[wasm] Use shared mutex for background compile token

This introduces a new {base::SharedMutex}, which mimics
{std::shared_mutex}, available in C++17.
It is being used for synchronizing the WebAssembly background compile
tasks. This removes a lot of unnecessary contention, leading to
synchronization of background tasks that should be able to run in
parallel ideally.
Locally, this reduces Liftoff compilation time by more than 20 percent.

R=mstarzinger@chromium.org, mlippautz@chromium.org

Bug: chromium:924956
Cq-Include-Trybots: luci.v8.try:v8_linux64_tsan_rel
Change-Id: I8c522faf385149bfe2cf00d777a7942c537f9cd2
Reviewed-on: https://chromium-review.googlesource.com/c/1477731
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#59729}

src/base/platform/mutex.cc

@@ -155,6 +155,45 @@ bool RecursiveMutex::TryLock() {
   return true;
 }
 
+SharedMutex::SharedMutex() { pthread_rwlock_init(&native_handle_, nullptr); }
+
+SharedMutex::~SharedMutex() {
+  int result = pthread_rwlock_destroy(&native_handle_);
+  DCHECK_EQ(0, result);
+  USE(result);
+}
+
+void SharedMutex::LockShared() {
+  int result = pthread_rwlock_rdlock(&native_handle_);
+  DCHECK_EQ(0, result);
+  USE(result);
+}
+
+void SharedMutex::LockExclusive() {
+  int result = pthread_rwlock_wrlock(&native_handle_);
+  DCHECK_EQ(0, result);
+  USE(result);
+}
+
+void SharedMutex::UnlockShared() {
+  int result = pthread_rwlock_unlock(&native_handle_);
+  DCHECK_EQ(0, result);
+  USE(result);
+}
+
+void SharedMutex::UnlockExclusive() {
+  // Same code as {UnlockShared} on POSIX.
+  UnlockShared();
+}
+
+bool SharedMutex::TryLockShared() {
+  return pthread_rwlock_tryrdlock(&native_handle_) == 0;
+}
+
+bool SharedMutex::TryLockExclusive() {
+  return pthread_rwlock_trywrlock(&native_handle_) == 0;
+}
+
 #elif V8_OS_WIN
 
 Mutex::Mutex() : native_handle_(SRWLOCK_INIT) {
@@ -233,6 +272,28 @@ bool RecursiveMutex::TryLock() {
   return true;
 }
 
+SharedMutex::SharedMutex() : native_handle_(SRWLOCK_INIT) {}
+
+SharedMutex::~SharedMutex() {}
+
+void SharedMutex::LockShared() { AcquireSRWLockShared(&native_handle_); }
+
+void SharedMutex::LockExclusive() { AcquireSRWLockExclusive(&native_handle_); }
+
+void SharedMutex::UnlockShared() { ReleaseSRWLockShared(&native_handle_); }
+
+void SharedMutex::UnlockExclusive() {
+  ReleaseSRWLockExclusive(&native_handle_);
+}
+
+bool SharedMutex::TryLockShared() {
+  return TryAcquireSRWLockShared(&native_handle_);
+}
+
+bool SharedMutex::TryLockExclusive() {
+  return TryAcquireSRWLockExclusive(&native_handle_);
+}
+
 #endif  // V8_OS_POSIX
 
 }  // namespace base

src/base/platform/mutex.h

@@ -150,6 +150,7 @@ class V8_BASE_EXPORT RecursiveMutex final {
   // successfully locked.
   bool TryLock() V8_WARN_UNUSED_RESULT;
 
+ private:
   // The implementation-defined native handle type.
 #if V8_OS_POSIX
   typedef pthread_mutex_t NativeHandle;
@@ -157,7 +158,6 @@ class V8_BASE_EXPORT RecursiveMutex final {
   typedef CRITICAL_SECTION NativeHandle;
 #endif
 
- private:
   NativeHandle native_handle_;
 #ifdef DEBUG
   int level_;
@@ -183,6 +183,73 @@ typedef LazyStaticInstance<RecursiveMutex,
 
