• file
• revisions
• annotate
• diff
• comparison
• raw

src/share/vm/gc_implementation/parNew/parCardTableModRefBS.cpp@c69b1043dfb1 (annotated)

src/share/vm/gc_implementation/parNew/parCardTableModRefBS.cpp

Thu, 14 Apr 2011 12:10:15 -0700

author

ysr

date

Thu, 14 Apr 2011 12:10:15 -0700

changeset 2788

c69b1043dfb1

parent 2314

f95d63e2154a

child 2819

c48ad6ab8bdf

permissions

-rw-r--r--

7036482: clear argument is redundant and unused in cardtable methods
Summary: Removed the unused clear argument to various cardtbale methods and unused mod_oop_in_space_iterate method. Unrelated to synopsis, added a pair of clarifying parens in AllocationStats constructor.
Reviewed-by: brutisso, jcoomes

duke@435	1	/*
ysr@2788	2	* Copyright (c) 2007, 2011 Oracle and/or its affiliates. All rights reserved.
duke@435	3	* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435	4	*
duke@435	5	* This code is free software; you can redistribute it and/or modify it
duke@435	6	* under the terms of the GNU General Public License version 2 only, as
duke@435	7	* published by the Free Software Foundation.
duke@435	8	*
duke@435	9	* This code is distributed in the hope that it will be useful, but WITHOUT
duke@435	10	* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435	11	* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
duke@435	12	* version 2 for more details (a copy is included in the LICENSE file that
duke@435	13	* accompanied this code).
duke@435	14	*
duke@435	15	* You should have received a copy of the GNU General Public License version
duke@435	16	* 2 along with this work; if not, write to the Free Software Foundation,
duke@435	17	* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435	18	*
trims@1907	19	* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
trims@1907	20	* or visit www.oracle.com if you need additional information or have any
trims@1907	21	* questions.
duke@435	22	*
duke@435	23	*/
duke@435	24
stefank@2314	25	#include "precompiled.hpp"
stefank@2314	26	#include "memory/allocation.inline.hpp"
stefank@2314	27	#include "memory/cardTableModRefBS.hpp"
stefank@2314	28	#include "memory/cardTableRS.hpp"
stefank@2314	29	#include "memory/sharedHeap.hpp"
stefank@2314	30	#include "memory/space.inline.hpp"
stefank@2314	31	#include "memory/universe.hpp"
stefank@2314	32	#include "runtime/java.hpp"
stefank@2314	33	#include "runtime/mutexLocker.hpp"
stefank@2314	34	#include "runtime/virtualspace.hpp"
duke@435	35
duke@435	36	void CardTableModRefBS::par_non_clean_card_iterate_work(Space* sp, MemRegion mr,
duke@435	37	DirtyCardToOopClosure* dcto_cl,
duke@435	38	MemRegionClosure* cl,
duke@435	39	int n_threads) {
duke@435	40	if (n_threads > 0) {
ysr@1280	41	assert((n_threads == 1 && ParallelGCThreads == 0) ||
ysr@1280	42	n_threads <= (int)ParallelGCThreads,
ysr@1280	43	"# worker threads != # requested!");
ysr@1280	44	// Make sure the LNC array is valid for the space.
duke@435	45	jbyte** lowest_non_clean;
duke@435	46	uintptr_t lowest_non_clean_base_chunk_index;
duke@435	47	size_t lowest_non_clean_chunk_size;
duke@435	48	get_LNC_array_for_space(sp, lowest_non_clean,
duke@435	49	lowest_non_clean_base_chunk_index,
duke@435	50	lowest_non_clean_chunk_size);
duke@435	51
duke@435	52	int n_strides = n_threads * StridesPerThread;
duke@435	53	SequentialSubTasksDone* pst = sp->par_seq_tasks();
jmasa@2188	54	pst->set_n_threads(n_threads);
duke@435	55	pst->set_n_tasks(n_strides);
duke@435	56
duke@435	57	int stride = 0;
duke@435	58	while (!pst->is_task_claimed(/* reference */ stride)) {
ysr@2788	59	process_stride(sp, mr, stride, n_strides, dcto_cl, cl,
duke@435	60	lowest_non_clean,
duke@435	61	lowest_non_clean_base_chunk_index,
duke@435	62	lowest_non_clean_chunk_size);
duke@435	63	}
duke@435	64	if (pst->all_tasks_completed()) {
duke@435	65	// Clear lowest_non_clean array for next time.
duke@435	66	intptr_t first_chunk_index = addr_to_chunk_index(mr.start());
duke@435	67	uintptr_t last_chunk_index = addr_to_chunk_index(mr.last());
