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/symbolTable.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/atomic.inline.hpp"
37 #include "runtime/mutexLocker.hpp"
38 #include "utilities/hashtable.inline.hpp"
39
40 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
41
42 // --------------------------------------------------------------------------
43 // the number of buckets a thread claims
44 const int ClaimChunkSize = 32;
45
46 SymbolTable* SymbolTable::_the_table = NULL;
47 // Static arena for symbols that are not deallocated
48 Arena* SymbolTable::_arena = NULL;
49 bool SymbolTable::_needs_rehashing = false;
50
51 Symbol* SymbolTable::allocate_symbol(const u1* name, int len, bool c_heap, TRAPS) {
52 assert (len <= Symbol::max_length(), "should be checked by caller");
53
54 Symbol* sym;
55
56 if (DumpSharedSpaces) {
57 // Allocate all symbols to CLD shared metaspace
58 sym = new (len, ClassLoaderData::the_null_class_loader_data(), THREAD) Symbol(name, len, -1);
59 } else if (c_heap) {
60 // refcount starts as 1
61 sym = new (len, THREAD) Symbol(name, len, 1);
62 assert(sym != NULL, "new should call vm_exit_out_of_memory if C_HEAP is exhausted");
63 } else {
64 // Allocate to global arena
65 sym = new (len, arena(), THREAD) Symbol(name, len, -1);
66 }
67 return sym;
68 }
69
70 void SymbolTable::initialize_symbols(int arena_alloc_size) {
71 // Initialize the arena for global symbols, size passed in depends on CDS.
72 if (arena_alloc_size == 0) {
73 _arena = new (mtSymbol) Arena(mtSymbol);
74 } else {
75 _arena = new (mtSymbol) Arena(mtSymbol, arena_alloc_size);
76 }
77 }
78
79 // Call function for all symbols in the symbol table.
80 void SymbolTable::symbols_do(SymbolClosure *cl) {
81 const int n = the_table()->table_size();
82 for (int i = 0; i < n; i++) {
83 for (HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
84 p != NULL;
85 p = p->next()) {
86 cl->do_symbol(p->literal_addr());
87 }
88 }
89 }
90
91 int SymbolTable::_symbols_removed = 0;
92 int SymbolTable::_symbols_counted = 0;
93 volatile int SymbolTable::_parallel_claimed_idx = 0;
94
95 void SymbolTable::buckets_unlink(int start_idx, int end_idx, int* processed, int* removed, size_t* memory_total) {
96 for (int i = start_idx; i < end_idx; ++i) {
97 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
98 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
99 while (entry != NULL) {
100 // Shared entries are normally at the end of the bucket and if we run into
101 // a shared entry, then there is nothing more to remove. However, if we
102 // have rehashed the table, then the shared entries are no longer at the
103 // end of the bucket.
104 if (entry->is_shared() && !use_alternate_hashcode()) {
105 break;
106 }
107 Symbol* s = entry->literal();
108 (*memory_total) += s->size();
109 (*processed)++;
110 assert(s != NULL, "just checking");
111 // If reference count is zero, remove.
112 if (s->refcount() == 0) {
113 assert(!entry->is_shared(), "shared entries should be kept live");
114 delete s;
115 (*removed)++;
116 *p = entry->next();
117 the_table()->free_entry(entry);
118 } else {
119 p = entry->next_addr();
120 }
121 // get next entry
122 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
123 }
124 }
125 }
126
127 // Remove unreferenced symbols from the symbol table
128 // This is done late during GC.
129 void SymbolTable::unlink(int* processed, int* removed) {
130 size_t memory_total = 0;
131 buckets_unlink(0, the_table()->table_size(), processed, removed, &memory_total);
132 _symbols_removed += *removed;
133 _symbols_counted += *processed;
134 // Exclude printing for normal PrintGCDetails because people parse
135 // this output.
