Merge "Correct way to specify additional dependencies"

Previous one was not covering all the targets

Bug: 17548097
Change-Id: I425e0f8d6e5a43feabd29dcb327049b82754d249

.gitignore

@@ -1,2 +1,3 @@
 *.pyc
+*.*~
 libc/kernel/original

ABI-bugs.txt

@@ -0,0 +1,12 @@
+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
+
+  atexit(3) handlers registered by a shared library aren't called on
+  dlclose(3); this only affects ARM. http://b/4998315

Android.mk

@@ -14,6 +14,6 @@
 # limitations under the License.
 #
 
-ifneq ($(TARGET_SIMULATOR),true)
-  include $(call all-subdir-makefiles)
-endif
+LOCAL_PATH := $(call my-dir)
+
+include $(call all-subdir-makefiles)

CleanSpec.mk

@@ -0,0 +1,53 @@
+# Copyright (C) 2007 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.
+#
+
+# If you don't need to do a full clean build but would like to touch
+# a file or delete some intermediate files, add a clean step to the end
+# of the list.  These steps will only be run once, if they haven't been
+# run before.
+#
+# E.g.:
+#     $(call add-clean-step, touch -c external/sqlite/sqlite3.h)
+#     $(call add-clean-step, rm -rf $(PRODUCT_OUT)/obj/STATIC_LIBRARIES/libz_intermediates)
+#
+# Always use "touch -c" and "rm -f" or "rm -rf" to gracefully deal with
+# files that are missing or have been moved.
+#
+# Use $(PRODUCT_OUT) to get to the "out/target/product/blah/" directory.
+# Use $(OUT_DIR) to refer to the "out" directory.
+#
+# If you need to re-do something that's already mentioned, just copy
+# the command and add it to the bottom of the list.  E.g., if a change
+# that you made last week required touching a file and a change you
+# made today requires touching the same file, just copy the old
+# touch step and add it to the end of the list.
+#
+# ************************************************
+# NEWER CLEAN STEPS MUST BE AT THE END OF THE LIST
+# ************************************************
+
+# For example:
+#$(call add-clean-step, rm -rf $(OUT_DIR)/target/common/obj/APPS/AndroidTests_intermediates)
+#$(call add-clean-step, rm -rf $(OUT_DIR)/target/common/obj/JAVA_LIBRARIES/core_intermediates)
+#$(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 +1,246 @@
+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:
+
+  1. Run update-tzdata.py.
+
+
+Running the tests
+-----------------
+
+The tests are all built from the tests/ directory.
+
+### Device tests
+
+    $ mma
+    $ 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/nativetest/bionic-unit-tests/bionic-unit-tests64
+    $ adb shell \
+        /data/nativetest/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
+    # 64-bit tests for 64-bit targets, 32-bit otherwise.
+    $ mm bionic-unit-tests-run-on-host
+    # Only exists for 64-bit targets.
+    $ mm bionic-unit-tests-run-on-host32
+
+### 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.
+
+    $ mma
+    $ bionic-unit-tests-glibc32 # already in your path
+    $ 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`.

benchmarks/Android.mk

@@ -0,0 +1,79 @@
+#
+# 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.
+#
+
+ifneq ($(BUILD_TINY_ANDROID), true)
+
+LOCAL_PATH := $(call my-dir)
+
+# -----------------------------------------------------------------------------
+# Benchmarks.
+# -----------------------------------------------------------------------------
+
+benchmark_c_flags = \
+    -O2 \
+    -Wall -Wextra -Wunused \
+    -Werror \
+    -fno-builtin \
+    -std=gnu++11 \
+
+benchmark_src_files = \
+    benchmark_main.cpp \
+    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-benchmarks
+include $(CLEAR_VARS)
+LOCAL_MODULE := bionic-benchmarks
+LOCAL_MODULE_STEM_32 := bionic-benchmarks32
+LOCAL_MODULE_STEM_64 := bionic-benchmarks64
+LOCAL_MULTILIB := both
+LOCAL_ADDITIONAL_DEPENDENCIES := $(LOCAL_PATH)/Android.mk
+LOCAL_CFLAGS += $(benchmark_c_flags)
+LOCAL_SRC_FILES := $(benchmark_src_files)
+LOCAL_CXX_STL := libc++
+include $(BUILD_EXECUTABLE)
+
+ifeq ($(HOST_OS)-$(HOST_ARCH),$(filter $(HOST_OS)-$(HOST_ARCH),linux-x86 linux-x86_64))
+ifeq ($(TARGET_ARCH),x86)
+LINKER = linker
+NATIVE_SUFFIX=32
+else
+LINKER = linker64
+NATIVE_SUFFIX=64
+endif
+
+bionic-benchmarks-run-on-host: bionic-benchmarks $(TARGET_OUT_EXECUTABLES)/$(LINKER) $(TARGET_OUT_EXECUTABLES)/sh
+	if [ ! -d /system -o ! -d /system/bin ]; then \
+	  echo "Attempting to create /system/bin"; \
+	  sudo mkdir -p -m 0777 /system/bin; \
+	fi
+	mkdir -p $(TARGET_OUT_DATA)/local/tmp
+	cp $(TARGET_OUT_EXECUTABLES)/$(LINKER) /system/bin
+	cp $(TARGET_OUT_EXECUTABLES)/sh /system/bin
+	ANDROID_DATA=$(TARGET_OUT_DATA) \
+	ANDROID_ROOT=$(TARGET_OUT) \
+	LD_LIBRARY_PATH=$(TARGET_OUT_SHARED_LIBRARIES) \
+		$(TARGET_OUT_EXECUTABLES)/bionic-benchmarks$(NATIVE_SUFFIX) $(BIONIC_BENCHMARKS_FLAGS)
+endif # linux-x86
+
+endif # !BUILD_TINY_ANDROID

benchmarks/benchmark.h

@@ -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 <stdint.h>
+
+#include <vector>
+
+namespace testing {
+
+class Benchmark {
+ public:
+  Benchmark(const char* name, void (*fn)(int)) {
+    Register(name, fn, NULL);
+  }
+
+  Benchmark(const char* name, void (*fn_range)(int, int)) {
+    Register(name, NULL, fn_range);
+  }
+
+  Benchmark* Arg(int x);
+
+  const char* Name();
+
+  bool ShouldRun(int argc, char* argv[]);
+  void Run();
+
+ private:
+  const char* name_;
+
+  void (*fn_)(int);
+  void (*fn_range_)(int, int);
+
+  std::vector<int> args_;
+
+  void Register(const char* name, void (*fn)(int), void (*fn_range)(int, int));
+  void RunRepeatedlyWithArg(int iterations, int arg);
+  void RunWithArg(int arg);
+};
+
+}  // namespace testing
+
+void SetBenchmarkBytesProcessed(int64_t);
+void StopBenchmarkTiming();
+void StartBenchmarkTiming();
+
+#define BENCHMARK(f) \
+    static ::testing::Benchmark* _benchmark_##f __attribute__((unused)) = \
+        (new ::testing::Benchmark(#f, f))

benchmarks/benchmark_main.cpp

@@ -0,0 +1,226 @@
+/*
+ * 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 "benchmark.h"
+
+#include <regex.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <string>
+#include <map>
+
+#include <inttypes.h>
+
+static int64_t g_bytes_processed;
+static int64_t g_benchmark_total_time_ns;
+static int64_t g_benchmark_start_time_ns;
+
+typedef std::map<std::string, ::testing::Benchmark*> BenchmarkMap;
+typedef BenchmarkMap::iterator BenchmarkMapIt;
+static BenchmarkMap g_benchmarks;
+static int g_name_column_width = 20;
+
+static 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;
+}
+
+static int64_t NanoTime() {
+  struct timespec t;
+  t.tv_sec = t.tv_nsec = 0;
+  clock_gettime(CLOCK_MONOTONIC, &t);
+  return static_cast<int64_t>(t.tv_sec) * 1000000000LL + t.tv_nsec;
+}
+
+namespace testing {
+
+Benchmark* Benchmark::Arg(int arg) {
+  args_.push_back(arg);
+  return this;
+}
+
+const char* Benchmark::Name() {
+  return name_;
+}
+
+bool Benchmark::ShouldRun(int argc, char* argv[]) {
+  if (argc == 1) {
+    return true;  // With no arguments, we run all benchmarks.
+  }
+  // Otherwise, we interpret each argument as a regular expression and
+  // see if any of our benchmarks match.
+  for (int i = 1; i < argc; i++) {
+    regex_t re;
+    if (regcomp(&re, argv[i], 0) != 0) {
+      fprintf(stderr, "couldn't compile \"%s\" as a regular expression!\n", argv[i]);
+      exit(EXIT_FAILURE);
+    }
+    int match = regexec(&re, name_, 0, NULL, 0);
+    regfree(&re);
+    if (match != REG_NOMATCH) {
+      return true;
+    }
+  }
+  return false;
+}
+
+void Benchmark::Register(const char* name, void (*fn)(int), void (*fn_range)(int, int)) {
+  name_ = name;
+  fn_ = fn;
+  fn_range_ = fn_range;
+
+  if (fn_ == NULL && fn_range_ == NULL) {
+    fprintf(stderr, "%s: missing function\n", name_);
+    exit(EXIT_FAILURE);
+  }
+
+  g_benchmarks.insert(std::make_pair(name, this));
+}
+
+void Benchmark::Run() {
+  if (fn_ != NULL) {
+    RunWithArg(0);
+  } else {
+    if (args_.empty()) {
+      fprintf(stderr, "%s: no args!\n", name_);
+      exit(EXIT_FAILURE);
+    }
+    for (size_t i = 0; i < args_.size(); ++i) {
+      RunWithArg(args_[i]);
+    }
+  }
+}
+
+void Benchmark::RunRepeatedlyWithArg(int iterations, int arg) {
+  g_bytes_processed = 0;
+  g_benchmark_total_time_ns = 0;
+  g_benchmark_start_time_ns = NanoTime();
+  if (fn_ != NULL) {
+    fn_(iterations);
+  } else {
+    fn_range_(iterations, arg);
+  }
+  if (g_benchmark_start_time_ns != 0) {
+    g_benchmark_total_time_ns += NanoTime() - g_benchmark_start_time_ns;
+  }
+}
+
+void Benchmark::RunWithArg(int arg) {
+  // run once in case it's expensive
+  int iterations = 1;
+  RunRepeatedlyWithArg(iterations, arg);
+  while (g_benchmark_total_time_ns < 1e9 && iterations < 1e9) {
+    int last = iterations;
+    if (g_benchmark_total_time_ns/iterations == 0) {
+      iterations = 1e9;
+    } else {
+      iterations = 1e9 / (g_benchmark_total_time_ns/iterations);
+    }
+    iterations = std::max(last + 1, std::min(iterations + iterations/2, 100*last));
+    iterations = Round(iterations);
+    RunRepeatedlyWithArg(iterations, arg);
+  }
+
+  char throughput[100];
+  throughput[0] = '\0';
+  if (g_benchmark_total_time_ns > 0 && g_bytes_processed > 0) {
+    double mib_processed = static_cast<double>(g_bytes_processed)/1e6;
+    double seconds = static_cast<double>(g_benchmark_total_time_ns)/1e9;
+    snprintf(throughput, sizeof(throughput), " %8.2f MiB/s", mib_processed/seconds);
+  }
+
+  char full_name[100];
+  if (fn_range_ != NULL) {
+    if (arg >= (1<<20)) {
+      snprintf(full_name, sizeof(full_name), "%s/%dM", name_, arg/(1<<20));
+    } else if (arg >= (1<<10)) {
+      snprintf(full_name, sizeof(full_name), "%s/%dK", name_, arg/(1<<10));
+    } else {
+      snprintf(full_name, sizeof(full_name), "%s/%d", name_, arg);
+    }
+  } else {
+    snprintf(full_name, sizeof(full_name), "%s", name_);
+  }
+
+  printf("%-*s %10d %10" PRId64 "%s\n", g_name_column_width, full_name,
+         iterations, g_benchmark_total_time_ns/iterations, throughput);
+  fflush(stdout);
+}
+
+}  // namespace testing
+
+void SetBenchmarkBytesProcessed(int64_t x) {
+  g_bytes_processed = x;
+}
+
+void StopBenchmarkTiming() {
+  if (g_benchmark_start_time_ns != 0) {
+    g_benchmark_total_time_ns += NanoTime() - g_benchmark_start_time_ns;
+  }
+  g_benchmark_start_time_ns = 0;
+}
+
+void StartBenchmarkTiming() {
+  if (g_benchmark_start_time_ns == 0) {
+    g_benchmark_start_time_ns = NanoTime();
+  }
+}
+
+int main(int argc, char* argv[]) {
+  if (g_benchmarks.empty()) {
+    fprintf(stderr, "No benchmarks registered!\n");
+    exit(EXIT_FAILURE);
+  }
+
+  for (BenchmarkMapIt it = g_benchmarks.begin(); it != g_benchmarks.end(); ++it) {
+    int name_width = static_cast<int>(strlen(it->second->Name()));
+    g_name_column_width = std::max(g_name_column_width, name_width);
+  }
+
+  bool need_header = true;
+  for (BenchmarkMapIt it = g_benchmarks.begin(); it != g_benchmarks.end(); ++it) {
+    ::testing::Benchmark* b = it->second;
+    if (b->ShouldRun(argc, argv)) {
+      if (need_header) {
+        printf("%-*s %10s %10s\n", g_name_column_width, "", "iterations", "ns/op");
+        fflush(stdout);
+        need_header = false;
+      }
+      b->Run();
+    }
+  }
+
+  if (need_header) {
+    fprintf(stderr, "No matching benchmarks!\n");
+    fprintf(stderr, "Available benchmarks:\n");
+    for (BenchmarkMapIt it = g_benchmarks.begin(); it != g_benchmarks.end(); ++it) {
+      fprintf(stderr, "  %s\n", it->second->Name());
+    }
+    exit(EXIT_FAILURE);
+  }
+
+  return 0;
+}

benchmarks/math_benchmark.cpp

@@ -0,0 +1,211 @@
+/*
+ * 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 "benchmark.h"
+
+#include <fenv.h>
+#include <math.h>
+
+// Avoid optimization.
+volatile double d;
+volatile double v;
+
+static void BM_math_sqrt(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = 2.0;
+  for (int i = 0; i < iters; ++i) {
+    d += sqrt(v);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_math_sqrt);
+
+static void BM_math_log10(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = 1234.0;
+  for (int i = 0; i < iters; ++i) {
+    d += log10(v);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_math_log10);
+
+static void BM_math_logb(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = 1234.0;
+  for (int i = 0; i < iters; ++i) {
+    d += logb(v);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_math_logb);
+
+static void BM_math_isinf_NORMAL(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = 1234.0; // FP_NORMAL
+  for (int i = 0; i < iters; ++i) {
+    d += (isinf)(v);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_math_isinf_NORMAL);
+
+static void BM_math_isinf_NAN(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = nan(""); // FP_NAN
+  for (int i = 0; i < iters; ++i) {
+    d += (isinf)(v);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_math_isinf_NAN);
+
+static void BM_math_isinf_INFINITE(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = HUGE_VAL; // FP_INFINITE
+  for (int i = 0; i < iters; ++i) {
+    d += (isinf)(v);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_math_isinf_INFINITE);
+
+static void BM_math_isinf_ZERO(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = 0.0; // FP_ZERO
+  for (int i = 0; i < iters; ++i) {
+    d += (isinf)(v);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_math_isinf_ZERO);
+
+static void BM_math_sin_fast(int iters) {
+  StartBenchmarkTiming();
+
+  d = 1.0;
+  for (int i = 0; i < iters; ++i) {
+    d += sin(d);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_math_sin_fast);
+
+static void BM_math_sin_feupdateenv(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(BM_math_sin_feupdateenv);
+
+static void BM_math_sin_fesetenv(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(BM_math_sin_fesetenv);
+
+static void BM_math_fpclassify_NORMAL(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = 1234.0; // FP_NORMAL
+  for (int i = 0; i < iters; ++i) {
+    d += fpclassify(v);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_math_fpclassify_NORMAL);
+
+static void BM_math_fpclassify_NAN(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = nan(""); // FP_NAN
+  for (int i = 0; i < iters; ++i) {
+    d += fpclassify(v);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_math_fpclassify_NAN);
+
+static void BM_math_fpclassify_INFINITE(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = HUGE_VAL; // FP_INFINITE
+  for (int i = 0; i < iters; ++i) {
+    d += fpclassify(v);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_math_fpclassify_INFINITE);
+
+static void BM_math_fpclassify_ZERO(int iters) {
+  StartBenchmarkTiming();
+
+  d = 0.0;
+  v = 0.0; // FP_ZERO
+  for (int i = 0; i < iters; ++i) {
+    d += fpclassify(v);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_math_fpclassify_ZERO);

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 "benchmark.h"
+#include <errno.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+
+#define _REALLY_INCLUDE_SYS__SYSTEM_PROPERTIES_H_
+#include <sys/_system_properties.h>
+
+#include <vector>
+#include <string>
+
+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;
+};
+
+static void BM_property_get(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(BM_property_get)->TEST_NUM_PROPS;
+
+static void BM_property_find(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(BM_property_find)->TEST_NUM_PROPS;
+
+static void BM_property_read(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(BM_property_read)->TEST_NUM_PROPS;
+
+static void BM_property_serial(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;
+}
+BENCHMARK(BM_property_serial)->TEST_NUM_PROPS;

benchmarks/pthread_benchmark.cpp

@@ -0,0 +1,139 @@
+/*
+ * 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 "benchmark.h"
+
+#include <pthread.h>
+
+// Stop GCC optimizing out our pure function.
+/* Must not be static! */ pthread_t (*pthread_self_fp)() = pthread_self;
+
+static void BM_pthread_self(int iters) {
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_self_fp();
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_pthread_self);
+
+static void BM_pthread_getspecific(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(BM_pthread_getspecific);
+
+static void DummyPthreadOnceInitFunction() {
+}
+
+static void BM_pthread_once(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(BM_pthread_once);
+
+static void BM_pthread_mutex_lock(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(BM_pthread_mutex_lock);
+
+static void BM_pthread_mutex_lock_ERRORCHECK(int iters) {
+  StopBenchmarkTiming();
+  pthread_mutex_t mutex = PTHREAD_ERRORCHECK_MUTEX_INITIALIZER;
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_mutex_lock(&mutex);
+    pthread_mutex_unlock(&mutex);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_pthread_mutex_lock_ERRORCHECK);
+
+static void BM_pthread_mutex_lock_RECURSIVE(int iters) {
+  StopBenchmarkTiming();
+  pthread_mutex_t mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER;
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    pthread_mutex_lock(&mutex);
+    pthread_mutex_unlock(&mutex);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_pthread_mutex_lock_RECURSIVE);
+
+static void BM_pthread_rw_lock_read(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(BM_pthread_rw_lock_read);
+
+static void BM_pthread_rw_lock_write(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);
+}
+BENCHMARK(BM_pthread_rw_lock_write);

benchmarks/semaphore_benchmark.cpp

@@ -0,0 +1,49 @@
+/*
+ * 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 "benchmark.h"
+
+#include <semaphore.h>
+
+static void BM_semaphore_sem_getvalue(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(BM_semaphore_sem_getvalue);
+
+static void BM_semaphore_sem_wait_sem_post(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();
+}
+BENCHMARK(BM_semaphore_sem_wait_sem_post);

benchmarks/stdio_benchmark.cpp

@@ -0,0 +1,61 @@
+/*
+ * 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 "benchmark.h"
+
+#include <stdio.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)
+
+static void BM_stdio_fread(int iters, int chunk_size) {
+  StopBenchmarkTiming();
+  FILE* fp = fopen("/dev/zero", "rw");
+  char* buf = new char[chunk_size];
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    fread(buf, chunk_size, 1, fp);
+  }
+
+  StopBenchmarkTiming();
+  SetBenchmarkBytesProcessed(int64_t(iters) * int64_t(chunk_size));
+  delete[] buf;
+  fclose(fp);
+}
+BENCHMARK(BM_stdio_fread)->AT_COMMON_SIZES;
+
+
+static void BM_stdio_fwrite(int iters, int chunk_size) {
+  StopBenchmarkTiming();
+  FILE* fp = fopen("/dev/zero", "rw");
+  char* buf = new char[chunk_size];
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+      fwrite(buf, chunk_size, 1, fp);
+  }
+
+  StopBenchmarkTiming();
+  SetBenchmarkBytesProcessed(int64_t(iters) * int64_t(chunk_size));
+  delete[] buf;
+  fclose(fp);
+}
+BENCHMARK(BM_stdio_fwrite)->AT_COMMON_SIZES;

benchmarks/string_benchmark.cpp

@@ -0,0 +1,112 @@
+/*
+ * 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 "benchmark.h"
+
+#include <string.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.)
+
+static void BM_string_memcmp(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(int64_t(iters) * int64_t(nbytes));
+  delete[] src;
+  delete[] dst;
+}
+BENCHMARK(BM_string_memcmp)->AT_COMMON_SIZES;
+
+static void BM_string_memcpy(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(int64_t(iters) * int64_t(nbytes));
+  delete[] src;
+  delete[] dst;
+}
+BENCHMARK(BM_string_memcpy)->AT_COMMON_SIZES;
+
+static void BM_string_memmove(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(int64_t(iters) * int64_t(nbytes));
+  delete[] buf;
+}
+BENCHMARK(BM_string_memmove)->AT_COMMON_SIZES;
+
+static void BM_string_memset(int iters, int nbytes) {
+  StopBenchmarkTiming();
+  char* dst = new char[nbytes];
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    memset(dst, 0, nbytes);
+  }
+
+  StopBenchmarkTiming();
+  SetBenchmarkBytesProcessed(int64_t(iters) * int64_t(nbytes));
+  delete[] dst;
+}
+BENCHMARK(BM_string_memset)->AT_COMMON_SIZES;
+
+static void BM_string_strlen(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(int64_t(iters) * int64_t(nbytes));
+  delete[] s;
+}
+BENCHMARK(BM_string_strlen)->AT_COMMON_SIZES;

benchmarks/time_benchmark.cpp

@@ -0,0 +1,79 @@
+/*
+ * 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 "benchmark.h"
+
+#include <sys/syscall.h>
+#include <time.h>
+
+static void BM_time_clock_gettime(int iters) {
+  StartBenchmarkTiming();
+
+  timespec t;
+  for (int i = 0; i < iters; ++i) {
+    clock_gettime(CLOCK_MONOTONIC, &t);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_time_clock_gettime);
+
+static void BM_time_clock_gettime_syscall(int iters) {
+  StartBenchmarkTiming();
+
+  timespec t;
+  for (int i = 0; i < iters; ++i) {
+    syscall(__NR_clock_gettime, CLOCK_MONOTONIC, &t);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_time_clock_gettime_syscall);
+
+static void BM_time_gettimeofday(int iters) {
+  StartBenchmarkTiming();
+
+  timeval tv;
+  for (int i = 0; i < iters; ++i) {
+    gettimeofday(&tv, NULL);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_time_gettimeofday);
+
+static void BM_time_gettimeofday_syscall(int iters) {
+  StartBenchmarkTiming();
+
+  timeval tv;
+  for (int i = 0; i < iters; ++i) {
+    syscall(__NR_gettimeofday, &tv, NULL);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_time_gettimeofday_syscall);
+
+static void BM_time_time(int iters) {
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    time(NULL);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_time_time);

benchmarks/unistd_benchmark.cpp

@@ -0,0 +1,67 @@
+/*
+ * 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 "benchmark.h"
+
+#include <sys/syscall.h>
+#include <unistd.h>
+
+static void BM_unistd_getpid(int iters) {
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    getpid();
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_unistd_getpid);
+
+static void BM_unistd_getpid_syscall(int iters) {
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    syscall(__NR_getpid);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_unistd_getpid_syscall);
+
+// Stop GCC optimizing out our pure function.
+/* Must not be static! */ pid_t (*gettid_fp)() = gettid;
+
+static void BM_unistd_gettid(int iters) {
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    gettid_fp();
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_unistd_gettid);
+
+static void BM_unistd_gettid_syscall(int iters) {
+  StartBenchmarkTiming();
+
+  for (int i = 0; i < iters; ++i) {
+    syscall(__NR_gettid);
+  }
+
+  StopBenchmarkTiming();
+}
+BENCHMARK(BM_unistd_gettid_syscall);

