2 * Copyright 2015 Facebook, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 * Two Read-Write spin lock implementations.
20 * Ref: http://locklessinc.com/articles/locks
22 * Both locks here are faster than pthread_rwlock and have very low
23 * overhead (usually 20-30ns). They don't use any system mutexes and
24 * are very compact (4/8 bytes), so are suitable for per-instance
25 * based locking, particularly when contention is not expected.
27 * In most cases, RWSpinLock is a reasonable choice. It has minimal
28 * overhead, and comparable contention performance when the number of
29 * competing threads is less than or equal to the number of logical
30 * CPUs. Even as the number of threads gets larger, RWSpinLock can
31 * still be very competitive in READ, although it is slower on WRITE,
32 * and also inherently unfair to writers.
34 * RWTicketSpinLock shows more balanced READ/WRITE performance. If
35 * your application really needs a lot more threads, and a
36 * higher-priority writer, prefer one of the RWTicketSpinLock locks.
40 * RWTicketSpinLock locks can only be used with GCC on x86/x86-64
43 * RWTicketSpinLock<32> only allows up to 2^8 - 1 concurrent
44 * readers and writers.
46 * RWTicketSpinLock<64> only allows up to 2^16 - 1 concurrent
47 * readers and writers.
49 * RWTicketSpinLock<..., true> (kFavorWriter = true, that is, strict
50 * writer priority) is NOT reentrant, even for lock_shared().
52 * The lock will not grant any new shared (read) accesses while a thread
53 * attempting to acquire the lock in write mode is blocked. (That is,
54 * if the lock is held in shared mode by N threads, and a thread attempts
55 * to acquire it in write mode, no one else can acquire it in shared mode
56 * until these N threads release the lock and then the blocked thread
57 * acquires and releases the exclusive lock.) This also applies for
58 * attempts to reacquire the lock in shared mode by threads that already
59 * hold it in shared mode, making the lock non-reentrant.
61 * RWSpinLock handles 2^30 - 1 concurrent readers.
63 * @author Xin Liu <xliux@fb.com>
66 #ifndef FOLLY_RWSPINLOCK_H_
67 #define FOLLY_RWSPINLOCK_H_
70 ========================================================================
71 Benchmark on (Intel(R) Xeon(R) CPU L5630 @ 2.13GHz) 8 cores(16 HTs)
72 ========================================================================
74 ------------------------------------------------------------------------------
75 1. Single thread benchmark (read/write lock + unlock overhead)
76 Benchmark Iters Total t t/iter iter/sec
77 -------------------------------------------------------------------------------
78 * BM_RWSpinLockRead 100000 1.786 ms 17.86 ns 53.4M
79 +30.5% BM_RWSpinLockWrite 100000 2.331 ms 23.31 ns 40.91M
80 +85.7% BM_RWTicketSpinLock32Read 100000 3.317 ms 33.17 ns 28.75M
81 +96.0% BM_RWTicketSpinLock32Write 100000 3.5 ms 35 ns 27.25M
82 +85.6% BM_RWTicketSpinLock64Read 100000 3.315 ms 33.15 ns 28.77M
83 +96.0% BM_RWTicketSpinLock64Write 100000 3.5 ms 35 ns 27.25M
84 +85.7% BM_RWTicketSpinLock32FavorWriterRead 100000 3.317 ms 33.17 ns 28.75M
85 +29.7% BM_RWTicketSpinLock32FavorWriterWrite 100000 2.316 ms 23.16 ns 41.18M
86 +85.3% BM_RWTicketSpinLock64FavorWriterRead 100000 3.309 ms 33.09 ns 28.82M
87 +30.2% BM_RWTicketSpinLock64FavorWriterWrite 100000 2.325 ms 23.25 ns 41.02M
88 + 175% BM_PThreadRWMutexRead 100000 4.917 ms 49.17 ns 19.4M