136 if (PrintGCDetails && Verbose && WizardMode) {
137 gclog_or_tty->print(" [Symbols=%d size=" SIZE_FORMAT "K] ", *processed,
138 (memory_total*HeapWordSize)/1024);
139 }
140 }
141
142 void SymbolTable::possibly_parallel_unlink(int* processed, int* removed) {
143 const int limit = the_table()->table_size();
144
145 size_t memory_total = 0;
146
147 for (;;) {
148 // Grab next set of buckets to scan
149 int start_idx = Atomic::add(ClaimChunkSize, &_parallel_claimed_idx) - ClaimChunkSize;
150 if (start_idx >= limit) {
151 // End of table
152 break;
153 }
154
155 int end_idx = MIN2(limit, start_idx + ClaimChunkSize);
156 buckets_unlink(start_idx, end_idx, processed, removed, &memory_total);
157 }
158 Atomic::add(*processed, &_symbols_counted);
159 Atomic::add(*removed, &_symbols_removed);
160 // Exclude printing for normal PrintGCDetails because people parse
161 // this output.
162 if (PrintGCDetails && Verbose && WizardMode) {
163 gclog_or_tty->print(" [Symbols: scanned=%d removed=%d size=" SIZE_FORMAT "K] ", *processed, *removed,
164 (memory_total*HeapWordSize)/1024);
165 }
166 }
167
168 // Create a new table and using alternate hash code, populate the new table
169 // with the existing strings. Set flag to use the alternate hash code afterwards.
170 void SymbolTable::rehash_table() {
171 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
172 // This should never happen with -Xshare:dump but it might in testing mode.
173 if (DumpSharedSpaces) return;
174 // Create a new symbol table
175 SymbolTable* new_table = new SymbolTable();
176
177 the_table()->move_to(new_table);
178
179 // Delete the table and buckets (entries are reused in new table).
180 delete _the_table;
181 // Don't check if we need rehashing until the table gets unbalanced again.
182 // Then rehash with a new global seed.
183 _needs_rehashing = false;
184 _the_table = new_table;
185 }
186
187 // Lookup a symbol in a bucket.
188
189 Symbol* SymbolTable::lookup(int index, const char* name,
190 int len, unsigned int hash) {
191 int count = 0;
192 for (HashtableEntry<Symbol*, mtSymbol>* e = bucket(index); e != NULL; e = e->next()) {
193 count++; // count all entries in this bucket, not just ones with same hash
194 if (e->hash() == hash) {
195 Symbol* sym = e->literal();
196 if (sym->equals(name, len)) {
197 // something is referencing this symbol now.
198 sym->increment_refcount();
199 return sym;
200 }
201 }
202 }
203 // If the bucket size is too deep check if this hash code is insufficient.
204 if (count >= rehash_count && !needs_rehashing()) {
205 _needs_rehashing = check_rehash_table(count);
206 }
207 return NULL;
208 }
209
210 // Pick hashing algorithm.
211 unsigned int SymbolTable::hash_symbol(const char* s, int len) {
212 return use_alternate_hashcode() ?
213 AltHashing::murmur3_32(seed(), (const jbyte*)s, len) :
214 java_lang_String::hash_code(s, len);
215 }
216
217
218 // We take care not to be blocking while holding the
219 // SymbolTable_lock. Otherwise, the system might deadlock, since the
220 // symboltable is used during compilation (VM_thread) The lock free
221 // synchronization is simplified by the fact that we do not delete
222 // entries in the symbol table during normal execution (only during
223 // safepoints).
224
225 Symbol* SymbolTable::lookup(const char* name, int len, TRAPS) {
226 unsigned int hashValue = hash_symbol(name, len);
227 int index = the_table()->hash_to_index(hashValue);
228
229 Symbol* s = the_table()->lookup(index, name, len, hashValue);
230
231 // Found
232 if (s != NULL) return s;
233
234 // Grab SymbolTable_lock first.
235 MutexLocker ml(SymbolTable_lock, THREAD);
236
237 // Otherwise, add to symbol to table
238 return the_table()->basic_add(index, (u1*)name, len, hashValue, true, CHECK_NULL);
239 }
240
241 Symbol* SymbolTable::lookup(const Symbol* sym, int begin, int end, TRAPS) {
242 char* buffer;
243 int index, len;
244 unsigned int hashValue;
245 char* name;
246 {
247 debug_only(No_Safepoint_Verifier nsv;)
248
249 name = (char*)sym->base() + begin;
250 len = end - begin;
251 hashValue = hash_symbol(name, len);
252 index = the_table()->hash_to_index(hashValue);
253 Symbol* s = the_table()->lookup(index, name, len, hashValue);
254
255 // Found
256 if (s != NULL) return s;
257 }
258
259 // Otherwise, add to symbol to table. Copy to a C string first.
