Merge remote-tracking branch 'aosp/master' into HEAD

benchmarks/Android.mk

@@ -20,12 +20,29 @@ LOCAL_PATH := $(call my-dir)
 # Benchmarks library, usable by projects outside this directory.
 # -----------------------------------------------------------------------------
 
+benchmark_cflags := \
+    -O2 \
+    -fno-builtin \
+    -Wall \
+    -Wextra \
+    -Werror \
+    -Wunused \
+
+benchmark_cppflags := \
+    -std=gnu++11 \
+
+benchmarklib_src_files := \
+    Benchmark.cpp \
+    utils.cpp \
+    main.cpp \
+
 include $(CLEAR_VARS)
 LOCAL_MODULE := libbenchmark
-LOCAL_CFLAGS += -O2 -Wall -Wextra -Werror
+LOCAL_CFLAGS := $(benchmark_cflags)
-LOCAL_SRC_FILES := benchmark_main.cpp
+LOCAL_CPPFLAGS := $(benchmark_cppflags)
-LOCAL_C_INCLUDES := $(LOCAL_PATH)/include
+LOCAL_SRC_FILES := $(benchmarklib_src_files)
-LOCAL_EXPORT_C_INCLUDE_DIRS := $(LOCAL_PATH)/include
+LOCAL_C_INCLUDES := $(benchmark_c_includes)
+LOCAL_STATIC_LIBRARIES := libbase
 include $(BUILD_STATIC_LIBRARY)
 
 # Only supported on linux systems.
@@ -33,11 +50,12 @@ ifeq ($(HOST_OS),linux)
 
 include $(CLEAR_VARS)
 LOCAL_MODULE := libbenchmark
-LOCAL_CFLAGS += -O2 -Wall -Wextra -Werror
+LOCAL_CFLAGS := $(benchmark_cflags)
-LOCAL_SRC_FILES := benchmark_main.cpp
+LOCAL_CPPFLAGS := $(benchmark_cppflags)
-LOCAL_C_INCLUDES := $(LOCAL_PATH)/include
+LOCAL_SRC_FILES := $(benchmarklib_src_files)
-LOCAL_EXPORT_C_INCLUDE_DIRS := $(LOCAL_PATH)/include
+LOCAL_C_INCLUDES := $(benchmark_c_includes)
 LOCAL_MULTILIB := both
+LOCAL_STATIC_LIBRARIES := libbase
 include $(BUILD_HOST_STATIC_LIBRARY)
 
 endif
@@ -45,16 +63,9 @@ endif
 # -----------------------------------------------------------------------------
 # Benchmarks.
 # -----------------------------------------------------------------------------
-
+benchmark_src_files := \
-benchmark_c_flags = \
-    -O2 \
-    -Wall -Wextra -Wunused \
-    -Werror \
-    -fno-builtin \
-    -std=gnu++11 \
-
-benchmark_src_files = \
     math_benchmark.cpp \
+    property_benchmark.cpp \
     pthread_benchmark.cpp \
     semaphore_benchmark.cpp \
     stdio_benchmark.cpp \
@@ -70,9 +81,10 @@ LOCAL_MODULE := bionic-benchmarks
 LOCAL_MODULE_STEM_32 := bionic-benchmarks32
 LOCAL_MODULE_STEM_64 := bionic-benchmarks64
 LOCAL_MULTILIB := both
-LOCAL_CFLAGS += $(benchmark_c_flags)
+LOCAL_CFLAGS := $(benchmark_cflags)
-LOCAL_SRC_FILES := $(benchmark_src_files) property_benchmark.cpp
+LOCAL_CPPFLAGS := $(benchmark_cppflags)
-LOCAL_STATIC_LIBRARIES += libbenchmark
+LOCAL_SRC_FILES := $(benchmark_src_files)
+LOCAL_STATIC_LIBRARIES := libbenchmark libbase
 include $(BUILD_EXECUTABLE)
 
 # We don't build a static benchmark executable because it's not usually
@@ -90,10 +102,11 @@ LOCAL_MODULE := bionic-benchmarks-glibc
 LOCAL_MODULE_STEM_32 := bionic-benchmarks-glibc32
 LOCAL_MODULE_STEM_64 := bionic-benchmarks-glibc64
 LOCAL_MULTILIB := both
-LOCAL_CFLAGS += $(benchmark_c_flags)
+LOCAL_CFLAGS := $(benchmark_cflags)
-LOCAL_LDFLAGS += -lrt
+LOCAL_CPPFLAGS := $(benchmark_cppflags)
+LOCAL_LDFLAGS := -lrt
 LOCAL_SRC_FILES := $(benchmark_src_files)
-LOCAL_STATIC_LIBRARIES += libbenchmark
+LOCAL_STATIC_LIBRARIES := libbenchmark libbase
 include $(BUILD_HOST_EXECUTABLE)
 
 endif

benchmarks/Benchmark.cpp

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

benchmarks/benchmark/Benchmark.h

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

benchmarks/benchmark_main.cpp

@@ -1,266 +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 <time.h>
-
-#include <string>
-#include <vector>
-
-#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;
-static int g_name_column_width = 20;
-
-typedef std::vector<::testing::Benchmark*> BenchmarkList;
-
-static BenchmarkList& Benchmarks() {
-  static BenchmarkList benchmarks;
-  return benchmarks;
-}
-
-// Similar to the code in art, but supporting both binary and decimal prefixes.
-static std::string PrettyInt(uint64_t count, size_t base) {
-  if (base != 2 && base != 10) abort();
-
-  // The byte thresholds at which we display amounts. A count is displayed
-  // in unit U when kUnitThresholds[U] <= bytes < kUnitThresholds[U+1].
-  static const uint64_t kUnitThresholds2[] = {
-    1024*1024*1024 /* Gi */, 2*1024*1024 /* Mi */, 3*1024 /* Ki */, 0,
-  };
-  static const uint64_t kUnitThresholds10[] = {
-    1000*1000*1000 /* G */, 2*1000*1000 /* M */, 3*1000 /* k */, 0,
-  };
-  static const uint64_t kAmountPerUnit2[] = { 1024*1024*1024, 1024*1024, 1024, 1 };
-  static const uint64_t kAmountPerUnit10[] = { 1000*1000*1000, 1000*1000, 1000, 1 };
-  static const char* const kUnitStrings2[] = { "Gi", "Mi", "Ki", "" };
-  static const char* const kUnitStrings10[] = { "G", "M", "k", "" };
-
-  // Which set are we using?
-  const uint64_t* kUnitThresholds = ((base == 2) ? kUnitThresholds2 : kUnitThresholds10);
-  const uint64_t* kAmountPerUnit = ((base == 2) ? kAmountPerUnit2 : kAmountPerUnit10);
-  const char* const* kUnitStrings = ((base == 2) ? kUnitStrings2 : kUnitStrings10);
-
-  size_t i = 0;
-  for (; kUnitThresholds[i] != 0; ++i) {
-    if (count >= kUnitThresholds[i]) {
-      break;
-    }
-  }
-  char* s = NULL;
-  asprintf(&s, "%" PRId64 "%s", count / kAmountPerUnit[i], kUnitStrings[i]);
-  std::string result(s);
-  free(s);
-  return result;
-}
-
-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);
-  }
-
-  Benchmarks().push_back(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;
-  int64_t realStartTime = NanoTime();
-  RunRepeatedlyWithArg(iterations, arg);
-  int64_t realTotalTime = NanoTime() - realStartTime;
-  while (realTotalTime < 1e9 && iterations < 1e8) {
-    int last = iterations;
-    if (realTotalTime/iterations == 0) {
-      iterations = 1e9;
-    } else {
-      iterations = 1e9 / (realTotalTime/iterations);
-    }
-    iterations = std::max(last + 1, std::min(iterations + iterations/2, 100*last));
-    iterations = Round(iterations);
-    realStartTime = NanoTime();
-    RunRepeatedlyWithArg(iterations, arg);
-    realTotalTime = NanoTime() - realStartTime;
-  }
-
-  char throughput[100];
-  throughput[0] = '\0';
-
-  if (g_benchmark_total_time_ns > 0 && g_bytes_processed > 0) {
-    double gib_processed = static_cast<double>(g_bytes_processed)/1e9;
-    double seconds = static_cast<double>(g_benchmark_total_time_ns)/1e9;
-    snprintf(throughput, sizeof(throughput), " %8.3f GiB/s", gib_processed/seconds);
-  }
-
-  char full_name[100];
-  if (fn_range_ != NULL) {
-    snprintf(full_name, sizeof(full_name), "%s/%s", name_, PrettyInt(arg, 2).c_str());
-  } else {
-    snprintf(full_name, sizeof(full_name), "%s", name_);
-  }
-
-  printf("%-*s %10s %10" PRId64 "%s\n",
-         g_name_column_width, full_name,
-         PrettyInt(iterations, 10).c_str(),
-         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 (Benchmarks().empty()) {
-    fprintf(stderr, "No benchmarks registered!\n");
-    exit(EXIT_FAILURE);
-  }
-
-  for (auto& b : Benchmarks()) {
-    int name_width = static_cast<int>(strlen(b->Name()));
-    g_name_column_width = std::max(g_name_column_width, name_width);
-  }
-
-  bool need_header = true;
-  for (auto& b : Benchmarks()) {
-    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 (auto& b : Benchmarks()) {
-      fprintf(stderr, "  %s\n", b->Name());
-    }
-    exit(EXIT_FAILURE);
-  }
-
-  return 0;
-}

benchmarks/include/benchmark.h

@@ -1,65 +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.
- */
-
-#ifndef BENCHMARKS_BENCHMARK_H_
-#define BENCHMARKS_BENCHMARK_H_
-
-#include <stdint.h>
-#include <vector>
-
-namespace testing {
-
-class Benchmark {
- public:
-  Benchmark(const char* name, void (*fn)(int)) {
-    Register(name, fn, NULL);
-  }
-
-  Benchmark(const char* name, void (*fn_range)(int, int)) {
-    Register(name, NULL, fn_range);
-  }
-
-  Benchmark* Arg(int x);
-
-  const char* Name();
-
-  bool ShouldRun(int argc, char* argv[]);
-  void Run();
-
- private:
-  const char* name_;
-
-  void (*fn_)(int);
-  void (*fn_range_)(int, int);
-
-  std::vector<int> args_;
-
-  void Register(const char* name, void (*fn)(int), void (*fn_range)(int, int));
-  void RunRepeatedlyWithArg(int iterations, int arg);
-  void RunWithArg(int arg);
-};
-
-}  // namespace testing
-
-void SetBenchmarkBytesProcessed(int64_t);
-void StopBenchmarkTiming();
-void StartBenchmarkTiming();
-
-#define BENCHMARK(f) \
-    static ::testing::Benchmark* _benchmark_##f __attribute__((unused)) = \
-        (new ::testing::Benchmark(#f, f))
-
-#endif

benchmarks/main.cpp

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

benchmarks/math_benchmark.cpp

@@ -14,16 +14,20 @@
  * limitations under the License.
  */
 
-#include "benchmark.h"
-
 #include <fenv.h>
 #include <math.h>
 
+#include <benchmark/Benchmark.h>
+
+#define AT_COMMON_VALS \
+    Arg(1234.0)->Arg(nan(""))->Arg(HUGE_VAL)->Arg(0.0)
+
 // Avoid optimization.
 volatile double d;
 volatile double v;
 
-static void BM_math_sqrt(int iters) {
+BENCHMARK_NO_ARG(BM_math_sqrt);
+void BM_math_sqrt::Run(int iters) {
   StartBenchmarkTiming();
 
   d = 0.0;
@@ -34,9 +38,9 @@ static void BM_math_sqrt(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_math_sqrt);
 
-static void BM_math_log10(int iters) {
+BENCHMARK_NO_ARG(BM_math_log10);
+void BM_math_log10::Run(int iters) {
   StartBenchmarkTiming();
 
   d = 0.0;
@@ -47,9 +51,9 @@ static void BM_math_log10(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_math_log10);
 
-static void BM_math_logb(int iters) {
+BENCHMARK_NO_ARG(BM_math_logb);
+void BM_math_logb::Run(int iters) {
   StartBenchmarkTiming();
 
   d = 0.0;
@@ -60,61 +64,22 @@ static void BM_math_logb(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_math_logb);
 
-static void BM_math_isinf_NORMAL(int iters) {
+BENCHMARK_WITH_ARG(BM_math_isinf, double)->AT_COMMON_VALS;
+void BM_math_isinf::Run(int iters, double value) {
   StartBenchmarkTiming();
 
   d = 0.0;
-  v = 1234.0; // FP_NORMAL
+  v = value;
   for (int i = 0; i < iters; ++i) {
     d += (isinf)(v);
   }
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_math_isinf_NORMAL);
 
-static void BM_math_isinf_NAN(int iters) {
+BENCHMARK_NO_ARG(BM_math_sin_fast);
-  StartBenchmarkTiming();
+void BM_math_sin_fast::Run(int iters) {
-
-  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;
@@ -124,9 +89,9 @@ static void BM_math_sin_fast(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_math_sin_fast);
 
-static void BM_math_sin_feupdateenv(int iters) {
+BENCHMARK_NO_ARG(BM_math_sin_feupdateenv);
+void BM_math_sin_feupdateenv::Run(int iters) {
   StartBenchmarkTiming();
 
   d = 1.0;
@@ -140,9 +105,9 @@ static void BM_math_sin_feupdateenv(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_math_sin_feupdateenv);
 
-static void BM_math_sin_fesetenv(int iters) {
+BENCHMARK_NO_ARG(BM_math_sin_fesetenv);
+void BM_math_sin_fesetenv::Run(int iters) {
   StartBenchmarkTiming();
 
   d = 1.0;
@@ -156,56 +121,16 @@ static void BM_math_sin_fesetenv(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_math_sin_fesetenv);
 
-static void BM_math_fpclassify_NORMAL(int iters) {
+BENCHMARK_WITH_ARG(BM_math_fpclassify, double)->AT_COMMON_VALS;
+void BM_math_fpclassify::Run(int iters, double value) {
   StartBenchmarkTiming();
 
   d = 0.0;
-  v = 1234.0; // FP_NORMAL
+  v = value;
   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

@@ -14,19 +14,21 @@
  * limitations under the License.
  */
 
-#include "benchmark.h"
 #include <errno.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <unistd.h>
 
+#include <string>
+
+#if defined(__BIONIC__)
+
 #define _REALLY_INCLUDE_SYS__SYSTEM_PROPERTIES_H_
 #include <sys/_system_properties.h>
 
-#include <vector>
+#include <benchmark/Benchmark.h>
-#include <string>
 
-extern void *__system_property_area__;
+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.
@@ -34,200 +36,198 @@ extern void *__system_property_area__;
     Arg(1)->Arg(4)->Arg(16)->Arg(64)->Arg(128)->Arg(256)->Arg(512)
 
 struct LocalPropertyTestState {
-    LocalPropertyTestState(int nprops) : nprops(nprops), valid(false) {
+  LocalPropertyTestState(int nprops) : nprops(nprops), valid(false) {
-        static const char prop_name_chars[] = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ-_.";
+    static const char prop_name_chars[] = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ-_.";
 
-        const char* android_data = getenv("ANDROID_DATA");
+    const char* android_data = getenv("ANDROID_DATA");
-        if (android_data == NULL) {
+    if (android_data == NULL) {
-          printf("ANDROID_DATA environment variable not set\n");
+      printf("ANDROID_DATA environment variable not set\n");
-          return;
+      return;
-        }
+    }
-        char dir_template[PATH_MAX];
+    char dir_template[PATH_MAX];
-        snprintf(dir_template, sizeof(dir_template), "%s/local/tmp/prop-XXXXXX", android_data);
+    snprintf(dir_template, sizeof(dir_template), "%s/local/tmp/prop-XXXXXX", android_data);
-        char *dirname = mkdtemp(dir_template);
+    char* dirname = mkdtemp(dir_template);
-        if (!dirname) {
+    if (!dirname) {
-            printf("making temp file for test state failed (is %s/local/tmp writable?): %s\n",
+      printf("making temp file for test state failed (is %s/local/tmp writable?): %s\n",
-                   android_data, strerror(errno));
+             android_data, strerror(errno));
-            return;
+      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() {
+    old_pa = __system_property_area__;
-        if (!valid)
+    __system_property_area__ = NULL;
-            return;
 
-        __system_property_area__ = old_pa;
+    pa_dirname = dirname;
+    pa_filename = pa_dirname + "/__properties__";
 
-        __system_property_set_filename(PROP_FILENAME);
+    __system_property_set_filename(pa_filename.c_str());
-        unlink(pa_filename.c_str());
+    __system_property_area_init();
-        rmdir(pa_dirname.c_str());
 
-        for (int i = 0; i < nprops; i++) {
+    names = new char* [nprops];
-            delete names[i];
+    name_lens = new int[nprops];
-            delete values[i];
+    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--;
         }
-        delete[] names;
+        names[i][j] = prop_name_chars[random() % prop_name_len];
-        delete[] name_lens;
+      }
-        delete[] values;
+      names[i][name_lens[i]] = 0;
-        delete[] value_lens;
+
+      // 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;
+  const int nprops;
-    char **names;
+  char** names;
-    int *name_lens;
+  int* name_lens;
-    char **values;
+  char** values;
-    int *value_lens;
+  int* value_lens;
-    bool valid;
+  bool valid;
 
 private:
-    std::string pa_dirname;
+  std::string pa_dirname;
-    std::string pa_filename;
+  std::string pa_filename;
-    void *old_pa;
+  void* old_pa;
 };
 
-static void BM_property_get(int iters, int nprops)
+BENCHMARK_WITH_ARG(BM_property_get, int)->TEST_NUM_PROPS;
-{
+void BM_property_get::Run(int iters, int nprops) {
-    StopBenchmarkTiming();
+  StopBenchmarkTiming();
 
-    LocalPropertyTestState pa(nprops);
+  LocalPropertyTestState pa(nprops);
-    char value[PROP_VALUE_MAX];
+  char value[PROP_VALUE_MAX];
 
-    if (!pa.valid)
+  if (!pa.valid)
-        return;
+    return;
 
-    srandom(iters * nprops);
+  srandom(iters * nprops);
 
-    StartBenchmarkTiming();
+  StartBenchmarkTiming();
 
-    for (int i = 0; i < iters; i++) {
+  for (int i = 0; i < iters; i++) {
-        __system_property_get(pa.names[random() % nprops], value);
+    __system_property_get(pa.names[random() % nprops], value);
-    }
+  }
-    StopBenchmarkTiming();
+  StopBenchmarkTiming();
 }
-BENCHMARK(BM_property_get)->TEST_NUM_PROPS;
 
-static void BM_property_find(int iters, int nprops)
+BENCHMARK_WITH_ARG(BM_property_find, int)->TEST_NUM_PROPS;
-{
+void BM_property_find::Run(int iters, int nprops) {
-    StopBenchmarkTiming();
+  StopBenchmarkTiming();
 
-    LocalPropertyTestState pa(nprops);
+  LocalPropertyTestState pa(nprops);
 
-    if (!pa.valid)
+  if (!pa.valid)
-        return;
+    return;
 
-    srandom(iters * nprops);
+  srandom(iters * nprops);
 
-    StartBenchmarkTiming();
+  StartBenchmarkTiming();
 
-    for (int i = 0; i < iters; i++) {
+  for (int i = 0; i < iters; i++) {
-        __system_property_find(pa.names[random() % nprops]);
+    __system_property_find(pa.names[random() % nprops]);
-    }
+  }
-    StopBenchmarkTiming();
+  StopBenchmarkTiming();
 }
-BENCHMARK(BM_property_find)->TEST_NUM_PROPS;
 
-static void BM_property_read(int iters, int nprops)
+BENCHMARK_WITH_ARG(BM_property_read, int)->TEST_NUM_PROPS;
-{
+void BM_property_read::Run(int iters, int nprops) {
-    StopBenchmarkTiming();
+  StopBenchmarkTiming();
 
-    LocalPropertyTestState pa(nprops);
+  LocalPropertyTestState pa(nprops);
 
-    if (!pa.valid)
+  if (!pa.valid)
-        return;
+    return;
 
-    srandom(iters * nprops);
+  srandom(iters * nprops);
-    const prop_info** pinfo = new const prop_info*[iters];
+  const prop_info** pinfo = new const prop_info*[iters];
-    char propvalue[PROP_VALUE_MAX];
+  char propvalue[PROP_VALUE_MAX];
 
-    for (int i = 0; i < iters; i++) {
+  for (int i = 0; i < iters; i++) {
-        pinfo[i] = __system_property_find(pa.names[random() % nprops]);
+    pinfo[i] = __system_property_find(pa.names[random() % nprops]);
-    }
+  }
 
-    StartBenchmarkTiming();
+  StartBenchmarkTiming();
-    for (int i = 0; i < iters; i++) {
+  for (int i = 0; i < iters; i++) {
-        __system_property_read(pinfo[i], 0, propvalue);
+    __system_property_read(pinfo[i], 0, propvalue);
-    }
+  }
-    StopBenchmarkTiming();
+  StopBenchmarkTiming();
 
-    delete[] pinfo;
+  delete[] pinfo;
 }
-BENCHMARK(BM_property_read)->TEST_NUM_PROPS;
 
-static void BM_property_serial(int iters, int nprops)
+BENCHMARK_WITH_ARG(BM_property_serial, int)->TEST_NUM_PROPS;
-{
+void BM_property_serial::Run(int iters, int nprops) {
-    StopBenchmarkTiming();
+  StopBenchmarkTiming();
 
-    LocalPropertyTestState pa(nprops);
+  LocalPropertyTestState pa(nprops);
 
-    if (!pa.valid)
+  if (!pa.valid)
-        return;
+    return;
 
-    srandom(iters * nprops);
+  srandom(iters * nprops);
-    const prop_info** pinfo = new const prop_info*[iters];
+  const prop_info** pinfo = new const prop_info*[iters];
 
-    for (int i = 0; i < iters; i++) {
+  for (int i = 0; i < iters; i++) {
-        pinfo[i] = __system_property_find(pa.names[random() % nprops]);
+    pinfo[i] = __system_property_find(pa.names[random() % nprops]);
-    }
+  }
 
-    StartBenchmarkTiming();
+  StartBenchmarkTiming();
-    for (int i = 0; i < iters; i++) {
+  for (int i = 0; i < iters; i++) {
-        __system_property_serial(pinfo[i]);
+    __system_property_serial(pinfo[i]);
-    }
+  }
-    StopBenchmarkTiming();
+  StopBenchmarkTiming();
 
-    delete[] pinfo;
+  delete[] pinfo;
 }
-BENCHMARK(BM_property_serial)->TEST_NUM_PROPS;
+
+#endif  // __BIONIC__

benchmarks/pthread_benchmark.cpp

@@ -14,14 +14,15 @@
  * limitations under the License.
  */
 
-#include "benchmark.h"
-
 #include <pthread.h>
 
+#include <benchmark/Benchmark.h>
+
 // Stop GCC optimizing out our pure function.
 /* Must not be static! */ pthread_t (*pthread_self_fp)() = pthread_self;
 
-static void BM_pthread_self(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_self);
+void BM_pthread_self::Run(int iters) {
   StartBenchmarkTiming();
 
   for (int i = 0; i < iters; ++i) {
@@ -30,9 +31,9 @@ static void BM_pthread_self(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_pthread_self);
 
-static void BM_pthread_getspecific(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_getspecific);
+void BM_pthread_getspecific::Run(int iters) {
   StopBenchmarkTiming();
   pthread_key_t key;
   pthread_key_create(&key, NULL);
@@ -45,9 +46,9 @@ static void BM_pthread_getspecific(int iters) {
   StopBenchmarkTiming();
   pthread_key_delete(key);
 }
-BENCHMARK(BM_pthread_getspecific);
 
-static void BM_pthread_setspecific(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_setspecific);
+void BM_pthread_setspecific::Run(int iters) {
   StopBenchmarkTiming();
   pthread_key_t key;
   pthread_key_create(&key, NULL);
@@ -60,12 +61,12 @@ static void BM_pthread_setspecific(int iters) {
   StopBenchmarkTiming();
   pthread_key_delete(key);
 }
-BENCHMARK(BM_pthread_setspecific);
 
 static void DummyPthreadOnceInitFunction() {
 }
 
-static void BM_pthread_once(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_once);
+void BM_pthread_once::Run(int iters) {
   StopBenchmarkTiming();
   pthread_once_t once = PTHREAD_ONCE_INIT;
   pthread_once(&once, DummyPthreadOnceInitFunction);
@@ -77,9 +78,9 @@ static void BM_pthread_once(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_pthread_once);
 
-static void BM_pthread_mutex_lock(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_mutex_lock);
+void BM_pthread_mutex_lock::Run(int iters) {
   StopBenchmarkTiming();
   pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
   StartBenchmarkTiming();
@@ -91,9 +92,9 @@ static void BM_pthread_mutex_lock(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_pthread_mutex_lock);
 
-static void BM_pthread_mutex_lock_ERRORCHECK(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_mutex_lock_ERRORCHECK);
+void BM_pthread_mutex_lock_ERRORCHECK::Run(int iters) {
   StopBenchmarkTiming();
   pthread_mutex_t mutex = PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP;
   StartBenchmarkTiming();
@@ -105,9 +106,9 @@ static void BM_pthread_mutex_lock_ERRORCHECK(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_pthread_mutex_lock_ERRORCHECK);
 
-static void BM_pthread_mutex_lock_RECURSIVE(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_mutex_lock_RECURSIVE);
+void BM_pthread_mutex_lock_RECURSIVE::Run(int iters) {
   StopBenchmarkTiming();
   pthread_mutex_t mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
   StartBenchmarkTiming();
@@ -119,9 +120,9 @@ static void BM_pthread_mutex_lock_RECURSIVE(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_pthread_mutex_lock_RECURSIVE);
 
-static void BM_pthread_rw_lock_read(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_rw_lock_read);
+void BM_pthread_rw_lock_read::Run(int iters) {
   StopBenchmarkTiming();
   pthread_rwlock_t lock;
   pthread_rwlock_init(&lock, NULL);
@@ -135,9 +136,9 @@ static void BM_pthread_rw_lock_read(int iters) {
   StopBenchmarkTiming();
   pthread_rwlock_destroy(&lock);
 }
-BENCHMARK(BM_pthread_rw_lock_read);
 
-static void BM_pthread_rw_lock_write(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_rw_lock_write);
+void BM_pthread_rw_lock_write::Run(int iters) {
   StopBenchmarkTiming();
   pthread_rwlock_t lock;
   pthread_rwlock_init(&lock, NULL);
@@ -151,13 +152,13 @@ static void BM_pthread_rw_lock_write(int iters) {
   StopBenchmarkTiming();
   pthread_rwlock_destroy(&lock);
 }
-BENCHMARK(BM_pthread_rw_lock_write);
 
 static void* IdleThread(void*) {
   return NULL;
 }
 
-static void BM_pthread_create(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_create);
+void BM_pthread_create::Run(int iters) {
   StopBenchmarkTiming();
   pthread_t thread;
 
@@ -168,43 +169,45 @@ static void BM_pthread_create(int iters) {
     pthread_join(thread, NULL);
   }
 }
-BENCHMARK(BM_pthread_create);
 
-static void* RunThread(void*) {
+static void* RunThread(void* arg) {
-  StopBenchmarkTiming();
+  ::testing::Benchmark* benchmark = reinterpret_cast<::testing::Benchmark*>(arg);
+  benchmark->StopBenchmarkTiming();
   return NULL;
 }
 
-static void BM_pthread_create_and_run(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_create_and_run);
+void BM_pthread_create_and_run::Run(int iters) {
   StopBenchmarkTiming();
   pthread_t thread;
 
   for (int i = 0; i < iters; ++i) {
     StartBenchmarkTiming();
-    pthread_create(&thread, NULL, RunThread, NULL);
+    pthread_create(&thread, NULL, RunThread, this);
     pthread_join(thread, NULL);
   }
 }
-BENCHMARK(BM_pthread_create_and_run);
 
-static void* ExitThread(void*) {
+static void* ExitThread(void* arg) {
-  StartBenchmarkTiming();
+  ::testing::Benchmark* benchmark = reinterpret_cast<::testing::Benchmark*>(arg);
+  benchmark->StartBenchmarkTiming();
   pthread_exit(NULL);
 }
 
-static void BM_pthread_exit_and_join(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_exit_and_join);
+void BM_pthread_exit_and_join::Run(int iters) {
   StopBenchmarkTiming();
   pthread_t thread;
 
   for (int i = 0; i < iters; ++i) {
-    pthread_create(&thread, NULL, ExitThread, NULL);
+    pthread_create(&thread, NULL, ExitThread, this);
     pthread_join(thread, NULL);
     StopBenchmarkTiming();
   }
 }
-BENCHMARK(BM_pthread_exit_and_join);
 
-static void BM_pthread_key_create(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_key_create);
+void BM_pthread_key_create::Run(int iters) {
   StopBenchmarkTiming();
   pthread_key_t key;
 
@@ -215,9 +218,9 @@ static void BM_pthread_key_create(int iters) {
     pthread_key_delete(key);
   }
 }
-BENCHMARK(BM_pthread_key_create);
 
-static void BM_pthread_key_delete(int iters) {
+BENCHMARK_NO_ARG(BM_pthread_key_delete);
+void BM_pthread_key_delete::Run(int iters) {
   StopBenchmarkTiming();
   pthread_key_t key;
 
@@ -228,4 +231,3 @@ static void BM_pthread_key_delete(int iters) {
     StopBenchmarkTiming();
   }
 }
-BENCHMARK(BM_pthread_key_delete);

benchmarks/semaphore_benchmark.cpp

@@ -14,14 +14,15 @@
  * limitations under the License.
  */
 
-#include "benchmark.h"
-
 #include <pthread.h>
 #include <semaphore.h>
 #include <stdatomic.h>
 #include <stdio.h>
 
-static void BM_semaphore_sem_getvalue(int iters) {
+#include <benchmark/Benchmark.h>
+
+BENCHMARK_NO_ARG(BM_semaphore_sem_getvalue);
+void BM_semaphore_sem_getvalue::Run(int iters) {
   StopBenchmarkTiming();
   sem_t semaphore;
   sem_init(&semaphore, 1, 1);
@@ -34,9 +35,9 @@ static void BM_semaphore_sem_getvalue(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_semaphore_sem_getvalue);
 
-static void BM_semaphore_sem_wait_sem_post(int iters) {
+BENCHMARK_NO_ARG(BM_semaphore_sem_wait_sem_post);
+void BM_semaphore_sem_wait_sem_post::Run(int iters) {
   StopBenchmarkTiming();
   sem_t semaphore;
   sem_init(&semaphore, 1, 1);
@@ -49,7 +50,6 @@ static void BM_semaphore_sem_wait_sem_post(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_semaphore_sem_wait_sem_post);
 
 /*
  *    This test reports the overhead of the underlying futex wake syscall on
@@ -87,7 +87,8 @@ static void *BM_semaphore_sem_post_start_thread(void *obj) {
     return NULL;
 }
 