89 + 166% BM_PThreadRWMutexWrite 100000 4.757 ms 47.57 ns 20.05M
91 ------------------------------------------------------------------------------
92 2. Contention Benchmark 90% read 10% write
93 Benchmark hits average min max sigma
94 ------------------------------------------------------------------------------
95 ---------- 8 threads ------------
96 RWSpinLock Write 142666 220ns 78ns 40.8us 269ns
97 RWSpinLock Read 1282297 222ns 80ns 37.7us 248ns
98 RWTicketSpinLock Write 85692 209ns 71ns 17.9us 252ns
99 RWTicketSpinLock Read 769571 215ns 78ns 33.4us 251ns
100 pthread_rwlock_t Write 84248 2.48us 99ns 269us 8.19us
101 pthread_rwlock_t Read 761646 933ns 101ns 374us 3.25us
103 ---------- 16 threads ------------
104 RWSpinLock Write 124236 237ns 78ns 261us 801ns
105 RWSpinLock Read 1115807 236ns 78ns 2.27ms 2.17us
106 RWTicketSpinLock Write 81781 231ns 71ns 31.4us 351ns
107 RWTicketSpinLock Read 734518 238ns 78ns 73.6us 379ns
108 pthread_rwlock_t Write 83363 7.12us 99ns 785us 28.1us
109 pthread_rwlock_t Read 754978 2.18us 101ns 1.02ms 14.3us
111 ---------- 50 threads ------------
112 RWSpinLock Write 131142 1.37us 82ns 7.53ms 68.2us
113 RWSpinLock Read 1181240 262ns 78ns 6.62ms 12.7us
114 RWTicketSpinLock Write 83045 397ns 73ns 7.01ms 31.5us
115 RWTicketSpinLock Read 744133 386ns 78ns 11ms 31.4us
116 pthread_rwlock_t Write 80849 112us 103ns 4.52ms 263us
117 pthread_rwlock_t Read 728698 24us 101ns 7.28ms 194us
121 #include <folly/Portability.h>
123 #if defined(__GNUC__) && \
124 (defined(__i386) || FOLLY_X64 || \
126 # define RW_SPINLOCK_USE_X86_INTRINSIC_
127 # include <x86intrin.h>
128 #elif defined(_MSC_VER) && defined(FOLLY_X64)
129 # define RW_SPINLOCK_USE_X86_INTRINSIC_
131 # undef RW_SPINLOCK_USE_X86_INTRINSIC_
134 // iOS doesn't define _mm_cvtsi64_si128 and friends
135 #if (FOLLY_SSE >= 2) && !TARGET_OS_IPHONE
136 #define RW_SPINLOCK_USE_SSE_INSTRUCTIONS_
138 #undef RW_SPINLOCK_USE_SSE_INSTRUCTIONS_
144 #include <boost/noncopyable.hpp>
147 #include <glog/logging.h>
149 #include <folly/Likely.h>
154 * A simple, small (4-bytes), but unfair rwlock. Use it when you want
155 * a nice writer and don't expect a lot of write/read contention, or
156 * when you need small rwlocks since you are creating a large number
159 * Note that the unfairness here is extreme: if the lock is
160 * continually accessed for read, writers will never get a chance. If
161 * the lock can be that highly contended this class is probably not an
162 * ideal choice anyway.
164 * It currently implements most of the Lockable, SharedLockable and
165 * UpgradeLockable concepts except the TimedLockable related locking/unlocking
169 enum : int32_t { READER = 4, UPGRADED = 2, WRITER = 1 };
171 constexpr RWSpinLock() : bits_(0) {}
173 RWSpinLock(RWSpinLock const&) = delete;
174 RWSpinLock& operator=(RWSpinLock const&) = delete;
179 while (!LIKELY(try_lock())) {
180 if (++count > 1000) sched_yield();
184 // Writer is responsible for clearing up both the UPGRADED and WRITER bits.
186 static_assert(READER > WRITER + UPGRADED, "wrong bits!");
187 bits_.fetch_and(~(WRITER | UPGRADED), std::memory_order_release);
190 // SharedLockable Concept
193 while (!LIKELY(try_lock_shared())) {
194 if (++count > 1000) sched_yield();
198 void unlock_shared() {
199 bits_.fetch_add(-READER, std::memory_order_release);
202 // Downgrade the lock from writer status to reader status.
