1 /*
   2  * Copyright (c) 2016, 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.hpp"
  27 #include "gc/g1/g1CollectionSet.hpp"
  28 #include "gc/g1/g1CollectorState.hpp"
  29 #include "gc/g1/g1Policy.hpp"
  30 #include "gc/g1/heapRegion.inline.hpp"
  31 #include "gc/g1/heapRegionRemSet.hpp"
  32 #include "gc/g1/heapRegionSet.hpp"
  33 #include "utilities/debug.hpp"
  34 
  35 G1CollectorState* G1CollectionSet::collector_state() {
  36   return _g1->collector_state();
  37 }
  38 
  39 G1GCPhaseTimes* G1CollectionSet::phase_times() {
  40   return _policy->phase_times();
  41 }
  42 
  43 CollectionSetChooser* G1CollectionSet::cset_chooser() {
  44   return _cset_chooser;
  45 }
  46 
  47 double G1CollectionSet::predict_region_elapsed_time_ms(HeapRegion* hr) {
  48   return _policy->predict_region_elapsed_time_ms(hr, collector_state()->gcs_are_young());
  49 }
  50 
  51 G1CollectionSet::G1CollectionSet(G1CollectedHeap* g1h, G1Policy* policy) :
  52   _g1(g1h),
  53   _policy(policy),
  54   _cset_chooser(new CollectionSetChooser()),
  55   _eden_region_length(0),
  56   _survivor_region_length(0),
  57   _old_region_length(0),
  58 
  59   _head(NULL),
  60   _bytes_used_before(0),
  61   _recorded_rs_lengths(0),
  62   // Incremental CSet attributes
  63   _inc_build_state(Inactive),
  64   _inc_head(NULL),
  65   _inc_tail(NULL),
  66   _inc_bytes_used_before(0),
  67   _inc_recorded_rs_lengths(0),
  68   _inc_recorded_rs_lengths_diffs(0),
  69   _inc_predicted_elapsed_time_ms(0.0),
  70   _inc_predicted_elapsed_time_ms_diffs(0.0),
  71   _inc_region_length(0) {}
  72 
  73 G1CollectionSet::~G1CollectionSet() {
  74   delete _cset_chooser;
  75 }
  76 
  77 void G1CollectionSet::init_region_lengths(uint eden_cset_region_length,
  78                                           uint survivor_cset_region_length) {
  79   _eden_region_length     = eden_cset_region_length;
  80   _survivor_region_length = survivor_cset_region_length;
  81 
  82   assert(young_region_length() == _inc_region_length, "should match %u == %u", young_region_length(), _inc_region_length);
  83 
  84   _old_region_length      = 0;
  85 }
  86 
  87 void G1CollectionSet::set_recorded_rs_lengths(size_t rs_lengths) {
  88   _recorded_rs_lengths = rs_lengths;
  89 }
  90 
  91 // Add the heap region at the head of the non-incremental collection set
  92 void G1CollectionSet::add_old_region(HeapRegion* hr) {
  93   assert(_inc_build_state == Active, "Precondition");
  94   assert(hr->is_old(), "the region should be old");
  95 
  96   assert(!hr->in_collection_set(), "should not already be in the CSet");
  97   _g1->register_old_region_with_cset(hr);
  98   hr->set_next_in_collection_set(_head);
  99   _head = hr;
 100   _bytes_used_before += hr->used();
 101   size_t rs_length = hr->rem_set()->occupied();
 102   _recorded_rs_lengths += rs_length;
 103   _old_region_length += 1;
 104 }
 105 
 106 // Initialize the per-collection-set information
 107 void G1CollectionSet::start_incremental_building() {
 108   assert(_inc_build_state == Inactive, "Precondition");
 109 
 110   _inc_head = NULL;
 111   _inc_tail = NULL;
 112   _inc_bytes_used_before = 0;
 113   _inc_region_length = 0;
 114 
 115   _inc_recorded_rs_lengths = 0;
 116   _inc_recorded_rs_lengths_diffs = 0;
 117   _inc_predicted_elapsed_time_ms = 0.0;
 118   _inc_predicted_elapsed_time_ms_diffs = 0.0;
 119   _inc_build_state = Active;
 120 }
 121 
 122 void G1CollectionSet::finalize_incremental_building() {
 123   assert(_inc_build_state == Active, "Precondition");
 124   assert(SafepointSynchronize::is_at_safepoint(), "should be at a safepoint");
 125 
 126   // The two "main" fields, _inc_recorded_rs_lengths and
 127   // _inc_predicted_elapsed_time_ms, are updated by the thread
 128   // that adds a new region to the CSet. Further updates by the
 129   // concurrent refinement thread that samples the young RSet lengths
 130   // are accumulated in the *_diffs fields. Here we add the diffs to
 131   // the "main" fields.
