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/safepoint.hpp"
  41 #include "runtime/thread.inline.hpp"
  42 #include "runtime/threadSMR.hpp"
  43 
  44 // Closure used for updating remembered sets and recording references that
  45 // point into the collection set while the mutator is running.
  46 // Assumed to be only executed concurrently with the mutator. Yields via
  47 // SuspendibleThreadSet after every card.
  48 class G1RefineCardConcurrentlyClosure: public G1CardTableEntryClosure {
  49 public:
  50   bool do_card_ptr(CardValue* card_ptr, uint worker_i) {
  51     G1CollectedHeap::heap()->rem_set()->refine_card_concurrently(card_ptr, worker_i);
  52 
  53     if (SuspendibleThreadSet::should_yield()) {
  54       // Caller will actually yield.
  55       return false;
  56     }
  57     // Otherwise, we finished successfully; return true.
  58     return true;
  59   }
  60 };
  61 
  62 G1DirtyCardQueue::G1DirtyCardQueue(G1DirtyCardQueueSet* qset) :
  63   // Dirty card queues are always active, so we create them with their
  64   // active field set to true.
  65   PtrQueue(qset, true /* active */)
  66 { }
  67 
  68 G1DirtyCardQueue::~G1DirtyCardQueue() {
  69   flush();
  70 }
  71 
  72 void G1DirtyCardQueue::handle_completed_buffer() {
  73   assert(_buf != NULL, "precondition");
  74   BufferNode* node = BufferNode::make_node_from_buffer(_buf, index());
  75   G1DirtyCardQueueSet* dcqs = dirty_card_qset();
  76   if (dcqs->process_or_enqueue_completed_buffer(node)) {
  77     reset();                    // Buffer fully processed, reset index.
  78   } else {
  79     allocate_buffer();          // Buffer enqueued, get a new one.
  80   }
  81 }
  82 
  83 G1DirtyCardQueueSet::G1DirtyCardQueueSet(bool notify_when_complete) :
  84   PtrQueueSet(),
  85   _cbl_mon(NULL),
  86   _completed_buffers_head(NULL),
  87   _completed_buffers_tail(NULL),
  88   _num_entries_in_completed_buffers(0),
  89   _process_completed_buffers_threshold(ProcessCompletedBuffersThresholdNever),
  90   _process_completed_buffers(false),
  91   _notify_when_complete(notify_when_complete),
  92   _max_completed_buffers(MaxCompletedBuffersUnlimited),
  93   _completed_buffers_padding(0),
  94   _free_ids(0, num_par_ids()),
  95   _processed_buffers_mut(0),
  96   _processed_buffers_rs_thread(0)
  97 {
  98   _all_active = true;
  99 }
 100 
 101 G1DirtyCardQueueSet::~G1DirtyCardQueueSet() {
 102   abandon_completed_buffers();
 103 }
 104 
 105 // Determines how many mutator threads can process the buffers in parallel.
 106 uint G1DirtyCardQueueSet::num_par_ids() {
 107   return (uint)os::initial_active_processor_count();
 108 }
 109 
 110 void G1DirtyCardQueueSet::initialize(Monitor* cbl_mon,
 111                                      BufferNode::Allocator* allocator) {
 112   PtrQueueSet::initialize(allocator);
 113   assert(_cbl_mon == NULL, "Init order issue?");
 114   _cbl_mon = cbl_mon;
 115 }
 116 
 117 void G1DirtyCardQueueSet::handle_zero_index_for_thread(Thread* t) {
 118   G1ThreadLocalData::dirty_card_queue(t).handle_zero_index();
 119 }
 120 
 121 void G1DirtyCardQueueSet::enqueue_completed_buffer(BufferNode* cbn) {
 122   MutexLocker x(_cbl_mon, Mutex::_no_safepoint_check_flag);
 123   cbn->set_next(NULL);
 124   if (_completed_buffers_tail == NULL) {
 125     assert(_completed_buffers_head == NULL, "Well-formedness");
 126     _completed_buffers_head = cbn;
 127     _completed_buffers_tail = cbn;
 128   } else {
 129     _completed_buffers_tail->set_next(cbn);
 130     _completed_buffers_tail = cbn;
 131   }
 132   _num_entries_in_completed_buffers += buffer_size() - cbn->index();
 133 
 134   if (!process_completed_buffers() &&
 135       (num_completed_buffers() > process_completed_buffers_threshold())) {
 136     set_process_completed_buffers(true);
 137     if (_notify_when_complete) {
 138       _cbl_mon->notify_all();
 139     }
 140   }
 141   verify_num_entries_in_completed_buffers();
 142 }
 143 
 144 BufferNode* G1DirtyCardQueueSet::get_completed_buffer(size_t stop_at) {
 145   MutexLocker x(_cbl_mon, Mutex::_no_safepoint_check_flag);
 146 
