1 /*
2 * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "classfile/altHashing.hpp"
27 #include "classfile/javaClasses.hpp"
28 #include "classfile/stringTable.hpp"
29 #include "classfile/systemDictionary.hpp"
30 #include "gc_interface/collectedHeap.inline.hpp"
31 #include "memory/allocation.inline.hpp"
32 #include "memory/filemap.hpp"
33 #include "memory/gcLocker.inline.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "oops/oop.inline2.hpp"
36 #include "runtime/mutexLocker.hpp"
37 #include "utilities/hashtable.inline.hpp"
38 #if INCLUDE_ALL_GCS
39 #include "gc_implementation/g1/g1StringDedup.hpp"
40 #endif
41
42 // the number of buckets a thread claims
43 const int ClaimChunkSize = 32;
44
45 #ifdef ASSERT
46 class StableMemoryChecker : public StackObj {
47 enum { _bufsize = wordSize*4 };
48
49 address _region;
50 jint _size;
51 u1 _save_buf[_bufsize];
52
53 int sample(u1* save_buf) {
54 if (_size <= _bufsize) {
55 memcpy(save_buf, _region, _size);
56 return _size;
57 } else {
58 // copy head and tail
59 memcpy(&save_buf[0], _region, _bufsize/2);
60 memcpy(&save_buf[_bufsize/2], _region + _size - _bufsize/2, _bufsize/2);
61 return (_bufsize/2)*2;
62 }
63 }
64
65 public:
66 StableMemoryChecker(const void* region, jint size) {
67 _region = (address) region;
68 _size = size;
69 sample(_save_buf);
70 }
71
72 bool verify() {
73 u1 check_buf[sizeof(_save_buf)];
74 int check_size = sample(check_buf);
75 return (0 == memcmp(_save_buf, check_buf, check_size));
76 }
77
78 void set_region(const void* region) { _region = (address) region; }
79 };
80 #endif
81
82
83 // --------------------------------------------------------------------------
84 StringTable* StringTable::_the_table = NULL;
85
86 bool StringTable::_needs_rehashing = false;
87
88 volatile int StringTable::_parallel_claimed_idx = 0;
89
90 // Pick hashing algorithm
91 unsigned int StringTable::hash_string(const jchar* s, int len) {
92 return use_alternate_hashcode() ? AltHashing::murmur3_32(seed(), s, len) :
93 java_lang_String::hash_code(s, len);
94 }
95
96 oop StringTable::lookup(int index, jchar* name,
97 int len, unsigned int hash) {
98 int count = 0;
99 for (HashtableEntry<oop, mtSymbol>* l = bucket(index); l != NULL; l = l->next()) {
100 count++;
101 if (l->hash() == hash) {
102 if (java_lang_String::equals(l->literal(), name, len)) {
103 return l->literal();
104 }
105 }
106 }
107 // If the bucket size is too deep check if this hash code is insufficient.
108 if (count >= BasicHashtable<mtSymbol>::rehash_count && !needs_rehashing()) {
109 _needs_rehashing = check_rehash_table(count);
110 }
111 return NULL;
112 }
113
114
115 oop StringTable::basic_add(int index_arg, Handle string, jchar* name,
116 int len, unsigned int hashValue_arg, TRAPS) {
117
118 assert(java_lang_String::equals(string(), name, len),
119 "string must be properly initialized");
120 // Cannot hit a safepoint in this function because the "this" pointer can move.
121 No_Safepoint_Verifier nsv;
122
123 // Check if the symbol table has been rehashed, if so, need to recalculate
124 // the hash value and index before second lookup.
125 unsigned int hashValue;
126 int index;
127 if (use_alternate_hashcode()) {
128 hashValue = hash_string(name, len);
129 index = hash_to_index(hashValue);
130 } else {
131 hashValue = hashValue_arg;
132 index = index_arg;
133 }
134
135 // Since look-up was done lock-free, we need to check if another
136 // thread beat us in the race to insert the symbol.
