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// @author Nathan Bronson (ngbronson@fb.com)
18
19#pragma once
20
21#include <stdint.h>
22
23#include <atomic>
24#include <thread>
25#include <type_traits>
26
27#include <folly/CPortability.h>
28#include <folly/Likely.h>
29#include <folly/concurrency/CacheLocality.h>
30#include <folly/detail/Futex.h>
31#include <folly/portability/Asm.h>
32#include <folly/portability/SysResource.h>
33#include <folly/synchronization/SanitizeThread.h>
34
35// SharedMutex is a reader-writer lock. It is small, very fast, scalable
36// on multi-core, and suitable for use when readers or writers may block.
37// Unlike most other reader-writer locks, its throughput with concurrent
38// readers scales linearly; it is able to acquire and release the lock
39// in shared mode without cache line ping-ponging. It is suitable for
40// a wide range of lock hold times because it starts with spinning,
41// proceeds to using sched_yield with a preemption heuristic, and then
42// waits using futex and precise wakeups.
43//
44// SharedMutex provides all of the methods of folly::RWSpinLock,
45// boost::shared_mutex, boost::upgrade_mutex, and C++14's
46// std::shared_timed_mutex. All operations that can block are available
47// in try, try-for, and try-until (system_clock or steady_clock) versions.
48//
49// SharedMutexReadPriority gives priority to readers,
50// SharedMutexWritePriority gives priority to writers. SharedMutex is an
51// alias for SharedMutexWritePriority, because writer starvation is more
52// likely than reader starvation for the read-heavy workloads targetted
53// by SharedMutex.
54//
55// In my tests SharedMutex is as good or better than the other
56// reader-writer locks in use at Facebook for almost all use cases,
57// sometimes by a wide margin. (If it is rare that there are actually
58// concurrent readers then RWSpinLock can be a few nanoseconds faster.)
59// I compared it to folly::RWSpinLock, folly::RWTicketSpinLock64,
60// boost::shared_mutex, pthread_rwlock_t, and a RWLock that internally uses
61// spinlocks to guard state and pthread_mutex_t+pthread_cond_t to block.
62// (Thrift's ReadWriteMutex is based underneath on pthread_rwlock_t.)
63// It is generally as good or better than the rest when evaluating size,
64// speed, scalability, or latency outliers. In the corner cases where
65// it is not the fastest (such as single-threaded use or heavy write
66// contention) it is never very much worse than the best. See the bottom
67// of folly/test/SharedMutexTest.cpp for lots of microbenchmark results.
68//
69// Comparison to folly::RWSpinLock:
70//
71// * SharedMutex is faster than RWSpinLock when there are actually
72// concurrent read accesses (sometimes much faster), and ~5 nanoseconds
73// slower when there is not actually any contention. SharedMutex is
74// faster in every (benchmarked) scenario where the shared mode of
75// the lock is actually useful.
76//
77// * Concurrent shared access to SharedMutex scales linearly, while total
78// RWSpinLock throughput drops as more threads try to access the lock
79// in shared mode. Under very heavy read contention SharedMutex can
80// be two orders of magnitude faster than RWSpinLock (or any reader
81// writer lock that doesn't use striping or deferral).
82//
83// * SharedMutex can safely protect blocking calls, because after an
84// initial period of spinning it waits using futex().
85//
86// * RWSpinLock prioritizes readers, SharedMutex has both reader- and
87// writer-priority variants, but defaults to write priority.
88//
89// * RWSpinLock's upgradeable mode blocks new readers, while SharedMutex's
90// doesn't. Both semantics are reasonable. The boost documentation
91// doesn't explicitly talk about this behavior (except by omitting
92// any statement that those lock modes conflict), but the boost
93// implementations do allow new readers while the upgradeable mode
94// is held. See https://github.com/boostorg/thread/blob/master/
95// include/boost/thread/pthread/shared_mutex.hpp
96//
97// * RWSpinLock::UpgradedHolder maps to SharedMutex::UpgradeHolder
98// (UpgradeableHolder would be even more pedantically correct).
99// SharedMutex's holders have fewer methods (no reset) and are less
100// tolerant (promotion and downgrade crash if the donor doesn't own
101// the lock, and you must use the default constructor rather than
102// passing a nullptr to the pointer constructor).
103//
104// Both SharedMutex and RWSpinLock provide "exclusive", "upgrade",
105// and "shared" modes. At all times num_threads_holding_exclusive +
106// num_threads_holding_upgrade <= 1, and num_threads_holding_exclusive ==
107// 0 || num_threads_holding_shared == 0. RWSpinLock has the additional
108// constraint that num_threads_holding_shared cannot increase while
109// num_threads_holding_upgrade is non-zero.
110//
111// Comparison to the internal RWLock:
112//
113// * SharedMutex doesn't allow a maximum reader count to be configured,
114// so it can't be used as a semaphore in the same way as RWLock.
115//
116// * SharedMutex is 4 bytes, RWLock is 256.
117//
118// * SharedMutex is as fast or faster than RWLock in all of my
119// microbenchmarks, and has positive rather than negative scalability.
120//
121// * RWLock and SharedMutex are both writer priority locks.
122//
123// * SharedMutex avoids latency outliers as well as RWLock.
124//
125// * SharedMutex uses different names (t != 0 below):
126//
127// RWLock::lock(0) => SharedMutex::lock()
128//
129// RWLock::lock(t) => SharedMutex::try_lock_for(milliseconds(t))
130//
131// RWLock::tryLock() => SharedMutex::try_lock()
132//
133// RWLock::unlock() => SharedMutex::unlock()
134//
135// RWLock::enter(0) => SharedMutex::lock_shared()
136//
137// RWLock::enter(t) =>
138// SharedMutex::try_lock_shared_for(milliseconds(t))
139//
140// RWLock::tryEnter() => SharedMutex::try_lock_shared()
141//
142// RWLock::leave() => SharedMutex::unlock_shared()
143//
144// * RWLock allows the reader count to be adjusted by a value other
145// than 1 during enter() or leave(). SharedMutex doesn't currently
146// implement this feature.
147//
148// * RWLock's methods are marked const, SharedMutex's aren't.
149//
150// Reader-writer locks have the potential to allow concurrent access
151// to shared read-mostly data, but in practice they often provide no
152// improvement over a mutex. The problem is the cache coherence protocol
153// of modern CPUs. Coherence is provided by making sure that when a cache
154// line is written it is present in only one core's cache. Since a memory
155// write is required to acquire a reader-writer lock in shared mode, the
156// cache line holding the lock is invalidated in all of the other caches.
157// This leads to cache misses when another thread wants to acquire or
158// release the lock concurrently. When the RWLock is colocated with the
159// data it protects (common), cache misses can also continue occur when
160// a thread that already holds the lock tries to read the protected data.
161//
162// Ideally, a reader-writer lock would allow multiple cores to acquire
163// and release the lock in shared mode without incurring any cache misses.
164// This requires that each core records its shared access in a cache line
165// that isn't read or written by other read-locking cores. (Writers will
166// have to check all of the cache lines.) Typical server hardware when
167// this comment was written has 16 L1 caches and cache lines of 64 bytes,
168// so a lock striped over all L1 caches would occupy a prohibitive 1024
169// bytes. Nothing says that we need a separate set of per-core memory
170// locations for each lock, however. Each SharedMutex instance is only
171// 4 bytes, but all locks together share a 2K area in which they make a
172// core-local record of lock acquisitions.
173//
174// SharedMutex's strategy of using a shared set of core-local stripes has
175// a potential downside, because it means that acquisition of any lock in
176// write mode can conflict with acquisition of any lock in shared mode.
177// If a lock instance doesn't actually experience concurrency then this
178// downside will outweight the upside of improved scalability for readers.
179// To avoid this problem we dynamically detect concurrent accesses to
180// SharedMutex, and don't start using the deferred mode unless we actually
181// observe concurrency. See kNumSharedToStartDeferring.
182//
183// It is explicitly allowed to call unlock_shared() from a different
184// thread than lock_shared(), so long as they are properly paired.
185// unlock_shared() needs to find the location at which lock_shared()
186// recorded the lock, which might be in the lock itself or in any of
187// the shared slots. If you can conveniently pass state from lock
188// acquisition to release then the fastest mechanism is to std::move
189// the SharedMutex::ReadHolder instance or an SharedMutex::Token (using
190// lock_shared(Token&) and unlock_shared(Token&)). The guard or token
191// will tell unlock_shared where in deferredReaders[] to look for the
192// deferred lock. The Token-less version of unlock_shared() works in all
193// cases, but is optimized for the common (no inter-thread handoff) case.
194//
195// In both read- and write-priority mode, a waiting lock() (exclusive mode)
196// only blocks readers after it has waited for an active upgrade lock to be
197// released; until the upgrade lock is released (or upgraded or downgraded)
198// readers will still be able to enter. Preferences about lock acquisition
199// are not guaranteed to be enforced perfectly (even if they were, there
200// is theoretically the chance that a thread could be arbitrarily suspended
201// between calling lock() and SharedMutex code actually getting executed).
