	r2, t1, #LSB
-	and	r0, w2, #LSB
-	cmp	r0, #1
-	it	cs
-	cmpcs	r0, r2
-	itt	eq
-	SHFT2LSBEQ	t1, t1, #8
-	SHFT2LSBEQ	w2, w2, #8
-	beq	8b
-	sub	r0, r2, r0
-	ldr	r4, [sp], #4
-	ldr	r5, [sp], #4
-	bx	lr
-END(strcmp)

libc/arch-arm/bionic/strcmp.a15.S

@@ -1,787 +0,0 @@
-/*
- * Copyright (c) 2013 ARM Ltd
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- * 3. The name of the company may not be used to endorse or promote
- *    products derived from this software without specific prior written
- *    permission.
- *
- * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
- * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
- * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
- * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
- * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
- * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
- * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
- * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
- * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- */
-
-#include "arm_asm.h"
-
-#ifdef __ARMEB__
-#define S2LOMEM lsl
-#define S2LOMEMEQ lsleq
-#define S2HIMEM lsr
-#define MSB 0x000000ff
-#define LSB 0xff000000
-#define BYTE0_OFFSET 24
-#define BYTE1_OFFSET 16
-#define BYTE2_OFFSET 8
-#define BYTE3_OFFSET 0
-#else /* not  __ARMEB__ */
-#define S2LOMEM lsr
-#define S2LOMEMEQ lsreq
-#define S2HIMEM lsl
-#define BYTE0_OFFSET 0
-#define BYTE1_OFFSET 8
-#define BYTE2_OFFSET 16
-#define BYTE3_OFFSET 24
-#define MSB 0xff000000
-#define LSB 0x000000ff
-#endif /* not  __ARMEB__ */
-
-.syntax         unified
-
-#if defined (__thumb__)
-        .thumb
-        .thumb_func
-#endif
-        .global strcmp
-        .type   strcmp, %function
-strcmp:
-
-#if (defined (__thumb__) && !defined (__thumb2__))
-1:
-        ldrb    r2, [r0]
-        ldrb    r3, [r1]
-        adds    r0, r0, #1
-        adds    r1, r1, #1
-        cmp     r2, #0
-        beq     2f
-        cmp     r2, r3
-        beq     1b
-2:
-        subs    r0, r2, r3
-        bx      lr
-#elif (defined (__OPTIMIZE_SIZE__) || defined (PREFER_SIZE_OVER_SPEED))
-1:
-        ldrb    r2, [r0], #1
-        ldrb    r3, [r1], #1
-        cmp     r2, #1
-        it      cs
-        cmpcs   r2, r3
-        beq     1b
-        subs    r0, r2, r3
-        RETURN
-
-
-#elif (defined (_ISA_THUMB_2) || defined (_ISA_ARM_6))
-      /* Use LDRD whenever possible.  */
-
-/* The main thing to look out for when comparing large blocks is that
-   the loads do not cross a page boundary when loading past the index
-   of the byte with the first difference or the first string-terminator.
-
-   For example, if the strings are identical and the string-terminator
-   is at index k, byte by byte comparison will not load beyond address
-   s1+k and s2+k; word by word comparison may load up to 3 bytes beyond
-   k; double word - up to 7 bytes.  If the load of these bytes crosses
-   a page boundary, it might cause a memory fault (if the page is not mapped)
-   that would not have happened in byte by byte comparison.
-
-   If an address is (double) word aligned, then a load of a (double) word
-   from that address will not cross a page boundary.
-   Therefore, the algorithm below considers word and double-word alignment
-   of strings separately.  */
-
-/* High-level description of the algorithm.
-
-   * The fast path: if both strings are double-word aligned,
-     use LDRD to load two words from each string in every loop iteration.
-   * If the strings have the same offset from a word boundary,
-     use LDRB to load and compare byte by byte until
-     the first string is aligned to a word boundary (at most 3 bytes).
-     This is optimized for quick return on short unaligned strings.
-   * If the strings have the same offset from a double-word boundary,
-     use LDRD to load two words from each string in every loop iteration, as in the fast path.
-   * If the strings do not have the same offset from a double-word boundary,
-     load a word from the second string before the loop to initialize the queue.
-     Use LDRD to load two words from every string in every loop iteration.
-     Inside the loop, load the second word from the second string only after comparing
-     the first word, using the queued value, to guarantee safety across page boundaries.
-   * If the strings do not have the same offset from a word boundary,
-     use LDR and a shift queue. Order of loads and comparisons matters,
-     similarly to the previous case.
-
-   * Use UADD8 and SEL to compare words, and use REV and CLZ to compute the return value.
-   * The only difference between ARM and Thumb modes is the use of CBZ instruction.
-   * The only difference between big and little endian is the use of REV in little endian
-     to compute the return value, instead of MOV.
-   * No preload. [TODO.]
-*/
-
-        .macro m_cbz reg label
-#ifdef __thumb2__
-        cbz     \reg, \label
-#else   /* not defined __thumb2__ */
-        cmp     \reg, #0
-        beq     \label
-#endif /* not defined __thumb2__ */
-        .endm /* m_cbz */
-
-        .macro m_cbnz reg label
-#ifdef __thumb2__
-        cbnz    \reg, \label
-#else   /* not defined __thumb2__ */
-        cmp     \reg, #0
-        bne     \label
-#endif /* not defined __thumb2__ */
-        .endm /* m_cbnz */
-
-        .macro  init
-        /* Macro to save temporary registers and prepare magic values.  */
-        subs    sp, sp, #16
-        strd    r4, r5, [sp, #8]
-        strd    r6, r7, [sp]
-        mvn     r6, #0  /* all F */
-        mov     r7, #0  /* all 0 */
-        .endm   /* init */
-
-        .macro  magic_compare_and_branch w1 w2 label
-        /* Macro to compare registers w1 and w2 and conditionally branch to label.  */
-        cmp     \w1, \w2        /* Are w1 and w2 the same?  */
-        magic_find_zero_bytes \w1
-        it      eq
-        cmpeq   ip, #0          /* Is there a zero byte in w1?  */
-        bne     \label
-        .endm /* magic_compare_and_branch */
-
-        .macro  magic_find_zero_bytes w1
-        /* Macro to find all-zero bytes in w1, result is in ip.  */
-#if (defined (__ARM_FEATURE_DSP))
-        uadd8   ip, \w1, r6
-        sel     ip, r7, r6
-#else /* not defined (__ARM_FEATURE_DSP) */
-        /* __ARM_FEATURE_DSP is not defined for some Cortex-M processors.
-        Coincidently, these processors only have Thumb-2 mode, where we can use the
-        the (large) magic constant available directly as an immediate in instructions.
-        Note that we cannot use the magic constant in ARM mode, where we need
-        to create the constant in a register.  */
-        sub     ip, \w1, #0x01010101
-        bic     ip, ip, \w1
-        and     ip, ip, #0x80808080
-#endif /* not defined (__ARM_FEATURE_DSP) */
-        .endm /* magic_find_zero_bytes */
-
-        .macro  setup_return w1 w2
-#ifdef __ARMEB__
-        mov     r1, \w1
-        mov     r2, \w2
-#else /* not  __ARMEB__ */
-        rev     r1, \w1
-        rev     r2, \w2
-#endif /* not  __ARMEB__ */
-        .endm /* setup_return */
-
-        /*
-        optpld r0, #0
-        optpld r1, #0
-        */
-
-        /* Are both strings double-word aligned?  */
-        orr     ip, r0, r1
-        tst     ip, #7
-        bne     do_align
-
-        /* Fast path.  */
-        init
-
-doubleword_aligned:
-
-        /* Get here when the strings to compare are double-word aligned.  */
-        /* Compare two words in every iteration.  */
-        .p2align        2
-2:
-        /*
-        optpld r0, #16
-        optpld r1, #16
-        */
-
-        /* Load the next double-word from each string.  */
-        ldrd    r2, r3, [r0], #8
-        ldrd    r4, r5, [r1], #8
-
-        magic_compare_and_branch w1=r2, w2=r4, label=return_24
-        magic_compare_and_branch w1=r3, w2=r5, label=return_35
-        b       2b
-
-do_align:
-        /* Is the first string word-aligned?  */
-        ands    ip, r0, #3
-        beq     word_aligned_r0
-
-        /* Fast compare byte by byte until the first string is word-aligned.  */
-        /* The offset of r0 from a word boundary is in ip. Thus, the number of bytes
-        to read until the next word boudnary is 4-ip.  */
-        bic     r0, r0, #3
-        ldr     r2, [r0], #4
-        lsls    ip, ip, #31
-        beq     byte2
-        bcs     byte3
-
-byte1:
-        ldrb    ip, [r1], #1
-        uxtb    r3, r2, ror #BYTE1_OFFSET
-        subs    ip, r3, ip
-        bne     fast_return
-        m_cbz   reg=r3, label=fast_return
-
-byte2:
-        ldrb    ip, [r1], #1
-        uxtb    r3, r2, ror #BYTE2_OFFSET
-        subs    ip, r3, ip
-        bne     fast_return
-        m_cbz   reg=r3, label=fast_return
-
-byte3:
-        ldrb    ip, [r1], #1
-        uxtb    r3, r2, ror #BYTE3_OFFSET
-        subs    ip, r3, ip
-        bne     fast_return
-        m_cbnz  reg=r3, label=word_aligned_r0
-
-fast_return:
-        mov     r0, ip
-        bx      lr
-
-word_aligned_r0:
-        init
-        /* The first string is word-aligned.  */
-        /* Is the second string word-aligned?  */
-        ands    ip, r1, #3
-        bne     strcmp_unaligned
-
-word_aligned:
-        /* The strings are word-aligned. */
-        /* Is the first string double-word aligned?  */
-        tst     r0, #4
-        beq     doubleword_aligned_r0
-
-        /* If r0 is not double-word aligned yet, align it by loading
-        and comparing the next word from each string.  */
-        ldr     r2, [r0], #4
-        ldr     r4, [r1], #4
-        magic_compare_and_branch w1=r2 w2=r4 label=return_24
-
-doubleword_aligned_r0:
-        /* Get here when r0 is double-word aligned.  */
-        /* Is r1 doubleword_aligned?  */
-        tst     r1, #4
-        beq     doubleword_aligned
-
-        /* Get here when the strings to compare are word-aligned,
-        r0 is double-word aligned, but r1 is not double-word aligned.  */
-
-        /* Initialize the queue.  */
-        ldr     r5, [r1], #4
-
-        /* Compare two words in every iteration.  */
-        .p2align        2
-3:
-        /*
-        optpld r0, #16
-        optpld r1, #16
-        */
-
-        /* Load the next double-word from each string and compare.  */
-        ldrd    r2, r3, [r0], #8
-        magic_compare_and_branch w1=r2 w2=r5 label=return_25
-        ldrd    r4, r5, [r1], #8
-        magic_compare_and_branch w1=r3 w2=r4 label=return_34
-        b       3b
-
-        .macro miscmp_word offsetlo offsethi
-        /* Macro to compare misaligned strings.  */
-        /* r0, r1 are word-aligned, and at least one of the strings
-        is not double-word aligned.  */
-        /* Compare one word in every loop iteration.  */
-        /* OFFSETLO is the original bit-offset of r1 from a word-boundary,
-        OFFSETHI is 32 - OFFSETLO (i.e., offset from the next word).  */
-
-        /* Initialize the shift queue.  */
-        ldr     r5, [r1], #4
-
-        /* Compare one word from each string in every loop iteration.  */
-        .p2align        2
-7:
-        ldr     r3, [r0], #4
-        S2LOMEM r5, r5, #\offsetlo
-        magic_find_zero_bytes w1=r3
-        cmp     r7, ip, S2HIMEM #\offsetlo
-        and     r2, r3, r6, S2LOMEM #\offsetlo
-        it      eq
-        cmpeq   r2, r5
-        bne     return_25
-        ldr     r5, [r1], #4
-        cmp     ip, #0
-        eor r3, r2, r3
-        S2HIMEM r2, r5, #\offsethi
-        it      eq
-        cmpeq   r3, r2
-        bne     return_32
-        b       7b
-        .endm /* miscmp_word */
-
-strcmp_unaligned:
-        /* r0 is word-aligned, r1 is at offset ip from a word.  */
-        /* Align r1 to the (previous) word-boundary.  */
-        bic     r1, r1, #3
-
-        /* Unaligned comparison word by word using LDRs. */
-        cmp     ip, #2
-        beq     miscmp_word_16                    /* If ip == 2.  */
-        bge     miscmp_word_24                    /* If ip == 3.  */
-        miscmp_word offsetlo=8 offsethi=24        /* If ip == 1.  */
-miscmp_word_16:  miscmp_word offsetlo=16 offsethi=16
-miscmp_word_24:  miscmp_word offsetlo=24 offsethi=8
-
-
-return_32:
-        setup_return w1=r3, w2=r2
-        b       do_return
-return_34:
-        setup_return w1=r3, w2=r4
-        b       do_return
-return_25:
-        setup_return w1=r2, w2=r5
-        b       do_return
-return_35:
-        setup_return w1=r3, w2=r5
-        b       do_return
-return_24:
-        setup_return w1=r2, w2=r4
-
-do_return:
-
-#ifdef __ARMEB__
-        mov     r0, ip
-#else /* not  __ARMEB__ */
-        rev     r0, ip
-#endif /* not  __ARMEB__ */
-
-        /* Restore temporaries early, before computing the return value.  */
-        ldrd    r6, r7, [sp]
-        ldrd    r4, r5, [sp, #8]
-        adds    sp, sp, #16
-
-        /* There is a zero or a different byte between r1 and r2.  */
-        /* r0 contains a mask of all-zero bytes in r1.  */
-        /* Using r0 and not ip here because cbz requires low register.  */
-        m_cbz   reg=r0, label=compute_return_value
-        clz     r0, r0
-        /* r0 contains the number of bits on the left of the first all-zero byte in r1.  */
-        rsb     r0, r0, #24
-        /* Here, r0 contains the number of bits on the right of the first all-zero byte in r1.  */
-        lsr     r1, r1, r0
-        lsr     r2, r2, r0
-
-compute_return_value:
-        movs    r0, #1
-        cmp     r1, r2
-        /* The return value is computed as follows.
-        If r1>r2 then (C==1 and Z==0) and LS doesn't hold and r0 is #1 at return.
-        If r1<r2 then (C==0 and Z==0) and we execute SBC with carry_in=0,
-        which means r0:=r0-r0-1 and r0 is #-1 at return.
-        If r1=r2 then (C==1 and Z==1) and we execute SBC with carry_in=1,
-        which means r0:=r0-r0 and r0 is #0 at return.
-        (C==0 and Z==1) cannot happen because the carry bit is "not borrow".  */
-        it      ls
-        sbcls   r0, r0, r0
-        bx      lr
-
-
-#else   /* !(defined (_ISA_THUMB_2) || defined (_ISA_ARM_6)
-             defined (__OPTIMIZE_SIZE__) || defined (PREFER_SIZE_OVER_SPEED) ||
-             (defined (__thumb__) && !defined (__thumb2__))) */
-
-        /* Use LDR whenever possible. */
-
-#ifdef __thumb2__
-#define magic1(REG) 0x01010101
-#define magic2(REG) 0x80808080
-#else
-#define magic1(REG) REG
-#define magic2(REG) REG, lsl #7
-#endif
-
-        optpld  r0
-        optpld  r1
-        eor     r2, r0, r1
-        tst     r2, #3
-        /* Strings not at same byte offset from a word boundary.  */
-        bne     strcmp_unaligned
-        ands    r2, r0, #3
-        bic     r0, r0, #3
-        bic     r1, r1, #3
-        ldr     ip, [r0], #4
-        it      eq
-        ldreq   r3, [r1], #4
-        beq     1f
-        /* Although s1 and s2 have identical initial alignment, they are
-        not currently word aligned.  Rather than comparing bytes,
-    make sure that any bytes fetched from before the addressed
-    bytes are forced to 0xff.  Then they will always compare
-    equal.  */
-        eor     r2, r2, #3
-        lsl     r2, r2, #3
-        mvn     r3, MSB
-        S2LOMEM        r2, r3, r2
-        ldr     r3, [r1], #4
-        orr     ip, ip, r2
-        orr     r3, r3, r2
-1:
-#ifndef __thumb2__
-              /* Load the 'magic' constant 0x01010101.  */
-        str     r4, [sp, #-4]!
-        mov     r4, #1
-        orr     r4, r4, r4, lsl #8
-        orr     r4, r4, r4, lsl #16
-#endif
-        .p2align        2
-4:
-        optpld  r0, #8
-        optpld  r1, #8
-        sub     r2, ip, magic1(r4)
-        cmp     ip, r3
-        itttt   eq
-        /* check for any zero bytes in first word */
-        biceq   r2, r2, ip
-        tsteq   r2, magic2(r4)
-        ldreq   ip, [r0], #4
-        ldreq   r3, [r1], #4
-        beq     4b
-2:
-        /* There's a zero or a different byte in the word */
-        S2HIMEM  r0, ip, #24
-        S2LOMEM  ip, ip, #8
-        cmp     r0, #1
-        it      cs
-        cmpcs   r0, r3, S2HIMEM #24
-        it      eq
-        S2LOMEMEQ r3, r3, #8
-        beq     2b
-        /* On a big-endian machine, r0 contains the desired byte in bits
-        0-7; on a little-endian machine they are in bits 24-31.  In
-        both cases the other bits in r0 are all zero.  For r3 the
-        interesting byte is at the other end of the word, but the
-        other bits are not necessarily zero.  We need a signed result
-        representing the differnece in the unsigned bytes, so for the
-        little-endian case we can't just shift the interesting bits
-        up.  */
-#ifdef __ARMEB__
-        sub     r0, r0, r3, lsr #24
-#else
-        and     r3, r3, #255
-#ifdef __thumb2__
-        /* No RSB instruction in Thumb2 */
-        lsr     r0, r0, #24
-        sub     r0, r0, r3
-#else
-        rsb     r0, r3, r0, lsr #24
-#endif
-#endif
-#ifndef __thumb2__
-        ldr     r4, [sp], #4
-#endif
-        RETURN
-
-
-strcmp_unaligned:
-
-#if 0
-        /* The assembly code below is based on the following alogrithm.  */
-#ifdef __ARMEB__
-#define RSHIFT <<
-#define LSHIFT >>
-#else
-#define RSHIFT >>
-#define LSHIFT <<
-#endif
-
-#define body(shift)                         \
-  mask = 0xffffffffU RSHIFT shift;                  \
-  w1 = *wp1++;                              \
-  w2 = *wp2++;                              \
-  do                                    \
-    {                                   \
-      t1 = w1 & mask;                           \
-      if (__builtin_expect(t1 != w2 RSHIFT shift, 0))           \
-    {                               \
-      w2 RSHIFT= shift;                     \
-      break;                            \
-    }                               \
-      if (__builtin_expect(((w1 - b1) & ~w1) & (b1 << 7), 0))       \
-    {                               \
-      /* See comment in assembler below re syndrome on big-endian */\
-      if ((((w1 - b1) & ~w1) & (b1 << 7)) & mask)           \
-        w2 RSHIFT= shift;                       \
-      else                              \
-        {                               \
-          w2 = *wp2;                        \
-          t1 = w1 RSHIFT (32 - shift);              \
-          w2 = (w2 LSHIFT (32 - shift)) RSHIFT (32 - shift);    \
-        }                               \
-      break;                            \
-    }                               \
-      w2 = *wp2++;                          \
-      t1 ^= w1;                             \
-      if (__builtin_expect(t1 != w2 LSHIFT (32 - shift), 0))        \
-    {                               \
-      t1 = w1 >> (32 - shift);                  \
-      w2 = (w2 << (32 - shift)) RSHIFT (32 - shift);        \
-      break;                            \
-    }                               \
-      w1 = *wp1++;                          \
-    } while (1)
-
-  const unsigned* wp1;
-  const unsigned* wp2;
-  unsigned w1, w2;
-  unsigned mask;
-  unsigned shift;
-  unsigned b1 = 0x01010101;
-  char c1, c2;
-  unsigned t1;
-
-  while (((unsigned) s1) & 3)
-    {
-      c1 = *s1++;
-      c2 = *s2++;
-      if (c1 == 0 || c1 != c2)
-    return c1 - (int)c2;
-    }
-  wp1 = (unsigned*) (((unsigned)s1) & ~3);
-  wp2 = (unsigned*) (((unsigned)s2) & ~3);
-  t1 = ((unsigned) s2) & 3;
-  if (t1 == 1)
-    {
-      body(8);
-    }
-  else if (t1 == 2)
-    {
-      body(16);
-    }
-  else
-    {
-      body (24);
-    }
-
-  do
-    {
-#ifdef __ARMEB__
-      c1 = (char) t1 >> 24;
-      c2 = (char) w2 >> 24;
-#else /* not  __ARMEB__ */
-      c1 = (char) t1;
-      c2 = (char) w2;
-#endif /* not  __ARMEB__ */
-      t1 RSHIFT= 8;
-      w2 RSHIFT= 8;
-    } while (c1 != 0 && c1 == c2);
-  return c1 - c2;
-#endif /* 0 */
-
-
-        wp1 .req r0
-        wp2 .req r1
-        b1  .req r2
-        w1  .req r4
-        w2  .req r5
-        t1  .req ip
-        @ r3 is scratch
-
-        /* First of all, compare bytes until wp1(sp1) is word-aligned. */
-1:
-        tst     wp1, #3
-        beq     2f
-        ldrb    r2, [wp1], #1
-        ldrb    r3, [wp2], #1
-        cmp     r2, #1
-        it      cs
-        cmpcs   r2, r3
-        beq     1b
-        sub     r0, r2, r3
-        RETURN
-
-2:
-        str     r5, [sp, #-4]!
-        str     r4, [sp, #-4]!
-        //stmfd   sp!, {r4, r5}
-        mov     b1, #1
-        orr     b1, b1, b1, lsl #8
-        orr     b1, b1, b1, lsl #16
-
-        and     t1, wp2, #3
-        bic     wp2, wp2, #3
-        ldr     w1, [wp1], #4
-        ldr     w2, [wp2], #4
-        cmp     t1, #2
-        beq     2f
-        bhi     3f
-
-        /* Critical inner Loop: Block with 3 bytes initial overlap */
-        .p2align        2
-1:
-        bic     t1, w1, MSB
-        cmp     t1, w2, S2LOMEM #8
-        sub     r3, w1, b1
-        bic     r3, r3, w1
-        bne     4f
-        ands    r3, r3, b1, lsl #7
-        it      eq
-        ldreq   w2, [wp2], #4
-        bne     5f
-        eor     t1, t1, w1
-        cmp     t1, w2, S2HIMEM #24
-        bne     6f
-        ldr     w1, [wp1], #4
-        b       1b
-4:
-        S2LOMEM        w2, w2, #8
-        b       8f
-
-5:
-#ifdef __ARMEB__
-        /* The syndrome value may contain false ones if the string ends
-        with the bytes 0x01 0x00 */
-        tst     w1, #0xff000000
-        itt     ne
-        tstne   w1, #0x00ff0000
-        tstne   w1, #0x0000ff00
-        beq     7f
-#else
-        bics    r3, r3, #0xff000000
-        bne     7f
-#endif
-        ldrb    w2, [wp2]
-        S2LOMEM  t1, w1, #24
-#ifdef __ARMEB__
-        lsl     w2, w2, #24
-#endif
-        b       8f
-
-6:
-        S2LOMEM  t1, w1, #24
-        and     w2, w2, LSB
-        b       8f
-
-        /* Critical inner Loop: Block with 2 bytes initial overlap */
-        .p2align        2
-2:
-        S2HIMEM  t1, w1, #16
-        sub     r3, w1, b1
-        S2LOMEM  t1, t1, #16
-        bic     r3, r3, w1
-        cmp     t1, w2, S2LOMEM #16
-        bne     4f
-        ands    r3, r3, b1, lsl #7
-        it      eq
-        ldreq   w2, [wp2], #4
-        bne     5f
-        eor     t1, t1, w1
-        cmp     t1, w2, S2HIMEM #16
-        bne     6f
-        ldr     w1, [wp1], #4
-        b       2b
-
-5:
-#ifdef __ARMEB__
-        /* The syndrome value may contain false ones if the string ends
-        with the bytes 0x01 0x00 */
-        tst     w1, #0xff000000
-        it      ne
-        tstne   w1, #0x00ff0000
-        beq     7f
-#else
-        lsls    r3, r3, #16
-        bne     7f
-#endif
-        ldrh    w2, [wp2]
-        S2LOMEM  t1, w1, #16
-#ifdef __ARMEB__
-        lsl     w2, w2, #16
-#endif
-        b       8f
-
-6:
-        S2HIMEM  w2, w2, #16
-        S2LOMEM  t1, w1, #16
-4:
-        S2LOMEM  w2, w2, #16
-        b       8f
-
-        /* Critical inner Loop: Block with 1 byte initial overlap */
-        .p2align        2
-3:
-        and     t1, w1, LSB
-        cmp     t1, w2, S2LOMEM #24
-        sub     r3, w1, b1
-        bic     r3, r3, w1
-        bne     4f
-        ands    r3, r3, b1, lsl #7
-        it      eq
-        ldreq   w2, [wp2], #4
-        bne     5f
-        eor     t1, t1, w1
-        cmp     t1, w2, S2HIMEM #8
-        bne     6f
-        ldr     w1, [wp1], #4
-        b       3b
-4:
-        S2LOMEM  w2, w2, #24
-        b       8f
-5:
-        /* The syndrome value may contain false ones if the string ends
-        with the bytes 0x01 0x00 */
-        tst     w1, LSB
-        beq     7f
-        ldr     w2, [wp2], #4
-6:
-        S2LOMEM  t1, w1, #8
-        bic     w2, w2, MSB
-        b       8f
-7:
-        mov     r0, #0
-        //ldmfd   sp!, {r4, r5}
-        ldr     r4, [sp], #4
-        ldr     r5, [sp], #4
-        RETURN
-8:
-        and     r2, t1, LSB
-        and     r0, w2, LSB
-        cmp     r0, #1
-        it      cs
-        cmpcs   r0, r2
-        itt     eq
-        S2LOMEMEQ        t1, t1, #8
-        S2LOMEMEQ        w2, w2, #8
-        beq     8b
-        sub     r0, r2, r0
-        //ldmfd   sp!, {r4, r5}
-        ldr     r4, [sp], #4
-        ldr     r5, [sp], #4
-        RETURN
-
-#endif /* !(defined (_ISA_THUMB_2) || defined (_ISA_ARM_6)
-            defined (__OPTIMIZE_SIZE__) || defined (PREFER_SIZE_OVER_SPEED) ||
-            (defined (__thumb__) && !defined (__thumb2__))) */

libc/arch-arm/bionic/syscall.S

@@ -26,12 +26,16 @@
  * SUCH DAMAGE.
  */
 
-#include <linux/err.h>
-#include <machine/asm.h>
+#include <private/bionic_asm.h>
 
