1 /*
2 * Copyright (c) 2019, 2025, 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 "cds/archiveBuilder.hpp"
26 #include "cds/archiveHeapLoader.inline.hpp"
27 #include "cds/archiveUtils.hpp"
28 #include "cds/cdsConfig.hpp"
29 #include "cds/classListParser.hpp"
30 #include "cds/classListWriter.hpp"
31 #include "cds/dynamicArchive.hpp"
32 #include "cds/filemap.hpp"
33 #include "cds/heapShared.hpp"
34 #include "cds/lambdaProxyClassDictionary.hpp"
35 #include "cds/metaspaceShared.hpp"
36 #include "classfile/systemDictionaryShared.hpp"
37 #include "classfile/vmClasses.hpp"
38 #include "interpreter/bootstrapInfo.hpp"
39 #include "memory/metaspaceUtils.hpp"
40 #include "memory/resourceArea.hpp"
41 #include "oops/compressedOops.inline.hpp"
42 #include "oops/klass.inline.hpp"
43 #include "runtime/arguments.hpp"
44 #include "utilities/bitMap.inline.hpp"
45 #include "utilities/debug.hpp"
46 #include "utilities/formatBuffer.hpp"
47 #include "utilities/globalDefinitions.hpp"
48 #include "utilities/spinYield.hpp"
49
50 CHeapBitMap* ArchivePtrMarker::_ptrmap = nullptr;
51 CHeapBitMap* ArchivePtrMarker::_rw_ptrmap = nullptr;
52 CHeapBitMap* ArchivePtrMarker::_ro_ptrmap = nullptr;
53 VirtualSpace* ArchivePtrMarker::_vs;
54
55 bool ArchivePtrMarker::_compacted;
56
57 void ArchivePtrMarker::initialize(CHeapBitMap* ptrmap, VirtualSpace* vs) {
58 assert(_ptrmap == nullptr, "initialize only once");
59 assert(_rw_ptrmap == nullptr, "initialize only once");
60 assert(_ro_ptrmap == nullptr, "initialize only once");
61 _vs = vs;
62 _compacted = false;
63 _ptrmap = ptrmap;
64
65 // Use this as initial guesstimate. We should need less space in the
66 // archive, but if we're wrong the bitmap will be expanded automatically.
67 size_t estimated_archive_size = MetaspaceGC::capacity_until_GC();
68 // But set it smaller in debug builds so we always test the expansion code.
69 // (Default archive is about 12MB).
70 DEBUG_ONLY(estimated_archive_size = 6 * M);
71
72 // We need one bit per pointer in the archive.
73 _ptrmap->initialize(estimated_archive_size / sizeof(intptr_t));
74 }
75
76 void ArchivePtrMarker::initialize_rw_ro_maps(CHeapBitMap* rw_ptrmap, CHeapBitMap* ro_ptrmap) {
77 address* buff_bottom = (address*)ArchiveBuilder::current()->buffer_bottom();
78 address* rw_bottom = (address*)ArchiveBuilder::current()->rw_region()->base();
79 address* ro_bottom = (address*)ArchiveBuilder::current()->ro_region()->base();
80
81 // The bit in _ptrmap that cover the very first word in the rw/ro regions.
82 size_t rw_start = rw_bottom - buff_bottom;
83 size_t ro_start = ro_bottom - buff_bottom;
84
85 // The number of bits used by the rw/ro ptrmaps. We might have lots of zero
86 // bits at the bottom and top of rw/ro ptrmaps, but these zeros will be
87 // removed by FileMapInfo::write_bitmap_region().
88 size_t rw_size = ArchiveBuilder::current()->rw_region()->used() / sizeof(address);
89 size_t ro_size = ArchiveBuilder::current()->ro_region()->used() / sizeof(address);
90
91 // The last (exclusive) bit in _ptrmap that covers the rw/ro regions.
92 // Note: _ptrmap is dynamically expanded only when an actual pointer is written, so
93 // it may not be as large as we want.
