Merge "resolved conflicts for merge of 5d0ad38c to jb-mr2-dev" into jb-mr2-dev

Change-Id: I606f6d86aad46c5eda0dc0d907fe691ca0eea826

ABI-bugs.txt

@@ -8,5 +8,8 @@ KNOWN ABI BUGS
   sigset_t is too small on ARM and x86 (but correct on MIPS), so support
   for real-time signals is broken. http://b/5828899
 
+  Too few TLS slots mean we can't allocate 128 pthread_key_t instances,
+  which POSIX says should be the minimum.
+
   atexit(3) handlers registered by a shared library aren't called on
   dlclose(3); this only affects ARM. http://b/4998315

CleanSpec.mk

@@ -44,10 +44,6 @@
 #$(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
 # ************************************************

HACKING.txt

@@ -1,162 +0,0 @@
-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/?
-----------------
-
-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.
-
-
-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.
-

benchmarks/Android.mk

@@ -1,80 +0,0 @@
-#
-# 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_C_INCLUDES += external/stlport/stlport bionic/ bionic/libstdc++/include
-LOCAL_SHARED_LIBRARIES += libstlport
-LOCAL_SRC_FILES := $(benchmark_src_files)
-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_main.cpp

@@ -1,226 +0,0 @@
-/*
- * 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

@@ -1,211 +0,0 @@
-/*
- * 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.
-double d;
-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

@@ -1,233 +0,0 @@
-/*
- * 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

@@ -1,107 +0,0 @@
-/*
- * 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);

benchmarks/semaphore_benchmark.cpp

@@ -1,49 +0,0 @@
-/*
- * 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

@@ -1,61 +0,0 @@
-/*
- * 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/time_benchmark.cpp

@@ -1,79 +0,0 @@
-/*
- * 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

@@ -1,67 +0,0 @@
-/*
- * 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

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

libc/SYSCALLS.TXT

@@ -1,334 +1,321 @@
-# This file is used to automatically generate bionic's system call stubs.
+# This file is used to automatically generate bionic's the system calls stubs.
 #
-# Each non-blank, non-comment line has the following format:
+# Each non comment line has the following format:
 #
-# return_type func_name[|alias_list][:syscall_name[:socketcall_id]]([parameter_list]) arch_list
+# return_type    func_name[:syscall_name[:call_id]]([parameter_list])  (1|-1|"stub")
 #
-# where:
-#       arch_list ::= "all" | arch+
-#       arch      ::= "arm" | "arm64" | "mips" | "mips64" | "x86" | "x86_64"
-#
-# Note:
+# Note that:
 #      - 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)
-#
-#      - alias_list is optional comma separated list of function aliases.
-#
-#      - The call_id parameter, given that func_name and syscall_name have
+#        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 in 32 bits.
+#      - each parameter type is assumed to be stored on 32 bits, there is no plan to support
+#        64-bit architectures at the moment
+#
+#      - the final field can be "1", meaning: generate a stub for each architecture,
+#        taking the constants from the kernel header files.
+#
+#      - the final field can be "stub" meaning: do not generate any stubs ---
+#        in this case, a hand-written custom stub must be provided.
+#        TODO: replace this with something like "custom" or "none", or remove
+#        it entirely.
+#
+#      - the final field can be a three-element list of 1s and -1 meaning:
+#        this system call is only available on some of the architectures (1),
+#        and no stub should be generated for those architectures marked with -1.
+#        the order is arm,x86,mips.
+#        TODO: replace this with something more readable like "-arm,-mips" (meaning x86 only).
 #
 # This file is processed by a python script named gensyscalls.py.
+#
 
-int     execve(const char*, char* const*, char* const*)  all
+# process management
+void    _exit:exit_group (int)      1
+void    _exit_thread:exit (int)     1
+pid_t   __fork:fork (void)           1
+pid_t   _waitpid:waitpid (pid_t, int*, int, struct rusage*)   -1,1,1
+int     __waitid:waitid(int, pid_t, struct siginfo_t*, int,void*)          1
+pid_t   wait4(pid_t pid, int *status, int options, struct rusage *rusage)   1
 
-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
+# NOTE: this system call is never called directly, but we list it there
+#       to have __NR_clone properly defined.
+#
+pid_t   __sys_clone:clone (int, void*, int*, void*, int*) 1
 
-# <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     execve (const char*, char* const*, char* const*)  1
 
-int     setgroups:setgroups32(int, const gid_t*)   arm,x86
-int     setgroups:setgroups(int, const gid_t*)     arm64,mips,mips64,x86_64
-int     setpgid(pid_t, pid_t)  all
-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
+int     __setuid:setuid32 (uid_t)    1,1,-1
+int     __setuid:setuid (uid_t)   -1,-1,1
+uid_t   getuid:getuid32 ()         1,1,-1
+uid_t   getuid:getuid ()           -1,-1,1
+gid_t   getgid:getgid32 ()         1,1,-1
+gid_t   getgid:getgid ()           -1,-1,1
+uid_t   geteuid:geteuid32 ()       1,1,-1
+uid_t   geteuid:geteuid ()         -1,-1,1
+gid_t   getegid:getegid32 ()       1,1,-1
+gid_t   getegid:getegid ()         -1,-1,1
+uid_t   getresuid:getresuid32 (uid_t *ruid, uid_t *euid, uid_t *suid)   1,1,-1
+uid_t   getresuid:getresuid (uid_t *ruid, uid_t *euid, uid_t *suid)     -1,-1,1
+gid_t   getresgid:getresgid32 (gid_t *rgid, gid_t *egid, gid_t *sgid)   1,1,-1
+gid_t   getresgid:getresgid (gid_t *rgid, gid_t *egid, gid_t *sgid)     -1,-1,1
+pid_t   gettid()                   1
+ssize_t readahead(int, off64_t, size_t)     1
+int     getgroups:getgroups32(int, gid_t *)    1,1,-1
+int     getgroups:getgroups(int, gid_t *)      -1,-1,1
+pid_t   getpgid(pid_t)             1
+pid_t   getppid()                  1
+pid_t   getsid(pid_t)              1
+pid_t   setsid()                   1
+int     setgid:setgid32(gid_t)     1,1,-1
+int     setgid:setgid(gid_t)       -1,-1,1
+int     seteuid:seteuid32(uid_t)   stub
+int     __setreuid:setreuid32(uid_t, uid_t)   1,1,-1
+int     __setreuid:setreuid(uid_t, uid_t)     -1,-1,1
+int     __setresuid:setresuid32(uid_t, uid_t, uid_t)   1,1,-1
+int     __setresuid:setresuid(uid_t, uid_t, uid_t)     -1,-1,1
+int     setresgid:setresgid32(gid_t, gid_t, gid_t)   1,1,-1
+int     setresgid:setresgid(gid_t, gid_t, gid_t)     -1,-1,1
+void*   __brk:brk(void*)           1
+# see comments in arch-arm/bionic/kill.S to understand why we don't generate an ARM stub for kill/tkill
+int     kill(pid_t, int)           -1,1,1
+int     tkill(pid_t tid, int sig)  -1,1,1
+int     tgkill(pid_t tgid, pid_t tid, int sig)  -1,1,1
+int     __ptrace:ptrace(int request, int pid, void* addr, void* data)  1
+int     __set_thread_area:set_thread_area(void*  user_desc)  -1,1,1
+int     __getpriority:getpriority(int, int)  1
+int     setpriority(int, int, int)   1
+int     setrlimit(int resource, const struct rlimit *rlp)  1
+int     getrlimit:ugetrlimit(int resource, struct rlimit *rlp)  1,1,-1
+int     getrlimit:getrlimit(int resource, struct rlimit *rlp)  -1,-1,1
+int     getrusage(int who, struct rusage*  r_usage)  1
+int     setgroups:setgroups32(int, const gid_t *)   1,1,-1
+int     setgroups:setgroups(int, const gid_t *)     -1,-1,1
+pid_t   getpgrp(void)  stub
+int     setpgid(pid_t, pid_t)  1
+pid_t   vfork(void)  1,-1,-1
+int     setregid:setregid32(gid_t, gid_t)  1,1,-1
+int     setregid:setregid(gid_t, gid_t)    -1,-1,1
+int     chroot(const char *)  1
+# IMPORTANT: Even though <sys/prctl.h> declares prctl(int,...), the syscall stub must take 6 arguments
+#            to match the kernel implementation.
+int     prctl(int option, unsigned int arg2, unsigned int arg3, unsigned int arg4, unsigned int arg5) 1
+int     capget(cap_user_header_t header, cap_user_data_t data) 1
+int     capset(cap_user_header_t header, const cap_user_data_t data) 1
+int     sigaltstack(const stack_t*, stack_t*) 1
+int     acct(const char*  filepath)  1
 
 # file descriptors
-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
+ssize_t     read (int, void*, size_t)        1
+ssize_t     write (int, const void*, size_t)       1
+ssize_t     pread64 (int, void *, size_t, off64_t) 1
+ssize_t     pwrite64 (int, void *, size_t, off64_t) 1
+int         __open:open (const char*, int, mode_t)  1
+int         __openat:openat (int, const char*, int, mode_t) 1
+int         close (int)                      1
+int         creat(const char*, mode_t)       stub
+off_t       lseek(int, off_t, int)           1
+int         __llseek:_llseek (int, unsigned long, unsigned long, loff_t*, int)  1
+pid_t       getpid ()    1
+void *      mmap(void *, size_t, int, int, int, long)  stub
+void *      __mmap2:mmap2(void*, size_t, int, int, int, long)   1
+int         munmap(void *, size_t)  1
+void *      mremap(void *, size_t, size_t, unsigned long)  1
+int         msync(const void *, size_t, int)    1
+int         mprotect(const void *, size_t, int)  1
+int         madvise(const void *, size_t, int)  1
+int         mlock(const void *addr, size_t len)    1
+int         munlock(const void *addr, size_t len)   1
+int         mlockall(int flags)   1
+int         munlockall()   1
+int         mincore(void*  start, size_t  length, unsigned char*  vec)   1
+int         __ioctl:ioctl(int, int, void *)  1
+int         readv(int, const struct iovec *, int)   1
+int         writev(int, const struct iovec *, int)  1
+int         __fcntl:fcntl(int, int, void*)  1
+int         flock(int, int)   1
+int         fchmod(int, mode_t)  1
+int         dup(int)  1
+int         pipe(int *)  1,1,-1
+int         pipe2(int *, int) 1
+int         dup2(int, int)   1
+int         select:_newselect(int, struct fd_set *, struct fd_set *, struct fd_set *, struct timeval *)  1
+int         ftruncate(int, off_t)  1
+int         ftruncate64(int, off64_t) 1
+int         getdents:getdents64(unsigned int, struct dirent *, unsigned int)   1
+int         fsync(int)  1
+int         fdatasync(int) 1
+int         fchown:fchown32(int, uid_t, gid_t)  1,1,-1
+int         fchown:fchown(int, uid_t, gid_t)    -1,-1,1
+void        sync(void)  1
+int         __fcntl64:fcntl64(int, int, void *)  1
+int         __fstatfs64:fstatfs64(int, size_t, struct statfs *)  1
+ssize_t     sendfile(int out_fd, int in_fd, off_t *offset, size_t count)  1
+int         fstatat:fstatat64(int dirfd, const char *path, struct stat *buf, int flags)   1
+int         mkdirat(int dirfd, const char *pathname, mode_t mode)  1
+int         fchownat(int dirfd, const char *path, uid_t owner, gid_t group, int flags)  1
+int         fchmodat(int dirfd, const char *path, mode_t mode, int flags)  1
+int         renameat(int olddirfd, const char *oldpath, int newdirfd, const char *newpath)  1
+int         fsetxattr(int, const char *, const void *, size_t, int) 1
+ssize_t     fgetxattr(int, const char *, void *, size_t) 1
+ssize_t     flistxattr(int, char *, size_t) 1
+int         fremovexattr(int, const char *) 1
 
 # file system
-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
+int     link (const char*, const char*)  1
+int     unlink (const char*)             1
+int     unlinkat (int, const char *, int)   1
+int     chdir (const char*)              1
+int     mknod (const char*, mode_t, dev_t)  1
+int     chmod (const char*,mode_t)          1
+int     chown:chown32(const char *, uid_t, gid_t)  1,1,-1
+int     chown:chown(const char *, uid_t, gid_t)    -1,-1,1
+int     lchown:lchown32 (const char*, uid_t, gid_t)  1,1,-1
+int     lchown:lchown (const char*, uid_t, gid_t)  -1,-1,1
+int     mount (const char*, const char*, const char*, unsigned long, const void*)  1
+int     umount(const char*)  stub
+int     umount2 (const char*, int)  1
+int     fstat:fstat64(int, struct stat*)    1
+int     stat:stat64(const char *, struct stat *)  1
+int     lstat:lstat64(const char *, struct stat *)  1
+int     mkdir(const char *, mode_t) 1
+int     readlink(const char *, char *, size_t)  1
+int     rmdir(const char *)  1
+int     rename(const char *, const char *)  1
+int     __getcwd:getcwd(char * buf, size_t size)  1
+int     access(const char *, int)  1
+int     faccessat(int, const char *, int, int)  1
+int     symlink(const char *, const char *)  1
+int     fchdir(int)    1
+int     truncate(const char*, off_t)    1
+int     setxattr(const char *, const char *, const void *, size_t, int) 1
+int     lsetxattr(const char *, const char *, const void *, size_t, int) 1
+ssize_t getxattr(const char *, const char *, void *, size_t) 1
+ssize_t lgetxattr(const char *, const char *, void *, size_t) 1
+ssize_t listxattr(const char *, char *, size_t) 1
+ssize_t llistxattr(const char *, char *, size_t) 1
+int     removexattr(const char *, const char *) 1
+int     lremovexattr(const char *, const char *) 1
+int     __statfs64:statfs64(const char *, size_t, struct statfs *)  1
+long    unshare(unsigned long)  1
 
 # time
-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 clk_id, const struct timespec* tp)  all
-int           clock_getres(clockid_t clk_id, struct timespec* res)   all
-int           clock_nanosleep(clockid_t clock_id, int flags, const struct timespec* req, struct timespec* rem)  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
+int           pause ()                       1
+int           gettimeofday(struct timeval*, struct timezone*)       1
+int           settimeofday(const struct timeval*, const struct timezone*)   1
+clock_t       times(struct tms *)       1
+int           nanosleep(const struct timespec *, struct timespec *)   1
+int           clock_gettime(clockid_t clk_id, struct timespec *tp)    1
+int           clock_settime(clockid_t clk_id, const struct timespec *tp)  1
+int           clock_getres(clockid_t clk_id, struct timespec *res)   1
+int           clock_nanosleep(clockid_t clock_id, int flags, const struct timespec *req, struct timespec *rem)  1
+int           getitimer(int, const struct itimerval *)   1
+int           setitimer(int, const struct itimerval *, struct itimerval *)  1
+int           __timer_create:timer_create(clockid_t clockid, struct sigevent *evp, timer_t *timerid)    1
+int           __timer_settime:timer_settime(timer_t, int, const struct itimerspec*, struct itimerspec*) 1
+int           __timer_gettime:timer_gettime(timer_t, struct itimerspec*)                                1
+int           __timer_getoverrun:timer_getoverrun(timer_t)                                              1
+int           __timer_delete:timer_delete(timer_t)                                                      1
+int           utimes(const char*, const struct timeval tvp[2])                          1
+int           utimensat(int, const char *, const struct timespec times[2], int)         1
 
 # signals
-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
+int     sigaction(int, const struct sigaction *, struct sigaction *)  1
+int     sigprocmask(int, const sigset_t *, sigset_t *)  1
+int     __sigsuspend:sigsuspend(int unused1, int unused2, unsigned mask)  1,1,-1
+int     __sigsuspend:sigsuspend(const sigset_t *mask)  -1,-1,1
+int     __rt_sigaction:rt_sigaction (int sig, const struct sigaction *act, struct sigaction *oact, size_t sigsetsize)  1
+int     __rt_sigprocmask:rt_sigprocmask (int  how, const sigset_t *set, sigset_t *oset, size_t sigsetsize)  1
+int     __rt_sigtimedwait:rt_sigtimedwait(const sigset_t *set, struct siginfo_t  *info, struct timespec_t  *timeout, size_t  sigset_size)  1
+int     sigpending(sigset_t *)  1
+int     signalfd4(int fd, const sigset_t *mask, size_t sizemask, int flags)  1
 
 # sockets
-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
+int           socket(int, int, int)              1,-1,1
+int           socketpair(int, int, int, int*)    1,-1,1
+int           bind(int, struct sockaddr *, int)  1,-1,1
+int           connect(int, struct sockaddr *, socklen_t)   1,-1,1
+int           listen(int, int)                   1,-1,1
+int           accept(int, struct sockaddr *, socklen_t *)  1,-1,1
+int           getsockname(int, struct sockaddr *, socklen_t *)  1,-1,1
+int           getpeername(int, struct sockaddr *, socklen_t *)  1,-1,1
+int           sendto(int, const void *, size_t, int, const struct sockaddr *, socklen_t)  1,-1,1
+int           recvfrom(int, void *, size_t, unsigned int, struct sockaddr *, socklen_t *)  1,-1,1
+int           shutdown(int, int)  1,-1,1
+int           setsockopt(int, int, int, const void *, socklen_t)  1,-1,1
+int           getsockopt(int, int, int, void *, socklen_t *)    1,-1,1
+int           sendmsg(int, const struct msghdr *, unsigned int)  1,-1,1
+int           recvmsg(int, struct msghdr *, unsigned int)   1,-1,1
 