 ENTRY(syscall)
     mov     ip, sp
     stmfd   sp!, {r4, r5, r6, r7}
+    .cfi_def_cfa_offset 16
+    .cfi_rel_offset r4, 0
+    .cfi_rel_offset r5, 4
+    .cfi_rel_offset r6, 8
+    .cfi_rel_offset r7, 12
     mov     r7, r0
     mov     r0, r1
     mov     r1, r2
@@ -39,8 +43,9 @@ ENTRY(syscall)
     ldmfd   ip, {r3, r4, r5, r6}
     swi     #0
     ldmfd   sp!, {r4, r5, r6, r7}
+    .cfi_def_cfa_offset 0
     cmn     r0, #(MAX_ERRNO + 1)
     bxls    lr
     neg     r0, r0
-    b       __set_errno
+    b       __set_errno_internal
 END(syscall)

libc/arch-arm/bionic/tgkill.S

@@ -1,51 +0,0 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include <linux/err.h>
-#include <asm/unistd.h>
-#include <machine/asm.h>
-
-/* unlike our auto-generated syscall stubs, this code saves lr
-   on the stack, as well as a few other registers. this makes
-   our stack unwinder happy, when we generate debug stack
-   traces after the C library or other parts of the system
-   abort due to a fatal runtime error (e.g. detection
-   of a corrupted malloc heap).
-*/
-
-ENTRY(tgkill)
-    .save   {r4-r7, ip, lr}
-    stmfd   sp!, {r4-r7, ip, lr}
-    ldr     r7, =__NR_tgkill
-    swi     #0
-    ldmfd   sp!, {r4-r7, ip, lr}
-    cmn     r0, #(MAX_ERRNO + 1)
-    bxls    lr
-    neg     r0, r0
-    b       __set_errno
-END(tgkill)

libc/arch-arm/bionic/tkill.S

@@ -1,50 +0,0 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include <linux/err.h>
-#include <asm/unistd.h>
-#include <machine/asm.h>
-
-/* unlike our auto-generated syscall stubs, this code saves lr
-   on the stack, as well as a few other registers. this makes
-   our stack unwinder happy, when we generate debug stack
-   traces after the C library or other parts of the system
-   abort due to a fatal runtime error (e.g. detection
-   of a corrupted malloc heap).
-*/
-
-ENTRY(tkill)
-    stmfd   sp!, {r4-r7, ip, lr}
-    ldr     r7, =__NR_tkill
-    swi     #0
-    ldmfd   sp!, {r4-r7, ip, lr}
-    cmn     r0, #(MAX_ERRNO + 1)
-    bxls    lr
-    neg     r0, r0
-    b       __set_errno
-END(tkill)

libc/arch-arm/bionic/vfork.S

@@ -0,0 +1,46 @@
+/*
+ * Copyright (C) 2015 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <private/bionic_asm.h>
+
+ENTRY(vfork)
+    // __get_tls()[TLS_SLOT_THREAD_ID]->cached_pid_ = 0
+    mrc     p15, 0, r3, c13, c0, 3
+    ldr     r3, [r3, #4]
+    mov     r0, #0
+    str     r0, [r3, #12]
+
+    mov     ip, r7
+    ldr     r7, =__NR_vfork
+    swi     #0
+    mov     r7, ip
+    cmn     r0, #(MAX_ERRNO + 1)
+    bxls    lr
+    neg     r0, r0
+    b       __set_errno_internal
+END(vfork)

libc/arch-arm/cortex-a15/bionic/__strcat_chk.S

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2013 The Android Open Source Project
+ * Copyright (C) 2015 The Android Open Source Project
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -26,202 +26,7 @@
  * SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
-#include "libc_events.h"
+// Indicate which memcpy base file to include.
+#define MEMCPY_BASE "memcpy_base.S"
 
-    .syntax unified
-
-    .thumb
-    .thumb_func
-
-// Get the length of src string, then get the source of the dst string.
-// Check that the two lengths together don't exceed the threshold, then
-// do a memcpy of the data.
-ENTRY(__strcat_chk)
-    .cfi_startproc
-    pld     [r0, #0]
-    push    {r0, lr}
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-    push    {r4, r5}
-    .save   {r4, r5}
-    .cfi_adjust_cfa_offset 8
-    .cfi_rel_offset r4, 0
-    .cfi_rel_offset r5, 4
-
-    mov     lr, r2
-
-    // Save the dst register to r5
-    mov     r5, r0
-
-    // Zero out r4
-    eor     r4, r4, r4
-
-    // r1 contains the address of the string to count.
-.L_strlen_start:
-    mov     r0, r1
-    ands    r3, r1, #7
-    beq     .L_mainloop
-
-    // Align to a double word (64 bits).
-    rsb     r3, r3, #8
-    lsls    ip, r3, #31
-    beq     .L_align_to_32
-
-    ldrb    r2, [r1], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_32:
-    bcc     .L_align_to_64
-    ands    ip, r3, #2
-    beq     .L_align_to_64
-
-    ldrb    r2, [r1], #1
-    cbz     r2, .L_update_count_and_finish
-    ldrb    r2, [r1], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_64:
-    tst     r3, #4
-    beq     .L_mainloop
-    ldr     r3, [r1], #4
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-
-    .p2align 2
-.L_mainloop:
-    ldrd    r2, r3, [r1], #8
-
-    pld     [r1, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-    b       .L_mainloop
-
-.L_update_count_and_finish:
-    sub     r3, r1, r0
-    sub     r3, r3, #1
-    b       .L_finish
-
-.L_zero_in_first_register:
-    sub     r3, r1, r0
-    lsls    r2, ip, #17
-    bne     .L_sub8_and_finish
-    bcs     .L_sub7_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub6_and_finish
-
-    sub     r3, r3, #5
-    b       .L_finish
-
-.L_sub8_and_finish:
-    sub     r3, r3, #8
-    b       .L_finish
-
-.L_sub7_and_finish:
-    sub     r3, r3, #7
-    b       .L_finish
-
-.L_sub6_and_finish:
-    sub     r3, r3, #6
-    b       .L_finish
-
-.L_zero_in_second_register:
-    sub     r3, r1, r0
-    lsls    r2, ip, #17
-    bne     .L_sub4_and_finish
-    bcs     .L_sub3_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub2_and_finish
-
-    sub     r3, r3, #1
-    b       .L_finish
-
-.L_sub4_and_finish:
-    sub     r3, r3, #4
-    b       .L_finish
-
-.L_sub3_and_finish:
-    sub     r3, r3, #3
-    b       .L_finish
-
-.L_sub2_and_finish:
-    sub     r3, r3, #2
-
-.L_finish:
-    cmp     r4, #0
-    bne     .L_strlen_done
-
-    // Time to get the dst string length.
-    mov     r1, r5
-
-    // Save the original source address to r5.
-    mov     r5, r0
-
-    // Save the current length (adding 1 for the terminator).
-    add     r4, r3, #1
-    b       .L_strlen_start
-
-    // r0 holds the pointer to the dst string.
-    // r3 holds the dst string length.
-    // r4 holds the src string length + 1.
-.L_strlen_done:
-    add     r2, r3, r4
-    cmp     r2, lr
-    bhi     __strcat_chk_failed
-
-    // Set up the registers for the memcpy code.
-    mov     r1, r5
-    pld     [r1, #64]
-    mov     r2, r4
-    add     r0, r0, r3
-    pop     {r4, r5}
-
-    .cfi_endproc
-END(__strcat_chk)
-
-#define MEMCPY_BASE         __strcat_chk_memcpy_base
-#define MEMCPY_BASE_ALIGNED __strcat_chk_memcpy_base_aligned
-
-#include "memcpy_base.S"
-
-ENTRY(__strcat_chk_failed)
-    .cfi_startproc
-    .save   {r0, lr}
-    .save   {r4, r5}
-
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-    .cfi_adjust_cfa_offset 8
-    .cfi_rel_offset r4, 0
-    .cfi_rel_offset r5, 4
-
-    ldr     r0, error_message
-    ldr     r1, error_code
-1:
-    add     r0, pc
-    bl      __fortify_chk_fail
-error_code:
-    .word   BIONIC_EVENT_STRCAT_BUFFER_OVERFLOW
-error_message:
-    .word   error_string-(1b+4)
-
-    .cfi_endproc
-END(__strcat_chk_failed)
-
-    .data
-error_string:
-    .string "strcat buffer overflow"
+#include "__strcat_chk_common.S"

libc/arch-arm/cortex-a15/bionic/__strcat_chk_common.S

@@ -0,0 +1,212 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <private/bionic_asm.h>
+#include <private/libc_events.h>
+
+    .syntax unified
+
+    .thumb
+    .thumb_func
+
+// Get the length of src string, then get the source of the dst string.
+// Check that the two lengths together don't exceed the threshold, then
+// do a memcpy of the data.
+ENTRY(__strcat_chk)
+    pld     [r0, #0]
+    push    {r0, lr}
+    .cfi_def_cfa_offset 8
+    .cfi_rel_offset r0, 0
+    .cfi_rel_offset lr, 4
+    push    {r4, r5}
+    .cfi_adjust_cfa_offset 8
+    .cfi_rel_offset r4, 0
+    .cfi_rel_offset r5, 4
+
+    mov     lr, r2
+
+    // Save the dst register to r5
+    mov     r5, r0
+
+    // Zero out r4
+    eor     r4, r4, r4
+
+    // r1 contains the address of the string to count.
+.L_strlen_start:
+    mov     r0, r1
+    ands    r3, r1, #7
+    beq     .L_mainloop
+
+    // Align to a double word (64 bits).
+    rsb     r3, r3, #8
+    lsls    ip, r3, #31
+    beq     .L_align_to_32
+
+    ldrb    r2, [r1], #1
+    cbz     r2, .L_update_count_and_finish
+
+.L_align_to_32:
+    bcc     .L_align_to_64
+    ands    ip, r3, #2
+    beq     .L_align_to_64
+
+    ldrb    r2, [r1], #1
+    cbz     r2, .L_update_count_and_finish
+    ldrb    r2, [r1], #1
+    cbz     r2, .L_update_count_and_finish
+
+.L_align_to_64:
+    tst     r3, #4
+    beq     .L_mainloop
+    ldr     r3, [r1], #4
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_second_register
+
+    .p2align 2
+.L_mainloop:
+    ldrd    r2, r3, [r1], #8
+
+    pld     [r1, #64]
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_first_register
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_second_register
+    b       .L_mainloop
+
+.L_update_count_and_finish:
+    sub     r3, r1, r0
+    sub     r3, r3, #1
+    b       .L_finish
+
+.L_zero_in_first_register:
+    sub     r3, r1, r0
+    lsls    r2, ip, #17
+    bne     .L_sub8_and_finish
+    bcs     .L_sub7_and_finish
+    lsls    ip, ip, #1
+    bne     .L_sub6_and_finish
+
+    sub     r3, r3, #5
+    b       .L_finish
+
+.L_sub8_and_finish:
+    sub     r3, r3, #8
+    b       .L_finish
+
+.L_sub7_and_finish:
+    sub     r3, r3, #7
+    b       .L_finish
+
+.L_sub6_and_finish:
+    sub     r3, r3, #6
+    b       .L_finish
+
+.L_zero_in_second_register:
+    sub     r3, r1, r0
+    lsls    r2, ip, #17
+    bne     .L_sub4_and_finish
+    bcs     .L_sub3_and_finish
+    lsls    ip, ip, #1
+    bne     .L_sub2_and_finish
+
+    sub     r3, r3, #1
+    b       .L_finish
+
+.L_sub4_and_finish:
+    sub     r3, r3, #4
+    b       .L_finish
+
+.L_sub3_and_finish:
+    sub     r3, r3, #3
+    b       .L_finish
+
+.L_sub2_and_finish:
+    sub     r3, r3, #2
+
+.L_finish:
+    cmp     r4, #0
+    bne     .L_strlen_done
+
+    // Time to get the dst string length.
+    mov     r1, r5
+
+    // Save the original source address to r5.
+    mov     r5, r0
+
+    // Save the current length (adding 1 for the terminator).
+    add     r4, r3, #1
+    b       .L_strlen_start
+
+    // r0 holds the pointer to the dst string.
+    // r3 holds the dst string length.
+    // r4 holds the src string length + 1.
+.L_strlen_done:
+    add     r2, r3, r4
+    cmp     r2, lr
+    bhi     .L_strcat_chk_failed
+
+    // Set up the registers for the memcpy code.
+    mov     r1, r5
+    pld     [r1, #64]
+    mov     r2, r4
+    add     r0, r0, r3
+    pop     {r4, r5}
+    .cfi_adjust_cfa_offset -8
+    .cfi_restore r4
+    .cfi_restore r5
+
+#include MEMCPY_BASE
+
+    // Undo the above cfi directives
+    .cfi_adjust_cfa_offset 8
+    .cfi_rel_offset r4, 0
+    .cfi_rel_offset r5, 4
+.L_strcat_chk_failed:
+    ldr     r0, error_message
+    ldr     r1, error_code
+1:
+    add     r0, pc
+    bl      __fortify_chk_fail
+error_code:
+    .word   BIONIC_EVENT_STRCAT_BUFFER_OVERFLOW
+error_message:
+    .word   error_string-(1b+4)
+END(__strcat_chk)
+
+    .data
+error_string:
+    .string "strcat: prevented write past end of buffer"

libc/arch-arm/cortex-a15/bionic/__strcpy_chk.S

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2013 The Android Open Source Project
+ * Copyright (C) 2015 The Android Open Source Project
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -26,163 +26,7 @@
  * SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
-#include "libc_events.h"
+// Indicate which memcpy base file to include.
+#define MEMCPY_BASE "memcpy_base.S"
 
-    .syntax unified
-
-    .thumb
-    .thumb_func
-
-// Get the length of the source string first, then do a memcpy of the data
-// instead of a strcpy.
-ENTRY(__strcpy_chk)
-    .cfi_startproc
-    pld     [r0, #0]
-    push    {r0, lr}
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-
-    mov     lr, r2
-    mov     r0, r1
-
-    ands    r3, r1, #7
-    beq     .L_mainloop
-
-    // Align to a double word (64 bits).
-    rsb     r3, r3, #8
-    lsls    ip, r3, #31
-    beq     .L_align_to_32
-
-    ldrb    r2, [r0], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_32:
-    bcc     .L_align_to_64
-    ands    ip, r3, #2
-    beq     .L_align_to_64
-
-    ldrb    r2, [r0], #1
-    cbz     r2, .L_update_count_and_finish
-    ldrb    r2, [r0], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_64:
-    tst     r3, #4
-    beq     .L_mainloop
-    ldr     r3, [r0], #4
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-
-    .p2align 2
-.L_mainloop:
-    ldrd    r2, r3, [r0], #8
-
-    pld     [r0, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-    b       .L_mainloop
-
-.L_update_count_and_finish:
-    sub     r3, r0, r1
-    sub     r3, r3, #1
-    b       .L_check_size
-
-.L_zero_in_first_register:
-    sub     r3, r0, r1
-    lsls    r2, ip, #17
-    bne     .L_sub8_and_finish
-    bcs     .L_sub7_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub6_and_finish
-
-    sub     r3, r3, #5
-    b       .L_check_size
-
-.L_sub8_and_finish:
-    sub     r3, r3, #8
-    b       .L_check_size
-
-.L_sub7_and_finish:
-    sub     r3, r3, #7
-    b       .L_check_size
-
-.L_sub6_and_finish:
-    sub     r3, r3, #6
-    b       .L_check_size
-
-.L_zero_in_second_register:
-    sub     r3, r0, r1
-    lsls    r2, ip, #17
-    bne     .L_sub4_and_finish
-    bcs     .L_sub3_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub2_and_finish
-
-    sub     r3, r3, #1
-    b       .L_check_size
-
-.L_sub4_and_finish:
-    sub     r3, r3, #4
-    b       .L_check_size
-
-.L_sub3_and_finish:
-    sub     r3, r3, #3
-    b       .L_check_size
-
-.L_sub2_and_finish:
-    sub     r3, r3, #2
-
-.L_check_size:
-    pld     [r1, #0]
-    pld     [r1, #64]
-    ldr     r0, [sp]
-    cmp     r3, lr
-    bhs     __strcpy_chk_failed
-
-    // Add 1 for copy length to get the string terminator.
-    add     r2, r3, #1
-
-    .cfi_endproc
-END(__strcpy_chk)
-
-#define MEMCPY_BASE         __strcpy_chk_memcpy_base
-#define MEMCPY_BASE_ALIGNED __strcpy_chk_memcpy_base_aligned
-#include "memcpy_base.S"
-
-ENTRY(__strcpy_chk_failed)
-    .cfi_startproc
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-
-    ldr     r0, error_message
-    ldr     r1, error_code
-1:
-    add     r0, pc
-    bl      __fortify_chk_fail
-error_code:
-    .word   BIONIC_EVENT_STRCPY_BUFFER_OVERFLOW
-error_message:
-    .word   error_string-(1b+4)
-
-    .cfi_endproc
-END(__strcpy_chk_failed)
-
-    .data
-error_string:
-    .string "strcpy buffer overflow"
+#include "__strcpy_chk_common.S"

libc/arch-arm/cortex-a15/bionic/__strcpy_chk_common.S

@@ -0,0 +1,173 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <private/bionic_asm.h>
+#include <private/libc_events.h>
+
+    .syntax unified
+
+    .thumb
+    .thumb_func
+
+// Get the length of the source string first, then do a memcpy of the data
+// instead of a strcpy.
+ENTRY(__strcpy_chk)
+    pld     [r0, #0]
+    push    {r0, lr}
+    .cfi_def_cfa_offset 8
+    .cfi_rel_offset r0, 0
+    .cfi_rel_offset lr, 4
+
+    mov     lr, r2
+    mov     r0, r1
+
+    ands    r3, r1, #7
+    beq     .L_mainloop
+
+    // Align to a double word (64 bits).
+    rsb     r3, r3, #8
+    lsls    ip, r3, #31
+    beq     .L_align_to_32
+
+    ldrb    r2, [r0], #1
+    cbz     r2, .L_update_count_and_finish
+
+.L_align_to_32:
+    bcc     .L_align_to_64
+    ands    ip, r3, #2
+    beq     .L_align_to_64
+
+    ldrb    r2, [r0], #1
+    cbz     r2, .L_update_count_and_finish
+    ldrb    r2, [r0], #1
+    cbz     r2, .L_update_count_and_finish
+
+.L_align_to_64:
+    tst     r3, #4
+    beq     .L_mainloop
+    ldr     r3, [r0], #4
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_second_register
+
+    .p2align 2
+.L_mainloop:
+    ldrd    r2, r3, [r0], #8
+
+    pld     [r0, #64]
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_first_register
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_second_register
+    b       .L_mainloop
+
+.L_update_count_and_finish:
+    sub     r3, r0, r1
+    sub     r3, r3, #1
+    b       .L_check_size
+
+.L_zero_in_first_register:
+    sub     r3, r0, r1
+    lsls    r2, ip, #17
+    bne     .L_sub8_and_finish
+    bcs     .L_sub7_and_finish
+    lsls    ip, ip, #1
+    bne     .L_sub6_and_finish
+
+    sub     r3, r3, #5
+    b       .L_check_size
+
+.L_sub8_and_finish:
+    sub     r3, r3, #8
+    b       .L_check_size
+
+.L_sub7_and_finish:
+    sub     r3, r3, #7
+    b       .L_check_size
+
+.L_sub6_and_finish:
+    sub     r3, r3, #6
+    b       .L_check_size
+
+.L_zero_in_second_register:
+    sub     r3, r0, r1
+    lsls    r2, ip, #17
+    bne     .L_sub4_and_finish
+    bcs     .L_sub3_and_finish
+    lsls    ip, ip, #1
+    bne     .L_sub2_and_finish
+
+    sub     r3, r3, #1
+    b       .L_check_size
+
+.L_sub4_and_finish:
+    sub     r3, r3, #4
+    b       .L_check_size
+
+.L_sub3_and_finish:
+    sub     r3, r3, #3
+    b       .L_check_size
+
+.L_sub2_and_finish:
+    sub     r3, r3, #2
+
+.L_check_size:
+    pld     [r1, #0]
+    pld     [r1, #64]
+    ldr     r0, [sp]
+    cmp     r3, lr
+    bhs     .L_strcpy_chk_failed
+
+    // Add 1 for copy length to get the string terminator.
+    add     r2, r3, #1
+
+#include MEMCPY_BASE
+
+.L_strcpy_chk_failed:
+    ldr     r0, error_message
+    ldr     r1, error_code
+1:
+    add     r0, pc
+    bl      __fortify_chk_fail
+error_code:
+    .word   BIONIC_EVENT_STRCPY_BUFFER_OVERFLOW
+error_message:
+    .word   error_string-(1b+4)
+END(__strcpy_chk)
+
+    .data
+error_string:
+    .string "strcpy: prevented write past end of buffer"

libc/arch-arm/cortex-a15/bionic/memcpy.S

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2008 The Android Open Source Project
+ * Copyright (C) 2015 The Android Open Source Project
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -25,88 +25,8 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
-/*
- * Copyright (c) 2013 ARM Ltd
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- * 3. The name of the company may not be used to endorse or promote
- *    products derived from this software without specific prior written
- *    permission.
- *
- * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
- * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
- * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
- * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
- * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
- * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
- * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
- * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
- * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- */
 
-// Prototype: void *memcpy (void *dst, const void *src, size_t count).
+// Indicate which memcpy base file to include.
+#define MEMCPY_BASE "memcpy_base.S"
 
