platform_bionic/tests/pthread_test.cpp

2975 lines
99 KiB
C++
Raw Normal View History

/*
* 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 <gtest/gtest.h>
#include <errno.h>
#include <inttypes.h>
#include <limits.h>
#include <malloc.h>
#include <pthread.h>
#include <signal.h>
#include <stdio.h>
#include <sys/mman.h>
#include <sys/prctl.h>
#include <sys/resource.h>
#include <sys/syscall.h>
#include <time.h>
#include <unistd.h>
#include <unwind.h>
#include <atomic>
#include <future>
#include <vector>
#include <android-base/macros.h>
#include <android-base/parseint.h>
#include <android-base/scopeguard.h>
#include <android-base/strings.h>
#include "private/bionic_constants.h"
#include "BionicDeathTest.h"
#include "SignalUtils.h"
#include "utils.h"
TEST(pthread, pthread_key_create) {
pthread_key_t key;
ASSERT_EQ(0, pthread_key_create(&key, nullptr));
ASSERT_EQ(0, pthread_key_delete(key));
// Can't delete a key that's already been deleted.
ASSERT_EQ(EINVAL, pthread_key_delete(key));
}
TEST(pthread, pthread_keys_max) {
// POSIX says PTHREAD_KEYS_MAX should be at least _POSIX_THREAD_KEYS_MAX.
ASSERT_GE(PTHREAD_KEYS_MAX, _POSIX_THREAD_KEYS_MAX);
}
TEST(pthread, sysconf_SC_THREAD_KEYS_MAX_eq_PTHREAD_KEYS_MAX) {
int sysconf_max = sysconf(_SC_THREAD_KEYS_MAX);
ASSERT_EQ(sysconf_max, PTHREAD_KEYS_MAX);
}
TEST(pthread, pthread_key_many_distinct) {
// As gtest uses pthread keys, we can't allocate exactly PTHREAD_KEYS_MAX
// pthread keys, but We should be able to allocate at least this many keys.
int nkeys = PTHREAD_KEYS_MAX / 2;
std::vector<pthread_key_t> keys;
auto scope_guard = android::base::make_scope_guard([&keys] {
for (const auto& key : keys) {
EXPECT_EQ(0, pthread_key_delete(key));
}
});
for (int i = 0; i < nkeys; ++i) {
pthread_key_t key;
// If this fails, it's likely that LIBC_PTHREAD_KEY_RESERVED_COUNT is wrong.
ASSERT_EQ(0, pthread_key_create(&key, nullptr)) << i << " of " << nkeys;
keys.push_back(key);
ASSERT_EQ(0, pthread_setspecific(key, reinterpret_cast<void*>(i)));
}
for (int i = keys.size() - 1; i >= 0; --i) {
ASSERT_EQ(reinterpret_cast<void*>(i), pthread_getspecific(keys.back()));
pthread_key_t key = keys.back();
keys.pop_back();
ASSERT_EQ(0, pthread_key_delete(key));
}
}
TEST(pthread, pthread_key_not_exceed_PTHREAD_KEYS_MAX) {
std::vector<pthread_key_t> keys;
int rv = 0;
// Pthread keys are used by gtest, so PTHREAD_KEYS_MAX should
// be more than we are allowed to allocate now.
for (int i = 0; i < PTHREAD_KEYS_MAX; i++) {
pthread_key_t key;
rv = pthread_key_create(&key, nullptr);
if (rv == EAGAIN) {
break;
}
EXPECT_EQ(0, rv);
keys.push_back(key);
}
// Don't leak keys.
for (const auto& key : keys) {
EXPECT_EQ(0, pthread_key_delete(key));
}
keys.clear();
// We should have eventually reached the maximum number of keys and received
// EAGAIN.
ASSERT_EQ(EAGAIN, rv);
}
TEST(pthread, pthread_key_delete) {
void* expected = reinterpret_cast<void*>(1234);
pthread_key_t key;
ASSERT_EQ(0, pthread_key_create(&key, nullptr));
ASSERT_EQ(0, pthread_setspecific(key, expected));
ASSERT_EQ(expected, pthread_getspecific(key));
ASSERT_EQ(0, pthread_key_delete(key));
// After deletion, pthread_getspecific returns nullptr.
ASSERT_EQ(nullptr, pthread_getspecific(key));
// And you can't use pthread_setspecific with the deleted key.
ASSERT_EQ(EINVAL, pthread_setspecific(key, expected));
}
TEST(pthread, pthread_key_fork) {
void* expected = reinterpret_cast<void*>(1234);
pthread_key_t key;
ASSERT_EQ(0, pthread_key_create(&key, nullptr));
ASSERT_EQ(0, pthread_setspecific(key, expected));
ASSERT_EQ(expected, pthread_getspecific(key));
pid_t pid = fork();
ASSERT_NE(-1, pid) << strerror(errno);
if (pid == 0) {
// The surviving thread inherits all the forking thread's TLS values...
ASSERT_EQ(expected, pthread_getspecific(key));
_exit(99);
}
AssertChildExited(pid, 99);
ASSERT_EQ(expected, pthread_getspecific(key));
ASSERT_EQ(0, pthread_key_delete(key));
}
static void* DirtyKeyFn(void* key) {
return pthread_getspecific(*reinterpret_cast<pthread_key_t*>(key));
}
TEST(pthread, pthread_key_dirty) {
pthread_key_t key;
ASSERT_EQ(0, pthread_key_create(&key, nullptr));
size_t stack_size = 640 * 1024;
void* stack = mmap(nullptr, stack_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
ASSERT_NE(MAP_FAILED, stack);
memset(stack, 0xff, stack_size);
pthread_attr_t attr;
ASSERT_EQ(0, pthread_attr_init(&attr));
ASSERT_EQ(0, pthread_attr_setstack(&attr, stack, stack_size));
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, &attr, DirtyKeyFn, &key));
void* result;
ASSERT_EQ(0, pthread_join(t, &result));
ASSERT_EQ(nullptr, result); // Not ~0!
ASSERT_EQ(0, munmap(stack, stack_size));
ASSERT_EQ(0, pthread_key_delete(key));
}
TEST(pthread, static_pthread_key_used_before_creation) {
#if defined(__BIONIC__)
// See http://b/19625804. The bug is about a static/global pthread key being used before creation.
// So here tests if the static/global default value 0 can be detected as invalid key.
static pthread_key_t key;
ASSERT_EQ(nullptr, pthread_getspecific(key));
ASSERT_EQ(EINVAL, pthread_setspecific(key, nullptr));
ASSERT_EQ(EINVAL, pthread_key_delete(key));
#else
GTEST_SKIP() << "bionic-only test";
#endif
}
static void* IdFn(void* arg) {
return arg;
}
class SpinFunctionHelper {
public:
SpinFunctionHelper() {
SpinFunctionHelper::spin_flag_ = true;
}
~SpinFunctionHelper() {
UnSpin();
}
auto GetFunction() -> void* (*)(void*) {
return SpinFunctionHelper::SpinFn;
}
void UnSpin() {
SpinFunctionHelper::spin_flag_ = false;
}
private:
static void* SpinFn(void*) {
while (spin_flag_) {}
return nullptr;
}
static std::atomic<bool> spin_flag_;
};
// It doesn't matter if spin_flag_ is used in several tests,
// because it is always set to false after each test. Each thread
// loops on spin_flag_ can find it becomes false at some time.
std::atomic<bool> SpinFunctionHelper::spin_flag_;
static void* JoinFn(void* arg) {
return reinterpret_cast<void*>(pthread_join(reinterpret_cast<pthread_t>(arg), nullptr));
}
static void AssertDetached(pthread_t t, bool is_detached) {
pthread_attr_t attr;
ASSERT_EQ(0, pthread_getattr_np(t, &attr));
int detach_state;
ASSERT_EQ(0, pthread_attr_getdetachstate(&attr, &detach_state));
pthread_attr_destroy(&attr);
ASSERT_EQ(is_detached, (detach_state == PTHREAD_CREATE_DETACHED));
}
static void MakeDeadThread(pthread_t& t) {
ASSERT_EQ(0, pthread_create(&t, nullptr, IdFn, nullptr));
ASSERT_EQ(0, pthread_join(t, nullptr));
}
TEST(pthread, pthread_create) {
void* expected_result = reinterpret_cast<void*>(123);
// Can we create a thread?
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, nullptr, IdFn, expected_result));
// If we join, do we get the expected value back?
void* result;
ASSERT_EQ(0, pthread_join(t, &result));
ASSERT_EQ(expected_result, result);
}
TEST(pthread, pthread_create_EAGAIN) {
pthread_attr_t attributes;
ASSERT_EQ(0, pthread_attr_init(&attributes));
ASSERT_EQ(0, pthread_attr_setstacksize(&attributes, static_cast<size_t>(-1) & ~(getpagesize() - 1)));
pthread_t t;
ASSERT_EQ(EAGAIN, pthread_create(&t, &attributes, IdFn, nullptr));
}
TEST(pthread, pthread_no_join_after_detach) {
SpinFunctionHelper spin_helper;
pthread_t t1;
ASSERT_EQ(0, pthread_create(&t1, nullptr, spin_helper.GetFunction(), nullptr));
// After a pthread_detach...
ASSERT_EQ(0, pthread_detach(t1));
AssertDetached(t1, true);
// ...pthread_join should fail.
ASSERT_EQ(EINVAL, pthread_join(t1, nullptr));
}
TEST(pthread, pthread_no_op_detach_after_join) {
SpinFunctionHelper spin_helper;
pthread_t t1;
ASSERT_EQ(0, pthread_create(&t1, nullptr, spin_helper.GetFunction(), nullptr));
// If thread 2 is already waiting to join thread 1...
pthread_t t2;
ASSERT_EQ(0, pthread_create(&t2, nullptr, JoinFn, reinterpret_cast<void*>(t1)));
sleep(1); // (Give t2 a chance to call pthread_join.)
#if defined(__BIONIC__)
ASSERT_EQ(EINVAL, pthread_detach(t1));
#else
ASSERT_EQ(0, pthread_detach(t1));
#endif
AssertDetached(t1, false);
spin_helper.UnSpin();
// ...but t2's join on t1 still goes ahead (which we can tell because our join on t2 finishes).
void* join_result;
ASSERT_EQ(0, pthread_join(t2, &join_result));
ASSERT_EQ(0U, reinterpret_cast<uintptr_t>(join_result));
}
TEST(pthread, pthread_join_self) {
ASSERT_EQ(EDEADLK, pthread_join(pthread_self(), nullptr));
}
struct TestBug37410 {
pthread_t main_thread;
pthread_mutex_t mutex;
static void main() {
TestBug37410 data;
data.main_thread = pthread_self();
ASSERT_EQ(0, pthread_mutex_init(&data.mutex, nullptr));
ASSERT_EQ(0, pthread_mutex_lock(&data.mutex));
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, nullptr, TestBug37410::thread_fn, reinterpret_cast<void*>(&data)));
// Wait for the thread to be running...
ASSERT_EQ(0, pthread_mutex_lock(&data.mutex));
ASSERT_EQ(0, pthread_mutex_unlock(&data.mutex));
// ...and exit.
pthread_exit(nullptr);
}
private:
static void* thread_fn(void* arg) {
TestBug37410* data = reinterpret_cast<TestBug37410*>(arg);
// Unlocking data->mutex will cause the main thread to exit, invalidating *data. Save the handle.
pthread_t main_thread = data->main_thread;
// Let the main thread know we're running.
pthread_mutex_unlock(&data->mutex);
// And wait for the main thread to exit.
pthread_join(main_thread, nullptr);
return nullptr;
}
};
// Even though this isn't really a death test, we have to say "DeathTest" here so gtest knows to
// run this test (which exits normally) in its own process.
class pthread_DeathTest : public BionicDeathTest {};
TEST_F(pthread_DeathTest, pthread_bug_37410) {
// http://code.google.com/p/android/issues/detail?id=37410
ASSERT_EXIT(TestBug37410::main(), ::testing::ExitedWithCode(0), "");
}
static void* SignalHandlerFn(void* arg) {
sigset64_t wait_set;
sigfillset64(&wait_set);
return reinterpret_cast<void*>(sigwait64(&wait_set, reinterpret_cast<int*>(arg)));
}
TEST(pthread, pthread_sigmask) {
// Check that SIGUSR1 isn't blocked.
sigset_t original_set;
sigemptyset(&original_set);
ASSERT_EQ(0, pthread_sigmask(SIG_BLOCK, nullptr, &original_set));
ASSERT_FALSE(sigismember(&original_set, SIGUSR1));
// Block SIGUSR1.
sigset_t set;
sigemptyset(&set);
sigaddset(&set, SIGUSR1);
ASSERT_EQ(0, pthread_sigmask(SIG_BLOCK, &set, nullptr));
// Check that SIGUSR1 is blocked.
sigset_t final_set;
sigemptyset(&final_set);
ASSERT_EQ(0, pthread_sigmask(SIG_BLOCK, nullptr, &final_set));
ASSERT_TRUE(sigismember(&final_set, SIGUSR1));
// ...and that sigprocmask agrees with pthread_sigmask.
sigemptyset(&final_set);
ASSERT_EQ(0, sigprocmask(SIG_BLOCK, nullptr, &final_set));
ASSERT_TRUE(sigismember(&final_set, SIGUSR1));
// Spawn a thread that calls sigwait and tells us what it received.
pthread_t signal_thread;
int received_signal = -1;
ASSERT_EQ(0, pthread_create(&signal_thread, nullptr, SignalHandlerFn, &received_signal));
// Send that thread SIGUSR1.
pthread_kill(signal_thread, SIGUSR1);
// See what it got.
void* join_result;
ASSERT_EQ(0, pthread_join(signal_thread, &join_result));
ASSERT_EQ(SIGUSR1, received_signal);
ASSERT_EQ(0U, reinterpret_cast<uintptr_t>(join_result));
// Restore the original signal mask.
ASSERT_EQ(0, pthread_sigmask(SIG_SETMASK, &original_set, nullptr));
}
TEST(pthread, pthread_sigmask64_SIGTRMIN) {
// Check that SIGRTMIN isn't blocked.
sigset64_t original_set;
sigemptyset64(&original_set);
ASSERT_EQ(0, pthread_sigmask64(SIG_BLOCK, nullptr, &original_set));
ASSERT_FALSE(sigismember64(&original_set, SIGRTMIN));
// Block SIGRTMIN.
sigset64_t set;
sigemptyset64(&set);
sigaddset64(&set, SIGRTMIN);
ASSERT_EQ(0, pthread_sigmask64(SIG_BLOCK, &set, nullptr));
// Check that SIGRTMIN is blocked.
sigset64_t final_set;
sigemptyset64(&final_set);
ASSERT_EQ(0, pthread_sigmask64(SIG_BLOCK, nullptr, &final_set));
ASSERT_TRUE(sigismember64(&final_set, SIGRTMIN));
// ...and that sigprocmask64 agrees with pthread_sigmask64.
sigemptyset64(&final_set);
ASSERT_EQ(0, sigprocmask64(SIG_BLOCK, nullptr, &final_set));
ASSERT_TRUE(sigismember64(&final_set, SIGRTMIN));
// Spawn a thread that calls sigwait64 and tells us what it received.
pthread_t signal_thread;
int received_signal = -1;
ASSERT_EQ(0, pthread_create(&signal_thread, nullptr, SignalHandlerFn, &received_signal));
// Send that thread SIGRTMIN.
pthread_kill(signal_thread, SIGRTMIN);
// See what it got.
void* join_result;
ASSERT_EQ(0, pthread_join(signal_thread, &join_result));
ASSERT_EQ(SIGRTMIN, received_signal);
ASSERT_EQ(0U, reinterpret_cast<uintptr_t>(join_result));
// Restore the original signal mask.
ASSERT_EQ(0, pthread_sigmask64(SIG_SETMASK, &original_set, nullptr));
}
static void test_pthread_setname_np__pthread_getname_np(pthread_t t) {
ASSERT_EQ(0, pthread_setname_np(t, "short"));
char name[32];
ASSERT_EQ(0, pthread_getname_np(t, name, sizeof(name)));
ASSERT_STREQ("short", name);
// The limit is 15 characters --- the kernel's buffer is 16, but includes a NUL.
ASSERT_EQ(0, pthread_setname_np(t, "123456789012345"));
ASSERT_EQ(0, pthread_getname_np(t, name, sizeof(name)));
ASSERT_STREQ("123456789012345", name);
ASSERT_EQ(ERANGE, pthread_setname_np(t, "1234567890123456"));
// The passed-in buffer should be at least 16 bytes.
ASSERT_EQ(0, pthread_getname_np(t, name, 16));
ASSERT_EQ(ERANGE, pthread_getname_np(t, name, 15));
}
TEST(pthread, pthread_setname_np__pthread_getname_np__self) {
test_pthread_setname_np__pthread_getname_np(pthread_self());
}
TEST(pthread, pthread_setname_np__pthread_getname_np__other) {
SpinFunctionHelper spin_helper;
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, nullptr, spin_helper.GetFunction(), nullptr));
test_pthread_setname_np__pthread_getname_np(t);
spin_helper.UnSpin();
ASSERT_EQ(0, pthread_join(t, nullptr));
}
// http://b/28051133: a kernel misfeature means that you can't change the
// name of another thread if you've set PR_SET_DUMPABLE to 0.
