Tue, 12 Dec 2017 10:30:27 +0800
#6345 sync is controled by UseSyncLevel instead of Use3A2000
Reviewed-by: fujie
aoqi@0 | 1 | /* |
aoqi@0 | 2 | * Copyright (c) 2001, 2014, Oracle and/or its affiliates. All rights reserved. |
aoqi@0 | 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
aoqi@0 | 4 | * |
aoqi@0 | 5 | * This code is free software; you can redistribute it and/or modify it |
aoqi@0 | 6 | * under the terms of the GNU General Public License version 2 only, as |
aoqi@0 | 7 | * published by the Free Software Foundation. |
aoqi@0 | 8 | * |
aoqi@0 | 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
aoqi@0 | 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
aoqi@0 | 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
aoqi@0 | 12 | * version 2 for more details (a copy is included in the LICENSE file that |
aoqi@0 | 13 | * accompanied this code). |
aoqi@0 | 14 | * |
aoqi@0 | 15 | * You should have received a copy of the GNU General Public License version |
aoqi@0 | 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
aoqi@0 | 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
aoqi@0 | 18 | * |
aoqi@0 | 19 | * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
aoqi@0 | 20 | * or visit www.oracle.com if you need additional information or have any |
aoqi@0 | 21 | * questions. |
aoqi@0 | 22 | * |
aoqi@0 | 23 | */ |
aoqi@0 | 24 | |
aoqi@0 | 25 | #include "precompiled.hpp" |
aoqi@0 | 26 | #include "memory/allocation.inline.hpp" |
aoqi@0 | 27 | #include "memory/cardTableRS.hpp" |
aoqi@0 | 28 | #include "memory/genCollectedHeap.hpp" |
aoqi@0 | 29 | #include "memory/generation.hpp" |
aoqi@0 | 30 | #include "memory/space.hpp" |
aoqi@0 | 31 | #include "oops/oop.inline.hpp" |
aoqi@0 | 32 | #include "runtime/java.hpp" |
aoqi@0 | 33 | #include "runtime/os.hpp" |
aoqi@0 | 34 | #include "utilities/macros.hpp" |
aoqi@0 | 35 | #if INCLUDE_ALL_GCS |
aoqi@0 | 36 | #include "gc_implementation/g1/concurrentMark.hpp" |
aoqi@0 | 37 | #include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp" |
aoqi@0 | 38 | #endif // INCLUDE_ALL_GCS |
aoqi@0 | 39 | |
aoqi@0 | 40 | CardTableRS::CardTableRS(MemRegion whole_heap, |
aoqi@0 | 41 | int max_covered_regions) : |
aoqi@0 | 42 | GenRemSet(), |
aoqi@0 | 43 | _cur_youngergen_card_val(youngergenP1_card), |
aoqi@0 | 44 | _regions_to_iterate(max_covered_regions - 1) |
aoqi@0 | 45 | { |
aoqi@0 | 46 | #if INCLUDE_ALL_GCS |
aoqi@0 | 47 | if (UseG1GC) { |
aoqi@0 | 48 | _ct_bs = new G1SATBCardTableLoggingModRefBS(whole_heap, |
aoqi@0 | 49 | max_covered_regions); |
aoqi@0 | 50 | } else { |
aoqi@0 | 51 | _ct_bs = new CardTableModRefBSForCTRS(whole_heap, max_covered_regions); |
aoqi@0 | 52 | } |
aoqi@0 | 53 | #else |
aoqi@0 | 54 | _ct_bs = new CardTableModRefBSForCTRS(whole_heap, max_covered_regions); |
aoqi@0 | 55 | #endif |
tschatzl@7051 | 56 | _ct_bs->initialize(); |
aoqi@0 | 57 | set_bs(_ct_bs); |
aoqi@0 | 58 | _last_cur_val_in_gen = NEW_C_HEAP_ARRAY3(jbyte, GenCollectedHeap::max_gens + 1, |
zgu@7074 | 59 | mtGC, CURRENT_PC, AllocFailStrategy::RETURN_NULL); |
aoqi@0 | 60 | if (_last_cur_val_in_gen == NULL) { |
aoqi@0 | 61 | vm_exit_during_initialization("Could not create last_cur_val_in_gen array."); |
aoqi@0 | 62 | } |
aoqi@0 | 63 | for (int i = 0; i < GenCollectedHeap::max_gens + 1; i++) { |
aoqi@0 | 64 | _last_cur_val_in_gen[i] = clean_card_val(); |
aoqi@0 | 65 | } |
aoqi@0 | 66 | _ct_bs->set_CTRS(this); |
aoqi@0 | 67 | } |
aoqi@0 | 68 | |
aoqi@0 | 69 | CardTableRS::~CardTableRS() { |
aoqi@0 | 70 | if (_ct_bs) { |
aoqi@0 | 71 | delete _ct_bs; |
aoqi@0 | 72 | _ct_bs = NULL; |
aoqi@0 | 73 | } |
aoqi@0 | 74 | if (_last_cur_val_in_gen) { |
aoqi@0 | 75 | FREE_C_HEAP_ARRAY(jbyte, _last_cur_val_in_gen, mtInternal); |
aoqi@0 | 76 | } |
aoqi@0 | 77 | } |
aoqi@0 | 78 | |
aoqi@0 | 79 | void CardTableRS::resize_covered_region(MemRegion new_region) { |
aoqi@0 | 80 | _ct_bs->resize_covered_region(new_region); |
aoqi@0 | 81 | } |
aoqi@0 | 82 | |
aoqi@0 | 83 | jbyte CardTableRS::find_unused_youngergenP_card_value() { |
aoqi@0 | 84 | for (jbyte v = youngergenP1_card; |
aoqi@0 | 85 | v < cur_youngergen_and_prev_nonclean_card; |
aoqi@0 | 86 | v++) { |
aoqi@0 | 87 | bool seen = false; |
aoqi@0 | 88 | for (int g = 0; g < _regions_to_iterate; g++) { |
aoqi@0 | 89 | if (_last_cur_val_in_gen[g] == v) { |
aoqi@0 | 90 | seen = true; |
aoqi@0 | 91 | break; |
aoqi@0 | 92 | } |
aoqi@0 | 93 | } |
