2 * Copyright 2014 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.
17 #ifndef FOLLY_SMALLLOCKS_H_
18 #define FOLLY_SMALLLOCKS_H_
21 * This header defines a few very small mutex types. These are useful
22 * in highly memory-constrained environments where contention is
25 * Note: these locks are for use when you aren't likely to contend on
26 * the critical section, or when the critical section is incredibly
27 * small. Given that, both of the locks defined in this header are
28 * inherently unfair: that is, the longer a thread is waiting, the
29 * longer it waits between attempts to acquire, so newer waiters are
30 * more likely to get the mutex. For the intended use-case this is
33 * @author Keith Adams <kma@fb.com>
34 * @author Jordan DeLong <delong.j@fb.com>
39 #include <type_traits>
41 #include <boost/noncopyable.hpp>
46 #include <glog/logging.h>
49 # error "SmallLocks.h is currently x64-only."
52 #include "folly/Portability.h"
56 //////////////////////////////////////////////////////////////////////
61 * A helper object for the condended case. Starts off with eager
62 * spinning, and falls back to sleeping for small quantums.
65 static const uint32_t kMaxActiveSpin = 4000;
70 Sleeper() : spinCount(0) {}
73 if (spinCount < kMaxActiveSpin) {
75 asm volatile("pause");
78 * Always sleep 0.5ms, assuming this will make the kernel put
79 * us down for whatever its minimum timer resolution is (in
80 * linux this varies by kernel version from 1ms to 10ms).
82 struct timespec ts = { 0, 500000 };
90 //////////////////////////////////////////////////////////////////////
93 * A really, *really* small spinlock for fine-grained locking of lots
96 * Zero initializing these is guaranteed to be as good as calling
97 * init(), since the free state is guaranteed to be all-bits zero.
99 * This class should be kept a POD, so we can used it in other packed
100 * structs (gcc does not allow __attribute__((packed)) on structs that
101 * contain non-POD data). This means avoid adding a constructor, or
102 * making some members private, etc.
104 struct MicroSpinLock {
105 enum { FREE = 0, LOCKED = 1 };
109 * Atomically move lock_ from "compare" to "newval". Return boolean
110 * success. Do not play on or around.
112 bool cas(uint8_t compare, uint8_t newVal) {
114 bool memVal; // only set if the cmpxchg fails
115 asm volatile("lock; cmpxchgb %[newVal], (%[lockPtr]);"
117 : [output] "=r" (out), "=a" (memVal)
118 : "a" (compare), // cmpxchgb constrains this to be in %al
119 [newVal] "q" (newVal), // Needs to be byte-accessible
120 [lockPtr] "r" (&lock_)
121 : "memory", "flags");
125 // Initialize this MSL. It is unnecessary to call this if you
126 // zero-initialize the MicroSpinLock.
132 return cas(FREE, LOCKED);
136 detail::Sleeper sleeper;
138 while (lock_ != FREE) {
139 asm volatile("" : : : "memory");
142 } while (!try_lock());
143 DCHECK(lock_ == LOCKED);
147 CHECK(lock_ == LOCKED);
148 asm volatile("" : : : "memory");
149 lock_ = FREE; // release barrier on x86
153 //////////////////////////////////////////////////////////////////////
156 * Spin lock on a single bit in an integral type. You can use this
157 * with 16, 32, or 64-bit integral types.
159 * This is useful if you want a small lock and already have an int
160 * with a bit in it that you aren't using. But note that it can't be
161 * as small as MicroSpinLock (1 byte), if you don't already have a
162 * convenient int with an unused bit lying around to put it on.
164 * To construct these, either use init() or zero initialize. We don't
165 * have a real constructor because we want this to be a POD type so we
166 * can put it into packed structs.
168 template<class IntType, int Bit = sizeof(IntType) * 8 - 1>
169 struct PicoSpinLock {
170 // Internally we deal with the unsigned version of the type.