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/Jamfile

@@ -1,441 +0,0 @@
-# This file is used to build the Bionic library with the Jam build
-# tool. For info, see www.perforce.com/jam/jam.html
-#
-
-BIONIC_TOP ?= $(DOT) ;
-
-DEBUG = 1 ;
-
-# pattern used for automatic heade inclusion detection
-HDRPATTERN = "^[ 	]*#[ 	]*include[ 	]*[<\"]([^\">]*)[\">].*$" ;
-
-
-# debugging support, simply define the DEBUG variable to activate verbose output
-rule Debug
-{
-    if $(DEBUG) {
-        Echo $(1) ;
-    }
-}
-
-# return all elements from $(1) that are not in $(2)
-rule Filter  list : filter
-{
-    local result = ;
-    local item ;
-    for item in $(list) {
-        if ! $(item) in $(filter) {
-            result += $(item) ;
-        }
-    }
-    return $(result) ;
-}
-
-
-# reverse a list of elements
-rule Reverse  list
-{
-    local  result = ;
-    local  item ;
-
-    for item in $(list) {
-        result = $(item) $(result) ;
-    }
-    return $(result) ;
-}
-
-
-# decompose a path into a list of elements
-rule PathDecompose  dir
-{
-    local  result ;
-
-    while $(dir:D)
-    {
-        if ! $(dir:BS) {  # for rooted paths like "/foo"
-            break ;
-        }
-        result = $(dir:BS) $(result) ;
-        dir    = $(dir:D) ;
-    }
-    result = $(dir) $(result) ;
-    return $(result) ;
-}
-
-
-# simply a file path, i.e. get rid of . or .. when possible
-rule _PathSimplify  dir
-{
-    local  result = ;
-    local  dir2 d ;
-
-    dir  = [ PathDecompose $(dir) ] ;
-
-    # get rid of any single dot
-    dir2 = ;
-    for d in $(dir) {
-        if $(d) = "." {
-            continue ;
-        }
-        dir2 += $(d) ;
-    }
-
-    # get rid of .. when possible
-    for d in $(dir2) {
-        if $(d) = ".." && $(result) {
-            result = $(result[2-]) ;
-        }
-        else
-            result = $(d) $(result) ;
-    }
-
-    # now invert the result
-    result = [ Reverse $(result) ] ;
-    if ! $(result) {
-        result = "." ;
-    }
-    return $(result:J="/") ;
-}
-
-
-rule PathSimplify  dirs
-{
-    local result ;
-    local d ;
-    for d in $(dirs) {
-        result += [ _PathSimplify $(d) ] ;
-    }
-    return $(result) ;
-}
-
-
-# retrieve list of subdirectories
-rule ListSubDirs  paths
-{
-    local  result  = ;
-    local  entry ;
-    for entry in [ Glob $(paths) : * ] {
-        if ! $(entry:S) {
-            result += $(entry) ;
-        }
-    }
-    return [ PathSimplify $(result) ] ;
-}
-
-
-# retrieve list of sources in a given directory
-rule ListSources  path
-{
-    return [ Glob $(path) : *.S *.c ] ;
-}
-
-
-# find the prebuilt directory
-#
-if ! $(TOP) {
-    Echo "Please define TOP as the root of your device build tree" ;
-    Exit ;
-}
-
-Debug "OS is" $(OS) ;
-Debug "CPU is" $(CPU) ;
-
-if $(OS) = LINUX
-{
-    PREBUILT = $(TOP)/prebuilt/Linux ;
-}
-else if $(OS) = MACOSX
-{
-    switch $(CPU) {
-        case i386 : PREBUILT = $(TOP)/prebuilt/darwin-x86 ; break ;
-        case ppc  : PREBUILT = $(TOP)/prebuilt/darwin-ppc ; break ;
-        case *    : Echo "unsupported CPU" "$(CPU) !!" ;
-                    Echo "Please contact digit@google.com for help" ;
-                    Exit ;
-    }
-}
-else
-{
-    Echo "Unsupported operating system" $(OS) ;
-    Echo "Please contact digit@google.com for help" ;
-    Exit ;
-}
-
-Debug "TOP is" $(TOP) ;
-Debug "PREBUILT is" $(PREBUILT) ;
-
-
-# check architectures and setup toolchain variables
-#
-SUPPORTED_ARCHS = x86 arm ;
-
-ARCH ?= $(SUPPORTED_ARCHS) ;
-
-if ! $(ARCH) in $(SUPPORTED_ARCHS) {
-    Echo "The variable ARCH contains an unsupported value, use one or more of these instead" ;
-    Echo "separated by spaces:" $(SUPPORTED_ARCHS) ;
-    Exit ;
-}
-
-x86_TOOLSET_PREFIX ?= "" ;
-arm_TOOLSET_PREFIX ?= $(TOP)/prebuilt/Linux/toolchain-4.1.1/bin/arm-elf- ;
-
-for arch in $(ARCH) {
-    CC_$(arch)  = $($(arch)_TOOLSET_PREFIX)gcc ;
-    C++_$(arch) = $($(arch)_TOOLSET_PREFIX)g++ ;
-    AR_$(arch)  = $($(arch)_TOOLSET_PREFIX)ar ;
-}
-
-
-# the list of arch-independent source subdirectories
-BIONIC_SRC_SUBDIRS = string ;
-BIONIC_x86_SUBDIRS = ;
-BIONIC_arm_SUBDIRS = ;
-
-CFLAGS   = -O0 -g -W ;
-
-
-
-# find sources in a given list of subdirectories
-rule FindSources  dirs
-{
-    local dir ;
-
-    for dir in $(dirs)
-    {
-        local LOCAL_SRC NO_LOCAL_SRC ;
-
-        if [ Glob $(dir) : rules.jam ] {
-            include $(dir)/rules.jam ;
-            if $(LOCAL_SRC) {
-                _sources = $(LOCAL_SRC) ;
-            }
-            else {
-                _sources = [ Glob $(dir) : *.S *.c ] ;
-                _sources = $(_sources:BS) ;
-            }
-            if $(NO_LOCAL_SRC) {
-                _sources = [ Filter $(_sources) : $(NO_LOCAL_SRC) ] ;
-            }
-            sources += $(dir)/$(_sources) ;
-        }
-        else
-            sources += [ ListSources $(dir) ] ;
-    }
-}
-
-# Compile a given object file from a source
-rule Compile  object : source
-{
-    Depends $(object) : $(source) ;
-    Depends bionic : $(object) ;
-    Clean clean : $(object) ;
-
-    MakeLocate $(object) : $(OUT) ;
-
-
-    CC on $(object)       = $(CC_$(arch)) ;
-    CFLAGS on $(object)   = $(CFLAGS) ;
-    INCLUDES on $(object) = $(INCLUDES) ;
-    DEFINES on $(object)  = $(DEFINES) ;
-
-    HDRRULE on $(>) = HdrRule ;
-    HDRSCAN on $(>) = $(HDRPATTERN) ;
-    HDRSEARCH on $(>) = $(INCLUDES) ;
-    HDRGRIST on $(>) = $(HDRGRIST) ;
-}
-
-
-actions Compile
-{
-    $(CC) -c -o $(1) $(CFLAGS) -I$(INCLUDES) -D$(DEFINES) $(2)
-}
-
-
-rule RmTemps
-{
-    Temporary $(2) ;
-}
-
-actions quietly updated piecemeal together RmTemps
-{
-    rm -f $(2)
-}
-
-actions Archive
-{
-    $(AR) ru $(1) $(2)
-}
-
-rule Library  library : objects
-{
-    local  obj ;
-
-    if ! $(library:S) {
-        library = $(library:S=.a) ;
-    }
-    library = $(library:G=<$(arch)>) ;
-
-    Depends all : $(library) ;
-
-    if ! $(library:D) {
-        MakeLocate $(library) $(library)($(objects:BS)) : $(OUT) ;
-    }
-
-    Depends $(library) : $(library)($(objects:BS)) ;
-    for obj in $(objects) {
-        Depends $(library)($(obj:BS)) : $(obj) ;
-    }
-
-    Clean clean : $(library) ;
-
-    AR on $(library) = $(AR_$(arch)) ;
-    Archive $(library) : $(objects) ;
-
-    RmTemps $(library) : $(objects) ;
-}
-
-
-rule  ProcessDir
-{
-    local CFLAGS   = $(CFLAGS) ;
-    local DEFINES  = $(DEFINES) ;
-    local INCLUDES = $(INCLUDES) ;
-    local local_rules = [ Glob $(1) : rules.jam ] ;
-    local source sources ;
-
-    if $(local_rules) {
-        local LOCAL_CFLAGS LOCAL_DEFINES LOCAL_INCLUDES LOCAL_SRC NO_LOCAL_SRC ;
-
-        include $(local_rules) ;
-        CFLAGS   += $(LOCAL_CFLAGS) ;
-        DEFINES  += $(LOCAL_DEFINES) ;
-        INCLUDES += $(LOCAL_INCLUDES) ;
-
-        if $(LOCAL_SRC) {
-            sources = $(LOCAL_SRC) ;
-        }
-        else {
-            sources = [ Glob $(1) : *.S *.c ] ;
-            sources = $(sources:BS) ;
-        }
-
-        if $(NO_LOCAL_SRC) {
-            sources = [ Filter $(sources) : $(NO_LOCAL_SRC) ] ;
-        }
-
-        sources = $(1)/$(sources) ;
-    }
-    else
-        sources = [ Glob $(1) : *.S *.c ] ;
-
-    for source in $(sources) {
-        local name = $(source:B) ;
-
-        if $(source:S) = ".S" {
-            # record the list of assembler sources
-            ASSEMBLER_SOURCES += $(name) ;
-        }
-        else if $(source:S) = ".c" && $(name) in $(ASSEMBLER_SOURCES) {
-            # skip C source file if corresponding assembler exists
-            continue ;
-        }
-
-        objname = <$(arch)>$(name).o  ;
-
-        Compile $(objname) : $(source) ;
-        ALL_OBJECTS += $(objname) ;
-    }
-}
-
-rule ProcessDirs
-{
-    local  dir ;
-    for dir in $(1) {
-        ProcessDir $(dir) ;
-    }
-}
-
-INCLUDES_x86 = /usr/src/linux/include ;
-
-INCLUDES_arm = ../kernel_headers
-               include/arch/arm
-               include/bits32
-               ;
-
-INCLUDES = include stdio string stdlib .
-           ../msun/include
-           ;
-
-DEFINES  = ANDROID_CHANGES
-           USE_LOCKS
-           REALLOC_ZERO_BYTES_FREES
-           _LIBC=1
-           SOFTFLOAT
-           FLOATING_POINT
-           NEED_PSELECT=1
-           ANDROID
-           ;
-
-CFLAGS_x86 = ;
-
-
-for arch in $(ARCH)
-{
-    local ARCH_DIR = $(BIONIC_TOP)/arch-$(arch) ;
-    local INCLUDES = $(INCLUDES_$(arch)) $(ARCH_DIR)/include $(INCLUDES) ;
-    local DEFINES  = $(DEFINES_$(arch)) $(DEFINES) ARCH=$(arch)  ;
-    local CFLAGS   = $(CFLAGS) $(CFLAGS_$(arch)) ;
-    local OUT      = out/$(arch) ;
-    local ASSEMBLER_SOURCES ALL_OBJECTS ;
-
-    ProcessDirs [ ListSubDirs $(ARCH_DIR) ] ;
-    ProcessDirs stdlib stdio unistd string tzcode inet ;
-    ProcessDirs [ ListSubDirs netbsd ] ;
-    ProcessDirs bionic ;
-
-    Library bionic : $(ALL_OBJECTS) ;
-}
-
-BIONIC_SEARCH = $(BIONIC_TOP)/include ;
-
-
-
-# /HdrRule source : headers ;
-#
-# Arranges the proper dependencies when the file _source_ includes the files
-# _headers_ through the #include C preprocessor directive
-#
-# this rule is not intendend to be called explicitely. It is called
-# automatically during header scanning on sources handled by the @Object
-# rule (e.g. sources in @Main or @Library rules)
-#
-rule HdrRule
-{
-    # HdrRule source : headers ;
-
-    # N.B.  This rule is called during binding, potentially after
-    # the fate of many targets has been determined, and must be
-    # used with caution: don't add dependencies to unrelated
-    # targets, and don't set variables on $(<).
-
-    # Tell Jam that anything depending on $(<) also depends on $(>),
-    # set SEARCH so Jam can find the headers, but then say we don't
-    # care if we can't actually find the headers (they may have been
-    # within ifdefs),
-
-    local s = $(>:G=$(HDRGRIST:E)) ;
-
-    Includes $(<) : $(s) ;
-    SEARCH on $(s) = $(HDRSEARCH) ;
-    NoCare $(s) ;
-
-    # Propagate on $(<) to $(>)
-
-    HDRSEARCH on $(s) = $(HDRSEARCH) ;
-    HDRSCAN on $(s) = $(HDRSCAN) ;
-    HDRRULE on $(s) = $(HDRRULE) ;
-    HDRGRIST on $(s) = $(HDRGRIST) ;
-}
-
-

libc/README

@@ -1,53 +0,0 @@
-Welcome to Bionic, Android small and custom C library for the Android platform
-
-Bionic is mainly a port of the BSD C library to our Linux kernel with the
-following additions/changes:
-
-- no support for locales
-- no support for wide chars (i.e. multi-byte characters)
-- its own smallish implementation of pthreads based on Linux futexes
-- support for x86, ARM and ARM thumb CPU instruction sets and kernel interfaces
-
-Bionic is released under the standard 3-clause BSD License
-
-Bionic doesn't want to implement all features of a traditional C library, we only
-add features to it as we need them, and we try to keep things as simple and small
-as possible. Our goal is not to support scaling to thousands of concurrent threads
-on multi-processors machines; we're running this on cell-phones, damnit !!
-
-Note that Bionic doesn't provide a libthread_db or a libm implementation.
-
-
-Adding new syscalls:
-====================
-
-Bionic provides the gensyscalls.py Python script to automatically generate syscall
-stubs from the list defined in the file SYSCALLS.TXT. You can thus add a new syscall
-by doing the following:
-
-- edit SYSCALLS.TXT
-- add a new line describing your syscall, it should look like:
-
-   return_type  syscall_name(parameters)    syscall_number
-
-- in the event where you want to differentiate the syscall function from its entry name,
-  use the alternate:
-
-   return_type  funcname:syscall_name(parameters)  syscall_number
-
-- additionally, if the syscall number is different between ARM and x86, use:
-
-   return_type  funcname[:syscall_name](parameters)   arm_number,x86_number
-
-- a syscall number can be -1 to indicate that the syscall is not implemented on
-  a given platform, for example:
-
-   void   __set_tls(void*)   arm_number,-1
-
-
-the comments in SYSCALLS.TXT contain more information about the line format
-
-You can also use the 'checksyscalls.py' script to check that all the syscall
-numbers you entered are correct. It does so by looking at the values defined in
-your Linux kernel headers. The script indicates where the values are incorrect
-and what is expected instead.

libc/SYSCALLS.TXT

@@ -1,245 +1,334 @@
-# this file is used to list all the syscalls that will be supported by
-# the Bionic C library. It is used to automatically generate the syscall
-# stubs, the list of syscall constants (__NR_xxxx) and the content of <linux/_unitsd.h>
+# 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])  (#syscall_number|stub)
+# return_type func_name[|alias_list][:syscall_name[:socketcall_id]]([parameter_list]) arch_list
 #
-# note that:
-#      - syscall_name correspond to the name of the syscall, which may differ from
+# 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
-#
-#      - it there is "stub" instead of a syscall number, the tool will not generate any
-#        assembler template for the syscall; it's up to the bionic implementation to provide
-#        a relevant C stub
-#
-# the file is processed by a python script named gensyscalls.py
+#      - 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)      248,252
-void    _exit_thread:exit (int)	    1
-pid_t   __fork:fork (void)           2
-pid_t   _waitpid:waitpid (pid_t, int*, int, struct rusage*)   -1,7
-int     waitid(int, pid_t, struct siginfo_t*, int,void*)          280,284
-pid_t   __clone:clone(int (*fn)(void*), void *child_stack, int flags, void *arg)  120
-int     execve (const char*, char* const*, char* const*)  11
+int     execve(const char*, char* const*, char* const*)  all
 
-int     setuid:setuid32 (uid_t)    213
-uid_t   getuid:getuid32 ()         199
-gid_t   getgid:getgid32 ()         200
-uid_t   geteuid:geteuid32 ()       201
-gid_t   getegid:getegid32 ()       202
-uid_t   getresuid:getresuid32 ()   209
-gid_t   getresgid:getresgid32 ()   211
-pid_t   gettid()                   224
-int     getgroups:getgroups32(int, gid_t *)    205
-pid_t   getpgid(pid_t)             132
-pid_t   getppid()		   64
-pid_t   setsid()                   66
-int     setgid:setgid32(gid_t)     214
-int     seteuid:seteuid32(uid_t)   stub
-int     setreuid:setreuid32(uid_t, uid_t)   203
-int     setresuid:setresuid32(uid_t, uid_t, uid_t)   208
-int     setresgid:setresgid32(gid_t, gid_t, gid_t)   210
-void*   __brk:brk(void*)           45
-# 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)           -1,37
-int     tkill(pid_t tid, int sig)  -1,238
-int     __ptrace:ptrace(int request, int pid, void* addr, void* data)  26
-int     __set_thread_area:set_thread_area(void*  user_desc)  -1,243
-int     __getpriority:getpriority(int, int)  96
-int     setpriority(int, int, int)   97
-int     setrlimit(int resource, const struct rlimit *rlp)  75
-int     getrlimit:ugetrlimit(int resource, struct rlimit *rlp)  191
-int     getrusage(int who, struct rusage*  r_usage)  77
-int     setgroups:setgroups32(int, const gid_t *)   206
-pid_t   getpgrp(void)  stub
-int     setpgid(pid_t, pid_t)  57
-pid_t   vfork(void)  190,-1
-int     setregid:setregid32(gid_t, gid_t)  204
-int     chroot(const char *)  61
-int     prctl(int option, unsigned int arg2, unsigned int arg3, unsigned int arg4, unsigned int arg5)  172
-int     capget(cap_user_header_t header, cap_user_data_t data) 184
-int     capset(cap_user_header_t header, const cap_user_data_t data) 185
-int     acct(const char*  filepath)  51
+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
+
+# <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     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
+pid_t   vfork(void)  arm
+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)        3
-ssize_t     write (int, const void*, size_t)       4
-ssize_t     __pread64:pread64 (int, void *, size_t, off_t, off_t) 180
-ssize_t     __pwrite64:pwrite64 (int, void *, size_t, off_t, off_t) 181
-int         __open:open (const char*, int, mode_t)  5
-int         __openat:openat (int, const char*, int, mode_t)  322,295
-int         close (int)                      6
-int         creat(const char*, mode_t)       stub
-off_t       lseek(int, off_t, int)           19
-int         __llseek:_llseek (int, unsigned long, unsigned long, loff_t*, int)  140
-pid_t       getpid ()    20
-void *      mmap(void *, size_t, int, int, int, long)  stub
-void *      __mmap2:mmap2(void*, size_t, int, int, int, long)   192
-int         munmap(void *, size_t)  91
-void *      mremap(void *, size_t, size_t, unsigned long)  163
-int         msync(const void *, size_t, int)    144
-int         mprotect(const void *, size_t, int)  125
-int         madvise(const void *, size_t, int)  220,219
-int         mlock(const void *addr, size_t len)    150
-int         munlock(const void *addr, size_t len)   151
-int         mincore(void*  start, size_t  length, unsigned char*  vec)   219,218
-int         __ioctl:ioctl(int, int, void *)  54
-int         readv(int, const struct iovec *, int)   145
-int         writev(int, const struct iovec *, int)  146
-int         __fcntl:fcntl(int, int, void*)  55
-int         flock(int, int)   143
-int         fchmod(int, mode_t)  94
-int         dup(int)  41
-int         pipe(int *)  42
-int         dup2(int, int)   63
-int         select:_newselect(int, struct fd_set *, struct fd_set *, struct fd_set *, struct timeval *)  142
-int         ftruncate(int, off_t)  93
-int         getdents:getdents64(unsigned int, struct dirent *, unsigned int)   217,220
-int         fsync(int)  118
-int         fchown:fchown32(int, uid_t, gid_t)  207
-void        sync(void)  36
-int         __fcntl64:fcntl64(int, int, void *)  221
-int         fstatfs:fstatfs64(int, size_t, struct statfs *)  267,269
-ssize_t     sendfile(int out_fd, int in_fd, off_t *offset, size_t count)  187
-int         fstatat:fstatat64(int dirfd, const char *path, struct stat *buf, int flags)   327,300
-int         mkdirat(int dirfd, const char *pathname, mode_t mode)  323,296
-int         fchownat(int dirfd, const char *path, uid_t owner, gid_t group, int flags)  325,298
-int         fchmodat(int dirfd, const char *path, mode_t mode, int flags)  333,306
-int         renameat(int olddirfd, const char *oldpath, int newdirfd, const char *newpath)  329,302
+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
+int         close(int)                      all
+pid_t       __getpid:getpid()  all
+int         munmap(void*, size_t)  all
+void*       mremap(void*, size_t, size_t, unsigned long)  all
+int         msync(const void*, size_t, int)    all
+int         mprotect(const void*, size_t, int)  all
+int         madvise(const 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(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(int, const char*, const void*, size_t, int) all
+ssize_t     fgetxattr(int, const char*, void*, size_t) all
+ssize_t     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(int, const char*, int, int)  all
+int fchmodat(int, const char*, mode_t, int)  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,mips64,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 ftruncate(int, off_t) arm,mips,x86
+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 fstatfs64|fstatfs:fstatfs(int, struct statfs*)  arm64,mips64,x86_64
+int __statfs64:statfs64(const char*, size_t, struct statfs*)  arm,mips,x86
+int statfs64|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,mips64,x86_64
 
 # file system
-int     link (const char*, const char*)  9
-int     unlink (const char*)             10
-int     unlinkat (int, const char *, int)   328,301
-int     chdir (const char*)              12
-int     mknod (const char*, mode_t, dev_t)  14
-int     chmod (const char*,mode_t)          15
-int     chown:chown32(const char *, uid_t, gid_t)  212
-int     lchown:lchown32 (const char*, uid_t, gid_t)  198
-int     mount (const char*, const char*, const char*, unsigned long, const void*)  21
-int     umount(const char*)  stub
-int     umount2 (const char*, int)  52
-int     fstat:fstat64(int, struct stat*)    197
-int     stat:stat64(const char *, struct stat *)  195
-int     lstat:lstat64(const char *, struct stat *)  196
-int     mkdir(const char *, mode_t) 39
-int     readlink(const char *, char *, size_t)  85
-int     rmdir(const char *)  40
-int     rename(const char *, const char *)  38
-int     __getcwd:getcwd(char * buf, size_t size)  183
-int     access(const char *, int)  33
-int     symlink(const char *, const char *)  83
-int     fchdir(int)    133
-int     truncate(const char*, off_t)    92
-int     __statfs64:statfs64(const char *, size_t, struct statfs *)  266,268
+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 ()                       29
-int           gettimeofday(struct timeval*, struct timezone*)       78
-int           settimeofday(const struct timeval*, const struct timezone*)   79
-clock_t       times(struct tms *)       43
-int           nanosleep(const struct timespec *, struct timespec *)   162
-int           clock_gettime(clockid_t clk_id, struct timespec *tp)    263,265
-int           clock_settime(clockid_t clk_id, const struct timespec *tp)  262,264
-int           clock_getres(clockid_t clk_id, struct timespec *res)   264,266
-int           clock_nanosleep(const struct timespec *req, struct timespec *rem)  265,267
-int           getitimer(int, const struct itimerval *)   105
-int           setitimer(int, const struct itimerval *, struct itimerval *)  104
-int           __timer_create:timer_create(clockid_t clockid, struct sigevent *evp, timer_t *timerid)    257,259
-int           __timer_settime:timer_settime(timer_t, int, const struct itimerspec*, struct itimerspec*) 258,260
-int           __timer_gettime:timer_gettime(timer_t, struct itimerspec*)                                259,261
-int           __timer_getoverrun:timer_getoverrun(timer_t)                                              260,262
-int           __timer_delete:timer_delete(timer_t)                                                      261,263
-int           utimes(const char*, const struct timeval tvp[2])                          269, 271
+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 *)  67
-int     sigprocmask(int, const sigset_t *, sigset_t *)  126
-int     __sigsuspend:sigsuspend(int unused1, int unused2, unsigned mask)  72
-int     __rt_sigaction:rt_sigaction (int sig, const struct sigaction *act, struct sigaction *oact, size_t sigsetsize)  174
-int     __rt_sigprocmask:rt_sigprocmask (int  how, const sigset_t *set, sigset_t *oset, size_t sigsetsize)  175
-int     __rt_sigtimedwait:rt_sigtimedwait(const sigset_t *set, struct siginfo_t  *info, struct timespec_t  *timeout, size_t  sigset_size)  177
-int     sigpending(sigset_t *)  73
+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     __signalfd4:signalfd4(int, const sigset_t*, size_t, int)  all
 
