1 /*
2 * Copyright (c) 2020, 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 #ifndef SHARE_CDS_ARCHIVEBUILDER_HPP
26 #define SHARE_CDS_ARCHIVEBUILDER_HPP
27
28 #include "cds/archiveUtils.hpp"
29 #include "cds/dumpAllocStats.hpp"
30 #include "memory/metaspace.hpp"
31 #include "memory/metaspaceClosure.hpp"
32 #include "oops/array.hpp"
33 #include "oops/klass.hpp"
34 #include "runtime/os.hpp"
35 #include "utilities/bitMap.hpp"
36 #include "utilities/growableArray.hpp"
37 #include "utilities/resizeableResourceHash.hpp"
38 #include "utilities/resourceHash.hpp"
39
40 class ArchiveHeapInfo;
41 class CHeapBitMap;
42 class FileMapInfo;
43 class Klass;
44 class MemRegion;
45 class Symbol;
46
47 // The minimum alignment for non-Klass objects inside the CDS archive. Klass objects need
48 // to follow CompressedKlassPointers::klass_alignment_in_bytes().
49 constexpr size_t SharedSpaceObjectAlignment = Metaspace::min_allocation_alignment_bytes;
50
51 // Overview of CDS archive creation (for both static and dynamic dump):
52 //
53 // [1] Load all classes (static dump: from the classlist, dynamic dump: as part of app execution)
54 // [2] Allocate "output buffer"
55 // [3] Copy contents of the 2 "core" regions (rw/ro) into the output buffer.
56 // - allocate the cpp vtables in rw (static dump only)
57 // - memcpy the MetaspaceObjs into rw/ro:
58 // dump_rw_region();
59 // dump_ro_region();
60 // - fix all the pointers in the MetaspaceObjs to point to the copies
61 // relocate_metaspaceobj_embedded_pointers()
62 // [4] Copy symbol table, dictionary, etc, into the ro region
63 // [5] Relocate all the pointers in rw/ro, so that the archive can be mapped to
64 // the "requested" location without runtime relocation. See relocate_to_requested()
65 //
66 // "source" vs "buffered" vs "requested"
67 //
68 // The ArchiveBuilder deals with three types of addresses.
69 //
70 // "source": These are the addresses of objects created in step [1] above. They are the actual
71 // InstanceKlass*, Method*, etc, of the Java classes that are loaded for executing
72 // Java bytecodes in the JVM process that's dumping the CDS archive.
73 //
74 // It may be necessary to contiue Java execution after ArchiveBuilder is finished.
75 // Therefore, we don't modify any of the "source" objects.
76 //
77 // "buffered": The "source" objects that are deemed archivable are copied into a temporary buffer.
78 // Objects in the buffer are modified in steps [2, 3, 4] (e.g., unshareable info is
79 // removed, pointers are relocated, etc) to prepare them to be loaded at runtime.
80 //
81 // "requested": These are the addreses where the "buffered" objects should be loaded at runtime.
82 // When the "buffered" objects are written into the archive file, their addresses
83 // are adjusted in step [5] such that the lowest of these objects would be mapped
84 // at SharedBaseAddress.
85 //
86 // Translation between "source" and "buffered" addresses is done with two hashtables:
87 // _src_obj_table : "source" -> "buffered"
88 // _buffered_to_src_table : "buffered" -> "source"
89 //
90 // Translation between "buffered" and "requested" addresses is done with a simple shift:
91 // buffered_address + _buffer_to_requested_delta == requested_address
92 //
93 class ArchiveBuilder : public StackObj {
94 protected:
95 DumpRegion* _current_dump_region;
96 address _buffer_bottom; // for writing the contents of rw/ro regions
97 address _last_verified_top;
98 int _num_dump_regions_used;
99 size_t _other_region_used_bytes;
100
101 // These are the addresses where we will request the static and dynamic archives to be
102 // mapped at run time. If the request fails (due to ASLR), we will map the archives at
103 // os-selected addresses.
104 address _requested_static_archive_bottom; // This is determined solely by the value of
105 // SharedBaseAddress during -Xshare:dump.
106 address _requested_static_archive_top;
107 address _requested_dynamic_archive_bottom; // Used only during dynamic dump. It's placed
108 // immediately above _requested_static_archive_top.
