zircon/system/utest/core/pthread/pthread.cc

// Copyright 2016 The Fuchsia Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

#include <errno.h>
#include <inttypes.h>
#include <lib/zx/clock.h>
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <unistd.h>
#include <zircon/syscalls.h>
#include <zircon/utc.h>

#include <atomic>

#include <zxtest/zxtest.h>

namespace {

static pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
static pthread_cond_t cond = PTHREAD_COND_INITIALIZER;

// These are accessed by by both the cond_threads, and the main thread. The latter does so without
// holding the mutex.
static std::atomic_int thread_waked = 0;
static std::atomic_int ready_count = 0;

static int thread_with_lock = 0;

static void log(const char* str) {
  struct timespec time;
  clock_gettime(CLOCK_REALTIME, &time);
  printf("[%08lu.%08lu]: %s", time.tv_sec, time.tv_nsec / 1000, str);
}

static void* mutex_thread_1(void* arg) {
  log("thread 1 grabbing mutex\n");
  pthread_mutex_lock(&mutex);
  log("thread 1 got mutex\n");
  thread_with_lock = 1;
  zx_nanosleep(zx_deadline_after(ZX_MSEC(300)));

  // Make sure no other thread woke up
  EXPECT_EQ(thread_with_lock, 1, "Only thread 1 should have woken up");
  log("thread 1 releasing mutex\n");
  pthread_mutex_unlock(&mutex);
  log("thread 1 done\n");
  return NULL;
}

static void* mutex_thread_2(void* arg) {
  zx_nanosleep(zx_deadline_after(ZX_MSEC(100)));
  log("thread 2 grabbing mutex\n");
  pthread_mutex_lock(&mutex);
  log("thread 2 got mutex\n");
  thread_with_lock = 2;

  zx_nanosleep(zx_deadline_after(ZX_MSEC(300)));

  // Make sure no other thread woke up
  EXPECT_EQ(thread_with_lock, 2, "Only thread 2 should have woken up");

  log("thread 2 releasing mutex\n");
  pthread_mutex_unlock(&mutex);
  log("thread 2 done\n");
  return NULL;
}

static void* mutex_thread_3(void* arg) {
  zx_nanosleep(zx_deadline_after(ZX_MSEC(100)));
  log("thread 3 grabbing mutex\n");
  pthread_mutex_lock(&mutex);
  log("thread 3 got mutex\n");
  thread_with_lock = 3;

  zx_nanosleep(zx_deadline_after(ZX_MSEC(300)));

  // Make sure no other thread woke up
  EXPECT_EQ(thread_with_lock, 3, "Only thread 3 should have woken up");

  log("thread 3 releasing mutex\n");
  pthread_mutex_unlock(&mutex);
  log("thread 3 done\n");
  return NULL;
}

static void* cond_thread1(void* arg) {
  pthread_mutex_lock(&mutex);
  log("thread 1 waiting on condition\n");
  ready_count.fetch_add(1);
  pthread_cond_wait(&cond, &mutex);
  log("thread 1 waiting again\n");
  ready_count.fetch_add(1);
  pthread_cond_wait(&cond, &mutex);
  thread_waked.fetch_add(1);
  pthread_mutex_unlock(&mutex);
  log("thread 1 done\n");
  return NULL;
}

static void* cond_thread2(void* arg) {
  pthread_mutex_lock(&mutex);
  log("thread 2 waiting on condition\n");
  ready_count.fetch_add(1);
  pthread_cond_wait(&cond, &mutex);
  log("thread 2 waiting again\n");
  ready_count.fetch_add(1);
  pthread_cond_wait(&cond, &mutex);
  thread_waked.fetch_add(1);
  pthread_mutex_unlock(&mutex);
  log("thread 2 done\n");
  return NULL;
}

static void* cond_thread3(void* arg) {
  pthread_mutex_lock(&mutex);
  log("thread 3 waiting on condition\n");
  ready_count.fetch_add(1);
  pthread_cond_wait(&cond, &mutex);
  log("thread 3 waiting again\n");
  ready_count.fetch_add(1);
  pthread_cond_wait(&cond, &mutex);
  thread_waked.fetch_add(1);
  pthread_mutex_unlock(&mutex);
  log("thread 3 done\n");
  return NULL;
}