202//
203// try_*_for methods always try at least once, even if the duration
204// is zero or negative. The duration type must be compatible with
205// std::chrono::steady_clock. try_*_until methods also always try at
206// least once. std::chrono::system_clock and std::chrono::steady_clock
207// are supported.
208//
209// If you have observed by profiling that your SharedMutex-s are getting
210// cache misses on deferredReaders[] due to another SharedMutex user, then
211// you can use the tag type to create your own instantiation of the type.
212// The contention threshold (see kNumSharedToStartDeferring) should make
213// this unnecessary in all but the most extreme cases. Make sure to check
214// that the increased icache and dcache footprint of the tagged result is
215// worth it.
216
217// SharedMutex's use of thread local storage is an optimization, so
218// for the case where thread local storage is not supported, define it
219// away.
220
221// Note about TSAN (ThreadSanitizer): the SharedMutexWritePriority version
222// (the default) of this mutex is annotated appropriately so that TSAN can
223// perform lock inversion analysis. However, the SharedMutexReadPriority version
224// is not annotated. This is because TSAN's lock order heuristic
225// assumes that two calls to lock_shared must be ordered, which leads
226// to too many false positives for the reader-priority case.
227//
228// Suppose thread A holds a SharedMutexWritePriority lock in shared mode and an
229// independent thread B is waiting for exclusive access. Then a thread C's
230// lock_shared can't proceed until A has released the lock. Discounting
231// situations that never use exclusive mode (so no lock is necessary at all)
232// this means that without higher-level reasoning it is not safe to ignore
233// reader <-> reader interactions.
234//
235// This reasoning does not apply to SharedMutexReadPriority, because there are
236// no actions by a thread B that can make C need to wait for A. Since the
237// overwhelming majority of SharedMutex instances use write priority, we
238// restrict the TSAN annotations to only SharedMutexWritePriority.
239
240#ifndef FOLLY_SHAREDMUTEX_TLS
241#if !FOLLY_MOBILE
242#define FOLLY_SHAREDMUTEX_TLS FOLLY_TLS
243#else
244#define FOLLY_SHAREDMUTEX_TLS
245#endif
246#endif
247
248namespace folly {
249
250struct SharedMutexToken {
251 enum class Type : uint16_t {
252 INVALID = 0,
253 INLINE_SHARED,
254 DEFERRED_SHARED,
255 };
256
257 Type type_;
258 uint16_t slot_;
259};
260
261namespace detail {
262// Returns a guard that gives permission for the current thread to
263// annotate, and adjust the annotation bits in, the SharedMutex at ptr.
264std::unique_lock<std::mutex> sharedMutexAnnotationGuard(void* ptr);
265} // namespace detail
266
267template <
268 bool ReaderPriority,
269 typename Tag_ = void,
270 template <typename> class Atom = std::atomic,
271 bool BlockImmediately = false,
272 bool AnnotateForThreadSanitizer = kIsSanitizeThread && !ReaderPriority>
273class SharedMutexImpl {
274 public:
275 static constexpr bool kReaderPriority = ReaderPriority;
276
277 typedef Tag_ Tag;
278
279 typedef SharedMutexToken Token;
280
281 class ReadHolder;
282 class UpgradeHolder;
283 class WriteHolder;
284
285 constexpr SharedMutexImpl() noexcept : state_(0) {}
286
287 SharedMutexImpl(const SharedMutexImpl&) = delete;
288 SharedMutexImpl(SharedMutexImpl&&) = delete;
289 SharedMutexImpl& operator=(const SharedMutexImpl&) = delete;
290 SharedMutexImpl& operator=(SharedMutexImpl&&) = delete;
291
292 // It is an error to destroy an SharedMutex that still has
293 // any outstanding locks. This is checked if NDEBUG isn't defined.
294 // SharedMutex's exclusive mode can be safely used to guard the lock's
295 // own destruction. If, for example, you acquire the lock in exclusive
296 // mode and then observe that the object containing the lock is no longer
297 // needed, you can unlock() and then immediately destroy the lock.
298 // See https://sourceware.org/bugzilla/show_bug.cgi?id=13690 for a
299 // description about why this property needs to be explicitly mentioned.
300 ~SharedMutexImpl() {
301 auto state = state_.load(std::memory_order_relaxed);
302 if (UNLIKELY((state & kHasS) != 0)) {
303 cleanupTokenlessSharedDeferred(state);
304 }
305
306#ifndef NDEBUG
307 // These asserts check that everybody has released the lock before it
308 // is destroyed. If you arrive here while debugging that is likely
309 // the problem. (You could also have general heap corruption.)
310
311 // if a futexWait fails to go to sleep because the value has been
312 // changed, we don't necessarily clean up the wait bits, so it is
313 // possible they will be set here in a correct system
314 assert((state & ~(kWaitingAny | kMayDefer | kAnnotationCreated)) == 0);
315 if ((state & kMayDefer) != 0) {
316 for (uint32_t slot = 0; slot < kMaxDeferredReaders; ++slot) {
317 auto slotValue = deferredReader(slot)->load(std::memory_order_relaxed);
318 assert(!slotValueIsThis(slotValue));
319 }
320 }
321#endif
322 annotateDestroy();
323 }
324
325 void lock() {
326 WaitForever ctx;
327 (void)lockExclusiveImpl(kHasSolo, ctx);
328 annotateAcquired(annotate_rwlock_level::wrlock);
329 }
330
331 bool try_lock() {
332 WaitNever ctx;
333 auto result = lockExclusiveImpl(kHasSolo, ctx);
334 annotateTryAcquired(result, annotate_rwlock_level::wrlock);
335 return result;
336 }
337
338 template <class Rep, class Period>
339 bool try_lock_for(const std::chrono::duration<Rep, Period>& duration) {
340 WaitForDuration<Rep, Period> ctx(duration);
341 auto result = lockExclusiveImpl(kHasSolo, ctx);
342 annotateTryAcquired(result, annotate_rwlock_level::wrlock);
343 return result;
344 }
345
346 template <class Clock, class Duration>
347 bool try_lock_until(
348 const std::chrono::time_point<Clock, Duration>& absDeadline) {
349 WaitUntilDeadline<Clock, Duration> ctx{absDeadline};
350 auto result = lockExclusiveImpl(kHasSolo, ctx);
351 annotateTryAcquired(result, annotate_rwlock_level::wrlock);
352 return result;
353 }
354
355 void unlock() {
356 annotateReleased(annotate_rwlock_level::wrlock);
357 // It is possible that we have a left-over kWaitingNotS if the last
358 // unlock_shared() that let our matching lock() complete finished
359 // releasing before lock()'s futexWait went to sleep. Clean it up now
360 auto state = (state_ &= ~(kWaitingNotS | kPrevDefer | kHasE));
361 assert((state & ~(kWaitingAny | kAnnotationCreated)) == 0);
362 wakeRegisteredWaiters(state, kWaitingE | kWaitingU | kWaitingS);
363 }
364
365 // Managing the token yourself makes unlock_shared a bit faster
366
367 void lock_shared() {
368 WaitForever ctx;
369 (void)lockSharedImpl(nullptr, ctx);
370 annotateAcquired(annotate_rwlock_level::rdlock);
371 }
372
373 void lock_shared(Token& token) {
374 WaitForever ctx;
375 (void)lockSharedImpl(&token, ctx);
376 annotateAcquired(annotate_rwlock_level::rdlock);
377 }
378
379 bool try_lock_shared() {
380 WaitNever ctx;
381 auto result = lockSharedImpl(nullptr, ctx);
382 annotateTryAcquired(result, annotate_rwlock_level::rdlock);
383 return result;
384 }
385
386 bool try_lock_shared(Token& token) {
387 WaitNever ctx;
388 auto result = lockSharedImpl(&token, ctx);
389 annotateTryAcquired(result, annotate_rwlock_level::rdlock);
390 return result;
391 }
392
393 template <class Rep, class Period>
394 bool try_lock_shared_for(const std::chrono::duration<Rep, Period>& duration) {
395 WaitForDuration<Rep, Period> ctx(duration);
396 auto result = lockSharedImpl(nullptr, ctx);
397 annotateTryAcquired(result, annotate_rwlock_level::rdlock);
398 return result;
399 }
400
401 template <class Rep, class Period>
402 bool try_lock_shared_for(
403 const std::chrono::duration<Rep, Period>& duration,
404 Token& token) {
405 WaitForDuration<Rep, Period> ctx(duration);
406 auto result = lockSharedImpl(&token, ctx);
407 annotateTryAcquired(result, annotate_rwlock_level::rdlock);
408 return result;
409 }
410
411 template <class Clock, class Duration>
412 bool try_lock_shared_until(
413 const std::chrono::time_point<Clock, Duration>& absDeadline) {
414 WaitUntilDeadline<Clock, Duration> ctx{absDeadline};
415 auto result = lockSharedImpl(nullptr, ctx);
416 annotateTryAcquired(result, annotate_rwlock_level::rdlock);
417 return result;
418 }
419
420 template <class Clock, class Duration>
421 bool try_lock_shared_until(
422 const std::chrono::time_point<Clock, Duration>& absDeadline,
423 Token& token) {
424 WaitUntilDeadline<Clock, Duration> ctx{absDeadline};
425 auto result = lockSharedImpl(&token, ctx);
426 annotateTryAcquired(result, annotate_rwlock_level::rdlock);
427 return result;
428 }
429
430 void unlock_shared() {
431 annotateReleased(annotate_rwlock_level::rdlock);
432
433 auto state = state_.load(std::memory_order_acquire);
434
435 // kPrevDefer can only be set if HasE or BegunE is set
436 assert((state & (kPrevDefer | kHasE | kBegunE)) != kPrevDefer);
437
438 // lock() strips kMayDefer immediately, but then copies it to
439 // kPrevDefer so we can tell if the pre-lock() lock_shared() might
440 // have deferred
441 if ((state & (kMayDefer | kPrevDefer)) == 0 ||
442 !tryUnlockTokenlessSharedDeferred()) {
443 // Matching lock_shared() couldn't have deferred, or the deferred
444 // lock has already been inlined by applyDeferredReaders()
445 unlockSharedInline();
446 }
447 }
448
449 void unlock_shared(Token& token) {
450 annotateReleased(annotate_rwlock_level::rdlock);
451
452 assert(
453 token.type_ == Token::Type::INLINE_SHARED ||
454 token.type_ == Token::Type::DEFERRED_SHARED);
455
456 if (token.type_ != Token::Type::DEFERRED_SHARED ||
457 !tryUnlockSharedDeferred(token.slot_)) {
458 unlockSharedInline();
459 }
460#ifndef NDEBUG
461 token.type_ = Token::Type::INVALID;
462#endif
463 }
464
465 void unlock_and_lock_shared() {
466 annotateReleased(annotate_rwlock_level::wrlock);
467 annotateAcquired(annotate_rwlock_level::rdlock);
468 // We can't use state_ -=, because we need to clear 2 bits (1 of which
469 // has an uncertain initial state) and set 1 other. We might as well
470 // clear the relevant wake bits at the same time. Note that since S
471 // doesn't block the beginning of a transition to E (writer priority
472 // can cut off new S, reader priority grabs BegunE and blocks deferred
473 // S) we need to wake E as well.
