1 /*
  2  * Copyright (c) 2019, 2026, 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/aotCompressedPointers.hpp"
 26 #include "cds/aotLogging.hpp"
 27 #include "cds/aotMetaspace.hpp"
 28 #include "cds/archiveBuilder.hpp"
 29 #include "cds/archiveUtils.hpp"
 30 #include "cds/cdsConfig.hpp"
 31 #include "cds/classListParser.hpp"
 32 #include "cds/classListWriter.hpp"
 33 #include "cds/dumpAllocStats.hpp"
 34 #include "cds/dynamicArchive.hpp"
 35 #include "cds/filemap.hpp"
 36 #include "cds/heapShared.hpp"
 37 #include "cds/lambdaProxyClassDictionary.hpp"
 38 #include "classfile/classLoader.hpp"
 39 #include "classfile/systemDictionaryShared.hpp"
 40 #include "classfile/vmClasses.hpp"
 41 #include "interpreter/bootstrapInfo.hpp"
 42 #include "memory/metaspaceUtils.hpp"
 43 #include "memory/resourceArea.hpp"
 44 #include "oops/compressedOops.inline.hpp"
 45 #include "oops/klass.inline.hpp"
 46 #include "runtime/arguments.hpp"
 47 #include "utilities/bitMap.inline.hpp"
 48 #include "utilities/debug.hpp"
 49 #include "utilities/formatBuffer.hpp"
 50 #include "utilities/globalDefinitions.hpp"
 51 #include "utilities/rbTree.inline.hpp"
 52 #include "utilities/spinYield.hpp"
 53 
 54 CHeapBitMap* ArchivePtrMarker::_ptrmap = nullptr;
 55 CHeapBitMap* ArchivePtrMarker::_rw_ptrmap = nullptr;
 56 CHeapBitMap* ArchivePtrMarker::_ro_ptrmap = nullptr;
 57 CHeapBitMap* ArchivePtrMarker::_ac_ptrmap = nullptr;
 58 VirtualSpace* ArchivePtrMarker::_vs;
 59 
 60 bool ArchivePtrMarker::_compacted;
 61 
 62 void ArchivePtrMarker::initialize(CHeapBitMap* ptrmap, VirtualSpace* vs) {
 63   assert(_ptrmap == nullptr, "initialize only once");
 64   assert(_rw_ptrmap == nullptr, "initialize only once");
 65   assert(_ro_ptrmap == nullptr, "initialize only once");
 66   assert(_ac_ptrmap == nullptr, "initialize only once");
 67   _vs = vs;
 68   _compacted = false;
 69   _ptrmap = ptrmap;
 70 
 71   // Use this as initial guesstimate. We should need less space in the
 72   // archive, but if we're wrong the bitmap will be expanded automatically.
 73   size_t estimated_archive_size = MetaspaceGC::capacity_until_GC();
 74   // But set it smaller in debug builds so we always test the expansion code.
 75   // (Default archive is about 12MB).
 76   DEBUG_ONLY(estimated_archive_size = 6 * M);
 77 
 78   // We need one bit per pointer in the archive.
 79   _ptrmap->initialize(estimated_archive_size / sizeof(intptr_t));
 80 }
 81 
 82 void ArchivePtrMarker::initialize_rw_ro_ac_maps(CHeapBitMap* rw_ptrmap, CHeapBitMap* ro_ptrmap, CHeapBitMap* ac_ptrmap) {
 83   address* buff_bottom = (address*)ArchiveBuilder::current()->buffer_bottom();
 84   address* rw_bottom   = (address*)ArchiveBuilder::current()->rw_region()->base();
 85   address* ro_bottom   = (address*)ArchiveBuilder::current()->ro_region()->base();
 86   address* ac_bottom   = (address*)ArchiveBuilder::current()->ac_region()->base();
 87 
 88   // The bit in _ptrmap that cover the very first word in the rw/ro/ac regions.
 89   size_t rw_start = rw_bottom - buff_bottom;
 90   size_t ro_start = ro_bottom - buff_bottom;
 91   size_t ac_start = ac_bottom - buff_bottom;
 92 
 93   // The number of bits used by the rw/ro ptrmaps. We might have lots of zero
 94   // bits at the bottom and top of rw/ro ptrmaps, but these zeros will be
 95   // removed by FileMapInfo::write_bitmap_region().