109 address _requested_dynamic_archive_top;
110
111 // (Used only during dynamic dump) where the static archive is actually mapped. This
112 // may be different than _requested_static_archive_{bottom,top} due to ASLR
113 address _mapped_static_archive_bottom;
114 address _mapped_static_archive_top;
115
116 intx _buffer_to_requested_delta;
117
118 DumpRegion* current_dump_region() const { return _current_dump_region; }
119
120 public:
121 enum FollowMode {
122 make_a_copy, point_to_it, set_to_null
123 };
124
125 private:
126 class SourceObjInfo {
127 uintx _ptrmap_start; // The bit-offset of the start of this object (inclusive)
128 uintx _ptrmap_end; // The bit-offset of the end of this object (exclusive)
129 bool _read_only;
130 bool _has_embedded_pointer;
131 FollowMode _follow_mode;
132 int _size_in_bytes;
133 int _id; // Each object has a unique serial ID, starting from zero. The ID is assigned
134 // when the object is added into _source_objs.
135 MetaspaceObj::Type _msotype;
136 address _source_addr; // The source object to be copied.
137 address _buffered_addr; // The copy of this object insider the buffer.
138 public:
139 SourceObjInfo(MetaspaceClosure::Ref* ref, bool read_only, FollowMode follow_mode) :
140 _ptrmap_start(0), _ptrmap_end(0), _read_only(read_only), _has_embedded_pointer(false), _follow_mode(follow_mode),
141 _size_in_bytes(ref->size() * BytesPerWord), _id(0), _msotype(ref->msotype()),
142 _source_addr(ref->obj()) {
143 if (follow_mode == point_to_it) {
144 _buffered_addr = ref->obj();
145 } else {
146 _buffered_addr = nullptr;
147 }
148 }
149 SourceObjInfo(address src, address buf) {
150 _source_addr = src;
151 _buffered_addr = buf;
152 }
153
154 // This constructor is only used for regenerated objects (created by LambdaFormInvokers, etc).
155 // src = address of a Method or InstanceKlass that has been regenerated.
156 // renegerated_obj_info = info for the regenerated version of src.
157 SourceObjInfo(address src, SourceObjInfo* renegerated_obj_info) :
158 _ptrmap_start(0), _ptrmap_end(0), _read_only(false),
159 _follow_mode(renegerated_obj_info->_follow_mode),
160 _size_in_bytes(0), _msotype(renegerated_obj_info->_msotype),
161 _source_addr(src), _buffered_addr(renegerated_obj_info->_buffered_addr) {}
162
163 bool should_copy() const { return _follow_mode == make_a_copy; }
164 void set_buffered_addr(address addr) {
165 assert(should_copy(), "must be");
166 assert(_buffered_addr == nullptr, "cannot be copied twice");
167 assert(addr != nullptr, "must be a valid copy");
168 _buffered_addr = addr;
169 }
170 void set_ptrmap_start(uintx v) { _ptrmap_start = v; }
171 void set_ptrmap_end(uintx v) { _ptrmap_end = v; }
172 uintx ptrmap_start() const { return _ptrmap_start; } // inclusive
173 uintx ptrmap_end() const { return _ptrmap_end; } // exclusive
174 bool read_only() const { return _read_only; }
175 bool has_embedded_pointer() const { return _has_embedded_pointer; }
176 void set_has_embedded_pointer() { _has_embedded_pointer = true; }
177 int size_in_bytes() const { return _size_in_bytes; }
178 int id() const { return _id; }
179 void set_id(int i) { _id = i; }
180 address source_addr() const { return _source_addr; }
181 address buffered_addr() const {
182 if (_follow_mode != set_to_null) {
183 assert(_buffered_addr != nullptr, "must be initialized");
184 }
185 return _buffered_addr;
186 }
187 MetaspaceObj::Type msotype() const { return _msotype; }
188 };
189
190 class SourceObjList {
191 uintx _total_bytes;
192 GrowableArray<SourceObjInfo*>* _objs; // Source objects to be archived
193 CHeapBitMap _ptrmap; // Marks the addresses of the pointer fields
194 // in the source objects
195 public:
196 SourceObjList();
197 ~SourceObjList();
198
199 GrowableArray<SourceObjInfo*>* objs() const { return _objs; }
200
201 void append(SourceObjInfo* src_info);
202 void remember_embedded_pointer(SourceObjInfo* pointing_obj, MetaspaceClosure::Ref* ref);
203 void relocate(int i, ArchiveBuilder* builder);
204
205 // convenience accessor
206 SourceObjInfo* at(int i) const { return objs()->at(i); }
207 };
208
209 class CDSMapLogger;
210
211 static const int INITIAL_TABLE_SIZE = 15889;
212 static const int MAX_TABLE_SIZE = 1000000;
213
214 ReservedSpace _shared_rs;
215 VirtualSpace _shared_vs;
216
217 DumpRegion _rw_region;
218 DumpRegion _ro_region;
219 DumpRegion _cc_region;
220
221 // Combined bitmap to track pointers in both RW and RO regions. This is updated
222 // as objects are copied into RW and RO.