474 auto state = state_.load(std::memory_order_acquire);
475 do {
476 assert(
477 (state & ~(kWaitingAny | kPrevDefer | kAnnotationCreated)) == kHasE);
478 } while (!state_.compare_exchange_strong(
479 state, (state & ~(kWaitingAny | kPrevDefer | kHasE)) + kIncrHasS));
480 if ((state & (kWaitingE | kWaitingU | kWaitingS)) != 0) {
481 futexWakeAll(kWaitingE | kWaitingU | kWaitingS);
482 }
483 }
484
485 void unlock_and_lock_shared(Token& token) {
486 unlock_and_lock_shared();
487 token.type_ = Token::Type::INLINE_SHARED;
488 }
489
490 void lock_upgrade() {
491 WaitForever ctx;
492 (void)lockUpgradeImpl(ctx);
493 // For TSAN: treat upgrade locks as equivalent to read locks
494 annotateAcquired(annotate_rwlock_level::rdlock);
495 }
496
497 bool try_lock_upgrade() {
498 WaitNever ctx;
499 auto result = lockUpgradeImpl(ctx);
500 annotateTryAcquired(result, annotate_rwlock_level::rdlock);
501 return result;
502 }
503
504 template <class Rep, class Period>
505 bool try_lock_upgrade_for(
506 const std::chrono::duration<Rep, Period>& duration) {
507 WaitForDuration<Rep, Period> ctx(duration);
508 auto result = lockUpgradeImpl(ctx);
509 annotateTryAcquired(result, annotate_rwlock_level::rdlock);
510 return result;
511 }
512
513 template <class Clock, class Duration>
514 bool try_lock_upgrade_until(
515 const std::chrono::time_point<Clock, Duration>& absDeadline) {
516 WaitUntilDeadline<Clock, Duration> ctx{absDeadline};
517 auto result = lockUpgradeImpl(ctx);
518 annotateTryAcquired(result, annotate_rwlock_level::rdlock);
519 return result;
520 }
521
522 void unlock_upgrade() {
523 annotateReleased(annotate_rwlock_level::rdlock);
524 auto state = (state_ -= kHasU);
525 assert((state & (kWaitingNotS | kHasSolo)) == 0);
526 wakeRegisteredWaiters(state, kWaitingE | kWaitingU);
527 }
528
529 void unlock_upgrade_and_lock() {
530 // no waiting necessary, so waitMask is empty
531 WaitForever ctx;
532 (void)lockExclusiveImpl(0, ctx);
533 annotateReleased(annotate_rwlock_level::rdlock);
534 annotateAcquired(annotate_rwlock_level::wrlock);
535 }
536
537 void unlock_upgrade_and_lock_shared() {
538 // No need to annotate for TSAN here because we model upgrade and shared
539 // locks as the same.
540 auto state = (state_ -= kHasU - kIncrHasS);
541 assert((state & (kWaitingNotS | kHasSolo)) == 0);
542 wakeRegisteredWaiters(state, kWaitingE | kWaitingU);
543 }
544
545 void unlock_upgrade_and_lock_shared(Token& token) {
546 unlock_upgrade_and_lock_shared();
547 token.type_ = Token::Type::INLINE_SHARED;
548 }
549
550 void unlock_and_lock_upgrade() {
551 annotateReleased(annotate_rwlock_level::wrlock);
552 annotateAcquired(annotate_rwlock_level::rdlock);
553 // We can't use state_ -=, because we need to clear 2 bits (1 of
554 // which has an uncertain initial state) and set 1 other. We might
555 // as well clear the relevant wake bits at the same time.
556 auto state = state_.load(std::memory_order_acquire);
557 while (true) {
558 assert(
559 (state & ~(kWaitingAny | kPrevDefer | kAnnotationCreated)) == kHasE);
560 auto after =
561 (state & ~(kWaitingNotS | kWaitingS | kPrevDefer | kHasE)) + kHasU;
562 if (state_.compare_exchange_strong(state, after)) {
563 if ((state & kWaitingS) != 0) {
564 futexWakeAll(kWaitingS);
565 }
566 return;
567 }
568 }
569 }
570
571 private:
572 typedef typename folly::detail::Futex<Atom> Futex;
573
574 // Internally we use four kinds of wait contexts. These are structs
575 // that provide a doWait method that returns true if a futex wake
576 // was issued that intersects with the waitMask, false if there was a
577 // timeout and no more waiting should be performed. Spinning occurs
578 // before the wait context is invoked.