-#include <machine/asm.h>
-#include "libc_events.h"
-
-        .text
-        .syntax unified
-        .fpu    neon
-
-ENTRY(__memcpy_chk)
-        .cfi_startproc
-        cmp     r2, r3
-        bhi     __memcpy_chk_fail
-
-        // Fall through to memcpy...
-        .cfi_endproc
-END(__memcpy_chk)
-
-ENTRY(memcpy)
-        .cfi_startproc
-        pld     [r1, #64]
-        push    {r0, lr}
-        .save   {r0, lr}
-        .cfi_def_cfa_offset 8
-        .cfi_rel_offset r0, 0
-        .cfi_rel_offset lr, 4
-
-        .cfi_endproc
-END(memcpy)
-
-#define MEMCPY_BASE         __memcpy_base
-#define MEMCPY_BASE_ALIGNED __memcpy_base_aligned
-#include "memcpy_base.S"
-
-ENTRY(__memcpy_chk_fail)
-        .cfi_startproc
-        // Preserve lr for backtrace.
-        push    {lr}
-        .save   {lr}
-        .cfi_def_cfa_offset 4
-        .cfi_rel_offset lr, 0
-
-        ldr     r0, error_message
-        ldr     r1, error_code
-1:
-        add     r0, pc
-        bl      __fortify_chk_fail
-error_code:
-        .word   BIONIC_EVENT_MEMCPY_BUFFER_OVERFLOW
-error_message:
-        .word   error_string-(1b+8)
-        .cfi_endproc
-END(__memcpy_chk_fail)
-
-        .data
-error_string:
-        .string "memcpy buffer overflow"
+#include "memcpy_common.S"

libc/arch-arm/cortex-a15/bionic/memcpy_base.S

@@ -53,13 +53,7 @@
  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-ENTRY(MEMCPY_BASE)
-        .cfi_startproc
-        .save   {r0, lr}
-        .cfi_def_cfa_offset 8
-        .cfi_rel_offset r0, 0
-        .cfi_rel_offset lr, 4
-
+.L_memcpy_base:
         // Assumes that n >= 0, and dst, src are valid pointers.
         // For any sizes less than 832 use the neon code that doesn't
         // care about the src alignment. This avoids any checks
@@ -171,31 +165,18 @@ ENTRY(MEMCPY_BASE)
         ands    r3, r3, #0x3
         bne     .L_copy_unknown_alignment
 
-        .cfi_endproc
-END(MEMCPY_BASE)
-
-ENTRY(MEMCPY_BASE_ALIGNED)
-        .cfi_startproc
-        .save   {r0, lr}
-        .cfi_def_cfa_offset 8
-        .cfi_rel_offset r0, 0
-        .cfi_rel_offset lr, 4
-
         // To try and improve performance, stack layout changed,
         // i.e., not keeping the stack looking like users expect
         // (highest numbered register at highest address).
         strd    r4, r5, [sp, #-8]!
-        .save   {r4, r5}
         .cfi_adjust_cfa_offset 8
         .cfi_rel_offset r4, 0
         .cfi_rel_offset r5, 4
         strd    r6, r7, [sp, #-8]!
-        .save   {r6, r7}
         .cfi_adjust_cfa_offset 8
         .cfi_rel_offset r6, 0
-        .cfi_rel_offset r7, 0
+        .cfi_rel_offset r7, 4
         strd    r8, r9, [sp, #-8]!
-        .save   {r8, r9}
         .cfi_adjust_cfa_offset 8
         .cfi_rel_offset r8, 0
         .cfi_rel_offset r9, 4
@@ -300,10 +281,28 @@ ENTRY(MEMCPY_BASE_ALIGNED)
 
         // Restore registers: optimized pop {r0, pc}
         ldrd    r8, r9, [sp], #8
+        .cfi_adjust_cfa_offset -8
+        .cfi_restore r8
+        .cfi_restore r9
         ldrd    r6, r7, [sp], #8
+        .cfi_adjust_cfa_offset -8
+        .cfi_restore r6
+        .cfi_restore r7
         ldrd    r4, r5, [sp], #8
+        .cfi_adjust_cfa_offset -8
+        .cfi_restore r4
+        .cfi_restore r5
         pop     {r0, pc}
 
+        // Put the cfi directives back for the below instructions.
+        .cfi_adjust_cfa_offset 24
+        .cfi_rel_offset r4, 0
+        .cfi_rel_offset r5, 4
+        .cfi_rel_offset r6, 8
+        .cfi_rel_offset r7, 12
+        .cfi_rel_offset r8, 16
+        .cfi_rel_offset r9, 20
+
 .L_dst_not_word_aligned:
         // Align dst to word.
         rsb     ip, ip, #4
@@ -325,5 +324,11 @@ ENTRY(MEMCPY_BASE_ALIGNED)
         // Src is guaranteed to be at least word aligned by this point.
         b       .L_word_aligned
 
-        .cfi_endproc
-END(MEMCPY_BASE_ALIGNED)
+        // Undo any cfi directives from above.
+        .cfi_adjust_cfa_offset -24
+        .cfi_restore r4
+        .cfi_restore r5
+        .cfi_restore r6
+        .cfi_restore r7
+        .cfi_restore r8
+        .cfi_restore r9

libc/arch-arm/cortex-a15/bionic/memcpy_common.S

@@ -0,0 +1,103 @@
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+/*
+ * Copyright (c) 2013 ARM Ltd
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. The name of the company may not be used to endorse or promote
+ *    products derived from this software without specific prior written
+ *    permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <private/bionic_asm.h>
+#include <private/libc_events.h>
+
+        .text
+        .syntax unified
+        .fpu    neon
+
+ENTRY(__memcpy_chk)
+        cmp     r2, r3
+        bhi     .L_memcpy_chk_fail
+
+        // Fall through to memcpy...
+END(__memcpy_chk)
+
+// Prototype: void *memcpy (void *dst, const void *src, size_t count).
+ENTRY(memcpy)
+        pld     [r1, #64]
+        push    {r0, lr}
+        .cfi_def_cfa_offset 8
+        .cfi_rel_offset r0, 0
+        .cfi_rel_offset lr, 4
+
+#include MEMCPY_BASE
+
+        // Undo the cfi instructions from above.
+        .cfi_def_cfa_offset 0
+        .cfi_restore r0
+        .cfi_restore lr
+.L_memcpy_chk_fail:
+        // Preserve lr for backtrace.
+        push    {lr}
+        .cfi_adjust_cfa_offset 4
+        .cfi_rel_offset lr, 0
+
+        ldr     r0, error_message
+        ldr     r1, error_code
+1:
+        add     r0, pc
+        bl      __fortify_chk_fail
+error_code:
+        .word   BIONIC_EVENT_MEMCPY_BUFFER_OVERFLOW
+error_message:
+        .word   error_string-(1b+8)
+END(memcpy)
+
+        .data
+error_string:
+        .string "memcpy: prevented write past end of buffer"

libc/arch-arm/cortex-a15/bionic/memset.S

@@ -27,8 +27,8 @@
  */
 
 #include <machine/cpu-features.h>
-#include <machine/asm.h>
-#include "libc_events.h"
+#include <private/bionic_asm.h>
+#include <private/libc_events.h>
 
         /*
          * Optimized memset() for ARM.
@@ -40,12 +40,10 @@
         .syntax     unified
 
 ENTRY(__memset_chk)
-        .cfi_startproc
         cmp         r2, r3
         bls         .L_done
 
         // Preserve lr for backtrace.
-        .save       {lr}
         push        {lr}
         .cfi_def_cfa_offset 4
         .cfi_rel_offset lr, 0
@@ -59,22 +57,16 @@ error_code:
         .word       BIONIC_EVENT_MEMSET_BUFFER_OVERFLOW
 error_message:
         .word       error_string-(1b+8)
-
-        .cfi_endproc
 END(__memset_chk)
 
 ENTRY(bzero)
-        .cfi_startproc
         mov         r2, r1
         mov         r1, #0
 .L_done:
         // Fall through to memset...
-        .cfi_endproc
 END(bzero)
 
 ENTRY(memset)
-        .cfi_startproc
-        .save       {r0}
         stmfd       sp!, {r0}
         .cfi_def_cfa_offset 4
         .cfi_rel_offset r0, 0
@@ -192,9 +184,8 @@ ENTRY(memset)
         strbcs      r1, [r0], #1
         ldmfd       sp!, {r0}
         bx          lr
-        .cfi_endproc
 END(memset)
 
         .data
 error_string:
-        .string     "memset buffer overflow"
+        .string     "memset: prevented write past end of buffer"

libc/arch-arm/cortex-a15/bionic/stpcpy.S

@@ -0,0 +1,30 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#define STPCPY
+#include "string_copy.S"

libc/arch-arm/cortex-a15/bionic/strcat.S

@@ -53,7 +53,7 @@
  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
+#include <private/bionic_asm.h>
 
     .syntax unified
 
@@ -70,7 +70,7 @@
 
     .macro m_scan_byte
     ldrb    r3, [r0]
-    cbz     r3, strcat_r0_scan_done
+    cbz     r3, .L_strcat_r0_scan_done
     add     r0, #1
     .endm // m_scan_byte
 
@@ -84,10 +84,10 @@ ENTRY(strcat)
     // Quick check to see if src is empty.
     ldrb    r2, [r1]
     pld     [r1, #0]
-    cbnz    r2, strcat_continue
+    cbnz    r2, .L_strcat_continue
     bx      lr
 
-strcat_continue:
+.L_strcat_continue:
     // To speed up really small dst strings, unroll checking the first 4 bytes.
     m_push
     m_scan_byte
@@ -96,95 +96,102 @@ strcat_continue:
     m_scan_byte
 
     ands    r3, r0, #7
-    beq     strcat_mainloop
+    beq     .L_strcat_mainloop
 
     // Align to a double word (64 bits).
     rsb     r3, r3, #8
     lsls    ip, r3, #31
-    beq     strcat_align_to_32
+    beq     .L_strcat_align_to_32
 
     ldrb    r5, [r0]
-    cbz     r5, strcat_r0_scan_done
+    cbz     r5, .L_strcat_r0_scan_done
     add     r0, r0, #1
 
-strcat_align_to_32:
-    bcc     strcat_align_to_64
+.L_strcat_align_to_32:
+    bcc     .L_strcat_align_to_64
 
     ldrb    r2, [r0]
-    cbz     r2, strcat_r0_scan_done
+    cbz     r2, .L_strcat_r0_scan_done
     add     r0, r0, #1
     ldrb    r4, [r0]
-    cbz     r4, strcat_r0_scan_done
+    cbz     r4, .L_strcat_r0_scan_done
     add     r0, r0, #1
 
-strcat_align_to_64:
+.L_strcat_align_to_64:
     tst     r3, #4
-    beq     strcat_mainloop
+    beq     .L_strcat_mainloop
     ldr     r3, [r0], #4
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcat_zero_in_second_register
-    b       strcat_mainloop
+    bne     .L_strcat_zero_in_second_register
+    b       .L_strcat_mainloop
 
-strcat_r0_scan_done:
+.L_strcat_r0_scan_done:
     // For short copies, hard-code checking the first 8 bytes since this
     // new code doesn't win until after about 8 bytes.
-    m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r5, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r5, cmd=cbnz, label=strcpy_continue
+    m_copy_byte reg=r2, cmd=cbz, label=.L_strcpy_finish
+    m_copy_byte reg=r3, cmd=cbz, label=.L_strcpy_finish
+    m_copy_byte reg=r4, cmd=cbz, label=.L_strcpy_finish
+    m_copy_byte reg=r5, cmd=cbz, label=.L_strcpy_finish
+    m_copy_byte reg=r2, cmd=cbz, label=.L_strcpy_finish
+    m_copy_byte reg=r3, cmd=cbz, label=.L_strcpy_finish
+    m_copy_byte reg=r4, cmd=cbz, label=.L_strcpy_finish
+    m_copy_byte reg=r5, cmd=cbnz, label=.L_strcpy_continue
 
-strcpy_finish:
+.L_strcpy_finish:
     m_pop
 
-strcpy_continue:
+.L_strcpy_continue:
     ands    r3, r0, #7
-    beq     strcpy_check_src_align
+    beq     .L_strcpy_check_src_align
 
     // Align to a double word (64 bits).
     rsb     r3, r3, #8
     lsls    ip, r3, #31
-    beq     strcpy_align_to_32
+    beq     .L_strcpy_align_to_32
 
     ldrb    r2, [r1], #1
     strb    r2, [r0], #1
-    cbz     r2, strcpy_complete
+    cbz     r2, .L_strcpy_complete
 
-strcpy_align_to_32:
-    bcc     strcpy_align_to_64
+.L_strcpy_align_to_32:
+    bcc     .L_strcpy_align_to_64
 
     ldrb    r2, [r1], #1
     strb    r2, [r0], #1
-    cbz     r2, strcpy_complete
+    cbz     r2, .L_strcpy_complete
     ldrb    r2, [r1], #1
     strb    r2, [r0], #1
-    cbz     r2, strcpy_complete
+    cbz     r2, .L_strcpy_complete
 
-strcpy_align_to_64:
+.L_strcpy_align_to_64:
     tst     r3, #4
-    beq     strcpy_check_src_align
-    ldr     r2, [r1], #4
+    beq     .L_strcpy_check_src_align
+    // Read one byte at a time since we don't know the src alignment
+    // and we don't want to read into a different page.
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .L_strcpy_complete
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .L_strcpy_complete
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .L_strcpy_complete
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .L_strcpy_complete
 
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-    str     r2, [r0], #4
-
-strcpy_check_src_align:
+.L_strcpy_check_src_align:
     // At this point dst is aligned to a double word, check if src
     // is also aligned to a double word.
     ands    r3, r1, #7
-    bne     strcpy_unaligned_copy
+    bne     .L_strcpy_unaligned_copy
 
     .p2align 2
-strcpy_mainloop:
+.L_strcpy_mainloop:
     ldrd    r2, r3, [r1], #8
 
     pld     [r1, #64]
@@ -192,128 +199,128 @@ strcpy_mainloop:
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
+    bne     .L_strcpy_zero_in_first_register
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
+    bne     .L_strcpy_zero_in_second_register
 
     strd    r2, r3, [r0], #8
-    b       strcpy_mainloop
+    b       .L_strcpy_mainloop
 
-strcpy_complete:
+.L_strcpy_complete:
     m_pop
 
-strcpy_zero_in_first_register:
+.L_strcpy_zero_in_first_register:
     lsls    lr, ip, #17
-    bne     strcpy_copy1byte
-    bcs     strcpy_copy2bytes
+    bne     .L_strcpy_copy1byte
+    bcs     .L_strcpy_copy2bytes
     lsls    ip, ip, #1
-    bne     strcpy_copy3bytes
+    bne     .L_strcpy_copy3bytes
 
-strcpy_copy4bytes:
+.L_strcpy_copy4bytes:
     // Copy 4 bytes to the destiniation.
     str     r2, [r0]
     m_pop
 
-strcpy_copy1byte:
+.L_strcpy_copy1byte:
     strb    r2, [r0]
     m_pop
 
-strcpy_copy2bytes:
+.L_strcpy_copy2bytes:
     strh    r2, [r0]
     m_pop
 
-strcpy_copy3bytes:
+.L_strcpy_copy3bytes:
     strh    r2, [r0], #2
     lsr     r2, #16
     strb    r2, [r0]
     m_pop
 
-strcpy_zero_in_second_register:
+.L_strcpy_zero_in_second_register:
     lsls    lr, ip, #17
-    bne     strcpy_copy5bytes
-    bcs     strcpy_copy6bytes
+    bne     .L_strcpy_copy5bytes
+    bcs     .L_strcpy_copy6bytes
     lsls    ip, ip, #1
-    bne     strcpy_copy7bytes
+    bne     .L_strcpy_copy7bytes
 
     // Copy 8 bytes to the destination.
     strd    r2, r3, [r0]
     m_pop
 
-strcpy_copy5bytes:
+.L_strcpy_copy5bytes:
     str     r2, [r0], #4
     strb    r3, [r0]
     m_pop
 
-strcpy_copy6bytes:
+.L_strcpy_copy6bytes:
     str     r2, [r0], #4
     strh    r3, [r0]
     m_pop
 
-strcpy_copy7bytes:
+.L_strcpy_copy7bytes:
     str     r2, [r0], #4
     strh    r3, [r0], #2
     lsr     r3, #16
     strb    r3, [r0]
     m_pop
 
-strcpy_unaligned_copy:
+.L_strcpy_unaligned_copy:
     // Dst is aligned to a double word, while src is at an unknown alignment.
     // There are 7 different versions of the unaligned copy code
     // to prevent overreading the src. The mainloop of every single version
     // will store 64 bits per loop. The difference is how much of src can
     // be read without potentially crossing a page boundary.
     tbb     [pc, r3]
-strcpy_unaligned_branchtable:
+.L_strcpy_unaligned_branchtable:
     .byte 0
-    .byte ((strcpy_unalign7 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign6 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign5 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign4 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign3 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign2 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign1 - strcpy_unaligned_branchtable)/2)
+    .byte ((.L_strcpy_unalign7 - .L_strcpy_unaligned_branchtable)/2)
+    .byte ((.L_strcpy_unalign6 - .L_strcpy_unaligned_branchtable)/2)
+    .byte ((.L_strcpy_unalign5 - .L_strcpy_unaligned_branchtable)/2)
+    .byte ((.L_strcpy_unalign4 - .L_strcpy_unaligned_branchtable)/2)
+    .byte ((.L_strcpy_unalign3 - .L_strcpy_unaligned_branchtable)/2)
+    .byte ((.L_strcpy_unalign2 - .L_strcpy_unaligned_branchtable)/2)
+    .byte ((.L_strcpy_unalign1 - .L_strcpy_unaligned_branchtable)/2)
 
     .p2align 2
     // Can read 7 bytes before possibly crossing a page.
-strcpy_unalign7:
+.L_strcpy_unalign7:
     ldr     r2, [r1], #4
 
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
+    bne     .L_strcpy_zero_in_first_register
 
     ldrb    r3, [r1]
-    cbz     r3, strcpy_unalign7_copy5bytes
+    cbz     r3, .L_strcpy_unalign7_copy5bytes
     ldrb    r4, [r1, #1]
-    cbz     r4, strcpy_unalign7_copy6bytes
+    cbz     r4, .L_strcpy_unalign7_copy6bytes
     ldrb    r5, [r1, #2]
-    cbz     r5, strcpy_unalign7_copy7bytes
+    cbz     r5, .L_strcpy_unalign7_copy7bytes
 
     ldr     r3, [r1], #4
     pld     [r1, #64]
 
     lsrs    ip, r3, #24
     strd    r2, r3, [r0], #8
-    beq     strcpy_unalign_return
-    b       strcpy_unalign7
+    beq     .L_strcpy_unalign_return
+    b       .L_strcpy_unalign7
 
-strcpy_unalign7_copy5bytes:
+.L_strcpy_unalign7_copy5bytes:
     str     r2, [r0], #4
     strb    r3, [r0]
-strcpy_unalign_return:
+.L_strcpy_unalign_return:
     m_pop
 
-strcpy_unalign7_copy6bytes:
+.L_strcpy_unalign7_copy6bytes:
     str     r2, [r0], #4
     strb    r3, [r0], #1
     strb    r4, [r0], #1
     m_pop
 
-strcpy_unalign7_copy7bytes:
+.L_strcpy_unalign7_copy7bytes:
     str     r2, [r0], #4
     strb    r3, [r0], #1
     strb    r4, [r0], #1
@@ -322,41 +329,41 @@ strcpy_unalign7_copy7bytes:
 
     .p2align 2
     // Can read 6 bytes before possibly crossing a page.
-strcpy_unalign6:
+.L_strcpy_unalign6:
     ldr     r2, [r1], #4
 
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
+    bne     .L_strcpy_zero_in_first_register
 
     ldrb    r4, [r1]
-    cbz     r4, strcpy_unalign_copy5bytes
+    cbz     r4, .L_strcpy_unalign_copy5bytes
     ldrb    r5, [r1, #1]
-    cbz     r5, strcpy_unalign_copy6bytes
+    cbz     r5, .L_strcpy_unalign_copy6bytes
 
     ldr     r3, [r1], #4
     pld     [r1, #64]
 
     tst     r3, #0xff0000
-    beq     strcpy_copy7bytes
+    beq     .L_strcpy_copy7bytes
     lsrs    ip, r3, #24
     strd    r2, r3, [r0], #8
-    beq     strcpy_unalign_return
-    b       strcpy_unalign6
+    beq     .L_strcpy_unalign_return
+    b       .L_strcpy_unalign6
 
     .p2align 2
     // Can read 5 bytes before possibly crossing a page.
-strcpy_unalign5:
+.L_strcpy_unalign5:
     ldr     r2, [r1], #4
 
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
+    bne     .L_strcpy_zero_in_first_register
 
     ldrb    r4, [r1]
-    cbz     r4, strcpy_unalign_copy5bytes
+    cbz     r4, .L_strcpy_unalign_copy5bytes
 
     ldr     r3, [r1], #4
 
@@ -365,17 +372,17 @@ strcpy_unalign5:
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
+    bne     .L_strcpy_zero_in_second_register
 
     strd    r2, r3, [r0], #8
-    b       strcpy_unalign5
+    b       .L_strcpy_unalign5
 
-strcpy_unalign_copy5bytes:
+.L_strcpy_unalign_copy5bytes:
     str     r2, [r0], #4
     strb    r4, [r0]
     m_pop
 
-strcpy_unalign_copy6bytes:
+.L_strcpy_unalign_copy6bytes:
     str     r2, [r0], #4
     strb    r4, [r0], #1
     strb    r5, [r0]
@@ -383,13 +390,13 @@ strcpy_unalign_copy6bytes:
 
     .p2align 2
     // Can read 4 bytes before possibly crossing a page.
-strcpy_unalign4:
+.L_strcpy_unalign4:
     ldr     r2, [r1], #4
 
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
+    bne     .L_strcpy_zero_in_first_register
 
     ldr     r3, [r1], #4
     pld     [r1, #64]
@@ -397,20 +404,20 @@ strcpy_unalign4:
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
+    bne     .L_strcpy_zero_in_second_register
 
     strd    r2, r3, [r0], #8
-    b       strcpy_unalign4
+    b       .L_strcpy_unalign4
 
     .p2align 2
     // Can read 3 bytes before possibly crossing a page.
-strcpy_unalign3:
+.L_strcpy_unalign3:
     ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign3_copy1byte
+    cbz     r2, .L_strcpy_unalign3_copy1byte
     ldrb    r3, [r1, #1]
-    cbz     r3, strcpy_unalign3_copy2bytes
+    cbz     r3, .L_strcpy_unalign3_copy2bytes
     ldrb    r4, [r1, #2]
-    cbz     r4, strcpy_unalign3_copy3bytes
+    cbz     r4, .L_strcpy_unalign3_copy3bytes
 
     ldr     r2, [r1], #4
     ldr     r3, [r1], #4
@@ -418,26 +425,26 @@ strcpy_unalign3:
     pld     [r1, #64]
 
     lsrs    lr, r2, #24
-    beq     strcpy_copy4bytes
+    beq     .L_strcpy_copy4bytes
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
+    bne     .L_strcpy_zero_in_second_register
 
     strd    r2, r3, [r0], #8
-    b       strcpy_unalign3
+    b       .L_strcpy_unalign3
 
-strcpy_unalign3_copy1byte:
+.L_strcpy_unalign3_copy1byte:
     strb    r2, [r0]
     m_pop
 
-strcpy_unalign3_copy2bytes:
+.L_strcpy_unalign3_copy2bytes:
     strb    r2, [r0], #1
     strb    r3, [r0]
     m_pop
 
-strcpy_unalign3_copy3bytes:
+.L_strcpy_unalign3_copy3bytes:
     strb    r2, [r0], #1
     strb    r3, [r0], #1
     strb    r4, [r0]
@@ -445,34 +452,34 @@ strcpy_unalign3_copy3bytes:
 