 132 
 133   if (_inc_recorded_rs_lengths_diffs >= 0) {
 134     _inc_recorded_rs_lengths += _inc_recorded_rs_lengths_diffs;
 135   } else {
 136     // This is defensive. The diff should in theory be always positive
 137     // as RSets can only grow between GCs. However, given that we
 138     // sample their size concurrently with other threads updating them
 139     // it's possible that we might get the wrong size back, which
 140     // could make the calculations somewhat inaccurate.
 141     size_t diffs = (size_t) (-_inc_recorded_rs_lengths_diffs);
 142     if (_inc_recorded_rs_lengths >= diffs) {
 143       _inc_recorded_rs_lengths -= diffs;
 144     } else {
 145       _inc_recorded_rs_lengths = 0;
 146     }
 147   }
 148   _inc_predicted_elapsed_time_ms += _inc_predicted_elapsed_time_ms_diffs;
 149 
 150   _inc_recorded_rs_lengths_diffs = 0;
 151   _inc_predicted_elapsed_time_ms_diffs = 0.0;
 152 }
 153 
 154 void G1CollectionSet::update_young_region_prediction(HeapRegion* hr,
 155                                                      size_t new_rs_length) {
 156   // Update the CSet information that is dependent on the new RS length
 157   assert(hr->is_young(), "Precondition");
 158   assert(!SafepointSynchronize::is_at_safepoint(), "should not be at a safepoint");
 159 
 160   // We could have updated _inc_recorded_rs_lengths and
 161   // _inc_predicted_elapsed_time_ms directly but we'd need to do
 162   // that atomically, as this code is executed by a concurrent
 163   // refinement thread, potentially concurrently with a mutator thread
 164   // allocating a new region and also updating the same fields. To
 165   // avoid the atomic operations we accumulate these updates on two
 166   // separate fields (*_diffs) and we'll just add them to the "main"
 167   // fields at the start of a GC.
 168 
 169   ssize_t old_rs_length = (ssize_t) hr->recorded_rs_length();
 170   ssize_t rs_lengths_diff = (ssize_t) new_rs_length - old_rs_length;
 171   _inc_recorded_rs_lengths_diffs += rs_lengths_diff;
 172 
 173   double old_elapsed_time_ms = hr->predicted_elapsed_time_ms();
 174   double new_region_elapsed_time_ms = predict_region_elapsed_time_ms(hr);
 175   double elapsed_ms_diff = new_region_elapsed_time_ms - old_elapsed_time_ms;
 176   _inc_predicted_elapsed_time_ms_diffs += elapsed_ms_diff;
 177 
 178   hr->set_recorded_rs_length(new_rs_length);
 179   hr->set_predicted_elapsed_time_ms(new_region_elapsed_time_ms);
 180 }
 181 
 182 void G1CollectionSet::add_young_region_common(HeapRegion* hr) {
 183   assert(hr->is_young(), "invariant");
 184   assert(_inc_build_state == Active, "Precondition");
 185 
 186   hr->set_young_index_in_cset(_inc_region_length);
 187   _inc_region_length++;
 188 
 189   // This routine is used when:
 190   // * adding survivor regions to the incremental cset at the end of an
 191   //   evacuation pause or
 192   // * adding the current allocation region to the incremental cset
 193   //   when it is retired.
 194   // Therefore this routine may be called at a safepoint by the
 195   // VM thread, or in-between safepoints by mutator threads (when
 196   // retiring the current allocation region)
 197   // We need to clear and set the cached recorded/cached collection set
 198   // information in the heap region here (before the region gets added
 199   // to the collection set). An individual heap region's cached values
 200   // are calculated, aggregated with the policy collection set info,
 201   // and cached in the heap region here (initially) and (subsequently)
 202   // by the Young List sampling code.