 # sockets for x86. These are done as an "indexed" call to socketcall syscall.
-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
+int           socket:socketcall:1 (int, int, int) -1,1,-1
+int           bind:socketcall:2 (int, struct sockaddr *, int)  -1,1,-1
+int           connect:socketcall:3(int, struct sockaddr *, socklen_t)   -1,1,-1
+int           listen:socketcall:4(int, int)                   -1,1,-1
+int           accept:socketcall:5(int, struct sockaddr *, socklen_t *)  -1,1,-1
+int           getsockname:socketcall:6(int, struct sockaddr *, socklen_t *)  -1,1,-1
+int           getpeername:socketcall:7(int, struct sockaddr *, socklen_t *)  -1,1,-1
+int           socketpair:socketcall:8(int, int, int, int*)    -1,1,-1
+int           sendto:socketcall:11(int, const void *, size_t, int, const struct sockaddr *, socklen_t)  -1,1,-1
+int           recvfrom:socketcall:12(int, void *, size_t, unsigned int, struct sockaddr *, socklen_t *)  -1,1,-1
+int           shutdown:socketcall:13(int, int)  -1,1,-1
+int           setsockopt:socketcall:14(int, int, int, const void *, socklen_t)  -1,1,-1
+int           getsockopt:socketcall:15(int, int, int, void *, socklen_t *)    -1,1,-1
+int           sendmsg:socketcall:16(int, const struct msghdr *, unsigned int)  -1,1,-1
+int           recvmsg:socketcall:17(int, struct msghdr *, unsigned int)   -1,1,-1
 
 # scheduler & real-time
-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
+int sched_setscheduler(pid_t pid, int policy, const struct sched_param *param)  1
+int sched_getscheduler(pid_t pid)  1
+int sched_yield(void)  1
+int sched_setparam(pid_t pid, const struct sched_param *param)  1
+int sched_getparam(pid_t pid, struct sched_param *param)  1
+int sched_get_priority_max(int policy)  1
+int sched_get_priority_min(int policy)  1
+int sched_rr_get_interval(pid_t pid, struct timespec *interval)  1
+int sched_setaffinity(pid_t pid, size_t setsize, const cpu_set_t* set) 1
+int __sched_getaffinity:sched_getaffinity(pid_t pid, size_t setsize, cpu_set_t* set)  1
+int __getcpu:getcpu(unsigned *cpu, unsigned *node, void *unused) 1
+
+# io priorities
+int ioprio_set(int which, int who, int ioprio) 1
+int ioprio_get(int which, int who) 1
 
 # other
-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
+int     uname(struct utsname *)  1
+mode_t  umask(mode_t)  1
+int      __reboot:reboot(int, int, int, void *)  1
+int     __syslog:syslog(int, char *, int)  1
+int     init_module(void *, unsigned long, const char *)  1
+int     delete_module(const char*, unsigned int)   1
+int     klogctl:syslog(int, char *, int)   1
+int     sysinfo(struct sysinfo *)  1
+int     personality(unsigned long)  1
+long    perf_event_open(struct perf_event_attr *attr_uptr, pid_t pid, int cpu, int group_fd, unsigned long flags) 1
 
-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
+# futex
+int	futex(void *, int, int, void *, void *, int) 1
 
-int epoll_create1(int)  all
-int epoll_ctl(int, int op, int, struct epoll_event*)  all
-int __epoll_pwait:epoll_pwait(int, struct epoll_event*, int, int, const sigset_t*, size_t)  all
+# epoll
+int     epoll_create(int size)     1
+int     epoll_ctl(int epfd, int op, int fd, struct epoll_event *event)    1
+int     epoll_wait(int epfd, struct epoll_event *events, int max, int timeout)   1
 
-int eventfd:eventfd2(unsigned int, int)  all
+int     inotify_init(void)      1
+int     inotify_add_watch(int, const char *, unsigned int)  1
+int     inotify_rm_watch(int, unsigned int)  1
 
-void _exit|_Exit:exit_group(int)  all
-void __exit:exit(int)  all
+int     poll(struct pollfd *, unsigned int, long)  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     eventfd:eventfd2(unsigned int, int)  1
 
-int __pselect6:pselect6(int, fd_set*, fd_set*, fd_set*, timespec*, void*)  all
-int __ppoll:ppoll(pollfd*, unsigned int, timespec*, const sigset_t*, size_t)  all
-
-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
+# ARM-specific ARM_NR_BASE == 0x0f0000 == 983040
+int     __set_tls:__ARM_NR_set_tls(void*)                                 1,-1,-1
+int     cacheflush:__ARM_NR_cacheflush(long start, long end, long flags)  1,-1,-1
 
 # MIPS-specific
-int     _flush_cache:cacheflush(char* addr, const int nbytes, const int op) mips,mips64
-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
+int     _flush_cache:cacheflush(char *addr, const int nbytes, const int op) -1,-1,1
+int     syscall(int number,...) -1,-1,1

libc/arch-arm/arm.mk

@@ -1,93 +1,35 @@
-# arm specific configs
-
-# These are used by the 32-bit targets, but not the 64-bit ones.
-libc_common_src_files_arm := \
-    bionic/legacy_32_bit_support.cpp \
-    bionic/ndk_cruft.cpp \
-    bionic/time64.c \
-    upstream-openbsd/lib/libc/stdio/putw.c \
-
-# These are shared by all the 32-bit targets, but not the 64-bit ones.
-libc_bionic_src_files_arm := \
-    bionic/mmap.cpp
-
-libc_common_src_files_arm += \
-    bionic/memchr.c \
-    bionic/memrchr.c \
-    bionic/strchr.cpp \
-    bionic/strnlen.c \
-    bionic/strrchr.cpp \
-    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 \
-    upstream-openbsd/lib/libc/string/bcopy.c \
-    upstream-openbsd/lib/libc/string/stpcpy.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 \
-
-# The C++ fortify function implementations for which there is an
-# arm assembler version.
-#
-# Fortify implementations of libc functions.
-# libc_common_src_files_arm +=
-#    bionic/__memcpy_chk.cpp \
-#    bionic/__memset_chk.cpp \
-#    bionic/__strcpy_chk.cpp \
-#    bionic/__strcat_chk.cpp \
-
-libc_common_cflags_arm := -DSOFTFLOAT
-
-##########################################
-### CPU specific source files
-libc_bionic_src_files_arm += \
+_LIBC_ARCH_COMMON_SRC_FILES := \
     arch-arm/bionic/abort_arm.S \
     arch-arm/bionic/atomics_arm.c \
-    arch-arm/bionic/__bionic_clone.S \
+    arch-arm/bionic/clone.S \
+    arch-arm/bionic/eabi.c \
     arch-arm/bionic/_exit_with_stack_teardown.S \
+    arch-arm/bionic/ffs.S \
+    arch-arm/bionic/futex_arm.S \
+    arch-arm/bionic/__get_sp.S \
+    arch-arm/bionic/kill.S \
     arch-arm/bionic/libgcc_compat.c \
+    arch-arm/bionic/memcmp16.S \
     arch-arm/bionic/memcmp.S \
     arch-arm/bionic/_setjmp.S \
     arch-arm/bionic/setjmp.S \
     arch-arm/bionic/sigsetjmp.S \
+    arch-arm/bionic/strcpy.S \
+    arch-arm/bionic/strlen.c.arm \
     arch-arm/bionic/syscall.S \
+    arch-arm/bionic/tgkill.S \
+    arch-arm/bionic/tkill.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
+# These are used by the static and dynamic versions of the libc
+# respectively.
+_LIBC_ARCH_STATIC_SRC_FILES := \
+    arch-arm/bionic/exidx_static.c
 
-libc_netbsd_src_files_arm := \
-    upstream-netbsd/common/lib/libc/hash/sha1/sha1.c \
+_LIBC_ARCH_DYNAMIC_SRC_FILES := \
+    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_$(my_2nd_arch_prefix)CPU_VARIANT is not defined)
+ifeq ($(strip $(wildcard bionic/libc/arch-arm/$(TARGET_CPU_VARIANT)/$(TARGET_CPU_VARIANT).mk)),)
+$(error "TARGET_CPU_VARIANT not set or set to an unknown value. Possible values are cortex-a9, cortex-a15, krait. Use generic for devices that do not have a CPU similar to any of the supported cpu variants.")
 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 := \
-    $(LOCAL_PATH)/arch-common/bionic/crtbegin_so.c
+include bionic/libc/arch-arm/$(TARGET_CPU_VARIANT)/$(TARGET_CPU_VARIANT).mk

libc/arch-arm/bionic/__aeabi.c

@@ -1,110 +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.
- */
-
-#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*);
-
-/* 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
- */
-
-// Make this a weak symbol to avoid a multiple definition error when linking with libstdc++-v3.
-int __attribute__((weak))
-__aeabi_atexit(void *object, void (*destructor) (void *), void *dso_handle) {
-    return __cxa_atexit(destructor, object, dso_handle);
-}
-
-
-void __aeabi_memcpy8(void *dest, const void *src, size_t n) {
-    memcpy(dest, src, n);
-}
-
-void __aeabi_memcpy4(void *dest, const void *src, size_t n) {
-    memcpy(dest, src, n);
-}
-
-void __aeabi_memcpy(void *dest, const void *src, size_t n) {
-    memcpy(dest, src, n);
-}
-
-
-void __aeabi_memmove8(void *dest, const void *src, size_t n) {
-    memmove(dest, src, n);
-}
-
-void __aeabi_memmove4(void *dest, const void *src, size_t n) {
-    memmove(dest, src, n);
-}
-
-void __aeabi_memmove(void *dest, const void *src, size_t n) {
-    memmove(dest, src, n);
-}
-
-/*
- * __aeabi_memset has the order of its second and third arguments reversed.
- *  This allows __aeabi_memclr to tail-call __aeabi_memset
- */
-
-void __aeabi_memset8(void *dest, size_t n, int c) {
-    memset(dest, c, n);
-}
-
-void __aeabi_memset4(void *dest, size_t n, int c) {
-    memset(dest, c, n);
-}
-
-void __aeabi_memset(void *dest, size_t n, int c) {
-    memset(dest, c, n);
-}
-
-
-void __aeabi_memclr8(void *dest, size_t n) {
-    __aeabi_memset8(dest, n, 0);
-}
-
-void __aeabi_memclr4(void *dest, size_t n) {
-    __aeabi_memset4(dest, n, 0);
-}
-
-void __aeabi_memclr(void *dest, size_t n) {
-    __aeabi_memset(dest, n, 0);
-}

libc/arch-arm/bionic/__bionic_clone.S

@@ -1,69 +0,0 @@
-/*
- * Copyright (C) 2008-2010 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-
-#include <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_sp.S

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

libc/arch-arm/bionic/_exit_with_stack_teardown.S

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

libc/arch-arm/bionic/_setjmp.S

@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  */
 
-#include <private/bionic_asm.h>
+#include <machine/asm.h>
 #include <machine/setjmp.h>
 #include <machine/cpu-features.h>
 
@@ -107,7 +107,7 @@ ENTRY(_longjmp)
 
 	/* validation failed, die die die. */
 botch:
-	bl	PIC_SYM(longjmperror, PLT)
-	bl	PIC_SYM(abort, PLT)
+	bl	PIC_SYM(_C_LABEL(longjmperror), PLT)
+	bl	PIC_SYM(_C_LABEL(abort), PLT)
 	b	. - 8		/* Cannot get here */
 END(_longjmp)

libc/arch-arm/bionic/abort_arm.S

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

libc/arch-arm/bionic/atexit.h

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

libc/arch-arm/bionic/atexit_legacy.c

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

libc/arch-arm/bionic/clone.S

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

libc/arch-arm/bionic/crtbegin.c

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

libc/arch-arm/bionic/crtbegin_so.c

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

libc/arch-arm/bionic/crtend.S

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

libc/arch-arm/bionic/crtend_so.S

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

libc/arch-arm/bionic/eabi.c

@@ -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.
+ */
+#include <stddef.h>
+#include <string.h>
+
+extern int  __cxa_atexit(void (*)(void*), void*, void* );
+
+/* The "C++ ABI for ARM" document states that static C++ constructors,
+ * which are called from the .init_array, should manually call
+ * __aeabi_atexit() to register static destructors explicitely.
+ *
+ * Note that 'dso_handle' is the address of a magic linker-generate
+ * variable from the shared object that contains the constructor/destructor
+ */
+
+/* Make this a weak symbol to avoid a multiple definition error when linking
+ * with libstdc++-v3.  */
+int __attribute__((weak))
+__aeabi_atexit (void *object, void (*destructor) (void *), void *dso_handle)
+{
+    return __cxa_atexit(destructor, object, dso_handle);
+}
+
+
+void __aeabi_memcpy8(void *dest, const void *src, size_t n) {
+    memcpy(dest, src, n);
+}
+
+void __aeabi_memcpy4(void *dest, const void *src, size_t n) {
+    memcpy(dest, src, n);
+}
+
+void __aeabi_memcpy(void *dest, const void *src, size_t n) {
+    memcpy(dest, src, n);
+}
+
+
+void __aeabi_memmove8(void *dest, const void *src, size_t n) {
+    memmove(dest, src, n);
+}
+
+void __aeabi_memmove4(void *dest, const void *src, size_t n) {
+    memmove(dest, src, n);
+}
+
+void __aeabi_memmove(void *dest, const void *src, size_t n) {
+    memmove(dest, src, n);
+}
+
+/*
+ * __aeabi_memset has the order of its second and third arguments reversed. 
+ *  This allows __aeabi_memclr to tail-call __aeabi_memset
+ */
+ 
+void __aeabi_memset8(void *dest, size_t n, int c) {
+    memset(dest, c, n);
+}
+
+void __aeabi_memset4(void *dest, size_t n, int c) {
+    memset(dest, c, n);
+}
+
+void __aeabi_memset(void *dest, size_t n, int c) {
+    memset(dest, c, n);
+}
+
+
+void __aeabi_memclr8(void *dest, size_t n) {
+    __aeabi_memset8(dest, n, 0);
+}
+
+void __aeabi_memclr4(void *dest, size_t n) {
+    __aeabi_memset4(dest, n, 0);
+}
+
+void __aeabi_memclr(void *dest, size_t n) {
+    __aeabi_memset(dest, n, 0);
+}

libc/arch-arm/bionic/ffs.S

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

libc/arch-arm/bionic/futex_arm.S

@@ -0,0 +1,73 @@
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <asm/unistd.h>
+#include <machine/asm.h>
+
+#define FUTEX_WAIT 0
+#define FUTEX_WAKE 1
+
+// __futex_syscall3(*ftx, op, val)
+ENTRY(__futex_syscall3)
+    stmdb   sp!, {r4, r7}
+    .save   {r4, r7}
+    ldr     r7, =__NR_futex
+    swi     #0
+    ldmia   sp!, {r4, r7}
+    bx      lr
+END(__futex_syscall3)
+
+// __futex_syscall4(*ftx, op, val, *timespec)
+ENTRY(__futex_syscall4)
+    b __futex_syscall3
+END(__futex_syscall4)
+
+// __futex_wait(*ftx, val, *timespec)
+ENTRY(__futex_wait)
+    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
+END(__futex_wait)
+
+// __futex_wake(*ftx, counter)
+ENTRY(__futex_wake)
+    .save   {r4, r7}
+    stmdb   sp!, {r4, r7}
+    mov     r2, r1
+    mov     r1, #FUTEX_WAKE
+    ldr     r7, =__NR_futex
+    swi     #0
+    ldmia   sp!, {r4, r7}
+    bx      lr
+END(__futex_wake)

libc/arch-arm/bionic/kill.S

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

libc/arch-arm/bionic/libgcc_compat.c

@@ -1,159 +1,161 @@
-/* Generated by genlibgcc_compat.py */
+/*
+ * 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.
+ */
 
-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;
+/* This file contains dummy references to libgcc.a functions to force the
+ * dynamic linker to copy their definition into the final libc.so binary.
+ *
+ * They are required to ensure backwards binary compatibility with
+ * Android 1.5, 1.6 and even 3.0  system images. Some applications built
+ * using the NDK require them to be here.
+ *
+ * Now, for a more elaborate description of the issue:
+ *
+ * libgcc.a is a compiler-specific library containing various helper
+ * functions used to implement certain operations that are not necessarily
+ * supported by the target CPU. For example, integer division doesn't have a
+ * corresponding CPU instruction on ARMv5, and is instead implemented in the
+ * compiler-generated machine code as a call to an __idiv helper function.
+ *
+ * Normally, one has to place libgcc.a in the link command used to generate
+ * target binaries (shared libraries and executables) after all objects and
+ * static libraries, but before dependent shared libraries, i.e. something
+ * like:
+ *         gcc <options> -o libfoo.so  foo.a libgcc.a -lc -lm
+ *
+ * This ensures that any helper function needed by the code in foo.a is copied
+ * into the final libfoo.so. Unfortunately, the Android build system has been
+ * using this instead:
+ *
+ *         gcc <options> -o libfoo.so foo.a -lc -lm libgcc.a
+ *
+ * The problem with this is that if one helper function needed by foo.a has
+ * already been copied into libc.so or libm.so, then nothing will be copied
+ * into libfoo.so. Instead, a symbol import definition will be added to it
+ * so libfoo.so can directly call the one in libc.so at runtime.
+ *
+ * When changing toolchains for 2.0, the set of helper functions copied to
+ * libc.so changed, which resulted in some native shared libraries generated
+ * with the NDK to fail to load properly.
+ *
+ * The NDK has been fixed after 1.6_r1 to use the correct link command, so
+ * any native shared library generated with it should now be safe from that
+ * problem. On the other hand, existing shared libraries distributed with
+ * applications that were generated with a previous version of the NDK
+ * still need all 1.5/1.6 helper functions in libc.so and libm.so
+ *
+ * After 3.2, the toolchain was updated again, adding __aeabi_f2uiz to the
+ * list of requirements. Technically, this is due to mis-linked NDK libraries
+ * but it is easier to add a single function here than asking several app
+ * developers to fix their build.
+ *
+ * Final note: some of the functions below should really be in libm.so to
+ *             completely reflect the state of 1.5/1.6 system images. However,
+ *             since libm.so depends on libc.so, it's easier to put all of
+ *             these in libc.so instead, since the dynamic linker will always
+ *             search in libc.so before libm.so for dependencies.
+ */
 
