Export of internal Abseil changes

--
465461299a9814aca325fee599cefbfe462f12fe by Abseil Team <absl-team@google.com>:

Optimize trivially copyable flags with a sequence lock

PiperOrigin-RevId: 349602779

--
73f39f959e21121684a51887243abad0814a335e by Abseil Team <absl-team@google.com>:

Internal change

PiperOrigin-RevId: 349590869

--
6b3106fa66b8f075a39a1a8f3265ae132b7e2c84 by Abseil Team <absl-team@google.com>:

Remove ABSL_DLL from `log_prefix_hook` and `abort_hook`.

PiperOrigin-RevId: 349560499

--
bb0d295e699a509f3284145e025d00036b70dbb2 by Abseil Team <absl-team@google.com>:

Tiny docstring fix

A small edit to make "use of this is useful" a little less redundant. :)

PiperOrigin-RevId: 349445689
GitOrigin-RevId: 465461299a9814aca325fee599cefbfe462f12fe
Change-Id: I08cc4091b8b95b68188cb9168ac622dacc5fa688

4 files changed, 425 insertions, 108 deletions

diff --git a/absl/flags/internal/flag.cc b/absl/flags/internal/flag.cc
index 1502e7f1..f83c1fe7 100644
--- a/absl/flags/internal/flag.cc
+++ b/absl/flags/internal/flag.cc
@@ -96,7 +96,8 @@ class FlagState : public flags_internal::FlagStateInterface {
         counter_(counter) {}
 
   ~FlagState() override {
-    if (flag_impl_.ValueStorageKind() != FlagValueStorageKind::kAlignedBuffer)
+    if (flag_impl_.ValueStorageKind() != FlagValueStorageKind::kAlignedBuffer &&
+        flag_impl_.ValueStorageKind() != FlagValueStorageKind::kSequenceLocked)
       return;
     flags_internal::Delete(flag_impl_.op_, value_.heap_allocated);
   }
@@ -118,11 +119,9 @@ class FlagState : public flags_internal::FlagStateInterface {
   union SavedValue {
     explicit SavedValue(void* v) : heap_allocated(v) {}
     explicit SavedValue(int64_t v) : one_word(v) {}
-    explicit SavedValue(flags_internal::AlignedTwoWords v) : two_words(v) {}
 
     void* heap_allocated;
     int64_t one_word;
-    flags_internal::AlignedTwoWords two_words;
   } value_;
   bool modified_;
   bool on_command_line_;
@@ -164,17 +163,15 @@ void FlagImpl::Init() {
                            std::memory_order_release);
       break;
     }
-    case FlagValueStorageKind::kTwoWordsAtomic: {
+    case FlagValueStorageKind::kSequenceLocked: {
       // For this storage kind the default_value_ always points to gen_func
       // during initialization.
       assert(def_kind == FlagDefaultKind::kGenFunc);
-      alignas(AlignedTwoWords) std::array<char, sizeof(AlignedTwoWords)> buf{};
-      (*default_value_.gen_func)(buf.data());
-      auto atomic_value = absl::bit_cast<AlignedTwoWords>(buf);
-      TwoWordsValue().store(atomic_value, std::memory_order_release);
+      (*default_value_.gen_func)(AtomicBufferValue());
       break;
     }
   }
+  seq_lock_.MarkInitialized();
 }
 
 absl::Mutex* FlagImpl::DataGuard() const {
@@ -231,23 +228,21 @@ void FlagImpl::StoreValue(const void* src) {
   switch (ValueStorageKind()) {
     case FlagValueStorageKind::kAlignedBuffer:
       Copy(op_, src, AlignedBufferValue());
+      seq_lock_.IncrementModificationCount();
       break;
     case FlagValueStorageKind::kOneWordAtomic: {
       int64_t one_word_val = 0;
       std::memcpy(&one_word_val, src, Sizeof(op_));
       OneWordValue().store(one_word_val, std::memory_order_release);
+      seq_lock_.IncrementModificationCount();
       break;
     }
-    case FlagValueStorageKind::kTwoWordsAtomic: {
-      AlignedTwoWords two_words_val{0, 0};
-      std::memcpy(&two_words_val, src, Sizeof(op_));
-      TwoWordsValue().store(two_words_val, std::memory_order_release);
+    case FlagValueStorageKind::kSequenceLocked: {
+      seq_lock_.Write(AtomicBufferValue(), src, Sizeof(op_));
       break;
     }
   }
-
   modified_ = true;
-  ++counter_;
   InvokeCallback();
 }
 