94 size_t rw_end = MIN2<size_t>(rw_start + rw_size, _ptrmap->size());
95 size_t ro_end = MIN2<size_t>(ro_start + ro_size, _ptrmap->size());
96
97 rw_ptrmap->initialize(rw_size);
98 ro_ptrmap->initialize(ro_size);
99
100 for (size_t rw_bit = rw_start; rw_bit < rw_end; rw_bit++) {
101 rw_ptrmap->at_put(rw_bit - rw_start, _ptrmap->at(rw_bit));
102 }
103
104 for(size_t ro_bit = ro_start; ro_bit < ro_end; ro_bit++) {
105 ro_ptrmap->at_put(ro_bit - ro_start, _ptrmap->at(ro_bit));
106 }
107
108 _rw_ptrmap = rw_ptrmap;
109 _ro_ptrmap = ro_ptrmap;
110 }
111
112 void ArchivePtrMarker::mark_pointer(address* ptr_loc) {
113 assert(_ptrmap != nullptr, "not initialized");
114 assert(!_compacted, "cannot mark anymore");
115
116 if (ptr_base() <= ptr_loc && ptr_loc < ptr_end()) {
117 address value = *ptr_loc;
118 // We don't want any pointer that points to very bottom of the archive, otherwise when
119 // MetaspaceShared::default_base_address()==0, we can't distinguish between a pointer
120 // to nothing (null) vs a pointer to an objects that happens to be at the very bottom
121 // of the archive.
122 assert(value != (address)ptr_base(), "don't point to the bottom of the archive");
123
124 if (value != nullptr) {
125 assert(uintx(ptr_loc) % sizeof(intptr_t) == 0, "pointers must be stored in aligned addresses");
126 size_t idx = ptr_loc - ptr_base();
127 if (_ptrmap->size() <= idx) {
128 _ptrmap->resize((idx + 1) * 2);
129 }
130 assert(idx < _ptrmap->size(), "must be");
131 _ptrmap->set_bit(idx);
132 //tty->print_cr("Marking pointer [" PTR_FORMAT "] -> " PTR_FORMAT " @ %5zu", p2i(ptr_loc), p2i(*ptr_loc), idx);
133 }
134 }
135 }
136
137 void ArchivePtrMarker::clear_pointer(address* ptr_loc) {
138 assert(_ptrmap != nullptr, "not initialized");
139 assert(!_compacted, "cannot clear anymore");
140
141 assert(ptr_base() <= ptr_loc && ptr_loc < ptr_end(), "must be");
142 assert(uintx(ptr_loc) % sizeof(intptr_t) == 0, "pointers must be stored in aligned addresses");
143 size_t idx = ptr_loc - ptr_base();
144 assert(idx < _ptrmap->size(), "cannot clear pointers that have not been marked");
145 _ptrmap->clear_bit(idx);
146 //tty->print_cr("Clearing pointer [" PTR_FORMAT "] -> " PTR_FORMAT " @ %5zu", p2i(ptr_loc), p2i(*ptr_loc), idx);
147 }
148
149 class ArchivePtrBitmapCleaner: public BitMapClosure {
150 CHeapBitMap* _ptrmap;
151 address* _ptr_base;
152 address _relocatable_base;
153 address _relocatable_end;
154 size_t _max_non_null_offset;
155
156 public:
157 ArchivePtrBitmapCleaner(CHeapBitMap* ptrmap, address* ptr_base, address relocatable_base, address relocatable_end) :
158 _ptrmap(ptrmap), _ptr_base(ptr_base),
159 _relocatable_base(relocatable_base), _relocatable_end(relocatable_end), _max_non_null_offset(0) {}
160
161 bool do_bit(size_t offset) {
162 address* ptr_loc = _ptr_base + offset;
163 address ptr_value = *ptr_loc;
164 if (ptr_value != nullptr) {
165 assert(_relocatable_base <= ptr_value && ptr_value < _relocatable_end, "do not point to arbitrary locations!");
166 if (_max_non_null_offset < offset) {
167 _max_non_null_offset = offset;
168 }
169 } else {
170 _ptrmap->clear_bit(offset);
171 DEBUG_ONLY(log_trace(cds, reloc)("Clearing pointer [" PTR_FORMAT "] -> null @ %9zu", p2i(ptr_loc), offset));
172 }
173
174 return true;
175 }
176
177 size_t max_non_null_offset() const { return _max_non_null_offset; }
178 };
179
180 void ArchivePtrMarker::compact(address relocatable_base, address relocatable_end) {
181 assert(!_compacted, "cannot compact again");
182 ArchivePtrBitmapCleaner cleaner(_ptrmap, ptr_base(), relocatable_base, relocatable_end);
183 _ptrmap->iterate(&cleaner);
184 compact(cleaner.max_non_null_offset());
185 }
186
187 void ArchivePtrMarker::compact(size_t max_non_null_offset) {