137
138 oop test = lookup(index, name, len, hashValue); // calls lookup(u1*, int)
139 if (test != NULL) {
140 // Entry already added
141 return test;
142 }
143
144 HashtableEntry<oop, mtSymbol>* entry = new_entry(hashValue, string());
145 add_entry(index, entry);
146 return string();
147 }
148
149
150 oop StringTable::lookup(Symbol* symbol) {
151 ResourceMark rm;
152 int length;
153 jchar* chars = symbol->as_unicode(length);
154 return lookup(chars, length);
155 }
156
157
158 oop StringTable::lookup(jchar* name, int len) {
159 unsigned int hash = hash_string(name, len);
160 int index = the_table()->hash_to_index(hash);
161 return the_table()->lookup(index, name, len, hash);
162 }
163
164
165 oop StringTable::intern(Handle string_or_null, jchar* name,
166 int len, TRAPS) {
167 unsigned int hashValue = hash_string(name, len);
168 int index = the_table()->hash_to_index(hashValue);
169 oop found_string = the_table()->lookup(index, name, len, hashValue);
170
171 // Found
172 if (found_string != NULL) return found_string;
173
174 debug_only(StableMemoryChecker smc(name, len * sizeof(name[0])));
175 assert(!Universe::heap()->is_in_reserved(name),
176 "proposed name of symbol must be stable");
177
178 Handle string;
179 // try to reuse the string if possible
180 if (!string_or_null.is_null()) {
181 string = string_or_null;
182 } else {
183 string = java_lang_String::create_from_unicode(name, len, CHECK_NULL);
184 }
185
186 #if INCLUDE_ALL_GCS
187 if (G1StringDedup::is_enabled()) {
188 // Deduplicate the string before it is interned. Note that we should never
189 // deduplicate a string after it has been interned. Doing so will counteract
190 // compiler optimizations done on e.g. interned string literals.
191 G1StringDedup::deduplicate(string());
192 }
193 #endif
194
195 // Grab the StringTable_lock before getting the_table() because it could
196 // change at safepoint.
197 MutexLocker ml(StringTable_lock, THREAD);
198
199 // Otherwise, add to symbol to table
200 return the_table()->basic_add(index, string, name, len,
201 hashValue, CHECK_NULL);
202 }
203
204 oop StringTable::intern(Symbol* symbol, TRAPS) {
205 if (symbol == NULL) return NULL;
206 ResourceMark rm(THREAD);
207 int length;
208 jchar* chars = symbol->as_unicode(length);
209 Handle string;
210 oop result = intern(string, chars, length, CHECK_NULL);
211 return result;
212 }
213
214
215 oop StringTable::intern(oop string, TRAPS)
216 {
217 if (string == NULL) return NULL;
218 ResourceMark rm(THREAD);
219 int length;
220 Handle h_string (THREAD, string);
221 jchar* chars = java_lang_String::as_unicode_string(string, length, CHECK_NULL);
222 oop result = intern(h_string, chars, length, CHECK_NULL);
223 return result;
224 }
225
226
227 oop StringTable::intern(const char* utf8_string, TRAPS) {
228 if (utf8_string == NULL) return NULL;
229 ResourceMark rm(THREAD);
230 int length = UTF8::unicode_length(utf8_string);
231 jchar* chars = NEW_RESOURCE_ARRAY(jchar, length);
232 UTF8::convert_to_unicode(utf8_string, chars, length);
233 Handle string;
234 oop result = intern(string, chars, length, CHECK_NULL);
235 return result;
236 }
237
238 void StringTable::unlink_or_oops_do(BoolObjectClosure* is_alive, OopClosure* f, int* processed, int* removed) {
239 buckets_unlink_or_oops_do(is_alive, f, 0, the_table()->table_size(), processed, removed);
240 }
241
242 void StringTable::possibly_parallel_unlink_or_oops_do(BoolObjectClosure* is_alive, OopClosure* f, int* processed, int* removed) {
243 // Readers of the table are unlocked, so we should only be removing
244 // entries at a safepoint.