// Polls for the correct wake count. This is expected to be fast. It
// polls since there's no great way to otherwise observe the scheduler
// state in this test. 10 seconds is chosen to be enough less than
// infinity to allow other tests to complete should this hang, and big
// enough to allow most plausible delays in scheduling.
static bool poll_waked_threads(int expected_count) {
  zx_time_t start = zx_clock_get_monotonic();
  for (int tries = 0; tries < (ZX_SEC(10) / ZX_MSEC(1)); ++tries) {
    zx_nanosleep(start + ZX_MSEC(tries));
    if (thread_waked.load() == expected_count) {
      return true;
    }
  }
  return false;
}

// Poll |var| until it has reached or exceeded |value|, then acquire and release |mutex|.
//
// The purpose of this function is to allow the caller to wait until |value| threads have entered a
// condition wait protected by |mutex|. For this to work, |var| must be incremented while holding
// the |mutex| before issuing the cond wait.
static void wait_for_count(const std::atomic_int& var, int value, pthread_mutex_t* mutex) {
  while (var.load() < value) {
    zx_nanosleep(zx_deadline_after(ZX_MSEC(100)));
  }
  pthread_mutex_lock(mutex);
  pthread_mutex_unlock(mutex);
}

class PthreadTest : public zxtest::Test {
 public:
  static constexpr int64_t kBackstopTime = 123456789;

  static void SetUpTestSuite() {
    // pthread_cond_timedwait relies on the system UTC clock which might not have been set or might
    // not have been started when these tests are run. Install a new test clock for the duration of
    // the test case.
    ASSERT_FALSE(clock_installed_);
    zx::clock test_clock;
    zx_clock_create_args_v1_t create_args{.backstop_time = kBackstopTime};
    ASSERT_OK(zx::clock::create(ZX_CLOCK_OPT_AUTO_START, &create_args, &test_clock));
    ASSERT_OK(zx_utc_reference_swap(test_clock.release(), runtime_clock_.reset_and_get_address()));
    clock_installed_ = true;
  }

  static void TearDownTestSuite() {
    // If we replaced the UTC clock reference, restore it back to the original.
    if (clock_installed_) {
      zx::clock release_me;
      zx_utc_reference_swap(runtime_clock_.release(), release_me.reset_and_get_address());
      clock_installed_ = false;
    } else {
      runtime_clock_.reset();
    }
  }

 private:
  inline static zx::clock runtime_clock_;
  inline static bool clock_installed_ = false;
};

TEST_F(PthreadTest, Basic) {
  pthread_t thread1, thread2, thread3;

  log("testing uncontested case\n");
  pthread_mutex_lock(&mutex);
  pthread_mutex_unlock(&mutex);
  log("mutex locked and unlocked\n");

  log("starting cond threads\n");
  pthread_create(&thread1, NULL, cond_thread1, NULL);
  pthread_create(&thread2, NULL, cond_thread2, NULL);
  pthread_create(&thread3, NULL, cond_thread3, NULL);

  // Wait for all 3 to reach the first cond wait before broadcasting.
  wait_for_count(ready_count, 3, &mutex);
  ready_count.store(0);

  log("calling pthread_cond_broadcast\n");
  pthread_cond_broadcast(&cond);

  // Wait until they all reach the second cond wait before signaling one at a time.
  wait_for_count(ready_count, 3, &mutex);
  ready_count.store(0);

  log("calling pthread_cond_signal\n");
  pthread_cond_signal(&cond);
  EXPECT_TRUE(poll_waked_threads(1), "Exactly 1 process should have woken up");

  log("calling pthread_cond_signal\n");
  pthread_cond_signal(&cond);
  EXPECT_TRUE(poll_waked_threads(2), "Exactly 2 processes should have woken up");

  log("calling pthread_cond_signal\n");
  pthread_cond_signal(&cond);
  EXPECT_TRUE(poll_waked_threads(3), "Exactly 3 processes should have woken up");

  log("joining cond threads\n");
  pthread_join(thread1, NULL);
  log("cond_thread 1 joined\n");
  pthread_join(thread2, NULL);
  log("cond_thread 2 joined\n");
  pthread_join(thread3, NULL);
  log("cond_thread 3 joined\n");

  pthread_mutex_lock(&mutex);
  log("waiting on condition with 2 second timeout\n");
  struct timespec delay;
  ASSERT_EQ(clock_gettime(CLOCK_REALTIME, &delay), 0);
  delay.tv_sec += 2;
  int result = pthread_cond_timedwait(&cond, &mutex, &delay);
  pthread_mutex_unlock(&mutex);
  log("pthread_cond_timedwait returned\n");