@@ -266,6 +261,10 @@ FlagFastTypeId FlagImpl::TypeId() const {
   return flags_internal::FastTypeId(op_);
 }
 
+int64_t FlagImpl::ModificationCount() const {
+  return seq_lock_.ModificationCount();
+}
+
 bool FlagImpl::IsSpecifiedOnCommandLine() const {
   absl::MutexLock l(DataGuard());
   return on_command_line_;
@@ -291,11 +290,11 @@ std::string FlagImpl::CurrentValue() const {
               OneWordValue().load(std::memory_order_acquire));
       return flags_internal::Unparse(op_, one_word_val.data());
     }
-    case FlagValueStorageKind::kTwoWordsAtomic: {
-      const auto two_words_val =
-          absl::bit_cast<std::array<char, sizeof(AlignedTwoWords)>>(
-              TwoWordsValue().load(std::memory_order_acquire));
-      return flags_internal::Unparse(op_, two_words_val.data());
+    case FlagValueStorageKind::kSequenceLocked: {
+      std::unique_ptr<void, DynValueDeleter> cloned(flags_internal::Alloc(op_),
+                                                    DynValueDeleter{op_});
+      ReadSequenceLockedData(cloned.get());
+      return flags_internal::Unparse(op_, cloned.get());
     }
   }
 
@@ -345,17 +344,22 @@ std::unique_ptr<FlagStateInterface> FlagImpl::SaveState() {
     case FlagValueStorageKind::kAlignedBuffer: {
       return absl::make_unique<FlagState>(
           *this, flags_internal::Clone(op_, AlignedBufferValue()), modified,
-          on_command_line, counter_);
+          on_command_line, ModificationCount());
     }
     case FlagValueStorageKind::kOneWordAtomic: {
       return absl::make_unique<FlagState>(
           *this, OneWordValue().load(std::memory_order_acquire), modified,
-          on_command_line, counter_);
+          on_command_line, ModificationCount());
     }
-    case FlagValueStorageKind::kTwoWordsAtomic: {
-      return absl::make_unique<FlagState>(
-          *this, TwoWordsValue().load(std::memory_order_acquire), modified,
-          on_command_line, counter_);
+    case FlagValueStorageKind::kSequenceLocked: {
+      void* cloned = flags_internal::Alloc(op_);
+      // Read is guaranteed to be successful because we hold the lock.
+      bool success =
+          seq_lock_.TryRead(cloned, AtomicBufferValue(), Sizeof(op_));
+      assert(success);
+      static_cast<void>(success);
+      return absl::make_unique<FlagState>(*this, cloned, modified,
+                                          on_command_line, ModificationCount());
     }
   }
   return nullptr;
@@ -363,21 +367,18 @@ std::unique_ptr<FlagStateInterface> FlagImpl::SaveState() {
 
 bool FlagImpl::RestoreState(const FlagState& flag_state) {
   absl::MutexLock l(DataGuard());
-
-  if (flag_state.counter_ == counter_) {
+  if (flag_state.counter_ == ModificationCount()) {
     return false;
   }
 
   switch (ValueStorageKind()) {
     case FlagValueStorageKind::kAlignedBuffer:
+    case FlagValueStorageKind::kSequenceLocked:
       StoreValue(flag_state.value_.heap_allocated);
       break;
     case FlagValueStorageKind::kOneWordAtomic:
       StoreValue(&flag_state.value_.one_word);
       break;
-    case FlagValueStorageKind::kTwoWordsAtomic:
-      StoreValue(&flag_state.value_.two_words);
-      break;
   }
 
   modified_ = flag_state.modified_;
@@ -400,16 +401,16 @@ void* FlagImpl::AlignedBufferValue() const {
   return OffsetValue<void>();
 }
 