579
580 struct WaitForever {
581 bool canBlock() {
582 return true;
583 }
584 bool canTimeOut() {
585 return false;
586 }
587 bool shouldTimeOut() {
588 return false;
589 }
590
591 bool doWait(Futex& futex, uint32_t expected, uint32_t waitMask) {
592 detail::futexWait(&futex, expected, waitMask);
593 return true;
594 }
595 };
596
597 struct WaitNever {
598 bool canBlock() {
599 return false;
600 }
601 bool canTimeOut() {
602 return true;
603 }
604 bool shouldTimeOut() {
605 return true;
606 }
607
608 bool doWait(
609 Futex& /* futex */,
610 uint32_t /* expected */,
611 uint32_t /* waitMask */) {
612 return false;
613 }
614 };
615
616 template <class Rep, class Period>
617 struct WaitForDuration {
618 std::chrono::duration<Rep, Period> duration_;
619 bool deadlineComputed_;
620 std::chrono::steady_clock::time_point deadline_;
621
622 explicit WaitForDuration(const std::chrono::duration<Rep, Period>& duration)
623 : duration_(duration), deadlineComputed_(false) {}
624
625 std::chrono::steady_clock::time_point deadline() {
626 if (!deadlineComputed_) {
627 deadline_ = std::chrono::steady_clock::now() + duration_;
628 deadlineComputed_ = true;
629 }
630 return deadline_;
631 }
632
633 bool canBlock() {
634 return duration_.count() > 0;
635 }
636 bool canTimeOut() {
637 return true;
638 }
639
640 bool shouldTimeOut() {
641 return std::chrono::steady_clock::now() > deadline();
642 }
643
644 bool doWait(Futex& futex, uint32_t expected, uint32_t waitMask) {
645 auto result =
646 detail::futexWaitUntil(&futex, expected, deadline(), waitMask);
647 return result != folly::detail::FutexResult::TIMEDOUT;
648 }
649 };
650
651 template <class Clock, class Duration>
652 struct WaitUntilDeadline {
653 std::chrono::time_point<Clock, Duration> absDeadline_;
654
655 bool canBlock() {
656 return true;
657 }
658 bool canTimeOut() {
659 return true;
660 }
661 bool shouldTimeOut() {
662 return Clock::now() > absDeadline_;
663 }
664
665 bool doWait(Futex& futex, uint32_t expected, uint32_t waitMask) {
666 auto result =
667 detail::futexWaitUntil(&futex, expected, absDeadline_, waitMask);
668 return result != folly::detail::FutexResult::TIMEDOUT;
669 }
670 };
671
672 void annotateLazyCreate() {
673 if (AnnotateForThreadSanitizer &&
674 (state_.load() & kAnnotationCreated) == 0) {
675 auto guard = detail::sharedMutexAnnotationGuard(this);
676 // check again
677 if ((state_.load() & kAnnotationCreated) == 0) {
678 state_.fetch_or(kAnnotationCreated);
679 annotate_benign_race_sized(
680 &state_, sizeof(state_), "init TSAN", __FILE__, __LINE__);
681 annotate_rwlock_create(this, __FILE__, __LINE__);
682 }
683 }
684 }
685
686 void annotateDestroy() {
687 if (AnnotateForThreadSanitizer) {
688 annotateLazyCreate();
689 annotate_rwlock_destroy(this, __FILE__, __LINE__);
690 }
691 }
692
693 void annotateAcquired(annotate_rwlock_level w) {
694 if (AnnotateForThreadSanitizer) {
695 annotateLazyCreate();
696 annotate_rwlock_acquired(this, w, __FILE__, __LINE__);
697 }
698 }
699
700 void annotateTryAcquired(bool result, annotate_rwlock_level w) {
701 if (AnnotateForThreadSanitizer) {
702 annotateLazyCreate();
703 annotate_rwlock_try_acquired(this, w, result, __FILE__, __LINE__);
704 }
705 }
706
707 void annotateReleased(annotate_rwlock_level w) {
708 if (AnnotateForThreadSanitizer) {
709 assert((state_.load() & kAnnotationCreated) != 0);
710 annotate_rwlock_released(this, w, __FILE__, __LINE__);
711 }
712 }
713
714 // 32 bits of state
715 Futex state_{};
716
717 // S count needs to be on the end, because we explicitly allow it to
718 // underflow. This can occur while we are in the middle of applying
719 // deferred locks (we remove them from deferredReaders[] before
720 // inlining them), or during token-less unlock_shared() if a racing
721 // lock_shared();unlock_shared() moves the deferredReaders slot while
722 // the first unlock_shared() is scanning. The former case is cleaned
723 // up before we finish applying the locks. The latter case can persist
724 // until destruction, when it is cleaned up.
725 static constexpr uint32_t kIncrHasS = 1 << 11;
726 static constexpr uint32_t kHasS = ~(kIncrHasS - 1);
727
728 // Set if annotation has been completed for this instance. That annotation
729 // (and setting this bit afterward) must be guarded by one of the mutexes in
730 // annotationCreationGuards.
731 static constexpr uint32_t kAnnotationCreated = 1 << 10;
732
733 // If false, then there are definitely no deferred read locks for this
734 // instance. Cleared after initialization and when exclusively locked.
735 static constexpr uint32_t kMayDefer = 1 << 9;
736
737 // lock() cleared kMayDefer as soon as it starts draining readers (so
738 // that it doesn't have to do a second CAS once drain completes), but
739 // unlock_shared() still needs to know whether to scan deferredReaders[]
740 // or not. We copy kMayDefer to kPrevDefer when setting kHasE or
741 // kBegunE, and clear it when clearing those bits.
742 static constexpr uint32_t kPrevDefer = 1 << 8;
743
744 // Exclusive-locked blocks all read locks and write locks. This bit
745 // may be set before all readers have finished, but in that case the
746 // thread that sets it won't return to the caller until all read locks
747 // have been released.
748 static constexpr uint32_t kHasE = 1 << 7;
749
750 // Exclusive-draining means that lock() is waiting for existing readers
751 // to leave, but that new readers may still acquire shared access.
752 // This is only used in reader priority mode. New readers during
753 // drain must be inline. The difference between this and kHasU is that
754 // kBegunE prevents kMayDefer from being set.
755 static constexpr uint32_t kBegunE = 1 << 6;
756
757 // At most one thread may have either exclusive or upgrade lock
758 // ownership. Unlike exclusive mode, ownership of the lock in upgrade
759 // mode doesn't preclude other threads holding the lock in shared mode.
760 // boost's concept for this doesn't explicitly say whether new shared
761 // locks can be acquired one lock_upgrade has succeeded, but doesn't
762 // list that as disallowed. RWSpinLock disallows new read locks after
763 // lock_upgrade has been acquired, but the boost implementation doesn't.
764 // We choose the latter.
765 static constexpr uint32_t kHasU = 1 << 5;
766
767 // There are three states that we consider to be "solo", in that they
768 // cannot coexist with other solo states. These are kHasE, kBegunE,
769 // and kHasU. Note that S doesn't conflict with any of these, because
770 // setting the kHasE is only one of the two steps needed to actually
771 // acquire the lock in exclusive mode (the other is draining the existing
772 // S holders).
773 static constexpr uint32_t kHasSolo = kHasE | kBegunE | kHasU;
774
775 // Once a thread sets kHasE it needs to wait for the current readers
776 // to exit the lock. We give this a separate wait identity from the
777 // waiting to set kHasE so that we can perform partial wakeups (wake
778 // one instead of wake all).
779 static constexpr uint32_t kWaitingNotS = 1 << 4;
780
781 // When waking writers we can either wake them all, in which case we
782 // can clear kWaitingE, or we can call futexWake(1). futexWake tells
783 // us if anybody woke up, but even if we detect that nobody woke up we
784 // can't clear the bit after the fact without issuing another wakeup.
785 // To avoid thundering herds when there are lots of pending lock()
786 // without needing to call futexWake twice when there is only one
787 // waiter, kWaitingE actually encodes if we have observed multiple
788 // concurrent waiters. Tricky: ABA issues on futexWait mean that when
789 // we see kWaitingESingle we can't assume that there is only one.
790 static constexpr uint32_t kWaitingESingle = 1 << 2;
791 static constexpr uint32_t kWaitingEMultiple = 1 << 3;
792 static constexpr uint32_t kWaitingE = kWaitingESingle | kWaitingEMultiple;
793
794 // kWaitingU is essentially a 1 bit saturating counter. It always
795 // requires a wakeAll.
796 static constexpr uint32_t kWaitingU = 1 << 1;
797
798 // All blocked lock_shared() should be awoken, so it is correct (not
799 // suboptimal) to wakeAll if there are any shared readers.
800 static constexpr uint32_t kWaitingS = 1 << 0;
801
802 // kWaitingAny is a mask of all of the bits that record the state of
803 // threads, rather than the state of the lock. It is convenient to be
804 // able to mask them off during asserts.
805 static constexpr uint32_t kWaitingAny =
806 kWaitingNotS | kWaitingE | kWaitingU | kWaitingS;
807
808 // The reader count at which a reader will attempt to use the lock
809 // in deferred mode. If this value is 2, then the second concurrent
810 // reader will set kMayDefer and use deferredReaders[]. kMayDefer is
811 // cleared during exclusive access, so this threshold must be reached
812 // each time a lock is held in exclusive mode.
813 static constexpr uint32_t kNumSharedToStartDeferring = 2;
814
815 // The typical number of spins that a thread will wait for a state
816 // transition. There is no bound on the number of threads that can wait
817 // for a writer, so we are pretty conservative here to limit the chance
818 // that we are starving the writer of CPU. Each spin is 6 or 7 nanos,
819 // almost all of which is in the pause instruction.
820 static constexpr uint32_t kMaxSpinCount = !BlockImmediately ? 1000 : 2;
821
822 // The maximum number of soft yields before falling back to futex.
823 // If the preemption heuristic is activated we will fall back before
824 // this. A soft yield takes ~900 nanos (two sched_yield plus a call
825 // to getrusage, with checks of the goal at each step). Soft yields
826 // aren't compatible with deterministic execution under test (unlike
827 // futexWaitUntil, which has a capricious but deterministic back end).
828 static constexpr uint32_t kMaxSoftYieldCount = !BlockImmediately ? 1000 : 0;
829
830 // If AccessSpreader assigns indexes from 0..k*n-1 on a system where some
831 // level of the memory hierarchy is symmetrically divided into k pieces
832 // (NUMA nodes, last-level caches, L1 caches, ...), then slot indexes
833 // that are the same after integer division by k share that resource.