 96   size_t rw_size = ArchiveBuilder::current()->rw_region()->used() / sizeof(address);
 97   size_t ro_size = ArchiveBuilder::current()->ro_region()->used() / sizeof(address);
 98   size_t ac_size = ArchiveBuilder::current()->ac_region()->used() / sizeof(address);
 99 
100   // The last (exclusive) bit in _ptrmap that covers the rw/ro regions.
101   // Note: _ptrmap is dynamically expanded only when an actual pointer is written, so
102   // it may not be as large as we want.
103   size_t rw_end = MIN2<size_t>(rw_start + rw_size, _ptrmap->size());
104   size_t ro_end = MIN2<size_t>(ro_start + ro_size, _ptrmap->size());
105   size_t ac_end = MIN2<size_t>(ac_start + ac_size, _ptrmap->size());
106 
107   rw_ptrmap->initialize(rw_size);
108   ro_ptrmap->initialize(ro_size);
109   ac_ptrmap->initialize(ac_size);
110 
111   for (size_t rw_bit = rw_start; rw_bit < rw_end; rw_bit++) {
112     rw_ptrmap->at_put(rw_bit - rw_start, _ptrmap->at(rw_bit));
113   }
114 
115   for(size_t ro_bit = ro_start; ro_bit < ro_end; ro_bit++) {
116     ro_ptrmap->at_put(ro_bit - ro_start, _ptrmap->at(ro_bit));
117   }
118 
119   for (size_t ac_bit = ac_start; ac_bit < ac_end; ac_bit++) {
120     ac_ptrmap->at_put(ac_bit - ac_start, _ptrmap->at(ac_bit));
121   }
122 
123   _rw_ptrmap = rw_ptrmap;
124   _ro_ptrmap = ro_ptrmap;
125   _ac_ptrmap = ac_ptrmap;
126 }
127 
128 void ArchivePtrMarker::mark_pointer(address* ptr_loc) {
129   assert(_ptrmap != nullptr, "not initialized");
130   assert(!_compacted, "cannot mark anymore");
131 
132   if (ptr_base() <= ptr_loc && ptr_loc < ptr_end()) {
133     address value = *ptr_loc;
134     if (value != nullptr) {
135       // We don't want any pointer that points to very bottom of the AOT metaspace, otherwise
136       // when AOTMetaspace::default_base_address()==0, we can't distinguish between a pointer
137       // to nothing (null) vs a pointer to an objects that happens to be at the very bottom
138       // of the AOT metaspace.
139       //
140       // This should never happen because the protection zone prevents any valid objects from
141       // being allocated at the bottom of the AOT metaspace.
142       assert(AOTMetaspace::protection_zone_size() > 0, "must be");
143       assert(ArchiveBuilder::current()->any_to_offset(value) > 0, "cannot point to bottom of AOT metaspace");
144 
145       assert(uintx(ptr_loc) % sizeof(intptr_t) == 0, "pointers must be stored in aligned addresses");
146       size_t idx = ptr_loc - ptr_base();
147       if (_ptrmap->size() <= idx) {
148         _ptrmap->resize((idx + 1) * 2);
149       }
150       assert(idx < _ptrmap->size(), "must be");
151       _ptrmap->set_bit(idx);
152     }
153   }
154 }
155 
156 void ArchivePtrMarker::clear_pointer(address* ptr_loc) {
157   assert(_ptrmap != nullptr, "not initialized");
158   assert(!_compacted, "cannot clear anymore");
159 
160   assert(ptr_base() <= ptr_loc && ptr_loc < ptr_end(), "must be");
161   assert(uintx(ptr_loc) % sizeof(intptr_t) == 0, "pointers must be stored in aligned addresses");
162   size_t idx = ptr_loc - ptr_base();
163   assert(idx < _ptrmap->size(), "cannot clear pointers that have not been marked");
164   _ptrmap->clear_bit(idx);
165 }
166 
167 class ArchivePtrBitmapCleaner: public BitMapClosure {
168   CHeapBitMap* _ptrmap;
169   address* _ptr_base;
170   address  _relocatable_base;
171   address  _relocatable_end;
172   size_t   _max_non_null_offset;
173 
174 public:
175   ArchivePtrBitmapCleaner(CHeapBitMap* ptrmap, address* ptr_base, address relocatable_base, address relocatable_end) :
176     _ptrmap(ptrmap), _ptr_base(ptr_base),