  EXPECT_EQ(result, ETIMEDOUT, "Lock should have timeout");

  log("creating mutex threads\n");
  pthread_create(&thread1, NULL, mutex_thread_1, NULL);
  pthread_create(&thread2, NULL, mutex_thread_2, NULL);
  pthread_create(&thread3, NULL, mutex_thread_3, NULL);

  log("joining mutex threads\n");
  pthread_join(thread1, NULL);
  log("thread 1 joined\n");
  pthread_join(thread2, NULL);
  log("thread 2 joined\n");
  pthread_join(thread3, NULL);
  log("thread 3 joined\n");
}

TEST_F(PthreadTest, SelfMainThread) {
  pthread_t self = pthread_self();
  pthread_t null = 0;
  ASSERT_NE(self, null, "pthread_self() was NULL");
}

// Pick a stack size well bigger than the default, which is <1MB.
constexpr size_t stack_size = 16u << 20;

static void big_stack_check() {
  // Stack allocate a lot, but less than the full stack size.
  volatile uint8_t buffer[stack_size / 2];
  uint8_t value = 0u;
  for (auto& byte : buffer)
    byte = value++;

  uint64_t sum = 0u;
  uint64_t expected_sum = 0u;
  value = 0u;
  for (auto& byte : buffer) {
    sum += byte;
    expected_sum += value++;
  }

  ASSERT_EQ(sum, expected_sum, "buffer corrupted");
}

static void* bigger_stack_thread(void* unused) {
  big_stack_check();
  return nullptr;
}

TEST_F(PthreadTest, BigStackSize) {
  pthread_t thread;
  pthread_attr_t attr;
  int result = pthread_attr_init(&attr);
  ASSERT_EQ(result, 0, "failed to initialize pthread attributes");
  result = pthread_attr_setstacksize(&attr, stack_size);
  ASSERT_EQ(result, 0, "failed to set stack size");
  result = pthread_create(&thread, &attr, bigger_stack_thread, nullptr);
  ASSERT_EQ(result, 0, "failed to start thread");
  result = pthread_join(thread, nullptr);
  ASSERT_EQ(result, 0, "failed to join thread");
}

static void pthread_getstack_check() {
  pthread_attr_t attr;
  int result = pthread_getattr_np(pthread_self(), &attr);
  ASSERT_EQ(result, 0, "pthread_getattr_np failed");

  void* stack_base;
  size_t stack_size;
  result = pthread_attr_getstack(&attr, &stack_base, &stack_size);
  ASSERT_EQ(result, 0, "pthread_attr_getstack failed");

  // Convert the reported bounds of the stack into something we can
  // compare against.
  uintptr_t low = reinterpret_cast<uintptr_t>(stack_base);
  uintptr_t high = low + stack_size;

  // This is just some arbitrary address known to be on our thread stack.
  // Note this is the "safe stack".  If using -fsanitize=safe-stack,
  // there is also an "unsafe stack", where e.g. &attr will reside.
  uintptr_t here = reinterpret_cast<uintptr_t>(__builtin_frame_address(0));

  printf("pthread_attr_getstack reports [%#" PRIxPTR ", %#" PRIxPTR "); SP ~= %#" PRIxPTR "\n", low,
         high, here);

  ASSERT_LT(low, here, "reported stack base not below actual SP");
  ASSERT_GT(high, here, "reported stack end not above actual SP");
}

TEST_F(PthreadTest, GetstackMainThread) { pthread_getstack_check(); }

static void* getstack_thread(void*) {
  pthread_getstack_check();
  return reinterpret_cast<void*>(static_cast<uintptr_t>(true));
}

static void pthread_getstack_on_new_thread(const pthread_attr_t* attr) {
  pthread_t thread;
  int result = pthread_create(&thread, attr, &getstack_thread, nullptr);
  ASSERT_EQ(result, 0, "pthread_create failed");
  void* thread_result;
  result = pthread_join(thread, &thread_result);
  ASSERT_EQ(result, 0, "pthread_join failed");

  bool ok = static_cast<bool>(reinterpret_cast<uintptr_t>(thread_result));
  ASSERT_TRUE(ok, "pthread_attr_getstack on another thread");
}

TEST_F(PthreadTest, GetstackOtherThread) { pthread_getstack_on_new_thread(nullptr); }

TEST_F(PthreadTest, GetstackOtherThreadExplicitSize) {
  pthread_attr_t attr;
  int result = pthread_attr_init(&attr);
  ASSERT_EQ(result, 0, "pthread_attr_init failed");
  result = pthread_attr_setstacksize(&attr, 1 << 20);
  ASSERT_EQ(result, 0, "pthread_attr_setstacksize failed");

  pthread_getstack_on_new_thread(&attr);
}

}  // namespace