-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,
-};
+#define   COMPAT_FUNCTIONS_LIST \
+    XX(__adddf3)             \
+    XX(__addsf3)             \
+    XX(__aeabi_cdcmpeq)      \
+    XX(__aeabi_cdcmple)      \
+    XX(__aeabi_cdrcmple)     \
+    XX(__aeabi_d2f)          \
+    XX(__aeabi_d2iz)         \
+    XX(__aeabi_dadd)         \
+    XX(__aeabi_dcmpeq)       \
+    XX(__aeabi_dcmpge)       \
+    XX(__aeabi_dcmpgt)       \
+    XX(__aeabi_dcmple)       \
+    XX(__aeabi_dcmplt)       \
+    XX(__aeabi_dcmpun)       \
+    XX(__aeabi_ddiv)         \
+    XX(__aeabi_dmul)         \
+    XX(__aeabi_drsub)        \
+    XX(__aeabi_dsub)         \
+    XX(__aeabi_f2d)          \
+    XX(__aeabi_f2iz)         \
+    XX(__aeabi_f2uiz)        \
+    XX(__aeabi_fadd)         \
+    XX(__aeabi_fcmpun)       \
+    XX(__aeabi_fdiv)         \
+    XX(__aeabi_fmul)         \
+    XX(__aeabi_frsub)        \
+    XX(__aeabi_fsub)         \
+    XX(__aeabi_i2d)          \
+    XX(__aeabi_i2f)          \
+    XX(__aeabi_idiv)         \
+    XX(__aeabi_l2d)          \
+    XX(__aeabi_l2f)          \
+    XX(__aeabi_lmul)         \
+    XX(__aeabi_llsl)         \
+    XX(__aeabi_llsr)         \
+    XX(__aeabi_ui2d)         \
+    XX(__aeabi_ui2f)         \
+    XX(__aeabi_ul2d)         \
+    XX(__aeabi_ul2f)         \
+    XX(__cmpdf2)             \
+    XX(__divdf3)             \
+    XX(__divsf3)             \
+    XX(__eqdf2)             \
+    XX(__extendsfdf2)        \
+    XX(__fixdfsi)            \
+    XX(__fixsfsi)            \
+    XX(__floatdidf)          \
+    XX(__floatdisf)          \
+    XX(__floatsidf)          \
+    XX(__floatsisf)          \
+    XX(__floatundidf)        \
+    XX(__floatundisf)        \
+    XX(__floatunsidf)        \
+    XX(__floatunsisf)        \
+    XX(__gedf2)              \
+    XX(__gtdf2)              \
+    XX(__ledf2)              \
+    XX(__ltdf2)              \
+    XX(__muldf3)             \
+    XX(__muldi3)             \
+    XX(__mulsf3)             \
+    XX(__nedf2)              \
+    XX(__subdf3)             \
+    XX(__subsf3)             \
+    XX(__truncdfsf2)         \
+    XX(__unorddf2)           \
+    XX(__unordsf2)           \
+
+#define  XX(f)    extern void f(void);
+COMPAT_FUNCTIONS_LIST
+#undef XX
+
+void  __bionic_libgcc_compat_hooks(void)
+{
+#define XX(f)    f();
+COMPAT_FUNCTIONS_LIST
+#undef XX
+}

libc/arch-arm/bionic/memcmp.S

@@ -27,7 +27,7 @@
  */
 
 #include <machine/cpu-features.h>
-#include <private/bionic_asm.h>
+#include <machine/asm.h>
 
 
 #ifdef HAVE_32_BYTE_CACHE_LINE
@@ -107,11 +107,9 @@ ENTRY(memcmp)
         bmi        10f
 #endif
 
+        .save {r4, lr}
         /* 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

libc/arch-arm/bionic/memcmp16.S

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

libc/arch-arm/bionic/memcpy.S

@@ -27,7 +27,7 @@
  */
 
 #include <machine/cpu-features.h>
-#include <private/bionic_asm.h>
+#include <machine/asm.h>
 
 #if defined(__ARM_NEON__) && !defined(ARCH_ARM_USE_NON_NEON_MEMCPY)
 
@@ -352,9 +352,9 @@ ENTRY(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
@@ -442,7 +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
@@ -563,7 +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
@@ -590,7 +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
@@ -617,7 +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

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

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

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

@@ -1,614 +0,0 @@
-/* Copyright (c) 2013, Linaro Limited
-   All rights reserved.
-
-   Redistribution and use in source and binary forms, with or without
-   modification, are permitted provided that the following conditions
-   are met:
-
-      * Redistributions of source code must retain the above copyright
-      notice, this list of conditions and the following disclaimer.
-
-      * Redistributions in binary form must reproduce the above copyright
-      notice, this list of conditions and the following disclaimer in the
-      documentation and/or other materials provided with the distribution.
-
-      * Neither the name of Linaro Limited nor the names of its
-      contributors may be used to endorse or promote products derived
-      from this software without specific prior written permission.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-   HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
- */
-
-/*
-   This memcpy routine is optimised for Cortex-A15 cores and takes advantage
-   of VFP or NEON when built with the appropriate flags.
-
-   Assumptions:
-
-    ARMv6 (ARMv7-a if using Neon)
-    ARM state
-    Unaligned accesses
-    LDRD/STRD support unaligned word accesses
-
- */
-
-#include <machine/cpu-features.h>
-#include <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

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

libc/arch-arm/bionic/setjmp.S

@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  */
 
-#include <private/bionic_asm.h>
+#include <machine/asm.h>
 #include <machine/setjmp.h>
 #include <machine/cpu-features.h>
 
@@ -51,16 +51,12 @@
 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(sigblock, PLT)
+	bl	PIC_SYM(_C_LABEL(sigblock), PLT)
 	mov	r1, r0
 
 	ldmfd	sp!, {r0, r14}
-	.cfi_def_cfa_offset 0
 
 	/* Store signal mask */
 	str	r1, [r0, #(_JB_SIGMASK * 4)]
@@ -100,20 +96,13 @@ ENTRY(longjmp)
 
 	/* 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(sigsetmask, PLT)
+	bl	PIC_SYM(_C_LABEL(sigsetmask), PLT)
 
 	add	sp, sp, #4	/* unalign the stack */
-	.cfi_adjust_cfa_offset -4
-	ldmfd	sp!, {r0, r1, r14}
-	.cfi_def_cfa_offset 0
+	ldmfd	sp!, {r0, r1, r14} 
 
 #ifdef __ARM_HAVE_VFP
 	/* Restore floating-point registers */
@@ -147,7 +136,7 @@ ENTRY(longjmp)
 
 	/* validation failed, die die die. */
 botch:
-	bl	PIC_SYM(longjmperror, PLT)
-	bl	PIC_SYM(abort, PLT)
+	bl	PIC_SYM(_C_LABEL(longjmperror), PLT)
+	bl	PIC_SYM(_C_LABEL(abort), PLT)
 	b	. - 8		/* Cannot get here */
 END(longjmp)

libc/arch-arm/bionic/sigsetjmp.S

@@ -35,7 +35,7 @@
 
 #define _ALIGN_TEXT .align 0
 
-#include <private/bionic_asm.h>
+#include <machine/asm.h>
 #include <machine/setjmp.h>
 
 /*
@@ -50,8 +50,8 @@
 
 ENTRY(sigsetjmp)
 	teq	r1, #0
-	beq	PIC_SYM(_setjmp, PLT)
-	b	PIC_SYM(setjmp, PLT)
+	beq	PIC_SYM(_C_LABEL(_setjmp), PLT)
+	b	PIC_SYM(_C_LABEL(setjmp), PLT)
 END(sigsetjmp)
 
 .L_setjmp_magic:
@@ -61,6 +61,6 @@ ENTRY(siglongjmp)
 	ldr	r2, .L_setjmp_magic
 	ldr	r3, [r0]
 	teq	r2, r3
-	beq	PIC_SYM(_longjmp, PLT)
-	b	PIC_SYM(longjmp, PLT)
+	beq	PIC_SYM(_C_LABEL(_longjmp), PLT)
+	b	PIC_SYM(_C_LABEL(longjmp), PLT)
 END(siglongjmp)

libc/arch-arm/bionic/strcmp.S

@@ -28,7 +28,7 @@
  */
 
 #include <machine/cpu-features.h>
-#include <private/bionic_asm.h>
+#include <machine/asm.h>
 
 	.text
 
@@ -52,8 +52,8 @@
 #define magic2(REG) REG, lsl #7
 
 ENTRY(strcmp)
-	pld	[r0, #0]
-	pld	[r1, #0]
+	PLD(r0, #0)
+	PLD(r1, #0)
 	eor	r2, r0, r1
 	tst	r2, #3
 
@@ -88,8 +88,8 @@ ENTRY(strcmp)
 	orr	r4, r4, r4, lsl #16
 	.p2align	2
 4:
-	pld	[r0, #8]
-	pld	[r1, #8]
+	PLD(r0, #8)
+	PLD(r1, #8)
 	sub	r2, ip, magic1(r4)
 	cmp	ip, r3
 	itttt	eq

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

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

libc/arch-arm/bionic/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 <machine/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/bionic/strlen.c

@@ -0,0 +1,129 @@
+/*
+ * 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

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

libc/arch-arm/bionic/tgkill.S

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

libc/arch-arm/bionic/tkill.S

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

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

@@ -1,221 +0,0 @@
-/*
- * 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}
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-    push    {r4, r5}
-    .save   {r4, r5}
-    .cfi_adjust_cfa_offset 8
-    .cfi_rel_offset r4, 0
-    .cfi_rel_offset r5, 4
-
-    mov     lr, r2
-
-    // Save the dst register to r5
-    mov     r5, r0
-
-    // Zero out r4
-    eor     r4, r4, r4
-
-    // r1 contains the address of the string to count.
-.L_strlen_start:
-    mov     r0, r1
-    ands    r3, r1, #7
-    beq     .L_mainloop
-
-    // Align to a double word (64 bits).
-    rsb     r3, r3, #8
-    lsls    ip, r3, #31
-    beq     .L_align_to_32
-
-    ldrb    r2, [r1], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_32:
-    bcc     .L_align_to_64
-    ands    ip, r3, #2
-    beq     .L_align_to_64
-
-    ldrb    r2, [r1], #1
-    cbz     r2, .L_update_count_and_finish
-    ldrb    r2, [r1], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_64:
-    tst     r3, #4
-    beq     .L_mainloop
-    ldr     r3, [r1], #4
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-
-    .p2align 2
-.L_mainloop:
-    ldrd    r2, r3, [r1], #8
-
-    pld     [r1, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-    b       .L_mainloop
-
-.L_update_count_and_finish:
-    sub     r3, r1, r0
-    sub     r3, r3, #1
-    b       .L_finish
-
-.L_zero_in_first_register:
-    sub     r3, r1, r0
-    lsls    r2, ip, #17
-    bne     .L_sub8_and_finish
-    bcs     .L_sub7_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub6_and_finish
-
-    sub     r3, r3, #5
-    b       .L_finish
-
-.L_sub8_and_finish:
-    sub     r3, r3, #8
-    b       .L_finish
-
-.L_sub7_and_finish:
-    sub     r3, r3, #7
-    b       .L_finish
-
-.L_sub6_and_finish:
-    sub     r3, r3, #6
-    b       .L_finish
-
-.L_zero_in_second_register:
-    sub     r3, r1, r0
-    lsls    r2, ip, #17
-    bne     .L_sub4_and_finish
-    bcs     .L_sub3_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub2_and_finish
-
-    sub     r3, r3, #1
-    b       .L_finish
-
-.L_sub4_and_finish:
-    sub     r3, r3, #4
-    b       .L_finish
-
-.L_sub3_and_finish:
-    sub     r3, r3, #3
-    b       .L_finish
-
-.L_sub2_and_finish:
-    sub     r3, r3, #2
-
-.L_finish:
-    cmp     r4, #0
-    bne     .L_strlen_done
-
-    // Time to get the dst string length.
-    mov     r1, r5
-
-    // Save the original source address to r5.
-    mov     r5, r0
-
-    // Save the current length (adding 1 for the terminator).
-    add     r4, r3, #1
-    b       .L_strlen_start
-
-    // r0 holds the pointer to the dst string.
-    // r3 holds the dst string length.
-    // r4 holds the src string length + 1.
-.L_strlen_done:
-    add     r2, r3, r4
-    cmp     r2, lr
-    bhi     __strcat_chk_failed
-
-    // Set up the registers for the memcpy code.
-    mov     r1, r5
-    pld     [r1, #64]
-    mov     r2, r4
-    add     r0, r0, r3
-    pop     {r4, r5}
-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)
-    .save   {r0, lr}
-    .save   {r4, r5}
-
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-    .cfi_adjust_cfa_offset 8
-    .cfi_rel_offset r4, 0
-    .cfi_rel_offset r5, 4
-
-    ldr     r0, error_message
-    ldr     r1, error_code
-1:
-    add     r0, pc
-    bl      __fortify_chk_fail
-error_code:
-    .word   BIONIC_EVENT_STRCAT_BUFFER_OVERFLOW
-error_message:
-    .word   error_string-(1b+4)
-END(__strcat_chk_failed)
-
-    .data
-error_string:
-    .string "strcat: prevented write past end of buffer"

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

@@ -1,182 +0,0 @@
-/*
- * 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}
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-
-    mov     lr, r2
-    mov     r0, r1
-
-    ands    r3, r1, #7
-    beq     .L_mainloop
-
-    // Align to a double word (64 bits).
-    rsb     r3, r3, #8
-    lsls    ip, r3, #31
-    beq     .L_align_to_32
-
-    ldrb    r2, [r0], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_32:
-    bcc     .L_align_to_64
-    ands    ip, r3, #2
-    beq     .L_align_to_64
-
-    ldrb    r2, [r0], #1
-    cbz     r2, .L_update_count_and_finish
-    ldrb    r2, [r0], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_64:
-    tst     r3, #4
-    beq     .L_mainloop
-    ldr     r3, [r0], #4
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-
-    .p2align 2
-.L_mainloop:
-    ldrd    r2, r3, [r0], #8
-
-    pld     [r0, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-    b       .L_mainloop
-
-.L_update_count_and_finish:
-    sub     r3, r0, r1
-    sub     r3, r3, #1
-    b       .L_check_size
-
-.L_zero_in_first_register:
-    sub     r3, r0, r1
-    lsls    r2, ip, #17
-    bne     .L_sub8_and_finish
-    bcs     .L_sub7_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub6_and_finish
-
-    sub     r3, r3, #5
-    b       .L_check_size
-
-.L_sub8_and_finish:
-    sub     r3, r3, #8
-    b       .L_check_size
-
-.L_sub7_and_finish:
-    sub     r3, r3, #7
-    b       .L_check_size
-
-.L_sub6_and_finish:
-    sub     r3, r3, #6
-    b       .L_check_size
-
-.L_zero_in_second_register:
-    sub     r3, r0, r1
-    lsls    r2, ip, #17
-    bne     .L_sub4_and_finish
-    bcs     .L_sub3_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub2_and_finish
-
-    sub     r3, r3, #1
-    b       .L_check_size
-
-.L_sub4_and_finish:
-    sub     r3, r3, #4
-    b       .L_check_size
-
-.L_sub3_and_finish:
-    sub     r3, r3, #3
-    b       .L_check_size
-
-.L_sub2_and_finish:
-    sub     r3, r3, #2
-
-.L_check_size:
-    pld     [r1, #0]
-    pld     [r1, #64]
-    ldr     r0, [sp]
-    cmp     r3, lr
-    bhs     __strcpy_chk_failed
-
-    // Add 1 for copy length to get the string terminator.
-    add     r2, r3, #1
-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)
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-
-    ldr     r0, error_message
-    ldr     r1, error_code
-1:
-    add     r0, pc
-    bl      __fortify_chk_fail
-error_code:
-    .word   BIONIC_EVENT_STRCPY_BUFFER_OVERFLOW
-error_message:
-    .word   error_string-(1b+4)
-END(__strcpy_chk_failed)
-
-    .data
-error_string:
-    .string "strcpy: prevented write past end of buffer"

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

@@ -25,81 +25,122 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
+
+/* Assumes neon instructions and a cache line size of 64 bytes. */
+
+#include <machine/cpu-features.h>
+#include <machine/asm.h>
+
 /*
- * 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.
+ * This code assumes it is running on a processor that supports all arm v7
+ * instructions, that supports neon instructions, and that has a 64 byte
+ * cache line.
  */
 
-// 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)
+#define CACHE_LINE_SIZE     64
 
 ENTRY(memcpy)
-        pld     [r1, #64]
-        push    {r0, lr}
-        .save   {r0, lr}
-        .cfi_def_cfa_offset 8
-        .cfi_rel_offset r0, 0
-        .cfi_rel_offset lr, 4
+        .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)]
+
+        /* do we have at least 16-bytes to copy (needed for alignment below) */
+        cmp         r2, #16
+        blo         5f
+
+        /* align destination to cache-line for the write-buffer */
+        rsb         r3, r0, #0
+        ands        r3, r3, #0xF
+        beq         0f
+
+        /* 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:
+
+0:      /* preload immediately the next cache line, which we may need */
+        pld         [r1, #(CACHE_LINE_SIZE*0)]
+        pld         [r1, #(CACHE_LINE_SIZE*1)]
+
+        /* make sure we have at least 64 bytes to copy */
+        subs        r2, r2, #64
+        blo         2f
+
+        /* Preload all the cache lines we need.
+         * NOTE: The number of pld below depends on CACHE_LINE_SIZE,
+         * ideally we 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.
+         * NOTE: The value CACHE_LINE_SIZE * 4 was chosen through
+         * experimentation.
+         */
+        pld         [r1, #(CACHE_LINE_SIZE*2)]
+        pld         [r1, #(CACHE_LINE_SIZE*3)]
+        pld         [r1, #(CACHE_LINE_SIZE*4)]
+
+1:      /* The main loop copies 64 bytes at a time */
+        vld1.8      {d0  - d3},   [r1]!
+        vld1.8      {d4  - d7},   [r1]!
+        pld         [r1, #(CACHE_LINE_SIZE*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 */
+        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
+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
 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}
-        .save   {lr}
-        .cfi_def_cfa_offset 4
-        .cfi_rel_offset lr, 0
-
-        ldr     r0, error_message
-        ldr     r1, error_code
-1:
-        add     r0, pc
-        bl      __fortify_chk_fail
-error_code:
-        .word   BIONIC_EVENT_MEMCPY_BUFFER_OVERFLOW
-error_message:
-        .word   error_string-(1b+8)
-END(__memcpy_chk_fail)
-
-        .data
-error_string:
-        .string "memcpy: prevented write past end of buffer"

