1 /*
  2  * Copyright (c) 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_ARCHIVEHEAPWRITER_HPP
 26 #define SHARE_CDS_ARCHIVEHEAPWRITER_HPP
 27 
 28 #include "cds/heapShared.hpp"
 29 #include "memory/allocation.hpp"
 30 #include "memory/allStatic.hpp"
 31 #include "oops/oopHandle.hpp"
 32 #include "utilities/bitMap.hpp"
 33 #include "utilities/exceptions.hpp"
 34 #include "utilities/growableArray.hpp"
 35 #include "utilities/macros.hpp"
 36 #include "utilities/resourceHash.hpp"
 37 
 38 class MemRegion;
 39 
 40 class ArchiveHeapInfo {
 41   MemRegion _buffer_region;             // Contains the archived objects to be written into the CDS archive.
 42   CHeapBitMap _oopmap;
 43   CHeapBitMap _ptrmap;
 44   HeapRootSegments _heap_root_segments;
 45 
 46 public:
 47   ArchiveHeapInfo() : _buffer_region(), _oopmap(128, mtClassShared), _ptrmap(128, mtClassShared) {}
 48   bool is_used() { return !_buffer_region.is_empty(); }
 49 
 50   MemRegion buffer_region() { return _buffer_region; }
 51   void set_buffer_region(MemRegion r) { _buffer_region = r; }
 52 
 53   char* buffer_start() { return (char*)_buffer_region.start(); }
 54   size_t buffer_byte_size() { return _buffer_region.byte_size();    }
 55 
 56   CHeapBitMap* oopmap() { return &_oopmap; }
 57   CHeapBitMap* ptrmap() { return &_ptrmap; }
 58 
 59   void set_heap_root_segments(HeapRootSegments segments) { _heap_root_segments = segments; };
 60   HeapRootSegments heap_root_segments() { return _heap_root_segments; }
 61 };
 62 
 63 #if INCLUDE_CDS_JAVA_HEAP
 64 class ArchiveHeapWriter : AllStatic {
 65   // ArchiveHeapWriter manipulates three types of addresses:
 66   //
 67   //     "source" vs "buffered" vs "requested"
 68   //
 69   // (Note: the design and convention is the same as for the archiving of Metaspace objects.
 70   //  See archiveBuilder.hpp.)
 71   //
 72   // - "source objects" are regular Java objects allocated during the execution
 73   //   of "java -Xshare:dump". They can be used as regular oops.
 74   //
 75   //   HeapShared::archive_objects() recursively searches for the oops that need to be
 76   //   stored into the CDS archive. These are entered into HeapShared::archived_object_cache().
 77   //
 78   // - "buffered objects" are copies of the "source objects", and are stored in into
 79   //   ArchiveHeapWriter::_buffer, which is a GrowableArray that sits outside of
 80   //   the valid heap range. Therefore we avoid using the addresses of these copies
 81   //   as oops. They are usually called "buffered_addr" in the code (of the type "address").
 82   //
 83   //   The buffered objects are stored contiguously, possibly with interleaving fillers
 84   //   to make sure no objects span across boundaries of MIN_GC_REGION_ALIGNMENT.
 85   //
 86   // - Each archived object has a "requested address" -- at run time, if the object
 87   //   can be mapped at this address, we can avoid relocation.
 88   //
 89   // The requested address is implemented differently depending on UseCompressedOops:
 90   //
 91   // UseCompressedOops == true:
 92   //   The archived objects are stored assuming that the runtime COOPS compression
 93   //   scheme is exactly the same as in dump time (or else a more expensive runtime relocation
 94   //   would be needed.)
 95   //
 96   //   At dump time, we assume that the runtime heap range is exactly the same as
 97   //   in dump time. The requested addresses of the archived objects are chosen such that
 98   //   they would occupy the top end of a G1 heap (TBD when dumping is supported by other
 99   //   collectors. See JDK-8298614).
100   //
101   // UseCompressedOops == false:
102   //   At runtime, the heap range is usually picked (randomly) by the OS, so we will almost always
103   //   need to perform relocation. Hence, the goal of the "requested address" is to ensure that
104   //   the contents of the archived objects are deterministic. I.e., the oop fields of archived
105   //   objects will always point to deterministic addresses.