-static void BM_semaphore_sem_post(int iters) {
+BENCHMARK_NO_ARG(BM_semaphore_sem_post);
+void BM_semaphore_sem_post::Run(int iters) {
   StopBenchmarkTiming();
 
   sem_t semaphore;
@@ -100,9 +101,6 @@ static void BM_semaphore_sem_post(int iters) {
   pthread_attr_setschedparam(&attr, &param);
   pthread_attr_setschedpolicy(&attr, SCHED_OTHER);
   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
-#ifdef PTHREAD_SET_INHERIT_SCHED
-  pthread_attr_setinheritsched(&attr, PTHREAD_EXPLICIT_SCHED);
-#endif
   pthread_t pthread;
   pthread_create(&pthread, &attr, BM_semaphore_sem_post_start_thread, &semaphore);
   pthread_attr_destroy(&attr);
@@ -143,99 +141,3 @@ static void BM_semaphore_sem_post(int iters) {
     sched_yield();
   } while (!BM_semaphore_sem_post_running);
 }
-BENCHMARK(BM_semaphore_sem_post);
-
-/*
- *    This test reports the overhead of sem_post to sem_wake. A circle of
- * num_semaphore - 1 threads are run on a set of semaphores to measure the
- * activity. One can calculate the sem_wake overhead alone by:
- *
- * BM_semaphore_sem_post_sem_wait - BM_semaphore_sem_post - BM_time_clock_gettime
- *
- * Differences will result if there are more threads than active processors,
- * there will be delay induced when scheduling the processes. This cost is
- * measured by trying different values of num_semaphore. The governor selected
- * will have a major impact on the results for a large number of threads.
- *
- *     To reduce the chances for threads racing ahead and not triggering the
- * futex, for example the background threads finish their job before the
- * sem_wait is hit in the main thread, the background threads will run at
- * batch priority and the main thread at fifo priority. This should generally
- * guarantee the main thread completes its task of priming itself with the
- * sem_wait before the other threads can start. In practice without the
- * sched mechanics here, this works on Android configured kernels, this is
- * insurance for wacky(tm) sched configurations.
- */
-static void *BM_semaphore_sem_post_sem_wait_start_thread(void *obj) {
-  sem_t *semaphore = reinterpret_cast<sem_t *>(obj);
-
-  while ((BM_semaphore_sem_post_running > 0) && !sem_wait(semaphore)) {
-    sem_post(semaphore + 1);
-  }
-  --BM_semaphore_sem_post_running;
-  return NULL;
-}
-
-static void BM_semaphore_sem_post_sem_wait_num(int iters, int num_semaphore) {
-  StopBenchmarkTiming();
-
-  sem_t semaphore[num_semaphore];
-
-  for (int i = 0; i < num_semaphore; ++i) {
-    sem_init(semaphore + i, 0, 0);
-  }
-
-  pthread_attr_t attr;
-  pthread_attr_init(&attr);
-  BM_semaphore_sem_post_running = 1;
-  struct sched_param param = { 0, };
-  pthread_attr_setschedparam(&attr, &param);
-  pthread_attr_setschedpolicy(&attr, SCHED_BATCH);
-  pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
-#ifdef PTHREAD_SET_INHERIT_SCHED
-  pthread_attr_setinheritsched(&attr, PTHREAD_EXPLICIT_SCHED);
-#endif
-  for (int i = 0; i < (num_semaphore - 1); ++i) {
-    pthread_t pthread;
-    pthread_create(&pthread, &attr, BM_semaphore_sem_post_sem_wait_start_thread, semaphore + i);
-  }
-  pthread_attr_destroy(&attr);
-  sched_yield();
-
-  param.sched_priority = 1;
-  sched_setscheduler((pid_t)0, SCHED_FIFO, &param);
-
-  StartBenchmarkTiming();
-
-  for (int i = 0; i < iters; i += num_semaphore) {
-    sem_post(semaphore);
-    sem_wait(semaphore + num_semaphore - 1);
-  }
-
-  StopBenchmarkTiming();
-
-  param.sched_priority = 0;
-  sched_setscheduler((pid_t)0, SCHED_OTHER, &param);
-
-  if (BM_semaphore_sem_post_running > 0) {
-    BM_semaphore_sem_post_running = 0;
-  }
-  for (int i = 0;
-       (i < (10 * num_semaphore)) && (BM_semaphore_sem_post_running > (1 - num_semaphore));
-       ++i) {
-    for (int j = 0; j < (num_semaphore - 1); ++j) {
-      sem_post(semaphore + j);
-    }
-    sched_yield();
-  }
-}
-
-static void BM_semaphore_sem_post_sem_wait_low(int iters) {
-    BM_semaphore_sem_post_sem_wait_num(iters, 2);
-}
-BENCHMARK(BM_semaphore_sem_post_sem_wait_low);
-
-static void BM_semaphore_sem_post_sem_wait_high(int iters) {
-    BM_semaphore_sem_post_sem_wait_num(iters, 100);
-}
-BENCHMARK(BM_semaphore_sem_post_sem_wait_high);

benchmarks/stdio_benchmark.cpp

@@ -14,11 +14,11 @@
  * limitations under the License.
  */
 
-#include "benchmark.h"
-
 #include <stdio.h>
 #include <stdio_ext.h>
 
+#include <benchmark/Benchmark.h>
+
 #define KB 1024
 #define MB 1024*KB
 
@@ -27,12 +27,12 @@
     Arg(1*KB)->Arg(4*KB)->Arg(8*KB)->Arg(16*KB)->Arg(64*KB)
 
 template <typename Fn>
-static void ReadWriteTest(int iters, int chunk_size, Fn f, bool buffered) {
+void ReadWriteTest(::testing::Benchmark* benchmark, int iters, int chunk_size, Fn f, bool buffered) {
-  StopBenchmarkTiming();
+  benchmark->StopBenchmarkTiming();
   FILE* fp = fopen("/dev/zero", "rw");
   __fsetlocking(fp, FSETLOCKING_BYCALLER);
   char* buf = new char[chunk_size];
-  StartBenchmarkTiming();
+  benchmark->StartBenchmarkTiming();
 
   if (!buffered) {
     setvbuf(fp, 0, _IONBF, 0);
@@ -42,31 +42,31 @@ static void ReadWriteTest(int iters, int chunk_size, Fn f, bool buffered) {
     f(buf, chunk_size, 1, fp);
   }
 
-  StopBenchmarkTiming();
+  benchmark->StopBenchmarkTiming();
-  SetBenchmarkBytesProcessed(int64_t(iters) * int64_t(chunk_size));
+  benchmark->SetBenchmarkBytesProcessed(int64_t(iters) * int64_t(chunk_size));
   delete[] buf;
   fclose(fp);
 }
 
-static void BM_stdio_fread(int iters, int chunk_size) {
+BENCHMARK_WITH_ARG(BM_stdio_fread, int)->AT_COMMON_SIZES;
-  ReadWriteTest(iters, chunk_size, fread, true);
+void BM_stdio_fread::Run(int iters, int chunk_size) {
+  ReadWriteTest(this, iters, chunk_size, fread, true);
 }
-BENCHMARK(BM_stdio_fread)->AT_COMMON_SIZES;
 
-static void BM_stdio_fwrite(int iters, int chunk_size) {
+BENCHMARK_WITH_ARG(BM_stdio_fwrite, int)->AT_COMMON_SIZES;
-  ReadWriteTest(iters, chunk_size, fwrite, true);
+void BM_stdio_fwrite::Run(int iters, int chunk_size) {
+  ReadWriteTest(this, iters, chunk_size, fwrite, true);
 }
-BENCHMARK(BM_stdio_fwrite)->AT_COMMON_SIZES;
 
-static void BM_stdio_fread_unbuffered(int iters, int chunk_size) {
+BENCHMARK_WITH_ARG(BM_stdio_fread_unbuffered, int)->AT_COMMON_SIZES;
-  ReadWriteTest(iters, chunk_size, fread, false);
+void BM_stdio_fread_unbuffered::Run(int iters, int chunk_size) {
+  ReadWriteTest(this, iters, chunk_size, fread, false);
 }
-BENCHMARK(BM_stdio_fread_unbuffered)->AT_COMMON_SIZES;
 
-static void BM_stdio_fwrite_unbuffered(int iters, int chunk_size) {
+BENCHMARK_WITH_ARG(BM_stdio_fwrite_unbuffered, int)->AT_COMMON_SIZES;
-  ReadWriteTest(iters, chunk_size, fwrite, false);
+void BM_stdio_fwrite_unbuffered::Run(int iters, int chunk_size) {
+  ReadWriteTest(this, iters, chunk_size, fwrite, false);
 }
-BENCHMARK(BM_stdio_fwrite_unbuffered)->AT_COMMON_SIZES;
 
 static void FopenFgetsFclose(int iters, bool no_locking) {
   char buf[1024];
@@ -78,12 +78,16 @@ static void FopenFgetsFclose(int iters, bool no_locking) {
   }
 }
 
-static void BM_stdio_fopen_fgets_fclose_locking(int iters) {
+BENCHMARK_NO_ARG(BM_stdio_fopen_fgets_fclose_locking);
+void BM_stdio_fopen_fgets_fclose_locking::Run(int iters) {
+  StartBenchmarkTiming();
   FopenFgetsFclose(iters, false);
+  StopBenchmarkTiming();
 }
-BENCHMARK(BM_stdio_fopen_fgets_fclose_locking);
 
-static void BM_stdio_fopen_fgets_fclose_no_locking(int iters) {
+BENCHMARK_NO_ARG(BM_stdio_fopen_fgets_fclose_no_locking);
+void BM_stdio_fopen_fgets_fclose_no_locking::Run(int iters) {
+  StartBenchmarkTiming();
   FopenFgetsFclose(iters, true);
+  StopBenchmarkTiming();
 }
-BENCHMARK(BM_stdio_fopen_fgets_fclose_no_locking);

benchmarks/string_benchmark.cpp

@@ -14,10 +14,11 @@
  * limitations under the License.
  */
 
-#include "benchmark.h"
+#include <stdint.h>
-
 #include <string.h>
 
+#include <benchmark/Benchmark.h>
+
 #define KB 1024
 #define MB 1024*KB
 
@@ -26,7 +27,8 @@
 
 // TODO: test unaligned operation too? (currently everything will be 8-byte aligned by malloc.)
 
-static void BM_string_memcmp(int iters, int nbytes) {
+BENCHMARK_WITH_ARG(BM_string_memcmp, int)->AT_COMMON_SIZES;
+void BM_string_memcmp::Run(int iters, int nbytes) {
   StopBenchmarkTiming();
   char* src = new char[nbytes]; char* dst = new char[nbytes];
   memset(src, 'x', nbytes);
@@ -39,13 +41,13 @@ static void BM_string_memcmp(int iters, int nbytes) {
   }
 
   StopBenchmarkTiming();
-  SetBenchmarkBytesProcessed(int64_t(iters) * int64_t(nbytes));
+  SetBenchmarkBytesProcessed(uint64_t(iters) * uint64_t(nbytes));
   delete[] src;
   delete[] dst;
 }
-BENCHMARK(BM_string_memcmp)->AT_COMMON_SIZES;
 
-static void BM_string_memcpy(int iters, int nbytes) {
+BENCHMARK_WITH_ARG(BM_string_memcpy, int)->AT_COMMON_SIZES;
+void BM_string_memcpy::Run(int iters, int nbytes) {
   StopBenchmarkTiming();
   char* src = new char[nbytes]; char* dst = new char[nbytes];
   memset(src, 'x', nbytes);
@@ -56,13 +58,13 @@ static void BM_string_memcpy(int iters, int nbytes) {
   }
 
   StopBenchmarkTiming();
-  SetBenchmarkBytesProcessed(int64_t(iters) * int64_t(nbytes));
+  SetBenchmarkBytesProcessed(uint64_t(iters) * uint64_t(nbytes));
   delete[] src;
   delete[] dst;
 }
-BENCHMARK(BM_string_memcpy)->AT_COMMON_SIZES;
 
-static void BM_string_memmove(int iters, int nbytes) {
+BENCHMARK_WITH_ARG(BM_string_memmove, int)->AT_COMMON_SIZES;
+void BM_string_memmove::Run(int iters, int nbytes) {
   StopBenchmarkTiming();
   char* buf = new char[nbytes + 64];
   memset(buf, 'x', nbytes + 64);
@@ -73,12 +75,12 @@ static void BM_string_memmove(int iters, int nbytes) {
   }
 
   StopBenchmarkTiming();
-  SetBenchmarkBytesProcessed(int64_t(iters) * int64_t(nbytes));
+  SetBenchmarkBytesProcessed(uint64_t(iters) * uint64_t(nbytes));
   delete[] buf;
 }
-BENCHMARK(BM_string_memmove)->AT_COMMON_SIZES;
 
-static void BM_string_memset(int iters, int nbytes) {
+BENCHMARK_WITH_ARG(BM_string_memset, int)->AT_COMMON_SIZES;
+void BM_string_memset::Run(int iters, int nbytes) {
   StopBenchmarkTiming();
   char* dst = new char[nbytes];
   StartBenchmarkTiming();
@@ -88,12 +90,12 @@ static void BM_string_memset(int iters, int nbytes) {
   }
 
   StopBenchmarkTiming();
-  SetBenchmarkBytesProcessed(int64_t(iters) * int64_t(nbytes));
+  SetBenchmarkBytesProcessed(uint64_t(iters) * uint64_t(nbytes));
   delete[] dst;
 }
-BENCHMARK(BM_string_memset)->AT_COMMON_SIZES;
 
-static void BM_string_strlen(int iters, int nbytes) {
+BENCHMARK_WITH_ARG(BM_string_strlen, int)->AT_COMMON_SIZES;
+void BM_string_strlen::Run(int iters, int nbytes) {
   StopBenchmarkTiming();
   char* s = new char[nbytes];
   memset(s, 'x', nbytes);
@@ -106,7 +108,6 @@ static void BM_string_strlen(int iters, int nbytes) {
   }
 
   StopBenchmarkTiming();
-  SetBenchmarkBytesProcessed(int64_t(iters) * int64_t(nbytes));
+  SetBenchmarkBytesProcessed(uint64_t(iters) * uint64_t(nbytes));
   delete[] s;
 }
-BENCHMARK(BM_string_strlen)->AT_COMMON_SIZES;

benchmarks/time_benchmark.cpp

@@ -14,14 +14,14 @@
  * limitations under the License.
  */
 
-#include "benchmark.h"
-
-#include <unistd.h>
 #include <sys/syscall.h>
 #include <sys/time.h>
 #include <time.h>
 
-static void BM_time_clock_gettime(int iters) {
+#include <benchmark/Benchmark.h>
+
+BENCHMARK_NO_ARG(BM_time_clock_gettime);
+void BM_time_clock_gettime::Run(int iters) {
   StartBenchmarkTiming();
 
   timespec t;
@@ -31,9 +31,9 @@ static void BM_time_clock_gettime(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_time_clock_gettime);
 
-static void BM_time_clock_gettime_syscall(int iters) {
+BENCHMARK_NO_ARG(BM_time_clock_gettime_syscall);
+void BM_time_clock_gettime_syscall::Run(int iters) {
   StartBenchmarkTiming();
 
   timespec t;
@@ -43,9 +43,9 @@ static void BM_time_clock_gettime_syscall(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_time_clock_gettime_syscall);
 
-static void BM_time_gettimeofday(int iters) {
+BENCHMARK_NO_ARG(BM_time_gettimeofday);
+void BM_time_gettimeofday::Run(int iters) {
   StartBenchmarkTiming();
 
   timeval tv;
@@ -55,9 +55,9 @@ static void BM_time_gettimeofday(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_time_gettimeofday);
 
-static void BM_time_gettimeofday_syscall(int iters) {
+BENCHMARK_NO_ARG(BM_time_gettimeofday_syscall);
+void BM_time_gettimeofday_syscall::Run(int iters) {
   StartBenchmarkTiming();
 
   timeval tv;
@@ -67,9 +67,9 @@ static void BM_time_gettimeofday_syscall(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_time_gettimeofday_syscall);
 
-static void BM_time_time(int iters) {
+BENCHMARK_NO_ARG(BM_time_time);
+void BM_time_time::Run(int iters) {
   StartBenchmarkTiming();
 
   for (int i = 0; i < iters; ++i) {
@@ -78,4 +78,3 @@ static void BM_time_time(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_time_time);

benchmarks/unistd_benchmark.cpp

@@ -14,12 +14,13 @@
  * limitations under the License.
  */
 
-#include "benchmark.h"
-
 #include <sys/syscall.h>
 #include <unistd.h>
 
-static void BM_unistd_getpid(int iters) {
+#include <benchmark/Benchmark.h>
+
+BENCHMARK_NO_ARG(BM_unistd_getpid);
+void BM_unistd_getpid::Run(int iters) {
   StartBenchmarkTiming();
 
   for (int i = 0; i < iters; ++i) {
@@ -28,9 +29,9 @@ static void BM_unistd_getpid(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_unistd_getpid);
 
-static void BM_unistd_getpid_syscall(int iters) {
+BENCHMARK_NO_ARG(BM_unistd_getpid_syscall);
+void BM_unistd_getpid_syscall::Run(int iters) {
   StartBenchmarkTiming();
 
   for (int i = 0; i < iters; ++i) {
@@ -39,14 +40,14 @@ static void BM_unistd_getpid_syscall(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_unistd_getpid_syscall);
 
 #if defined(__BIONIC__)
 
 // Stop GCC optimizing out our pure function.
 /* Must not be static! */ pid_t (*gettid_fp)() = gettid;
 
-static void BM_unistd_gettid(int iters) {
+BENCHMARK_NO_ARG(BM_unistd_gettid);
+void BM_unistd_gettid::Run(int iters) {
   StartBenchmarkTiming();
 
   for (int i = 0; i < iters; ++i) {
@@ -55,11 +56,11 @@ static void BM_unistd_gettid(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_unistd_gettid);
 
 #endif
 
-static void BM_unistd_gettid_syscall(int iters) {
+BENCHMARK_NO_ARG(BM_unistd_gettid_syscall);
+void BM_unistd_gettid_syscall::Run(int iters) {
   StartBenchmarkTiming();
 
   for (int i = 0; i < iters; ++i) {
@@ -68,4 +69,3 @@ static void BM_unistd_gettid_syscall(int iters) {
 
   StopBenchmarkTiming();
 }
-BENCHMARK(BM_unistd_gettid_syscall);

benchmarks/utils.cpp

@@ -0,0 +1,81 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <inttypes.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <stdlib.h>
+
+#include <string>
+
+#include "utils.h"
+
+int Round(int n) {
+  int base = 1;
+  while (base*10 < n) {
+    base *= 10;
+  }
+  if (n < 2*base) {
+    return 2*base;
+  }
+  if (n < 5*base) {
+    return 5*base;
+  }
+  return 10*base;
+}
+
+// Similar to the code in art, but supporting both binary and decimal prefixes.
+std::string PrettyInt(long value, size_t base) {
+  if (base != 2 && base != 10) abort();
+
+  uint64_t count = static_cast<uint64_t>(value);
+  bool negative_number = false;
+  if (value < 0) {
+    negative_number = true;
+    count = static_cast<uint64_t>(-value);
+  }
+
+  // The byte thresholds at which we display amounts. A count is displayed
+  // in unit U when kUnitThresholds[U] <= bytes < kUnitThresholds[U+1].
+  static const uint64_t kUnitThresholds2[] = {
+    1024*1024*1024 /* Gi */, 2*1024*1024 /* Mi */, 3*1024 /* Ki */, 0,
+  };
+  static const uint64_t kUnitThresholds10[] = {
+    1000*1000*1000 /* G */, 2*1000*1000 /* M */, 3*1000 /* k */, 0,
+  };
+  static const uint64_t kAmountPerUnit2[] = { 1024*1024*1024, 1024*1024, 1024, 1 };
+  static const uint64_t kAmountPerUnit10[] = { 1000*1000*1000, 1000*1000, 1000, 1 };
+  static const char* const kUnitStrings2[] = { "Gi", "Mi", "Ki", "" };
+  static const char* const kUnitStrings10[] = { "G", "M", "k", "" };
+
+  // Which set are we using?
+  const uint64_t* kUnitThresholds = ((base == 2) ? kUnitThresholds2 : kUnitThresholds10);
+  const uint64_t* kAmountPerUnit = ((base == 2) ? kAmountPerUnit2 : kAmountPerUnit10);
+  const char* const* kUnitStrings = ((base == 2) ? kUnitStrings2 : kUnitStrings10);
+
+  size_t i = 0;
+  for (; kUnitThresholds[i] != 0; ++i) {
+    if (count >= kUnitThresholds[i]) {
+      break;
+    }
+  }
+  char* s = NULL;
+  asprintf(&s, "%s%" PRId64 "%s", (negative_number ? "-" : ""),
+           count / kAmountPerUnit[i], kUnitStrings[i]);
+  std::string result(s);
+  free(s);
+  return result;
+}

benchmarks/utils.h

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

libc/Android.mk

@@ -85,7 +85,7 @@ libc_common_src_files += \
     bionic/__vsnprintf_chk.cpp \
     bionic/__vsprintf_chk.cpp \
 
-libc_bionic_src_files := \
+libc_bionic_ndk_src_files := \
     bionic/abort.cpp \
     bionic/accept.cpp \
     bionic/accept4.cpp \
@@ -116,16 +116,14 @@ libc_bionic_src_files := \
     bionic/error.cpp \
     bionic/eventfd_read.cpp \
     bionic/eventfd_write.cpp \
+    bionic/faccessat.cpp \
     bionic/fchmod.cpp \
     bionic/fchmodat.cpp \
     bionic/ffs.cpp \
     bionic/flockfile.cpp \
-    bionic/fork.cpp \
     bionic/fpclassify.cpp \
     bionic/futimens.cpp \
-    bionic/getauxval.cpp \
     bionic/getcwd.cpp \
-    bionic/getentropy_linux.c \
     bionic/gethostname.cpp \
     bionic/getpgrp.cpp \
     bionic/getpid.cpp \
@@ -146,6 +144,7 @@ libc_bionic_src_files := \
     bionic/mbrtoc16.cpp \
     bionic/mbrtoc32.cpp \
     bionic/mbstate.cpp \
+    bionic/mempcpy.cpp \
     bionic/mkdir.cpp \
     bionic/mkfifo.cpp \
     bionic/mknod.cpp \
@@ -160,27 +159,6 @@ libc_bionic_src_files := \
     bionic/posix_fallocate.cpp \
     bionic/posix_madvise.cpp \
     bionic/posix_timers.cpp \
-    bionic/pthread_atfork.cpp \
-    bionic/pthread_attr.cpp \
-    bionic/pthread_cond.cpp \
-    bionic/pthread_create.cpp \
-    bionic/pthread_detach.cpp \
-    bionic/pthread_equal.cpp \
-    bionic/pthread_exit.cpp \
-    bionic/pthread_getcpuclockid.cpp \
-    bionic/pthread_getschedparam.cpp \
-    bionic/pthread_gettid_np.cpp \
-    bionic/pthread_internals.cpp \
-    bionic/pthread_join.cpp \
-    bionic/pthread_key.cpp \
-    bionic/pthread_kill.cpp \
-    bionic/pthread_mutex.cpp \
-    bionic/pthread_once.cpp \
-    bionic/pthread_rwlock.cpp \
-    bionic/pthread_self.cpp \
-    bionic/pthread_setname_np.cpp \
-    bionic/pthread_setschedparam.cpp \
-    bionic/pthread_sigmask.cpp \
     bionic/ptrace.cpp \
     bionic/pty.cpp \
     bionic/raise.cpp \
@@ -223,7 +201,6 @@ libc_bionic_src_files := \
     bionic/strtold.cpp \
     bionic/stubs.cpp \
     bionic/symlink.cpp \
-    bionic/sysconf.cpp \
     bionic/sysinfo.cpp \
     bionic/syslog.cpp \
     bionic/sys_siglist.c \
@@ -236,10 +213,27 @@ libc_bionic_src_files := \
     bionic/umount.cpp \
     bionic/unlink.cpp \
     bionic/utimes.cpp \
-    bionic/vdso.cpp \
     bionic/wait.cpp \
     bionic/wchar.cpp \
     bionic/wctype.cpp \
+    bionic/wmempcpy.cpp \
+
+libc_bionic_src_files :=
+
+# The fork implementation depends on pthread data, so we can't include it in
+# libc_ndk.a.
+libc_bionic_src_files += bionic/fork.cpp
+
+# The data that backs getauxval is initialized in the libc init functions which
+# are invoked by the linker. If this file is included in libc_ndk.a, only one of
+# the copies of the global data will be initialized, resulting in nullptr
+# dereferences.
+libc_bionic_src_files += bionic/getauxval.cpp
+
+# These three require getauxval, which isn't available on older platforms.
+libc_bionic_src_files += bionic/getentropy_linux.c
+libc_bionic_src_files += bionic/sysconf.cpp
+libc_bionic_src_files += bionic/vdso.cpp
 
 libc_cxa_src_files := \
     bionic/__cxa_guard.cpp \
@@ -337,10 +331,13 @@ libc_upstream_openbsd_gdtoa_src_files_64 := \
     $(libc_upstream_openbsd_gdtoa_src_files) \
     upstream-openbsd/lib/libc/gdtoa/strtorQ.c \
 
+# These two depend on getentropy_linux.cpp, which isn't in libc_ndk.a.
 libc_upstream_openbsd_src_files := \
-    upstream-openbsd/lib/libc/compat-43/killpg.c \
     upstream-openbsd/lib/libc/crypt/arc4random.c \
     upstream-openbsd/lib/libc/crypt/arc4random_uniform.c \
+
+libc_upstream_openbsd_ndk_src_files := \
+    upstream-openbsd/lib/libc/compat-43/killpg.c \
     upstream-openbsd/lib/libc/gen/alarm.c \
     upstream-openbsd/lib/libc/gen/ctype_.c \
     upstream-openbsd/lib/libc/gen/daemon.c \
@@ -513,6 +510,29 @@ libc_upstream_openbsd_src_files := \
     upstream-openbsd/lib/libc/string/wcsstr.c \
     upstream-openbsd/lib/libc/string/wcswidth.c \
 
+libc_pthread_src_files := \
+    bionic/pthread_atfork.cpp \
+    bionic/pthread_attr.cpp \
+    bionic/pthread_cond.cpp \
+    bionic/pthread_create.cpp \
+    bionic/pthread_detach.cpp \
+    bionic/pthread_equal.cpp \
+    bionic/pthread_exit.cpp \
+    bionic/pthread_getcpuclockid.cpp \
+    bionic/pthread_getschedparam.cpp \
+    bionic/pthread_gettid_np.cpp \
+    bionic/pthread_internals.cpp \
+    bionic/pthread_join.cpp \
+    bionic/pthread_key.cpp \
+    bionic/pthread_kill.cpp \
+    bionic/pthread_mutex.cpp \
+    bionic/pthread_once.cpp \
+    bionic/pthread_rwlock.cpp \
+    bionic/pthread_self.cpp \
+    bionic/pthread_setname_np.cpp \
+    bionic/pthread_setschedparam.cpp \
+    bionic/pthread_sigmask.cpp \
+
 libc_arch_static_src_files := \
     bionic/dl_iterate_phdr_static.cpp \
 
@@ -785,6 +805,51 @@ $(eval $(call patch-up-arch-specific-flags,LOCAL_SRC_FILES,libc_netbsd_src_files
 include $(BUILD_STATIC_LIBRARY)
 
 
+# ========================================================
+# libc_openbsd_ndk.a - upstream OpenBSD C library code
+# that can be safely included in the libc_ndk.a (doesn't
+# contain any troublesome global data or constructors).
+# ========================================================
+#
+# These files are built with the openbsd-compat.h header file
+# automatically included.
+
+include $(CLEAR_VARS)
+
+LOCAL_SRC_FILES := $(libc_upstream_openbsd_ndk_src_files)
+ifneq (,$(filter $(TARGET_ARCH),x86 x86_64))
+  # Clang has wrong long double size or LDBL_MANT_DIG, http://b/17163651.
+  LOCAL_CLANG := false
+else
+  LOCAL_CLANG := $(use_clang)
+endif
+
+LOCAL_CFLAGS := \
+    $(libc_common_cflags) \
+    -Wno-sign-compare \
+    -Wno-uninitialized \
+    -Wno-unused-parameter \
+    -include openbsd-compat.h \
+
+LOCAL_CONLYFLAGS := $(libc_common_conlyflags)
+LOCAL_CPPFLAGS := $(libc_common_cppflags)
+LOCAL_C_INCLUDES := $(libc_common_c_includes) \
+    $(LOCAL_PATH)/private \
+    $(LOCAL_PATH)/upstream-openbsd/android/include \
+    $(LOCAL_PATH)/upstream-openbsd/lib/libc/include \
+    $(LOCAL_PATH)/upstream-openbsd/lib/libc/gdtoa/ \
+
+LOCAL_MODULE := libc_openbsd_ndk
+LOCAL_ADDITIONAL_DEPENDENCIES := $(libc_common_additional_dependencies)
+LOCAL_CXX_STL := none
+LOCAL_SYSTEM_SHARED_LIBRARIES :=
+LOCAL_ADDRESS_SANITIZER := false
+LOCAL_NATIVE_COVERAGE := $(bionic_coverage)
+
+$(eval $(call patch-up-arch-specific-flags,LOCAL_CFLAGS,libc_common_cflags))
+include $(BUILD_STATIC_LIBRARY)
+
+
 # ========================================================
 # libc_openbsd.a - upstream OpenBSD C library code
 # ========================================================
@@ -899,11 +964,80 @@ LOCAL_SYSTEM_SHARED_LIBRARIES :=
 LOCAL_ADDRESS_SANITIZER := false
 LOCAL_NATIVE_COVERAGE := $(bionic_coverage)
 
+$(eval $(call patch-up-arch-specific-flags,LOCAL_CFLAGS,libc_common_cflags))
+include $(BUILD_STATIC_LIBRARY)
+
+
+# ========================================================
+# libc_bionic_ndk.a - The portions of libc_bionic that can
+# be safely used in libc_ndk.a (no troublesome global data
+# or constructors).
+# ========================================================
+
+include $(CLEAR_VARS)
+
+LOCAL_SRC_FILES := $(libc_bionic_ndk_src_files)
+LOCAL_CFLAGS := $(libc_common_cflags) \
+    -Wframe-larger-than=2048 \
+
+# ssse3-strcmp-slm.S does not compile with Clang.
+LOCAL_CLANG_ASFLAGS_x86_64 += -no-integrated-as
+
+# memcpy.S, memchr.S, etc. do not compile with Clang.
+LOCAL_CLANG_ASFLAGS_arm += -no-integrated-as
+LOCAL_CLANG_ASFLAGS_arm64 += -no-integrated-as
+
+LOCAL_CONLYFLAGS := $(libc_common_conlyflags)
+LOCAL_CPPFLAGS := $(libc_common_cppflags) -Wold-style-cast
+LOCAL_C_INCLUDES := $(libc_common_c_includes) bionic/libstdc++/include
+LOCAL_MODULE := libc_bionic_ndk
+LOCAL_CLANG := $(use_clang)
+LOCAL_ADDITIONAL_DEPENDENCIES := $(libc_common_additional_dependencies)
+LOCAL_CXX_STL := none
+LOCAL_SYSTEM_SHARED_LIBRARIES :=
+LOCAL_ADDRESS_SANITIZER := false
+LOCAL_NATIVE_COVERAGE := $(bionic_coverage)
+
 $(eval $(call patch-up-arch-specific-flags,LOCAL_CFLAGS,libc_common_cflags))
 $(eval $(call patch-up-arch-specific-flags,LOCAL_SRC_FILES,libc_bionic_src_files))
 include $(BUILD_STATIC_LIBRARY)
 