aoqi@0 | 94 | if (!seen) return v; |
aoqi@0 | 95 | } |
aoqi@0 | 96 | ShouldNotReachHere(); |
aoqi@0 | 97 | return 0; |
aoqi@0 | 98 | } |
aoqi@0 | 99 | |
aoqi@0 | 100 | void CardTableRS::prepare_for_younger_refs_iterate(bool parallel) { |
aoqi@0 | 101 | // Parallel or sequential, we must always set the prev to equal the |
aoqi@0 | 102 | // last one written. |
aoqi@0 | 103 | if (parallel) { |
aoqi@0 | 104 | // Find a parallel value to be used next. |
aoqi@0 | 105 | jbyte next_val = find_unused_youngergenP_card_value(); |
aoqi@0 | 106 | set_cur_youngergen_card_val(next_val); |
aoqi@0 | 107 | |
aoqi@0 | 108 | } else { |
aoqi@0 | 109 | // In an sequential traversal we will always write youngergen, so that |
aoqi@0 | 110 | // the inline barrier is correct. |
aoqi@0 | 111 | set_cur_youngergen_card_val(youngergen_card); |
aoqi@0 | 112 | } |
aoqi@0 | 113 | } |
aoqi@0 | 114 | |
aoqi@0 | 115 | void CardTableRS::younger_refs_iterate(Generation* g, |
aoqi@0 | 116 | OopsInGenClosure* blk) { |
aoqi@0 | 117 | _last_cur_val_in_gen[g->level()+1] = cur_youngergen_card_val(); |
aoqi@0 | 118 | g->younger_refs_iterate(blk); |
aoqi@0 | 119 | } |
aoqi@0 | 120 | |
aoqi@0 | 121 | inline bool ClearNoncleanCardWrapper::clear_card(jbyte* entry) { |
aoqi@0 | 122 | if (_is_par) { |
aoqi@0 | 123 | return clear_card_parallel(entry); |
aoqi@0 | 124 | } else { |
aoqi@0 | 125 | return clear_card_serial(entry); |
aoqi@0 | 126 | } |
aoqi@0 | 127 | } |
aoqi@0 | 128 | |
aoqi@0 | 129 | inline bool ClearNoncleanCardWrapper::clear_card_parallel(jbyte* entry) { |
aoqi@0 | 130 | while (true) { |
aoqi@0 | 131 | // In the parallel case, we may have to do this several times. |
aoqi@0 | 132 | jbyte entry_val = *entry; |
aoqi@0 | 133 | assert(entry_val != CardTableRS::clean_card_val(), |
aoqi@0 | 134 | "We shouldn't be looking at clean cards, and this should " |
aoqi@0 | 135 | "be the only place they get cleaned."); |
aoqi@0 | 136 | if (CardTableRS::card_is_dirty_wrt_gen_iter(entry_val) |
aoqi@0 | 137 | || _ct->is_prev_youngergen_card_val(entry_val)) { |
aoqi@0 | 138 | jbyte res = |
aoqi@0 | 139 | Atomic::cmpxchg(CardTableRS::clean_card_val(), entry, entry_val); |
aoqi@0 | 140 | if (res == entry_val) { |
aoqi@0 | 141 | break; |
aoqi@0 | 142 | } else { |
aoqi@0 | 143 | assert(res == CardTableRS::cur_youngergen_and_prev_nonclean_card, |
aoqi@0 | 144 | "The CAS above should only fail if another thread did " |
aoqi@0 | 145 | "a GC write barrier."); |
aoqi@0 | 146 | } |
aoqi@0 | 147 | } else if (entry_val == |
aoqi@0 | 148 | CardTableRS::cur_youngergen_and_prev_nonclean_card) { |
aoqi@0 | 149 | // Parallelism shouldn't matter in this case. Only the thread |
aoqi@0 | 150 | // assigned to scan the card should change this value. |
aoqi@0 | 151 | *entry = _ct->cur_youngergen_card_val(); |
aoqi@0 | 152 | break; |
aoqi@0 | 153 | } else { |
aoqi@0 | 154 | assert(entry_val == _ct->cur_youngergen_card_val(), |
aoqi@0 | 155 | "Should be the only possibility."); |
aoqi@0 | 156 | // In this case, the card was clean before, and become |
aoqi@0 | 157 | // cur_youngergen only because of processing of a promoted object. |
aoqi@0 | 158 | // We don't have to look at the card. |
aoqi@0 | 159 | return false; |
aoqi@0 | 160 | } |
aoqi@0 | 161 | } |
aoqi@0 | 162 | return true; |
aoqi@0 | 163 | } |
aoqi@0 | 164 | |
aoqi@0 | 165 | |
aoqi@0 | 166 | inline bool ClearNoncleanCardWrapper::clear_card_serial(jbyte* entry) { |
aoqi@0 | 167 | jbyte entry_val = *entry; |
aoqi@0 | 168 | assert(entry_val != CardTableRS::clean_card_val(), |
aoqi@0 | 169 | "We shouldn't be looking at clean cards, and this should " |
aoqi@0 | 170 | "be the only place they get cleaned."); |
aoqi@0 | 171 | assert(entry_val != CardTableRS::cur_youngergen_and_prev_nonclean_card, |
aoqi@0 | 172 | "This should be possible in the sequential case."); |
aoqi@0 | 173 | *entry = CardTableRS::clean_card_val(); |
aoqi@0 | 174 | return true; |
aoqi@0 | 175 | } |
aoqi@0 | 176 | |
aoqi@0 | 177 | ClearNoncleanCardWrapper::ClearNoncleanCardWrapper( |
aoqi@0 | 178 | DirtyCardToOopClosure* dirty_card_closure, CardTableRS* ct) : |
aoqi@0 | 179 | _dirty_card_closure(dirty_card_closure), _ct(ct) { |
aoqi@0 | 180 | // Cannot yet substitute active_workers for n_par_threads |
aoqi@0 | 181 | // in the case where parallelism is being turned off by |
aoqi@0 | 182 | // setting n_par_threads to 0. |
aoqi@0 | 183 | _is_par = (SharedHeap::heap()->n_par_threads() > 0); |
