1 /*
   2  * Copyright (c) 2001, 2019, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "gc/g1/g1BufferNodeList.hpp"
  27 #include "gc/g1/g1CardTableEntryClosure.hpp"
  28 #include "gc/g1/g1CollectedHeap.inline.hpp"
  29 #include "gc/g1/g1DirtyCardQueue.hpp"
  30 #include "gc/g1/g1FreeIdSet.hpp"
  31 #include "gc/g1/g1RedirtyCardsQueue.hpp"
  32 #include "gc/g1/g1RemSet.hpp"
  33 #include "gc/g1/g1ThreadLocalData.hpp"
  34 #include "gc/g1/heapRegionRemSet.hpp"
  35 #include "gc/shared/suspendibleThreadSet.hpp"
  36 #include "gc/shared/workgroup.hpp"
  37 #include "runtime/atomic.hpp"
  38 #include "runtime/flags/flagSetting.hpp"
  39 #include "runtime/mutexLocker.hpp"
  40 #include "runtime/os.hpp"
  41 #include "runtime/safepoint.hpp"
  42 #include "runtime/thread.inline.hpp"
  43 #include "runtime/threadSMR.hpp"
  44 
  45 G1DirtyCardQueue::G1DirtyCardQueue(G1DirtyCardQueueSet* qset) :
  46   // Dirty card queues are always active, so we create them with their
  47   // active field set to true.
  48   PtrQueue(qset, true /* active */)
  49 { }
  50 
  51 G1DirtyCardQueue::~G1DirtyCardQueue() {
  52   flush();
  53 }
  54 
  55 void G1DirtyCardQueue::handle_completed_buffer() {
  56   assert(_buf != NULL, "precondition");
  57   BufferNode* node = BufferNode::make_node_from_buffer(_buf, index());
  58   G1DirtyCardQueueSet* dcqs = dirty_card_qset();
  59   if (dcqs->process_or_enqueue_completed_buffer(node)) {
  60     reset();                    // Buffer fully processed, reset index.
  61   } else {
  62     allocate_buffer();          // Buffer enqueued, get a new one.
  63   }
  64 }
  65 
  66 // Assumed to be zero by concurrent threads.
  67 static uint par_ids_start() { return 0; }
  68 
  69 G1DirtyCardQueueSet::G1DirtyCardQueueSet(Monitor* cbl_mon,
  70                                          BufferNode::Allocator* allocator) :
  71   PtrQueueSet(allocator),
  72   _cbl_mon(cbl_mon),
  73   _completed_buffers_head(NULL),
  74   _completed_buffers_tail(NULL),
  75   _num_cards(0),
  76   _process_cards_threshold(ProcessCardsThresholdNever),
  77   _process_completed_buffers(false),
  78   _max_cards(MaxCardsUnlimited),
  79   _max_cards_padding(0),
  80   _free_ids(par_ids_start(), num_par_ids()),
  81   _mutator_refined_cards_counters(NEW_C_HEAP_ARRAY(size_t, num_par_ids(), mtGC))
  82 {
  83   ::memset(_mutator_refined_cards_counters, 0, num_par_ids() * sizeof(size_t));
  84   _all_active = true;
  85 }
  86 
  87 G1DirtyCardQueueSet::~G1DirtyCardQueueSet() {
  88   abandon_completed_buffers();
  89   FREE_C_HEAP_ARRAY(size_t, _mutator_refined_cards_counters);
  90 }
  91 
  92 // Determines how many mutator threads can process the buffers in parallel.
