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/g1CollectedHeap.inline.hpp"
  27 #include "gc/g1/g1DirtyCardQueue.hpp"
  28 #include "gc/g1/g1FreeIdSet.hpp"
  29 #include "gc/g1/g1RemSet.hpp"
  30 #include "gc/g1/g1ThreadLocalData.hpp"
  31 #include "gc/g1/heapRegionRemSet.hpp"
  32 #include "gc/shared/suspendibleThreadSet.hpp"
  33 #include "gc/shared/workgroup.hpp"
  34 #include "runtime/atomic.hpp"
  35 #include "runtime/flags/flagSetting.hpp"
  36 #include "runtime/mutexLocker.hpp"
  37 #include "runtime/safepoint.hpp"
  38 #include "runtime/thread.inline.hpp"
  39 #include "runtime/threadSMR.hpp"
  40 
  41 // Closure used for updating remembered sets and recording references that
  42 // point into the collection set while the mutator is running.
  43 // Assumed to be only executed concurrently with the mutator. Yields via
  44 // SuspendibleThreadSet after every card.
  45 class G1RefineCardConcurrentlyClosure: public G1CardTableEntryClosure {
  46 public:
  47   bool do_card_ptr(CardValue* card_ptr, uint worker_i) {
  48     G1CollectedHeap::heap()->rem_set()->refine_card_concurrently(card_ptr, worker_i);
  49 
  50     if (SuspendibleThreadSet::should_yield()) {
  51       // Caller will actually yield.
  52       return false;
  53     }
  54     // Otherwise, we finished successfully; return true.
  55     return true;
  56   }
  57 };
  58 
  59 G1DirtyCardQueue::G1DirtyCardQueue(G1DirtyCardQueueSet* qset) :
  60   // Dirty card queues are always active, so we create them with their
  61   // active field set to true.
  62   PtrQueue(qset, true /* active */)
  63 { }
  64 
  65 G1DirtyCardQueue::~G1DirtyCardQueue() {
  66   flush();
  67 }
  68 
  69 void G1DirtyCardQueue::handle_completed_buffer() {
  70   assert(_buf != NULL, "precondition");
  71   BufferNode* node = BufferNode::make_node_from_buffer(_buf, index());
  72   G1DirtyCardQueueSet* dcqs = dirty_card_qset();
  73   if (dcqs->process_or_enqueue_completed_buffer(node)) {
  74     reset();                    // Buffer fully processed, reset index.
  75   } else {
  76     allocate_buffer();          // Buffer enqueued, get a new one.
  77   }
  78 }
  79 
  80 G1DirtyCardQueueSet::G1DirtyCardQueueSet(bool notify_when_complete) :
  81   PtrQueueSet(notify_when_complete),
  82   _max_completed_buffers(MaxCompletedBuffersUnlimited),
  83   _completed_buffers_padding(0),
  84   _free_ids(NULL),
  85   _processed_buffers_mut(0),
  86   _processed_buffers_rs_thread(0),
  87   _cur_par_buffer_node(NULL)
  88 {
  89   _all_active = true;
  90 }
  91 
  92 G1DirtyCardQueueSet::~G1DirtyCardQueueSet() {
  93   delete _free_ids;
  94 }
  95 
  96 // Determines how many mutator threads can process the buffers in parallel.
  97 uint G1DirtyCardQueueSet::num_par_ids() {
  98   return (uint)os::initial_active_processor_count();
  99 }
 100 
 101 void G1DirtyCardQueueSet::initialize(Monitor* cbl_mon,
 102                                      BufferNode::Allocator* allocator,
 103                                      bool init_free_ids) {
 104   PtrQueueSet::initialize(cbl_mon, allocator);
 105   if (init_free_ids) {
 106     _free_ids = new G1FreeIdSet(0, num_par_ids());
 107   }
 108 }
 109 
 110 void G1DirtyCardQueueSet::handle_zero_index_for_thread(Thread* t) {
 111   G1ThreadLocalData::dirty_card_queue(t).handle_zero_index();
 112 }
 113 
 114 bool G1DirtyCardQueueSet::apply_closure_to_buffer(G1CardTableEntryClosure* cl,
 115                                                   BufferNode* node,
 116                                                   bool consume,
 117                                                   uint worker_i) {
 118   if (cl == NULL) return true;
 119   bool result = true;
 120   void** buf = BufferNode::make_buffer_from_node(node);
 121   size_t i = node->index();
 122   size_t limit = buffer_size();
 123   for ( ; i < limit; ++i) {
 124     CardTable::CardValue* card_ptr = static_cast<CardTable::CardValue*>(buf[i]);
 125     assert(card_ptr != NULL, "invariant");
 126     if (!cl->do_card_ptr(card_ptr, worker_i)) {
 127       result = false;           // Incomplete processing.
 128       break;
 129     }
 130   }
 131   if (consume) {
 132     assert(i <= buffer_size(), "invariant");
 133     node->set_index(i);
 134   }
 135   return result;
 136 }
 137 
 138 #ifndef ASSERT
 139 #define assert_fully_consumed(node, buffer_size)
 140 #else
 141 #define assert_fully_consumed(node, buffer_size)                \
 142   do {                                                          \
 143     size_t _afc_index = (node)->index();                        \
 144     size_t _afc_size = (buffer_size);                           \
 145     assert(_afc_index == _afc_size,                             \
 146            "Buffer was not fully consumed as claimed: index: "  \
 147            SIZE_FORMAT ", size: " SIZE_FORMAT,                  \
 148             _afc_index, _afc_size);                             \
 149   } while (0)
 150 #endif // ASSERT
 151 
 152 bool G1DirtyCardQueueSet::process_or_enqueue_completed_buffer(BufferNode* node) {
 153   if (Thread::current()->is_Java_thread()) {
 154     // If the number of buffers exceeds the limit, make this Java
 155     // thread do the processing itself.  We don't lock to access
 156     // buffer count or padding; it is fine to be imprecise here.  The
 157     // add of padding could overflow, which is treated as unlimited.