 203 
 204   size_t rs_length = hr->rem_set()->occupied();
 205   double region_elapsed_time_ms = predict_region_elapsed_time_ms(hr);
 206 
 207   // Cache the values we have added to the aggregated information
 208   // in the heap region in case we have to remove this region from
 209   // the incremental collection set, or it is updated by the
 210   // rset sampling code
 211   hr->set_recorded_rs_length(rs_length);
 212   hr->set_predicted_elapsed_time_ms(region_elapsed_time_ms);
 213 
 214   size_t used_bytes = hr->used();
 215   _inc_recorded_rs_lengths += rs_length;
 216   _inc_predicted_elapsed_time_ms += region_elapsed_time_ms;
 217   _inc_bytes_used_before += used_bytes;
 218 
 219   assert(!hr->in_collection_set(), "invariant");
 220   _g1->register_young_region_with_cset(hr);
 221   assert(hr->next_in_collection_set() == NULL, "invariant");
 222 }
 223 
 224 // Add the region at the RHS of the incremental cset
 225 void G1CollectionSet::add_survivor_regions(HeapRegion* hr) {
 226   // We should only ever be appending survivors at the end of a pause
 227   assert(hr->is_survivor(), "Logic");
 228 
 229   // Do the 'common' stuff
 230   add_young_region_common(hr);
 231 
 232   // Now add the region at the right hand side
 233   if (_inc_tail == NULL) {
 234     assert(_inc_head == NULL, "invariant");
 235     _inc_head = hr;
 236   } else {
 237     _inc_tail->set_next_in_collection_set(hr);
 238   }
 239   _inc_tail = hr;
 240 }
 241 
 242 // Add the region to the LHS of the incremental cset
 243 void G1CollectionSet::add_eden_region(HeapRegion* hr) {
 244   // Survivors should be added to the RHS at the end of a pause
 245   assert(hr->is_eden(), "Logic");
 246 
 247   // Do the 'common' stuff
 248   add_young_region_common(hr);
 249 
 250   // Add the region at the left hand side
 251   hr->set_next_in_collection_set(_inc_head);
 252   if (_inc_head == NULL) {
 253     assert(_inc_tail == NULL, "Invariant");
 254     _inc_tail = hr;
 255   }
 256   _inc_head = hr;
 257 }
 258 
 259 #ifndef PRODUCT
 260 void G1CollectionSet::print(HeapRegion* list_head, outputStream* st) {
 261   assert(list_head == inc_head() || list_head == head(), "must be");
 262 
 263   st->print_cr("\nCollection_set:");
 264   HeapRegion* csr = list_head;
 265   while (csr != NULL) {
 266     HeapRegion* next = csr->next_in_collection_set();
 267     assert(csr->in_collection_set(), "bad CS");
 268     st->print_cr("  " HR_FORMAT ", P: " PTR_FORMAT "N: " PTR_FORMAT ", age: %4d",
 269                  HR_FORMAT_PARAMS(csr),
 270                  p2i(csr->prev_top_at_mark_start()), p2i(csr->next_top_at_mark_start()),
 271                  csr->age_in_surv_rate_group_cond());
 272     csr = next;
 273   }
 274 }
 275 #endif // !PRODUCT
 276 
 277 double G1CollectionSet::finalize_young_part(double target_pause_time_ms) {
 278   double young_start_time_sec = os::elapsedTime();
 279 
 280   YoungList* young_list = _g1->young_list();
 281   finalize_incremental_building();
 282 
 283   guarantee(target_pause_time_ms > 0.0,
 284             "target_pause_time_ms = %1.6lf should be positive", target_pause_time_ms);
 285   guarantee(_head == NULL, "Precondition");
 286 
 287   size_t pending_cards = _policy->pending_cards();
 288   double base_time_ms = _policy->predict_base_elapsed_time_ms(pending_cards);
 289   double time_remaining_ms = MAX2(target_pause_time_ms - base_time_ms, 0.0);
 290 
 291   log_trace(gc, ergo, cset)("Start choosing CSet. pending cards: " SIZE_FORMAT " predicted base time: %1.2fms remaining time: %1.2fms target pause time: %1.2fms",
 292                             pending_cards, base_time_ms, time_remaining_ms, target_pause_time_ms);
 293 
 294   collector_state()->set_last_gc_was_young(collector_state()->gcs_are_young());
 295 
 296   // The young list is laid with the survivor regions from the previous
 297   // pause are appended to the RHS of the young list, i.e.