177     _relocatable_base(relocatable_base), _relocatable_end(relocatable_end), _max_non_null_offset(0) {}
178 
179   bool do_bit(size_t offset) {
180     address* ptr_loc = _ptr_base + offset;
181     address  ptr_value = *ptr_loc;
182     if (ptr_value != nullptr) {
183       assert(_relocatable_base <= ptr_value && ptr_value < _relocatable_end, "do not point to arbitrary locations!");
184       if (_max_non_null_offset < offset) {
185         _max_non_null_offset = offset;
186       }
187     } else {
188       _ptrmap->clear_bit(offset);
189       DEBUG_ONLY(log_trace(aot, reloc)("Clearing pointer [" PTR_FORMAT  "] -> null @ %9zu", p2i(ptr_loc), offset));
190     }
191 
192     return true;
193   }
194 
195   size_t max_non_null_offset() const { return _max_non_null_offset; }
196 };
197 
198 void ArchivePtrMarker::compact(address relocatable_base, address relocatable_end) {
199   assert(!_compacted, "cannot compact again");
200   ArchivePtrBitmapCleaner cleaner(_ptrmap, ptr_base(), relocatable_base, relocatable_end);
201   _ptrmap->iterate(&cleaner);
202   compact(cleaner.max_non_null_offset());
203 }
204 
205 void ArchivePtrMarker::compact(size_t max_non_null_offset) {
206   assert(!_compacted, "cannot compact again");
207   _ptrmap->resize(max_non_null_offset + 1);
208   _compacted = true;
209 }
210 
211 char* DumpRegion::expand_top_to(char* newtop) {
212   assert(is_allocatable(), "must be initialized and not packed");
213   assert(newtop >= _top, "must not grow backwards");
214   if (newtop > _end) {
215     ArchiveBuilder::current()->report_out_of_space(_name, newtop - _top);
216     ShouldNotReachHere();
217   }
218 
219   commit_to(newtop);
220   _top = newtop;
221 
222   if (ArchiveBuilder::is_active() && ArchiveBuilder::current()->is_in_buffer_space(_base)) {
223     uintx delta = ArchiveBuilder::current()->buffer_to_offset((address)(newtop-1));
224     if (delta > AOTCompressedPointers::MaxMetadataOffsetBytes) {
225       // This is just a sanity check and should not appear in any real world usage. This
226       // happens only if you allocate more than 2GB of shared objects and would require
227       // millions of shared classes.
228       aot_log_error(aot)("Out of memory in the %s: Please reduce the number of shared classes.", CDSConfig::type_of_archive_being_written());
229       AOTMetaspace::unrecoverable_writing_error();
230     }
231   }
232 
233   return _top;
234 }
235 
236 void DumpRegion::commit_to(char* newtop) {
237   assert(CDSConfig::is_dumping_archive(), "sanity");
238   char* base = _rs->base();
239   size_t need_committed_size = newtop - base;
240   size_t has_committed_size = _vs->committed_size();
241   if (need_committed_size < has_committed_size) {
242     return;
243   }
244 
245   size_t min_bytes = need_committed_size - has_committed_size;
246   size_t preferred_bytes = 1 * M;
247   size_t uncommitted = _vs->reserved_size() - has_committed_size;
248 
249   size_t commit = MAX2(min_bytes, preferred_bytes);
250   commit = MIN2(commit, uncommitted);
251   assert(commit <= uncommitted, "sanity");
252 
253   if (!_vs->expand_by(commit, false)) {
254     aot_log_error(aot)("Failed to expand shared space to %zu bytes",
255                     need_committed_size);
256     AOTMetaspace::unrecoverable_writing_error();
257   }
258 
259   const char* which;
260   if (_rs->base() == (char*)AOTMetaspace::symbol_rs_base()) {
261     which = "symbol";
262   } else {
263     which = "shared";
264   }
265   log_debug(aot)("Expanding %s spaces by %7zu bytes [total %9zu bytes ending at %p]",
266                  which, commit, _vs->actual_committed_size(), _vs->high());
267 }
268 
269 // Basic allocation. Any alignment gaps will be wasted.