+std::atomic<uint64_t>* FlagImpl::AtomicBufferValue() const {
+  assert(ValueStorageKind() == FlagValueStorageKind::kSequenceLocked);
+  return OffsetValue<std::atomic<uint64_t>>();
+}
+
 std::atomic<int64_t>& FlagImpl::OneWordValue() const {
   assert(ValueStorageKind() == FlagValueStorageKind::kOneWordAtomic);
   return OffsetValue<FlagOneWordValue>()->value;
 }
 
-std::atomic<AlignedTwoWords>& FlagImpl::TwoWordsValue() const {
-  assert(ValueStorageKind() == FlagValueStorageKind::kTwoWordsAtomic);
-  return OffsetValue<FlagTwoWordsValue>()->value;
-}
-
 // Attempts to parse supplied `value` string using parsing routine in the `flag`
 // argument. If parsing successful, this function replaces the dst with newly
 // parsed value. In case if any error is encountered in either step, the error
@@ -443,15 +444,27 @@ void FlagImpl::Read(void* dst) const {
       std::memcpy(dst, &one_word_val, Sizeof(op_));
       break;
     }
-    case FlagValueStorageKind::kTwoWordsAtomic: {
-      const AlignedTwoWords two_words_val =
-          TwoWordsValue().load(std::memory_order_acquire);
-      std::memcpy(dst, &two_words_val, Sizeof(op_));
+    case FlagValueStorageKind::kSequenceLocked: {
+      ReadSequenceLockedData(dst);
       break;
     }
   }
 }
 
+void FlagImpl::ReadSequenceLockedData(void* dst) const {
+  int size = Sizeof(op_);
+  // Attempt to read using the sequence lock.
+  if (ABSL_PREDICT_TRUE(seq_lock_.TryRead(dst, AtomicBufferValue(), size))) {
+    return;
+  }
+  // We failed due to contention. Acquire the lock to prevent contention
+  // and try again.
+  absl::ReaderMutexLock l(DataGuard());
+  bool success = seq_lock_.TryRead(dst, AtomicBufferValue(), size);
+  assert(success);
+  static_cast<void>(success);
+}
+
 void FlagImpl::Write(const void* src) {
   absl::MutexLock l(DataGuard());
 
diff --git a/absl/flags/internal/flag.h b/absl/flags/internal/flag.h
index 370d8a02..83548143 100644
--- a/absl/flags/internal/flag.h
+++ b/absl/flags/internal/flag.h
@@ -36,6 +36,7 @@
 #include "absl/flags/config.h"
 #include "absl/flags/internal/commandlineflag.h"
 #include "absl/flags/internal/registry.h"
+#include "absl/flags/internal/sequence_lock.h"
 #include "absl/flags/marshalling.h"
 #include "absl/meta/type_traits.h"
 #include "absl/strings/string_view.h"
@@ -308,59 +309,23 @@ using FlagUseOneWordStorage = std::integral_constant<
     bool, absl::type_traits_internal::is_trivially_copyable<T>::value &&
               (sizeof(T) <= 8)>;
 
-#if defined(ABSL_FLAGS_INTERNAL_ATOMIC_DOUBLE_WORD)
-// Clang does not always produce cmpxchg16b instruction when alignment of a 16
-// bytes type is not 16.
-struct alignas(16) AlignedTwoWords {
-  int64_t first;
-  int64_t second;
-
-  bool IsInitialized() const {
-    return first != flags_internal::UninitializedFlagValue();
-  }
-};
-
-template <typename T>
-using FlagUseTwoWordsStorage = std::integral_constant<
+template <class T>
+using FlagShouldUseSequenceLock = std::integral_constant<
     bool, absl::type_traits_internal::is_trivially_copyable<T>::value &&
-              (sizeof(T) > 8) && (sizeof(T) <= 16)>;
-#else
-// This is actually unused and only here to avoid ifdefs in other palces.
-struct AlignedTwoWords {
-  constexpr AlignedTwoWords() noexcept : dummy() {}
-  constexpr AlignedTwoWords(int64_t, int64_t) noexcept : dummy() {}
-  char dummy;
-
-  bool IsInitialized() const {
-    std::abort();
-    return true;
-  }
-};
-
-// This trait should be type dependent, otherwise SFINAE below will fail
-template <typename T>
-using FlagUseTwoWordsStorage =
-    std::integral_constant<bool, sizeof(T) != sizeof(T)>;
-#endif
-
-template <typename T>
-using FlagUseBufferStorage =
-    std::integral_constant<bool, !FlagUseOneWordStorage<T>::value &&
-                                     !FlagUseTwoWordsStorage<T>::value>;
+              (sizeof(T) > 8)>;
 