203 void unlock_and_lock_shared() {
204 bits_.fetch_add(READER, std::memory_order_acquire);
208 // UpgradeLockable Concept
209 void lock_upgrade() {
211 while (!try_lock_upgrade()) {
212 if (++count > 1000) sched_yield();
216 void unlock_upgrade() {
217 bits_.fetch_add(-UPGRADED, std::memory_order_acq_rel);
220 // unlock upgrade and try to acquire write lock
221 void unlock_upgrade_and_lock() {
223 while (!try_unlock_upgrade_and_lock()) {
224 if (++count > 1000) sched_yield();
228 // unlock upgrade and read lock atomically
229 void unlock_upgrade_and_lock_shared() {
230 bits_.fetch_add(READER - UPGRADED, std::memory_order_acq_rel);
233 // write unlock and upgrade lock atomically
234 void unlock_and_lock_upgrade() {
235 // need to do it in two steps here -- as the UPGRADED bit might be OR-ed at
236 // the same time when other threads are trying do try_lock_upgrade().
237 bits_.fetch_or(UPGRADED, std::memory_order_acquire);
238 bits_.fetch_add(-WRITER, std::memory_order_release);
242 // Attempt to acquire writer permission. Return false if we didn't get it.
245 return bits_.compare_exchange_strong(expect, WRITER,
246 std::memory_order_acq_rel);
249 // Try to get reader permission on the lock. This can fail if we
250 // find out someone is a writer or upgrader.
251 // Setting the UPGRADED bit would allow a writer-to-be to indicate
252 // its intention to write and block any new readers while waiting
253 // for existing readers to finish and release their read locks. This
254 // helps avoid starving writers (promoted from upgraders).
255 bool try_lock_shared() {
256 // fetch_add is considerably (100%) faster than compare_exchange,
257 // so here we are optimizing for the common (lock success) case.
258 int32_t value = bits_.fetch_add(READER, std::memory_order_acquire);
259 if (UNLIKELY(value & (WRITER|UPGRADED))) {
260 bits_.fetch_add(-READER, std::memory_order_release);
266 // try to unlock upgrade and write lock atomically
267 bool try_unlock_upgrade_and_lock() {
268 int32_t expect = UPGRADED;
269 return bits_.compare_exchange_strong(expect, WRITER,
270 std::memory_order_acq_rel);
273 // try to acquire an upgradable lock.
274 bool try_lock_upgrade() {
275 int32_t value = bits_.fetch_or(UPGRADED, std::memory_order_acquire);
277 // Note: when failed, we cannot flip the UPGRADED bit back,
278 // as in this case there is either another upgrade lock or a write lock.
279 // If it's a write lock, the bit will get cleared up when that lock's done
281 return ((value & (UPGRADED | WRITER)) == 0);
284 // mainly for debugging purposes.
285 int32_t bits() const { return bits_.load(std::memory_order_acquire); }
288 class UpgradedHolder;
293 explicit ReadHolder(RWSpinLock* lock = nullptr) : lock_(lock) {
294 if (lock_) lock_->lock_shared();
297 explicit ReadHolder(RWSpinLock& lock) : lock_(&lock) {
298 lock_->lock_shared();
301 ReadHolder(ReadHolder&& other) noexcept : lock_(other.lock_) {
302 other.lock_ = nullptr;
306 explicit ReadHolder(UpgradedHolder&& upgraded) : lock_(upgraded.lock_) {
307 upgraded.lock_ = nullptr;
308 if (lock_) lock_->unlock_upgrade_and_lock_shared();
