1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/src/share/vm/gc_implementation/g1/concurrentMark.hpp Thu Jun 05 15:57:56 2008 -0700
1.3 @@ -0,0 +1,1049 @@
1.4 +/*
1.5 + * Copyright 2001-2007 Sun Microsystems, Inc. All Rights Reserved.
1.6 + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
1.7 + *
1.8 + * This code is free software; you can redistribute it and/or modify it
1.9 + * under the terms of the GNU General Public License version 2 only, as
1.10 + * published by the Free Software Foundation.
1.11 + *
1.12 + * This code is distributed in the hope that it will be useful, but WITHOUT
1.13 + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
1.14 + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
1.15 + * version 2 for more details (a copy is included in the LICENSE file that
1.16 + * accompanied this code).
1.17 + *
1.18 + * You should have received a copy of the GNU General Public License version
1.19 + * 2 along with this work; if not, write to the Free Software Foundation,
1.20 + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
1.21 + *
1.22 + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
1.23 + * CA 95054 USA or visit www.sun.com if you need additional information or
1.24 + * have any questions.
1.25 + *
1.26 + */
1.27 +
1.28 +class G1CollectedHeap;
1.29 +class CMTask;
1.30 +typedef GenericTaskQueue<oop> CMTaskQueue;
1.31 +typedef GenericTaskQueueSet<oop> CMTaskQueueSet;
1.32 +
1.33 +// A generic CM bit map. This is essentially a wrapper around the BitMap
1.34 +// class, with one bit per (1<<_shifter) HeapWords.
1.35 +
1.36 +class CMBitMapRO {
1.37 + protected:
1.38 + HeapWord* _bmStartWord; // base address of range covered by map
1.39 + size_t _bmWordSize; // map size (in #HeapWords covered)
1.40 + const int _shifter; // map to char or bit
1.41 + VirtualSpace _virtual_space; // underlying the bit map
1.42 + BitMap _bm; // the bit map itself
1.43 +
1.44 + public:
1.45 + // constructor
1.46 + CMBitMapRO(ReservedSpace rs, int shifter);
1.47 +
1.48 + enum { do_yield = true };
1.49 +
1.50 + // inquiries
1.51 + HeapWord* startWord() const { return _bmStartWord; }
1.52 + size_t sizeInWords() const { return _bmWordSize; }
1.53 + // the following is one past the last word in space
1.54 + HeapWord* endWord() const { return _bmStartWord + _bmWordSize; }
1.55 +
1.56 + // read marks
1.57 +
1.58 + bool isMarked(HeapWord* addr) const {
1.59 + assert(_bmStartWord <= addr && addr < (_bmStartWord + _bmWordSize),
1.60 + "outside underlying space?");
1.61 + return _bm.at(heapWordToOffset(addr));
1.62 + }
1.63 +
1.64 + // iteration
1.65 + bool iterate(BitMapClosure* cl) { return _bm.iterate(cl); }
1.66 + bool iterate(BitMapClosure* cl, MemRegion mr);
1.67 +
1.68 + // Return the address corresponding to the next marked bit at or after
1.69 + // "addr", and before "limit", if "limit" is non-NULL. If there is no
1.70 + // such bit, returns "limit" if that is non-NULL, or else "endWord()".
1.71 + HeapWord* getNextMarkedWordAddress(HeapWord* addr,
1.72 + HeapWord* limit = NULL) const;
1.73 + // Return the address corresponding to the next unmarked bit at or after
1.74 + // "addr", and before "limit", if "limit" is non-NULL. If there is no
1.75 + // such bit, returns "limit" if that is non-NULL, or else "endWord()".
1.76 + HeapWord* getNextUnmarkedWordAddress(HeapWord* addr,
1.77 + HeapWord* limit = NULL) const;
1.78 +
1.79 + // conversion utilities
1.80 + // XXX Fix these so that offsets are size_t's...
1.81 + HeapWord* offsetToHeapWord(size_t offset) const {
1.82 + return _bmStartWord + (offset << _shifter);
1.83 + }
1.84 + size_t heapWordToOffset(HeapWord* addr) const {
1.85 + return pointer_delta(addr, _bmStartWord) >> _shifter;
1.86 + }
1.87 + int heapWordDiffToOffsetDiff(size_t diff) const;
1.88 + HeapWord* nextWord(HeapWord* addr) {
1.89 + return offsetToHeapWord(heapWordToOffset(addr) + 1);
1.90 + }
1.91 +
1.92 + void mostly_disjoint_range_union(BitMap* from_bitmap,
1.93 + size_t from_start_index,
1.94 + HeapWord* to_start_word,
1.95 + size_t word_num);
1.96 +
1.97 + // debugging
1.98 + NOT_PRODUCT(bool covers(ReservedSpace rs) const;)
1.99 +};
1.100 +
1.101 +class CMBitMap : public CMBitMapRO {
1.102 +
1.103 + public:
1.104 + // constructor
1.105 + CMBitMap(ReservedSpace rs, int shifter) :
1.106 + CMBitMapRO(rs, shifter) {}
1.107 +
1.108 + // write marks
1.109 + void mark(HeapWord* addr) {
1.110 + assert(_bmStartWord <= addr && addr < (_bmStartWord + _bmWordSize),
1.111 + "outside underlying space?");
1.112 + _bm.at_put(heapWordToOffset(addr), true);
1.113 + }
1.114 + void clear(HeapWord* addr) {
1.115 + assert(_bmStartWord <= addr && addr < (_bmStartWord + _bmWordSize),
1.116 + "outside underlying space?");
1.117 + _bm.at_put(heapWordToOffset(addr), false);
1.118 + }
1.119 + bool parMark(HeapWord* addr) {
1.120 + assert(_bmStartWord <= addr && addr < (_bmStartWord + _bmWordSize),
1.121 + "outside underlying space?");
1.122 + return _bm.par_at_put(heapWordToOffset(addr), true);
1.123 + }
1.124 + bool parClear(HeapWord* addr) {
1.125 + assert(_bmStartWord <= addr && addr < (_bmStartWord + _bmWordSize),
1.126 + "outside underlying space?");
1.127 + return _bm.par_at_put(heapWordToOffset(addr), false);
1.128 + }
1.129 + void markRange(MemRegion mr);
1.130 + void clearAll();
1.131 + void clearRange(MemRegion mr);
1.132 +
1.133 + // Starting at the bit corresponding to "addr" (inclusive), find the next
1.134 + // "1" bit, if any. This bit starts some run of consecutive "1"'s; find
1.135 + // the end of this run (stopping at "end_addr"). Return the MemRegion
1.136 + // covering from the start of the region corresponding to the first bit
1.137 + // of the run to the end of the region corresponding to the last bit of
1.138 + // the run. If there is no "1" bit at or after "addr", return an empty
1.139 + // MemRegion.