834 // Our strategy for deferred readers is to probe up to numSlots/4 slots,
835 // using the full granularity of AccessSpreader for the start slot
836 // and then search outward. We can use AccessSpreader::current(n)
837 // without managing our own spreader if kMaxDeferredReaders <=
838 // AccessSpreader::kMaxCpus, which is currently 128.
839 //
840 // Our 2-socket E5-2660 machines have 8 L1 caches on each chip,
841 // with 64 byte cache lines. That means we need 64*16 bytes of
842 // deferredReaders[] to give each L1 its own playground. On x86_64
843 // each DeferredReaderSlot is 8 bytes, so we need kMaxDeferredReaders
844 // * kDeferredSeparationFactor >= 64 * 16 / 8 == 128. If
845 // kDeferredSearchDistance * kDeferredSeparationFactor <=
846 // 64 / 8 then we will search only within a single cache line, which
847 // guarantees we won't have inter-L1 contention. We give ourselves
848 // a factor of 2 on the core count, which should hold us for a couple
849 // processor generations. deferredReaders[] is 2048 bytes currently.
850 public:
851 static constexpr uint32_t kMaxDeferredReaders = 64;
852 static constexpr uint32_t kDeferredSearchDistance = 2;
853 static constexpr uint32_t kDeferredSeparationFactor = 4;
854
855 private:
856 static_assert(
857 !(kMaxDeferredReaders & (kMaxDeferredReaders - 1)),
858 "kMaxDeferredReaders must be a power of 2");
859 static_assert(
860 !(kDeferredSearchDistance & (kDeferredSearchDistance - 1)),
861 "kDeferredSearchDistance must be a power of 2");
862
863 // The number of deferred locks that can be simultaneously acquired
864 // by a thread via the token-less methods without performing any heap
865 // allocations. Each of these costs 3 pointers (24 bytes, probably)
866 // per thread. There's not much point in making this larger than
867 // kDeferredSearchDistance.
868 static constexpr uint32_t kTokenStackTLSCapacity = 2;
869
870 // We need to make sure that if there is a lock_shared()
871 // and lock_shared(token) followed by unlock_shared() and
872 // unlock_shared(token), the token-less unlock doesn't null
873 // out deferredReaders[token.slot_]. If we allowed that, then
874 // unlock_shared(token) wouldn't be able to assume that its lock
875 // had been inlined by applyDeferredReaders when it finds that
876 // deferredReaders[token.slot_] no longer points to this. We accomplish
877 // this by stealing bit 0 from the pointer to record that the slot's
878 // element has no token, hence our use of uintptr_t in deferredReaders[].
879 static constexpr uintptr_t kTokenless = 0x1;
880
881 // This is the starting location for Token-less unlock_shared().
882 static FOLLY_SHAREDMUTEX_TLS uint32_t tls_lastTokenlessSlot;
883
884 // Last deferred reader slot used.
885 static FOLLY_SHAREDMUTEX_TLS uint32_t tls_lastDeferredReaderSlot;
886
887 // Only indexes divisible by kDeferredSeparationFactor are used.
888 // If any of those elements points to a SharedMutexImpl, then it
889 // should be considered that there is a shared lock on that instance.
890 // See kTokenless.
891 public:
892 typedef Atom<uintptr_t> DeferredReaderSlot;
893
894 private:
895 alignas(hardware_destructive_interference_size) static DeferredReaderSlot
896 deferredReaders[kMaxDeferredReaders * kDeferredSeparationFactor];
897
898 // Performs an exclusive lock, waiting for state_ & waitMask to be
899 // zero first
900 template <class WaitContext>
901 bool lockExclusiveImpl(uint32_t preconditionGoalMask, WaitContext& ctx) {
902 uint32_t state = state_.load(std::memory_order_acquire);
903 if (LIKELY(
904 (state & (preconditionGoalMask | kMayDefer | kHasS)) == 0 &&
905 state_.compare_exchange_strong(state, (state | kHasE) & ~kHasU))) {
906 return true;
907 } else {
908 return lockExclusiveImpl(state, preconditionGoalMask, ctx);
909 }
910 }
911
912 template <class WaitContext>
913 bool lockExclusiveImpl(
914 uint32_t& state,
915 uint32_t preconditionGoalMask,
916 WaitContext& ctx) {
917 while (true) {
918 if (UNLIKELY((state & preconditionGoalMask) != 0) &&
919 !waitForZeroBits(state, preconditionGoalMask, kWaitingE, ctx) &&
920 ctx.canTimeOut()) {
921 return false;
922 }
923
924 uint32_t after = (state & kMayDefer) == 0 ? 0 : kPrevDefer;
925 if (!kReaderPriority || (state & (kMayDefer | kHasS)) == 0) {
926 // Block readers immediately, either because we are in write
927 // priority mode or because we can acquire the lock in one
928 // step. Note that if state has kHasU, then we are doing an
929 // unlock_upgrade_and_lock() and we should clear it (reader
930 // priority branch also does this).
931 after |= (state | kHasE) & ~(kHasU | kMayDefer);
932 } else {
933 after |= (state | kBegunE) & ~(kHasU | kMayDefer);
934 }
935 if (state_.compare_exchange_strong(state, after)) {
936 auto before = state;
937 state = after;
938
939 // If we set kHasE (writer priority) then no new readers can
940 // arrive. If we set kBegunE then they can still enter, but
941 // they must be inline. Either way we need to either spin on
942 // deferredReaders[] slots, or inline them so that we can wait on
943 // kHasS to zero itself. deferredReaders[] is pointers, which on
944 // x86_64 are bigger than futex() can handle, so we inline the
945 // deferred locks instead of trying to futexWait on each slot.
946 // Readers are responsible for rechecking state_ after recording
947 // a deferred read to avoid atomicity problems between the state_
948 // CAS and applyDeferredReader's reads of deferredReaders[].
949 if (UNLIKELY((before & kMayDefer) != 0)) {
950 applyDeferredReaders(state, ctx);
951 }
952 while (true) {
953 assert((state & (kHasE | kBegunE)) != 0 && (state & kHasU) == 0);
954 if (UNLIKELY((state & kHasS) != 0) &&
955 !waitForZeroBits(state, kHasS, kWaitingNotS, ctx) &&
956 ctx.canTimeOut()) {
957 // Ugh. We blocked new readers and other writers for a while,
958 // but were unable to complete. Move on. On the plus side
959 // we can clear kWaitingNotS because nobody else can piggyback
960 // on it.
961 state = (state_ &= ~(kPrevDefer | kHasE | kBegunE | kWaitingNotS));
962 wakeRegisteredWaiters(state, kWaitingE | kWaitingU | kWaitingS);
963 return false;
964 }
965
966 if (kReaderPriority && (state & kHasE) == 0) {
967 assert((state & kBegunE) != 0);
968 if (!state_.compare_exchange_strong(
969 state, (state & ~kBegunE) | kHasE)) {
970 continue;
971 }
972 }
973
974 return true;
975 }
976 }
977 }
978 }
979
980 template <class WaitContext>
981 bool waitForZeroBits(
982 uint32_t& state,
983 uint32_t goal,
984 uint32_t waitMask,
985 WaitContext& ctx) {
986 uint32_t spinCount = 0;
987 while (true) {
988 state = state_.load(std::memory_order_acquire);
989 if ((state & goal) == 0) {
990 return true;
991 }
992 asm_volatile_pause();
993 ++spinCount;
994 if (UNLIKELY(spinCount >= kMaxSpinCount)) {
995 return ctx.canBlock() &&
996 yieldWaitForZeroBits(state, goal, waitMask, ctx);
997 }
998 }
999 }
1000
1001 template <class WaitContext>
1002 bool yieldWaitForZeroBits(
1003 uint32_t& state,
1004 uint32_t goal,
1005 uint32_t waitMask,
1006 WaitContext& ctx) {
1007#ifdef RUSAGE_THREAD
1008 struct rusage usage;
1009 std::memset(&usage, 0, sizeof(usage));
1010 long before = -1;
1011#endif
1012 for (uint32_t yieldCount = 0; yieldCount < kMaxSoftYieldCount;
1013 ++yieldCount) {
1014 for (int softState = 0; softState < 3; ++softState) {
1015 if (softState < 2) {
1016 std::this_thread::yield();
1017 } else {
1018#ifdef RUSAGE_THREAD
1019 getrusage(RUSAGE_THREAD, &usage);
1020#endif
1021 }
1022 if (((state = state_.load(std::memory_order_acquire)) & goal) == 0) {
1023 return true;
1024 }
1025 if (ctx.shouldTimeOut()) {
1026 return false;
1027 }
1028 }
1029#ifdef RUSAGE_THREAD
1030 if (before >= 0 && usage.ru_nivcsw >= before + 2) {
1031 // One involuntary csw might just be occasional background work,
1032 // but if we get two in a row then we guess that there is someone
1033 // else who can profitably use this CPU. Fall back to futex
1034 break;
1035 }
1036 before = usage.ru_nivcsw;
1037#endif
1038 }
1039 return futexWaitForZeroBits(state, goal, waitMask, ctx);
1040 }
1041
1042 template <class WaitContext>
1043 bool futexWaitForZeroBits(
1044 uint32_t& state,
1045 uint32_t goal,
1046 uint32_t waitMask,
1047 WaitContext& ctx) {
1048 assert(
1049 waitMask == kWaitingNotS || waitMask == kWaitingE ||
1050 waitMask == kWaitingU || waitMask == kWaitingS);
1051
1052 while (true) {
1053 state = state_.load(std::memory_order_acquire);
1054 if ((state & goal) == 0) {
1055 return true;
1056 }
1057
1058 auto after = state;
1059 if (waitMask == kWaitingE) {
1060 if ((state & kWaitingESingle) != 0) {
1061 after |= kWaitingEMultiple;
1062 } else {
1063 after |= kWaitingESingle;
1064 }
1065 } else {
1066 after |= waitMask;
1067 }
1068
1069 // CAS is better than atomic |= here, because it lets us avoid
1070 // setting the wait flag when the goal is concurrently achieved
1071 if (after != state && !state_.compare_exchange_strong(state, after)) {
1072 continue;
1073 }
1074
1075 if (!ctx.doWait(state_, after, waitMask)) {
1076 // timed out
1077 return false;
1078 }
1079 }
1080 }
1081
1082 // Wakes up waiters registered in state_ as appropriate, clearing the
1083 // awaiting bits for anybody that was awoken. Tries to perform direct
1084 // single wakeup of an exclusive waiter if appropriate
1085 void wakeRegisteredWaiters(uint32_t& state, uint32_t wakeMask) {
1086 if (UNLIKELY((state & wakeMask) != 0)) {
1087 wakeRegisteredWaitersImpl(state, wakeMask);
1088 }
1089 }
1090
1091 void wakeRegisteredWaitersImpl(uint32_t& state, uint32_t wakeMask) {
1092 // If there are multiple lock() pending only one of them will actually
1093 // get to wake up, so issuing futexWakeAll will make a thundering herd.