  93 uint G1DirtyCardQueueSet::num_par_ids() {
  94   return (uint)os::initial_active_processor_count();
  95 }
  96 
  97 size_t G1DirtyCardQueueSet::total_mutator_refined_cards() const {
  98   size_t sum = 0;
  99   for (uint i = 0; i < num_par_ids(); ++i) {
 100     sum += _mutator_refined_cards_counters[i];
 101   }
 102   return sum;
 103 }
 104 
 105 void G1DirtyCardQueueSet::handle_zero_index_for_thread(Thread* t) {
 106   G1ThreadLocalData::dirty_card_queue(t).handle_zero_index();
 107 }
 108 
 109 void G1DirtyCardQueueSet::enqueue_completed_buffer(BufferNode* cbn) {
 110   MonitorLocker ml(_cbl_mon, Mutex::_no_safepoint_check_flag);
 111   cbn->set_next(NULL);
 112   if (_completed_buffers_tail == NULL) {
 113     assert(_completed_buffers_head == NULL, "Well-formedness");
 114     _completed_buffers_head = cbn;
 115     _completed_buffers_tail = cbn;
 116   } else {
 117     _completed_buffers_tail->set_next(cbn);
 118     _completed_buffers_tail = cbn;
 119   }
 120   _num_cards += buffer_size() - cbn->index();
 121 
 122   if (!process_completed_buffers() &&
 123       (num_cards() > process_cards_threshold())) {
 124     set_process_completed_buffers(true);
 125     ml.notify_all();
 126   }
 127   verify_num_cards();
 128 }
 129 
 130 BufferNode* G1DirtyCardQueueSet::get_completed_buffer(size_t stop_at) {
 131   MutexLocker x(_cbl_mon, Mutex::_no_safepoint_check_flag);
 132 
 133   if (num_cards() <= stop_at) {
 134     return NULL;
 135   }
 136 
 137   assert(num_cards() > 0, "invariant");
 138   assert(_completed_buffers_head != NULL, "invariant");
 139   assert(_completed_buffers_tail != NULL, "invariant");
 140 
 141   BufferNode* bn = _completed_buffers_head;
 142   _num_cards -= buffer_size() - bn->index();
 143   _completed_buffers_head = bn->next();
 144   if (_completed_buffers_head == NULL) {
 145     assert(num_cards() == 0, "invariant");
 146     _completed_buffers_tail = NULL;
 147     set_process_completed_buffers(false);
 148   }
 149   verify_num_cards();
 150   bn->set_next(NULL);
 151   return bn;
 152 }
 153 
 154 #ifdef ASSERT
 155 void G1DirtyCardQueueSet::verify_num_cards() const {
 156   size_t actual = 0;
 157   BufferNode* cur = _completed_buffers_head;
 158   while (cur != NULL) {
 159     actual += buffer_size() - cur->index();
 160     cur = cur->next();
 161   }
 162   assert(actual == _num_cards,
 163          "Num entries in completed buffers should be " SIZE_FORMAT " but are " SIZE_FORMAT,
 164          _num_cards, actual);
 165 }
 166 #endif
 167 
 168 void G1DirtyCardQueueSet::abandon_completed_buffers() {
 169   BufferNode* buffers_to_delete = NULL;
 170   {
 171     MutexLocker x(_cbl_mon, Mutex::_no_safepoint_check_flag);
 172     buffers_to_delete = _completed_buffers_head;
 173     _completed_buffers_head = NULL;
 174     _completed_buffers_tail = NULL;
 175     _num_cards = 0;
 176     set_process_completed_buffers(false);
 177   }
 178   while (buffers_to_delete != NULL) {
 179     BufferNode* bn = buffers_to_delete;
 180     buffers_to_delete = bn->next();
 181     bn->set_next(NULL);
 182     deallocate_buffer(bn);
 183   }
 184 }
 185 
 186 void G1DirtyCardQueueSet::notify_if_necessary() {
 187   MonitorLocker ml(_cbl_mon, Mutex::_no_safepoint_check_flag);
 188   if (num_cards() > process_cards_threshold()) {
 189     set_process_completed_buffers(true);
 190     ml.notify_all();
 191   }
 192 }
 193 
 194 // Merge lists of buffers. Notify the processing threads.
 195 // The source queue is emptied as a result. The queues
 196 // must share the monitor.
 197 void G1DirtyCardQueueSet::merge_bufferlists(G1RedirtyCardsQueueSet* src) {
 198   assert(allocator() == src->allocator(), "precondition");
 199   const G1BufferNodeList from = src->take_all_completed_buffers();
 200   if (from._head == NULL) return;
 201 
 202   MutexLocker x(_cbl_mon, Mutex::_no_safepoint_check_flag);
 203   if (_completed_buffers_tail == NULL) {
 204     assert(_completed_buffers_head == NULL, "Well-formedness");
 205     _completed_buffers_head = from._head;
 206     _completed_buffers_tail = from._tail;
 207   } else {
 208     assert(_completed_buffers_head != NULL, "Well formedness");
 209     _completed_buffers_tail->set_next(from._head);
 210     _completed_buffers_tail = from._tail;
 211   }
 212   _num_cards += from._entry_count;
 213 
 214   assert(_completed_buffers_head == NULL && _completed_buffers_tail == NULL ||
 215          _completed_buffers_head != NULL && _completed_buffers_tail != NULL,
 216          "Sanity");
 217   verify_num_cards();
 218 }
 219 
 220 G1BufferNodeList G1DirtyCardQueueSet::take_all_completed_buffers() {
 221   MutexLocker x(_cbl_mon, Mutex::_no_safepoint_check_flag);
 222   G1BufferNodeList result(_completed_buffers_head, _completed_buffers_tail, _num_cards);
 223   _completed_buffers_head = NULL;
 224   _completed_buffers_tail = NULL;
 225   _num_cards = 0;
 226   return result;