245 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
246 const int limit = the_table()->table_size();
247
248 for (;;) {
249 // Grab next set of buckets to scan
250 int start_idx = Atomic::add(ClaimChunkSize, &_parallel_claimed_idx) - ClaimChunkSize;
251 if (start_idx >= limit) {
252 // End of table
253 break;
254 }
255
256 int end_idx = MIN2(limit, start_idx + ClaimChunkSize);
257 buckets_unlink_or_oops_do(is_alive, f, start_idx, end_idx, processed, removed);
258 }
259 }
260
261 void StringTable::buckets_oops_do(OopClosure* f, int start_idx, int end_idx) {
262 const int limit = the_table()->table_size();
263
264 assert(0 <= start_idx && start_idx <= limit,
265 err_msg("start_idx (%d) is out of bounds", start_idx));
266 assert(0 <= end_idx && end_idx <= limit,
267 err_msg("end_idx (%d) is out of bounds", end_idx));
268 assert(start_idx <= end_idx,
269 err_msg("Index ordering: start_idx=%d, end_idx=%d",
270 start_idx, end_idx));
271
272 for (int i = start_idx; i < end_idx; i += 1) {
273 HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i);
274 while (entry != NULL) {
275 assert(!entry->is_shared(), "CDS not used for the StringTable");
276
277 f->do_oop((oop*)entry->literal_addr());
278
279 entry = entry->next();
280 }
281 }
282 }
283
284 void StringTable::buckets_unlink_or_oops_do(BoolObjectClosure* is_alive, OopClosure* f, int start_idx, int end_idx, int* processed, int* removed) {
285 const int limit = the_table()->table_size();
286
287 assert(0 <= start_idx && start_idx <= limit,
288 err_msg("start_idx (%d) is out of bounds", start_idx));
289 assert(0 <= end_idx && end_idx <= limit,
290 err_msg("end_idx (%d) is out of bounds", end_idx));
291 assert(start_idx <= end_idx,
292 err_msg("Index ordering: start_idx=%d, end_idx=%d",
293 start_idx, end_idx));
294
295 for (int i = start_idx; i < end_idx; ++i) {
296 HashtableEntry<oop, mtSymbol>** p = the_table()->bucket_addr(i);
297 HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i);
298 while (entry != NULL) {
299 assert(!entry->is_shared(), "CDS not used for the StringTable");
300
301 if (is_alive->do_object_b(entry->literal())) {
302 if (f != NULL) {
303 f->do_oop((oop*)entry->literal_addr());
304 }
305 p = entry->next_addr();
306 } else {
307 *p = entry->next();
308 the_table()->free_entry(entry);
309 (*removed)++;
310 }
311 (*processed)++;
312 entry = *p;
313 }
314 }
315 }
316
317 void StringTable::oops_do(OopClosure* f) {
318 buckets_oops_do(f, 0, the_table()->table_size());
319 }
320
321 void StringTable::possibly_parallel_oops_do(OopClosure* f) {
322 const int limit = the_table()->table_size();
323
324 for (;;) {
325 // Grab next set of buckets to scan
326 int start_idx = Atomic::add(ClaimChunkSize, &_parallel_claimed_idx) - ClaimChunkSize;
327 if (start_idx >= limit) {
328 // End of table
329 break;
330 }
331
332 int end_idx = MIN2(limit, start_idx + ClaimChunkSize);
333 buckets_oops_do(f, start_idx, end_idx);
334 }
335 }
336
337 // This verification is part of Universe::verify() and needs to be quick.
338 // See StringTable::verify_and_compare() below for exhaustive verification.