TEST(pthread, pthread_setname_np__pthread_getname_np__other_PR_SET_DUMPABLE) {
ASSERT_EQ(0, prctl(PR_SET_DUMPABLE, 0)) << strerror(errno);
SpinFunctionHelper spin_helper;
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, nullptr, spin_helper.GetFunction(), nullptr));
test_pthread_setname_np__pthread_getname_np(t);
spin_helper.UnSpin();
ASSERT_EQ(0, pthread_join(t, nullptr));
}
Be more strict about using invalid `pthread_t`s. Another release, another attempt to remove the global thread list. But this time, let's admit that it's not going away. We can switch to using a read/write lock for the global thread list, and to aborting rather than quietly returning ESRCH if we're given an invalid pthread_t. This change affects pthread_detach, pthread_getcpuclockid, pthread_getschedparam/pthread_setschedparam, pthread_join, and pthread_kill: instead of returning ESRCH when passed an invalid pthread_t, if you're targeting O or above, they'll abort with the message "attempt to use invalid pthread_t". Note that this doesn't change behavior as much as you might think: the old lookup only held the global thread list lock for the duration of the lookup, so there was still a race between that and the dereference in the caller, given that callers actually need the tid to pass to some syscall or other, and sometimes update fields in the pthread_internal_t struct too. (This patch replaces such users with calls to pthread_gettid_np, which at least makes the TOCTOU window smaller.) We can't check thread->tid against 0 to see whether a pthread_t is still valid because a dead thread gets its thread struct unmapped along with its stack, so the dereference isn't safe. Taking the affected functions one by one: * pthread_getcpuclockid and pthread_getschedparam/pthread_setschedparam should be fine. Unsafe calls to those seem highly unlikely. * Unsafe pthread_detach callers probably want to switch to pthread_attr_setdetachstate instead, or using pthread_detach(pthread_self()) from the new thread's start routine rather than doing the detach in the parent. * pthread_join calls should be safe anyway, because a joinable thread won't actually exit and unmap until it's joined. If you're joining an unjoinable thread, the fix is to stop marking it detached. If you're joining an already-joined thread, you need to rethink your design. * Unsafe pthread_kill calls aren't portably fixable. (And are obviously inherently non-portable as-is.) The best alternative on Android is to use pthread_gettid_np at some point that you know the thread to be alive, and then call kill/tgkill directly. That's still not completely safe because if you're too late, the tid may have been reused, but then your code is inherently unsafe anyway. Bug: http://b/19636317 Test: ran tests Change-Id: I0372c4428e8a7f1c3af5c9334f5d9c25f2c73f21
2017-02-14 02:59:29 +01:00
TEST_F(pthread_DeathTest, pthread_setname_np__no_such_thread) {
pthread_t dead_thread;
MakeDeadThread(dead_thread);
EXPECT_DEATH(pthread_setname_np(dead_thread, "short 3"),
"invalid pthread_t (.*) passed to pthread_setname_np");
}
TEST_F(pthread_DeathTest, pthread_setname_np__null_thread) {
pthread_t null_thread = 0;
EXPECT_EQ(ENOENT, pthread_setname_np(null_thread, "short 3"));
Be more strict about using invalid `pthread_t`s. Another release, another attempt to remove the global thread list. But this time, let's admit that it's not going away. We can switch to using a read/write lock for the global thread list, and to aborting rather than quietly returning ESRCH if we're given an invalid pthread_t. This change affects pthread_detach, pthread_getcpuclockid, pthread_getschedparam/pthread_setschedparam, pthread_join, and pthread_kill: instead of returning ESRCH when passed an invalid pthread_t, if you're targeting O or above, they'll abort with the message "attempt to use invalid pthread_t". Note that this doesn't change behavior as much as you might think: the old lookup only held the global thread list lock for the duration of the lookup, so there was still a race between that and the dereference in the caller, given that callers actually need the tid to pass to some syscall or other, and sometimes update fields in the pthread_internal_t struct too. (This patch replaces such users with calls to pthread_gettid_np, which at least makes the TOCTOU window smaller.) We can't check thread->tid against 0 to see whether a pthread_t is still valid because a dead thread gets its thread struct unmapped along with its stack, so the dereference isn't safe. Taking the affected functions one by one: * pthread_getcpuclockid and pthread_getschedparam/pthread_setschedparam should be fine. Unsafe calls to those seem highly unlikely. * Unsafe pthread_detach callers probably want to switch to pthread_attr_setdetachstate instead, or using pthread_detach(pthread_self()) from the new thread's start routine rather than doing the detach in the parent. * pthread_join calls should be safe anyway, because a joinable thread won't actually exit and unmap until it's joined. If you're joining an unjoinable thread, the fix is to stop marking it detached. If you're joining an already-joined thread, you need to rethink your design. * Unsafe pthread_kill calls aren't portably fixable. (And are obviously inherently non-portable as-is.) The best alternative on Android is to use pthread_gettid_np at some point that you know the thread to be alive, and then call kill/tgkill directly. That's still not completely safe because if you're too late, the tid may have been reused, but then your code is inherently unsafe anyway. Bug: http://b/19636317 Test: ran tests Change-Id: I0372c4428e8a7f1c3af5c9334f5d9c25f2c73f21
2017-02-14 02:59:29 +01:00
}
TEST_F(pthread_DeathTest, pthread_getname_np__no_such_thread) {
pthread_t dead_thread;
MakeDeadThread(dead_thread);
char name[64];
EXPECT_DEATH(pthread_getname_np(dead_thread, name, sizeof(name)),
"invalid pthread_t (.*) passed to pthread_getname_np");
}
TEST_F(pthread_DeathTest, pthread_getname_np__null_thread) {
pthread_t null_thread = 0;
char name[64];
EXPECT_EQ(ENOENT, pthread_getname_np(null_thread, name, sizeof(name)));
}
TEST(pthread, pthread_kill__0) {
// Signal 0 just tests that the thread exists, so it's safe to call on ourselves.
ASSERT_EQ(0, pthread_kill(pthread_self(), 0));
}
TEST(pthread, pthread_kill__invalid_signal) {
ASSERT_EQ(EINVAL, pthread_kill(pthread_self(), -1));
}
static void pthread_kill__in_signal_handler_helper(int signal_number) {
static int count = 0;
ASSERT_EQ(SIGALRM, signal_number);
if (++count == 1) {
// Can we call pthread_kill from a signal handler?
ASSERT_EQ(0, pthread_kill(pthread_self(), SIGALRM));
}
}
TEST(pthread, pthread_kill__in_signal_handler) {
ScopedSignalHandler ssh(SIGALRM, pthread_kill__in_signal_handler_helper);
ASSERT_EQ(0, pthread_kill(pthread_self(), SIGALRM));
}
TEST(pthread, pthread_kill__exited_thread) {
static std::promise<pid_t> tid_promise;
pthread_t thread;
ASSERT_EQ(0, pthread_create(&thread, nullptr,
[](void*) -> void* {
tid_promise.set_value(gettid());
return nullptr;
},
nullptr));
pid_t tid = tid_promise.get_future().get();
while (TEMP_FAILURE_RETRY(syscall(__NR_tgkill, getpid(), tid, 0)) != -1) {
continue;
}
ASSERT_EQ(ESRCH, errno);
ASSERT_EQ(ESRCH, pthread_kill(thread, 0));
}
Be more strict about using invalid `pthread_t`s. Another release, another attempt to remove the global thread list. But this time, let's admit that it's not going away. We can switch to using a read/write lock for the global thread list, and to aborting rather than quietly returning ESRCH if we're given an invalid pthread_t. This change affects pthread_detach, pthread_getcpuclockid, pthread_getschedparam/pthread_setschedparam, pthread_join, and pthread_kill: instead of returning ESRCH when passed an invalid pthread_t, if you're targeting O or above, they'll abort with the message "attempt to use invalid pthread_t". Note that this doesn't change behavior as much as you might think: the old lookup only held the global thread list lock for the duration of the lookup, so there was still a race between that and the dereference in the caller, given that callers actually need the tid to pass to some syscall or other, and sometimes update fields in the pthread_internal_t struct too. (This patch replaces such users with calls to pthread_gettid_np, which at least makes the TOCTOU window smaller.) We can't check thread->tid against 0 to see whether a pthread_t is still valid because a dead thread gets its thread struct unmapped along with its stack, so the dereference isn't safe. Taking the affected functions one by one: * pthread_getcpuclockid and pthread_getschedparam/pthread_setschedparam should be fine. Unsafe calls to those seem highly unlikely. * Unsafe pthread_detach callers probably want to switch to pthread_attr_setdetachstate instead, or using pthread_detach(pthread_self()) from the new thread's start routine rather than doing the detach in the parent. * pthread_join calls should be safe anyway, because a joinable thread won't actually exit and unmap until it's joined. If you're joining an unjoinable thread, the fix is to stop marking it detached. If you're joining an already-joined thread, you need to rethink your design. * Unsafe pthread_kill calls aren't portably fixable. (And are obviously inherently non-portable as-is.) The best alternative on Android is to use pthread_gettid_np at some point that you know the thread to be alive, and then call kill/tgkill directly. That's still not completely safe because if you're too late, the tid may have been reused, but then your code is inherently unsafe anyway. Bug: http://b/19636317 Test: ran tests Change-Id: I0372c4428e8a7f1c3af5c9334f5d9c25f2c73f21
2017-02-14 02:59:29 +01:00
TEST_F(pthread_DeathTest, pthread_detach__no_such_thread) {
pthread_t dead_thread;
MakeDeadThread(dead_thread);
EXPECT_DEATH(pthread_detach(dead_thread),
"invalid pthread_t (.*) passed to pthread_detach");
}
TEST_F(pthread_DeathTest, pthread_detach__null_thread) {
pthread_t null_thread = 0;
EXPECT_EQ(ESRCH, pthread_detach(null_thread));
}
TEST(pthread, pthread_getcpuclockid__clock_gettime) {
SpinFunctionHelper spin_helper;
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, nullptr, spin_helper.GetFunction(), nullptr));
clockid_t c;
ASSERT_EQ(0, pthread_getcpuclockid(t, &c));
timespec ts;
ASSERT_EQ(0, clock_gettime(c, &ts));
spin_helper.UnSpin();
ASSERT_EQ(0, pthread_join(t, nullptr));
}
Be more strict about using invalid `pthread_t`s. Another release, another attempt to remove the global thread list. But this time, let's admit that it's not going away. We can switch to using a read/write lock for the global thread list, and to aborting rather than quietly returning ESRCH if we're given an invalid pthread_t. This change affects pthread_detach, pthread_getcpuclockid, pthread_getschedparam/pthread_setschedparam, pthread_join, and pthread_kill: instead of returning ESRCH when passed an invalid pthread_t, if you're targeting O or above, they'll abort with the message "attempt to use invalid pthread_t". Note that this doesn't change behavior as much as you might think: the old lookup only held the global thread list lock for the duration of the lookup, so there was still a race between that and the dereference in the caller, given that callers actually need the tid to pass to some syscall or other, and sometimes update fields in the pthread_internal_t struct too. (This patch replaces such users with calls to pthread_gettid_np, which at least makes the TOCTOU window smaller.) We can't check thread->tid against 0 to see whether a pthread_t is still valid because a dead thread gets its thread struct unmapped along with its stack, so the dereference isn't safe. Taking the affected functions one by one: * pthread_getcpuclockid and pthread_getschedparam/pthread_setschedparam should be fine. Unsafe calls to those seem highly unlikely. * Unsafe pthread_detach callers probably want to switch to pthread_attr_setdetachstate instead, or using pthread_detach(pthread_self()) from the new thread's start routine rather than doing the detach in the parent. * pthread_join calls should be safe anyway, because a joinable thread won't actually exit and unmap until it's joined. If you're joining an unjoinable thread, the fix is to stop marking it detached. If you're joining an already-joined thread, you need to rethink your design. * Unsafe pthread_kill calls aren't portably fixable. (And are obviously inherently non-portable as-is.) The best alternative on Android is to use pthread_gettid_np at some point that you know the thread to be alive, and then call kill/tgkill directly. That's still not completely safe because if you're too late, the tid may have been reused, but then your code is inherently unsafe anyway. Bug: http://b/19636317 Test: ran tests Change-Id: I0372c4428e8a7f1c3af5c9334f5d9c25f2c73f21
2017-02-14 02:59:29 +01:00
TEST_F(pthread_DeathTest, pthread_getcpuclockid__no_such_thread) {
pthread_t dead_thread;
MakeDeadThread(dead_thread);
clockid_t c;
EXPECT_DEATH(pthread_getcpuclockid(dead_thread, &c),
"invalid pthread_t (.*) passed to pthread_getcpuclockid");
}
TEST_F(pthread_DeathTest, pthread_getcpuclockid__null_thread) {
pthread_t null_thread = 0;
clockid_t c;
EXPECT_EQ(ESRCH, pthread_getcpuclockid(null_thread, &c));
}
Be more strict about using invalid `pthread_t`s. Another release, another attempt to remove the global thread list. But this time, let's admit that it's not going away. We can switch to using a read/write lock for the global thread list, and to aborting rather than quietly returning ESRCH if we're given an invalid pthread_t. This change affects pthread_detach, pthread_getcpuclockid, pthread_getschedparam/pthread_setschedparam, pthread_join, and pthread_kill: instead of returning ESRCH when passed an invalid pthread_t, if you're targeting O or above, they'll abort with the message "attempt to use invalid pthread_t". Note that this doesn't change behavior as much as you might think: the old lookup only held the global thread list lock for the duration of the lookup, so there was still a race between that and the dereference in the caller, given that callers actually need the tid to pass to some syscall or other, and sometimes update fields in the pthread_internal_t struct too. (This patch replaces such users with calls to pthread_gettid_np, which at least makes the TOCTOU window smaller.) We can't check thread->tid against 0 to see whether a pthread_t is still valid because a dead thread gets its thread struct unmapped along with its stack, so the dereference isn't safe. Taking the affected functions one by one: * pthread_getcpuclockid and pthread_getschedparam/pthread_setschedparam should be fine. Unsafe calls to those seem highly unlikely. * Unsafe pthread_detach callers probably want to switch to pthread_attr_setdetachstate instead, or using pthread_detach(pthread_self()) from the new thread's start routine rather than doing the detach in the parent. * pthread_join calls should be safe anyway, because a joinable thread won't actually exit and unmap until it's joined. If you're joining an unjoinable thread, the fix is to stop marking it detached. If you're joining an already-joined thread, you need to rethink your design. * Unsafe pthread_kill calls aren't portably fixable. (And are obviously inherently non-portable as-is.) The best alternative on Android is to use pthread_gettid_np at some point that you know the thread to be alive, and then call kill/tgkill directly. That's still not completely safe because if you're too late, the tid may have been reused, but then your code is inherently unsafe anyway. Bug: http://b/19636317 Test: ran tests Change-Id: I0372c4428e8a7f1c3af5c9334f5d9c25f2c73f21
2017-02-14 02:59:29 +01:00
TEST_F(pthread_DeathTest, pthread_getschedparam__no_such_thread) {
pthread_t dead_thread;
MakeDeadThread(dead_thread);
int policy;
sched_param param;
EXPECT_DEATH(pthread_getschedparam(dead_thread, &policy, &param),
"invalid pthread_t (.*) passed to pthread_getschedparam");
}
TEST_F(pthread_DeathTest, pthread_getschedparam__null_thread) {
pthread_t null_thread = 0;
int policy;
sched_param param;
EXPECT_EQ(ESRCH, pthread_getschedparam(null_thread, &policy, &param));
}
Be more strict about using invalid `pthread_t`s. Another release, another attempt to remove the global thread list. But this time, let's admit that it's not going away. We can switch to using a read/write lock for the global thread list, and to aborting rather than quietly returning ESRCH if we're given an invalid pthread_t. This change affects pthread_detach, pthread_getcpuclockid, pthread_getschedparam/pthread_setschedparam, pthread_join, and pthread_kill: instead of returning ESRCH when passed an invalid pthread_t, if you're targeting O or above, they'll abort with the message "attempt to use invalid pthread_t". Note that this doesn't change behavior as much as you might think: the old lookup only held the global thread list lock for the duration of the lookup, so there was still a race between that and the dereference in the caller, given that callers actually need the tid to pass to some syscall or other, and sometimes update fields in the pthread_internal_t struct too. (This patch replaces such users with calls to pthread_gettid_np, which at least makes the TOCTOU window smaller.) We can't check thread->tid against 0 to see whether a pthread_t is still valid because a dead thread gets its thread struct unmapped along with its stack, so the dereference isn't safe. Taking the affected functions one by one: * pthread_getcpuclockid and pthread_getschedparam/pthread_setschedparam should be fine. Unsafe calls to those seem highly unlikely. * Unsafe pthread_detach callers probably want to switch to pthread_attr_setdetachstate instead, or using pthread_detach(pthread_self()) from the new thread's start routine rather than doing the detach in the parent. * pthread_join calls should be safe anyway, because a joinable thread won't actually exit and unmap until it's joined. If you're joining an unjoinable thread, the fix is to stop marking it detached. If you're joining an already-joined thread, you need to rethink your design. * Unsafe pthread_kill calls aren't portably fixable. (And are obviously inherently non-portable as-is.) The best alternative on Android is to use pthread_gettid_np at some point that you know the thread to be alive, and then call kill/tgkill directly. That's still not completely safe because if you're too late, the tid may have been reused, but then your code is inherently unsafe anyway. Bug: http://b/19636317 Test: ran tests Change-Id: I0372c4428e8a7f1c3af5c9334f5d9c25f2c73f21
2017-02-14 02:59:29 +01:00
TEST_F(pthread_DeathTest, pthread_setschedparam__no_such_thread) {
pthread_t dead_thread;
MakeDeadThread(dead_thread);
int policy = 0;
sched_param param;
EXPECT_DEATH(pthread_setschedparam(dead_thread, policy, &param),
"invalid pthread_t (.*) passed to pthread_setschedparam");
}
TEST_F(pthread_DeathTest, pthread_setschedparam__null_thread) {
pthread_t null_thread = 0;
int policy = 0;
sched_param param;
EXPECT_EQ(ESRCH, pthread_setschedparam(null_thread, policy, &param));
}
TEST_F(pthread_DeathTest, pthread_setschedprio__no_such_thread) {
pthread_t dead_thread;
MakeDeadThread(dead_thread);
EXPECT_DEATH(pthread_setschedprio(dead_thread, 123),
"invalid pthread_t (.*) passed to pthread_setschedprio");
}
TEST_F(pthread_DeathTest, pthread_setschedprio__null_thread) {
pthread_t null_thread = 0;
EXPECT_EQ(ESRCH, pthread_setschedprio(null_thread, 123));
}
Be more strict about using invalid `pthread_t`s. Another release, another attempt to remove the global thread list. But this time, let's admit that it's not going away. We can switch to using a read/write lock for the global thread list, and to aborting rather than quietly returning ESRCH if we're given an invalid pthread_t. This change affects pthread_detach, pthread_getcpuclockid, pthread_getschedparam/pthread_setschedparam, pthread_join, and pthread_kill: instead of returning ESRCH when passed an invalid pthread_t, if you're targeting O or above, they'll abort with the message "attempt to use invalid pthread_t". Note that this doesn't change behavior as much as you might think: the old lookup only held the global thread list lock for the duration of the lookup, so there was still a race between that and the dereference in the caller, given that callers actually need the tid to pass to some syscall or other, and sometimes update fields in the pthread_internal_t struct too. (This patch replaces such users with calls to pthread_gettid_np, which at least makes the TOCTOU window smaller.) We can't check thread->tid against 0 to see whether a pthread_t is still valid because a dead thread gets its thread struct unmapped along with its stack, so the dereference isn't safe. Taking the affected functions one by one: * pthread_getcpuclockid and pthread_getschedparam/pthread_setschedparam should be fine. Unsafe calls to those seem highly unlikely. * Unsafe pthread_detach callers probably want to switch to pthread_attr_setdetachstate instead, or using pthread_detach(pthread_self()) from the new thread's start routine rather than doing the detach in the parent. * pthread_join calls should be safe anyway, because a joinable thread won't actually exit and unmap until it's joined. If you're joining an unjoinable thread, the fix is to stop marking it detached. If you're joining an already-joined thread, you need to rethink your design. * Unsafe pthread_kill calls aren't portably fixable. (And are obviously inherently non-portable as-is.) The best alternative on Android is to use pthread_gettid_np at some point that you know the thread to be alive, and then call kill/tgkill directly. That's still not completely safe because if you're too late, the tid may have been reused, but then your code is inherently unsafe anyway. Bug: http://b/19636317 Test: ran tests Change-Id: I0372c4428e8a7f1c3af5c9334f5d9c25f2c73f21
2017-02-14 02:59:29 +01:00
TEST_F(pthread_DeathTest, pthread_join__no_such_thread) {
pthread_t dead_thread;
MakeDeadThread(dead_thread);
EXPECT_DEATH(pthread_join(dead_thread, nullptr),
"invalid pthread_t (.*) passed to pthread_join");
}
TEST_F(pthread_DeathTest, pthread_join__null_thread) {
pthread_t null_thread = 0;
EXPECT_EQ(ESRCH, pthread_join(null_thread, nullptr));
}
Be more strict about using invalid `pthread_t`s. Another release, another attempt to remove the global thread list. But this time, let's admit that it's not going away. We can switch to using a read/write lock for the global thread list, and to aborting rather than quietly returning ESRCH if we're given an invalid pthread_t. This change affects pthread_detach, pthread_getcpuclockid, pthread_getschedparam/pthread_setschedparam, pthread_join, and pthread_kill: instead of returning ESRCH when passed an invalid pthread_t, if you're targeting O or above, they'll abort with the message "attempt to use invalid pthread_t". Note that this doesn't change behavior as much as you might think: the old lookup only held the global thread list lock for the duration of the lookup, so there was still a race between that and the dereference in the caller, given that callers actually need the tid to pass to some syscall or other, and sometimes update fields in the pthread_internal_t struct too. (This patch replaces such users with calls to pthread_gettid_np, which at least makes the TOCTOU window smaller.) We can't check thread->tid against 0 to see whether a pthread_t is still valid because a dead thread gets its thread struct unmapped along with its stack, so the dereference isn't safe. Taking the affected functions one by one: * pthread_getcpuclockid and pthread_getschedparam/pthread_setschedparam should be fine. Unsafe calls to those seem highly unlikely. * Unsafe pthread_detach callers probably want to switch to pthread_attr_setdetachstate instead, or using pthread_detach(pthread_self()) from the new thread's start routine rather than doing the detach in the parent. * pthread_join calls should be safe anyway, because a joinable thread won't actually exit and unmap until it's joined. If you're joining an unjoinable thread, the fix is to stop marking it detached. If you're joining an already-joined thread, you need to rethink your design. * Unsafe pthread_kill calls aren't portably fixable. (And are obviously inherently non-portable as-is.) The best alternative on Android is to use pthread_gettid_np at some point that you know the thread to be alive, and then call kill/tgkill directly. That's still not completely safe because if you're too late, the tid may have been reused, but then your code is inherently unsafe anyway. Bug: http://b/19636317 Test: ran tests Change-Id: I0372c4428e8a7f1c3af5c9334f5d9c25f2c73f21
2017-02-14 02:59:29 +01:00
TEST_F(pthread_DeathTest, pthread_kill__no_such_thread) {
pthread_t dead_thread;
MakeDeadThread(dead_thread);
EXPECT_DEATH(pthread_kill(dead_thread, 0),
"invalid pthread_t (.*) passed to pthread_kill");
}
TEST_F(pthread_DeathTest, pthread_kill__null_thread) {
pthread_t null_thread = 0;
EXPECT_EQ(ESRCH, pthread_kill(null_thread, 0));
}
TEST(pthread, pthread_join__multijoin) {
SpinFunctionHelper spin_helper;
pthread_t t1;
ASSERT_EQ(0, pthread_create(&t1, nullptr, spin_helper.GetFunction(), nullptr));
pthread_t t2;
ASSERT_EQ(0, pthread_create(&t2, nullptr, JoinFn, reinterpret_cast<void*>(t1)));
sleep(1); // (Give t2 a chance to call pthread_join.)