1.140 + MemRegion getAndClearMarkedRegion(HeapWord* addr, HeapWord* end_addr);
1.141 +};
1.142 +
1.143 +// Represents a marking stack used by the CM collector.
1.144 +// Ideally this should be GrowableArray<> just like MSC's marking stack(s).
1.145 +class CMMarkStack {
1.146 + ConcurrentMark* _cm;
1.147 + oop* _base; // bottom of stack
1.148 + jint _index; // one more than last occupied index
1.149 + jint _capacity; // max #elements
1.150 + jint _oops_do_bound; // Number of elements to include in next iteration.
1.151 + NOT_PRODUCT(jint _max_depth;) // max depth plumbed during run
1.152 +
1.153 + bool _overflow;
1.154 + DEBUG_ONLY(bool _drain_in_progress;)
1.155 + DEBUG_ONLY(bool _drain_in_progress_yields;)
1.156 +
1.157 + public:
1.158 + CMMarkStack(ConcurrentMark* cm);
1.159 + ~CMMarkStack();
1.160 +
1.161 + void allocate(size_t size);
1.162 +
1.163 + oop pop() {
1.164 + if (!isEmpty()) {
1.165 + return _base[--_index] ;
1.166 + }
1.167 + return NULL;
1.168 + }
1.169 +
1.170 + // If overflow happens, don't do the push, and record the overflow.
1.171 + // *Requires* that "ptr" is already marked.
1.172 + void push(oop ptr) {
1.173 + if (isFull()) {
1.174 + // Record overflow.
1.175 + _overflow = true;
1.176 + return;
1.177 + } else {
1.178 + _base[_index++] = ptr;
1.179 + NOT_PRODUCT(_max_depth = MAX2(_max_depth, _index));
1.180 + }
1.181 + }
1.182 + // Non-block impl. Note: concurrency is allowed only with other
1.183 + // "par_push" operations, not with "pop" or "drain". We would need
1.184 + // parallel versions of them if such concurrency was desired.
1.185 + void par_push(oop ptr);
1.186 +
1.187 + // Pushes the first "n" elements of "ptr_arr" on the stack.
1.188 + // Non-block impl. Note: concurrency is allowed only with other
1.189 + // "par_adjoin_arr" or "push" operations, not with "pop" or "drain".
1.190 + void par_adjoin_arr(oop* ptr_arr, int n);
1.191 +
1.192 + // Pushes the first "n" elements of "ptr_arr" on the stack.
1.193 + // Locking impl: concurrency is allowed only with
1.194 + // "par_push_arr" and/or "par_pop_arr" operations, which use the same
1.195 + // locking strategy.
1.196 + void par_push_arr(oop* ptr_arr, int n);
1.197 +
1.198 + // If returns false, the array was empty. Otherwise, removes up to "max"
1.199 + // elements from the stack, and transfers them to "ptr_arr" in an
1.200 + // unspecified order. The actual number transferred is given in "n" ("n
1.201 + // == 0" is deliberately redundant with the return value.) Locking impl:
1.202 + // concurrency is allowed only with "par_push_arr" and/or "par_pop_arr"
1.203 + // operations, which use the same locking strategy.
1.204 + bool par_pop_arr(oop* ptr_arr, int max, int* n);
1.205 +
1.206 + // Drain the mark stack, applying the given closure to all fields of
1.207 + // objects on the stack. (That is, continue until the stack is empty,
1.208 + // even if closure applications add entries to the stack.) The "bm"
1.209 + // argument, if non-null, may be used to verify that only marked objects
1.210 + // are on the mark stack. If "yield_after" is "true", then the
1.211 + // concurrent marker performing the drain offers to yield after
1.212 + // processing each object. If a yield occurs, stops the drain operation
1.213 + // and returns false. Otherwise, returns true.
1.214 + template<class OopClosureClass>
1.215 + bool drain(OopClosureClass* cl, CMBitMap* bm, bool yield_after = false);
1.216 +
1.217 + bool isEmpty() { return _index == 0; }
1.218 + bool isFull() { return _index == _capacity; }
1.219 + int maxElems() { return _capacity; }
1.220 +
1.221 + bool overflow() { return _overflow; }
1.222 + void clear_overflow() { _overflow = false; }
1.223 +
1.224 + int size() { return _index; }
1.225 +
1.226 + void setEmpty() { _index = 0; clear_overflow(); }
1.227 +
1.228 + // Record the current size; a subsequent "oops_do" will iterate only over
1.229 + // indices valid at the time of this call.
1.230 + void set_oops_do_bound(jint bound = -1) {
1.231 + if (bound == -1) {
1.232 + _oops_do_bound = _index;
1.233 + } else {
1.234 + _oops_do_bound = bound;
1.235 + }
1.236 + }
1.237 + jint oops_do_bound() { return _oops_do_bound; }
1.238 + // iterate over the oops in the mark stack, up to the bound recorded via
1.239 + // the call above.
1.240 + void oops_do(OopClosure* f);
1.241 +};
1.242 +
1.243 +class CMRegionStack {
1.244 + MemRegion* _base;
1.245 + jint _capacity;
1.246 + jint _index;
1.247 + jint _oops_do_bound;
1.248 + bool _overflow;
1.249 +public:
1.250 + CMRegionStack();
1.251 + ~CMRegionStack();
1.252 + void allocate(size_t size);
1.253 +
1.254 + // This is lock-free; assumes that it will only be called in parallel
1.255 + // with other "push" operations (no pops).