223 CHeapBitMap _ptrmap;
224
225 // _ptrmap is split into these two bitmaps which are written into the archive.
226 CHeapBitMap _rw_ptrmap; // marks pointers in the RW region
227 CHeapBitMap _ro_ptrmap; // marks pointers in the RO region
228 CHeapBitMap _cc_ptrmap; // marks pointers in the CC region
229
230 SourceObjList _rw_src_objs; // objs to put in rw region
231 SourceObjList _ro_src_objs; // objs to put in ro region
232 ResizeableResourceHashtable<address, SourceObjInfo, AnyObj::C_HEAP, mtClassShared> _src_obj_table;
233 ResizeableResourceHashtable<address, address, AnyObj::C_HEAP, mtClassShared> _buffered_to_src_table;
234 GrowableArray<Klass*>* _klasses;
235 GrowableArray<Symbol*>* _symbols;
236 unsigned int _entropy_seed;
237
238 // statistics
239 DumpAllocStats _alloc_stats;
240 size_t _total_heap_region_size;
241
242 void print_region_stats(FileMapInfo *map_info, ArchiveHeapInfo* heap_info);
243 void print_bitmap_region_stats(size_t size, size_t total_size);
244 void print_heap_region_stats(ArchiveHeapInfo* heap_info, size_t total_size);
245
246 // For global access.
247 static ArchiveBuilder* _current;
248
249 public:
250 // Use this when you allocate space outside of ArchiveBuilder::dump_{rw,ro}_region.
251 // These are usually for misc tables that are allocated in the RO space.
252 class OtherROAllocMark {
253 char* _oldtop;
254 public:
255 OtherROAllocMark() {
256 _oldtop = _current->_ro_region.top();
257 }
258 ~OtherROAllocMark();
259 };
260
261 private:
262 FollowMode get_follow_mode(MetaspaceClosure::Ref *ref);
263
264 void iterate_sorted_roots(MetaspaceClosure* it);
265 void sort_klasses();
266 static int compare_symbols_by_address(Symbol** a, Symbol** b);
267 static int compare_klass_by_name(Klass** a, Klass** b);
268 void update_hidden_class_loader_type(InstanceKlass* ik) NOT_CDS_JAVA_HEAP_RETURN;
269
270 void make_shallow_copies(DumpRegion *dump_region, const SourceObjList* src_objs);
271 void make_shallow_copy(DumpRegion *dump_region, SourceObjInfo* src_info);
272
273 void relocate_embedded_pointers(SourceObjList* src_objs);
274
275 bool is_excluded(Klass* k);
276 void clean_up_src_obj_table();
277
278 protected:
279 virtual void iterate_roots(MetaspaceClosure* it) = 0;
280
281 // Conservative estimate for number of bytes needed for:
282 size_t _estimated_metaspaceobj_bytes; // all archived MetaspaceObj's.