270 char* DumpRegion::allocate(size_t num_bytes, size_t alignment) {
271   // Always align to at least minimum alignment
272   alignment = MAX2(SharedSpaceObjectAlignment, alignment);
273   char* p = (char*)align_up(_top, alignment);
274   char* newtop = p + align_up(num_bytes, SharedSpaceObjectAlignment);
275   expand_top_to(newtop);
276   memset(p, 0, newtop - p);
277   return p;
278 }
279 
280 class DumpRegion::AllocGap {
281   size_t _gap_bytes;   // size of this gap in bytes
282   char* _gap_bottom;   // must be SharedSpaceObjectAlignment aligned
283 public:
284   size_t gap_bytes() const { return _gap_bytes; }
285   char* gap_bottom() const { return _gap_bottom; }
286 
287   AllocGap(size_t bytes, char* bottom) : _gap_bytes(bytes), _gap_bottom(bottom) {
288     precond(is_aligned(gap_bytes(), SharedSpaceObjectAlignment));
289     precond(is_aligned(gap_bottom(), SharedSpaceObjectAlignment));
290   }
291 };
292 
293 struct DumpRegion::AllocGapCmp {
294   static RBTreeOrdering cmp(AllocGap a, AllocGap b) {
295     RBTreeOrdering order = rbtree_primitive_cmp(a.gap_bytes(), b.gap_bytes());
296     if (order == RBTreeOrdering::EQ) {
297       order = rbtree_primitive_cmp(a.gap_bottom(), b.gap_bottom());
298     }
299     return order;
300   }
301 };
302 
303 struct Empty {};
304 using AllocGapNode = RBNode<DumpRegion::AllocGap, Empty>;
305 
306 class DumpRegion::AllocGapTree : public RBTreeCHeap<AllocGap, Empty, AllocGapCmp, mtClassShared> {
307 public:
308   size_t add_gap(char* gap_bottom, char* gap_top) {
309     precond(gap_bottom < gap_top);
310     size_t gap_bytes = pointer_delta(gap_top, gap_bottom, 1);
311     precond(gap_bytes > 0);
312 
313     _total_gap_bytes += gap_bytes;
314 
315     AllocGap gap(gap_bytes, gap_bottom); // constructor checks alignment
316     AllocGapNode* node = allocate_node(gap, Empty{});
317     insert(gap, node);
318 
319     log_trace(aot, alloc)("adding a gap of %zu bytes @ %p (total = %zu) in %zu blocks", gap_bytes, gap_bottom, _total_gap_bytes, size());
320     return gap_bytes;
321   }
322 
323   char* allocate_from_gap(size_t num_bytes) {
324     // The gaps are sorted in ascending order of their sizes. When two gaps have the same
325     // size, the one with a lower gap_bottom comes first.
326     //
327     // Find the first gap that's big enough, with the lowest gap_bottom.
328     AllocGap target(num_bytes, nullptr);
329     AllocGapNode* node = closest_ge(target);
330     if (node == nullptr) {
331       return nullptr; // Didn't find any usable gap.
332     }
333 
334     size_t gap_bytes = node->key().gap_bytes();
335     char* gap_bottom = node->key().gap_bottom();
336     char* result = gap_bottom;
337     precond(is_aligned(result, SharedSpaceObjectAlignment));
338 
339     remove(node);
340 
341     precond(_total_gap_bytes >= num_bytes);
342     _total_gap_bytes -= num_bytes;
343     _total_gap_bytes_used += num_bytes;
344     _total_gap_allocs++;
345     DEBUG_ONLY(node = nullptr); // Don't use it anymore!
346 
347     precond(gap_bytes >= num_bytes);
348     if (gap_bytes > num_bytes) {
349       gap_bytes -= num_bytes;
350       gap_bottom += num_bytes;
351 
352       AllocGap gap(gap_bytes, gap_bottom); // constructor checks alignment
353       AllocGapNode* new_node = allocate_node(gap, Empty{});
354       insert(gap, new_node);
355     }
356     log_trace(aot, alloc)("%zu bytes @ %p in a gap of %zu bytes (used gaps %zu times, remain gap = %zu bytes in %zu blocks)",
357                           num_bytes, result, gap_bytes, _total_gap_allocs, _total_gap_bytes, size());
358     return result;
359   }
360 };
361 
362 size_t DumpRegion::_total_gap_bytes = 0;
363 size_t DumpRegion::_total_gap_bytes_used = 0;
364 size_t DumpRegion::_total_gap_allocs = 0;
365 DumpRegion::AllocGapTree DumpRegion::_gap_tree;
366 
367 // Alignment gaps happen only for the RW space. Collect the gaps into the _gap_tree so they can be
368 // used for future small object allocation.