 enum class FlagValueStorageKind : uint8_t {
   kAlignedBuffer = 0,
   kOneWordAtomic = 1,
-  kTwoWordsAtomic = 2
+  kSequenceLocked = 2,
 };
 
 template <typename T>
 static constexpr FlagValueStorageKind StorageKind() {
-  return FlagUseBufferStorage<T>::value
-             ? FlagValueStorageKind::kAlignedBuffer
-             : FlagUseOneWordStorage<T>::value
-                   ? FlagValueStorageKind::kOneWordAtomic
-                   : FlagValueStorageKind::kTwoWordsAtomic;
+  return FlagUseOneWordStorage<T>::value ? FlagValueStorageKind::kOneWordAtomic
+         : FlagShouldUseSequenceLock<T>::value
+             ? FlagValueStorageKind::kSequenceLocked
+             : FlagValueStorageKind::kAlignedBuffer;
 }
 
 struct FlagOneWordValue {
@@ -369,27 +334,20 @@ struct FlagOneWordValue {
   std::atomic<int64_t> value;
 };
 
-struct FlagTwoWordsValue {
-  constexpr FlagTwoWordsValue()
-      : value(AlignedTwoWords{UninitializedFlagValue(), 0}) {}
-
-  std::atomic<AlignedTwoWords> value;
-};
-
 template <typename T,
           FlagValueStorageKind Kind = flags_internal::StorageKind<T>()>
 struct FlagValue;
 
 template <typename T>
 struct FlagValue<T, FlagValueStorageKind::kAlignedBuffer> {
-  bool Get(T&) const { return false; }
+  bool Get(const SequenceLock&, T&) const { return false; }
 
   alignas(T) char value[sizeof(T)];
 };
 
 template <typename T>
 struct FlagValue<T, FlagValueStorageKind::kOneWordAtomic> : FlagOneWordValue {
-  bool Get(T& dst) const {
+  bool Get(const SequenceLock&, T& dst) const {
     int64_t one_word_val = value.load(std::memory_order_acquire);
     if (ABSL_PREDICT_FALSE(one_word_val == UninitializedFlagValue())) {
       return false;
@@ -400,15 +358,16 @@ struct FlagValue<T, FlagValueStorageKind::kOneWordAtomic> : FlagOneWordValue {
 };
 
 template <typename T>
-struct FlagValue<T, FlagValueStorageKind::kTwoWordsAtomic> : FlagTwoWordsValue {
-  bool Get(T& dst) const {
-    AlignedTwoWords two_words_val = value.load(std::memory_order_acquire);
-    if (ABSL_PREDICT_FALSE(!two_words_val.IsInitialized())) {
-      return false;
-    }
-    std::memcpy(&dst, static_cast<const void*>(&two_words_val), sizeof(T));
-    return true;
+struct FlagValue<T, FlagValueStorageKind::kSequenceLocked> {
+  bool Get(const SequenceLock& lock, T& dst) const {
+    return lock.TryRead(&dst, value_words, sizeof(T));
   }
+
+  static constexpr int kNumWords =
+      flags_internal::AlignUp(sizeof(T), sizeof(uint64_t)) / sizeof(uint64_t);
+
+  alignas(T) alignas(
+      std::atomic<uint64_t>) std::atomic<uint64_t> value_words[kNumWords];
 };
 