 227 }
 228 
 229 bool G1DirtyCardQueueSet::refine_buffer(BufferNode* node,
 230                                         uint worker_id,
 231                                         size_t* total_refined_cards) {
 232   G1RemSet* rem_set = G1CollectedHeap::heap()->rem_set();
 233   size_t size = buffer_size();
 234   void** buffer = BufferNode::make_buffer_from_node(node);
 235   size_t i = node->index();
 236   assert(i <= size, "invariant");
 237   for ( ; (i < size) && !SuspendibleThreadSet::should_yield(); ++i) {
 238     CardTable::CardValue* cp = static_cast<CardTable::CardValue*>(buffer[i]);
 239     rem_set->refine_card_concurrently(cp, worker_id);
 240   }
 241   *total_refined_cards += (i - node->index());
 242   node->set_index(i);
 243   return i == size;
 244 }
 245 
 246 #ifndef ASSERT
 247 #define assert_fully_consumed(node, buffer_size)
 248 #else
 249 #define assert_fully_consumed(node, buffer_size)                \
 250   do {                                                          \
 251     size_t _afc_index = (node)->index();                        \
 252     size_t _afc_size = (buffer_size);                           \
 253     assert(_afc_index == _afc_size,                             \
 254            "Buffer was not fully consumed as claimed: index: "  \
 255            SIZE_FORMAT ", size: " SIZE_FORMAT,                  \
 256             _afc_index, _afc_size);                             \
 257   } while (0)
 258 #endif // ASSERT
 259 
 260 bool G1DirtyCardQueueSet::process_or_enqueue_completed_buffer(BufferNode* node) {
 261   if (Thread::current()->is_Java_thread()) {
 262     // If the number of buffers exceeds the limit, make this Java
 263     // thread do the processing itself.  We don't lock to access
 264     // buffer count or padding; it is fine to be imprecise here.  The
 265     // add of padding could overflow, which is treated as unlimited.
 266     size_t limit = max_cards() + max_cards_padding();
 267     if ((num_cards() > limit) && (limit >= max_cards())) {
 268       if (mut_process_buffer(node)) {
 269         return true;
 270       }
 271     }
 272   }
 273   enqueue_completed_buffer(node);
 274   return false;
 275 }
 276 
 277 bool G1DirtyCardQueueSet::mut_process_buffer(BufferNode* node) {
 278   uint worker_id = _free_ids.claim_par_id(); // temporarily claim an id
 279   uint counter_index = worker_id - par_ids_start();
 280   size_t* counter = &_mutator_refined_cards_counters[counter_index];
 281   bool result = refine_buffer(node, worker_id, counter);
 282   _free_ids.release_par_id(worker_id); // release the id
 283 
 284   if (result) {
 285     assert_fully_consumed(node, buffer_size());
 286   }
 287   return result;
 288 }
 289 
 290 bool G1DirtyCardQueueSet::refine_completed_buffer_concurrently(uint worker_id,
 291                                                                size_t stop_at,
 292                                                                size_t* total_refined_cards) {
 293   BufferNode* node = get_completed_buffer(stop_at);
 294   if (node == NULL) {
 295     return false;
 296   } else if (refine_buffer(node, worker_id, total_refined_cards)) {
 297     assert_fully_consumed(node, buffer_size());
 298     // Done with fully processed buffer.
 299     deallocate_buffer(node);
 300     return true;
 301   } else {
 302     // Return partially processed buffer to the queue.
 303     enqueue_completed_buffer(node);
 304     return true;
 305   }
 306 }
 307 
 308 void G1DirtyCardQueueSet::abandon_logs() {
 309   assert(SafepointSynchronize::is_at_safepoint(), "Must be at safepoint.");
 310   abandon_completed_buffers();
 311 
 312   // Since abandon is done only at safepoints, we can safely manipulate
 313   // these queues.
 314   struct AbandonThreadLogClosure : public ThreadClosure {
 315     virtual void do_thread(Thread* t) {
 316       G1ThreadLocalData::dirty_card_queue(t).reset();
 317     }
 318   } closure;
 319   Threads::threads_do(&closure);
 320 
 321   G1BarrierSet::shared_dirty_card_queue().reset();
 322 }
 323 
 324 void G1DirtyCardQueueSet::concatenate_logs() {
 325   // Iterate over all the threads, if we find a partial log add it to
 326   // the global list of logs.  Temporarily turn off the limit on the number
 327   // of outstanding buffers.
 328   assert(SafepointSynchronize::is_at_safepoint(), "Must be at safepoint.");
 329   size_t old_limit = max_cards();
 330   set_max_cards(MaxCardsUnlimited);
 331 
 332   struct ConcatenateThreadLogClosure : public ThreadClosure {
 333     virtual void do_thread(Thread* t) {
 334       G1DirtyCardQueue& dcq = G1ThreadLocalData::dirty_card_queue(t);
 335       if (!dcq.is_empty()) {
 336         dcq.flush();
 337       }
 338     }
 339   } closure;
 340   Threads::threads_do(&closure);
 341 
 342   G1BarrierSet::shared_dirty_card_queue().flush();
 343   set_max_cards(old_limit);
 344 }