 158     size_t max_buffers = max_completed_buffers();
 159     size_t limit = max_buffers + completed_buffers_padding();
 160     if ((completed_buffers_num() > limit) && (limit >= max_buffers)) {
 161       if (mut_process_buffer(node)) {
 162         return true;
 163       }
 164     }
 165   }
 166   enqueue_completed_buffer(node);
 167   return false;
 168 }
 169 
 170 bool G1DirtyCardQueueSet::mut_process_buffer(BufferNode* node) {
 171   guarantee(_free_ids != NULL, "must be");
 172 
 173   uint worker_i = _free_ids->claim_par_id(); // temporarily claim an id
 174   G1RefineCardConcurrentlyClosure cl;
 175   bool result = apply_closure_to_buffer(&cl, node, true, worker_i);
 176   _free_ids->release_par_id(worker_i); // release the id
 177 
 178   if (result) {
 179     assert_fully_consumed(node, buffer_size());
 180     Atomic::inc(&_processed_buffers_mut);
 181   }
 182   return result;
 183 }
 184 
 185 bool G1DirtyCardQueueSet::refine_completed_buffer_concurrently(uint worker_i, size_t stop_at) {
 186   G1RefineCardConcurrentlyClosure cl;
 187   return apply_closure_to_completed_buffer(&cl, worker_i, stop_at, false);
 188 }
 189 
 190 bool G1DirtyCardQueueSet::apply_closure_during_gc(G1CardTableEntryClosure* cl, uint worker_i) {
 191   assert_at_safepoint();
 192   return apply_closure_to_completed_buffer(cl, worker_i, 0, true);
 193 }
 194 
 195 bool G1DirtyCardQueueSet::apply_closure_to_completed_buffer(G1CardTableEntryClosure* cl,
 196                                                             uint worker_i,
 197                                                             size_t stop_at,
 198                                                             bool during_pause) {
 199   assert(!during_pause || stop_at == 0, "Should not leave any completed buffers during a pause");
 200   BufferNode* nd = get_completed_buffer(stop_at);
 201   if (nd == NULL) {
 202     return false;
 203   } else {
 204     if (apply_closure_to_buffer(cl, nd, true, worker_i)) {
 205       assert_fully_consumed(nd, buffer_size());
 206       // Done with fully processed buffer.
 207       deallocate_buffer(nd);
 208       Atomic::inc(&_processed_buffers_rs_thread);
 209     } else {
 210       // Return partially processed buffer to the queue.
 211       guarantee(!during_pause, "Should never stop early");
 212       enqueue_completed_buffer(nd);
 213     }
 214     return true;
 215   }
 216 }
 217 
 218 void G1DirtyCardQueueSet::par_apply_closure_to_all_completed_buffers(G1CardTableEntryClosure* cl) {
 219   BufferNode* nd = _cur_par_buffer_node;
 220   while (nd != NULL) {
 221     BufferNode* next = nd->next();
 222     BufferNode* actual = Atomic::cmpxchg(next, &_cur_par_buffer_node, nd);
 223     if (actual == nd) {
 224       bool b = apply_closure_to_buffer(cl, nd, false);
 225       guarantee(b, "Should not stop early.");
 226       nd = next;
 227     } else {
 228       nd = actual;
 229     }
 230   }
 231 }
 232 
 233 void G1DirtyCardQueueSet::abandon_logs() {
 234   assert(SafepointSynchronize::is_at_safepoint(), "Must be at safepoint.");
 235   abandon_completed_buffers();
 236 
 237   // Since abandon is done only at safepoints, we can safely manipulate
 238   // these queues.
 239   struct AbandonThreadLogClosure : public ThreadClosure {
 240     virtual void do_thread(Thread* t) {
 241       G1ThreadLocalData::dirty_card_queue(t).reset();
 242     }
 243   } closure;
 244   Threads::threads_do(&closure);
 245 
 246   G1BarrierSet::shared_dirty_card_queue().reset();
 247 }
 248 
 249 void G1DirtyCardQueueSet::concatenate_logs() {
 250   // Iterate over all the threads, if we find a partial log add it to
 251   // the global list of logs.  Temporarily turn off the limit on the number
 252   // of outstanding buffers.
 253   assert(SafepointSynchronize::is_at_safepoint(), "Must be at safepoint.");
 254   size_t old_limit = max_completed_buffers();
 255   set_max_completed_buffers(MaxCompletedBuffersUnlimited);
 256 
 257   struct ConcatenateThreadLogClosure : public ThreadClosure {
 258     virtual void do_thread(Thread* t) {
 259       G1DirtyCardQueue& dcq = G1ThreadLocalData::dirty_card_queue(t);
 260       if (!dcq.is_empty()) {
 261         dcq.flush();
 262       }
 263     }
 264   } closure;
 265   Threads::threads_do(&closure);
 266 
 267   G1BarrierSet::shared_dirty_card_queue().flush();
 268   set_max_completed_buffers(old_limit);
 269 }