     .p2align 2
     // Can read 2 bytes before possibly crossing a page.
-strcpy_unalign2:
+.L_strcpy_unalign2:
     ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign_copy1byte
+    cbz     r2, .L_strcpy_unalign_copy1byte
     ldrb    r4, [r1, #1]
-    cbz     r4, strcpy_unalign_copy2bytes
+    cbz     r4, .L_strcpy_unalign_copy2bytes
 
     ldr     r2, [r1], #4
     ldr     r3, [r1], #4
     pld     [r1, #64]
 
     tst     r2, #0xff0000
-    beq     strcpy_copy3bytes
+    beq     .L_strcpy_copy3bytes
     lsrs    ip, r2, #24
-    beq     strcpy_copy4bytes
+    beq     .L_strcpy_copy4bytes
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
+    bne     .L_strcpy_zero_in_second_register
 
     strd    r2, r3, [r0], #8
-    b       strcpy_unalign2
+    b       .L_strcpy_unalign2
 
     .p2align 2
     // Can read 1 byte before possibly crossing a page.
-strcpy_unalign1:
+.L_strcpy_unalign1:
     ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign_copy1byte
+    cbz     r2, .L_strcpy_unalign_copy1byte
 
     ldr     r2, [r1], #4
     ldr     r3, [r1], #4
@@ -482,27 +489,27 @@ strcpy_unalign1:
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
+    bne     .L_strcpy_zero_in_first_register
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
+    bne     .L_strcpy_zero_in_second_register
 
     strd    r2, r3, [r0], #8
-    b       strcpy_unalign1
+    b       .L_strcpy_unalign1
 
-strcpy_unalign_copy1byte:
+.L_strcpy_unalign_copy1byte:
     strb    r2, [r0]
     m_pop
 
-strcpy_unalign_copy2bytes:
+.L_strcpy_unalign_copy2bytes:
     strb    r2, [r0], #1
     strb    r4, [r0]
     m_pop
 
     .p2align 2
-strcat_mainloop:
+.L_strcat_mainloop:
     ldrd    r2, r3, [r0], #8
 
     pld     [r0, #64]
@@ -510,59 +517,59 @@ strcat_mainloop:
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcat_zero_in_first_register
+    bne     .L_strcat_zero_in_first_register
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcat_zero_in_second_register
-    b       strcat_mainloop
+    bne     .L_strcat_zero_in_second_register
+    b       .L_strcat_mainloop
 
-strcat_zero_in_first_register:
+.L_strcat_zero_in_first_register:
     // Prefetch the src now, it's going to be used soon.
     pld     [r1, #0]
     lsls    lr, ip, #17
-    bne     strcat_sub8
-    bcs     strcat_sub7
+    bne     .L_strcat_sub8
+    bcs     .L_strcat_sub7
     lsls    ip, ip, #1
-    bne     strcat_sub6
+    bne     .L_strcat_sub6
 
     sub     r0, r0, #5
-    b       strcat_r0_scan_done
+    b       .L_strcat_r0_scan_done
 
-strcat_sub8:
+.L_strcat_sub8:
     sub     r0, r0, #8
-    b       strcat_r0_scan_done
+    b       .L_strcat_r0_scan_done
 
-strcat_sub7:
+.L_strcat_sub7:
     sub     r0, r0, #7
-    b       strcat_r0_scan_done
+    b       .L_strcat_r0_scan_done
 
-strcat_sub6:
+.L_strcat_sub6:
     sub     r0, r0, #6
-    b       strcat_r0_scan_done
+    b       .L_strcat_r0_scan_done
 
-strcat_zero_in_second_register:
+.L_strcat_zero_in_second_register:
     // Prefetch the src now, it's going to be used soon.
     pld     [r1, #0]
     lsls    lr, ip, #17
-    bne     strcat_sub4
-    bcs     strcat_sub3
+    bne     .L_strcat_sub4
+    bcs     .L_strcat_sub3
     lsls    ip, ip, #1
-    bne     strcat_sub2
+    bne     .L_strcat_sub2
 
     sub     r0, r0, #1
-    b       strcat_r0_scan_done
+    b       .L_strcat_r0_scan_done
 
-strcat_sub4:
+.L_strcat_sub4:
     sub     r0, r0, #4
-    b       strcat_r0_scan_done
+    b       .L_strcat_r0_scan_done
 
-strcat_sub3:
+.L_strcat_sub3:
     sub     r0, r0, #3
-    b       strcat_r0_scan_done
+    b       .L_strcat_r0_scan_done
 
-strcat_sub2:
+.L_strcat_sub2:
     sub     r0, r0, #2
-    b       strcat_r0_scan_done
+    b       .L_strcat_r0_scan_done
 END(strcat)

libc/arch-arm/cortex-a15/bionic/strcmp.S

@@ -27,7 +27,7 @@
  */
 
 #include <machine/cpu-features.h>
-#include <machine/asm.h>
+#include <private/bionic_asm.h>
 
 #ifdef __ARMEB__
 #define S2LOMEM lsl
@@ -145,19 +145,8 @@ ENTRY(strcmp)
 
         .macro  magic_find_zero_bytes w1
         /* Macro to find all-zero bytes in w1, result is in ip.  */
-#if (defined (__ARM_FEATURE_DSP))
         uadd8   ip, \w1, r6
         sel     ip, r7, r6
-#else /* not defined (__ARM_FEATURE_DSP) */
-        /* __ARM_FEATURE_DSP is not defined for some Cortex-M processors.
-        Coincidently, these processors only have Thumb-2 mode, where we can use the
-        the (large) magic constant available directly as an immediate in instructions.
-        Note that we cannot use the magic constant in ARM mode, where we need
-        to create the constant in a register.  */
-        sub     ip, \w1, #0x01010101
-        bic     ip, ip, \w1
-        and     ip, ip, #0x80808080
-#endif /* not defined (__ARM_FEATURE_DSP) */
         .endm /* magic_find_zero_bytes */
 
         .macro  setup_return w1 w2
@@ -170,7 +159,6 @@ ENTRY(strcmp)
 #endif /* not  __ARMEB__ */
         .endm /* setup_return */
 
-        .cfi_startproc
         pld [r0, #0]
         pld [r1, #0]
 
@@ -180,7 +168,6 @@ ENTRY(strcmp)
         bne     .L_do_align
 
         /* Fast path.  */
-        .save   {r4-r7}
         init
 
 .L_doubleword_aligned:
@@ -386,5 +373,4 @@ ENTRY(strcmp)
         it      ls
         sbcls   r0, r0, r0
         bx      lr
-        .cfi_endproc
 END(strcmp)

libc/arch-arm/cortex-a15/bionic/strcpy.S

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2013 The Android Open Source Project
+ * Copyright (C) 2014 The Android Open Source Project
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -25,427 +25,6 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
-/*
- * Copyright (c) 2013 ARM Ltd
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- * 3. The name of the company may not be used to endorse or promote
- *    products derived from this software without specific prior written
- *    permission.
- *
- * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
- * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
- * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
- * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
- * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
- * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
- * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
- * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
- * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- */
 
-#include <machine/asm.h>
-
-    .syntax unified
-
-    .thumb
-    .thumb_func
-
-    .macro m_push
-    push    {r0, r4, r5, lr}
-    .endm // m_push
-
-    .macro m_pop
-    pop     {r0, r4, r5, pc}
-    .endm // m_pop
-
-    .macro m_copy_byte reg, cmd, label
-    ldrb    \reg, [r1], #1
-    strb    \reg, [r0], #1
-    \cmd    \reg, \label
-    .endm // m_copy_byte
-
-ENTRY(strcpy)
-    // For short copies, hard-code checking the first 8 bytes since this
-    // new code doesn't win until after about 8 bytes.
-    m_push
-    m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r5, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r5, cmd=cbnz, label=strcpy_continue
-
-strcpy_finish:
-    m_pop
-
-strcpy_continue:
-    pld     [r1, #0]
-    ands    r3, r0, #7
-    beq     strcpy_check_src_align
-
-    // Align to a double word (64 bits).
-    rsb     r3, r3, #8
-    lsls    ip, r3, #31
-    beq     strcpy_align_to_32
-
-    ldrb    r2, [r1], #1
-    strb    r2, [r0], #1
-    cbz     r2, strcpy_complete
-
-strcpy_align_to_32:
-    bcc     strcpy_align_to_64
-
-    ldrb    r2, [r1], #1
-    strb    r2, [r0], #1
-    cbz     r2, strcpy_complete
-    ldrb    r2, [r1], #1
-    strb    r2, [r0], #1
-    cbz     r2, strcpy_complete
-
-strcpy_align_to_64:
-    tst     r3, #4
-    beq     strcpy_check_src_align
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-    str     r2, [r0], #4
-
-strcpy_check_src_align:
-    // At this point dst is aligned to a double word, check if src
-    // is also aligned to a double word.
-    ands    r3, r1, #7
-    bne     strcpy_unaligned_copy
-
-    .p2align 2
-strcpy_mainloop:
-    ldrd    r2, r3, [r1], #8
-
-    pld     [r1, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    strd    r2, r3, [r0], #8
-    b       strcpy_mainloop
-
-strcpy_complete:
-    m_pop
-
-strcpy_zero_in_first_register:
-    lsls    lr, ip, #17
-    bne     strcpy_copy1byte
-    bcs     strcpy_copy2bytes
-    lsls    ip, ip, #1
-    bne     strcpy_copy3bytes
-
-strcpy_copy4bytes:
-    // Copy 4 bytes to the destiniation.
-    str     r2, [r0]
-    m_pop
-
-strcpy_copy1byte:
-    strb    r2, [r0]
-    m_pop
-
-strcpy_copy2bytes:
-    strh    r2, [r0]
-    m_pop
-
-strcpy_copy3bytes:
-    strh    r2, [r0], #2
-    lsr     r2, #16
-    strb    r2, [r0]
-    m_pop
-
-strcpy_zero_in_second_register:
-    lsls    lr, ip, #17
-    bne     strcpy_copy5bytes
-    bcs     strcpy_copy6bytes
-    lsls    ip, ip, #1
-    bne     strcpy_copy7bytes
-
-    // Copy 8 bytes to the destination.
-    strd    r2, r3, [r0]
-    m_pop
-
-strcpy_copy5bytes:
-    str     r2, [r0], #4
-    strb    r3, [r0]
-    m_pop
-
-strcpy_copy6bytes:
-    str     r2, [r0], #4
-    strh    r3, [r0]
-    m_pop
-
-strcpy_copy7bytes:
-    str     r2, [r0], #4
-    strh    r3, [r0], #2
-    lsr     r3, #16
-    strb    r3, [r0]
-    m_pop
-
-strcpy_unaligned_copy:
-    // Dst is aligned to a double word, while src is at an unknown alignment.
-    // There are 7 different versions of the unaligned copy code
-    // to prevent overreading the src. The mainloop of every single version
-    // will store 64 bits per loop. The difference is how much of src can
-    // be read without potentially crossing a page boundary.
-    tbb     [pc, r3]
-strcpy_unaligned_branchtable:
-    .byte 0
-    .byte ((strcpy_unalign7 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign6 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign5 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign4 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign3 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign2 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign1 - strcpy_unaligned_branchtable)/2)
-
-    .p2align 2
-    // Can read 7 bytes before possibly crossing a page.
-strcpy_unalign7:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldrb    r3, [r1]
-    cbz     r3, strcpy_unalign7_copy5bytes
-    ldrb    r4, [r1, #1]
-    cbz     r4, strcpy_unalign7_copy6bytes
-    ldrb    r5, [r1, #2]
-    cbz     r5, strcpy_unalign7_copy7bytes
-
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    lsrs    ip, r3, #24
-    strd    r2, r3, [r0], #8
-    beq     strcpy_unalign_return
-    b       strcpy_unalign7
-
-strcpy_unalign7_copy5bytes:
-    str     r2, [r0], #4
-    strb    r3, [r0]
-strcpy_unalign_return:
-    m_pop
-
-strcpy_unalign7_copy6bytes:
-    str     r2, [r0], #4
-    strb    r3, [r0], #1
-    strb    r4, [r0], #1
-    m_pop
-
-strcpy_unalign7_copy7bytes:
-    str     r2, [r0], #4
-    strb    r3, [r0], #1
-    strb    r4, [r0], #1
-    strb    r5, [r0], #1
-    m_pop
-
-    .p2align 2
-    // Can read 6 bytes before possibly crossing a page.
-strcpy_unalign6:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldrb    r4, [r1]
-    cbz     r4, strcpy_unalign_copy5bytes
-    ldrb    r5, [r1, #1]
-    cbz     r5, strcpy_unalign_copy6bytes
-
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    tst     r3, #0xff0000
-    beq     strcpy_copy7bytes
-    lsrs    ip, r3, #24
-    strd    r2, r3, [r0], #8
-    beq     strcpy_unalign_return
-    b       strcpy_unalign6
-
-    .p2align 2
-    // Can read 5 bytes before possibly crossing a page.
-strcpy_unalign5:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldrb    r4, [r1]
-    cbz     r4, strcpy_unalign_copy5bytes
-
-    ldr     r3, [r1], #4
-
-    pld     [r1, #64]
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    strd    r2, r3, [r0], #8
-    b       strcpy_unalign5
-
-strcpy_unalign_copy5bytes:
-    str     r2, [r0], #4
-    strb    r4, [r0]
-    m_pop
-
-strcpy_unalign_copy6bytes:
-    str     r2, [r0], #4
-    strb    r4, [r0], #1
-    strb    r5, [r0]
-    m_pop
-
-    .p2align 2
-    // Can read 4 bytes before possibly crossing a page.
-strcpy_unalign4:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    strd    r2, r3, [r0], #8
-    b       strcpy_unalign4
-
-    .p2align 2
-    // Can read 3 bytes before possibly crossing a page.
-strcpy_unalign3:
-    ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign3_copy1byte
-    ldrb    r3, [r1, #1]
-    cbz     r3, strcpy_unalign3_copy2bytes
-    ldrb    r4, [r1, #2]
-    cbz     r4, strcpy_unalign3_copy3bytes
-
-    ldr     r2, [r1], #4
-    ldr     r3, [r1], #4
-
-    pld     [r1, #64]
-
-    lsrs    lr, r2, #24
-    beq     strcpy_copy4bytes
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    strd    r2, r3, [r0], #8
-    b       strcpy_unalign3
-
-strcpy_unalign3_copy1byte:
-    strb    r2, [r0]
-    m_pop
-
-strcpy_unalign3_copy2bytes:
-    strb    r2, [r0], #1
-    strb    r3, [r0]
-    m_pop
-
-strcpy_unalign3_copy3bytes:
-    strb    r2, [r0], #1
-    strb    r3, [r0], #1
-    strb    r4, [r0]
-    m_pop
-
-    .p2align 2
-    // Can read 2 bytes before possibly crossing a page.
-strcpy_unalign2:
-    ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign_copy1byte
-    ldrb    r4, [r1, #1]
-    cbz     r4, strcpy_unalign_copy2bytes
-
-    ldr     r2, [r1], #4
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    tst     r2, #0xff0000
-    beq     strcpy_copy3bytes
-    lsrs    ip, r2, #24
-    beq     strcpy_copy4bytes
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    strd    r2, r3, [r0], #8
-    b       strcpy_unalign2
-
-    .p2align 2
-    // Can read 1 byte before possibly crossing a page.
-strcpy_unalign1:
-    ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign_copy1byte
-
-    ldr     r2, [r1], #4
-    ldr     r3, [r1], #4
-
-    pld     [r1, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    strd    r2, r3, [r0], #8
-    b       strcpy_unalign1
-
-strcpy_unalign_copy1byte:
-    strb    r2, [r0]
-    m_pop
-
-strcpy_unalign_copy2bytes:
-    strb    r2, [r0], #1
-    strb    r4, [r0]
-    m_pop
-END(strcpy)
+#define STRCPY
+#include "string_copy.S"

libc/arch-arm/cortex-a15/bionic/string_copy.S

@@ -0,0 +1,520 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+/*
+ * Copyright (c) 2013 ARM Ltd
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. The name of the company may not be used to endorse or promote
+ *    products derived from this software without specific prior written
+ *    permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#if !defined(STPCPY) && !defined(STRCPY)
+#error "Either STPCPY or STRCPY must be defined."
+#endif
+
+#include <private/bionic_asm.h>
+
+    .syntax unified
+
+    .thumb
+    .thumb_func
+
+#if defined(STPCPY)
+    .macro m_push
+    push    {r4, r5, lr}
+    .cfi_def_cfa_offset 12
+    .cfi_rel_offset r4, 0
+    .cfi_rel_offset r5, 4
+    .cfi_rel_offset lr, 8
+    .endm // m_push
+#else
+    .macro m_push
+    push    {r0, r4, r5, lr}
+    .cfi_def_cfa_offset 16
+    .cfi_rel_offset r0, 0
+    .cfi_rel_offset r4, 4
+    .cfi_rel_offset r5, 8
+    .cfi_rel_offset lr, 12
+    .endm // m_push
+#endif
+
+#if defined(STPCPY)
+    .macro m_pop
+    pop     {r4, r5, pc}
+    .endm // m_pop
+#else
+    .macro m_pop
+    pop     {r0, r4, r5, pc}
+    .endm // m_pop
+#endif
+
+    .macro m_copy_byte reg, cmd, label
+    ldrb    \reg, [r1], #1
+    strb    \reg, [r0], #1
+    \cmd    \reg, \label
+    .endm // m_copy_byte
+
+#if defined(STPCPY)
+ENTRY(stpcpy)
+#else
+ENTRY(strcpy)
+#endif
+    // For short copies, hard-code checking the first 8 bytes since this
+    // new code doesn't win until after about 8 bytes.
+    m_push
+    m_copy_byte reg=r2, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r3, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r4, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r5, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r2, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r3, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r4, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r5, cmd=cbnz, label=.Lstringcopy_continue
+
+.Lstringcopy_finish:
+#if defined(STPCPY)
+    sub     r0, r0, #1
+#endif
+    m_pop
+
+.Lstringcopy_continue:
+    pld     [r1, #0]
+    ands    r3, r0, #7
+    beq     .Lstringcopy_check_src_align
+
+    // Align to a double word (64 bits).
+    rsb     r3, r3, #8
+    lsls    ip, r3, #31
+    beq     .Lstringcopy_align_to_32
+
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .Lstringcopy_complete
+
+.Lstringcopy_align_to_32:
+    bcc     .Lstringcopy_align_to_64
+
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .Lstringcopy_complete
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .Lstringcopy_complete
+
+.Lstringcopy_align_to_64:
+    tst     r3, #4
+    beq     .Lstringcopy_check_src_align
+    // Read one byte at a time since we don't have any idea about the alignment
+    // of the source and we don't want to read into a different page.
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .Lstringcopy_complete
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .Lstringcopy_complete
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .Lstringcopy_complete
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .Lstringcopy_complete
+
+.Lstringcopy_check_src_align:
+    // At this point dst is aligned to a double word, check if src
+    // is also aligned to a double word.
+    ands    r3, r1, #7
+    bne     .Lstringcopy_unaligned_copy
+
+    .p2align 2
+.Lstringcopy_mainloop:
+    ldrd    r2, r3, [r1], #8
+
+    pld     [r1, #64]
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_second_register
+
+    strd    r2, r3, [r0], #8
+    b       .Lstringcopy_mainloop
+
+.Lstringcopy_complete:
+#if defined(STPCPY)
+    sub     r0, r0, #1
+#endif
+    m_pop
+
+.Lstringcopy_zero_in_first_register:
+    lsls    lr, ip, #17
+    bne     .Lstringcopy_copy1byte
+    bcs     .Lstringcopy_copy2bytes
+    lsls    ip, ip, #1
+    bne     .Lstringcopy_copy3bytes
+
+.Lstringcopy_copy4bytes:
+    // Copy 4 bytes to the destiniation.
+#if defined(STPCPY)
+    str     r2, [r0], #3
+#else
+    str     r2, [r0]
+#endif
+    m_pop
+
+.Lstringcopy_copy1byte:
+    strb    r2, [r0]
+    m_pop
+
+.Lstringcopy_copy2bytes:
+#if defined(STPCPY)
+    strh    r2, [r0], #1
+#else
+    strh    r2, [r0]
+#endif
+    m_pop
+
+.Lstringcopy_copy3bytes:
+    strh    r2, [r0], #2
+    lsr     r2, #16
+    strb    r2, [r0]
+    m_pop
+
+.Lstringcopy_zero_in_second_register:
+    lsls    lr, ip, #17
+    bne     .Lstringcopy_copy5bytes
+    bcs     .Lstringcopy_copy6bytes
+    lsls    ip, ip, #1
+    bne     .Lstringcopy_copy7bytes
+
+    // Copy 8 bytes to the destination.
+    strd    r2, r3, [r0]
+#if defined(STPCPY)
+    add     r0, r0, #7
+#endif
+    m_pop
+
+.Lstringcopy_copy5bytes:
+    str     r2, [r0], #4
+    strb    r3, [r0]
+    m_pop
+
+.Lstringcopy_copy6bytes:
+    str     r2, [r0], #4
+#if defined(STPCPY)
+    strh    r3, [r0], #1
+#else
+    strh    r3, [r0]
+#endif
+    m_pop
+
+.Lstringcopy_copy7bytes:
+    str     r2, [r0], #4
+    strh    r3, [r0], #2
+    lsr     r3, #16
+    strb    r3, [r0]
+    m_pop
+
+.Lstringcopy_unaligned_copy:
+    // Dst is aligned to a double word, while src is at an unknown alignment.
+    // There are 7 different versions of the unaligned copy code
+    // to prevent overreading the src. The mainloop of every single version
+    // will store 64 bits per loop. The difference is how much of src can
+    // be read without potentially crossing a page boundary.
+    tbb     [pc, r3]
+.Lstringcopy_unaligned_branchtable:
+    .byte 0
+    .byte ((.Lstringcopy_unalign7 - .Lstringcopy_unaligned_branchtable)/2)
+    .byte ((.Lstringcopy_unalign6 - .Lstringcopy_unaligned_branchtable)/2)
+    .byte ((.Lstringcopy_unalign5 - .Lstringcopy_unaligned_branchtable)/2)
+    .byte ((.Lstringcopy_unalign4 - .Lstringcopy_unaligned_branchtable)/2)
+    .byte ((.Lstringcopy_unalign3 - .Lstringcopy_unaligned_branchtable)/2)
+    .byte ((.Lstringcopy_unalign2 - .Lstringcopy_unaligned_branchtable)/2)
+    .byte ((.Lstringcopy_unalign1 - .Lstringcopy_unaligned_branchtable)/2)
+
+    .p2align 2
+    // Can read 7 bytes before possibly crossing a page.
+.Lstringcopy_unalign7:
+    ldr     r2, [r1], #4
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+
+    ldrb    r3, [r1]
+    cbz     r3, .Lstringcopy_unalign7_copy5bytes
+    ldrb    r4, [r1, #1]
+    cbz     r4, .Lstringcopy_unalign7_copy6bytes
+    ldrb    r5, [r1, #2]
+    cbz     r5, .Lstringcopy_unalign7_copy7bytes
+
+    ldr     r3, [r1], #4
+    pld     [r1, #64]
+
+    lsrs    ip, r3, #24
+    strd    r2, r3, [r0], #8
+#if defined(STPCPY)
+    beq     .Lstringcopy_finish
+#else
+    beq     .Lstringcopy_unalign_return
+#endif
+    b       .Lstringcopy_unalign7
+
+.Lstringcopy_unalign7_copy5bytes:
+    str     r2, [r0], #4
+    strb    r3, [r0]
+.Lstringcopy_unalign_return:
+    m_pop
+
+.Lstringcopy_unalign7_copy6bytes:
+    str     r2, [r0], #4
+    strb    r3, [r0], #1
+    strb    r4, [r0]
+    m_pop
+
+.Lstringcopy_unalign7_copy7bytes:
+    str     r2, [r0], #4
+    strb    r3, [r0], #1
+    strb    r4, [r0], #1
+    strb    r5, [r0]
+    m_pop
+
+    .p2align 2
+    // Can read 6 bytes before possibly crossing a page.
+.Lstringcopy_unalign6:
+    ldr     r2, [r1], #4
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+
+    ldrb    r4, [r1]
+    cbz     r4, .Lstringcopy_unalign_copy5bytes
+    ldrb    r5, [r1, #1]
+    cbz     r5, .Lstringcopy_unalign_copy6bytes
+
+    ldr     r3, [r1], #4
+    pld     [r1, #64]
+
+    tst     r3, #0xff0000
+    beq     .Lstringcopy_copy7bytes
+    lsrs    ip, r3, #24
+    strd    r2, r3, [r0], #8
+#if defined(STPCPY)
+    beq     .Lstringcopy_finish
+#else
+    beq     .Lstringcopy_unalign_return
+#endif
+    b       .Lstringcopy_unalign6
+
+    .p2align 2
+    // Can read 5 bytes before possibly crossing a page.
+.Lstringcopy_unalign5:
+    ldr     r2, [r1], #4
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+
+    ldrb    r4, [r1]
+    cbz     r4, .Lstringcopy_unalign_copy5bytes
+
+    ldr     r3, [r1], #4
+
+    pld     [r1, #64]
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_second_register
+
+    strd    r2, r3, [r0], #8
+    b       .Lstringcopy_unalign5
+
+.Lstringcopy_unalign_copy5bytes:
+    str     r2, [r0], #4
+    strb    r4, [r0]
+    m_pop
+
+.Lstringcopy_unalign_copy6bytes:
+    str     r2, [r0], #4
+    strb    r4, [r0], #1
+    strb    r5, [r0]
+    m_pop
+
+    .p2align 2
+    // Can read 4 bytes before possibly crossing a page.
+.Lstringcopy_unalign4:
+    ldr     r2, [r1], #4
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+
+    ldr     r3, [r1], #4
+    pld     [r1, #64]
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_second_register
+
+    strd    r2, r3, [r0], #8
+    b       .Lstringcopy_unalign4
+
+    .p2align 2
+    // Can read 3 bytes before possibly crossing a page.
+.Lstringcopy_unalign3:
+    ldrb    r2, [r1]
+    cbz     r2, .Lstringcopy_unalign3_copy1byte
+    ldrb    r3, [r1, #1]
+    cbz     r3, .Lstringcopy_unalign3_copy2bytes
+    ldrb    r4, [r1, #2]
+    cbz     r4, .Lstringcopy_unalign3_copy3bytes
+
+    ldr     r2, [r1], #4
+    ldr     r3, [r1], #4
+
+    pld     [r1, #64]
+
+    lsrs    lr, r2, #24
+    beq     .Lstringcopy_copy4bytes
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_second_register
+
+    strd    r2, r3, [r0], #8
+    b       .Lstringcopy_unalign3
+
+.Lstringcopy_unalign3_copy1byte:
+    strb    r2, [r0]
+    m_pop
+
+.Lstringcopy_unalign3_copy2bytes:
+    strb    r2, [r0], #1
+    strb    r3, [r0]
+    m_pop
+
+.Lstringcopy_unalign3_copy3bytes:
+    strb    r2, [r0], #1
+    strb    r3, [r0], #1
+    strb    r4, [r0]
+    m_pop
+
+    .p2align 2
+    // Can read 2 bytes before possibly crossing a page.
+.Lstringcopy_unalign2:
+    ldrb    r2, [r1]
+    cbz     r2, .Lstringcopy_unalign_copy1byte
+    ldrb    r4, [r1, #1]
+    cbz     r4, .Lstringcopy_unalign_copy2bytes
+
+    ldr     r2, [r1], #4
+    ldr     r3, [r1], #4
+    pld     [r1, #64]
+
+    tst     r2, #0xff0000
+    beq     .Lstringcopy_copy3bytes
+    lsrs    ip, r2, #24
+    beq     .Lstringcopy_copy4bytes
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_second_register
+
+    strd    r2, r3, [r0], #8
+    b       .Lstringcopy_unalign2
+
+    .p2align 2
+    // Can read 1 byte before possibly crossing a page.
+.Lstringcopy_unalign1:
+    ldrb    r2, [r1]
+    cbz     r2, .Lstringcopy_unalign_copy1byte
+
+    ldr     r2, [r1], #4
+    ldr     r3, [r1], #4
+
+    pld     [r1, #64]
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_second_register
+
+    strd    r2, r3, [r0], #8
+    b       .Lstringcopy_unalign1
+
+.Lstringcopy_unalign_copy1byte:
+    strb    r2, [r0]
+    m_pop
+
+.Lstringcopy_unalign_copy2bytes:
+    strb    r2, [r0], #1
+    strb    r4, [r0]
+    m_pop
+#if defined(STPCPY)
+END(stpcpy)
+#else
+END(strcpy)
+#endif