 298   //   [Newly Young Regions ++ Survivors from last pause].
 299 
 300   uint survivor_region_length = young_list->survivor_length();
 301   uint eden_region_length = young_list->eden_length();
 302   init_region_lengths(eden_region_length, survivor_region_length);
 303 
 304   const GrowableArray<HeapRegion*>* survivor_regions = _g1->young_list()->survivor_regions();
 305   for (GrowableArrayIterator<HeapRegion*> it = survivor_regions->begin();
 306        it != survivor_regions->end();
 307        ++it) {
 308     HeapRegion* hr = *it;
 309     assert(hr->is_survivor(), "badly formed young list");
 310     // There is a convention that all the young regions in the CSet
 311     // are tagged as "eden", so we do this for the survivors here. We
 312     // use the special set_eden_pre_gc() as it doesn't check that the
 313     // region is free (which is not the case here).
 314     hr->set_eden_pre_gc();
 315   }
 316 
 317   verify_young_cset_indices();
 318 
 319   // Clear the fields that point to the survivor list - they are all young now.
 320   young_list->clear_survivors();
 321 
 322   _head = _inc_head;
 323   _bytes_used_before = _inc_bytes_used_before;
 324   time_remaining_ms = MAX2(time_remaining_ms - _inc_predicted_elapsed_time_ms, 0.0);
 325 
 326   log_trace(gc, ergo, cset)("Add young regions to CSet. eden: %u regions, survivors: %u regions, predicted young region time: %1.2fms, target pause time: %1.2fms",
 327                             eden_region_length, survivor_region_length, _inc_predicted_elapsed_time_ms, target_pause_time_ms);
 328 
 329   // The number of recorded young regions is the incremental
 330   // collection set's current size
 331   set_recorded_rs_lengths(_inc_recorded_rs_lengths);
 332 
 333   double young_end_time_sec = os::elapsedTime();
 334   phase_times()->record_young_cset_choice_time_ms((young_end_time_sec - young_start_time_sec) * 1000.0);
 335 
 336   return time_remaining_ms;
 337 }
 338 
 339 void G1CollectionSet::finalize_old_part(double time_remaining_ms) {
 340   double non_young_start_time_sec = os::elapsedTime();
 341   double predicted_old_time_ms = 0.0;
 342 
 343   if (!collector_state()->gcs_are_young()) {
 344     cset_chooser()->verify();
 345     const uint min_old_cset_length = _policy->calc_min_old_cset_length();
 346     const uint max_old_cset_length = _policy->calc_max_old_cset_length();
 347 
 348     uint expensive_region_num = 0;
 349     bool check_time_remaining = _policy->adaptive_young_list_length();
 350 
 351     HeapRegion* hr = cset_chooser()->peek();
 352     while (hr != NULL) {
 353       if (old_region_length() >= max_old_cset_length) {
 354         // Added maximum number of old regions to the CSet.
 355         log_debug(gc, ergo, cset)("Finish adding old regions to CSet (old CSet region num reached max). old %u regions, max %u regions",
 356                                   old_region_length(), max_old_cset_length);
 357         break;
 358       }
 359 
 360       // Stop adding regions if the remaining reclaimable space is
 361       // not above G1HeapWastePercent.
 362       size_t reclaimable_bytes = cset_chooser()->remaining_reclaimable_bytes();
 363       double reclaimable_perc = _policy->reclaimable_bytes_perc(reclaimable_bytes);
 364       double threshold = (double) G1HeapWastePercent;
 365       if (reclaimable_perc <= threshold) {
 366         // We've added enough old regions that the amount of uncollected
 367         // reclaimable space is at or below the waste threshold. Stop
 368         // adding old regions to the CSet.