369 char* DumpRegion::allocate_metaspace_obj(size_t num_bytes, address src, MetaspaceClosureType type, bool read_only, DumpAllocStats* stats) {
370   num_bytes = align_up(num_bytes, SharedSpaceObjectAlignment);
371   size_t alignment = SharedSpaceObjectAlignment; // alignment for the dest pointer
372   bool is_class = (type == MetaspaceClosureType::ClassType);
373   bool is_instance_class = is_class && ((Klass*)src)->is_instance_klass();
374 
375 #ifdef _LP64
376   // More strict alignments needed for UseCompressedClassPointers
377   if (is_class && UseCompressedClassPointers) {
378     size_t klass_alignment = checked_cast<size_t>(nth_bit(ArchiveBuilder::precomputed_narrow_klass_shift()));
379     alignment = MAX2(alignment, klass_alignment);
380     precond(is_aligned(alignment, SharedSpaceObjectAlignment));
381   }
382 #endif
383 
384   if (alignment == SharedSpaceObjectAlignment && type != MetaspaceClosureType::SymbolType) {
385     // The addresses of Symbols must be in the same order as they are in ArchiveBuilder::SourceObjList.
386     // If we put them in gaps, their order will change.
387     //
388     // We have enough small objects that all gaps are usually filled.
389     char* p = _gap_tree.allocate_from_gap(num_bytes);
390     if (p != nullptr) {
391       // Already memset to 0 when adding the gap
392       stats->record(type, checked_cast<int>(num_bytes), /*read_only=*/false); // all gaps are from RW space (for classes)
393       return p;
394     }
395   }
396 
397   // Reserve space for a pointer directly in front of the buffered InstanceKlass, so
398   // we can do a quick lookup from InstanceKlass* -> RunTimeClassInfo*
399   // without building another hashtable. See RunTimeClassInfo::get_for()
400   // in systemDictionaryShared.cpp.
401   const size_t RuntimeClassInfoPtrSize = is_instance_class ? sizeof(address) : 0;
402 
403   if (is_class && !is_aligned(top() + RuntimeClassInfoPtrSize, alignment)) {
404     // We need to add a gap to align the buffered Klass. Save the gap for future small allocations.
405     assert(read_only == false, "only gaps in RW region are reusable");
406     char* gap_bottom = top();
407     char* gap_top = align_up(gap_bottom + RuntimeClassInfoPtrSize, alignment) - RuntimeClassInfoPtrSize;
408     size_t gap_bytes = _gap_tree.add_gap(gap_bottom, gap_top);
409     allocate(gap_bytes);
410   }
411 
412   char* oldtop = top();
413   if (is_instance_class) {
414     SystemDictionaryShared::validate_before_archiving((InstanceKlass*)src);
415     allocate(RuntimeClassInfoPtrSize);
416   }
417 
418   precond(is_aligned(top(), alignment));
419   char* result = allocate(num_bytes);
420   log_trace(aot, alloc)("%zu bytes @ %p", num_bytes, result);
421   stats->record(type, pointer_delta_as_int(top(), oldtop), read_only); // includes RuntimeClassInfoPtrSize for classes
422 
423   return result;
424 }
425 
426 // Usually we have no gaps left.
427 void DumpRegion::report_gaps(DumpAllocStats* stats) {
428   _gap_tree.visit_in_order([&](const AllocGapNode* node) {
429         stats->record_gap(checked_cast<int>(node->key().gap_bytes()));
430         return true;
431       });
432   if (_gap_tree.size() > 0) {
433     log_warning(aot)("Unexpected %zu gaps (%zu bytes) for Klass alignment",
434                      _gap_tree.size(), _total_gap_bytes);
435   }
436   if (_total_gap_allocs > 0) {
437     log_info(aot)("Allocated %zu objects of %zu bytes in gaps (remain = %zu bytes)",
438                   _total_gap_allocs, _total_gap_bytes_used, _total_gap_bytes);
439   }
440 }
441 
442 void DumpRegion::append_intptr_t(intptr_t n, bool need_to_mark) {
443   assert(is_aligned(_top, sizeof(intptr_t)), "bad alignment");
444   intptr_t *p = (intptr_t*)_top;
445   char* newtop = _top + sizeof(intptr_t);
446   expand_top_to(newtop);
447   *p = n;
448   if (need_to_mark) {
449     ArchivePtrMarker::mark_pointer(p);
450   }
451 }
452 
453 void DumpRegion::print(size_t total_bytes) const {
454   char* base = used() > 0 ? ArchiveBuilder::current()->to_requested(_base) : nullptr;
455   log_debug(aot)("%s space: %9zu [ %4.1f%% of total] out of %9zu bytes [%5.1f%% used] at " INTPTR_FORMAT,
456                  _name, used(), percent_of(used(), total_bytes), reserved(), percent_of(used(), reserved()),
457                  p2i(base));
458 }
459 
460 void DumpRegion::print_out_of_space_msg(const char* failing_region, size_t needed_bytes) {
461   aot_log_error(aot)("[%-8s] " PTR_FORMAT " - " PTR_FORMAT " capacity =%9d, allocated =%9d",
462                  _name, p2i(_base), p2i(_top), int(_end - _base), int(_top - _base));
463   if (strcmp(_name, failing_region) == 0) {
464     aot_log_error(aot)(" required = %d", int(needed_bytes));
465   }
466 }
467 
468 void DumpRegion::init(ReservedSpace* rs, VirtualSpace* vs) {
469   _rs = rs;
470   _vs = vs;
471   // Start with 0 committed bytes. The memory will be committed as needed.