1.256 + void push(MemRegion mr);
1.257 +
1.258 + // Lock-free; assumes that it will only be called in parallel
1.259 + // with other "pop" operations (no pushes).
1.260 + MemRegion pop();
1.261 +
1.262 + bool isEmpty() { return _index == 0; }
1.263 + bool isFull() { return _index == _capacity; }
1.264 +
1.265 + bool overflow() { return _overflow; }
1.266 + void clear_overflow() { _overflow = false; }
1.267 +
1.268 + int size() { return _index; }
1.269 +
1.270 + // It iterates over the entries in the region stack and it
1.271 + // invalidates (i.e. assigns MemRegion()) the ones that point to
1.272 + // regions in the collection set.
1.273 + bool invalidate_entries_into_cset();
1.274 +
1.275 + // This gives an upper bound up to which the iteration in
1.276 + // invalidate_entries_into_cset() will reach. This prevents
1.277 + // newly-added entries to be unnecessarily scanned.
1.278 + void set_oops_do_bound() {
1.279 + _oops_do_bound = _index;
1.280 + }
1.281 +
1.282 + void setEmpty() { _index = 0; clear_overflow(); }
1.283 +};
1.284 +
1.285 +// this will enable a variety of different statistics per GC task
1.286 +#define _MARKING_STATS_ 0
1.287 +// this will enable the higher verbose levels
1.288 +#define _MARKING_VERBOSE_ 0
1.289 +
1.290 +#if _MARKING_STATS_
1.291 +#define statsOnly(statement) \
1.292 +do { \
1.293 + statement ; \
1.294 +} while (0)
1.295 +#else // _MARKING_STATS_
1.296 +#define statsOnly(statement) \
1.297 +do { \
1.298 +} while (0)
1.299 +#endif // _MARKING_STATS_
1.300 +
1.301 +// Some extra guarantees that I like to also enable in optimised mode
1.302 +// when debugging. If you want to enable them, comment out the assert
1.303 +// macro and uncomment out the guaratee macro
1.304 +// #define tmp_guarantee_CM(expr, str) guarantee(expr, str)
1.305 +#define tmp_guarantee_CM(expr, str) assert(expr, str)
1.306 +
1.307 +typedef enum {
1.308 + no_verbose = 0, // verbose turned off
1.309 + stats_verbose, // only prints stats at the end of marking
1.310 + low_verbose, // low verbose, mostly per region and per major event
1.311 + medium_verbose, // a bit more detailed than low
1.312 + high_verbose // per object verbose
1.313 +} CMVerboseLevel;
1.314 +
1.315 +
1.316 +class ConcurrentMarkThread;
1.317 +
1.318 +class ConcurrentMark {
1.319 + friend class ConcurrentMarkThread;
1.320 + friend class CMTask;
1.321 + friend class CMBitMapClosure;
1.322 + friend class CSMarkOopClosure;
1.323 + friend class CMGlobalObjectClosure;
1.324 + friend class CMRemarkTask;
1.325 + friend class CMConcurrentMarkingTask;
1.326 + friend class G1ParNoteEndTask;
1.327 + friend class CalcLiveObjectsClosure;
1.328 +
1.329 +protected:
1.330 + ConcurrentMarkThread* _cmThread; // the thread doing the work
1.331 + G1CollectedHeap* _g1h; // the heap.
1.332 + size_t _parallel_marking_threads; // the number of marking
1.333 + // threads we'll use
1.334 + double _sleep_factor; // how much we have to sleep, with
1.335 + // respect to the work we just did, to
1.336 + // meet the marking overhead goal
1.337 + double _marking_task_overhead; // marking target overhead for
1.338 + // a single task
1.339 +
1.340 + // same as the two above, but for the cleanup task
1.341 + double _cleanup_sleep_factor;
1.342 + double _cleanup_task_overhead;
1.343 +
1.344 + // Stuff related to age cohort processing.
1.345 + struct ParCleanupThreadState {
1.346 + char _pre[64];
1.347 + UncleanRegionList list;
1.348 + char _post[64];
1.349 + };
1.350 + ParCleanupThreadState** _par_cleanup_thread_state;
1.351 +
1.352 + // CMS marking support structures
1.353 + CMBitMap _markBitMap1;
1.354 + CMBitMap _markBitMap2;
1.355 + CMBitMapRO* _prevMarkBitMap; // completed mark bitmap
1.356 + CMBitMap* _nextMarkBitMap; // under-construction mark bitmap
1.357 + bool _at_least_one_mark_complete;
1.358 +
1.359 + BitMap _region_bm;
1.360 + BitMap _card_bm;
1.361 +
1.362 + // Heap bounds
1.363 + HeapWord* _heap_start;
1.364 + HeapWord* _heap_end;
1.365 +
1.366 + // For gray objects
1.367 + CMMarkStack _markStack; // Grey objects behind global finger.
1.368 + CMRegionStack _regionStack; // Grey regions behind global finger.
1.369 + HeapWord* volatile _finger; // the global finger, region aligned,
1.370 + // always points to the end of the
1.371 + // last claimed region
1.372 +
1.373 + // marking tasks
1.374 + size_t _max_task_num; // maximum task number
1.375 + size_t _active_tasks; // task num currently active
1.376 + CMTask** _tasks; // task queue array (max_task_num len)
1.377 + CMTaskQueueSet* _task_queues; // task queue set
1.378 + ParallelTaskTerminator _terminator; // for termination
1.379 +
1.380 + // Two sync barriers that are used to synchronise tasks when an
1.381 + // overflow occurs. The algorithm is the following. All tasks enter
1.382 + // the first one to ensure that they have all stopped manipulating
1.383 + // the global data structures. After they exit it, they re-initialise
1.384 + // their data structures and task 0 re-initialises the global data
1.385 + // structures. Then, they enter the second sync barrier. This
1.386 + // ensure, that no task starts doing work before all data
1.387 + // structures (local and global) have been re-initialised. When they
1.388 + // exit it, they are free to start working again.