311 explicit ReadHolder(WriteHolder&& writer) : lock_(writer.lock_) {
312 writer.lock_ = nullptr;
313 if (lock_) lock_->unlock_and_lock_shared();
316 ReadHolder& operator=(ReadHolder&& other) {
318 swap(lock_, other.lock_);
322 ReadHolder(const ReadHolder& other) = delete;
323 ReadHolder& operator=(const ReadHolder& other) = delete;
325 ~ReadHolder() { if (lock_) lock_->unlock_shared(); }
327 void reset(RWSpinLock* lock = nullptr) {
328 if (lock == lock_) return;
329 if (lock_) lock_->unlock_shared();
331 if (lock_) lock_->lock_shared();
334 void swap(ReadHolder* other) {
335 std::swap(lock_, other->lock_);
339 friend class UpgradedHolder;
340 friend class WriteHolder;
344 class UpgradedHolder {
346 explicit UpgradedHolder(RWSpinLock* lock = nullptr) : lock_(lock) {
347 if (lock_) lock_->lock_upgrade();
350 explicit UpgradedHolder(RWSpinLock& lock) : lock_(&lock) {
351 lock_->lock_upgrade();
354 explicit UpgradedHolder(WriteHolder&& writer) {
355 lock_ = writer.lock_;
356 writer.lock_ = nullptr;
357 if (lock_) lock_->unlock_and_lock_upgrade();
360 UpgradedHolder(UpgradedHolder&& other) noexcept : lock_(other.lock_) {
361 other.lock_ = nullptr;
364 UpgradedHolder& operator =(UpgradedHolder&& other) {
366 swap(lock_, other.lock_);
370 UpgradedHolder(const UpgradedHolder& other) = delete;
371 UpgradedHolder& operator =(const UpgradedHolder& other) = delete;
373 ~UpgradedHolder() { if (lock_) lock_->unlock_upgrade(); }
375 void reset(RWSpinLock* lock = nullptr) {
376 if (lock == lock_) return;
377 if (lock_) lock_->unlock_upgrade();
379 if (lock_) lock_->lock_upgrade();
382 void swap(UpgradedHolder* other) {
384 swap(lock_, other->lock_);
388 friend class WriteHolder;
389 friend class ReadHolder;
395 explicit WriteHolder(RWSpinLock* lock = nullptr) : lock_(lock) {
396 if (lock_) lock_->lock();
399 explicit WriteHolder(RWSpinLock& lock) : lock_(&lock) {
403 // promoted from an upgrade lock holder
404 explicit WriteHolder(UpgradedHolder&& upgraded) {
405 lock_ = upgraded.lock_;
406 upgraded.lock_ = nullptr;
407 if (lock_) lock_->unlock_upgrade_and_lock();
410 WriteHolder(WriteHolder&& other) noexcept : lock_(other.lock_) {
411 other.lock_ = nullptr;
414 WriteHolder& operator =(WriteHolder&& other) {
416 swap(lock_, other.lock_);
420 WriteHolder(const WriteHolder& other) = delete;
421 WriteHolder& operator =(const WriteHolder& other) = delete;
423 ~WriteHolder () { if (lock_) lock_->unlock(); }
425 void reset(RWSpinLock* lock = nullptr) {
426 if (lock == lock_) return;
427 if (lock_) lock_->unlock();
429 if (lock_) lock_->lock();
432 void swap(WriteHolder* other) {
434 swap(lock_, other->lock_);
438 friend class ReadHolder;
439 friend class UpgradedHolder;
443 // Synchronized<> adaptors
444 friend void acquireRead(RWSpinLock& l) { return l.lock_shared(); }
445 friend void acquireReadWrite(RWSpinLock& l) { return l.lock(); }
446 friend void releaseRead(RWSpinLock& l) { return l.unlock_shared(); }
447 friend void releaseReadWrite(RWSpinLock& l) { return l.unlock(); }
450 std::atomic<int32_t> bits_;
454 #ifdef RW_SPINLOCK_USE_X86_INTRINSIC_
455 // A more balanced Read-Write spin lock implemented based on GCC intrinsics.