1094 // There's nothing stopping us from issuing futexWake(1) instead,
1095 // so long as the wait bits are still an accurate reflection of
1096 // the waiters. If we notice (via futexWake's return value) that
1097 // nobody woke up then we can try again with the normal wake-all path.
1098 // Note that we can't just clear the bits at that point; we need to
1099 // clear the bits and then issue another wakeup.
1100 //
1101 // It is possible that we wake an E waiter but an outside S grabs the
1102 // lock instead, at which point we should wake pending U and S waiters.
1103 // Rather than tracking state to make the failing E regenerate the
1104 // wakeup, we just disable the optimization in the case that there
1105 // are waiting U or S that we are eligible to wake.
1106 if ((wakeMask & kWaitingE) == kWaitingE &&
1107 (state & wakeMask) == kWaitingE &&
1108 detail::futexWake(&state_, 1, kWaitingE) > 0) {
1109 // somebody woke up, so leave state_ as is and clear it later
1110 return;
1111 }
1112
1113 if ((state & wakeMask) != 0) {
1114 auto prev = state_.fetch_and(~wakeMask);
1115 if ((prev & wakeMask) != 0) {
1116 futexWakeAll(wakeMask);
1117 }
1118 state = prev & ~wakeMask;
1119 }
1120 }
1121
1122 void futexWakeAll(uint32_t wakeMask) {
1123 detail::futexWake(&state_, std::numeric_limits<int>::max(), wakeMask);
1124 }
1125
1126 DeferredReaderSlot* deferredReader(uint32_t slot) {
1127 return &deferredReaders[slot * kDeferredSeparationFactor];
1128 }
1129
1130 uintptr_t tokenfulSlotValue() {
1131 return reinterpret_cast<uintptr_t>(this);
1132 }
1133
1134 uintptr_t tokenlessSlotValue() {
1135 return tokenfulSlotValue() | kTokenless;
1136 }
1137
1138 bool slotValueIsThis(uintptr_t slotValue) {
1139 return (slotValue & ~kTokenless) == tokenfulSlotValue();
1140 }
1141
1142 // Clears any deferredReaders[] that point to this, adjusting the inline
1143 // shared lock count to compensate. Does some spinning and yielding
1144 // to avoid the work. Always finishes the application, even if ctx
1145 // times out.
1146 template <class WaitContext>
1147 void applyDeferredReaders(uint32_t& state, WaitContext& ctx) {
1148 uint32_t slot = 0;
1149
1150 uint32_t spinCount = 0;
1151 while (true) {
1152 while (!slotValueIsThis(
1153 deferredReader(slot)->load(std::memory_order_acquire))) {
1154 if (++slot == kMaxDeferredReaders) {
1155 return;
1156 }
1157 }
1158 asm_volatile_pause();
1159 if (UNLIKELY(++spinCount >= kMaxSpinCount)) {
1160 applyDeferredReaders(state, ctx, slot);
1161 return;
1162 }
1163 }
1164 }
1165
1166 template <class WaitContext>
1167 void applyDeferredReaders(uint32_t& state, WaitContext& ctx, uint32_t slot) {
1168#ifdef RUSAGE_THREAD
1169 struct rusage usage;
1170 std::memset(&usage, 0, sizeof(usage));
1171 long before = -1;
1172#endif
1173 for (uint32_t yieldCount = 0; yieldCount < kMaxSoftYieldCount;
1174 ++yieldCount) {
1175 for (int softState = 0; softState < 3; ++softState) {
1176 if (softState < 2) {
1177 std::this_thread::yield();
1178 } else {
1179#ifdef RUSAGE_THREAD
1180 getrusage(RUSAGE_THREAD, &usage);
1181#endif
1182 }
1183 while (!slotValueIsThis(
1184 deferredReader(slot)->load(std::memory_order_acquire))) {
1185 if (++slot == kMaxDeferredReaders) {
1186 return;
1187 }
1188 }
1189 if (ctx.shouldTimeOut()) {
1190 // finish applying immediately on timeout
1191 break;
1192 }
1193 }
1194#ifdef RUSAGE_THREAD
1195 if (before >= 0 && usage.ru_nivcsw >= before + 2) {
1196 // heuristic says run queue is not empty
1197 break;
1198 }
1199 before = usage.ru_nivcsw;
1200#endif
1201 }
1202
1203 uint32_t movedSlotCount = 0;
1204 for (; slot < kMaxDeferredReaders; ++slot) {
1205 auto slotPtr = deferredReader(slot);
1206 auto slotValue = slotPtr->load(std::memory_order_acquire);
1207 if (slotValueIsThis(slotValue) &&
1208 slotPtr->compare_exchange_strong(slotValue, 0)) {
1209 ++movedSlotCount;
1210 }
1211 }
1212
1213 if (movedSlotCount > 0) {
1214 state = (state_ += movedSlotCount * kIncrHasS);
1215 }
1216 assert((state & (kHasE | kBegunE)) != 0);
1217
1218 // if state + kIncrHasS overflows (off the end of state) then either
1219 // we have 2^(32-9) readers (almost certainly an application bug)
1220 // or we had an underflow (also a bug)
1221 assert(state < state + kIncrHasS);
1222 }
1223
1224 // It is straightfoward to make a token-less lock_shared() and
1225 // unlock_shared() either by making the token-less version always use
1226 // INLINE_SHARED mode or by removing the token version. Supporting
1227 // deferred operation for both types is trickier than it appears, because
1228 // the purpose of the token it so that unlock_shared doesn't have to
1229 // look in other slots for its deferred lock. Token-less unlock_shared
1230 // might place a deferred lock in one place and then release a different
1231 // slot that was originally used by the token-ful version. If this was
1232 // important we could solve the problem by differentiating the deferred
1233 // locks so that cross-variety release wouldn't occur. The best way
1234 // is probably to steal a bit from the pointer, making deferredLocks[]
1235 // an array of Atom<uintptr_t>.