 
+# ========================================================
+# libc_pthread.a - pthreads parts that previously lived in
+# libc_bionic.a. Relocated to their own library because
+# they can't be included in libc_ndk.a (as they layout of
+# pthread_t has changed over the years and has ABI
+# compatibility issues).
+# ========================================================
+
+include $(CLEAR_VARS)
+
+LOCAL_SRC_FILES := $(libc_pthread_src_files)
+LOCAL_CFLAGS := $(libc_common_cflags) \
+    -Wframe-larger-than=2048 \
+
+# ssse3-strcmp-slm.S does not compile with Clang.
+LOCAL_CLANG_ASFLAGS_x86_64 += -no-integrated-as
+
+# memcpy.S, memchr.S, etc. do not compile with Clang.
+LOCAL_CLANG_ASFLAGS_arm += -no-integrated-as
+LOCAL_CLANG_ASFLAGS_arm64 += -no-integrated-as
+
+LOCAL_CONLYFLAGS := $(libc_common_conlyflags)
+LOCAL_CPPFLAGS := $(libc_common_cppflags) -Wold-style-cast
+LOCAL_C_INCLUDES := $(libc_common_c_includes)
+LOCAL_MODULE := libc_pthread
+LOCAL_CLANG := $(use_clang)
+LOCAL_ADDITIONAL_DEPENDENCIES := $(libc_common_additional_dependencies)
+LOCAL_CXX_STL := none
+LOCAL_SYSTEM_SHARED_LIBRARIES :=
+LOCAL_ADDRESS_SANITIZER := false
+LOCAL_NATIVE_COVERAGE := $(bionic_coverage)
+
+include $(BUILD_STATIC_LIBRARY)
+
+
 # ========================================================
 # libc_cxa.a - Things traditionally in libstdc++
 # ========================================================
@@ -992,9 +1126,56 @@ include $(BUILD_STATIC_LIBRARY)
 include $(CLEAR_VARS)
 
 LOCAL_MODULE := libc_ndk
-LOCAL_WHOLE_STATIC_LIBRARIES := libc_syscalls libm
+LOCAL_CLANG := $(use_clang)
-LOCAL_ADDITIONAL_DEPENDENCIES := $(libc_common_additional_dependencies)
+LOCAL_ASFLAGS := $(LOCAL_CFLAGS)
+LOCAL_CONLYFLAGS := $(libc_common_conlyflags)
+LOCAL_CFLAGS := $(libc_common_cflags) -fvisibility=hidden -O0
+LOCAL_CPPFLAGS := $(libc_common_cppflags)
+LOCAL_C_INCLUDES := $(libc_common_c_includes)
+LOCAL_ADDRESS_SANITIZER := false
+LOCAL_NATIVE_COVERAGE := $(bionic_coverage)
+LOCAL_SYSTEM_SHARED_LIBRARIES :=
 
+LOCAL_SRC_FILES := \
+    $(libc_common_src_files) \
+    $(libc_arch_dynamic_src_files) \
+    $(libc_ndk_stub_src_files) \
+    bionic/malloc_debug_common.cpp \
+
+LOCAL_SRC_FILES_arm += \
+    arch-common/bionic/crtbegin_so.c \
+    arch-arm/bionic/atexit_legacy.c \
+    arch-common/bionic/crtend_so.S \
+
+LOCAL_CFLAGS := $(libc_common_cflags) \
+    -DLIBC_STATIC \
+
+LOCAL_WHOLE_STATIC_LIBRARIES := \
+    libc_bionic_ndk \
+    libc_cxa \
+    libc_freebsd \
+    libc_gdtoa \
+    libc_malloc \
+    libc_netbsd \
+    libc_openbsd_ndk \
+    libc_stack_protector \
+    libc_syscalls \
+    libc_tzcode \
+    libm \
+
+LOCAL_WHOLE_STATIC_LIBRARIES_arm := libc_aeabi
+LOCAL_CXX_STL := none
+
+ifneq ($(MALLOC_IMPL),dlmalloc)
+LOCAL_WHOLE_STATIC_LIBRARIES += libjemalloc
+endif
+
+$(eval $(call patch-up-arch-specific-flags,LOCAL_CFLAGS,libc_common_cflags))
+$(eval $(call patch-up-arch-specific-flags,LOCAL_SRC_FILES,libc_common_src_files))
+$(eval $(call patch-up-arch-specific-flags,LOCAL_SRC_FILES,libc_arch_dynamic_src_files))
+$(eval $(call patch-up-arch-specific-flags,LOCAL_ASFLAGS,LOCAL_CFLAGS))
+
+LOCAL_ADDITIONAL_DEPENDENCIES := $(libc_common_additional_dependencies)
 include $(BUILD_STATIC_LIBRARY)
 
 # ========================================================
@@ -1013,6 +1194,7 @@ LOCAL_CLANG := $(use_clang)
 LOCAL_ADDITIONAL_DEPENDENCIES := $(libc_common_additional_dependencies)
 LOCAL_WHOLE_STATIC_LIBRARIES := \
     libc_bionic \
+    libc_bionic_ndk \
     libc_cxa \
     libc_dns \
     libc_freebsd \
@@ -1020,6 +1202,8 @@ LOCAL_WHOLE_STATIC_LIBRARIES := \
     libc_malloc \
     libc_netbsd \
     libc_openbsd \
+    libc_openbsd_ndk \
+    libc_pthread \
     libc_stack_protector \
     libc_syscalls \
     libc_tzcode \
@@ -1140,6 +1324,10 @@ include $(CLEAR_VARS)
 LOCAL_CFLAGS := $(libc_common_cflags)
 LOCAL_CONLYFLAGS := $(libc_common_conlyflags)
 LOCAL_CPPFLAGS := $(libc_common_cppflags)
+
+# TODO: This is to work around b/19059885. Remove after root cause is fixed
+LOCAL_LDFLAGS_arm := -Wl,--hash-style=sysv
+
 LOCAL_C_INCLUDES := $(libc_common_c_includes)
 LOCAL_SRC_FILES := \
     $(libc_arch_dynamic_src_files) \
@@ -1290,6 +1478,10 @@ include $(CLEAR_VARS)
 LOCAL_C_INCLUDES := $(libc_common_c_includes) bionic/libstdc++/include
 LOCAL_CFLAGS := $(libc_common_cflags)
 LOCAL_CPPFLAGS := $(libc_common_cppflags)
+
+# TODO: This is to work around b/19059885. Remove after root cause is fixed
+LOCAL_LDFLAGS_arm := -Wl,--hash-style=both
+
 LOCAL_SRC_FILES := $(libstdcxx_common_src_files)
 LOCAL_MODULE:= libstdc++
 LOCAL_ADDITIONAL_DEPENDENCIES := $(LOCAL_PATH)/Android.mk

libc/SYSCALLS.TXT

@@ -130,7 +130,7 @@ 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 ___faccessat:faccessat(int, const char*, int)  all
 int ___fchmodat:fchmodat(int, const char*, mode_t)  all
 int fchownat(int, const char*, uid_t, gid_t, int)  all
 int fstatat64|fstatat:fstatat64(int, const char*, struct stat*, int)   arm,mips,x86

libc/arch-arm/arm.mk

@@ -5,8 +5,6 @@
 #
 
 libc_bionic_src_files_arm += \
-    bionic/memchr.c \
-    bionic/memrchr.c \
     bionic/strchr.cpp \
     bionic/strnlen.c \
     bionic/strrchr.cpp \
@@ -22,6 +20,8 @@ libc_freebsd_src_files_arm += \
     upstream-freebsd/lib/libc/string/wmemmove.c \
 
 libc_openbsd_src_files_arm += \
+    upstream-openbsd/lib/libc/string/memchr.c \
+    upstream-openbsd/lib/libc/string/memrchr.c \
     upstream-openbsd/lib/libc/string/stpncpy.c \
     upstream-openbsd/lib/libc/string/strlcat.c \
     upstream-openbsd/lib/libc/string/strlcpy.c \

libc/arch-arm/syscalls/___faccessat.S

@@ -2,7 +2,7 @@
 
 #include <private/bionic_asm.h>
 
-ENTRY(faccessat)
+ENTRY(___faccessat)
     mov     ip, r7
     ldr     r7, =__NR_faccessat
     swi     #0
@@ -11,4 +11,5 @@ ENTRY(faccessat)
     bxls    lr
     neg     r0, r0
     b       __set_errno_internal
-END(faccessat)
+END(___faccessat)
+.hidden ___faccessat

libc/arch-arm64/arm64.mk

@@ -8,7 +8,6 @@ libc_bionic_src_files_arm64 += \
     bionic/__memset_chk.cpp \
     bionic/__strcpy_chk.cpp \
     bionic/__strcat_chk.cpp \
-    bionic/memrchr.c \
     bionic/strrchr.cpp \
 
 libc_freebsd_src_files_arm64 += \
@@ -21,6 +20,7 @@ libc_freebsd_src_files_arm64 += \
     upstream-freebsd/lib/libc/string/wmemcmp.c \
 
 libc_openbsd_src_files_arm64 += \
+    upstream-openbsd/lib/libc/string/memrchr.c \
     upstream-openbsd/lib/libc/string/stpncpy.c \
     upstream-openbsd/lib/libc/string/strcat.c \
     upstream-openbsd/lib/libc/string/strlcat.c \

libc/arch-arm64/syscalls/___faccessat.S

@@ -2,7 +2,7 @@
 
 #include <private/bionic_asm.h>
 
-ENTRY(faccessat)
+ENTRY(___faccessat)
     mov     x8, __NR_faccessat
     svc     #0
 
@@ -11,4 +11,5 @@ ENTRY(faccessat)
     b.hi    __set_errno_internal
 
     ret
-END(faccessat)
+END(___faccessat)
+.hidden ___faccessat

libc/arch-mips/bionic/setjmp.S

@@ -1,3 +1,30 @@
+/*
+ * Copyright (C) 2014-2015 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
 /*
  * Copyright (c) 2001-2002 Opsycon AB  (www.opsycon.se / www.opsycon.com)
  *
@@ -94,23 +121,31 @@
 #include <private/bionic_asm.h>
 #include <machine/setjmp.h>
 
-/* On Mips32, jmpbuf begins with optional 4-byte filler so that
+/* jmpbuf is declared to users as an array of longs, which is only
- *  all saved FP regs are aligned on 8-byte boundary, despite this whole
+ * 4-byte aligned in 32-bit builds.  The Mips jmpbuf begins with a
- *  struct being mis-declared to users as an array of (4-byte) longs.
+ * dynamically-sized 0- or 4-byte unused filler so that double-prec FP regs
- *  All the following offsets are then from the rounded-up base addr
+ * are saved to 8-byte-aligned mem cells.
+ * All the following jmpbuf offsets are from the rounded-DOWN base addr.
  */
 
 /* Fields of same size on all MIPS abis: */
-#define	SC_MAGIC        (0*4)		/* 4 bytes, identify jmpbuf */
+/*     	field:		byte offset:	size:						*/
-#define	SC_MASK		(1*4)		/* 4 bytes, saved signal mask */
+/*     	dynam filler	(0*4)		   0-4 bytes of rounddown filler, DON'T TOUCH!!
-#define	SC_FPSR		(2*4)		/* 4 bytes, floating point control/status reg */
+						often overlays user storage!!		*/
-/*     	filler2		(3*4)		   4 bytes, pad to 8-byte boundary */
+#define	SC_MAGIC_OFFSET	(1*4)		/* 4 bytes, identify jmpbuf, first actual field */
+#define	SC_FLAG_OFFSET	(2*4)		/* 4 bytes, savesigs flag */
+#define	SC_FPSR_OFFSET	(3*4)		/* 4 bytes, floating point control/status reg */
+/* following fields are 8-byte aligned */
+#define	SC_MASK_OFFSET	(4*4)		/* 16 bytes, mips32/mips64 version of sigset_t */
+#define	SC_SPARE_OFFSET	(8*4)		/* 8 bytes, reserved for future uses */
 
 /* Registers that are 4-byte on mips32 o32, and 8-byte on mips64 n64 abi */
-#define	SC_REGS_SAVED	12		/* ra,gp,sp,s0-s8 */
+#define	SC_REGS_OFFSET	(10*4)		/* SC_REGS_BYTES */
-#define	SC_REGS		(4*4)		/* SC_REGS_SAVED*REGSZ bytes */
+#define	SC_REGS_SAVED	12 /*regs*/	/* ra,s0-s8,gp,sp */
+#define	SC_REGS_BYTES   (SC_REGS_SAVED*REGSZ)
+#define	SC_REGS		SC_REGS_OFFSET
 
-/* Floating pt registers are 8-bytes on all abis,
+/* Double floating pt registers are 8-bytes on all abis,
  * but the number of saved fp regs varies for o32/n32 versus n64 abis:
  */
 
@@ -120,22 +155,20 @@
 #define	SC_FPREGS_SAVED	6  /* even fp regs f20,f22,f24,f26,f28,f30 */
 #endif
 
-#define	SC_FPREGS	(SC_REGS + SC_REGS_SAVED*REGSZ)  /* SC_FPREGS_SAVED*REGSZ_FP bytes */
+#define	SC_FPREGS_OFFSET (SC_REGS_OFFSET + SC_REGS_BYTES)  /* SC_FPREGS_BYTES */
+#define	SC_FPREGS_BYTES (SC_FPREGS_SAVED*REGSZ_FP)
+#define	SC_FPREGS	SC_FPREGS_OFFSET
 
-#define	SC_BYTES	(SC_FPREGS + SC_FPREGS_SAVED*REGSZ_FP)
+#define	SC_TOTAL_BYTES	(SC_FPREGS_OFFSET + SC_FPREGS_BYTES)
-#define	SC_LONGS	(SC_BYTES/REGSZ)
+#define	SC_TOTAL_LONGS	(SC_TOTAL_BYTES/REGSZ)
 
-#ifdef __LP64__
+#if SC_TOTAL_LONGS > _JBLEN
-/* SC_LONGS is 22, so _JBLEN should be 22 or larger */
+#error _JBLEN is too small
-#else
-/* SC_LONGS is 28, but must also allocate dynamic-roundup filler.
-   so _JBLEN should be 29 or larger */
 #endif
 
 /*
- * _setjmp, _longjmp (restoring signal state)
  *
- *  GPOFF and FRAMESIZE must be the same for both _setjmp and _longjmp!
+ *  GPOFF and FRAMESIZE must be the same for all setjmp/longjmp routines
  *
  */
 
@@ -145,30 +178,33 @@ A0OFF= FRAMESZ-3*REGSZ
 GPOFF= FRAMESZ-2*REGSZ
 RAOFF= FRAMESZ-1*REGSZ
 
-NON_LEAF(setjmp, FRAMESZ, ra)
+NON_LEAF(sigsetjmp, FRAMESZ, ra)
 	.mask	0x80000000, RAOFF
 	PTR_SUBU sp, FRAMESZ			# allocate stack frame
-	SETUP_GP64(GPOFF, setjmp)
+	SETUP_GP64(GPOFF, sigsetjmp)
 	SAVE_GP(GPOFF)
 	.set	reorder
 
+setjmp_common:
 #ifndef __LP64__
-	addiu   a0, 7				# roundup jmpbuf addr to 8-byte boundary
+	li	t0, ~7
-	li      t0, ~7
+	and	a0, t0				# round jmpbuf addr DOWN to 8-byte boundary
-	and     a0, t0
 #endif
+	sw	a1, SC_FLAG_OFFSET(a0)		# save savesigs flag
+	beqz	a1, 1f				# do saving of signal mask?
 
-	REG_S	ra, RAOFF(sp)			# save state
+	REG_S	ra, RAOFF(sp)			# spill state
 	REG_S	a0, A0OFF(sp)
-	move	a0, zero			# get current signal mask
+	# call sigprocmask(int how ignored, sigset_t* null, sigset_t* SC_MASK(a0)):
-	jal	sigblock
+	LA	a2, SC_MASK_OFFSET(a0)		# gets current signal mask
+	li	a0, 0				# how; ignored when new mask is null
+	li	a1, 0				# null new mask
+	jal	sigprocmask			# get current signal mask
 	REG_L	a0, A0OFF(sp)
 	REG_L	ra, RAOFF(sp)
-
+1:
-	REG_S	v0, SC_MASK(a0)			# save sc_mask = sigblock(0)
-
 	li	v0, 0xACEDBADE			# sigcontext magic number
-	sw	v0, SC_MAGIC(a0)
+	sw	v0, SC_MAGIC_OFFSET(a0)
 	# callee-saved long-sized regs:
 	REG_S	ra, SC_REGS+0*REGSZ(a0)
 	REG_S	s0, SC_REGS+1*REGSZ(a0)
@@ -181,9 +217,9 @@ NON_LEAF(setjmp, FRAMESZ, ra)
 	REG_S	s7, SC_REGS+8*REGSZ(a0)
 	REG_S	s8, SC_REGS+9*REGSZ(a0)
 	REG_L	v0, GPOFF(sp)
-	REG_S	v0, SC_REGS+10*REGSZ(a0)
+	REG_S	v0, SC_REGS+10*REGSZ(a0)	# save gp
 	PTR_ADDU v0, sp, FRAMESZ
-	REG_S	v0, SC_REGS+11*REGSZ(a0)
+	REG_S	v0, SC_REGS+11*REGSZ(a0)	# save orig sp
 
 	cfc1	v0, $31
 
@@ -199,7 +235,7 @@ NON_LEAF(setjmp, FRAMESZ, ra)
 	s.d	$f31, SC_FPREGS+7*REGSZ_FP(a0)
 #else
 	# callee-saved fp regs on mips o32 ABI are
-	#   the even-numbered fp regs $f20,$f22,...$f30
+	#   the even-numbered double fp regs $f20,$f22,...$f30
 	s.d	$f20, SC_FPREGS+0*REGSZ_FP(a0)
 	s.d	$f22, SC_FPREGS+1*REGSZ_FP(a0)
 	s.d	$f24, SC_FPREGS+2*REGSZ_FP(a0)
@@ -207,37 +243,68 @@ NON_LEAF(setjmp, FRAMESZ, ra)
 	s.d	$f28, SC_FPREGS+4*REGSZ_FP(a0)
 	s.d	$f30, SC_FPREGS+5*REGSZ_FP(a0)
 #endif
-	sw	v0, SC_FPSR(a0)
+	sw	v0, SC_FPSR_OFFSET(a0)
 	move	v0, zero
 	RESTORE_GP64
 	PTR_ADDU sp, FRAMESZ
 	j	ra
-END(setjmp)
+END(sigsetjmp)
 
-NON_LEAF(longjmp, FRAMESZ, ra)
+
+# Alternate entry points:
+
+NON_LEAF(setjmp, FRAMESZ, ra)
 	.mask	0x80000000, RAOFF
 	PTR_SUBU sp, FRAMESZ
-	SETUP_GP64(GPOFF, longjmp)
+	SETUP_GP64(GPOFF, setjmp)		# can't share sigsetjmp's gp code
+	SAVE_GP(GPOFF)
+	.set	reorder
+
+	li	a1, 1				# save/restore signals state
+	b	setjmp_common			# tail call
+END(setjmp)
+
+
+NON_LEAF(_setjmp, FRAMESZ, ra)
+	.mask	0x80000000, RAOFF
+	PTR_SUBU sp, FRAMESZ
+	SETUP_GP64(GPOFF, _setjmp)		# can't share sigsetjmp's gp code
+	SAVE_GP(GPOFF)
+	.set	reorder
+
+	li	a1, 0				# don't save/restore signals
+	b	setjmp_common			# tail call
+END(_setjmp)
+
+
+NON_LEAF(siglongjmp, FRAMESZ, ra)
+	.mask	0x80000000, RAOFF
+	PTR_SUBU sp, FRAMESZ
+	SETUP_GP64(GPOFF, siglongjmp)
 	SAVE_GP(GPOFF)
 	.set	reorder
 
 #ifndef __LP64__
-	addiu	a0, 7				# roundup jmpbuf addr to 8-byte boundary
+	li	t0, ~7
-	li      t0, ~7
+	and	a0, t0				# round jmpbuf addr DOWN to 8-byte boundary
-	and	a0, t0
 #endif
+	lw	v0, SC_MAGIC_OFFSET(a0)
+	li	t0, 0xACEDBADE
+	bne	v0, t0, longjmp_botch		# jump if error
 
-	REG_S	a1, A1OFF(sp)
+	lw	t0, SC_FLAG_OFFSET(a0)		# get savesigs flag
+	beqz	t0, 1f				# restore signal mask?
+
+	REG_S	a1, A1OFF(sp)			# temp spill
 	REG_S	a0, A0OFF(sp)
-	lw	a0, SC_MASK(a0)
+        # call sigprocmask(int how SIG_SETMASK, sigset_t* SC_MASK(a0), sigset_t* null):
-	jal	sigsetmask
+	LA	a1, SC_MASK_OFFSET(a0)		# signals being restored
+	li	a0, 3				# mips SIG_SETMASK
+	li	a2, 0				# null
+	jal	sigprocmask			# restore signal mask
 	REG_L	a0, A0OFF(sp)
 	REG_L	a1, A1OFF(sp)
-
+1:
-	lw	v0, SC_MAGIC(a0)
-	li	t0, 0xACEDBADE
-	bne	v0, t0, longjmp_botch			# jump if error
-
 	# callee-saved long-sized regs:
 	REG_L	ra, SC_REGS+0*REGSZ(a0)
 	REG_L	s0, SC_REGS+1*REGSZ(a0)
@@ -252,8 +319,8 @@ NON_LEAF(longjmp, FRAMESZ, ra)
 	REG_L	gp, SC_REGS+10*REGSZ(a0)
 	REG_L	sp, SC_REGS+11*REGSZ(a0)
 
-	lw	v0, SC_FPSR(a0)
+	lw	v0, SC_FPSR_OFFSET(a0)
-	ctc1	v0, $31
+	ctc1	v0, $31			# restore old fr mode before fp values
 #ifdef __LP64__
 	# callee-saved fp regs on mips n64 ABI are $f24..$f31
 	l.d	$f24, SC_FPREGS+0*REGSZ_FP(a0)
@@ -266,7 +333,7 @@ NON_LEAF(longjmp, FRAMESZ, ra)
 	l.d	$f31, SC_FPREGS+7*REGSZ_FP(a0)
 #else
 	# callee-saved fp regs on mips o32 ABI are
-	#   the even-numbered fp regs $f20,$f22,...$f30
+	#   the even-numbered double fp regs $f20,$f22,...$f30
 	l.d	$f20, SC_FPREGS+0*REGSZ_FP(a0)
 	l.d	$f22, SC_FPREGS+1*REGSZ_FP(a0)
 	l.d	$f24, SC_FPREGS+2*REGSZ_FP(a0)
@@ -278,192 +345,19 @@ NON_LEAF(longjmp, FRAMESZ, ra)
 	li	a1, 1			# never return 0!
 1:
 	move	v0, a1
-	j	ra
+	j	ra			# return to setjmp call site
 
 longjmp_botch:
 	jal	longjmperror
 	jal	abort
-	RESTORE_GP64
-	PTR_ADDU sp, FRAMESZ
-END(longjmp)
-
-
-/*
- * _setjmp, _longjmp (not restoring signal state)
- *
- *  GPOFF and FRAMESIZE must be the same for both _setjmp and _longjmp!
- *
- */
-
-FRAMESZ= MKFSIZ(0,4)
-GPOFF= FRAMESZ-2*REGSZ
-
-LEAF(_setjmp, FRAMESZ)
-	PTR_SUBU sp, FRAMESZ
-	SETUP_GP64(GPOFF, _setjmp)
-	SAVE_GP(GPOFF)
-	.set	reorder
-
-#ifndef __LP64__
-	addiu   a0, 7				# roundup jmpbuf addr to 8-byte boundary
-	li      t0, ~7
-	and     a0, t0
-#endif
-
-	# SC_MASK is unused here
-
-	li	v0, 0xACEDBADE			# sigcontext magic number
-	sw	v0, SC_MAGIC(a0)
-	# callee-saved long-sized regs:
-	REG_S	ra, SC_REGS+0*REGSZ(a0)
-	REG_S	s0, SC_REGS+1*REGSZ(a0)
-	REG_S	s1, SC_REGS+2*REGSZ(a0)
-	REG_S	s2, SC_REGS+3*REGSZ(a0)
-	REG_S	s3, SC_REGS+4*REGSZ(a0)
-	REG_S	s4, SC_REGS+5*REGSZ(a0)
-	REG_S	s5, SC_REGS+6*REGSZ(a0)
-	REG_S	s6, SC_REGS+7*REGSZ(a0)
-	REG_S	s7, SC_REGS+8*REGSZ(a0)
-	REG_S	s8, SC_REGS+9*REGSZ(a0)
-	REG_L	v0, GPOFF(sp)
-	REG_S	v0, SC_REGS+10*REGSZ(a0)
-	PTR_ADDU v0, sp, FRAMESZ
-	REG_S	v0, SC_REGS+11*REGSZ(a0)
-
-	cfc1	v0, $31
-
-#ifdef __LP64__
-	# callee-saved fp regs on mips n64 ABI are $f24..$f31
-	s.d	$f24, SC_FPREGS+0*REGSZ_FP(a0)
-	s.d	$f25, SC_FPREGS+1*REGSZ_FP(a0)
-	s.d	$f26, SC_FPREGS+2*REGSZ_FP(a0)
-	s.d	$f27, SC_FPREGS+3*REGSZ_FP(a0)
-	s.d	$f28, SC_FPREGS+4*REGSZ_FP(a0)
-	s.d	$f29, SC_FPREGS+5*REGSZ_FP(a0)
-	s.d	$f30, SC_FPREGS+6*REGSZ_FP(a0)
-	s.d	$f31, SC_FPREGS+7*REGSZ_FP(a0)
-#else
-	# callee-saved fp regs on mips o32 ABI are
-	#   the even-numbered fp regs $f20,$f22,...$f30
-	s.d	$f20, SC_FPREGS+0*REGSZ_FP(a0)
-	s.d	$f22, SC_FPREGS+1*REGSZ_FP(a0)
-	s.d	$f24, SC_FPREGS+2*REGSZ_FP(a0)
-	s.d	$f26, SC_FPREGS+3*REGSZ_FP(a0)
-	s.d	$f28, SC_FPREGS+4*REGSZ_FP(a0)
-	s.d	$f30, SC_FPREGS+5*REGSZ_FP(a0)
-#endif
-	sw	v0, SC_FPSR(a0)
-	move	v0, zero
-	RESTORE_GP64
-	PTR_ADDU sp, FRAMESZ
-	j	ra
-END(_setjmp)
-
-
-LEAF(_longjmp, FRAMESZ)
-	PTR_SUBU sp, FRAMESZ
-	SETUP_GP64(GPOFF, _longjmp)
-	SAVE_GP(GPOFF)
-	.set	reorder
-
-#ifndef __LP64__
-	addiu	a0, 7				# roundup jmpbuf addr to 8-byte boundary
-	li      t0, ~7
-	and	a0, t0
-#endif
-
-	# SC_MASK is unused here
-
-	lw	v0, SC_MAGIC(a0)
-	li	t0, 0xACEDBADE
-	bne	v0, t0, _longjmp_botch			# jump if error
-
-	# callee-saved long-sized regs:
-	REG_L	ra, SC_REGS+0*REGSZ(a0)
-	REG_L	s0, SC_REGS+1*REGSZ(a0)
-	REG_L	s1, SC_REGS+2*REGSZ(a0)
-	REG_L	s2, SC_REGS+3*REGSZ(a0)
-	REG_L	s3, SC_REGS+4*REGSZ(a0)
-	REG_L	s4, SC_REGS+5*REGSZ(a0)
-	REG_L	s5, SC_REGS+6*REGSZ(a0)
-	REG_L	s6, SC_REGS+7*REGSZ(a0)
-	REG_L	s7, SC_REGS+8*REGSZ(a0)
-	REG_L	s8, SC_REGS+9*REGSZ(a0)
-	REG_L	gp, SC_REGS+10*REGSZ(a0)
-	REG_L	sp, SC_REGS+11*REGSZ(a0)
-
-	lw	v0, SC_FPSR(a0)
-	ctc1	v0, $31
-#ifdef __LP64__
-	# callee-saved fp regs on mips n64 ABI are $f24..$f31
-	l.d	$f24, SC_FPREGS+0*REGSZ_FP(a0)
-	l.d	$f25, SC_FPREGS+1*REGSZ_FP(a0)
-	l.d	$f26, SC_FPREGS+2*REGSZ_FP(a0)
-	l.d	$f27, SC_FPREGS+3*REGSZ_FP(a0)
-	l.d	$f28, SC_FPREGS+4*REGSZ_FP(a0)
-	l.d	$f29, SC_FPREGS+5*REGSZ_FP(a0)
-	l.d	$f30, SC_FPREGS+6*REGSZ_FP(a0)
-	l.d	$f31, SC_FPREGS+7*REGSZ_FP(a0)
-#else
-	# callee-saved fp regs on mips o32 ABI are
-	#   the even-numbered fp regs $f20,$f22,...$f30
-	l.d	$f20, SC_FPREGS+0*REGSZ_FP(a0)
-	l.d	$f22, SC_FPREGS+1*REGSZ_FP(a0)
-	l.d	$f24, SC_FPREGS+2*REGSZ_FP(a0)
-	l.d	$f26, SC_FPREGS+3*REGSZ_FP(a0)
-	l.d	$f28, SC_FPREGS+4*REGSZ_FP(a0)
-	l.d	$f30, SC_FPREGS+5*REGSZ_FP(a0)
-#endif
-	bne	a1, zero, 1f
-	li	a1, 1			# never return 0!
-1:
-	move	v0, a1
-	j	ra
-
-_longjmp_botch:
-	jal	longjmperror
-	jal	abort
-	RESTORE_GP64
-	PTR_ADDU sp, FRAMESZ
-END(_longjmp)
-
-/*
- * trampolines for sigsetjmp and  siglongjmp save and restore mask.
- *
- */
-FRAMESZ= MKFSIZ(1,1)
-GPOFF= FRAMESZ-2*REGSZ
-
-LEAF(sigsetjmp, FRAMESZ)
-	PTR_SUBU sp, FRAMESZ
-	SETUP_GP64(GPOFF, sigsetjmp)
-	.set	reorder
-	sw	a1, _JBLEN*REGSZ(a0)		# save "savemask"
-	bne	a1, 0x0, 1f			# do saving of signal mask?
-	LA	t9, _setjmp
-	RESTORE_GP64
-	PTR_ADDU sp, FRAMESZ
-	jr t9
-
-1:	LA	t9, setjmp
-	RESTORE_GP64
-	PTR_ADDU sp, FRAMESZ
-	jr t9
-END(sigsetjmp)
-
-LEAF(siglongjmp, FRAMESZ)
-	PTR_SUBU sp, FRAMESZ
-	SETUP_GP64(GPOFF, siglongjmp)
-	.set	reorder
-	lw	t0, _JBLEN*REGSZ(a0)		# get "savemask"
-	bne	t0, 0x0, 1f			# restore signal mask?
-	LA	t9, _longjmp
-	RESTORE_GP64
-	PTR_ADDU sp, FRAMESZ
-	jr	t9
-1:
-	LA	t9, longjmp
-	RESTORE_GP64
-	PTR_ADDU sp, FRAMESZ
-	jr	t9
 END(siglongjmp)
+
+
+	.globl	longjmp
+	.type	longjmp, @function
+	.equ	longjmp, siglongjmp	# alias for siglongjmp
+
+
+	.globl	_longjmp
+	.type	_longjmp, @function
+	.equ	_longjmp, siglongjmp	# alias for siglongjmp

libc/arch-mips/include/machine/setjmp.h

@@ -6,9 +6,10 @@
 #define _MIPS_SETJMP_H_
 
 #ifdef __LP64__
-#define	_JBLEN	22		/* size, in 8-byte longs, of a mips64 jmp_buf */
+#define	_JBLEN	25	/* size, in 8-byte longs, of a mips64 jmp_buf/sigjmp_buf */
 #else
-#define	_JBLEN	29		/* size, in 4-byte longs, of a mips32 jmp_buf */
+#define	_JBLEN	157	/* historical size, in 4-byte longs, of a mips32 jmp_buf */
+			/* actual used size is 34 */
 #endif
 