1.389 + WorkGangBarrierSync _first_overflow_barrier_sync;
1.390 + WorkGangBarrierSync _second_overflow_barrier_sync;
1.391 +
1.392 +
1.393 + // this is set by any task, when an overflow on the global data
1.394 + // structures is detected.
1.395 + volatile bool _has_overflown;
1.396 + // true: marking is concurrent, false: we're in remark
1.397 + volatile bool _concurrent;
1.398 + // set at the end of a Full GC so that marking aborts
1.399 + volatile bool _has_aborted;
1.400 + // used when remark aborts due to an overflow to indicate that
1.401 + // another concurrent marking phase should start
1.402 + volatile bool _restart_for_overflow;
1.403 +
1.404 + // This is true from the very start of concurrent marking until the
1.405 + // point when all the tasks complete their work. It is really used
1.406 + // to determine the points between the end of concurrent marking and
1.407 + // time of remark.
1.408 + volatile bool _concurrent_marking_in_progress;
1.409 +
1.410 + // verbose level
1.411 + CMVerboseLevel _verbose_level;
1.412 +
1.413 + COTracker _cleanup_co_tracker;
1.414 +
1.415 + // These two fields are used to implement the optimisation that
1.416 + // avoids pushing objects on the global/region stack if there are
1.417 + // no collection set regions above the lowest finger.
1.418 +
1.419 + // This is the lowest finger (among the global and local fingers),
1.420 + // which is calculated before a new collection set is chosen.
1.421 + HeapWord* _min_finger;
1.422 + // If this flag is true, objects/regions that are marked below the
1.423 + // finger should be pushed on the stack(s). If this is flag is
1.424 + // false, it is safe not to push them on the stack(s).
1.425 + bool _should_gray_objects;
1.426 +
1.427 + // All of these times are in ms.
1.428 + NumberSeq _init_times;
1.429 + NumberSeq _remark_times;
1.430 + NumberSeq _remark_mark_times;
1.431 + NumberSeq _remark_weak_ref_times;
1.432 + NumberSeq _cleanup_times;
1.433 + double _total_counting_time;
1.434 + double _total_rs_scrub_time;
1.435 +
1.436 + double* _accum_task_vtime; // accumulated task vtime
1.437 +
1.438 + WorkGang* _parallel_workers;
1.439 +
1.440 + void weakRefsWork(bool clear_all_soft_refs);
1.441 +
1.442 + void swapMarkBitMaps();
1.443 +
1.444 + // It resets the global marking data structures, as well as the
1.445 + // task local ones; should be called during initial mark.
1.446 + void reset();
1.447 + // It resets all the marking data structures.
1.448 + void clear_marking_state();
1.449 +
1.450 + // It should be called to indicate which phase we're in (concurrent
1.451 + // mark or remark) and how many threads are currently active.
1.452 + void set_phase(size_t active_tasks, bool concurrent);
1.453 + // We do this after we're done with marking so that the marking data
1.454 + // structures are initialised to a sensible and predictable state.
1.455 + void set_non_marking_state();
1.456 +
1.457 + // prints all gathered CM-related statistics
1.458 + void print_stats();
1.459 +
1.460 + // accessor methods
1.461 + size_t parallel_marking_threads() { return _parallel_marking_threads; }
1.462 + double sleep_factor() { return _sleep_factor; }
1.463 + double marking_task_overhead() { return _marking_task_overhead;}
1.464 + double cleanup_sleep_factor() { return _cleanup_sleep_factor; }
1.465 + double cleanup_task_overhead() { return _cleanup_task_overhead;}
1.466 +
1.467 + HeapWord* finger() { return _finger; }
1.468 + bool concurrent() { return _concurrent; }
1.469 + size_t active_tasks() { return _active_tasks; }
1.470 + ParallelTaskTerminator* terminator() { return &_terminator; }
1.471 +
1.472 + // It claims the next available region to be scanned by a marking
1.473 + // task. It might return NULL if the next region is empty or we have
1.474 + // run out of regions. In the latter case, out_of_regions()
1.475 + // determines whether we've really run out of regions or the task
1.476 + // should call claim_region() again. This might seem a bit
1.477 + // awkward. Originally, the code was written so that claim_region()
1.478 + // either successfully returned with a non-empty region or there
1.479 + // were no more regions to be claimed. The problem with this was
1.480 + // that, in certain circumstances, it iterated over large chunks of
1.481 + // the heap finding only empty regions and, while it was working, it
1.482 + // was preventing the calling task to call its regular clock
1.483 + // method. So, this way, each task will spend very little time in
1.484 + // claim_region() and is allowed to call the regular clock method
1.485 + // frequently.
1.486 + HeapRegion* claim_region(int task);
1.487 +
1.488 + // It determines whether we've run out of regions to scan.
1.489 + bool out_of_regions() { return _finger == _heap_end; }
1.490 +
1.491 + // Returns the task with the given id
1.492 + CMTask* task(int id) {
1.493 + guarantee( 0 <= id && id < (int) _active_tasks, "task id not within "
1.494 + "active bounds" );
1.495 + return _tasks[id];
1.496 + }
1.497 +
1.498 + // Returns the task queue with the given id
1.499 + CMTaskQueue* task_queue(int id) {
1.500 + guarantee( 0 <= id && id < (int) _active_tasks, "task queue id not within "
1.501 + "active bounds" );
1.502 + return (CMTaskQueue*) _task_queues->queue(id);
1.503 + }
1.504 +
1.505 + // Returns the task queue set
1.506 + CMTaskQueueSet* task_queues() { return _task_queues; }
1.507 +
1.508 + // Access / manipulation of the overflow flag which is set to
1.509 + // indicate that the global stack or region stack has overflown
1.510 + bool has_overflown() { return _has_overflown; }
1.511 + void set_has_overflown() { _has_overflown = true; }
1.512 + void clear_has_overflown() { _has_overflown = false; }
1.513 +
1.514 + bool has_aborted() { return _has_aborted; }
1.515 + bool restart_for_overflow() { return _restart_for_overflow; }
1.516 +
1.517 + // Methods to enter the two overflow sync barriers
1.518 + void enter_first_sync_barrier(int task_num);
1.519 + void enter_second_sync_barrier(int task_num);
1.520 +
1.521 +public:
1.522 + // Manipulation of the global mark stack.
