Summary: [Folly] Opportunity to use `exchange` in `ScopeGuardImpl`.

Reviewed By: Orvid

Differential Revision: D10119838

fbshipit-source-id: 64a00f4f60d913f8af248ebf891f730d2c69fec8

Summary: then to thenValue to remove calls to deprecated functions.

Reviewed By: Orvid

Differential Revision: D10362133

fbshipit-source-id: e638d24e95c5c6c46d0366eabcd1ff9537421bc6

Summary:
Just require that MSVC is at least MSVC 2017

Closes: https://github.com/facebook/folly/issues/949

Reviewed By: yfeldblum

Differential Revision: D10355264

fbshipit-source-id: 52ac7489200285f06c5f4fb3d176f6427974b163

Summary:
Make StringPiece and MutableStringPiece implicitly convertible
to std::string_view.  This diff also adds FOLLY_HAS_STRING_VIEW to
Portability.h, which will be defined to be 0 or 1.

Reviewed By: vitaut

Differential Revision: D10187660

fbshipit-source-id: 90e6f3ac82e47eae423dab8f1de1d39e5ac997c4

Summary: [Folly] `exchange` in `DelayedDestruction` move-ctor.

Reviewed By: knekritz

Differential Revision: D10320776

fbshipit-source-id: b6b1f910b927098e61f6f6c8e414fdfcca843033

Summary: Clear the zerocopy maps in AsyncSocket::failAllWrites if the fd_ has been closed

Reviewed By: djwatson

Differential Revision: D10347608

fbshipit-source-id: a60d0450b13d4989d6849ca4cbe1057ecb6b92dd

Summary: Future::then to Future::thenValue to remove calls to deprecated functions.

Reviewed By: Orvid

Differential Revision: D10346695

fbshipit-source-id: 7f9641ef5b7f93ac4376a52724d42502eb1703b9

Summary: MSVC doesn't support inline assembly.

Reviewed By: yfeldblum, aary

Differential Revision: D10345334

fbshipit-source-id: 5b95d9f6bd8158fe426e01fa176ecfa732cf5bc3

Summary:
DistributedMutex is a small, exclusive-only mutex that distributes the
bookkeeping required for mutual exclusion in the stacks of threads that are
contending for it.  It tries to come at a lower space cost than std::mutex
while still trying to maintain the fairness benefits that come from using
std::mutex.  DistributedMutex provides the entire API included in
std::mutex, and more, with slight modifications.  DistributedMutex is the
same width as a single pointer (8 bytes on most platforms), where on the
other hand, std::mutex and pthread_mutex_t are both 40 bytes.  It is larger
than some of the other smaller locks, but the wide majority of cases using
the small locks are wasting the difference in alignment padding anyway

Benchmark results are good - at the time of writing in the common
uncontended case, it is 30% faster than some of the other small mutexes in
folly and as fast as std::mutex, which recently optimized its uncontended
path.  In the contended case, it is about 4-5x faster than some of the
smaller locks in folly, ~2x faster than std::mutex in clang and ~1.8x
faster in gcc.  DistributedMutex is also resistent to tail latency
pathalogies unlike many of the other small mutexes.  Which sleep for large
time quantums to reduce spin churn, this causes elevated latencies for
threads that enter the sleep cycle.  The tail latency of lock acquisition
on average up to 10x better with DistributedMutex

DistributedMutex reduces cache line contention by making each thread wait
on a thread local spinlock and futex.  This allows threads to keep working
only on their own cache lines without requiring cache coherence operations
when a mutex heavy contention.  This strategy does not require sequential
ordering on the centralized atomic storage for wakeup operations as each
thread assigned its own wait state