260 char stack_buf[128];
261 ResourceMark rm(THREAD);
262 if (len <= 128) {
263 buffer = stack_buf;
264 } else {
265 buffer = NEW_RESOURCE_ARRAY_IN_THREAD(THREAD, char, len);
266 }
267 for (int i=0; i<len; i++) {
268 buffer[i] = name[i];
269 }
270 // Make sure there is no safepoint in the code above since name can't move.
271 // We can't include the code in No_Safepoint_Verifier because of the
272 // ResourceMark.
273
274 // Grab SymbolTable_lock first.
275 MutexLocker ml(SymbolTable_lock, THREAD);
276
277 return the_table()->basic_add(index, (u1*)buffer, len, hashValue, true, CHECK_NULL);
278 }
279
280 Symbol* SymbolTable::lookup_only(const char* name, int len,
281 unsigned int& hash) {
282 hash = hash_symbol(name, len);
283 int index = the_table()->hash_to_index(hash);
284
285 Symbol* s = the_table()->lookup(index, name, len, hash);
286 return s;
287 }
288
289 // Look up the address of the literal in the SymbolTable for this Symbol*
290 // Do not create any new symbols
291 // Do not increment the reference count to keep this alive
292 Symbol** SymbolTable::lookup_symbol_addr(Symbol* sym){
293 unsigned int hash = hash_symbol((char*)sym->bytes(), sym->utf8_length());
294 int index = the_table()->hash_to_index(hash);
295
296 for (HashtableEntry<Symbol*, mtSymbol>* e = the_table()->bucket(index); e != NULL; e = e->next()) {
297 if (e->hash() == hash) {
298 Symbol* literal_sym = e->literal();
299 if (sym == literal_sym) {
300 return e->literal_addr();
301 }
302 }
303 }
304 return NULL;
305 }
306
307 // Suggestion: Push unicode-based lookup all the way into the hashing
308 // and probing logic, so there is no need for convert_to_utf8 until
309 // an actual new Symbol* is created.
310 Symbol* SymbolTable::lookup_unicode(const jchar* name, int utf16_length, TRAPS) {
311 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
312 char stack_buf[128];
313 if (utf8_length < (int) sizeof(stack_buf)) {
314 char* chars = stack_buf;
315 UNICODE::convert_to_utf8(name, utf16_length, chars);
316 return lookup(chars, utf8_length, THREAD);
317 } else {
318 ResourceMark rm(THREAD);
319 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
320 UNICODE::convert_to_utf8(name, utf16_length, chars);
321 return lookup(chars, utf8_length, THREAD);
322 }
323 }
324
325 Symbol* SymbolTable::lookup_only_unicode(const jchar* name, int utf16_length,
326 unsigned int& hash) {
327 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
328 char stack_buf[128];
329 if (utf8_length < (int) sizeof(stack_buf)) {
330 char* chars = stack_buf;
331 UNICODE::convert_to_utf8(name, utf16_length, chars);
332 return lookup_only(chars, utf8_length, hash);
333 } else {
334 ResourceMark rm;
335 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
336 UNICODE::convert_to_utf8(name, utf16_length, chars);
337 return lookup_only(chars, utf8_length, hash);
338 }
339 }
340
341 void SymbolTable::add(ClassLoaderData* loader_data, constantPoolHandle cp,
342 int names_count,
343 const char** names, int* lengths, int* cp_indices,
344 unsigned int* hashValues, TRAPS) {
345 // Grab SymbolTable_lock first.
346 MutexLocker ml(SymbolTable_lock, THREAD);
347
348 SymbolTable* table = the_table();
349 bool added = table->basic_add(loader_data, cp, names_count, names, lengths,
350 cp_indices, hashValues, CHECK);
351 if (!added) {
352 // do it the hard way
353 for (int i=0; i<names_count; i++) {
354 int index = table->hash_to_index(hashValues[i]);
355 bool c_heap = !loader_data->is_the_null_class_loader_data();
356 Symbol* sym = table->basic_add(index, (u1*)names[i], lengths[i], hashValues[i], c_heap, CHECK);
357 cp->symbol_at_put(cp_indices[i], sym);
358 }
359 }
360 }
361
362 Symbol* SymbolTable::new_permanent_symbol(const char* name, TRAPS) {
363 unsigned int hash;
364 Symbol* result = SymbolTable::lookup_only((char*)name, (int)strlen(name), hash);
365 if (result != NULL) {
366 return result;
367 }
368 // Grab SymbolTable_lock first.