duke@435	68	for (uintptr_t ch = first_chunk_index; ch <= last_chunk_index; ch++) {
duke@435	69	intptr_t ind = ch - lowest_non_clean_base_chunk_index;
duke@435	70	assert(0 <= ind && ind < (intptr_t)lowest_non_clean_chunk_size,
duke@435	71	"Bounds error");
duke@435	72	lowest_non_clean[ind] = NULL;
duke@435	73	}
duke@435	74	}
duke@435	75	}
duke@435	76	}
duke@435	77
duke@435	78	void
duke@435	79	CardTableModRefBS::
duke@435	80	process_stride(Space* sp,
duke@435	81	MemRegion used,
duke@435	82	jint stride, int n_strides,
duke@435	83	DirtyCardToOopClosure* dcto_cl,
duke@435	84	MemRegionClosure* cl,
duke@435	85	jbyte** lowest_non_clean,
duke@435	86	uintptr_t lowest_non_clean_base_chunk_index,
duke@435	87	size_t lowest_non_clean_chunk_size) {
duke@435	88	// We don't have to go downwards here; it wouldn't help anyway,
duke@435	89	// because of parallelism.
duke@435	90
duke@435	91	// Find the first card address of the first chunk in the stride that is
duke@435	92	// at least "bottom" of the used region.
duke@435	93	jbyte* start_card = byte_for(used.start());
duke@435	94	jbyte* end_card = byte_after(used.last());
duke@435	95	uintptr_t start_chunk = addr_to_chunk_index(used.start());
duke@435	96	uintptr_t start_chunk_stride_num = start_chunk % n_strides;
duke@435	97	jbyte* chunk_card_start;
duke@435	98
duke@435	99	if ((uintptr_t)stride >= start_chunk_stride_num) {
duke@435	100	chunk_card_start = (jbyte*)(start_card +
duke@435	101	(stride - start_chunk_stride_num) *
duke@435	102	CardsPerStrideChunk);
duke@435	103	} else {
duke@435	104	// Go ahead to the next chunk group boundary, then to the requested stride.
duke@435	105	chunk_card_start = (jbyte*)(start_card +
duke@435	106	(n_strides - start_chunk_stride_num + stride) *
duke@435	107	CardsPerStrideChunk);
duke@435	108	}
duke@435	109
duke@435	110	while (chunk_card_start < end_card) {
duke@435	111	// We don't have to go downwards here; it wouldn't help anyway,
duke@435	112	// because of parallelism. (We take care with "min_done"; see below.)
duke@435	113	// Invariant: chunk_mr should be fully contained within the "used" region.
duke@435	114	jbyte* chunk_card_end = chunk_card_start + CardsPerStrideChunk;
duke@435	115	MemRegion chunk_mr = MemRegion(addr_for(chunk_card_start),
duke@435	116	chunk_card_end >= end_card ?
duke@435	117	used.end() : addr_for(chunk_card_end));
duke@435	118	assert(chunk_mr.word_size() > 0, "[chunk_card_start > used_end)");
duke@435	119	assert(used.contains(chunk_mr), "chunk_mr should be subset of used");
duke@435	120
duke@435	121	// Process the chunk.
duke@435	122	process_chunk_boundaries(sp,
duke@435	123	dcto_cl,
duke@435	124	chunk_mr,
duke@435	125	used,
duke@435	126	lowest_non_clean,
duke@435	127	lowest_non_clean_base_chunk_index,
duke@435	128	lowest_non_clean_chunk_size);
duke@435	129
ysr@2788	130	non_clean_card_iterate_work(chunk_mr, cl);
duke@435	131
duke@435	132	// Find the next chunk of the stride.
duke@435	133	chunk_card_start += CardsPerStrideChunk * n_strides;
duke@435	134	}
duke@435	135	}
duke@435	136
duke@435	137	void
duke@435	138	CardTableModRefBS::
duke@435	139	process_chunk_boundaries(Space* sp,
duke@435	140	DirtyCardToOopClosure* dcto_cl,
duke@435	141	MemRegion chunk_mr,
duke@435	142	MemRegion used,
duke@435	143	jbyte** lowest_non_clean,
duke@435	144	uintptr_t lowest_non_clean_base_chunk_index,
duke@435	145	size_t lowest_non_clean_chunk_size)
duke@435	146	{
duke@435	147	// We must worry about the chunk boundaries.
duke@435	148
duke@435	149	// First, set our max_to_do:
duke@435	150	HeapWord* max_to_do = NULL;
duke@435	151	uintptr_t cur_chunk_index = addr_to_chunk_index(chunk_mr.start());
duke@435	152	cur_chunk_index = cur_chunk_index - lowest_non_clean_base_chunk_index;
duke@435	153
duke@435	154	if (chunk_mr.end() < used.end()) {
duke@435	155	// This is not the last chunk in the used region. What is the last
duke@435	156	// object?
duke@435	157	HeapWord* last_block = sp->block_start(chunk_mr.end());
duke@435	158	assert(last_block <= chunk_mr.end(), "In case this property changes.");