Non-timed mutex acquisitions are scheduled through intrusive LIFO
contention chains.  Each thread starts by spinning for a short quantum and
falls back to two phased sleeping.  Enqueue operations are lock free and
are piggybacked off mutex acquisition attempts.  The LIFO behavior of a
contention chain is good in the case where the mutex is held for a short
amount of time, as the head of the chain is likely to not have slept on
futex() after exhausting its spin quantum.  This allow us to avoid
unnecessary traversal and syscalls in the fast path with a higher
probability.  Even though the contention chains are LIFO, the mutex itself
does not adhere to that scheduling policy globally.  During contention,
threads that fail to lock the mutex form a LIFO chain on the central mutex
state, this chain is broken when a wakeup is scheduled, and future enqueue
operations form a new chain.  This makes the chains themselves LIFO, but
preserves global fairness through a constant factor which is limited to the
number of concurrent failed mutex acquisition attempts.  This binds the
last in first out behavior to the number of contending threads and helps
prevent starvation and latency outliers

This strategy of waking up wakers one by one in a queue does not scale well
when the number of threads goes past the number of cores.  At which point
preemption causes elevated lock acquisition latencies.  DistributedMutex
implements a hardware timestamp publishing heuristic to detect and adapt to
preemption.

DistributedMutex does not have the typical mutex API - it does not satisfy
the Lockable concept.  It requires the user to maintain ephemeral
bookkeeping and pass that bookkeeping around to unlock() calls.  The API
overhead, however, comes for free when you wrap this mutex for usage with
folly::Synchronized or std::unique_lock, which is the recommended usage
(std::lock_guard, in optimized mode, has no performance benefit over
std::unique_lock, so has been omitted).  A benefit of this API is that it
disallows incorrect usage where a thread unlocks a mutex that it does not
own, thinking a mutex is functionally identical to a binary semaphore,
which, unlike a mutex, is a suitable primitive for that usage

Timed locking through DistributedMutex is implemented through a centralized
algorithm - all waiters wait on the central mutex state, by setting and
resetting bits within the pointer-length word.  Since pointer length atomic
integers are incompatible with futex(FUTEX_WAIT) on most systems, a
non-standard implementation of futex() is used, where wait queues are
managed in user-space.  See p1135r0 and folly::ParkingLot

Reviewed By: djwatson

Differential Revision: D8949918

fbshipit-source-id: a772a70114e943ff68525c990da45e32ad1a5077

Summary:
[Folly] Reimplement `SCOPE_SUCCESS` and `SCOPE_FAIL` on top of `ScopeGuardImpl`.

Allows the underlying type to be safely move-constructed, in case the compiler does not do RVO copy elision in pre-C++17 builds.

Fixes #940.

Reviewed By: ericniebler

Differential Revision: D10120199

fbshipit-source-id: 889b867041338d6b76a0af5f04076b16e3120f63

Summary:
[Folly] Disable `-Wdeprecated-declarations` in cmake builds.

To hide deprecation spam in unit-tests for uses of deprecated declarations which cannot yet be removed.

Reviewed By: Orvid

Differential Revision: D10287167

fbshipit-source-id: 2dddd9f69ca1dcced61e4a2d6cd16bad54b2a2ad

Summary:
Remove the form of Future::then that could take a continuation with no parameters from elsewhere in Future.h as it is deprecated, and pending deletion.

In particular this required splitting the multi forms of the functions which still allow flexible typing at this point in time.

Reviewed By: yfeldblum

Differential Revision: D10206118

fbshipit-source-id: bb2ae241d3bfd5b89cae077e667cb39c4cc62797

Summary: Do not check zeroCopyEnabled_ in AsyncSocket::isZeroCopyMsg to avoid an IOBuf leak when zeroCopyEnabled_ changes from true to false

Reviewed By: yfeldblum

Differential Revision: D10285664

fbshipit-source-id: 5f13426f6140668d029a89786eb9a92466d4b1f4

Summary:
I was reading through about the various approaches to small-string
optimization: https://shaharmike.com/cpp/std-string/.
I noticed that Clang can store up to 22 bytes inline but FBString claims it
can do 23, so I read through the code to figure out where the extra byte came
from.