1.523 + // Notice that the first mark_stack_push is CAS-based, whereas the
1.524 + // two below are Mutex-based. This is OK since the first one is only
1.525 + // called during evacuation pauses and doesn't compete with the
1.526 + // other two (which are called by the marking tasks during
1.527 + // concurrent marking or remark).
1.528 + bool mark_stack_push(oop p) {
1.529 + _markStack.par_push(p);
1.530 + if (_markStack.overflow()) {
1.531 + set_has_overflown();
1.532 + return false;
1.533 + }
1.534 + return true;
1.535 + }
1.536 + bool mark_stack_push(oop* arr, int n) {
1.537 + _markStack.par_push_arr(arr, n);
1.538 + if (_markStack.overflow()) {
1.539 + set_has_overflown();
1.540 + return false;
1.541 + }
1.542 + return true;
1.543 + }
1.544 + void mark_stack_pop(oop* arr, int max, int* n) {
1.545 + _markStack.par_pop_arr(arr, max, n);
1.546 + }
1.547 + size_t mark_stack_size() { return _markStack.size(); }
1.548 + size_t partial_mark_stack_size_target() { return _markStack.maxElems()/3; }
1.549 + bool mark_stack_overflow() { return _markStack.overflow(); }
1.550 + bool mark_stack_empty() { return _markStack.isEmpty(); }
1.551 +
1.552 + // Manipulation of the region stack
1.553 + bool region_stack_push(MemRegion mr) {
1.554 + _regionStack.push(mr);
1.555 + if (_regionStack.overflow()) {
1.556 + set_has_overflown();
1.557 + return false;
1.558 + }
1.559 + return true;
1.560 + }
1.561 + MemRegion region_stack_pop() { return _regionStack.pop(); }
1.562 + int region_stack_size() { return _regionStack.size(); }
1.563 + bool region_stack_overflow() { return _regionStack.overflow(); }
1.564 + bool region_stack_empty() { return _regionStack.isEmpty(); }
1.565 +
1.566 + bool concurrent_marking_in_progress() {
1.567 + return _concurrent_marking_in_progress;
1.568 + }
1.569 + void set_concurrent_marking_in_progress() {
1.570 + _concurrent_marking_in_progress = true;
1.571 + }
1.572 + void clear_concurrent_marking_in_progress() {
1.573 + _concurrent_marking_in_progress = false;
1.574 + }
1.575 +
1.576 + void update_accum_task_vtime(int i, double vtime) {
1.577 + _accum_task_vtime[i] += vtime;
1.578 + }
1.579 +
1.580 + double all_task_accum_vtime() {
1.581 + double ret = 0.0;
1.582 + for (int i = 0; i < (int)_max_task_num; ++i)
1.583 + ret += _accum_task_vtime[i];
1.584 + return ret;
1.585 + }
1.586 +
1.587 + // Attempts to steal an object from the task queues of other tasks
1.588 + bool try_stealing(int task_num, int* hash_seed, oop& obj) {
1.589 + return _task_queues->steal(task_num, hash_seed, obj);
1.590 + }
1.591 +
1.592 + // It grays an object by first marking it. Then, if it's behind the
1.593 + // global finger, it also pushes it on the global stack.
1.594 + void deal_with_reference(oop obj);
1.595 +
1.596 + ConcurrentMark(ReservedSpace rs, int max_regions);
1.597 + ~ConcurrentMark();
1.598 + ConcurrentMarkThread* cmThread() { return _cmThread; }
1.599 +
1.600 + CMBitMapRO* prevMarkBitMap() const { return _prevMarkBitMap; }
1.601 + CMBitMap* nextMarkBitMap() const { return _nextMarkBitMap; }
1.602 +
1.603 + // The following three are interaction between CM and
1.604 + // G1CollectedHeap
1.605 +
1.606 + // This notifies CM that a root during initial-mark needs to be
1.607 + // grayed and it's MT-safe. Currently, we just mark it. But, in the
1.608 + // future, we can experiment with pushing it on the stack and we can
1.609 + // do this without changing G1CollectedHeap.
1.610 + void grayRoot(oop p);
1.611 + // It's used during evacuation pauses to gray a region, if
1.612 + // necessary, and it's MT-safe. It assumes that the caller has
1.613 + // marked any objects on that region. If _should_gray_objects is
1.614 + // true and we're still doing concurrent marking, the region is
1.615 + // pushed on the region stack, if it is located below the global
1.616 + // finger, otherwise we do nothing.
1.617 + void grayRegionIfNecessary(MemRegion mr);
1.618 + // It's used during evacuation pauses to mark and, if necessary,
1.619 + // gray a single object and it's MT-safe. It assumes the caller did
1.620 + // not mark the object. If _should_gray_objects is true and we're
1.621 + // still doing concurrent marking, the objects is pushed on the
1.622 + // global stack, if it is located below the global finger, otherwise
1.623 + // we do nothing.
1.624 + void markAndGrayObjectIfNecessary(oop p);
1.625 +
1.626 + // This iterates over the bitmap of the previous marking and prints
1.627 + // out all objects that are marked on the bitmap and indicates
1.628 + // whether what they point to is also marked or not.
1.629 + void print_prev_bitmap_reachable();
1.630 +
1.631 + // Clear the next marking bitmap (will be called concurrently).
1.632 + void clearNextBitmap();
1.633 +
1.634 + // main CMS steps and related support
1.635 + void checkpointRootsInitial();
1.636 +
1.637 + // These two do the work that needs to be done before and after the
1.638 + // initial root checkpoint. Since this checkpoint can be done at two
1.639 + // different points (i.e. an explicit pause or piggy-backed on a
1.640 + // young collection), then it's nice to be able to easily share the
1.641 + // pre/post code. It might be the case that we can put everything in
1.642 + // the post method. TP
1.643 + void checkpointRootsInitialPre();
1.644 + void checkpointRootsInitialPost();
1.645 +
1.646 + // Do concurrent phase of marking, to a tentative transitive closure.