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

@@ -1,323 +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.
- */
-/*
- * 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)
-        .save   {r0, lr}
-        .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)
-        .save   {r0, lr}
-        .cfi_def_cfa_offset 8
-        .cfi_rel_offset r0, 0
-        .cfi_rel_offset lr, 4
-
-        // To try and improve performance, stack layout changed,
-        // i.e., not keeping the stack looking like users expect
-        // (highest numbered register at highest address).
-        strd    r4, r5, [sp, #-8]!
-        .save   {r4, r5}
-        .cfi_adjust_cfa_offset 8
-        .cfi_rel_offset r4, 0
-        .cfi_rel_offset r5, 4
-        strd    r6, r7, [sp, #-8]!
-        .save   {r6, r7}
-        .cfi_adjust_cfa_offset 8
-        .cfi_rel_offset r6, 0
-        .cfi_rel_offset r7, 0
-        strd    r8, r9, [sp, #-8]!
-        .save   {r8, r9}
-        .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

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2013 The Android Open Source Project
+ * Copyright (C) 2008 The Android Open Source Project
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -27,167 +27,80 @@
  */
 
 #include <machine/cpu-features.h>
-#include <private/bionic_asm.h>
-#include <private/libc_events.h>
+#include <machine/asm.h>
 
-        /*
-         * Optimized memset() for ARM.
+		/*
+		 * 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.
-        .save       {lr}
-        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)
+    .fpu    neon
 
 ENTRY(bzero)
-        mov         r2, r1
-        mov         r1, #0
-.L_done:
+        mov     r2, r1
+        mov     r1, #0
         // Fall through to memset...
 END(bzero)
 
 ENTRY(memset)
         .save       {r0}
         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.
+        vdup.8      q0, r1
+
+        /* do we have at least 16-bytes to write (needed for alignment below) */
         cmp         r2, #16
-        blo         .L_set_less_than_16_unknown_align
+        blo         3f
 
-        // 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
+        /* align destination to 16 bytes for the write-buffer */
+        rsb         r3, r0, #0
+        ands        r3, r3, #0xF
+        beq         2f
 
-        // 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
+        /* write up to 15-bytes (count in r3) */
+        sub         r2, r2, r3
+        movs        ip, r3, lsl #31
+        strmib      r1, [r0], #1
+        strcsb      r1, [r0], #1
+        strcsb      r1, [r0], #1
+        movs        ip, r3, lsl #29
+        bge         1f
 
-.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
+        // writes 4 bytes, 32-bits aligned
+        vst1.32     {d0[0]}, [r0, :32]!
+1:      bcc         2f
 
-.L_double_word_aligned:
-        mov         r0, r1
+        // writes 8 bytes, 64-bits aligned
+        vst1.8      {d0}, [r0, :64]!
+2:
+        /* make sure we have at least 32 bytes to write */
+        subs        r2, r2, #32
+        blo         2f
+        vmov        q1, q0
 
-        subs        r2, #64
-        blo         .L_set_less_than_64
+1:      /* The main loop writes 32 bytes at a time */
+        subs        r2, r2, #32
+        vst1.8      {d0 - d3}, [r0, :128]!
+        bhs         1b
 
-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
+2:      /* less than 32 left */
+        add         r2, r2, #32
+        tst         r2, #0x10
+        beq         3f
 
-        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
+        // writes 16 bytes, 128-bits aligned
+        vst1.8      {d0, d1}, [r0, :128]!
+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
-        it          mi
-        strbmi      r1, [r0], #1
-        itt         cs
-        strbcs      r1, [r0], #1
-        strbcs      r1, [r0], #1
+        strmib      r1, [r0], #1
+        strcsb      r1, [r0], #1
+        strcsb      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/strcat.S

@@ -1,568 +0,0 @@
-/*
- * 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

@@ -27,7 +27,7 @@
  */
 
 #include <machine/cpu-features.h>
-#include <private/bionic_asm.h>
+#include <machine/asm.h>
 
 #ifdef __ARMEB__
 #define S2LOMEM lsl
@@ -123,13 +123,8 @@ ENTRY(strcmp)
         .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 */
@@ -145,8 +140,19 @@ ENTRY(strcmp)
 
         .macro  magic_find_zero_bytes w1
         /* Macro to find all-zero bytes in w1, result is in ip.  */
+#if (defined (__ARM_FEATURE_DSP))
         uadd8   ip, \w1, r6
         sel     ip, r7, r6
+#else /* not defined (__ARM_FEATURE_DSP) */
+        /* __ARM_FEATURE_DSP is not defined for some Cortex-M processors.
+        Coincidently, these processors only have Thumb-2 mode, where we can use the
+        the (large) magic constant available directly as an immediate in instructions.
+        Note that we cannot use the magic constant in ARM mode, where we need
+        to create the constant in a register.  */
+        sub     ip, \w1, #0x01010101
+        bic     ip, ip, \w1
+        and     ip, ip, #0x80808080
+#endif /* not defined (__ARM_FEATURE_DSP) */
         .endm /* magic_find_zero_bytes */
 
         .macro  setup_return w1 w2
@@ -165,13 +171,12 @@ ENTRY(strcmp)
         /* Are both strings double-word aligned?  */
         orr     ip, r0, r1
         tst     ip, #7
-        bne     .L_do_align
+        bne     do_align
 
         /* Fast path.  */
-        .save   {r4-r7}
         init
 
-.L_doubleword_aligned:
+doubleword_aligned:
 
         /* Get here when the strings to compare are double-word aligned.  */
         /* Compare two words in every iteration.  */
@@ -184,14 +189,14 @@ ENTRY(strcmp)
         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
+        magic_compare_and_branch w1=r2, w2=r4, label=return_24
+        magic_compare_and_branch w1=r3, w2=r5, label=return_35
         b       2b
 
-.L_do_align:
+do_align:
         /* Is the first string word-aligned?  */
         ands    ip, r0, #3
-        beq     .L_word_aligned_r0
+        beq     word_aligned_r0
 
         /* Fast compare byte by byte until the first string is word-aligned.  */
         /* The offset of r0 from a word boundary is in ip. Thus, the number of bytes
@@ -199,58 +204,58 @@ ENTRY(strcmp)
         bic     r0, r0, #3
         ldr     r2, [r0], #4
         lsls    ip, ip, #31
-        beq     .L_byte2
-        bcs     .L_byte3
+        beq     byte2
+        bcs     byte3
 
-.L_byte1:
+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
+        bne     fast_return
+        m_cbz   reg=r3, label=fast_return
 
-.L_byte2:
+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
+        bne     fast_return
+        m_cbz   reg=r3, label=fast_return
 
-.L_byte3:
+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
+        bne     fast_return
+        m_cbnz  reg=r3, label=word_aligned_r0
 
-.L_fast_return:
+fast_return:
         mov     r0, ip
         bx      lr
 
-.L_word_aligned_r0:
+word_aligned_r0:
         init
         /* The first string is word-aligned.  */
         /* Is the second string word-aligned?  */
         ands    ip, r1, #3
-        bne     .L_strcmp_unaligned
+        bne     strcmp_unaligned
 
-.L_word_aligned:
+word_aligned:
         /* The strings are word-aligned. */
         /* Is the first string double-word aligned?  */
         tst     r0, #4
-        beq     .L_doubleword_aligned_r0
+        beq     doubleword_aligned_r0
 
         /* If r0 is not double-word aligned yet, align it by loading
         and comparing the next word from each string.  */
         ldr     r2, [r0], #4
         ldr     r4, [r1], #4
-        magic_compare_and_branch w1=r2 w2=r4 label=.L_return_24
+        magic_compare_and_branch w1=r2 w2=r4 label=return_24
 
-.L_doubleword_aligned_r0:
+doubleword_aligned_r0:
         /* Get here when r0 is double-word aligned.  */
         /* Is r1 doubleword_aligned?  */
         tst     r1, #4
-        beq     .L_doubleword_aligned
+        beq     doubleword_aligned
 
         /* Get here when the strings to compare are word-aligned,
         r0 is double-word aligned, but r1 is not double-word aligned.  */
@@ -266,9 +271,9 @@ ENTRY(strcmp)
 
         /* 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
+        magic_compare_and_branch w1=r2 w2=r5 label=return_25
         ldrd    r4, r5, [r1], #8
-        magic_compare_and_branch w1=r3 w2=r4 label=.L_return_34
+        magic_compare_and_branch w1=r3 w2=r4 label=return_34
         b       3b
 
         .macro miscmp_word offsetlo offsethi
@@ -292,47 +297,47 @@ ENTRY(strcmp)
         and     r2, r3, r6, S2LOMEM #\offsetlo
         it      eq
         cmpeq   r2, r5
-        bne     .L_return_25
+        bne     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
+        bne     return_32
         b       7b
         .endm /* miscmp_word */
 
-.L_strcmp_unaligned:
+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.  */
+        beq     miscmp_word_16                    /* If ip == 2.  */
+        bge     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
+miscmp_word_16:  miscmp_word offsetlo=16 offsethi=16
+miscmp_word_24:  miscmp_word offsetlo=24 offsethi=8
 
 
-.L_return_32:
+return_32:
         setup_return w1=r3, w2=r2
-        b       .L_do_return
-.L_return_34:
+        b       do_return
+return_34:
         setup_return w1=r3, w2=r4
-        b       .L_do_return
-.L_return_25:
+        b       do_return
+return_25:
         setup_return w1=r2, w2=r5
-        b       .L_do_return
-.L_return_35:
+        b       do_return
+return_35:
         setup_return w1=r3, w2=r5
-        b       .L_do_return
-.L_return_24:
+        b       do_return
+return_24:
         setup_return w1=r2, w2=r4
 
-.L_do_return:
+do_return:
 
 #ifdef __ARMEB__
         mov     r0, ip
@@ -344,16 +349,11 @@ ENTRY(strcmp)
         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
+        m_cbz   reg=r0, label=compute_return_value
         clz     r0, r0
         /* r0 contains the number of bits on the left of the first all-zero byte in r1.  */
         rsb     r0, r0, #24
@@ -361,7 +361,7 @@ ENTRY(strcmp)
         lsr     r1, r1, r0
         lsr     r2, r2, r0
 
-.L_compute_return_value:
+compute_return_value:
         movs    r0, #1
         cmp     r1, r2
         /* The return value is computed as follows.

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

@@ -1,451 +0,0 @@
-/*
- * 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_copy_byte reg, cmd, label
-    ldrb    \reg, [r1], #1
-    strb    \reg, [r0], #1
-    \cmd    \reg, \label
-    .endm // m_copy_byte
-
-ENTRY(strcpy)
-    // For short copies, hard-code checking the first 8 bytes since this
-    // new code doesn't win until after about 8 bytes.
-    m_push
-    m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r5, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r5, cmd=cbnz, label=strcpy_continue
-
-strcpy_finish:
-    m_pop
-
-strcpy_continue:
-    pld     [r1, #0]
-    ands    r3, r0, #7
-    beq     strcpy_check_src_align
-
-    // Align to a double word (64 bits).
-    rsb     r3, r3, #8
-    lsls    ip, r3, #31
-    beq     strcpy_align_to_32
-
-    ldrb    r2, [r1], #1
-    strb    r2, [r0], #1
-    cbz     r2, strcpy_complete
-
-strcpy_align_to_32:
-    bcc     strcpy_align_to_64
-
-    ldrb    r2, [r1], #1
-    strb    r2, [r0], #1
-    cbz     r2, strcpy_complete
-    ldrb    r2, [r1], #1
-    strb    r2, [r0], #1
-    cbz     r2, strcpy_complete
-
-strcpy_align_to_64:
-    tst     r3, #4
-    beq     strcpy_check_src_align
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-    str     r2, [r0], #4
-
-strcpy_check_src_align:
-    // At this point dst is aligned to a double word, check if src
-    // is also aligned to a double word.
-    ands    r3, r1, #7
-    bne     strcpy_unaligned_copy
-
-    .p2align 2
-strcpy_mainloop:
-    ldrd    r2, r3, [r1], #8
-
-    pld     [r1, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    strd    r2, r3, [r0], #8
-    b       strcpy_mainloop
-
-strcpy_complete:
-    m_pop
-
-strcpy_zero_in_first_register:
-    lsls    lr, ip, #17
-    bne     strcpy_copy1byte
-    bcs     strcpy_copy2bytes
-    lsls    ip, ip, #1
-    bne     strcpy_copy3bytes
-
-strcpy_copy4bytes:
-    // Copy 4 bytes to the destiniation.
-    str     r2, [r0]
-    m_pop
-
-strcpy_copy1byte:
-    strb    r2, [r0]
-    m_pop
-
-strcpy_copy2bytes:
-    strh    r2, [r0]
-    m_pop
-
-strcpy_copy3bytes:
-    strh    r2, [r0], #2
-    lsr     r2, #16
-    strb    r2, [r0]
-    m_pop
-
-strcpy_zero_in_second_register:
-    lsls    lr, ip, #17
-    bne     strcpy_copy5bytes
-    bcs     strcpy_copy6bytes
-    lsls    ip, ip, #1
-    bne     strcpy_copy7bytes
-
-    // Copy 8 bytes to the destination.
-    strd    r2, r3, [r0]
-    m_pop
-
-strcpy_copy5bytes:
-    str     r2, [r0], #4
-    strb    r3, [r0]
-    m_pop
-
-strcpy_copy6bytes:
-    str     r2, [r0], #4
-    strh    r3, [r0]
-    m_pop
-
-strcpy_copy7bytes:
-    str     r2, [r0], #4
-    strh    r3, [r0], #2
-    lsr     r3, #16
-    strb    r3, [r0]
-    m_pop
-
-strcpy_unaligned_copy:
-    // Dst is aligned to a double word, while src is at an unknown alignment.
-    // There are 7 different versions of the unaligned copy code
-    // to prevent overreading the src. The mainloop of every single version
-    // will store 64 bits per loop. The difference is how much of src can
-    // be read without potentially crossing a page boundary.
-    tbb     [pc, r3]
-strcpy_unaligned_branchtable:
-    .byte 0
-    .byte ((strcpy_unalign7 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign6 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign5 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign4 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign3 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign2 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign1 - strcpy_unaligned_branchtable)/2)
-
-    .p2align 2
-    // Can read 7 bytes before possibly crossing a page.
-strcpy_unalign7:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldrb    r3, [r1]
-    cbz     r3, strcpy_unalign7_copy5bytes
-    ldrb    r4, [r1, #1]
-    cbz     r4, strcpy_unalign7_copy6bytes
-    ldrb    r5, [r1, #2]
-    cbz     r5, strcpy_unalign7_copy7bytes
-
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    lsrs    ip, r3, #24
-    strd    r2, r3, [r0], #8
-    beq     strcpy_unalign_return
-    b       strcpy_unalign7
-
-strcpy_unalign7_copy5bytes:
-    str     r2, [r0], #4
-    strb    r3, [r0]
-strcpy_unalign_return:
-    m_pop
-
-strcpy_unalign7_copy6bytes:
-    str     r2, [r0], #4
-    strb    r3, [r0], #1
-    strb    r4, [r0], #1
-    m_pop
-
-strcpy_unalign7_copy7bytes:
-    str     r2, [r0], #4
-    strb    r3, [r0], #1
-    strb    r4, [r0], #1
-    strb    r5, [r0], #1
-    m_pop
-
-    .p2align 2
-    // Can read 6 bytes before possibly crossing a page.
-strcpy_unalign6:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldrb    r4, [r1]
-    cbz     r4, strcpy_unalign_copy5bytes
-    ldrb    r5, [r1, #1]
-    cbz     r5, strcpy_unalign_copy6bytes
-
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    tst     r3, #0xff0000
-    beq     strcpy_copy7bytes
-    lsrs    ip, r3, #24
-    strd    r2, r3, [r0], #8
-    beq     strcpy_unalign_return
-    b       strcpy_unalign6
-
-    .p2align 2
-    // Can read 5 bytes before possibly crossing a page.
-strcpy_unalign5:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldrb    r4, [r1]
-    cbz     r4, strcpy_unalign_copy5bytes
-
-    ldr     r3, [r1], #4
-
-    pld     [r1, #64]
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    strd    r2, r3, [r0], #8
-    b       strcpy_unalign5
-
-strcpy_unalign_copy5bytes:
-    str     r2, [r0], #4
-    strb    r4, [r0]
-    m_pop
-
-strcpy_unalign_copy6bytes:
-    str     r2, [r0], #4
-    strb    r4, [r0], #1
-    strb    r5, [r0]
-    m_pop
-
-    .p2align 2
-    // Can read 4 bytes before possibly crossing a page.
-strcpy_unalign4:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    strd    r2, r3, [r0], #8
-    b       strcpy_unalign4
-
-    .p2align 2
-    // Can read 3 bytes before possibly crossing a page.
-strcpy_unalign3:
-    ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign3_copy1byte
-    ldrb    r3, [r1, #1]
-    cbz     r3, strcpy_unalign3_copy2bytes
-    ldrb    r4, [r1, #2]
-    cbz     r4, strcpy_unalign3_copy3bytes
-
-    ldr     r2, [r1], #4
-    ldr     r3, [r1], #4
-
-    pld     [r1, #64]
-
-    lsrs    lr, r2, #24
-    beq     strcpy_copy4bytes
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    strd    r2, r3, [r0], #8
-    b       strcpy_unalign3
-
-strcpy_unalign3_copy1byte:
-    strb    r2, [r0]
-    m_pop
-
-strcpy_unalign3_copy2bytes:
-    strb    r2, [r0], #1
-    strb    r3, [r0]
-    m_pop
-
-strcpy_unalign3_copy3bytes:
-    strb    r2, [r0], #1
-    strb    r3, [r0], #1
-    strb    r4, [r0]
-    m_pop
-
-    .p2align 2
-    // Can read 2 bytes before possibly crossing a page.
-strcpy_unalign2:
-    ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign_copy1byte
-    ldrb    r4, [r1, #1]
-    cbz     r4, strcpy_unalign_copy2bytes
-
-    ldr     r2, [r1], #4
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    tst     r2, #0xff0000
-    beq     strcpy_copy3bytes
-    lsrs    ip, r2, #24
-    beq     strcpy_copy4bytes
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    strd    r2, r3, [r0], #8
-    b       strcpy_unalign2
-
-    .p2align 2
-    // Can read 1 byte before possibly crossing a page.
-strcpy_unalign1:
-    ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign_copy1byte
-
-    ldr     r2, [r1], #4
-    ldr     r3, [r1], #4
-
-    pld     [r1, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    strd    r2, r3, [r0], #8
-    b       strcpy_unalign1
-
-strcpy_unalign_copy1byte:
-    strb    r2, [r0]
-    m_pop
-
-strcpy_unalign_copy2bytes:
-    strb    r2, [r0], #1
-    strb    r4, [r0]
-    m_pop
-END(strcpy)