 # sockets
-int           socket(int, int, int)              281,-1
-int           socketpair(int, int, int, int*)    288,-1
-int           bind(int, struct sockaddr *, int)  282,-1
-int           connect(int, struct sockaddr *, socklen_t)   283,-1
-int           listen(int, int)                   284,-1
-int           accept(int, struct sockaddr *, socklen_t *)  285,-1
-int           getsockname(int, struct sockaddr *, socklen_t *)  286,-1
-int           getpeername(int, struct sockaddr *, socklen_t *)  287,-1
-int           sendto(int, const void *, size_t, int, const struct sockaddr *, socklen_t)  290,-1
-int           recvfrom(int, void *, size_t, unsigned int, struct sockaddr *, socklen_t *)  292,-1
-int           shutdown(int, int)  293,-1
-int           setsockopt(int, int, int, const void *, socklen_t)  294,-1
-int           getsockopt(int, int, int, void *, socklen_t *)    295,-1
-int           sendmsg(int, const struct msghdr *, unsigned int)  296,-1
-int           recvmsg(int, struct msghdr *, unsigned int)   297,-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) -1,102
-int           bind:socketcall:2 (int, struct sockaddr *, int)  -1,102
-int           connect:socketcall:3(int, struct sockaddr *, socklen_t)   -1,102
-int           listen:socketcall:4(int, int)                   -1,102
-int           accept:socketcall:5(int, struct sockaddr *, socklen_t *)  -1,102
-int           getsockname:socketcall:6(int, struct sockaddr *, socklen_t *)  -1,102
-int           getpeername:socketcall:7(int, struct sockaddr *, socklen_t *)  -1,102
-int           socketpair:socketcall:8(int, int, int, int*)    -1,102
-int           sendto:socketcall:11(int, const void *, size_t, int, const struct sockaddr *, socklen_t)  -1,102
-int           recvfrom:socketcall:12(int, void *, size_t, unsigned int, struct sockaddr *, socklen_t *)  -1,102
-int           shutdown:socketcall:13(int, int)  -1,102
-int           setsockopt:socketcall:14(int, int, int, const void *, socklen_t)  -1,102
-int           getsockopt:socketcall:15(int, int, int, void *, socklen_t *)    -1,102
-int           sendmsg:socketcall:16(int, const struct msghdr *, unsigned int)  -1,102
-int           recvmsg:socketcall:17(int, struct msghdr *, unsigned int)   -1,102
+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)  156
-int sched_getscheduler(pid_t pid)  157
-int sched_yield(void)  158
-int sched_setparam(pid_t pid, const struct sched_param *param)  154
-int sched_getparam(pid_t pid, struct sched_param *param)  155
-int sched_get_priority_max(int policy)  159
-int sched_get_priority_min(int policy)  160
-int sched_rr_get_interval(pid_t pid, struct timespec *interval)  161
+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 *)  122
-pid_t   __wait4:wait4(pid_t pid, int *status, int options, struct rusage *rusage)   114
-mode_t  umask(mode_t)  60
-int      __reboot:reboot(int, int, int, void *)  88
-int     __syslog:syslog(int, char *, int)  103
-int     init_module(void *, unsigned long, const char *)  128
-int     delete_module(const char*, unsigned int)   129
-int     klogctl:syslog(int, char *, int)   103
+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) 240
+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)     250,254
-int     epoll_ctl(int epfd, int op, int fd, struct epoll_event *event)    251,255
-int     epoll_wait(int epfd, struct epoll_event *events, int max, int timeout)   252,256
+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)      316,291
-int     inotify_add_watch(int, const char *, unsigned int)  317,292
-int     inotify_rm_watch(int, unsigned int)  318,293
+int eventfd:eventfd2(unsigned int, int)  all
 
-int     poll(struct pollfd *, unsigned int, long)  168
+void _exit|_Exit:exit_group(int)  all
+void __exit:exit(int)  all
 
-# ARM-specific ARM_NR_BASE == 0x0f0000 == 983040
-int     __set_tls:ARM_set_tls(void*)                                 983045,-1
-int     cacheflush:ARM_cacheflush(long start, long end, long flags)  983042,-1
+int inotify_init1(int)  all
+int inotify_add_watch(int, const char*, unsigned int)  all
+int inotify_rm_watch(int, unsigned int)  all
+
+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
+
+int __set_tid_address:set_tid_address(int*)  all
+
+int setfsgid(gid_t)  all
+int setfsuid(uid_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) 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,x86
+int __clock_gettime:clock_gettime(clockid_t, timespec*) arm64,x86_64
+int gettimeofday(timeval*, timezone*)                   arm,mips,mips64,x86
+int __gettimeofday:gettimeofday(timeval*, timezone*)    arm64,x86_64

libc/arch-arm/arm.mk

@@ -0,0 +1,95 @@
+# 32-bit arm.
+
+#
+# Various kinds of LP32 cruft.
+#
+
+libc_bionic_src_files_arm += \
+    bionic/mmap.cpp \
+
+libc_common_src_files_arm += \
+    bionic/legacy_32_bit_support.cpp \
+    bionic/ndk_cruft.cpp \
+    bionic/time64.c \
+
+libc_netbsd_src_files_arm += \
+    upstream-netbsd/common/lib/libc/hash/sha1/sha1.c \
+
+libc_openbsd_src_files_arm += \
+    upstream-openbsd/lib/libc/stdio/putw.c \
+
+#
+# Default implementations of functions that are commonly optimized.
+#
+
+libc_bionic_src_files_arm += \
+    bionic/memchr.c \
+    bionic/memrchr.c \
+    bionic/strchr.cpp \
+    bionic/strnlen.c \
+    bionic/strrchr.cpp \
+
+libc_freebsd_src_files_arm += \
+    upstream-freebsd/lib/libc/string/wcscat.c \
+    upstream-freebsd/lib/libc/string/wcschr.c \
+    upstream-freebsd/lib/libc/string/wcscmp.c \
+    upstream-freebsd/lib/libc/string/wcscpy.c \
+    upstream-freebsd/lib/libc/string/wcslen.c \
+    upstream-freebsd/lib/libc/string/wcsrchr.c \
+    upstream-freebsd/lib/libc/string/wmemcmp.c \
+    upstream-freebsd/lib/libc/string/wmemmove.c \
+
+libc_openbsd_src_files_arm += \
+    upstream-openbsd/lib/libc/string/bcopy.c \
+    upstream-openbsd/lib/libc/string/stpncpy.c \
+    upstream-openbsd/lib/libc/string/strlcat.c \
+    upstream-openbsd/lib/libc/string/strlcpy.c \
+    upstream-openbsd/lib/libc/string/strncat.c \
+    upstream-openbsd/lib/libc/string/strncmp.c \
+    upstream-openbsd/lib/libc/string/strncpy.c \
+
+#
+# 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/memcmp.S \
+    arch-arm/bionic/__restore.S \
+    arch-arm/bionic/_setjmp.S \
+    arch-arm/bionic/setjmp.S \
+    arch-arm/bionic/sigsetjmp.S \
+    arch-arm/bionic/syscall.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
+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 += $(cpu_variant_mk)
+
+cpu_variant_mk :=
+
+
+libc_crt_target_cflags_arm := \
+    -I$(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 +1,113 @@
+/*
+ * 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(void *object, void (*destructor) (void *), void *dso_handle) {
+    return __cxa_atexit(destructor, object, dso_handle);
+}
+
+
+void __attribute__((weak))
+__aeabi_memcpy8(void *dest, const void *src, size_t n) {
+    memcpy(dest, src, n);
+}
+
+void __attribute__((weak)) __aeabi_memcpy4(void *dest, const void *src, size_t n) {
+    memcpy(dest, src, n);
+}
+
+void __attribute__((weak)) __aeabi_memcpy(void *dest, const void *src, size_t n) {
+    memcpy(dest, src, n);
+}
+
+
+void __attribute__((weak)) __aeabi_memmove8(void *dest, const void *src, size_t n) {
+    memmove(dest, src, n);
+}
+
+void __attribute__((weak)) __aeabi_memmove4(void *dest, const void *src, size_t n) {
+    memmove(dest, src, n);
+}
+
+void __attribute__((weak)) __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 __attribute__((weak)) __aeabi_memset8(void *dest, size_t n, int c) {
+    memset(dest, c, n);
+}
+
+void __attribute__((weak)) __aeabi_memset4(void *dest, size_t n, int c) {
+    memset(dest, c, n);
+}
+
+void __attribute__((weak)) __aeabi_memset(void *dest, size_t n, int c) {
+    memset(dest, c, n);
+}
+
+
+void __attribute__((weak)) __aeabi_memclr8(void *dest, size_t n) {
+    __aeabi_memset8(dest, n, 0);
+}
+
+void __attribute__((weak)) __aeabi_memclr4(void *dest, size_t n) {
+    __aeabi_memset4(dest, n, 0);
+}
+
+void __attribute__((weak)) __aeabi_memclr(void *dest, size_t n) {
+    __aeabi_memset(dest, n, 0);
+}

libc/arch-arm/bionic/__bionic_clone.S

@@ -0,0 +1,69 @@
+/*
+ * 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 <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}
+
+    # 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_internal
+
+1:  # The child.
+    # Setting lr to 0 will make the unwinder stop at __start_thread
+    mov    lr, #0
+    ldr    r0, [sp, #-4]
+    ldr    r1, [sp, #-8]
+    b      __start_thread
+END(__bionic_clone)
+.hidden __bionic_clone

libc/arch-arm/bionic/__get_pc.S

@@ -1,33 +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.
- */
-.global __get_pc
-
-__get_pc:
-	mov r0, pc
-	bx lr
-

libc/arch-arm/bionic/__get_sp.S

@@ -1,33 +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.
- */
-.global __get_sp
-
-__get_sp:
-	mov r0, sp
-	bx lr
-

libc/arch-arm/bionic/__restore.S

@@ -0,0 +1,61 @@
+/*
+ * 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 DWARF information to find out where it landed) gets the right answer.
+
+// 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(__restore)
+  // This function must have exactly this instruction sequence.
+  mov r7, #__NR_sigreturn
+  swi #0
+END(__restore)
+
+  .fnstart
+  .save {r0-r15}
+  .pad #160
+  nop
+ENTRY_PRIVATE(__restore_rt)
+  // This function must have exactly this instruction sequence.
+  mov r7, #__NR_rt_sigreturn
+  swi #0
+END(__restore_rt)

libc/arch-arm/bionic/_exit_with_stack_teardown.S

@@ -25,31 +25,17 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
-#include <asm/unistd.h>
 
-.text
-.type _exit_with_stack_teardown, #function
-.globl _exit_with_stack_teardown
-.align 4
+#include <private/bionic_asm.h>
 
-@ void _exit_with_stack_teardown(void * stackBase, int stackSize, int retCode)
-
-_exit_with_stack_teardown:
-
-#if __ARM_EABI__
-    mov     lr, r2
+// void _exit_with_stack_teardown(void* stackBase, size_t stackSize)
+ENTRY_PRIVATE(_exit_with_stack_teardown)
   ldr r7, =__NR_munmap
-    swi     #0              @ the stack is destroyed by this call
-    mov     r0, lr
+  swi #0
+  // If munmap failed, we ignore the failure and exit anyway.
+
+  mov r0, #0
   ldr r7, =__NR_exit
   swi #0
-#else
-    mov     lr, r2
-    swi     # __NR_munmap   @ the stack is destroyed by this call
-    mov     r0, lr
-    swi     # __NR_exit
-#endif
-
-    @ exit() should never return, cause a crash if it does
-    mov		r0, #0
-    ldr		r0, [r0]
+  // The exit syscall does not return.
+END(_exit_with_stack_teardown)

libc/arch-arm/bionic/_setjmp.S

@@ -3,6 +3,7 @@
 
 /*
  * 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
@@ -33,8 +34,9 @@
  * SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
+#include <private/bionic_asm.h>
 #include <machine/setjmp.h>
+#include <machine/cpu-features.h>
 
 /*
  * C library -- _setjmp, _longjmp
@@ -51,42 +53,46 @@
 
 ENTRY(_setjmp)
 	ldr	r1, .L_setjmp_magic
-	str	r1, [r0], #4
-#ifdef SOFTFLOAT
-	add	r0, r0, #52
-#else
-	/* Store fp registers */
-	sfm	f4, 4, [r0], #48
-	/* Store fpsr */
-	rfs	r1
-	str	r1, [r0], #0x0004
-#endif	/* SOFTFLOAT */
-	/* Store integer registers */
-        stmia	r0, {r4-r14}
+	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], #4
+	ldr	r3, [r0, #(_JB_MAGIC * 4)]
 	teq	r2, r3
 	bne	botch
 
-#ifdef SOFTFLOAT
-	add	r0, r0, #52
-#else
-	/* Restore fp registers */
-	lfm	f4, 4, [r0], #48
-	/* Restore fpsr */
-	ldr	r4, [r0], #0x0004
-	wfs	r4
-#endif	/* SOFTFLOAT */
-       	/* Restore integer registers */
-        ldmia	r0, {r4-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
@@ -101,6 +107,7 @@ ENTRY(_longjmp)
 
 	/* validation failed, die die die. */
 botch:
-	bl	PIC_SYM(_C_LABEL(longjmperror), PLT)
-	bl	PIC_SYM(_C_LABEL(abort), PLT)
+	bl	longjmperror
+	bl	abort
 	b	. - 8		/* Cannot get here */
+END(_longjmp)

libc/arch-arm/bionic/abort_arm.S

@@ -0,0 +1,44 @@
+/*
+ * 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 <private/bionic_asm.h>
+
+/*
+ * Coding the abort function in assembly so that registers are guaranteed to
+ * be preserved properly regardless of GCC's assumption on the "noreturn"
+ * attribute. When the registers are not properly preserved we won't be able
+ * to unwind the stack all the way to the bottom to fully reveal the call
+ * sequence when the crash happens.
+ */
+ENTRY(abort)
+    stmfd   sp!, {r3, r14}
+    .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_legacy.c

@@ -0,0 +1,60 @@
+/*
+ * 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 <sys/types.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
+ * implementation of atexit().
+ */
+
+extern int __cxa_atexit(void (*func)(void *), void *arg, void *dso);
+
+/*
+ * Register a function to be performed at exit.
+ */
+int
+atexit(void (*func)(void))
+{
+    /*
+     * Exit functions queued by this version of atexit will not be called
+     * on dlclose(), and when they are called (at program exit), the
+     * calling library may have been dlclose()'d, causing the program to
+     * crash.
+     */
+    static char const warning[] = "WARNING: generic atexit() called from legacy shared library\n";
+
+    __libc_format_log(ANDROID_LOG_WARN, "libc", warning);
+    fprintf(stderr, warning);
+
+    return (__cxa_atexit((void (*)(void *))func, NULL, NULL));
+}

libc/arch-arm/bionic/atomics_arm.S

@@ -1,178 +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 <sys/linux-syscalls.h>
-
-.global __atomic_cmpxchg
-.global __atomic_swap
-.global __atomic_dec
-.global __atomic_inc
-.global __futex_wait
-.global __futex_wake
-
-#define FUTEX_WAIT 0
-#define FUTEX_WAKE 1
-
-#if 1
-   .equ     kernel_cmpxchg, 0xFFFF0FC0
-   .equ     kernel_atomic_base, 0xFFFF0FFF
-__atomic_dec:
-    stmdb   sp!, {r4, lr}
-    mov     r2, r0
-1: @ atomic_dec
-    ldr     r0, [r2]
-    mov     r3, #kernel_atomic_base
-    add     lr, pc, #4
-    sub     r1, r0, #1
-    add     pc, r3, #(kernel_cmpxchg - kernel_atomic_base)
-    bcc     1b
-    add     r0, r1, #1
-    ldmia   sp!, {r4, lr}
-    bx      lr
-
-__atomic_inc:
-    stmdb   sp!, {r4, lr}
-    mov     r2, r0
-1: @ atomic_inc
-    ldr     r0, [r2]
-    mov     r3, #kernel_atomic_base
-    add     lr, pc, #4
-    add     r1, r0, #1
-    add     pc, r3, #(kernel_cmpxchg - kernel_atomic_base)
-    bcc     1b
-    sub     r0, r1, #1
-    ldmia   sp!, {r4, lr}
-    bx      lr
-
-/* r0(old) r1(new) r2(addr) -> r0(zero_if_succeeded) */
-__atomic_cmpxchg:
-    stmdb   sp!, {r4, lr}
-    mov     r4, r0          /* r4 = save oldvalue */
-1: @ atomic_cmpxchg
-    mov     r3, #kernel_atomic_base
-    add     lr, pc, #4
-    mov     r0, r4          /* r0 = oldvalue */
-    add     pc, r3, #(kernel_cmpxchg - kernel_atomic_base)
-    bcs     2f              /* swap was made. we're good, return. */
-    ldr     r3, [r2]        /* swap not made, see if it's because *ptr!=oldvalue */
-    cmp     r3, r4
-    beq     1b
-2: @ atomic_cmpxchg
-    ldmia   sp!, {r4, lr}
-    bx      lr
-#else
-#define KUSER_CMPXCHG 0xffffffc0
-
-/* r0(old) r1(new) r2(addr) -> r0(zero_if_succeeded) */
-__atomic_cmpxchg:
-    stmdb   sp!, {r4, lr}
-    mov     r4, r0          /* r4 = save oldvalue */
-1:  add     lr, pc, #4
-    mov     r0, r4          /* r0 = oldvalue */
-    mov     pc, #KUSER_CMPXCHG
-    bcs     2f              /* swap was made. we're good, return. */
-    ldr     r3, [r2]        /* swap not made, see if it's because *ptr!=oldvalue */
-    cmp     r3, r4
-    beq     1b
-2:  ldmia   sp!, {r4, lr}
-    bx      lr
-
-/* r0(addr) -> r0(old) */
-__atomic_dec:
-    stmdb   sp!, {r4, lr}
-    mov     r2, r0          /* address */
-1:  ldr     r0, [r2]        /* oldvalue */
-    add     lr, pc, #4
-    sub     r1, r0, #1      /* newvalue = oldvalue - 1 */
-    mov     pc, #KUSER_CMPXCHG
-    bcc     1b              /* no swap, try again until we get it right */
-    mov     r0, ip          /* swapped, return the old value */
-    ldmia   sp!, {r4, lr}
-    bx      lr
-
-/* r0(addr) -> r0(old) */
-__atomic_inc:
-    stmdb   sp!, {r4, lr}
-    mov     r2, r0          /* address */
-1:  ldr     r0, [r2]        /* oldvalue */
-    add     lr, pc, #4
-    add     r1, r0, #1      /* newvalue = oldvalue + 1 */
-    mov     pc, #KUSER_CMPXCHG
-    bcc     1b              /* no swap, try again until we get it right */
-    mov     r0, ip          /* swapped, return the old value */
-    ldmia   sp!, {r4, lr}
-    bx      lr
-#endif
-
-/* r0(new) r1(addr) -> r0(old) */
-__atomic_swap:
-    swp     r0, r0, [r1]
-    bx      lr
-
-/* __futex_wait(*ftx, val, *timespec) */
-/* __futex_syscall(*ftx, op, val, *timespec, *addr2, val3) */
-
-#if __ARM_EABI__
-
-__futex_wait:
-    .fnstart
-    stmdb   sp!, {r4, r7}
-    .save   {r4, r7}
-    mov     r3, r2
-    mov     r2, r1
-    mov     r1, #FUTEX_WAIT
-    ldr     r7, =__NR_futex
-    swi     #0
-    ldmia   sp!, {r4, r7}
-    bx      lr
-    .fnend
-
-__futex_wake:
-    stmdb   sp!, {r4, r7}
-    mov     r2, r1
-    mov     r1, #FUTEX_WAKE
-    ldr     r7, =__NR_futex
-    swi     #0
-    ldmia   sp!, {r4, r7}
-    bx      lr
-
-#else
-
-__futex_wait:
-    mov     r3, r2
-    mov     r2, r1
-    mov     r1, #FUTEX_WAIT
-    swi     #__NR_futex
-    bx      lr
-
-__futex_wake:
-    mov     r2, r1
-    mov     r1, #FUTEX_WAKE
-    swi     #__NR_futex
-    bx      lr
-
-#endif

libc/arch-arm/bionic/atomics_arm.c

@@ -0,0 +1,87 @@
+/*
+ * Copyright (C) 2011 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.
+ */
+
+
+/* The purpose of this file is to export a small set of atomic-related
+ * functions from the C library, to ensure binary ABI compatibility for
+ * the NDK.
+ *
+ * These functions were initially exposed by the NDK through <sys/atomics.h>,
+ * which was unfortunate because their implementation didn't provide any
+ * memory barriers at all.
+ *
+ * This wasn't a problem for the platform code that used them, because it
+ * used explicit barrier instructions around them. On the other hand, it means
+ * that any NDK-generated machine code that linked against them would not
+ * perform correctly when running on multi-core devices.
+ *
+ * To fix this, the platform code was first modified to not use any of these
+ * functions (everything is now inlined through assembly statements, see
+ * libc/private/bionic_arm_inline.h and the headers it includes.
+ *
+ * The functions here are thus only for the benefit of NDK applications,
+ * and now includes full memory barriers to prevent any random memory ordering
+ * issue from cropping.
+ *
+ * Note that we also provide an updated <sys/atomics.h> header that defines
+ * always_inlined versions of the functions that use the GCC builtin
+ * intrinsics to perform the same thing.
+ *
+ * NOTE: There is no need for a similar file for non-ARM platforms.
+ */
+
+/* DO NOT INCLUDE <sys/atomics.h> HERE ! */
+
+int
+__atomic_cmpxchg(int old, int _new, volatile int *ptr)
+{
+    /* We must return 0 on success */
+    return __sync_val_compare_and_swap(ptr, old, _new) != old;
+}
+
+int
+__atomic_swap(int _new, volatile int *ptr)
+{
+    int prev;
+    do {
+        prev = *ptr;
+    } while (__sync_val_compare_and_swap(ptr, prev, _new) != prev);
+    return prev;
+}
+
+int
+__atomic_dec(volatile int *ptr)
+{
+  return __sync_fetch_and_sub (ptr, 1);
+}
+
+int
+__atomic_inc(volatile int *ptr)
+{
+  return __sync_fetch_and_add (ptr, 1);
+}

libc/arch-arm/bionic/clone.S

@@ -1,74 +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 <sys/linux-syscalls.h>
-
-	.text
-	.type __pthread_clone, #function
-	.global __pthread_clone
-	.align 4
-    
-__pthread_clone:
-	@ insert the args onto the new stack
-	str     r0, [r1, #-4]
-	str     r3, [r1, #-8]
-
-	@ do the system call
-	@ get flags
-	
-    mov     r0, r2
-	
-    @ new sp is already in r1
-
-#if __ARM_EABI__
-    stmfd   sp!, {r4, r7}
-    ldr     r7, =__NR_clone
-	swi     #0
-#else
-	swi     #__NR_clone
-#endif
-
-	movs    r0, r0
-#if __ARM_EABI__
-    ldmnefd sp!, {r4, r7}
-#endif
-	blt     __error
-	bxne    lr
-
-
-	@ pick the function arg and call address off the stack and jump
-	@ to the C __thread_entry function which does some setup and then
-	@ calls the thread's start function
-
-	ldr     r0, [sp, #-4]
-	ldr     r1, [sp, #-8]
-	mov     r2, sp			@ __thread_entry needs the TLS pointer
-	b       __thread_entry
-
-__error:
-	mov     r0, #-1
-	bx      lr