// Multiple joins to the same thread should fail.
ASSERT_EQ(EINVAL, pthread_join(t1, nullptr));
spin_helper.UnSpin();
// ...but t2's join on t1 still goes ahead (which we can tell because our join on t2 finishes).
void* join_result;
ASSERT_EQ(0, pthread_join(t2, &join_result));
ASSERT_EQ(0U, reinterpret_cast<uintptr_t>(join_result));
}
TEST(pthread, pthread_join__race) {
// http://b/11693195 --- pthread_join could return before the thread had actually exited.
// If the joiner unmapped the thread's stack, that could lead to SIGSEGV in the thread.
for (size_t i = 0; i < 1024; ++i) {
size_t stack_size = 640*1024;
void* stack = mmap(nullptr, stack_size, PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE, -1, 0);
pthread_attr_t a;
pthread_attr_init(&a);
pthread_attr_setstack(&a, stack, stack_size);
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, &a, IdFn, nullptr));
ASSERT_EQ(0, pthread_join(t, nullptr));
ASSERT_EQ(0, munmap(stack, stack_size));
}
}
static void* GetActualGuardSizeFn(void* arg) {
pthread_attr_t attributes;
pthread_getattr_np(pthread_self(), &attributes);
pthread_attr_getguardsize(&attributes, reinterpret_cast<size_t*>(arg));
return nullptr;
}
static size_t GetActualGuardSize(const pthread_attr_t& attributes) {
size_t result;
pthread_t t;
pthread_create(&t, &attributes, GetActualGuardSizeFn, &result);
pthread_join(t, nullptr);
return result;
}
static void* GetActualStackSizeFn(void* arg) {
pthread_attr_t attributes;
pthread_getattr_np(pthread_self(), &attributes);
pthread_attr_getstacksize(&attributes, reinterpret_cast<size_t*>(arg));
return nullptr;
}
static size_t GetActualStackSize(const pthread_attr_t& attributes) {
size_t result;
pthread_t t;
pthread_create(&t, &attributes, GetActualStackSizeFn, &result);
pthread_join(t, nullptr);
return result;
}
TEST(pthread, pthread_attr_setguardsize_tiny) {
pthread_attr_t attributes;
ASSERT_EQ(0, pthread_attr_init(&attributes));
// No such thing as too small: will be rounded up to one page by pthread_create.
ASSERT_EQ(0, pthread_attr_setguardsize(&attributes, 128));
size_t guard_size;
ASSERT_EQ(0, pthread_attr_getguardsize(&attributes, &guard_size));
ASSERT_EQ(128U, guard_size);
ASSERT_EQ(4096U, GetActualGuardSize(attributes));
}
TEST(pthread, pthread_attr_setguardsize_reasonable) {
pthread_attr_t attributes;
ASSERT_EQ(0, pthread_attr_init(&attributes));
// Large enough and a multiple of the page size.
ASSERT_EQ(0, pthread_attr_setguardsize(&attributes, 32*1024));
size_t guard_size;
ASSERT_EQ(0, pthread_attr_getguardsize(&attributes, &guard_size));
ASSERT_EQ(32*1024U, guard_size);
ASSERT_EQ(32*1024U, GetActualGuardSize(attributes));
}
TEST(pthread, pthread_attr_setguardsize_needs_rounding) {
pthread_attr_t attributes;
ASSERT_EQ(0, pthread_attr_init(&attributes));
// Large enough but not a multiple of the page size.
ASSERT_EQ(0, pthread_attr_setguardsize(&attributes, 32*1024 + 1));
size_t guard_size;
ASSERT_EQ(0, pthread_attr_getguardsize(&attributes, &guard_size));
ASSERT_EQ(32*1024U + 1, guard_size);
ASSERT_EQ(36*1024U, GetActualGuardSize(attributes));
}
TEST(pthread, pthread_attr_setguardsize_enormous) {
pthread_attr_t attributes;
ASSERT_EQ(0, pthread_attr_init(&attributes));
// Larger than the stack itself. (Historically we mistakenly carved
// the guard out of the stack itself, rather than adding it after the
// end.)
ASSERT_EQ(0, pthread_attr_setguardsize(&attributes, 32*1024*1024));
size_t guard_size;
ASSERT_EQ(0, pthread_attr_getguardsize(&attributes, &guard_size));
ASSERT_EQ(32*1024*1024U, guard_size);
ASSERT_EQ(32*1024*1024U, GetActualGuardSize(attributes));
}
TEST(pthread, pthread_attr_setstacksize) {
pthread_attr_t attributes;
ASSERT_EQ(0, pthread_attr_init(&attributes));
// Get the default stack size.
size_t default_stack_size;
ASSERT_EQ(0, pthread_attr_getstacksize(&attributes, &default_stack_size));
// Too small.
ASSERT_EQ(EINVAL, pthread_attr_setstacksize(&attributes, 128));
size_t stack_size;
ASSERT_EQ(0, pthread_attr_getstacksize(&attributes, &stack_size));
ASSERT_EQ(default_stack_size, stack_size);
ASSERT_GE(GetActualStackSize(attributes), default_stack_size);
// Large enough and a multiple of the page size; may be rounded up by pthread_create.
ASSERT_EQ(0, pthread_attr_setstacksize(&attributes, 32*1024));
ASSERT_EQ(0, pthread_attr_getstacksize(&attributes, &stack_size));
ASSERT_EQ(32*1024U, stack_size);
ASSERT_GE(GetActualStackSize(attributes), 32*1024U);
// Large enough but not aligned; will be rounded up by pthread_create.
ASSERT_EQ(0, pthread_attr_setstacksize(&attributes, 32*1024 + 1));
ASSERT_EQ(0, pthread_attr_getstacksize(&attributes, &stack_size));
ASSERT_EQ(32*1024U + 1, stack_size);
#if defined(__BIONIC__)
ASSERT_GT(GetActualStackSize(attributes), 32*1024U + 1);
#else // __BIONIC__
// glibc rounds down, in violation of POSIX. They document this in their BUGS section.
ASSERT_EQ(GetActualStackSize(attributes), 32*1024U);
#endif // __BIONIC__
}
TEST(pthread, pthread_rwlockattr_smoke) {
pthread_rwlockattr_t attr;
ASSERT_EQ(0, pthread_rwlockattr_init(&attr));
int pshared_value_array[] = {PTHREAD_PROCESS_PRIVATE, PTHREAD_PROCESS_SHARED};
for (size_t i = 0; i < sizeof(pshared_value_array) / sizeof(pshared_value_array[0]); ++i) {
ASSERT_EQ(0, pthread_rwlockattr_setpshared(&attr, pshared_value_array[i]));
int pshared;
ASSERT_EQ(0, pthread_rwlockattr_getpshared(&attr, &pshared));
ASSERT_EQ(pshared_value_array[i], pshared);
}
int kind_array[] = {PTHREAD_RWLOCK_PREFER_READER_NP,
PTHREAD_RWLOCK_PREFER_WRITER_NONRECURSIVE_NP};
for (size_t i = 0; i < sizeof(kind_array) / sizeof(kind_array[0]); ++i) {
ASSERT_EQ(0, pthread_rwlockattr_setkind_np(&attr, kind_array[i]));
int kind;
ASSERT_EQ(0, pthread_rwlockattr_getkind_np(&attr, &kind));
ASSERT_EQ(kind_array[i], kind);
}
ASSERT_EQ(0, pthread_rwlockattr_destroy(&attr));
}
TEST(pthread, pthread_rwlock_init_same_as_PTHREAD_RWLOCK_INITIALIZER) {
pthread_rwlock_t lock1 = PTHREAD_RWLOCK_INITIALIZER;
pthread_rwlock_t lock2;
ASSERT_EQ(0, pthread_rwlock_init(&lock2, nullptr));
ASSERT_EQ(0, memcmp(&lock1, &lock2, sizeof(lock1)));
}
TEST(pthread, pthread_rwlock_smoke) {
pthread_rwlock_t l;
ASSERT_EQ(0, pthread_rwlock_init(&l, nullptr));
// Single read lock
ASSERT_EQ(0, pthread_rwlock_rdlock(&l));
ASSERT_EQ(0, pthread_rwlock_unlock(&l));
// Multiple read lock
ASSERT_EQ(0, pthread_rwlock_rdlock(&l));
ASSERT_EQ(0, pthread_rwlock_rdlock(&l));
ASSERT_EQ(0, pthread_rwlock_unlock(&l));
ASSERT_EQ(0, pthread_rwlock_unlock(&l));
// Write lock
ASSERT_EQ(0, pthread_rwlock_wrlock(&l));
ASSERT_EQ(0, pthread_rwlock_unlock(&l));
// Try writer lock
ASSERT_EQ(0, pthread_rwlock_trywrlock(&l));
ASSERT_EQ(EBUSY, pthread_rwlock_trywrlock(&l));
ASSERT_EQ(EBUSY, pthread_rwlock_tryrdlock(&l));
ASSERT_EQ(0, pthread_rwlock_unlock(&l));
// Try reader lock
ASSERT_EQ(0, pthread_rwlock_tryrdlock(&l));
ASSERT_EQ(0, pthread_rwlock_tryrdlock(&l));
ASSERT_EQ(EBUSY, pthread_rwlock_trywrlock(&l));
ASSERT_EQ(0, pthread_rwlock_unlock(&l));
ASSERT_EQ(0, pthread_rwlock_unlock(&l));
// Try writer lock after unlock
ASSERT_EQ(0, pthread_rwlock_wrlock(&l));
ASSERT_EQ(0, pthread_rwlock_unlock(&l));
// EDEADLK in "read after write"
ASSERT_EQ(0, pthread_rwlock_wrlock(&l));
ASSERT_EQ(EDEADLK, pthread_rwlock_rdlock(&l));
ASSERT_EQ(0, pthread_rwlock_unlock(&l));
// EDEADLK in "write after write"
ASSERT_EQ(0, pthread_rwlock_wrlock(&l));
ASSERT_EQ(EDEADLK, pthread_rwlock_wrlock(&l));
ASSERT_EQ(0, pthread_rwlock_unlock(&l));
ASSERT_EQ(0, pthread_rwlock_destroy(&l));
}
struct RwlockWakeupHelperArg {
pthread_rwlock_t lock;
enum Progress {
LOCK_INITIALIZED,
LOCK_WAITING,
LOCK_RELEASED,
LOCK_ACCESSED,
LOCK_TIMEDOUT,
};
std::atomic<Progress> progress;
std::atomic<pid_t> tid;
std::function<int (pthread_rwlock_t*)> trylock_function;
std::function<int (pthread_rwlock_t*)> lock_function;
std::function<int (pthread_rwlock_t*, const timespec*)> timed_lock_function;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
clockid_t clock;
};
static void pthread_rwlock_wakeup_helper(RwlockWakeupHelperArg* arg) {
arg->tid = gettid();
ASSERT_EQ(RwlockWakeupHelperArg::LOCK_INITIALIZED, arg->progress);
arg->progress = RwlockWakeupHelperArg::LOCK_WAITING;
ASSERT_EQ(EBUSY, arg->trylock_function(&arg->lock));
ASSERT_EQ(0, arg->lock_function(&arg->lock));
ASSERT_EQ(RwlockWakeupHelperArg::LOCK_RELEASED, arg->progress);
ASSERT_EQ(0, pthread_rwlock_unlock(&arg->lock));
arg->progress = RwlockWakeupHelperArg::LOCK_ACCESSED;
}
static void test_pthread_rwlock_reader_wakeup_writer(std::function<int (pthread_rwlock_t*)> lock_function) {
RwlockWakeupHelperArg wakeup_arg;
ASSERT_EQ(0, pthread_rwlock_init(&wakeup_arg.lock, nullptr));
ASSERT_EQ(0, pthread_rwlock_rdlock(&wakeup_arg.lock));
wakeup_arg.progress = RwlockWakeupHelperArg::LOCK_INITIALIZED;
wakeup_arg.tid = 0;
wakeup_arg.trylock_function = &pthread_rwlock_trywrlock;
wakeup_arg.lock_function = lock_function;
pthread_t thread;
ASSERT_EQ(0, pthread_create(&thread, nullptr,
reinterpret_cast<void* (*)(void*)>(pthread_rwlock_wakeup_helper), &wakeup_arg));
WaitUntilThreadSleep(wakeup_arg.tid);
ASSERT_EQ(RwlockWakeupHelperArg::LOCK_WAITING, wakeup_arg.progress);
wakeup_arg.progress = RwlockWakeupHelperArg::LOCK_RELEASED;
ASSERT_EQ(0, pthread_rwlock_unlock(&wakeup_arg.lock));
ASSERT_EQ(0, pthread_join(thread, nullptr));
ASSERT_EQ(RwlockWakeupHelperArg::LOCK_ACCESSED, wakeup_arg.progress);
ASSERT_EQ(0, pthread_rwlock_destroy(&wakeup_arg.lock));
}
TEST(pthread, pthread_rwlock_reader_wakeup_writer) {
test_pthread_rwlock_reader_wakeup_writer(pthread_rwlock_wrlock);
}
TEST(pthread, pthread_rwlock_reader_wakeup_writer_timedwait) {
timespec ts;
ASSERT_EQ(0, clock_gettime(CLOCK_REALTIME, &ts));
ts.tv_sec += 1;
test_pthread_rwlock_reader_wakeup_writer([&](pthread_rwlock_t* lock) {
return pthread_rwlock_timedwrlock(lock, &ts);
});
}
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
TEST(pthread, pthread_rwlock_reader_wakeup_writer_timedwait_monotonic_np) {
#if defined(__BIONIC__)
timespec ts;
ASSERT_EQ(0, clock_gettime(CLOCK_MONOTONIC, &ts));
ts.tv_sec += 1;
test_pthread_rwlock_reader_wakeup_writer(
[&](pthread_rwlock_t* lock) { return pthread_rwlock_timedwrlock_monotonic_np(lock, &ts); });
#else // __BIONIC__
GTEST_SKIP() << "pthread_rwlock_timedwrlock_monotonic_np not available";
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
#endif // __BIONIC__
}
TEST(pthread, pthread_rwlock_reader_wakeup_writer_clockwait) {
#if defined(__BIONIC__)
timespec ts;
ASSERT_EQ(0, clock_gettime(CLOCK_MONOTONIC, &ts));
ts.tv_sec += 1;
test_pthread_rwlock_reader_wakeup_writer([&](pthread_rwlock_t* lock) {
return pthread_rwlock_clockwrlock(lock, CLOCK_MONOTONIC, &ts);
});
ASSERT_EQ(0, clock_gettime(CLOCK_REALTIME, &ts));
ts.tv_sec += 1;
test_pthread_rwlock_reader_wakeup_writer([&](pthread_rwlock_t* lock) {
return pthread_rwlock_clockwrlock(lock, CLOCK_REALTIME, &ts);
});
#else // __BIONIC__
GTEST_SKIP() << "pthread_rwlock_clockwrlock not available";
#endif // __BIONIC__
}
static void test_pthread_rwlock_writer_wakeup_reader(std::function<int (pthread_rwlock_t*)> lock_function) {
RwlockWakeupHelperArg wakeup_arg;
ASSERT_EQ(0, pthread_rwlock_init(&wakeup_arg.lock, nullptr));
ASSERT_EQ(0, pthread_rwlock_wrlock(&wakeup_arg.lock));
wakeup_arg.progress = RwlockWakeupHelperArg::LOCK_INITIALIZED;
wakeup_arg.tid = 0;
wakeup_arg.trylock_function = &pthread_rwlock_tryrdlock;
wakeup_arg.lock_function = lock_function;
pthread_t thread;
ASSERT_EQ(0, pthread_create(&thread, nullptr,
reinterpret_cast<void* (*)(void*)>(pthread_rwlock_wakeup_helper), &wakeup_arg));
WaitUntilThreadSleep(wakeup_arg.tid);
ASSERT_EQ(RwlockWakeupHelperArg::LOCK_WAITING, wakeup_arg.progress);
wakeup_arg.progress = RwlockWakeupHelperArg::LOCK_RELEASED;
ASSERT_EQ(0, pthread_rwlock_unlock(&wakeup_arg.lock));
ASSERT_EQ(0, pthread_join(thread, nullptr));
ASSERT_EQ(RwlockWakeupHelperArg::LOCK_ACCESSED, wakeup_arg.progress);
ASSERT_EQ(0, pthread_rwlock_destroy(&wakeup_arg.lock));
}
TEST(pthread, pthread_rwlock_writer_wakeup_reader) {
test_pthread_rwlock_writer_wakeup_reader(pthread_rwlock_rdlock);
}
TEST(pthread, pthread_rwlock_writer_wakeup_reader_timedwait) {
timespec ts;
ASSERT_EQ(0, clock_gettime(CLOCK_REALTIME, &ts));
ts.tv_sec += 1;
test_pthread_rwlock_writer_wakeup_reader([&](pthread_rwlock_t* lock) {
return pthread_rwlock_timedrdlock(lock, &ts);
});
}
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
TEST(pthread, pthread_rwlock_writer_wakeup_reader_timedwait_monotonic_np) {