aoqi@0 | 184 | assert(!_is_par || |
aoqi@0 | 185 | (SharedHeap::heap()->n_par_threads() == |
aoqi@0 | 186 | SharedHeap::heap()->workers()->active_workers()), "Mismatch"); |
aoqi@0 | 187 | } |
aoqi@0 | 188 | |
aoqi@0 | 189 | bool ClearNoncleanCardWrapper::is_word_aligned(jbyte* entry) { |
aoqi@0 | 190 | return (((intptr_t)entry) & (BytesPerWord-1)) == 0; |
aoqi@0 | 191 | } |
aoqi@0 | 192 | |
aoqi@0 | 193 | void ClearNoncleanCardWrapper::do_MemRegion(MemRegion mr) { |
aoqi@0 | 194 | assert(mr.word_size() > 0, "Error"); |
aoqi@0 | 195 | assert(_ct->is_aligned(mr.start()), "mr.start() should be card aligned"); |
aoqi@0 | 196 | // mr.end() may not necessarily be card aligned. |
aoqi@0 | 197 | jbyte* cur_entry = _ct->byte_for(mr.last()); |
aoqi@0 | 198 | const jbyte* limit = _ct->byte_for(mr.start()); |
aoqi@0 | 199 | HeapWord* end_of_non_clean = mr.end(); |
aoqi@0 | 200 | HeapWord* start_of_non_clean = end_of_non_clean; |
aoqi@0 | 201 | while (cur_entry >= limit) { |
aoqi@0 | 202 | HeapWord* cur_hw = _ct->addr_for(cur_entry); |
aoqi@0 | 203 | if ((*cur_entry != CardTableRS::clean_card_val()) && clear_card(cur_entry)) { |
aoqi@0 | 204 | // Continue the dirty range by opening the |
aoqi@0 | 205 | // dirty window one card to the left. |
aoqi@0 | 206 | start_of_non_clean = cur_hw; |
aoqi@0 | 207 | } else { |
aoqi@0 | 208 | // We hit a "clean" card; process any non-empty |
aoqi@0 | 209 | // "dirty" range accumulated so far. |
aoqi@0 | 210 | if (start_of_non_clean < end_of_non_clean) { |
aoqi@0 | 211 | const MemRegion mrd(start_of_non_clean, end_of_non_clean); |
aoqi@0 | 212 | _dirty_card_closure->do_MemRegion(mrd); |
aoqi@0 | 213 | } |
aoqi@0 | 214 | |
aoqi@0 | 215 | // fast forward through potential continuous whole-word range of clean cards beginning at a word-boundary |
aoqi@0 | 216 | if (is_word_aligned(cur_entry)) { |
aoqi@0 | 217 | jbyte* cur_row = cur_entry - BytesPerWord; |
aoqi@0 | 218 | while (cur_row >= limit && *((intptr_t*)cur_row) == CardTableRS::clean_card_row()) { |
aoqi@0 | 219 | cur_row -= BytesPerWord; |
aoqi@0 | 220 | } |
aoqi@0 | 221 | cur_entry = cur_row + BytesPerWord; |
aoqi@0 | 222 | cur_hw = _ct->addr_for(cur_entry); |
aoqi@0 | 223 | } |
aoqi@0 | 224 | |
aoqi@0 | 225 | // Reset the dirty window, while continuing to look |
aoqi@0 | 226 | // for the next dirty card that will start a |
aoqi@0 | 227 | // new dirty window. |
aoqi@0 | 228 | end_of_non_clean = cur_hw; |
aoqi@0 | 229 | start_of_non_clean = cur_hw; |
aoqi@0 | 230 | } |
aoqi@0 | 231 | // Note that "cur_entry" leads "start_of_non_clean" in |
aoqi@0 | 232 | // its leftward excursion after this point |
aoqi@0 | 233 | // in the loop and, when we hit the left end of "mr", |
aoqi@0 | 234 | // will point off of the left end of the card-table |
aoqi@0 | 235 | // for "mr". |
aoqi@0 | 236 | cur_entry--; |
aoqi@0 | 237 | } |
aoqi@0 | 238 | // If the first card of "mr" was dirty, we will have |
aoqi@0 | 239 | // been left with a dirty window, co-initial with "mr", |
aoqi@0 | 240 | // which we now process. |
aoqi@0 | 241 | if (start_of_non_clean < end_of_non_clean) { |
aoqi@0 | 242 | const MemRegion mrd(start_of_non_clean, end_of_non_clean); |
aoqi@0 | 243 | _dirty_card_closure->do_MemRegion(mrd); |
aoqi@0 | 244 | } |
aoqi@0 | 245 | } |
aoqi@0 | 246 | |
aoqi@0 | 247 | // clean (by dirty->clean before) ==> cur_younger_gen |
aoqi@0 | 248 | // dirty ==> cur_youngergen_and_prev_nonclean_card |
aoqi@0 | 249 | // precleaned ==> cur_youngergen_and_prev_nonclean_card |
aoqi@0 | 250 | // prev-younger-gen ==> cur_youngergen_and_prev_nonclean_card |
aoqi@0 | 251 | // cur-younger-gen ==> cur_younger_gen |
aoqi@0 | 252 | // cur_youngergen_and_prev_nonclean_card ==> no change. |
aoqi@0 | 253 | void CardTableRS::write_ref_field_gc_par(void* field, oop new_val) { |
aoqi@0 | 254 | jbyte* entry = ct_bs()->byte_for(field); |
fujie@116 | 255 | #ifdef MIPS64 |
aoqi@8019 | 256 | if (UseSyncLevel >= 2000) OrderAccess::fence(); |
fujie@116 | 257 | #endif |
aoqi@0 | 258 | do { |
aoqi@0 | 259 | jbyte entry_val = *entry; |
aoqi@0 | 260 | // We put this first because it's probably the most common case. |
aoqi@0 | 261 | if (entry_val == clean_card_val()) { |
aoqi@0 | 262 | // No threat of contention with cleaning threads. |
aoqi@0 | 263 | *entry = cur_youngergen_card_val(); |
aoqi@0 | 264 | return; |
aoqi@0 | 265 | } else if (card_is_dirty_wrt_gen_iter(entry_val) |