1.647 + void markFromRoots();
1.648 +
1.649 + // Process all unprocessed SATB buffers. It is called at the
1.650 + // beginning of an evacuation pause.
1.651 + void drainAllSATBBuffers();
1.652 +
1.653 + void checkpointRootsFinal(bool clear_all_soft_refs);
1.654 + void checkpointRootsFinalWork();
1.655 + void calcDesiredRegions();
1.656 + void cleanup();
1.657 + void completeCleanup();
1.658 +
1.659 + // Mark in the previous bitmap. NB: this is usually read-only, so use
1.660 + // this carefully!
1.661 + void markPrev(oop p);
1.662 + void clear(oop p);
1.663 + // Clears marks for all objects in the given range, for both prev and
1.664 + // next bitmaps. NB: the previous bitmap is usually read-only, so use
1.665 + // this carefully!
1.666 + void clearRangeBothMaps(MemRegion mr);
1.667 +
1.668 + // Record the current top of the mark and region stacks; a
1.669 + // subsequent oops_do() on the mark stack and
1.670 + // invalidate_entries_into_cset() on the region stack will iterate
1.671 + // only over indices valid at the time of this call.
1.672 + void set_oops_do_bound() {
1.673 + _markStack.set_oops_do_bound();
1.674 + _regionStack.set_oops_do_bound();
1.675 + }
1.676 + // Iterate over the oops in the mark stack and all local queues. It
1.677 + // also calls invalidate_entries_into_cset() on the region stack.
1.678 + void oops_do(OopClosure* f);
1.679 + // It is called at the end of an evacuation pause during marking so
1.680 + // that CM is notified of where the new end of the heap is. It
1.681 + // doesn't do anything if concurrent_marking_in_progress() is false,
1.682 + // unless the force parameter is true.
1.683 + void update_g1_committed(bool force = false);
1.684 +
1.685 + void complete_marking_in_collection_set();
1.686 +
1.687 + // It indicates that a new collection set is being chosen.
1.688 + void newCSet();
1.689 + // It registers a collection set heap region with CM. This is used
1.690 + // to determine whether any heap regions are located above the finger.
1.691 + void registerCSetRegion(HeapRegion* hr);
1.692 +
1.693 + // Returns "true" if at least one mark has been completed.
1.694 + bool at_least_one_mark_complete() { return _at_least_one_mark_complete; }
1.695 +
1.696 + bool isMarked(oop p) const {
1.697 + assert(p != NULL && p->is_oop(), "expected an oop");
1.698 + HeapWord* addr = (HeapWord*)p;
1.699 + assert(addr >= _nextMarkBitMap->startWord() ||
1.700 + addr < _nextMarkBitMap->endWord(), "in a region");
1.701 +
1.702 + return _nextMarkBitMap->isMarked(addr);
1.703 + }
1.704 +
1.705 + inline bool not_yet_marked(oop p) const;
1.706 +
1.707 + // XXX Debug code
1.708 + bool containing_card_is_marked(void* p);
1.709 + bool containing_cards_are_marked(void* start, void* last);
1.710 +
1.711 + bool isPrevMarked(oop p) const {
1.712 + assert(p != NULL && p->is_oop(), "expected an oop");
1.713 + HeapWord* addr = (HeapWord*)p;
1.714 + assert(addr >= _prevMarkBitMap->startWord() ||
1.715 + addr < _prevMarkBitMap->endWord(), "in a region");
1.716 +
1.717 + return _prevMarkBitMap->isMarked(addr);
1.718 + }
1.719 +
1.720 + inline bool do_yield_check(int worker_i = 0);
1.721 + inline bool should_yield();
1.722 +
1.723 + // Called to abort the marking cycle after a Full GC takes palce.
1.724 + void abort();
1.725 +
1.726 + void disable_co_trackers();
1.727 +
1.728 + // This prints the global/local fingers. It is used for debugging.
1.729 + NOT_PRODUCT(void print_finger();)
1.730 +
1.731 + void print_summary_info();
1.732 +
1.733 + // The following indicate whether a given verbose level has been
1.734 + // set. Notice that anything above stats is conditional to
1.735 + // _MARKING_VERBOSE_ having been set to 1
1.736 + bool verbose_stats()
1.737 + { return _verbose_level >= stats_verbose; }
1.738 + bool verbose_low()
1.739 + { return _MARKING_VERBOSE_ && _verbose_level >= low_verbose; }
1.740 + bool verbose_medium()
1.741 + { return _MARKING_VERBOSE_ && _verbose_level >= medium_verbose; }
1.742 + bool verbose_high()
1.743 + { return _MARKING_VERBOSE_ && _verbose_level >= high_verbose; }
1.744 +};
1.745 +
1.746 +// A class representing a marking task.