Specifically, I wasn't sure how it could have space to store the size as well
as ensure the buffer was null-terminated. After playing with the code for a bit
(this was further complicated because running the code under ASAN changes the
behavior, which I didn't realize before I started this exploration), I saw
how we don't store the size, but `maxSmallSize - size`, so that null-termination
works out even when `size==23`.

This updates the docs to hopefully save someone else this same exploration.

(Note: this ignores all push blocking failures!)

Reviewed By: ot

Differential Revision: D10258831

fbshipit-source-id: bfc0dd7ae55518af4173625bd719cfd4778180cc

Summary: [Folly] Fix cmake tests build after {D10244149}, which added `instructions::dispatch`.

Reviewed By: ot

Differential Revision: D10287193

fbshipit-source-id: f555229c8fcebccf61cb5577f5bf3edee9c46163

Summary:
[Folly] DRY template args in `Future::thenImplementation`.

Explicitly passing the template arguments which are also the types of non-template arguments is weird. Either limit the explicit template arguments to those which are not also the types of non-template arguments, or pass everything as non-template arguments.

Reviewed By: andriigrynenko

Differential Revision: D10247910

fbshipit-source-id: 64a56c5ea2072803781143b5247145ec2ee09ae0

Summary: as described in title

Reviewed By: Orvid

Differential Revision: D10279251

fbshipit-source-id: 0ab66033193026ccb0b6a06e57557a822ccb8a70

Summary:
- A couple of tests pragma push/pop to avoid `-Wself-move` warning in
  code that does a move-assignment with itself.
- Use `static_cast<T&&>` instead of `std::move` which will suppress
  the warning still without the need for pragma push/pop.
Pull Request resolved: https://github.com/facebook/folly/pull/946

Reviewed By: Orvid

Differential Revision: D10253159

Pulled By: yfeldblum

fbshipit-source-id: 9d75db95018e6115a195f246dbd93b0a5935d80d

Summary: [Folly] Enable `Fingerprint` test in CMake builds.

Reviewed By: Orvid

Differential Revision: D10081558

fbshipit-source-id: e4f753feed5b2698dd485cf32403c1c4d15d0c25

Summary:
TSAN complains about a lock inversion between folly/io/async/Request.cpp
lines 242 and 277.  The code as 242 was changed to use folly::acquireLocked() in
D9797715, but the code at 277 needs to be changed in the same way (except for
requesting a write lock on the child)

Reviewed By: kennyyu

Differential Revision: D10251470

fbshipit-source-id: c476a3664bfe83edbddbd402cf568eac885b9123

Summary:
These were failing in the watchman build

```
$ cd build && python fbcode_builder/shell_builder.py > ~/run.sh && bash ~/run.sh
Traceback (most recent call last):
  File "fbcode_builder/shell_builder.py", line 110, in <module>
    steps = make_steps(builder)
  File "/home/travis/build/facebook/watchman/build/fbcode_builder/utils.py", line 94, in <lambda>
    steps_for_spec(builder, config['fbcode_builder_spec'](builder))
  File "/home/travis/build/facebook/watchman/build/fbcode_builder/fbcode_builder.py", line 144, in build
    return [self.setup(), self.diagnostics()] + steps
  File "fbcode_builder/shell_builder.py", line 61, in setup
    ).format(ccache_dir=ccache_dir)
  File "/home/travis/build/facebook/watchman/build/fbcode_builder/shell_quoting.py", line 64, in format
    (k, shell_quote(v).do_not_use_raw_str) for k, v in kwargs.items()
KeyError: u'CC'
```

Reviewed By: snarkmaster

fbshipit-source-id: 614723e631a82f277739765a920731c872700c45

Summary: Add a simple function to dispatch the `instructions` types based on current CPU (overridable in tests).

Reviewed By: luciang

Differential Revision: D10244149

fbshipit-source-id: ac0c1ff0d14402fa77c939361398db6122dea728

Summary: For determining proper use of `strerror_r` based on available decl, move messy `#if` logic out of `String.cpp` and into buck config.