libc/arch-arm/bionic/crtbegin_dynamic.S

@@ -1,92 +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.
- */
-	.text
-	.align 4
-	.type _start,#function
-	.globl _start
-
-# this is the small startup code that is first run when
-# any executable that is statically-linked with Bionic
-# runs.
-#
-# it's purpose is to call __libc_init with appropriate
-# arguments, which are:
-#
-#    - the address of the raw data block setup by the Linux
-#      kernel ELF loader
-#
-#    - address of an "onexit" function, not used on any
-#      platform supported by Bionic
-#
-#    - address of the "main" function of the program. We
-#      can't hard-code it in the adr pseudo instruction
-#      so we use a tiny trampoline that will get relocated
-#      by the dynamic linker before this code runs
-#
-#    - address of the constructor list
-#
-_start:	
-	mov	r0, sp
-	mov	r1, #0
-	adr r2, 0f
-	adr r3, 1f
-	b	__libc_init
-
-0:  b   main
-
-1:  .long   __PREINIT_ARRAY__
-    .long   __INIT_ARRAY__
-    .long   __FINI_ARRAY__
-    .long   __CTOR_LIST__
-      
-# the .ctors section contains a list of pointers to "constructor"
-# functions that need to be called in order during C library initialization,
-# just before the program is being run. This is a C++ requirement
-#
-# the last entry shall be 0, and is defined in crtend.S
-#
-	.section .preinit_array, "aw"
-	.globl __PREINIT_ARRAY__
-__PREINIT_ARRAY__:
-	.long -1
-
-	.section .init_array, "aw"
-	.globl __INIT_ARRAY__
-__INIT_ARRAY__:
-	.long -1
-
-	.section .fini_array, "aw"
-	.globl __FINI_ARRAY__
-__FINI_ARRAY__:
-	.long -1
-
-	.section .ctors, "aw"
-	.globl __CTOR_LIST__
-__CTOR_LIST__:
-	.long -1
-

libc/arch-arm/bionic/crtbegin_static.S

@@ -1,92 +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.
- */
-	.text
-	.align 4
-	.type _start,#function
-	.globl _start
-
-# this is the small startup code that is first run when
-# any executable that is statically-linked with Bionic
-# runs.
-#
-# it's purpose is to call __libc_init with appropriate
-# arguments, which are:
-#
-#    - the address of the raw data block setup by the Linux
-#      kernel ELF loader
-#
-#    - address of an "onexit" function, not used on any
-#      platform supported by Bionic
-#
-#    - address of the "main" function of the program. We
-#      can't hard-code it in the adr pseudo instruction
-#      so we use a tiny trampoline that will get relocated
-#      by the dynamic linker before this code runs
-#
-#    - address of the constructor list
-#
-_start:	
-	mov	r0, sp
-	mov	r1, #0
-	adr r2, 0f
-	adr r3, 1f
-	b	__libc_init
-
-0:  b   main
-
-1:  .long   __PREINIT_ARRAY__
-    .long   __INIT_ARRAY__
-    .long   __FINI_ARRAY__
-    .long   __CTOR_LIST__
-      
-# the .ctors section contains a list of pointers to "constructor"
-# functions that need to be called in order during C library initialization,
-# just before the program is being run. This is a C++ requirement
-#
-# the last entry shall be 0, and is defined in crtend.S
-#
-	.section .preinit_array, "aw"
-	.globl __PREINIT_ARRAY__
-__PREINIT_ARRAY__:
-	.long -1
-
-	.section .init_array, "aw"
-	.globl __INIT_ARRAY__
-__INIT_ARRAY__:
-	.long -1
-
-	.section .fini_array, "aw"
-	.globl __FINI_ARRAY__
-__FINI_ARRAY__:
-	.long -1
-
-	.section .ctors, "aw"
-	.globl __CTOR_LIST__
-__CTOR_LIST__:
-	.long -1
-

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
-
-	.section .ctors, "aw"
-	.long 0
-

libc/arch-arm/bionic/exidx_dynamic.c

@@ -25,12 +25,8 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
-typedef long unsigned int *_Unwind_Ptr;
 
-/* Stubbed out in libdl and defined in the dynamic linker.
- * Same semantics as __gnu_Unwind_Find_exidx().
- */
-extern _Unwind_Ptr dl_unwind_find_exidx(_Unwind_Ptr pc, int *pcount);
+#include <link.h>
 
 /* For a given PC, find the .so that it belongs to.
  * Returns the base address of the .ARM.exidx section

libc/arch-arm/bionic/exidx_static.c

@@ -25,7 +25,8 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
-typedef long unsigned int *_Unwind_Ptr;
+
+#include <link.h>
 
 /* Find the .ARM.exidx section (which in the case of a static executable
  * can be identified through its start and end symbols), and return its

libc/arch-arm/bionic/ffs.S

@@ -1,82 +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>
-
-/*
- * ffs - find first set bit, this algorithm isolates the first set
- * bit, then multiplies the number by 0x0450fbaf which leaves the top
- * 6 bits as an index into the table.  This algorithm should be a win
- * over the checking each bit in turn as per the C compiled version.
- *
- * under ARMv5 there's an instruction called CLZ (count leading Zero's) that
- * could be used
- *
- * This is the ffs algorithm devised by d.seal and posted to comp.sys.arm on
- * 16 Feb 1994.
- */
-
-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
-#ifndef __ARM_ARCH_5__
-	/*
-	 * now r0 has at most one set bit, call this X
-	 * if X = 0, all further instructions are skipped
-	 */
-	adrne   r2, .L_ffs_table
-	orrne   r0, r0, r0, lsl #4  /* r0 = X * 0x11 */ 
-	orrne   r0, r0, r0, lsl #6  /* r0 = X * 0x451 */
-	rsbne   r0, r0, r0, lsl #16 /* r0 = X * 0x0450fbaf */
-              
-	/* now lookup in table indexed on top 6 bits of r0 */
-	ldrneb  r0, [ r2, r0, lsr #26 ]
-
-	bx		lr
-
-.text;
-.type .L_ffs_table, _ASM_TYPE_OBJECT;
-.L_ffs_table:
-/*               0   1   2   3   4   5   6   7           */
-	.byte	 0,  1,  2, 13,  3,  7,  0, 14  /*  0- 7 */
-	.byte	 4,  0,  8,  0,  0,  0,  0, 15  /*  8-15 */
-	.byte	11,  5,  0,  0,  9,  0,  0, 26  /* 16-23 */
-	.byte	 0,  0,  0,  0,  0, 22, 28, 16  /* 24-31 */
-	.byte	32, 12,  6,  0,  0,  0,  0,  0	/* 32-39 */
-	.byte	10,  0,  0, 25,  0,  0, 21, 27  /* 40-47 */
-	.byte	31,  0,  0,  0,  0, 24,  0, 20  /* 48-55 */
-	.byte   30,  0, 23, 19, 29, 18, 17,  0  /* 56-63 */
-#else
-	clzne	r0, r0
-	rsbne	r0, r0, #32
-	bx		lr
-#endif
-

libc/arch-arm/bionic/kill.S

@@ -1,53 +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.
- */
-/* 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).
-*/
-#include <sys/linux-syscalls.h>
-
-#ifndef __NR_kill
-#define __NR_kill   37
-#endif
-
-    .text
-    .type kill, #function
-    .globl kill
-    .align 4
-
-kill:
-    stmfd   sp!, {r4-r7, ip, lr}
-    ldr     r7, =__NR_kill
-    swi     #0
-    ldmfd   sp!, {r4-r7, ip, lr}
-    movs    r0, r0
-    bxpl    lr
-    b       __set_syscall_errno

libc/arch-arm/bionic/libgcc_compat.c

@@ -0,0 +1,159 @@
+/* Generated by genlibgcc_compat.py */
+
+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;
+
+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

@@ -27,49 +27,96 @@
  */
 
 #include <machine/cpu-features.h>
+#include <private/bionic_asm.h>
 
-    .text
 
-    .global memcmp
-    .type memcmp, %function
-    .align 4
+#ifdef HAVE_32_BYTE_CACHE_LINE
+#define CACHE_LINE_SIZE     32
+#else
+#define CACHE_LINE_SIZE     64
+#endif
 
 /*
- * Optimized memcmp() 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
+ * Optimized memcmp() for Cortex-A9.
  */
 
-memcmp:
-        PLD         (r0, #0)
-        PLD         (r1, #0)
+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
-        cmpne       r2, #0
         moveq       r0, #0
         bxeq        lr
 
-        /* save registers */
-        stmfd       sp!, {r4, lr}
-        
-        PLD         (r0, #32)
-        PLD         (r1, #32)
-
-        /* since r0 hold the result, move the first source
-         * pointer somewhere else
-         */
-         
-         mov        r4, r0
+        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        8f
+        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 registers */
+        stmfd       sp!, {r4, lr}
+        .cfi_def_cfa_offset 8
+        .cfi_rel_offset r4, 0
+        .cfi_rel_offset lr, 4
+
+        /* since r0 hold the result, move the first source
+         * pointer somewhere else
+         */
+         mov        r4, r0
 
         /* align first pointer to word boundary
          * offset = -src & 3
@@ -107,8 +154,8 @@ memcmp:
         subs        r2, r2, #(32 + 4)
         bmi         1f
 
-0:      PLD         (r4, #64)
-        PLD         (r1, #64)
+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
@@ -175,9 +222,19 @@ memcmp:
         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
@@ -195,8 +252,8 @@ memcmp:
         bic         r1, r1, #3
         ldr         lr, [r1], #4
 
-6:      PLD         (r1, #64)
-        PLD         (r4, #64)
+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
@@ -283,3 +340,4 @@ memcmp:
         mov         r2, #4
 		ldmfd		sp!, {r5, r6, r7}
         b           8b
+END(memcmp)

libc/arch-arm/bionic/memcmp16.S

@@ -1,236 +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>
-
-    .text
-
-    .global __memcmp16
-    .type __memcmp16, %function
-    .align 4
-
-/*
- * 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
- */
-
-__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 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

libc/arch-arm/bionic/memcpy.S

@@ -27,12 +27,310 @@
  */
 
 #include <machine/cpu-features.h>
+#include <private/bionic_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 */
 
-    .global memcpy
-    .type memcpy, %function
-    .align 4
 
 		/*
 		 * Optimized memcpy() for ARM.
@@ -41,12 +339,11 @@
 		 * so we have to preserve R0.
 		 */
 
-memcpy:	
+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
 		 */
-        .fnstart
         .save       {r0, r4, lr}
         stmfd       sp!, {r0, r4, lr}
         /* Making room for r5-r11 which will be spilled later */
@@ -55,9 +352,9 @@ memcpy:
 
         // 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)
+        pld         [r0, #0]
+        pld         [r1, #0]
+        pld         [r1, #32]
 
 		/* it simplifies things to take care of len<4 early */
 		cmp			r2, #4
@@ -145,7 +442,7 @@ cached_aligned32:
         add         r12, r12, #64
 
 1:      ldmia       r1!, { r4-r11 }
-        PLD         (r12, #64)
+        pld         [r12, #64]
         subs        r2, r2, #32
 
         // NOTE: if r12 is more than 64 ahead of r1, the following ldrhi
@@ -266,7 +563,7 @@ loop16:
         ldr         r12, [r1], #4
 1:      mov         r4, r12
 		ldmia		r1!, {   r5,r6,r7,  r8,r9,r10,r11}
-        PLD         (r1, #64)
+        pld         [r1, #64]
         subs        r2, r2, #32
         ldrhs       r12, [r1], #4
 		orr			r3, r3, r4,		lsl #16
@@ -293,7 +590,7 @@ loop8:
         ldr         r12, [r1], #4
 1:      mov         r4, r12
 		ldmia		r1!, {   r5,r6,r7,  r8,r9,r10,r11}
-        PLD         (r1, #64)
+        pld         [r1, #64]
 		subs		r2, r2, #32
         ldrhs       r12, [r1], #4
 		orr			r3, r3, r4,		lsl #24
@@ -320,7 +617,7 @@ loop24:
         ldr         r12, [r1], #4
 1:      mov         r4, r12
 		ldmia		r1!, {   r5,r6,r7,  r8,r9,r10,r11}
-        PLD         (r1, #64)
+        pld         [r1, #64]
 		subs		r2, r2, #32
         ldrhs       r12, [r1], #4
 		orr			r3, r3, r4,		lsl #8
@@ -383,5 +680,7 @@ copy_last_3_and_return:
         add         sp,  sp, #28
 		ldmfd		sp!, {r0, r4, lr}
 		bx			lr
-        .fnend
+END(memcpy)
 
+
+#endif    /* __ARM_ARCH__ < 7 */

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

@@ -0,0 +1,614 @@
+/* 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 <private/bionic_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

@@ -1,117 +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.
- */
-	.text
-
-    .global memset
-    .type memset, %function
-
-    .global bzero
-    .type bzero, %function
-
-    .align
-	
-		/*
-		 * Optimized memset() for ARM.
-         *
-         * memset() returns its first argument.
-		 */
-	
-bzero:
-        mov     r2, r1
-        mov     r1, #0
-
-memset:	
-		/* compute the offset to align the destination
-		 * offset = (4-(src&3))&3 = -src & 3
-		 */
-        .fnstart
-        .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         aligned32
-		cmp         r3, r2
-		andhi		r3, r2, #0x1C
-		sub         r2, r2, r3
-
-		/* conditionnaly 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
-
-aligned32:
-        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
-
-		/* conditionnaly 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
-        .fnend
-    

libc/arch-arm/bionic/setjmp.S

@@ -3,6 +3,7 @@
 
 /*
  * 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
@@ -33,8 +34,9 @@
  * SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
+#include <private/bionic_asm.h>
 #include <machine/setjmp.h>
+#include <machine/cpu-features.h>
 
 /*
  * C library -- setjmp, longjmp
@@ -49,32 +51,39 @@
 ENTRY(setjmp)
 	/* Block all signals and retrieve the old signal mask */
 	stmfd	sp!, {r0, r14}
+	.cfi_def_cfa_offset 8
+	.cfi_rel_offset r0, 0
+	.cfi_rel_offset r14, 4
 	mov	r0, #0x00000000
 
-	bl	PIC_SYM(_C_LABEL(sigblock), PLT)
+	bl	sigblock
 	mov	r1, r0
 
 	ldmfd	sp!, {r0, r14}
+	.cfi_def_cfa_offset 0
 
 	/* Store signal mask */
-	str	r1, [r0, #(25 * 4)]
+	str	r1, [r0, #(_JB_SIGMASK * 4)]
 
 	ldr	r1, .Lsetjmp_magic
-	str	r1, [r0], #4
+	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 */
 
-#ifdef SOFTFLOAT
-	add	r0, r0, #52
-#else
-	/* Store fp registers */
-	sfm	f4, 4, [r0], #48
-	/* Store fpsr */
-	rfs	r1
-	str	r1, [r0], #0x0004
-#endif	/*SOFTFLOAT*/
-	/* Store integer registers */
-        stmia	r0, {r4-r14}
 	mov	r0, #0x00000000
 	bx      lr
+END(setjmp)
 
 .Lsetjmp_magic:
 	.word	_JB_MAGIC_SETJMP
@@ -82,35 +91,42 @@ ENTRY(setjmp)
 
 ENTRY(longjmp)
 	ldr	r2, .Lsetjmp_magic
-	ldr	r3, [r0]
+	ldr	r3, [r0, #(_JB_MAGIC * 4)]
 	teq	r2, r3
 	bne	botch
 
 	/* Fetch signal mask */
-	ldr	r2, [r0, #(25 * 4)]
+	ldr	r2, [r0, #(_JB_SIGMASK * 4)]
 
 	/* Set signal mask */
 	stmfd	sp!, {r0, r1, r14}
+	.cfi_def_cfa_offset 12
+	.cfi_rel_offset r0, 0
+	.cfi_rel_offset r1, 4
+	.cfi_rel_offset r14, 8
 	sub	sp, sp, #4	/* align the stack */
+	.cfi_adjust_cfa_offset 4
 
 	mov	r0, r2
-	bl	PIC_SYM(_C_LABEL(sigsetmask), PLT)
+	bl	sigsetmask
 
 	add	sp, sp, #4	/* unalign the stack */
+	.cfi_adjust_cfa_offset -4
 	ldmfd	sp!, {r0, r1, r14}
+	.cfi_def_cfa_offset 0
 
-	add	r0, r0, #4
-#ifdef SOFTFLOAT
-	add	r0, r0, #52
-#else
-	/* Restore fp registers */
-	lfm	f4, 4, [r0], #48
-	/* Restore FPSR */
-	ldr	r4, [r0], #0x0004
-	wfs	r4
-#endif	/* SOFTFLOAT */
-	/* Restore integer registers */
-        ldmia	r0, {r4-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
@@ -131,6 +147,7 @@ ENTRY(longjmp)
 
 	/* validation failed, die die die. */
 botch:
-	bl	PIC_SYM(_C_LABEL(longjmperror), PLT)
-	bl	PIC_SYM(_C_LABEL(abort), PLT)
+	bl	longjmperror
+	bl	abort
 	b	. - 8		/* Cannot get here */
+END(longjmp)

libc/arch-arm/bionic/sigsetjmp.S

@@ -33,7 +33,7 @@
  * SUCH DAMAGE.
  */
 
-#include <machine/asm.h>
+#include <private/bionic_asm.h>
 #include <machine/setjmp.h>
 
 /*
@@ -48,8 +48,9 @@
 
 ENTRY(sigsetjmp)
 	teq	r1, #0
-	beq	PIC_SYM(_C_LABEL(_setjmp), PLT)
-	b	PIC_SYM(_C_LABEL(setjmp), PLT)
+	beq	_setjmp
+	b	setjmp
+END(sigsetjmp)
 
 .L_setjmp_magic:
 	.word	_JB_MAGIC__SETJMP
@@ -58,5 +59,6 @@ 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)
+	beq	_longjmp
+	b	longjmp
+END(siglongjmp)

libc/arch-arm/bionic/strcmp.S

@@ -0,0 +1,317 @@
+/*
+ * 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 <private/bionic_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/strlen.c

@@ -1,129 +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 <string.h>
-#include <stdint.h>
-#include <machine/cpu-features.h>
-
-size_t strlen(const char *s)
-{
-    __builtin_prefetch(s);
-    __builtin_prefetch(s+32);
-    
-    union {
-        const char      *b;
-        const uint32_t  *w;
-        uintptr_t       i;
-    } u;
-    
-    // these are some scratch variables for the asm code below
-    uint32_t v, t;
-    
-    // initialize the string length to zero
-    size_t l = 0;
-
-    // align the pointer to a 32-bit word boundary
-    u.b = s;
-    while (u.i & 0x3)  {
-        if (__builtin_expect(*u.b++ == 0, 0)) {
-            goto done;
-        }
-        l++;
-    }
-
-    // loop for each word, testing if it contains a zero byte
-    // if so, exit the loop and update the length.
-    // We need to process 32 bytes per loop to schedule PLD properly
-    // and achieve the maximum bus speed.
-    asm(
-        "ldr     %[v], [ %[s] ], #4         \n"
-        "sub     %[l], %[l], %[s]           \n"
-        "0:                                 \n"
-#if __ARM_HAVE_PLD
-        "pld     [ %[s], #64 ]              \n"
-#endif
-        "sub     %[t], %[v], %[mask], lsr #7\n"
-        "and     %[t], %[t], %[mask]        \n"
-        "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
-#if !defined(__OPTIMIZE_SIZE__)
-        "bne     1f                         \n"
-        "sub     %[t], %[v], %[mask], lsr #7\n"
-        "and     %[t], %[t], %[mask]        \n"
-        "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
-        "bne     1f                         \n"
-        "sub     %[t], %[v], %[mask], lsr #7\n"
-        "and     %[t], %[t], %[mask]        \n"
-        "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
-        "bne     1f                         \n"
-        "sub     %[t], %[v], %[mask], lsr #7\n"
-        "and     %[t], %[t], %[mask]        \n"
-        "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
-        "bne     1f                         \n"
-        "sub     %[t], %[v], %[mask], lsr #7\n"
-        "and     %[t], %[t], %[mask]        \n"
-        "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
-        "bne     1f                         \n"
-        "sub     %[t], %[v], %[mask], lsr #7\n"
-        "and     %[t], %[t], %[mask]        \n"
-        "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
-        "bne     1f                         \n"
-        "sub     %[t], %[v], %[mask], lsr #7\n"
-        "and     %[t], %[t], %[mask]        \n"
-        "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
-        "bne     1f                         \n"
-        "sub     %[t], %[v], %[mask], lsr #7\n"
-        "and     %[t], %[t], %[mask]        \n"
-        "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
-#endif
-        "beq     0b                         \n"
-        "1:                                 \n"
-        "add     %[l], %[l], %[s]           \n"
-        "tst     %[v], #0xFF                \n"
-        "beq     2f                         \n"
-        "add     %[l], %[l], #1             \n"
-        "tst     %[v], #0xFF00              \n"
-        "beq     2f                         \n"
-        "add     %[l], %[l], #1             \n"
-        "tst     %[v], #0xFF0000            \n"
-        "addne   %[l], %[l], #1             \n"
-        "2:                                 \n"
-        : [l]"=&r"(l), [v]"=&r"(v), [t]"=&r"(t), [s]"=&r"(u.b)
-        : "%[l]"(l), "%[s]"(u.b), [mask]"r"(0x80808080UL)
-        : "cc"
-    );
-    
-done:
-    return l;
-}

libc/arch-arm/bionic/syscall.S

@@ -25,22 +25,17 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
-#include <sys/linux-syscalls.h>
 
+#include <private/bionic_asm.h>
 
-	.text
-	.align 4
-	.type syscall,#function
-	.globl syscall
- 
-	.text
-	.align
-
-#if __ARM_EABI__
-
-syscall:
+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
@@ -48,24 +43,9 @@ syscall:
     ldmfd   ip, {r3, r4, r5, r6}
     swi     #0
     ldmfd   sp!, {r4, r5, r6, r7}
-    movs    r0, r0
-    bxpl    lr
-    b       __set_syscall_errno
-
-#else
-
-#ifndef __NR_syscall
-#define __NR_syscall    113
-#endif
-
-syscall:
-    stmfd   sp!, {r4, r5, lr}
-    ldr     r4, [sp, #12]
-    ldr     r5, [sp, #16]
-    swi     __NR_syscall
-    ldmfd   sp!, {r4, r5, lr}
-    movs    r0, r0
-    bxpl    lr
-    b       __set_syscall_errno
-
-#endif
+    .cfi_def_cfa_offset 0
+    cmn     r0, #(MAX_ERRNO + 1)
+    bxls    lr
+    neg     r0, r0
+    b       __set_errno_internal
+END(syscall)

libc/arch-arm/bionic/tkill.S

@@ -1,53 +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.
- */
-/* 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).
-*/
-#include <sys/linux-syscalls.h>
-
-#ifndef __NR_tkill
-#define __NR_tkill  238
-#endif
-
-    .text
-    .type tkill, #function
-    .globl tkill
-    .align 4
-
-tkill:
-    stmfd   sp!, {r4-r7, ip, lr}
-    ldr     r7, =__NR_tkill
-    swi     #0
-    ldmfd   sp!, {r4-r7, ip, lr}
-    movs    r0, r0
-    bxpl    lr
-    b       __set_syscall_errno

libc/arch-arm/cortex-a15/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/cortex-a15/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/cortex-a15/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"

