rtc_base/critical_section_unittest.cc

/*
 *  Copyright 2014 The WebRTC Project Authors. All rights reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#include "rtc_base/critical_section.h"

#include <stddef.h>
#include <stdint.h>

#include <memory>
#include <set>
#include <utility>
#include <vector>

#include "rtc_base/arraysize.h"
#include "rtc_base/atomic_ops.h"
#include "rtc_base/checks.h"
#include "rtc_base/event.h"
#include "rtc_base/location.h"
#include "rtc_base/message_handler.h"
#include "rtc_base/message_queue.h"
#include "rtc_base/platform_thread.h"
#include "rtc_base/thread.h"
#include "test/gtest.h"

namespace rtc {

namespace {

const int kLongTime = 10000;  // 10 seconds
const int kNumThreads = 16;
const int kOperationsToRun = 1000;

class UniqueValueVerifier {
 public:
  void Verify(const std::vector<int>& values) {
    for (size_t i = 0; i < values.size(); ++i) {
      std::pair<std::set<int>::iterator, bool> result =
          all_values_.insert(values[i]);
      // Each value should only be taken by one thread, so if this value
      // has already been added, something went wrong.
      EXPECT_TRUE(result.second)
          << " Thread=" << Thread::Current() << " value=" << values[i];
    }
  }

  void Finalize() {}

 private:
  std::set<int> all_values_;
};

class CompareAndSwapVerifier {
 public:
  CompareAndSwapVerifier() : zero_count_(0) {}

  void Verify(const std::vector<int>& values) {
    for (auto v : values) {
      if (v == 0) {
        EXPECT_EQ(0, zero_count_) << "Thread=" << Thread::Current();
        ++zero_count_;
      } else {
        EXPECT_EQ(1, v) << " Thread=" << Thread::Current();
      }
    }
  }

  void Finalize() { EXPECT_EQ(1, zero_count_); }

 private:
  int zero_count_;
};

class RunnerBase : public MessageHandler {
 public:
  explicit RunnerBase(int value)
      : threads_active_(0),
        start_event_(true, false),
        done_event_(true, false),
        shared_value_(value) {}

  bool Run() {
    // Signal all threads to start.
    start_event_.Set();

    // Wait for all threads to finish.
    return done_event_.Wait(kLongTime);
  }

  void SetExpectedThreadCount(int count) { threads_active_ = count; }

  int shared_value() const { return shared_value_; }

 protected:
  // Derived classes must override OnMessage, and call BeforeStart and AfterEnd
  // at the beginning and the end of OnMessage respectively.
  void BeforeStart() { ASSERT_TRUE(start_event_.Wait(kLongTime)); }

  // Returns true if all threads have finished.
  bool AfterEnd() {
    if (AtomicOps::Decrement(&threads_active_) == 0) {
      done_event_.Set();
      return true;
    }
    return false;
  }

  int threads_active_;
  Event start_event_;
  Event done_event_;
  int shared_value_;
};

class RTC_LOCKABLE CriticalSectionLock {
 public:
  void Lock() RTC_EXCLUSIVE_LOCK_FUNCTION() { cs_.Enter(); }
  void Unlock() RTC_UNLOCK_FUNCTION() { cs_.Leave(); }

 private:
  CriticalSection cs_;
};

template <class Lock>
class LockRunner : public RunnerBase {
 public:
  LockRunner() : RunnerBase(0) {}

  void OnMessage(Message* msg) override {
    BeforeStart();

    lock_.Lock();

    EXPECT_EQ(0, shared_value_);
    int old = shared_value_;

    // Use a loop to increase the chance of race.
    for (int i = 0; i < kOperationsToRun; ++i) {
      ++shared_value_;
    }
    EXPECT_EQ(old + kOperationsToRun, shared_value_);
    shared_value_ = 0;

    lock_.Unlock();

    AfterEnd();
  }

 private:
  Lock lock_;
};

template <class Op, class Verifier>
class AtomicOpRunner : public RunnerBase {
 public:
  explicit AtomicOpRunner(int initial_value) : RunnerBase(initial_value) {}

  void OnMessage(Message* msg) override {
    BeforeStart();

    std::vector<int> values;
    values.reserve(kOperationsToRun);

    // Generate a bunch of values by updating shared_value_ atomically.
    for (int i = 0; i < kOperationsToRun; ++i) {
      values.push_back(Op::AtomicOp(&shared_value_));
    }