Reviewed By: yfeldblum

Differential Revision: D10243924

fbshipit-source-id: 6c10fc43562765c158d13ed74961277b633bb606

Summary:
- LLVM 7 emits a warning about self-assignment.
- Silence the warning by using a static_cast.
Pull Request resolved: https://github.com/facebook/folly/pull/945

Reviewed By: Orvid

Differential Revision: D10236656

Pulled By: yfeldblum

fbshipit-source-id: 43b46563357d89db9e0911b894f8fead0a14d660

Summary:
Apply fixes that work with new unified headers and remain compatible with old per-API+arch headers.

I do this to unclutter the diff that enables unified headers, so that it only contains things that are incompatible with old per-API+arch headers.

Reviewed By: yfeldblum

Differential Revision: D10225372

fbshipit-source-id: b92cb3ef45cf23f2233c39520236c4531dba715b

Summary:
Use `FOLLY_ATTR_WEAK` for the annotate functions.

(Note: this ignores all push blocking failures!)

Differential Revision: D10234953

fbshipit-source-id: 534cb1d2d7d6053bc000e4c7ae3c6df16fc7a958

Summary:
`shell_builder.py` allows running the fbcode_builder logic
on the host rather than in a container.

It emits a bash script with `set -exo pipefail` configured such that
any failing step will cause the script to exit with failure.

Refs: https://github.com/facebook/watchman/pull/639

Reviewed By: simpkins

Differential Revision: D9552411

fbshipit-source-id: c7835deedf07ea342dcb3de61d576a4fb5439985

Summary: As titled, purely for uniformity.

Reviewed By: yfeldblum

Differential Revision: D10098066

fbshipit-source-id: 12d78fcc60805284d3d85f405475e63a85799041

Summary:
This adds a missing feature. Nested exceptions are cool, see examples in https://en.cppreference.com/w/cpp/error/rethrow_if_nested (bottom of the page).

In short, nested exceptions allow people to add "contextual" information to exceptions. Previously we had to write the `try { ew.throw_exception(); } catch (...) { std::throw_with_nested(xxx); }` boilerplate if we used an exception_wrapper (e.g. in futures-oriented code); with this diff, we're able to just run `ew.throw_with_nested(xxx)`.

Reviewed By: yfeldblum

Differential Revision: D10219994

fbshipit-source-id: cdf0fbe8e4608f6c87d326631e8ffb4919c20711

Summary: Pull Request resolved: https://github.com/facebook/folly/pull/944

Reviewed By: Orvid

Differential Revision: D10231693

Pulled By: yfeldblum

fbshipit-source-id: 9c0c5d7c4c1e4d9d9860c71e608e47d661a41fdd

Summary:
icc accepts the syntax, but doesn't emit the definition into the
object file.

Reviewed By: ricklavoie, yfeldblum

Differential Revision: D10228495

fbshipit-source-id: 73c1676deb4963a9c30886414cc1722d44a79214

Summary:
Remove falls to Future::then that take a nullary lambda. This is deprecated and will be deleted.

This change replaces them with thenValue([](auto&&){...}) or similar.

This does not replace all calls to then with thenValue - other forms will be replaced when those forms of then are due for deletion.

Reviewed By: yfeldblum

Differential Revision: D10219070

fbshipit-source-id: 13b6f64ee2d7c8741537fe1131bdf3b251361413

Summary:
Similar to `--dynamic_cputhreadpoolexecutor`, this is a temporary
kill switch until all outstanding issues are addressed that would allow us
to use dynamic thread pools unconditionally.

Reviewed By: djwatson

Differential Revision: D10231230

fbshipit-source-id: b0abd77ff2a6a7ce31727e14eca799a6f3b5535a

Summary:
There were some read-modify-write operations that were behaving as read and
then modify.  This is incorrect when you use DeterministicAtomic without a
backing schedule.