 #endif /* !_MIPS_SETJMP_H_ */

libc/arch-mips/mips.mk

@@ -10,8 +10,6 @@ libc_bionic_src_files_mips += \
     bionic/__memset_chk.cpp \
     bionic/__strcpy_chk.cpp \
     bionic/__strcat_chk.cpp \
-    bionic/memchr.c \
-    bionic/memrchr.c \
     bionic/strchr.cpp \
     bionic/strnlen.c \
     bionic/strrchr.cpp \
@@ -27,7 +25,9 @@ libc_freebsd_src_files_mips += \
     upstream-freebsd/lib/libc/string/wmemmove.c \
 
 libc_openbsd_src_files_mips += \
+    upstream-openbsd/lib/libc/string/memchr.c \
     upstream-openbsd/lib/libc/string/memmove.c \
+    upstream-openbsd/lib/libc/string/memrchr.c \
     upstream-openbsd/lib/libc/string/stpcpy.c \
     upstream-openbsd/lib/libc/string/stpncpy.c \
     upstream-openbsd/lib/libc/string/strcat.c \

libc/arch-mips/syscalls/___faccessat.S

@@ -2,7 +2,7 @@
 
 #include <private/bionic_asm.h>
 
-ENTRY(faccessat)
+ENTRY(___faccessat)
     .set noreorder
     .cpload t9
     li v0, __NR_faccessat
@@ -16,4 +16,5 @@ ENTRY(faccessat)
     j t9
     nop
     .set reorder
-END(faccessat)
+END(___faccessat)
+.hidden ___faccessat

libc/arch-mips64/bionic/stat.cpp

@@ -29,7 +29,7 @@
 #include <sys/stat.h>
 #include <sys/types.h>
 #include <sys/syscall.h>
-#include <asm/unistd.h>
+#include <unistd.h>
 
 struct kernel_stat {
  unsigned int st_dev;

libc/arch-mips64/mips64.mk

@@ -9,8 +9,6 @@ libc_bionic_src_files_mips64 += \
     bionic/__memset_chk.cpp \
     bionic/__strcpy_chk.cpp \
     bionic/__strcat_chk.cpp \
-    bionic/memchr.c \
-    bionic/memrchr.c \
     bionic/strchr.cpp \
     bionic/strnlen.c \
     bionic/strrchr.cpp \
@@ -30,7 +28,9 @@ libc_freebsd_src_files_mips64 += \
     upstream-freebsd/lib/libc/string/wmemmove.c \
 
 libc_openbsd_src_files_mips64 += \
+    upstream-openbsd/lib/libc/string/memchr.c \
     upstream-openbsd/lib/libc/string/memmove.c \
+    upstream-openbsd/lib/libc/string/memrchr.c \
     upstream-openbsd/lib/libc/string/stpcpy.c \
     upstream-openbsd/lib/libc/string/stpncpy.c \
     upstream-openbsd/lib/libc/string/strcat.c \

libc/arch-mips64/syscalls/___faccessat.S

@@ -2,7 +2,7 @@
 
 #include <private/bionic_asm.h>
 
-ENTRY(faccessat)
+ENTRY(___faccessat)
     .set push
     .set noreorder
     li v0, __NR_faccessat
@@ -22,4 +22,5 @@ ENTRY(faccessat)
     j t9
     move ra, t0
     .set pop
-END(faccessat)
+END(___faccessat)
+.hidden ___faccessat

libc/arch-x86/syscalls/___faccessat.S

@@ -2,7 +2,7 @@
 
 #include <private/bionic_asm.h>
 
-ENTRY(faccessat)
+ENTRY(___faccessat)
     pushl   %ebx
     .cfi_def_cfa_offset 8
     .cfi_rel_offset ebx, 0
@@ -12,13 +12,9 @@ ENTRY(faccessat)
     pushl   %edx
     .cfi_adjust_cfa_offset 4
     .cfi_rel_offset edx, 0
-    pushl   %esi
+    mov     16(%esp), %ebx
-    .cfi_adjust_cfa_offset 4
+    mov     20(%esp), %ecx
-    .cfi_rel_offset esi, 0
+    mov     24(%esp), %edx
-    mov     20(%esp), %ebx
-    mov     24(%esp), %ecx
-    mov     28(%esp), %edx
-    mov     32(%esp), %esi
     movl    $__NR_faccessat, %eax
     int     $0x80
     cmpl    $-MAX_ERRNO, %eax
@@ -28,9 +24,9 @@ ENTRY(faccessat)
     call    __set_errno_internal
     addl    $4, %esp
 1:
-    popl    %esi
     popl    %edx
     popl    %ecx
     popl    %ebx
     ret
-END(faccessat)
+END(___faccessat)
+.hidden ___faccessat

libc/arch-x86_64/syscalls/___faccessat.S

@@ -2,8 +2,7 @@
 
 #include <private/bionic_asm.h>
 
-ENTRY(faccessat)
+ENTRY(___faccessat)
-    movq    %rcx, %r10
     movl    $__NR_faccessat, %eax
     syscall
     cmpq    $-MAX_ERRNO, %rax
@@ -13,4 +12,5 @@ ENTRY(faccessat)
     call    __set_errno_internal
 1:
     ret
-END(faccessat)
+END(___faccessat)
+.hidden ___faccessat

libc/arch-x86_64/x86_64.mk

@@ -9,8 +9,6 @@ libc_bionic_src_files_x86_64 += \
     bionic/__memset_chk.cpp \
     bionic/__strcpy_chk.cpp \
     bionic/__strcat_chk.cpp \
-    bionic/memchr.c \
-    bionic/memrchr.c \
     bionic/strchr.cpp \
     bionic/strnlen.c \
     bionic/strrchr.cpp \
@@ -25,6 +23,10 @@ libc_freebsd_src_files_x86_64 += \
     upstream-freebsd/lib/libc/string/wmemcmp.c \
     upstream-freebsd/lib/libc/string/wmemmove.c \
 
+libc_openbsd_src_files_x86_64 += \
+    upstream-openbsd/lib/libc/string/memchr.c \
+    upstream-openbsd/lib/libc/string/memrchr.c \
+
 #
 # Inherently architecture-specific code.
 #

libc/bionic/dup2.cpp

@@ -26,8 +26,19 @@
  * SUCH DAMAGE.
  */
 
+#include <fcntl.h>
 #include <unistd.h>
 
 int dup2(int old_fd, int new_fd) {
+  // If old_fd is equal to new_fd and a valid file descriptor, dup2 returns
+  // old_fd without closing it. This is not true of dup3, so we have to
+  // handle this case ourselves.
+  if (old_fd == new_fd) {
+    if (fcntl(old_fd, F_GETFD) == -1) {
+      return -1;
+    }
+    return old_fd;
+  }
+
   return dup3(old_fd, new_fd, 0);
 }

libc/bionic/faccessat.cpp

@@ -0,0 +1,59 @@
+/*
+ * Copyright (C) 2015 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <fcntl.h>
+#include <unistd.h>
+#include <errno.h>
+
+extern "C" int ___faccessat(int, const char*, int);
+
+int faccessat(int dirfd, const char* pathname, int mode, int flags) {
+  // "The mode specifies the accessibility check(s) to be performed,
+  // and is either the value F_OK, or a mask consisting of the
+  // bitwise OR of one or more of R_OK, W_OK, and X_OK."
+  if ((mode != F_OK) && ((mode & ~(R_OK | W_OK | X_OK)) != 0) &&
+      ((mode & (R_OK | W_OK | X_OK)) == 0)) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  if (flags != 0) {
+    // We deliberately don't support AT_SYMLINK_NOFOLLOW, a glibc
+    // only feature which is error prone and dangerous.
+    //
+    // AT_EACCESS isn't supported either. Android doesn't have setuid
+    // programs, and never runs code with euid!=uid. It could be
+    // implemented in an expensive way, following the model at
+    // https://gitlab.com/bminor/musl/commit/0a05eace163cee9b08571d2ff9d90f5e82d9c228
+    // but not worth it.
+    errno = EINVAL;
+    return -1;
+  }
+
+  return ___faccessat(dirfd, pathname, mode);
+}

libc/bionic/libc_init_common.cpp

@@ -88,7 +88,6 @@ void __libc_init_tls(KernelArgumentBlock& args) {
   // The main thread has no mmap allocated space for stack or pthread_internal_t.
   main_thread.mmap_size = 0;
   pthread_attr_init(&main_thread.attr);
-  main_thread.attr.flags = PTHREAD_ATTR_FLAG_MAIN_THREAD;
   main_thread.attr.guard_size = 0; // The main thread has no guard page.
   main_thread.attr.stack_size = 0; // User code should never see this; we'll compute it when asked.
   // TODO: the main thread's sched_policy and sched_priority need to be queried.

libc/bionic/locale.cpp

@@ -58,15 +58,18 @@ struct __locale_t {
   DISALLOW_COPY_AND_ASSIGN(__locale_t);
 };
 
+size_t __ctype_get_mb_cur_max() {
+  locale_t l = uselocale(NULL);
+  if (l == LC_GLOBAL_LOCALE) {
+    return __bionic_current_locale_is_utf8 ? 4 : 1;
+  } else {
+    return l->mb_cur_max;
+  }
+}
+
 static pthread_once_t g_locale_once = PTHREAD_ONCE_INIT;
 static lconv g_locale;
 
-// We don't use pthread_once for this so that we know when the resource (a TLS slot) will be taken.
-static pthread_key_t g_uselocale_key;
-__attribute__((constructor)) static void __bionic_tls_uselocale_key_init() {
-  pthread_key_create(&g_uselocale_key, NULL);
-}
-
 static void __locale_init() {
   g_locale.decimal_point = const_cast<char*>(".");
 
@@ -97,15 +100,6 @@ static void __locale_init() {
   g_locale.int_n_sign_posn = CHAR_MAX;
 }
 
-size_t __ctype_get_mb_cur_max() {
-  locale_t l = reinterpret_cast<locale_t>(pthread_getspecific(g_uselocale_key));
-  if (l == nullptr || l == LC_GLOBAL_LOCALE) {
-    return __bionic_current_locale_is_utf8 ? 4 : 1;
-  } else {
-    return l->mb_cur_max;
-  }
-}
-
 static bool __is_supported_locale(const char* locale) {
   return (strcmp(locale, "") == 0 ||
           strcmp(locale, "C") == 0 ||
@@ -162,7 +156,16 @@ char* setlocale(int category, const char* locale_name) {
   return const_cast<char*>(__bionic_current_locale_is_utf8 ? "C.UTF-8" : "C");
 }
 
+// We can't use a constructor to create g_uselocal_key, because it may be used in constructors.
+static pthread_once_t g_uselocale_once = PTHREAD_ONCE_INIT;
+static pthread_key_t g_uselocale_key;
+
+static void g_uselocale_key_init() {
+  pthread_key_create(&g_uselocale_key, NULL);
+}
+
 locale_t uselocale(locale_t new_locale) {
+  pthread_once(&g_uselocale_once, g_uselocale_key_init);
   locale_t old_locale = static_cast<locale_t>(pthread_getspecific(g_uselocale_key));
 
   // If this is the first call to uselocale(3) on this thread, we return LC_GLOBAL_LOCALE.

libc/bionic/mempcpy.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2008 The Android Open Source Project
+ * Copyright (C) 2015 The Android Open Source Project
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -25,22 +25,9 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
-#include <stddef.h>
+
 #include <string.h>
 
-void *memchr(const void *s, int c, size_t n)
+void* mempcpy(void* dst, const void* src, size_t n) {
-{
+  return reinterpret_cast<char*>(memcpy(dst, src, n)) + n;
-    const unsigned char*  p   = s;
-    const unsigned char*  end = p + n;
-
-    for (;;) {
-        if (p >= end || p[0] == c) break; p++;
-        if (p >= end || p[0] == c) break; p++;
-        if (p >= end || p[0] == c) break; p++;
-        if (p >= end || p[0] == c) break; p++;
-    }
-    if (p >= end)
-        return NULL;
-    else
-        return (void*) p;
 }

libc/bionic/ndk_cruft.cpp

@@ -26,11 +26,11 @@
  * SUCH DAMAGE.
  */
 
-// This file perpetuates the mistakes of the past, but only for 32-bit targets.
+// This file perpetuates the mistakes of the past.
-#if !defined(__LP64__)
 
 #include <ctype.h>
 #include <dirent.h>
+#include <errno.h>
 #include <inttypes.h>
 #include <pthread.h>
 #include <signal.h>
@@ -45,6 +45,11 @@
 #include <unistd.h>
 #include <wchar.h>
 
+#include "private/libc_logging.h"
+
+// The part is only for 32-bit targets.
+#if !defined(__LP64__)
+
 // These were accidentally declared in <unistd.h> because we stupidly used to inline
 // getpagesize() and __getpageshift(). Needed for backwards compatibility with old NDK apps.
 extern "C" {
@@ -341,4 +346,18 @@ extern "C" void* dlmalloc(size_t size) {
   return malloc(size);
 }
 
-#endif
+#endif // !defined(__LP64__)
+
+// This is never implemented in bionic, only needed for ABI compatibility with the NDK.
+extern "C" char* getusershell() {
+  return NULL;
+}
+
+// This is never implemented in bionic, only needed for ABI compatibility with the NDK.
+extern "C" void setusershell() { }
+
+// This is never implemented in bionic, only needed for ABI compatibility with the NDK.
+extern "C" void endusershell() { }
+
+// This is never implemented in bionic, only needed for ABI compatibility with the NDK.
+extern "C" void endpwent() { }

libc/bionic/pthread_attr.cpp

@@ -152,9 +152,6 @@ static int __pthread_attr_getstack_main_thread(void** stack_base, size_t* stack_
 }
 
 int pthread_attr_getstack(const pthread_attr_t* attr, void** stack_base, size_t* stack_size) {
-  if ((attr->flags & PTHREAD_ATTR_FLAG_MAIN_THREAD) != 0) {
-    return __pthread_attr_getstack_main_thread(stack_base, stack_size);
-  }
   *stack_base = attr->stack_base;
   *stack_size = attr->stack_size;
   return 0;
@@ -171,7 +168,13 @@ int pthread_attr_getguardsize(const pthread_attr_t* attr, size_t* guard_size) {
 }
 
 int pthread_getattr_np(pthread_t t, pthread_attr_t* attr) {
-  *attr = reinterpret_cast<pthread_internal_t*>(t)->attr;
+  pthread_internal_t* thread = reinterpret_cast<pthread_internal_t*>(t);
+  *attr = thread->attr;
+  // The main thread's stack information is not stored in thread->attr, and we need to
+  // collect that at runtime.
+  if (thread->tid == getpid()) {
+    return __pthread_attr_getstack_main_thread(&attr->stack_base, &attr->stack_size);
+  }
   return 0;
 }
 

libc/bionic/pthread_cond.cpp

@@ -41,6 +41,13 @@
 #include "private/bionic_time_conversions.h"
 #include "private/bionic_tls.h"
 
+// XXX *technically* there is a race condition that could allow
+// XXX a signal to be missed.  If thread A is preempted in _wait()
+// XXX after unlocking the mutex and before waiting, and if other
+// XXX threads call signal or broadcast UINT_MAX/2 times (exactly),
+// XXX before thread A is scheduled again and calls futex_wait(),
+// XXX then the signal will be lost.
+
 // We use one bit in pthread_condattr_t (long) values as the 'shared' flag
 // and one bit for the clock type (CLOCK_REALTIME is ((clockid_t) 1), and
 // CLOCK_MONOTONIC is ((clockid_t) 0).). The rest of the bits are a counter.
@@ -57,7 +64,6 @@
 #define COND_GET_CLOCK(c) (((c) & COND_CLOCK_MASK) >> 1)
 #define COND_SET_CLOCK(attr, c) ((attr) | (c << 1))
 
-
 int pthread_condattr_init(pthread_condattr_t* attr) {
   *attr = 0;
   *attr |= PTHREAD_PROCESS_PRIVATE;
@@ -98,47 +104,50 @@ int pthread_condattr_destroy(pthread_condattr_t* attr) {
   return 0;
 }
 
-static inline atomic_uint* COND_TO_ATOMIC_POINTER(pthread_cond_t* cond) {
+struct pthread_cond_internal_t {
-  static_assert(sizeof(atomic_uint) == sizeof(cond->value),
+  atomic_uint state;
-                "cond->value should actually be atomic_uint in implementation.");
 
-  // We prefer casting to atomic_uint instead of declaring cond->value to be atomic_uint directly.
+  bool process_shared() {
-  // Because using the second method pollutes pthread.h, and causes an error when compiling libcxx.
+    return COND_IS_SHARED(atomic_load_explicit(&state, memory_order_relaxed));
-  return reinterpret_cast<atomic_uint*>(&cond->value);
+  }
+
+  int get_clock() {
+    return COND_GET_CLOCK(atomic_load_explicit(&state, memory_order_relaxed));
+  }
+
+#if defined(__LP64__)
+  char __reserved[44];
+#endif
+};
+
+static pthread_cond_internal_t* __get_internal_cond(pthread_cond_t* cond_interface) {
+  static_assert(sizeof(pthread_cond_t) == sizeof(pthread_cond_internal_t),
+                "pthread_cond_t should actually be pthread_cond_internal_t in implementation.");
+  return reinterpret_cast<pthread_cond_internal_t*>(cond_interface);
 }
 
-// XXX *technically* there is a race condition that could allow
+int pthread_cond_init(pthread_cond_t* cond_interface, const pthread_condattr_t* attr) {
-// XXX a signal to be missed.  If thread A is preempted in _wait()
+  pthread_cond_internal_t* cond = __get_internal_cond(cond_interface);
-// XXX after unlocking the mutex and before waiting, and if other
-// XXX threads call signal or broadcast UINT_MAX/2 times (exactly),
-// XXX before thread A is scheduled again and calls futex_wait(),
-// XXX then the signal will be lost.
-
-int pthread_cond_init(pthread_cond_t* cond, const pthread_condattr_t* attr) {
-  atomic_uint* cond_value_ptr = COND_TO_ATOMIC_POINTER(cond);
-  unsigned int init_value = 0;
 
+  unsigned int init_state = 0;
   if (attr != NULL) {
-    init_value = (*attr & COND_FLAGS_MASK);
+    init_state = (*attr & COND_FLAGS_MASK);
   }
-  atomic_init(cond_value_ptr, init_value);
+  atomic_init(&cond->state, init_state);
 
   return 0;
 }
 
-int pthread_cond_destroy(pthread_cond_t* cond) {
+int pthread_cond_destroy(pthread_cond_t* cond_interface) {
-  atomic_uint* cond_value_ptr = COND_TO_ATOMIC_POINTER(cond);
+  pthread_cond_internal_t* cond = __get_internal_cond(cond_interface);
-  atomic_store_explicit(cond_value_ptr, 0xdeadc04d, memory_order_relaxed);
+  atomic_store_explicit(&cond->state, 0xdeadc04d, memory_order_relaxed);
   return 0;
 }
 
 // This function is used by pthread_cond_broadcast and
 // pthread_cond_signal to atomically decrement the counter
 // then wake up thread_count threads.
-static int __pthread_cond_pulse(atomic_uint* cond_value_ptr, int thread_count) {
+static int __pthread_cond_pulse(pthread_cond_internal_t* cond, int thread_count) {
-  unsigned int old_value = atomic_load_explicit(cond_value_ptr, memory_order_relaxed);
-  bool shared = COND_IS_SHARED(old_value);
-
   // We don't use a release/seq_cst fence here. Because pthread_cond_wait/signal can't be
   // used as a method for memory synchronization by itself. It should always be used with
   // pthread mutexes. Note that Spurious wakeups from pthread_cond_wait/timedwait may occur,
@@ -149,20 +158,18 @@ static int __pthread_cond_pulse(atomic_uint* cond_value_ptr, int thread_count) {
   // synchronization. And it doesn't help even if we use any fence here.
 
   // The increase of value should leave flags alone, even if the value can overflows.
-  atomic_fetch_add_explicit(cond_value_ptr, COND_COUNTER_STEP, memory_order_relaxed);
+  atomic_fetch_add_explicit(&cond->state, COND_COUNTER_STEP, memory_order_relaxed);
 
-  __futex_wake_ex(cond_value_ptr, shared, thread_count);
+  __futex_wake_ex(&cond->state, cond->process_shared(), thread_count);
   return 0;
 }
 
-__LIBC_HIDDEN__
+static int __pthread_cond_timedwait_relative(pthread_cond_internal_t* cond, pthread_mutex_t* mutex,
-int __pthread_cond_timedwait_relative(atomic_uint* cond_value_ptr, pthread_mutex_t* mutex,
+                                             const timespec* rel_timeout_or_null) {
-                                      const timespec* reltime) {
+  unsigned int old_state = atomic_load_explicit(&cond->state, memory_order_relaxed);
-  unsigned int old_value = atomic_load_explicit(cond_value_ptr, memory_order_relaxed);
-  bool shared = COND_IS_SHARED(old_value);
 
   pthread_mutex_unlock(mutex);
-  int status = __futex_wait_ex(cond_value_ptr, shared, old_value, reltime);
+  int status = __futex_wait_ex(&cond->state, cond->process_shared(), old_state, rel_timeout_or_null);
   pthread_mutex_lock(mutex);
 
   if (status == -ETIMEDOUT) {
@@ -171,67 +178,68 @@ int __pthread_cond_timedwait_relative(atomic_uint* cond_value_ptr, pthread_mutex
   return 0;
 }
 
-__LIBC_HIDDEN__
+static int __pthread_cond_timedwait(pthread_cond_internal_t* cond, pthread_mutex_t* mutex,
-int __pthread_cond_timedwait(atomic_uint* cond_value_ptr, pthread_mutex_t* mutex,
+                                    const timespec* abs_timeout_or_null, clockid_t clock) {
-                             const timespec* abs_ts, clockid_t clock) {
   timespec ts;
-  timespec* tsp;
+  timespec* rel_timeout = NULL;
 
-  if (abs_ts != NULL) {
+  if (abs_timeout_or_null != NULL) {
-    if (!timespec_from_absolute_timespec(ts, *abs_ts, clock)) {
+    rel_timeout = &ts;
+    if (!timespec_from_absolute_timespec(*rel_timeout, *abs_timeout_or_null, clock)) {
       return ETIMEDOUT;
     }
-    tsp = &ts;
-  } else {
-    tsp = NULL;
   }
 
-  return __pthread_cond_timedwait_relative(cond_value_ptr, mutex, tsp);
+  return __pthread_cond_timedwait_relative(cond, mutex, rel_timeout);
 }
 
-int pthread_cond_broadcast(pthread_cond_t* cond) {
+int pthread_cond_broadcast(pthread_cond_t* cond_interface) {
-  atomic_uint* cond_value_ptr = COND_TO_ATOMIC_POINTER(cond);
+  return __pthread_cond_pulse(__get_internal_cond(cond_interface), INT_MAX);
-  return __pthread_cond_pulse(cond_value_ptr, INT_MAX);
 }
 
-int pthread_cond_signal(pthread_cond_t* cond) {
+int pthread_cond_signal(pthread_cond_t* cond_interface) {
-  atomic_uint* cond_value_ptr = COND_TO_ATOMIC_POINTER(cond);
+  return __pthread_cond_pulse(__get_internal_cond(cond_interface), 1);
-  return __pthread_cond_pulse(cond_value_ptr, 1);
 }
 
-int pthread_cond_wait(pthread_cond_t* cond, pthread_mutex_t* mutex) {
+int pthread_cond_wait(pthread_cond_t* cond_interface, pthread_mutex_t* mutex) {
-  atomic_uint* cond_value_ptr = COND_TO_ATOMIC_POINTER(cond);
+  pthread_cond_internal_t* cond = __get_internal_cond(cond_interface);
-  return __pthread_cond_timedwait(cond_value_ptr, mutex, NULL,
+  return __pthread_cond_timedwait(cond, mutex, NULL, cond->get_clock());
-           COND_GET_CLOCK(atomic_load_explicit(cond_value_ptr, memory_order_relaxed)));
 }
 
-int pthread_cond_timedwait(pthread_cond_t *cond, pthread_mutex_t * mutex, const timespec *abstime) {
+int pthread_cond_timedwait(pthread_cond_t *cond_interface, pthread_mutex_t * mutex,
-  atomic_uint* cond_value_ptr = COND_TO_ATOMIC_POINTER(cond);
+                           const timespec *abstime) {
-  return __pthread_cond_timedwait(cond_value_ptr, mutex, abstime,
+
-           COND_GET_CLOCK(atomic_load_explicit(cond_value_ptr, memory_order_relaxed)));
+  pthread_cond_internal_t* cond = __get_internal_cond(cond_interface);
+  return __pthread_cond_timedwait(cond, mutex, abstime, cond->get_clock());
 }
 
 #if !defined(__LP64__)
 // TODO: this exists only for backward binary compatibility on 32 bit platforms.
-extern "C" int pthread_cond_timedwait_monotonic(pthread_cond_t* cond, pthread_mutex_t* mutex, const timespec* abstime) {
+extern "C" int pthread_cond_timedwait_monotonic(pthread_cond_t* cond_interface,
-  atomic_uint* cond_value_ptr = COND_TO_ATOMIC_POINTER(cond);
+                                                pthread_mutex_t* mutex,
-  return __pthread_cond_timedwait(cond_value_ptr, mutex, abstime, CLOCK_MONOTONIC);
+                                                const timespec* abs_timeout) {
+
+  return __pthread_cond_timedwait(__get_internal_cond(cond_interface), mutex, abs_timeout,
+                                  CLOCK_MONOTONIC);
 }
 
-extern "C" int pthread_cond_timedwait_monotonic_np(pthread_cond_t* cond, pthread_mutex_t* mutex, const timespec* abstime) {
+extern "C" int pthread_cond_timedwait_monotonic_np(pthread_cond_t* cond_interface,
-  atomic_uint* cond_value_ptr = COND_TO_ATOMIC_POINTER(cond);
+                                                   pthread_mutex_t* mutex,
-  return __pthread_cond_timedwait(cond_value_ptr, mutex, abstime, CLOCK_MONOTONIC);
+                                                   const timespec* abs_timeout) {
+  return pthread_cond_timedwait_monotonic(cond_interface, mutex, abs_timeout);
 }
 
-extern "C" int pthread_cond_timedwait_relative_np(pthread_cond_t* cond, pthread_mutex_t* mutex, const timespec* reltime) {
+extern "C" int pthread_cond_timedwait_relative_np(pthread_cond_t* cond_interface,
-  atomic_uint* cond_value_ptr = COND_TO_ATOMIC_POINTER(cond);
+                                                  pthread_mutex_t* mutex,
-  return __pthread_cond_timedwait_relative(cond_value_ptr, mutex, reltime);
+                                                  const timespec* rel_timeout) {
+
+  return __pthread_cond_timedwait_relative(__get_internal_cond(cond_interface), mutex, rel_timeout);
 }
 
-extern "C" int pthread_cond_timeout_np(pthread_cond_t* cond, pthread_mutex_t* mutex, unsigned ms) {
+extern "C" int pthread_cond_timeout_np(pthread_cond_t* cond_interface,
+                                       pthread_mutex_t* mutex, unsigned ms) {
   timespec ts;
   timespec_from_ms(ts, ms);
-  atomic_uint* cond_value_ptr = COND_TO_ATOMIC_POINTER(cond);
+  return pthread_cond_timedwait_relative_np(cond_interface, mutex, &ts);
-  return __pthread_cond_timedwait_relative(cond_value_ptr, mutex, &ts);
 }
 #endif // !defined(__LP64__)

libc/bionic/pthread_create.cpp

@@ -56,7 +56,8 @@ void __init_tls(pthread_internal_t* thread) {
   if (thread->mmap_size == 0) {
     // If the TLS area was not allocated by mmap(), it may not have been cleared to zero.
     // So assume the worst and zero the TLS area.
-    memset(&thread->tls[0], 0, BIONIC_TLS_SLOTS * sizeof(void*));
+    memset(thread->tls, 0, sizeof(thread->tls));
+    memset(thread->key_data, 0, sizeof(thread->key_data));
   }
 
   // Slot 0 must point to itself. The x86 Linux kernel reads the TLS from %fs:0.
@@ -155,7 +156,7 @@ static int __allocate_thread(pthread_attr_t* attr, pthread_internal_t** threadp,
   }
 
   // Mapped space(or user allocated stack) is used for:
-  //   thread_internal_t (including tls array)
+  //   thread_internal_t
   //   thread stack (including guard page)
   stack_top -= sizeof(pthread_internal_t);
   pthread_internal_t* thread = reinterpret_cast<pthread_internal_t*>(stack_top);

libc/bionic/pthread_internal.h

@@ -41,8 +41,10 @@
 /* Did the thread exit without freeing pthread_internal_t? */
 #define PTHREAD_ATTR_FLAG_ZOMBIE 0x00000004
 