 #define LAZY_RECURSIVE_MUTEX_INITIALIZER LAZY_STATIC_INSTANCE_INITIALIZER
 
+// ----------------------------------------------------------------------------
+// SharedMutex - a replacement for std::shared_mutex
+//
+// This class is a synchronization primitive that can be used to protect shared
+// data from being simultaneously accessed by multiple threads. In contrast to
+// other mutex types which facilitate exclusive access, a shared_mutex has two
+// levels of access:
+// - shared: several threads can share ownership of the same mutex.
+// - exclusive: only one thread can own the mutex.
+// Shared mutexes are usually used in situations when multiple readers can
+// access the same resource at the same time without causing data races, but
+// only one writer can do so.
+// The SharedMutex class is non-copyable.
+
+class V8_BASE_EXPORT SharedMutex final {
+ public:
+  SharedMutex();
+  ~SharedMutex();
+
+  // Acquires shared ownership of the {SharedMutex}. If another thread is
+  // holding the mutex in exclusive ownership, a call to {LockShared()} will
+  // block execution until shared ownership can be acquired.
+  // If {LockShared()} is called by a thread that already owns the mutex in any
+  // mode (exclusive or shared), the behavior is undefined.
+  void LockShared();
+
+  // Locks the SharedMutex. If another thread has already locked the mutex, a
+  // call to {LockExclusive()} will block execution until the lock is acquired.
+  // If {LockExclusive()} is called by a thread that already owns the mutex in
+  // any mode (shared or exclusive), the behavior is undefined.
+  void LockExclusive();
+
+  // Releases the {SharedMutex} from shared ownership by the calling thread.
+  // The mutex must be locked by the current thread of execution in shared mode,
+  // otherwise, the behavior is undefined.
+  void UnlockShared();
+
+  // Unlocks the {SharedMutex}. It must be locked by the current thread of
+  // execution, otherwise, the behavior is undefined.
+  void UnlockExclusive();
+
+  // Tries to lock the {SharedMutex} in shared mode. Returns immediately. On
+  // successful lock acquisition returns true, otherwise returns false.
+  // This function is allowed to fail spuriously and return false even if the
+  // mutex is not currenly exclusively locked by any other thread.
+  bool TryLockShared() V8_WARN_UNUSED_RESULT;
+
+  // Tries to lock the {SharedMutex}. Returns immediately. On successful lock
+  // acquisition returns true, otherwise returns false.
+  // This function is allowed to fail spuriously and return false even if the
+  // mutex is not currently locked by any other thread.
+  // If try_lock is called by a thread that already owns the mutex in any mode
+  // (shared or exclusive), the behavior is undefined.
+  bool TryLockExclusive() V8_WARN_UNUSED_RESULT;
+
+ private:
+  // The implementation-defined native handle type.
+#if V8_OS_POSIX
+  typedef pthread_rwlock_t NativeHandle;
+#elif V8_OS_WIN
+  typedef SRWLOCK NativeHandle;
+#endif
+
+  NativeHandle native_handle_;
+
+  DISALLOW_COPY_AND_ASSIGN(SharedMutex);
+};
 
 // -----------------------------------------------------------------------------
 // LockGuard
@@ -222,6 +289,41 @@ class LockGuard final {
 
 using MutexGuard = LockGuard<Mutex>;
 
+enum MutexSharedType : bool { kShared = true, kExclusive = false };
+
+template <MutexSharedType kIsShared,
+          NullBehavior Behavior = NullBehavior::kRequireNotNull>
+class SharedMutexGuard final {
+ public:
+  explicit SharedMutexGuard(SharedMutex* mutex) : mutex_(mutex) {
+    if (!has_mutex()) return;
+    if (kIsShared) {
+      mutex_->LockShared();
+    } else {
+      mutex_->LockExclusive();
+    }
+  }
+  ~SharedMutexGuard() {
+    if (!has_mutex()) return;
+    if (kIsShared) {
+      mutex_->UnlockShared();
+    } else {
+      mutex_->UnlockExclusive();
+    }
+  }
+
+ private:
+  SharedMutex* const mutex_;
+
+  bool V8_INLINE has_mutex() const {
+    DCHECK_IMPLIES(Behavior == NullBehavior::kRequireNotNull,
+                   mutex_ != nullptr);
+    return Behavior == NullBehavior::kRequireNotNull || mutex_ != nullptr;
+  }
+
+  DISALLOW_COPY_AND_ASSIGN(SharedMutexGuard);
+};
+
 }  // namespace base
 }  // namespace v8
 

src/wasm/module-compiler.cc

@@ -65,24 +65,21 @@ class BackgroundCompileToken {
       : native_module_(native_module) {}
 
   void Cancel() {
-    base::MutexGuard mutex_guard(&mutex_);
+    base::SharedMutexGuard<base::kExclusive> mutex_guard(&mutex_);
     native_module_ = nullptr;
   }
 
-  // Only call this while holding the {mutex_}.
-  void CancelLocked() { native_module_ = nullptr; }
-
  private:
   friend class BackgroundCompileScope;
-  base::Mutex mutex_;
+  base::SharedMutex mutex_;
   NativeModule* native_module_;
 
   NativeModule* StartScope() {
-    mutex_.Lock();
+    mutex_.LockShared();
     return native_module_;
   }
 
-  void ExitScope() { mutex_.Unlock(); }
+  void ExitScope() { mutex_.UnlockShared(); }
 };
 
 class CompilationStateImpl;
@@ -681,6 +678,7 @@ class BackgroundCompileTask : public CancelableTask {
           &env.value(), wire_bytes, async_counters_.get(), &detected_features);
 
       // Step 3 (synchronized): Publish the compilation result.
+      bool cancel_compilation = false;
       {
         BackgroundCompileScope compile_scope(token_);
         if (compile_scope.cancelled()) return;
@@ -690,9 +688,8 @@ class BackgroundCompileTask : public CancelableTask {
           compile_scope.compilation_state()->OnBackgroundTaskStopped(
               detected_features);
           // Also, cancel all remaining compilation.
-          token_->CancelLocked();
-          return;
-        }
+          cancel_compilation = true;
+        } else {
           compile_scope.compilation_state()->OnFinishedUnit(
               unit->requested_tier(), code);
           if (deadline < MonotonicallyIncreasingTimeInMs()) {
@@ -703,6 +700,11 @@ class BackgroundCompileTask : public CancelableTask {
           }
         }
       }
+      if (cancel_compilation) {
+        token_->Cancel();
+        return;
+      }
+    }
     UNREACHABLE();  // Loop exits via explicit return.
   }