#if defined(__BIONIC__)
timespec ts;
ASSERT_EQ(0, clock_gettime(CLOCK_MONOTONIC, &ts));
ts.tv_sec += 1;
test_pthread_rwlock_writer_wakeup_reader(
[&](pthread_rwlock_t* lock) { return pthread_rwlock_timedrdlock_monotonic_np(lock, &ts); });
#else // __BIONIC__
GTEST_SKIP() << "pthread_rwlock_timedrdlock_monotonic_np not available";
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
#endif // __BIONIC__
}
TEST(pthread, pthread_rwlock_writer_wakeup_reader_clockwait) {
#if defined(__BIONIC__)
timespec ts;
ASSERT_EQ(0, clock_gettime(CLOCK_MONOTONIC, &ts));
ts.tv_sec += 1;
test_pthread_rwlock_writer_wakeup_reader([&](pthread_rwlock_t* lock) {
return pthread_rwlock_clockrdlock(lock, CLOCK_MONOTONIC, &ts);
});
ASSERT_EQ(0, clock_gettime(CLOCK_REALTIME, &ts));
ts.tv_sec += 1;
test_pthread_rwlock_writer_wakeup_reader([&](pthread_rwlock_t* lock) {
return pthread_rwlock_clockrdlock(lock, CLOCK_REALTIME, &ts);
});
#else // __BIONIC__
GTEST_SKIP() << "pthread_rwlock_clockrdlock not available";
#endif // __BIONIC__
}
static void pthread_rwlock_wakeup_timeout_helper(RwlockWakeupHelperArg* arg) {
arg->tid = gettid();
ASSERT_EQ(RwlockWakeupHelperArg::LOCK_INITIALIZED, arg->progress);
arg->progress = RwlockWakeupHelperArg::LOCK_WAITING;
ASSERT_EQ(EBUSY, arg->trylock_function(&arg->lock));
timespec ts;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EQ(0, clock_gettime(arg->clock, &ts));
ASSERT_EQ(ETIMEDOUT, arg->timed_lock_function(&arg->lock, &ts));
ts.tv_nsec = -1;
ASSERT_EQ(EINVAL, arg->timed_lock_function(&arg->lock, &ts));
ts.tv_nsec = NS_PER_S;
ASSERT_EQ(EINVAL, arg->timed_lock_function(&arg->lock, &ts));
ts.tv_nsec = NS_PER_S - 1;
ts.tv_sec = -1;
ASSERT_EQ(ETIMEDOUT, arg->timed_lock_function(&arg->lock, &ts));
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EQ(0, clock_gettime(arg->clock, &ts));
ts.tv_sec += 1;
ASSERT_EQ(ETIMEDOUT, arg->timed_lock_function(&arg->lock, &ts));
ASSERT_EQ(RwlockWakeupHelperArg::LOCK_WAITING, arg->progress);
arg->progress = RwlockWakeupHelperArg::LOCK_TIMEDOUT;
}
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
static void pthread_rwlock_timedrdlock_timeout_helper(
clockid_t clock, int (*lock_function)(pthread_rwlock_t* __rwlock, const timespec* __timeout)) {
RwlockWakeupHelperArg wakeup_arg;
ASSERT_EQ(0, pthread_rwlock_init(&wakeup_arg.lock, nullptr));
ASSERT_EQ(0, pthread_rwlock_wrlock(&wakeup_arg.lock));
wakeup_arg.progress = RwlockWakeupHelperArg::LOCK_INITIALIZED;
wakeup_arg.tid = 0;
wakeup_arg.trylock_function = &pthread_rwlock_tryrdlock;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
wakeup_arg.timed_lock_function = lock_function;
wakeup_arg.clock = clock;
pthread_t thread;
ASSERT_EQ(0, pthread_create(&thread, nullptr,
reinterpret_cast<void* (*)(void*)>(pthread_rwlock_wakeup_timeout_helper), &wakeup_arg));
WaitUntilThreadSleep(wakeup_arg.tid);
ASSERT_EQ(RwlockWakeupHelperArg::LOCK_WAITING, wakeup_arg.progress);
ASSERT_EQ(0, pthread_join(thread, nullptr));
ASSERT_EQ(RwlockWakeupHelperArg::LOCK_TIMEDOUT, wakeup_arg.progress);
ASSERT_EQ(0, pthread_rwlock_unlock(&wakeup_arg.lock));
ASSERT_EQ(0, pthread_rwlock_destroy(&wakeup_arg.lock));
}
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
TEST(pthread, pthread_rwlock_timedrdlock_timeout) {
pthread_rwlock_timedrdlock_timeout_helper(CLOCK_REALTIME, pthread_rwlock_timedrdlock);
}
TEST(pthread, pthread_rwlock_timedrdlock_monotonic_np_timeout) {
#if defined(__BIONIC__)
pthread_rwlock_timedrdlock_timeout_helper(CLOCK_MONOTONIC,
pthread_rwlock_timedrdlock_monotonic_np);
#else // __BIONIC__
GTEST_SKIP() << "pthread_rwlock_timedrdlock_monotonic_np not available";
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
#endif // __BIONIC__
}
TEST(pthread, pthread_rwlock_clockrdlock_monotonic_timeout) {
#if defined(__BIONIC__)
pthread_rwlock_timedrdlock_timeout_helper(
CLOCK_MONOTONIC, [](pthread_rwlock_t* __rwlock, const timespec* __timeout) {
return pthread_rwlock_clockrdlock(__rwlock, CLOCK_MONOTONIC, __timeout);
});
#else // __BIONIC__
GTEST_SKIP() << "pthread_rwlock_clockrdlock not available";
#endif // __BIONIC__
}
TEST(pthread, pthread_rwlock_clockrdlock_realtime_timeout) {
#if defined(__BIONIC__)
pthread_rwlock_timedrdlock_timeout_helper(
CLOCK_REALTIME, [](pthread_rwlock_t* __rwlock, const timespec* __timeout) {
return pthread_rwlock_clockrdlock(__rwlock, CLOCK_REALTIME, __timeout);
});
#else // __BIONIC__
GTEST_SKIP() << "pthread_rwlock_clockrdlock not available";
#endif // __BIONIC__
}
TEST(pthread, pthread_rwlock_clockrdlock_invalid) {
#if defined(__BIONIC__)
pthread_rwlock_t lock = PTHREAD_RWLOCK_INITIALIZER;
timespec ts;
EXPECT_EQ(EINVAL, pthread_rwlock_clockrdlock(&lock, CLOCK_PROCESS_CPUTIME_ID, &ts));
#else // __BIONIC__
GTEST_SKIP() << "pthread_rwlock_clockrdlock not available";
#endif // __BIONIC__
}
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
static void pthread_rwlock_timedwrlock_timeout_helper(
clockid_t clock, int (*lock_function)(pthread_rwlock_t* __rwlock, const timespec* __timeout)) {
RwlockWakeupHelperArg wakeup_arg;
ASSERT_EQ(0, pthread_rwlock_init(&wakeup_arg.lock, nullptr));
ASSERT_EQ(0, pthread_rwlock_rdlock(&wakeup_arg.lock));
wakeup_arg.progress = RwlockWakeupHelperArg::LOCK_INITIALIZED;
wakeup_arg.tid = 0;
wakeup_arg.trylock_function = &pthread_rwlock_trywrlock;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
wakeup_arg.timed_lock_function = lock_function;
wakeup_arg.clock = clock;
pthread_t thread;
ASSERT_EQ(0, pthread_create(&thread, nullptr,
reinterpret_cast<void* (*)(void*)>(pthread_rwlock_wakeup_timeout_helper), &wakeup_arg));
WaitUntilThreadSleep(wakeup_arg.tid);
ASSERT_EQ(RwlockWakeupHelperArg::LOCK_WAITING, wakeup_arg.progress);
ASSERT_EQ(0, pthread_join(thread, nullptr));
ASSERT_EQ(RwlockWakeupHelperArg::LOCK_TIMEDOUT, wakeup_arg.progress);
ASSERT_EQ(0, pthread_rwlock_unlock(&wakeup_arg.lock));
ASSERT_EQ(0, pthread_rwlock_destroy(&wakeup_arg.lock));
}
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
TEST(pthread, pthread_rwlock_timedwrlock_timeout) {
pthread_rwlock_timedwrlock_timeout_helper(CLOCK_REALTIME, pthread_rwlock_timedwrlock);
}
TEST(pthread, pthread_rwlock_timedwrlock_monotonic_np_timeout) {
#if defined(__BIONIC__)
pthread_rwlock_timedwrlock_timeout_helper(CLOCK_MONOTONIC,
pthread_rwlock_timedwrlock_monotonic_np);
#else // __BIONIC__
GTEST_SKIP() << "pthread_rwlock_timedwrlock_monotonic_np not available";
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
#endif // __BIONIC__
}
TEST(pthread, pthread_rwlock_clockwrlock_monotonic_timeout) {
#if defined(__BIONIC__)
pthread_rwlock_timedwrlock_timeout_helper(
CLOCK_MONOTONIC, [](pthread_rwlock_t* __rwlock, const timespec* __timeout) {
return pthread_rwlock_clockwrlock(__rwlock, CLOCK_MONOTONIC, __timeout);
});
#else // __BIONIC__
GTEST_SKIP() << "pthread_rwlock_clockwrlock not available";
#endif // __BIONIC__
}
TEST(pthread, pthread_rwlock_clockwrlock_realtime_timeout) {
#if defined(__BIONIC__)
pthread_rwlock_timedwrlock_timeout_helper(
CLOCK_REALTIME, [](pthread_rwlock_t* __rwlock, const timespec* __timeout) {
return pthread_rwlock_clockwrlock(__rwlock, CLOCK_REALTIME, __timeout);
});
#else // __BIONIC__
GTEST_SKIP() << "pthread_rwlock_clockwrlock not available";
#endif // __BIONIC__
}
TEST(pthread, pthread_rwlock_clockwrlock_invalid) {
#if defined(__BIONIC__)
pthread_rwlock_t lock = PTHREAD_RWLOCK_INITIALIZER;
timespec ts;
EXPECT_EQ(EINVAL, pthread_rwlock_clockwrlock(&lock, CLOCK_PROCESS_CPUTIME_ID, &ts));
#else // __BIONIC__
GTEST_SKIP() << "pthread_rwlock_clockrwlock not available";
#endif // __BIONIC__
}
class RwlockKindTestHelper {
private:
struct ThreadArg {
RwlockKindTestHelper* helper;
std::atomic<pid_t>& tid;
ThreadArg(RwlockKindTestHelper* helper, std::atomic<pid_t>& tid)
: helper(helper), tid(tid) { }
};
public:
pthread_rwlock_t lock;
public:
explicit RwlockKindTestHelper(int kind_type) {
InitRwlock(kind_type);
}
~RwlockKindTestHelper() {
DestroyRwlock();
}
void CreateWriterThread(pthread_t& thread, std::atomic<pid_t>& tid) {
tid = 0;
ThreadArg* arg = new ThreadArg(this, tid);
ASSERT_EQ(0, pthread_create(&thread, nullptr,
reinterpret_cast<void* (*)(void*)>(WriterThreadFn), arg));
}
void CreateReaderThread(pthread_t& thread, std::atomic<pid_t>& tid) {
tid = 0;
ThreadArg* arg = new ThreadArg(this, tid);
ASSERT_EQ(0, pthread_create(&thread, nullptr,
reinterpret_cast<void* (*)(void*)>(ReaderThreadFn), arg));
}
private:
void InitRwlock(int kind_type) {
pthread_rwlockattr_t attr;
ASSERT_EQ(0, pthread_rwlockattr_init(&attr));
ASSERT_EQ(0, pthread_rwlockattr_setkind_np(&attr, kind_type));
ASSERT_EQ(0, pthread_rwlock_init(&lock, &attr));
ASSERT_EQ(0, pthread_rwlockattr_destroy(&attr));
}
void DestroyRwlock() {
ASSERT_EQ(0, pthread_rwlock_destroy(&lock));
}
static void WriterThreadFn(ThreadArg* arg) {
arg->tid = gettid();
RwlockKindTestHelper* helper = arg->helper;
ASSERT_EQ(0, pthread_rwlock_wrlock(&helper->lock));
ASSERT_EQ(0, pthread_rwlock_unlock(&helper->lock));
delete arg;
}
static void ReaderThreadFn(ThreadArg* arg) {
arg->tid = gettid();
RwlockKindTestHelper* helper = arg->helper;
ASSERT_EQ(0, pthread_rwlock_rdlock(&helper->lock));
ASSERT_EQ(0, pthread_rwlock_unlock(&helper->lock));
delete arg;
}
};
TEST(pthread, pthread_rwlock_kind_PTHREAD_RWLOCK_PREFER_READER_NP) {
RwlockKindTestHelper helper(PTHREAD_RWLOCK_PREFER_READER_NP);
ASSERT_EQ(0, pthread_rwlock_rdlock(&helper.lock));
pthread_t writer_thread;
std::atomic<pid_t> writer_tid;
helper.CreateWriterThread(writer_thread, writer_tid);
WaitUntilThreadSleep(writer_tid);
pthread_t reader_thread;
std::atomic<pid_t> reader_tid;
helper.CreateReaderThread(reader_thread, reader_tid);
ASSERT_EQ(0, pthread_join(reader_thread, nullptr));
ASSERT_EQ(0, pthread_rwlock_unlock(&helper.lock));
ASSERT_EQ(0, pthread_join(writer_thread, nullptr));
}
TEST(pthread, pthread_rwlock_kind_PTHREAD_RWLOCK_PREFER_WRITER_NONRECURSIVE_NP) {
RwlockKindTestHelper helper(PTHREAD_RWLOCK_PREFER_WRITER_NONRECURSIVE_NP);
ASSERT_EQ(0, pthread_rwlock_rdlock(&helper.lock));
pthread_t writer_thread;
std::atomic<pid_t> writer_tid;
helper.CreateWriterThread(writer_thread, writer_tid);
WaitUntilThreadSleep(writer_tid);
pthread_t reader_thread;
std::atomic<pid_t> reader_tid;
helper.CreateReaderThread(reader_thread, reader_tid);
WaitUntilThreadSleep(reader_tid);
ASSERT_EQ(0, pthread_rwlock_unlock(&helper.lock));
ASSERT_EQ(0, pthread_join(writer_thread, nullptr));
ASSERT_EQ(0, pthread_join(reader_thread, nullptr));
}
static int g_once_fn_call_count = 0;
static void OnceFn() {
++g_once_fn_call_count;
}
TEST(pthread, pthread_once_smoke) {
pthread_once_t once_control = PTHREAD_ONCE_INIT;
ASSERT_EQ(0, pthread_once(&once_control, OnceFn));
ASSERT_EQ(0, pthread_once(&once_control, OnceFn));
ASSERT_EQ(1, g_once_fn_call_count);
}
static std::string pthread_once_1934122_result = "";
static void Routine2() {
pthread_once_1934122_result += "2";
}
static void Routine1() {
pthread_once_t once_control_2 = PTHREAD_ONCE_INIT;
pthread_once_1934122_result += "1";
pthread_once(&once_control_2, &Routine2);
}
TEST(pthread, pthread_once_1934122) {
// Very old versions of Android couldn't call pthread_once from a
// pthread_once init routine. http://b/1934122.
pthread_once_t once_control_1 = PTHREAD_ONCE_INIT;
ASSERT_EQ(0, pthread_once(&once_control_1, &Routine1));
ASSERT_EQ("12", pthread_once_1934122_result);
}
static int g_atfork_prepare_calls = 0;
static void AtForkPrepare1() { g_atfork_prepare_calls = (g_atfork_prepare_calls * 10) + 1; }
static void AtForkPrepare2() { g_atfork_prepare_calls = (g_atfork_prepare_calls * 10) + 2; }
static int g_atfork_parent_calls = 0;
static void AtForkParent1() { g_atfork_parent_calls = (g_atfork_parent_calls * 10) + 1; }
static void AtForkParent2() { g_atfork_parent_calls = (g_atfork_parent_calls * 10) + 2; }
static int g_atfork_child_calls = 0;
static void AtForkChild1() { g_atfork_child_calls = (g_atfork_child_calls * 10) + 1; }
static void AtForkChild2() { g_atfork_child_calls = (g_atfork_child_calls * 10) + 2; }
TEST(pthread, pthread_atfork_smoke) {
ASSERT_EQ(0, pthread_atfork(AtForkPrepare1, AtForkParent1, AtForkChild1));
ASSERT_EQ(0, pthread_atfork(AtForkPrepare2, AtForkParent2, AtForkChild2));
pid_t pid = fork();
ASSERT_NE(-1, pid) << strerror(errno);
// Child and parent calls are made in the order they were registered.
if (pid == 0) {
ASSERT_EQ(12, g_atfork_child_calls);
_exit(0);
}
ASSERT_EQ(12, g_atfork_parent_calls);
// Prepare calls are made in the reverse order.