    {  // Add them all to the set.
      CritScope cs(&all_values_crit_);
      verifier_.Verify(values);
    }

    if (AfterEnd()) {
      verifier_.Finalize();
    }
  }

 private:
  CriticalSection all_values_crit_;
  Verifier verifier_;
};

struct IncrementOp {
  static int AtomicOp(int* i) { return AtomicOps::Increment(i); }
};

struct DecrementOp {
  static int AtomicOp(int* i) { return AtomicOps::Decrement(i); }
};

struct CompareAndSwapOp {
  static int AtomicOp(int* i) { return AtomicOps::CompareAndSwap(i, 0, 1); }
};

void StartThreads(std::vector<std::unique_ptr<Thread>>* threads,
                  MessageHandler* handler) {
  for (int i = 0; i < kNumThreads; ++i) {
    std::unique_ptr<Thread> thread(Thread::Create());
    thread->Start();
    thread->Post(RTC_FROM_HERE, handler);
    threads->push_back(std::move(thread));
  }
}

}  // namespace

TEST(AtomicOpsTest, Simple) {
  int value = 0;
  EXPECT_EQ(1, AtomicOps::Increment(&value));
  EXPECT_EQ(1, value);
  EXPECT_EQ(2, AtomicOps::Increment(&value));
  EXPECT_EQ(2, value);
  EXPECT_EQ(1, AtomicOps::Decrement(&value));
  EXPECT_EQ(1, value);
  EXPECT_EQ(0, AtomicOps::Decrement(&value));
  EXPECT_EQ(0, value);
}

TEST(AtomicOpsTest, SimplePtr) {
  class Foo {};
  Foo* volatile foo = nullptr;
  std::unique_ptr<Foo> a(new Foo());
  std::unique_ptr<Foo> b(new Foo());
  // Reading the initial value should work as expected.
  EXPECT_TRUE(rtc::AtomicOps::AcquireLoadPtr(&foo) == nullptr);
  // Setting using compare and swap should work.
  EXPECT_TRUE(rtc::AtomicOps::CompareAndSwapPtr(
                  &foo, static_cast<Foo*>(nullptr), a.get()) == nullptr);
  EXPECT_TRUE(rtc::AtomicOps::AcquireLoadPtr(&foo) == a.get());
  // Setting another value but with the wrong previous pointer should fail
  // (remain a).
  EXPECT_TRUE(rtc::AtomicOps::CompareAndSwapPtr(
                  &foo, static_cast<Foo*>(nullptr), b.get()) == a.get());
  EXPECT_TRUE(rtc::AtomicOps::AcquireLoadPtr(&foo) == a.get());
  // Replacing a with b should work.
  EXPECT_TRUE(rtc::AtomicOps::CompareAndSwapPtr(&foo, a.get(), b.get()) ==
              a.get());
  EXPECT_TRUE(rtc::AtomicOps::AcquireLoadPtr(&foo) == b.get());
}

TEST(AtomicOpsTest, Increment) {
  // Create and start lots of threads.
  AtomicOpRunner<IncrementOp, UniqueValueVerifier> runner(0);
  std::vector<std::unique_ptr<Thread>> threads;
  StartThreads(&threads, &runner);
  runner.SetExpectedThreadCount(kNumThreads);

  // Release the hounds!
  EXPECT_TRUE(runner.Run());
  EXPECT_EQ(kOperationsToRun * kNumThreads, runner.shared_value());
}

TEST(AtomicOpsTest, Decrement) {
  // Create and start lots of threads.
  AtomicOpRunner<DecrementOp, UniqueValueVerifier> runner(kOperationsToRun *
                                                          kNumThreads);
  std::vector<std::unique_ptr<Thread>> threads;
  StartThreads(&threads, &runner);
  runner.SetExpectedThreadCount(kNumThreads);

  // Release the hounds!
  EXPECT_TRUE(runner.Run());
  EXPECT_EQ(0, runner.shared_value());
}

TEST(AtomicOpsTest, CompareAndSwap) {
  // Create and start lots of threads.
  AtomicOpRunner<CompareAndSwapOp, CompareAndSwapVerifier> runner(0);
  std::vector<std::unique_ptr<Thread>> threads;
  StartThreads(&threads, &runner);
  runner.SetExpectedThreadCount(kNumThreads);

  // Release the hounds!
  EXPECT_TRUE(runner.Run());
  EXPECT_EQ(1, runner.shared_value());
}