339 void StringTable::verify() {
340 for (int i = 0; i < the_table()->table_size(); ++i) {
341 HashtableEntry<oop, mtSymbol>* p = the_table()->bucket(i);
342 for ( ; p != NULL; p = p->next()) {
343 oop s = p->literal();
344 guarantee(s != NULL, "interned string is NULL");
345 unsigned int h = java_lang_String::hash_string(s);
346 guarantee(p->hash() == h, "broken hash in string table entry");
347 guarantee(the_table()->hash_to_index(h) == i,
348 "wrong index in string table");
349 }
350 }
351 }
352
353 void StringTable::dump(outputStream* st) {
354 the_table()->dump_table(st, "StringTable");
355 }
356
357 StringTable::VerifyRetTypes StringTable::compare_entries(
358 int bkt1, int e_cnt1,
359 HashtableEntry<oop, mtSymbol>* e_ptr1,
360 int bkt2, int e_cnt2,
361 HashtableEntry<oop, mtSymbol>* e_ptr2) {
362 // These entries are sanity checked by verify_and_compare_entries()
363 // before this function is called.
364 oop str1 = e_ptr1->literal();
365 oop str2 = e_ptr2->literal();
366
367 if (str1 == str2) {
368 tty->print_cr("ERROR: identical oop values (0x" PTR_FORMAT ") "
369 "in entry @ bucket[%d][%d] and entry @ bucket[%d][%d]",
370 (void *)str1, bkt1, e_cnt1, bkt2, e_cnt2);
371 return _verify_fail_continue;
372 }
373
374 if (java_lang_String::equals(str1, str2)) {
375 tty->print_cr("ERROR: identical String values in entry @ "
376 "bucket[%d][%d] and entry @ bucket[%d][%d]",
377 bkt1, e_cnt1, bkt2, e_cnt2);
378 return _verify_fail_continue;
379 }
380
381 return _verify_pass;
382 }
383
384 StringTable::VerifyRetTypes StringTable::verify_entry(int bkt, int e_cnt,
385 HashtableEntry<oop, mtSymbol>* e_ptr,
386 StringTable::VerifyMesgModes mesg_mode) {
387
388 VerifyRetTypes ret = _verify_pass; // be optimistic
389
390 oop str = e_ptr->literal();
391 if (str == NULL) {
392 if (mesg_mode == _verify_with_mesgs) {
393 tty->print_cr("ERROR: NULL oop value in entry @ bucket[%d][%d]", bkt,
394 e_cnt);
395 }
396 // NULL oop means no more verifications are possible
397 return _verify_fail_done;
398 }
399
400 if (str->klass() != SystemDictionary::String_klass()) {
401 if (mesg_mode == _verify_with_mesgs) {
402 tty->print_cr("ERROR: oop is not a String in entry @ bucket[%d][%d]",
403 bkt, e_cnt);
404 }
405 // not a String means no more verifications are possible
406 return _verify_fail_done;
407 }
408
409 unsigned int h = java_lang_String::hash_string(str);
410 if (e_ptr->hash() != h) {
411 if (mesg_mode == _verify_with_mesgs) {
412 tty->print_cr("ERROR: broken hash value in entry @ bucket[%d][%d], "
413 "bkt_hash=%d, str_hash=%d", bkt, e_cnt, e_ptr->hash(), h);
414 }
415 ret = _verify_fail_continue;
416 }
417
418 if (the_table()->hash_to_index(h) != bkt) {
419 if (mesg_mode == _verify_with_mesgs) {
420 tty->print_cr("ERROR: wrong index value for entry @ bucket[%d][%d], "
421 "str_hash=%d, hash_to_index=%d", bkt, e_cnt, h,
422 the_table()->hash_to_index(h));
423 }
424 ret = _verify_fail_continue;
425 }
426
427 return ret;
428 }
429
430 // See StringTable::verify() above for the quick verification that is
431 // part of Universe::verify(). This verification is exhaustive and
432 // reports on every issue that is found. StringTable::verify() only
433 // reports on the first issue that is found.