-/* Is this the main thread? */
+struct pthread_key_data_t {
-#define PTHREAD_ATTR_FLAG_MAIN_THREAD 0x80000000
+  uintptr_t seq; // Use uintptr_t just for alignment, as we use pointer below.
+  void* data;
+};
 
 struct pthread_internal_t {
   struct pthread_internal_t* next;
@@ -86,6 +88,8 @@ struct pthread_internal_t {
 
   void* tls[BIONIC_TLS_SLOTS];
 
+  pthread_key_data_t key_data[BIONIC_PTHREAD_KEY_COUNT];
+
   /*
    * The dynamic linker implements dlerror(3), which makes it hard for us to implement this
    * per-thread buffer by simply using malloc(3) and free(3).

libc/bionic/pthread_key.cpp

@@ -28,175 +28,98 @@
 
 #include <errno.h>
 #include <pthread.h>
+#include <stdatomic.h>
 
 #include "private/bionic_tls.h"
 #include "pthread_internal.h"
 
-/* A technical note regarding our thread-local-storage (TLS) implementation:
- *
- * There can be up to BIONIC_TLS_SLOTS independent TLS keys in a given process,
- * The keys below TLS_SLOT_FIRST_USER_SLOT are reserved for Bionic to hold
- * special thread-specific variables like errno or a pointer to
- * the current thread's descriptor. These entries cannot be accessed through
- * pthread_getspecific() / pthread_setspecific() or pthread_key_delete()
- *
- * The 'tls_map_t' type defined below implements a shared global map of
- * currently created/allocated TLS keys and the destructors associated
- * with them.
- *
- * The global TLS map simply contains a bitmap of allocated keys, and
- * an array of destructors.
- *
- * Each thread has a TLS area that is a simple array of BIONIC_TLS_SLOTS void*
- * pointers. the TLS area of the main thread is stack-allocated in
- * __libc_init_common, while the TLS area of other threads is placed at
- * the top of their stack in pthread_create.
- *
- * When pthread_key_delete() is called it will erase the key's bitmap bit
- * and its destructor, and will also clear the key data in the TLS area of
- * all created threads. As mandated by Posix, it is the responsibility of
- * the caller of pthread_key_delete() to properly reclaim the objects that
- * were pointed to by these data fields (either before or after the call).
- */
-
-#define TLSMAP_BITS       32
-#define TLSMAP_WORDS      ((BIONIC_TLS_SLOTS+TLSMAP_BITS-1)/TLSMAP_BITS)
-#define TLSMAP_WORD(m,k)  (m).map[(k)/TLSMAP_BITS]
-#define TLSMAP_MASK(k)    (1U << ((k)&(TLSMAP_BITS-1)))
-
-static inline bool IsValidUserKey(pthread_key_t key) {
-  return (key >= TLS_SLOT_FIRST_USER_SLOT && key < BIONIC_TLS_SLOTS);
-}
-
 typedef void (*key_destructor_t)(void*);
 
-struct tls_map_t {
+#define SEQ_KEY_IN_USE_BIT     0
-  bool is_initialized;
 
-  /* bitmap of allocated keys */
+#define SEQ_INCREMENT_STEP  (1 << SEQ_KEY_IN_USE_BIT)
-  uint32_t map[TLSMAP_WORDS];
 
-  key_destructor_t key_destructors[BIONIC_TLS_SLOTS];
+// pthread_key_internal_t records the use of each pthread key slot:
+//   seq records the state of the slot.
+//      bit 0 is 1 when the key is in use, 0 when it is unused. Each time we create or delete the
+//      pthread key in the slot, we increse the seq by 1 (which inverts bit 0). The reason to use
+//      a sequence number instead of a boolean value here is that when the key slot is deleted and
+//      reused for a new key, pthread_getspecific will not return stale data.
+//   key_destructor records the destructor called at thread exit.
+struct pthread_key_internal_t {
+  atomic_uintptr_t seq;
+  atomic_uintptr_t key_destructor;
 };
 
-class ScopedTlsMapAccess {
+static pthread_key_internal_t key_map[BIONIC_PTHREAD_KEY_COUNT];
- public:
-  ScopedTlsMapAccess() {
-    Lock();
 
-    // If this is the first time the TLS map has been accessed,
+static inline bool SeqOfKeyInUse(uintptr_t seq) {
-    // mark the slots belonging to well-known keys as being in use.
+  return seq & (1 << SEQ_KEY_IN_USE_BIT);
-    // This isn't currently necessary because the well-known keys
+}
-    // can only be accessed directly by bionic itself, do not have
-    // destructors, and all the functions that touch the TLS map
-    // start after the maximum well-known slot.
-    if (!s_tls_map_.is_initialized) {
-      for (pthread_key_t key = 0; key < TLS_SLOT_FIRST_USER_SLOT; ++key) {
-        SetInUse(key, NULL);
-      }
-      s_tls_map_.is_initialized = true;
-    }
-  }
 
-  ~ScopedTlsMapAccess() {
+static inline bool KeyInValidRange(pthread_key_t key) {
-    Unlock();
+  return key >= 0 && key < BIONIC_PTHREAD_KEY_COUNT;
-  }
+}
-
-  int CreateKey(pthread_key_t* result, void (*key_destructor)(void*)) {
-    // Take the first unallocated key.
-    for (int key = 0; key < BIONIC_TLS_SLOTS; ++key) {
-      if (!IsInUse(key)) {
-        SetInUse(key, key_destructor);
-        *result = key;
-        return 0;
-      }
-    }
-
-    // We hit PTHREAD_KEYS_MAX. POSIX says EAGAIN for this case.
-    return EAGAIN;
-  }
-
-  void DeleteKey(pthread_key_t key) {
-    TLSMAP_WORD(s_tls_map_, key) &= ~TLSMAP_MASK(key);
-    s_tls_map_.key_destructors[key] = NULL;
-  }
-
-  bool IsInUse(pthread_key_t key) {
-    return (TLSMAP_WORD(s_tls_map_, key) & TLSMAP_MASK(key)) != 0;
-  }
-
-  void SetInUse(pthread_key_t key, void (*key_destructor)(void*)) {
-    TLSMAP_WORD(s_tls_map_, key) |= TLSMAP_MASK(key);
-    s_tls_map_.key_destructors[key] = key_destructor;
-  }
-
-  // Called from pthread_exit() to remove all TLS key data
-  // from this thread's TLS area. This must call the destructor of all keys
-  // that have a non-NULL data value and a non-NULL destructor.
-  void CleanAll() {
-    void** tls = __get_tls();
-
-    // Because destructors can do funky things like deleting/creating other
-    // keys, we need to implement this in a loop.
-    for (int rounds = PTHREAD_DESTRUCTOR_ITERATIONS; rounds > 0; --rounds) {
-      size_t called_destructor_count = 0;
-      for (int key = 0; key < BIONIC_TLS_SLOTS; ++key) {
-        if (IsInUse(key)) {
-          void* data = tls[key];
-          void (*key_destructor)(void*) = s_tls_map_.key_destructors[key];
-
-          if (data != NULL && key_destructor != NULL) {
-            // we need to clear the key data now, this will prevent the
-            // destructor (or a later one) from seeing the old value if
-            // it calls pthread_getspecific() for some odd reason
-
-            // we do not do this if 'key_destructor == NULL' just in case another
-            // destructor function might be responsible for manually
-            // releasing the corresponding data.
-            tls[key] = NULL;
-
-            // because the destructor is free to call pthread_key_create
-            // and/or pthread_key_delete, we need to temporarily unlock
-            // the TLS map
-            Unlock();
-            (*key_destructor)(data);
-            Lock();
-            ++called_destructor_count;
-          }
-        }
-      }
-
-      // If we didn't call any destructors, there is no need to check the TLS data again.
-      if (called_destructor_count == 0) {
-        break;
-      }
-    }
-  }
-
- private:
-  static tls_map_t s_tls_map_;
-  static pthread_mutex_t s_tls_map_lock_;
-
-  void Lock() {
-    pthread_mutex_lock(&s_tls_map_lock_);
-  }
-
-  void Unlock() {
-    pthread_mutex_unlock(&s_tls_map_lock_);
-  }
-};
-
-__LIBC_HIDDEN__ tls_map_t ScopedTlsMapAccess::s_tls_map_;
-__LIBC_HIDDEN__ pthread_mutex_t ScopedTlsMapAccess::s_tls_map_lock_;
 
+// Called from pthread_exit() to remove all pthread keys. This must call the destructor of
+// all keys that have a non-NULL data value and a non-NULL destructor.
 __LIBC_HIDDEN__ void pthread_key_clean_all() {
-  ScopedTlsMapAccess tls_map;
+  // Because destructors can do funky things like deleting/creating other keys,
-  tls_map.CleanAll();
+  // we need to implement this in a loop.
+  pthread_key_data_t* key_data = __get_thread()->key_data;
+  for (size_t rounds = PTHREAD_DESTRUCTOR_ITERATIONS; rounds > 0; --rounds) {
+    size_t called_destructor_count = 0;
+    for (size_t i = 0; i < BIONIC_PTHREAD_KEY_COUNT; ++i) {
+      uintptr_t seq = atomic_load_explicit(&key_map[i].seq, memory_order_relaxed);
+      if (SeqOfKeyInUse(seq) && seq == key_data[i].seq && key_data[i].data != NULL) {
+        // Other threads may be calling pthread_key_delete/pthread_key_create while current thread
+        // is exiting. So we need to ensure we read the right key_destructor.
+        // We can rely on a user-established happens-before relationship between the creation and
+        // use of pthread key to ensure that we're not getting an earlier key_destructor.
+        // To avoid using the key_destructor of the newly created key in the same slot, we need to
+        // recheck the sequence number after reading key_destructor. As a result, we either see the
+        // right key_destructor, or the sequence number must have changed when we reread it below.
+        key_destructor_t key_destructor = reinterpret_cast<key_destructor_t>(
+          atomic_load_explicit(&key_map[i].key_destructor, memory_order_relaxed));
+        if (key_destructor == NULL) {
+          continue;
+        }
+        atomic_thread_fence(memory_order_acquire);
+        if (atomic_load_explicit(&key_map[i].seq, memory_order_relaxed) != seq) {
+           continue;
+        }
+
+        // We need to clear the key data now, this will prevent the destructor (or a later one)
+        // from seeing the old value if it calls pthread_getspecific().
+        // We don't do this if 'key_destructor == NULL' just in case another destructor
+        // function is responsible for manually releasing the corresponding data.
+        void* data = key_data[i].data;
+        key_data[i].data = NULL;
+
+        (*key_destructor)(data);
+        ++called_destructor_count;
+      }
+    }
+
+    // If we didn't call any destructors, there is no need to check the pthread keys again.
+    if (called_destructor_count == 0) {
+      break;
+    }
+  }
 }
 
 int pthread_key_create(pthread_key_t* key, void (*key_destructor)(void*)) {
-  ScopedTlsMapAccess tls_map;
+  for (size_t i = 0; i < BIONIC_PTHREAD_KEY_COUNT; ++i) {
-  return tls_map.CreateKey(key, key_destructor);
+    uintptr_t seq = atomic_load_explicit(&key_map[i].seq, memory_order_relaxed);
+    while (!SeqOfKeyInUse(seq)) {
+      if (atomic_compare_exchange_weak(&key_map[i].seq, &seq, seq + SEQ_INCREMENT_STEP)) {
+        atomic_store(&key_map[i].key_destructor, reinterpret_cast<uintptr_t>(key_destructor));
+        *key = i;
+        return 0;
+      }
+    }
+  }
+  return EAGAIN;
 }
 
 // Deletes a pthread_key_t. note that the standard mandates that this does
@@ -204,42 +127,44 @@ int pthread_key_create(pthread_key_t* key, void (*key_destructor)(void*)) {
 // responsibility of the caller to properly dispose of the corresponding data
 // and resources, using any means it finds suitable.
 int pthread_key_delete(pthread_key_t key) {
-  ScopedTlsMapAccess tls_map;
+  if (!KeyInValidRange(key)) {
-
-  if (!IsValidUserKey(key) || !tls_map.IsInUse(key)) {
     return EINVAL;
   }
-
+  // Increase seq to invalidate values in all threads.
-  // Clear value in all threads.
+  uintptr_t seq = atomic_load_explicit(&key_map[key].seq, memory_order_relaxed);
-  pthread_mutex_lock(&g_thread_list_lock);
+  if (SeqOfKeyInUse(seq)) {
-  for (pthread_internal_t*  t = g_thread_list; t != NULL; t = t->next) {
+    if (atomic_compare_exchange_strong(&key_map[key].seq, &seq, seq + SEQ_INCREMENT_STEP)) {
-    t->tls[key] = NULL;
+      return 0;
+    }
   }
-  tls_map.DeleteKey(key);
+  return EINVAL;
-
-  pthread_mutex_unlock(&g_thread_list_lock);
-  return 0;
 }
 
 void* pthread_getspecific(pthread_key_t key) {
-  if (!IsValidUserKey(key)) {
+  if (!KeyInValidRange(key)) {
     return NULL;
   }
-
+  uintptr_t seq = atomic_load_explicit(&key_map[key].seq, memory_order_relaxed);
-  // For performance reasons, we do not lock/unlock the global TLS map
+  pthread_key_data_t* data = &(__get_thread()->key_data[key]);
-  // to check that the key is properly allocated. If the key was not
+  // It is user's responsibility to synchornize between the creation and use of pthread keys,
-  // allocated, the value read from the TLS should always be NULL
+  // so we use memory_order_relaxed when checking the sequence number.
-  // due to pthread_key_delete() clearing the values for all threads.
+  if (__predict_true(SeqOfKeyInUse(seq) && data->seq == seq)) {
-  return __get_tls()[key];
+    return data->data;
+  }
+  data->data = NULL;
+  return NULL;
 }
 
 int pthread_setspecific(pthread_key_t key, const void* ptr) {
-  ScopedTlsMapAccess tls_map;
+  if (!KeyInValidRange(key)) {
-
-  if (!IsValidUserKey(key) || !tls_map.IsInUse(key)) {
     return EINVAL;
   }
-
+  uintptr_t seq = atomic_load_explicit(&key_map[key].seq, memory_order_relaxed);
-  __get_tls()[key] = const_cast<void*>(ptr);
+  if (SeqOfKeyInUse(seq)) {
-  return 0;
+    pthread_key_data_t* data = &(__get_thread()->key_data[key]);
+    data->seq = seq;
+    data->data = const_cast<void*>(ptr);
+    return 0;
+  }
+  return EINVAL;
 }

libc/bionic/pthread_rwlock.cpp

@@ -27,6 +27,7 @@
  */
 
 #include <errno.h>
+#include <stdatomic.h>
 
 #include "pthread_internal.h"
 #include "private/bionic_futex.h"
@@ -52,11 +53,6 @@
  *  - This implementation will return EDEADLK in "write after write" and "read after
  *    write" cases and will deadlock in write after read case.
  *
- * TODO: VERY CAREFULLY convert this to use C++11 atomics when possible. All volatile
- * members of pthread_rwlock_t should be converted to atomics<> and __sync_bool_compare_and_swap
- * should be changed to compare_exchange_strong accompanied by the proper ordering
- * constraints (comments have been added with the intending ordering across the code).
- *
  * TODO: As it stands now, pending_readers and pending_writers could be merged into a
  * a single waiters variable.  Keeping them separate adds a bit of clarity and keeps
  * the door open for a writer-biased implementation.
@@ -66,18 +62,6 @@
 #define RWLOCKATTR_DEFAULT     0
 #define RWLOCKATTR_SHARED_MASK 0x0010
 
-static inline bool rwlock_is_shared(const pthread_rwlock_t* rwlock) {
-  return rwlock->attr == PTHREAD_PROCESS_SHARED;
-}
-
-static bool timespec_from_absolute(timespec* rel_timeout, const timespec* abs_timeout) {
-  if (abs_timeout != NULL) {
-    if (!timespec_from_absolute_timespec(*rel_timeout, *abs_timeout, CLOCK_REALTIME)) {
-      return false;
-    }
-  }
-  return true;
-}
 
 int pthread_rwlockattr_init(pthread_rwlockattr_t* attr) {
   *attr = PTHREAD_PROCESS_PRIVATE;
@@ -105,8 +89,36 @@ int pthread_rwlockattr_getpshared(const pthread_rwlockattr_t* attr, int* pshared
   return 0;
 }
 
-int pthread_rwlock_init(pthread_rwlock_t* rwlock, const pthread_rwlockattr_t* attr) {
+struct pthread_rwlock_internal_t {
-  if (attr != NULL) {
+  atomic_int state; // 0=unlock, -1=writer lock, +n=reader lock
+  atomic_int writer_thread_id;
+  atomic_uint pending_readers;
+  atomic_uint pending_writers;
+  int32_t attr;
+
+  bool process_shared() const {
+    return attr == PTHREAD_PROCESS_SHARED;
+  }
+
+#if defined(__LP64__)
+  char __reserved[36];
+#else
+  char __reserved[20];
+#endif
+};
+
+static inline pthread_rwlock_internal_t* __get_internal_rwlock(pthread_rwlock_t* rwlock_interface) {
+  static_assert(sizeof(pthread_rwlock_t) == sizeof(pthread_rwlock_internal_t),
+                "pthread_rwlock_t should actually be pthread_rwlock_internal_t in implementation.");
+  return reinterpret_cast<pthread_rwlock_internal_t*>(rwlock_interface);
+}
+
+int pthread_rwlock_init(pthread_rwlock_t* rwlock_interface, const pthread_rwlockattr_t* attr) {
+  pthread_rwlock_internal_t* rwlock = __get_internal_rwlock(rwlock_interface);
+
+  if (__predict_true(attr == NULL)) {
+    rwlock->attr = 0;
+  } else {
     switch (*attr) {
       case PTHREAD_PROCESS_SHARED:
       case PTHREAD_PROCESS_PRIVATE:
@@ -117,165 +129,214 @@ int pthread_rwlock_init(pthread_rwlock_t* rwlock, const pthread_rwlockattr_t* at
     }
   }
 
-  rwlock->state = 0;
+  atomic_init(&rwlock->state, 0);
-  rwlock->pending_readers = 0;
+  atomic_init(&rwlock->writer_thread_id, 0);
-  rwlock->pending_writers = 0;
+  atomic_init(&rwlock->pending_readers, 0);
-  rwlock->writer_thread_id = 0;
+  atomic_init(&rwlock->pending_writers, 0);
 
   return 0;
 }
 
-int pthread_rwlock_destroy(pthread_rwlock_t* rwlock) {
+int pthread_rwlock_destroy(pthread_rwlock_t* rwlock_interface) {
-  if (rwlock->state != 0) {
+  pthread_rwlock_internal_t* rwlock = __get_internal_rwlock(rwlock_interface);
+
+  if (atomic_load_explicit(&rwlock->state, memory_order_relaxed) != 0) {
     return EBUSY;
   }
   return 0;
 }
 
-static int __pthread_rwlock_timedrdlock(pthread_rwlock_t* rwlock, const timespec* abs_timeout) {
+static int __pthread_rwlock_timedrdlock(pthread_rwlock_internal_t* rwlock,
-  if (__predict_false(__get_thread()->tid == rwlock->writer_thread_id)) {
+                                        const timespec* abs_timeout_or_null) {
+
+  if (__predict_false(__get_thread()->tid == atomic_load_explicit(&rwlock->writer_thread_id,
+                                                                  memory_order_relaxed))) {
     return EDEADLK;
   }
 
-  timespec ts;
+  while (true) {
-  timespec* rel_timeout = (abs_timeout == NULL) ? NULL : &ts;
+    int old_state = atomic_load_explicit(&rwlock->state, memory_order_relaxed);
-  bool done = false;
+    if (__predict_true(old_state >= 0)) {
-  do {
+      if (atomic_compare_exchange_weak_explicit(&rwlock->state, &old_state, old_state + 1,
-    // This is actually a race read as there's nothing that guarantees the atomicity of integer
+                                                memory_order_acquire, memory_order_relaxed)) {
-    // reads / writes. However, in practice this "never" happens so until we switch to C++11 this
+        return 0;
-    // should work fine. The same applies in the other places this idiom is used.
-    int32_t cur_state = rwlock->state;  // C++11 relaxed atomic read
-    if (__predict_true(cur_state >= 0)) {
-      // Add as an extra reader.
-      done = __sync_bool_compare_and_swap(&rwlock->state, cur_state, cur_state + 1);  // C++11 memory_order_aquire
-    } else {
-      if (!timespec_from_absolute(rel_timeout, abs_timeout)) {
-        return ETIMEDOUT;
       }
-      // Owner holds it in write mode, hang up.
+    } else {
-      // To avoid losing wake ups the pending_readers update and the state read should be
+      timespec ts;
-      // sequentially consistent. (currently enforced by __sync_fetch_and_add which creates a full barrier)
+      timespec* rel_timeout = NULL;
-      __sync_fetch_and_add(&rwlock->pending_readers, 1);  // C++11 memory_order_relaxed (if the futex_wait ensures the ordering)
+
-      int ret = __futex_wait_ex(&rwlock->state, rwlock_is_shared(rwlock), cur_state, rel_timeout);
+      if (abs_timeout_or_null != NULL) {
-      __sync_fetch_and_sub(&rwlock->pending_readers, 1);  // C++11 memory_order_relaxed
+        rel_timeout = &ts;
+        if (!timespec_from_absolute_timespec(*rel_timeout, *abs_timeout_or_null, CLOCK_REALTIME)) {
+          return ETIMEDOUT;
+        }
+      }
+
+      // To avoid losing wake ups, the pending_readers increment should be observed before
+      // futex_wait by all threads. A seq_cst fence instead of a seq_cst operation is used
+      // here. Because only a seq_cst fence can ensure sequential consistency for non-atomic
+      // operations in futex_wait.
+      atomic_fetch_add_explicit(&rwlock->pending_readers, 1, memory_order_relaxed);
+
+      atomic_thread_fence(memory_order_seq_cst);
+
+      int ret = __futex_wait_ex(&rwlock->state, rwlock->process_shared(), old_state,
+                                rel_timeout);
+
+      atomic_fetch_sub_explicit(&rwlock->pending_readers, 1, memory_order_relaxed);
+
       if (ret == -ETIMEDOUT) {
         return ETIMEDOUT;
       }
     }
-  } while (!done);
+  }
-
-  return 0;
 }
 
-static int __pthread_rwlock_timedwrlock(pthread_rwlock_t* rwlock, const timespec* abs_timeout) {
+static int __pthread_rwlock_timedwrlock(pthread_rwlock_internal_t* rwlock,
-  int tid = __get_thread()->tid;
+                                        const timespec* abs_timeout_or_null) {
-  if (__predict_false(tid == rwlock->writer_thread_id)) {
+
+  if (__predict_false(__get_thread()->tid == atomic_load_explicit(&rwlock->writer_thread_id,
+                                                                  memory_order_relaxed))) {
     return EDEADLK;
   }
 
-  timespec ts;
+  while (true) {
-  timespec* rel_timeout = (abs_timeout == NULL) ? NULL : &ts;
+    int old_state = atomic_load_explicit(&rwlock->state, memory_order_relaxed);
-  bool done = false;
+    if (__predict_true(old_state == 0)) {
-  do {
+      if (atomic_compare_exchange_weak_explicit(&rwlock->state, &old_state, -1,
-    int32_t cur_state = rwlock->state;
+                                                memory_order_acquire, memory_order_relaxed)) {
-    if (__predict_true(cur_state == 0)) {
+        // writer_thread_id is protected by rwlock and can only be modified in rwlock write
-      // Change state from 0 to -1.
+        // owner thread. Other threads may read it for EDEADLK error checking, atomic operation
-      done =  __sync_bool_compare_and_swap(&rwlock->state, 0 /* cur state */, -1 /* new state */);  // C++11 memory_order_aquire
+        // is safe enough for it.
-    } else {
+        atomic_store_explicit(&rwlock->writer_thread_id, __get_thread()->tid, memory_order_relaxed);
-      if (!timespec_from_absolute(rel_timeout, abs_timeout)) {
+        return 0;
-        return ETIMEDOUT;
       }
-      // Failed to acquire, hang up.
+    } else {
-      // To avoid losing wake ups the pending_writers update and the state read should be
+      timespec ts;
-      // sequentially consistent. (currently enforced by __sync_fetch_and_add which creates a full barrier)
+      timespec* rel_timeout = NULL;
-      __sync_fetch_and_add(&rwlock->pending_writers, 1);  // C++11 memory_order_relaxed (if the futex_wait ensures the ordering)
+
-      int ret = __futex_wait_ex(&rwlock->state, rwlock_is_shared(rwlock), cur_state, rel_timeout);
+      if (abs_timeout_or_null != NULL) {
-      __sync_fetch_and_sub(&rwlock->pending_writers, 1);  // C++11 memory_order_relaxed
+        rel_timeout = &ts;
+        if (!timespec_from_absolute_timespec(*rel_timeout, *abs_timeout_or_null, CLOCK_REALTIME)) {
+          return ETIMEDOUT;
+        }
+      }
+
+      // To avoid losing wake ups, the pending_writers increment should be observed before
+      // futex_wait by all threads. A seq_cst fence instead of a seq_cst operation is used
+      // here. Because only a seq_cst fence can ensure sequential consistency for non-atomic
+      // operations in futex_wait.
+      atomic_fetch_add_explicit(&rwlock->pending_writers, 1, memory_order_relaxed);
+
+      atomic_thread_fence(memory_order_seq_cst);
+
+      int ret = __futex_wait_ex(&rwlock->state, rwlock->process_shared(), old_state,
+                                rel_timeout);
+
+      atomic_fetch_sub_explicit(&rwlock->pending_writers, 1, memory_order_relaxed);
+
       if (ret == -ETIMEDOUT) {
         return ETIMEDOUT;
       }
     }
-  } while (!done);
+  }
-
-  rwlock->writer_thread_id = tid;
-  return 0;
 }
 
-int pthread_rwlock_rdlock(pthread_rwlock_t* rwlock) {
+int pthread_rwlock_rdlock(pthread_rwlock_t* rwlock_interface) {
+  pthread_rwlock_internal_t* rwlock = __get_internal_rwlock(rwlock_interface);
+
   return __pthread_rwlock_timedrdlock(rwlock, NULL);
 }
 
-int pthread_rwlock_timedrdlock(pthread_rwlock_t* rwlock, const timespec* abs_timeout) {
+int pthread_rwlock_timedrdlock(pthread_rwlock_t* rwlock_interface, const timespec* abs_timeout) {
+  pthread_rwlock_internal_t* rwlock = __get_internal_rwlock(rwlock_interface);
+
   return __pthread_rwlock_timedrdlock(rwlock, abs_timeout);
 }
 
-int pthread_rwlock_tryrdlock(pthread_rwlock_t* rwlock) {
+int pthread_rwlock_tryrdlock(pthread_rwlock_t* rwlock_interface) {
-  int32_t cur_state = rwlock->state;
+  pthread_rwlock_internal_t* rwlock = __get_internal_rwlock(rwlock_interface);
-  if ((cur_state >= 0) &&
+
-      __sync_bool_compare_and_swap(&rwlock->state, cur_state, cur_state + 1)) {  // C++11 memory_order_acquire
+  int old_state = atomic_load_explicit(&rwlock->state, memory_order_relaxed);
-    return 0;
+
+  while (old_state >= 0 && !atomic_compare_exchange_weak_explicit(&rwlock->state, &old_state,
+                             old_state + 1, memory_order_acquire, memory_order_relaxed)) {
   }
-  return EBUSY;
+  return (old_state >= 0) ? 0 : EBUSY;
 }
 
-int pthread_rwlock_wrlock(pthread_rwlock_t* rwlock) {
+int pthread_rwlock_wrlock(pthread_rwlock_t* rwlock_interface) {
+  pthread_rwlock_internal_t* rwlock = __get_internal_rwlock(rwlock_interface);
+
   return __pthread_rwlock_timedwrlock(rwlock, NULL);
 }
 
-int pthread_rwlock_timedwrlock(pthread_rwlock_t* rwlock, const timespec* abs_timeout) {
+int pthread_rwlock_timedwrlock(pthread_rwlock_t* rwlock_interface, const timespec* abs_timeout) {
+  pthread_rwlock_internal_t* rwlock = __get_internal_rwlock(rwlock_interface);
+
   return __pthread_rwlock_timedwrlock(rwlock, abs_timeout);
 }
 
-int pthread_rwlock_trywrlock(pthread_rwlock_t* rwlock) {
+int pthread_rwlock_trywrlock(pthread_rwlock_t* rwlock_interface) {
-  int tid = __get_thread()->tid;
+  pthread_rwlock_internal_t* rwlock = __get_internal_rwlock(rwlock_interface);
-  int32_t cur_state = rwlock->state;
+
-  if ((cur_state == 0) &&
+  int old_state = atomic_load_explicit(&rwlock->state, memory_order_relaxed);
-      __sync_bool_compare_and_swap(&rwlock->state, 0 /* cur state */, -1 /* new state */)) {  // C++11 memory_order_acquire
+
-    rwlock->writer_thread_id = tid;
+  while (old_state == 0 && !atomic_compare_exchange_weak_explicit(&rwlock->state, &old_state, -1,
+                                              memory_order_acquire, memory_order_relaxed)) {
+  }
+  if (old_state == 0) {
+    atomic_store_explicit(&rwlock->writer_thread_id, __get_thread()->tid, memory_order_relaxed);
     return 0;
   }
   return EBUSY;
 }
 
 
-int pthread_rwlock_unlock(pthread_rwlock_t* rwlock) {
+int pthread_rwlock_unlock(pthread_rwlock_t* rwlock_interface) {
-  int tid = __get_thread()->tid;
+  pthread_rwlock_internal_t* rwlock = __get_internal_rwlock(rwlock_interface);
-  bool done = false;
+
-  do {
+  int old_state = atomic_load_explicit(&rwlock->state, memory_order_relaxed);
-    int32_t cur_state = rwlock->state;
+  if (__predict_false(old_state == 0)) {
-    if (cur_state == 0) {
+    return EPERM;
+  } else if (old_state == -1) {
+    if (atomic_load_explicit(&rwlock->writer_thread_id, memory_order_relaxed) != __get_thread()->tid) {
       return EPERM;
     }
-    if (cur_state == -1) {
+    // We're no longer the owner.
-      if (rwlock->writer_thread_id != tid) {
+    atomic_store_explicit(&rwlock->writer_thread_id, 0, memory_order_relaxed);
-        return EPERM;
+    // Change state from -1 to 0.
-      }
+    atomic_store_explicit(&rwlock->state, 0, memory_order_release);
-      // We're no longer the owner.
-      rwlock->writer_thread_id = 0;
-      // Change state from -1 to 0.
-      // We use __sync_bool_compare_and_swap to achieve sequential consistency of the state store and
-      // the following pendingX loads. A simple store with memory_order_release semantics
-      // is not enough to guarantee that the pendingX loads are not reordered before the
-      // store (which may lead to a lost wakeup).
-      __sync_bool_compare_and_swap( &rwlock->state, -1 /* cur state*/, 0 /* new state */);  // C++11 maybe memory_order_seq_cst?
 