This is okay if the test is operating in the presence of a
DeterministicSchedule schedule, but is probably worth fixing anyway.  If using
DeterministicAtomic without a backing schedule is considered a bug, we can add
an assertion in beforeSharedAccess() and afterSharedAccess() so users know not
to try that

Reviewed By: djwatson, nbronson

Differential Revision: D10190682

fbshipit-source-id: 6a3173d904a40bbc5c96b934a76aff90a48c8608

Summary:
icc gets confused on the parameter unpacking otherwise.

https://godbolt.org/z/rSRZYn

Reviewed By: yfeldblum

Differential Revision: D10147211

fbshipit-source-id: 85de581ffa4c487523a5e9f173636f74863a82ac

Summary:
When they come before the explicit instantiations of Rob, icc fails to
compile certain combintions of calls.

Reviewed By: alexeyt

Differential Revision: D10144943

fbshipit-source-id: ce63a4d289f64d1d644b116ab99b45b9da013162

Summary:
`folly::SharedMutex` allows `unlock_shared()` to be called in a different thread from the one that locked it
There are some places in folly that use this property.

Example: In `ObserverManager`, the calling thread

1. acquires the read lock
2. then it schedules some async work and gives ownership of the lock guard to the lambda
3. the thread that performs the async work will unlock the mutex

However, this causes false positives for TSAN. It looks like this breaks TSAN's assumptions that a read-write mutex is always read locked and unlocked in the same thread.
In the example above, TSAN thinks the calling thread in step (1) still owns the mutex, even though semantically the mutex is now owned by the thread in step (3).

This results in false positives for running unit tests with an annotated version of `folly::SharedMutex`.
TSAN reports a lock inversion between the `folly::Singleton vaule_.state_` mutex, and the read lock referenced in `ObserverManager`.

To suppress this error, this change annotates ObserverManager to let TSAN know that the the thread in step 1 has dropped
the lock, and the thread in step 3 now owns the lock.

Reviewed By: andriigrynenko

Differential Revision: D9797937

fbshipit-source-id: c0e5277bb9cb2f404df8abd3c3d6ea4a16460c78

Summary:
TSAN flagged the following lock inversion below. The reason for this lock inversion
is because of the following example usage:

1. set request context B, save old one A (lock B, lock A)
2. later on, reset request context back to A, get the old one back B (lock A, lock B)

In the two calls, we always locked the the new context first, resulting
in a lock inversion. The fix is to always acquire them in a consistent
order using `folly::acquireLocked`

