1 /*
2 * Copyright (c) 2019, 2024, 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 "cds/archiveBuilder.hpp"
27 #include "cds/archiveHeapLoader.inline.hpp"
28 #include "cds/archiveUtils.hpp"
29 #include "cds/cdsConfig.hpp"
30 #include "cds/classListParser.hpp"
31 #include "cds/classListWriter.hpp"
32 #include "cds/dynamicArchive.hpp"
33 #include "cds/filemap.hpp"
34 #include "cds/heapShared.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* rw_bottom = (address*)ArchiveBuilder::current()->rw_region()->base();
78 address* ro_bottom = (address*)ArchiveBuilder::current()->ro_region()->base();
79
80 _rw_ptrmap = rw_ptrmap;
81 _ro_ptrmap = ro_ptrmap;
82
83 size_t rw_size = ArchiveBuilder::current()->rw_region()->used() / sizeof(address);
84 size_t ro_size = ArchiveBuilder::current()->ro_region()->used() / sizeof(address);
85 // ro_start is the first bit in _ptrmap that covers the pointer that would sit at ro_bottom.
86 // E.g., if rw_bottom = (address*)100
87 // ro_bottom = (address*)116
88 // then for 64-bit platform:
89 // ro_start = ro_bottom - rw_bottom = (116 - 100) / sizeof(address) = 2;
90 size_t ro_start = ro_bottom - rw_bottom;
91
92 // Note: ptrmap is big enough only to cover the last pointer in ro_region.
93 // See ArchivePtrMarker::compact()
94 _rw_ptrmap->initialize(rw_size);
95 _ro_ptrmap->initialize(_ptrmap->size() - ro_start);
96
97 for (size_t rw_bit = 0; rw_bit < _rw_ptrmap->size(); rw_bit++) {
98 _rw_ptrmap->at_put(rw_bit, _ptrmap->at(rw_bit));
99 }
100
101 for(size_t ro_bit = ro_start; ro_bit < _ptrmap->size(); ro_bit++) {
102 _ro_ptrmap->at_put(ro_bit-ro_start, _ptrmap->at(ro_bit));
103 }
104 assert(_ptrmap->size() - ro_start == _ro_ptrmap->size(), "must be");
105 }
106
107 void ArchivePtrMarker::mark_pointer(address* ptr_loc) {
108 assert(_ptrmap != nullptr, "not initialized");
109 assert(!_compacted, "cannot mark anymore");
110
111 if (ptr_base() <= ptr_loc && ptr_loc < ptr_end()) {
112 address value = *ptr_loc;
113 // We don't want any pointer that points to very bottom of the archive, otherwise when
114 // MetaspaceShared::default_base_address()==0, we can't distinguish between a pointer
115 // to nothing (null) vs a pointer to an objects that happens to be at the very bottom
116 // of the archive.
117 assert(value != (address)ptr_base(), "don't point to the bottom of the archive");
118
119 if (value != nullptr) {
120 assert(uintx(ptr_loc) % sizeof(intptr_t) == 0, "pointers must be stored in aligned addresses");
121 size_t idx = ptr_loc - ptr_base();
122 if (_ptrmap->size() <= idx) {
123 _ptrmap->resize((idx + 1) * 2);
124 }
125 assert(idx < _ptrmap->size(), "must be");
126 _ptrmap->set_bit(idx);
127 //tty->print_cr("Marking pointer [" PTR_FORMAT "] -> " PTR_FORMAT " @ " SIZE_FORMAT_W(5), p2i(ptr_loc), p2i(*ptr_loc), idx);
128 }
129 }
130 }
131
132 void ArchivePtrMarker::clear_pointer(address* ptr_loc) {
133 assert(_ptrmap != nullptr, "not initialized");
134 assert(!_compacted, "cannot clear anymore");
135
136 assert(ptr_base() <= ptr_loc && ptr_loc < ptr_end(), "must be");
137 assert(uintx(ptr_loc) % sizeof(intptr_t) == 0, "pointers must be stored in aligned addresses");
138 size_t idx = ptr_loc - ptr_base();
139 assert(idx < _ptrmap->size(), "cannot clear pointers that have not been marked");
140 _ptrmap->clear_bit(idx);
141 //tty->print_cr("Clearing pointer [" PTR_FORMAT "] -> " PTR_FORMAT " @ " SIZE_FORMAT_W(5), p2i(ptr_loc), p2i(*ptr_loc), idx);
142 }
143
144 class ArchivePtrBitmapCleaner: public BitMapClosure {
145 CHeapBitMap* _ptrmap;
146 address* _ptr_base;
147 address _relocatable_base;
148 address _relocatable_end;
149 size_t _max_non_null_offset;
150
151 public:
152 ArchivePtrBitmapCleaner(CHeapBitMap* ptrmap, address* ptr_base, address relocatable_base, address relocatable_end) :