libc/arch-arm/cortex-a15/bionic/strlen.S

@@ -53,7 +53,7 @@
  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
+#include <private/bionic_asm.h>
 
     .syntax unified
 
@@ -65,38 +65,38 @@ ENTRY(strlen)
     mov     r1, r0
 
     ands    r3, r0, #7
-    beq     mainloop
+    beq     .L_mainloop
 
     // Align to a double word (64 bits).
     rsb     r3, r3, #8
     lsls    ip, r3, #31
-    beq     align_to_32
+    beq     .L_align_to_32
 
     ldrb    r2, [r1], #1
-    cbz     r2, update_count_and_return
+    cbz     r2, .L_update_count_and_return
 
-align_to_32:
-    bcc     align_to_64
+.L_align_to_32:
+    bcc     .L_align_to_64
     ands    ip, r3, #2
-    beq     align_to_64
+    beq     .L_align_to_64
 
     ldrb    r2, [r1], #1
-    cbz     r2, update_count_and_return
+    cbz     r2, .L_update_count_and_return
     ldrb    r2, [r1], #1
-    cbz     r2, update_count_and_return
+    cbz     r2, .L_update_count_and_return
 
-align_to_64:
+.L_align_to_64:
     tst     r3, #4
-    beq     mainloop
+    beq     .L_mainloop
     ldr     r3, [r1], #4
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     zero_in_second_register
+    bne     .L_zero_in_second_register
 
     .p2align 2
-mainloop:
+.L_mainloop:
     ldrd    r2, r3, [r1], #8
 
     pld     [r1, #64]
@@ -104,62 +104,62 @@ mainloop:
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     zero_in_first_register
+    bne     .L_zero_in_first_register
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     zero_in_second_register
-    b       mainloop
+    bne     .L_zero_in_second_register
+    b       .L_mainloop
 
-update_count_and_return:
+.L_update_count_and_return:
     sub     r0, r1, r0
     sub     r0, r0, #1
     bx      lr
 
-zero_in_first_register:
+.L_zero_in_first_register:
     sub     r0, r1, r0
     lsls    r3, ip, #17
-    bne     sub8_and_return
-    bcs     sub7_and_return
+    bne     .L_sub8_and_return
+    bcs     .L_sub7_and_return
     lsls    ip, ip, #1
-    bne     sub6_and_return
+    bne     .L_sub6_and_return
 
     sub     r0, r0, #5
     bx      lr
 
-sub8_and_return:
+.L_sub8_and_return:
     sub     r0, r0, #8
     bx      lr
 
-sub7_and_return:
+.L_sub7_and_return:
     sub     r0, r0, #7
     bx      lr
 
-sub6_and_return:
+.L_sub6_and_return:
     sub     r0, r0, #6
     bx      lr
 
-zero_in_second_register:
+.L_zero_in_second_register:
     sub     r0, r1, r0
     lsls    r3, ip, #17
-    bne     sub4_and_return
-    bcs     sub3_and_return
+    bne     .L_sub4_and_return
+    bcs     .L_sub3_and_return
     lsls    ip, ip, #1
-    bne     sub2_and_return
+    bne     .L_sub2_and_return
 
     sub     r0, r0, #1
     bx      lr
 
-sub4_and_return:
+.L_sub4_and_return:
     sub     r0, r0, #4
     bx      lr
 
-sub3_and_return:
+.L_sub3_and_return:
     sub     r0, r0, #3
     bx      lr
 
-sub2_and_return:
+.L_sub2_and_return:
     sub     r0, r0, #2
     bx      lr
 END(strlen)

libc/arch-arm/cortex-a15/cortex-a15.mk

@@ -1,10 +1,27 @@
-$(call libc-add-cpu-variant-src,MEMCPY,arch-arm/cortex-a15/bionic/memcpy.S)
-$(call libc-add-cpu-variant-src,MEMSET,arch-arm/cortex-a15/bionic/memset.S)
-$(call libc-add-cpu-variant-src,STRCAT,arch-arm/cortex-a15/bionic/strcat.S)
-$(call libc-add-cpu-variant-src,STRCMP,arch-arm/cortex-a15/bionic/strcmp.S)
-$(call libc-add-cpu-variant-src,STRCPY,arch-arm/cortex-a15/bionic/strcpy.S)
-$(call libc-add-cpu-variant-src,STRLEN,arch-arm/cortex-a15/bionic/strlen.S)
-$(call libc-add-cpu-variant-src,__STRCAT_CHK,arch-arm/cortex-a15/bionic/__strcat_chk.S)
-$(call libc-add-cpu-variant-src,__STRCPY_CHK,arch-arm/cortex-a15/bionic/__strcpy_chk.S)
+libc_openbsd_src_files_exclude_arm += \
+    upstream-openbsd/lib/libc/string/memmove.c \
+    upstream-openbsd/lib/libc/string/stpcpy.c \
+    upstream-openbsd/lib/libc/string/strcat.c \
 
-include bionic/libc/arch-arm/generic/generic.mk
+libc_bionic_src_files_exclude_arm += \
+    arch-arm/generic/bionic/memcpy.S \
+    arch-arm/generic/bionic/memset.S \
+    arch-arm/generic/bionic/strcmp.S \
+    arch-arm/generic/bionic/strcpy.S \
+    arch-arm/generic/bionic/strlen.c \
+    bionic/__strcat_chk.cpp \
+    bionic/__strcpy_chk.cpp \
+
+libc_bionic_src_files_arm += \
+    arch-arm/cortex-a15/bionic/memcpy.S \
+    arch-arm/cortex-a15/bionic/memset.S \
+    arch-arm/cortex-a15/bionic/stpcpy.S \
+    arch-arm/cortex-a15/bionic/strcat.S \
+    arch-arm/cortex-a15/bionic/__strcat_chk.S \
+    arch-arm/cortex-a15/bionic/strcmp.S \
+    arch-arm/cortex-a15/bionic/strcpy.S \
+    arch-arm/cortex-a15/bionic/__strcpy_chk.S \
+    arch-arm/cortex-a15/bionic/strlen.S \
+
+libc_bionic_src_files_arm += \
+    arch-arm/denver/bionic/memmove.S \

libc/arch-arm/cortex-a53.a57/cortex-a53.a57.mk

@@ -0,0 +1,32 @@
+# This file represents the best optimized routines that are the middle
+# ground when running on a big/little system that is cortex-a57/cortex-a53.
+# The cortex-a7 optimized routines, and the cortex-a53 optimized routines
+# decrease performance on cortex-a57 processors by as much as 20%.
+
+libc_openbsd_src_files_exclude_arm += \
+    upstream-openbsd/lib/libc/string/memmove.c \
+    upstream-openbsd/lib/libc/string/stpcpy.c \
+    upstream-openbsd/lib/libc/string/strcat.c \
+
+libc_bionic_src_files_exclude_arm += \
+    arch-arm/generic/bionic/memcpy.S \
+    arch-arm/generic/bionic/memset.S \
+    arch-arm/generic/bionic/strcmp.S \
+    arch-arm/generic/bionic/strcpy.S \
+    arch-arm/generic/bionic/strlen.c \
+    bionic/__strcat_chk.cpp \
+    bionic/__strcpy_chk.cpp \
+
+libc_bionic_src_files_arm += \
+    arch-arm/cortex-a15/bionic/memcpy.S \
+    arch-arm/cortex-a15/bionic/memset.S \
+    arch-arm/cortex-a15/bionic/stpcpy.S \
+    arch-arm/cortex-a15/bionic/strcat.S \
+    arch-arm/cortex-a15/bionic/__strcat_chk.S \
+    arch-arm/cortex-a15/bionic/strcmp.S \
+    arch-arm/cortex-a15/bionic/strcpy.S \
+    arch-arm/cortex-a15/bionic/__strcpy_chk.S \
+    arch-arm/cortex-a15/bionic/strlen.S \
+
+libc_bionic_src_files_arm += \
+    arch-arm/denver/bionic/memmove.S \

libc/arch-arm/cortex-a53/bionic/__strcat_chk.S

@@ -0,0 +1,32 @@
+/*
+ * Copyright (C) 2015 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+// Indicate which memcpy base file to include.
+#define MEMCPY_BASE "arch-arm/cortex-a53/bionic/memcpy_base.S"
+
+#include "arch-arm/cortex-a15/bionic/__strcat_chk_common.S"

libc/arch-arm/cortex-a53/bionic/__strcpy_chk.S

@@ -0,0 +1,32 @@
+/*
+ * Copyright (C) 2015 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+// Indicate which memcpy base file to include.
+#define MEMCPY_BASE "arch-arm/cortex-a53/bionic/memcpy_base.S"
+
+#include "arch-arm/cortex-a15/bionic/__strcpy_chk_common.S"

libc/arch-arm/cortex-a53/bionic/memcpy.S

@@ -0,0 +1,32 @@
+/*
+ * Copyright (C) 2015 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+// Indicate which memcpy base file to include.
+#define MEMCPY_BASE "arch-arm/cortex-a53/bionic/memcpy_base.S"
+
+#include "arch-arm/cortex-a15/bionic/memcpy_common.S"

libc/arch-arm/cortex-a53/bionic/memcpy_base.S

@@ -0,0 +1,143 @@
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+/*
+ * Copyright (c) 2013 ARM Ltd
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. The name of the company may not be used to endorse or promote
+ *    products derived from this software without specific prior written
+ *    permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+.L_memcpy_base:
+        // Assumes that n >= 0, and dst, src are valid pointers.
+        cmp     r2, #16
+        blo     .L_copy_less_than_16_unknown_align
+
+.L_copy_unknown_alignment:
+        // Unknown alignment of src and dst.
+        // Assumes that the first few bytes have already been prefetched.
+
+        // Align destination to 128 bits. The mainloop store instructions
+        // require this alignment or they will throw an exception.
+        rsb         r3, r0, #0
+        ands        r3, r3, #0xF
+        beq         2f
+
+        // Copy up to 15 bytes (count in r3).
+        sub         r2, r2, r3
+        movs        ip, r3, lsl #31
+
+        itt         mi
+        ldrbmi      lr, [r1], #1
+        strbmi      lr, [r0], #1
+        itttt       cs
+        ldrbcs      ip, [r1], #1
+        ldrbcs      lr, [r1], #1
+        strbcs      ip, [r0], #1
+        strbcs      lr, [r0], #1
+
+        movs        ip, r3, lsl #29
+        bge         1f
+        // Copies 4 bytes, dst 32 bits aligned before, at least 64 bits after.
+        vld4.8      {d0[0], d1[0], d2[0], d3[0]}, [r1]!
+        vst4.8      {d0[0], d1[0], d2[0], d3[0]}, [r0, :32]!
+1:      bcc         2f
+        // Copies 8 bytes, dst 64 bits aligned before, at least 128 bits after.
+        vld1.8      {d0}, [r1]!
+        vst1.8      {d0}, [r0, :64]!
+
+2:      // Make sure we have at least 64 bytes to copy.
+        subs        r2, r2, #64
+        blo         2f
+
+1:      // The main loop copies 64 bytes at a time.
+        vld1.8      {d0  - d3},   [r1]!
+        vld1.8      {d4  - d7},   [r1]!
+        subs        r2, r2, #64
+        vstmia      r0!, {d0 - d7}
+        pld         [r1, #(64*10)]
+        bhs         1b
+
+2:      // Fix-up the remaining count and make sure we have >= 32 bytes left.
+        adds        r2, r2, #32
+        blo         3f
+
+        // 32 bytes. These cache lines were already preloaded.
+        vld1.8      {d0 - d3},  [r1]!
+        sub         r2, r2, #32
+        vst1.8      {d0 - d3},  [r0, :128]!
+3:      // Less than 32 left.
+        add         r2, r2, #32
+        tst         r2, #0x10
+        beq         .L_copy_less_than_16_unknown_align
+        // Copies 16 bytes, destination 128 bits aligned.
+        vld1.8      {d0, d1}, [r1]!
+        vst1.8      {d0, d1}, [r0, :128]!
+
+.L_copy_less_than_16_unknown_align:
+        // Copy up to 15 bytes (count in r2).
+        movs        ip, r2, lsl #29
+        bcc         1f
+        vld1.8      {d0}, [r1]!
+        vst1.8      {d0}, [r0]!
+1:      bge         2f
+        vld4.8      {d0[0], d1[0], d2[0], d3[0]}, [r1]!
+        vst4.8      {d0[0], d1[0], d2[0], d3[0]}, [r0]!
+
+2:      // Copy 0 to 4 bytes.
+        lsls        r2, r2, #31
+        itt         ne
+        ldrbne      lr, [r1], #1
+        strbne      lr, [r0], #1
+        itttt       cs
+        ldrbcs      ip, [r1], #1
+        ldrbcs      lr, [r1]
+        strbcs      ip, [r0], #1
+        strbcs      lr, [r0]
+
+        pop         {r0, pc}

libc/arch-arm/cortex-a53/cortex-a53.mk

@@ -0,0 +1,31 @@
+libc_openbsd_src_files_exclude_arm += \
+    upstream-openbsd/lib/libc/string/memmove.c \
+    upstream-openbsd/lib/libc/string/stpcpy.c \
+    upstream-openbsd/lib/libc/string/strcat.c \
+
+libc_bionic_src_files_exclude_arm += \
+    arch-arm/generic/bionic/memcpy.S \
+    arch-arm/generic/bionic/memset.S \
+    arch-arm/generic/bionic/strcmp.S \
+    arch-arm/generic/bionic/strcpy.S \
+    arch-arm/generic/bionic/strlen.c \
+    bionic/__strcat_chk.cpp \
+    bionic/__strcpy_chk.cpp \
+
+libc_bionic_src_files_arm += \
+    arch-arm/cortex-a53/bionic/memcpy.S \
+    arch-arm/cortex-a53/bionic/__strcat_chk.S \
+    arch-arm/cortex-a53/bionic/__strcpy_chk.S \
+
+libc_bionic_src_files_arm += \
+    arch-arm/cortex-a7/bionic/memset.S \
+
+libc_bionic_src_files_arm += \
+    arch-arm/cortex-a15/bionic/stpcpy.S \
+    arch-arm/cortex-a15/bionic/strcat.S \
+    arch-arm/cortex-a15/bionic/strcmp.S \
+    arch-arm/cortex-a15/bionic/strcpy.S \
+    arch-arm/cortex-a15/bionic/strlen.S \
+
+libc_bionic_src_files_arm += \
+    arch-arm/denver/bionic/memmove.S \

libc/arch-arm/cortex-a7/bionic/memset.S

@@ -0,0 +1,180 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <machine/cpu-features.h>
+#include <private/bionic_asm.h>
+#include <private/libc_events.h>
+
+        /*
+         * Optimized memset() for ARM.
+         *
+         * memset() returns its first argument.
+         */
+
+        .fpu        neon
+        .syntax     unified
+
+ENTRY(__memset_chk)
+        cmp         r2, r3
+        bls         .L_done
+
+        // Preserve lr for backtrace.
+        push        {lr}
+        .cfi_def_cfa_offset 4
+        .cfi_rel_offset lr, 0
+
+        ldr         r0, error_message
+        ldr         r1, error_code
+1:
+        add         r0, pc
+        bl          __fortify_chk_fail
+error_code:
+        .word       BIONIC_EVENT_MEMSET_BUFFER_OVERFLOW
+error_message:
+        .word       error_string-(1b+8)
+END(__memset_chk)
+
+ENTRY(bzero)
+        mov         r2, r1
+        mov         r1, #0
+.L_done:
+        // Fall through to memset...
+END(bzero)
+
+ENTRY(memset)
+        mov         r3, r0
+        // At this point only d0, d1 are going to be used below.
+        vdup.8      q0, r1
+        cmp         r2, #16
+        blo         .L_set_less_than_16_unknown_align
+
+.L_check_alignment:
+        // Align destination to a double word to avoid the store crossing
+        // a cache line boundary.
+        ands        ip, r3, #7
+        bne         .L_do_double_word_align
+
+.L_double_word_aligned:
+        // Duplicate since the less than 64 can use d2, d3.
+        vmov        q1, q0
+        subs        r2, #64
+        blo         .L_set_less_than_64
+
+        // Duplicate the copy value so that we can store 64 bytes at a time.
+        vmov        q2, q0
+        vmov        q3, q0
+
+1:      // Main loop stores 64 bytes at a time.
+        subs        r2, #64
+        vstmia      r3!, {d0 - d7}
+        bge         1b
+
+.L_set_less_than_64:
+        // Restore r2 to the count of bytes left to set.
+        add         r2, #64
+        lsls        ip, r2, #27
+        bcc         .L_set_less_than_32
+        // Set 32 bytes.
+        vstmia      r3!, {d0 - d3}
+
+.L_set_less_than_32:
+        bpl         .L_set_less_than_16
+        // Set 16 bytes.
+        vstmia      r3!, {d0, d1}
+
+.L_set_less_than_16:
+        // Less than 16 bytes to set.
+        lsls        ip, r2, #29
+        bcc         .L_set_less_than_8
+
+        // Set 8 bytes.
+        vstmia      r3!, {d0}
+
+.L_set_less_than_8:
+        bpl         .L_set_less_than_4
+        // Set 4 bytes
+        vst1.32     {d0[0]}, [r3]!
+
+.L_set_less_than_4:
+        lsls        ip, r2, #31
+        it          ne
+        strbne      r1, [r3], #1
+        itt         cs
+        strbcs      r1, [r3], #1
+        strbcs      r1, [r3]
+        bx          lr
+
+.L_do_double_word_align:
+        rsb         ip, ip, #8
+        sub         r2, r2, ip
+
+        // Do this comparison now, otherwise we'll need to save a
+        // register to the stack since we've used all available
+        // registers.
+        cmp         ip, #4
+        blo         1f
+
+        // Need to do a four byte copy.
+        movs        ip, ip, lsl #31
+        it          mi
+        strbmi      r1, [r3], #1
+        itt         cs
+        strbcs      r1, [r3], #1
+        strbcs      r1, [r3], #1
+        vst1.32     {d0[0]}, [r3]!
+        b           .L_double_word_aligned
+
+1:
+        // No four byte copy.
+        movs        ip, ip, lsl #31
+        it          mi
+        strbmi      r1, [r3], #1
+        itt         cs
+        strbcs      r1, [r3], #1
+        strbcs      r1, [r3], #1
+        b           .L_double_word_aligned
+
+.L_set_less_than_16_unknown_align:
+        // Set up to 15 bytes.
+        movs        ip, r2, lsl #29
+        bcc         1f
+        vst1.8      {d0}, [r3]!
+1:      bge         2f
+        vst1.32     {d0[0]}, [r3]!
+2:      movs        ip, r2, lsl #31
+        it          mi
+        strbmi      r1, [r3], #1
+        itt         cs
+        strbcs      r1, [r3], #1
+        strbcs      r1, [r3], #1
+        bx          lr
+END(memset)
+
+        .data
+error_string:
+        .string     "memset: prevented write past end of buffer"

libc/arch-arm/cortex-a7/cortex-a7.mk

@@ -1 +1,29 @@
-include bionic/libc/arch-arm/cortex-a15/cortex-a15.mk
+libc_openbsd_src_files_exclude_arm += \
+    upstream-openbsd/lib/libc/string/memmove.c \
+    upstream-openbsd/lib/libc/string/stpcpy.c \
+    upstream-openbsd/lib/libc/string/strcat.c \
+
+libc_bionic_src_files_exclude_arm += \
+    arch-arm/generic/bionic/memcpy.S \
+    arch-arm/generic/bionic/memset.S \
+    arch-arm/generic/bionic/strcmp.S \
+    arch-arm/generic/bionic/strcpy.S \
+    arch-arm/generic/bionic/strlen.c \
+    bionic/__strcat_chk.cpp \
+    bionic/__strcpy_chk.cpp \
+
+libc_bionic_src_files_arm += \
+    arch-arm/cortex-a7/bionic/memset.S \
+
+libc_bionic_src_files_arm += \
+    arch-arm/cortex-a15/bionic/memcpy.S \
+    arch-arm/cortex-a15/bionic/stpcpy.S \
+    arch-arm/cortex-a15/bionic/strcat.S \
+    arch-arm/cortex-a15/bionic/__strcat_chk.S \
+    arch-arm/cortex-a15/bionic/strcmp.S \
+    arch-arm/cortex-a15/bionic/strcpy.S \
+    arch-arm/cortex-a15/bionic/__strcpy_chk.S \
+    arch-arm/cortex-a15/bionic/strlen.S \
+
+libc_bionic_src_files_arm += \
+    arch-arm/denver/bionic/memmove.S \

libc/arch-arm/cortex-a9/bionic/__strcat_chk.S

@@ -26,8 +26,8 @@
  * SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
-#include "libc_events.h"
+#include <private/bionic_asm.h>
+#include <private/libc_events.h>
 