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

@@ -1,165 +0,0 @@
-/*
- * 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

@@ -1,10 +1,5 @@
-libc_bionic_src_files_arm += \
-    arch-arm/cortex-a15/bionic/memcpy.S \
-    arch-arm/cortex-a15/bionic/memset.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 \
-    arch-arm/cortex-a15/bionic/__strcat_chk.S \
-    arch-arm/cortex-a15/bionic/__strcpy_chk.S \
-    bionic/memmove.c \
+$(call libc-add-cpu-variant-src,MEMCPY,arch-arm/cortex-a15/bionic/memcpy.S)
+$(call libc-add-cpu-variant-src,MEMSET,arch-arm/cortex-a15/bionic/memset.S)
+$(call libc-add-cpu-variant-src,STRCMP,arch-arm/cortex-a15/bionic/strcmp.S)
+
+include bionic/libc/arch-arm/generic/generic.mk

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

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

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

@@ -1,224 +0,0 @@
-/*
- * 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}
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-    push    {r4, r5}
-    .save   {r4, r5}
-    .cfi_adjust_cfa_offset 8
-    .cfi_rel_offset r4, 0
-    .cfi_rel_offset r5, 4
-
-    mov     lr, r2
-
-    // Save the dst register to r5
-    mov     r5, r0
-
-    // Zero out r4
-    eor     r4, r4, r4
-
-    // r1 contains the address of the string to count.
-.L_strlen_start:
-    mov     r0, r1
-
-    ands    r3, 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)
-    .save   {r0, lr}
-    .save   {r4, r5}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-    .cfi_adjust_cfa_offset 8
-    .cfi_rel_offset r4, 0
-    .cfi_rel_offset r5, 4
-
-    ldr     r0, error_message
-    ldr     r1, error_code
-1:
-    add     r0, pc
-    bl      __fortify_chk_fail
-error_code:
-    .word   BIONIC_EVENT_STRCAT_BUFFER_OVERFLOW
-error_message:
-    .word   error_string-(1b+4)
-END(__strcat_chk_fail)
-
-    .data
-error_string:
-    .string "strcat: prevented write past end of buffer"

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

@@ -1,187 +0,0 @@
-/*
- * 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}
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-
-    mov     lr, r2
-    mov     r0, r1
-
-    ands    r3, 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)
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-
-    ldr     r0, error_message
-    ldr     r1, error_code
-1:
-    add     r0, pc
-    bl      __fortify_chk_fail
-
-error_code:
-    .word   BIONIC_EVENT_STRCPY_BUFFER_OVERFLOW
-error_message:
-    .word   error_string-(1b+4)
-END(__strcpy_chk_fail)
-
-    .data
-error_string:
-    .string "strcpy: prevented write past end of buffer"

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

@@ -26,8 +26,8 @@
  * SUCH DAMAGE.
  */
 
-#include <private/bionic_asm.h>
-#include <private/libc_events.h>
+#include <machine/cpu-features.h>
+#include <machine/asm.h>
 
 /*
  * This code assumes it is running on a processor that supports all arm v7
@@ -35,50 +35,177 @@
  * cache line.
  */
 
-        .syntax unified
+        .text
         .fpu    neon
-        .thumb
-        .thumb_func
 
-ENTRY(__memcpy_chk)
-        cmp         r2, r3
-        bhi         __memcpy_chk_fail
-
-        // Fall through to memcpy...
-END(__memcpy_chk)
+#define CACHE_LINE_SIZE     32
 
 ENTRY(memcpy)
-        pld     [r1, #0]
-        stmfd   sp!, {r0, lr}
-        .save   {r0, lr}
-        .cfi_def_cfa_offset 8
-        .cfi_rel_offset r0, 0
-        .cfi_rel_offset lr, 4
-        pld     [r1, #64]
+        .save       {r0, lr}
+        /* start preloading as early as possible */
+        pld         [r1, #(CACHE_LINE_SIZE * 0)]
+        stmfd       sp!, {r0, lr}
+        pld         [r1, #(CACHE_LINE_SIZE * 2)]
+
+        // 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         11f
+
+        /* 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
+        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 * 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, #(CACHE_LINE_SIZE * 4)]
+        pld         [r1, #(CACHE_LINE_SIZE * 6)]
+
+1:      /* The main loop copies 64 bytes at a time */
+        vld1.8      {d0 - d3}, [r1]!
+        vld1.8      {d4 - d7}, [r1]!
+        pld         [r1, #(CACHE_LINE_SIZE * 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
+        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
+11:
+        /* Simple arm-only copy loop to handle aligned copy operations */
+        stmfd       sp!, {r4, r5, r6, r7, r8}
+        pld         [r1, #(CACHE_LINE_SIZE * 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 */
+        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, r2, #64
+        blt         4f
+
+3:      /* Main copy loop, copying 64 bytes at a time */
+        pld         [r1, #(CACHE_LINE_SIZE * 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
+        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)
-
-#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}
-        .save   {lr}
-        .cfi_def_cfa_offset 4
-        .cfi_rel_offset lr, 0
-
-        ldr     r0, error_message
-        ldr     r1, error_code
-1:
-        add     r0, pc
-        bl      __fortify_chk_fail
-error_code:
-        .word   BIONIC_EVENT_MEMCPY_BUFFER_OVERFLOW
-error_message:
-        .word   error_string-(1b+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

@@ -1,226 +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.
- */
-
-/*
- * 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)
-        .save       {r0, lr}
-        .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)
-        .save       {r0, lr}
-        .cfi_def_cfa_offset 8
-        .cfi_rel_offset r0, 0
-        .cfi_rel_offset lr, 4
-
-        /* Simple arm-only copy loop to handle aligned copy operations */
-        stmfd       sp!, {r4-r8}
-        .save       {r4-r8}
-        .cfi_adjust_cfa_offset 20
-        .cfi_rel_offset r4, 0
-        .cfi_rel_offset r5, 4
-        .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

@@ -26,8 +26,8 @@
  * SUCH DAMAGE.
  */
 
-#include <private/bionic_asm.h>
-#include <private/libc_events.h>
+#include <machine/cpu-features.h>
+#include <machine/asm.h>
 
 /*
  * This code assumes it is running on a processor that supports all arm v7
@@ -36,45 +36,19 @@
 
     .fpu    neon
 
-ENTRY(__memset_chk)
-        cmp         r2, r3
-        bls         .L_done
-
-        // Preserve lr for backtrace.
-        push        {lr}
-        .save       {lr}
-        .cfi_def_cfa_offset 4
-        .cfi_rel_offset lr, 0
-
-        ldr         r0, error_message
-        ldr         r1, error_code
-1:
-        add         r0, pc
-        bl          __fortify_chk_fail
-error_code:
-        .word       BIONIC_EVENT_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.
+        # The neon memset only wins for less than 132.
         cmp         r2, #132
-        bhi         __memset_large_copy
+        bhi         11f
 
-        stmfd       sp!, {r0}
         .save       {r0}
-        .cfi_def_cfa_offset 4
-        .cfi_rel_offset r0, 0
+        stmfd       sp!, {r0}
 
         vdup.8      q0, r1
 
@@ -107,22 +81,13 @@ ENTRY(memset)
         strcsb      r1, [r0], #1
         ldmfd       sp!, {r0}
         bx          lr
-END(memset)
-
-ENTRY_PRIVATE(__memset_large_copy)
+11:
         /* compute the offset to align the destination
          * offset = (4-(src&3))&3 = -src & 3
          */
-        stmfd       sp!, {r0, r4-r7, lr}
-        .save       {r0, r4-r7, lr}
-        .cfi_def_cfa_offset 24
-        .cfi_rel_offset r0, 0
-        .cfi_rel_offset r4, 4
-        .cfi_rel_offset r5, 8
-        .cfi_rel_offset r6, 12
-        .cfi_rel_offset r7, 16
-        .cfi_rel_offset lr, 20
 
+        .save       {r0, r4-r7, lr}
+        stmfd       sp!, {r0, r4-r7, lr}
         rsb         r3, r0, #0
         ands        r3, r3, #3
         cmp         r3, r2
@@ -184,8 +149,4 @@ ENTRY_PRIVATE(__memset_large_copy)
         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"
+END(memset)

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

@@ -1,548 +0,0 @@
-/*
- * 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

@@ -27,7 +27,7 @@
  */
 
 #include <machine/cpu-features.h>
-#include <private/bionic_asm.h>
+#include <machine/asm.h>
 
 #ifdef __ARMEB__
 #define S2LOMEM lsl
@@ -123,13 +123,8 @@ ENTRY(strcmp)
         .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 */
@@ -145,8 +140,19 @@ ENTRY(strcmp)
 
         .macro  magic_find_zero_bytes w1
         /* Macro to find all-zero bytes in w1, result is in ip.  */
+#if (defined (__ARM_FEATURE_DSP))
         uadd8   ip, \w1, r6
         sel     ip, r7, r6
+#else /* not defined (__ARM_FEATURE_DSP) */
+        /* __ARM_FEATURE_DSP is not defined for some Cortex-M processors.
+        Coincidently, these processors only have Thumb-2 mode, where we can use the
+        the (large) magic constant available directly as an immediate in instructions.
+        Note that we cannot use the magic constant in ARM mode, where we need
+        to create the constant in a register.  */
+        sub     ip, \w1, #0x01010101
+        bic     ip, ip, \w1
+        and     ip, ip, #0x80808080
+#endif /* not defined (__ARM_FEATURE_DSP) */
         .endm /* magic_find_zero_bytes */
 
         .macro  setup_return w1 w2
@@ -165,13 +171,12 @@ ENTRY(strcmp)
         /* Are both strings double-word aligned?  */
         orr     ip, r0, r1
         tst     ip, #7
-        bne     .L_do_align
+        bne     do_align
 
         /* Fast path.  */
-        .save   {r4-r7}
         init
 
-.L_doubleword_aligned:
+doubleword_aligned:
 
         /* Get here when the strings to compare are double-word aligned.  */
         /* Compare two words in every iteration.  */
@@ -184,14 +189,14 @@ ENTRY(strcmp)
         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
+        magic_compare_and_branch w1=r2, w2=r4, label=return_24
+        magic_compare_and_branch w1=r3, w2=r5, label=return_35
         b       2b
 
-.L_do_align:
+do_align:
         /* Is the first string word-aligned?  */
         ands    ip, r0, #3
-        beq     .L_word_aligned_r0
+        beq     word_aligned_r0
 
         /* Fast compare byte by byte until the first string is word-aligned.  */
         /* The offset of r0 from a word boundary is in ip. Thus, the number of bytes
@@ -199,58 +204,58 @@ ENTRY(strcmp)
         bic     r0, r0, #3
         ldr     r2, [r0], #4
         lsls    ip, ip, #31
-        beq     .L_byte2
-        bcs     .L_byte3
+        beq     byte2
+        bcs     byte3
 
-.L_byte1:
+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
+        bne     fast_return
+        m_cbz   reg=r3, label=fast_return
 
-.L_byte2:
+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
+        bne     fast_return
+        m_cbz   reg=r3, label=fast_return
 
-.L_byte3:
+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
+        bne     fast_return
+        m_cbnz  reg=r3, label=word_aligned_r0
 
-.L_fast_return:
+fast_return:
         mov     r0, ip
         bx      lr
 
-.L_word_aligned_r0:
+word_aligned_r0:
         init
         /* The first string is word-aligned.  */
         /* Is the second string word-aligned?  */
         ands    ip, r1, #3
-        bne     .L_strcmp_unaligned
+        bne     strcmp_unaligned
 
-.L_word_aligned:
+word_aligned:
         /* The strings are word-aligned. */
         /* Is the first string double-word aligned?  */
         tst     r0, #4
-        beq     .L_doubleword_aligned_r0
+        beq     doubleword_aligned_r0
 
         /* If r0 is not double-word aligned yet, align it by loading
         and comparing the next word from each string.  */
         ldr     r2, [r0], #4
         ldr     r4, [r1], #4
-        magic_compare_and_branch w1=r2 w2=r4 label=.L_return_24
+        magic_compare_and_branch w1=r2 w2=r4 label=return_24
 
-.L_doubleword_aligned_r0:
+doubleword_aligned_r0:
         /* Get here when r0 is double-word aligned.  */
         /* Is r1 doubleword_aligned?  */
         tst     r1, #4
-        beq     .L_doubleword_aligned
+        beq     doubleword_aligned
 
         /* Get here when the strings to compare are word-aligned,
         r0 is double-word aligned, but r1 is not double-word aligned.  */
@@ -266,9 +271,9 @@ ENTRY(strcmp)
 
         /* 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
+        magic_compare_and_branch w1=r2 w2=r5 label=return_25
         ldrd    r4, r5, [r1], #8
-        magic_compare_and_branch w1=r3 w2=r4 label=.L_return_34
+        magic_compare_and_branch w1=r3 w2=r4 label=return_34
         b       3b
 
         .macro miscmp_word offsetlo offsethi
@@ -292,33 +297,33 @@ ENTRY(strcmp)
         and     r2, r3, r6, S2LOMEM #\offsetlo
         it      eq
         cmpeq   r2, r5
-        bne     .L_return_25
+        bne     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
+        bne     return_32
         b       7b
         .endm /* miscmp_word */
 
-.L_return_32:
+return_32:
         setup_return w1=r3, w2=r2
-        b       .L_do_return
-.L_return_34:
+        b       do_return
+return_34:
         setup_return w1=r3, w2=r4
-        b       .L_do_return
-.L_return_25:
+        b       do_return
+return_25:
         setup_return w1=r2, w2=r5
-        b       .L_do_return
-.L_return_35:
+        b       do_return
+return_35:
         setup_return w1=r3, w2=r5
-        b       .L_do_return
-.L_return_24:
+        b       do_return
+return_24:
         setup_return w1=r2, w2=r4
 
-.L_do_return:
+do_return:
 
 #ifdef __ARMEB__
         mov     r0, ip
@@ -330,16 +335,11 @@ ENTRY(strcmp)
         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
+        m_cbz   reg=r0, label=compute_return_value
         clz     r0, r0
         /* r0 contains the number of bits on the left of the first all-zero byte in r1.  */
         rsb     r0, r0, #24
@@ -347,7 +347,7 @@ ENTRY(strcmp)
         lsr     r1, r1, r0
         lsr     r2, r2, r0
 
-.L_compute_return_value:
+compute_return_value:
         movs    r0, #1
         cmp     r1, r2
         /* The return value is computed as follows.
@@ -367,7 +367,7 @@ ENTRY(strcmp)
      * bionic/libc/arch-arm/cortex-a15/bionic/strcmp.S for the unedited
      * version of the code.
      */
-.L_strcmp_unaligned:
+strcmp_unaligned:
 	wp1 .req r0
 	wp2 .req r1
 	b1  .req r2
@@ -520,11 +520,6 @@ ENTRY(strcmp)
     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
 