aoqi@0 | 266 | || is_prev_youngergen_card_val(entry_val)) { |
aoqi@0 | 267 | // Mark it as both cur and prev youngergen; card cleaning thread will |
aoqi@0 | 268 | // eventually remove the previous stuff. |
aoqi@0 | 269 | jbyte new_val = cur_youngergen_and_prev_nonclean_card; |
aoqi@0 | 270 | jbyte res = Atomic::cmpxchg(new_val, entry, entry_val); |
aoqi@0 | 271 | // Did the CAS succeed? |
fujie@116 | 272 | if (res == entry_val) { |
fujie@116 | 273 | #ifdef MIPS64 |
aoqi@8019 | 274 | if (UseSyncLevel >= 2000) OrderAccess::fence(); |
fujie@116 | 275 | #endif |
fujie@116 | 276 | return; |
fujie@116 | 277 | } |
aoqi@0 | 278 | // Otherwise, retry, to see the new value. |
aoqi@0 | 279 | continue; |
aoqi@0 | 280 | } else { |
aoqi@0 | 281 | assert(entry_val == cur_youngergen_and_prev_nonclean_card |
aoqi@0 | 282 | || entry_val == cur_youngergen_card_val(), |
aoqi@0 | 283 | "should be only possibilities."); |
aoqi@0 | 284 | return; |
aoqi@0 | 285 | } |
aoqi@0 | 286 | } while (true); |
aoqi@0 | 287 | } |
aoqi@0 | 288 | |
aoqi@0 | 289 | void CardTableRS::younger_refs_in_space_iterate(Space* sp, |
aoqi@0 | 290 | OopsInGenClosure* cl) { |
aoqi@0 | 291 | const MemRegion urasm = sp->used_region_at_save_marks(); |
aoqi@0 | 292 | #ifdef ASSERT |
aoqi@0 | 293 | // Convert the assertion check to a warning if we are running |
aoqi@0 | 294 | // CMS+ParNew until related bug is fixed. |
aoqi@0 | 295 | MemRegion ur = sp->used_region(); |
aoqi@0 | 296 | assert(ur.contains(urasm) || (UseConcMarkSweepGC && UseParNewGC), |
aoqi@0 | 297 | err_msg("Did you forget to call save_marks()? " |
aoqi@0 | 298 | "[" PTR_FORMAT ", " PTR_FORMAT ") is not contained in " |
aoqi@0 | 299 | "[" PTR_FORMAT ", " PTR_FORMAT ")", |
aoqi@0 | 300 | p2i(urasm.start()), p2i(urasm.end()), p2i(ur.start()), p2i(ur.end()))); |
aoqi@0 | 301 | // In the case of CMS+ParNew, issue a warning |
aoqi@0 | 302 | if (!ur.contains(urasm)) { |
aoqi@0 | 303 | assert(UseConcMarkSweepGC && UseParNewGC, "Tautology: see assert above"); |
aoqi@0 | 304 | warning("CMS+ParNew: Did you forget to call save_marks()? " |
aoqi@0 | 305 | "[" PTR_FORMAT ", " PTR_FORMAT ") is not contained in " |
aoqi@0 | 306 | "[" PTR_FORMAT ", " PTR_FORMAT ")", |
aoqi@0 | 307 | p2i(urasm.start()), p2i(urasm.end()), p2i(ur.start()), p2i(ur.end())); |
aoqi@0 | 308 | MemRegion ur2 = sp->used_region(); |
aoqi@0 | 309 | MemRegion urasm2 = sp->used_region_at_save_marks(); |
aoqi@0 | 310 | if (!ur.equals(ur2)) { |
aoqi@0 | 311 | warning("CMS+ParNew: Flickering used_region()!!"); |
aoqi@0 | 312 | } |
aoqi@0 | 313 | if (!urasm.equals(urasm2)) { |
aoqi@0 | 314 | warning("CMS+ParNew: Flickering used_region_at_save_marks()!!"); |
aoqi@0 | 315 | } |
aoqi@0 | 316 | ShouldNotReachHere(); |
aoqi@0 | 317 | } |
aoqi@0 | 318 | #endif |
aoqi@0 | 319 | _ct_bs->non_clean_card_iterate_possibly_parallel(sp, urasm, cl, this); |
aoqi@0 | 320 | } |
aoqi@0 | 321 | |
aoqi@0 | 322 | void CardTableRS::clear_into_younger(Generation* old_gen) { |
aoqi@0 | 323 | assert(old_gen->level() == 1, "Should only be called for the old generation"); |
aoqi@0 | 324 | // The card tables for the youngest gen need never be cleared. |
aoqi@0 | 325 | // There's a bit of subtlety in the clear() and invalidate() |
aoqi@0 | 326 | // methods that we exploit here and in invalidate_or_clear() |
aoqi@0 | 327 | // below to avoid missing cards at the fringes. If clear() or |
aoqi@0 | 328 | // invalidate() are changed in the future, this code should |
aoqi@0 | 329 | // be revisited. 20040107.ysr |
aoqi@0 | 330 | clear(old_gen->prev_used_region()); |
aoqi@0 | 331 | } |
aoqi@0 | 332 | |
aoqi@0 | 333 | void CardTableRS::invalidate_or_clear(Generation* old_gen) { |
aoqi@0 | 334 | assert(old_gen->level() == 1, "Should only be called for the old generation"); |
aoqi@0 | 335 | // Invalidate the cards for the currently occupied part of |
aoqi@0 | 336 | // the old generation and clear the cards for the |
aoqi@0 | 337 | // unoccupied part of the generation (if any, making use |
aoqi@0 | 338 | // of that generation's prev_used_region to determine that |
aoqi@0 | 339 | // region). No need to do anything for the youngest |
aoqi@0 | 340 | // generation. Also see note#20040107.ysr above. |
aoqi@0 | 341 | MemRegion used_mr = old_gen->used_region(); |
aoqi@0 | 342 | MemRegion to_be_cleared_mr = old_gen->prev_used_region().minus(used_mr); |
aoqi@0 | 343 | if (!to_be_cleared_mr.is_empty()) { |