171 typedef typename std::make_unsigned<IntType>::type UIntType;
173 static_assert(std::is_integral<IntType>::value,
174 "PicoSpinLock needs an integral type");
175 static_assert(sizeof(IntType) == 2 || sizeof(IntType) == 4 ||
176 sizeof(IntType) == 8,
177 "PicoSpinLock can't work on integers smaller than 2 bytes");
180 static const UIntType kLockBitMask_ = UIntType(1) << Bit;
184 * You must call this function before using this class, if you
185 * default constructed it. If you zero-initialized it you can
186 * assume the PicoSpinLock is in a valid unlocked state with
189 * (This doesn't use a constructor because we want to be a POD.)
191 void init(IntType initialValue = 0) {
192 CHECK(!(initialValue & kLockBitMask_));
193 lock_ = initialValue;
197 * Returns the value of the integer we using for our lock, except
198 * with the bit we are using as a lock cleared, regardless of
199 * whether the lock is held.
201 * It is 'safe' to call this without holding the lock. (As in: you
202 * get the same guarantees for simultaneous accesses to an integer
203 * as you normally get.)
205 IntType getData() const {
206 return static_cast<IntType>(lock_ & ~kLockBitMask_);
210 * Set the value of the other bits in our integer.
212 * Don't use this when you aren't holding the lock, unless it can be
213 * guaranteed that no other threads may be trying to use this.
215 void setData(IntType w) {
216 CHECK(!(w & kLockBitMask_));
217 lock_ = (lock_ & kLockBitMask_) | w;
221 * Try to get the lock without blocking: returns whether or not we
224 bool try_lock() const {
227 #define FB_DOBTS(size) \
228 asm volatile("lock; bts" #size " %1, (%2); setnc %0" \
234 switch (sizeof(IntType)) {
235 case 2: FB_DOBTS(w); break;
236 case 4: FB_DOBTS(l); break;
237 case 8: FB_DOBTS(q); break;
246 * Block until we can acquire the lock. Uses Sleeper to wait.
249 detail::Sleeper sleeper;
250 while (!try_lock()) {
256 * Release the lock, without changing the value of the rest of the
259 void unlock() const {
260 #define FB_DOBTR(size) \
261 asm volatile("lock; btr" #size " %0, (%1)" \
268 // Reads and writes can not be reordered wrt locked instructions,
269 // so we don't need a memory fence here.
270 switch (sizeof(IntType)) {
271 case 2: FB_DOBTR(w); break;
272 case 4: FB_DOBTR(l); break;
273 case 8: FB_DOBTR(q); break;
280 //////////////////////////////////////////////////////////////////////
283 * Array of spinlocks where each one is padded to prevent false sharing.
284 * Useful for shard-based locking implementations in environments where
285 * contention is unlikely.
288 // TODO: generate it from configure (`getconf LEVEL1_DCACHE_LINESIZE`)
289 #define FOLLY_CACHE_LINE_SIZE 64
291 template <class T, size_t N>
292 struct SpinLockArray {
293 T& operator[](size_t i) {
294 return data_[i].lock;
297 const T& operator[](size_t i) const {
298 return data_[i].lock;
301 constexpr size_t size() const { return N; }
304 struct PaddedSpinLock {
305 PaddedSpinLock() : lock() { }
307 char padding[FOLLY_CACHE_LINE_SIZE - sizeof(T)];
309 static_assert(sizeof(PaddedSpinLock) == FOLLY_CACHE_LINE_SIZE,
310 "Invalid size of PaddedSpinLock");
312 // Check if T can theoretically cross a cache line.
313 // NOTE: It should be alignof(std::max_align_t), but max_align_t
314 // isn't supported by gcc 4.6.2.
315 static_assert(alignof(MaxAlign) > 0 &&
316 FOLLY_CACHE_LINE_SIZE % alignof(MaxAlign) == 0 &&
317 sizeof(T) <= alignof(MaxAlign),
318 "T can cross cache line boundaries");
320 char padding_[FOLLY_CACHE_LINE_SIZE];
321 std::array<PaddedSpinLock, N> data_;
322 } __attribute__((aligned));
324 //////////////////////////////////////////////////////////////////////
326 typedef std::lock_guard<MicroSpinLock> MSLGuard;
328 //////////////////////////////////////////////////////////////////////