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

@@ -1,456 +0,0 @@
-/*
- * 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_copy_byte reg, cmd, label
-    ldrb    \reg, [r1], #1
-    strb    \reg, [r0], #1
-    \cmd    \reg, \label
-    .endm // m_copy_byte
-
-ENTRY(strcpy)
-    // Unroll the first 8 bytes that will be copied.
-    m_push
-    m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r5, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish
-    m_copy_byte reg=r5, cmd=cbnz, label=strcpy_continue
-
-strcpy_finish:
-    m_ret   inst=pop
-
-strcpy_continue:
-    pld     [r1, #0]
-    ands    r3, r0, #7
-    bne     strcpy_align_dst
-
-strcpy_check_src_align:
-    // At this point dst is aligned to a double word, check if src
-    // is also aligned to a double word.
-    ands    r3, r1, #7
-    bne     strcpy_unaligned_copy
-
-    .p2align 2
-strcpy_mainloop:
-    ldmia   r1!, {r2, r3}
-
-    pld     [r1, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    stmia   r0!, {r2, r3}
-    b       strcpy_mainloop
-
-strcpy_zero_in_first_register:
-    lsls    lr, ip, #17
-    itt     ne
-    strbne  r2, [r0]
-    m_ret   inst=popne
-    itt     cs
-    strhcs  r2, [r0]
-    m_ret   inst=popcs
-    lsls    ip, ip, #1
-    itt     eq
-    streq   r2, [r0]
-    m_ret   inst=popeq
-    strh    r2, [r0], #2
-    lsr     r3, r2, #16
-    strb    r3, [r0]
-    m_ret   inst=pop
-
-strcpy_zero_in_second_register:
-    lsls    lr, ip, #17
-    ittt    ne
-    stmiane r0!, {r2}
-    strbne  r3, [r0]
-    m_ret   inst=popne
-    ittt    cs
-    strcs   r2, [r0], #4
-    strhcs  r3, [r0]
-    m_ret   inst=popcs
-    lsls    ip, ip, #1
-    itt     eq
-    stmiaeq r0, {r2, r3}
-    m_ret   inst=popeq
-    stmia   r0!, {r2}
-    strh    r3, [r0], #2
-    lsr     r4, r3, #16
-    strb    r4, [r0]
-    m_ret   inst=pop
-
-strcpy_align_dst:
-    // Align to a double word (64 bits).
-    rsb     r3, r3, #8
-    lsls    ip, r3, #31
-    beq     strcpy_align_to_32
-
-    ldrb    r2, [r1], #1
-    strb    r2, [r0], #1
-    cbz     r2, strcpy_complete
-
-strcpy_align_to_32:
-    bcc     strcpy_align_to_64
-
-    ldrb    r4, [r1], #1
-    strb    r4, [r0], #1
-    cmp     r4, #0
-    it      eq
-    m_ret   inst=popeq
-    ldrb    r5, [r1], #1
-    strb    r5, [r0], #1
-    cmp     r5, #0
-    it      eq
-    m_ret   inst=popeq
-
-strcpy_align_to_64:
-    tst     r3, #4
-    beq     strcpy_check_src_align
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-    stmia   r0!, {r2}
-    b       strcpy_check_src_align
-
-strcpy_complete:
-    m_ret   inst=pop
-
-strcpy_unaligned_copy:
-    // Dst is aligned to a double word, while src is at an unknown alignment.
-    // There are 7 different versions of the unaligned copy code
-    // to prevent overreading the src. The mainloop of every single version
-    // will store 64 bits per loop. The difference is how much of src can
-    // be read without potentially crossing a page boundary.
-    tbb     [pc, r3]
-strcpy_unaligned_branchtable:
-    .byte 0
-    .byte ((strcpy_unalign7 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign6 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign5 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign4 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign3 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign2 - strcpy_unaligned_branchtable)/2)
-    .byte ((strcpy_unalign1 - strcpy_unaligned_branchtable)/2)
-
-    .p2align 2
-    // Can read 7 bytes before possibly crossing a page.
-strcpy_unalign7:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldrb    r3, [r1]
-    cbz     r3, strcpy_unalign7_copy5bytes
-    ldrb    r4, [r1, #1]
-    cbz     r4, strcpy_unalign7_copy6bytes
-    ldrb    r5, [r1, #2]
-    cbz     r5, strcpy_unalign7_copy7bytes
-
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    lsrs    ip, r3, #24
-    stmia   r0!, {r2, r3}
-    beq     strcpy_unalign_return
-    b       strcpy_unalign7
-
-strcpy_unalign7_copy5bytes:
-    stmia   r0!, {r2}
-    strb    r3, [r0]
-strcpy_unalign_return:
-    m_ret   inst=pop
-
-strcpy_unalign7_copy6bytes:
-    stmia   r0!, {r2}
-    strb    r3, [r0], #1
-    strb    r4, [r0], #1
-    m_ret   inst=pop
-
-strcpy_unalign7_copy7bytes:
-    stmia   r0!, {r2}
-    strb    r3, [r0], #1
-    strb    r4, [r0], #1
-    strb    r5, [r0], #1
-    m_ret   inst=pop
-
-    .p2align 2
-    // Can read 6 bytes before possibly crossing a page.
-strcpy_unalign6:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldrb    r4, [r1]
-    cbz     r4, strcpy_unalign_copy5bytes
-    ldrb    r5, [r1, #1]
-    cbz     r5, strcpy_unalign_copy6bytes
-
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    tst     r3, #0xff0000
-    beq     strcpy_unalign6_copy7bytes
-    lsrs    ip, r3, #24
-    stmia   r0!, {r2, r3}
-    beq     strcpy_unalign_return
-    b       strcpy_unalign6
-
-strcpy_unalign6_copy7bytes:
-    stmia   r0!, {r2}
-    strh    r3, [r0], #2
-    lsr     r3, #16
-    strb    r3, [r0]
-    m_ret   inst=pop
-
-    .p2align 2
-    // Can read 5 bytes before possibly crossing a page.
-strcpy_unalign5:
-    ldr     r2, [r1], #4
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldrb    r4, [r1]
-    cbz     r4, strcpy_unalign_copy5bytes
-
-    ldr     r3, [r1], #4
-
-    pld     [r1, #64]
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    stmia   r0!, {r2, r3}
-    b       strcpy_unalign5
-
-strcpy_unalign_copy5bytes:
-    stmia   r0!, {r2}
-    strb    r4, [r0]
-    m_ret   inst=pop
-
-strcpy_unalign_copy6bytes:
-    stmia   r0!, {r2}
-    strb    r4, [r0], #1
-    strb    r5, [r0]
-    m_ret   inst=pop
-
-    .p2align 2
-    // Can read 4 bytes before possibly crossing a page.
-strcpy_unalign4:
-    ldmia   r1!, {r2}
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    ldmia   r1!, {r3}
-    pld     [r1, #64]
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    stmia   r0!, {r2, r3}
-    b       strcpy_unalign4
-
-    .p2align 2
-    // Can read 3 bytes before possibly crossing a page.
-strcpy_unalign3:
-    ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign3_copy1byte
-    ldrb    r3, [r1, #1]
-    cbz     r3, strcpy_unalign3_copy2bytes
-    ldrb    r4, [r1, #2]
-    cbz     r4, strcpy_unalign3_copy3bytes
-
-    ldr     r2, [r1], #4
-    ldr     r3, [r1], #4
-
-    pld     [r1, #64]
-
-    lsrs    lr, r2, #24
-    beq     strcpy_unalign_copy4bytes
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    stmia   r0!, {r2, r3}
-    b       strcpy_unalign3
-
-strcpy_unalign3_copy1byte:
-    strb    r2, [r0]
-    m_ret   inst=pop
-
-strcpy_unalign3_copy2bytes:
-    strb    r2, [r0], #1
-    strb    r3, [r0]
-    m_ret   inst=pop
-
-strcpy_unalign3_copy3bytes:
-    strb    r2, [r0], #1
-    strb    r3, [r0], #1
-    strb    r4, [r0]
-    m_ret   inst=pop
-
-    .p2align 2
-    // Can read 2 bytes before possibly crossing a page.
-strcpy_unalign2:
-    ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign_copy1byte
-    ldrb    r3, [r1, #1]
-    cbz     r3, strcpy_unalign_copy2bytes
-
-    ldr     r2, [r1], #4
-    ldr     r3, [r1], #4
-    pld     [r1, #64]
-
-    tst     r2, #0xff0000
-    beq     strcpy_unalign_copy3bytes
-    lsrs    ip, r2, #24
-    beq     strcpy_unalign_copy4bytes
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    stmia   r0!, {r2, r3}
-    b       strcpy_unalign2
-
-    .p2align 2
-    // Can read 1 byte before possibly crossing a page.
-strcpy_unalign1:
-    ldrb    r2, [r1]
-    cbz     r2, strcpy_unalign_copy1byte
-
-    ldr     r2, [r1], #4
-    ldr     r3, [r1], #4
-
-    pld     [r1, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     strcpy_zero_in_second_register
-
-    stmia   r0!, {r2, r3}
-    b       strcpy_unalign1
-
-strcpy_unalign_copy1byte:
-    strb    r2, [r0]
-    m_ret   inst=pop
-
-strcpy_unalign_copy2bytes:
-    strb    r2, [r0], #1
-    strb    r3, [r0]
-    m_ret   inst=pop
-
-strcpy_unalign_copy3bytes:
-    strh    r2, [r0], #2
-    lsr     r2, #16
-    strb    r2, [r0]
-    m_ret   inst=pop
-
-strcpy_unalign_copy4bytes:
-    stmia   r0, {r2}
-    m_ret   inst=pop
-END(strcpy)

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

@@ -1,167 +0,0 @@
-/*
- * 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

@@ -1,10 +1,5 @@
-libc_bionic_src_files_arm += \
-    arch-arm/cortex-a9/bionic/memcpy.S \
-    arch-arm/cortex-a9/bionic/memset.S \
-    arch-arm/cortex-a9/bionic/strcat.S \
-    arch-arm/cortex-a9/bionic/strcmp.S \
-    arch-arm/cortex-a9/bionic/strcpy.S \
-    arch-arm/cortex-a9/bionic/strlen.S \
-    arch-arm/cortex-a9/bionic/__strcat_chk.S \
-    arch-arm/cortex-a9/bionic/__strcpy_chk.S \
-    bionic/memmove.c \
+$(call libc-add-cpu-variant-src,MEMCPY,arch-arm/cortex-a9/bionic/memcpy.S)
+$(call libc-add-cpu-variant-src,MEMSET,arch-arm/cortex-a9/bionic/memset.S)
+$(call libc-add-cpu-variant-src,STRCMP,arch-arm/cortex-a9/bionic/strcmp.S)
+
+include bionic/libc/arch-arm/generic/generic.mk

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

@@ -1,221 +0,0 @@
-/*
- * 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}
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-    push    {r4, r5}
-    .save   {r4, r5}
-    .cfi_adjust_cfa_offset 8
-    .cfi_rel_offset r4, 0
-    .cfi_rel_offset r5, 4
-
-    mov     lr, r2
-
-    // Save the dst register to r5
-    mov     r5, r0
-
-    // Zero out r4
-    eor     r4, r4, r4
-
-    // r1 contains the address of the string to count.
-.L_strlen_start:
-    mov     r0, r1
-    ands    r3, r1, #7
-    beq     .L_mainloop
-
-    // Align to a double word (64 bits).
-    rsb     r3, r3, #8
-    lsls    ip, r3, #31
-    beq     .L_align_to_32
-
-    ldrb    r2, [r1], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_32:
-    bcc     .L_align_to_64
-    ands    ip, r3, #2
-    beq     .L_align_to_64
-
-    ldrb    r2, [r1], #1
-    cbz     r2, .L_update_count_and_finish
-    ldrb    r2, [r1], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_64:
-    tst     r3, #4
-    beq     .L_mainloop
-    ldr     r3, [r1], #4
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-
-    .p2align 2
-.L_mainloop:
-    ldrd    r2, r3, [r1], #8
-
-    pld     [r1, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-    b       .L_mainloop
-
-.L_update_count_and_finish:
-    sub     r3, r1, r0
-    sub     r3, r3, #1
-    b       .L_finish
-
-.L_zero_in_first_register:
-    sub     r3, r1, r0
-    lsls    r2, ip, #17
-    bne     .L_sub8_and_finish
-    bcs     .L_sub7_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub6_and_finish
-
-    sub     r3, r3, #5
-    b       .L_finish
-
-.L_sub8_and_finish:
-    sub     r3, r3, #8
-    b       .L_finish
-
-.L_sub7_and_finish:
-    sub     r3, r3, #7
-    b       .L_finish
-
-.L_sub6_and_finish:
-    sub     r3, r3, #6
-    b       .L_finish
-
-.L_zero_in_second_register:
-    sub     r3, r1, r0
-    lsls    r2, ip, #17
-    bne     .L_sub4_and_finish
-    bcs     .L_sub3_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub2_and_finish
-
-    sub     r3, r3, #1
-    b       .L_finish
-
-.L_sub4_and_finish:
-    sub     r3, r3, #4
-    b       .L_finish
-
-.L_sub3_and_finish:
-    sub     r3, r3, #3
-    b       .L_finish
-
-.L_sub2_and_finish:
-    sub     r3, r3, #2
-
-.L_finish:
-    cmp     r4, #0
-    bne     .L_strlen_done
-
-    // Time to get the dst string length.
-    mov     r1, r5
-
-    // Save the original source address to r5.
-    mov     r5, r0
-
-    // Save the current length (adding 1 for the terminator).
-    add     r4, r3, #1
-    b       .L_strlen_start
-
-    // r0 holds the pointer to the dst string.
-    // r3 holds the dst string length.
-    // r4 holds the src string length + 1.
-.L_strlen_done:
-    add     r2, r3, r4
-    cmp     r2, lr
-    bhi     __strcat_chk_failed
-
-    // Set up the registers for the memcpy code.
-    mov     r1, r5
-    pld     [r1, #64]
-    mov     r2, r4
-    add     r0, r0, r3
-    pop     {r4, r5}
-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)
-    .save   {r0, lr}
-    .save   {r4, r5}
-
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-    .cfi_adjust_cfa_offset 8
-    .cfi_rel_offset r4, 0
-    .cfi_rel_offset r5, 4
-
-    ldr     r0, error_message
-    ldr     r1, error_code
-1:
-    add     r0, pc
-    bl      __fortify_chk_fail
-error_code:
-    .word   BIONIC_EVENT_STRCAT_BUFFER_OVERFLOW
-error_message:
-    .word   error_string-(1b+4)
-END(__strcat_chk_failed)
-
-    .data
-error_string:
-    .string "strcat: prevented write past end of buffer"

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

@@ -1,182 +0,0 @@
-/*
- * 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}
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-
-    mov     lr, r2
-    mov     r0, r1
-
-    ands    r3, r1, #7
-    beq     .L_mainloop
-
-    // Align to a double word (64 bits).
-    rsb     r3, r3, #8
-    lsls    ip, r3, #31
-    beq     .L_align_to_32
-
-    ldrb    r2, [r0], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_32:
-    bcc     .L_align_to_64
-    ands    ip, r3, #2
-    beq     .L_align_to_64
-
-    ldrb    r2, [r0], #1
-    cbz     r2, .L_update_count_and_finish
-    ldrb    r2, [r0], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_64:
-    tst     r3, #4
-    beq     .L_mainloop
-    ldr     r3, [r0], #4
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-
-    .p2align 2
-.L_mainloop:
-    ldrd    r2, r3, [r0], #8
-
-    pld     [r0, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-    b       .L_mainloop
-
-.L_update_count_and_finish:
-    sub     r3, r0, r1
-    sub     r3, r3, #1
-    b       .L_check_size
-
-.L_zero_in_first_register:
-    sub     r3, r0, r1
-    lsls    r2, ip, #17
-    bne     .L_sub8_and_finish
-    bcs     .L_sub7_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub6_and_finish
-
-    sub     r3, r3, #5
-    b       .L_check_size
-
-.L_sub8_and_finish:
-    sub     r3, r3, #8
-    b       .L_check_size
-
-.L_sub7_and_finish:
-    sub     r3, r3, #7
-    b       .L_check_size
-
-.L_sub6_and_finish:
-    sub     r3, r3, #6
-    b       .L_check_size
-
-.L_zero_in_second_register:
-    sub     r3, r0, r1
-    lsls    r2, ip, #17
-    bne     .L_sub4_and_finish
-    bcs     .L_sub3_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub2_and_finish
-
-    sub     r3, r3, #1
-    b       .L_check_size
-
-.L_sub4_and_finish:
-    sub     r3, r3, #4
-    b       .L_check_size
-
-.L_sub3_and_finish:
-    sub     r3, r3, #3
-    b       .L_check_size
-
-.L_sub2_and_finish:
-    sub     r3, r3, #2
-
-.L_check_size:
-    pld     [r1, #0]
-    pld     [r1, #64]
-    ldr     r0, [sp]
-    cmp     r3, lr
-    bhs     __strcpy_chk_failed
-
-    // Add 1 for copy length to get the string terminator.
-    add     r2, r3, #1
-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)
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-
-    ldr     r0, error_message
-    ldr     r1, error_code
-1:
-    add     r0, pc
-    bl      __fortify_chk_fail
-error_code:
-    .word   BIONIC_EVENT_STRCPY_BUFFER_OVERFLOW
-error_message:
-    .word   error_string-(1b+4)
-END(__strcpy_chk_failed)
-
-    .data
-error_string:
-    .string "strcpy: prevented write past end of buffer"

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

@@ -1,105 +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.
- */
-/*
- * 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}
-        .save   {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}
-        .save   {lr}
-        .cfi_def_cfa_offset 4
-        .cfi_rel_offset lr, 0
-
-        ldr     r0, error_message
-        ldr     r1, error_code
-1:
-        add     r0, pc
-        bl      __fortify_chk_fail
-error_code:
-        .word   BIONIC_EVENT_MEMCPY_BUFFER_OVERFLOW
-error_message:
-        .word   error_string-(1b+8)
-END(__memcpy_chk_fail)
-
-        .data
-error_string:
-        .string "memcpy: prevented write past end of buffer"

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

@@ -1,234 +0,0 @@
-/*
- * 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

@@ -1,281 +0,0 @@
-/*
- * 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

@@ -1,207 +0,0 @@
-/*
- * 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

@@ -1,13 +0,0 @@
-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/strcat.S \
-    arch-arm/cortex-a15/bionic/strcpy.S \
-    arch-arm/cortex-a15/bionic/strlen.S \
-    arch-arm/cortex-a15/bionic/strcmp.S \

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

@@ -27,8 +27,7 @@
  */
 
 #include <machine/cpu-features.h>
-#include <private/bionic_asm.h>
-#include <private/libc_events.h>
+#include <machine/asm.h>
 
         /*
          * Optimized memcpy() for ARM.
@@ -37,13 +36,6 @@
          * 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
@@ -57,9 +49,9 @@ ENTRY(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
@@ -147,7 +139,7 @@ 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
@@ -268,7 +260,7 @@ 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
@@ -295,7 +287,7 @@ 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
@@ -322,7 +314,7 @@ 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
@@ -385,20 +377,4 @@ copy_last_3_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

@@ -26,8 +26,7 @@
  * SUCH DAMAGE.
  */
 
-#include <private/bionic_asm.h>
-#include <private/libc_events.h>
+#include <machine/asm.h>
 
         /*
          * Optimized memset() for ARM.
@@ -35,28 +34,9 @@
          * 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)
@@ -127,7 +107,3 @@ ENTRY(memset)
         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