TSAN report:

```
WARNING: ThreadSanitizer: lock-order-inversion (potential deadlock) (pid=1574307)
  Cycle in lock order graph: M159 (0x7b1400000138) => M157 (0x7b14000000e8) => M159

  Mutex M157 acquired here while holding mutex M159 in main thread:
    #0 0x48f18c in AnnotateRWLockAcquired ??:?
    #1 0x315495 in folly::SharedMutexImpl<false, void, std::atomic, false>::lock_shared() ./folly/SharedMutex.h:385
    #2 0x1b0eb8 in folly::detail::LockTraitsImpl<folly::SharedMutexImpl<false, void, std::atomic, false>, (folly::detail::MutexLevel)1, false>::lock_shared(folly::SharedMutexImpl<false, void, std::atomic, false>&) ./folly/LockTraits.h:157
    #3 0x1b0e68 in std::integral_constant<bool, true> folly::LockPolicyShared::lock<folly::SharedMutexImpl<false, void, std::atomic, false> >(folly::SharedMutexImpl<false, void, std::atomic, false>&) ./folly/LockTraits.h:493
    #4 0x1d63a9 in LockedPtrBase ./folly/Synchronized.h:1026
    #5 0x1d6258 in LockedPtr ./folly/Synchronized.h:1326
    #6 0x1c9a17 in folly::SynchronizedBase<folly::Synchronized<folly::RequestContext::State, folly::SharedMutexImpl<false, void, std::atomic, false> >, (folly::detail::MutexLevel)1>::rlock() const ./folly/Synchronized.h:123
    #7 0x1c7530 in folly::RequestContext::setContext(std::shared_ptr<folly::RequestContext>) ./folly/io/async/Request.cpp:243
    #8 0x36a7b in folly::RequestContext::create() ./folly/io/async/Request.h:119
    #9 0x34916 in RequestContextScopeGuard ./folly/io/async/Request.h:269

  Mutex M159 previously acquired by the same thread here:
    #0 0x48f18c in AnnotateRWLockAcquired ??:?
    #1 0x315495 in folly::SharedMutexImpl<false, void, std::atomic, false>::lock_shared() ./folly/SharedMutex.h:385
    #2 0x1b0eb8 in folly::detail::LockTraitsImpl<folly::SharedMutexImpl<false, void, std::atomic, false>, (folly::detail::MutexLevel)1, false>::lock_shared(folly::SharedMutexImpl<false, void, std::atomic, false>&) ./folly/LockTraits.h:157
    #3 0x1b0e68 in std::integral_constant<bool, true> folly::LockPolicyShared::lock<folly::SharedMutexImpl<false, void, std::atomic, false> >(folly::SharedMutexImpl<false, void, std::atomic, false>&) ./folly/LockTraits.h:493
    #4 0x1d63a9 in LockedPtrBase ./folly/Synchronized.h:1026
    #5 0x1d6258 in LockedPtr ./folly/Synchronized.h:1326
    #6 0x1c9a17 in folly::SynchronizedBase<folly::Synchronized<folly::RequestContext::State, folly::SharedMutexImpl<false, void, std::atomic, false> >, (folly::detail::MutexLevel)1>::rlock() const ./folly/Synchronized.h:123
    #7 0x1c7516 in folly::RequestContext::setContext(std::shared_ptr<folly::RequestContext>) ./folly/io/async/Request.cpp:242
    #8 0x36a7b in folly::RequestContext::create() ./folly/io/async/Request.h:119
    #9 0x34916 in RequestContextScopeGuard ./folly/io/async/Request.h:269

  Mutex M159 acquired here while holding mutex M157 in main thread:
    #0 0x48f18c in AnnotateRWLockAcquired ??:?
    #1 0x315495 in folly::SharedMutexImpl<false, void, std::atomic, false>::lock_shared() ./folly/SharedMutex.h:385
    #2 0x1b0eb8 in folly::detail::LockTraitsImpl<folly::SharedMutexImpl<false, void, std::atomic, false>, (folly::detail::MutexLevel)1, false>::lock_shared(folly::SharedMutexImpl<false, void, std::atomic, false>&) ./folly/LockTraits.h:157
    #3 0x1b0e68 in std::integral_constant<bool, true> folly::LockPolicyShared::lock<folly::SharedMutexImpl<false, void, std::atomic, false> >(folly::SharedMutexImpl<false, void, std::atomic, false>&) ./folly/LockTraits.h:493
    #4 0x1d63a9 in LockedPtrBase ./folly/Synchronized.h:1026
    #5 0x1d6258 in LockedPtr ./folly/Synchronized.h:1326
    #6 0x1c9a17 in folly::SynchronizedBase<folly::Synchronized<folly::RequestContext::State, folly::SharedMutexImpl<false, void, std::atomic, false> >, (folly::detail::MutexLevel)1>::rlock() const ./folly/Synchronized.h:123
    #7 0x1c7530 in folly::RequestContext::setContext(std::shared_ptr<folly::RequestContext>) ./folly/io/async/Request.cpp:243
    #8 0x34b99 in ~RequestContextScopeGuard ./folly/io/async/Request.h:278

  Mutex M157 previously acquired by the same thread here:
    #0 0x48f18c in AnnotateRWLockAcquired ??:?
    #1 0x315495 in folly::SharedMutexImpl<false, void, std::atomic, false>::lock_shared() ./folly/SharedMutex.h:385
    #2 0x1b0eb8 in folly::detail::LockTraitsImpl<folly::SharedMutexImpl<false, void, std::atomic, false>, (folly::detail::MutexLevel)1, false>::lock_shared(folly::SharedMutexImpl<false, void, std::atomic, false>&) ./folly/LockTraits.h:157
    #3 0x1b0e68 in std::integral_constant<bool, true> folly::LockPolicyShared::lock<folly::SharedMutexImpl<false, void, std::atomic, false> >(folly::SharedMutexImpl<false, void, std::atomic, false>&) ./folly/LockTraits.h:493
    #4 0x1d63a9 in LockedPtrBase ./folly/Synchronized.h:1026
    #5 0x1d6258 in LockedPtr ./folly/Synchronized.h:1326
    #6 0x1c9a17 in folly::SynchronizedBase<folly::Synchronized<folly::RequestContext::State, folly::SharedMutexImpl<false, void, std::atomic, false> >, (folly::detail::MutexLevel)1>::rlock() const ./folly/Synchronized.h:123
    #7 0x1c7516 in folly::RequestContext::setContext(std::shared_ptr<folly::RequestContext>) ./folly/io/async/Request.cpp:242
    #8 0x34b99 in ~RequestContextScopeGuard ./folly/io/async/Request.h:278