188 assert(!_compacted, "cannot compact again");
189 _ptrmap->resize(max_non_null_offset + 1);
190 _compacted = true;
191 }
192
193 char* DumpRegion::expand_top_to(char* newtop) {
194 assert(is_allocatable(), "must be initialized and not packed");
195 assert(newtop >= _top, "must not grow backwards");
196 if (newtop > _end) {
197 ArchiveBuilder::current()->report_out_of_space(_name, newtop - _top);
198 ShouldNotReachHere();
199 }
200
201 commit_to(newtop);
202 _top = newtop;
203
204 if (_max_delta > 0) {
205 uintx delta = ArchiveBuilder::current()->buffer_to_offset((address)(newtop-1));
206 if (delta > _max_delta) {
207 // This is just a sanity check and should not appear in any real world usage. This
208 // happens only if you allocate more than 2GB of shared objects and would require
209 // millions of shared classes.
210 log_error(cds)("Out of memory in the CDS archive: Please reduce the number of shared classes.");
211 MetaspaceShared::unrecoverable_writing_error();
212 }
213 }
214
215 return _top;
216 }
217
218 void DumpRegion::commit_to(char* newtop) {
219 assert(CDSConfig::is_dumping_archive(), "sanity");
220 char* base = _rs->base();
221 size_t need_committed_size = newtop - base;
222 size_t has_committed_size = _vs->committed_size();
223 if (need_committed_size < has_committed_size) {
224 return;
225 }
226
227 size_t min_bytes = need_committed_size - has_committed_size;
228 size_t preferred_bytes = 1 * M;
229 size_t uncommitted = _vs->reserved_size() - has_committed_size;
230
231 size_t commit = MAX2(min_bytes, preferred_bytes);
232 commit = MIN2(commit, uncommitted);
233 assert(commit <= uncommitted, "sanity");
234
235 if (!_vs->expand_by(commit, false)) {
236 log_error(cds)("Failed to expand shared space to %zu bytes",
237 need_committed_size);
238 MetaspaceShared::unrecoverable_writing_error();
239 }
240
241 const char* which;
242 if (_rs->base() == (char*)MetaspaceShared::symbol_rs_base()) {
243 which = "symbol";
244 } else {
245 which = "shared";
246 }
247 log_debug(cds)("Expanding %s spaces by %7zu bytes [total %9zu bytes ending at %p]",
248 which, commit, _vs->actual_committed_size(), _vs->high());
249 }
250
251 char* DumpRegion::allocate(size_t num_bytes, size_t alignment) {
252 // Always align to at least minimum alignment
253 alignment = MAX2(SharedSpaceObjectAlignment, alignment);
254 char* p = (char*)align_up(_top, alignment);
255 char* newtop = p + align_up(num_bytes, (size_t)SharedSpaceObjectAlignment);
256 expand_top_to(newtop);
257 memset(p, 0, newtop - p);
258 return p;
259 }
260
261 void DumpRegion::append_intptr_t(intptr_t n, bool need_to_mark) {
262 assert(is_aligned(_top, sizeof(intptr_t)), "bad alignment");
263 intptr_t *p = (intptr_t*)_top;
264 char* newtop = _top + sizeof(intptr_t);
265 expand_top_to(newtop);
266 *p = n;
267 if (need_to_mark) {
268 ArchivePtrMarker::mark_pointer(p);
269 }
270 }
271
272 void DumpRegion::print(size_t total_bytes) const {
273 log_debug(cds)("%s space: %9zu [ %4.1f%% of total] out of %9zu bytes [%5.1f%% used] at " INTPTR_FORMAT,
274 _name, used(), percent_of(used(), total_bytes), reserved(), percent_of(used(), reserved()),
275 p2i(ArchiveBuilder::current()->to_requested(_base)));
276 }
277
278 void DumpRegion::print_out_of_space_msg(const char* failing_region, size_t needed_bytes) {
279 log_error(cds)("[%-8s] " PTR_FORMAT " - " PTR_FORMAT " capacity =%9d, allocated =%9d",
280 _name, p2i(_base), p2i(_top), int(_end - _base), int(_top - _base));
281 if (strcmp(_name, failing_region) == 0) {
282 log_error(cds)(" required = %d", int(needed_bytes));
283 }
284 }
285
286 void DumpRegion::init(ReservedSpace* rs, VirtualSpace* vs) {
287 _rs = rs;
288 _vs = vs;
289 // Start with 0 committed bytes. The memory will be committed as needed.