@@ -28,7 +28,7 @@
  */
 
 #include <machine/cpu-features.h>
-#include <private/bionic_asm.h>
+#include <machine/asm.h>
 
 	.text
 
@@ -52,8 +52,8 @@
 #define magic2(REG) REG, lsl #7
 
 ENTRY(strcmp)
-	pld	[r0, #0]
-	pld	[r1, #0]
+	PLD(r0, #0)
+	PLD(r1, #0)
 	eor	r2, r0, r1
 	tst	r2, #3
 
@@ -88,8 +88,8 @@ ENTRY(strcmp)
 	orr	r4, r4, r4, lsl #16
 	.p2align	2
 4:
-	pld	[r0, #8]
-	pld	[r1, #8]
+	PLD(r0, #8)
+	PLD(r1, #8)
 	sub	r2, ip, magic1(r4)
 	cmp	ip, r3
 	itttt	eq

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

@@ -1,133 +0,0 @@
-/*
- * 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

@@ -1,136 +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"
-        "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

@@ -1,10 +1,3 @@
-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/strcat.c \
+$(call libc-add-cpu-variant-src,MEMCPY,arch-arm/generic/bionic/memcpy.S)
+$(call libc-add-cpu-variant-src,MEMSET,arch-arm/generic/bionic/memset.S)
+$(call libc-add-cpu-variant-src,STRCMP,arch-arm/generic/bionic/strcmp.S)

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

@@ -0,0 +1,110 @@
+/*	$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_
+
+/* 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 double			__double_t;
+typedef float			__float_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;
+
+#endif	/* _ARM__TYPES_H_ */

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

@@ -38,22 +38,107 @@
 #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
+
+#ifdef __STDC__
+# define __CONCAT(x,y)	x ## y
+# define __STRING(x)	#x
+#else
+# define __CONCAT(x,y)	x/**/y
+# define __STRING(x)	"x"
+#endif
+
 #ifndef _ALIGN_TEXT
 # define _ALIGN_TEXT .align 0
 #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
+/*
+ * 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: .fnstart
 
-#undef __bionic_asm_function_type
-#define __bionic_asm_function_type #function
+#define _ASM_SIZE(x)	.size x, .-x;
 
-#if defined(__ELF__) && defined(PIC)
-#define PIC_SYM(x,y) x ## ( ## y ## )
+#define _END(x) \
+	.fnend; \
+	_ASM_SIZE(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 PIC_SYM(x,y) x
+# define _PROF_PROLOGUE
 #endif
 
+#define	ENTRY(y)	_ENTRY(_C_LABEL(y)); _PROF_PROLOGUE
+#define	ENTRY_NP(y)	_ENTRY(_C_LABEL(y))
+#define	END(y)		_END(_C_LABEL(y))
+#define	ASENTRY(y)	_ENTRY(_ASM_LABEL(y)); _PROF_PROLOGUE
+#define	ASENTRY_NP(y)	_ENTRY(_ASM_LABEL(y))
+#define	ASEND(y)	_END(_ASM_LABEL(y))
+
+#ifdef __ELF__
+#define ENTRY_PRIVATE(y)  ENTRY(y); .hidden _C_LABEL(y)
+#else
+#define ENTRY_PRIVATE(y)  ENTRY(y)
+#endif
+
+#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__ */
+
 #endif /* !_ARM_ASM_H_ */

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

@@ -0,0 +1,19 @@
+/*	$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

@@ -34,29 +34,133 @@
  *
  * 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__ is a number corresponding to the ARM revision
- * we're going to support. Our toolchain doesn't define __ARM_ARCH__
+ * we're going to support
+ *
+ * it looks like 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__
+      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__
+      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
  */
-#define  __ARM_HAVE_HALFWORD_MULTIPLY  1
+#if __ARM_HAVE_5TE
+#  define  __ARM_HAVE_HALFWORD_MULTIPLY  1
+#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'
+ */
+#if __ARM_HAVE_5TE
+#  define  __ARM_HAVE_PAIR_LOAD_STORE 1
+#endif
+
+/* define __ARM_HAVE_SATURATED_ARITHMETIC is you have the saturated integer
+ * arithmetic instructions: qdd, qdadd, qsub, qdsub
+ */
+#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 explicitly changes the
+ * value of the pc register, including ldm { ...,pc } or 'add pc, #offset'
+ */
+#if __ARM_ARCH__ >= 5
+#  define __ARM_HAVE_PC_INTERWORK
+#endif
+
+/* define __ARM_HAVE_LDREX_STREX for ARMv6 and ARMv7 architecture to be
+ * used in replacement of deprecated swp instruction
+ */
+#if __ARM_ARCH__ >= 6
+#  define __ARM_HAVE_LDREX_STREX
+#endif
+
+/* define __ARM_HAVE_DMB for ARMv7 architecture
+ */
+#if __ARM_ARCH__ >= 7
+#  define __ARM_HAVE_DMB
+#endif
 
 /* define __ARM_HAVE_LDREXD for ARMv7 architecture
  * (also present in ARMv6K, but not implemented in ARMv7-M, neither of which
@@ -80,4 +184,18 @@
 #  define __ARM_HAVE_NEON
 #endif
 
+/* Assembly-only macros */
+#ifdef __ASSEMBLY__
+
+/* 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 */
+#endif
+
+#endif /* ! __ASSEMBLY__ */
+
 #endif /* _ARM_MACHINE_CPU_FEATURES_H */

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

@@ -33,6 +33,15 @@
 
 #ifdef __GNUC__
 
+/*
+ * REV and REV16 weren't available on ARM5 or ARM4.
+ * We don't include <machine/cpu-features.h> because it pollutes the
+ * namespace with macros like PLD.
+ */
+#if !defined __ARM_ARCH_5__ && !defined __ARM_ARCH_5T__ && \
+    !defined __ARM_ARCH_5TE__ && !defined __ARM_ARCH_5TEJ__ && \
+    !defined __ARM_ARCH_4T__ && !defined __ARM_ARCH_4__
+
 /* 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.
  *
@@ -46,13 +55,13 @@
  */
 #define __swap16md(x) ({                                        \
     register u_int16_t _x = (x);                                \
-    __asm__ __volatile__("rev16 %0, %0" : "+l" (_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));             \
+    __asm volatile ("rev %0, %0" : "+l" (_x));                  \
     _x;                                                         \
 })
 
@@ -65,6 +74,7 @@
 /* Tell sys/endian.h we have MD variants of the swap macros.  */
 #define MD_SWAP
 
+#endif  /* __ARM_ARCH__ */
 #endif  /* __GNUC__ */
 
 #if defined(__ARMEB__)

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

@@ -0,0 +1,191 @@
+/*	$OpenBSD: ieee.h,v 1.1 2004/02/01 05:09:49 drahn Exp $	*/
+/*	$NetBSD: ieee.h,v 1.2 2001/02/21 17:43:50 bjh21 Exp $	*/
+
+/*
+ * Copyright (c) 1992, 1993
+ *	The Regents of the University of California.  All rights reserved.
+ *
+ * This software was developed by the Computer Systems Engineering group
+ * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
+ * contributed to Berkeley.
+ *
+ * All advertising materials mentioning features or use of this software
+ * must display the following acknowledgement:
+ *	This product includes software developed by the University of
+ *	California, Lawrence Berkeley Laboratory.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *	This product includes software developed by the University of
+ *	California, Berkeley and its contributors.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE 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.
+ *
+ *	@(#)ieee.h	8.1 (Berkeley) 6/11/93
+ */
+
+/*
+ * ieee.h defines the machine-dependent layout of the machine's IEEE
+ * floating point.
+ */
+
+/*
+ * Define the number of bits in each fraction and exponent.
+ *
+ *		     k	         k+1
+ * Note that  1.0 x 2  == 0.1 x 2      and that denorms are represented
+ *
+ *					  (-exp_bias+1)
+ * as fractions that look like 0.fffff x 2             .  This means that
+ *
+ *			 -126
+ * the number 0.10000 x 2    , for instance, is the same as the normalized
+ *
+ *		-127			   -128
+ * float 1.0 x 2    .  Thus, to represent 2    , we need one leading zero
+ *
+ *				  -129
+ * in the fraction; to represent 2    , we need two, and so on.  This
+ *
+ *						     (-exp_bias-fracbits+1)
+ * implies that the smallest denormalized number is 2
+ *
+ * for whichever format we are talking about: for single precision, for
+ *
+ *						-126		-149
+ * instance, we get .00000000000000000000001 x 2    , or 1.0 x 2    , and
+ *
+ * -149 == -127 - 23 + 1.
+ */
+
+/*
+ * The ARM has two sets of FP data formats.  The FPA supports 32-bit, 64-bit
+ * and 96-bit IEEE formats, with the words in big-endian order.  VFP supports
+ * 32-bin and 64-bit IEEE formats with the words in the CPU's native byte
+ * order.
+ *
+ * The FPA also has two packed decimal formats, but we ignore them here.
+ */
+
+#define	SNG_EXPBITS	8
+#define	SNG_FRACBITS	23
+
+#define	DBL_EXPBITS	11
+#define	DBL_FRACBITS	52
+
+#ifndef __VFP_FP__
+#define	E80_EXPBITS	15
+#define	E80_FRACBITS	64
+
+#define	EXT_EXPBITS	15
+#define	EXT_FRACBITS	112
+#endif
+
+struct ieee_single {
+	u_int	sng_frac:23;
+	u_int	sng_exponent:8;
+	u_int	sng_sign:1;
+};
+
+#ifdef __VFP_FP__
+struct ieee_double {
+#ifdef __ARMEB__
+	u_int	dbl_sign:1;
+	u_int	dbl_exp:11;
+	u_int	dbl_frach:20;
+	u_int	dbl_fracl;
+#else /* !__ARMEB__ */
+	u_int	dbl_fracl;
+	u_int	dbl_frach:20;
+	u_int	dbl_exp:11;
+	u_int	dbl_sign:1;
+#endif /* !__ARMEB__ */
+};
+#else /* !__VFP_FP__ */
+struct ieee_double {
+	u_int	dbl_frach:20;
+	u_int	dbl_exp:11;
+	u_int	dbl_sign:1;
+	u_int	dbl_fracl;
+};
+
+union ieee_double_u {
+	double                  dblu_d;
+	struct ieee_double      dblu_dbl;
+};
+
+
+struct ieee_e80 {
+	u_int	e80_exp:15;
+	u_int	e80_zero:16;
+	u_int	e80_sign:1;
+	u_int	e80_frach:31;
+	u_int	e80_j:1;
+	u_int	e80_fracl;
+};
+
+struct ieee_ext {
+	u_int	ext_frach:16;
+	u_int	ext_exp:15;
+	u_int	ext_sign:1;
+	u_int	ext_frachm;
+	u_int	ext_fraclm;
+	u_int	ext_fracl;
+};
+#endif /* !__VFP_FP__ */
+
+/*
+ * Floats whose exponent is in [1..INFNAN) (of whatever type) are
+ * `normal'.  Floats whose exponent is INFNAN are either Inf or NaN.
+ * Floats whose exponent is zero are either zero (iff all fraction
+ * bits are zero) or subnormal values.
+ *
+ * A NaN is a `signalling NaN' if its QUIETNAN bit is clear in its
+ * high fraction; if the bit is set, it is a `quiet NaN'.
+ */
+#define	SNG_EXP_INFNAN	255
+#define	DBL_EXP_INFNAN	2047
+#ifndef __VFP_FP__
+#define	E80_EXP_INFNAN	32767
+#define	EXT_EXP_INFNAN	32767
+#endif /* !__VFP_FP__ */
+
+#if 0
+#define	SNG_QUIETNAN	(1 << 22)
+#define	DBL_QUIETNAN	(1 << 19)
+#ifndef __VFP_FP__
+#define	E80_QUIETNAN	(1 << 15)
+#define	EXT_QUIETNAN	(1 << 15)
+#endif /* !__VFP_FP__ */
+#endif
+
+/*
+ * Exponent biases.
+ */
+#define	SNG_EXP_BIAS	127
+#define	DBL_EXP_BIAS	1023
+#ifndef __VFP_FP__
+#define	E80_EXP_BIAS	16383
+#define	EXT_EXP_BIAS	16383
+#endif /* !__VFP_FP__ */

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

@@ -0,0 +1,10 @@
+/* $OpenBSD: internal_types.h,v 1.2 2004/05/06 15:53:39 drahn Exp $ */
+/* Public domain */
+#ifndef _ARM_INTERNAL_TYPES_H_
+#define _ARM_INTERNAL_TYPES_H_
+
+#ifdef __CHAR_UNSIGNED__
+#define __machine_has_unsigned_chars
+#endif
+
+#endif

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

@@ -0,0 +1,42 @@
+/*
+ * 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.
+ */
+#ifndef _ARCH_ARM_KERNEL_H
+#define _ARCH_ARM_KERNEL_H
+
+/* this file contains kernel-specific definitions that were optimized out of
+   our processed kernel headers, but still useful nonetheless... */
+
+typedef unsigned long   __kernel_blkcnt_t;
+typedef unsigned long   __kernel_blksize_t;
+
+/* these aren't really defined by the kernel headers though... */
+typedef unsigned long   __kernel_fsblkcnt_t;
+typedef unsigned long   __kernel_fsfilcnt_t;
+typedef unsigned int    __kernel_id_t;
+
+#endif /* _ARCH_ARM_KERNEL_H */

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

@@ -0,0 +1,63 @@
+/*	$OpenBSD: limits.h,v 1.3 2006/01/06 22:48:46 millert Exp $	*/
+/*	$NetBSD: limits.h,v 1.4 2003/04/28 23:16:18 bjh21 Exp $	*/
+
+/*
+ * Copyright (c) 1988 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.
+ *
+ *	from: @(#)limits.h	7.2 (Berkeley) 6/28/90
+ */
+
+#ifndef	_ARM32_LIMITS_H_
+#define	_ARM32_LIMITS_H_
+
+#include <sys/cdefs.h>
+
+#define	MB_LEN_MAX	1		/* no multibyte characters */
+
+#ifndef	SIZE_MAX
+#define	SIZE_MAX	UINT_MAX	/* max value for a size_t */
+#endif
+#ifndef SSIZE_MAX
+#define	SSIZE_MAX	INT_MAX		/* max value for a ssize_t */
+#endif
+
+#if __BSD_VISIBLE
+#define	SIZE_T_MAX	UINT_MAX	/* max value for a size_t (historic) */
+
+#define	UQUAD_MAX	0xffffffffffffffffULL		/* max unsigned quad */
+#define	QUAD_MAX	0x7fffffffffffffffLL		/* max signed quad */
+#define	QUAD_MIN	(-0x7fffffffffffffffLL-1)	/* min signed quad */
+
+#endif /* __BSD_VISIBLE */
+
+#define LONGLONG_BIT    64
+#define LONGLONG_MIN    (-9223372036854775807LL-1)
+#define LONGLONG_MAX    9223372036854775807LL
+#define ULONGLONG_MAX   18446744073709551615ULL
+
+#endif	/* _ARM32_LIMITS_H_ */

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

@@ -1,219 +0,0 @@
-/*
- * 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}
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-    push    {r4, r5}
-    .save   {r4, r5}
-    .cfi_adjust_cfa_offset 8
-    .cfi_rel_offset r4, 0
-    .cfi_rel_offset r5, 4
-
-    mov     lr, r2
-
-    // Save the dst register to r5
-    mov     r5, r0
-
-    // Zero out r4
-    eor     r4, r4, r4
-
-    // r1 contains the address of the string to count.
-.L_strlen_start:
-    mov     r0, r1
-    ands    r3, r1, #7
-    beq     .L_mainloop
-
-    // Align to a double word (64 bits).
-    rsb     r3, r3, #8
-    lsls    ip, r3, #31
-    beq     .L_align_to_32
-
-    ldrb    r2, [r1], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_32:
-    bcc     .L_align_to_64
-    ands    ip, r3, #2
-    beq     .L_align_to_64
-
-    ldrb    r2, [r1], #1
-    cbz     r2, .L_update_count_and_finish
-    ldrb    r2, [r1], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_64:
-    tst     r3, #4
-    beq     .L_mainloop
-    ldr     r3, [r1], #4
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-
-    .p2align 2
-.L_mainloop:
-    ldrd    r2, r3, [r1], #8
-
-    pld     [r1, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-    b       .L_mainloop
-
-.L_update_count_and_finish:
-    sub     r3, r1, r0
-    sub     r3, r3, #1
-    b       .L_finish
-
-.L_zero_in_first_register:
-    sub     r3, r1, r0
-    lsls    r2, ip, #17
-    bne     .L_sub8_and_finish
-    bcs     .L_sub7_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub6_and_finish
-
-    sub     r3, r3, #5
-    b       .L_finish
-
-.L_sub8_and_finish:
-    sub     r3, r3, #8
-    b       .L_finish
-
-.L_sub7_and_finish:
-    sub     r3, r3, #7
-    b       .L_finish
-
-.L_sub6_and_finish:
-    sub     r3, r3, #6
-    b       .L_finish
-
-.L_zero_in_second_register:
-    sub     r3, r1, r0
-    lsls    r2, ip, #17
-    bne     .L_sub4_and_finish
-    bcs     .L_sub3_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub2_and_finish
-
-    sub     r3, r3, #1
-    b       .L_finish
-
-.L_sub4_and_finish:
-    sub     r3, r3, #4
-    b       .L_finish
-
-.L_sub3_and_finish:
-    sub     r3, r3, #3
-    b       .L_finish
-
-.L_sub2_and_finish:
-    sub     r3, r3, #2
-
-.L_finish:
-    cmp     r4, #0
-    bne     .L_strlen_done
-
-    // Time to get the dst string length.
-    mov     r1, r5
-
-    // Save the original source address to r5.
-    mov     r5, r0
-
-    // Save the current length (adding 1 for the terminator).
-    add     r4, r3, #1
-    b       .L_strlen_start
-
-    // r0 holds the pointer to the dst string.
-    // r3 holds the dst string length.
-    // r4 holds the src string length + 1.
-.L_strlen_done:
-    add     r2, r3, r4
-    cmp     r2, lr
-    bhi     __strcat_chk_failed
-
-    // Set up the registers for the memcpy code.
-    mov     r1, r5
-    pld     [r1, #64]
-    mov     r2, r4
-    add     r0, r0, r3
-    pop     {r4, r5}
-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)
-    .save   {r0, lr}
-    .save   {r4, r5}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-    .cfi_adjust_cfa_offset 8
-    .cfi_rel_offset r4, 0
-    .cfi_rel_offset r5, 4
-
-    ldr     r0, error_message
-    ldr     r1, error_code
-1:
-    add     r0, pc
-    bl      __fortify_chk_fail
-error_code:
-    .word   BIONIC_EVENT_STRCAT_BUFFER_OVERFLOW
-error_message:
-    .word   error_string-(1b+4)
-END(__strcat_chk_failed)
-
-    .data
-error_string:
-    .string "strcat: prevented write past end of buffer"