153 _ptrmap(ptrmap), _ptr_base(ptr_base),
154 _relocatable_base(relocatable_base), _relocatable_end(relocatable_end), _max_non_null_offset(0) {}
155
156 bool do_bit(size_t offset) {
157 address* ptr_loc = _ptr_base + offset;
158 address ptr_value = *ptr_loc;
159 if (ptr_value != nullptr) {
160 assert(_relocatable_base <= ptr_value && ptr_value < _relocatable_end, "do not point to arbitrary locations!");
161 if (_max_non_null_offset < offset) {
162 _max_non_null_offset = offset;
163 }
164 } else {
165 _ptrmap->clear_bit(offset);
166 DEBUG_ONLY(log_trace(cds, reloc)("Clearing pointer [" PTR_FORMAT "] -> null @ " SIZE_FORMAT_W(9), p2i(ptr_loc), offset));
167 }
168
169 return true;
170 }
171
172 size_t max_non_null_offset() const { return _max_non_null_offset; }
173 };
174
175 void ArchivePtrMarker::compact(address relocatable_base, address relocatable_end) {
176 assert(!_compacted, "cannot compact again");
177 ArchivePtrBitmapCleaner cleaner(_ptrmap, ptr_base(), relocatable_base, relocatable_end);
178 _ptrmap->iterate(&cleaner);
179 compact(cleaner.max_non_null_offset());
180 }
181
182 void ArchivePtrMarker::compact(size_t max_non_null_offset) {
183 assert(!_compacted, "cannot compact again");
184 _ptrmap->resize(max_non_null_offset + 1);
185 _compacted = true;
186 }
187
188 char* DumpRegion::expand_top_to(char* newtop) {
189 assert(is_allocatable(), "must be initialized and not packed");
190 assert(newtop >= _top, "must not grow backwards");
191 if (newtop > _end) {
192 ArchiveBuilder::current()->report_out_of_space(_name, newtop - _top);
193 ShouldNotReachHere();
194 }
195
196 commit_to(newtop);
197 _top = newtop;
198
199 if (_max_delta > 0) {
200 uintx delta = ArchiveBuilder::current()->buffer_to_offset((address)(newtop-1));
201 if (delta > _max_delta) {
202 // This is just a sanity check and should not appear in any real world usage. This
203 // happens only if you allocate more than 2GB of shared objects and would require
204 // millions of shared classes.
205 log_error(cds)("Out of memory in the CDS archive: Please reduce the number of shared classes.");
206 MetaspaceShared::unrecoverable_writing_error();
207 }
208 }
209
210 return _top;
211 }
212
213 void DumpRegion::commit_to(char* newtop) {
214 assert(CDSConfig::is_dumping_archive(), "sanity");
215 char* base = _rs->base();
216 size_t need_committed_size = newtop - base;
217 size_t has_committed_size = _vs->committed_size();
218 if (need_committed_size < has_committed_size) {
219 return;
220 }
221
222 size_t min_bytes = need_committed_size - has_committed_size;
223 size_t preferred_bytes = 1 * M;
224 size_t uncommitted = _vs->reserved_size() - has_committed_size;
225
226 size_t commit = MAX2(min_bytes, preferred_bytes);
227 commit = MIN2(commit, uncommitted);
228 assert(commit <= uncommitted, "sanity");
229
230 if (!_vs->expand_by(commit, false)) {
231 log_error(cds)("Failed to expand shared space to " SIZE_FORMAT " bytes",
232 need_committed_size);
233 MetaspaceShared::unrecoverable_writing_error();
234 }
235
236 const char* which;
237 if (_rs->base() == (char*)MetaspaceShared::symbol_rs_base()) {
238 which = "symbol";
239 } else {
240 which = "shared";
241 }
242 log_debug(cds)("Expanding %s spaces by " SIZE_FORMAT_W(7) " bytes [total " SIZE_FORMAT_W(9) " bytes ending at %p]",
243 which, commit, _vs->actual_committed_size(), _vs->high());
244 }
245
246 char* DumpRegion::allocate(size_t num_bytes, size_t alignment) {
247 // Always align to at least minimum alignment
248 alignment = MAX2(SharedSpaceObjectAlignment, alignment);
249 char* p = (char*)align_up(_top, alignment);
250 char* newtop = p + align_up(num_bytes, (size_t)SharedSpaceObjectAlignment);
251 expand_top_to(newtop);
252 memset(p, 0, newtop - p);
253 return p;
254 }
255
256 void DumpRegion::append_intptr_t(intptr_t n, bool need_to_mark) {