```

Reviewed By: igorsugak

Differential Revision: D9797715

fbshipit-source-id: a0f991ae0ec8f3e9d33af6e05db49eceaf3a5174

Summary:
There is a lock inversion between VirtualEventBase and the fibers library:

1. During program shutdown, `GlobalCache::getImpl` acquires its `mutex_`, then it calls `VirtualEventBase::runOnDestruction` which acquires the SharedMutex on `onDestructionCallbacks_`
2. During normal program execution, `VirtualEventBase::destroyImpl()` acquires the `onDestructionCallbacks_` lock and then invokes the user-supplied callbacks, which will invoke `GlobalCache::eraseImpl` and acquire `mutex_`

The reason for this lock inversion is because in step 2, we are holding a mutex while invoking user-supplied callbacks.
The fix is to NOT hold the lock while we invoke these callbacks.

Reviewed By: andriigrynenko

Differential Revision: D9797816

fbshipit-source-id: 861a915f22138f0c8d3f8bca4d7bf4b9a0aa3d26

Summary:
When running tests with a TSAN-annotated version of folly::SharedMutex,
I discovered at least 3 types of lock inversions in folly::Singleton code.
These inversions are probably benign and deadlock is not possible, but we need to
suppress these inversions to prevent too much noise from TSAN.

Lock inversions:

1. The users can supply arbitrary create() functions for Singletons. If the create() function
   happens to acquire a mutex, then this introduces a lock ordering from
   SingletonVault -> mutex. Later on, when the user code runs in normal operations, we might
   hold the mutex and try to create a singleton, resulting in the mutex -> SingletonVault ordering
2. SingletonVault::destroyInstances() holds a mutex, and then invokes the user-supplied destroy()
   functions. In the user supplied destroy functions, we might attempt to create a singleton and
   acquire the lock in SingletonHolder, resulting in SingletonVault -> SingletonHolder lock ordering.
   In normal operations, whenever a singleton is created, we have the SingletonHolder -> SingletonVault
   lock ordering in createInstance()
3. doEagerInit() holds the eagerInitSingleton mutex and then acquires the SingletonVault lock
   during createInstance(). addEagerInitSingleton() holds the SingletonVault lock and then acquires
   the eagerInitSingleton lock.

The source of these errors is that we are invoking user-supplied callbacks while holding mutexes.
However, these lock inversions cannot actually deadlock based on higher-level knowledge of folly::Singleton
state transitions.

To suppress these errors, use the new `folly::SharedMutexSuppressTSAN` mutex instead, which will
not be annotated by TSAN. If we are not building with TSAN, then this mutex is equivalent to a normal
`folly::SharedMutex`.

Reviewed By: andriigrynenko

Differential Revision: D9797988

fbshipit-source-id: 82b5850fe189f7d1dcaca0e3562fabcfafd0cef5