aoqi@0 | 344 | clear(to_be_cleared_mr); |
aoqi@0 | 345 | } |
aoqi@0 | 346 | invalidate(used_mr); |
aoqi@0 | 347 | } |
aoqi@0 | 348 | |
aoqi@0 | 349 | |
aoqi@0 | 350 | class VerifyCleanCardClosure: public OopClosure { |
aoqi@0 | 351 | private: |
aoqi@0 | 352 | HeapWord* _boundary; |
aoqi@0 | 353 | HeapWord* _begin; |
aoqi@0 | 354 | HeapWord* _end; |
aoqi@0 | 355 | protected: |
aoqi@0 | 356 | template <class T> void do_oop_work(T* p) { |
aoqi@0 | 357 | HeapWord* jp = (HeapWord*)p; |
aoqi@0 | 358 | assert(jp >= _begin && jp < _end, |
aoqi@0 | 359 | err_msg("Error: jp " PTR_FORMAT " should be within " |
aoqi@0 | 360 | "[_begin, _end) = [" PTR_FORMAT "," PTR_FORMAT ")", |
aoqi@0 | 361 | p2i(jp), p2i(_begin), p2i(_end))); |
aoqi@0 | 362 | oop obj = oopDesc::load_decode_heap_oop(p); |
aoqi@0 | 363 | guarantee(obj == NULL || (HeapWord*)obj >= _boundary, |
aoqi@0 | 364 | err_msg("pointer " PTR_FORMAT " at " PTR_FORMAT " on " |
aoqi@0 | 365 | "clean card crosses boundary" PTR_FORMAT, |
aoqi@0 | 366 | p2i((HeapWord*)obj), p2i(jp), p2i(_boundary))); |
aoqi@0 | 367 | } |
aoqi@0 | 368 | |
aoqi@0 | 369 | public: |
aoqi@0 | 370 | VerifyCleanCardClosure(HeapWord* b, HeapWord* begin, HeapWord* end) : |
aoqi@0 | 371 | _boundary(b), _begin(begin), _end(end) { |
aoqi@0 | 372 | assert(b <= begin, |
aoqi@0 | 373 | err_msg("Error: boundary " PTR_FORMAT " should be at or below begin " PTR_FORMAT, |
aoqi@0 | 374 | p2i(b), p2i(begin))); |
aoqi@0 | 375 | assert(begin <= end, |
aoqi@0 | 376 | err_msg("Error: begin " PTR_FORMAT " should be strictly below end " PTR_FORMAT, |
aoqi@0 | 377 | p2i(begin), p2i(end))); |
aoqi@0 | 378 | } |
aoqi@0 | 379 | |
aoqi@0 | 380 | virtual void do_oop(oop* p) { VerifyCleanCardClosure::do_oop_work(p); } |
aoqi@0 | 381 | virtual void do_oop(narrowOop* p) { VerifyCleanCardClosure::do_oop_work(p); } |
aoqi@0 | 382 | }; |
aoqi@0 | 383 | |
aoqi@0 | 384 | class VerifyCTSpaceClosure: public SpaceClosure { |
aoqi@0 | 385 | private: |
aoqi@0 | 386 | CardTableRS* _ct; |
aoqi@0 | 387 | HeapWord* _boundary; |
aoqi@0 | 388 | public: |
aoqi@0 | 389 | VerifyCTSpaceClosure(CardTableRS* ct, HeapWord* boundary) : |
aoqi@0 | 390 | _ct(ct), _boundary(boundary) {} |
aoqi@0 | 391 | virtual void do_space(Space* s) { _ct->verify_space(s, _boundary); } |
aoqi@0 | 392 | }; |
aoqi@0 | 393 | |
aoqi@0 | 394 | class VerifyCTGenClosure: public GenCollectedHeap::GenClosure { |
aoqi@0 | 395 | CardTableRS* _ct; |
aoqi@0 | 396 | public: |
aoqi@0 | 397 | VerifyCTGenClosure(CardTableRS* ct) : _ct(ct) {} |
aoqi@0 | 398 | void do_generation(Generation* gen) { |
aoqi@0 | 399 | // Skip the youngest generation. |
aoqi@0 | 400 | if (gen->level() == 0) return; |
aoqi@0 | 401 | // Normally, we're interested in pointers to younger generations. |
aoqi@0 | 402 | VerifyCTSpaceClosure blk(_ct, gen->reserved().start()); |
aoqi@0 | 403 | gen->space_iterate(&blk, true); |
aoqi@0 | 404 | } |
aoqi@0 | 405 | }; |
aoqi@0 | 406 | |
aoqi@0 | 407 | void CardTableRS::verify_space(Space* s, HeapWord* gen_boundary) { |
aoqi@0 | 408 | // We don't need to do young-gen spaces. |
aoqi@0 | 409 | if (s->end() <= gen_boundary) return; |
aoqi@0 | 410 | MemRegion used = s->used_region(); |
aoqi@0 | 411 | |
aoqi@0 | 412 | jbyte* cur_entry = byte_for(used.start()); |
aoqi@0 | 413 | jbyte* limit = byte_after(used.last()); |
aoqi@0 | 414 | while (cur_entry < limit) { |
aoqi@0 | 415 | if (*cur_entry == CardTableModRefBS::clean_card) { |
aoqi@0 | 416 | jbyte* first_dirty = cur_entry+1; |
aoqi@0 | 417 | while (first_dirty < limit && |
aoqi@0 | 418 | *first_dirty == CardTableModRefBS::clean_card) { |
aoqi@0 | 419 | first_dirty++; |
aoqi@0 | 420 | } |
aoqi@0 | 421 | // If the first object is a regular object, and it has a |
aoqi@0 | 422 | // young-to-old field, that would mark the previous card. |
aoqi@0 | 423 | HeapWord* boundary = addr_for(cur_entry); |
aoqi@0 | 424 | HeapWord* end = (first_dirty >= limit) ? used.end() : addr_for(first_dirty); |
aoqi@0 | 425 | HeapWord* boundary_block = s->block_start(boundary); |
aoqi@0 | 426 | HeapWord* begin = boundary; // Until proven otherwise. |
aoqi@0 | 427 | HeapWord* start_block = boundary_block; // Until proven otherwise. |
aoqi@0 | 428 | if (boundary_block < boundary) { |
aoqi@0 | 429 | if (s->block_is_obj(boundary_block) && s->obj_is_alive(boundary_block)) { |
aoqi@0 | 430 | oop boundary_obj = oop(boundary_block); |