     .syntax unified
     .fpu    neon
@@ -38,15 +38,12 @@
 // Check that the two lengths together don't exceed the threshold, then
 // do a memcpy of the data.
 ENTRY(__strcat_chk)
-    .cfi_startproc
     pld     [r0, #0]
     push    {r0, lr}
-    .save   {r0, lr}
     .cfi_def_cfa_offset 8
     .cfi_rel_offset r0, 0
     .cfi_rel_offset lr, 4
     push    {r4, r5}
-    .save   {r4, r5}
     .cfi_adjust_cfa_offset 8
     .cfi_rel_offset r4, 0
     .cfi_rel_offset r5, 4
@@ -193,18 +190,13 @@ ENTRY(__strcat_chk)
     pop     {r4, r5}
 
     // Fall through into the memcpy_base function.
-    .cfi_endproc
 END(__strcat_chk)
 
 #define MEMCPY_BASE         __strcat_chk_memcpy_base
 #define MEMCPY_BASE_ALIGNED __strcat_chk_memcpy_base_aligned
 #include "memcpy_base.S"
 
-ENTRY(__strcat_chk_fail)
-    .cfi_startproc
-
-    .save   {r0, lr}
-    .save   {r4, r5}
+ENTRY_PRIVATE(__strcat_chk_fail)
     .cfi_def_cfa_offset 8
     .cfi_rel_offset r0, 0
     .cfi_rel_offset lr, 4
@@ -221,10 +213,8 @@ error_code:
     .word   BIONIC_EVENT_STRCAT_BUFFER_OVERFLOW
 error_message:
     .word   error_string-(1b+4)
-
-    .cfi_endproc
 END(__strcat_chk_fail)
 
     .data
 error_string:
-    .string "strcat buffer overflow"
+    .string "strcat: prevented write past end of buffer"

libc/arch-arm/cortex-a9/bionic/__strcpy_chk.S

@@ -26,8 +26,8 @@
  * SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
-#include "libc_events.h"
+#include <private/bionic_asm.h>
+#include <private/libc_events.h>
 
     .syntax unified
     .fpu    neon
@@ -37,10 +37,8 @@
 // Get the length of the source string first, then do a memcpy of the data
 // instead of a strcpy.
 ENTRY(__strcpy_chk)
-    .cfi_startproc
     pld     [r0, #0]
     push    {r0, lr}
-    .save   {r0, lr}
     .cfi_def_cfa_offset 8
     .cfi_rel_offset r0, 0
     .cfi_rel_offset lr, 4
@@ -158,8 +156,6 @@ ENTRY(__strcpy_chk)
     // Add 1 for copy length to get the string terminator.
     add     r2, r3, #1
 
-    .cfi_endproc
-
     // Fall through into the memcpy_base function.
 END(__strcpy_chk)
 
@@ -167,10 +163,7 @@ END(__strcpy_chk)
 #define MEMCPY_BASE_ALIGNED __strcpy_chk_memcpy_base_aligned
 #include "memcpy_base.S"
 
-ENTRY(__strcpy_chk_fail)
-    .cfi_startproc
-
-    .save   {r0, lr}
+ENTRY_PRIVATE(__strcpy_chk_fail)
     .cfi_def_cfa_offset 8
     .cfi_rel_offset r0, 0
     .cfi_rel_offset lr, 4
@@ -185,10 +178,8 @@ error_code:
     .word   BIONIC_EVENT_STRCPY_BUFFER_OVERFLOW
 error_message:
     .word   error_string-(1b+4)
-
-    .cfi_endproc
 END(__strcpy_chk_fail)
 
     .data
 error_string:
-    .string "strcpy buffer overflow"
+    .string "strcpy: prevented write past end of buffer"

libc/arch-arm/cortex-a9/bionic/memcpy.S

@@ -26,8 +26,8 @@
  * SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
-#include "libc_events.h"
+#include <private/bionic_asm.h>
+#include <private/libc_events.h>
 
 /*
  * This code assumes it is running on a processor that supports all arm v7
@@ -41,37 +41,28 @@
         .thumb_func
 
 ENTRY(__memcpy_chk)
-        .cfi_startproc
         cmp         r2, r3
         bhi         __memcpy_chk_fail
 
         // Fall through to memcpy...
-        .cfi_endproc
 END(__memcpy_chk)
 
 ENTRY(memcpy)
-        .cfi_startproc
-
         pld     [r1, #0]
         stmfd   sp!, {r0, lr}
-        .save   {r0, lr}
         .cfi_def_cfa_offset 8
         .cfi_rel_offset r0, 0
         .cfi_rel_offset lr, 4
         pld     [r1, #64]
-
-        .cfi_endproc
 END(memcpy)
 
 #define MEMCPY_BASE         __memcpy_base
 #define MEMCPY_BASE_ALIGNED __memcpy_base_aligned
 #include "memcpy_base.S"
 
-ENTRY(__memcpy_chk_fail)
-        .cfi_startproc
+ENTRY_PRIVATE(__memcpy_chk_fail)
         // Preserve lr for backtrace.
         push    {lr}
-        .save   {lr}
         .cfi_def_cfa_offset 4
         .cfi_rel_offset lr, 0
 
@@ -84,9 +75,8 @@ error_code:
         .word   BIONIC_EVENT_MEMCPY_BUFFER_OVERFLOW
 error_message:
         .word   error_string-(1b+4)
-        .cfi_endproc
 END(__memcpy_chk_fail)
 
         .data
 error_string:
-        .string     "memcpy buffer overflow"
+        .string     "memcpy: prevented write past end of buffer"

libc/arch-arm/cortex-a9/bionic/memcpy_base.S

@@ -32,9 +32,7 @@
  * cache line.
  */
 
-ENTRY(MEMCPY_BASE)
-        .cfi_startproc
-        .save       {r0, lr}
+ENTRY_PRIVATE(MEMCPY_BASE)
         .cfi_def_cfa_offset 8
         .cfi_rel_offset r0, 0
         .cfi_rel_offset lr, 4
@@ -46,7 +44,7 @@ ENTRY(MEMCPY_BASE)
         /* check if buffers are aligned. If so, run arm-only version */
         eor         r3, r0, r1
         ands        r3, r3, #0x3
-        beq         __memcpy_base_aligned
+        beq         MEMCPY_BASE_ALIGNED
 
         /* Check the upper size limit for Neon unaligned memory access in memcpy */
         cmp         r2, #224
@@ -135,23 +133,16 @@ ENTRY(MEMCPY_BASE)
         strbcs      ip, [r0], #1
         strbcs      lr, [r0], #1
 
-        ldmfd       sp!, {r0, lr}
-        bx          lr
-
-        .cfi_endproc
+        ldmfd       sp!, {r0, pc}
 END(MEMCPY_BASE)
 
-ENTRY(MEMCPY_BASE_ALIGNED)
-        .cfi_startproc
-
-        .save       {r0, lr}
+ENTRY_PRIVATE(MEMCPY_BASE_ALIGNED)
         .cfi_def_cfa_offset 8
         .cfi_rel_offset r0, 0
         .cfi_rel_offset lr, 4
 
         /* Simple arm-only copy loop to handle aligned copy operations */
         stmfd       sp!, {r4-r8}
-        .save       {r4-r8}
         .cfi_adjust_cfa_offset 20
         .cfi_rel_offset r4, 0
         .cfi_rel_offset r5, 4
@@ -228,6 +219,4 @@ ENTRY(MEMCPY_BASE_ALIGNED)
 6:
         ldmfd       sp!, {r4-r8}
         ldmfd       sp!, {r0, pc}
-
-        .cfi_endproc
 END(MEMCPY_BASE_ALIGNED)

libc/arch-arm/cortex-a9/bionic/memset.S

@@ -26,9 +26,8 @@
  * SUCH DAMAGE.
  */
 
-#include <machine/cpu-features.h>
-#include <machine/asm.h>
-#include "libc_events.h"
+#include <private/bionic_asm.h>
+#include <private/libc_events.h>
 
 /*
  * This code assumes it is running on a processor that supports all arm v7
@@ -36,15 +35,14 @@
  */
 
     .fpu    neon
+    .syntax unified
 
 ENTRY(__memset_chk)
-        .cfi_startproc
         cmp         r2, r3
         bls         .L_done
 
         // Preserve lr for backtrace.
         push        {lr}
-        .save       {lr}
         .cfi_def_cfa_offset 4
         .cfi_rel_offset lr, 0
 
@@ -57,33 +55,23 @@ error_code:
         .word       BIONIC_EVENT_MEMSET_BUFFER_OVERFLOW
 error_message:
         .word       error_string-(1b+8)
-
-        .cfi_endproc
 END(__memset_chk)
 
 ENTRY(bzero)
-        .cfi_startproc
         mov     r2, r1
         mov     r1, #0
 
 .L_done:
         // Fall through to memset...
-        .cfi_endproc
 END(bzero)
 
 /* memset() returns its first argument.  */
 ENTRY(memset)
-        .cfi_startproc
-
-        # The neon memset only wins for less than 132.
+        // The neon memset only wins for less than 132.
         cmp         r2, #132
-        bhi         __memset_large_copy
-
-        stmfd       sp!, {r0}
-        .save       {r0}
-        .cfi_def_cfa_offset 4
-        .cfi_rel_offset r0, 0
+        bhi         .L_memset_large_copy
 
+        mov         r3, r0
         vdup.8      q0, r1
 
         /* make sure we have at least 32 bytes to write */
@@ -93,7 +81,7 @@ ENTRY(memset)
 
 1:      /* The main loop writes 32 bytes at a time */
         subs        r2, r2, #32
-        vst1.8      {d0 - d3}, [r0]!
+        vst1.8      {d0 - d3}, [r3]!
         bhs         1b
 
 2:      /* less than 32 left */
@@ -102,31 +90,24 @@ ENTRY(memset)
         beq         3f
 
         // writes 16 bytes, 128-bits aligned
-        vst1.8      {d0, d1}, [r0]!
+        vst1.8      {d0, d1}, [r3]!
 3:      /* write up to 15-bytes (count in r2) */
         movs        ip, r2, lsl #29
         bcc         1f
-        vst1.8      {d0}, [r0]!
+        vst1.8      {d0}, [r3]!
 1:      bge         2f
-        vst1.32     {d0[0]}, [r0]!
+        vst1.32     {d0[0]}, [r3]!
 2:      movs        ip, r2, lsl #31
-        strmib      r1, [r0], #1
-        strcsb      r1, [r0], #1
-        strcsb      r1, [r0], #1
-        ldmfd       sp!, {r0}
+        strbmi      r1, [r3], #1
+        strbcs      r1, [r3], #1
+        strbcs      r1, [r3], #1
         bx          lr
 
-        .cfi_endproc
-END(memset)
-
-ENTRY(__memset_large_copy)
-        .cfi_startproc
-
+.L_memset_large_copy:
         /* compute the offset to align the destination
          * offset = (4-(src&3))&3 = -src & 3
          */
         stmfd       sp!, {r0, r4-r7, lr}
-        .save       {r0, r4-r7, lr}
         .cfi_def_cfa_offset 24
         .cfi_rel_offset r0, 0
         .cfi_rel_offset r4, 4
@@ -146,12 +127,11 @@ ENTRY(__memset_large_copy)
         orr         r1, r1, r1, lsr #16
 
         movs        r12, r3, lsl #31
-        strcsb      r1, [r0], #1    /* can't use strh (alignment unknown) */
-        strcsb      r1, [r0], #1
-        strmib      r1, [r0], #1
+        strbcs      r1, [r0], #1    /* can't use strh (alignment unknown) */
+        strbcs      r1, [r0], #1
+        strbmi      r1, [r0], #1
         subs        r2, r2, r3
-        ldmlsfd     sp!, {r0, r4-r7, lr}   /* return */
-        bxls        lr
+        popls       {r0, r4-r7, pc}   /* return */
 
         /* align the destination to a cache-line */
         mov         r12, r1
@@ -170,9 +150,9 @@ ENTRY(__memset_large_copy)
 
         /* conditionally writes 0 to 7 words (length in r3) */
         movs        r3, r3, lsl #28
-        stmcsia     r0!, {r1, lr}
-        stmcsia     r0!, {r1, lr}
-        stmmiia     r0!, {r1, lr}
+        stmcs       r0!, {r1, lr}
+        stmcs       r0!, {r1, lr}
+        stmmi       r0!, {r1, lr}
         movs        r3, r3, lsl #2
         strcs       r1, [r0], #4
 
@@ -187,18 +167,16 @@ ENTRY(__memset_large_copy)
 
         /* conditionally stores 0 to 31 bytes */
         movs        r2, r2, lsl #28
-        stmcsia     r0!, {r1,r3,r12,lr}
-        stmmiia     r0!, {r1, lr}
+        stmcs       r0!, {r1,r3,r12,lr}
+        stmmi       r0!, {r1, lr}
         movs        r2, r2, lsl #2
         strcs       r1, [r0], #4
-        strmih      r1, [r0], #2
+        strhmi      r1, [r0], #2
         movs        r2, r2, lsl #2
-        strcsb      r1, [r0]
-        ldmfd       sp!, {r0, r4-r7, lr}
-        bx          lr
-        .cfi_endproc
-END(__memset_large_copy)
+        strbcs      r1, [r0]
+        ldmfd       sp!, {r0, r4-r7, pc}
+END(memset)
 
         .data
 error_string:
-        .string     "memset buffer overflow"
+        .string     "memset: prevented write past end of buffer"

libc/arch-arm/cortex-a9/bionic/stpcpy.S

@@ -0,0 +1,30 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#define STPCPY
+#include "string_copy.S"

libc/arch-arm/cortex-a9/bionic/strcat.S

@@ -53,7 +53,7 @@
  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
+#include <private/bionic_asm.h>
 
     .syntax unified
 
@@ -70,7 +70,7 @@
 
     .macro m_scan_byte
     ldrb    r3, [r0]
-    cbz     r3, strcat_r0_scan_done
+    cbz     r3, .Lstrcat_r0_scan_done
     add     r0, #1
     .endm // m_scan_byte
 
@@ -84,10 +84,10 @@ ENTRY(strcat)
     // Quick check to see if src is empty.
     ldrb        r2, [r1]
     pld         [r1, #0]
-    cbnz        r2, strcat_continue
+    cbnz        r2, .Lstrcat_continue
     bx          lr
 
-strcat_continue:
+.Lstrcat_continue:
     // To speed up really small dst strings, unroll checking the first 4 bytes.
     m_push
     m_scan_byte
@@ -96,10 +96,10 @@ strcat_continue:
     m_scan_byte
 
     ands    r3, r0, #7
-    bne     strcat_align_src
+    bne     .Lstrcat_align_src
 
     .p2align 2
-strcat_mainloop:
+.Lstrcat_mainloop:
     ldmia   r0!, {r2, r3}
 
     pld     [r0, #64]
@@ -107,28 +107,28 @@ strcat_mainloop:
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcat_zero_in_first_register
+    bne     .Lstrcat_zero_in_first_register
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcat_zero_in_second_register
-    b       strcat_mainloop
+    bne     .Lstrcat_zero_in_second_register
+    b       .Lstrcat_mainloop
 
-strcat_zero_in_first_register:
+.Lstrcat_zero_in_first_register:
     sub     r0, r0, #4
 
-strcat_zero_in_second_register:
+.Lstrcat_zero_in_second_register:
     // Check for zero in byte 0.
     tst     ip, #0x80
     it      ne
     subne   r0, r0, #4
-    bne     strcat_r0_scan_done
+    bne     .Lstrcat_r0_scan_done
     // Check for zero in byte 1.
     tst     ip, #0x8000
     it      ne
     subne   r0, r0, #3
-    bne     strcat_r0_scan_done
+    bne     .Lstrcat_r0_scan_done
     // Check for zero in byte 2.
     tst     ip, #0x800000
     it      ne
@@ -137,33 +137,33 @@ strcat_zero_in_second_register:
     // Zero is in byte 3.
     subeq   r0, r0, #1
 
-strcat_r0_scan_done:
+.Lstrcat_r0_scan_done:
     // Unroll the first 8 bytes that will be copied.
-    m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r5, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r5, cmd=cbnz, label=strcpy_continue
+    m_copy_byte reg=r2, cmd=cbz, label=.Lstrcpy_finish
+    m_copy_byte reg=r3, cmd=cbz, label=.Lstrcpy_finish
+    m_copy_byte reg=r4, cmd=cbz, label=.Lstrcpy_finish
+    m_copy_byte reg=r5, cmd=cbz, label=.Lstrcpy_finish
+    m_copy_byte reg=r2, cmd=cbz, label=.Lstrcpy_finish
+    m_copy_byte reg=r3, cmd=cbz, label=.Lstrcpy_finish
+    m_copy_byte reg=r4, cmd=cbz, label=.Lstrcpy_finish
+    m_copy_byte reg=r5, cmd=cbnz, label=.Lstrcpy_continue
 
-strcpy_finish:
+.Lstrcpy_finish:
     m_ret   inst=pop
 
-strcpy_continue:
+.Lstrcpy_continue:
     pld     [r1, #0]
     ands    r3, r0, #7
-    bne     strcpy_align_dst
+    bne     .Lstrcpy_align_dst
 
-strcpy_check_src_align:
+.Lstrcpy_check_src_align:
     // At this point dst is aligned to a double word, check if src
     // is also aligned to a double word.
     ands    r3, r1, #7
-    bne     strcpy_unaligned_copy
+    bne     .Lstrcpy_unaligned_copy
 
     .p2align 2
-strcpy_mainloop:
+.Lstrcpy_mainloop:
     ldmia   r1!, {r2, r3}
 
     pld     [r1, #64]
@@ -171,17 +171,17 @@ strcpy_mainloop:
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
+    bne     .Lstrcpy_zero_in_first_register
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
+    bne     .Lstrcpy_zero_in_second_register
 
     stmia   r0!, {r2, r3}
-    b       strcpy_mainloop
+    b       .Lstrcpy_mainloop
 
-strcpy_zero_in_first_register:
+.Lstrcpy_zero_in_first_register:
     lsls    lr, ip, #17
     itt     ne
     strbne  r2, [r0]
@@ -198,7 +198,7 @@ strcpy_zero_in_first_register:
     strb    r3, [r0]
     m_ret   inst=pop
 
-strcpy_zero_in_second_register:
+.Lstrcpy_zero_in_second_register:
     lsls    lr, ip, #17
     ittt    ne
     stmiane r0!, {r2}
@@ -218,18 +218,18 @@ strcpy_zero_in_second_register:
     strb    r4, [r0]
     m_ret   inst=pop
 
-strcpy_align_dst:
+.Lstrcpy_align_dst:
     // Align to a double word (64 bits).
     rsb     r3, r3, #8
     lsls    ip, r3, #31
-    beq     strcpy_align_to_32
+    beq     .Lstrcpy_align_to_32
 
     ldrb    r2, [r1], #1
     strb    r2, [r0], #1
-    cbz     r2, strcpy_complete
+    cbz     r2, .Lstrcpy_complete
 
-strcpy_align_to_32:
-    bcc     strcpy_align_to_64
+.Lstrcpy_align_to_32:
+    bcc     .Lstrcpy_align_to_64
 
     ldrb    r4, [r1], #1
     strb    r4, [r0], #1
@@ -242,76 +242,83 @@ strcpy_align_to_32:
     it      eq
     m_ret   inst=popeq
 
-strcpy_align_to_64:
+.Lstrcpy_align_to_64:
     tst     r3, #4
-    beq     strcpy_check_src_align
-    ldr     r2, [r1], #4
+    beq     .Lstrcpy_check_src_align
+    // Read one byte at a time since we don't know the src alignment
+    // and we don't want to read into a different page.
+    ldrb    r4, [r1], #1
+    strb    r4, [r0], #1
+    cbz     r4, .Lstrcpy_complete
+    ldrb    r5, [r1], #1
+    strb    r5, [r0], #1
+    cbz     r5, .Lstrcpy_complete
+    ldrb    r4, [r1], #1
+    strb    r4, [r0], #1
+    cbz     r4, .Lstrcpy_complete
+    ldrb    r5, [r1], #1
+    strb    r5, [r0], #1
+    cbz     r5, .Lstrcpy_complete
+    b       .Lstrcpy_check_src_align
 
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-    stmia   r0!, {r2}
-    b       strcpy_check_src_align
-
-strcpy_complete:
+.Lstrcpy_complete:
     m_ret   inst=pop
 
-strcpy_unaligned_copy:
+.Lstrcpy_unaligned_copy:
     // Dst is aligned to a double word, while src is at an unknown alignment.
     // There are 7 different versions of the unaligned copy code
     // to prevent overreading the src. The mainloop of every single version
     // will store 64 bits per loop. The difference is how much of src can
     // be read without potentially crossing a page boundary.
     tbb     [pc, r3]
-strcpy_unaligned_branchtable:
+.Lstrcpy_unaligned_branchtable:
     .byte 0
-    .byte ((strcpy_unalign7 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign6 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign5 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign4 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign3 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign2 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign1 - strcpy_unaligned_branchtable)/2)
+    .byte ((.Lstrcpy_unalign7 - .Lstrcpy_unaligned_branchtable)/2)
+    .byte ((.Lstrcpy_unalign6 - .Lstrcpy_unaligned_branchtable)/2)
+    .byte ((.Lstrcpy_unalign5 - .Lstrcpy_unaligned_branchtable)/2)
+    .byte ((.Lstrcpy_unalign4 - .Lstrcpy_unaligned_branchtable)/2)
+    .byte ((.Lstrcpy_unalign3 - .Lstrcpy_unaligned_branchtable)/2)
+    .byte ((.Lstrcpy_unalign2 - .Lstrcpy_unaligned_branchtable)/2)
+    .byte ((.Lstrcpy_unalign1 - .Lstrcpy_unaligned_branchtable)/2)
 
     .p2align 2
     // Can read 7 bytes before possibly crossing a page.
-strcpy_unalign7:
+.Lstrcpy_unalign7:
     ldr     r2, [r1], #4
 
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
+    bne     .Lstrcpy_zero_in_first_register
 
     ldrb    r3, [r1]
-    cbz     r3, strcpy_unalign7_copy5bytes
+    cbz     r3, .Lstrcpy_unalign7_copy5bytes
     ldrb    r4, [r1, #1]
-    cbz     r4, strcpy_unalign7_copy6bytes
+    cbz     r4, .Lstrcpy_unalign7_copy6bytes
     ldrb    r5, [r1, #2]
-    cbz     r5, strcpy_unalign7_copy7bytes
+    cbz     r5, .Lstrcpy_unalign7_copy7bytes
 
     ldr     r3, [r1], #4
     pld     [r1, #64]
 
     lsrs    ip, r3, #24
     stmia   r0!, {r2, r3}
-    beq     strcpy_unalign_return
-    b       strcpy_unalign7
+    beq     .Lstrcpy_unalign_return
+    b       .Lstrcpy_unalign7
 
-strcpy_unalign7_copy5bytes:
+.Lstrcpy_unalign7_copy5bytes:
     stmia   r0!, {r2}
     strb    r3, [r0]
-strcpy_unalign_return:
+.Lstrcpy_unalign_return:
     m_ret   inst=pop
 
-strcpy_unalign7_copy6bytes:
+.Lstrcpy_unalign7_copy6bytes:
     stmia   r0!, {r2}
     strb    r3, [r0], #1
     strb    r4, [r0], #1
     m_ret   inst=pop
 
-strcpy_unalign7_copy7bytes:
+.Lstrcpy_unalign7_copy7bytes:
     stmia   r0!, {r2}
     strb    r3, [r0], #1
     strb    r4, [r0], #1
@@ -320,30 +327,30 @@ strcpy_unalign7_copy7bytes:
 
