8195098: Low latency hashtable for read-mostly scenarios

This implement a concurrent hashtable using chaining and the GlobalCounter for ABA problems.

Reviewed-by: acorn, coleenp, dcubed, eosterlund, gziemski, mlarsson

src/hotspot/share/utilities/concurrentHashTableTasks.inline.hpp

@@ -0,0 +1,226 @@
+/*
+ * Copyright (c) 2018, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#ifndef SHARE_UTILITIES_CONCURRENT_HASH_TABLE_TASKS_INLINE_HPP
+#define SHARE_UTILITIES_CONCURRENT_HASH_TABLE_TASKS_INLINE_HPP
+
+#include "utilities/concurrentHashTable.inline.hpp"
+
+// This inline file contains BulkDeleteTask and GrowTasks which are both bucket
+// operations, which they are serialized with each other.
+
+// Base class for pause and/or parallel bulk operations.
+template <typename VALUE, typename CONFIG, MEMFLAGS F>
+class ConcurrentHashTable<VALUE, CONFIG, F>::BucketsOperation {
+ protected:
+  ConcurrentHashTable<VALUE, CONFIG, F>* _cht;
+
+  // Default size of _task_size_log2
+  static const size_t DEFAULT_TASK_SIZE_LOG2 = 12;
+
+  // The table is split into ranges, every increment is one range.
+  volatile size_t _next_to_claim;
+  size_t _task_size_log2; // Number of buckets.
+  size_t _stop_task;      // Last task
+  size_t _size_log2;      // Table size.
+
+  BucketsOperation(ConcurrentHashTable<VALUE, CONFIG, F>* cht)
+    : _cht(cht), _next_to_claim(0), _task_size_log2(DEFAULT_TASK_SIZE_LOG2),
+    _stop_task(0), _size_log2(0) {}
+
+  // Returns true if you succeeded to claim the range start -> (stop-1).
+  bool claim(size_t* start, size_t* stop) {
+    size_t claimed = Atomic::add((size_t)1, &_next_to_claim) - 1;
+    if (claimed >= _stop_task) {
+      return false;
+    }
+    *start = claimed * (((size_t)1) << _task_size_log2);
+    *stop  = ((*start) + (((size_t)1) << _task_size_log2));
+    return true;
+  }
+
+  // Calculate starting values.
+  void setup() {
+    _size_log2 = _cht->_table->_log2_size;
+    size_t tmp = _size_log2 > _task_size_log2 ?
+                 _size_log2 - _task_size_log2 : 0;
+    _stop_task = (((size_t)1) << tmp);
+  }
+
+  // Returns false if all ranges are claimed.
+  bool have_more_work() {
+    return OrderAccess::load_acquire(&_next_to_claim) >= _stop_task;
+  }
+
+  // If we have changed size.
+  bool is_same_table() {
+    // Not entirely true.
+    return _size_log2 != _cht->_table->_log2_size;
+  }
+
+  void thread_owns_resize_lock(Thread* thread) {
+    assert(BucketsOperation::_cht->_resize_lock_owner == thread,
+           "Should be locked by me");
+    assert(BucketsOperation::_cht->_resize_lock->owned_by_self(),
+           "Operations lock not held");
+  }
+  void thread_owns_only_state_lock(Thread* thread) {
+    assert(BucketsOperation::_cht->_resize_lock_owner == thread,
+           "Should be locked by me");
+    assert(!BucketsOperation::_cht->_resize_lock->owned_by_self(),
+           "Operations lock held");
+  }
+  void thread_do_not_own_resize_lock(Thread* thread) {
+    assert(!BucketsOperation::_cht->_resize_lock->owned_by_self(),
+           "Operations lock held");
+    assert(BucketsOperation::_cht->_resize_lock_owner != thread,
+           "Should not be locked by me");
+  }
+};
+
+// For doing pausable/parallel bulk delete.
+template <typename VALUE, typename CONFIG, MEMFLAGS F>
+class ConcurrentHashTable<VALUE, CONFIG, F>::BulkDeleteTask :
+  public BucketsOperation
+{
+ public:
+  BulkDeleteTask(ConcurrentHashTable<VALUE, CONFIG, F>* cht)
+    : BucketsOperation(cht) {
+  }
+  // Before start prepare must be called.
+  bool prepare(Thread* thread) {
+    bool lock = BucketsOperation::_cht->try_resize_lock(thread);
+    if (!lock) {
+      return false;