1236
1237 template <class WaitContext>
1238 bool lockSharedImpl(Token* token, WaitContext& ctx) {
1239 uint32_t state = state_.load(std::memory_order_relaxed);
1240 if ((state & (kHasS | kMayDefer | kHasE)) == 0 &&
1241 state_.compare_exchange_strong(state, state + kIncrHasS)) {
1242 if (token != nullptr) {
1243 token->type_ = Token::Type::INLINE_SHARED;
1244 }
1245 return true;
1246 }
1247 return lockSharedImpl(state, token, ctx);
1248 }
1249
1250 template <class WaitContext>
1251 bool lockSharedImpl(uint32_t& state, Token* token, WaitContext& ctx);
1252
1253 // Updates the state in/out argument as if the locks were made inline,
1254 // but does not update state_
1255 void cleanupTokenlessSharedDeferred(uint32_t& state) {
1256 for (uint32_t i = 0; i < kMaxDeferredReaders; ++i) {
1257 auto slotPtr = deferredReader(i);
1258 auto slotValue = slotPtr->load(std::memory_order_relaxed);
1259 if (slotValue == tokenlessSlotValue()) {
1260 slotPtr->store(0, std::memory_order_relaxed);
1261 state += kIncrHasS;
1262 if ((state & kHasS) == 0) {
1263 break;
1264 }
1265 }
1266 }
1267 }
1268
1269 bool tryUnlockTokenlessSharedDeferred();
1270
1271 bool tryUnlockSharedDeferred(uint32_t slot) {
1272 assert(slot < kMaxDeferredReaders);
1273 auto slotValue = tokenfulSlotValue();
1274 return deferredReader(slot)->compare_exchange_strong(slotValue, 0);
1275 }
1276
1277 uint32_t unlockSharedInline() {
1278 uint32_t state = (state_ -= kIncrHasS);
1279 assert(
1280 (state & (kHasE | kBegunE | kMayDefer)) != 0 ||
1281 state < state + kIncrHasS);
1282 if ((state & kHasS) == 0) {
1283 // Only the second half of lock() can be blocked by a non-zero
1284 // reader count, so that's the only thing we need to wake
1285 wakeRegisteredWaiters(state, kWaitingNotS);
1286 }
1287 return state;
1288 }
1289
1290 template <class WaitContext>
1291 bool lockUpgradeImpl(WaitContext& ctx) {
1292 uint32_t state;
1293 do {
1294 if (!waitForZeroBits(state, kHasSolo, kWaitingU, ctx)) {
1295 return false;
1296 }
1297 } while (!state_.compare_exchange_strong(state, state | kHasU));
1298 return true;
1299 }
1300
1301 public:
1302 class ReadHolder {
1303 ReadHolder() : lock_(nullptr) {}
1304
1305 public:
1306 explicit ReadHolder(const SharedMutexImpl* lock)
1307 : lock_(const_cast<SharedMutexImpl*>(lock)) {
1308 if (lock_) {
1309 lock_->lock_shared(token_);
1310 }
1311 }
1312
1313 explicit ReadHolder(const SharedMutexImpl& lock)
1314 : lock_(const_cast<SharedMutexImpl*>(&lock)) {
1315 lock_->lock_shared(token_);
1316 }
1317
1318 ReadHolder(ReadHolder&& rhs) noexcept
1319 : lock_(rhs.lock_), token_(rhs.token_) {
1320 rhs.lock_ = nullptr;
1321 }
1322
1323 // Downgrade from upgrade mode
1324 explicit ReadHolder(UpgradeHolder&& upgraded) : lock_(upgraded.lock_) {
1325 assert(upgraded.lock_ != nullptr);
1326 upgraded.lock_ = nullptr;
1327 lock_->unlock_upgrade_and_lock_shared(token_);
1328 }
1329
1330 // Downgrade from exclusive mode
1331 explicit ReadHolder(WriteHolder&& writer) : lock_(writer.lock_) {
1332 assert(writer.lock_ != nullptr);
1333 writer.lock_ = nullptr;
1334 lock_->unlock_and_lock_shared(token_);
1335 }
1336
1337 ReadHolder& operator=(ReadHolder&& rhs) noexcept {
1338 std::swap(lock_, rhs.lock_);
1339 std::swap(token_, rhs.token_);
1340 return *this;
1341 }
1342
1343 ReadHolder(const ReadHolder& rhs) = delete;
1344 ReadHolder& operator=(const ReadHolder& rhs) = delete;
1345
1346 ~ReadHolder() {
1347 unlock();
1348 }
1349
1350 void unlock() {
1351 if (lock_) {
1352 lock_->unlock_shared(token_);
1353 lock_ = nullptr;
1354 }
1355 }
1356
1357 private:
1358 friend class UpgradeHolder;
1359 friend class WriteHolder;
1360 SharedMutexImpl* lock_;
1361 SharedMutexToken token_;
1362 };
1363
1364 class UpgradeHolder {
1365 UpgradeHolder() : lock_(nullptr) {}
1366
1367 public:
1368 explicit UpgradeHolder(SharedMutexImpl* lock) : lock_(lock) {
1369 if (lock_) {
1370 lock_->lock_upgrade();
1371 }
1372 }
1373
1374 explicit UpgradeHolder(SharedMutexImpl& lock) : lock_(&lock) {
1375 lock_->lock_upgrade();
1376 }
1377
1378 // Downgrade from exclusive mode
1379 explicit UpgradeHolder(WriteHolder&& writer) : lock_(writer.lock_) {
1380 assert(writer.lock_ != nullptr);
1381 writer.lock_ = nullptr;
1382 lock_->unlock_and_lock_upgrade();
1383 }
1384
1385 UpgradeHolder(UpgradeHolder&& rhs) noexcept : lock_(rhs.lock_) {
1386 rhs.lock_ = nullptr;
1387 }
1388
1389 UpgradeHolder& operator=(UpgradeHolder&& rhs) noexcept {
1390 std::swap(lock_, rhs.lock_);
1391 return *this;
1392 }
1393
1394 UpgradeHolder(const UpgradeHolder& rhs) = delete;
1395 UpgradeHolder& operator=(const UpgradeHolder& rhs) = delete;
1396
1397 ~UpgradeHolder() {
1398 unlock();
1399 }
1400
1401 void unlock() {
1402 if (lock_) {
1403 lock_->unlock_upgrade();
1404 lock_ = nullptr;
1405 }
1406 }
1407
1408 private:
1409 friend class WriteHolder;
1410 friend class ReadHolder;
1411 SharedMutexImpl* lock_;
1412 };
1413
1414 class WriteHolder {
1415 WriteHolder() : lock_(nullptr) {}
1416
1417 public:
1418 explicit WriteHolder(SharedMutexImpl* lock) : lock_(lock) {
1419 if (lock_) {
1420 lock_->lock();
1421 }
1422 }
1423
1424 explicit WriteHolder(SharedMutexImpl& lock) : lock_(&lock) {
1425 lock_->lock();
1426 }
1427
1428 // Promotion from upgrade mode
1429 explicit WriteHolder(UpgradeHolder&& upgrade) : lock_(upgrade.lock_) {
1430 assert(upgrade.lock_ != nullptr);
1431 upgrade.lock_ = nullptr;
1432 lock_->unlock_upgrade_and_lock();
1433 }
1434
1435 // README:
1436 //
1437 // It is intended that WriteHolder(ReadHolder&& rhs) do not exist.
1438 //
1439 // Shared locks (read) can not safely upgrade to unique locks (write).
1440 // That upgrade path is a well-known recipe for deadlock, so we explicitly
1441 // disallow it.
1442 //
1443 // If you need to do a conditional mutation, you have a few options:
1444 // 1. Check the condition under a shared lock and release it.
1445 // Then maybe check the condition again under a unique lock and maybe do
1446 // the mutation.
1447 // 2. Check the condition once under an upgradeable lock.
1448 // Then maybe upgrade the lock to a unique lock and do the mutation.
1449 // 3. Check the condition and maybe perform the mutation under a unique
1450 // lock.
1451 //
1452 // Relevant upgradeable lock notes:
1453 // * At most one upgradeable lock can be held at a time for a given shared
1454 // mutex, just like a unique lock.
1455 // * An upgradeable lock may be held concurrently with any number of shared
1456 // locks.
1457 // * An upgradeable lock may be upgraded atomically to a unique lock.
1458
1459 WriteHolder(WriteHolder&& rhs) noexcept : lock_(rhs.lock_) {
1460 rhs.lock_ = nullptr;
1461 }
1462
1463 WriteHolder& operator=(WriteHolder&& rhs) noexcept {
1464 std::swap(lock_, rhs.lock_);
1465 return *this;
1466 }
1467
1468 WriteHolder(const WriteHolder& rhs) = delete;
1469 WriteHolder& operator=(const WriteHolder& rhs) = delete;
1470
1471 ~WriteHolder() {
1472 unlock();
1473 }
1474
1475 void unlock() {
1476 if (lock_) {
1477 lock_->unlock();
1478 lock_ = nullptr;
1479 }
1480 }
1481
1482 private:
1483 friend class ReadHolder;
1484 friend class UpgradeHolder;
1485 SharedMutexImpl* lock_;
1486 };
1487
1488 // Adapters for Synchronized<>
1489 friend void acquireRead(SharedMutexImpl& lock) {
1490 lock.lock_shared();
1491 }
1492 friend void acquireReadWrite(SharedMutexImpl& lock) {
1493 lock.lock();
1494 }
1495 friend void releaseRead(SharedMutexImpl& lock) {
1496 lock.unlock_shared();
1497 }
1498 friend void releaseReadWrite(SharedMutexImpl& lock) {
1499 lock.unlock();
1500 }
1501 friend bool acquireRead(SharedMutexImpl& lock, unsigned int ms) {
1502 return lock.try_lock_shared_for(std::chrono::milliseconds(ms));
1503 }
1504 friend bool acquireReadWrite(SharedMutexImpl& lock, unsigned int ms) {
1505 return lock.try_lock_for(std::chrono::milliseconds(ms));
1506 }
1507};
1508
1509typedef SharedMutexImpl<true> SharedMutexReadPriority;
1510typedef SharedMutexImpl<false> SharedMutexWritePriority;
1511typedef SharedMutexWritePriority SharedMutex;
1512typedef SharedMutexImpl<false, void, std::atomic, false, false>
1513 SharedMutexSuppressTSAN;
1514
1515// Prevent the compiler from instantiating these in other translation units.