472   if (!_vs->initialize(*_rs, 0)) {
473     fatal("Unable to allocate memory for shared space");
474   }
475   _base = _top = _rs->base();
476   _end = _rs->end();
477 }
478 
479 void DumpRegion::pack(DumpRegion* next) {
480   if (!is_packed()) {
481     _end = (char*)align_up(_top, AOTMetaspace::core_region_alignment());
482     _is_packed = true;
483   }


484   if (next != nullptr) {
485     next->_rs = _rs;
486     next->_vs = _vs;
487     next->_base = next->_top = this->_end;
488     next->_end = _rs->end();
489   }
490 }
491 
492 void WriteClosure::do_ptr(void** p) {
493   address ptr = *(address*)p;
494   AOTCompressedPointers::narrowPtr narrowp = AOTCompressedPointers::encode(ptr);
495   _dump_region->append_intptr_t(checked_cast<intptr_t>(narrowp), false);
496 }
497 
498 void ReadClosure::do_ptr(void** p) {
499   assert(*p == nullptr, "initializing previous initialized pointer.");
500   u4 narrowp = checked_cast<u4>(nextPtr());
501   *p = AOTCompressedPointers::decode<void*>(cast_from_u4(narrowp), _base_address);
502 }
503 
504 void ReadClosure::do_u4(u4* p) {
505   intptr_t obj = nextPtr();
506   *p = (u4)(uintx(obj));
507 }
508 
509 void ReadClosure::do_int(int* p) {
510   intptr_t obj = nextPtr();
511   *p = (int)(intx(obj));
512 }
513 
514 void ReadClosure::do_bool(bool* p) {
515   intptr_t obj = nextPtr();
516   *p = (bool)(uintx(obj));
517 }
518 
519 void ReadClosure::do_tag(int tag) {
520   int old_tag;
521   old_tag = (int)(intptr_t)nextPtr();
522   // do_int(&old_tag);
523   assert(tag == old_tag, "tag doesn't match (%d, expected %d)", old_tag, tag);
524   FileMapInfo::assert_mark(tag == old_tag);
525 }
526 
527 void ArchiveUtils::log_to_classlist(BootstrapInfo* bootstrap_specifier, TRAPS) {
528   if (ClassListWriter::is_enabled()) {
529     if (LambdaProxyClassDictionary::is_supported_invokedynamic(bootstrap_specifier)) {
530       const constantPoolHandle& pool = bootstrap_specifier->pool();
531       if (SystemDictionaryShared::is_builtin_loader(pool->pool_holder()->class_loader_data())) {
532         // Currently lambda proxy classes are supported only for the built-in loaders.