     .p2align 2
     // Can read 6 bytes before possibly crossing a page.
-strcpy_unalign6:
+.Lstrcpy_unalign6:
     ldr     r2, [r1], #4
 
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
+    bne     .Lstrcpy_zero_in_first_register
 
     ldrb    r4, [r1]
-    cbz     r4, strcpy_unalign_copy5bytes
+    cbz     r4, .Lstrcpy_unalign_copy5bytes
     ldrb    r5, [r1, #1]
-    cbz     r5, strcpy_unalign_copy6bytes
+    cbz     r5, .Lstrcpy_unalign_copy6bytes
 
     ldr     r3, [r1], #4
     pld     [r1, #64]
 
     tst     r3, #0xff0000
-    beq     strcpy_unalign6_copy7bytes
+    beq     .Lstrcpy_unalign6_copy7bytes
     lsrs    ip, r3, #24
     stmia   r0!, {r2, r3}
-    beq     strcpy_unalign_return
-    b       strcpy_unalign6
+    beq     .Lstrcpy_unalign_return
+    b       .Lstrcpy_unalign6
 
-strcpy_unalign6_copy7bytes:
+.Lstrcpy_unalign6_copy7bytes:
     stmia   r0!, {r2}
     strh    r3, [r0], #2
     lsr     r3, #16
@@ -352,16 +359,16 @@ strcpy_unalign6_copy7bytes:
 
     .p2align 2
     // Can read 5 bytes before possibly crossing a page.
-strcpy_unalign5:
+.Lstrcpy_unalign5:
     ldr     r2, [r1], #4
 
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
+    bne     .Lstrcpy_zero_in_first_register
 
     ldrb    r4, [r1]
-    cbz     r4, strcpy_unalign_copy5bytes
+    cbz     r4, .Lstrcpy_unalign_copy5bytes
 
     ldr     r3, [r1], #4
 
@@ -370,17 +377,17 @@ strcpy_unalign5:
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
+    bne     .Lstrcpy_zero_in_second_register
 
     stmia   r0!, {r2, r3}
-    b       strcpy_unalign5
+    b       .Lstrcpy_unalign5
 
-strcpy_unalign_copy5bytes:
+.Lstrcpy_unalign_copy5bytes:
     stmia   r0!, {r2}
     strb    r4, [r0]
     m_ret   inst=pop
 
-strcpy_unalign_copy6bytes:
+.Lstrcpy_unalign_copy6bytes:
     stmia   r0!, {r2}
     strb    r4, [r0], #1
     strb    r5, [r0]
@@ -388,13 +395,13 @@ strcpy_unalign_copy6bytes:
 
     .p2align 2
     // Can read 4 bytes before possibly crossing a page.
-strcpy_unalign4:
+.Lstrcpy_unalign4:
     ldmia   r1!, {r2}
 
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
+    bne     .Lstrcpy_zero_in_first_register
 
     ldmia   r1!, {r3}
     pld     [r1, #64]
@@ -402,20 +409,20 @@ strcpy_unalign4:
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
+    bne     .Lstrcpy_zero_in_second_register
 
     stmia   r0!, {r2, r3}
-    b       strcpy_unalign4
+    b       .Lstrcpy_unalign4
 
     .p2align 2
     // Can read 3 bytes before possibly crossing a page.
-strcpy_unalign3:
+.Lstrcpy_unalign3:
     ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign3_copy1byte
+    cbz     r2, .Lstrcpy_unalign3_copy1byte
     ldrb    r3, [r1, #1]
-    cbz     r3, strcpy_unalign3_copy2bytes
+    cbz     r3, .Lstrcpy_unalign3_copy2bytes
     ldrb    r4, [r1, #2]
-    cbz     r4, strcpy_unalign3_copy3bytes
+    cbz     r4, .Lstrcpy_unalign3_copy3bytes
 
     ldr     r2, [r1], #4
     ldr     r3, [r1], #4
@@ -423,26 +430,26 @@ strcpy_unalign3:
     pld     [r1, #64]
 
     lsrs    lr, r2, #24
-    beq     strcpy_unalign_copy4bytes
+    beq     .Lstrcpy_unalign_copy4bytes
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
+    bne     .Lstrcpy_zero_in_second_register
 
     stmia   r0!, {r2, r3}
-    b       strcpy_unalign3
+    b       .Lstrcpy_unalign3
 
-strcpy_unalign3_copy1byte:
+.Lstrcpy_unalign3_copy1byte:
     strb    r2, [r0]
     m_ret   inst=pop
 
-strcpy_unalign3_copy2bytes:
+.Lstrcpy_unalign3_copy2bytes:
     strb    r2, [r0], #1
     strb    r3, [r0]
     m_ret   inst=pop
 
-strcpy_unalign3_copy3bytes:
+.Lstrcpy_unalign3_copy3bytes:
     strb    r2, [r0], #1
     strb    r3, [r0], #1
     strb    r4, [r0]
@@ -450,34 +457,34 @@ strcpy_unalign3_copy3bytes:
 
     .p2align 2
     // Can read 2 bytes before possibly crossing a page.
-strcpy_unalign2:
+.Lstrcpy_unalign2:
     ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign_copy1byte
+    cbz     r2, .Lstrcpy_unalign_copy1byte
     ldrb    r3, [r1, #1]
-    cbz     r3, strcpy_unalign_copy2bytes
+    cbz     r3, .Lstrcpy_unalign_copy2bytes
 
     ldr     r2, [r1], #4
     ldr     r3, [r1], #4
     pld     [r1, #64]
 
     tst     r2, #0xff0000
-    beq     strcpy_unalign_copy3bytes
+    beq     .Lstrcpy_unalign_copy3bytes
     lsrs    ip, r2, #24
-    beq     strcpy_unalign_copy4bytes
+    beq     .Lstrcpy_unalign_copy4bytes
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
+    bne     .Lstrcpy_zero_in_second_register
 
     stmia   r0!, {r2, r3}
-    b       strcpy_unalign2
+    b       .Lstrcpy_unalign2
 
     .p2align 2
     // Can read 1 byte before possibly crossing a page.
-strcpy_unalign1:
+.Lstrcpy_unalign1:
     ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign_copy1byte
+    cbz     r2, .Lstrcpy_unalign_copy1byte
 
     ldr     r2, [r1], #4
     ldr     r3, [r1], #4
@@ -487,62 +494,62 @@ strcpy_unalign1:
     sub     ip, r2, #0x01010101
     bic     ip, ip, r2
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
+    bne     .Lstrcpy_zero_in_first_register
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
+    bne     .Lstrcpy_zero_in_second_register
 
     stmia   r0!, {r2, r3}
-    b       strcpy_unalign1
+    b       .Lstrcpy_unalign1
 
-strcpy_unalign_copy1byte:
+.Lstrcpy_unalign_copy1byte:
     strb    r2, [r0]
     m_ret   inst=pop
 
-strcpy_unalign_copy2bytes:
+.Lstrcpy_unalign_copy2bytes:
     strb    r2, [r0], #1
     strb    r3, [r0]
     m_ret   inst=pop
 
-strcpy_unalign_copy3bytes:
+.Lstrcpy_unalign_copy3bytes:
     strh    r2, [r0], #2
     lsr     r2, #16
     strb    r2, [r0]
     m_ret   inst=pop
 
-strcpy_unalign_copy4bytes:
+.Lstrcpy_unalign_copy4bytes:
     stmia   r0, {r2}
     m_ret   inst=pop
 
-strcat_align_src:
+.Lstrcat_align_src:
     // Align to a double word (64 bits).
     rsb     r3, r3, #8
     lsls    ip, r3, #31
-    beq     strcat_align_to_32
+    beq     .Lstrcat_align_to_32
     ldrb    r2, [r0], #1
-    cbz     r2, strcat_r0_update
+    cbz     r2, .Lstrcat_r0_update
 
-strcat_align_to_32:
-    bcc     strcat_align_to_64
+.Lstrcat_align_to_32:
+    bcc     .Lstrcat_align_to_64
     ldrb    r2, [r0], #1
-    cbz     r2, strcat_r0_update
+    cbz     r2, .Lstrcat_r0_update
     ldrb    r2, [r0], #1
-    cbz     r2, strcat_r0_update
+    cbz     r2, .Lstrcat_r0_update
 
-strcat_align_to_64:
+.Lstrcat_align_to_64:
     tst     r3, #4
-    beq     strcat_mainloop
+    beq     .Lstrcat_mainloop
     ldr     r3, [r0], #4
 
     sub     ip, r3, #0x01010101
     bic     ip, ip, r3
     ands    ip, ip, #0x80808080
-    bne     strcat_zero_in_second_register
-    b       strcat_mainloop
+    bne     .Lstrcat_zero_in_second_register
+    b       .Lstrcat_mainloop
 
-strcat_r0_update:
+.Lstrcat_r0_update:
     sub     r0, r0, #1
-    b strcat_r0_scan_done
+    b .Lstrcat_r0_scan_done
 END(strcat)

libc/arch-arm/cortex-a9/bionic/strcmp.S

@@ -27,7 +27,7 @@
  */
 
 #include <machine/cpu-features.h>
-#include <machine/asm.h>
+#include <private/bionic_asm.h>
 
 #ifdef __ARMEB__
 #define S2LOMEM lsl
@@ -145,19 +145,8 @@ ENTRY(strcmp)
 
         .macro  magic_find_zero_bytes w1
         /* Macro to find all-zero bytes in w1, result is in ip.  */
-#if (defined (__ARM_FEATURE_DSP))
         uadd8   ip, \w1, r6
         sel     ip, r7, r6
-#else /* not defined (__ARM_FEATURE_DSP) */
-        /* __ARM_FEATURE_DSP is not defined for some Cortex-M processors.
-        Coincidently, these processors only have Thumb-2 mode, where we can use the
-        the (large) magic constant available directly as an immediate in instructions.
-        Note that we cannot use the magic constant in ARM mode, where we need
-        to create the constant in a register.  */
-        sub     ip, \w1, #0x01010101
-        bic     ip, ip, \w1
-        and     ip, ip, #0x80808080
-#endif /* not defined (__ARM_FEATURE_DSP) */
         .endm /* magic_find_zero_bytes */
 
         .macro  setup_return w1 w2
@@ -170,7 +159,6 @@ ENTRY(strcmp)
 #endif /* not  __ARMEB__ */
         .endm /* setup_return */
 
-        .cfi_startproc
         pld [r0, #0]
         pld [r1, #0]
 
@@ -180,7 +168,6 @@ ENTRY(strcmp)
         bne     .L_do_align
 
         /* Fast path.  */
-        .save   {r4-r7}
         init
 
 .L_doubleword_aligned:
@@ -558,5 +545,4 @@ ENTRY(strcmp)
     adds    sp, sp, #16
 
 	bx	lr
-    .cfi_endproc
 END(strcmp)

libc/arch-arm/cortex-a9/bionic/strcpy.S

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2013 The Android Open Source Project
+ * Copyright (C) 2014 The Android Open Source Project
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -25,432 +25,6 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
-/*
- * Copyright (c) 2013 ARM Ltd
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- * 3. The name of the company may not be used to endorse or promote
- *    products derived from this software without specific prior written
- *    permission.
- *
- * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
- * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
- * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
- * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
- * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
- * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
- * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
- * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
- * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- */
 
-#include <machine/asm.h>
-
-    .syntax unified
-
-    .thumb
-    .thumb_func
-
-    .macro m_push
-    push    {r0, r4, r5, lr}
-    .endm // m_push
-
-    .macro m_ret inst
-    \inst   {r0, r4, r5, pc}
-    .endm // m_ret
-
-    .macro m_copy_byte reg, cmd, label
-    ldrb    \reg, [r1], #1
-    strb    \reg, [r0], #1
-    \cmd    \reg, \label
-    .endm // m_copy_byte
-
-ENTRY(strcpy)
-    // Unroll the first 8 bytes that will be copied.
-    m_push
-    m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r5, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r5, cmd=cbnz, label=strcpy_continue
-
-strcpy_finish:
-    m_ret   inst=pop
-
-strcpy_continue:
-    pld     [r1, #0]
-    ands    r3, r0, #7
-    bne     strcpy_align_dst
-
-strcpy_check_src_align:
-    // At this point dst is aligned to a double word, check if src
-    // is also aligned to a double word.
-    ands    r3, r1, #7
-    bne     strcpy_unaligned_copy
-
-    .p2align 2
-strcpy_mainloop:
-    ldmia   r1!, {r2, r3}
-
-    pld     [r1, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    stmia   r0!, {r2, r3}
-    b       strcpy_mainloop
-
-strcpy_zero_in_first_register:
-    lsls    lr, ip, #17
-    itt     ne
-    strbne  r2, [r0]
-    m_ret   inst=popne
-    itt     cs
-    strhcs  r2, [r0]
-    m_ret   inst=popcs
-    lsls    ip, ip, #1
-    itt     eq
-    streq   r2, [r0]
-    m_ret   inst=popeq
-    strh    r2, [r0], #2
-    lsr     r3, r2, #16
-    strb    r3, [r0]
-    m_ret   inst=pop
-
-strcpy_zero_in_second_register:
-    lsls    lr, ip, #17
-    ittt    ne
-    stmiane r0!, {r2}
-    strbne  r3, [r0]
-    m_ret   inst=popne
-    ittt    cs
-    strcs   r2, [r0], #4
-    strhcs  r3, [r0]
-    m_ret   inst=popcs
-    lsls    ip, ip, #1
-    itt     eq
-    stmiaeq r0, {r2, r3}
-    m_ret   inst=popeq
-    stmia   r0!, {r2}
-    strh    r3, [r0], #2
-    lsr     r4, r3, #16
-    strb    r4, [r0]
-    m_ret   inst=pop
-
-strcpy_align_dst:
-    // Align to a double word (64 bits).
-    rsb     r3, r3, #8
-    lsls    ip, r3, #31
-    beq     strcpy_align_to_32
-
-    ldrb    r2, [r1], #1
-    strb    r2, [r0], #1
-    cbz     r2, strcpy_complete
-
-strcpy_align_to_32:
-    bcc     strcpy_align_to_64
-
-    ldrb    r4, [r1], #1
-    strb    r4, [r0], #1
-    cmp     r4, #0
-    it      eq
-    m_ret   inst=popeq
-    ldrb    r5, [r1], #1
-    strb    r5, [r0], #1
-    cmp     r5, #0
-    it      eq
-    m_ret   inst=popeq
-
-strcpy_align_to_64:
-    tst     r3, #4
-    beq     strcpy_check_src_align
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-    stmia   r0!, {r2}
-    b       strcpy_check_src_align
-
-strcpy_complete:
-    m_ret   inst=pop
-
-strcpy_unaligned_copy:
-    // Dst is aligned to a double word, while src is at an unknown alignment.
-    // There are 7 different versions of the unaligned copy code
-    // to prevent overreading the src. The mainloop of every single version
-    // will store 64 bits per loop. The difference is how much of src can
-    // be read without potentially crossing a page boundary.
-    tbb     [pc, r3]
-strcpy_unaligned_branchtable:
-    .byte 0
-    .byte ((strcpy_unalign7 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign6 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign5 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign4 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign3 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign2 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign1 - strcpy_unaligned_branchtable)/2)
-
-    .p2align 2
-    // Can read 7 bytes before possibly crossing a page.
-strcpy_unalign7:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldrb    r3, [r1]
-    cbz     r3, strcpy_unalign7_copy5bytes
-    ldrb    r4, [r1, #1]
-    cbz     r4, strcpy_unalign7_copy6bytes
-    ldrb    r5, [r1, #2]
-    cbz     r5, strcpy_unalign7_copy7bytes
-
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    lsrs    ip, r3, #24
-    stmia   r0!, {r2, r3}
-    beq     strcpy_unalign_return
-    b       strcpy_unalign7
-
-strcpy_unalign7_copy5bytes:
-    stmia   r0!, {r2}
-    strb    r3, [r0]
-strcpy_unalign_return:
-    m_ret   inst=pop
-
-strcpy_unalign7_copy6bytes:
-    stmia   r0!, {r2}
-    strb    r3, [r0], #1
-    strb    r4, [r0], #1
-    m_ret   inst=pop
-
-strcpy_unalign7_copy7bytes:
-    stmia   r0!, {r2}
-    strb    r3, [r0], #1
-    strb    r4, [r0], #1
-    strb    r5, [r0], #1
-    m_ret   inst=pop
-
-    .p2align 2
-    // Can read 6 bytes before possibly crossing a page.
-strcpy_unalign6:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldrb    r4, [r1]
-    cbz     r4, strcpy_unalign_copy5bytes
-    ldrb    r5, [r1, #1]
-    cbz     r5, strcpy_unalign_copy6bytes
-
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    tst     r3, #0xff0000
-    beq     strcpy_unalign6_copy7bytes
-    lsrs    ip, r3, #24
-    stmia   r0!, {r2, r3}
-    beq     strcpy_unalign_return
-    b       strcpy_unalign6
-
-strcpy_unalign6_copy7bytes:
-    stmia   r0!, {r2}
-    strh    r3, [r0], #2
-    lsr     r3, #16
-    strb    r3, [r0]
-    m_ret   inst=pop
-
-    .p2align 2
-    // Can read 5 bytes before possibly crossing a page.
-strcpy_unalign5:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldrb    r4, [r1]
-    cbz     r4, strcpy_unalign_copy5bytes
-
-    ldr     r3, [r1], #4
-
-    pld     [r1, #64]
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    stmia   r0!, {r2, r3}
-    b       strcpy_unalign5
-
-strcpy_unalign_copy5bytes:
-    stmia   r0!, {r2}
-    strb    r4, [r0]
-    m_ret   inst=pop
-
-strcpy_unalign_copy6bytes:
-    stmia   r0!, {r2}
-    strb    r4, [r0], #1
-    strb    r5, [r0]
-    m_ret   inst=pop
-
-    .p2align 2
-    // Can read 4 bytes before possibly crossing a page.
-strcpy_unalign4:
-    ldmia   r1!, {r2}
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldmia   r1!, {r3}
-    pld     [r1, #64]
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    stmia   r0!, {r2, r3}
-    b       strcpy_unalign4
-
-    .p2align 2
-    // Can read 3 bytes before possibly crossing a page.
-strcpy_unalign3:
-    ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign3_copy1byte
-    ldrb    r3, [r1, #1]
-    cbz     r3, strcpy_unalign3_copy2bytes
-    ldrb    r4, [r1, #2]
-    cbz     r4, strcpy_unalign3_copy3bytes
-
-    ldr     r2, [r1], #4
-    ldr     r3, [r1], #4
-
-    pld     [r1, #64]
-
-    lsrs    lr, r2, #24
-    beq     strcpy_unalign_copy4bytes
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    stmia   r0!, {r2, r3}
-    b       strcpy_unalign3
-
-strcpy_unalign3_copy1byte:
-    strb    r2, [r0]
-    m_ret   inst=pop
-
-strcpy_unalign3_copy2bytes:
-    strb    r2, [r0], #1
-    strb    r3, [r0]
-    m_ret   inst=pop
-
-strcpy_unalign3_copy3bytes:
-    strb    r2, [r0], #1
-    strb    r3, [r0], #1
-    strb    r4, [r0]
-    m_ret   inst=pop
-
-    .p2align 2
-    // Can read 2 bytes before possibly crossing a page.
-strcpy_unalign2:
-    ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign_copy1byte
-    ldrb    r3, [r1, #1]
-    cbz     r3, strcpy_unalign_copy2bytes
-
-    ldr     r2, [r1], #4
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    tst     r2, #0xff0000
-    beq     strcpy_unalign_copy3bytes
-    lsrs    ip, r2, #24
-    beq     strcpy_unalign_copy4bytes
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    stmia   r0!, {r2, r3}
-    b       strcpy_unalign2
-
-    .p2align 2
-    // Can read 1 byte before possibly crossing a page.
-strcpy_unalign1:
-    ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign_copy1byte
-
-    ldr     r2, [r1], #4
-    ldr     r3, [r1], #4
-
-    pld     [r1, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    stmia   r0!, {r2, r3}
-    b       strcpy_unalign1
-
-strcpy_unalign_copy1byte:
-    strb    r2, [r0]
-    m_ret   inst=pop
-
-strcpy_unalign_copy2bytes:
-    strb    r2, [r0], #1
-    strb    r3, [r0]
-    m_ret   inst=pop
-
-strcpy_unalign_copy3bytes:
-    strh    r2, [r0], #2
-    lsr     r2, #16
-    strb    r2, [r0]
-    m_ret   inst=pop
-
-strcpy_unalign_copy4bytes:
-    stmia   r0, {r2}
-    m_ret   inst=pop
-END(strcpy)
+#define STRCPY
+#include "string_copy.S"