TEST(GlobalLockTest, Basic) {
  // Create and start lots of threads.
  LockRunner<GlobalLock> runner;
  std::vector<std::unique_ptr<Thread>> threads;
  StartThreads(&threads, &runner);
  runner.SetExpectedThreadCount(kNumThreads);

  // Release the hounds!
  EXPECT_TRUE(runner.Run());
  EXPECT_EQ(0, runner.shared_value());
}

TEST(CriticalSectionTest, Basic) {
  // Create and start lots of threads.
  LockRunner<CriticalSectionLock> runner;
  std::vector<std::unique_ptr<Thread>> threads;
  StartThreads(&threads, &runner);
  runner.SetExpectedThreadCount(kNumThreads);

  // Release the hounds!
  EXPECT_TRUE(runner.Run());
  EXPECT_EQ(0, runner.shared_value());
}

class PerfTestData {
 public:
  PerfTestData(int expected_count, Event* event)
      : cache_line_barrier_1_(),
        cache_line_barrier_2_(),
        expected_count_(expected_count),
        event_(event) {
    cache_line_barrier_1_[0]++;  // Avoid 'is not used'.
    cache_line_barrier_2_[0]++;  // Avoid 'is not used'.
  }
  ~PerfTestData() {}

  void AddToCounter(int add) {
    rtc::CritScope cs(&lock_);
    my_counter_ += add;
    if (my_counter_ == expected_count_)
      event_->Set();
  }

  int64_t total() const {
    // Assume that only one thread is running now.
    return my_counter_;
  }

 private:
  uint8_t cache_line_barrier_1_[64];
  CriticalSection lock_;
  uint8_t cache_line_barrier_2_[64];
  int64_t my_counter_ = 0;
  const int expected_count_;
  Event* const event_;
};

class PerfTestThread {
 public:
  PerfTestThread() : thread_(&ThreadFunc, this, "CsPerf") {}

  void Start(PerfTestData* data, int repeats, int id) {
    RTC_DCHECK(!thread_.IsRunning());
    RTC_DCHECK(!data_);
    data_ = data;
    repeats_ = repeats;
    my_id_ = id;
    thread_.Start();
  }

  void Stop() {
    RTC_DCHECK(thread_.IsRunning());
    RTC_DCHECK(data_);
    thread_.Stop();
    repeats_ = 0;
    data_ = nullptr;
    my_id_ = 0;
  }

 private:
  static void ThreadFunc(void* param) {
    PerfTestThread* me = static_cast<PerfTestThread*>(param);
    for (int i = 0; i < me->repeats_; ++i)
      me->data_->AddToCounter(me->my_id_);
  }

  PlatformThread thread_;
  PerfTestData* data_ = nullptr;
  int repeats_ = 0;
  int my_id_ = 0;
};

// Comparison of output of this test as tested on a MacBook Pro, 13-inch,
// 2017, 3,5 GHz Intel Core i7, 16 GB 2133 MHz LPDDR3,
// running macOS Mojave, 10.14.3.
//
// Native mutex implementation using fair policy (previously macOS default):
// Approximate CPU usage:
// real    4m54.612s
// user    1m20.575s
// sys     3m48.872s
// Unit test output:
// [       OK ] CriticalSectionTest.Performance (294375 ms)
//
// Native mutex implementation using first fit policy (current macOS default):
// Approximate CPU usage:
// real    0m11.535s
// user    0m12.738s
// sys     0m31.207s
// Unit test output:
// [       OK ] CriticalSectionTest.Performance (11444 ms)
//
// Special partially spin lock based implementation:
// Approximate CPU usage:
// real    0m2.113s
// user    0m3.014s
// sys     0m4.495s
// Unit test output:
// [       OK ] CriticalSectionTest.Performance (1885 ms)
//
// The test is disabled by default to avoid unecessarily loading the bots.
TEST(CriticalSectionTest, DISABLED_Performance) {
  PerfTestThread threads[8];
  Event event;

  static const int kThreadRepeats = 10000000;
  static const int kExpectedCount = kThreadRepeats * arraysize(threads);
  PerfTestData test_data(kExpectedCount, &event);

  for (auto& t : threads)
    t.Start(&test_data, kThreadRepeats, 1);

  event.Wait(Event::kForever);

  for (auto& t : threads)
    t.Stop();
}

}  // namespace rtc