533         ResourceMark rm(THREAD);
534         int pool_index = bootstrap_specifier->bss_index();
535         ClassListWriter w;
536         w.stream()->print("%s %s", ClassListParser::lambda_proxy_tag(), pool->pool_holder()->name()->as_C_string());
537         CDSIndyInfo cii;
538         ClassListParser::populate_cds_indy_info(pool, pool_index, &cii, CHECK);
539         GrowableArray<const char*>* indy_items = cii.items();
540         for (int i = 0; i < indy_items->length(); i++) {
541           w.stream()->print(" %s", indy_items->at(i));
542         }
543         w.stream()->cr();
544       }
545     }
546   }
547 }
548 
549 
550 // "boot", "platform", "app" or nullptr
551 const char* ArchiveUtils::builtin_loader_name_or_null(oop loader) {
552   if (loader == nullptr) {
553     return "boot";
554   } else if (loader == SystemDictionary::java_platform_loader()) {
555     return "platform";
556   } else if (loader == SystemDictionary::java_system_loader()) {
557     return "app";
558   } else {
559     return nullptr;
560   }
561 }
562 
563 // "boot", "platform", "app". Asserts if not a built-in-loader
564 const char* ArchiveUtils::builtin_loader_name(oop loader) {
565   const char* name = builtin_loader_name_or_null(loader);
566   assert(name != nullptr, "must be a built-in loader");
567   return name;
568 }
569 
570 bool ArchiveUtils::builtin_loader_from_type(const char* loader_type, oop* value_ret) {
571   if (strcmp(loader_type, "boot") == 0) {
572     *value_ret = nullptr;
573     return true;
574   } else if (strcmp(loader_type, "platform") == 0) {
575     *value_ret = SystemDictionary::java_platform_loader();
576     return true;
577   } else if (strcmp(loader_type, "app") == 0) {
578     *value_ret = SystemDictionary::java_system_loader();
579     return true;
580   } else {
581     DEBUG_ONLY(*value_ret = cast_to_oop((void*)badOopVal));
582     return false;
583   }
584 }
585 
586 oop ArchiveUtils::builtin_loader_from_type(int loader_type) {
587   if (loader_type == ClassLoader::BOOT_LOADER) {
588     return nullptr;
589   } else if (loader_type == ClassLoader::PLATFORM_LOADER)  {
590     return SystemDictionary::java_platform_loader();
591   } else if (loader_type == ClassLoader::APP_LOADER) {
592     return SystemDictionary::java_system_loader();
593   } else {
594     ShouldNotReachHere();
595     return nullptr;
596   }
597 }
598 
599 bool ArchiveUtils::has_aot_initialized_mirror(InstanceKlass* src_ik) {
600   if (!ArchiveBuilder::current()->has_been_archived(src_ik)) {
601     return false;
602   }
603   return ArchiveBuilder::current()->get_buffered_addr(src_ik)->has_aot_initialized_mirror();
604 }
605 
606 size_t HeapRootSegments::size_in_bytes(size_t seg_idx) {
607   assert(seg_idx < _count, "In range");
608   return objArrayOopDesc::object_size(size_in_elems(seg_idx)) * HeapWordSize;
609 }
610 
611 int HeapRootSegments::size_in_elems(size_t seg_idx) {
612   assert(seg_idx < _count, "In range");
613   if (seg_idx != _count - 1) {
614     return _max_size_in_elems;
615   } else {
616     // Last slice, leftover
617     return _roots_count % _max_size_in_elems;
618   }
619 }
620 
621 size_t HeapRootSegments::segment_offset(size_t seg_idx) {
622   assert(seg_idx < _count, "In range");
623   return _base_offset + seg_idx * _max_size_in_bytes;
624 }
625 
626 ArchiveWorkers::ArchiveWorkers() :
627         _end_semaphore(0),
628         _num_workers(max_workers()),
629         _started_workers(0),
630         _finish_tokens(0),
631         _state(UNUSED),
632         _task(nullptr) {}
633 
634 ArchiveWorkers::~ArchiveWorkers() {
635   assert(AtomicAccess::load(&_state) != WORKING, "Should not be working");
636 }
637 
638 int ArchiveWorkers::max_workers() {
639   // The pool is used for short-lived bursty tasks. We do not want to spend
640   // too much time creating and waking up threads unnecessarily. Plus, we do
641   // not want to overwhelm large machines. This is why we want to be very
642   // conservative about the number of workers actually needed.
643   return MAX2(0, log2i_graceful(os::active_processor_count()));
644 }
645 
646 bool ArchiveWorkers::is_parallel() {
647   return _num_workers > 0;
648 }
649 
650 void ArchiveWorkers::start_worker_if_needed() {
651   while (true) {
652     int cur = AtomicAccess::load(&_started_workers);
653     if (cur >= _num_workers) {
654       return;
655     }
656     if (AtomicAccess::cmpxchg(&_started_workers, cur, cur + 1, memory_order_relaxed) == cur) {
657       new ArchiveWorkerThread(this);
658       return;
659     }
660   }
661 }
662 
663 void ArchiveWorkers::run_task(ArchiveWorkerTask* task) {
664   assert(AtomicAccess::load(&_state) == UNUSED, "Should be unused yet");
665   assert(AtomicAccess::load(&_task) == nullptr, "Should not have running tasks");
666   AtomicAccess::store(&_state, WORKING);
667 
668   if (is_parallel()) {
669     run_task_multi(task);
670   } else {
671     run_task_single(task);
672   }
673 
674   assert(AtomicAccess::load(&_state) == WORKING, "Should be working");
675   AtomicAccess::store(&_state, SHUTDOWN);
676 }
677 
678 void ArchiveWorkers::run_task_single(ArchiveWorkerTask* task) {
679   // Single thread needs no chunking.