283 size_t _estimated_hashtable_bytes; // symbol table and dictionaries
284
285 static const int _total_dump_regions = 2;
286
287 size_t estimate_archive_size();
288
289 void start_dump_region(DumpRegion* next);
290 void verify_estimate_size(size_t estimate, const char* which);
291
292 public:
293 address reserve_buffer();
294
295 address buffer_bottom() const { return _buffer_bottom; }
296 address buffer_top() const { return (address)current_dump_region()->top(); }
297 address requested_static_archive_bottom() const { return _requested_static_archive_bottom; }
298 address mapped_static_archive_bottom() const { return _mapped_static_archive_bottom; }
299 intx buffer_to_requested_delta() const { return _buffer_to_requested_delta; }
300
301 bool is_in_buffer_space(address p) const {
302 return (buffer_bottom() != nullptr && buffer_bottom() <= p && p < buffer_top());
303 }
304
305 template <typename T> bool is_in_requested_static_archive(T p) const {
306 return _requested_static_archive_bottom <= (address)p && (address)p < _requested_static_archive_top;
307 }
308
309 template <typename T> bool is_in_mapped_static_archive(T p) const {
310 return _mapped_static_archive_bottom <= (address)p && (address)p < _mapped_static_archive_top;
311 }
312
313 template <typename T> bool is_in_buffer_space(T obj) const {
314 return is_in_buffer_space(address(obj));
315 }
316
317 template <typename T> T to_requested(T obj) const {
318 assert(is_in_buffer_space(obj), "must be");
319 return (T)(address(obj) + _buffer_to_requested_delta);
320 }
321
322 static intx get_buffer_to_requested_delta() {
323 return current()->buffer_to_requested_delta();
324 }
325
326 inline static u4 to_offset_u4(uintx offset) {
327 guarantee(offset <= MAX_SHARED_DELTA, "must be 32-bit offset " INTPTR_FORMAT, offset);
328 return (u4)offset;
329 }
330
331 public:
332 static const uintx MAX_SHARED_DELTA = ArchiveUtils::MAX_SHARED_DELTA;;
333
334 // The address p points to an object inside the output buffer. When the archive is mapped
335 // at the requested address, what's the offset of this object from _requested_static_archive_bottom?
336 uintx buffer_to_offset(address p) const;
337
338 // Same as buffer_to_offset, except that the address p points to either (a) an object
339 // inside the output buffer, or (b), an object in the currently mapped static archive.
340 uintx any_to_offset(address p) const;
341
342 // The reverse of buffer_to_offset()
343 address offset_to_buffered_address(u4 offset) const;
344
345 template <typename T>
346 u4 buffer_to_offset_u4(T p) const {
347 uintx offset = buffer_to_offset((address)p);
348 return to_offset_u4(offset);
349 }
350
351 template <typename T>
352 u4 any_to_offset_u4(T p) const {
353 uintx offset = any_to_offset((address)p);
354 return to_offset_u4(offset);
355 }
356
357 template <typename T>
358 T offset_to_buffered(u4 offset) const {
359 return (T)offset_to_buffered_address(offset);
360 }
361
362 public:
363 ArchiveBuilder();
364 ~ArchiveBuilder();
365
366 int entropy();
367 void gather_klasses_and_symbols();
368 void gather_source_objs();
369 bool gather_klass_and_symbol(MetaspaceClosure::Ref* ref, bool read_only);
370 bool gather_one_source_obj(MetaspaceClosure::Ref* ref, bool read_only);
371 void remember_embedded_pointer_in_enclosing_obj(MetaspaceClosure::Ref* ref);
372 static void serialize_dynamic_archivable_items(SerializeClosure* soc);
373
374 DumpRegion* rw_region() { return &_rw_region; }
375 DumpRegion* ro_region() { return &_ro_region; }
376 DumpRegion* cc_region() { return &_cc_region; }
377
378 void start_cc_region();
379 void end_cc_region();
380
381 static char* rw_region_alloc(size_t num_bytes) {
382 return current()->rw_region()->allocate(num_bytes);
383 }
384 static char* ro_region_alloc(size_t num_bytes) {
385 return current()->ro_region()->allocate(num_bytes);
386 }
387 static char* cc_region_alloc(size_t num_bytes) {
388 return current()->cc_region()->allocate(num_bytes);
389 }
390
391 template <typename T>
392 static Array<T>* new_ro_array(int length) {
393 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
394 Array<T>* array = (Array<T>*)ro_region_alloc(byte_size);
395 array->initialize(length);
396 return array;
397 }
398
399 template <typename T>
400 static Array<T>* new_rw_array(int length) {
401 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
402 Array<T>* array = (Array<T>*)rw_region_alloc(byte_size);
403 array->initialize(length);
404 return array;
405 }
406
407 template <typename T>
408 static size_t ro_array_bytesize(int length) {
409 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
410 return align_up(byte_size, SharedSpaceObjectAlignment);