+    }
+    this->setup();
+    this->thread_owns_resize_lock(thread);
+    return true;
+  }
+
+  // Does one range destroying all matching EVALUATE_FUNC and
+  // DELETE_FUNC is called be destruction. Returns true if there is more work.
+  template <typename EVALUATE_FUNC, typename DELETE_FUNC>
+  bool doTask(Thread* thread, EVALUATE_FUNC& eval_f, DELETE_FUNC& del_f) {
+    size_t start, stop;
+    assert(BucketsOperation::_cht->_resize_lock_owner != NULL,
+           "Should be locked");
+    if (!this->claim(&start, &stop)) {
+      return false;
+    }
+    BucketsOperation::_cht->do_bulk_delete_locked_for(thread, start, stop,
+                                                      eval_f, del_f);
+    return true;
+  }
+
+  // Pauses this operations for a safepoint.
+  void pause(Thread* thread) {
+    this->thread_owns_resize_lock(thread);
+    // This leaves internal state locked.
+    BucketsOperation::_cht->unlock_resize_lock(thread);
+    this->thread_do_not_own_resize_lock(thread);
+  }
+
+  // Continues this operations after a safepoint.
+  bool cont(Thread* thread) {
+    this->thread_do_not_own_resize_lock(thread);
+    if (!BucketsOperation::_cht->try_resize_lock(thread)) {
+      this->thread_do_not_own_resize_lock(thread);
+      return false;
+    }
+    if (BucketsOperation::is_same_table()) {
+      BucketsOperation::_cht->unlock_resize_lock(thread);
+      this->thread_do_not_own_resize_lock(thread);
+      return false;
+    }
+    this->thread_owns_resize_lock(thread);
+    return true;
+  }
+
+  // Must be called after ranges are done.
+  void done(Thread* thread) {
+    this->thread_owns_resize_lock(thread);
+    BucketsOperation::_cht->unlock_resize_lock(thread);
+    this->thread_do_not_own_resize_lock(thread);
+  }
+};
+
+template <typename VALUE, typename CONFIG, MEMFLAGS F>
+class ConcurrentHashTable<VALUE, CONFIG, F>::GrowTask :
+  public BucketsOperation
+{
+ public:
+  GrowTask(ConcurrentHashTable<VALUE, CONFIG, F>* cht) : BucketsOperation(cht) {
+  }
+  // Before start prepare must be called.
+  bool prepare(Thread* thread) {
+    if (!BucketsOperation::_cht->internal_grow_prolog(
+          thread, BucketsOperation::_cht->_log2_size_limit)) {
+      return false;
+    }
+    this->thread_owns_resize_lock(thread);
+    BucketsOperation::setup();
+    return true;
+  }
+
+  // Re-sizes a portion of the table. Returns true if there is more work.
+  bool doTask(Thread* thread) {
+    size_t start, stop;
+    assert(BucketsOperation::_cht->_resize_lock_owner != NULL,
+           "Should be locked");
+    if (!this->claim(&start, &stop)) {
+      return false;
+    }
+    BucketsOperation::_cht->internal_grow_range(thread, start, stop);
+    assert(BucketsOperation::_cht->_resize_lock_owner != NULL,
+           "Should be locked");
+    return true;
+  }
+
+  // Pauses growing for safepoint
+  void pause(Thread* thread) {
+    // This leaves internal state locked.
+    this->thread_owns_resize_lock(thread);
+    BucketsOperation::_cht->_resize_lock->unlock();
+    this->thread_owns_only_state_lock(thread);
+  }
+
+  // Continues growing after safepoint.
+  void cont(Thread* thread) {
+    this->thread_owns_only_state_lock(thread);
+    // If someone slips in here directly after safepoint.
+    while (!BucketsOperation::_cht->_resize_lock->try_lock())
+      { /* for ever */ };
+    this->thread_owns_resize_lock(thread);
+  }
+
+  // Must be called after doTask returns false.
+  void done(Thread* thread) {
+    this->thread_owns_resize_lock(thread);
+    BucketsOperation::_cht->internal_grow_epilog(thread);
+    this->thread_do_not_own_resize_lock(thread);
+  }
+};
+
+#endif // include guard