duke@435	159	if (last_block == chunk_mr.end()
duke@435	160	|| !sp->block_is_obj(last_block)) {
duke@435	161	max_to_do = chunk_mr.end();
duke@435	162
duke@435	163	} else {
duke@435	164	// It is an object and starts before the end of the current chunk.
duke@435	165	// last_obj_card is the card corresponding to the start of the last object
duke@435	166	// in the chunk. Note that the last object may not start in
duke@435	167	// the chunk.
duke@435	168	jbyte* last_obj_card = byte_for(last_block);
duke@435	169	if (!card_may_have_been_dirty(*last_obj_card)) {
duke@435	170	// The card containing the head is not dirty. Any marks in
duke@435	171	// subsequent cards still in this chunk must have been made
duke@435	172	// precisely; we can cap processing at the end.
duke@435	173	max_to_do = chunk_mr.end();
duke@435	174	} else {
duke@435	175	// The last object must be considered dirty, and extends onto the
duke@435	176	// following chunk. Look for a dirty card in that chunk that will
duke@435	177	// bound our processing.
duke@435	178	jbyte* limit_card = NULL;
duke@435	179	size_t last_block_size = sp->block_size(last_block);
duke@435	180	jbyte* last_card_of_last_obj =
duke@435	181	byte_for(last_block + last_block_size - 1);
duke@435	182	jbyte* first_card_of_next_chunk = byte_for(chunk_mr.end());
duke@435	183	// This search potentially goes a long distance looking
duke@435	184	// for the next card that will be scanned. For example,
duke@435	185	// an object that is an array of primitives will not
duke@435	186	// have any cards covering regions interior to the array
duke@435	187	// that will need to be scanned. The scan can be terminated
duke@435	188	// at the last card of the next chunk. That would leave
duke@435	189	// limit_card as NULL and would result in "max_to_do"
duke@435	190	// being set with the LNC value or with the end
duke@435	191	// of the last block.
duke@435	192	jbyte* last_card_of_next_chunk = first_card_of_next_chunk +
duke@435	193	CardsPerStrideChunk;
duke@435	194	assert(byte_for(chunk_mr.end()) - byte_for(chunk_mr.start())
duke@435	195	== CardsPerStrideChunk, "last card of next chunk may be wrong");
duke@435	196	jbyte* last_card_to_check = (jbyte*) MIN2(last_card_of_last_obj,
duke@435	197	last_card_of_next_chunk);
duke@435	198	for (jbyte* cur = first_card_of_next_chunk;
duke@435	199	cur <= last_card_to_check; cur++) {
duke@435	200	if (card_will_be_scanned(*cur)) {
duke@435	201	limit_card = cur; break;
duke@435	202	}
duke@435	203	}
duke@435	204	assert(0 <= cur_chunk_index+1 &&
duke@435	205	cur_chunk_index+1 < lowest_non_clean_chunk_size,
duke@435	206	"Bounds error.");
duke@435	207	// LNC for the next chunk
duke@435	208	jbyte* lnc_card = lowest_non_clean[cur_chunk_index+1];
duke@435	209	if (limit_card == NULL) {
duke@435	210	limit_card = lnc_card;
duke@435	211	}
duke@435	212	if (limit_card != NULL) {
duke@435	213	if (lnc_card != NULL) {
duke@435	214	limit_card = (jbyte*)MIN2((intptr_t)limit_card,
duke@435	215	(intptr_t)lnc_card);
duke@435	216	}
duke@435	217	max_to_do = addr_for(limit_card);
duke@435	218	} else {
duke@435	219	max_to_do = last_block + last_block_size;
duke@435	220	}
duke@435	221	}
duke@435	222	}
duke@435	223	assert(max_to_do != NULL, "OOPS!");
duke@435	224	} else {
duke@435	225	max_to_do = used.end();
duke@435	226	}
duke@435	227	// Now we can set the closure we're using so it doesn't to beyond
duke@435	228	// max_to_do.
duke@435	229	dcto_cl->set_min_done(max_to_do);
duke@435	230	#ifndef PRODUCT
duke@435	231	dcto_cl->set_last_bottom(max_to_do);
duke@435	232	#endif
duke@435	233
duke@435	234	// Now we set *our" lowest_non_clean entry.
duke@435	235	// Find the object that spans our boundary, if one exists.
duke@435	236	// Nothing to do on the first chunk.
duke@435	237	if (chunk_mr.start() > used.start()) {
duke@435	238	// first_block is the block possibly spanning the chunk start
duke@435	239	HeapWord* first_block = sp->block_start(chunk_mr.start());
duke@435	240	// Does the block span the start of the chunk and is it
