| 1 | /* |
| 2 | * Copyright 2015-present Facebook, Inc. |
| 3 | * |
| 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 |
| 7 | * |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | * |
| 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. |
| 15 | */ |
| 16 | |
| 17 | /* |
| 18 | * N.B. You most likely do _not_ want to use PicoSpinLock or any other |
| 19 | * kind of spinlock. Consider MicroLock instead. |
| 20 | * |
| 21 | * In short, spinlocks in preemptive multi-tasking operating systems |
| 22 | * have serious problems and fast mutexes like std::mutex are almost |
| 23 | * certainly the better choice, because letting the OS scheduler put a |
| 24 | * thread to sleep is better for system responsiveness and throughput |
| 25 | * than wasting a timeslice repeatedly querying a lock held by a |
| 26 | * thread that's blocked, and you can't prevent userspace |
| 27 | * programs blocking. |
| 28 | * |
| 29 | * Spinlocks in an operating system kernel make much more sense than |
| 30 | * they do in userspace. |
| 31 | */ |
| 32 | |
| 33 | #pragma once |
| 34 | #define FOLLY_PICO_SPIN_LOCK_H_ |
| 35 | |
| 36 | /* |
| 37 | * @author Keith Adams <kma@fb.com> |
| 38 | * @author Jordan DeLong <delong.j@fb.com> |
| 39 | */ |
| 40 | |
| 41 | #include <array> |
| 42 | #include <atomic> |
| 43 | #include <cinttypes> |
| 44 | #include <cstdlib> |
| 45 | #include <mutex> |
| 46 | #include <type_traits> |
| 47 | |
| 48 | #include <glog/logging.h> |
| 49 | |
| 50 | #include <folly/Portability.h> |
| 51 | #include <folly/synchronization/AtomicUtil.h> |
| 52 | #include <folly/synchronization/detail/Sleeper.h> |
| 53 | |
| 54 | namespace folly { |
| 55 | |
| 56 | /* |
| 57 | * Spin lock on a single bit in an integral type. You can use this |
| 58 | * with 16, 32, or 64-bit integral types. |
| 59 | * |
| 60 | * This is useful if you want a small lock and already have an int |
| 61 | * with a bit in it that you aren't using. But note that it can't be |
| 62 | * as small as MicroSpinLock (1 byte), if you don't already have a |
| 63 | * convenient int with an unused bit lying around to put it on. |
| 64 | * |
| 65 | * To construct these, either use init() or zero initialize. We don't |
| 66 | * have a real constructor because we want this to be a POD type so we |
| 67 | * can put it into packed structs. |
| 68 | */ |
| 69 | template <class IntType, int Bit = sizeof(IntType) * 8 - 1> |
| 70 | struct PicoSpinLock { |
| 71 | // Internally we deal with the unsigned version of the type. |
| 72 | typedef typename std::make_unsigned<IntType>::type UIntType; |
| 73 | |
| 74 | static_assert( |
| 75 | std::is_integral<IntType>::value, |
| 76 | "PicoSpinLock needs an integral type" ); |
| 77 | static_assert( |
| 78 | sizeof(IntType) == 2 || sizeof(IntType) == 4 || sizeof(IntType) == 8, |
| 79 | "PicoSpinLock can't work on integers smaller than 2 bytes" ); |
| 80 | |
| 81 | public: |
| 82 | static const UIntType kLockBitMask_ = UIntType(1) << Bit; |
| 83 | mutable UIntType lock_; |
| 84 | |
| 85 | /* |
| 86 | * You must call this function before using this class, if you |
| 87 | * default constructed it. If you zero-initialized it you can |
| 88 | * assume the PicoSpinLock is in a valid unlocked state with |
| 89 | * getData() == 0. |
| 90 | * |
| 91 | * (This doesn't use a constructor because we want to be a POD.) |
| 92 | */ |
| 93 | void init(IntType initialValue = 0) { |
| 94 | CHECK(!(initialValue & kLockBitMask_)); |
| 95 | reinterpret_cast<std::atomic<UIntType>*>(&lock_)->store( |
| 96 | UIntType(initialValue), std::memory_order_release); |
| 97 | } |
| 98 | |
| 99 | /* |
| 100 | * Returns the value of the integer we using for our lock, except |
| 101 | * with the bit we are using as a lock cleared, regardless of |
| 102 | * whether the lock is held. |
| 103 | * |
| 104 | * It is 'safe' to call this without holding the lock. (As in: you |
| 105 | * get the same guarantees for simultaneous accesses to an integer |
| 106 | * as you normally get.) |
| 107 | */ |
| 108 | IntType getData() const { |
| 109 | auto res = reinterpret_cast<std::atomic<UIntType>*>(&lock_)->load( |
| 110 | std::memory_order_relaxed) & |
| 111 | ~kLockBitMask_; |
| 112 | return res; |
| 113 | } |
| 114 | |
| 115 | /* |
| 116 | * Set the value of the other bits in our integer. |
| 117 | * |
| 118 | * Don't use this when you aren't holding the lock, unless it can be |
| 119 | * guaranteed that no other threads may be trying to use this. |
| 120 | */ |
| 121 | void setData(IntType w) { |
| 122 | CHECK(!(w & kLockBitMask_)); |
| 123 | auto l = reinterpret_cast<std::atomic<UIntType>*>(&lock_); |
| 124 | l->store( |
| 125 | (l->load(std::memory_order_relaxed) & kLockBitMask_) | w, |
| 126 | std::memory_order_relaxed); |
| 127 | } |
| 128 | |
| 129 | /* |
| 130 | * Try to get the lock without blocking: returns whether or not we |
| 131 | * got it. |
| 132 | */ |
| 133 | bool try_lock() const { |
| 134 | auto previous = atomic_fetch_set( |
| 135 | *reinterpret_cast<std::atomic<UIntType>*>(&lock_), |
| 136 | Bit, |
| 137 | std::memory_order_acquire); |
| 138 | return !previous; |
| 139 | } |
| 140 | |
| 141 | /* |
| 142 | * Block until we can acquire the lock. Uses Sleeper to wait. |
| 143 | */ |
| 144 | void lock() const { |
| 145 | detail::Sleeper sleeper; |
| 146 | while (!try_lock()) { |
| 147 | sleeper.wait(); |
| 148 | } |
| 149 | } |
| 150 | |
| 151 | /* |
| 152 | * Release the lock, without changing the value of the rest of the |
| 153 | * integer. |
| 154 | */ |
| 155 | void unlock() const { |
| 156 | auto previous = atomic_fetch_reset( |
| 157 | *reinterpret_cast<std::atomic<UIntType>*>(&lock_), |
| 158 | Bit, |
| 159 | std::memory_order_release); |
| 160 | DCHECK(previous); |
| 161 | } |
| 162 | }; |
| 163 | |
| 164 | } // namespace folly |
| 165 | |