ASSERT_EQ(21, g_atfork_prepare_calls);
AssertChildExited(pid, 0);
}
TEST(pthread, pthread_attr_getscope) {
pthread_attr_t attr;
ASSERT_EQ(0, pthread_attr_init(&attr));
int scope;
ASSERT_EQ(0, pthread_attr_getscope(&attr, &scope));
ASSERT_EQ(PTHREAD_SCOPE_SYSTEM, scope);
}
TEST(pthread, pthread_condattr_init) {
pthread_condattr_t attr;
pthread_condattr_init(&attr);
clockid_t clock;
ASSERT_EQ(0, pthread_condattr_getclock(&attr, &clock));
ASSERT_EQ(CLOCK_REALTIME, clock);
int pshared;
ASSERT_EQ(0, pthread_condattr_getpshared(&attr, &pshared));
ASSERT_EQ(PTHREAD_PROCESS_PRIVATE, pshared);
}
TEST(pthread, pthread_condattr_setclock) {
pthread_condattr_t attr;
pthread_condattr_init(&attr);
ASSERT_EQ(0, pthread_condattr_setclock(&attr, CLOCK_REALTIME));
clockid_t clock;
ASSERT_EQ(0, pthread_condattr_getclock(&attr, &clock));
ASSERT_EQ(CLOCK_REALTIME, clock);
ASSERT_EQ(0, pthread_condattr_setclock(&attr, CLOCK_MONOTONIC));
ASSERT_EQ(0, pthread_condattr_getclock(&attr, &clock));
ASSERT_EQ(CLOCK_MONOTONIC, clock);
ASSERT_EQ(EINVAL, pthread_condattr_setclock(&attr, CLOCK_PROCESS_CPUTIME_ID));
}
TEST(pthread, pthread_cond_broadcast__preserves_condattr_flags) {
#if defined(__BIONIC__)
pthread_condattr_t attr;
pthread_condattr_init(&attr);
ASSERT_EQ(0, pthread_condattr_setclock(&attr, CLOCK_MONOTONIC));
ASSERT_EQ(0, pthread_condattr_setpshared(&attr, PTHREAD_PROCESS_SHARED));
pthread_cond_t cond_var;
ASSERT_EQ(0, pthread_cond_init(&cond_var, &attr));
ASSERT_EQ(0, pthread_cond_signal(&cond_var));
ASSERT_EQ(0, pthread_cond_broadcast(&cond_var));
attr = static_cast<pthread_condattr_t>(*reinterpret_cast<uint32_t*>(cond_var.__private));
clockid_t clock;
ASSERT_EQ(0, pthread_condattr_getclock(&attr, &clock));
ASSERT_EQ(CLOCK_MONOTONIC, clock);
int pshared;
ASSERT_EQ(0, pthread_condattr_getpshared(&attr, &pshared));
ASSERT_EQ(PTHREAD_PROCESS_SHARED, pshared);
#else // !defined(__BIONIC__)
GTEST_SKIP() << "bionic-only test";
#endif // !defined(__BIONIC__)
}
class pthread_CondWakeupTest : public ::testing::Test {
protected:
pthread_mutex_t mutex;
pthread_cond_t cond;
enum Progress {
INITIALIZED,
WAITING,
SIGNALED,
FINISHED,
};
std::atomic<Progress> progress;
pthread_t thread;
std::function<int (pthread_cond_t* cond, pthread_mutex_t* mutex)> wait_function;
protected:
void SetUp() override {
ASSERT_EQ(0, pthread_mutex_init(&mutex, nullptr));
}
void InitCond(clockid_t clock=CLOCK_REALTIME) {
pthread_condattr_t attr;
ASSERT_EQ(0, pthread_condattr_init(&attr));
ASSERT_EQ(0, pthread_condattr_setclock(&attr, clock));
ASSERT_EQ(0, pthread_cond_init(&cond, &attr));
ASSERT_EQ(0, pthread_condattr_destroy(&attr));
}
void StartWaitingThread(
std::function<int(pthread_cond_t* cond, pthread_mutex_t* mutex)> wait_function) {
progress = INITIALIZED;
this->wait_function = wait_function;
ASSERT_EQ(0, pthread_create(&thread, nullptr, reinterpret_cast<void* (*)(void*)>(WaitThreadFn),
this));
while (progress != WAITING) {
usleep(5000);
}
usleep(5000);
}
void RunTimedTest(
clockid_t clock,
std::function<int(pthread_cond_t* cond, pthread_mutex_t* mutex, const timespec* timeout)>
wait_function) {
timespec ts;
ASSERT_EQ(0, clock_gettime(clock, &ts));
ts.tv_sec += 1;
StartWaitingThread([&wait_function, &ts](pthread_cond_t* cond, pthread_mutex_t* mutex) {
return wait_function(cond, mutex, &ts);
});
progress = SIGNALED;
ASSERT_EQ(0, pthread_cond_signal(&cond));
}
void RunTimedTest(clockid_t clock, std::function<int(pthread_cond_t* cond, pthread_mutex_t* mutex,
clockid_t clock, const timespec* timeout)>
wait_function) {
RunTimedTest(clock, [clock, &wait_function](pthread_cond_t* cond, pthread_mutex_t* mutex,
const timespec* timeout) {
return wait_function(cond, mutex, clock, timeout);
});
}
void TearDown() override {
ASSERT_EQ(0, pthread_join(thread, nullptr));
ASSERT_EQ(FINISHED, progress);
ASSERT_EQ(0, pthread_cond_destroy(&cond));
ASSERT_EQ(0, pthread_mutex_destroy(&mutex));
}
private:
static void WaitThreadFn(pthread_CondWakeupTest* test) {
ASSERT_EQ(0, pthread_mutex_lock(&test->mutex));
test->progress = WAITING;
while (test->progress == WAITING) {
ASSERT_EQ(0, test->wait_function(&test->cond, &test->mutex));
}
ASSERT_EQ(SIGNALED, test->progress);
test->progress = FINISHED;
ASSERT_EQ(0, pthread_mutex_unlock(&test->mutex));
}
};
TEST_F(pthread_CondWakeupTest, signal_wait) {
InitCond();
StartWaitingThread([](pthread_cond_t* cond, pthread_mutex_t* mutex) {
return pthread_cond_wait(cond, mutex);
});
progress = SIGNALED;
ASSERT_EQ(0, pthread_cond_signal(&cond));
}
TEST_F(pthread_CondWakeupTest, broadcast_wait) {
InitCond();
StartWaitingThread([](pthread_cond_t* cond, pthread_mutex_t* mutex) {
return pthread_cond_wait(cond, mutex);
});
progress = SIGNALED;
ASSERT_EQ(0, pthread_cond_broadcast(&cond));
}
TEST_F(pthread_CondWakeupTest, signal_timedwait_CLOCK_REALTIME) {
InitCond(CLOCK_REALTIME);
RunTimedTest(CLOCK_REALTIME, pthread_cond_timedwait);
}
TEST_F(pthread_CondWakeupTest, signal_timedwait_CLOCK_MONOTONIC) {
InitCond(CLOCK_MONOTONIC);
RunTimedTest(CLOCK_MONOTONIC, pthread_cond_timedwait);
}
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
TEST_F(pthread_CondWakeupTest, signal_timedwait_CLOCK_MONOTONIC_np) {
#if defined(__BIONIC__)
InitCond(CLOCK_REALTIME);
RunTimedTest(CLOCK_MONOTONIC, pthread_cond_timedwait_monotonic_np);
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
#else // __BIONIC__
GTEST_SKIP() << "pthread_cond_timedwait_monotonic_np not available";
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
#endif // __BIONIC__
}
TEST_F(pthread_CondWakeupTest, signal_clockwait_monotonic_monotonic) {
#if defined(__BIONIC__)
InitCond(CLOCK_MONOTONIC);
RunTimedTest(CLOCK_MONOTONIC, pthread_cond_clockwait);
#else // __BIONIC__
GTEST_SKIP() << "pthread_cond_clockwait not available";
#endif // __BIONIC__
}
TEST_F(pthread_CondWakeupTest, signal_clockwait_monotonic_realtime) {
#if defined(__BIONIC__)
InitCond(CLOCK_MONOTONIC);
RunTimedTest(CLOCK_REALTIME, pthread_cond_clockwait);
#else // __BIONIC__
GTEST_SKIP() << "pthread_cond_clockwait not available";
#endif // __BIONIC__
}
TEST_F(pthread_CondWakeupTest, signal_clockwait_realtime_monotonic) {
#if defined(__BIONIC__)
InitCond(CLOCK_REALTIME);
RunTimedTest(CLOCK_MONOTONIC, pthread_cond_clockwait);
#else // __BIONIC__
GTEST_SKIP() << "pthread_cond_clockwait not available";
#endif // __BIONIC__
}
TEST_F(pthread_CondWakeupTest, signal_clockwait_realtime_realtime) {
#if defined(__BIONIC__)
InitCond(CLOCK_REALTIME);
RunTimedTest(CLOCK_REALTIME, pthread_cond_clockwait);
#else // __BIONIC__
GTEST_SKIP() << "pthread_cond_clockwait not available";
#endif // __BIONIC__
}
static void pthread_cond_timedwait_timeout_helper(bool init_monotonic, clockid_t clock,
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
int (*wait_function)(pthread_cond_t* __cond,
pthread_mutex_t* __mutex,
const timespec* __timeout)) {
pthread_mutex_t mutex;
ASSERT_EQ(0, pthread_mutex_init(&mutex, nullptr));
pthread_cond_t cond;
if (init_monotonic) {
pthread_condattr_t attr;
pthread_condattr_init(&attr);
ASSERT_EQ(0, pthread_condattr_setclock(&attr, CLOCK_MONOTONIC));
clockid_t clock;
ASSERT_EQ(0, pthread_condattr_getclock(&attr, &clock));
ASSERT_EQ(CLOCK_MONOTONIC, clock);
ASSERT_EQ(0, pthread_cond_init(&cond, &attr));
} else {
ASSERT_EQ(0, pthread_cond_init(&cond, nullptr));
}
ASSERT_EQ(0, pthread_mutex_lock(&mutex));
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
timespec ts;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EQ(0, clock_gettime(clock, &ts));
ASSERT_EQ(ETIMEDOUT, wait_function(&cond, &mutex, &ts));
ts.tv_nsec = -1;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EQ(EINVAL, wait_function(&cond, &mutex, &ts));
ts.tv_nsec = NS_PER_S;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EQ(EINVAL, wait_function(&cond, &mutex, &ts));
ts.tv_nsec = NS_PER_S - 1;
ts.tv_sec = -1;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EQ(ETIMEDOUT, wait_function(&cond, &mutex, &ts));
ASSERT_EQ(0, pthread_mutex_unlock(&mutex));
}
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
TEST(pthread, pthread_cond_timedwait_timeout) {
pthread_cond_timedwait_timeout_helper(false, CLOCK_REALTIME, pthread_cond_timedwait);
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
}
TEST(pthread, pthread_cond_timedwait_monotonic_np_timeout) {
#if defined(__BIONIC__)
pthread_cond_timedwait_timeout_helper(false, CLOCK_MONOTONIC, pthread_cond_timedwait_monotonic_np);
pthread_cond_timedwait_timeout_helper(true, CLOCK_MONOTONIC, pthread_cond_timedwait_monotonic_np);
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
#else // __BIONIC__
GTEST_SKIP() << "pthread_cond_timedwait_monotonic_np not available";
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
#endif // __BIONIC__
}
TEST(pthread, pthread_cond_clockwait_timeout) {
#if defined(__BIONIC__)
pthread_cond_timedwait_timeout_helper(
false, CLOCK_MONOTONIC,
[](pthread_cond_t* __cond, pthread_mutex_t* __mutex, const timespec* __timeout) {
return pthread_cond_clockwait(__cond, __mutex, CLOCK_MONOTONIC, __timeout);
});
pthread_cond_timedwait_timeout_helper(
true, CLOCK_MONOTONIC,
[](pthread_cond_t* __cond, pthread_mutex_t* __mutex, const timespec* __timeout) {
return pthread_cond_clockwait(__cond, __mutex, CLOCK_MONOTONIC, __timeout);
});
pthread_cond_timedwait_timeout_helper(
false, CLOCK_REALTIME,
[](pthread_cond_t* __cond, pthread_mutex_t* __mutex, const timespec* __timeout) {
return pthread_cond_clockwait(__cond, __mutex, CLOCK_REALTIME, __timeout);
});
pthread_cond_timedwait_timeout_helper(
true, CLOCK_REALTIME,
[](pthread_cond_t* __cond, pthread_mutex_t* __mutex, const timespec* __timeout) {
return pthread_cond_clockwait(__cond, __mutex, CLOCK_REALTIME, __timeout);
});
#else // __BIONIC__
GTEST_SKIP() << "pthread_cond_clockwait not available";
#endif // __BIONIC__
}
TEST(pthread, pthread_cond_clockwait_invalid) {
#if defined(__BIONIC__)
pthread_cond_t cond = PTHREAD_COND_INITIALIZER;
pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
timespec ts;
EXPECT_EQ(EINVAL, pthread_cond_clockwait(&cond, &mutex, CLOCK_PROCESS_CPUTIME_ID, &ts));
#else // __BIONIC__
GTEST_SKIP() << "pthread_cond_clockwait not available";
#endif // __BIONIC__
}
TEST(pthread, pthread_attr_getstack__main_thread) {
// This test is only meaningful for the main thread, so make sure we're running on it!
ASSERT_EQ(getpid(), syscall(__NR_gettid));
// Get the main thread's attributes.
pthread_attr_t attributes;
ASSERT_EQ(0, pthread_getattr_np(pthread_self(), &attributes));
// Check that we correctly report that the main thread has no guard page.
size_t guard_size;
ASSERT_EQ(0, pthread_attr_getguardsize(&attributes, &guard_size));
ASSERT_EQ(0U, guard_size); // The main thread has no guard page.
// Get the stack base and the stack size (both ways).
void* stack_base;
size_t stack_size;
ASSERT_EQ(0, pthread_attr_getstack(&attributes, &stack_base, &stack_size));
size_t stack_size2;
ASSERT_EQ(0, pthread_attr_getstacksize(&attributes, &stack_size2));
// The two methods of asking for the stack size should agree.
EXPECT_EQ(stack_size, stack_size2);
#if defined(__BIONIC__)
// Find stack in /proc/self/maps using a pointer to the stack.
//
// We do not use "[stack]" label because in native-bridge environment it is not
// guaranteed to point to the right stack. A native bridge implementation may
// keep separate stack for the guest code.
void* maps_stack_hi = nullptr;
std::vector<map_record> maps;
ASSERT_TRUE(Maps::parse_maps(&maps));
uintptr_t stack_address = reinterpret_cast<uintptr_t>(untag_address(&maps_stack_hi));
for (const auto& map : maps) {
if (map.addr_start <= stack_address && map.addr_end > stack_address){
maps_stack_hi = reinterpret_cast<void*>(map.addr_end);
break;
}
}
// The high address of the /proc/self/maps stack region should equal stack_base + stack_size.
// Remember that the stack grows down (and is mapped in on demand), so the low address of the
// region isn't very interesting.
EXPECT_EQ(maps_stack_hi, reinterpret_cast<uint8_t*>(stack_base) + stack_size);
// The stack size should correspond to RLIMIT_STACK.
rlimit rl;
ASSERT_EQ(0, getrlimit(RLIMIT_STACK, &rl));
uint64_t original_rlim_cur = rl.rlim_cur;
if (rl.rlim_cur == RLIM_INFINITY) {
rl.rlim_cur = 8 * 1024 * 1024; // Bionic reports unlimited stacks as 8MiB.
}
EXPECT_EQ(rl.rlim_cur, stack_size);
auto guard = android::base::make_scope_guard([&rl, original_rlim_cur]() {
rl.rlim_cur = original_rlim_cur;
ASSERT_EQ(0, setrlimit(RLIMIT_STACK, &rl));
});
//
// What if RLIMIT_STACK is smaller than the stack's current extent?
//
rl.rlim_cur = rl.rlim_max = 1024; // 1KiB. We know the stack must be at least a page already.
rl.rlim_max = RLIM_INFINITY;
ASSERT_EQ(0, setrlimit(RLIMIT_STACK, &rl));
ASSERT_EQ(0, pthread_getattr_np(pthread_self(), &attributes));
ASSERT_EQ(0, pthread_attr_getstack(&attributes, &stack_base, &stack_size));
ASSERT_EQ(0, pthread_attr_getstacksize(&attributes, &stack_size2));
EXPECT_EQ(stack_size, stack_size2);
ASSERT_EQ(1024U, stack_size);
//
// What if RLIMIT_STACK isn't a whole number of pages?
//
rl.rlim_cur = rl.rlim_max = 6666; // Not a whole number of pages.
rl.rlim_max = RLIM_INFINITY;
ASSERT_EQ(0, setrlimit(RLIMIT_STACK, &rl));
ASSERT_EQ(0, pthread_getattr_np(pthread_self(), &attributes));
ASSERT_EQ(0, pthread_attr_getstack(&attributes, &stack_base, &stack_size));
ASSERT_EQ(0, pthread_attr_getstacksize(&attributes, &stack_size2));
EXPECT_EQ(stack_size, stack_size2);
ASSERT_EQ(6666U, stack_size);
#endif
}
struct GetStackSignalHandlerArg {
volatile bool done;
void* signal_stack_base;
size_t signal_stack_size;
void* main_stack_base;
size_t main_stack_size;
};
static GetStackSignalHandlerArg getstack_signal_handler_arg;
static void getstack_signal_handler(int sig) {
ASSERT_EQ(SIGUSR1, sig);
// Use sleep() to make current thread be switched out by the kernel to provoke the error.
sleep(1);
pthread_attr_t attr;
ASSERT_EQ(0, pthread_getattr_np(pthread_self(), &attr));
void* stack_base;
size_t stack_size;
ASSERT_EQ(0, pthread_attr_getstack(&attr, &stack_base, &stack_size));
// Verify if the stack used by the signal handler is the alternate stack just registered.