290 if (!_vs->initialize(*_rs, 0)) {
291 fatal("Unable to allocate memory for shared space");
292 }
293 _base = _top = _rs->base();
294 _end = _rs->end();
295 }
296
297 void DumpRegion::pack(DumpRegion* next) {
298 assert(!is_packed(), "sanity");
299 _end = (char*)align_up(_top, MetaspaceShared::core_region_alignment());
300 _is_packed = true;
301 if (next != nullptr) {
302 next->_rs = _rs;
303 next->_vs = _vs;
304 next->_base = next->_top = this->_end;
305 next->_end = _rs->end();
306 }
307 }
308
309 void WriteClosure::do_ptr(void** p) {
310 // Write ptr into the archive; ptr can be:
311 // (a) null -> written as 0
312 // (b) a "buffered" address -> written as is
313 // (c) a "source" address -> convert to "buffered" and write
314 // The common case is (c). E.g., when writing the vmClasses into the archive.
315 // We have (b) only when we don't have a corresponding source object. E.g.,
316 // the archived c++ vtable entries.
317 address ptr = *(address*)p;
318 if (ptr != nullptr && !ArchiveBuilder::current()->is_in_buffer_space(ptr)) {
319 ptr = ArchiveBuilder::current()->get_buffered_addr(ptr);
320 }
321 // null pointers do not need to be converted to offsets
322 if (ptr != nullptr) {
323 ptr = (address)ArchiveBuilder::current()->buffer_to_offset(ptr);
324 }
325 _dump_region->append_intptr_t((intptr_t)ptr, false);
326 }
327
328 void ReadClosure::do_ptr(void** p) {
329 assert(*p == nullptr, "initializing previous initialized pointer.");
330 intptr_t obj = nextPtr();
331 assert(obj >= 0, "sanity.");
332 *p = (obj != 0) ? (void*)(_base_address + obj) : (void*)obj;
333 }
334
335 void ReadClosure::do_u4(u4* p) {
336 intptr_t obj = nextPtr();
337 *p = (u4)(uintx(obj));
338 }
339
340 void ReadClosure::do_int(int* p) {
341 intptr_t obj = nextPtr();
342 *p = (int)(intx(obj));
343 }
344
345 void ReadClosure::do_bool(bool* p) {
346 intptr_t obj = nextPtr();
347 *p = (bool)(uintx(obj));
348 }
349
350 void ReadClosure::do_tag(int tag) {
351 int old_tag;
352 old_tag = (int)(intptr_t)nextPtr();
353 // do_int(&old_tag);
354 assert(tag == old_tag, "tag doesn't match (%d, expected %d)", old_tag, tag);
355 FileMapInfo::assert_mark(tag == old_tag);
356 }
357
358 void ArchiveUtils::log_to_classlist(BootstrapInfo* bootstrap_specifier, TRAPS) {
359 if (ClassListWriter::is_enabled()) {
360 if (LambdaProxyClassDictionary::is_supported_invokedynamic(bootstrap_specifier)) {
361 const constantPoolHandle& pool = bootstrap_specifier->pool();
362 if (SystemDictionaryShared::is_builtin_loader(pool->pool_holder()->class_loader_data())) {
363 // Currently lambda proxy classes are supported only for the built-in loaders.