434 //
435 // StringTable::verify_entry() checks:
436 // - oop value != NULL (same as verify())
437 // - oop value is a String
438 // - hash(String) == hash in entry (same as verify())
439 // - index for hash == index of entry (same as verify())
440 //
441 // StringTable::compare_entries() checks:
442 // - oops are unique across all entries
443 // - String values are unique across all entries
444 //
445 int StringTable::verify_and_compare_entries() {
446 assert(StringTable_lock->is_locked(), "sanity check");
447
448 int fail_cnt = 0;
449
450 // first, verify all the entries individually:
451 for (int bkt = 0; bkt < the_table()->table_size(); bkt++) {
452 HashtableEntry<oop, mtSymbol>* e_ptr = the_table()->bucket(bkt);
453 for (int e_cnt = 0; e_ptr != NULL; e_ptr = e_ptr->next(), e_cnt++) {
454 VerifyRetTypes ret = verify_entry(bkt, e_cnt, e_ptr, _verify_with_mesgs);
455 if (ret != _verify_pass) {
456 fail_cnt++;
457 }
458 }
459 }
460
461 // Optimization: if the above check did not find any failures, then
462 // the comparison loop below does not need to call verify_entry()
463 // before calling compare_entries(). If there were failures, then we
464 // have to call verify_entry() to see if the entry can be passed to
465 // compare_entries() safely. When we call verify_entry() in the loop
466 // below, we do so quietly to void duplicate messages and we don't
467 // increment fail_cnt because the failures have already been counted.
468 bool need_entry_verify = (fail_cnt != 0);
469
470 // second, verify all entries relative to each other:
471 for (int bkt1 = 0; bkt1 < the_table()->table_size(); bkt1++) {
472 HashtableEntry<oop, mtSymbol>* e_ptr1 = the_table()->bucket(bkt1);
473 for (int e_cnt1 = 0; e_ptr1 != NULL; e_ptr1 = e_ptr1->next(), e_cnt1++) {
474 if (need_entry_verify) {
475 VerifyRetTypes ret = verify_entry(bkt1, e_cnt1, e_ptr1,
476 _verify_quietly);
477 if (ret == _verify_fail_done) {
478 // cannot use the current entry to compare against other entries
479 continue;
480 }
481 }
482
483 for (int bkt2 = bkt1; bkt2 < the_table()->table_size(); bkt2++) {
484 HashtableEntry<oop, mtSymbol>* e_ptr2 = the_table()->bucket(bkt2);
485 int e_cnt2;
486 for (e_cnt2 = 0; e_ptr2 != NULL; e_ptr2 = e_ptr2->next(), e_cnt2++) {
487 if (bkt1 == bkt2 && e_cnt2 <= e_cnt1) {
488 // skip the entries up to and including the one that
489 // we're comparing against
490 continue;
491 }
492
493 if (need_entry_verify) {
494 VerifyRetTypes ret = verify_entry(bkt2, e_cnt2, e_ptr2,
495 _verify_quietly);
496 if (ret == _verify_fail_done) {
497 // cannot compare against this entry
498 continue;
499 }
500 }
501
502 // compare two entries, report and count any failures:
503 if (compare_entries(bkt1, e_cnt1, e_ptr1, bkt2, e_cnt2, e_ptr2)
504 != _verify_pass) {
505 fail_cnt++;
506 }
507 }
508 }
509 }
510 }
511 return fail_cnt;
512 }
513
514 // Create a new table and using alternate hash code, populate the new table
515 // with the existing strings. Set flag to use the alternate hash code afterwards.
516 void StringTable::rehash_table() {
517 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
518 // This should never happen with -Xshare:dump but it might in testing mode.
519 if (DumpSharedSpaces) return;
520 StringTable* new_table = new StringTable();
521
522 // Rehash the table
523 the_table()->move_to(new_table);
524
525 // Delete the table and buckets (entries are reused in new table).
526 delete _the_table;
527 // Don't check if we need rehashing until the table gets unbalanced again.
528 // Then rehash with a new global seed.
529 _needs_rehashing = false;
530 _the_table = new_table;
531 }