458 template <size_t kBitWidth> struct RWTicketIntTrait {
459 static_assert(kBitWidth == 32 || kBitWidth == 64,
460 "bit width has to be either 32 or 64 ");
464 struct RWTicketIntTrait<64> {
465 typedef uint64_t FullInt;
466 typedef uint32_t HalfInt;
467 typedef uint16_t QuarterInt;
469 #ifdef RW_SPINLOCK_USE_SSE_INSTRUCTIONS_
470 static __m128i make128(const uint16_t v[4]) {
471 return _mm_set_epi16(0, 0, 0, 0, v[3], v[2], v[1], v[0]);
473 static inline __m128i fromInteger(uint64_t from) {
474 return _mm_cvtsi64_si128(from);
476 static inline uint64_t toInteger(__m128i in) {
477 return _mm_cvtsi128_si64(in);
479 static inline uint64_t addParallel(__m128i in, __m128i kDelta) {
480 return toInteger(_mm_add_epi16(in, kDelta));
486 struct RWTicketIntTrait<32> {
487 typedef uint32_t FullInt;
488 typedef uint16_t HalfInt;
489 typedef uint8_t QuarterInt;
491 #ifdef RW_SPINLOCK_USE_SSE_INSTRUCTIONS_
492 static __m128i make128(const uint8_t v[4]) {
493 return _mm_set_epi8(0, 0, 0, 0, 0, 0, 0, 0,
494 0, 0, 0, 0, v[3], v[2], v[1], v[0]);
496 static inline __m128i fromInteger(uint32_t from) {
497 return _mm_cvtsi32_si128(from);
499 static inline uint32_t toInteger(__m128i in) {
500 return _mm_cvtsi128_si32(in);
502 static inline uint32_t addParallel(__m128i in, __m128i kDelta) {
503 return toInteger(_mm_add_epi8(in, kDelta));
510 template<size_t kBitWidth, bool kFavorWriter=false>
511 class RWTicketSpinLockT {
512 typedef detail::RWTicketIntTrait<kBitWidth> IntTraitType;
513 typedef typename detail::RWTicketIntTrait<kBitWidth>::FullInt FullInt;
514 typedef typename detail::RWTicketIntTrait<kBitWidth>::HalfInt HalfInt;
515 typedef typename detail::RWTicketIntTrait<kBitWidth>::QuarterInt
519 constexpr RWTicket() : whole(0) {}
522 __extension__ struct {
529 private: // Some x64-specific utilities for atomic access to ticket.
530 template<class T> static T load_acquire(T* addr) {
531 T t = *addr; // acquire barrier
532 asm_volatile_memory();
537 static void store_release(T* addr, T v) {
538 asm_volatile_memory();
539 *addr = v; // release barrier
544 constexpr RWTicketSpinLockT() {}
546 RWTicketSpinLockT(RWTicketSpinLockT const&) = delete;
547 RWTicketSpinLockT& operator=(RWTicketSpinLockT const&) = delete;
551 writeLockAggressive();
558 * Both try_lock and try_lock_shared diverge in our implementation from the
559 * lock algorithm described in the link above.
561 * In the read case, it is undesirable that the readers could wait
562 * for another reader (before increasing ticket.read in the other
563 * implementation). Our approach gives up on
564 * first-come-first-serve, but our benchmarks showed improve
565 * performance for both readers and writers under heavily contended
566 * cases, particularly when the number of threads exceeds the number
569 * We have writeLockAggressive() using the original implementation
570 * for a writer, which gives some advantage to the writer over the
571 * readers---for that path it is guaranteed that the writer will
572 * acquire the lock after all the existing readers exit.
576 FullInt old = t.whole = load_acquire(&ticket.whole);
577 if (t.users != t.write) return false;
579 return __sync_bool_compare_and_swap(&ticket.whole, old, t.whole);
583 * Call this if you want to prioritize writer to avoid starvation.
584 * Unlike writeLockNice, immediately acquires the write lock when
585 * the existing readers (arriving before the writer) finish their
588 void writeLockAggressive() {
589 // sched_yield() is needed here to avoid a pathology if the number
590 // of threads attempting concurrent writes is >= the number of real
591 // cores allocated to this process. This is less likely than the
592 // corresponding situation in lock_shared(), but we still want to
595 QuarterInt val = __sync_fetch_and_add(&ticket.users, 1);
596 while (val != load_acquire(&ticket.write)) {
597 asm_volatile_pause();
598 if (UNLIKELY(++count > 1000)) sched_yield();
602 // Call this when the writer should be nicer to the readers.
603 void writeLockNice() {
604 // Here it doesn't cpu-relax the writer.
606 // This is because usually we have many more readers than the
607 // writers, so the writer has less chance to get the lock when
608 // there are a lot of competing readers. The aggressive spinning
609 // can help to avoid starving writers.
611 // We don't worry about sched_yield() here because the caller
612 // has already explicitly abandoned fairness.
613 while (!try_lock()) {}
616 // Atomically unlock the write-lock from writer and acquire the read-lock.
617 void unlock_and_lock_shared() {
618 QuarterInt val = __sync_fetch_and_add(&ticket.read, 1);
621 // Release writer permission on the lock.