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

@@ -1,182 +0,0 @@
-/*
- * 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}
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-
-    mov     lr, r2
-    mov     r0, r1
-
-    ands    r3, r1, #7
-    beq     .L_mainloop
-
-    // Align to a double word (64 bits).
-    rsb     r3, r3, #8
-    lsls    ip, r3, #31
-    beq     .L_align_to_32
-
-    ldrb    r2, [r0], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_32:
-    bcc     .L_align_to_64
-    ands    ip, r3, #2
-    beq     .L_align_to_64
-
-    ldrb    r2, [r0], #1
-    cbz     r2, .L_update_count_and_finish
-    ldrb    r2, [r0], #1
-    cbz     r2, .L_update_count_and_finish
-
-.L_align_to_64:
-    tst     r3, #4
-    beq     .L_mainloop
-    ldr     r3, [r0], #4
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-
-    .p2align 2
-.L_mainloop:
-    ldrd    r2, r3, [r0], #8
-
-    pld     [r0, #64]
-
-    sub     ip, r2, #0x01010101
-    bic     ip, ip, r2
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_first_register
-
-    sub     ip, r3, #0x01010101
-    bic     ip, ip, r3
-    ands    ip, ip, #0x80808080
-    bne     .L_zero_in_second_register
-    b       .L_mainloop
-
-.L_update_count_and_finish:
-    sub     r3, r0, r1
-    sub     r3, r3, #1
-    b       .L_check_size
-
-.L_zero_in_first_register:
-    sub     r3, r0, r1
-    lsls    r2, ip, #17
-    bne     .L_sub8_and_finish
-    bcs     .L_sub7_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub6_and_finish
-
-    sub     r3, r3, #5
-    b       .L_check_size
-
-.L_sub8_and_finish:
-    sub     r3, r3, #8
-    b       .L_check_size
-
-.L_sub7_and_finish:
-    sub     r3, r3, #7
-    b       .L_check_size
-
-.L_sub6_and_finish:
-    sub     r3, r3, #6
-    b       .L_check_size
-
-.L_zero_in_second_register:
-    sub     r3, r0, r1
-    lsls    r2, ip, #17
-    bne     .L_sub4_and_finish
-    bcs     .L_sub3_and_finish
-    lsls    ip, ip, #1
-    bne     .L_sub2_and_finish
-
-    sub     r3, r3, #1
-    b       .L_check_size
-
-.L_sub4_and_finish:
-    sub     r3, r3, #4
-    b       .L_check_size
-
-.L_sub3_and_finish:
-    sub     r3, r3, #3
-    b       .L_check_size
-
-.L_sub2_and_finish:
-    sub     r3, r3, #2
-
-.L_check_size:
-    pld     [r1, #0]
-    pld     [r1, #64]
-    ldr     r0, [sp]
-    cmp     r3, lr
-    bhs     __strcpy_chk_failed
-
-    // Add 1 for copy length to get the string terminator.
-    add     r2, r3, #1
-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)
-    .save   {r0, lr}
-    .cfi_def_cfa_offset 8
-    .cfi_rel_offset r0, 0
-    .cfi_rel_offset lr, 4
-
-    ldr     r0, error_message
-    ldr     r1, error_code
-1:
-    add     r0, pc
-    bl      __fortify_chk_fail
-error_code:
-    .word   BIONIC_EVENT_STRCPY_BUFFER_OVERFLOW
-error_message:
-    .word   error_string-(1b+4)
-END(__strcpy_chk_failed)
-
-    .data
-error_string:
-    .string "strcpy: prevented write past end of buffer"

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

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2013 The Android Open Source Project
+ * Copyright (C) 2008 The Android Open Source Project
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -28,8 +28,8 @@
 
 /* Assumes neon instructions and a cache line size of 32 bytes. */
 
-#include <private/bionic_asm.h>
-#include <private/libc_events.h>
+#include <machine/cpu-features.h>
+#include <machine/asm.h>
 
 /*
  * This code assumes it is running on a processor that supports all arm v7
@@ -38,49 +38,109 @@
  */
 
         .text
-        .syntax unified
         .fpu    neon
-        .thumb
-        .thumb_func
 
-ENTRY(__memcpy_chk)
-        cmp         r2, r3
-        bhi         __memcpy_chk_fail
-
-        // Fall through to memcpy...
-END(__memcpy_chk)
+#define CACHE_LINE_SIZE     32
 
 ENTRY(memcpy)
-        pld     [r1, #64]
-        stmfd   sp!, {r0, lr}
-        .save   {r0, lr}
-        .cfi_def_cfa_offset 8
-        .cfi_rel_offset r0, 0
-        .cfi_rel_offset lr, 4
+        .save       {r0, lr}
+        /* start preloading as early as possible */
+        pld         [r1, #(CACHE_LINE_SIZE*0)]
+        stmfd       sp!, {r0, lr}
+        pld         [r1, #(CACHE_LINE_SIZE*2)]
+
+        /* do we have at least 16-bytes to copy (needed for alignment below) */
+        cmp         r2, #16
+        blo         5f
+
+        /* align destination to cache-line for the write-buffer */
+        rsb         r3, r0, #0
+        ands        r3, r3, #0xF
+        beq         0f
+
+        /* 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:
+
+0:      /* preload immediately the next cache line, which we may need */
+        pld         [r1, #(CACHE_LINE_SIZE*0)]
+        pld         [r1, #(CACHE_LINE_SIZE*2)]
+
+        /* make sure we have at least 64 bytes to copy */
+        subs        r2, r2, #64
+        blo         2f
+
+        /* Preload all the cache lines we need.
+         * NOTE: The number of pld below depends on CACHE_LINE_SIZE,
+         * ideally we 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.
+         * NOTE: The value CACHE_LINE_SIZE * 8 was chosen through
+         * experimentation.
+         */
+        pld         [r1, #(CACHE_LINE_SIZE*4)]
+        pld         [r1, #(CACHE_LINE_SIZE*6)]
+        pld         [r1, #(CACHE_LINE_SIZE*8)]
+
+1:      /* The main loop copies 64 bytes at a time */
+        vld1.8      {d0  - d3},   [r1]!
+        vld1.8      {d4  - d7},   [r1]!
+        pld         [r1, #(CACHE_LINE_SIZE*8)]
+        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
+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
 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}
-        .save   {lr}
-        .cfi_def_cfa_offset 4
-        .cfi_rel_offset lr, 0
-
-        ldr     r0, error_message
-        ldr     r1, error_code
-1:
-        add     r0, pc
-        bl      __fortify_chk_fail
-error_code:
-        .word   BIONIC_EVENT_MEMCPY_BUFFER_OVERFLOW
-error_message:
-        .word   error_string-(1b+4)
-END(__memcpy_chk_fail)
-
-        .data
-error_string:
-        .string     "memcpy: prevented write past end of buffer"

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

@@ -1,124 +0,0 @@
-/*
- * 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.
- */
-
-
-/*
- * 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.
- */
-
-// Assumes neon instructions and a cache line size of 32 bytes.
-
-ENTRY_PRIVATE(MEMCPY_BASE)
-        .save {r0, lr}
-        .cfi_def_cfa_offset 8
-        .cfi_rel_offset r0, 0
-        .cfi_rel_offset lr, 4
-
-        /* do we have at least 16-bytes to copy (needed for alignment below) */
-        cmp         r2, #16
-        blo         5f
-
-        /* 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
-        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
-        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:      /* 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, #(32*8)]
-        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         4f
-
-        /* Copy 32 bytes. These cache lines were already preloaded */
-        vld1.8      {d0 - d3},  [r1]!
-        sub         r2, r2, #32
-        vst1.8      {d0 - d3},  [r0, :128]!
-
-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
-        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)

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

@@ -27,8 +27,7 @@
  */
 
 #include <machine/cpu-features.h>
-#include <private/bionic_asm.h>
-#include <private/libc_events.h>
+#include <machine/asm.h>
 
 /*
  * This code assumes it is running on a processor that supports all arm v7
@@ -38,41 +37,15 @@
 
     .fpu    neon
 
-ENTRY(__memset_chk)
-        cmp         r2, r3
-        bls         .L_done
-
-        // Preserve lr for backtrace.
-        .save       {lr}
-        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)
         .save       {r0}
         stmfd       sp!, {r0}
-        .cfi_def_cfa_offset 4
-        .cfi_rel_offset r0, 0
 
         vdup.8      q0, r1
 
@@ -106,7 +79,3 @@ ENTRY(memset)
         ldmfd       sp!, {r0}
         bx          lr
 END(memset)
-
-        .data
-error_string:
-        .string     "memset: prevented write past end of buffer"

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

@@ -27,7 +27,7 @@
  */
 
 #include <machine/cpu-features.h>
-#include <private/bionic_asm.h>
+#include <machine/asm.h>
 
 #ifdef __ARMEB__
 #define S2LOMEM lsl
@@ -123,13 +123,8 @@ ENTRY(strcmp)
         .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 */
@@ -145,8 +140,19 @@ ENTRY(strcmp)
 
         .macro  magic_find_zero_bytes w1
         /* Macro to find all-zero bytes in w1, result is in ip.  */
+#if (defined (__ARM_FEATURE_DSP))
         uadd8   ip, \w1, r6
         sel     ip, r7, r6
+#else /* not defined (__ARM_FEATURE_DSP) */
+        /* __ARM_FEATURE_DSP is not defined for some Cortex-M processors.
+        Coincidently, these processors only have Thumb-2 mode, where we can use the
+        the (large) magic constant available directly as an immediate in instructions.
+        Note that we cannot use the magic constant in ARM mode, where we need
+        to create the constant in a register.  */
+        sub     ip, \w1, #0x01010101
+        bic     ip, ip, \w1
+        and     ip, ip, #0x80808080
+#endif /* not defined (__ARM_FEATURE_DSP) */
         .endm /* magic_find_zero_bytes */
 
         .macro  setup_return w1 w2
@@ -165,13 +171,12 @@ ENTRY(strcmp)
         /* Are both strings double-word aligned?  */
         orr     ip, r0, r1
         tst     ip, #7
-        bne     .L_do_align
+        bne     do_align
 
         /* Fast path.  */
-        .save   {r4-r7}
         init
 
-.L_doubleword_aligned:
+doubleword_aligned:
 
         /* Get here when the strings to compare are double-word aligned.  */
         /* Compare two words in every iteration.  */
@@ -184,14 +189,14 @@ ENTRY(strcmp)
         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
+        magic_compare_and_branch w1=r2, w2=r4, label=return_24
+        magic_compare_and_branch w1=r3, w2=r5, label=return_35
         b       2b
 
-.L_do_align:
+do_align:
         /* Is the first string word-aligned?  */
         ands    ip, r0, #3
-        beq     .L_word_aligned_r0
+        beq     word_aligned_r0
 
         /* Fast compare byte by byte until the first string is word-aligned.  */
         /* The offset of r0 from a word boundary is in ip. Thus, the number of bytes
@@ -199,58 +204,58 @@ ENTRY(strcmp)
         bic     r0, r0, #3
         ldr     r2, [r0], #4
         lsls    ip, ip, #31
-        beq     .L_byte2
-        bcs     .L_byte3
+        beq     byte2
+        bcs     byte3
 
-.L_byte1:
+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
+        bne     fast_return
+        m_cbz   reg=r3, label=fast_return
 
-.L_byte2:
+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
+        bne     fast_return
+        m_cbz   reg=r3, label=fast_return
 
-.L_byte3:
+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
+        bne     fast_return
+        m_cbnz  reg=r3, label=word_aligned_r0
 
-.L_fast_return:
+fast_return:
         mov     r0, ip
         bx      lr
 
-.L_word_aligned_r0:
+word_aligned_r0:
         init
         /* The first string is word-aligned.  */
         /* Is the second string word-aligned?  */
         ands    ip, r1, #3
-        bne     .L_strcmp_unaligned
+        bne     strcmp_unaligned
 
-.L_word_aligned:
+word_aligned:
         /* The strings are word-aligned. */
         /* Is the first string double-word aligned?  */
         tst     r0, #4
-        beq     .L_doubleword_aligned_r0
+        beq     doubleword_aligned_r0
 
         /* If r0 is not double-word aligned yet, align it by loading
         and comparing the next word from each string.  */
         ldr     r2, [r0], #4
         ldr     r4, [r1], #4
-        magic_compare_and_branch w1=r2 w2=r4 label=.L_return_24
+        magic_compare_and_branch w1=r2 w2=r4 label=return_24
 
-.L_doubleword_aligned_r0:
+doubleword_aligned_r0:
         /* Get here when r0 is double-word aligned.  */
         /* Is r1 doubleword_aligned?  */
         tst     r1, #4
-        beq     .L_doubleword_aligned
+        beq     doubleword_aligned
 
         /* Get here when the strings to compare are word-aligned,
         r0 is double-word aligned, but r1 is not double-word aligned.  */
@@ -266,9 +271,9 @@ ENTRY(strcmp)
 
         /* 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
+        magic_compare_and_branch w1=r2 w2=r5 label=return_25
         ldrd    r4, r5, [r1], #8
-        magic_compare_and_branch w1=r3 w2=r4 label=.L_return_34
+        magic_compare_and_branch w1=r3 w2=r4 label=return_34
         b       3b
 
         .macro miscmp_word offsetlo offsethi
@@ -292,46 +297,46 @@ ENTRY(strcmp)
         and     r2, r3, r6, S2LOMEM #\offsetlo
         it      eq
         cmpeq   r2, r5
-        bne     .L_return_25
+        bne     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
+        bne     return_32
         b       7b
         .endm /* miscmp_word */
 
-.L_strcmp_unaligned:
+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.  */
+        beq     miscmp_word_16                    /* If ip == 2.  */
+        bge     miscmp_word_24                    /* If ip == 3.  */
         miscmp_word offsetlo=8 offsethi=24        /* If ip == 1.  */
-.L_miscmp_word_24:  miscmp_word offsetlo=24 offsethi=8
+miscmp_word_24:  miscmp_word offsetlo=24 offsethi=8
 
 
-.L_return_32:
+return_32:
         setup_return w1=r3, w2=r2
-        b       .L_do_return
-.L_return_34:
+        b       do_return
+return_34:
         setup_return w1=r3, w2=r4
-        b       .L_do_return
-.L_return_25:
+        b       do_return
+return_25:
         setup_return w1=r2, w2=r5
-        b       .L_do_return
-.L_return_35:
+        b       do_return
+return_35:
         setup_return w1=r3, w2=r5
-        b       .L_do_return
-.L_return_24:
+        b       do_return
+return_24:
         setup_return w1=r2, w2=r4
 
-.L_do_return:
+do_return:
 
 #ifdef __ARMEB__
         mov     r0, ip
@@ -343,16 +348,11 @@ ENTRY(strcmp)
         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
+        m_cbz   reg=r0, label=compute_return_value
         clz     r0, r0
         /* r0 contains the number of bits on the left of the first all-zero byte in r1.  */
         rsb     r0, r0, #24
@@ -360,7 +360,7 @@ ENTRY(strcmp)
         lsr     r1, r1, r0
         lsr     r2, r2, r0
 
-.L_compute_return_value:
+compute_return_value:
         movs    r0, #1
         cmp     r1, r2
         /* The return value is computed as follows.
@@ -380,7 +380,7 @@ ENTRY(strcmp)
      * previous version. See bionic/libc/arch-arm/cortex-a15/bionic/strcmp.S
      * for the unedited version of this code.
      */
-.L_miscmp_word_16:
+miscmp_word_16:
 	wp1 .req r0
 	wp2 .req r1
 	b1  .req r2
@@ -453,11 +453,6 @@ ENTRY(strcmp)
     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
 
@@ -477,11 +472,6 @@ ENTRY(strcmp)
     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
 END(strcmp)

libc/arch-arm/krait/krait.mk

@@ -1,13 +1,5 @@
-libc_bionic_src_files_arm += \
-    arch-arm/krait/bionic/memcpy.S \
-    arch-arm/krait/bionic/memset.S \
-    arch-arm/krait/bionic/strcmp.S \
-    arch-arm/krait/bionic/__strcat_chk.S \
-    arch-arm/krait/bionic/__strcpy_chk.S \
+$(call libc-add-cpu-variant-src,MEMCPY,arch-arm/krait/bionic/memcpy.S)
+$(call libc-add-cpu-variant-src,MEMSET,arch-arm/krait/bionic/memset.S)
+$(call libc-add-cpu-variant-src,STRCMP,arch-arm/krait/bionic/strcmp.S)
 
-# Use cortex-a15 versions of strcat/strcpy/strlen and standard memmove
-libc_bionic_src_files_arm += \
-    arch-arm/cortex-a15/bionic/strcat.S \
-    arch-arm/cortex-a15/bionic/strcpy.S \
-    arch-arm/cortex-a15/bionic/strlen.S \
-    bionic/memmove.c \
+include bionic/libc/arch-arm/generic/generic.mk