 ///////////////////////////////////////////////////////////////////////////////
@@ -451,7 +410,6 @@ class FlagImpl final : public CommandLineFlag {
         def_kind_(static_cast<uint8_t>(default_arg.kind)),
         modified_(false),
         on_command_line_(false),
-        counter_(0),
         callback_(nullptr),
         default_value_(default_arg.source),
         data_guard_{} {}
@@ -498,15 +456,17 @@ class FlagImpl final : public CommandLineFlag {
   // flag.cc, we can define it in that file as well.
   template <typename StorageT>
   StorageT* OffsetValue() const;
-  // This is an accessor for a value stored in an aligned buffer storage.
+  // This is an accessor for a value stored in an aligned buffer storage
+  // used for non-trivially-copyable data types.
   // Returns a mutable pointer to the start of a buffer.
   void* AlignedBufferValue() const;
+
+  // The same as above, but used for sequencelock-protected storage.
+  std::atomic<uint64_t>* AtomicBufferValue() const;
+
   // This is an accessor for a value stored as one word atomic. Returns a
   // mutable reference to an atomic value.
   std::atomic<int64_t>& OneWordValue() const;
-  // This is an accessor for a value stored as two words atomic. Returns a
-  // mutable reference to an atomic value.
-  std::atomic<AlignedTwoWords>& TwoWordsValue() const;
 
   // Attempts to parse supplied `value` string. If parsing is successful,
   // returns new value. Otherwise returns nullptr.
@@ -516,6 +476,12 @@ class FlagImpl final : public CommandLineFlag {
   // Stores the flag value based on the pointer to the source.
   void StoreValue(const void* src) ABSL_EXCLUSIVE_LOCKS_REQUIRED(*DataGuard());
 
+  // Copy the flag data, protected by `seq_lock_` into `dst`.
+  //
+  // REQUIRES: ValueStorageKind() == kSequenceLocked.
+  void ReadSequenceLockedData(void* dst) const
+      ABSL_LOCKS_EXCLUDED(*DataGuard());
+
   FlagHelpKind HelpSourceKind() const {
     return static_cast<FlagHelpKind>(help_source_kind_);
   }
@@ -541,6 +507,8 @@ class FlagImpl final : public CommandLineFlag {
   void CheckDefaultValueParsingRoundtrip() const override
       ABSL_LOCKS_EXCLUDED(*DataGuard());
 
+  int64_t ModificationCount() const ABSL_EXCLUSIVE_LOCKS_REQUIRED(*DataGuard());
+
   // Interfaces to save and restore flags to/from persistent state.
   // Returns current flag state or nullptr if flag does not support
   // saving and restoring a state.
@@ -587,8 +555,9 @@ class FlagImpl final : public CommandLineFlag {
   // Unique tag for absl::call_once call to initialize this flag.
   absl::once_flag init_control_;
 
-  // Mutation counter
-  int64_t counter_ ABSL_GUARDED_BY(*DataGuard());
+  // Sequence lock / mutation counter.
+  flags_internal::SequenceLock seq_lock_;
+
   // Optional flag's callback and absl::Mutex to guard the invocations.
   FlagCallback* callback_ ABSL_GUARDED_BY(*DataGuard());
   // Either a pointer to the function generating the default value based on the
@@ -649,7 +618,9 @@ class Flag {
     impl_.AssertValidType(base_internal::FastTypeId<T>(), &GenRuntimeTypeId<T>);
 #endif
 