 147   if (num_completed_buffers() <= stop_at) {
 148     return NULL;
 149   }
 150 
 151   assert(num_completed_buffers() > 0, "invariant");
 152   assert(_completed_buffers_head != NULL, "invariant");
 153   assert(_completed_buffers_tail != NULL, "invariant");
 154 
 155   BufferNode* bn = _completed_buffers_head;
 156   _num_entries_in_completed_buffers -= buffer_size() - bn->index();
 157   _completed_buffers_head = bn->next();
 158   if (_completed_buffers_head == NULL) {
 159     assert(num_completed_buffers() == 0, "invariant");
 160     _completed_buffers_tail = NULL;
 161     set_process_completed_buffers(false);
 162   }
 163   verify_num_entries_in_completed_buffers();
 164   bn->set_next(NULL);
 165   return bn;
 166 }
 167 
 168 #ifdef ASSERT
 169 void G1DirtyCardQueueSet::verify_num_entries_in_completed_buffers() const {
 170   size_t actual = 0;
 171   BufferNode* cur = _completed_buffers_head;
 172   while (cur != NULL) {
 173     actual += buffer_size() - cur->index();
 174     cur = cur->next();
 175   }
 176   assert(actual == _num_entries_in_completed_buffers,
 177          "Num entries in completed buffers should be " SIZE_FORMAT " but are " SIZE_FORMAT,
 178          _num_entries_in_completed_buffers, actual);
 179 }
 180 #endif
 181 
 182 void G1DirtyCardQueueSet::abandon_completed_buffers() {
 183   BufferNode* buffers_to_delete = NULL;
 184   {
 185     MutexLocker x(_cbl_mon, Mutex::_no_safepoint_check_flag);
 186     buffers_to_delete = _completed_buffers_head;
 187     _completed_buffers_head = NULL;
 188     _completed_buffers_tail = NULL;
 189     _num_entries_in_completed_buffers = 0;
 190     set_process_completed_buffers(false);
 191   }
 192   while (buffers_to_delete != NULL) {
 193     BufferNode* bn = buffers_to_delete;
 194     buffers_to_delete = bn->next();
 195     bn->set_next(NULL);
 196     deallocate_buffer(bn);
 197   }
 198 }
 199 
 200 void G1DirtyCardQueueSet::notify_if_necessary() {
 201   MutexLocker x(_cbl_mon, Mutex::_no_safepoint_check_flag);
 202   if (num_completed_buffers() > process_completed_buffers_threshold()) {
 203     set_process_completed_buffers(true);
 204     if (_notify_when_complete)
 205       _cbl_mon->notify();
 206   }
 207 }
 208 
 209 // Merge lists of buffers. Notify the processing threads.
 210 // The source queue is emptied as a result. The queues
 211 // must share the monitor.
 212 void G1DirtyCardQueueSet::merge_bufferlists(G1RedirtyCardsQueueSet* src) {
 213   assert(allocator() == src->allocator(), "precondition");
 214   const G1BufferNodeList from = src->take_all_completed_buffers();
 215   if (from._head == NULL) return;
 216 
 217   MutexLocker x(_cbl_mon, Mutex::_no_safepoint_check_flag);
 218   if (_completed_buffers_tail == NULL) {
 219     assert(_completed_buffers_head == NULL, "Well-formedness");
 220     _completed_buffers_head = from._head;
 221     _completed_buffers_tail = from._tail;
 222   } else {
 223     assert(_completed_buffers_head != NULL, "Well formedness");
 224     _completed_buffers_tail->set_next(from._head);
 225     _completed_buffers_tail = from._tail;
 226   }
 227   _num_entries_in_completed_buffers += from._entry_count;
 228 
 229   assert(_completed_buffers_head == NULL && _completed_buffers_tail == NULL ||
 230          _completed_buffers_head != NULL && _completed_buffers_tail != NULL,
 231          "Sanity");
 232   verify_num_entries_in_completed_buffers();
 233 }
 234 
 235 bool G1DirtyCardQueueSet::apply_closure_to_buffer(G1CardTableEntryClosure* cl,
 236                                                   BufferNode* node,
 237                                                   uint worker_i) {
 238   if (cl == NULL) return true;
 239   bool result = true;
 240   void** buf = BufferNode::make_buffer_from_node(node);
 241   size_t i = node->index();
 242   size_t limit = buffer_size();
 243   for ( ; i < limit; ++i) {
 244     CardTable::CardValue* card_ptr = static_cast<CardTable::CardValue*>(buf[i]);
 245     assert(card_ptr != NULL, "invariant");
 246     if (!cl->do_card_ptr(card_ptr, worker_i)) {
 247       result = false;           // Incomplete processing.