369 MutexLocker ml(SymbolTable_lock, THREAD);
370
371 SymbolTable* table = the_table();
372 int index = table->hash_to_index(hash);
373 return table->basic_add(index, (u1*)name, (int)strlen(name), hash, false, THREAD);
374 }
375
376 Symbol* SymbolTable::basic_add(int index_arg, u1 *name, int len,
377 unsigned int hashValue_arg, bool c_heap, TRAPS) {
378 assert(!Universe::heap()->is_in_reserved(name),
379 "proposed name of symbol must be stable");
380
381 // Don't allow symbols to be created which cannot fit in a Symbol*.
382 if (len > Symbol::max_length()) {
383 THROW_MSG_0(vmSymbols::java_lang_InternalError(),
384 "name is too long to represent");
385 }
386
387 // Cannot hit a safepoint in this function because the "this" pointer can move.
388 No_Safepoint_Verifier nsv;
389
390 // Check if the symbol table has been rehashed, if so, need to recalculate
391 // the hash value and index.
392 unsigned int hashValue;
393 int index;
394 if (use_alternate_hashcode()) {
395 hashValue = hash_symbol((const char*)name, len);
396 index = hash_to_index(hashValue);
397 } else {
398 hashValue = hashValue_arg;
399 index = index_arg;
400 }
401
402 // Since look-up was done lock-free, we need to check if another
403 // thread beat us in the race to insert the symbol.
404 Symbol* test = lookup(index, (char*)name, len, hashValue);
405 if (test != NULL) {
406 // A race occurred and another thread introduced the symbol.
407 assert(test->refcount() != 0, "lookup should have incremented the count");
408 return test;
409 }
410
411 // Create a new symbol.
412 Symbol* sym = allocate_symbol(name, len, c_heap, CHECK_NULL);
413 assert(sym->equals((char*)name, len), "symbol must be properly initialized");
414
415 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
416 add_entry(index, entry);
417 return sym;
418 }
419
420 // This version of basic_add adds symbols in batch from the constant pool
421 // parsing.
422 bool SymbolTable::basic_add(ClassLoaderData* loader_data, constantPoolHandle cp,
423 int names_count,
424 const char** names, int* lengths,
425 int* cp_indices, unsigned int* hashValues,
426 TRAPS) {
427
428 // Check symbol names are not too long. If any are too long, don't add any.
429 for (int i = 0; i< names_count; i++) {
430 if (lengths[i] > Symbol::max_length()) {
431 THROW_MSG_0(vmSymbols::java_lang_InternalError(),
432 "name is too long to represent");
433 }
434 }
435
436 // Cannot hit a safepoint in this function because the "this" pointer can move.
437 No_Safepoint_Verifier nsv;
438
439 for (int i=0; i<names_count; i++) {
440 // Check if the symbol table has been rehashed, if so, need to recalculate
441 // the hash value.
442 unsigned int hashValue;
443 if (use_alternate_hashcode()) {
444 hashValue = hash_symbol(names[i], lengths[i]);
445 } else {
446 hashValue = hashValues[i];
447 }
448 // Since look-up was done lock-free, we need to check if another
449 // thread beat us in the race to insert the symbol.
450 int index = hash_to_index(hashValue);
451 Symbol* test = lookup(index, names[i], lengths[i], hashValue);
452 if (test != NULL) {
453 // A race occurred and another thread introduced the symbol, this one
454 // will be dropped and collected. Use test instead.
455 cp->symbol_at_put(cp_indices[i], test);
456 assert(test->refcount() != 0, "lookup should have incremented the count");
457 } else {
458 // Create a new symbol. The null class loader is never unloaded so these
459 // are allocated specially in a permanent arena.
460 bool c_heap = !loader_data->is_the_null_class_loader_data();
461 Symbol* sym = allocate_symbol((const u1*)names[i], lengths[i], c_heap, CHECK_(false));
462 assert(sym->equals(names[i], lengths[i]), "symbol must be properly initialized"); // why wouldn't it be???