ASSERT_LE(getstack_signal_handler_arg.signal_stack_base, &attr);
ASSERT_LT(static_cast<void*>(untag_address(&attr)),
static_cast<char*>(getstack_signal_handler_arg.signal_stack_base) +
getstack_signal_handler_arg.signal_stack_size);
// Verify if the main thread's stack got in the signal handler is correct.
ASSERT_EQ(getstack_signal_handler_arg.main_stack_base, stack_base);
ASSERT_LE(getstack_signal_handler_arg.main_stack_size, stack_size);
getstack_signal_handler_arg.done = true;
}
// The previous code obtained the main thread's stack by reading the entry in
// /proc/self/task/<pid>/maps that was labeled [stack]. Unfortunately, on x86/x86_64, the kernel
// relies on sp0 in task state segment(tss) to label the stack map with [stack]. If the kernel
// switches a process while the main thread is in an alternate stack, then the kernel will label
// the wrong map with [stack]. This test verifies that when the above situation happens, the main
// thread's stack is found correctly.
TEST(pthread, pthread_attr_getstack_in_signal_handler) {
// This test is only meaningful for the main thread, so make sure we're running on it!
ASSERT_EQ(getpid(), syscall(__NR_gettid));
const size_t sig_stack_size = 16 * 1024;
void* sig_stack = mmap(nullptr, sig_stack_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS,
-1, 0);
ASSERT_NE(MAP_FAILED, sig_stack);
stack_t ss;
ss.ss_sp = sig_stack;
ss.ss_size = sig_stack_size;
ss.ss_flags = 0;
stack_t oss;
ASSERT_EQ(0, sigaltstack(&ss, &oss));
pthread_attr_t attr;
ASSERT_EQ(0, pthread_getattr_np(pthread_self(), &attr));
void* main_stack_base;
size_t main_stack_size;
ASSERT_EQ(0, pthread_attr_getstack(&attr, &main_stack_base, &main_stack_size));
ScopedSignalHandler handler(SIGUSR1, getstack_signal_handler, SA_ONSTACK);
getstack_signal_handler_arg.done = false;
getstack_signal_handler_arg.signal_stack_base = sig_stack;
getstack_signal_handler_arg.signal_stack_size = sig_stack_size;
getstack_signal_handler_arg.main_stack_base = main_stack_base;
getstack_signal_handler_arg.main_stack_size = main_stack_size;
kill(getpid(), SIGUSR1);
ASSERT_EQ(true, getstack_signal_handler_arg.done);
ASSERT_EQ(0, sigaltstack(&oss, nullptr));
ASSERT_EQ(0, munmap(sig_stack, sig_stack_size));
}
static void pthread_attr_getstack_18908062_helper(void*) {
char local_variable;
pthread_attr_t attributes;
pthread_getattr_np(pthread_self(), &attributes);
void* stack_base;
size_t stack_size;
pthread_attr_getstack(&attributes, &stack_base, &stack_size);
// Test whether &local_variable is in [stack_base, stack_base + stack_size).
ASSERT_LE(reinterpret_cast<char*>(stack_base), &local_variable);
ASSERT_LT(untag_address(&local_variable), reinterpret_cast<char*>(stack_base) + stack_size);
}
// Check whether something on stack is in the range of
// [stack_base, stack_base + stack_size). see b/18908062.
TEST(pthread, pthread_attr_getstack_18908062) {
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, nullptr,
reinterpret_cast<void* (*)(void*)>(pthread_attr_getstack_18908062_helper),
nullptr));
ASSERT_EQ(0, pthread_join(t, nullptr));
}
#if defined(__BIONIC__)
static pthread_mutex_t pthread_gettid_np_mutex = PTHREAD_MUTEX_INITIALIZER;
static void* pthread_gettid_np_helper(void* arg) {
*reinterpret_cast<pid_t*>(arg) = gettid();
// Wait for our parent to call pthread_gettid_np on us before exiting.
pthread_mutex_lock(&pthread_gettid_np_mutex);
pthread_mutex_unlock(&pthread_gettid_np_mutex);
return nullptr;
}
#endif
TEST(pthread, pthread_gettid_np) {
#if defined(__BIONIC__)
ASSERT_EQ(gettid(), pthread_gettid_np(pthread_self()));
// Ensure the other thread doesn't exit until after we've called
// pthread_gettid_np on it.
pthread_mutex_lock(&pthread_gettid_np_mutex);
pid_t t_gettid_result;
pthread_t t;
pthread_create(&t, nullptr, pthread_gettid_np_helper, &t_gettid_result);
pid_t t_pthread_gettid_np_result = pthread_gettid_np(t);
// Release the other thread and wait for it to exit.
pthread_mutex_unlock(&pthread_gettid_np_mutex);
ASSERT_EQ(0, pthread_join(t, nullptr));
ASSERT_EQ(t_gettid_result, t_pthread_gettid_np_result);
#else
GTEST_SKIP() << "pthread_gettid_np not available";
#endif
}
static size_t cleanup_counter = 0;
static void AbortCleanupRoutine(void*) {
abort();
}
static void CountCleanupRoutine(void*) {
++cleanup_counter;
}
static void PthreadCleanupTester() {
pthread_cleanup_push(CountCleanupRoutine, nullptr);
pthread_cleanup_push(CountCleanupRoutine, nullptr);
pthread_cleanup_push(AbortCleanupRoutine, nullptr);
pthread_cleanup_pop(0); // Pop the abort without executing it.
pthread_cleanup_pop(1); // Pop one count while executing it.
ASSERT_EQ(1U, cleanup_counter);
// Exit while the other count is still on the cleanup stack.
pthread_exit(nullptr);
// Calls to pthread_cleanup_pop/pthread_cleanup_push must always be balanced.
pthread_cleanup_pop(0);
}
static void* PthreadCleanupStartRoutine(void*) {
PthreadCleanupTester();
return nullptr;
}
TEST(pthread, pthread_cleanup_push__pthread_cleanup_pop) {
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, nullptr, PthreadCleanupStartRoutine, nullptr));
ASSERT_EQ(0, pthread_join(t, nullptr));
ASSERT_EQ(2U, cleanup_counter);
}
TEST(pthread, PTHREAD_MUTEX_DEFAULT_is_PTHREAD_MUTEX_NORMAL) {
ASSERT_EQ(PTHREAD_MUTEX_NORMAL, PTHREAD_MUTEX_DEFAULT);
}
TEST(pthread, pthread_mutexattr_gettype) {
pthread_mutexattr_t attr;
ASSERT_EQ(0, pthread_mutexattr_init(&attr));
int attr_type;
ASSERT_EQ(0, pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_NORMAL));
ASSERT_EQ(0, pthread_mutexattr_gettype(&attr, &attr_type));
ASSERT_EQ(PTHREAD_MUTEX_NORMAL, attr_type);
ASSERT_EQ(0, pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_ERRORCHECK));
ASSERT_EQ(0, pthread_mutexattr_gettype(&attr, &attr_type));
ASSERT_EQ(PTHREAD_MUTEX_ERRORCHECK, attr_type);
ASSERT_EQ(0, pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE));
ASSERT_EQ(0, pthread_mutexattr_gettype(&attr, &attr_type));
ASSERT_EQ(PTHREAD_MUTEX_RECURSIVE, attr_type);
ASSERT_EQ(0, pthread_mutexattr_destroy(&attr));
}
TEST(pthread, pthread_mutexattr_protocol) {
pthread_mutexattr_t attr;
ASSERT_EQ(0, pthread_mutexattr_init(&attr));
int protocol;
ASSERT_EQ(0, pthread_mutexattr_getprotocol(&attr, &protocol));
ASSERT_EQ(PTHREAD_PRIO_NONE, protocol);
for (size_t repeat = 0; repeat < 2; ++repeat) {
for (int set_protocol : {PTHREAD_PRIO_NONE, PTHREAD_PRIO_INHERIT}) {
ASSERT_EQ(0, pthread_mutexattr_setprotocol(&attr, set_protocol));
ASSERT_EQ(0, pthread_mutexattr_getprotocol(&attr, &protocol));
ASSERT_EQ(protocol, set_protocol);
}
}
}
struct PthreadMutex {
pthread_mutex_t lock;
explicit PthreadMutex(int mutex_type, int protocol = PTHREAD_PRIO_NONE) {
init(mutex_type, protocol);
}
~PthreadMutex() {
destroy();
}
private:
void init(int mutex_type, int protocol) {
pthread_mutexattr_t attr;
ASSERT_EQ(0, pthread_mutexattr_init(&attr));
ASSERT_EQ(0, pthread_mutexattr_settype(&attr, mutex_type));
ASSERT_EQ(0, pthread_mutexattr_setprotocol(&attr, protocol));
ASSERT_EQ(0, pthread_mutex_init(&lock, &attr));
ASSERT_EQ(0, pthread_mutexattr_destroy(&attr));
}
void destroy() {
ASSERT_EQ(0, pthread_mutex_destroy(&lock));
}
DISALLOW_COPY_AND_ASSIGN(PthreadMutex);
};
static int UnlockFromAnotherThread(pthread_mutex_t* mutex) {
pthread_t thread;
pthread_create(&thread, nullptr, [](void* mutex_voidp) -> void* {
pthread_mutex_t* mutex = static_cast<pthread_mutex_t*>(mutex_voidp);
intptr_t result = pthread_mutex_unlock(mutex);
return reinterpret_cast<void*>(result);
}, mutex);
void* result;
EXPECT_EQ(0, pthread_join(thread, &result));
return reinterpret_cast<intptr_t>(result);
};
static void TestPthreadMutexLockNormal(int protocol) {
PthreadMutex m(PTHREAD_MUTEX_NORMAL, protocol);
ASSERT_EQ(0, pthread_mutex_lock(&m.lock));
if (protocol == PTHREAD_PRIO_INHERIT) {
ASSERT_EQ(EPERM, UnlockFromAnotherThread(&m.lock));
}
ASSERT_EQ(0, pthread_mutex_unlock(&m.lock));
ASSERT_EQ(0, pthread_mutex_trylock(&m.lock));
ASSERT_EQ(EBUSY, pthread_mutex_trylock(&m.lock));
ASSERT_EQ(0, pthread_mutex_unlock(&m.lock));
}
static void TestPthreadMutexLockErrorCheck(int protocol) {
PthreadMutex m(PTHREAD_MUTEX_ERRORCHECK, protocol);
ASSERT_EQ(0, pthread_mutex_lock(&m.lock));
ASSERT_EQ(EPERM, UnlockFromAnotherThread(&m.lock));
ASSERT_EQ(EDEADLK, pthread_mutex_lock(&m.lock));
ASSERT_EQ(0, pthread_mutex_unlock(&m.lock));
ASSERT_EQ(0, pthread_mutex_trylock(&m.lock));
if (protocol == PTHREAD_PRIO_NONE) {
ASSERT_EQ(EBUSY, pthread_mutex_trylock(&m.lock));
} else {
ASSERT_EQ(EDEADLK, pthread_mutex_trylock(&m.lock));
}
ASSERT_EQ(0, pthread_mutex_unlock(&m.lock));
ASSERT_EQ(EPERM, pthread_mutex_unlock(&m.lock));
}
static void TestPthreadMutexLockRecursive(int protocol) {
PthreadMutex m(PTHREAD_MUTEX_RECURSIVE, protocol);
ASSERT_EQ(0, pthread_mutex_lock(&m.lock));
ASSERT_EQ(EPERM, UnlockFromAnotherThread(&m.lock));
ASSERT_EQ(0, pthread_mutex_lock(&m.lock));
ASSERT_EQ(EPERM, UnlockFromAnotherThread(&m.lock));
ASSERT_EQ(0, pthread_mutex_unlock(&m.lock));
ASSERT_EQ(0, pthread_mutex_unlock(&m.lock));
ASSERT_EQ(0, pthread_mutex_trylock(&m.lock));
ASSERT_EQ(0, pthread_mutex_trylock(&m.lock));
ASSERT_EQ(0, pthread_mutex_unlock(&m.lock));
ASSERT_EQ(0, pthread_mutex_unlock(&m.lock));
ASSERT_EQ(EPERM, pthread_mutex_unlock(&m.lock));
}
TEST(pthread, pthread_mutex_lock_NORMAL) {
TestPthreadMutexLockNormal(PTHREAD_PRIO_NONE);
}
TEST(pthread, pthread_mutex_lock_ERRORCHECK) {
TestPthreadMutexLockErrorCheck(PTHREAD_PRIO_NONE);
}
TEST(pthread, pthread_mutex_lock_RECURSIVE) {
TestPthreadMutexLockRecursive(PTHREAD_PRIO_NONE);
}
TEST(pthread, pthread_mutex_lock_pi) {
TestPthreadMutexLockNormal(PTHREAD_PRIO_INHERIT);
TestPthreadMutexLockErrorCheck(PTHREAD_PRIO_INHERIT);
TestPthreadMutexLockRecursive(PTHREAD_PRIO_INHERIT);
}
TEST(pthread, pthread_mutex_pi_count_limit) {
#if defined(__BIONIC__) && !defined(__LP64__)
// Bionic only supports 65536 pi mutexes in 32-bit programs.
pthread_mutexattr_t attr;
ASSERT_EQ(0, pthread_mutexattr_init(&attr));
ASSERT_EQ(0, pthread_mutexattr_setprotocol(&attr, PTHREAD_PRIO_INHERIT));
std::vector<pthread_mutex_t> mutexes(65536);
// Test if we can use 65536 pi mutexes at the same time.
// Run 2 times to check if freed pi mutexes can be recycled.
for (int repeat = 0; repeat < 2; ++repeat) {
for (auto& m : mutexes) {
ASSERT_EQ(0, pthread_mutex_init(&m, &attr));
}
pthread_mutex_t m;
ASSERT_EQ(ENOMEM, pthread_mutex_init(&m, &attr));
for (auto& m : mutexes) {
ASSERT_EQ(0, pthread_mutex_lock(&m));
}
for (auto& m : mutexes) {
ASSERT_EQ(0, pthread_mutex_unlock(&m));
}
for (auto& m : mutexes) {
ASSERT_EQ(0, pthread_mutex_destroy(&m));
}
}
ASSERT_EQ(0, pthread_mutexattr_destroy(&attr));
#else
GTEST_SKIP() << "pi mutex count not limited to 64Ki";
#endif
}
TEST(pthread, pthread_mutex_init_same_as_static_initializers) {
pthread_mutex_t lock_normal = PTHREAD_MUTEX_INITIALIZER;
PthreadMutex m1(PTHREAD_MUTEX_NORMAL);
ASSERT_EQ(0, memcmp(&lock_normal, &m1.lock, sizeof(pthread_mutex_t)));
pthread_mutex_destroy(&lock_normal);
pthread_mutex_t lock_errorcheck = PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP;
PthreadMutex m2(PTHREAD_MUTEX_ERRORCHECK);
ASSERT_EQ(0, memcmp(&lock_errorcheck, &m2.lock, sizeof(pthread_mutex_t)));
pthread_mutex_destroy(&lock_errorcheck);
pthread_mutex_t lock_recursive = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
PthreadMutex m3(PTHREAD_MUTEX_RECURSIVE);
ASSERT_EQ(0, memcmp(&lock_recursive, &m3.lock, sizeof(pthread_mutex_t)));
ASSERT_EQ(0, pthread_mutex_destroy(&lock_recursive));
}
class MutexWakeupHelper {
private:
PthreadMutex m;
enum Progress {
LOCK_INITIALIZED,
LOCK_WAITING,
LOCK_RELEASED,
LOCK_ACCESSED
};
std::atomic<Progress> progress;
std::atomic<pid_t> tid;
static void thread_fn(MutexWakeupHelper* helper) {
helper->tid = gettid();
ASSERT_EQ(LOCK_INITIALIZED, helper->progress);
helper->progress = LOCK_WAITING;
ASSERT_EQ(0, pthread_mutex_lock(&helper->m.lock));
ASSERT_EQ(LOCK_RELEASED, helper->progress);
ASSERT_EQ(0, pthread_mutex_unlock(&helper->m.lock));
helper->progress = LOCK_ACCESSED;
}
public:
explicit MutexWakeupHelper(int mutex_type) : m(mutex_type) {
}
void test() {
ASSERT_EQ(0, pthread_mutex_lock(&m.lock));
progress = LOCK_INITIALIZED;
tid = 0;
pthread_t thread;
ASSERT_EQ(0, pthread_create(&thread, nullptr,
reinterpret_cast<void* (*)(void*)>(MutexWakeupHelper::thread_fn), this));
WaitUntilThreadSleep(tid);
ASSERT_EQ(LOCK_WAITING, progress);
progress = LOCK_RELEASED;
ASSERT_EQ(0, pthread_mutex_unlock(&m.lock));
ASSERT_EQ(0, pthread_join(thread, nullptr));
ASSERT_EQ(LOCK_ACCESSED, progress);
}
};
TEST(pthread, pthread_mutex_NORMAL_wakeup) {
MutexWakeupHelper helper(PTHREAD_MUTEX_NORMAL);
helper.test();
}
TEST(pthread, pthread_mutex_ERRORCHECK_wakeup) {
MutexWakeupHelper helper(PTHREAD_MUTEX_ERRORCHECK);
helper.test();
}
TEST(pthread, pthread_mutex_RECURSIVE_wakeup) {
MutexWakeupHelper helper(PTHREAD_MUTEX_RECURSIVE);
helper.test();
}
static int GetThreadPriority(pid_t tid) {
// sched_getparam() returns the static priority of a thread, which can't reflect a thread's
// priority after priority inheritance. So read /proc/<pid>/stat to get the dynamic priority.
std::string filename = android::base::StringPrintf("/proc/%d/stat", tid);
std::string content;
int result = INT_MAX;
if (!android::base::ReadFileToString(filename, &content)) {
return result;
}
std::vector<std::string> strs = android::base::Split(content, " ");
if (strs.size() < 18) {
return result;
}
if (!android::base::ParseInt(strs[17], &result)) {
return INT_MAX;
}
return result;
}
class PIMutexWakeupHelper {
private:
PthreadMutex m;
int protocol;
enum Progress {
LOCK_INITIALIZED,
LOCK_CHILD_READY,
LOCK_WAITING,
LOCK_RELEASED,
};
std::atomic<Progress> progress;
std::atomic<pid_t> main_tid;
std::atomic<pid_t> child_tid;
PthreadMutex start_thread_m;
static void thread_fn(PIMutexWakeupHelper* helper) {
helper->child_tid = gettid();
ASSERT_EQ(LOCK_INITIALIZED, helper->progress);
ASSERT_EQ(0, setpriority(PRIO_PROCESS, gettid(), 1));
ASSERT_EQ(21, GetThreadPriority(gettid()));
ASSERT_EQ(0, pthread_mutex_lock(&helper->m.lock));
helper->progress = LOCK_CHILD_READY;
ASSERT_EQ(0, pthread_mutex_lock(&helper->start_thread_m.lock));
ASSERT_EQ(0, pthread_mutex_unlock(&helper->start_thread_m.lock));
WaitUntilThreadSleep(helper->main_tid);
ASSERT_EQ(LOCK_WAITING, helper->progress);
if (helper->protocol == PTHREAD_PRIO_INHERIT) {
ASSERT_EQ(20, GetThreadPriority(gettid()));
} else {
ASSERT_EQ(21, GetThreadPriority(gettid()));
}
helper->progress = LOCK_RELEASED;
ASSERT_EQ(0, pthread_mutex_unlock(&helper->m.lock));
}
public:
explicit PIMutexWakeupHelper(int mutex_type, int protocol)
: m(mutex_type, protocol), protocol(protocol), start_thread_m(PTHREAD_MUTEX_NORMAL) {
}
void test() {
ASSERT_EQ(0, pthread_mutex_lock(&start_thread_m.lock));
main_tid = gettid();
ASSERT_EQ(20, GetThreadPriority(main_tid));
progress = LOCK_INITIALIZED;
child_tid = 0;
pthread_t thread;
ASSERT_EQ(0, pthread_create(&thread, nullptr,
reinterpret_cast<void* (*)(void*)>(PIMutexWakeupHelper::thread_fn), this));
WaitUntilThreadSleep(child_tid);
ASSERT_EQ(LOCK_CHILD_READY, progress);
ASSERT_EQ(0, pthread_mutex_unlock(&start_thread_m.lock));
progress = LOCK_WAITING;
ASSERT_EQ(0, pthread_mutex_lock(&m.lock));
ASSERT_EQ(LOCK_RELEASED, progress);
ASSERT_EQ(0, pthread_mutex_unlock(&m.lock));
ASSERT_EQ(0, pthread_join(thread, nullptr));
}
};
TEST(pthread, pthread_mutex_pi_wakeup) {
for (int type : {PTHREAD_MUTEX_NORMAL, PTHREAD_MUTEX_RECURSIVE, PTHREAD_MUTEX_ERRORCHECK}) {
for (int protocol : {PTHREAD_PRIO_INHERIT}) {
PIMutexWakeupHelper helper(type, protocol);
helper.test();
}
}
}
TEST(pthread, pthread_mutex_owner_tid_limit) {
#if defined(__BIONIC__) && !defined(__LP64__)
FILE* fp = fopen("/proc/sys/kernel/pid_max", "r");
ASSERT_TRUE(fp != nullptr);
long pid_max;
ASSERT_EQ(1, fscanf(fp, "%ld", &pid_max));
fclose(fp);
// Bionic's pthread_mutex implementation on 32-bit devices uses 16 bits to represent owner tid.
