2 * Copyright 2016 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.
22 #include <folly/detail/Futex.h>
23 #include <folly/Portability.h>
28 * Tiny exclusive lock that packs four lock slots into a single
29 * byte. Each slot is an independent real, sleeping lock. The default
30 * lock and unlock functions operate on slot zero, which modifies only
31 * the low two bits of the host byte.
33 * You should zero-initialize the bits of a MicroLock that you intend
36 * If you're not space-constrained, prefer std::mutex, which will
37 * likely be faster, since it has more than two bits of information to
40 * You are free to put a MicroLock in a union with some other object.
41 * If, for example, you want to use the bottom two bits of a pointer
42 * as a lock, you can put a MicroLock in a union with the pointer and
43 * limit yourself to MicroLock slot zero, which will use the two
44 * least-significant bits in the bottom byte.
46 * MicroLock uses a dirty trick: it actually operates on the full
47 * word-size, word-aligned bit of memory into which it is embedded.
48 * It never modifies bits outside the ones it's defined to modify, but
49 * it _accesses_ all the bits in the word for purposes of
52 * The MaxSpins template parameter controls the number of times we
53 * spin trying to acquire the lock. MaxYields controls the number of
54 * times we call sched_yield; once we've tried to acquire the lock
55 * MaxSpins + MaxYields times, we sleep on the lock futex.
56 * By adjusting these parameters, you can make MicroLock behave as
57 * much or as little like a conventional spinlock as you'd like.
62 * With the default template options, the timings for uncontended
63 * acquire-then-release come out as follows on Intel(R) Xeon(R) CPU
64 * E5-2660 0 @ 2.20GHz, in @mode/opt, as of the master tree at Tue, 01
67 * ========================================================================
68 * folly/test/SmallLocksBenchmark.cpp relative time/iter iters/s
69 * ========================================================================
70 * MicroSpinLockUncontendedBenchmark 13.46ns 74.28M
71 * PicoSpinLockUncontendedBenchmark 14.99ns 66.71M
72 * MicroLockUncontendedBenchmark 27.06ns 36.96M
73 * StdMutexUncontendedBenchmark 25.18ns 39.72M
74 * VirtualFunctionCall 1.72ns 579.78M
75 * ========================================================================
77 * (The virtual dispatch benchmark is provided for scale.)
79 * The contended case for MicroLock is likely to be worse compared to
80 * std::mutex than the contended case is. Make sure to benchmark your
81 * particular workload.
88 inline detail::Futex<>* word() const;
89 inline uint32_t baseShift(unsigned slot) const;
90 inline uint32_t heldBit(unsigned slot) const;
91 inline uint32_t waitBit(unsigned slot) const;
92 static void lockSlowPath(uint32_t oldWord,
93 detail::Futex<>* wordPtr,
99 inline void unlock(unsigned slot);
100 inline void unlock() { unlock(0); }
101 inline void init(unsigned slot) { lock_ &= ~(3U << (2 * slot)); }
102 inline void init() { init(0); }
105 inline detail::Futex<>* MicroLockCore::word() const {
106 uintptr_t lockptr = (uintptr_t)&lock_;
107 lockptr &= ~(sizeof(uint32_t) - 1);
108 return (detail::Futex<>*)lockptr;
111 inline unsigned MicroLockCore::baseShift(unsigned slot) const {
112 assert(slot < CHAR_BIT / 2);
113 uintptr_t offset_bytes = (uintptr_t)&lock_ - (uintptr_t)word();
114 assert(offset_bytes < sizeof(uint32_t));
116 return kIsLittleEndian
117 ? (unsigned)offset_bytes * CHAR_BIT + slot * 2
118 : CHAR_BIT * (sizeof(uint32_t) - offset_bytes - 1) + slot * 2;
121 inline uint32_t MicroLockCore::heldBit(unsigned slot) const {
122 return 1U << (baseShift(slot) + 0);
125 inline uint32_t MicroLockCore::waitBit(unsigned slot) const {
126 return 1U << (baseShift(slot) + 1);
129 void MicroLockCore::unlock(unsigned slot) {
130 detail::Futex<>* wordPtr = word();
134 oldWord = wordPtr->load(std::memory_order_relaxed);
136 assert(oldWord & heldBit(slot));
137 newWord = oldWord & ~(heldBit(slot) | waitBit(slot));
138 } while (!wordPtr->compare_exchange_weak(
139 oldWord, newWord, std::memory_order_release, std::memory_order_relaxed));
141 if (oldWord & waitBit(slot)) {
142 // We don't track the number of waiters, so wake everyone
143 (void)wordPtr->futexWake(std::numeric_limits<int>::max(), heldBit(slot));
147 template <unsigned MaxSpins = 1000, unsigned MaxYields = 0>
148 class MicroLockBase : public MicroLockCore {
150 inline void lock(unsigned slot);
151 inline void lock() { lock(0); }
152 inline bool try_lock(unsigned slot);
153 inline bool try_lock() { return try_lock(0); }
156 template <unsigned MaxSpins, unsigned MaxYields>
157 bool MicroLockBase<MaxSpins, MaxYields>::try_lock(unsigned slot) {
159 // N.B. You might think that try_lock is just the fast path of lock,
160 // but you'd be wrong. Keep in mind that other parts of our host
161 // word might be changing while we take the lock! We're not allowed
162 // to fail spuriously if the lock is in fact not held, even if other
163 // people are concurrently modifying other parts of the word.
165 // We need to loop until we either see firm evidence that somebody
166 // else has the lock (by looking at heldBit) or see our CAS succeed.
167 // A failed CAS by itself does not indicate lock-acquire failure.
169 detail::Futex<>* wordPtr = word();
170 uint32_t oldWord = wordPtr->load(std::memory_order_relaxed);
172 if (oldWord & heldBit(slot)) {
175 } while (!wordPtr->compare_exchange_weak(oldWord,
176 oldWord | heldBit(slot),
177 std::memory_order_acquire,
178 std::memory_order_relaxed));
183 template <unsigned MaxSpins, unsigned MaxYields>
184 void MicroLockBase<MaxSpins, MaxYields>::lock(unsigned slot) {
186 static_assert(MaxSpins + MaxYields < (unsigned)-1, "overflow");
188 detail::Futex<>* wordPtr = word();
190 oldWord = wordPtr->load(std::memory_order_relaxed);
191 if ((oldWord & heldBit(slot)) == 0 &&
192 wordPtr->compare_exchange_weak(oldWord,
193 oldWord | heldBit(slot),
194 std::memory_order_acquire,
195 std::memory_order_relaxed)) {
196 // Fast uncontended case: seq_cst above is our memory barrier
198 // lockSlowPath doesn't have any slot-dependent computation; it
199 // just shifts the input bit. Make sure its shifting produces the
200 // same result a call to waitBit for our slot would.
201 assert(heldBit(slot) << 1 == waitBit(slot));
202 // lockSlowPath emits its own memory barrier
203 lockSlowPath(oldWord, wordPtr, heldBit(slot), MaxSpins, MaxYields);
207 typedef MicroLockBase<> MicroLock;