duke@435	241	// an object?
duke@435	242	if (first_block < chunk_mr.start() &&
duke@435	243	sp->block_is_obj(first_block)) {
duke@435	244	jbyte* first_dirty_card = NULL;
duke@435	245	jbyte* last_card_of_first_obj =
duke@435	246	byte_for(first_block + sp->block_size(first_block) - 1);
duke@435	247	jbyte* first_card_of_cur_chunk = byte_for(chunk_mr.start());
duke@435	248	jbyte* last_card_of_cur_chunk = byte_for(chunk_mr.last());
duke@435	249	jbyte* last_card_to_check =
duke@435	250	(jbyte*) MIN2((intptr_t) last_card_of_cur_chunk,
duke@435	251	(intptr_t) last_card_of_first_obj);
duke@435	252	for (jbyte* cur = first_card_of_cur_chunk;
duke@435	253	cur <= last_card_to_check; cur++) {
duke@435	254	if (card_will_be_scanned(*cur)) {
duke@435	255	first_dirty_card = cur; break;
duke@435	256	}
duke@435	257	}
duke@435	258	if (first_dirty_card != NULL) {
duke@435	259	assert(0 <= cur_chunk_index &&
duke@435	260	cur_chunk_index < lowest_non_clean_chunk_size,
duke@435	261	"Bounds error.");
duke@435	262	lowest_non_clean[cur_chunk_index] = first_dirty_card;
duke@435	263	}
duke@435	264	}
duke@435	265	}
duke@435	266	}
duke@435	267
duke@435	268	void
duke@435	269	CardTableModRefBS::
duke@435	270	get_LNC_array_for_space(Space* sp,
duke@435	271	jbyte**& lowest_non_clean,
duke@435	272	uintptr_t& lowest_non_clean_base_chunk_index,
duke@435	273	size_t& lowest_non_clean_chunk_size) {
duke@435	274
duke@435	275	int i = find_covering_region_containing(sp->bottom());
duke@435	276	MemRegion covered = _covered[i];
duke@435	277	size_t n_chunks = chunks_to_cover(covered);
duke@435	278
duke@435	279	// Only the first thread to obtain the lock will resize the
duke@435	280	// LNC array for the covered region. Any later expansion can't affect
duke@435	281	// the used_at_save_marks region.
duke@435	282	// (I observed a bug in which the first thread to execute this would
duke@435	283	// resize, and then it would cause "expand_and_allocates" that would
duke@435	284	// Increase the number of chunks in the covered region. Then a second
duke@435	285	// thread would come and execute this, see that the size didn't match,
duke@435	286	// and free and allocate again. So the first thread would be using a
duke@435	287	// freed "_lowest_non_clean" array.)
duke@435	288
duke@435	289	// Do a dirty read here. If we pass the conditional then take the rare
duke@435	290	// event lock and do the read again in case some other thread had already
duke@435	291	// succeeded and done the resize.
duke@435	292	int cur_collection = Universe::heap()->total_collections();
duke@435	293	if (_last_LNC_resizing_collection[i] != cur_collection) {
duke@435	294	MutexLocker x(ParGCRareEvent_lock);
duke@435	295	if (_last_LNC_resizing_collection[i] != cur_collection) {
duke@435	296	if (_lowest_non_clean[i] == NULL ||
duke@435	297	n_chunks != _lowest_non_clean_chunk_size[i]) {
duke@435	298
duke@435	299	// Should we delete the old?
duke@435	300	if (_lowest_non_clean[i] != NULL) {
duke@435	301	assert(n_chunks != _lowest_non_clean_chunk_size[i],
duke@435	302	"logical consequence");
duke@435	303	FREE_C_HEAP_ARRAY(CardPtr, _lowest_non_clean[i]);
duke@435	304	_lowest_non_clean[i] = NULL;
duke@435	305	}
duke@435	306	// Now allocate a new one if necessary.
duke@435	307	if (_lowest_non_clean[i] == NULL) {
duke@435	308	_lowest_non_clean[i] = NEW_C_HEAP_ARRAY(CardPtr, n_chunks);
duke@435	309	_lowest_non_clean_chunk_size[i] = n_chunks;
duke@435	310	_lowest_non_clean_base_chunk_index[i] = addr_to_chunk_index(covered.start());
duke@435	311	for (int j = 0; j < (int)n_chunks; j++)
duke@435	312	_lowest_non_clean[i][j] = NULL;
duke@435	313	}
duke@435	314	}
duke@435	315	_last_LNC_resizing_collection[i] = cur_collection;
duke@435	316	}
duke@435	317	}
duke@435	318	// In any case, now do the initialization.
duke@435	319	lowest_non_clean = _lowest_non_clean[i];
duke@435	320	lowest_non_clean_base_chunk_index = _lowest_non_clean_base_chunk_index[i];
duke@435	321	lowest_non_clean_chunk_size = _lowest_non_clean_chunk_size[i];
duke@435	322	}