411 }
412
413 char* ro_strdup(const char* s);
414
415 static int compare_src_objs(SourceObjInfo** a, SourceObjInfo** b);
416 void sort_metadata_objs();
417 void dump_rw_metadata();
418 void dump_ro_metadata();
419 void relocate_metaspaceobj_embedded_pointers();
420 void record_regenerated_object(address orig_src_obj, address regen_src_obj);
421 void make_klasses_shareable();
422 void make_training_data_shareable();
423 void relocate_to_requested();
424 void write_archive(FileMapInfo* mapinfo, ArchiveHeapInfo* heap_info);
425 void write_region(FileMapInfo* mapinfo, int region_idx, DumpRegion* dump_region,
426 bool read_only, bool allow_exec);
427
428 void write_pointer_in_buffer(address* ptr_location, address src_addr);
429 template <typename T> void write_pointer_in_buffer(T* ptr_location, T src_addr) {
430 write_pointer_in_buffer((address*)ptr_location, (address)src_addr);
431 }
432
433 void mark_and_relocate_to_buffered_addr(address* ptr_location);
434 template <typename T> void mark_and_relocate_to_buffered_addr(T ptr_location) {
435 mark_and_relocate_to_buffered_addr((address*)ptr_location);
436 }
437
438 bool has_been_archived(address src_addr) const;
439
440 bool has_been_buffered(address src_addr) const;
441 template <typename T> bool has_been_buffered(T src_addr) const {
442 return has_been_buffered((address)src_addr);
443 }
444 address get_buffered_addr(address src_addr) const;
445 template <typename T> T get_buffered_addr(T src_addr) const {
446 CDS_ONLY(return (T)get_buffered_addr((address)src_addr);)
447 NOT_CDS(return nullptr;)
448 }
449
450 address get_source_addr(address buffered_addr) const;
451 template <typename T> T get_source_addr(T buffered_addr) const {
452 return (T)get_source_addr((address)buffered_addr);
453 }
454
455 // All klasses and symbols that will be copied into the archive
456 GrowableArray<Klass*>* klasses() const { return _klasses; }
457 GrowableArray<Symbol*>* symbols() const { return _symbols; }
458
459 static bool is_active() {
460 CDS_ONLY(return (_current != nullptr));
461 NOT_CDS(return false;)
462 }
463
464 static ArchiveBuilder* current() {
465 assert(_current != nullptr, "ArchiveBuilder must be active");
466 return _current;
467 }
468
469 static DumpAllocStats* alloc_stats() {
470 return &(current()->_alloc_stats);
471 }
472
473 static CompactHashtableStats* symbol_stats() {
474 return alloc_stats()->symbol_stats();
475 }
476
477 static CompactHashtableStats* string_stats() {
478 return alloc_stats()->string_stats();
479 }
480
481 narrowKlass get_requested_narrow_klass(Klass* k);
482
483 static Klass* get_buffered_klass(Klass* src_klass) {
484 Klass* klass = (Klass*)current()->get_buffered_addr((address)src_klass);
485 assert(klass != nullptr && klass->is_klass(), "must be");
486 return klass;
487 }
488
489 static Symbol* get_buffered_symbol(Symbol* src_symbol) {
490 return (Symbol*)current()->get_buffered_addr((address)src_symbol);
491 }
492
493 void print_stats();
494 void report_out_of_space(const char* name, size_t needed_bytes);
495
496 #ifdef _LP64
497 // The CDS archive contains pre-computed narrow Klass IDs. It carries them in the headers of
498 // archived heap objects. With +UseCompactObjectHeaders, it also carries them in prototypes
499 // in Klass.
500 // When generating the archive, these narrow Klass IDs are computed using the following scheme:
501 // 1) The future encoding base is assumed to point to the first address of the generated mapping.
502 // That means that at runtime, the narrow Klass encoding must be set up with base pointing to
503 // the start address of the mapped CDS metadata archive (wherever that may be). This precludes
504 // zero-based encoding.
505 // 2) The shift must be large enough to result in an encoding range that covers the future assumed
506 // runtime Klass range. That future Klass range will contain both the CDS metadata archive and
507 // the future runtime class space. Since we do not know the size of the future class space, we
508 // need to chose an encoding base/shift combination that will result in a "large enough" size.
509 // The details depend on whether we use compact object headers or legacy object headers.
510 // In Legacy Mode, a narrow Klass ID is 32 bit. This gives us an encoding range size of 4G even
511 // with shift = 0, which is all we need. Therefore, we use a shift=0 for pre-calculating the
512 // narrow Klass IDs.
513 // TinyClassPointer Mode:
514 // We use the highest possible shift value to maximize the encoding range size.
515 static int precomputed_narrow_klass_shift();
516 #endif // _LP64
517
518 };
519
520 #endif // SHARE_CDS_ARCHIVEBUILDER_HPP