257 assert(is_aligned(_top, sizeof(intptr_t)), "bad alignment");
258 intptr_t *p = (intptr_t*)_top;
259 char* newtop = _top + sizeof(intptr_t);
260 expand_top_to(newtop);
261 *p = n;
262 if (need_to_mark) {
263 ArchivePtrMarker::mark_pointer(p);
264 }
265 }
266
267 void DumpRegion::print(size_t total_bytes) const {
268 log_debug(cds)("%s space: " SIZE_FORMAT_W(9) " [ %4.1f%% of total] out of " SIZE_FORMAT_W(9) " bytes [%5.1f%% used] at " INTPTR_FORMAT,
269 _name, used(), percent_of(used(), total_bytes), reserved(), percent_of(used(), reserved()),
270 p2i(ArchiveBuilder::current()->to_requested(_base)));
271 }
272
273 void DumpRegion::print_out_of_space_msg(const char* failing_region, size_t needed_bytes) {
274 log_error(cds)("[%-8s] " PTR_FORMAT " - " PTR_FORMAT " capacity =%9d, allocated =%9d",
275 _name, p2i(_base), p2i(_top), int(_end - _base), int(_top - _base));
276 if (strcmp(_name, failing_region) == 0) {
277 log_error(cds)(" required = %d", int(needed_bytes));
278 }
279 }
280
281 void DumpRegion::init(ReservedSpace* rs, VirtualSpace* vs) {
282 _rs = rs;
283 _vs = vs;
284 // Start with 0 committed bytes. The memory will be committed as needed.
285 if (!_vs->initialize(*_rs, 0)) {
286 fatal("Unable to allocate memory for shared space");
287 }
288 _base = _top = _rs->base();
289 _end = _rs->end();
290 }
291
292 void DumpRegion::pack(DumpRegion* next) {
293 assert(!is_packed(), "sanity");
294 _end = (char*)align_up(_top, MetaspaceShared::core_region_alignment());
295 _is_packed = true;
296 if (next != nullptr) {
297 next->_rs = _rs;
298 next->_vs = _vs;
299 next->_base = next->_top = this->_end;
300 next->_end = _rs->end();
301 }
302 }
303
304 void WriteClosure::do_ptr(void** p) {
305 // Write ptr into the archive; ptr can be:
306 // (a) null -> written as 0
307 // (b) a "buffered" address -> written as is
308 // (c) a "source" address -> convert to "buffered" and write
309 // The common case is (c). E.g., when writing the vmClasses into the archive.
310 // We have (b) only when we don't have a corresponding source object. E.g.,
311 // the archived c++ vtable entries.
312 address ptr = *(address*)p;
313 if (ptr != nullptr && !ArchiveBuilder::current()->is_in_buffer_space(ptr)) {
314 ptr = ArchiveBuilder::current()->get_buffered_addr(ptr);
315 }
316 // null pointers do not need to be converted to offsets
317 if (ptr != nullptr) {
318 ptr = (address)ArchiveBuilder::current()->buffer_to_offset(ptr);
319 }
320 _dump_region->append_intptr_t((intptr_t)ptr, false);
321 }
322
323 void ReadClosure::do_ptr(void** p) {
324 assert(*p == nullptr, "initializing previous initialized pointer.");
325 intptr_t obj = nextPtr();
326 assert(obj >= 0, "sanity.");
327 *p = (obj != 0) ? (void*)(_base_address + obj) : (void*)obj;
328 }
329
330 void ReadClosure::do_u4(u4* p) {
331 intptr_t obj = nextPtr();
332 *p = (u4)(uintx(obj));
333 }
334
335 void ReadClosure::do_int(int* p) {
336 intptr_t obj = nextPtr();
337 *p = (int)(intx(obj));
338 }
339
340 void ReadClosure::do_bool(bool* p) {
341 intptr_t obj = nextPtr();
342 *p = (bool)(uintx(obj));
343 }
344
345 void ReadClosure::do_tag(int tag) {
346 int old_tag;
347 old_tag = (int)(intptr_t)nextPtr();
348 // do_int(&old_tag);
349 assert(tag == old_tag, "tag doesn't match (%d, expected %d)", old_tag, tag);
350 FileMapInfo::assert_mark(tag == old_tag);
351 }
352
353 void ArchiveUtils::log_to_classlist(BootstrapInfo* bootstrap_specifier, TRAPS) {
354 if (ClassListWriter::is_enabled()) {
355 if (SystemDictionaryShared::is_supported_invokedynamic(bootstrap_specifier)) {
356 const constantPoolHandle& pool = bootstrap_specifier->pool();
357 if (SystemDictionaryShared::is_builtin_loader(pool->pool_holder()->class_loader_data())) {
358 // Currently lambda proxy classes are supported only for the built-in loaders.