-    if (!value_.Get(u.value)) impl_.Read(&u.value);
+    if (ABSL_PREDICT_FALSE(!value_.Get(impl_.seq_lock_, u.value))) {
+      impl_.Read(&u.value);
+    }
     return std::move(u.value);
   }
   void Set(const T& v) {
diff --git a/absl/flags/internal/sequence_lock.h b/absl/flags/internal/sequence_lock.h
new file mode 100644
index 00000000..807b2a73
--- /dev/null
+++ b/absl/flags/internal/sequence_lock.h
@@ -0,0 +1,187 @@
+//
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#ifndef ABSL_FLAGS_INTERNAL_SEQUENCE_LOCK_H_
+#define ABSL_FLAGS_INTERNAL_SEQUENCE_LOCK_H_
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include <atomic>
+#include <cassert>
+#include <cstring>
+
+#include "absl/base/optimization.h"
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+namespace flags_internal {
+
+// Align 'x' up to the nearest 'align' bytes.
+inline constexpr size_t AlignUp(size_t x, size_t align) {
+  return align * ((x + align - 1) / align);
+}
+
+// A SequenceLock implements lock-free reads. A sequence counter is incremented
+// before and after each write, and readers access the counter before and after
+// accessing the protected data. If the counter is verified to not change during
+// the access, and the sequence counter value was even, then the reader knows
+// that the read was race-free and valid. Otherwise, the reader must fall back
+// to a Mutex-based code path.
+//
+// This particular SequenceLock starts in an "uninitialized" state in which
+// TryRead() returns false. It must be enabled by calling MarkInitialized().
+// This serves as a marker that the associated flag value has not yet been
+// initialized and a slow path needs to be taken.
+//
+// The memory reads and writes protected by this lock must use the provided
+// `TryRead()` and `Write()` functions. These functions behave similarly to
+// `memcpy()`, with one oddity: the protected data must be an array of
+// `std::atomic<int64>`. This is to comply with the C++ standard, which
+// considers data races on non-atomic objects to be undefined behavior. See "Can
+// Seqlocks Get Along With Programming Language Memory Models?"[1] by Hans J.
+// Boehm for more details.
+//
+// [1] https://www.hpl.hp.com/techreports/2012/HPL-2012-68.pdf
+class SequenceLock {
+ public:
+  constexpr SequenceLock() : lock_(kUninitialized) {}
+
+  // Mark that this lock is ready for use.
+  void MarkInitialized() {
+    assert(lock_.load(std::memory_order_relaxed) == kUninitialized);
+    lock_.store(0, std::memory_order_release);
+  }
+
+  // Copy "size" bytes of data from "src" to "dst", protected as a read-side
+  // critical section of the sequence lock.
+  //
+  // Unlike traditional sequence lock implementations which loop until getting a
+  // clean read, this implementation returns false in the case of concurrent
+  // calls to `Write`. In such a case, the caller should fall back to a
+  // locking-based slow path.
+  //
+  // Returns false if the sequence lock was not yet marked as initialized.
+  //
+  // NOTE: If this returns false, "dst" may be overwritten with undefined
+  // (potentially uninitialized) data.
+  bool TryRead(void* dst, const std::atomic<uint64_t>* src, size_t size) const {
+    // Acquire barrier ensures that no loads done by f() are reordered
+    // above the first load of the sequence counter.
+    int64_t seq_before = lock_.load(std::memory_order_acquire);
+    if (ABSL_PREDICT_FALSE(seq_before & 1) == 1) return false;
+    RelaxedCopyFromAtomic(dst, src, size);
+    // Another acquire fence ensures that the load of 'lock_' below is
+    // strictly ordered after the RelaxedCopyToAtomic call above.
+    std::atomic_thread_fence(std::memory_order_acquire);
+    int64_t seq_after = lock_.load(std::memory_order_relaxed);
+    return ABSL_PREDICT_TRUE(seq_before == seq_after);
+  }
+
+  // Copy "size" bytes from "src" to "dst" as a write-side critical section
+  // of the sequence lock. Any concurrent readers will be forced to retry
+  // until they get a read that does not conflict with this write.
+  //
+  // This call must be externally synchronized against other calls to Write,
+  // but may proceed concurrently with reads.
+  void Write(std::atomic<uint64_t>* dst, const void* src, size_t size) {