ASSERT_LE(pid_max, 65536);
#else
GTEST_SKIP() << "pthread_mutex supports 32-bit tid";
#endif
}
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
static void pthread_mutex_timedlock_helper(clockid_t clock,
int (*lock_function)(pthread_mutex_t* __mutex,
const timespec* __timeout)) {
pthread_mutex_t m;
ASSERT_EQ(0, pthread_mutex_init(&m, nullptr));
// If the mutex is already locked, pthread_mutex_timedlock should time out.
ASSERT_EQ(0, pthread_mutex_lock(&m));
timespec ts;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EQ(0, clock_gettime(clock, &ts));
ASSERT_EQ(ETIMEDOUT, lock_function(&m, &ts));
ts.tv_nsec = -1;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EQ(EINVAL, lock_function(&m, &ts));
ts.tv_nsec = NS_PER_S;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EQ(EINVAL, lock_function(&m, &ts));
ts.tv_nsec = NS_PER_S - 1;
ts.tv_sec = -1;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EQ(ETIMEDOUT, lock_function(&m, &ts));
// If the mutex is unlocked, pthread_mutex_timedlock should succeed.
ASSERT_EQ(0, pthread_mutex_unlock(&m));
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EQ(0, clock_gettime(clock, &ts));
ts.tv_sec += 1;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EQ(0, lock_function(&m, &ts));
ASSERT_EQ(0, pthread_mutex_unlock(&m));
ASSERT_EQ(0, pthread_mutex_destroy(&m));
}
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
TEST(pthread, pthread_mutex_timedlock) {
pthread_mutex_timedlock_helper(CLOCK_REALTIME, pthread_mutex_timedlock);
}
TEST(pthread, pthread_mutex_timedlock_monotonic_np) {
#if defined(__BIONIC__)
pthread_mutex_timedlock_helper(CLOCK_MONOTONIC, pthread_mutex_timedlock_monotonic_np);
#else // __BIONIC__
GTEST_SKIP() << "pthread_mutex_timedlock_monotonic_np not available";
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
#endif // __BIONIC__
}
TEST(pthread, pthread_mutex_clocklock) {
#if defined(__BIONIC__)
pthread_mutex_timedlock_helper(
CLOCK_MONOTONIC, [](pthread_mutex_t* __mutex, const timespec* __timeout) {
return pthread_mutex_clocklock(__mutex, CLOCK_MONOTONIC, __timeout);
});
pthread_mutex_timedlock_helper(
CLOCK_REALTIME, [](pthread_mutex_t* __mutex, const timespec* __timeout) {
return pthread_mutex_clocklock(__mutex, CLOCK_REALTIME, __timeout);
});
#else // __BIONIC__
GTEST_SKIP() << "pthread_mutex_clocklock not available";
#endif // __BIONIC__
}
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
static void pthread_mutex_timedlock_pi_helper(clockid_t clock,
int (*lock_function)(pthread_mutex_t* __mutex,
const timespec* __timeout)) {
PthreadMutex m(PTHREAD_MUTEX_NORMAL, PTHREAD_PRIO_INHERIT);
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
timespec ts;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
clock_gettime(clock, &ts);
ts.tv_sec += 1;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EQ(0, lock_function(&m.lock, &ts));
struct ThreadArgs {
clockid_t clock;
int (*lock_function)(pthread_mutex_t* __mutex, const timespec* __timeout);
PthreadMutex& m;
};
ThreadArgs thread_args = {
.clock = clock,
.lock_function = lock_function,
.m = m,
};
auto ThreadFn = [](void* arg) -> void* {
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
auto args = static_cast<ThreadArgs*>(arg);
timespec ts;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
clock_gettime(args->clock, &ts);
ts.tv_sec += 1;
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
intptr_t result = args->lock_function(&args->m.lock, &ts);
return reinterpret_cast<void*>(result);
};
pthread_t thread;
ASSERT_EQ(0, pthread_create(&thread, nullptr, ThreadFn, &thread_args));
void* result;
ASSERT_EQ(0, pthread_join(thread, &result));
ASSERT_EQ(ETIMEDOUT, reinterpret_cast<intptr_t>(result));
ASSERT_EQ(0, pthread_mutex_unlock(&m.lock));
}
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
TEST(pthread, pthread_mutex_timedlock_pi) {
pthread_mutex_timedlock_pi_helper(CLOCK_REALTIME, pthread_mutex_timedlock);
}
TEST(pthread, pthread_mutex_timedlock_monotonic_np_pi) {
#if defined(__BIONIC__)
pthread_mutex_timedlock_pi_helper(CLOCK_MONOTONIC, pthread_mutex_timedlock_monotonic_np);
#else // __BIONIC__
GTEST_SKIP() << "pthread_mutex_timedlock_monotonic_np not available";
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
#endif // __BIONIC__
}
TEST(pthread, pthread_mutex_clocklock_pi) {
#if defined(__BIONIC__)
pthread_mutex_timedlock_pi_helper(
CLOCK_MONOTONIC, [](pthread_mutex_t* __mutex, const timespec* __timeout) {
return pthread_mutex_clocklock(__mutex, CLOCK_MONOTONIC, __timeout);
});
pthread_mutex_timedlock_pi_helper(
CLOCK_REALTIME, [](pthread_mutex_t* __mutex, const timespec* __timeout) {
return pthread_mutex_clocklock(__mutex, CLOCK_REALTIME, __timeout);
});
#else // __BIONIC__
GTEST_SKIP() << "pthread_mutex_clocklock not available";
#endif // __BIONIC__
}
TEST(pthread, pthread_mutex_clocklock_invalid) {
#if defined(__BIONIC__)
pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
timespec ts;
EXPECT_EQ(EINVAL, pthread_mutex_clocklock(&mutex, CLOCK_PROCESS_CPUTIME_ID, &ts));
#else // __BIONIC__
GTEST_SKIP() << "pthread_mutex_clocklock not available";
#endif // __BIONIC__
}
TEST(pthread, pthread_mutex_using_destroyed_mutex) {
#if defined(__BIONIC__)
pthread_mutex_t m;
ASSERT_EQ(0, pthread_mutex_init(&m, nullptr));
ASSERT_EQ(0, pthread_mutex_destroy(&m));
ASSERT_EXIT(pthread_mutex_lock(&m), ::testing::KilledBySignal(SIGABRT),
"pthread_mutex_lock called on a destroyed mutex");
ASSERT_EXIT(pthread_mutex_unlock(&m), ::testing::KilledBySignal(SIGABRT),
"pthread_mutex_unlock called on a destroyed mutex");
ASSERT_EXIT(pthread_mutex_trylock(&m), ::testing::KilledBySignal(SIGABRT),
"pthread_mutex_trylock called on a destroyed mutex");
timespec ts;
ASSERT_EXIT(pthread_mutex_timedlock(&m, &ts), ::testing::KilledBySignal(SIGABRT),
"pthread_mutex_timedlock called on a destroyed mutex");
Add _monotonic_np versions of timed wait functions As a follow up to Ibba98f5d88be1c306d14e9b9366302ecbef6d534, where we added a work around to convert the CLOCK_REALTIME timeouts to CLOCK_MONOTONIC for pthread and semaphore timed wait functions, we're introducing a set of _monotonic_np versions of each of these functions that wait on CLOCK_MONOTONIC directly. The primary motivation here is that while the above work around helps for 3rd party code, it creates a dilemma when implementing new code that would use these functions: either one implements code with these functions knowing there is a race condition possible or one avoids these functions and reinvent their own waiting/signaling mechanisms. Neither are satisfactory, so we create a third option to use these Android specific _monotonic_np functions that completely remove the race condition while keeping the rest of the interface. Specifically this adds the below functions: pthread_mutex_timedlock_monotonic_np() pthread_cond_timedwait_monotonic_np() pthread_rwlock_timedrdlock_monotonic_np() pthread_rwlock_timedwrlock_monotonic_np() sem_timedwait_monotonic_np() Note that pthread_cond_timedwait_monotonic_np() previously existed and was removed since it's possible to initialize a condition variable to use CLOCK_MONOTONIC. It is added back for a mix of reasons, 1) Symmetry with the rest of the functions we're adding 2) libc++ cannot easily take advantage of the new initializer, but will be able to use this function in order to wait on std::steady_clock 3) Frankly, it's a better API to specify the clock in the waiter function than to specify the clock when the condition variable is initialized. Bug: 73951740 Test: new unit tests Change-Id: I23aa5c204e36a194237d41e064c5c8ccaa4204e3
2018-03-05 23:14:44 +01:00
ASSERT_EXIT(pthread_mutex_timedlock_monotonic_np(&m, &ts), ::testing::KilledBySignal(SIGABRT),
"pthread_mutex_timedlock_monotonic_np called on a destroyed mutex");
ASSERT_EXIT(pthread_mutex_clocklock(&m, CLOCK_MONOTONIC, &ts), ::testing::KilledBySignal(SIGABRT),
"pthread_mutex_clocklock called on a destroyed mutex");
ASSERT_EXIT(pthread_mutex_clocklock(&m, CLOCK_REALTIME, &ts), ::testing::KilledBySignal(SIGABRT),
"pthread_mutex_clocklock called on a destroyed mutex");
ASSERT_EXIT(pthread_mutex_clocklock(&m, CLOCK_PROCESS_CPUTIME_ID, &ts),
::testing::KilledBySignal(SIGABRT),
"pthread_mutex_clocklock called on a destroyed mutex");
ASSERT_EXIT(pthread_mutex_destroy(&m), ::testing::KilledBySignal(SIGABRT),
"pthread_mutex_destroy called on a destroyed mutex");
#else
GTEST_SKIP() << "bionic-only test";
#endif
}
class StrictAlignmentAllocator {
public:
void* allocate(size_t size, size_t alignment) {
char* p = new char[size + alignment * 2];
allocated_array.push_back(p);
while (!is_strict_aligned(p, alignment)) {
++p;
}
return p;
}
~StrictAlignmentAllocator() {
for (const auto& p : allocated_array) {
delete[] p;
}
}
private:
bool is_strict_aligned(char* p, size_t alignment) {
return (reinterpret_cast<uintptr_t>(p) % (alignment * 2)) == alignment;
}
std::vector<char*> allocated_array;
};
TEST(pthread, pthread_types_allow_four_bytes_alignment) {
#if defined(__BIONIC__)
// For binary compatibility with old version, we need to allow 4-byte aligned data for pthread types.
StrictAlignmentAllocator allocator;
pthread_mutex_t* mutex = reinterpret_cast<pthread_mutex_t*>(
allocator.allocate(sizeof(pthread_mutex_t), 4));
ASSERT_EQ(0, pthread_mutex_init(mutex, nullptr));
ASSERT_EQ(0, pthread_mutex_lock(mutex));
ASSERT_EQ(0, pthread_mutex_unlock(mutex));
ASSERT_EQ(0, pthread_mutex_destroy(mutex));
pthread_cond_t* cond = reinterpret_cast<pthread_cond_t*>(
allocator.allocate(sizeof(pthread_cond_t), 4));
ASSERT_EQ(0, pthread_cond_init(cond, nullptr));
ASSERT_EQ(0, pthread_cond_signal(cond));
ASSERT_EQ(0, pthread_cond_broadcast(cond));
ASSERT_EQ(0, pthread_cond_destroy(cond));
pthread_rwlock_t* rwlock = reinterpret_cast<pthread_rwlock_t*>(
allocator.allocate(sizeof(pthread_rwlock_t), 4));
ASSERT_EQ(0, pthread_rwlock_init(rwlock, nullptr));
ASSERT_EQ(0, pthread_rwlock_rdlock(rwlock));
ASSERT_EQ(0, pthread_rwlock_unlock(rwlock));
ASSERT_EQ(0, pthread_rwlock_wrlock(rwlock));
ASSERT_EQ(0, pthread_rwlock_unlock(rwlock));
ASSERT_EQ(0, pthread_rwlock_destroy(rwlock));
#else
GTEST_SKIP() << "bionic-only test";
#endif
}
TEST(pthread, pthread_mutex_lock_null_32) {
#if defined(__BIONIC__) && !defined(__LP64__)
// For LP32, the pthread lock/unlock functions allow a NULL mutex and return
// EINVAL in that case: http://b/19995172.
//
// We decorate the public defintion with _Nonnull so that people recompiling
// their code with get a warning and might fix their bug, but need to pass
// NULL here to test that we remain compatible.
pthread_mutex_t* null_value = nullptr;
ASSERT_EQ(EINVAL, pthread_mutex_lock(null_value));
#else
GTEST_SKIP() << "32-bit bionic-only test";
#endif
}
TEST(pthread, pthread_mutex_unlock_null_32) {
#if defined(__BIONIC__) && !defined(__LP64__)
// For LP32, the pthread lock/unlock functions allow a NULL mutex and return
// EINVAL in that case: http://b/19995172.
//
// We decorate the public defintion with _Nonnull so that people recompiling
// their code with get a warning and might fix their bug, but need to pass
// NULL here to test that we remain compatible.
pthread_mutex_t* null_value = nullptr;
ASSERT_EQ(EINVAL, pthread_mutex_unlock(null_value));
#else
GTEST_SKIP() << "32-bit bionic-only test";
#endif
}
TEST_F(pthread_DeathTest, pthread_mutex_lock_null_64) {
#if defined(__BIONIC__) && defined(__LP64__)
pthread_mutex_t* null_value = nullptr;
ASSERT_EXIT(pthread_mutex_lock(null_value), testing::KilledBySignal(SIGSEGV), "");
#else
GTEST_SKIP() << "64-bit bionic-only test";
#endif
}
TEST_F(pthread_DeathTest, pthread_mutex_unlock_null_64) {
#if defined(__BIONIC__) && defined(__LP64__)
pthread_mutex_t* null_value = nullptr;
ASSERT_EXIT(pthread_mutex_unlock(null_value), testing::KilledBySignal(SIGSEGV), "");
#else
GTEST_SKIP() << "64-bit bionic-only test";
#endif
}
extern _Unwind_Reason_Code FrameCounter(_Unwind_Context* ctx, void* arg);
static volatile bool signal_handler_on_altstack_done;
__attribute__((__noinline__))
static void signal_handler_backtrace() {
// Check if we have enough stack space for unwinding.
int count = 0;
_Unwind_Backtrace(FrameCounter, &count);
ASSERT_GT(count, 0);
}
__attribute__((__noinline__))
static void signal_handler_logging() {
// Check if we have enough stack space for logging.
std::string s(2048, '*');
GTEST_LOG_(INFO) << s;
signal_handler_on_altstack_done = true;
}
__attribute__((__noinline__))
static void signal_handler_snprintf() {
// Check if we have enough stack space for snprintf to a PATH_MAX buffer, plus some extra.