1516// They are instantiated once in SharedMutex.cpp
1517extern template class SharedMutexImpl<true>;
1518extern template class SharedMutexImpl<false>;
1519
1520template <
1521 bool ReaderPriority,
1522 typename Tag_,
1523 template <typename> class Atom,
1524 bool BlockImmediately,
1525 bool AnnotateForThreadSanitizer>
1526alignas(hardware_destructive_interference_size) typename SharedMutexImpl<
1527 ReaderPriority,
1528 Tag_,
1529 Atom,
1530 BlockImmediately,
1531 AnnotateForThreadSanitizer>::DeferredReaderSlot
1532 SharedMutexImpl<
1533 ReaderPriority,
1534 Tag_,
1535 Atom,
1536 BlockImmediately,
1537 AnnotateForThreadSanitizer>::deferredReaders
1538 [kMaxDeferredReaders * kDeferredSeparationFactor] = {};
1539
1540template <
1541 bool ReaderPriority,
1542 typename Tag_,
1543 template <typename> class Atom,
1544 bool BlockImmediately,
1545 bool AnnotateForThreadSanitizer>
1546FOLLY_SHAREDMUTEX_TLS uint32_t SharedMutexImpl<
1547 ReaderPriority,
1548 Tag_,
1549 Atom,
1550 BlockImmediately,
1551 AnnotateForThreadSanitizer>::tls_lastTokenlessSlot = 0;
1552
1553template <
1554 bool ReaderPriority,
1555 typename Tag_,
1556 template <typename> class Atom,
1557 bool BlockImmediately,
1558 bool AnnotateForThreadSanitizer>
1559FOLLY_SHAREDMUTEX_TLS uint32_t SharedMutexImpl<
1560 ReaderPriority,
1561 Tag_,
1562 Atom,
1563 BlockImmediately,
1564 AnnotateForThreadSanitizer>::tls_lastDeferredReaderSlot = 0;
1565
1566template <
1567 bool ReaderPriority,
1568 typename Tag_,
1569 template <typename> class Atom,
1570 bool BlockImmediately,
1571 bool AnnotateForThreadSanitizer>
1572bool SharedMutexImpl<
1573 ReaderPriority,
1574 Tag_,
1575 Atom,
1576 BlockImmediately,
1577 AnnotateForThreadSanitizer>::tryUnlockTokenlessSharedDeferred() {
1578 auto bestSlot = tls_lastTokenlessSlot;
1579 for (uint32_t i = 0; i < kMaxDeferredReaders; ++i) {
1580 auto slotPtr = deferredReader(bestSlot ^ i);
1581 auto slotValue = slotPtr->load(std::memory_order_relaxed);
1582 if (slotValue == tokenlessSlotValue() &&
1583 slotPtr->compare_exchange_strong(slotValue, 0)) {
1584 tls_lastTokenlessSlot = bestSlot ^ i;
1585 return true;
1586 }
1587 }
1588 return false;
1589}
1590
1591template <
1592 bool ReaderPriority,
1593 typename Tag_,
1594 template <typename> class Atom,
1595 bool BlockImmediately,
1596 bool AnnotateForThreadSanitizer>
1597template <class WaitContext>
1598bool SharedMutexImpl<
1599 ReaderPriority,
1600 Tag_,
1601 Atom,
1602 BlockImmediately,
1603 AnnotateForThreadSanitizer>::
1604 lockSharedImpl(uint32_t& state, Token* token, WaitContext& ctx) {
1605 while (true) {
1606 if (UNLIKELY((state & kHasE) != 0) &&
1607 !waitForZeroBits(state, kHasE, kWaitingS, ctx) && ctx.canTimeOut()) {
1608 return false;
1609 }
1610
1611 uint32_t slot = tls_lastDeferredReaderSlot;
1612 uintptr_t slotValue = 1; // any non-zero value will do
1613
1614 bool canAlreadyDefer = (state & kMayDefer) != 0;
1615 bool aboveDeferThreshold =
1616 (state & kHasS) >= (kNumSharedToStartDeferring - 1) * kIncrHasS;
1617 bool drainInProgress = ReaderPriority && (state & kBegunE) != 0;
1618 if (canAlreadyDefer || (aboveDeferThreshold && !drainInProgress)) {
1619 /* Try using the most recent slot first. */
1620 slotValue = deferredReader(slot)->load(std::memory_order_relaxed);
1621 if (slotValue != 0) {
1622 // starting point for our empty-slot search, can change after
1623 // calling waitForZeroBits
1624 uint32_t bestSlot =
1625 (uint32_t)folly::AccessSpreader<Atom>::current(kMaxDeferredReaders);
1626
1627 // deferred readers are already enabled, or it is time to
1628 // enable them if we can find a slot
1629 for (uint32_t i = 0; i < kDeferredSearchDistance; ++i) {
1630 slot = bestSlot ^ i;
1631 assert(slot < kMaxDeferredReaders);
1632 slotValue = deferredReader(slot)->load(std::memory_order_relaxed);
1633 if (slotValue == 0) {
1634 // found empty slot
1635 tls_lastDeferredReaderSlot = slot;
1636 break;
1637 }
1638 }
1639 }
1640 }
1641
1642 if (slotValue != 0) {
1643 // not yet deferred, or no empty slots
1644 if (state_.compare_exchange_strong(state, state + kIncrHasS)) {
1645 // successfully recorded the read lock inline
1646 if (token != nullptr) {
1647 token->type_ = Token::Type::INLINE_SHARED;
1648 }
1649 return true;
1650 }
1651 // state is updated, try again
1652 continue;
1653 }
1654
1655 // record that deferred readers might be in use if necessary
1656 if ((state & kMayDefer) == 0) {
1657 if (!state_.compare_exchange_strong(state, state | kMayDefer)) {
1658 // keep going if CAS failed because somebody else set the bit
1659 // for us
1660 if ((state & (kHasE | kMayDefer)) != kMayDefer) {
1661 continue;
1662 }
1663 }
1664 // state = state | kMayDefer;
1665 }
1666
1667 // try to use the slot
1668 bool gotSlot = deferredReader(slot)->compare_exchange_strong(
1669 slotValue,
1670 token == nullptr ? tokenlessSlotValue() : tokenfulSlotValue());
1671
1672 // If we got the slot, we need to verify that an exclusive lock
1673 // didn't happen since we last checked. If we didn't get the slot we
1674 // need to recheck state_ anyway to make sure we don't waste too much
1675 // work. It is also possible that since we checked state_ someone
1676 // has acquired and released the write lock, clearing kMayDefer.
1677 // Both cases are covered by looking for the readers-possible bit,
1678 // because it is off when the exclusive lock bit is set.
1679 state = state_.load(std::memory_order_acquire);
1680
1681 if (!gotSlot) {
1682 continue;
1683 }
1684
1685 if (token == nullptr) {
1686 tls_lastTokenlessSlot = slot;
1687 }
1688
1689 if ((state & kMayDefer) != 0) {
1690 assert((state & kHasE) == 0);
1691 // success
1692 if (token != nullptr) {
1693 token->type_ = Token::Type::DEFERRED_SHARED;
1694 token->slot_ = (uint16_t)slot;
1695 }
1696 return true;
1697 }
1698
1699 // release the slot before retrying
1700 if (token == nullptr) {
1701 // We can't rely on slot. Token-less slot values can be freed by
1702 // any unlock_shared(), so we need to do the full deferredReader
1703 // search during unlock. Unlike unlock_shared(), we can't trust
1704 // kPrevDefer here. This deferred lock isn't visible to lock()
1705 // (that's the whole reason we're undoing it) so there might have
1706 // subsequently been an unlock() and lock() with no intervening
1707 // transition to deferred mode.
1708 if (!tryUnlockTokenlessSharedDeferred()) {
1709 unlockSharedInline();
1710 }
1711 } else {
1712 if (!tryUnlockSharedDeferred(slot)) {
1713 unlockSharedInline();
1714 }
1715 }
1716
1717 // We got here not because the lock was unavailable, but because
1718 // we lost a compare-and-swap. Try-lock is typically allowed to
1719 // have spurious failures, but there is no lock efficiency gain
1720 // from exploiting that freedom here.
1721 }
1722}
1723
1724} // namespace folly
1725