-      // Wake any waiters.
+  } else { // old_state > 0
-      if (__predict_false(rwlock->pending_readers > 0 || rwlock->pending_writers > 0)) {
+    // Reduce state by 1.
-        __futex_wake_ex(&rwlock->state, rwlock_is_shared(rwlock), INT_MAX);
+    while (old_state > 0 && !atomic_compare_exchange_weak_explicit(&rwlock->state, &old_state,
-      }
+                               old_state - 1, memory_order_release, memory_order_relaxed)) {
-      done = true;
-    } else { // cur_state > 0
-      // Reduce state by 1.
-      // See the comment above on why we need __sync_bool_compare_and_swap.
-      done = __sync_bool_compare_and_swap(&rwlock->state, cur_state, cur_state - 1);  // C++11 maybe memory_order_seq_cst?
-      if (done && (cur_state - 1) == 0) {
-        // There are no more readers, wake any waiters.
-        if (__predict_false(rwlock->pending_readers > 0 || rwlock->pending_writers > 0)) {
-          __futex_wake_ex(&rwlock->state, rwlock_is_shared(rwlock), INT_MAX);
-        }
-      }
     }
-  } while (!done);
 
+    if (old_state <= 0) {
+      return EPERM;
+    } else if (old_state > 1) {
+      return 0;
+    }
+    // old_state = 1, which means the last reader calling unlock. It has to wake up waiters.
+  }
+
+  // If having waiters, wake up them.
+  // To avoid losing wake ups, the update of state should be observed before reading
+  // pending_readers/pending_writers by all threads. Use read locking as an example:
+  //     read locking thread                        unlocking thread
+  //      pending_readers++;                         state = 0;
+  //      seq_cst fence                              seq_cst fence
+  //      read state for futex_wait                  read pending_readers for futex_wake
+  //
+  // So when locking and unlocking threads are running in parallel, we will not get
+  // in a situation that the locking thread reads state as negative and needs to wait,
+  // while the unlocking thread reads pending_readers as zero and doesn't need to wake up waiters.
+  atomic_thread_fence(memory_order_seq_cst);
+  if (__predict_false(atomic_load_explicit(&rwlock->pending_readers, memory_order_relaxed) > 0 ||
+                      atomic_load_explicit(&rwlock->pending_writers, memory_order_relaxed) > 0)) {
+    __futex_wake_ex(&rwlock->state, rwlock->process_shared(), INT_MAX);
+  }
   return 0;
 }

libc/bionic/stubs.cpp

@@ -446,31 +446,6 @@ protoent* getprotobynumber(int /*proto*/) {
   return NULL;
 }
 
-static void unimplemented_stub(const char* function) {
-  const char* fmt = "%s(3) is not implemented on Android\n";
-  __libc_format_log(ANDROID_LOG_WARN, "libc", fmt, function);
-  fprintf(stderr, fmt, function);
-}
-
-#define UNIMPLEMENTED unimplemented_stub(__PRETTY_FUNCTION__)
-
-void endpwent() {
-  UNIMPLEMENTED;
-}
-
-char* getusershell() {
-  UNIMPLEMENTED;
-  return NULL;
-}
-
-void setusershell() {
-  UNIMPLEMENTED;
-}
-
-void endusershell() {
-  UNIMPLEMENTED;
-}
-
 // Portable code should use sysconf(_SC_PAGE_SIZE) directly instead.
 int getpagesize() {
   // We dont use sysconf(3) here because that drags in stdio, which makes static binaries fat.

libc/bionic/sysconf.cpp

@@ -33,6 +33,7 @@
 #include <pthread.h>
 #include <stdio.h>  // For FOPEN_MAX.
 #include <sys/auxv.h>
+#include <sys/resource.h>
 #include <sys/sysconf.h>
 #include <sys/sysinfo.h>
 #include <time.h>
@@ -50,6 +51,12 @@ static bool __sysconf_has_clock(clockid_t clock_id) {
   return clock_getres(clock_id, NULL) == 0;
 }
 
+static long __sysconf_rlimit(int resource) {
+  rlimit rl;
+  getrlimit(resource, &rl);
+  return rl.rlim_cur;
+}
+
 long sysconf(int name) {
   switch (name) {
     case _SC_ARG_MAX:           return ARG_MAX;
@@ -57,13 +64,13 @@ long sysconf(int name) {
     case _SC_BC_DIM_MAX:        return _POSIX2_BC_DIM_MAX;    // Minimum requirement.
     case _SC_BC_SCALE_MAX:      return _POSIX2_BC_SCALE_MAX;  // Minimum requirement.
     case _SC_BC_STRING_MAX:     return _POSIX2_BC_STRING_MAX; // Minimum requirement.
-    case _SC_CHILD_MAX:         return CHILD_MAX;
+    case _SC_CHILD_MAX:         return __sysconf_rlimit(RLIMIT_NPROC);
     case _SC_CLK_TCK:           return static_cast<long>(getauxval(AT_CLKTCK));
     case _SC_COLL_WEIGHTS_MAX:  return _POSIX2_COLL_WEIGHTS_MAX;  // Minimum requirement.
     case _SC_EXPR_NEST_MAX:     return _POSIX2_EXPR_NEST_MAX;     // Minimum requirement.
     case _SC_LINE_MAX:          return _POSIX2_LINE_MAX;          // Minimum requirement.
     case _SC_NGROUPS_MAX:       return NGROUPS_MAX;
-    case _SC_OPEN_MAX:          return OPEN_MAX;
+    case _SC_OPEN_MAX:          return __sysconf_rlimit(RLIMIT_NOFILE);
     case _SC_PASS_MAX:          return PASS_MAX;
     case _SC_2_C_BIND:          return _POSIX2_C_BIND;
     case _SC_2_C_DEV:           return _POSIX2_C_DEV;

libc/bionic/system_properties.cpp

@@ -51,7 +51,6 @@
 #include <sys/_system_properties.h>
 #include <sys/system_properties.h>
 
-#include "private/bionic_atomic_inline.h"
 #include "private/bionic_futex.h"
 #include "private/bionic_macros.h"
 
@@ -80,22 +79,26 @@ struct prop_bt {
     uint8_t namelen;
     uint8_t reserved[3];
 
-    // TODO: The following fields should be declared as atomic_uint32_t.
+    // The property trie is updated only by the init process (single threaded) which provides
-    // They should be assigned to with release semantics, instead of using
+    // property service. And it can be read by multiple threads at the same time.
-    // explicit fences.  Unfortunately, the read accesses are generally
+    // As the property trie is not protected by locks, we use atomic_uint_least32_t types for the
-    // followed by more dependent read accesses, and the dependence
+    // left, right, children "pointers" in the trie node. To make sure readers who see the
-    // is assumed to enforce memory ordering.  Which it does on supported
+    // change of "pointers" can also notice the change of prop_bt structure contents pointed by
-    // hardware.  This technically should use memory_order_consume, if
+    // the "pointers", we always use release-consume ordering pair when accessing these "pointers".
-    // that worked as intended.
+
+    // prop "points" to prop_info structure if there is a propery associated with the trie node.
+    // Its situation is similar to the left, right, children "pointers". So we use
+    // atomic_uint_least32_t and release-consume ordering to protect it as well.
+
     // We should also avoid rereading these fields redundantly, since not
     // all processor implementations ensure that multiple loads from the
     // same field are carried out in the right order.
-    volatile uint32_t prop;
+    atomic_uint_least32_t prop;
 
-    volatile uint32_t left;
+    atomic_uint_least32_t left;
-    volatile uint32_t right;
+    atomic_uint_least32_t right;
 
-    volatile uint32_t children;
+    atomic_uint_least32_t children;
 
     char name[0];
 
@@ -103,8 +106,6 @@ struct prop_bt {
         this->namelen = name_length;
         memcpy(this->name, name, name_length);
         this->name[name_length] = '\0';
-        ANDROID_MEMBAR_FULL();  // TODO: Instead use a release store
-                                // for subsequent pointer assignment.
     }
 
 private:
@@ -143,8 +144,6 @@ struct prop_info {
         atomic_init(&this->serial, valuelen << 24);
         memcpy(this->value, value, valuelen);
         this->value[valuelen] = '\0';
-        ANDROID_MEMBAR_FULL();  // TODO: Instead use a release store
-                                // for subsequent point assignment.
     }
 private:
     DISALLOW_COPY_AND_ASSIGN(prop_info);
@@ -291,10 +290,10 @@ static int map_prop_area()
     return map_result;
 }
 
-static void *allocate_obj(const size_t size, uint32_t *const off)
+static void *allocate_obj(const size_t size, uint_least32_t *const off)
 {
     prop_area *pa = __system_property_area__;
-    const size_t aligned = BIONIC_ALIGN(size, sizeof(uint32_t));
+    const size_t aligned = BIONIC_ALIGN(size, sizeof(uint_least32_t));
     if (pa->bytes_used + aligned > pa_data_size) {
         return NULL;
     }
@@ -304,12 +303,12 @@ static void *allocate_obj(const size_t size, uint32_t *const off)
     return pa->data + *off;
 }
 
-static prop_bt *new_prop_bt(const char *name, uint8_t namelen, uint32_t *const off)
+static prop_bt *new_prop_bt(const char *name, uint8_t namelen, uint_least32_t *const off)
 {
-    uint32_t new_offset;
+    uint_least32_t new_offset;
-    void *const offset = allocate_obj(sizeof(prop_bt) + namelen + 1, &new_offset);
+    void *const p = allocate_obj(sizeof(prop_bt) + namelen + 1, &new_offset);
-    if (offset) {
+    if (p != NULL) {
-        prop_bt* bt = new(offset) prop_bt(name, namelen);
+        prop_bt* bt = new(p) prop_bt(name, namelen);
         *off = new_offset;
         return bt;
     }
@@ -318,20 +317,20 @@ static prop_bt *new_prop_bt(const char *name, uint8_t namelen, uint32_t *const o
 }
 
 static prop_info *new_prop_info(const char *name, uint8_t namelen,
-        const char *value, uint8_t valuelen, uint32_t *const off)
+        const char *value, uint8_t valuelen, uint_least32_t *const off)
 {
-    uint32_t off_tmp;
+    uint_least32_t new_offset;
-    void* const offset = allocate_obj(sizeof(prop_info) + namelen + 1, &off_tmp);
+    void* const p = allocate_obj(sizeof(prop_info) + namelen + 1, &new_offset);
-    if (offset) {
+    if (p != NULL) {
-        prop_info* info = new(offset) prop_info(name, namelen, value, valuelen);
+        prop_info* info = new(p) prop_info(name, namelen, value, valuelen);
-        *off = off_tmp;
+        *off = new_offset;
         return info;
     }
 
     return NULL;
 }
 
-static void *to_prop_obj(const uint32_t off)
+static void *to_prop_obj(uint_least32_t off)
 {
     if (off > pa_data_size)
         return NULL;
@@ -341,7 +340,17 @@ static void *to_prop_obj(const uint32_t off)
     return (__system_property_area__->data + off);
 }
 
-static prop_bt *root_node()
+static inline prop_bt *to_prop_bt(atomic_uint_least32_t* off_p) {
+  uint_least32_t off = atomic_load_explicit(off_p, memory_order_consume);
+  return reinterpret_cast<prop_bt*>(to_prop_obj(off));
+}
+
+static inline prop_info *to_prop_info(atomic_uint_least32_t* off_p) {
+  uint_least32_t off = atomic_load_explicit(off_p, memory_order_consume);
+  return reinterpret_cast<prop_info*>(to_prop_obj(off));
+}
+
+static inline prop_bt *root_node()
 {
     return reinterpret_cast<prop_bt*>(to_prop_obj(0));
 }
@@ -373,36 +382,34 @@ static prop_bt *find_prop_bt(prop_bt *const bt, const char *name,
         }
 
         if (ret < 0) {
-            if (current->left) {
+            uint_least32_t left_offset = atomic_load_explicit(&current->left, memory_order_relaxed);
-                current = reinterpret_cast<prop_bt*>(to_prop_obj(current->left));
+            if (left_offset != 0) {
+                current = to_prop_bt(&current->left);
             } else {
                 if (!alloc_if_needed) {
                    return NULL;
                 }
 
-                // Note that there isn't a race condition here. "clients" never
+                uint_least32_t new_offset;
-                // reach this code-path since It's only the (single threaded) server
-                // that allocates new nodes. Though "bt->left" is volatile, it can't
-                // have changed since the last value was last read.
-                uint32_t new_offset = 0;
                 prop_bt* new_bt = new_prop_bt(name, namelen, &new_offset);
                 if (new_bt) {
-                    current->left = new_offset;
+                    atomic_store_explicit(&current->left, new_offset, memory_order_release);
                 }
                 return new_bt;
             }
         } else {
-            if (current->right) {
+            uint_least32_t right_offset = atomic_load_explicit(&current->right, memory_order_relaxed);
-                current = reinterpret_cast<prop_bt*>(to_prop_obj(current->right));
+            if (right_offset != 0) {
+                current = to_prop_bt(&current->right);
             } else {
                 if (!alloc_if_needed) {
                    return NULL;
                 }
 
-                uint32_t new_offset;
+                uint_least32_t new_offset;
                 prop_bt* new_bt = new_prop_bt(name, namelen, &new_offset);
                 if (new_bt) {
-                    current->right = new_offset;
+                    atomic_store_explicit(&current->right, new_offset, memory_order_release);
                 }
                 return new_bt;
             }
@@ -429,13 +436,14 @@ static const prop_info *find_property(prop_bt *const trie, const char *name,
         }
 
         prop_bt* root = NULL;
-        if (current->children) {
+        uint_least32_t children_offset = atomic_load_explicit(&current->children, memory_order_relaxed);
-            root = reinterpret_cast<prop_bt*>(to_prop_obj(current->children));
+        if (children_offset != 0) {
+            root = to_prop_bt(&current->children);
         } else if (alloc_if_needed) {
-            uint32_t new_bt_offset;
+            uint_least32_t new_offset;
-            root = new_prop_bt(remaining_name, substr_size, &new_bt_offset);
+            root = new_prop_bt(remaining_name, substr_size, &new_offset);
             if (root) {
-                current->children = new_bt_offset;
+                atomic_store_explicit(&current->children, new_offset, memory_order_release);
             }
         }
 
@@ -454,13 +462,14 @@ static const prop_info *find_property(prop_bt *const trie, const char *name,
         remaining_name = sep + 1;
     }
 
-    if (current->prop) {
+    uint_least32_t prop_offset = atomic_load_explicit(&current->prop, memory_order_relaxed);
-        return reinterpret_cast<prop_info*>(to_prop_obj(current->prop));
+    if (prop_offset != 0) {
+        return to_prop_info(&current->prop);
     } else if (alloc_if_needed) {
-        uint32_t new_info_offset;
+        uint_least32_t new_offset;
-        prop_info* new_info = new_prop_info(name, namelen, value, valuelen, &new_info_offset);
+        prop_info* new_info = new_prop_info(name, namelen, value, valuelen, &new_offset);
         if (new_info) {
-            current->prop = new_info_offset;
+            atomic_store_explicit(&current->prop, new_offset, memory_order_release);
         }
 
         return new_info;
@@ -534,31 +543,34 @@ static void find_nth_fn(const prop_info *pi, void *ptr)
     cookie->count++;
 }
 
-static int foreach_property(const uint32_t off,
+static int foreach_property(prop_bt *const trie,
         void (*propfn)(const prop_info *pi, void *cookie), void *cookie)
 {
-    prop_bt *trie = reinterpret_cast<prop_bt*>(to_prop_obj(off));
     if (!trie)
         return -1;
 
-    if (trie->left) {
+    uint_least32_t left_offset = atomic_load_explicit(&trie->left, memory_order_relaxed);
-        const int err = foreach_property(trie->left, propfn, cookie);
+    if (left_offset != 0) {
+        const int err = foreach_property(to_prop_bt(&trie->left), propfn, cookie);
         if (err < 0)
             return -1;
     }
-    if (trie->prop) {
+    uint_least32_t prop_offset = atomic_load_explicit(&trie->prop, memory_order_relaxed);
-        prop_info *info = reinterpret_cast<prop_info*>(to_prop_obj(trie->prop));
+    if (prop_offset != 0) {
+        prop_info *info = to_prop_info(&trie->prop);
         if (!info)
             return -1;
         propfn(info, cookie);
     }
-    if (trie->children) {
+    uint_least32_t children_offset = atomic_load_explicit(&trie->children, memory_order_relaxed);
-        const int err = foreach_property(trie->children, propfn, cookie);
+    if (children_offset != 0) {
+        const int err = foreach_property(to_prop_bt(&trie->children), propfn, cookie);
         if (err < 0)
             return -1;
     }
-    if (trie->right) {
+    uint_least32_t right_offset = atomic_load_explicit(&trie->right, memory_order_relaxed);
-        const int err = foreach_property(trie->right, propfn, cookie);
+    if (right_offset != 0) {
+        const int err = foreach_property(to_prop_bt(&trie->right), propfn, cookie);
         if (err < 0)
             return -1;
     }
@@ -766,5 +778,5 @@ int __system_property_foreach(void (*propfn)(const prop_info *pi, void *cookie),
         return __system_property_foreach_compat(propfn, cookie);
     }
 
-    return foreach_property(0, propfn, cookie);
+    return foreach_property(root_node(), propfn, cookie);
 }

libc/bionic/wmempcpy.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2008 The Android Open Source Project
+ * Copyright (C) 2015 The Android Open Source Project
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -25,23 +25,9 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
-#include <stddef.h>
-#include <string.h>
 
-void *memrchr(const void *s, int c, size_t n)
+#include <wchar.h>
-{
-    if (n > 0) {
-        const char*  p = (const char*) s;
-        const char*  q = p + n;
 
-        while (1) {
+wchar_t* wmempcpy(wchar_t* dst, const wchar_t* src, size_t n) {
-            q--; if (q < p || q[0] == (char) c) break;
+  return wmemcpy(dst, src, n) + n;
-            q--; if (q < p || q[0] == (char) c) break;
-            q--; if (q < p || q[0] == (char) c) break;
-            q--; if (q < p || q[0] == (char) c) break;
-        }
-        if (q >= p)
-            return (void*)q;
-    }
-    return NULL;
 }

libc/dns/resolv/res_send.c

@@ -402,6 +402,10 @@ res_nsend(res_state statp,
 	}
 
 	if (statp->nscount == 0) {
+		// We have no nameservers configured, so there's no point trying.
+		// Tell the cache the query failed, or any retries and anyone else asking the same
+		// question will block for PENDING_REQUEST_TIMEOUT seconds instead of failing fast.
+		_resolv_cache_query_failed(statp->netid, buf, buflen);
 		errno = ESRCH;
 		return (-1);
 	}

libc/dns/resolv/res_state.c

@@ -39,6 +39,8 @@
 #define _REALLY_INCLUDE_SYS__SYSTEM_PROPERTIES_H_
 #include <sys/_system_properties.h>
 
+#include "private/ThreadLocalBuffer.h"
+
 /* Set to 1 to enable debug traces */
 #define DEBUG 0
 
@@ -50,8 +52,6 @@
 #  define D(...)  do{}while(0)
 #endif
 
-static pthread_key_t   _res_key;
-
 typedef struct {
     int                  _h_errno;
     // TODO: Have one __res_state per network so we don't have to repopulate frequently.
@@ -105,12 +105,7 @@ _res_thread_free( void*  _rt )
     free(rt);
 }
 
-__attribute__((constructor))
+BIONIC_PTHREAD_KEY_WITH_CONSTRUCTOR(_res_key, _res_thread_free);
-static void
-_res_init_key( void )
-{
-    pthread_key_create( &_res_key, _res_thread_free );
-}
 
 static _res_thread*
 _res_thread_get(void)

libc/include/elf.h

@@ -94,6 +94,13 @@ typedef struct {
 #define DT_PREINIT_ARRAY 32
 #define DT_PREINIT_ARRAYSZ 33
 
+/* Android compressed rel/rela sections */
+#define DT_ANDROID_REL (DT_LOOS + 2)
+#define DT_ANDROID_RELSZ (DT_LOOS + 3)
+
+#define DT_ANDROID_RELA (DT_LOOS + 4)
+#define DT_ANDROID_RELASZ (DT_LOOS + 5)
+
 /* gnu hash entry */
 #define DT_GNU_HASH 0x6ffffef5
 
@@ -106,6 +113,9 @@ typedef struct {
 #define STB_LOPROC    13
 #define STB_HIPROC    15
 
+#define SHT_LOOS    0x60000000
+#define SHT_HIOS    0x6fffffff
+
 #define STT_GNU_IFUNC 10
 #define STT_LOOS      10
 #define STT_HIOS      12
@@ -115,4 +125,8 @@ typedef struct {
 /* The kernel uses NT_PRFPREG but glibc also offers NT_FPREGSET */
 #define NT_FPREGSET NT_PRFPREG
 
+#define ELF_NOTE_GNU "GNU"
+
+#define NT_GNU_BUILD_ID 3
+
 #endif /* _ELF_H */

libc/include/fcntl.h

@@ -59,22 +59,29 @@ __BEGIN_DECLS
 
 extern int creat(const char*, mode_t);
 extern int creat64(const char*, mode_t);
-extern int fallocate64(int, int, off64_t, off64_t);
-extern int fallocate(int, int, off_t, off_t);
 extern int fcntl(int, int, ...);
 extern int openat(int, const char*, int, ...);
 extern int openat64(int, const char*, int, ...);
 extern int open(const char*, int, ...);
 extern int open64(const char*, int, ...);
-extern int posix_fadvise64(int, off64_t, off64_t, int);
-extern int posix_fadvise(int, off_t, off_t, int);
-extern int posix_fallocate64(int, off64_t, off64_t);
-extern int posix_fallocate(int, off_t, off_t);
 extern ssize_t splice(int, off64_t*, int, off64_t*, size_t, unsigned int);
 extern ssize_t tee(int, int, size_t, unsigned int);
 extern int unlinkat(int, const char*, int);
 extern ssize_t vmsplice(int, const struct iovec*, size_t, unsigned int);
 
+#if defined(__USE_FILE_OFFSET64)
+extern int fallocate(int, int, off_t, off_t) __RENAME(fallocate64);
+extern int posix_fadvise(int, off_t, off_t, int) __RENAME(posix_fadvise64);
+extern int posix_fallocate(int, off_t, off_t) __RENAME(posix_fallocate);
+#else
+extern int fallocate(int, int, off_t, off_t);
+extern int posix_fadvise(int, off_t, off_t, int);
+extern int posix_fallocate(int, off_t, off_t);
+#endif
+extern int fallocate64(int, int, off64_t, off64_t);
+extern int posix_fadvise64(int, off64_t, off64_t, int);
+extern int posix_fallocate64(int, off64_t, off64_t);
+
 extern int __open_2(const char*, int);
 extern int __open_real(const char*, int, ...) __RENAME(open);
 extern int __openat_2(int, const char*, int);

libc/include/grp.h

@@ -54,9 +54,9 @@ __BEGIN_DECLS
 struct group	*getgrgid(gid_t);
 struct group	*getgrnam(const char *);
 #if __POSIX_VISIBLE >= 200112 || __XPG_VISIBLE
-struct group	*getgrent(void);
+struct group	*getgrent(void)  __attribute__((deprecated("getgrent is meaningless on Android")));
-void		 setgrent(void);
+void setgrent(void) __attribute__((deprecated("setgrent is meaningless on Android")));
-void		 endgrent(void);
+void endgrent(void) __attribute__((deprecated("endgrent is meaningless on Android")));
 int		 getgrgid_r(gid_t, struct group *, char *,
 		    size_t, struct group **);
 int		 getgrnam_r(const char *, struct group *, char *,

libc/include/paths.h

@@ -39,7 +39,6 @@
 #define	_PATH_CONSOLE	"/dev/console"
 #define	_PATH_DEVNULL	"/dev/null"
 #define	_PATH_KLOG	"/proc/kmsg"
-#define	_PATH_MEM	"/dev/mem"
 
 #define	_PATH_MOUNTED	"/proc/mounts"
 #define	_PATH_TTY	"/dev/tty"

libc/include/pthread.h

@@ -73,13 +73,14 @@ enum {
 };
 
 typedef struct {
-  unsigned int value;
+#if defined(__LP64__)
-#ifdef __LP64__
+  char __private[48];
-  char __reserved[44];
+#else
+  char __private[4];
 #endif
-} pthread_cond_t;
+} pthread_cond_t __attribute__((aligned(8)));
 
-#define PTHREAD_COND_INITIALIZER  {0 __RESERVED_INITIALIZER}
+#define PTHREAD_COND_INITIALIZER  { { 0 } }
 
 typedef long pthread_mutexattr_t;
 typedef long pthread_condattr_t;
@@ -87,28 +88,14 @@ typedef long pthread_condattr_t;
 typedef long pthread_rwlockattr_t;
 
 typedef struct {
-#if !defined(__LP64__)
+#if defined(__LP64__)
-  pthread_mutex_t __unused_lock;
+  char __private[56];
-  pthread_cond_t __unused_cond;
-#endif
-  volatile int32_t state; // 0=unlock, -1=writer lock, +n=reader lock
-  volatile int32_t writer_thread_id;
-  volatile int32_t pending_readers;
-  volatile int32_t pending_writers;
-  int32_t attr;
-#ifdef __LP64__
-  char __reserved[36];
 #else
-  char __reserved[12];
+  char __private[40];
 #endif
+} pthread_rwlock_t __attribute__((aligned(8)));
 
-} pthread_rwlock_t;
+#define PTHREAD_RWLOCK_INITIALIZER  { { 0 } }
-
-#ifdef __LP64__
-  #define PTHREAD_RWLOCK_INITIALIZER  { 0, 0, 0, 0, 0, { 0 } }
-#else
-  #define PTHREAD_RWLOCK_INITIALIZER  { PTHREAD_MUTEX_INITIALIZER, PTHREAD_COND_INITIALIZER, 0, 0, 0, 0, 0, { 0 } }
-#endif
 
 typedef int pthread_key_t;
 

libc/include/pwd.h

@@ -119,10 +119,6 @@ struct passwd* getpwuid(uid_t);
 int getpwnam_r(const char*, struct passwd*, char*, size_t, struct passwd**);
 int getpwuid_r(uid_t, struct passwd*, char*, size_t, struct passwd**);
 
-void endpwent(void);
-struct passwd* getpwent(void);
-int setpwent(void);
-
 __END_DECLS
 
 #endif

libc/include/stdio.h

@@ -57,8 +57,6 @@
 
 __BEGIN_DECLS
 
-#define	_FSTDIO			/* Define for new stdio with functions. */
-
 typedef off_t fpos_t;		/* stdio file position type */
 
 /*
@@ -266,27 +264,38 @@ int vdprintf(int, const char * __restrict, __va_list) __printflike(2, 0);
 
 #ifndef __AUDIT__
 #if !defined(__STDC_VERSION__) || __STDC_VERSION__ < 201112L
-char* gets(char*) __warnattr("gets is very unsafe; consider using fgets");
+char* gets(char*) __attribute__((deprecated("gets is unsafe, use fgets instead")));
 #endif
 int sprintf(char* __restrict, const char* __restrict, ...)
     __printflike(2, 3) __warnattr("sprintf is often misused; please use snprintf");
-char* tmpnam(char*) __warnattr("tmpnam possibly used unsafely; consider using mkstemp");
 int vsprintf(char* __restrict, const char* __restrict, __va_list)
     __printflike(2, 0) __warnattr("vsprintf is often misused; please use vsnprintf");
+char* tmpnam(char*) __attribute__((deprecated("tmpnam is unsafe, use mkstemp or tmpfile instead")));
 #if __XPG_VISIBLE
 char* tempnam(const char*, const char*)
-    __warnattr("tempnam possibly used unsafely; consider using mkstemp");
+    __attribute__((deprecated("tempnam is unsafe, use mkstemp or tmpfile instead")));
 #endif
 #endif
 
 extern int rename(const char*, const char*);
 extern int renameat(int, const char*, int, const char*);
 
+#if defined(__USE_FILE_OFFSET64)
+/* Not possible. */
+int	 fgetpos(FILE * __restrict, fpos_t * __restrict)
+	__attribute__((__error__("not available with _FILE_OFFSET_BITS=64")));
+int	 fsetpos(FILE *, const fpos_t *)
+	__attribute__((__error__("not available with _FILE_OFFSET_BITS=64")));
+int	 fseeko(FILE *, off_t, int)
+	__attribute__((__error__("not available with _FILE_OFFSET_BITS=64")));
+off_t	 ftello(FILE *)
+	__attribute__((__error__("not available with _FILE_OFFSET_BITS=64")));
+#else
 int	 fgetpos(FILE * __restrict, fpos_t * __restrict);
 int	 fsetpos(FILE *, const fpos_t *);
-
 int	 fseeko(FILE *, off_t, int);
 off_t	 ftello(FILE *);
+#endif
 
 #if __ISO_C_VISIBLE >= 1999 || __BSD_VISIBLE
 int	 snprintf(char * __restrict, size_t, const char * __restrict, ...)

libc/include/stdlib.h

@@ -58,7 +58,7 @@ extern int unsetenv(const char*);
 extern int clearenv(void);
 
 extern char* mkdtemp(char*);
-extern char* mktemp(char*) __warnattr("mktemp possibly used unsafely; consider using mkstemp");
+extern char* mktemp(char*) __attribute__((deprecated("mktemp is unsafe, use mkstemp or tmpfile instead")));
 
 extern int mkostemp64(char*, int);
 extern int mkostemp(char*, int);

libc/include/string.h

@@ -44,6 +44,9 @@ extern void*  memchr(const void *, int, size_t) __purefunc;
 extern void*  memrchr(const void *, int, size_t) __purefunc;
 extern int    memcmp(const void *, const void *, size_t) __purefunc;
 extern void*  memcpy(void* __restrict, const void* __restrict, size_t);
+#if defined(__USE_GNU)
+extern void*  mempcpy(void* __restrict, const void* __restrict, size_t);
+#endif
 extern void*  memmove(void *, const void *, size_t);
 extern void*  memset(void *, int, size_t);
 extern void*  memmem(const void *, size_t, const void *, size_t) __purefunc;

libc/include/sys/cdefs.h

@@ -335,13 +335,15 @@
 #endif
 
 #if __GNUC_PREREQ(4, 3)
-#define __errordecl(name, msg) extern void name(void) __attribute__((__error__(msg)))
+#define __errorattr(msg) __attribute__((__error__(msg)))
 #define __warnattr(msg) __attribute__((__warning__(msg)))
 #else
-#define __errordecl(name, msg) extern void name(void)
+#define __errorattr(msg)
 #define __warnattr(msg)
 #endif
 