364 ResourceMark rm(THREAD);
365 int pool_index = bootstrap_specifier->bss_index();
366 ClassListWriter w;
367 w.stream()->print("%s %s", ClassListParser::lambda_proxy_tag(), pool->pool_holder()->name()->as_C_string());
368 CDSIndyInfo cii;
369 ClassListParser::populate_cds_indy_info(pool, pool_index, &cii, CHECK);
370 GrowableArray<const char*>* indy_items = cii.items();
371 for (int i = 0; i < indy_items->length(); i++) {
372 w.stream()->print(" %s", indy_items->at(i));
373 }
374 w.stream()->cr();
375 }
376 }
377 }
378 }
379
380 bool ArchiveUtils::has_aot_initialized_mirror(InstanceKlass* src_ik) {
381 if (SystemDictionaryShared::is_excluded_class(src_ik)) {
382 assert(!ArchiveBuilder::current()->has_been_buffered(src_ik), "sanity");
383 return false;
384 }
385 return ArchiveBuilder::current()->get_buffered_addr(src_ik)->has_aot_initialized_mirror();
386 }
387
388 size_t HeapRootSegments::size_in_bytes(size_t seg_idx) {
389 assert(seg_idx < _count, "In range");
390 return objArrayOopDesc::object_size(size_in_elems(seg_idx)) * HeapWordSize;
391 }
392
393 int HeapRootSegments::size_in_elems(size_t seg_idx) {
394 assert(seg_idx < _count, "In range");
395 if (seg_idx != _count - 1) {
396 return _max_size_in_elems;
397 } else {
398 // Last slice, leftover
399 return _roots_count % _max_size_in_elems;
400 }
401 }
402
403 size_t HeapRootSegments::segment_offset(size_t seg_idx) {
404 assert(seg_idx < _count, "In range");
405 return _base_offset + seg_idx * _max_size_in_bytes;
406 }
407
408 ArchiveWorkers::ArchiveWorkers() :
409 _end_semaphore(0),
410 _num_workers(max_workers()),
411 _started_workers(0),
412 _finish_tokens(0),
413 _state(UNUSED),
414 _task(nullptr) {}
415
416 ArchiveWorkers::~ArchiveWorkers() {
417 assert(Atomic::load(&_state) != WORKING, "Should not be working");
418 }
419
420 int ArchiveWorkers::max_workers() {
421 // The pool is used for short-lived bursty tasks. We do not want to spend
422 // too much time creating and waking up threads unnecessarily. Plus, we do
423 // not want to overwhelm large machines. This is why we want to be very
424 // conservative about the number of workers actually needed.
425 return MAX2(0, log2i_graceful(os::active_processor_count()));
426 }
427
428 bool ArchiveWorkers::is_parallel() {
429 return _num_workers > 0;
430 }
431
432 void ArchiveWorkers::start_worker_if_needed() {
433 while (true) {
434 int cur = Atomic::load(&_started_workers);
435 if (cur >= _num_workers) {
436 return;
437 }
438 if (Atomic::cmpxchg(&_started_workers, cur, cur + 1, memory_order_relaxed) == cur) {
439 new ArchiveWorkerThread(this);
440 return;
441 }
442 }
443 }
444
445 void ArchiveWorkers::run_task(ArchiveWorkerTask* task) {
446 assert(Atomic::load(&_state) == UNUSED, "Should be unused yet");
447 assert(Atomic::load(&_task) == nullptr, "Should not have running tasks");
448 Atomic::store(&_state, WORKING);
449
450 if (is_parallel()) {
451 run_task_multi(task);
452 } else {
453 run_task_single(task);
454 }
455
456 assert(Atomic::load(&_state) == WORKING, "Should be working");
457 Atomic::store(&_state, SHUTDOWN);
458 }
459
460 void ArchiveWorkers::run_task_single(ArchiveWorkerTask* task) {
461 // Single thread needs no chunking.