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

@@ -0,0 +1,318 @@
+/*
+ * 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.
+ */
+
+ENTRY_PRIVATE(MEMCPY_BASE)
+        .cfi_def_cfa_offset 8
+        .cfi_rel_offset r0, 0
+        .cfi_rel_offset lr, 4
+
+        // 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
+        // for src alignment, and offers the best improvement since
+        // smaller sized copies are dominated by the overhead of
+        // the pre and post main loop.
+        // For larger copies, if src and dst cannot both be aligned to
+        // word boundaries, use the neon code.
+        // For all other copies, align dst to a double word boundary
+        // and copy using LDRD/STRD instructions.
+
+        cmp     r2, #16
+        blo     .L_copy_less_than_16_unknown_align
+
+        // TODO: The aligned copy code is extremely slow copying some large
+        //       buffers so always go through the unaligned path for now.
+        //cmp     r2, #832
+        //bge     .L_check_alignment
+
+.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]!
+        pld         [r1, #(64*4)]
+        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.
+        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}
+
+.L_check_alignment:
+        // If src and dst cannot both be aligned to a word boundary,
+        // use the unaligned copy version.
+        eor     r3, r0, r1
+        ands    r3, r3, #0x3
+        bne     .L_copy_unknown_alignment
+END(MEMCPY_BASE)
+
+ENTRY_PRIVATE(MEMCPY_BASE_ALIGNED)
+        .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]!
+        .cfi_adjust_cfa_offset 8
+        .cfi_rel_offset r4, 0
+        .cfi_rel_offset r5, 4
+        strd    r6, r7, [sp, #-8]!
+        .cfi_adjust_cfa_offset 8
+        .cfi_rel_offset r6, 0
+        .cfi_rel_offset r7, 0
+        strd    r8, r9, [sp, #-8]!
+        .cfi_adjust_cfa_offset 8
+        .cfi_rel_offset r8, 0
+        .cfi_rel_offset r9, 4
+
+        // Optimized for already aligned dst code.
+        ands    ip, r0, #3
+        bne     .L_dst_not_word_aligned
+
+.L_word_aligned:
+        // Align the destination buffer to 8 bytes, to make sure double
+        // loads and stores don't cross a 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 a cacheline boundary.
+        // This check and realignment are only done if there is >= 832
+        // bytes to copy.
+
+        // Dst is word aligned, but check if it is already double word aligned.
+        ands    r3, r0, #4
+        beq     1f
+        ldr     r3, [r1], #4
+        str     r3, [r0], #4
+        sub     r2, #4
+
+1:      // Can only get here if > 64 bytes to copy, so don't do check r2.
+        sub     r2, #64
+
+2:      // Every loop iteration copies 64 bytes.
+        .irp    offset, #0, #8, #16, #24, #32
+        ldrd    r4, r5, [r1, \offset]
+        strd    r4, r5, [r0, \offset]
+        .endr
+
+        ldrd    r4, r5, [r1, #40]
+        ldrd    r6, r7, [r1, #48]
+        ldrd    r8, r9, [r1, #56]
+
+        // Keep the pld as far from the next load as possible.
+        // The amount to prefetch was determined experimentally using
+        // large sizes, and verifying the prefetch size does not affect
+        // the smaller copies too much.
+        // WARNING: If the ldrd and strd instructions get too far away
+        //          from each other, performance suffers. Three loads
+        //          in a row is the best tradeoff.
+        pld     [r1, #(64*16)]
+        strd    r4, r5, [r0, #40]
+        strd    r6, r7, [r0, #48]
+        strd    r8, r9, [r0, #56]
+
+        add     r0, r0, #64
+        add     r1, r1, #64
+        subs    r2, r2, #64
+        bge     2b
+
+        // Fix-up the remaining count and make sure we have >= 32 bytes left.
+        adds    r2, r2, #32
+        blo     4f
+
+        // Copy 32 bytes. These cache lines were already preloaded.
+        .irp    offset, #0, #8, #16, #24
+        ldrd    r4, r5, [r1, \offset]
+        strd    r4, r5, [r0, \offset]
+        .endr
+        add     r1, r1, #32
+        add     r0, r0, #32
+        sub     r2, r2, #32
+4:      // Less than 32 left.
+        add     r2, r2, #32
+        tst     r2, #0x10
+        beq     5f
+        // Copy 16 bytes.
+        .irp    offset, #0, #8
+        ldrd    r4, r5, [r1, \offset]
+        strd    r4, r5, [r0, \offset]
+        .endr
+        add     r1, r1, #16
+        add     r0, r0, #16
+
+5:      // Copy up to 15 bytes (count in r2).
+        movs    ip, r2, lsl #29
+        bcc     1f
+        // Copy 8 bytes.
+        ldrd    r4, r5, [r1], #8
+        strd    r4, r5, [r0], #8
+1:      bge         2f
+        // Copy 4 bytes.
+        ldr     r4, [r1], #4
+        str     r4, [r0], #4
+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]
+
+        // Restore registers: optimized pop {r0, pc}
+        ldrd    r8, r9, [sp], #8
+        ldrd    r6, r7, [sp], #8
+        ldrd    r4, r5, [sp], #8
+        pop     {r0, pc}
+
+.L_dst_not_word_aligned:
+        // Align dst to word.
+        rsb     ip, ip, #4
+        cmp     ip, #2
+
+        itt     gt
+        ldrbgt  lr, [r1], #1
+        strbgt  lr, [r0], #1
+
+        itt     ge
+        ldrbge  lr, [r1], #1
+        strbge  lr, [r0], #1
+
+        ldrb    lr, [r1], #1
+        strb    lr, [r0], #1
+
+        sub     r2, r2, ip
+
+        // Src is guaranteed to be at least word aligned by this point.
+        b       .L_word_aligned
+END(MEMCPY_BASE_ALIGNED)

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

@@ -0,0 +1,191 @@
+/*
+ * 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)
+        stmfd       sp!, {r0}
+        .cfi_def_cfa_offset 4
+        .cfi_rel_offset r0, 0
+
+        // The new algorithm is slower for copies < 16 so use the old
+        // neon code in that case.
+        cmp         r2, #16
+        blo         .L_set_less_than_16_unknown_align
+
+        // Use strd which requires an even and odd register so move the
+        // values so that:
+        //   r0 and r1 contain the memset value
+        //   r2 is the number of bytes to set
+        //   r3 is the destination pointer
+        mov         r3, r0
+
+        // Copy the byte value in every byte of r1.
+        mov         r1, r1, lsl #24
+        orr         r1, r1, r1, lsr #8
+        orr         r1, r1, r1, lsr #16
+
+.L_check_alignment:
+        // Align destination to a double word to avoid the strd crossing
+        // a cache line boundary.
+        ands        ip, r3, #7
+        bne         .L_do_double_word_align
+
+.L_double_word_aligned:
+        mov         r0, r1
+
+        subs        r2, #64
+        blo         .L_set_less_than_64
+
+1:      // Main loop sets 64 bytes at a time.
+        .irp        offset, #0, #8, #16, #24, #32, #40, #48, #56
+        strd        r0, r1, [r3, \offset]
+        .endr
+
+        add         r3, #64
+        subs        r2, #64
+        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.
+        .irp        offset, #0, #8, #16, #24
+        strd        r0, r1, [r3, \offset]
+        .endr
+        add         r3, #32
+
+.L_set_less_than_32:
+        bpl         .L_set_less_than_16
+        // Set 16 bytes.
+        .irp        offset, #0, #8
+        strd        r0, r1, [r3, \offset]
+        .endr
+        add         r3, #16
+
+.L_set_less_than_16:
+        // Less than 16 bytes to set.
+        lsls        ip, r2, #29
+        bcc         .L_set_less_than_8
+
+        // Set 8 bytes.
+        strd        r0, r1, [r3], #8
+
+.L_set_less_than_8:
+        bpl         .L_set_less_than_4
+        // Set 4 bytes
+        str         r1, [r3], #4
+
+.L_set_less_than_4:
+        lsls        ip, r2, #31
+        it          ne
+        strbne      r1, [r3], #1
+        itt         cs
+        strbcs      r1, [r3], #1
+        strbcs      r1, [r3]
+
+        ldmfd       sp!, {r0}
+        bx          lr
+
+.L_do_double_word_align:
+        rsb         ip, ip, #8
+        sub         r2, r2, ip
+        movs        r0, ip, lsl #31
+        it          mi
+        strbmi      r1, [r3], #1
+        itt         cs
+        strbcs      r1, [r3], #1
+        strbcs      r1, [r3], #1
+
+        // Dst is at least word aligned by this point.
+        cmp         ip, #4
+        blo         .L_double_word_aligned
+        str         r1, [r3], #4
+        b           .L_double_word_aligned
+
+.L_set_less_than_16_unknown_align:
+        // Set up to 15 bytes.
+        vdup.8      d0, r1
+        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
+        it          mi
+        strbmi      r1, [r0], #1
+        itt         cs
+        strbcs      r1, [r0], #1
+        strbcs      r1, [r0], #1
+        ldmfd       sp!, {r0}
+        bx          lr
+END(memset)
+
+        .data
+error_string:
+        .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

@@ -0,0 +1,568 @@
+/*
+ * 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.
+ */
+
+#include <private/bionic_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_scan_byte
+    ldrb    r3, [r0]
+    cbz     r3, strcat_r0_scan_done
+    add     r0, #1
+    .endm // m_scan_byte
+
+    .macro m_copy_byte reg, cmd, label
+    ldrb    \reg, [r1], #1
+    strb    \reg, [r0], #1
+    \cmd    \reg, \label
+    .endm // m_copy_byte
+
+ENTRY(strcat)
+    // Quick check to see if src is empty.
+    ldrb    r2, [r1]
+    pld     [r1, #0]
+    cbnz    r2, strcat_continue
+    bx      lr
+
+strcat_continue:
+    // To speed up really small dst strings, unroll checking the first 4 bytes.
+    m_push
+    m_scan_byte
+    m_scan_byte
+    m_scan_byte
+    m_scan_byte
+
+    ands    r3, r0, #7
+    beq     strcat_mainloop
+
+    // Align to a double word (64 bits).
+    rsb     r3, r3, #8
+    lsls    ip, r3, #31
+    beq     strcat_align_to_32
+
+    ldrb    r5, [r0]
+    cbz     r5, strcat_r0_scan_done
+    add     r0, r0, #1
+
+strcat_align_to_32:
+    bcc     strcat_align_to_64
+
+    ldrb    r2, [r0]
+    cbz     r2, strcat_r0_scan_done
+    add     r0, r0, #1
+    ldrb    r4, [r0]
+    cbz     r4, strcat_r0_scan_done
+    add     r0, r0, #1
+
+strcat_align_to_64:
+    tst     r3, #4
+    beq     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
+
+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
+
+strcpy_finish:
+    m_pop
+
+strcpy_continue:
+    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
+
+    .p2align 2
+strcat_mainloop:
+    ldrd    r2, r3, [r0], #8
+
+    pld     [r0, #64]
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     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
+
+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
+    lsls    ip, ip, #1
+    bne     strcat_sub6
+
+    sub     r0, r0, #5
+    b       strcat_r0_scan_done
+
+strcat_sub8:
+    sub     r0, r0, #8
+    b       strcat_r0_scan_done
+
+strcat_sub7:
+    sub     r0, r0, #7
+    b       strcat_r0_scan_done
+
+strcat_sub6:
+    sub     r0, r0, #6
+    b       strcat_r0_scan_done
+
+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
+    lsls    ip, ip, #1
+    bne     strcat_sub2
+
+    sub     r0, r0, #1
+    b       strcat_r0_scan_done
+
+strcat_sub4:
+    sub     r0, r0, #4
+    b       strcat_r0_scan_done
+
+strcat_sub3:
+    sub     r0, r0, #3
+    b       strcat_r0_scan_done
+
+strcat_sub2:
+    sub     r0, r0, #2
+    b       strcat_r0_scan_done
+END(strcat)

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

@@ -0,0 +1,376 @@
+/*
+ * 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/cpu-features.h>
+#include <private/bionic_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
+
+ENTRY(strcmp)
+      /* 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.
+*/
+
+        .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
+        .cfi_def_cfa_offset 16
+        strd    r4, r5, [sp, #8]
+        .cfi_rel_offset r4, 0
+        .cfi_rel_offset r5, 4
+        strd    r6, r7, [sp]
+        .cfi_rel_offset r6, 8
+        .cfi_rel_offset r7, 12
+        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.  */
+        uadd8   ip, \w1, r6
+        sel     ip, r7, r6
+        .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 */
+
+        pld [r0, #0]
+        pld [r1, #0]
+
+        /* Are both strings double-word aligned?  */
+        orr     ip, r0, r1
+        tst     ip, #7
+        bne     .L_do_align
+
+        /* Fast path.  */
+        init
+
+.L_doubleword_aligned:
+
+        /* Get here when the strings to compare are double-word aligned.  */
+        /* Compare two words in every iteration.  */
+        .p2align        2
+2:
+        pld [r0, #16]
+        pld [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=.L_return_24
+        magic_compare_and_branch w1=r3, w2=r5, label=.L_return_35
+        b       2b
+
+.L_do_align:
+        /* Is the first string word-aligned?  */
+        ands    ip, r0, #3
+        beq     .L_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 boundary is 4-ip.  */
+        bic     r0, r0, #3
+        ldr     r2, [r0], #4
+        lsls    ip, ip, #31
+        beq     .L_byte2
+        bcs     .L_byte3
+
+.L_byte1:
+        ldrb    ip, [r1], #1
+        uxtb    r3, r2, ror #BYTE1_OFFSET
+        subs    ip, r3, ip
+        bne     .L_fast_return
+        m_cbz   reg=r3, label=.L_fast_return
+
+.L_byte2:
+        ldrb    ip, [r1], #1
+        uxtb    r3, r2, ror #BYTE2_OFFSET
+        subs    ip, r3, ip
+        bne     .L_fast_return
+        m_cbz   reg=r3, label=.L_fast_return
+
+.L_byte3:
+        ldrb    ip, [r1], #1
+        uxtb    r3, r2, ror #BYTE3_OFFSET
+        subs    ip, r3, ip
+        bne     .L_fast_return
+        m_cbnz  reg=r3, label=.L_word_aligned_r0
+
+.L_fast_return:
+        mov     r0, ip
+        bx      lr
+
+.L_word_aligned_r0:
+        init
+        /* The first string is word-aligned.  */
+        /* Is the second string word-aligned?  */
+        ands    ip, r1, #3
+        bne     .L_strcmp_unaligned
+
+.L_word_aligned:
+        /* The strings are word-aligned. */
+        /* Is the first string double-word aligned?  */
+        tst     r0, #4
+        beq     .L_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=.L_return_24
+
+.L_doubleword_aligned_r0:
+        /* Get here when r0 is double-word aligned.  */
+        /* Is r1 doubleword_aligned?  */
+        tst     r1, #4
+        beq     .L_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:
+        pld [r0, #16]
+        pld [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=.L_return_25
+        ldrd    r4, r5, [r1], #8
+        magic_compare_and_branch w1=r3 w2=r4 label=.L_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     .L_return_25
+        ldr     r5, [r1], #4
+        cmp     ip, #0
+        eor r3, r2, r3
+        S2HIMEM r2, r5, #\offsethi
+        it      eq
+        cmpeq   r3, r2
+        bne     .L_return_32
+        b       7b
+        .endm /* miscmp_word */
+
+.L_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     .L_miscmp_word_16                 /* If ip == 2.  */
+        bge     .L_miscmp_word_24                 /* If ip == 3.  */
+        miscmp_word offsetlo=8 offsethi=24        /* If ip == 1.  */
+.L_miscmp_word_16:  miscmp_word offsetlo=16 offsethi=16
+.L_miscmp_word_24:  miscmp_word offsetlo=24 offsethi=8
+
+
+.L_return_32:
+        setup_return w1=r3, w2=r2
+        b       .L_do_return
+.L_return_34:
+        setup_return w1=r3, w2=r4
+        b       .L_do_return
+.L_return_25:
+        setup_return w1=r2, w2=r5
+        b       .L_do_return
+.L_return_35:
+        setup_return w1=r3, w2=r5
+        b       .L_do_return
+.L_return_24:
+        setup_return w1=r2, w2=r4
+
+.L_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
+        .cfi_def_cfa_offset 0
+        .cfi_restore r4
+        .cfi_restore r5
+        .cfi_restore r6
+        .cfi_restore r7
+
+        /* 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=.L_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
+
+.L_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
+END(strcmp)

libc/arch-arm/cortex-a15/bionic/strcpy.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 STRCPY
+#include "string_copy.S"

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

@@ -0,0 +1,513 @@
+/*
+ * 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
+    ldr     r2, [r1], #4
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+    str     r2, [r0], #4
+
+.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

@@ -0,0 +1,165 @@
+/*
+ * 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.
+ */
+
+#include <private/bionic_asm.h>
+
+    .syntax unified
+
+    .thumb
+    .thumb_func
+
+ENTRY(strlen)
+    pld     [r0, #0]
+    mov     r1, r0
+
+    ands    r3, r0, #7
+    beq     mainloop
+
+    // Align to a double word (64 bits).
+    rsb     r3, r3, #8
+    lsls    ip, r3, #31
+    beq     align_to_32
+
+    ldrb    r2, [r1], #1
+    cbz     r2, update_count_and_return
+
+align_to_32:
+    bcc     align_to_64
+    ands    ip, r3, #2
+    beq     align_to_64
+
+    ldrb    r2, [r1], #1
+    cbz     r2, update_count_and_return
+    ldrb    r2, [r1], #1
+    cbz     r2, update_count_and_return
+
+align_to_64:
+    tst     r3, #4
+    beq     mainloop
+    ldr     r3, [r1], #4
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     zero_in_second_register
+
+    .p2align 2
+mainloop:
+    ldrd    r2, r3, [r1], #8
+
+    pld     [r1, #64]
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     zero_in_first_register
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     zero_in_second_register
+    b       mainloop
+
+update_count_and_return:
+    sub     r0, r1, r0
+    sub     r0, r0, #1
+    bx      lr
+
+zero_in_first_register:
+    sub     r0, r1, r0
+    lsls    r3, ip, #17
+    bne     sub8_and_return
+    bcs     sub7_and_return
+    lsls    ip, ip, #1
+    bne     sub6_and_return
+
+    sub     r0, r0, #5
+    bx      lr
+
+sub8_and_return:
+    sub     r0, r0, #8
+    bx      lr
+
+sub7_and_return:
+    sub     r0, r0, #7
+    bx      lr
+
+sub6_and_return:
+    sub     r0, r0, #6
+    bx      lr
+
+zero_in_second_register:
+    sub     r0, r1, r0
+    lsls    r3, ip, #17
+    bne     sub4_and_return
+    bcs     sub3_and_return
+    lsls    ip, ip, #1
+    bne     sub2_and_return
+
+    sub     r0, r0, #1
+    bx      lr
+
+sub4_and_return:
+    sub     r0, r0, #4
+    bx      lr
+
+sub3_and_return:
+    sub     r0, r0, #3
+    bx      lr
+
+sub2_and_return:
+    sub     r0, r0, #2
+    bx      lr
+END(strlen)

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

@@ -0,0 +1,11 @@
+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 \
+    bionic/memmove.c \

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

@@ -0,0 +1 @@
+include bionic/libc/arch-arm/cortex-a15/cortex-a15.mk

libc/arch-arm/cortex-a8/cortex-a8.mk

@@ -0,0 +1 @@
+include bionic/libc/arch-arm/cortex-a15/cortex-a15.mk

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

@@ -0,0 +1,220 @@
+/*
+ * 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
+    .fpu    neon
+    .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, r0, #7
+    bne     .L_align_src
+
+    .p2align 2
+.L_mainloop:
+    ldmia   r1!, {r2, r3}
+
+    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_zero_in_first_register:
+    sub     r3, r1, r0
+    // Check for zero in byte 0.
+    lsls    r2, ip, #17
+    beq     .L_check_byte1_reg1
+
+    sub     r3, r3, #8
+    b       .L_finish
+
+.L_check_byte1_reg1:
+    bcc     .L_check_byte2_reg1
+
+    sub     r3, r3, #7
+    b       .L_finish
+
+.L_check_byte2_reg1:
+    // Check for zero in byte 2.
+    tst     ip, #0x800000
+    it      ne
+    subne   r3, r3, #6
+    bne     .L_finish
+    sub     r3, r3, #5
+    b       .L_finish
+
+.L_zero_in_second_register:
+    sub     r3, r1, r0
+    // Check for zero in byte 0.
+    lsls    r2, ip, #17
+    beq     .L_check_byte1_reg2
+
+    sub     r3, r3, #4
+    b       .L_finish
+
+.L_check_byte1_reg2:
+    bcc     .L_check_byte2_reg2
+
+    sub     r3, r3, #3
+    b       .L_finish
+
+.L_check_byte2_reg2:
+    // Check for zero in byte 2.
+    tst     ip, #0x800000
+    it      ne
+    subne   r3, r3, #2
+    bne     .L_finish
+    sub     r3, r3, #1
+    b       .L_finish
+
+.L_align_src:
+    // 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_done
+
+.L_align_to_32:
+    bcc     .L_align_to_64
+
+    ldrb    r2, [r1], #1
+    cbz     r2, .L_done
+    ldrb    r2, [r1], #1
+    cbz     r2, .L_done
+
+.L_align_to_64:
+    tst     r3, #4
+    beq     .L_mainloop
+    ldr     r2, [r1], #4
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_second_register
+    b       .L_mainloop
+
+.L_done:
+    sub     r3, r1, r0
+    sub     r3, r3, #1
+
+.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_fail
+
+    // Set up the registers for the memcpy code.
+    mov     r1, r5
+    pld     [r1, #64]
+    mov     r2, r4
+    add     r0, r0, r3
+    pop     {r4, r5}
+
+    // Fall through into the memcpy_base function.
+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_fail)
+    .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_fail)
+
+    .data
+error_string:
+    .string "strcat: prevented write past end of buffer"

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

@@ -0,0 +1,185 @@
+/*
+ * 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
+    .fpu    neon
+    .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, r0, #7
+    bne     .L_align_src
+
+    .p2align 2
+.L_mainloop:
+    ldmia   r0!, {r2, r3}
+
+    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_zero_in_first_register:
+    sub     r3, r0, r1
+    // Check for zero in byte 0.
+    lsls    r2, ip, #17
+    beq     .L_check_byte1_reg1
+
+    sub     r3, r3, #8
+    b       .L_check_size
+
+.L_check_byte1_reg1:
+    bcc     .L_check_byte2_reg1
+
+    sub     r3, r3, #7
+    b       .L_check_size
+
+.L_check_byte2_reg1:
+    // Check for zero in byte 2.
+    tst     ip, #0x800000
+    it      ne
+    subne   r3, r3, #6
+    bne     .L_check_size
+    sub     r3, r3, #5
+    b       .L_check_size
+
+.L_zero_in_second_register:
+    sub     r3, r0, r1
+    // Check for zero in byte 0.
+    lsls    r2, ip, #17
+    beq     .L_check_byte1_reg2
+
+    sub     r3, r3, #4
+    b       .L_check_size
+
+.L_check_byte1_reg2:
+    bcc     .L_check_byte2_reg2
+
+    sub     r3, r3, #3
+    b       .L_check_size
+
+.L_check_byte2_reg2:
+    // Check for zero in byte 2.
+    tst     ip, #0x800000
+    it      ne
+    subne   r3, r3, #2
+    bne     .L_check_size
+    sub     r3, r3, #1
+    b       .L_check_size
+
+.L_align_src:
+    // 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_done
+
+.L_align_to_32:
+    bcc     .L_align_to_64
+
+    ldrb    r2, [r0], #1
+    cbz     r2, .L_done
+    ldrb    r2, [r0], #1
+    cbz     r2, .L_done
+
+.L_align_to_64:
+    tst     r3, #4
+    beq     .L_mainloop
+    ldr     r2, [r0], #4
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .L_zero_in_second_register
+    b       .L_mainloop
+
+.L_done:
+    sub     r3, r0, r1
+    sub     r3, r3, #1
+
+.L_check_size:
+    pld     [r1, #0]
+    pld     [r1, #64]
+    ldr     r0, [sp]
+    cmp     r3, lr
+    bhs     __strcpy_chk_fail
+
+    // Add 1 for copy length to get the string terminator.
+    add     r2, r3, #1
+
+    // Fall through into the memcpy_base function.
+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_fail)
+    .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_fail)
+
+    .data
+error_string:
+    .string "strcpy: prevented write past end of buffer"