aoqi@0 | 431 | if (!boundary_obj->is_objArray() && |
aoqi@0 | 432 | !boundary_obj->is_typeArray()) { |
aoqi@0 | 433 | guarantee(cur_entry > byte_for(used.start()), |
aoqi@0 | 434 | "else boundary would be boundary_block"); |
aoqi@0 | 435 | if (*byte_for(boundary_block) != CardTableModRefBS::clean_card) { |
aoqi@0 | 436 | begin = boundary_block + s->block_size(boundary_block); |
aoqi@0 | 437 | start_block = begin; |
aoqi@0 | 438 | } |
aoqi@0 | 439 | } |
aoqi@0 | 440 | } |
aoqi@0 | 441 | } |
aoqi@0 | 442 | // Now traverse objects until end. |
aoqi@0 | 443 | if (begin < end) { |
aoqi@0 | 444 | MemRegion mr(begin, end); |
aoqi@0 | 445 | VerifyCleanCardClosure verify_blk(gen_boundary, begin, end); |
aoqi@0 | 446 | for (HeapWord* cur = start_block; cur < end; cur += s->block_size(cur)) { |
aoqi@0 | 447 | if (s->block_is_obj(cur) && s->obj_is_alive(cur)) { |
aoqi@0 | 448 | oop(cur)->oop_iterate_no_header(&verify_blk, mr); |
aoqi@0 | 449 | } |
aoqi@0 | 450 | } |
aoqi@0 | 451 | } |
aoqi@0 | 452 | cur_entry = first_dirty; |
aoqi@0 | 453 | } else { |
aoqi@0 | 454 | // We'd normally expect that cur_youngergen_and_prev_nonclean_card |
aoqi@0 | 455 | // is a transient value, that cannot be in the card table |
aoqi@0 | 456 | // except during GC, and thus assert that: |
aoqi@0 | 457 | // guarantee(*cur_entry != cur_youngergen_and_prev_nonclean_card, |
aoqi@0 | 458 | // "Illegal CT value"); |
aoqi@0 | 459 | // That however, need not hold, as will become clear in the |
aoqi@0 | 460 | // following... |
aoqi@0 | 461 | |
aoqi@0 | 462 | // We'd normally expect that if we are in the parallel case, |
aoqi@0 | 463 | // we can't have left a prev value (which would be different |
aoqi@0 | 464 | // from the current value) in the card table, and so we'd like to |
aoqi@0 | 465 | // assert that: |
aoqi@0 | 466 | // guarantee(cur_youngergen_card_val() == youngergen_card |
aoqi@0 | 467 | // || !is_prev_youngergen_card_val(*cur_entry), |
aoqi@0 | 468 | // "Illegal CT value"); |
aoqi@0 | 469 | // That, however, may not hold occasionally, because of |
aoqi@0 | 470 | // CMS or MSC in the old gen. To wit, consider the |
aoqi@0 | 471 | // following two simple illustrative scenarios: |
aoqi@0 | 472 | // (a) CMS: Consider the case where a large object L |
aoqi@0 | 473 | // spanning several cards is allocated in the old |
aoqi@0 | 474 | // gen, and has a young gen reference stored in it, dirtying |
aoqi@0 | 475 | // some interior cards. A young collection scans the card, |
aoqi@0 | 476 | // finds a young ref and installs a youngergenP_n value. |
aoqi@0 | 477 | // L then goes dead. Now a CMS collection starts, |
aoqi@0 | 478 | // finds L dead and sweeps it up. Assume that L is |
aoqi@0 | 479 | // abutting _unallocated_blk, so _unallocated_blk is |
aoqi@0 | 480 | // adjusted down to (below) L. Assume further that |
aoqi@0 | 481 | // no young collection intervenes during this CMS cycle. |
aoqi@0 | 482 | // The next young gen cycle will not get to look at this |
aoqi@0 | 483 | // youngergenP_n card since it lies in the unoccupied |
aoqi@0 | 484 | // part of the space. |
aoqi@0 | 485 | // Some young collections later the blocks on this |
aoqi@0 | 486 | // card can be re-allocated either due to direct allocation |
aoqi@0 | 487 | // or due to absorbing promotions. At this time, the |
aoqi@0 | 488 | // before-gc verification will fail the above assert. |
aoqi@0 | 489 | // (b) MSC: In this case, an object L with a young reference |
aoqi@0 | 490 | // is on a card that (therefore) holds a youngergen_n value. |
aoqi@0 | 491 | // Suppose also that L lies towards the end of the used |
aoqi@0 | 492 | // the used space before GC. An MSC collection |
aoqi@0 | 493 | // occurs that compacts to such an extent that this |
aoqi@0 | 494 | // card is no longer in the occupied part of the space. |
aoqi@0 | 495 | // Since current code in MSC does not always clear cards |
aoqi@0 | 496 | // in the unused part of old gen, this stale youngergen_n |
aoqi@0 | 497 | // value is left behind and can later be covered by |
aoqi@0 | 498 | // an object when promotion or direct allocation |
aoqi@0 | 499 | // re-allocates that part of the heap. |
aoqi@0 | 500 | // |
aoqi@0 | 501 | // Fortunately, the presence of such stale card values is |
aoqi@0 | 502 | // "only" a minor annoyance in that subsequent young collections |
aoqi@0 | 503 | // might needlessly scan such cards, but would still never corrupt |