1.747 +class CMTask : public TerminatorTerminator {
1.748 +private:
1.749 + enum PrivateConstants {
1.750 + // the regular clock call is called once the scanned words reaches
1.751 + // this limit
1.752 + words_scanned_period = 12*1024,
1.753 + // the regular clock call is called once the number of visited
1.754 + // references reaches this limit
1.755 + refs_reached_period = 384,
1.756 + // initial value for the hash seed, used in the work stealing code
1.757 + init_hash_seed = 17,
1.758 + // how many entries will be transferred between global stack and
1.759 + // local queues
1.760 + global_stack_transfer_size = 16
1.761 + };
1.762 +
1.763 + int _task_id;
1.764 + G1CollectedHeap* _g1h;
1.765 + ConcurrentMark* _cm;
1.766 + CMBitMap* _nextMarkBitMap;
1.767 + // the task queue of this task
1.768 + CMTaskQueue* _task_queue;
1.769 + // the task queue set---needed for stealing
1.770 + CMTaskQueueSet* _task_queues;
1.771 + // indicates whether the task has been claimed---this is only for
1.772 + // debugging purposes
1.773 + bool _claimed;
1.774 +
1.775 + // number of calls to this task
1.776 + int _calls;
1.777 +
1.778 + // concurrent overhead over a single CPU for this task
1.779 + COTracker _co_tracker;
1.780 +
1.781 + // when the virtual timer reaches this time, the marking step should
1.782 + // exit
1.783 + double _time_target_ms;
1.784 + // the start time of the current marking step
1.785 + double _start_time_ms;
1.786 +
1.787 + // the oop closure used for iterations over oops
1.788 + OopClosure* _oop_closure;
1.789 +
1.790 + // the region this task is scanning, NULL if we're not scanning any
1.791 + HeapRegion* _curr_region;
1.792 + // the local finger of this task, NULL if we're not scanning a region
1.793 + HeapWord* _finger;
1.794 + // limit of the region this task is scanning, NULL if we're not scanning one
1.795 + HeapWord* _region_limit;
1.796 +
1.797 + // This is used only when we scan regions popped from the region
1.798 + // stack. It records what the last object on such a region we
1.799 + // scanned was. It is used to ensure that, if we abort region
1.800 + // iteration, we do not rescan the first part of the region. This
1.801 + // should be NULL when we're not scanning a region from the region
1.802 + // stack.
1.803 + HeapWord* _region_finger;
1.804 +
1.805 + // the number of words this task has scanned
1.806 + size_t _words_scanned;
1.807 + // When _words_scanned reaches this limit, the regular clock is
1.808 + // called. Notice that this might be decreased under certain
1.809 + // circumstances (i.e. when we believe that we did an expensive
1.810 + // operation).
1.811 + size_t _words_scanned_limit;
1.812 + // the initial value of _words_scanned_limit (i.e. what it was
1.813 + // before it was decreased).
1.814 + size_t _real_words_scanned_limit;
1.815 +
1.816 + // the number of references this task has visited
1.817 + size_t _refs_reached;
1.818 + // When _refs_reached reaches this limit, the regular clock is
1.819 + // called. Notice this this might be decreased under certain
1.820 + // circumstances (i.e. when we believe that we did an expensive
1.821 + // operation).
1.822 + size_t _refs_reached_limit;
1.823 + // the initial value of _refs_reached_limit (i.e. what it was before
1.824 + // it was decreased).
1.825 + size_t _real_refs_reached_limit;
1.826 +
1.827 + // used by the work stealing stuff
1.828 + int _hash_seed;
1.829 + // if this is true, then the task has aborted for some reason
1.830 + bool _has_aborted;
1.831 + // set when the task aborts because it has met its time quota
1.832 + bool _has_aborted_timed_out;
1.833 + // true when we're draining SATB buffers; this avoids the task
1.834 + // aborting due to SATB buffers being available (as we're already
1.835 + // dealing with them)
1.836 + bool _draining_satb_buffers;
1.837 +
1.838 + // number sequence of past step times
1.839 + NumberSeq _step_times_ms;
1.840 + // elapsed time of this task
1.841 + double _elapsed_time_ms;
1.842 + // termination time of this task
1.843 + double _termination_time_ms;
1.844 + // when this task got into the termination protocol
1.845 + double _termination_start_time_ms;
1.846 +
1.847 + // true when the task is during a concurrent phase, false when it is
1.848 + // in the remark phase (so, in the latter case, we do not have to
1.849 + // check all the things that we have to check during the concurrent
1.850 + // phase, i.e. SATB buffer availability...)
1.851 + bool _concurrent;
1.852 +
1.853 + TruncatedSeq _marking_step_diffs_ms;
1.854 +
1.855 + // LOTS of statistics related with this task
1.856 +#if _MARKING_STATS_
1.857 + NumberSeq _all_clock_intervals_ms;
1.858 + double _interval_start_time_ms;
1.859 +
1.860 + int _aborted;
1.861 + int _aborted_overflow;
1.862 + int _aborted_cm_aborted;
1.863 + int _aborted_yield;
1.864 + int _aborted_timed_out;
1.865 + int _aborted_satb;
1.866 + int _aborted_termination;
1.867 +
1.868 + int _steal_attempts;
1.869 + int _steals;
1.870 +
1.871 + int _clock_due_to_marking;
1.872 + int _clock_due_to_scanning;
1.873 +
1.874 + int _local_pushes;
1.875 + int _local_pops;
1.876 + int _local_max_size;
1.877 + int _objs_scanned;
1.878 +
1.879 + int _global_pushes;
1.880 + int _global_pops;
1.881 + int _global_max_size;
1.882 +
1.883 + int _global_transfers_to;
1.884 + int _global_transfers_from;
1.885 +
1.886 + int _region_stack_pops;
1.887 +
1.888 + int _regions_claimed;
1.889 + int _objs_found_on_bitmap;
1.890 +
1.891 + int _satb_buffers_processed;
1.892 +#endif // _MARKING_STATS_
1.893 +
1.894 + // it updates the local fields after this task has claimed
1.895 + // a new region to scan
1.896 + void setup_for_region(HeapRegion* hr);
1.897 + // it brings up-to-date the limit of the region
1.898 + void update_region_limit();
1.899 + // it resets the local fields after a task has finished scanning a
1.900 + // region
1.901 + void giveup_current_region();
1.902 +
1.903 + // called when either the words scanned or the refs visited limit
1.904 + // has been reached
1.905 + void reached_limit();
1.906 + // recalculates the words scanned and refs visited limits
1.907 + void recalculate_limits();
1.908 + // decreases the words scanned and refs visited limits when we reach
1.909 + // an expensive operation
1.910 + void decrease_limits();
1.911 + // it checks whether the words scanned or refs visited reached their
1.912 + // respective limit and calls reached_limit() if they have
1.913 + void check_limits() {
1.914 + if (_words_scanned >= _words_scanned_limit ||
1.915 + _refs_reached >= _refs_reached_limit)
1.916 + reached_limit();
1.917 + }
1.918 + // this is supposed to be called regularly during a marking step as
1.919 + // it checks a bunch of conditions that might cause the marking step
1.920 + // to abort
1.921 + void regular_clock_call();
1.922 + bool concurrent() { return _concurrent; }
1.923 +
1.924 +public:
1.925 + // It resets the task; it should be called right at the beginning of
1.926 + // a marking phase.