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

@@ -0,0 +1,82 @@
+/*
+ * 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>
+#include <private/libc_events.h>
+
+/*
+ * This code assumes it is running on a processor that supports all arm v7
+ * instructions, that supports neon instructions, and that has a 32 byte
+ * cache line.
+ */
+
+        .syntax unified
+        .fpu    neon
+        .thumb
+        .thumb_func
+
+ENTRY(__memcpy_chk)
+        cmp         r2, r3
+        bhi         __memcpy_chk_fail
+
+        // Fall through to memcpy...
+END(__memcpy_chk)
+
+ENTRY(memcpy)
+        pld     [r1, #0]
+        stmfd   sp!, {r0, lr}
+        .cfi_def_cfa_offset 8
+        .cfi_rel_offset r0, 0
+        .cfi_rel_offset lr, 4
+        pld     [r1, #64]
+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+4)
+END(__memcpy_chk_fail)
+
+        .data
+error_string:
+        .string     "memcpy: prevented write past end of buffer"

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

@@ -0,0 +1,223 @@
+/*
+ * 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.
+ */
+
+/*
+ * This code assumes it is running on a processor that supports all arm v7
+ * instructions, that supports neon instructions, and that has a 32 byte
+ * cache line.
+ */
+
+ENTRY_PRIVATE(MEMCPY_BASE)
+        .cfi_def_cfa_offset 8
+        .cfi_rel_offset r0, 0
+        .cfi_rel_offset lr, 4
+
+        // Check so divider is at least 16 bytes, needed for alignment code.
+        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         __memcpy_base_aligned
+
+        /* Check the upper size limit for Neon unaligned memory access in memcpy */
+        cmp         r2, #224
+        blo         3f
+
+        /* 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
+        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, 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, #0]
+        pld         [r1, #(32 * 2)]
+3:
+        /* 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, #(32 * 4)]
+        pld         [r1, #(32 * 6)]
+
+1:      /* The main loop copies 64 bytes at a time */
+        vld1.8      {d0 - d3}, [r1]!
+        vld1.8      {d4 - d7}, [r1]!
+        pld         [r1, #(32 * 6)]
+        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
+        itt         mi
+        ldrbmi      r3, [r1], #1
+        strbmi      r3, [r0], #1
+        itttt       cs
+        ldrbcs      ip, [r1], #1
+        ldrbcs      lr, [r1], #1
+        strbcs      ip, [r0], #1
+        strbcs      lr, [r0], #1
+
+        ldmfd       sp!, {r0, lr}
+        bx          lr
+END(MEMCPY_BASE)
+
+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}
+        .cfi_adjust_cfa_offset 20
+        .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
+        pld         [r1, #(32 * 4)]
+
+        /* 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 */
+        itt         mi
+        ldrbmi      r3, [r1], #1
+        strbmi      r3, [r0], #1
+        itttt       cs
+        ldrbcs      r4, [r1], #1
+        ldrbcs      r5, [r1], #1
+        strbcs      r4, [r0], #1
+        strbcs      r5, [r0], #1
+
+2:
+        subs        r2, r2, #64
+        blt         4f
+
+3:      /* Main copy loop, copying 64 bytes at a time */
+        pld         [r1, #(32 * 8)]
+        ldmia       r1!, {r3, r4, r5, r6, r7, r8, r12, lr}
+        stmia       r0!, {r3, r4, r5, r6, r7, r8, r12, lr}
+        ldmia       r1!, {r3, r4, r5, r6, r7, r8, r12, lr}
+        stmia       r0!, {r3, r4, r5, r6, r7, r8, r12, lr}
+        subs        r2, r2, #64
+        bge         3b
+
+4:      /* Check if there are > 32 bytes left */
+        adds        r2, r2, #64
+        subs        r2, r2, #32
+        blt         5f
+
+        /* Copy 32 bytes */
+        ldmia       r1!, {r3, r4, r5, r6, r7, r8, r12, lr}
+        stmia       r0!, {r3, r4, r5, r6, r7, r8, r12, lr}
+        subs        r2, #32
+
+5:      /* Handle any remaining bytes */
+        adds        r2, #32
+        beq         6f
+
+        movs        r12, r2, lsl #28
+        itt         cs
+        ldmiacs     r1!, {r3, r4, r5, r6}   /* 16 bytes */
+        stmiacs     r0!, {r3, r4, r5, r6}
+        itt         mi
+        ldmiami     r1!, {r7, r8}           /*  8 bytes */
+        stmiami     r0!, {r7, r8}
+        movs        r12, r2, lsl #30
+        itt         cs
+        ldrcs       r3, [r1], #4            /*  4 bytes */
+        strcs       r3, [r0], #4
+        itt         mi
+        ldrhmi      r4, [r1], #2            /*  2 bytes */
+        strhmi      r4, [r0], #2
+        tst         r2, #0x1
+        itt         ne
+        ldrbne      r3, [r1]                /*  last byte  */
+        strbne      r3, [r0]
+6:
+        ldmfd       sp!, {r4-r8}
+        ldmfd       sp!, {r0, pc}
+END(MEMCPY_BASE_ALIGNED)

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

@@ -0,0 +1,188 @@
+/*
+ * 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>
+#include <private/libc_events.h>
+
+/*
+ * This code assumes it is running on a processor that supports all arm v7
+ * instructions and that supports neon instructions.
+ */
+
+    .fpu    neon
+
+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)
+
+/* memset() returns its first argument.  */
+ENTRY(memset)
+        // The neon memset only wins for less than 132.
+        cmp         r2, #132
+        bhi         __memset_large_copy
+
+        stmfd       sp!, {r0}
+        .cfi_def_cfa_offset 4
+        .cfi_rel_offset r0, 0
+
+        vdup.8      q0, r1
+
+        /* 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
+        vst1.8      {d0 - d3}, [r0]!
+        bhs         1b
+
+2:      /* less than 32 left */
+        add         r2, r2, #32
+        tst         r2, #0x10
+        beq         3f
+
+        // writes 16 bytes, 128-bits aligned
+        vst1.8      {d0, d1}, [r0]!
+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
+END(memset)
+
+ENTRY_PRIVATE(__memset_large_copy)
+        /* compute the offset to align the destination
+         * offset = (4-(src&3))&3 = -src & 3
+         */
+        stmfd       sp!, {r0, r4-r7, lr}
+        .cfi_def_cfa_offset 24
+        .cfi_rel_offset r0, 0
+        .cfi_rel_offset r4, 4
+        .cfi_rel_offset r5, 8
+        .cfi_rel_offset r6, 12
+        .cfi_rel_offset r7, 16
+        .cfi_rel_offset lr, 20
+
+        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_large_copy)
+
+        .data
+error_string:
+        .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

@@ -0,0 +1,548 @@
+/*
+ * 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.
+ */
+
+#include <private/bionic_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_scan_byte
+    ldrb    r3, [r0]
+    cbz     r3, strcat_r0_scan_done
+    add     r0, #1
+    .endm // m_scan_byte
+
+    .macro m_copy_byte reg, cmd, label
+    ldrb    \reg, [r1], #1
+    strb    \reg, [r0], #1
+    \cmd    \reg, \label
+    .endm // m_copy_byte
+
+ENTRY(strcat)
+    // Quick check to see if src is empty.
+    ldrb        r2, [r1]
+    pld         [r1, #0]
+    cbnz        r2, strcat_continue
+    bx          lr
+
+strcat_continue:
+    // To speed up really small dst strings, unroll checking the first 4 bytes.
+    m_push
+    m_scan_byte
+    m_scan_byte
+    m_scan_byte
+    m_scan_byte
+
+    ands    r3, r0, #7
+    bne     strcat_align_src
+
+    .p2align 2
+strcat_mainloop:
+    ldmia   r0!, {r2, r3}
+
+    pld     [r0, #64]
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     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
+
+strcat_zero_in_first_register:
+    sub     r0, r0, #4
+
+strcat_zero_in_second_register:
+    // Check for zero in byte 0.
+    tst     ip, #0x80
+    it      ne
+    subne   r0, r0, #4
+    bne     strcat_r0_scan_done
+    // Check for zero in byte 1.
+    tst     ip, #0x8000
+    it      ne
+    subne   r0, r0, #3
+    bne     strcat_r0_scan_done
+    // Check for zero in byte 2.
+    tst     ip, #0x800000
+    it      ne
+    subne   r0, r0, #2
+    it      eq
+    // Zero is in byte 3.
+    subeq   r0, r0, #1
+
+strcat_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
+
+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
+
+strcat_align_src:
+    // Align to a double word (64 bits).
+    rsb     r3, r3, #8
+    lsls    ip, r3, #31
+    beq     strcat_align_to_32
+    ldrb    r2, [r0], #1
+    cbz     r2, strcat_r0_update
+
+strcat_align_to_32:
+    bcc     strcat_align_to_64
+    ldrb    r2, [r0], #1
+    cbz     r2, strcat_r0_update
+    ldrb    r2, [r0], #1
+    cbz     r2, strcat_r0_update
+
+strcat_align_to_64:
+    tst     r3, #4
+    beq     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
+
+strcat_r0_update:
+    sub     r0, r0, #1
+    b strcat_r0_scan_done
+END(strcat)

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

@@ -0,0 +1,548 @@
+/*
+ * 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/cpu-features.h>
+#include <private/bionic_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
+
+ENTRY(strcmp)
+      /* 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.
+*/
+
+        .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
+        .cfi_def_cfa_offset 16
+        strd    r4, r5, [sp, #8]
+        .cfi_rel_offset r4, 0
+        .cfi_rel_offset r5, 4
+        strd    r6, r7, [sp]
+        .cfi_rel_offset r6, 8
+        .cfi_rel_offset r7, 12
+        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.  */
+        uadd8   ip, \w1, r6
+        sel     ip, r7, r6
+        .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 */
+
+        pld [r0, #0]
+        pld [r1, #0]
+
+        /* Are both strings double-word aligned?  */
+        orr     ip, r0, r1
+        tst     ip, #7
+        bne     .L_do_align
+
+        /* Fast path.  */
+        init
+
+.L_doubleword_aligned:
+
+        /* Get here when the strings to compare are double-word aligned.  */
+        /* Compare two words in every iteration.  */
+        .p2align        2
+2:
+        pld [r0, #16]
+        pld [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=.L_return_24
+        magic_compare_and_branch w1=r3, w2=r5, label=.L_return_35
+        b       2b
+
+.L_do_align:
+        /* Is the first string word-aligned?  */
+        ands    ip, r0, #3
+        beq     .L_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 boundary is 4-ip.  */
+        bic     r0, r0, #3
+        ldr     r2, [r0], #4
+        lsls    ip, ip, #31
+        beq     .L_byte2
+        bcs     .L_byte3
+
+.L_byte1:
+        ldrb    ip, [r1], #1
+        uxtb    r3, r2, ror #BYTE1_OFFSET
+        subs    ip, r3, ip
+        bne     .L_fast_return
+        m_cbz   reg=r3, label=.L_fast_return
+
+.L_byte2:
+        ldrb    ip, [r1], #1
+        uxtb    r3, r2, ror #BYTE2_OFFSET
+        subs    ip, r3, ip
+        bne     .L_fast_return
+        m_cbz   reg=r3, label=.L_fast_return
+
+.L_byte3:
+        ldrb    ip, [r1], #1
+        uxtb    r3, r2, ror #BYTE3_OFFSET
+        subs    ip, r3, ip
+        bne     .L_fast_return
+        m_cbnz  reg=r3, label=.L_word_aligned_r0
+
+.L_fast_return:
+        mov     r0, ip
+        bx      lr
+
+.L_word_aligned_r0:
+        init
+        /* The first string is word-aligned.  */
+        /* Is the second string word-aligned?  */
+        ands    ip, r1, #3
+        bne     .L_strcmp_unaligned
+
+.L_word_aligned:
+        /* The strings are word-aligned. */
+        /* Is the first string double-word aligned?  */
+        tst     r0, #4
+        beq     .L_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=.L_return_24
+
+.L_doubleword_aligned_r0:
+        /* Get here when r0 is double-word aligned.  */
+        /* Is r1 doubleword_aligned?  */
+        tst     r1, #4
+        beq     .L_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:
+        pld [r0, #16]
+        pld [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=.L_return_25
+        ldrd    r4, r5, [r1], #8
+        magic_compare_and_branch w1=r3 w2=r4 label=.L_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     .L_return_25
+        ldr     r5, [r1], #4
+        cmp     ip, #0
+        eor r3, r2, r3
+        S2HIMEM r2, r5, #\offsethi
+        it      eq
+        cmpeq   r3, r2
+        bne     .L_return_32
+        b       7b
+        .endm /* miscmp_word */
+
+.L_return_32:
+        setup_return w1=r3, w2=r2
+        b       .L_do_return
+.L_return_34:
+        setup_return w1=r3, w2=r4
+        b       .L_do_return
+.L_return_25:
+        setup_return w1=r2, w2=r5
+        b       .L_do_return
+.L_return_35:
+        setup_return w1=r3, w2=r5
+        b       .L_do_return
+.L_return_24:
+        setup_return w1=r2, w2=r4
+
+.L_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
+        .cfi_def_cfa_offset 0
+        .cfi_restore r4
+        .cfi_restore r5
+        .cfi_restore r6
+        .cfi_restore r7
+
+        /* 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=.L_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
+
+.L_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
+
+    /* The code from the previous version of strcmp.S handles all of the
+     * cases where the first string and seconds string cannot both be
+     * aligned to a word boundary faster than the new algorithm. See
+     * bionic/libc/arch-arm/cortex-a15/bionic/strcmp.S for the unedited
+     * version of the code.
+     */
+.L_strcmp_unaligned:
+	wp1 .req r0
+	wp2 .req r1
+	b1  .req r2
+	w1  .req r4
+	w2  .req r5
+	t1  .req ip
+	@ r3 is scratch
+
+2:
+	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
+
+    /* Restore registers and stack. */
+    ldrd    r6, r7, [sp]
+    ldrd    r4, r5, [sp, #8]
+    adds    sp, sp, #16
+    .cfi_def_cfa_offset 0
+    .cfi_restore r4
+    .cfi_restore r5
+    .cfi_restore r6
+    .cfi_restore r7
+
+	bx	lr
+
+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
+
+    /* Restore registers and stack. */
+    ldrd    r6, r7, [sp]
+    ldrd    r4, r5, [sp, #8]
+    adds    sp, sp, #16
+
+	bx	lr
+END(strcmp)

libc/arch-arm/cortex-a9/bionic/strcpy.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 STRCPY
+#include "string_copy.S"

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

@@ -0,0 +1,535 @@
+/*
+ * 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
+    ldr     r2, [r1], #4
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     .Lstringcopy_zero_in_first_register
+    stmia   r0!, {r2}
+    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

@@ -0,0 +1,167 @@
+/*
+ * 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.
+ */
+
+#include <private/bionic_asm.h>
+
+    .syntax unified
+
+    .thumb
+    .thumb_func
+
+ENTRY(strlen)
+    pld     [r0, #0]
+    mov     r1, r0
+
+    ands    r3, r0, #7
+    bne     align_src
+
+    .p2align 2
+mainloop:
+    ldmia   r1!, {r2, r3}
+
+    pld     [r1, #64]
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     zero_in_first_register
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     zero_in_second_register
+    b       mainloop
+
+zero_in_first_register:
+    sub     r0, r1, r0
+    // Check for zero in byte 0.
+    lsls    r2, ip, #17
+    beq     check_byte1_reg1
+
+    sub     r0, r0, #8
+    bx      lr
+
+check_byte1_reg1:
+    bcc     check_byte2_reg1
+
+    sub     r0, r0, #7
+    bx      lr
+
+check_byte2_reg1:
+    // Check for zero in byte 2.
+    tst     ip, #0x800000
+    itt     ne
+    subne   r0, r0, #6
+    bxne    lr
+    sub     r0, r0, #5
+    bx      lr
+
+zero_in_second_register:
+    sub     r0, r1, r0
+    // Check for zero in byte 0.
+    lsls    r2, ip, #17
+    beq     check_byte1_reg2
+
+    sub     r0, r0, #4
+    bx      lr
+
+check_byte1_reg2:
+    bcc     check_byte2_reg2
+
+    sub     r0, r0, #3
+    bx      lr
+
+check_byte2_reg2:
+    // Check for zero in byte 2.
+    tst     ip, #0x800000
+    itt     ne
+    subne   r0, r0, #2
+    bxne    lr
+    sub     r0, r0, #1
+    bx      lr
+
+align_src:
+    // Align to a double word (64 bits).
+    rsb     r3, r3, #8
+    lsls    ip, r3, #31
+    beq     align_to_32
+
+    ldrb    r2, [r1], #1
+    cbz     r2, done
+
+align_to_32:
+    bcc     align_to_64
+
+    ldrb    r2, [r1], #1
+    cbz     r2, done
+    ldrb    r2, [r1], #1
+    cbz     r2, done
+
+align_to_64:
+    tst     r3, #4
+    beq     mainloop
+    ldr     r2, [r1], #4
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     zero_in_second_register
+    b       mainloop
+
+done:
+    sub     r0, r1, r0
+    sub     r0, r0, #1
+    bx      lr
+END(strlen)

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

@@ -0,0 +1,11 @@
+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 \
+    bionic/memmove.c \

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"

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

@@ -0,0 +1,234 @@
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ * All rights reserved.
+ * Copyright (c) 2013-2014, NVIDIA Corporation.  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 CACHE_LINE_SIZE         (64)
+#define PREFETCH_DISTANCE       (CACHE_LINE_SIZE*6)
+
+ENTRY_PRIVATE(MEMCPY_BASE)
+        .cfi_def_cfa_offset 8
+        .cfi_rel_offset r0, 0
+        .cfi_rel_offset lr, 4
+
+        cmp         r2, #0
+        beq         .L_memcpy_done
+        cmp         r0, r1
+        beq         .L_memcpy_done
+
+        /* preload next cache line */
+        pld         [r1, #CACHE_LINE_SIZE*1]
+
+        /* Deal with very small blocks (< 32bytes) asap */
+        cmp         r2, #32
+        blo         .L_memcpy_lt_32bytes
+        /* no need to align if len < 128 bytes */
+        cmp         r2, #128
+        blo         .L_memcpy_lt_128bytes
+
+        /* large copy, align dest to 64 byte boundry */
+        pld         [r1, #CACHE_LINE_SIZE*2]
+        rsb         r3, r0, #0
+        ands        r3, r3, #0x3F
+        pld         [r1, #CACHE_LINE_SIZE*3]
+        beq         .L_memcpy_dispatch
+        sub         r2, r2, r3
+        /* copy 1 byte */
+        movs        ip, r3, lsl #31
+        itt         mi
+        ldrbmi      ip, [r1], #1
+        strbmi      ip, [r0], #1
+        /* copy 2 bytes */
+        itt         cs
+        ldrhcs      ip, [r1], #2
+        strhcs      ip, [r0], #2
+        /* copy 4 bytes */
+        movs        ip, r3, lsl #29
+        itt         mi
+        ldrmi       ip, [r1], #4
+        strmi       ip, [r0], #4
+        /* copy 8 bytes */
+        bcc         1f
+        vld1.8      {d0}, [r1]!
+        vst1.8      {d0}, [r0, :64]!
+1:      /* copy 16 bytes */
+        movs        ip, r3, lsl #27
+        bpl         1f
+        vld1.8      {q0}, [r1]!
+        vst1.8      {q0}, [r0, :128]!
+1:      /* copy 32 bytes */
+        bcc         .L_memcpy_dispatch
+        vld1.8      {q0, q1}, [r1]!
+        vst1.8      {q0, q1}, [r0, :256]!
+
+.L_memcpy_dispatch:
+        // pre-decrement by 128 to detect nearly-done condition easily, but
+        // also need to check if we have less than 128 bytes left at this
+        // point due to alignment code above
+        subs        r2, r2, #128
+        blo         .L_memcpy_lt_128presub
+
+        // Denver does better if both source and dest are aligned so
+        // we'll special-case that even though the code is virually identical
+        tst         r1, #0xF
+        bne         .L_memcpy_neon_unalign_src_pld
+
+        // DRAM memcpy should be throttled slightly to get full bandwidth
+        //
+        cmp         r2, #32768
+        bhi         .L_memcpy_neon_unalign_src_pld
+        .align      4
+1:
+        /* copy 128 bytes in each loop */
+        subs        r2, r2, #128
+
+        /* preload a cache line */
+        pld         [r1, #PREFETCH_DISTANCE]
+        /* copy a cache line */
+        vld1.8      {q0, q1}, [r1, :128]!
+        vst1.8      {q0, q1}, [r0, :256]!
+        vld1.8      {q0, q1}, [r1, :128]!
+        vst1.8      {q0, q1}, [r0, :256]!
+        /* preload a cache line */
+        pld         [r1, #PREFETCH_DISTANCE]
+        /* copy a cache line */
+        vld1.8      {q0, q1}, [r1, :128]!
+        vst1.8      {q0, q1}, [r0, :256]!
+        vld1.8      {q0, q1}, [r1, :128]!
+        vst1.8      {q0, q1}, [r0, :256]!
+
+        bhs         1b
+        adds        r2, r2, #128
+        bne         .L_memcpy_lt_128bytes_align
+        pop         {r0, pc}
+
+        .align      4
+.L_memcpy_neon_unalign_src_pld:
+1:
+        /* copy 128 bytes in each loop */
+        subs        r2, r2, #128
+
+        /* preload a cache line */
+        pld         [r1, #PREFETCH_DISTANCE]
+        /* copy a cache line */
+        vld1.8      {q0, q1}, [r1]!
+        vst1.8      {q0, q1}, [r0, :256]!
+        vld1.8      {q0, q1}, [r1]!
+        vst1.8      {q0, q1}, [r0, :256]!
+        /* preload a cache line */
+        pld         [r1, #PREFETCH_DISTANCE]
+        /* copy a cache line */
+        vld1.8      {q0, q1}, [r1]!
+        vst1.8      {q0, q1}, [r0, :256]!
+        vld1.8      {q0, q1}, [r1]!
+        vst1.8      {q0, q1}, [r0, :256]!
+
+        bhs         1b
+        adds        r2, r2, #128
+        bne         .L_memcpy_lt_128bytes_align
+        pop         {r0, pc}
+
+.L_memcpy_lt_128presub:
+        add         r2, r2, #128
+.L_memcpy_lt_128bytes_align:
+        /* copy 64 bytes */
+        movs        ip, r2, lsl #26
+        bcc         1f
+        vld1.8      {q0, q1}, [r1]!
+        vst1.8      {q0, q1}, [r0, :256]!
+        vld1.8      {q0, q1}, [r1]!
+        vst1.8      {q0, q1}, [r0, :256]!
+1:      /* copy 32 bytes */
+        bpl         1f
+        vld1.8      {q0, q1}, [r1]!
+        vst1.8      {q0, q1}, [r0, :256]!
+1:      /* copy 16 bytes */
+        movs        ip, r2, lsl #28
+        bcc         1f
+        vld1.8      {q0}, [r1]!
+        vst1.8      {q0}, [r0, :128]!
+1:      /* copy 8 bytes */
+        bpl         1f
+        vld1.8      {d0}, [r1]!
+        vst1.8      {d0}, [r0, :64]!
+1:      /* copy 4 bytes */
+        tst         r2, #4
+        itt         ne
+        ldrne       ip, [r1], #4
+        strne       ip, [r0], #4
+        /* copy 2 bytes */
+        movs        ip, r2, lsl #31
+        itt         cs
+        ldrhcs      ip, [r1], #2
+        strhcs      ip, [r0], #2
+        /* copy 1 byte */
+        itt         mi
+        ldrbmi      ip, [r1]
+        strbmi      ip, [r0]
+
+        pop         {r0, pc}
+
+.L_memcpy_lt_128bytes:
+        /* copy 64 bytes */
+        movs        ip, r2, lsl #26
+        bcc         1f
+        vld1.8      {q0, q1}, [r1]!
+        vst1.8      {q0, q1}, [r0]!
+        vld1.8      {q0, q1}, [r1]!
+        vst1.8      {q0, q1}, [r0]!
+1:      /* copy 32 bytes */
+        bpl	    .L_memcpy_lt_32bytes
+        vld1.8      {q0, q1}, [r1]!
+        vst1.8      {q0, q1}, [r0]!
+.L_memcpy_lt_32bytes:
+        /* copy 16 bytes */
+        movs        ip, r2, lsl #28
+        bcc         1f
+        vld1.8      {q0}, [r1]!
+        vst1.8      {q0}, [r0]!
+1:      /* copy 8 bytes */
+        bpl         1f
+        vld1.8      {d0}, [r1]!
+        vst1.8      {d0}, [r0]!
+1:      /* copy 4 bytes */
+        tst         r2, #4
+        itt         ne
+        ldrne       ip, [r1], #4
+        strne       ip, [r0], #4
+        /* copy 2 bytes */
+        movs        ip, r2, lsl #31
+        itt         cs
+        ldrhcs      ip, [r1], #2
+        strhcs      ip, [r0], #2
+        /* copy 1 byte */
+        itt         mi
+        ldrbmi      ip, [r1]
+        strbmi      ip, [r0]
+
+.L_memcpy_done:
+        pop         {r0, pc}
+END(MEMCPY_BASE)