463 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
464 add_entry(index, entry);
465 cp->symbol_at_put(cp_indices[i], sym);
466 }
467 }
468 return true;
469 }
470
471
472 void SymbolTable::verify() {
473 for (int i = 0; i < the_table()->table_size(); ++i) {
474 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
475 for ( ; p != NULL; p = p->next()) {
476 Symbol* s = (Symbol*)(p->literal());
477 guarantee(s != NULL, "symbol is NULL");
478 unsigned int h = hash_symbol((char*)s->bytes(), s->utf8_length());
479 guarantee(p->hash() == h, "broken hash in symbol table entry");
480 guarantee(the_table()->hash_to_index(h) == i,
481 "wrong index in symbol table");
482 }
483 }
484 }
485
486 void SymbolTable::dump(outputStream* st) {
487 the_table()->dump_table(st, "SymbolTable");
488 }
489
490
491 //---------------------------------------------------------------------------
492 // Non-product code
493
494 #ifndef PRODUCT
495
496 void SymbolTable::print_histogram() {
497 MutexLocker ml(SymbolTable_lock);
498 const int results_length = 100;
499 int counts[results_length];
500 int sizes[results_length];
501 int i,j;
502
503 // initialize results to zero
504 for (j = 0; j < results_length; j++) {
505 counts[j] = 0;
506 sizes[j] = 0;
507 }
508
509 int total_size = 0;
510 int total_count = 0;
511 int total_length = 0;
512 int max_length = 0;
513 int out_of_range_count = 0;
514 int out_of_range_size = 0;
515 for (i = 0; i < the_table()->table_size(); i++) {
516 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
517 for ( ; p != NULL; p = p->next()) {
518 int size = p->literal()->size();
519 int len = p->literal()->utf8_length();
520 if (len < results_length) {
521 counts[len]++;
522 sizes[len] += size;
523 } else {
524 out_of_range_count++;
525 out_of_range_size += size;
526 }
527 total_count++;
528 total_size += size;
529 total_length += len;
530 max_length = MAX2(max_length, len);
531 }
532 }
533 tty->print_cr("Symbol Table Histogram:");
534 tty->print_cr(" Total number of symbols %7d", total_count);
535 tty->print_cr(" Total size in memory %7dK",
536 (total_size*HeapWordSize)/1024);
537 tty->print_cr(" Total counted %7d", _symbols_counted);
538 tty->print_cr(" Total removed %7d", _symbols_removed);
539 if (_symbols_counted > 0) {
540 tty->print_cr(" Percent removed %3.2f",
541 ((float)_symbols_removed/(float)_symbols_counted)* 100);
542 }
543 tty->print_cr(" Reference counts %7d", Symbol::_total_count);
544 tty->print_cr(" Symbol arena used %7dK", arena()->used()/1024);
545 tty->print_cr(" Symbol arena size %7dK", arena()->size_in_bytes()/1024);
546 tty->print_cr(" Total symbol length %7d", total_length);
547 tty->print_cr(" Maximum symbol length %7d", max_length);
548 tty->print_cr(" Average symbol length %7.2f", ((float) total_length / (float) total_count));
549 tty->print_cr(" Symbol length histogram:");
550 tty->print_cr(" %6s %10s %10s", "Length", "#Symbols", "Size");
551 for (i = 0; i < results_length; i++) {
552 if (counts[i] > 0) {
553 tty->print_cr(" %6d %10d %10dK", i, counts[i], (sizes[i]*HeapWordSize)/1024);
554 }
555 }
556 tty->print_cr(" >=%6d %10d %10dK\n", results_length,
557 out_of_range_count, (out_of_range_size*HeapWordSize)/1024);
558 }
559
560 void SymbolTable::print() {
561 for (int i = 0; i < the_table()->table_size(); ++i) {
562 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
563 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
564 if (entry != NULL) {
565 while (entry != NULL) {
566 tty->print(PTR_FORMAT " ", entry->literal());
567 entry->literal()->print();
568 tty->print(" %d", entry->literal()->refcount());
569 p = entry->next_addr();
570 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
571 }
572 tty->cr();
573 }
574 }
575 }
576 #endif // PRODUCT