1.927 + void reset(CMBitMap* _nextMarkBitMap);
1.928 + // it clears all the fields that correspond to a claimed region.
1.929 + void clear_region_fields();
1.930 +
1.931 + void set_concurrent(bool concurrent) { _concurrent = concurrent; }
1.932 +
1.933 + void enable_co_tracker() {
1.934 + guarantee( !_co_tracker.enabled(), "invariant" );
1.935 + _co_tracker.enable();
1.936 + }
1.937 + void disable_co_tracker() {
1.938 + guarantee( _co_tracker.enabled(), "invariant" );
1.939 + _co_tracker.disable();
1.940 + }
1.941 + bool co_tracker_enabled() {
1.942 + return _co_tracker.enabled();
1.943 + }
1.944 + void reset_co_tracker(double starting_conc_overhead = 0.0) {
1.945 + _co_tracker.reset(starting_conc_overhead);
1.946 + }
1.947 + void start_co_tracker() {
1.948 + _co_tracker.start();
1.949 + }
1.950 + void update_co_tracker(bool force_end = false) {
1.951 + _co_tracker.update(force_end);
1.952 + }
1.953 +
1.954 + // The main method of this class which performs a marking step
1.955 + // trying not to exceed the given duration. However, it might exit
1.956 + // prematurely, according to some conditions (i.e. SATB buffers are
1.957 + // available for processing).
1.958 + void do_marking_step(double target_ms);
1.959 +
1.960 + // These two calls start and stop the timer
1.961 + void record_start_time() {
1.962 + _elapsed_time_ms = os::elapsedTime() * 1000.0;
1.963 + }
1.964 + void record_end_time() {
1.965 + _elapsed_time_ms = os::elapsedTime() * 1000.0 - _elapsed_time_ms;
1.966 + }
1.967 +
1.968 + // returns the task ID
1.969 + int task_id() { return _task_id; }
1.970 +
1.971 + // From TerminatorTerminator. It determines whether this task should
1.972 + // exit the termination protocol after it's entered it.
1.973 + virtual bool should_exit_termination();
1.974 +
1.975 + HeapWord* finger() { return _finger; }
1.976 +
1.977 + bool has_aborted() { return _has_aborted; }
1.978 + void set_has_aborted() { _has_aborted = true; }
1.979 + void clear_has_aborted() { _has_aborted = false; }
1.980 + bool claimed() { return _claimed; }
1.981 +
1.982 + void set_oop_closure(OopClosure* oop_closure) {
1.983 + _oop_closure = oop_closure;
1.984 + }
1.985 +
1.986 + // It grays the object by marking it and, if necessary, pushing it
1.987 + // on the local queue
1.988 + void deal_with_reference(oop obj);
1.989 +
1.990 + // It scans an object and visits its children.
1.991 + void scan_object(oop obj) {
1.992 + tmp_guarantee_CM( _nextMarkBitMap->isMarked((HeapWord*) obj),
1.993 + "invariant" );
1.994 +
1.995 + if (_cm->verbose_high())
1.996 + gclog_or_tty->print_cr("[%d] we're scanning object "PTR_FORMAT,
1.997 + _task_id, (void*) obj);
1.998 +
1.999 + size_t obj_size = obj->size();
1.1000 + _words_scanned += obj_size;
1.1001 +
1.1002 + obj->oop_iterate(_oop_closure);
1.1003 + statsOnly( ++_objs_scanned );
1.1004 + check_limits();
1.1005 + }
1.1006 +
1.1007 + // It pushes an object on the local queue.
1.1008 + void push(oop obj);
1.1009 +
1.1010 + // These two move entries to/from the global stack.
1.1011 + void move_entries_to_global_stack();
1.1012 + void get_entries_from_global_stack();
1.1013 +
1.1014 + // It pops and scans objects from the local queue. If partially is
1.1015 + // true, then it stops when the queue size is of a given limit. If
1.1016 + // partially is false, then it stops when the queue is empty.
1.1017 + void drain_local_queue(bool partially);
1.1018 + // It moves entries from the global stack to the local queue and
1.1019 + // drains the local queue. If partially is true, then it stops when
1.1020 + // both the global stack and the local queue reach a given size. If
1.1021 + // partially if false, it tries to empty them totally.
1.1022 + void drain_global_stack(bool partially);
1.1023 + // It keeps picking SATB buffers and processing them until no SATB
1.1024 + // buffers are available.
1.1025 + void drain_satb_buffers();
1.1026 + // It keeps popping regions from the region stack and processing
1.1027 + // them until the region stack is empty.
1.1028 + void drain_region_stack(BitMapClosure* closure);
1.1029 +
1.1030 + // moves the local finger to a new location
1.1031 + inline void move_finger_to(HeapWord* new_finger) {
1.1032 + tmp_guarantee_CM( new_finger >= _finger && new_finger < _region_limit,
1.1033 + "invariant" );
1.1034 + _finger = new_finger;
1.1035 + }
1.1036 +
1.1037 + // moves the region finger to a new location
1.1038 + inline void move_region_finger_to(HeapWord* new_finger) {
1.1039 + tmp_guarantee_CM( new_finger < _cm->finger(), "invariant" );
1.1040 + _region_finger = new_finger;
1.1041 + }
1.1042 +
1.1043 + CMTask(int task_num, ConcurrentMark *cm,
1.1044 + CMTaskQueue* task_queue, CMTaskQueueSet* task_queues);
1.1045 +
1.1046 + // it prints statistics associated with this task
1.1047 + void print_stats();
1.1048 +
1.1049 +#if _MARKING_STATS_
1.1050 + void increase_objs_found_on_bitmap() { ++_objs_found_on_bitmap; }
1.1051 +#endif // _MARKING_STATS_
1.1052 +};