359 ResourceMark rm(THREAD);
360 int pool_index = bootstrap_specifier->bss_index();
361 ClassListWriter w;
362 w.stream()->print("%s %s", ClassListParser::lambda_proxy_tag(), pool->pool_holder()->name()->as_C_string());
363 CDSIndyInfo cii;
364 ClassListParser::populate_cds_indy_info(pool, pool_index, &cii, CHECK);
365 GrowableArray<const char*>* indy_items = cii.items();
366 for (int i = 0; i < indy_items->length(); i++) {
367 w.stream()->print(" %s", indy_items->at(i));
368 }
369 w.stream()->cr();
370 }
371 }
372 }
373 }
374
375 bool ArchiveUtils::has_aot_initialized_mirror(InstanceKlass* src_ik) {
376 if (SystemDictionaryShared::is_excluded_class(src_ik)) {
377 assert(!ArchiveBuilder::current()->has_been_buffered(src_ik), "sanity");
378 return false;
379 }
380 return ArchiveBuilder::current()->get_buffered_addr(src_ik)->has_aot_initialized_mirror();
381 }
382
383 size_t HeapRootSegments::size_in_bytes(size_t seg_idx) {
384 assert(seg_idx < _count, "In range");
385 return objArrayOopDesc::object_size(size_in_elems(seg_idx)) * HeapWordSize;
386 }
387
388 int HeapRootSegments::size_in_elems(size_t seg_idx) {
389 assert(seg_idx < _count, "In range");
390 if (seg_idx != _count - 1) {
391 return _max_size_in_elems;
392 } else {
393 // Last slice, leftover
394 return _roots_count % _max_size_in_elems;
395 }
396 }
397
398 size_t HeapRootSegments::segment_offset(size_t seg_idx) {
399 assert(seg_idx < _count, "In range");
400 return _base_offset + seg_idx * _max_size_in_bytes;
401 }
402
403 ArchiveWorkers::ArchiveWorkers() :
404 _end_semaphore(0),
405 _num_workers(max_workers()),
406 _started_workers(0),
407 _finish_tokens(0),
408 _state(UNUSED),
409 _task(nullptr) {}
410
411 ArchiveWorkers::~ArchiveWorkers() {
412 assert(Atomic::load(&_state) != WORKING, "Should not be working");
413 }
414
415 int ArchiveWorkers::max_workers() {
416 // The pool is used for short-lived bursty tasks. We do not want to spend
417 // too much time creating and waking up threads unnecessarily. Plus, we do
418 // not want to overwhelm large machines. This is why we want to be very
419 // conservative about the number of workers actually needed.
420 return MAX2(0, log2i_graceful(os::active_processor_count()));
421 }
422
423 bool ArchiveWorkers::is_parallel() {
424 return _num_workers > 0;
425 }
426
427 void ArchiveWorkers::start_worker_if_needed() {
428 while (true) {
429 int cur = Atomic::load(&_started_workers);
430 if (cur >= _num_workers) {
431 return;
432 }
433 if (Atomic::cmpxchg(&_started_workers, cur, cur + 1, memory_order_relaxed) == cur) {
434 new ArchiveWorkerThread(this);
435 return;
436 }
437 }
438 }
439
440 void ArchiveWorkers::run_task(ArchiveWorkerTask* task) {
441 assert(Atomic::load(&_state) == UNUSED, "Should be unused yet");
442 assert(Atomic::load(&_task) == nullptr, "Should not have running tasks");
443 Atomic::store(&_state, WORKING);
444
445 if (is_parallel()) {
446 run_task_multi(task);
447 } else {
448 run_task_single(task);
449 }
450
451 assert(Atomic::load(&_state) == WORKING, "Should be working");
452 Atomic::store(&_state, SHUTDOWN);
453 }
454
455 void ArchiveWorkers::run_task_single(ArchiveWorkerTask* task) {
456 // Single thread needs no chunking.