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

@@ -0,0 +1,281 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ * All rights reserved.
+ * Copyright (c) 2013-2014 NVIDIA Corporation.  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>
+
+        .text
+        .syntax unified
+        .fpu    neon
+
+#define CACHE_LINE_SIZE         (64)
+#define MEMCPY_BLOCK_SIZE_SMALL (32768)
+#define MEMCPY_BLOCK_SIZE_MID   (1048576)
+#define PREFETCH_DISTANCE_NEAR  (CACHE_LINE_SIZE*4)
+#define PREFETCH_DISTANCE_MID   (CACHE_LINE_SIZE*4)
+#define PREFETCH_DISTANCE_FAR   (CACHE_LINE_SIZE*16)
+
+ENTRY(memmove)
+        cmp         r2, #0
+        cmpne       r0, r1
+        bxeq        lr
+        subs        r3, r0, r1
+        bls         .L_jump_to_memcpy
+        cmp         r2, r3
+        bhi         .L_reversed_memcpy
+
+.L_jump_to_memcpy:
+        b           memcpy
+
+.L_reversed_memcpy:
+        push        {r0, lr}
+        .cfi_def_cfa_offset 8
+        .cfi_rel_offset r0, 0
+        .cfi_rel_offset lr, 4
+
+        add         r0, r0, r2
+        add         r1, r1, r2
+
+        /* preload next cache line */
+        pld         [r1, #-CACHE_LINE_SIZE]
+        pld         [r1, #-CACHE_LINE_SIZE*2]
+
+.L_reversed_memcpy_align_dest:
+        /* Deal with very small blocks (< 32bytes) asap */
+        cmp         r2, #32
+        blo         .L_reversed_memcpy_lt_32bytes
+        /* no need to align if len < 128 bytes */
+        cmp         r2, #128
+        blo         .L_reversed_memcpy_lt_128bytes
+        /* align destination to 64 bytes (1 cache line) */
+        ands        r3, r0, #0x3f
+        beq         .L_reversed_memcpy_dispatch
+        sub         r2, r2, r3
+0:      /* copy 1 byte */
+        movs        ip, r3, lsl #31
+        ldrbmi      ip, [r1, #-1]!
+        strbmi      ip, [r0, #-1]!
+1:      /* copy 2 bytes */
+        ldrbcs      ip, [r1, #-1]!
+        strbcs      ip, [r0, #-1]!
+        ldrbcs      ip, [r1, #-1]!
+        strbcs      ip, [r0, #-1]!
+2:      /* copy 4 bytes */
+        movs        ip, r3, lsl #29
+        bpl         3f
+        sub         r1, r1, #4
+        sub         r0, r0, #4
+        vld4.8      {d0[0], d1[0], d2[0], d3[0]}, [r1]
+        vst4.8      {d0[0], d1[0], d2[0], d3[0]}, [r0, :32]
+3:      /* copy 8 bytes */
+        bcc         4f
+        sub         r1, r1, #8
+        sub         r0, r0, #8
+        vld1.8      {d0}, [r1]
+        vst1.8      {d0}, [r0, :64]
+4:      /* copy 16 bytes */
+        movs        ip, r3, lsl #27
+        bpl         5f
+        sub         r1, r1, #16
+        sub         r0, r0, #16
+        vld1.8      {q0}, [r1]
+        vst1.8      {q0}, [r0, :128]
+5:      /* copy 32 bytes */
+        bcc         .L_reversed_memcpy_dispatch
+        sub         r1, r1, #32
+        sub         r0, r0, #32
+        vld1.8      {q0, q1}, [r1]
+        vst1.8      {q0, q1}, [r0, :256]
+
+.L_reversed_memcpy_dispatch:
+        /* preload more cache lines */
+        pld         [r1, #-CACHE_LINE_SIZE*3]
+        pld         [r1, #-CACHE_LINE_SIZE*4]
+
+        cmp         r2, #MEMCPY_BLOCK_SIZE_SMALL
+        blo         .L_reversed_memcpy_neon_pld_near
+        cmp         r2, #MEMCPY_BLOCK_SIZE_MID
+        blo         .L_reversed_memcpy_neon_pld_mid
+        b           .L_reversed_memcpy_neon_pld_far
+
+.L_reversed_memcpy_neon_pld_near:
+        /* less than 128 bytes? */
+        subs        r2, r2, #128
+        blo         1f
+        sub         r1, r1, #32
+        sub         r0, r0, #32
+        mov         r3, #-32
+        .align      4
+0:
+        /* copy 128 bytes in each loop */
+        subs        r2, r2, #128
+
+        /* preload to cache */
+        pld         [r1, #-(PREFETCH_DISTANCE_NEAR+CACHE_LINE_SIZE*2)+32]
+        /* copy a cache line */
+        vld1.8      {q0, q1}, [r1], r3
+        vst1.8      {q0, q1}, [r0, :256], r3
+        vld1.8      {q0, q1}, [r1], r3
+        vst1.8      {q0, q1}, [r0, :256], r3
+
+        /* preload to cache */
+        pld         [r1, #-(PREFETCH_DISTANCE_NEAR+CACHE_LINE_SIZE*2)+32]
+        /* copy a cache line */
+        vld1.8      {q0, q1}, [r1], r3
+        vst1.8      {q0, q1}, [r0, :256], r3
+        vld1.8      {q0, q1}, [r1], r3
+        vst1.8      {q0, q1}, [r0, :256], r3
+
+        bhs         0b
+        add         r1, r1, #32
+        add         r0, r0, #32
+1:
+        adds        r2, r2, #128
+        bne         .L_reversed_memcpy_lt_128bytes
+        pop         {r0, pc}
+
+.L_reversed_memcpy_neon_pld_mid:
+        subs        r2, r2, #128
+        sub         r1, r1, #32
+        sub         r0, r0, #32
+        mov         r3, #-32
+        .align      4
+0:
+        /* copy 128 bytes in each loop */
+        subs        r2, r2, #128
+
+        /* preload to cache */
+        pld         [r1, #-(PREFETCH_DISTANCE_MID+CACHE_LINE_SIZE)+32]
+        /* copy a cache line */
+        vld1.8      {q0, q1}, [r1], r3
+        vst1.8      {q0, q1}, [r0, :256], r3
+        vld1.8      {q0, q1}, [r1], r3
+        vst1.8      {q0, q1}, [r0, :256], r3
+
+        /* preload to cache */
+        pld         [r1, #-(PREFETCH_DISTANCE_MID+CACHE_LINE_SIZE)+32]
+        /* copy a cache line */
+        vld1.8      {q0, q1}, [r1], r3
+        vst1.8      {q0, q1}, [r0, :256], r3
+        vld1.8      {q0, q1}, [r1], r3
+        vst1.8      {q0, q1}, [r0, :256], r3
+
+        bhs         0b
+        add         r1, r1, #32
+        add         r0, r0, #32
+1:
+        adds        r2, r2, #128
+        bne         .L_reversed_memcpy_lt_128bytes
+        pop         {r0, pc}
+
+.L_reversed_memcpy_neon_pld_far:
+        sub         r2, r2, #128
+        sub         r0, r0, #128
+        sub         r1, r1, #128
+        .align      4
+0:
+        /* copy 128 bytes in each loop */
+        subs        r2, r2, #128
+
+        /* preload to cache */
+        pld         [r1, #-(PREFETCH_DISTANCE_FAR+CACHE_LINE_SIZE*2)+128]
+        pld         [r1, #-(PREFETCH_DISTANCE_FAR+CACHE_LINE_SIZE)+128]
+        /* read */
+        vld1.8      {q0, q1}, [r1]!
+        vld1.8      {q2, q3}, [r1]!
+        vld1.8      {q8, q9}, [r1]!
+        vld1.8      {q10, q11}, [r1]!
+        /* write */
+        vst1.8      {q0, q1}, [r0, :256]!
+        vst1.8      {q2, q3}, [r0, :256]!
+        vst1.8      {q8, q9}, [r0, :256]!
+        vst1.8      {q10, q11}, [r0, :256]!
+
+        sub         r0, r0, #256
+        sub         r1, r1, #256
+        bhs         0b
+        add         r0, r0, #128
+        add         r1, r1, #128
+1:
+        adds        r2, r2, #128
+        bne         .L_reversed_memcpy_lt_128bytes
+        pop         {r0, pc}
+
+.L_reversed_memcpy_lt_128bytes:
+6:      /* copy 64 bytes */
+        movs        ip, r2, lsl #26
+        bcc         5f
+        sub         r1, r1, #32
+        sub         r0, r0, #32
+        vld1.8      {q0, q1}, [r1]
+        vst1.8      {q0, q1}, [r0]
+        sub         r1, r1, #32
+        sub         r0, r0, #32
+        vld1.8      {q0, q1}, [r1]
+        vst1.8      {q0, q1}, [r0]
+5:      /* copy 32 bytes */
+        bpl         4f
+        sub         r1, r1, #32
+        sub         r0, r0, #32
+        vld1.8      {q0, q1}, [r1]
+        vst1.8      {q0, q1}, [r0]
+.L_reversed_memcpy_lt_32bytes:
+4:      /* copy 16 bytes */
+        movs        ip, r2, lsl #28
+        bcc         3f
+        sub         r1, r1, #16
+        sub         r0, r0, #16
+        vld1.8      {q0}, [r1]
+        vst1.8      {q0}, [r0]
+3:      /* copy 8 bytes */
+        bpl         2f
+        sub         r1, r1, #8
+        sub         r0, r0, #8
+        vld1.8      {d0}, [r1]
+        vst1.8      {d0}, [r0]
+2:      /* copy 4 bytes */
+        ands        ip, r2, #0x4
+        beq         1f
+        sub         r1, r1, #4
+        sub         r0, r0, #4
+        vld4.8      {d0[0], d1[0], d2[0], d3[0]}, [r1]
+        vst4.8      {d0[0], d1[0], d2[0], d3[0]}, [r0]
+1:      /* copy 2 bytes */
+        movs        ip, r2, lsl #31
+        ldrbcs      ip, [r1, #-1]!
+        strbcs      ip, [r0, #-1]!
+        ldrbcs      ip, [r1, #-1]!
+        strbcs      ip, [r0, #-1]!
+0:      /* copy 1 byte */
+        ldrbmi      ip, [r1, #-1]!
+        strbmi      ip, [r0, #-1]!
+
+        pop         {r0, pc}
+
+END(memmove)

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

@@ -0,0 +1,207 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ * Copyright (c) 2014, NVIDIA CORPORATION.  All rights reserved.
+ * 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)
+        pldw        [r0]
+        mov         r3, r0
+
+        // Duplicate the low byte of r1
+        mov         r1, r1, lsl #24
+        orr         r1, r1, r1, lsr #8
+        orr         r1, r1, r1, lsr #16
+
+        cmp         r2, #16
+        blo         .L_less_than_16
+
+        // This section handles regions 16 bytes or larger
+        //
+        // Use aligned vst1.8 and vstm when possible.  Register values will be:
+        //   ip is scratch
+        //   q0, q1, and r1 contain the memset value
+        //   r2 is the number of bytes to set
+        //   r3 is the advancing destination pointer
+        vdup.32     q0, r1
+
+        ands        ip, r3, 0xF
+        beq         .L_memset_aligned
+
+        // Align dest pointer to 16-byte boundary.
+        pldw        [r0, #64]
+        rsb         ip, ip, #16
+
+        // Pre-adjust the byte count to reflect post-aligment value.  Expecting
+        // 8-byte alignment to be rather common so we special case that one.
+        sub         r2, r2, ip
+
+        /* set 1 byte */
+        tst         ip, #1
+        it          ne
+        strbne      r1, [r3], #1
+        /* set 2 bytes */
+        tst         ip, #2
+        it          ne
+        strhne      r1, [r3], #2
+        /* set 4 bytes */
+        movs        ip, ip, lsl #29
+        it          mi
+        strmi       r1, [r3], #4
+        /* set 8 bytes */
+        itt         cs
+        strcs       r1, [r3], #4
+        strcs       r1, [r3], #4
+
+.L_memset_aligned:
+        // Destination is now 16-byte aligned.  Determine how to handle
+        // remaining bytes.
+        vmov        q1, q0
+        cmp         r2, #128
+        blo         .L_less_than_128
+
+        // We need to set a larger block of memory.  Use four Q regs to
+        // set a full cache line in one instruction.  Pre-decrement
+        // r2 to simplify end-of-loop detection
+        vmov        q2, q0
+        vmov        q3, q0
+        pldw        [r0, #128]
+        sub         r2, r2, #128
+        .align 4
+.L_memset_loop_128:
+        pldw        [r3, #192]
+        vstm        r3!, {q0, q1, q2, q3}
+        vstm        r3!, {q0, q1, q2, q3}
+        subs        r2, r2, #128
+        bhs         .L_memset_loop_128
+
+        // Un-bias r2 so it contains the number of bytes left.  Early
+        // exit if we are done.
+        adds        r2, r2, #128
+        beq         2f
+
+        .align 4
+.L_less_than_128:
+        // set 64 bytes
+        movs        ip, r2, lsl #26
+        bcc         1f
+        vst1.8      {q0, q1}, [r3, :128]!
+        vst1.8      {q0, q1}, [r3, :128]!
+        beq         2f
+1:
+        // set 32 bytes
+        bpl         1f
+        vst1.8      {q0, q1}, [r3, :128]!
+1:
+        // set 16 bytes
+        movs        ip, r2, lsl #28
+        bcc         1f
+        vst1.8      {q0}, [r3, :128]!
+        beq         2f
+1:
+        // set 8 bytes
+        bpl         1f
+        vst1.8      {d0}, [r3, :64]!
+1:
+        // set 4 bytes
+        tst         r2, #4
+        it          ne
+        strne       r1, [r3], #4
+1:
+        // set 2 bytes
+        movs        ip, r2, lsl #31
+        it          cs
+        strhcs      r1, [r3], #2
+        // set 1 byte
+        it          mi
+        strbmi      r1, [r3]
+2:
+        bx          lr
+
+.L_less_than_16:
+        // Store up to 15 bytes without worrying about byte alignment
+        movs        ip, r2, lsl #29
+        bcc         1f
+        str         r1, [r3], #4
+        str         r1, [r3], #4
+        beq         2f
+1:
+        it          mi
+        strmi       r1, [r3], #4
+        movs        ip, r2, lsl #31
+        it          mi
+        strbmi      r1, [r3], #1
+        itt         cs
+        strbcs      r1, [r3], #1
+        strbcs      r1, [r3]
+2:
+        bx          lr
+END(memset)
+
+        .data
+error_string:
+        .string     "memset: prevented write past end of buffer"

libc/arch-arm/denver/denver.mk

@@ -0,0 +1,14 @@
+libc_bionic_src_files_arm += \
+    arch-arm/denver/bionic/memcpy.S \
+    arch-arm/denver/bionic/memmove.S \
+    arch-arm/denver/bionic/memset.S \
+    arch-arm/denver/bionic/__strcat_chk.S \
+    arch-arm/denver/bionic/__strcpy_chk.S \
+
+# Use cortex-a15 versions of strcat/strcpy/strlen.
+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/arch-arm/generic/bionic/memcpy.S

@@ -0,0 +1,404 @@
+/*
+ * 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 <private/bionic_asm.h>
+#include <private/libc_events.h>
+
+        /*
+         * Optimized memcpy() for ARM.
+         *
+         * note that memcpy() always returns the destination pointer,
+         * so we have to preserve R0.
+         */
+
+ENTRY(__memcpy_chk)
+        cmp         r2, r3
+        bgt         fortify_check_failed
+
+        // Fall through to memcpy...
+END(__memcpy_chk)
+
+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
+
+        /* conditionally 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
+
+        // Only reached when the __memcpy_chk check fails.
+fortify_check_failed:
+        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/generic/bionic/memset.S

@@ -0,0 +1,133 @@
+/*
+ * 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>
+#include <private/libc_events.h>
+
+        /*
+         * Optimized memset() for ARM.
+         *
+         * memset() returns its first argument.
+         */
+
+ENTRY(__memset_chk)
+        cmp         r2, r3
+        bls         done
+
+        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
+
+done:
+        // Fall through to memset...
+END(bzero)
+
+ENTRY(memset)
+        /* 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)
+
+        .data
+error_string:
+        .string     "memset: prevented write past end of buffer"

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

@@ -0,0 +1,317 @@
+/*
+ * 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 <private/bionic_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/generic/bionic/strcpy.S

@@ -0,0 +1,133 @@
+/*
+ * Copyright (C) 2010 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.
+ *
+ * Android adaptation and tweak by Jim Huang <jserv@0xlab.org>.
+ */
+
+#include <machine/cpu-features.h>
+#include <private/bionic_asm.h>
+
+ENTRY(strcpy)
+	pld	[r1, #0]
+	eor	r2, r0, r1
+	mov	ip, r0
+	tst	r2, #3
+	bne	4f
+	tst	r1, #3
+	bne	3f
+5:
+	str	r5, [sp, #-4]!
+	mov	r5, #0x01
+	orr	r5, r5, r5, lsl #8
+	orr	r5, r5, r5, lsl #16
+
+	str	r4, [sp, #-4]!
+	tst	r1, #4
+	ldr	r3, [r1], #4
+	beq	2f
+	sub	r2, r3, r5
+	bics	r2, r2, r3
+	tst	r2, r5, lsl #7
+	itt	eq
+	streq	r3, [ip], #4
+	ldreq	r3, [r1], #4
+	bne	1f
+       /* Inner loop.  We now know that r1 is 64-bit aligned, so we
+	  can safely fetch up to two words.  This allows us to avoid
+	  load stalls.  */
+	.p2align 2
+2:
+	pld	[r1, #8]
+	ldr	r4, [r1], #4
+	sub	r2, r3, r5
+	bics	r2, r2, r3
+	tst	r2, r5, lsl #7
+	sub	r2, r4, r5
+	bne	1f
+	str	r3, [ip], #4
+	bics	r2, r2, r4
+	tst	r2, r5, lsl #7
+	itt	eq
+	ldreq	r3, [r1], #4
+	streq	r4, [ip], #4
+	beq	2b
+	mov	r3, r4
+1:
+#ifdef __ARMEB__
+	rors	r3, r3, #24
+#endif
+	strb	r3, [ip], #1
+	tst	r3, #0xff
+#ifdef __ARMEL__
+	ror	r3, r3, #8
+#endif
+	bne	1b
+	ldr	r4, [sp], #4
+	ldr	r5, [sp], #4
+	bx	lr
+
+       /* Strings have the same offset from word alignment, but it's
+	  not zero.  */
+3:
+	tst	r1, #1
+	beq	1f
+	ldrb	r2, [r1], #1
+	strb	r2, [ip], #1
+	cmp	r2, #0
+	it	eq
+	bxeq	lr
+1:
+	tst	r1, #2
+	beq	5b
+	ldrh	r2, [r1], #2
+#ifdef __ARMEB__
+	tst	r2, #0xff00
+	iteet	ne
+	strneh	r2, [ip], #2
+	lsreq	r2, r2, #8
+	streqb	r2, [ip]
+	tstne	r2, #0xff
+#else
+	tst	r2, #0xff
+	itet	ne
+	strneh	r2, [ip], #2
+	streqb	r2, [ip]
+	tstne	r2, #0xff00
+#endif
+	bne	5b
+	bx	lr
+
+       /* src and dst do not have a common word-alignement.  Fall back to
+	  byte copying.  */
+4:
+	ldrb	r2, [r1], #1
+	strb	r2, [ip], #1
+	cmp	r2, #0
+	bne	4b
+	bx	lr
+END(strcpy)

libc/arch-arm/generic/bionic/strlen.c

@@ -0,0 +1,136 @@
+/*
+ * 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 <string.h>
+#include <stdint.h>
+#include <machine/cpu-features.h>
+
+size_t strlen(const char *s)
+{
+    __builtin_prefetch(s);
+    __builtin_prefetch(s+32);
+
+    union {
+        const char      *b;
+        const uint32_t  *w;
+        uintptr_t       i;
+    } u;
+
+    // these are some scratch variables for the asm code below
+    uint32_t v, t;
+
+    // initialize the string length to zero
+    size_t l = 0;
+
+    // align the pointer to a 32-bit word boundary
+    u.b = s;
+    while (u.i & 0x3)  {
+        if (__builtin_expect(*u.b++ == 0, 0)) {
+            goto done;
+        }
+        l++;
+    }
+
+    // loop for each word, testing if it contains a zero byte
+    // if so, exit the loop and update the length.
+    // We need to process 32 bytes per loop to schedule PLD properly
+    // and achieve the maximum bus speed.
+    asm(
+        "ldr     %[v], [%[s]], #4           \n"
+        "sub     %[l], %[l], %[s]           \n"
+        "0:                                 \n"
+        "pld     [%[s], #64]                \n"
+        "sub     %[t], %[v], %[mask], lsr #7\n"
+        "and     %[t], %[t], %[mask]        \n"
+        "bics    %[t], %[t], %[v]           \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
+#if !defined(__OPTIMIZE_SIZE__)
+        "bne     1f                         \n"
+        "sub     %[t], %[v], %[mask], lsr #7\n"
+        "and     %[t], %[t], %[mask]        \n"
+        "bics    %[t], %[t], %[v]           \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
+        "bne     1f                         \n"
+        "sub     %[t], %[v], %[mask], lsr #7\n"
+        "and     %[t], %[t], %[mask]        \n"
+        "bics    %[t], %[t], %[v]           \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
+        "bne     1f                         \n"
+        "sub     %[t], %[v], %[mask], lsr #7\n"
+        "and     %[t], %[t], %[mask]        \n"
+        "bics    %[t], %[t], %[v]           \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
+        "bne     1f                         \n"
+        "sub     %[t], %[v], %[mask], lsr #7\n"
+        "and     %[t], %[t], %[mask]        \n"
+        "bics    %[t], %[t], %[v]           \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
+        "bne     1f                         \n"
+        "sub     %[t], %[v], %[mask], lsr #7\n"
+        "and     %[t], %[t], %[mask]        \n"
+        "bics    %[t], %[t], %[v]           \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
+        "bne     1f                         \n"
+        "sub     %[t], %[v], %[mask], lsr #7\n"
+        "and     %[t], %[t], %[mask]        \n"
+        "bics    %[t], %[t], %[v]           \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
+        "bne     1f                         \n"
+        "sub     %[t], %[v], %[mask], lsr #7\n"
+        "and     %[t], %[t], %[mask]        \n"
+        "bics    %[t], %[t], %[v]           \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
+#endif
+        "beq     0b                         \n"
+        "1:                                 \n"
+        "add     %[l], %[l], %[s]           \n"
+        "tst     %[v], #0xFF                \n"
+        "beq     2f                         \n"
+        "add     %[l], %[l], #1             \n"
+        "tst     %[v], #0xFF00              \n"
+        "beq     2f                         \n"
+        "add     %[l], %[l], #1             \n"
+        "tst     %[v], #0xFF0000            \n"
+        "it      ne                         \n"
+        "addne   %[l], %[l], #1             \n"
+        "2:                                 \n"
+        : [l]"=&r"(l), [v]"=&r"(v), [t]"=&r"(t), [s]"=&r"(u.b)
+        : "%[l]"(l), "%[s]"(u.b), [mask]"r"(0x80808080UL)
+        : "cc"
+    );
+
+done:
+    return l;
+}