+    // We can use relaxed instructions to increment the counter since we
+    // are extenally synchronized. The std::atomic_thread_fence below
+    // ensures that the counter updates don't get interleaved with the
+    // copy to the data.
+    int64_t orig_seq = lock_.load(std::memory_order_relaxed);
+    assert((orig_seq & 1) == 0);  // Must be initially unlocked.
+    lock_.store(orig_seq + 1, std::memory_order_relaxed);
+
+    // We put a release fence between update to lock_ and writes to shared data.
+    // Thus all stores to shared data are effectively release operations and
+    // update to lock_ above cannot be re-ordered past any of them. Note that
+    // this barrier is not for the fetch_add above.  A release barrier for the
+    // fetch_add would be before it, not after.
+    std::atomic_thread_fence(std::memory_order_release);
+    RelaxedCopyToAtomic(dst, src, size);
+    // "Release" semantics ensure that none of the writes done by
+    // RelaxedCopyToAtomic() can be reordered after the following modification.
+    lock_.store(orig_seq + 2, std::memory_order_release);
+  }
+
+  // Return the number of times that Write() has been called.
+  //
+  // REQUIRES: This must be externally synchronized against concurrent calls to
+  // `Write()` or `IncrementModificationCount()`.
+  // REQUIRES: `MarkInitialized()` must have been previously called.
+  int64_t ModificationCount() const {
+    int64_t val = lock_.load(std::memory_order_relaxed);
+    assert(val != kUninitialized && (val & 1) == 0);
+    return val / 2;
+  }
+
+  // REQUIRES: This must be externally synchronized against concurrent calls to
+  // `Write()` or `ModificationCount()`.
+  // REQUIRES: `MarkInitialized()` must have been previously called.
+  void IncrementModificationCount() {
+    int64_t val = lock_.load(std::memory_order_relaxed);
+    assert(val != kUninitialized);
+    lock_.store(val + 2, std::memory_order_relaxed);
+  }
+
+ private:
+  // Perform the equivalent of "memcpy(dst, src, size)", but using relaxed
+  // atomics.
+  static void RelaxedCopyFromAtomic(void* dst, const std::atomic<uint64_t>* src,
+                                    size_t size) {
+    char* dst_byte = static_cast<char*>(dst);
+    while (size >= sizeof(uint64_t)) {
+      uint64_t word = src->load(std::memory_order_relaxed);
+      std::memcpy(dst_byte, &word, sizeof(word));
+      dst_byte += sizeof(word);
+      src++;
+      size -= sizeof(word);
+    }
+    if (size > 0) {
+      uint64_t word = src->load(std::memory_order_relaxed);
+      std::memcpy(dst_byte, &word, size);
+    }
+  }
+
+  // Perform the equivalent of "memcpy(dst, src, size)", but using relaxed
+  // atomics.
+  static void RelaxedCopyToAtomic(std::atomic<uint64_t>* dst, const void* src,
+                                  size_t size) {
+    const char* src_byte = static_cast<const char*>(src);
+    while (size >= sizeof(uint64_t)) {
+      uint64_t word;
+      std::memcpy(&word, src_byte, sizeof(word));
+      dst->store(word, std::memory_order_relaxed);
+      src_byte += sizeof(word);
+      dst++;
+      size -= sizeof(word);
+    }
+    if (size > 0) {
+      uint64_t word = 0;
+      std::memcpy(&word, src_byte, size);
+      dst->store(word, std::memory_order_relaxed);
+    }
+  }
+
+  static constexpr int64_t kUninitialized = -1;
+  std::atomic<int64_t> lock_;
+};
+
+}  // namespace flags_internal
+ABSL_NAMESPACE_END
+}  // namespace absl
+
+#endif  // ABSL_FLAGS_INTERNAL_SEQUENCE_LOCK_H_
diff --git a/absl/flags/internal/sequence_lock_test.cc b/absl/flags/internal/sequence_lock_test.cc
new file mode 100644
index 00000000..9aff1edc
--- /dev/null
+++ b/absl/flags/internal/sequence_lock_test.cc
@@ -0,0 +1,146 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+#include "absl/flags/internal/sequence_lock.h"
+
+#include <atomic>
+#include <thread>  // NOLINT(build/c++11)
+#include <tuple>
+#include <vector>
+
+#include "gtest/gtest.h"
+#include "absl/base/internal/sysinfo.h"
+#include "absl/container/fixed_array.h"
+#include "absl/time/clock.h"
+
+namespace {
+
+namespace flags = absl::flags_internal;
+
+class ConcurrentSequenceLockTest
+    : public testing::TestWithParam<std::tuple<int, int>> {
+ public:
+  ConcurrentSequenceLockTest()
+      : buf_bytes_(std::get<0>(GetParam())),