624 t.whole = load_acquire(&ticket.whole);
625 FullInt old = t.whole;
627 #ifdef RW_SPINLOCK_USE_SSE_INSTRUCTIONS_
628 // SSE2 can reduce the lock and unlock overhead by 10%
629 static const QuarterInt kDeltaBuf[4] = { 1, 1, 0, 0 }; // write/read/user
630 static const __m128i kDelta = IntTraitType::make128(kDeltaBuf);
631 __m128i m = IntTraitType::fromInteger(old);
632 t.whole = IntTraitType::addParallel(m, kDelta);
637 store_release(&ticket.readWrite, t.readWrite);
641 // sched_yield() is important here because we can't grab the
642 // shared lock if there is a pending writeLockAggressive, so we
643 // need to let threads that already have a shared lock complete
645 while (!LIKELY(try_lock_shared())) {
646 asm_volatile_pause();
647 if (UNLIKELY((++count & 1023) == 0)) sched_yield();
651 bool try_lock_shared() {
653 old.whole = t.whole = load_acquire(&ticket.whole);
654 old.users = old.read;
655 #ifdef RW_SPINLOCK_USE_SSE_INSTRUCTIONS_
656 // SSE2 may reduce the total lock and unlock overhead by 10%
657 static const QuarterInt kDeltaBuf[4] = { 0, 1, 1, 0 }; // write/read/user
658 static const __m128i kDelta = IntTraitType::make128(kDeltaBuf);
659 __m128i m = IntTraitType::fromInteger(old.whole);
660 t.whole = IntTraitType::addParallel(m, kDelta);
665 return __sync_bool_compare_and_swap(&ticket.whole, old.whole, t.whole);
668 void unlock_shared() {
669 QuarterInt val = __sync_fetch_and_add(&ticket.write, 1);
674 typedef RWTicketSpinLockT<kBitWidth, kFavorWriter> RWSpinLock;
675 class ReadHolder : boost::noncopyable {
677 explicit ReadHolder(RWSpinLock *lock = nullptr) :
679 if (lock_) lock_->lock_shared();
682 explicit ReadHolder(RWSpinLock &lock) : lock_ (&lock) {
683 if (lock_) lock_->lock_shared();
686 // atomically unlock the write-lock from writer and acquire the read-lock
687 explicit ReadHolder(WriteHolder *writer) : lock_(nullptr) {
688 std::swap(this->lock_, writer->lock_);
690 lock_->unlock_and_lock_shared();
695 if (lock_) lock_->unlock_shared();
698 void reset(RWSpinLock *lock = nullptr) {
699 if (lock_) lock_->unlock_shared();
701 if (lock_) lock_->lock_shared();
704 void swap(ReadHolder *other) {
705 std::swap(this->lock_, other->lock_);
712 class WriteHolder : boost::noncopyable {
714 explicit WriteHolder(RWSpinLock *lock = nullptr) : lock_(lock) {
715 if (lock_) lock_->lock();
717 explicit WriteHolder(RWSpinLock &lock) : lock_ (&lock) {
718 if (lock_) lock_->lock();
722 if (lock_) lock_->unlock();
725 void reset(RWSpinLock *lock = nullptr) {
726 if (lock == lock_) return;
727 if (lock_) lock_->unlock();
729 if (lock_) lock_->lock();
732 void swap(WriteHolder *other) {
733 std::swap(this->lock_, other->lock_);
737 friend class ReadHolder;
741 // Synchronized<> adaptors.
742 friend void acquireRead(RWTicketSpinLockT& mutex) {
745 friend void acquireReadWrite(RWTicketSpinLockT& mutex) {
748 friend void releaseRead(RWTicketSpinLockT& mutex) {
749 mutex.unlock_shared();
751 friend void releaseReadWrite(RWTicketSpinLockT& mutex) {
756 typedef RWTicketSpinLockT<32> RWTicketSpinLock32;
757 typedef RWTicketSpinLockT<64> RWTicketSpinLock64;
759 #endif // RW_SPINLOCK_USE_X86_INTRINSIC_
763 #ifdef RW_SPINLOCK_USE_X86_INTRINSIC_
764 #undef RW_SPINLOCK_USE_X86_INTRINSIC_
767 #endif // FOLLY_RWSPINLOCK_H_