libc/arch-arm/generic/generic.mk

@@ -0,0 +1,11 @@
+libc_bionic_src_files_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/memmove.c \
+    bionic/__strcat_chk.cpp \
+    bionic/__strcpy_chk.cpp \
+    upstream-openbsd/lib/libc/string/stpcpy.c \
+    upstream-openbsd/lib/libc/string/strcat.c \

libc/arch-arm/include/endian.h

@@ -1,10 +0,0 @@
-/*	$OpenBSD: endian.h,v 1.3 2005/12/13 00:35:23 millert Exp $	*/
-
-#ifdef __ARMEB__
-#define _BYTE_ORDER _BIG_ENDIAN
-#else
-#define _BYTE_ORDER _LITTLE_ENDIAN
-#endif
-#define	__STRICT_ALIGNMENT
-#include <sys/types.h>
-#include <sys/endian.h>

libc/arch-arm/include/machine/_types.h

@@ -1,127 +0,0 @@
-/*	$OpenBSD: _types.h,v 1.3 2006/02/14 18:12:58 miod Exp $	*/
-
-/*-
- * Copyright (c) 1990, 1993
- *	The Regents of the University of California.  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. 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 REGENTS 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 REGENTS 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.
- *
- *	@(#)types.h	8.3 (Berkeley) 1/5/94
- *	@(#)ansi.h	8.2 (Berkeley) 1/4/94
- */
-
-#ifndef _ARM__TYPES_H_
-#define _ARM__TYPES_H_
-
-
-#if !defined(__ARM_EABI__)
-/* the kernel defines size_t as unsigned int, but g++ wants it to be unsigned long */
-#define _SIZE_T
-#define _SSIZE_T
-#define _PTRDIFF_T
-typedef unsigned long  size_t;
-typedef long           ssize_t;
-typedef long           ptrdiff_t;
-#endif
-
-//#include <linux/types.h>
-
-/* 7.18.1.1 Exact-width integer types */
-typedef	__signed char		__int8_t;
-typedef	unsigned char		__uint8_t;
-typedef	short			__int16_t;
-typedef	unsigned short		__uint16_t;
-typedef	int			__int32_t;
-typedef	unsigned int		__uint32_t;
-/* LONGLONG */
-typedef	long long		__int64_t;
-/* LONGLONG */
-typedef	unsigned long long	__uint64_t;
-
-/* 7.18.1.2 Minimum-width integer types */
-typedef	__int8_t		__int_least8_t;
-typedef	__uint8_t		__uint_least8_t;
-typedef	__int16_t		__int_least16_t;
-typedef	__uint16_t		__uint_least16_t;
-typedef	__int32_t		__int_least32_t;
-typedef	__uint32_t		__uint_least32_t;
-typedef	__int64_t		__int_least64_t;
-typedef	__uint64_t		__uint_least64_t;
-
-/* 7.18.1.3 Fastest minimum-width integer types */
-typedef	__int32_t		__int_fast8_t;
-typedef	__uint32_t		__uint_fast8_t;
-typedef	__int32_t		__int_fast16_t;
-typedef	__uint32_t		__uint_fast16_t;
-typedef	__int32_t		__int_fast32_t;
-typedef	__uint32_t		__uint_fast32_t;
-typedef	__int64_t		__int_fast64_t;
-typedef	__uint64_t		__uint_fast64_t;
-
-/* 7.18.1.4 Integer types capable of holding object pointers */
-typedef	int 			__intptr_t;
-typedef	unsigned int 		__uintptr_t;
-
-/* 7.18.1.5 Greatest-width integer types */
-typedef	__int64_t		__intmax_t;
-typedef	__uint64_t		__uintmax_t;
-
-/* Register size */
-typedef __int32_t		__register_t;
-
-/* VM system types */
-typedef unsigned long		__vaddr_t;
-typedef unsigned long		__paddr_t;
-typedef unsigned long		__vsize_t;
-typedef unsigned long		__psize_t;
-
-/* Standard system types */
-typedef int			__clock_t;
-typedef int			__clockid_t;
-typedef long			__ptrdiff_t;
-typedef	int			__time_t;
-typedef int			__timer_t;
-#if defined(__GNUC__) && __GNUC__ >= 3
-typedef	__builtin_va_list	__va_list;
-#else
-typedef	char *			__va_list;
-#endif
-
-/* Wide character support types */
-#ifndef __cplusplus
-typedef	int			__wchar_t;
-#endif
-typedef int			__wint_t;
-typedef	int			__rune_t;
-typedef	void *			__wctrans_t;
-typedef	void *			__wctype_t;
-
-#ifdef __ARMEB__
-#define _BYTE_ORDER _BIG_ENDIAN
-#else
-#define _BYTE_ORDER _LITTLE_ENDIAN
-#endif
-
-#endif	/* _ARM__TYPES_H_ */

libc/arch-arm/include/machine/asm.h

@@ -38,93 +38,14 @@
 #ifndef _ARM32_ASM_H_
 #define _ARM32_ASM_H_
 
-#ifdef __ELF__
-# define _C_LABEL(x)	x
-#else
-# ifdef __STDC__
-#  define _C_LABEL(x)	_ ## x
-# else
-#  define _C_LABEL(x)	_/**/x
-# endif
-#endif
-#define	_ASM_LABEL(x)	x
+#define __bionic_asm_align 0
 
-#ifdef __STDC__
-# define __CONCAT(x,y)	x ## y
-# define __STRING(x)	#x
-#else
-# define __CONCAT(x,y)	x/**/y
-# define __STRING(x)	"x"
-#endif
+#undef __bionic_asm_custom_entry
+#undef __bionic_asm_custom_end
+#define __bionic_asm_custom_entry(f) .fnstart
+#define __bionic_asm_custom_end(f) .fnend
 
-#ifndef _ALIGN_TEXT
-# define _ALIGN_TEXT .align 0
-#endif
-
-/*
- * gas/arm uses @ as a single comment character and thus cannot be used here
- * Instead it recognised the # instead of an @ symbols in .type directives
- * We define a couple of macros so that assembly code will not be dependant
- * on one or the other.
- */
-#define _ASM_TYPE_FUNCTION	#function
-#define _ASM_TYPE_OBJECT	#object
-#define _ENTRY(x) \
-	.text; _ALIGN_TEXT; .globl x; .type x,_ASM_TYPE_FUNCTION; x:
-
-#ifdef GPROF
-# ifdef __ELF__
-#  define _PROF_PROLOGUE	\
-	mov ip, lr; bl __mcount
-# else
-#  define _PROF_PROLOGUE	\
-	mov ip,lr; bl mcount
-# endif
-#else
-# define _PROF_PROLOGUE
-#endif
-
-#define	ENTRY(y)	_ENTRY(_C_LABEL(y)); _PROF_PROLOGUE
-#define	ENTRY_NP(y)	_ENTRY(_C_LABEL(y))
-#define	ASENTRY(y)	_ENTRY(_ASM_LABEL(y)); _PROF_PROLOGUE
-#define	ASENTRY_NP(y)	_ENTRY(_ASM_LABEL(y))
-
-#define	ASMSTR		.asciz
-
-#if defined(__ELF__) && defined(PIC)
-#ifdef __STDC__
-#define	PIC_SYM(x,y)	x ## ( ## y ## )
-#else
-#define	PIC_SYM(x,y)	x/**/(/**/y/**/)
-#endif
-#else
-#define	PIC_SYM(x,y)	x
-#endif
-
-#ifdef __ELF__
-#define RCSID(x)	.section ".ident"; .asciz x
-#else
-#define RCSID(x)	.text; .asciz x
-#endif
-
-#ifdef __ELF__
-#define	WEAK_ALIAS(alias,sym)						\
-	.weak alias;							\
-	alias = sym
-#endif
-
-#ifdef __STDC__
-#define	WARN_REFERENCES(sym,msg)					\
-	.stabs msg ## ,30,0,0,0 ;					\
-	.stabs __STRING(_C_LABEL(sym)) ## ,1,0,0,0
-#elif defined(__ELF__)
-#define	WARN_REFERENCES(sym,msg)					\
-	.stabs msg,30,0,0,0 ;						\
-	.stabs __STRING(sym),1,0,0,0
-#else
-#define	WARN_REFERENCES(sym,msg)					\
-	.stabs msg,30,0,0,0 ;						\
-	.stabs __STRING(_/**/sym),1,0,0,0
-#endif /* __STDC__ */
+#undef __bionic_asm_function_type
+#define __bionic_asm_function_type #function
 
 #endif /* !_ARM_ASM_H_ */

libc/arch-arm/include/machine/cdefs.h

@@ -1,19 +0,0 @@
-/*	$OpenBSD: cdefs.h,v 1.2 2005/11/24 20:46:44 deraadt Exp $	*/
-
-#ifndef	_MACHINE_CDEFS_H_
-#define	_MACHINE_CDEFS_H_
-
-#if defined(lint)
-#define __indr_reference(sym,alias)	__lint_equal__(sym,alias)
-#define __warn_references(sym,msg)
-#define __weak_alias(alias,sym)		__lint_equal__(sym,alias)
-#elif defined(__GNUC__) && defined(__STDC__)
-#define __weak_alias(alias,sym)					\
-	__asm__(".weak " __STRING(alias) " ; " __STRING(alias)	\
-	    " = " __STRING(sym));
-#define	__warn_references(sym,msg)				\
-	__asm__(".section .gnu.warning." __STRING(sym)		\
-	    " ; .ascii \"" msg "\" ; .text");
-#endif
-
-#endif /* !_MACHINE_CDEFS_H_ */

libc/arch-arm/include/machine/cpu-features.h

@@ -34,131 +34,50 @@
  *
  * This is done to abstract us from the various ARM Architecture
  * quirks and alphabet soup.
- *
- * IMPORTANT: We have no intention to support anything below an ARMv4T !
  */
 
-/* _ARM_ARCH_REVISION is a number corresponding to the ARM revision
- * we're going to support
- *
- * it looks like our toolchain doesn't define __ARM_ARCH__
+/* __ARM_ARCH__ is a number corresponding to the ARM revision
+ * we're going to support. Our toolchain doesn't define __ARM_ARCH__
  * so try to guess it.
- *
- *
- *
  */
 #ifndef __ARM_ARCH__
-
 #  if defined __ARM_ARCH_7__   || defined __ARM_ARCH_7A__ || \
         defined __ARM_ARCH_7R__  || defined __ARM_ARCH_7M__
-
 #    define __ARM_ARCH__ 7
-
 #  elif defined __ARM_ARCH_6__   || defined __ARM_ARCH_6J__ || \
         defined __ARM_ARCH_6K__  || defined __ARM_ARCH_6Z__ || \
         defined __ARM_ARCH_6KZ__ || defined __ARM_ARCH_6T2__
-#
 #    define __ARM_ARCH__ 6
-#
-#  elif defined __ARM_ARCH_5__ || defined __ARM_ARCH_5T__ || \
-        defined __ARM_ARCH_5TE__ || defined __ARM_ARCH_5TEJ__
-#
-#    define __ARM_ARCH__ 5
-#
-#  elif defined __ARM_ARCH_4T__
-#
-#    define __ARM_ARCH__ 4
-#
-#  elif defined __ARM_ARCH_4__
-#    error ARMv4 is not supported, please use ARMv4T at a minimum
 #  else
 #    error Unknown or unsupported ARM architecture
 #  endif
 #endif
 
-/* experimental feature used to check that our ARMv4 workarounds
- * work correctly without a real ARMv4 machine */
-#ifdef BIONIC_EXPERIMENTAL_FORCE_ARMV4
-#  undef  __ARM_ARCH__
-#  define __ARM_ARCH__  4
-#endif
-
-/* define __ARM_HAVE_5TE if we have the ARMv5TE instructions */
-#if __ARM_ARCH__ > 5
-#  define  __ARM_HAVE_5TE  1
-#elif __ARM_ARCH__ == 5
-#  if defined __ARM_ARCH_5TE__ || defined __ARM_ARCH_5TEJ__
-#    define __ARM_HAVE_5TE  1
-#  endif
-#endif
-
-/* instructions introduced in ARMv5 */
-#if __ARM_ARCH__ >= 5
-#  define  __ARM_HAVE_BLX  1
-#  define  __ARM_HAVE_CLZ  1
-#  define  __ARM_HAVE_LDC2 1
-#  define  __ARM_HAVE_MCR2 1
-#  define  __ARM_HAVE_MRC2 1
-#  define  __ARM_HAVE_STC2 1
-#endif
-
-/* ARMv5TE introduces a few instructions */
-#if __ARM_HAVE_5TE
-#  define  __ARM_HAVE_PLD   1
-#  define  __ARM_HAVE_MCRR  1
-#  define  __ARM_HAVE_MRRC  1
-#endif
-
 /* define __ARM_HAVE_HALFWORD_MULTIPLY when half-word multiply instructions
  * this means variants of: smul, smulw, smla, smlaw, smlal
  */
-#if __ARM_HAVE_5TE
-#  define  __ARM_HAVE_HALFWORD_MULTIPLY  1
+#define  __ARM_HAVE_HALFWORD_MULTIPLY  1
+
+/* define __ARM_HAVE_LDREXD for ARMv7 architecture
+ * (also present in ARMv6K, but not implemented in ARMv7-M, neither of which
+ * we care about)
+ */
+#if __ARM_ARCH__ >= 7
+#  define __ARM_HAVE_LDREXD
 #endif
 
-/* define __ARM_HAVE_PAIR_LOAD_STORE when 64-bit memory loads and stored
- * into/from a pair of 32-bit registers is supported throuhg 'ldrd' and 'strd'
+/* define _ARM_HAVE_VFP if we have VFPv3
  */
-#if __ARM_HAVE_5TE
-#  define  __ARM_HAVE_PAIR_LOAD_STORE 1
+#if __ARM_ARCH__ >= 7 && defined __VFP_FP__
+#  define __ARM_HAVE_VFP
 #endif
 
-/* define __ARM_HAVE_SATURATED_ARITHMETIC is you have the saturated integer
- * arithmetic instructions: qdd, qdadd, qsub, qdsub
+/* define _ARM_HAVE_NEON for ARMv7 architecture if we support the
+ * Neon SIMD instruction set extensions. This also implies
+ * that VFPv3-D32 is supported.
  */
-#if __ARM_HAVE_5TE
-#  define  __ARM_HAVE_SATURATED_ARITHMETIC 1
-#endif
-
-/* define __ARM_HAVE_PC_INTERWORK when a direct assignment to the
- * pc register will switch into thumb/ARM mode depending on bit 0
- * of the new instruction address. Before ARMv5, this was not the
- * case, and you have to write:
- *
- *     mov  r0, [<some address>]
- *     bx   r0
- *
- * instead of:
- *
- *     ldr  pc, [<some address>]
- *
- * note that this affects any instruction that explicitely changes the
- * value of the pc register, including ldm { ...,pc } or 'add pc, #offset'
- */
-#if __ARM_ARCH__ >= 5
-#  define __ARM_HAVE_PC_INTERWORK
-#endif
-
-
-/* Assembly-only macros */
-
-/* define a handy PLD(address) macro since the cache preload
- * is an optional opcode
- */
-#if __ARM_HAVE_PLD
-#  define  PLD(reg,offset)    pld    [reg, offset]
-#else
-#  define  PLD(reg,offset)    /* nothing */
+#if __ARM_ARCH__ >= 7 && defined __ARM_NEON__
+#  define __ARM_HAVE_NEON
 #endif
 
 #endif /* _ARM_MACHINE_CPU_FEATURES_H */

libc/arch-arm/include/machine/elf_machdep.h

@@ -0,0 +1,138 @@
+/*	$NetBSD: elf_machdep.h,v 1.10 2012/08/05 04:12:46 matt Exp $	*/
+
+#ifndef _ARM_ELF_MACHDEP_H_
+#define _ARM_ELF_MACHDEP_H_
+
+#if defined(__ARMEB__)
+#define ELF32_MACHDEP_ENDIANNESS	ELFDATA2MSB
+#else
+#define ELF32_MACHDEP_ENDIANNESS	ELFDATA2LSB
+#endif
+
+#define ELF64_MACHDEP_ENDIANNESS	XXX	/* break compilation */
+#define ELF64_MACHDEP_ID_CASES                                          \
+		/* no 64-bit ELF machine types supported */
+
+/* Processor specific flags for the ELF header e_flags field.  */
+#define EF_ARM_RELEXEC		0x00000001
+#define EF_ARM_HASENTRY		0x00000002
+#define EF_ARM_INTERWORK	0x00000004 /* GNU binutils 000413 */
+#define EF_ARM_SYMSARESORTED	0x00000004 /* ARM ELF A08 */
+#define EF_ARM_APCS_26		0x00000008 /* GNU binutils 000413 */
+#define EF_ARM_DYNSYMSUSESEGIDX	0x00000008 /* ARM ELF B01 */
+#define EF_ARM_APCS_FLOAT	0x00000010 /* GNU binutils 000413 */
+#define EF_ARM_MAPSYMSFIRST	0x00000010 /* ARM ELF B01 */
+#define EF_ARM_PIC		0x00000020
+#define EF_ARM_ALIGN8		0x00000040 /* 8-bit structure alignment.  */
+#define EF_ARM_NEW_ABI		0x00000080
+#define EF_ARM_OLD_ABI		0x00000100
+#define EF_ARM_SOFT_FLOAT	0x00000200
+#define EF_ARM_EABIMASK		0xff000000
+#define	EF_ARM_EABI_VER1	0x01000000
+#define	EF_ARM_EABI_VER2	0x02000000
+#define	EF_ARM_EABI_VER3	0x03000000
+#define	EF_ARM_EABI_VER4	0x04000000
+#define	EF_ARM_EABI_VER5	0x05000000
+
+#define	ELF32_MACHDEP_ID_CASES						\
+		case EM_ARM:						\
+			break;
+
+#define	ELF32_MACHDEP_ID	EM_ARM
+
+#define ARCH_ELFSIZE		32	/* MD native binary size */
+
+/* Processor specific relocation types */
+
+#define R_ARM_NONE		0
+#define R_ARM_PC24		1
+#define R_ARM_ABS32		2
+#define R_ARM_REL32		3
+#define R_ARM_PC13		4
+#define R_ARM_ABS16		5
+#define R_ARM_ABS12		6
+#define R_ARM_THM_ABS5		7
+#define R_ARM_ABS8		8
+#define R_ARM_SBREL32		9
+#define R_ARM_THM_PC22		10
+#define R_ARM_THM_PC8		11
+#define R_ARM_AMP_VCALL9	12
+#define R_ARM_SWI24		13
+#define R_ARM_THM_SWI8		14
+#define R_ARM_XPC25		15
+#define R_ARM_THM_XPC22		16
+
+/* TLS relocations */
+#define R_ARM_TLS_DTPMOD32	17	/* ID of module containing symbol */
+#define R_ARM_TLS_DTPOFF32	18	/* Offset in TLS block */
+#define R_ARM_TLS_TPOFF32	19	/* Offset in static TLS block */
+
+/* 20-31 are reserved for ARM Linux. */
+#define R_ARM_COPY		20
+#define R_ARM_GLOB_DAT		21
+#define	R_ARM_JUMP_SLOT		22
+#define R_ARM_RELATIVE		23
+#define	R_ARM_GOTOFF		24
+#define R_ARM_GOTPC		25
+#define R_ARM_GOT32		26
+#define R_ARM_PLT32		27
+
+#define R_ARM_ALU_PCREL_7_0	32
+#define R_ARM_ALU_PCREL_15_8	33
+#define R_ARM_ALU_PCREL_23_15	34
+#define R_ARM_ALU_SBREL_11_0	35
+#define R_ARM_ALU_SBREL_19_12	36
+#define R_ARM_ALU_SBREL_27_20	37
+
+/* 96-111 are reserved to G++. */
+#define R_ARM_GNU_VTENTRY	100
+#define R_ARM_GNU_VTINHERIT	101
+#define R_ARM_THM_PC11		102
+#define R_ARM_THM_PC9		103
+
+/* More TLS relocations */
+#define R_ARM_TLS_GD32		104	/* PC-rel 32 bit for global dynamic */
+#define R_ARM_TLS_LDM32		105	/* PC-rel 32 bit for local dynamic */
+#define R_ARM_TLS_LDO32		106	/* 32 bit offset relative to TLS */
+#define R_ARM_TLS_IE32		107	/* PC-rel 32 bit for GOT entry of */
+#define R_ARM_TLS_LE32		108
+#define R_ARM_TLS_LDO12		109
+#define R_ARM_TLS_LE12		110
+#define R_ARM_TLS_IE12GP	111
+
+/* 112-127 are reserved for private experiments. */
+
+#define R_ARM_RXPC25		249
+#define R_ARM_RSBREL32		250
+#define R_ARM_THM_RPC22		251
+#define R_ARM_RREL32		252
+#define R_ARM_RABS32		253
+#define R_ARM_RPC24		254
+#define R_ARM_RBASE		255
+
+#define R_TYPE(name)		__CONCAT(R_ARM_,name)
+
+/* Processor specific program header flags */
+#define PF_ARM_SB		0x10000000
+#define PF_ARM_PI		0x20000000
+#define PF_ARM_ENTRY		0x80000000
+
+/* Processor specific section header flags */
+#define SHF_ENTRYSECT		0x10000000
+#define SHF_COMDEF		0x80000000
+
+/* Processor specific symbol types */
+#define STT_ARM_TFUNC		STT_LOPROC
+
+#ifdef _KERNEL
+#ifdef ELFSIZE
+#define	ELF_MD_PROBE_FUNC	ELFNAME2(arm_netbsd,probe)
+#endif
+
+struct exec_package;
+
+int arm_netbsd_elf32_probe(struct lwp *, struct exec_package *, void *, char *,
+	vaddr_t *);
+#endif
+
+#endif /* _ARM_ELF_MACHDEP_H_ */

libc/arch-arm/include/machine/endian.h

@@ -0,0 +1,79 @@
+/*	$OpenBSD: endian.h,v 1.3 2005/12/13 00:35:23 millert Exp $	*/
+
+/*
+ * 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.
+ */
+
+#ifndef _ARM_ENDIAN_H_
+#define _ARM_ENDIAN_H_
+
+#ifdef __GNUC__
+
+/* According to RealView Assembler User's Guide, REV and REV16 are available
+ * in Thumb code and 16-bit instructions when used in Thumb-2 code.
+ *
+ * REV Rd, Rm
+ *   Rd and Rm must both be Lo registers.
+ *
+ * REV16 Rd, Rm
+ *   Rd and Rm must both be Lo registers.
+ *
+ * The +l constraint takes care of this without constraining us in ARM mode.
+ */
+#define __swap16md(x) ({                                        \
+    register u_int16_t _x = (x);                                \
+    __asm__ __volatile__("rev16 %0, %0" : "+l" (_x));           \
+    _x;                                                         \
+})
+
+#define __swap32md(x) ({                                        \
+    register u_int32_t _x = (x);                                \
+    __asm__ __volatile__("rev %0, %0" : "+l" (_x));             \
+    _x;                                                         \
+})
+
+#define __swap64md(x) ({                                        \
+    u_int64_t _swap64md_x = (x);                                \
+    (u_int64_t) __swap32md(_swap64md_x >> 32) |                 \
+        (u_int64_t) __swap32md(_swap64md_x & 0xffffffff) << 32; \
+})
+
+/* Tell sys/endian.h we have MD variants of the swap macros.  */
+#define MD_SWAP
+
+#endif  /* __GNUC__ */
+
+#if defined(__ARMEB__)
+#define _BYTE_ORDER _BIG_ENDIAN
+#else
+#define _BYTE_ORDER _LITTLE_ENDIAN
+#endif
+#define __STRICT_ALIGNMENT
+#include <sys/types.h>
+#include <sys/endian.h>
+
+#endif  /* !_ARM_ENDIAN_H_ */