+        num_threads_(std::get<1>(GetParam())) {}
+
+ protected:
+  const int buf_bytes_;
+  const int num_threads_;
+};
+
+TEST_P(ConcurrentSequenceLockTest, ReadAndWrite) {
+  const int buf_words =
+      flags::AlignUp(buf_bytes_, sizeof(uint64_t)) / sizeof(uint64_t);
+
+  // The buffer that will be protected by the SequenceLock.
+  absl::FixedArray<std::atomic<uint64_t>> protected_buf(buf_words);
+  for (auto& v : protected_buf) v = -1;
+
+  flags::SequenceLock seq_lock;
+  std::atomic<bool> stop{false};
+  std::atomic<int64_t> bad_reads{0};
+  std::atomic<int64_t> good_reads{0};
+  std::atomic<int64_t> unsuccessful_reads{0};
+
+  // Start a bunch of threads which read 'protected_buf' under the sequence
+  // lock. The main thread will concurrently update 'protected_buf'. The updates
+  // always consist of an array of identical integers. The reader ensures that
+  // any data it reads matches that pattern (i.e. the reads are not "torn").
+  std::vector<std::thread> threads;
+  for (int i = 0; i < num_threads_; i++) {
+    threads.emplace_back([&]() {
+      absl::FixedArray<char> local_buf(buf_bytes_);
+      while (!stop.load(std::memory_order_relaxed)) {
+        if (seq_lock.TryRead(local_buf.data(), protected_buf.data(),
+                             buf_bytes_)) {
+          bool good = true;
+          for (const auto& v : local_buf) {
+            if (v != local_buf[0]) good = false;
+          }
+          if (good) {
+            good_reads.fetch_add(1, std::memory_order_relaxed);
+          } else {
+            bad_reads.fetch_add(1, std::memory_order_relaxed);
+          }
+        } else {
+          unsuccessful_reads.fetch_add(1, std::memory_order_relaxed);
+        }
+      }
+    });
+  }
+  while (unsuccessful_reads.load(std::memory_order_relaxed) < num_threads_) {
+    absl::SleepFor(absl::Milliseconds(1));
+  }
+  seq_lock.MarkInitialized();
+
+  // Run a maximum of 5 seconds. On Windows, the scheduler behavior seems
+  // somewhat unfair and without an explicit timeout for this loop, the tests
+  // can run a long time.
+  absl::Time deadline = absl::Now() + absl::Seconds(5);
+  for (int i = 0; i < 100 && absl::Now() < deadline; i++) {
+    absl::FixedArray<char> writer_buf(buf_bytes_);
+    for (auto& v : writer_buf) v = i;
+    seq_lock.Write(protected_buf.data(), writer_buf.data(), buf_bytes_);
+    absl::SleepFor(absl::Microseconds(10));
+  }
+  stop.store(true, std::memory_order_relaxed);
+  for (auto& t : threads) t.join();
+  ASSERT_GE(good_reads, 0);
+  ASSERT_EQ(bad_reads, 0);
+}
+
+// Simple helper for generating a range of thread counts.
+// Generates [low, low*scale, low*scale^2, ...high)
+// (even if high is between low*scale^k and low*scale^(k+1)).
+std::vector<int> MultiplicativeRange(int low, int high, int scale) {
+  std::vector<int> result;
+  for (int current = low; current < high; current *= scale) {
+    result.push_back(current);
+  }
+  result.push_back(high);
+  return result;
+}
+
+INSTANTIATE_TEST_SUITE_P(TestManyByteSizes, ConcurrentSequenceLockTest,
+                         testing::Combine(
+                             // Buffer size (bytes).
+                             testing::Range(1, 128),
+                             // Number of reader threads.
+                             testing::ValuesIn(MultiplicativeRange(
+                                 1, absl::base_internal::NumCPUs(), 2))));
+
+// Simple single-threaded test, parameterized by the size of the buffer to be
+// protected.
+class SequenceLockTest : public testing::TestWithParam<int> {};
+
+TEST_P(SequenceLockTest, SingleThreaded) {
+  const int size = GetParam();
+  absl::FixedArray<std::atomic<uint64_t>> protected_buf(
+      flags::AlignUp(size, sizeof(uint64_t)) / sizeof(uint64_t));
+
+  flags::SequenceLock seq_lock;
+  seq_lock.MarkInitialized();
+
+  std::vector<char> src_buf(size, 'x');
+  seq_lock.Write(protected_buf.data(), src_buf.data(), size);
+
+  std::vector<char> dst_buf(size, '0');
+  ASSERT_TRUE(seq_lock.TryRead(dst_buf.data(), protected_buf.data(), size));
+  ASSERT_EQ(src_buf, dst_buf);
+}
+INSTANTIATE_TEST_SUITE_P(TestManyByteSizes, SequenceLockTest,
+                         // Buffer size (bytes).
+                         testing::Range(1, 128));
+
+}  // namespace