+#define __errordecl(name, msg) extern void name(void) __errorattr(msg)
+
 /*
  * Some BSD source needs these macros.
  * Originally they embedded the rcs versions of each source file
@@ -378,6 +380,15 @@
 # define __USE_BSD 1
 #endif
 
+/*
+ * _FILE_OFFSET_BITS 64 support.
+ */
+#if !defined(__LP64__) && defined(_FILE_OFFSET_BITS)
+#if _FILE_OFFSET_BITS == 64
+#define __USE_FILE_OFFSET64 1
+#endif
+#endif
+
 /*-
  * POSIX.1 requires that the macros we test be defined before any standard
  * header file is included.

libc/include/sys/file.h

@@ -25,14 +25,19 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
+
 #ifndef _SYS_FILE_H_
 #define _SYS_FILE_H_
 
 #include <sys/cdefs.h>
 #include <sys/types.h>
 
-/* ANDROID: needed for flock() */
-#include <unistd.h>
 #include <fcntl.h>
 
+__BEGIN_DECLS
+
+int flock(int, int);
+
+__END_DECLS
+
 #endif /* _SYS_FILE_H_ */

libc/include/sys/mman.h

@@ -49,8 +49,13 @@ __BEGIN_DECLS
 #define POSIX_MADV_WILLNEED   MADV_WILLNEED
 #define POSIX_MADV_DONTNEED   MADV_DONTNEED
 
+#if defined(__USE_FILE_OFFSET64)
+extern void* mmap(void*, size_t, int, int, int, off_t) __RENAME(mmap64);
+#else
 extern void* mmap(void*, size_t, int, int, int, off_t);
+#endif
 extern void* mmap64(void*, size_t, int, int, int, off64_t);
+
 extern int munmap(void*, size_t);
 extern int msync(const void*, size_t, int);
 extern int mprotect(const void*, size_t, int);

libc/include/sys/mount.h

@@ -25,6 +25,7 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
+
 #ifndef _SYS_MOUNT_H
 #define _SYS_MOUNT_H
 
@@ -35,9 +36,10 @@
 __BEGIN_DECLS
 
 /* umount2 flags. */
-#define MNT_FORCE	1	/* Forcibly unmount */
+#define MNT_FORCE 1
-#define MNT_DETACH	2	/* Detach from tree only */
+#define MNT_DETACH 2
-#define MNT_EXPIRE	4	/* Mark for expiry */
+#define MNT_EXPIRE 4
+#define UMOUNT_NOFOLLOW 8
 
 extern int mount(const char*, const char*, const char*, unsigned long, const void*);
 extern int umount(const char*);

libc/include/sys/resource.h

@@ -36,6 +36,10 @@
 
 __BEGIN_DECLS
 
+/* The kernel header doesn't have these, but POSIX does. */
+#define RLIM_SAVED_CUR RLIM_INFINITY
+#define RLIM_SAVED_MAX RLIM_INFINITY
+
 typedef unsigned long rlim_t;
 
 extern int getrlimit(int, struct rlimit*);

libc/include/sys/sendfile.h

@@ -34,7 +34,11 @@
 
 __BEGIN_DECLS
 
+#if defined(__USE_FILE_OFFSET64)
+extern ssize_t sendfile(int out_fd, int in_fd, off_t* offset, size_t count) __RENAME(sendfile64);
+#else
 extern ssize_t sendfile(int out_fd, int in_fd, off_t* offset, size_t count);
+#endif
 extern ssize_t sendfile64(int out_fd, int in_fd, off64_t* offset, size_t count);
 
 __END_DECLS

libc/include/sys/syscall.h

@@ -25,20 +25,13 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
+
 #ifndef _SYS_SYSCALL_H_
 #define _SYS_SYSCALL_H_
 
-#include <errno.h>
+#include <asm/unistd.h> /* Linux kernel __NR_* names. */
-#include <sys/cdefs.h>
+#include <sys/glibc-syscalls.h> /* glibc-compatible SYS_* aliases. */
-#include <sys/types.h>
-#include <asm/unistd.h>
 
-#include <sys/glibc-syscalls.h> /* glibc-compatible SYS_* aliases for our __NR_* names. */
+/* The syscall function itself is declared in <unistd.h>, not here. */
-
-__BEGIN_DECLS
-
-long syscall(long number, ...);
-
-__END_DECLS
 
 #endif /* _SYS_SYSCALL_H_ */

libc/include/sys/types.h

@@ -90,16 +90,14 @@ typedef uint64_t dev_t;
 typedef __kernel_time_t __time_t;
 typedef __time_t time_t;
 
-/* This historical accident means that we had a 32-bit off_t on 32-bit architectures. */
+#if defined(__USE_FILE_OFFSET64) || defined(__LP64__)
-#if !defined(__LP64__)
+typedef int64_t off_t;
-typedef __kernel_off_t off_t;
+typedef off_t loff_t;
-typedef __kernel_loff_t loff_t;
 typedef loff_t off64_t;
 #else
-/* We could re-use the LP32 definitions, but that would mean that although off_t and loff_t/off64_t
+/* This historical accident means that we had a 32-bit off_t on 32-bit architectures. */
- * would be the same size, they wouldn't actually be the same type, which can lead to warnings. */
 typedef __kernel_off_t off_t;
-typedef off_t loff_t;
+typedef __kernel_loff_t loff_t;
 typedef loff_t off64_t;
 #endif
 

libc/include/syslog.h

@@ -47,6 +47,7 @@ __BEGIN_DECLS
 
 #define LOG_PRIMASK 7
 #define LOG_PRI(x) ((x) & LOG_PRIMASK)
+#define LOG_MAKEPRI(fac, pri) ((fac) | (pri))
 
 /* Facilities are currently ignored on Android. */
 #define LOG_KERN     0000

libc/include/unistd.h

@@ -116,10 +116,6 @@ extern int setresgid(gid_t, gid_t, gid_t);
 extern int getresuid(uid_t *ruid, uid_t *euid, uid_t *suid);
 extern int getresgid(gid_t *rgid, gid_t *egid, gid_t *sgid);
 extern char* getlogin(void);
-extern char* getusershell(void);
-extern void setusershell(void);
-extern void endusershell(void);
-
 
 extern long fpathconf(int, int);
 extern long pathconf(const char*, int);
@@ -146,32 +142,40 @@ extern int chown(const char *, uid_t, gid_t);
 extern int fchown(int, uid_t, gid_t);
 extern int fchownat(int, const char*, uid_t, gid_t, int);
 extern int lchown(const char *, uid_t, gid_t);
-extern int truncate(const char *, off_t);
-extern int truncate64(const char *, off64_t);
 extern char *getcwd(char *, size_t);
 
 extern int sync(void);
 
 extern int close(int);
-extern off_t lseek(int, off_t, int);
-extern off64_t lseek64(int, off64_t, int);
 
 extern ssize_t read(int, void *, size_t);
 extern ssize_t write(int, const void *, size_t);
-extern ssize_t pread(int, void *, size_t, off_t);
-extern ssize_t pread64(int, void *, size_t, off64_t);
-extern ssize_t pwrite(int, const void *, size_t, off_t);
-extern ssize_t pwrite64(int, const void *, size_t, off64_t);
 
 extern int dup(int);
 extern int dup2(int, int);
 extern int dup3(int, int, int);
 extern int fcntl(int, int, ...);
 extern int ioctl(int, int, ...);
-extern int flock(int, int);
 extern int fsync(int);
 extern int fdatasync(int);
+
+#if defined(__USE_FILE_OFFSET64)
+extern int truncate(const char *, off_t) __RENAME(truncate64);
+extern off_t lseek(int, off_t, int) __RENAME(lseek64);
+extern ssize_t pread(int, void *, size_t, off_t) __RENAME(pread64);
+extern ssize_t pwrite(int, const void *, size_t, off_t) __RENAME(pwrite64);
+extern int ftruncate(int, off_t) __RENAME(ftruncate64);
+#else
+extern int truncate(const char *, off_t);
+extern off_t lseek(int, off_t, int);
+extern ssize_t pread(int, void *, size_t, off_t);
+extern ssize_t pwrite(int, const void *, size_t, off_t);
 extern int ftruncate(int, off_t);
+#endif
+extern int truncate64(const char *, off64_t);
+extern off64_t lseek64(int, off64_t, int);
+extern ssize_t pread64(int, void *, size_t, off64_t);
+extern ssize_t pwrite64(int, const void *, size_t, off64_t);
 extern int ftruncate64(int, off64_t);
 
 extern int pause(void);
@@ -200,6 +204,8 @@ int getpagesize(void);
 
 long sysconf(int);
 
+long syscall(long number, ...);
+
 extern int daemon(int, int);
 
 #if defined(__arm__) || (defined(__mips__) && !defined(__LP64__))

libc/include/wchar.h

@@ -151,6 +151,9 @@ extern int               wcwidth(wchar_t);
 extern wchar_t          *wmemchr(const wchar_t *, wchar_t, size_t);
 extern int               wmemcmp(const wchar_t *, const wchar_t *, size_t);
 extern wchar_t          *wmemcpy(wchar_t *, const wchar_t *, size_t);
+#if defined(__USE_GNU)
+extern wchar_t          *wmempcpy(wchar_t *, const wchar_t *, size_t);
+#endif
 extern wchar_t          *wmemmove(wchar_t *, const wchar_t *, size_t);
 extern wchar_t          *wmemset(wchar_t *, wchar_t, size_t);
 extern int               wprintf(const wchar_t *, ...);

libc/private/ThreadLocalBuffer.h

@@ -38,15 +38,17 @@
 
 // We used to use pthread_once to initialize the keys, but life is more predictable
 // if we allocate them all up front when the C library starts up, via __constructor__.
+#define BIONIC_PTHREAD_KEY_WITH_CONSTRUCTOR(key_name, key_destructor) \
+  static pthread_key_t key_name; \
+  __attribute__((constructor)) static void __bionic_tls_ ## key_name ## _key_init() { \
+    pthread_key_create(&key_name, key_destructor); \
+  }
 
 #define GLOBAL_INIT_THREAD_LOCAL_BUFFER(name) \
-  static pthread_key_t __bionic_tls_ ## name ## _key; \
   static void __bionic_tls_ ## name ## _key_destroy(void* buffer) { \
     free(buffer); \
   } \
-  __attribute__((constructor)) static void __bionic_tls_ ## name ## _key_init() { \
+  BIONIC_PTHREAD_KEY_WITH_CONSTRUCTOR(__bionic_tls_ ## name ## _key, __bionic_tls_ ## name ## _key_destroy)
-    pthread_key_create(&__bionic_tls_ ## name ## _key, __bionic_tls_ ## name ## _key_destroy); \
-  }
 
 // Leaves "name_tls_buffer" and "name_tls_buffer_size" defined and initialized.
 #define LOCAL_INIT_THREAD_LOCAL_BUFFER(type, name, byte_count) \

libc/private/bionic_asm.h

@@ -57,4 +57,8 @@
     ENTRY(f); \
     .hidden f \
 
+#define ALIAS_SYMBOL(alias, original) \
+    .globl alias; \
+    .equ alias, original
+
 #endif /* _PRIVATE_BIONIC_ASM_H_ */

libc/private/bionic_atomic_arm.h

@@ -1,74 +0,0 @@
-/*
- * Copyright (C) 2011 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-#ifndef BIONIC_ATOMIC_ARM_H
-#define BIONIC_ATOMIC_ARM_H
-
-__ATOMIC_INLINE__ void __bionic_memory_barrier() {
-  __asm__ __volatile__ ( "dmb ish" : : : "memory" );
-}
-
-/* Compare-and-swap, without any explicit barriers. Note that this function
- * returns 0 on success, and 1 on failure. The opposite convention is typically
- * used on other platforms.
- */
-__ATOMIC_INLINE__ int __bionic_cmpxchg(int32_t old_value, int32_t new_value, volatile int32_t* ptr) {
-  int32_t prev, status;
-  do {
-    __asm__ __volatile__ (
-          "ldrex %0, [%3]\n"
-          "mov %1, #0\n"
-          "teq %0, %4\n"
-#ifdef __thumb2__
-          "it eq\n"
-#endif
-          "strexeq %1, %5, [%3]"
-          : "=&r" (prev), "=&r" (status), "+m"(*ptr)
-          : "r" (ptr), "Ir" (old_value), "r" (new_value)
-          : "cc");
-  } while (__builtin_expect(status != 0, 0));
-  return prev != old_value;
-}
-
-/* Swap, without any explicit barriers. */
-__ATOMIC_INLINE__ int32_t __bionic_swap(int32_t new_value, volatile int32_t* ptr) {
-  int32_t prev, status;
-  do {
-    __asm__ __volatile__ (
-          "ldrex %0, [%3]\n"
-          "strex %1, %4, [%3]"
-          : "=&r" (prev), "=&r" (status), "+m" (*ptr)
-          : "r" (ptr), "r" (new_value)
-          : "cc");
-  } while (__builtin_expect(status != 0, 0));
-  return prev;
-}
-
-/* Atomic decrement, without explicit barriers. */
-__ATOMIC_INLINE__ int32_t __bionic_atomic_dec(volatile int32_t* ptr) {
-  int32_t prev, tmp, status;
-  do {
-    __asm__ __volatile__ (
-          "ldrex %0, [%4]\n"
-          "sub %1, %0, #1\n"
-          "strex %2, %1, [%4]"
-          : "=&r" (prev), "=&r" (tmp), "=&r" (status), "+m"(*ptr)
-          : "r" (ptr)
-          : "cc");
-  } while (__builtin_expect(status != 0, 0));
-  return prev;
-}
-
-#endif /* SYS_ATOMICS_ARM_H */

libc/private/bionic_atomic_arm64.h

@@ -1,72 +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.
- */
-#ifndef BIONIC_ATOMIC_AARCH64_H
-#define BIONIC_ATOMIC_AARCH64_H
-
-/* For ARMv8, we can use the 'dmb' instruction directly */
-__ATOMIC_INLINE__ void __bionic_memory_barrier() {
-  __asm__ __volatile__ ( "dmb ish" : : : "memory" );
-}
-
-/* Compare-and-swap, without any explicit barriers. Note that this function
- * returns 0 on success, and 1 on failure. The opposite convention is typically
- * used on other platforms.
- */
-__ATOMIC_INLINE__ int __bionic_cmpxchg(int32_t old_value, int32_t new_value, volatile int32_t* ptr) {
-  int32_t tmp, oldval;
-  __asm__ __volatile__ (
-      "// atomic_cmpxchg\n"
-      "1:  ldxr %w1, [%3]\n"
-      "    cmp %w1, %w4\n"
-      "    b.ne 2f\n"
-      "    stxr %w0, %w5, [%3]\n"
-      "    cbnz  %w0, 1b\n"
-      "2:"
-      : "=&r" (tmp), "=&r" (oldval), "+o"(*ptr)
-      : "r" (ptr), "Ir" (old_value), "r" (new_value)
-      : "cc", "memory");
-  return oldval != old_value;
-}
-
-/* Swap, without any explicit barriers.  */
-__ATOMIC_INLINE__ int32_t __bionic_swap(int32_t new_value, volatile int32_t* ptr) {
-  int32_t prev, status;
-  __asm__ __volatile__ (
-      "// atomic_swap\n"
-      "1:  ldxr %w0, [%3]\n"
-      "    stxr %w1, %w4, [%3]\n"
-      "    cbnz %w1, 1b\n"
-      : "=&r" (prev), "=&r" (status), "+o" (*ptr)
-      : "r" (ptr), "r" (new_value)
-      : "cc", "memory");
-  return prev;
-}
-
-/* Atomic decrement, without explicit barriers.  */
-__ATOMIC_INLINE__ int32_t __bionic_atomic_dec(volatile int32_t* ptr) {
-  int32_t prev, tmp, status;
-  __asm__ __volatile__ (
-      "1:  ldxr %w0, [%4]\n"
-      "    sub %w1, %w0, #1\n"
-      "    stxr %w2, %w1, [%4]\n"
-      "    cbnz %w2, 1b"
-      : "=&r" (prev), "=&r" (tmp), "=&r" (status), "+m"(*ptr)
-      : "r" (ptr)
-      : "cc", "memory");
-  return prev;
-}
-
-#endif /* BIONIC_ATOMICS_AARCH64_H */

libc/private/bionic_atomic_gcc_builtin.h

@@ -1,50 +0,0 @@
-/*
- * Copyright (C) 2011 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-#ifndef BIONIC_ATOMIC_GCC_BUILTIN_H
-#define BIONIC_ATOMIC_GCC_BUILTIN_H
-
-/*
- * This header file is used by default if we don't have optimized atomic
- * routines for a given platform. See bionic_atomic_arm.h and
- * bionic_atomic_x86.h for examples.
- *
- * Note that the GCC builtins include barriers that aren't present in
- * the architecture-specific assembler versions.
- */
-
-__ATOMIC_INLINE__ void __bionic_memory_barrier() {
-  __sync_synchronize();
-}
-
-__ATOMIC_INLINE__ int __bionic_cmpxchg(int32_t old_value, int32_t new_value, volatile int32_t* ptr) {
-  /* We must return 0 on success. */
-  return __sync_val_compare_and_swap(ptr, old_value, new_value) != old_value;
-}
-
-__ATOMIC_INLINE__ int32_t __bionic_swap(int32_t new_value, volatile int32_t* ptr) {
-  int32_t old_value;
-  do {
-    old_value = *ptr;
-  } while (__sync_val_compare_and_swap(ptr, old_value, new_value) != old_value);
-  return old_value;
-}
-
-__ATOMIC_INLINE__ int32_t __bionic_atomic_dec(volatile int32_t* ptr) {
-  /* We must return the old value. */
-  return __sync_fetch_and_add(ptr, -1);
-}
-
-#endif /* BIONIC_ATOMIC_GCC_BUILTIN_H */

libc/private/bionic_atomic_inline.h

@@ -1,61 +0,0 @@
-/*
- * Copyright (C) 2010 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef BIONIC_ATOMIC_INLINE_H
-#define BIONIC_ATOMIC_INLINE_H
-
-/*
- * Inline declarations and macros for some special-purpose atomic
- * operations.  These are intended for rare circumstances where a
- * memory barrier needs to be issued inline rather than as a function
- * call.
- *
- * Macros defined in this header:
- *
- * void ANDROID_MEMBAR_FULL()
- *   Full memory barrier.  Provides a compiler reordering barrier, and
- *   on SMP systems emits an appropriate instruction.
- */
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-/* Define __ATOMIC_INLINE__ to control the inlining of all atomics
- * functions declared here. For a slight performance boost, we want
- * all of them to be always_inline
- */
-#define  __ATOMIC_INLINE__  static __inline__ __attribute__((always_inline))
-
-#if defined(__arm__)
-#  include "bionic_atomic_arm.h"
-#elif defined(__aarch64__)
-#  include "bionic_atomic_arm64.h"
-#elif defined(__i386__)
-#  include "bionic_atomic_x86.h"
-#elif defined(__mips__)
-#  include "bionic_atomic_mips.h"
-#else
-#  include "bionic_atomic_gcc_builtin.h"
-#endif
-
-#define ANDROID_MEMBAR_FULL  __bionic_memory_barrier
-
-#ifdef __cplusplus
-} // extern "C"
-#endif
-
-#endif // BIONIC_ATOMIC_INLINE_H

libc/private/bionic_atomic_mips.h

@@ -1,71 +0,0 @@
-/*
- * Copyright (C) 2011 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-#ifndef BIONIC_ATOMIC_MIPS_H
-#define BIONIC_ATOMIC_MIPS_H
-
-/* Define a full memory barrier, this is only needed if we build the
- * platform for a multi-core device.
- */
-
-__ATOMIC_INLINE__ void __bionic_memory_barrier() {
-  __asm__ __volatile__ ( "sync" : : : "memory" );
-}
-
-/* Compare-and-swap, without any explicit barriers. Note that this function
- * returns 0 on success, and 1 on failure. The opposite convention is typically
- * used on other platforms.
- */
-__ATOMIC_INLINE__ int __bionic_cmpxchg(int32_t old_value, int32_t new_value, volatile int32_t* ptr) {
-  int32_t prev, status;
-  __asm__ __volatile__ ("1: move %[status], %[new_value]  \n"
-                        "   ll %[prev], 0(%[ptr])         \n"
-                        "   bne %[old_value], %[prev], 2f \n"
-                        "   sc   %[status], 0(%[ptr])     \n"
-                        "   beqz %[status], 1b            \n"
-                        "2:                               \n"
-                        : [prev]"=&r"(prev), [status]"=&r"(status), "+m"(*ptr)
-                        : [new_value]"r"(new_value), [old_value]"r"(old_value), [ptr]"r"(ptr)
-                        : "memory");
-  return prev != old_value;
-}
-
-/* Swap, without any explicit barriers. */
-__ATOMIC_INLINE__ int32_t __bionic_swap(int32_t new_value, volatile int32_t* ptr) {
-  int32_t prev, status;
-  __asm__ __volatile__ ("1:  move %[status], %[new_value] \n"
-                        "    ll %[prev], 0(%[ptr])        \n"
-                        "    sc %[status], 0(%[ptr])      \n"
-                        "    beqz %[status], 1b           \n"
-                        : [prev]"=&r"(prev), [status]"=&r"(status), "+m"(*ptr)
-                        : [ptr]"r"(ptr), [new_value]"r"(new_value)
-                        : "memory");
-  return prev;
-}
-
-/* Atomic decrement, without explicit barriers. */
-__ATOMIC_INLINE__ int32_t __bionic_atomic_dec(volatile int32_t* ptr) {
-  int32_t prev, status;
-  __asm__ __volatile__ ("1:  ll %[prev], 0(%[ptr])        \n"
-                        "    addiu %[status], %[prev], -1 \n"
-                        "    sc   %[status], 0(%[ptr])    \n"
-                        "    beqz %[status], 1b           \n"
-                        : [prev]"=&r" (prev), [status]"=&r"(status), "+m" (*ptr)
-                        : [ptr]"r"(ptr)
-                        : "memory");
-  return prev;
-}
-
-#endif /* BIONIC_ATOMIC_MIPS_H */

libc/private/bionic_atomic_x86.h

@@ -1,58 +0,0 @@
-/*
- * Copyright (C) 2011 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-#ifndef BIONIC_ATOMIC_X86_H
-#define BIONIC_ATOMIC_X86_H
-
-/* Define a full memory barrier, this is only needed if we build the
- * platform for a multi-core device.
- */
-__ATOMIC_INLINE__ void __bionic_memory_barrier() {
-  __asm__ __volatile__ ( "mfence" : : : "memory" );
-}
-
-/* Compare-and-swap, without any explicit barriers. Note that this function
- * returns 0 on success, and 1 on failure. The opposite convention is typically
- * used on other platforms.
- */
-__ATOMIC_INLINE__ int __bionic_cmpxchg(int32_t old_value, int32_t new_value, volatile int32_t* ptr) {
-    int32_t prev;
-    __asm__ __volatile__ ("lock; cmpxchgl %1, %2"
-                          : "=a" (prev)
-                          : "q" (new_value), "m" (*ptr), "0" (old_value)
-                          : "memory");
-    return prev != old_value;
-}
-
-/* Swap, without any explicit barriers. */
-__ATOMIC_INLINE__ int32_t __bionic_swap(int32_t new_value, volatile int32_t *ptr) {
-  __asm__ __volatile__ ("xchgl %1, %0"
-                        : "=r" (new_value)
-                        : "m" (*ptr), "0" (new_value)
-                        : "memory");
-  return new_value;
-}
-
-/* Atomic decrement, without explicit barriers. */
-__ATOMIC_INLINE__ int32_t __bionic_atomic_dec(volatile int32_t* ptr) {
-  int increment = -1;
-  __asm__ __volatile__ ("lock; xaddl %0, %1"
-                        : "+r" (increment), "+m" (*ptr)
-                        : : "memory");
-  /* increment now holds the old value of *ptr */
-  return increment;
-}
-
-#endif /* BIONIC_ATOMIC_X86_H */

libc/private/bionic_futex.h

@@ -34,6 +34,7 @@
 #include <stddef.h>
 #include <sys/cdefs.h>
 #include <sys/syscall.h>
+#include <unistd.h>
 
 __BEGIN_DECLS
 

libc/private/bionic_tls.h

@@ -67,18 +67,18 @@ enum {
   TLS_SLOT_STACK_GUARD = 5, // GCC requires this specific slot for x86.
   TLS_SLOT_DLERROR,
 
-  TLS_SLOT_FIRST_USER_SLOT // Must come last!
+  BIONIC_TLS_SLOTS // Must come last!
 };
 
 /*
- * There are two kinds of slot used internally by bionic --- there are the well-known slots
+ * Bionic uses some pthread keys internally. All pthread keys used internally
- * enumerated above, and then there are those that are allocated during startup by calls to
+ * should be created in constructors, except for keys that may be used in or before constructors.
- * pthread_key_create; grep for GLOBAL_INIT_THREAD_LOCAL_BUFFER to find those. We need to manually
+ * We need to manually maintain the count of pthread keys used internally, but
- * maintain that second number, but pthread_test will fail if we forget.
+ * pthread_test should fail if we forget.
- * Following are current pthread keys used internally:
+ * Following are current pthread keys used internally by libc:
  *  basename               libc (GLOBAL_INIT_THREAD_LOCAL_BUFFER)
  *  dirname                libc (GLOBAL_INIT_THREAD_LOCAL_BUFFER)
- *  uselocale              libc
+ *  uselocale              libc (can be used in constructors)
  *  getmntent_mntent       libc (GLOBAL_INIT_THREAD_LOCAL_BUFFER)
  *  getmntent_strings      libc (GLOBAL_INIT_THREAD_LOCAL_BUFFER)
  *  ptsname                libc (GLOBAL_INIT_THREAD_LOCAL_BUFFER)
@@ -87,29 +87,30 @@ enum {
  *  strsignal              libc (GLOBAL_INIT_THREAD_LOCAL_BUFFER)
  *  passwd                 libc (GLOBAL_INIT_THREAD_LOCAL_BUFFER)
  *  group                  libc (GLOBAL_INIT_THREAD_LOCAL_BUFFER)
- *  _res_key               libc
+ *  _res_key               libc (BIONIC_PTHREAD_KEY_WITH_CONSTRUCTOR)
+ */
+
+#define LIBC_PTHREAD_KEY_RESERVED_COUNT 12
+
+#if defined(USE_JEMALLOC)
+/* Following are current pthread keys used internally by jemalloc:
  * je_thread_allocated_tsd jemalloc
  * je_arenas_tsd           jemalloc
  * je_tcache_tsd           jemalloc
  * je_tcache_enabled_tsd   jemalloc
  * je_quarantine_tsd       jemalloc
- *
  */
-
+#define JEMALLOC_PTHREAD_KEY_RESERVED_COUNT 5
-#define LIBC_TLS_RESERVED_SLOTS 12
+#define BIONIC_PTHREAD_KEY_RESERVED_COUNT (LIBC_PTHREAD_KEY_RESERVED_COUNT + JEMALLOC_PTHREAD_KEY_RESERVED_COUNT)
-
-#if defined(USE_JEMALLOC)
-/* jemalloc uses 5 keys for itself. */
-#define BIONIC_TLS_RESERVED_SLOTS (LIBC_TLS_RESERVED_SLOTS + 5)
 #else
-#define BIONIC_TLS_RESERVED_SLOTS LIBC_TLS_RESERVED_SLOTS
+#define BIONIC_PTHREAD_KEY_RESERVED_COUNT LIBC_PTHREAD_KEY_RESERVED_COUNT
 #endif
 
 /*
- * Maximum number of elements in the TLS array.
+ * Maximum number of pthread keys allocated.
- * This includes space for pthread keys and our own internal slots.
+ * This includes pthread keys used internally and externally.
  */
-#define BIONIC_TLS_SLOTS (PTHREAD_KEYS_MAX + TLS_SLOT_FIRST_USER_SLOT + BIONIC_TLS_RESERVED_SLOTS)
+#define BIONIC_PTHREAD_KEY_COUNT (BIONIC_PTHREAD_KEY_RESERVED_COUNT + PTHREAD_KEYS_MAX)
 