aoqi@0 | 504 | // the heap as a result. However, it's likely not to be a significant |
aoqi@0 | 505 | // performance inhibitor in practice. For instance, |
aoqi@0 | 506 | // some recent measurements with unoccupied cards eagerly cleared |
aoqi@0 | 507 | // out to maintain this invariant, showed next to no |
aoqi@0 | 508 | // change in young collection times; of course one can construct |
aoqi@0 | 509 | // degenerate examples where the cost can be significant.) |
aoqi@0 | 510 | // Note, in particular, that if the "stale" card is modified |
aoqi@0 | 511 | // after re-allocation, it would be dirty, not "stale". Thus, |
aoqi@0 | 512 | // we can never have a younger ref in such a card and it is |
aoqi@0 | 513 | // safe not to scan that card in any collection. [As we see |
aoqi@0 | 514 | // below, we do some unnecessary scanning |
aoqi@0 | 515 | // in some cases in the current parallel scanning algorithm.] |
aoqi@0 | 516 | // |
aoqi@0 | 517 | // The main point below is that the parallel card scanning code |
aoqi@0 | 518 | // deals correctly with these stale card values. There are two main |
aoqi@0 | 519 | // cases to consider where we have a stale "younger gen" value and a |
aoqi@0 | 520 | // "derivative" case to consider, where we have a stale |
aoqi@0 | 521 | // "cur_younger_gen_and_prev_non_clean" value, as will become |
aoqi@0 | 522 | // apparent in the case analysis below. |
aoqi@0 | 523 | // o Case 1. If the stale value corresponds to a younger_gen_n |
aoqi@0 | 524 | // value other than the cur_younger_gen value then the code |
aoqi@0 | 525 | // treats this as being tantamount to a prev_younger_gen |
aoqi@0 | 526 | // card. This means that the card may be unnecessarily scanned. |
aoqi@0 | 527 | // There are two sub-cases to consider: |
aoqi@0 | 528 | // o Case 1a. Let us say that the card is in the occupied part |
aoqi@0 | 529 | // of the generation at the time the collection begins. In |
aoqi@0 | 530 | // that case the card will be either cleared when it is scanned |
aoqi@0 | 531 | // for young pointers, or will be set to cur_younger_gen as a |
aoqi@0 | 532 | // result of promotion. (We have elided the normal case where |
aoqi@0 | 533 | // the scanning thread and the promoting thread interleave |
aoqi@0 | 534 | // possibly resulting in a transient |
aoqi@0 | 535 | // cur_younger_gen_and_prev_non_clean value before settling |
aoqi@0 | 536 | // to cur_younger_gen. [End Case 1a.] |
aoqi@0 | 537 | // o Case 1b. Consider now the case when the card is in the unoccupied |
aoqi@0 | 538 | // part of the space which becomes occupied because of promotions |
aoqi@0 | 539 | // into it during the current young GC. In this case the card |
aoqi@0 | 540 | // will never be scanned for young references. The current |
aoqi@0 | 541 | // code will set the card value to either |
aoqi@0 | 542 | // cur_younger_gen_and_prev_non_clean or leave |
aoqi@0 | 543 | // it with its stale value -- because the promotions didn't |
aoqi@0 | 544 | // result in any younger refs on that card. Of these two |
aoqi@0 | 545 | // cases, the latter will be covered in Case 1a during |
aoqi@0 | 546 | // a subsequent scan. To deal with the former case, we need |
aoqi@0 | 547 | // to further consider how we deal with a stale value of |
aoqi@0 | 548 | // cur_younger_gen_and_prev_non_clean in our case analysis |
aoqi@0 | 549 | // below. This we do in Case 3 below. [End Case 1b] |
aoqi@0 | 550 | // [End Case 1] |
aoqi@0 | 551 | // o Case 2. If the stale value corresponds to cur_younger_gen being |
aoqi@0 | 552 | // a value not necessarily written by a current promotion, the |
aoqi@0 | 553 | // card will not be scanned by the younger refs scanning code. |
aoqi@0 | 554 | // (This is OK since as we argued above such cards cannot contain |
aoqi@0 | 555 | // any younger refs.) The result is that this value will be |
aoqi@0 | 556 | // treated as a prev_younger_gen value in a subsequent collection, |
aoqi@0 | 557 | // which is addressed in Case 1 above. [End Case 2] |
aoqi@0 | 558 | // o Case 3. We here consider the "derivative" case from Case 1b. above |
aoqi@0 | 559 | // because of which we may find a stale |
aoqi@0 | 560 | // cur_younger_gen_and_prev_non_clean card value in the table. |
aoqi@0 | 561 | // Once again, as in Case 1, we consider two subcases, depending |
aoqi@0 | 562 | // on whether the card lies in the occupied or unoccupied part |