462 task->configure_max_chunks(1);
463
464 // Execute the task ourselves, as there are no workers.
465 task->work(0, 1);
466 }
467
468 void ArchiveWorkers::run_task_multi(ArchiveWorkerTask* task) {
469 // Multiple threads can work with multiple chunks.
470 task->configure_max_chunks(_num_workers * CHUNKS_PER_WORKER);
471
472 // Set up the run and publish the task. Issue one additional finish token
473 // to cover the semaphore shutdown path, see below.
474 Atomic::store(&_finish_tokens, _num_workers + 1);
475 Atomic::release_store(&_task, task);
476
477 // Kick off pool startup by starting a single worker, and proceed
478 // immediately to executing the task locally.
479 start_worker_if_needed();
480
481 // Execute the task ourselves, while workers are catching up.
482 // This allows us to hide parts of task handoff latency.
483 task->run();
484
485 // Done executing task locally, wait for any remaining workers to complete.
486 // Once all workers report, we can proceed to termination. To do this safely,
487 // we need to make sure every worker has left. A spin-wait alone would suffice,
488 // but we do not want to burn cycles on it. A semaphore alone would not be safe,
489 // since workers can still be inside it as we proceed from wait here. So we block
490 // on semaphore first, and then spin-wait for all workers to terminate.
491 _end_semaphore.wait();
492 SpinYield spin;
493 while (Atomic::load(&_finish_tokens) != 0) {
494 spin.wait();
495 }
496
497 OrderAccess::fence();
498
499 assert(Atomic::load(&_finish_tokens) == 0, "All tokens are consumed");
500 }
501
502 void ArchiveWorkers::run_as_worker() {
503 assert(is_parallel(), "Should be in parallel mode");
504
505 ArchiveWorkerTask* task = Atomic::load_acquire(&_task);
506 task->run();
507
508 // All work done in threads should be visible to caller.
509 OrderAccess::fence();
510
511 // Signal the pool the work is complete, and we are exiting.
512 // Worker cannot do anything else with the pool after this.
513 if (Atomic::sub(&_finish_tokens, 1, memory_order_relaxed) == 1) {
514 // Last worker leaving. Notify the pool it can unblock to spin-wait.
515 // Then consume the last token and leave.
516 _end_semaphore.signal();
517 int last = Atomic::sub(&_finish_tokens, 1, memory_order_relaxed);
518 assert(last == 0, "Should be");
519 }
520 }
521
522 void ArchiveWorkerTask::run() {
523 while (true) {
524 int chunk = Atomic::load(&_chunk);
525 if (chunk >= _max_chunks) {
526 return;
527 }
528 if (Atomic::cmpxchg(&_chunk, chunk, chunk + 1, memory_order_relaxed) == chunk) {
529 assert(0 <= chunk && chunk < _max_chunks, "Sanity");
530 work(chunk, _max_chunks);
531 }
532 }
533 }
534
535 void ArchiveWorkerTask::configure_max_chunks(int max_chunks) {
536 if (_max_chunks == 0) {
537 _max_chunks = max_chunks;
538 }
539 }
540
541 ArchiveWorkerThread::ArchiveWorkerThread(ArchiveWorkers* pool) : NamedThread(), _pool(pool) {
542 set_name("ArchiveWorkerThread");
543 if (os::create_thread(this, os::os_thread)) {
544 os::start_thread(this);
545 } else {
546 vm_exit_during_initialization("Unable to create archive worker",
547 os::native_thread_creation_failed_msg());
548 }
549 }
550
551 void ArchiveWorkerThread::run() {
552 // Avalanche startup: each worker starts two others.
553 _pool->start_worker_if_needed();
554 _pool->start_worker_if_needed();
555
556 // Set ourselves up.
557 os::set_priority(this, NearMaxPriority);
558
559 // Work.
560 _pool->run_as_worker();
561 }
562
563 void ArchiveWorkerThread::post_run() {
564 this->NamedThread::post_run();
565 delete this;
566 }