 __END_DECLS
 

libc/stdio/fread.c

@@ -102,6 +102,12 @@ fread(void *buf, size_t size, size_t count, FILE *fp)
 		 * avoid copying it through the buffer?
 		 */
 		if (total > (size_t) fp->_bf._size) {
+			/*
+			 * Make sure that fseek doesn't think it can
+			 * reuse the buffer since we are going to read
+			 * directly from the file descriptor.
+			 */
+			fp->_flags |= __SMOD;
 			break;
 		}
 

libc/upstream-openbsd/lib/libc/string/memchr.c

@@ -0,0 +1,48 @@
+/*	$OpenBSD: memchr.c,v 1.7 2005/08/08 08:05:37 espie Exp $ */
+/*-
+ * Copyright (c) 1990 The Regents of the University of California.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to Berkeley by
+ * Chris Torek.
+ *
+ * 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.
+ */
+
+#include <string.h>
+
+void *
+memchr(const void *s, int c, size_t n)
+{
+	if (n != 0) {
+		const unsigned char *p = s;
+
+		do {
+			if (*p++ == (unsigned char)c)
+				return ((void *)(p - 1));
+		} while (--n != 0);
+	}
+	return (NULL);
+}

libc/upstream-openbsd/lib/libc/string/memrchr.c

@@ -0,0 +1,38 @@
+/*	$OpenBSD: memrchr.c,v 1.2 2007/11/27 16:22:12 martynas Exp $	*/
+
+/*
+ * Copyright (c) 2007 Todd C. Miller <Todd.Miller@courtesan.com>
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include <string.h>
+
+/*
+ * Reverse memchr()
+ * Find the last occurrence of 'c' in the buffer 's' of size 'n'.
+ */
+void *
+memrchr(const void *s, int c, size_t n)
+{
+	const unsigned char *cp;
+
+	if (n != 0) {
+		cp = (unsigned char *)s + n;
+		do {
+			if (*(--cp) == (unsigned char)c)
+				return((void *)cp);
+		} while (--n != 0);
+	}
+	return(NULL);
+}

libm/Android.mk

@@ -21,19 +21,14 @@ LOCAL_SRC_FILES := \
     upstream-freebsd/lib/msun/bsdsrc/b_tgamma.c \
     upstream-freebsd/lib/msun/src/catrig.c \
     upstream-freebsd/lib/msun/src/catrigf.c \
-    upstream-freebsd/lib/msun/src/e_acos.c \
     upstream-freebsd/lib/msun/src/e_acosf.c \
     upstream-freebsd/lib/msun/src/e_acosh.c \
     upstream-freebsd/lib/msun/src/e_acoshf.c \
-    upstream-freebsd/lib/msun/src/e_asin.c \
     upstream-freebsd/lib/msun/src/e_asinf.c \
-    upstream-freebsd/lib/msun/src/e_atan2.c \
     upstream-freebsd/lib/msun/src/e_atan2f.c \
     upstream-freebsd/lib/msun/src/e_atanh.c \
     upstream-freebsd/lib/msun/src/e_atanhf.c \
-    upstream-freebsd/lib/msun/src/e_cosh.c \
     upstream-freebsd/lib/msun/src/e_coshf.c \
-    upstream-freebsd/lib/msun/src/e_exp.c \
     upstream-freebsd/lib/msun/src/e_expf.c \
     upstream-freebsd/lib/msun/src/e_fmod.c \
     upstream-freebsd/lib/msun/src/e_fmodf.c \
@@ -41,7 +36,6 @@ LOCAL_SRC_FILES := \
     upstream-freebsd/lib/msun/src/e_gammaf.c \
     upstream-freebsd/lib/msun/src/e_gammaf_r.c \
     upstream-freebsd/lib/msun/src/e_gamma_r.c \
-    upstream-freebsd/lib/msun/src/e_hypot.c \
     upstream-freebsd/lib/msun/src/e_hypotf.c \
     upstream-freebsd/lib/msun/src/e_j0.c \
     upstream-freebsd/lib/msun/src/e_j0f.c \
@@ -53,13 +47,10 @@ LOCAL_SRC_FILES := \
     upstream-freebsd/lib/msun/src/e_lgammaf.c \
     upstream-freebsd/lib/msun/src/e_lgammaf_r.c \
     upstream-freebsd/lib/msun/src/e_lgamma_r.c \
-    upstream-freebsd/lib/msun/src/e_log10.c \
     upstream-freebsd/lib/msun/src/e_log10f.c \
     upstream-freebsd/lib/msun/src/e_log2.c \
     upstream-freebsd/lib/msun/src/e_log2f.c \
-    upstream-freebsd/lib/msun/src/e_log.c \
     upstream-freebsd/lib/msun/src/e_logf.c \
-    upstream-freebsd/lib/msun/src/e_pow.c \
     upstream-freebsd/lib/msun/src/e_powf.c \
     upstream-freebsd/lib/msun/src/e_remainder.c \
     upstream-freebsd/lib/msun/src/e_remainderf.c \
@@ -67,10 +58,7 @@ LOCAL_SRC_FILES := \
     upstream-freebsd/lib/msun/src/e_rem_pio2f.c \
     upstream-freebsd/lib/msun/src/e_scalb.c \
     upstream-freebsd/lib/msun/src/e_scalbf.c \
-    upstream-freebsd/lib/msun/src/e_sinh.c \
     upstream-freebsd/lib/msun/src/e_sinhf.c \
-    upstream-freebsd/lib/msun/src/e_sqrt.c \
-    upstream-freebsd/lib/msun/src/e_sqrtf.c \
     upstream-freebsd/lib/msun/src/imprecise.c \
     upstream-freebsd/lib/msun/src/k_cos.c \
     upstream-freebsd/lib/msun/src/k_cosf.c \
@@ -83,17 +71,13 @@ LOCAL_SRC_FILES := \
     upstream-freebsd/lib/msun/src/k_tanf.c \
     upstream-freebsd/lib/msun/src/s_asinh.c \
     upstream-freebsd/lib/msun/src/s_asinhf.c \
-    upstream-freebsd/lib/msun/src/s_atan.c \
     upstream-freebsd/lib/msun/src/s_atanf.c \
     upstream-freebsd/lib/msun/src/s_carg.c \
     upstream-freebsd/lib/msun/src/s_cargf.c \
     upstream-freebsd/lib/msun/src/s_cargl.c \
-    upstream-freebsd/lib/msun/src/s_cbrt.c \
     upstream-freebsd/lib/msun/src/s_cbrtf.c \
     upstream-freebsd/lib/msun/src/s_ccosh.c \
     upstream-freebsd/lib/msun/src/s_ccoshf.c \
-    upstream-freebsd/lib/msun/src/s_ceil.c \
-    upstream-freebsd/lib/msun/src/s_ceilf.c \
     upstream-freebsd/lib/msun/src/s_cexp.c \
     upstream-freebsd/lib/msun/src/s_cexpf.c \
     upstream-freebsd/lib/msun/src/s_cimag.c \
@@ -104,7 +88,6 @@ LOCAL_SRC_FILES := \
     upstream-freebsd/lib/msun/src/s_conjl.c \
     upstream-freebsd/lib/msun/src/s_copysign.c \
     upstream-freebsd/lib/msun/src/s_copysignf.c \
-    upstream-freebsd/lib/msun/src/s_cos.c \
     upstream-freebsd/lib/msun/src/s_cosf.c \
     upstream-freebsd/lib/msun/src/s_cproj.c \
     upstream-freebsd/lib/msun/src/s_cprojf.c \
@@ -123,17 +106,12 @@ LOCAL_SRC_FILES := \
     upstream-freebsd/lib/msun/src/s_erff.c \
     upstream-freebsd/lib/msun/src/s_exp2.c \
     upstream-freebsd/lib/msun/src/s_exp2f.c \
-    upstream-freebsd/lib/msun/src/s_expm1.c \
     upstream-freebsd/lib/msun/src/s_expm1f.c \
     upstream-freebsd/lib/msun/src/s_fabs.c \
     upstream-freebsd/lib/msun/src/s_fabsf.c \
     upstream-freebsd/lib/msun/src/s_fdim.c \
     upstream-freebsd/lib/msun/src/s_finite.c \
     upstream-freebsd/lib/msun/src/s_finitef.c \
-    upstream-freebsd/lib/msun/src/s_floor.c \
-    upstream-freebsd/lib/msun/src/s_floorf.c \
-    upstream-freebsd/lib/msun/src/s_fma.c \
-    upstream-freebsd/lib/msun/src/s_fmaf.c \
     upstream-freebsd/lib/msun/src/s_fmax.c \
     upstream-freebsd/lib/msun/src/s_fmaxf.c \
     upstream-freebsd/lib/msun/src/s_fmin.c \
@@ -142,16 +120,11 @@ LOCAL_SRC_FILES := \
     upstream-freebsd/lib/msun/src/s_frexpf.c \
     upstream-freebsd/lib/msun/src/s_ilogb.c \
     upstream-freebsd/lib/msun/src/s_ilogbf.c \
-    upstream-freebsd/lib/msun/src/s_llrint.c \
-    upstream-freebsd/lib/msun/src/s_llrintf.c \
     upstream-freebsd/lib/msun/src/s_llround.c \
     upstream-freebsd/lib/msun/src/s_llroundf.c \
-    upstream-freebsd/lib/msun/src/s_log1p.c \
     upstream-freebsd/lib/msun/src/s_log1pf.c \
     upstream-freebsd/lib/msun/src/s_logb.c \
     upstream-freebsd/lib/msun/src/s_logbf.c \
-    upstream-freebsd/lib/msun/src/s_lrint.c \
-    upstream-freebsd/lib/msun/src/s_lrintf.c \
     upstream-freebsd/lib/msun/src/s_lround.c \
     upstream-freebsd/lib/msun/src/s_lroundf.c \
     upstream-freebsd/lib/msun/src/s_modf.c \
@@ -162,8 +135,6 @@ LOCAL_SRC_FILES := \
     upstream-freebsd/lib/msun/src/s_nextafterf.c \
     upstream-freebsd/lib/msun/src/s_remquo.c \
     upstream-freebsd/lib/msun/src/s_remquof.c \
-    upstream-freebsd/lib/msun/src/s_rint.c \
-    upstream-freebsd/lib/msun/src/s_rintf.c \
     upstream-freebsd/lib/msun/src/s_round.c \
     upstream-freebsd/lib/msun/src/s_roundf.c \
     upstream-freebsd/lib/msun/src/s_scalbln.c \
@@ -172,15 +143,10 @@ LOCAL_SRC_FILES := \
     upstream-freebsd/lib/msun/src/s_signgam.c \
     upstream-freebsd/lib/msun/src/s_significand.c \
     upstream-freebsd/lib/msun/src/s_significandf.c \
-    upstream-freebsd/lib/msun/src/s_sin.c \
     upstream-freebsd/lib/msun/src/s_sinf.c \
-    upstream-freebsd/lib/msun/src/s_tan.c \
     upstream-freebsd/lib/msun/src/s_tanf.c \
-    upstream-freebsd/lib/msun/src/s_tanh.c \
     upstream-freebsd/lib/msun/src/s_tanhf.c \
     upstream-freebsd/lib/msun/src/s_tgammaf.c \
-    upstream-freebsd/lib/msun/src/s_trunc.c \
-    upstream-freebsd/lib/msun/src/s_truncf.c \
     upstream-freebsd/lib/msun/src/w_cabs.c \
     upstream-freebsd/lib/msun/src/w_cabsf.c \
     upstream-freebsd/lib/msun/src/w_cabsl.c \
@@ -261,23 +227,251 @@ LOCAL_SRC_FILES += \
 LOCAL_SRC_FILES += \
     signbit.c \
 
+# Arch specific optimizations.
+
+# -----------------------------------------------------------------------------
+# arm
+# -----------------------------------------------------------------------------
 LOCAL_SRC_FILES_arm += \
     arm/fenv.c \
+    upstream-freebsd/lib/msun/src/e_acos.c \
+    upstream-freebsd/lib/msun/src/e_asin.c \
+    upstream-freebsd/lib/msun/src/e_atan2.c \
+    upstream-freebsd/lib/msun/src/e_cosh.c \
+    upstream-freebsd/lib/msun/src/e_exp.c \
+    upstream-freebsd/lib/msun/src/e_hypot.c \
+    upstream-freebsd/lib/msun/src/e_log.c \
+    upstream-freebsd/lib/msun/src/e_log10.c \
+    upstream-freebsd/lib/msun/src/e_pow.c \
+    upstream-freebsd/lib/msun/src/e_sinh.c \
+    upstream-freebsd/lib/msun/src/s_atan.c \
+    upstream-freebsd/lib/msun/src/s_cbrt.c \
+    upstream-freebsd/lib/msun/src/s_ceil.c \
+    upstream-freebsd/lib/msun/src/s_ceilf.c \
+    upstream-freebsd/lib/msun/src/s_cos.c \
+    upstream-freebsd/lib/msun/src/s_fma.c \
+    upstream-freebsd/lib/msun/src/s_fmaf.c \
+    upstream-freebsd/lib/msun/src/s_floorf.c \
+    upstream-freebsd/lib/msun/src/s_expm1.c \
+    upstream-freebsd/lib/msun/src/s_llrint.c \
+    upstream-freebsd/lib/msun/src/s_llrintf.c \
+    upstream-freebsd/lib/msun/src/s_log1p.c \
+    upstream-freebsd/lib/msun/src/s_lrint.c \
+    upstream-freebsd/lib/msun/src/s_lrintf.c \
+    upstream-freebsd/lib/msun/src/s_rint.c \
+    upstream-freebsd/lib/msun/src/s_rintf.c \
+    upstream-freebsd/lib/msun/src/s_sin.c \
+    upstream-freebsd/lib/msun/src/s_tan.c \
+    upstream-freebsd/lib/msun/src/s_tanh.c \
+    upstream-freebsd/lib/msun/src/s_trunc.c \
+    upstream-freebsd/lib/msun/src/s_truncf.c \
 
+# s_floor.S requires neon instructions.
+ifdef TARGET_2ND_ARCH
+arch_variant := $(TARGET_2ND_ARCH_VARIANT)
+else
+arch_variant := $(TARGET_ARCH_VARIANT)
+endif
+
+# Use the C version on armv7-a since it doesn't support neon instructions.
+ifeq ($(arch_variant),armv7-a)
+LOCAL_SRC_FILES_arm += \
+    upstream-freebsd/lib/msun/src/e_sqrt.c \
+    upstream-freebsd/lib/msun/src/e_sqrtf.c \
+    upstream-freebsd/lib/msun/src/s_floor.c \
+
+else
+LOCAL_SRC_FILES_arm += \
+    arm/e_sqrt.S \
+    arm/e_sqrtf.S \
+    arm/s_floor.S \
+
+endif
+
+# -----------------------------------------------------------------------------
+# arm64
+# -----------------------------------------------------------------------------
 LOCAL_SRC_FILES_arm64 += \
+    arm64/ceil.S \
     arm64/fenv.c \
+    arm64/fma.S \
+    arm64/floor.S \
+    arm64/lrint.S \
+    arm64/rint.S \
+    arm64/sqrt.S \
+    arm64/trunc.S \
+    upstream-freebsd/lib/msun/src/e_acos.c \
+    upstream-freebsd/lib/msun/src/e_asin.c \
+    upstream-freebsd/lib/msun/src/e_atan2.c \
+    upstream-freebsd/lib/msun/src/e_cosh.c \
+    upstream-freebsd/lib/msun/src/e_exp.c \
+    upstream-freebsd/lib/msun/src/e_hypot.c \
+    upstream-freebsd/lib/msun/src/e_log.c \
+    upstream-freebsd/lib/msun/src/e_log10.c \
+    upstream-freebsd/lib/msun/src/e_pow.c \
+    upstream-freebsd/lib/msun/src/e_sinh.c \
+    upstream-freebsd/lib/msun/src/s_atan.c \
+    upstream-freebsd/lib/msun/src/s_cbrt.c \
+    upstream-freebsd/lib/msun/src/s_cos.c \
+    upstream-freebsd/lib/msun/src/s_expm1.c \
+    upstream-freebsd/lib/msun/src/s_log1p.c \
+    upstream-freebsd/lib/msun/src/s_sin.c \
+    upstream-freebsd/lib/msun/src/s_tan.c \
+    upstream-freebsd/lib/msun/src/s_tanh.c \
 
-LOCAL_SRC_FILES_mips += \
+# -----------------------------------------------------------------------------
+# mips
+# -----------------------------------------------------------------------------
+libm_mips_arch_files := \
     mips/fenv.c \
+    upstream-freebsd/lib/msun/src/e_acos.c \
+    upstream-freebsd/lib/msun/src/e_asin.c \
+    upstream-freebsd/lib/msun/src/e_atan2.c \
+    upstream-freebsd/lib/msun/src/e_cosh.c \
+    upstream-freebsd/lib/msun/src/e_exp.c \
+    upstream-freebsd/lib/msun/src/e_hypot.c \
+    upstream-freebsd/lib/msun/src/e_log.c \
+    upstream-freebsd/lib/msun/src/e_log10.c \
+    upstream-freebsd/lib/msun/src/e_pow.c \
+    upstream-freebsd/lib/msun/src/e_sinh.c \
+    upstream-freebsd/lib/msun/src/e_sqrt.c \
+    upstream-freebsd/lib/msun/src/e_sqrtf.c \
+    upstream-freebsd/lib/msun/src/s_atan.c \
+    upstream-freebsd/lib/msun/src/s_cbrt.c \
+    upstream-freebsd/lib/msun/src/s_ceil.c \
+    upstream-freebsd/lib/msun/src/s_ceilf.c \
+    upstream-freebsd/lib/msun/src/s_cos.c \
+    upstream-freebsd/lib/msun/src/s_fma.c \
+    upstream-freebsd/lib/msun/src/s_fmaf.c \
+    upstream-freebsd/lib/msun/src/s_floor.c \
+    upstream-freebsd/lib/msun/src/s_floorf.c \
+    upstream-freebsd/lib/msun/src/s_expm1.c \
+    upstream-freebsd/lib/msun/src/s_llrint.c \
+    upstream-freebsd/lib/msun/src/s_llrintf.c \
+    upstream-freebsd/lib/msun/src/s_log1p.c \
+    upstream-freebsd/lib/msun/src/s_lrint.c \
+    upstream-freebsd/lib/msun/src/s_lrintf.c \
+    upstream-freebsd/lib/msun/src/s_rint.c \
+    upstream-freebsd/lib/msun/src/s_rintf.c \
+    upstream-freebsd/lib/msun/src/s_sin.c \
+    upstream-freebsd/lib/msun/src/s_tan.c \
+    upstream-freebsd/lib/msun/src/s_tanh.c \
+    upstream-freebsd/lib/msun/src/s_trunc.c \
+    upstream-freebsd/lib/msun/src/s_truncf.c \
 
-LOCAL_SRC_FILES_mips64 += \
+LOCAL_SRC_FILES_mips += $(libm_mips_arch_files)
-    mips/fenv.c \
+LOCAL_SRC_FILES_mips64 += $(libm_mips_arch_files)
 
+# -----------------------------------------------------------------------------
+# x86
+# -----------------------------------------------------------------------------
 LOCAL_SRC_FILES_x86 += \
     i387/fenv.c \
+    upstream-freebsd/lib/msun/src/s_fma.c \
+    upstream-freebsd/lib/msun/src/s_fmaf.c \
+    upstream-freebsd/lib/msun/src/s_llrint.c \
+    upstream-freebsd/lib/msun/src/s_llrintf.c \
+    upstream-freebsd/lib/msun/src/s_lrint.c \
+    upstream-freebsd/lib/msun/src/s_lrintf.c \
+    upstream-freebsd/lib/msun/src/s_rint.c \
+    upstream-freebsd/lib/msun/src/s_rintf.c \
+    x86/sqrt.S \
+    x86/sqrtf.S \
+    x86/e_acos.S \
+    x86/e_asin.S \
+    x86/e_atan2.S \
+    x86/e_cosh.S \
+    x86/e_exp.S \
+    x86/e_hypot.S \
+    x86/e_log10.S \
+    x86/e_log.S \
+    x86/e_pow.S \
+    x86/e_sinh.S \
+    x86/libm_reduce_pi04l.S \
+    x86/libm_sincos_huge.S \
+    x86/libm_tancot_huge.S \
+    x86/s_atan.S \
+    x86/s_cbrt.S \
+    x86/s_cos.S \
+    x86/s_expm1.S \
+    x86/s_log1p.S \
+    x86/s_sin.S \
+    x86/s_tanh.S \
+    x86/s_tan.S \
 
+ifeq ($(ARCH_X86_HAVE_SSE4_1),true)
+LOCAL_SRC_FILES_x86 += \
+    x86/ceil.S \
+    x86/ceilf.S \
+    x86/floor.S \
+    x86/floorf.S \
+    x86/trunc.S \
+    x86/truncf.S \
+
+else
+LOCAL_SRC_FILES_x86 += \
+    upstream-freebsd/lib/msun/src/s_ceil.c \
+    upstream-freebsd/lib/msun/src/s_ceilf.c \
+    upstream-freebsd/lib/msun/src/s_floor.c \
+    upstream-freebsd/lib/msun/src/s_floorf.c \
+    upstream-freebsd/lib/msun/src/s_trunc.c \
+    upstream-freebsd/lib/msun/src/s_truncf.c \
+
+endif
+
+# -----------------------------------------------------------------------------
+# x86_64
+# -----------------------------------------------------------------------------
 LOCAL_SRC_FILES_x86_64 += \
     amd64/fenv.c \
+    upstream-freebsd/lib/msun/src/s_fma.c \
+    upstream-freebsd/lib/msun/src/s_fmaf.c \
+    upstream-freebsd/lib/msun/src/s_llrint.c \
+    upstream-freebsd/lib/msun/src/s_llrintf.c \
+    upstream-freebsd/lib/msun/src/s_lrint.c \
+    upstream-freebsd/lib/msun/src/s_lrintf.c \
+    upstream-freebsd/lib/msun/src/s_rint.c \
+    upstream-freebsd/lib/msun/src/s_rintf.c \
+    x86_64/sqrt.S \
+    x86_64/sqrtf.S \
+    x86_64/e_acos.S \
+    x86_64/e_asin.S \
+    x86_64/e_atan2.S \
+    x86_64/e_cosh.S \
+    x86_64/e_exp.S \
+    x86_64/e_hypot.S \
+    x86_64/e_log10.S \
+    x86_64/e_log.S \
+    x86_64/e_pow.S \
+    x86_64/e_sinh.S \
+    x86_64/s_atan.S \
+    x86_64/s_cbrt.S \
+    x86_64/s_cos.S \
+    x86_64/s_expm1.S \
+    x86_64/s_log1p.S \
+    x86_64/s_sin.S \
+    x86_64/s_tanh.S \
+    x86_64/s_tan.S \
+
+ifeq ($(ARCH_X86_HAVE_SSE4_1),true)
+LOCAL_SRC_FILES_x86_64 += \
+    x86_64/ceil.S \
+    x86_64/ceilf.S \
+    x86_64/floor.S \
+    x86_64/floorf.S \
+    x86_64/trunc.S \
+    x86_64/truncf.S \
+
+else
+LOCAL_SRC_FILES_x86_64 += \
+    upstream-freebsd/lib/msun/src/s_ceil.c \
+    upstream-freebsd/lib/msun/src/s_ceilf.c \
+    upstream-freebsd/lib/msun/src/s_floor.c \
+    upstream-freebsd/lib/msun/src/s_floorf.c \
+    upstream-freebsd/lib/msun/src/s_trunc.c \
+    upstream-freebsd/lib/msun/src/s_truncf.c \
+
+endif
 
 LOCAL_C_INCLUDES_x86 += $(LOCAL_PATH)/i387
 
@@ -285,7 +479,6 @@ LOCAL_C_INCLUDES += $(LOCAL_PATH)/upstream-freebsd/lib/msun/src/
 LOCAL_C_INCLUDES_64 += $(LOCAL_PATH)/upstream-freebsd/lib/msun/ld128/
 
 LOCAL_CLANG := $(libm_clang)
-LOCAL_ADDITIONAL_DEPENDENCIES := $(LOCAL_PATH)/Android.mk
 LOCAL_ARM_MODE := arm
 LOCAL_CFLAGS := \
     -DFLT_EVAL_METHOD=0 \
@@ -297,6 +490,9 @@ LOCAL_CFLAGS := \
     -Wno-unknown-pragmas \
     -fvisibility=hidden \
 
+LOCAL_ASFLAGS := \
+    -Ibionic/libc \
+
 # Workaround the GCC "(long)fn -> lfn" optimization bug which will result in
 # self recursions for lrint, lrintf, and lrintl.
 # BUG: 14225968
@@ -317,9 +513,11 @@ include $(BUILD_STATIC_LIBRARY)
 # -----------------------------------------------------------------------------
 include $(CLEAR_VARS)
 
+# TODO: This is to work around b/19059885. Remove after root cause is fixed
+LOCAL_LDFLAGS_arm := -Wl,--hash-style=sysv
+
 LOCAL_MODULE := libm
 LOCAL_CLANG := $(libm_clang)
-LOCAL_ADDITIONAL_DEPENDENCIES := $(LOCAL_PATH)/Android.mk
 LOCAL_SYSTEM_SHARED_LIBRARIES := libc
 LOCAL_WHOLE_STATIC_LIBRARIES := libm
 

libm/arm/e_sqrt.S

@@ -0,0 +1,42 @@
+/*
+ * Copyright (c) 2013-2014, NVIDIA Corporation.  All rights reserved.
+ * Johnny Qiu <joqiu@nvidia.com>
+ * Shu Zhang <chazhang@nvidia.com>
+ *
+ * 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 The Linux Foundation 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 "AS IS" AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT
+ * 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 <float.h>
+#include <private/bionic_asm.h>
+
+ENTRY(sqrt)
+    vmov.f64    d0, r0, r1
+    vsqrt.f64   d0, d0
+    vmov.f64    r0, r1, d0
+    bx          lr
+END(sqrt)
+
+ALIAS_SYMBOL(sqrtl, sqrt);

libm/arm/e_sqrtf.S

@@ -0,0 +1,39 @@
+/*
+ * Copyright (c) 2013-2014, NVIDIA Corporation.  All rights reserved.
+ * Johhnny Qiu <joqiu@nvidia.com>
+ * Shu Zhang <chazhang@nvidia.com>
+ *
+ * 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 The Linux Foundation 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 "AS IS" AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
+ * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+ * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
+ * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
+ * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <private/bionic_asm.h>
+
+ENTRY(sqrtf)
+    vmov.f32    s0, r0
+    vsqrt.f32   s0, s0
+    vmov.f32    r0, s0
+    bx          lr
+END(sqrtf)

libm/arm/s_floor.S

@@ -0,0 +1,139 @@
+/*
+ * Copyright (c) 2013-2014, NVIDIA Corporation.  All rights reserved.
+ * Johnny Qiu <joqiu@nvidia.com>
+ * Shu Zhang <chazhang@nvidia.com>
+ *
+ * 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 The Linux Foundation 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 "AS IS" AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT
+ * 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 <float.h>
+#include <private/bionic_asm.h>
+
+ENTRY(floor)    /* x in r0, r1 */
+
+        and             r3, r1, #0x80000000     /* sign(x) */
+        bic             r1, r1, #0x80000000     /* x = abs(x) */
+
+        /* extract exp of x */
+        lsr             r2, r1, #20
+        sub             r2, r2, #0x3fc
+        subs            r2, r2, #0x3            /* r2 <- exp */
+
+        /* |x| < 1.0? */
+        blt             .Lx_lt_one
+
+        /* x < 0? */
+        cmp             r3, #0
+        bne             .Lclr_frac_neg
+
+        /* |x| <= 2^20? */
+        cmp             r2, #20
+        ble             .Lclr_frac_r1
+
+        /* |x| < 2^52? */
+        cmp             r2, #52
+        blt             .Lclr_frac_r0
+
+        /* return x */
+        bx              lr
+
+.Lclr_frac_r1:
+        rsb             r2, r2, #20
+        lsr             r1, r1, r2
+        lsl             r1, r1, r2
+        mov             r0, #0
+        bx              lr
+
+.Lclr_frac_r0:
+        rsb             r2, r2, #52
+        lsr             r0, r0, r2
+        lsl             r0, r0, r2
+        bx              lr
+
+.Lclr_frac_neg:
+        /* |x| <= 2^20? */
+        cmp             r2, #20
+        ble             .Lclr_frac_r1_neg
+
+        /* |x| < 2^52? */
+        cmp             r2, #52
+        blt             .Lclr_frac_r0_neg
+
+        /* return x */
+        orr             r1, r1, #0x80000000
+        bx              lr
+
+.Lclr_frac_r1_neg:
+        rsb             r2, r2, #20
+        mov             r3, #1
+        lsl             r3, r3, r2
+        sub             r3, r3, #1
+        and             r3, r1, r3
+        orr             r3, r3, r0
+        lsr             r1, r1, r2
+        lsl             r1, r1, r2
+        mov             r0, #0
+        b               .Lreturn_x_neg
+
+.Lclr_frac_r0_neg:
+        rsb             r2, r2, #52
+        mov             r3, #1
+        lsl             r3, r3, r2
+        sub             r3, r3, #1
+        and             r3, r0, r3
+        lsr             r0, r0, r2
+        lsl             r0, r0, r2
+        b               .Lreturn_x_neg
+
+.Lx_lt_one:
+        /* x == +-0? */
+        cmp             r0, #0
+        cmpeq           r1, #0
+        orreq           r1, r1, r3
+        bxeq            lr
+
+        /* (x > 0) ? 0 : -1 */
+        mov             r1, #0x00100000
+        mov             r0, #0
+        cmp             r3, #0
+        movne           r1, #0xc0000000
+        sub             r1, r1, #0x00100000
+        bx              lr
+
+.Lreturn_x_neg:
+        cmp             r3, #0
+        orr             r1, r1, #0x80000000
+        bxeq            lr
+
+        vmov            d16, r0, r1
+        vmov.f64        d18, #1.0
+        vsub.f64        d16, d16, d18
+        vmov            r0, r1, d16
+        bx              lr
+
+END(floor)
+
+ALIAS_SYMBOL(floorl, floor);

libm/arm64/ceil.S

@@ -0,0 +1,27 @@
+/*
+ * Copyright 2015, 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 <private/bionic_asm.h>
+
+ENTRY(ceil)
+  frintP d0, d0
+  ret
+END(ceil)
+
+ENTRY(ceilf)
+  frintP s0, s0
+  ret
+END(ceilf)

libm/arm64/floor.S

@@ -0,0 +1,27 @@
+/*
+ * Copyright 2015, 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 <private/bionic_asm.h>
+
+ENTRY(floor)
+  frintM d0, d0
+  ret
+END(floor)
+
+ENTRY(floorf)
+  frintM s0, s0
+  ret
+END(floorf)

libm/arm64/fma.S

@@ -0,0 +1,27 @@
+/*
+ * Copyright 2015, 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 <private/bionic_asm.h>
+
+ENTRY(fma)
+  fmadd d0, d0, d1, d2
+  ret
+END(fma)
+
+ENTRY(fmaf)
+  fmadd s0, s0, s1, s2
+  ret
+END(fmaf)

libm/arm64/lrint.S

@@ -0,0 +1,34 @@
+/*
+ * Copyright 2015, 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 <private/bionic_asm.h>
+
+ENTRY(lrint)
+  frintX d0, d0
+  fcvtzs x0, d0
+  ret
+END(lrint)
+
+ENTRY(lrintf)
+  frintX s0, s0
+  fcvtzs x0, s0
+  ret
+END(lrintf)
+
+// sizeof(long) and sizeof(long long) are the same for aarch64
+ALIAS_SYMBOL(llrint, lrint);
+
+ALIAS_SYMBOL(llrintf, lrintf);

libm/arm64/rint.S

@@ -0,0 +1,27 @@
+/*
+ * Copyright 2015, 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 <private/bionic_asm.h>
+
+ENTRY(rint)
+  frintX d0, d0
+  ret
+END(rint)
+
+ENTRY(rintf)
+  frintX s0, s0
+  ret
+END(rintf)

libm/arm64/sqrt.S

@@ -0,0 +1,27 @@
+/*
+ * Copyright 2015, 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 <private/bionic_asm.h>
+
+ENTRY(sqrt)
+  fsqrt d0, d0
+  ret
+END(sqrt)
+
+ENTRY(sqrtf)
+  fsqrt s0, s0
+  ret
+END(sqrtf)

libm/arm64/trunc.S

@@ -0,0 +1,27 @@
+/*
+ * Copyright 2015, 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 <private/bionic_asm.h>
+
+ENTRY(trunc)
+  frintZ d0, d0
+  ret
+END(trunc)
+
+ENTRY(truncf)
+  frintZ s0, s0
+  ret
+END(truncf)

libm/x86/ceil.S

@@ -0,0 +1,43 @@
+/*
+Copyright (c) 2014, Intel 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.
+
+    * Neither the name of Intel Corporation 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 OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#include <private/bionic_asm.h>
+
+ENTRY(ceil)
+	mov    %esp,%eax
+	and    $0xfffffff8,%eax
+	movsd  0x4(%esp),%xmm0
+	roundsd $0x2,%xmm0,%xmm0
+	movlpd %xmm0,-0x8(%eax)
+	fldl   -0x8(%eax)
+	ret
+END(ceil)
+
+ALIAS_SYMBOL(ceill, ceil);

libm/x86/ceilf.S

@@ -0,0 +1,39 @@
+/*
+Copyright (c) 2014, Intel 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.
+
+    * Neither the name of Intel Corporation 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 OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#include <private/bionic_asm.h>
+
+ENTRY(ceilf)
+	movss  0x4(%esp),%xmm0
+	roundss $0x2,%xmm0,%xmm0
+	movss  %xmm0,-0x4(%esp)
+	flds   -0x4(%esp)
+	ret
+END(ceilf)