char buf[PATH_MAX + 2048];
ASSERT_GT(snprintf(buf, sizeof(buf), "/proc/%d/status", getpid()), 0);
}
static void SignalHandlerOnAltStack(int signo, siginfo_t*, void*) {
ASSERT_EQ(SIGUSR1, signo);
signal_handler_backtrace();
signal_handler_logging();
signal_handler_snprintf();
}
TEST(pthread, big_enough_signal_stack) {
signal_handler_on_altstack_done = false;
ScopedSignalHandler handler(SIGUSR1, SignalHandlerOnAltStack, SA_SIGINFO | SA_ONSTACK);
kill(getpid(), SIGUSR1);
ASSERT_TRUE(signal_handler_on_altstack_done);
}
TEST(pthread, pthread_barrierattr_smoke) {
pthread_barrierattr_t attr;
ASSERT_EQ(0, pthread_barrierattr_init(&attr));
int pshared;
ASSERT_EQ(0, pthread_barrierattr_getpshared(&attr, &pshared));
ASSERT_EQ(PTHREAD_PROCESS_PRIVATE, pshared);
ASSERT_EQ(0, pthread_barrierattr_setpshared(&attr, PTHREAD_PROCESS_SHARED));
ASSERT_EQ(0, pthread_barrierattr_getpshared(&attr, &pshared));
ASSERT_EQ(PTHREAD_PROCESS_SHARED, pshared);
ASSERT_EQ(0, pthread_barrierattr_destroy(&attr));
}
struct BarrierTestHelperData {
size_t thread_count;
pthread_barrier_t barrier;
std::atomic<int> finished_mask;
std::atomic<int> serial_thread_count;
size_t iteration_count;
std::atomic<size_t> finished_iteration_count;
BarrierTestHelperData(size_t thread_count, size_t iteration_count)
: thread_count(thread_count), finished_mask(0), serial_thread_count(0),
iteration_count(iteration_count), finished_iteration_count(0) {
}
};
struct BarrierTestHelperArg {
int id;
BarrierTestHelperData* data;
};
static void BarrierTestHelper(BarrierTestHelperArg* arg) {
for (size_t i = 0; i < arg->data->iteration_count; ++i) {
int result = pthread_barrier_wait(&arg->data->barrier);
if (result == PTHREAD_BARRIER_SERIAL_THREAD) {
arg->data->serial_thread_count++;
} else {
ASSERT_EQ(0, result);
}
int mask = arg->data->finished_mask.fetch_or(1 << arg->id);
mask |= 1 << arg->id;
if (mask == ((1 << arg->data->thread_count) - 1)) {
ASSERT_EQ(1, arg->data->serial_thread_count);
arg->data->finished_iteration_count++;
arg->data->finished_mask = 0;
arg->data->serial_thread_count = 0;
}
}
}
TEST(pthread, pthread_barrier_smoke) {
const size_t BARRIER_ITERATION_COUNT = 10;
const size_t BARRIER_THREAD_COUNT = 10;
BarrierTestHelperData data(BARRIER_THREAD_COUNT, BARRIER_ITERATION_COUNT);
ASSERT_EQ(0, pthread_barrier_init(&data.barrier, nullptr, data.thread_count));
std::vector<pthread_t> threads(data.thread_count);
std::vector<BarrierTestHelperArg> args(threads.size());
for (size_t i = 0; i < threads.size(); ++i) {
args[i].id = i;
args[i].data = &data;
ASSERT_EQ(0, pthread_create(&threads[i], nullptr,
reinterpret_cast<void* (*)(void*)>(BarrierTestHelper), &args[i]));
}
for (size_t i = 0; i < threads.size(); ++i) {
ASSERT_EQ(0, pthread_join(threads[i], nullptr));
}
ASSERT_EQ(data.iteration_count, data.finished_iteration_count);
ASSERT_EQ(0, pthread_barrier_destroy(&data.barrier));
}
struct BarrierDestroyTestArg {
std::atomic<int> tid;
pthread_barrier_t* barrier;
};
static void BarrierDestroyTestHelper(BarrierDestroyTestArg* arg) {
arg->tid = gettid();
ASSERT_EQ(0, pthread_barrier_wait(arg->barrier));
}
TEST(pthread, pthread_barrier_destroy) {
pthread_barrier_t barrier;
ASSERT_EQ(0, pthread_barrier_init(&barrier, nullptr, 2));
pthread_t thread;
BarrierDestroyTestArg arg;
arg.tid = 0;
arg.barrier = &barrier;
ASSERT_EQ(0, pthread_create(&thread, nullptr,
reinterpret_cast<void* (*)(void*)>(BarrierDestroyTestHelper), &arg));
WaitUntilThreadSleep(arg.tid);
ASSERT_EQ(EBUSY, pthread_barrier_destroy(&barrier));
ASSERT_EQ(PTHREAD_BARRIER_SERIAL_THREAD, pthread_barrier_wait(&barrier));
// Verify if the barrier can be destroyed directly after pthread_barrier_wait().
ASSERT_EQ(0, pthread_barrier_destroy(&barrier));
ASSERT_EQ(0, pthread_join(thread, nullptr));
#if defined(__BIONIC__)
ASSERT_EQ(EINVAL, pthread_barrier_destroy(&barrier));
#endif
}
struct BarrierOrderingTestHelperArg {
pthread_barrier_t* barrier;
size_t* array;
size_t array_length;
size_t id;
};
void BarrierOrderingTestHelper(BarrierOrderingTestHelperArg* arg) {
const size_t ITERATION_COUNT = 10000;
for (size_t i = 1; i <= ITERATION_COUNT; ++i) {
arg->array[arg->id] = i;
int result = pthread_barrier_wait(arg->barrier);
ASSERT_TRUE(result == 0 || result == PTHREAD_BARRIER_SERIAL_THREAD);
for (size_t j = 0; j < arg->array_length; ++j) {
ASSERT_EQ(i, arg->array[j]);
}
result = pthread_barrier_wait(arg->barrier);
ASSERT_TRUE(result == 0 || result == PTHREAD_BARRIER_SERIAL_THREAD);
}
}
TEST(pthread, pthread_barrier_check_ordering) {
const size_t THREAD_COUNT = 4;
pthread_barrier_t barrier;
ASSERT_EQ(0, pthread_barrier_init(&barrier, nullptr, THREAD_COUNT));
size_t array[THREAD_COUNT];
std::vector<pthread_t> threads(THREAD_COUNT);
std::vector<BarrierOrderingTestHelperArg> args(THREAD_COUNT);
for (size_t i = 0; i < THREAD_COUNT; ++i) {
args[i].barrier = &barrier;
args[i].array = array;
args[i].array_length = THREAD_COUNT;
args[i].id = i;
ASSERT_EQ(0, pthread_create(&threads[i], nullptr,
reinterpret_cast<void* (*)(void*)>(BarrierOrderingTestHelper),
&args[i]));
}
for (size_t i = 0; i < THREAD_COUNT; ++i) {
ASSERT_EQ(0, pthread_join(threads[i], nullptr));
}
}
TEST(pthread, pthread_barrier_init_zero_count) {
pthread_barrier_t barrier;
ASSERT_EQ(EINVAL, pthread_barrier_init(&barrier, nullptr, 0));
}
TEST(pthread, pthread_spinlock_smoke) {
pthread_spinlock_t lock;
ASSERT_EQ(0, pthread_spin_init(&lock, 0));
ASSERT_EQ(0, pthread_spin_trylock(&lock));
ASSERT_EQ(0, pthread_spin_unlock(&lock));
ASSERT_EQ(0, pthread_spin_lock(&lock));
ASSERT_EQ(EBUSY, pthread_spin_trylock(&lock));
ASSERT_EQ(0, pthread_spin_unlock(&lock));
ASSERT_EQ(0, pthread_spin_destroy(&lock));
}
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
TEST(pthread, pthread_attr_getdetachstate__pthread_attr_setdetachstate) {
pthread_attr_t attr;
ASSERT_EQ(0, pthread_attr_init(&attr));
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
int state;
ASSERT_EQ(0, pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED));
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
ASSERT_EQ(0, pthread_attr_getdetachstate(&attr, &state));
ASSERT_EQ(PTHREAD_CREATE_DETACHED, state);
ASSERT_EQ(0, pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE));
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
ASSERT_EQ(0, pthread_attr_getdetachstate(&attr, &state));
ASSERT_EQ(PTHREAD_CREATE_JOINABLE, state);
ASSERT_EQ(EINVAL, pthread_attr_setdetachstate(&attr, 123));
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
ASSERT_EQ(0, pthread_attr_getdetachstate(&attr, &state));
ASSERT_EQ(PTHREAD_CREATE_JOINABLE, state);
}
TEST(pthread, pthread_create__mmap_failures) {
pthread_attr_t attr;
ASSERT_EQ(0, pthread_attr_init(&attr));
ASSERT_EQ(0, pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED));
const auto kPageSize = sysconf(_SC_PAGE_SIZE);
// Use up all the VMAs. By default this is 64Ki (though some will already be in use).
std::vector<void*> pages;
pages.reserve(64 * 1024);
int prot = PROT_NONE;
while (true) {
void* page = mmap(nullptr, kPageSize, prot, MAP_ANON|MAP_PRIVATE, -1, 0);
if (page == MAP_FAILED) break;
pages.push_back(page);
prot = (prot == PROT_NONE) ? PROT_READ : PROT_NONE;
}
// Try creating threads, freeing up a page each time we fail.
size_t EAGAIN_count = 0;
size_t i = 0;
for (; i < pages.size(); ++i) {
pthread_t t;
int status = pthread_create(&t, &attr, IdFn, nullptr);
if (status != EAGAIN) break;
++EAGAIN_count;
ASSERT_EQ(0, munmap(pages[i], kPageSize));
}
Reorganize static TLS memory for ELF TLS For ELF TLS "local-exec" accesses, the static linker assumes that an executable's TLS segment is located at a statically-known offset from the thread pointer (i.e. "variant 1" for ARM and "variant 2" for x86). Because these layouts are incompatible, Bionic generally needs to allocate its TLS slots differently between different architectures. To allow per-architecture TLS slots: - Replace the TLS_SLOT_xxx enumerators with macros. New ARM slots are generally negative, while new x86 slots are generally positive. - Define a bionic_tcb struct that provides two things: - a void* raw_slots_storage[BIONIC_TLS_SLOTS] field - an inline accessor function: void*& tls_slot(size_t tpindex); For ELF TLS, it's necessary to allocate a temporary TCB (i.e. TLS slots), because the runtime linker doesn't know how large the static TLS area is until after it has loaded all of the initial solibs. To accommodate Golang, it's necessary to allocate the pthread keys at a fixed, small, positive offset from the thread pointer. This CL moves the pthread keys into bionic_tls, then allocates a single mapping per thread that looks like so: - stack guard - stack [omitted for main thread and with pthread_attr_setstack] - static TLS: - bionic_tcb [exec TLS will either precede or succeed the TCB] - bionic_tls [prefixed by the pthread keys] - [solib TLS segments will be placed here] - guard page As before, if the new mapping includes a stack, the pthread_internal_t is allocated on it. At startup, Bionic allocates a temporary bionic_tcb object on the stack, then allocates a temporary bionic_tls object using mmap. This mmap is delayed because the linker can't currently call async_safe_fatal() before relocating itself. Later, Bionic allocates a stack-less thread mapping for the main thread, and copies slots from the temporary TCB to the new TCB. (See *::copy_from_bootstrap methods.) Bug: http://b/78026329 Test: bionic unit tests Test: verify that a Golang app still works Test: verify that a Golang app crashes if bionic_{tls,tcb} are swapped Merged-In: I6543063752f4ec8ef6dc9c7f2a06ce2a18fc5af3 Change-Id: I6543063752f4ec8ef6dc9c7f2a06ce2a18fc5af3 (cherry picked from commit 1e660b70da625fcbf1e43dfae09b7b4817fa1660)
2019-01-03 11:51:30 +01:00
// Creating a thread uses at least three VMAs: the combined stack and TLS, and a guard on each
// side. So we should have seen at least three failures.
ASSERT_GE(EAGAIN_count, 3U);
for (; i < pages.size(); ++i) {
ASSERT_EQ(0, munmap(pages[i], kPageSize));
}
}
TEST(pthread, pthread_setschedparam) {
sched_param p = { .sched_priority = INT_MIN };
ASSERT_EQ(EINVAL, pthread_setschedparam(pthread_self(), INT_MIN, &p));
}
TEST(pthread, pthread_setschedprio) {
ASSERT_EQ(EINVAL, pthread_setschedprio(pthread_self(), INT_MIN));
}
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
TEST(pthread, pthread_attr_getinheritsched__pthread_attr_setinheritsched) {
pthread_attr_t attr;
ASSERT_EQ(0, pthread_attr_init(&attr));
int state;
ASSERT_EQ(0, pthread_attr_setinheritsched(&attr, PTHREAD_INHERIT_SCHED));
ASSERT_EQ(0, pthread_attr_getinheritsched(&attr, &state));
ASSERT_EQ(PTHREAD_INHERIT_SCHED, state);
ASSERT_EQ(0, pthread_attr_setinheritsched(&attr, PTHREAD_EXPLICIT_SCHED));
ASSERT_EQ(0, pthread_attr_getinheritsched(&attr, &state));
ASSERT_EQ(PTHREAD_EXPLICIT_SCHED, state);
ASSERT_EQ(EINVAL, pthread_attr_setinheritsched(&attr, 123));
ASSERT_EQ(0, pthread_attr_getinheritsched(&attr, &state));
ASSERT_EQ(PTHREAD_EXPLICIT_SCHED, state);
}
TEST(pthread, pthread_attr_setinheritsched__PTHREAD_INHERIT_SCHED__PTHREAD_EXPLICIT_SCHED) {
pthread_attr_t attr;
ASSERT_EQ(0, pthread_attr_init(&attr));
// If we set invalid scheduling attributes but choose to inherit, everything's fine...
sched_param param = { .sched_priority = sched_get_priority_max(SCHED_FIFO) + 1 };
ASSERT_EQ(0, pthread_attr_setschedparam(&attr, &param));
ASSERT_EQ(0, pthread_attr_setschedpolicy(&attr, SCHED_FIFO));
ASSERT_EQ(0, pthread_attr_setinheritsched(&attr, PTHREAD_INHERIT_SCHED));
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, &attr, IdFn, nullptr));
ASSERT_EQ(0, pthread_join(t, nullptr));
#if defined(__LP64__)
// If we ask to use them, though, we'll see a failure...
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
ASSERT_EQ(0, pthread_attr_setinheritsched(&attr, PTHREAD_EXPLICIT_SCHED));
ASSERT_EQ(EINVAL, pthread_create(&t, &attr, IdFn, nullptr));
#else
// For backwards compatibility with broken apps, we just ignore failures
// to set scheduler attributes on LP32.
#endif
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
}
TEST(pthread, pthread_attr_setinheritsched_PTHREAD_INHERIT_SCHED_takes_effect) {
sched_param param = { .sched_priority = sched_get_priority_min(SCHED_FIFO) };
int rc = pthread_setschedparam(pthread_self(), SCHED_FIFO, &param);
if (rc == EPERM) GTEST_SKIP() << "pthread_setschedparam failed with EPERM";
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
ASSERT_EQ(0, rc);
pthread_attr_t attr;
ASSERT_EQ(0, pthread_attr_init(&attr));
ASSERT_EQ(0, pthread_attr_setinheritsched(&attr, PTHREAD_INHERIT_SCHED));
SpinFunctionHelper spin_helper;
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, &attr, spin_helper.GetFunction(), nullptr));
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
int actual_policy;
sched_param actual_param;
ASSERT_EQ(0, pthread_getschedparam(t, &actual_policy, &actual_param));
ASSERT_EQ(SCHED_FIFO, actual_policy);
spin_helper.UnSpin();
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
ASSERT_EQ(0, pthread_join(t, nullptr));
}
TEST(pthread, pthread_attr_setinheritsched_PTHREAD_EXPLICIT_SCHED_takes_effect) {
sched_param param = { .sched_priority = sched_get_priority_min(SCHED_FIFO) };
int rc = pthread_setschedparam(pthread_self(), SCHED_FIFO, &param);
if (rc == EPERM) GTEST_SKIP() << "pthread_setschedparam failed with EPERM";
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
ASSERT_EQ(0, rc);
pthread_attr_t attr;
ASSERT_EQ(0, pthread_attr_init(&attr));
ASSERT_EQ(0, pthread_attr_setinheritsched(&attr, PTHREAD_EXPLICIT_SCHED));
ASSERT_EQ(0, pthread_attr_setschedpolicy(&attr, SCHED_OTHER));
SpinFunctionHelper spin_helper;
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, &attr, spin_helper.GetFunction(), nullptr));
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
int actual_policy;
sched_param actual_param;
ASSERT_EQ(0, pthread_getschedparam(t, &actual_policy, &actual_param));
ASSERT_EQ(SCHED_OTHER, actual_policy);
spin_helper.UnSpin();
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
ASSERT_EQ(0, pthread_join(t, nullptr));
}
TEST(pthread, pthread_attr_setinheritsched__takes_effect_despite_SCHED_RESET_ON_FORK) {
sched_param param = { .sched_priority = sched_get_priority_min(SCHED_FIFO) };
int rc = pthread_setschedparam(pthread_self(), SCHED_FIFO | SCHED_RESET_ON_FORK, &param);
if (rc == EPERM) GTEST_SKIP() << "pthread_setschedparam failed with EPERM";
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
ASSERT_EQ(0, rc);
pthread_attr_t attr;
ASSERT_EQ(0, pthread_attr_init(&attr));
ASSERT_EQ(0, pthread_attr_setinheritsched(&attr, PTHREAD_INHERIT_SCHED));
SpinFunctionHelper spin_helper;
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, &attr, spin_helper.GetFunction(), nullptr));
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
int actual_policy;
sched_param actual_param;
ASSERT_EQ(0, pthread_getschedparam(t, &actual_policy, &actual_param));
ASSERT_EQ(SCHED_FIFO | SCHED_RESET_ON_FORK, actual_policy);
spin_helper.UnSpin();
Implement pthread_attr_getinheritsched/pthread_attr_setinheritsched. Historically, Android defaulted to EXPLICIT but with a special case because SCHED_NORMAL/priority 0 was awkward. Because the code couldn't actually tell whether SCHED_NORMAL/priority 0 was a genuine attempt to explicitly set those attributes (because the parent thread is SCHED_FIFO, say) or just because the pthread_attr_t was left at its defaults. Now we support INHERIT, we could call sched_getscheduler to see whether we actually need to call sched_setscheduler, but since the major cost is the fixed syscall overhead, we may as well just conservatively call sched_setscheduler and let the kernel decide whether it's a no-op. (Especially because we'd then have to add both sched_getscheduler and sched_setscheduler to any seccomp filter.) Platform code (or app code that only needs to support >= P) can actually add a call to pthread_attr_setinheritsched to say that they just want to inherit (if they know that none of their threads actually mess with scheduler attributes at all), which will save them a sched_setscheduler call except in the doubly-special case of SCHED_RESET_ON_FORK (which we do handle). An alternative would be "make pthread_attr_setschedparams and pthread_attr_setschedprio set EXPLICIT and change the platform default to INHERIT", but even though I can only think of weird pathological examples where anyone would notice that change, that behavior -- of pthread_attr_setschedparams/pthread_attr_setschedprio overriding an earlier call to pthread_attr_setinheritsched -- isn't allowed by POSIX (whereas defaulting to EXPLICIT is). If we have a lot of trouble with this change in the app compatibility testing phase, though, we'll want to reconsider this decision! -*- This change also removes a comment about setting the scheduler attributes in main_thread because we'd have to actually keep them up to date, and it's not clear that doing so would be worth the trouble. Also make async_safe_format_log preserve errno so we don't have to be so careful around it. Bug: http://b/67471710 Test: ran tests Change-Id: Idd026c4ce78a536656adcb57aa2e7b2c616eeddf
2017-10-18 00:34:41 +02:00
ASSERT_EQ(0, pthread_join(t, nullptr));
}