680   task->configure_max_chunks(1);
681 
682   // Execute the task ourselves, as there are no workers.
683   task->work(0, 1);
684 }
685 
686 void ArchiveWorkers::run_task_multi(ArchiveWorkerTask* task) {
687   // Multiple threads can work with multiple chunks.
688   task->configure_max_chunks(_num_workers * CHUNKS_PER_WORKER);
689 
690   // Set up the run and publish the task. Issue one additional finish token
691   // to cover the semaphore shutdown path, see below.
692   AtomicAccess::store(&_finish_tokens, _num_workers + 1);
693   AtomicAccess::release_store(&_task, task);
694 
695   // Kick off pool startup by starting a single worker, and proceed
696   // immediately to executing the task locally.
697   start_worker_if_needed();
698 
699   // Execute the task ourselves, while workers are catching up.
700   // This allows us to hide parts of task handoff latency.
701   task->run();
702 
703   // Done executing task locally, wait for any remaining workers to complete.
704   // Once all workers report, we can proceed to termination. To do this safely,
705   // we need to make sure every worker has left. A spin-wait alone would suffice,
706   // but we do not want to burn cycles on it. A semaphore alone would not be safe,
707   // since workers can still be inside it as we proceed from wait here. So we block
708   // on semaphore first, and then spin-wait for all workers to terminate.
709   _end_semaphore.wait();
710   SpinYield spin;
711   while (AtomicAccess::load(&_finish_tokens) != 0) {
712     spin.wait();
713   }
714 
715   OrderAccess::fence();
716 
717   assert(AtomicAccess::load(&_finish_tokens) == 0, "All tokens are consumed");
718 }
719 
720 void ArchiveWorkers::run_as_worker() {
721   assert(is_parallel(), "Should be in parallel mode");
722 
723   ArchiveWorkerTask* task = AtomicAccess::load_acquire(&_task);
724   task->run();
725 
726   // All work done in threads should be visible to caller.
727   OrderAccess::fence();
728 
729   // Signal the pool the work is complete, and we are exiting.
730   // Worker cannot do anything else with the pool after this.
731   if (AtomicAccess::sub(&_finish_tokens, 1, memory_order_relaxed) == 1) {
732     // Last worker leaving. Notify the pool it can unblock to spin-wait.
733     // Then consume the last token and leave.
734     _end_semaphore.signal();
735     int last = AtomicAccess::sub(&_finish_tokens, 1, memory_order_relaxed);
736     assert(last == 0, "Should be");
737   }
738 }
739 
740 void ArchiveWorkerTask::run() {
741   while (true) {
742     int chunk = AtomicAccess::load(&_chunk);
743     if (chunk >= _max_chunks) {
744       return;
745     }
746     if (AtomicAccess::cmpxchg(&_chunk, chunk, chunk + 1, memory_order_relaxed) == chunk) {
747       assert(0 <= chunk && chunk < _max_chunks, "Sanity");
748       work(chunk, _max_chunks);
749     }
750   }
751 }
752 
753 void ArchiveWorkerTask::configure_max_chunks(int max_chunks) {
754   if (_max_chunks == 0) {
755     _max_chunks = max_chunks;
756   }
757 }
758 
759 ArchiveWorkerThread::ArchiveWorkerThread(ArchiveWorkers* pool) : NamedThread(), _pool(pool) {
760   set_name("ArchiveWorkerThread");
761   if (os::create_thread(this, os::os_thread)) {
762     os::start_thread(this);
763   } else {
764     vm_exit_during_initialization("Unable to create archive worker",
765                                   os::native_thread_creation_failed_msg());
766   }
767 }
768 
769 void ArchiveWorkerThread::run() {
770   // Avalanche startup: each worker starts two others.
771   _pool->start_worker_if_needed();
772   _pool->start_worker_if_needed();
773 
774   // Set ourselves up.
775   os::set_priority(this, NearMaxPriority);
776 
777   // Work.
778   _pool->run_as_worker();
779 }
780 
781 void ArchiveWorkerThread::post_run() {
782   this->NamedThread::post_run();
783   delete this;
784 }
--- EOF ---