aoqi@0 | 563 | // of the space at the start of the young collection. |
aoqi@0 | 564 | // o Case 3a. Let us say the card is in the occupied part of |
aoqi@0 | 565 | // the old gen at the start of the young collection. In that |
aoqi@0 | 566 | // case, the card will be scanned by the younger refs scanning |
aoqi@0 | 567 | // code which will set it to cur_younger_gen. In a subsequent |
aoqi@0 | 568 | // scan, the card will be considered again and get its final |
aoqi@0 | 569 | // correct value. [End Case 3a] |
aoqi@0 | 570 | // o Case 3b. Now consider the case where the card is in the |
aoqi@0 | 571 | // unoccupied part of the old gen, and is occupied as a result |
aoqi@0 | 572 | // of promotions during thus young gc. In that case, |
aoqi@0 | 573 | // the card will not be scanned for younger refs. The presence |
aoqi@0 | 574 | // of newly promoted objects on the card will then result in |
aoqi@0 | 575 | // its keeping the value cur_younger_gen_and_prev_non_clean |
aoqi@0 | 576 | // value, which we have dealt with in Case 3 here. [End Case 3b] |
aoqi@0 | 577 | // [End Case 3] |
aoqi@0 | 578 | // |
aoqi@0 | 579 | // (Please refer to the code in the helper class |
aoqi@0 | 580 | // ClearNonCleanCardWrapper and in CardTableModRefBS for details.) |
aoqi@0 | 581 | // |
aoqi@0 | 582 | // The informal arguments above can be tightened into a formal |
aoqi@0 | 583 | // correctness proof and it behooves us to write up such a proof, |
aoqi@0 | 584 | // or to use model checking to prove that there are no lingering |
aoqi@0 | 585 | // concerns. |
aoqi@0 | 586 | // |
aoqi@0 | 587 | // Clearly because of Case 3b one cannot bound the time for |
aoqi@0 | 588 | // which a card will retain what we have called a "stale" value. |
aoqi@0 | 589 | // However, one can obtain a Loose upper bound on the redundant |
aoqi@0 | 590 | // work as a result of such stale values. Note first that any |
aoqi@0 | 591 | // time a stale card lies in the occupied part of the space at |
aoqi@0 | 592 | // the start of the collection, it is scanned by younger refs |
aoqi@0 | 593 | // code and we can define a rank function on card values that |
aoqi@0 | 594 | // declines when this is so. Note also that when a card does not |
aoqi@0 | 595 | // lie in the occupied part of the space at the beginning of a |
aoqi@0 | 596 | // young collection, its rank can either decline or stay unchanged. |
aoqi@0 | 597 | // In this case, no extra work is done in terms of redundant |
aoqi@0 | 598 | // younger refs scanning of that card. |
aoqi@0 | 599 | // Then, the case analysis above reveals that, in the worst case, |
aoqi@0 | 600 | // any such stale card will be scanned unnecessarily at most twice. |
aoqi@0 | 601 | // |
aoqi@0 | 602 | // It is nonethelss advisable to try and get rid of some of this |
aoqi@0 | 603 | // redundant work in a subsequent (low priority) re-design of |
aoqi@0 | 604 | // the card-scanning code, if only to simplify the underlying |
aoqi@0 | 605 | // state machine analysis/proof. ysr 1/28/2002. XXX |
aoqi@0 | 606 | cur_entry++; |
aoqi@0 | 607 | } |
aoqi@0 | 608 | } |
aoqi@0 | 609 | } |
aoqi@0 | 610 | |
aoqi@0 | 611 | void CardTableRS::verify() { |
aoqi@0 | 612 | // At present, we only know how to verify the card table RS for |
aoqi@0 | 613 | // generational heaps. |
aoqi@0 | 614 | VerifyCTGenClosure blk(this); |
aoqi@0 | 615 | CollectedHeap* ch = Universe::heap(); |
aoqi@0 | 616 | |
aoqi@0 | 617 | if (ch->kind() == CollectedHeap::GenCollectedHeap) { |
aoqi@0 | 618 | GenCollectedHeap::heap()->generation_iterate(&blk, false); |
aoqi@0 | 619 | _ct_bs->verify(); |
aoqi@0 | 620 | } |
aoqi@0 | 621 | } |
aoqi@0 | 622 | |
aoqi@0 | 623 | |
aoqi@0 | 624 | void CardTableRS::verify_aligned_region_empty(MemRegion mr) { |
aoqi@0 | 625 | if (!mr.is_empty()) { |
aoqi@0 | 626 | jbyte* cur_entry = byte_for(mr.start()); |
aoqi@0 | 627 | jbyte* limit = byte_after(mr.last()); |
aoqi@0 | 628 | // The region mr may not start on a card boundary so |
aoqi@0 | 629 | // the first card may reflect a write to the space |
aoqi@0 | 630 | // just prior to mr. |
aoqi@0 | 631 | if (!is_aligned(mr.start())) { |
aoqi@0 | 632 | cur_entry++; |
aoqi@0 | 633 | } |
aoqi@0 | 634 | for (;cur_entry < limit; cur_entry++) { |
aoqi@0 | 635 | guarantee(*cur_entry == CardTableModRefBS::clean_card, |
aoqi@0 | 636 | "Unexpected dirty card found"); |
aoqi@0 | 637 | } |
aoqi@0 | 638 | } |
aoqi@0 | 639 | } |