 369         log_debug(gc, ergo, cset)("Finish adding old regions to CSet (reclaimable percentage not over threshold). "
 370                                   "old %u regions, max %u regions, reclaimable: " SIZE_FORMAT "B (%1.2f%%) threshold: " UINTX_FORMAT "%%",
 371                                   old_region_length(), max_old_cset_length, reclaimable_bytes, reclaimable_perc, G1HeapWastePercent);
 372         break;
 373       }
 374 
 375       double predicted_time_ms = predict_region_elapsed_time_ms(hr);
 376       if (check_time_remaining) {
 377         if (predicted_time_ms > time_remaining_ms) {
 378           // Too expensive for the current CSet.
 379 
 380           if (old_region_length() >= min_old_cset_length) {
 381             // We have added the minimum number of old regions to the CSet,
 382             // we are done with this CSet.
 383             log_debug(gc, ergo, cset)("Finish adding old regions to CSet (predicted time is too high). "
 384                                       "predicted time: %1.2fms, remaining time: %1.2fms old %u regions, min %u regions",
 385                                       predicted_time_ms, time_remaining_ms, old_region_length(), min_old_cset_length);
 386             break;
 387           }
 388 
 389           // We'll add it anyway given that we haven't reached the
 390           // minimum number of old regions.
 391           expensive_region_num += 1;
 392         }
 393       } else {
 394         if (old_region_length() >= min_old_cset_length) {
 395           // In the non-auto-tuning case, we'll finish adding regions
 396           // to the CSet if we reach the minimum.
 397 
 398           log_debug(gc, ergo, cset)("Finish adding old regions to CSet (old CSet region num reached min). old %u regions, min %u regions",
 399                                     old_region_length(), min_old_cset_length);
 400           break;
 401         }
 402       }
 403 
 404       // We will add this region to the CSet.
 405       time_remaining_ms = MAX2(time_remaining_ms - predicted_time_ms, 0.0);
 406       predicted_old_time_ms += predicted_time_ms;
 407       cset_chooser()->pop(); // already have region via peek()
 408       _g1->old_set_remove(hr);
 409       add_old_region(hr);
 410 
 411       hr = cset_chooser()->peek();
 412     }
 413     if (hr == NULL) {
 414       log_debug(gc, ergo, cset)("Finish adding old regions to CSet (candidate old regions not available)");
 415     }
 416 
 417     if (expensive_region_num > 0) {
 418       // We print the information once here at the end, predicated on
 419       // whether we added any apparently expensive regions or not, to
 420       // avoid generating output per region.
 421       log_debug(gc, ergo, cset)("Added expensive regions to CSet (old CSet region num not reached min)."
 422                                 "old: %u regions, expensive: %u regions, min: %u regions, remaining time: %1.2fms",
 423                                 old_region_length(), expensive_region_num, min_old_cset_length, time_remaining_ms);
 424     }
 425 
 426     cset_chooser()->verify();
 427   }
 428 
 429   stop_incremental_building();
 430 
 431   log_debug(gc, ergo, cset)("Finish choosing CSet. old: %u regions, predicted old region time: %1.2fms, time remaining: %1.2f",
 432                             old_region_length(), predicted_old_time_ms, time_remaining_ms);
 433 
 434   double non_young_end_time_sec = os::elapsedTime();
 435   phase_times()->record_non_young_cset_choice_time_ms((non_young_end_time_sec - non_young_start_time_sec) * 1000.0);
 436 }
 437 
 438 #ifdef ASSERT
 439 void G1CollectionSet::verify_young_cset_indices() const {
 440   ResourceMark rm;
 441   uint* heap_region_indices = NEW_RESOURCE_ARRAY(uint, young_region_length());
 442   for (uint i = 0; i < young_region_length(); ++i) {
 443     heap_region_indices[i] = (uint)-1;
 444   }
 445 
 446   for (HeapRegion* hr = _inc_head; hr != NULL; hr = hr->next_in_collection_set()) {
 447     const int idx = hr->young_index_in_cset();
 448     assert(idx > -1, "must be set for all inc cset regions");
 449     assert((uint)idx < young_region_length(), "young cset index too large");
 450 
 451     assert(heap_region_indices[idx] == (uint)-1,
 452            "index %d used by multiple regions, first use by %u, second by %u",
 453            idx, heap_region_indices[idx], hr->hrm_index());
 454 
 455     heap_region_indices[idx] = hr->hrm_index();
 456   }
 457 }
 458 #endif