457 task->configure_max_chunks(1);
458
459 // Execute the task ourselves, as there are no workers.
460 task->work(0, 1);
461 }
462
463 void ArchiveWorkers::run_task_multi(ArchiveWorkerTask* task) {
464 // Multiple threads can work with multiple chunks.
465 task->configure_max_chunks(_num_workers * CHUNKS_PER_WORKER);
466
467 // Set up the run and publish the task. Issue one additional finish token
468 // to cover the semaphore shutdown path, see below.
469 Atomic::store(&_finish_tokens, _num_workers + 1);
470 Atomic::release_store(&_task, task);
471
472 // Kick off pool startup by starting a single worker, and proceed
473 // immediately to executing the task locally.
474 start_worker_if_needed();
475
476 // Execute the task ourselves, while workers are catching up.
477 // This allows us to hide parts of task handoff latency.
478 task->run();
479
480 // Done executing task locally, wait for any remaining workers to complete.
481 // Once all workers report, we can proceed to termination. To do this safely,
482 // we need to make sure every worker has left. A spin-wait alone would suffice,
483 // but we do not want to burn cycles on it. A semaphore alone would not be safe,
484 // since workers can still be inside it as we proceed from wait here. So we block
485 // on semaphore first, and then spin-wait for all workers to terminate.
486 _end_semaphore.wait();
487 SpinYield spin;
488 while (Atomic::load(&_finish_tokens) != 0) {
489 spin.wait();
490 }
491
492 OrderAccess::fence();
493
494 assert(Atomic::load(&_finish_tokens) == 0, "All tokens are consumed");
495 }
496
497 void ArchiveWorkers::run_as_worker() {
498 assert(is_parallel(), "Should be in parallel mode");
499
500 ArchiveWorkerTask* task = Atomic::load_acquire(&_task);
501 task->run();
502
503 // All work done in threads should be visible to caller.
504 OrderAccess::fence();
505
506 // Signal the pool the work is complete, and we are exiting.
507 // Worker cannot do anything else with the pool after this.
508 if (Atomic::sub(&_finish_tokens, 1, memory_order_relaxed) == 1) {
509 // Last worker leaving. Notify the pool it can unblock to spin-wait.
510 // Then consume the last token and leave.
511 _end_semaphore.signal();
512 int last = Atomic::sub(&_finish_tokens, 1, memory_order_relaxed);
513 assert(last == 0, "Should be");
514 }
515 }
516
517 void ArchiveWorkerTask::run() {
518 while (true) {
519 int chunk = Atomic::load(&_chunk);
520 if (chunk >= _max_chunks) {
521 return;
522 }
523 if (Atomic::cmpxchg(&_chunk, chunk, chunk + 1, memory_order_relaxed) == chunk) {
524 assert(0 <= chunk && chunk < _max_chunks, "Sanity");
525 work(chunk, _max_chunks);
526 }
527 }
528 }
529
530 void ArchiveWorkerTask::configure_max_chunks(int max_chunks) {
531 if (_max_chunks == 0) {
532 _max_chunks = max_chunks;
533 }
534 }
535
536 ArchiveWorkerThread::ArchiveWorkerThread(ArchiveWorkers* pool) : NamedThread(), _pool(pool) {
537 set_name("ArchiveWorkerThread");
538 if (os::create_thread(this, os::os_thread)) {
539 os::start_thread(this);
540 } else {
541 vm_exit_during_initialization("Unable to create archive worker",
542 os::native_thread_creation_failed_msg());
543 }
544 }
545
546 void ArchiveWorkerThread::run() {
547 // Avalanche startup: each worker starts two others.
548 _pool->start_worker_if_needed();
549 _pool->start_worker_if_needed();
550
551 // Set ourselves up.
552 os::set_priority(this, NearMaxPriority);
553
554 // Work.
555 _pool->run_as_worker();
556 }
557
558 void ArchiveWorkerThread::post_run() {
559 this->NamedThread::post_run();
560 delete this;
561 }