libc/arch-arm/cortex-a9/bionic/string_copy.S

@@ -0,0 +1,542 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+/*
+ * Copyright (c) 2013 ARM Ltd
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. The name of the company may not be used to endorse or promote
+ *    products derived from this software without specific prior written
+ *    permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#if !defined(STPCPY) && !defined(STRCPY)
+#error "Either STPCPY or STRCPY must be defined."
+#endif
+
+#include <private/bionic_asm.h>
+
+    .syntax unified
+
+    .thumb
+    .thumb_func
+
+#if defined(STPCPY)
+    .macro m_push
+    push    {r4, r5, lr}
+    .cfi_def_cfa_offset 12
+    .cfi_rel_offset r4, 0
+    .cfi_rel_offset r5, 4
+    .cfi_rel_offset lr, 8
+    .endm // m_push
+#else
+    .macro m_push
+    push    {r0, r4, r5, lr}
+    .cfi_def_cfa_offset 16
+    .cfi_rel_offset r0, 0
+    .cfi_rel_offset r4, 4
+    .cfi_rel_offset r5, 8
+    .cfi_rel_offset lr, 12
+    .endm // m_push
+#endif
+
+#if defined(STPCPY)
+    .macro m_ret inst
+    \inst   {r4, r5, pc}
+    .endm // m_ret
+#else
+    .macro m_ret inst
+    \inst   {r0, r4, r5, pc}
+    .endm // m_ret
+#endif
+
+    .macro m_copy_byte reg, cmd, label
+    ldrb    \reg, [r1], #1
+    strb    \reg, [r0], #1
+    \cmd    \reg, \label
+    .endm // m_copy_byte
+
+#if defined(STPCPY)
+ENTRY(stpcpy)
+#else
+ENTRY(strcpy)
+#endif
+    // Unroll the first 8 bytes that will be copied.
+    m_push
+    m_copy_byte reg=r2, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r3, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r4, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r5, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r2, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r3, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r4, cmd=cbz, label=.Lstringcopy_finish
+    m_copy_byte reg=r5, cmd=cbnz, label=.Lstringcopy_continue
+
+.Lstringcopy_finish:
+#if defined(STPCPY)
+    sub     r0, r0, #1
+#endif
+    m_ret   inst=pop
+
+.Lstringcopy_continue:
+    pld     [r1, #0]
+    ands    r3, r0, #7
+    bne     .Lstringcopy_align_dst
+
+.Lstringcopy_check_src_align:
+    // At this point dst is aligned to a double word, check if src
+    // is also aligned to a double word.
+    ands    r3, r1, #7
+    bne     .Lstringcopy_unaligned_copy
+
+    .p2align 2
+.Lstringcopy_mainloop:
+    ldmia   r1!, {r2, r3}
+
+    pld     [r1, #64]
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_second_register
+
+    stmia   r0!, {r2, r3}
+    b       .Lstringcopy_mainloop
+
+.Lstringcopy_zero_in_first_register:
+    lsls    lr, ip, #17
+    itt     ne
+    strbne  r2, [r0]
+    m_ret   inst=popne
+    itt     cs
+#if defined(STPCPY)
+    strhcs  r2, [r0], #1
+#else
+    strhcs  r2, [r0]
+#endif
+    m_ret   inst=popcs
+    lsls    ip, ip, #1
+    itt     eq
+#if defined(STPCPY)
+    streq   r2, [r0], #3
+#else
+    streq   r2, [r0]
+#endif
+    m_ret   inst=popeq
+    strh    r2, [r0], #2
+    lsr     r3, r2, #16
+    strb    r3, [r0]
+    m_ret   inst=pop
+
+.Lstringcopy_zero_in_second_register:
+    lsls    lr, ip, #17
+    ittt    ne
+    stmiane r0!, {r2}
+    strbne  r3, [r0]
+    m_ret   inst=popne
+    ittt    cs
+    strcs   r2, [r0], #4
+#if defined(STPCPY)
+    strhcs  r3, [r0], #1
+#else
+    strhcs  r3, [r0]
+#endif
+    m_ret   inst=popcs
+    lsls    ip, ip, #1
+#if defined(STPCPY)
+    ittt    eq
+#else
+    itt     eq
+#endif
+    stmiaeq r0, {r2, r3}
+#if defined(STPCPY)
+    addeq   r0, r0, #7
+#endif
+    m_ret   inst=popeq
+    stmia   r0!, {r2}
+    strh    r3, [r0], #2
+    lsr     r4, r3, #16
+    strb    r4, [r0]
+    m_ret   inst=pop
+
+.Lstringcopy_align_dst:
+    // Align to a double word (64 bits).
+    rsb     r3, r3, #8
+    lsls    ip, r3, #31
+    beq     .Lstringcopy_align_to_32
+
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .Lstringcopy_complete
+
+.Lstringcopy_align_to_32:
+    bcc     .Lstringcopy_align_to_64
+
+    ldrb    r4, [r1], #1
+    strb    r4, [r0], #1
+    cmp     r4, #0
+#if defined(STPCPY)
+    itt     eq
+    subeq   r0, r0, #1
+#else
+    it      eq
+#endif
+    m_ret   inst=popeq
+    ldrb    r5, [r1], #1
+    strb    r5, [r0], #1
+    cmp     r5, #0
+#if defined(STPCPY)
+    itt     eq
+    subeq   r0, r0, #1
+#else
+    it      eq
+#endif
+    m_ret   inst=popeq
+
+.Lstringcopy_align_to_64:
+    tst     r3, #4
+    beq     .Lstringcopy_check_src_align
+    // Read one byte at a time since we don't have any idea about the alignment
+    // of the source and we don't want to read into a different page.
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .Lstringcopy_complete
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .Lstringcopy_complete
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .Lstringcopy_complete
+    ldrb    r2, [r1], #1
+    strb    r2, [r0], #1
+    cbz     r2, .Lstringcopy_complete
+    b       .Lstringcopy_check_src_align
+
+.Lstringcopy_complete:
+#if defined(STPCPY)
+    sub     r0, r0, #1
+#endif
+    m_ret   inst=pop
+
+.Lstringcopy_unaligned_copy:
+    // Dst is aligned to a double word, while src is at an unknown alignment.
+    // There are 7 different versions of the unaligned copy code
+    // to prevent overreading the src. The mainloop of every single version
+    // will store 64 bits per loop. The difference is how much of src can
+    // be read without potentially crossing a page boundary.
+    tbb     [pc, r3]
+.Lstringcopy_unaligned_branchtable:
+    .byte 0
+    .byte ((.Lstringcopy_unalign7 - .Lstringcopy_unaligned_branchtable)/2)
+    .byte ((.Lstringcopy_unalign6 - .Lstringcopy_unaligned_branchtable)/2)
+    .byte ((.Lstringcopy_unalign5 - .Lstringcopy_unaligned_branchtable)/2)
+    .byte ((.Lstringcopy_unalign4 - .Lstringcopy_unaligned_branchtable)/2)
+    .byte ((.Lstringcopy_unalign3 - .Lstringcopy_unaligned_branchtable)/2)
+    .byte ((.Lstringcopy_unalign2 - .Lstringcopy_unaligned_branchtable)/2)
+    .byte ((.Lstringcopy_unalign1 - .Lstringcopy_unaligned_branchtable)/2)
+
+    .p2align 2
+    // Can read 7 bytes before possibly crossing a page.
+.Lstringcopy_unalign7:
+    ldr     r2, [r1], #4
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+
+    ldrb    r3, [r1]
+    cbz     r3, .Lstringcopy_unalign7_copy5bytes
+    ldrb    r4, [r1, #1]
+    cbz     r4, .Lstringcopy_unalign7_copy6bytes
+    ldrb    r5, [r1, #2]
+    cbz     r5, .Lstringcopy_unalign7_copy7bytes
+
+    ldr     r3, [r1], #4
+    pld     [r1, #64]
+
+    lsrs    ip, r3, #24
+    stmia   r0!, {r2, r3}
+#if defined(STPCPY)
+    beq     .Lstringcopy_finish
+#else
+    beq     .Lstringcopy_unalign_return
+#endif
+    b       .Lstringcopy_unalign7
+
+.Lstringcopy_unalign7_copy5bytes:
+    stmia   r0!, {r2}
+    strb    r3, [r0]
+.Lstringcopy_unalign_return:
+    m_ret   inst=pop
+
+.Lstringcopy_unalign7_copy6bytes:
+    stmia   r0!, {r2}
+    strb    r3, [r0], #1
+    strb    r4, [r0]
+    m_ret   inst=pop
+
+.Lstringcopy_unalign7_copy7bytes:
+    stmia   r0!, {r2}
+    strb    r3, [r0], #1
+    strb    r4, [r0], #1
+    strb    r5, [r0]
+    m_ret   inst=pop
+
+    .p2align 2
+    // Can read 6 bytes before possibly crossing a page.
+.Lstringcopy_unalign6:
+    ldr     r2, [r1], #4
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+
+    ldrb    r4, [r1]
+    cbz     r4, .Lstringcopy_unalign_copy5bytes
+    ldrb    r5, [r1, #1]
+    cbz     r5, .Lstringcopy_unalign_copy6bytes
+
+    ldr     r3, [r1], #4
+    pld     [r1, #64]
+
+    tst     r3, #0xff0000
+    beq     .Lstringcopy_unalign6_copy7bytes
+    lsrs    ip, r3, #24
+    stmia   r0!, {r2, r3}
+#if defined(STPCPY)
+    beq     .Lstringcopy_finish
+#else
+    beq     .Lstringcopy_unalign_return
+#endif
+    b       .Lstringcopy_unalign6
+
+.Lstringcopy_unalign6_copy7bytes:
+    stmia   r0!, {r2}
+    strh    r3, [r0], #2
+    lsr     r3, #16
+    strb    r3, [r0]
+    m_ret   inst=pop
+
+    .p2align 2
+    // Can read 5 bytes before possibly crossing a page.
+.Lstringcopy_unalign5:
+    ldr     r2, [r1], #4
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+
+    ldrb    r4, [r1]
+    cbz     r4, .Lstringcopy_unalign_copy5bytes
+
+    ldr     r3, [r1], #4
+
+    pld     [r1, #64]
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_second_register
+
+    stmia   r0!, {r2, r3}
+    b       .Lstringcopy_unalign5
+
+.Lstringcopy_unalign_copy5bytes:
+    stmia   r0!, {r2}
+    strb    r4, [r0]
+    m_ret   inst=pop
+
+.Lstringcopy_unalign_copy6bytes:
+    stmia   r0!, {r2}
+    strb    r4, [r0], #1
+    strb    r5, [r0]
+    m_ret   inst=pop
+
+    .p2align 2
+    // Can read 4 bytes before possibly crossing a page.
+.Lstringcopy_unalign4:
+    ldmia   r1!, {r2}
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+
+    ldmia   r1!, {r3}
+    pld     [r1, #64]
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_second_register
+
+    stmia   r0!, {r2, r3}
+    b       .Lstringcopy_unalign4
+
+    .p2align 2
+    // Can read 3 bytes before possibly crossing a page.
+.Lstringcopy_unalign3:
+    ldrb    r2, [r1]
+    cbz     r2, .Lstringcopy_unalign3_copy1byte
+    ldrb    r3, [r1, #1]
+    cbz     r3, .Lstringcopy_unalign3_copy2bytes
+    ldrb    r4, [r1, #2]
+    cbz     r4, .Lstringcopy_unalign3_copy3bytes
+
+    ldr     r2, [r1], #4
+    ldr     r3, [r1], #4
+
+    pld     [r1, #64]
+
+    lsrs    lr, r2, #24
+    beq     .Lstringcopy_unalign_copy4bytes
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_second_register
+
+    stmia   r0!, {r2, r3}
+    b       .Lstringcopy_unalign3
+
+.Lstringcopy_unalign3_copy1byte:
+    strb    r2, [r0]
+    m_ret   inst=pop
+
+.Lstringcopy_unalign3_copy2bytes:
+    strb    r2, [r0], #1
+    strb    r3, [r0]
+    m_ret   inst=pop
+
+.Lstringcopy_unalign3_copy3bytes:
+    strb    r2, [r0], #1
+    strb    r3, [r0], #1
+    strb    r4, [r0]
+    m_ret   inst=pop
+
+    .p2align 2
+    // Can read 2 bytes before possibly crossing a page.
+.Lstringcopy_unalign2:
+    ldrb    r2, [r1]
+    cbz     r2, .Lstringcopy_unalign_copy1byte
+    ldrb    r3, [r1, #1]
+    cbz     r3, .Lstringcopy_unalign_copy2bytes
+
+    ldr     r2, [r1], #4
+    ldr     r3, [r1], #4
+    pld     [r1, #64]
+
+    tst     r2, #0xff0000
+    beq     .Lstringcopy_unalign_copy3bytes
+    lsrs    ip, r2, #24
+    beq     .Lstringcopy_unalign_copy4bytes
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_second_register
+
+    stmia   r0!, {r2, r3}
+    b       .Lstringcopy_unalign2
+
+    .p2align 2
+    // Can read 1 byte before possibly crossing a page.
+.Lstringcopy_unalign1:
+    ldrb    r2, [r1]
+    cbz     r2, .Lstringcopy_unalign_copy1byte
+
+    ldr     r2, [r1], #4
+    ldr     r3, [r1], #4
+
+    pld     [r1, #64]
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_second_register
+
+    stmia   r0!, {r2, r3}
+    b       .Lstringcopy_unalign1
+
+.Lstringcopy_unalign_copy1byte:
+    strb    r2, [r0]
+    m_ret   inst=pop
+
+.Lstringcopy_unalign_copy2bytes:
+    strb    r2, [r0], #1
+    strb    r3, [r0]
+    m_ret   inst=pop
+
+.Lstringcopy_unalign_copy3bytes:
+    strh    r2, [r0], #2
+    lsr     r2, #16
+    strb    r2, [r0]
+    m_ret   inst=pop
+
+.Lstringcopy_unalign_copy4bytes:
+    stmia   r0, {r2}
+#if defined(STPCPY)
+    add     r0, r0, #3
+#endif
+    m_ret   inst=pop
+#if defined(STPCPY)
+END(stpcpy)
+#else
+END(strcpy)
+#endif

libc/arch-arm/cortex-a9/bionic/strlen.S

@@ -53,7 +53,7 @@
  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
+#include <private/bionic_asm.h>
 
     .syntax unified
 

libc/arch-arm/cortex-a9/cortex-a9.mk

@@ -1,10 +1,28 @@
-$(call libc-add-cpu-variant-src,MEMCPY,arch-arm/cortex-a9/bionic/memcpy.S)
-$(call libc-add-cpu-variant-src,MEMSET,arch-arm/cortex-a9/bionic/memset.S)
-$(call libc-add-cpu-variant-src,STRCAT,arch-arm/cortex-a9/bionic/strcat.S)
-$(call libc-add-cpu-variant-src,STRCMP,arch-arm/cortex-a9/bionic/strcmp.S)
-$(call libc-add-cpu-variant-src,STRCPY,arch-arm/cortex-a9/bionic/strcpy.S)
-$(call libc-add-cpu-variant-src,STRLEN,arch-arm/cortex-a9/bionic/strlen.S)
-$(call libc-add-cpu-variant-src,__STRCAT_CHK,arch-arm/cortex-a9/bionic/__strcat_chk.S)
-$(call libc-add-cpu-variant-src,__STRCPY_CHK,arch-arm/cortex-a9/bionic/__strcpy_chk.S)
+libc_openbsd_src_files_exclude_arm += \
+    upstream-openbsd/lib/libc/string/memmove.c \
+    upstream-openbsd/lib/libc/string/stpcpy.c \
+    upstream-openbsd/lib/libc/string/strcat.c \
+    upstream-openbsd/lib/libc/string/strcpy.c \
 
-include bionic/libc/arch-arm/generic/generic.mk
+libc_bionic_src_files_exclude_arm += \
+    arch-arm/generic/bionic/memcpy.S \
+    arch-arm/generic/bionic/memset.S \
+    arch-arm/generic/bionic/strcmp.S \
+    arch-arm/generic/bionic/strcpy.S \
+    arch-arm/generic/bionic/strlen.c \
+    bionic/__strcat_chk.cpp \
+    bionic/__strcpy_chk.cpp \
+
+libc_bionic_src_files_arm += \
+    arch-arm/cortex-a9/bionic/memcpy.S \
+    arch-arm/cortex-a9/bionic/memset.S \
+    arch-arm/cortex-a9/bionic/stpcpy.S \
+    arch-arm/cortex-a9/bionic/strcat.S \
+    arch-arm/cortex-a9/bionic/__strcat_chk.S \
+    arch-arm/cortex-a9/bionic/strcmp.S \
+    arch-arm/cortex-a9/bionic/strcpy.S \
+    arch-arm/cortex-a9/bionic/__strcpy_chk.S \
+    arch-arm/cortex-a9/bionic/strlen.S \
+
+libc_bionic_src_files_arm += \
+    arch-arm/denver/bionic/memmove.S \

libc/arch-arm/denver/bionic/__strcat_chk.S

@@ -0,0 +1,216 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <private/bionic_asm.h>
+#include <private/libc_events.h>
+
+    .syntax unified
+
+    .thumb
+    .thumb_func
+
+// Get the length of src string, then get the source of the dst string.
+// Check that the two lengths together don't exceed the threshold, then
+// do a memcpy of the data.
+ENTRY(__strcat_chk)
+    pld     [r0, #0]
+    push    {r0, lr}
+    .cfi_def_cfa_offset 8
+    .cfi_rel_offset r0, 0
+    .cfi_rel_offset lr, 4
+    push    {r4, r5}
+    .cfi_adjust_cfa_offset 8
+    .cfi_rel_offset r4, 0
+    .cfi_rel_offset r5, 4
+
+    mov     lr, r2
+
+    // Save the dst register to r5
+    mov     r5, r0
+
+    // Zero out r4
+    eor     r4, r4, r4
+
+    // r1 contains the address of the string to count.
+.L_strlen_start:
+    mov     r0, r1
+    ands    r3, r1, #7
+    beq     .L_mainloop
+
+    // Align to a double word (64 bits).
+    rsb     r3, r3, #8
+    lsls    ip, r3, #31
+    beq     .L_align_to_32
+
+    ldrb    r2, [r1], #1
+    cbz     r2, .L_update_count_and_finish
+
+.L_align_to_32:
+    bcc     .L_align_to_64
+    ands    ip, r3, #2
+    beq     .L_align_to_64
+
+    ldrb    r2, [r1], #1
+    cbz     r2, .L_update_count_and_finish
+    ldrb    r2, [r1], #1
+    cbz     r2, .L_update_count_and_finish
+
+.L_align_to_64:
+    tst     r3, #4
+    beq     .L_mainloop
+    ldr     r3, [r1], #4
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_second_register
+
+    .p2align 2
+.L_mainloop:
+    ldrd    r2, r3, [r1], #8
+
+    pld     [r1, #64]
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_first_register
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_second_register
+    b       .L_mainloop
+
+.L_update_count_and_finish:
+    sub     r3, r1, r0
+    sub     r3, r3, #1
+    b       .L_finish
+
+.L_zero_in_first_register:
+    sub     r3, r1, r0
+    lsls    r2, ip, #17
+    bne     .L_sub8_and_finish
+    bcs     .L_sub7_and_finish
+    lsls    ip, ip, #1
+    bne     .L_sub6_and_finish
+
+    sub     r3, r3, #5
+    b       .L_finish
+
+.L_sub8_and_finish:
+    sub     r3, r3, #8
+    b       .L_finish
+
+.L_sub7_and_finish:
+    sub     r3, r3, #7
+    b       .L_finish
+
+.L_sub6_and_finish:
+    sub     r3, r3, #6
+    b       .L_finish
+
+.L_zero_in_second_register:
+    sub     r3, r1, r0
+    lsls    r2, ip, #17
+    bne     .L_sub4_and_finish
+    bcs     .L_sub3_and_finish
+    lsls    ip, ip, #1
+    bne     .L_sub2_and_finish
+
+    sub     r3, r3, #1
+    b       .L_finish
+
+.L_sub4_and_finish:
+    sub     r3, r3, #4
+    b       .L_finish
+
+.L_sub3_and_finish:
+    sub     r3, r3, #3
+    b       .L_finish
+
+.L_sub2_and_finish:
+    sub     r3, r3, #2
+
+.L_finish:
+    cmp     r4, #0
+    bne     .L_strlen_done
+
+    // Time to get the dst string length.
+    mov     r1, r5
+
+    // Save the original source address to r5.
+    mov     r5, r0
+
+    // Save the current length (adding 1 for the terminator).
+    add     r4, r3, #1
+    b       .L_strlen_start
+
+    // r0 holds the pointer to the dst string.
+    // r3 holds the dst string length.
+    // r4 holds the src string length + 1.
+.L_strlen_done:
+    add     r2, r3, r4
+    cmp     r2, lr
+    bhi     __strcat_chk_failed
+
+    // Set up the registers for the memcpy code.
+    mov     r1, r5
+    pld     [r1, #64]
+    mov     r2, r4
+    add     r0, r0, r3
+    pop     {r4, r5}
+END(__strcat_chk)
+
+#define MEMCPY_BASE         __strcat_chk_memcpy_base
+#define MEMCPY_BASE_ALIGNED __strcat_chk_memcpy_base_aligned
+
+#include "memcpy_base.S"
+
+ENTRY_PRIVATE(__strcat_chk_failed)
+    .cfi_def_cfa_offset 8
+    .cfi_rel_offset r0, 0
+    .cfi_rel_offset lr, 4
+    .cfi_adjust_cfa_offset 8
+    .cfi_rel_offset r4, 0
+    .cfi_rel_offset r5, 4
+
+    ldr     r0, error_message
+    ldr     r1, error_code
+1:
+    add     r0, pc
+    bl      __fortify_chk_fail
+error_code:
+    .word   BIONIC_EVENT_STRCAT_BUFFER_OVERFLOW
+error_message:
+    .word   error_string-(1b+4)
+END(__strcat_chk_failed)
+
+    .data
+error_string:
+    .string "strcat: prevented write past end of buffer"

libc/arch-arm/denver/bionic/__strcpy_chk.S

@@ -0,0 +1,180 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <private/bionic_asm.h>
+#include <private/libc_events.h>
+
+    .syntax unified
+
+    .thumb
+    .thumb_func
+
+// Get the length of the source string first, then do a memcpy of the data
+// instead of a strcpy.
+ENTRY(__strcpy_chk)
+    pld     [r0, #0]
+    push    {r0, lr}
+    .cfi_def_cfa_offset 8
+    .cfi_rel_offset r0, 0
+    .cfi_rel_offset lr, 4
+
+    mov     lr, r2
+    mov     r0, r1
+
+    ands    r3, r1, #7
+    beq     .L_mainloop
+
+    // Align to a double word (64 bits).
+    rsb     r3, r3, #8
+    lsls    ip, r3, #31
+    beq     .L_align_to_32
+
+    ldrb    r2, [r0], #1
+    cbz     r2, .L_update_count_and_finish
+
+.L_align_to_32:
+    bcc     .L_align_to_64
+    ands    ip, r3, #2
+    beq     .L_align_to_64
+
+    ldrb    r2, [r0], #1
+    cbz     r2, .L_update_count_and_finish
+    ldrb    r2, [r0], #1
+    cbz     r2, .L_update_count_and_finish
+
+.L_align_to_64:
+    tst     r3, #4
+    beq     .L_mainloop
+    ldr     r3, [r0], #4
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_second_register
+
+    .p2align 2
+.L_mainloop:
+    ldrd    r2, r3, [r0], #8
+
+    pld     [r0, #64]
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_first_register
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_second_register
+    b       .L_mainloop
+
+.L_update_count_and_finish:
+    sub     r3, r0, r1
+    sub     r3, r3, #1
+    b       .L_check_size
+
+.L_zero_in_first_register:
+    sub     r3, r0, r1
+    lsls    r2, ip, #17
+    bne     .L_sub8_and_finish
+    bcs     .L_sub7_and_finish
+    lsls    ip, ip, #1
+    bne     .L_sub6_and_finish
+
+    sub     r3, r3, #5
+    b       .L_check_size
+
+.L_sub8_and_finish:
+    sub     r3, r3, #8
+    b       .L_check_size
+
+.L_sub7_and_finish:
+    sub     r3, r3, #7
+    b       .L_check_size
+
+.L_sub6_and_finish:
+    sub     r3, r3, #6
+    b       .L_check_size
+
+.L_zero_in_second_register:
+    sub     r3, r0, r1
+    lsls    r2, ip, #17
+    bne     .L_sub4_and_finish
+    bcs     .L_sub3_and_finish
+    lsls    ip, ip, #1
+    bne     .L_sub2_and_finish
+
+    sub     r3, r3, #1
+    b       .L_check_size
+
+.L_sub4_and_finish:
+    sub     r3, r3, #4
+    b       .L_check_size
+
+.L_sub3_and_finish:
+    sub     r3, r3, #3
+    b       .L_check_size
+
+.L_sub2_and_finish:
+    sub     r3, r3, #2
+
+.L_check_size:
+    pld     [r1, #0]
+    pld     [r1, #64]
+    ldr     r0, [sp]
+    cmp     r3, lr
+    bhs     __strcpy_chk_failed
+
+    // Add 1 for copy length to get the string terminator.
+    add     r2, r3, #1
+END(__strcpy_chk)
+
+#define MEMCPY_BASE         __strcpy_chk_memcpy_base
+#define MEMCPY_BASE_ALIGNED __strcpy_chk_memcpy_base_aligned
+#include "memcpy_base.S"
+
+ENTRY_PRIVATE(__strcpy_chk_failed)
+    .cfi_def_cfa_offset 8
+    .cfi_rel_offset r0, 0
+    .cfi_rel_offset lr, 4
+
+    ldr     r0, error_message
+    ldr     r1, error_code
+1:
+    add     r0, pc
+    bl      __fortify_chk_fail
+error_code:
+    .word   BIONIC_EVENT_STRCPY_BUFFER_OVERFLOW
+error_message:
+    .word   error_string-(1b+4)
+END(__strcpy_chk_failed)
+
+    .data
+error_string:
+    .string "strcpy: prevented write past end of buffer"

libc/arch-arm/denver/bionic/memcpy.S

@@ -0,0 +1,103 @@
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+/*
+ * Copyright (c) 2013 ARM Ltd
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. The name of the company may not be used to endorse or promote
+ *    products derived from this software without specific prior written
+ *    permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+// Prototype: void *memcpy (void *dst, const void *src, size_t count).
+
+#include <private/bionic_asm.h>
+#include <private/libc_events.h>
+
+        .text
+        .syntax unified
+        .fpu    neon
+
+ENTRY(__memcpy_chk)
+        cmp     r2, r3
+        bhi     __memcpy_chk_fail
+
+        // Fall through to memcpy...
+END(__memcpy_chk)
+
+ENTRY(memcpy)
+        pld     [r1, #64]
+        push    {r0, lr}
+        .cfi_def_cfa_offset 8
+        .cfi_rel_offset r0, 0
+        .cfi_rel_offset lr, 4
+END(memcpy)
+
+#define MEMCPY_BASE         __memcpy_base
+#define MEMCPY_BASE_ALIGNED __memcpy_base_aligned
+#include "memcpy_base.S"
+
+ENTRY_PRIVATE(__memcpy_chk_fail)
+        // Preserve lr for backtrace.
+        push    {lr}
+        .cfi_def_cfa_offset 4
+        .cfi_rel_offset lr, 0
+
+        ldr     r0, error_message
+        ldr     r1, error_code
+1:
+        add     r0, pc
+        bl      __fortify_chk_fail
+error_code:
+        .word   BIONIC_EVENT_MEMCPY_BUFFER_OVERFLOW
+error_message:
+        .word   error_string-(1b+8)
+END(__memcpy_chk_fail)
+
+        .data
+error_string:
+        .string "memcpy: prevented write past end of buffer"