 248       break;
 249     }
 250   }
 251   assert(i <= buffer_size(), "invariant");
 252   node->set_index(i);
 253   return result;
 254 }
 255 
 256 #ifndef ASSERT
 257 #define assert_fully_consumed(node, buffer_size)
 258 #else
 259 #define assert_fully_consumed(node, buffer_size)                \
 260   do {                                                          \
 261     size_t _afc_index = (node)->index();                        \
 262     size_t _afc_size = (buffer_size);                           \
 263     assert(_afc_index == _afc_size,                             \
 264            "Buffer was not fully consumed as claimed: index: "  \
 265            SIZE_FORMAT ", size: " SIZE_FORMAT,                  \
 266             _afc_index, _afc_size);                             \
 267   } while (0)
 268 #endif // ASSERT
 269 
 270 bool G1DirtyCardQueueSet::process_or_enqueue_completed_buffer(BufferNode* node) {
 271   if (Thread::current()->is_Java_thread()) {
 272     // If the number of buffers exceeds the limit, make this Java
 273     // thread do the processing itself.  We don't lock to access
 274     // buffer count or padding; it is fine to be imprecise here.  The
 275     // add of padding could overflow, which is treated as unlimited.
 276     size_t max_buffers = max_completed_buffers();
 277     size_t limit = max_buffers + completed_buffers_padding();
 278     if ((num_completed_buffers() > limit) && (limit >= max_buffers)) {
 279       if (mut_process_buffer(node)) {
 280         return true;
 281       }
 282     }
 283   }
 284   enqueue_completed_buffer(node);
 285   return false;
 286 }
 287 
 288 bool G1DirtyCardQueueSet::mut_process_buffer(BufferNode* node) {
 289   uint worker_id = _free_ids.claim_par_id(); // temporarily claim an id
 290   G1RefineCardConcurrentlyClosure cl;
 291   bool result = apply_closure_to_buffer(&cl, node, worker_id);
 292   _free_ids.release_par_id(worker_id); // release the id
 293 
 294   if (result) {
 295     assert_fully_consumed(node, buffer_size());
 296     Atomic::inc(&_processed_buffers_mut);
 297   }
 298   return result;
 299 }
 300 
 301 bool G1DirtyCardQueueSet::refine_completed_buffer_concurrently(uint worker_i, size_t stop_at) {
 302   G1RefineCardConcurrentlyClosure cl;
 303   return apply_closure_to_completed_buffer(&cl, worker_i, stop_at, false);
 304 }
 305 
 306 bool G1DirtyCardQueueSet::apply_closure_during_gc(G1CardTableEntryClosure* cl, uint worker_i) {
 307   assert_at_safepoint();
 308   return apply_closure_to_completed_buffer(cl, worker_i, 0, true);
 309 }
 310 
 311 bool G1DirtyCardQueueSet::apply_closure_to_completed_buffer(G1CardTableEntryClosure* cl,
 312                                                             uint worker_i,
 313                                                             size_t stop_at,
 314                                                             bool during_pause) {
 315   assert(!during_pause || stop_at == 0, "Should not leave any completed buffers during a pause");
 316   BufferNode* nd = get_completed_buffer(stop_at);
 317   if (nd == NULL) {
 318     return false;
 319   } else {
 320     if (apply_closure_to_buffer(cl, nd, worker_i)) {
 321       assert_fully_consumed(nd, buffer_size());
 322       // Done with fully processed buffer.
 323       deallocate_buffer(nd);
 324       Atomic::inc(&_processed_buffers_rs_thread);
 325     } else {
 326       // Return partially processed buffer to the queue.
 327       guarantee(!during_pause, "Should never stop early");
 328       enqueue_completed_buffer(nd);
 329     }
 330     return true;
 331   }
 332 }
 333 
 334 void G1DirtyCardQueueSet::abandon_logs() {
 335   assert(SafepointSynchronize::is_at_safepoint(), "Must be at safepoint.");
 336   abandon_completed_buffers();
 337 
 338   // Since abandon is done only at safepoints, we can safely manipulate
 339   // these queues.
 340   struct AbandonThreadLogClosure : public ThreadClosure {
 341     virtual void do_thread(Thread* t) {
 342       G1ThreadLocalData::dirty_card_queue(t).reset();
 343     }
 344   } closure;
 345   Threads::threads_do(&closure);
 346 
 347   G1BarrierSet::shared_dirty_card_queue().reset();
 348 }
 349 
 350 void G1DirtyCardQueueSet::concatenate_logs() {
 351   // Iterate over all the threads, if we find a partial log add it to
 352   // the global list of logs.  Temporarily turn off the limit on the number
 353   // of outstanding buffers.
 354   assert(SafepointSynchronize::is_at_safepoint(), "Must be at safepoint.");
 355   size_t old_limit = max_completed_buffers();
 356   set_max_completed_buffers(MaxCompletedBuffersUnlimited);
 357 
 358   struct ConcatenateThreadLogClosure : public ThreadClosure {
 359     virtual void do_thread(Thread* t) {
 360       G1DirtyCardQueue& dcq = G1ThreadLocalData::dirty_card_queue(t);
 361       if (!dcq.is_empty()) {
 362         dcq.flush();
 363       }
 364     }
 365   } closure;
 366   Threads::threads_do(&closure);
 367 
 368   G1BarrierSet::shared_dirty_card_queue().flush();
 369   set_max_completed_buffers(old_limit);
 370 }