106   //
107   //   For G1, the archived heap is written such that the lowest archived object is placed
108   //   at NOCOOPS_REQUESTED_BASE. (TBD after JDK-8298614).
109   // ----------------------------------------------------------------------
110 
111 public:
112   static const intptr_t NOCOOPS_REQUESTED_BASE = 0x10000000;
113 
114   // The minimum region size of all collectors that are supported by CDS.
115   // G1 heap region size can never be smaller than 1M.
116   // Shenandoah heap region size can never be smaller than 256K.
117   static constexpr int MIN_GC_REGION_ALIGNMENT = 256 * K;
118 
119 private:
120   class EmbeddedOopRelocator;
121   struct NativePointerInfo {
122     oop _src_obj;
123     int _field_offset;
124   };
125 
126   static GrowableArrayCHeap<u1, mtClassShared>* _buffer;
127 
128   // The number of bytes that have written into _buffer (may be smaller than _buffer->length()).
129   static size_t _buffer_used;
130 
131   // The heap root segments information.
132   static HeapRootSegments _heap_root_segments;
133 
134   // The address range of the requested location of the archived heap objects.
135   static address _requested_bottom;
136   static address _requested_top;
137 
138   static GrowableArrayCHeap<NativePointerInfo, mtClassShared>* _native_pointers;
139   static GrowableArrayCHeap<oop, mtClassShared>* _source_objs;
140 
141   // We sort _source_objs_order to minimize the number of bits in ptrmap and oopmap.
142   // See comments near the body of ArchiveHeapWriter::compare_objs_by_oop_fields().
143   // The objects will be written in the order of:
144   //_source_objs->at(_source_objs_order->at(0)._index)
145   // source_objs->at(_source_objs_order->at(1)._index)
146   // source_objs->at(_source_objs_order->at(2)._index)
147   // ...
148   struct HeapObjOrder {
149     int _index;    // The location of this object in _source_objs
150     int _rank;     // A lower rank means the object will be written at a lower location.
151   };
152   static GrowableArrayCHeap<HeapObjOrder, mtClassShared>* _source_objs_order;
153 
154   typedef ResizeableResourceHashtable<size_t, oop,
155       AnyObj::C_HEAP,
156       mtClassShared> BufferOffsetToSourceObjectTable;
157   static BufferOffsetToSourceObjectTable* _buffer_offset_to_source_obj_table;
158 
159   static void allocate_buffer();
160   static void ensure_buffer_space(size_t min_bytes);
161 
162   // Both Java bytearray and GrowableArraty use int indices and lengths. Do a safe typecast with range check
163   static int to_array_index(size_t i) {
164     assert(i <= (size_t)max_jint, "must be");
165     return (int)i;
166   }
167   static int to_array_length(size_t n) {
168     return to_array_index(n);
169   }
170 
171   template <typename T> static T offset_to_buffered_address(size_t offset) {
172     return (T)(_buffer->adr_at(to_array_index(offset)));
173   }
174 
175   static address buffer_bottom() {
176     return offset_to_buffered_address<address>(0);
177   }
178 
179   // The exclusive end of the last object that was copied into the buffer.
180   static address buffer_top() {
181     return buffer_bottom() + _buffer_used;
182   }
183 
184   static bool in_buffer(address buffered_addr) {
185     return (buffer_bottom() <= buffered_addr) && (buffered_addr < buffer_top());
186   }
187 
188   static size_t buffered_address_to_offset(address buffered_addr) {
189     assert(in_buffer(buffered_addr), "sanity");
190     return buffered_addr - buffer_bottom();
191   }
192 
193   static void root_segment_at_put(objArrayOop segment, int index, oop root);
194   static objArrayOop allocate_root_segment(size_t offset, int element_count);
195   static void copy_roots_to_buffer(GrowableArrayCHeap<oop, mtClassShared>* roots);
196   static void copy_source_objs_to_buffer(GrowableArrayCHeap<oop, mtClassShared>* roots);
197   static size_t copy_one_source_obj_to_buffer(oop src_obj);
198 
199   static void maybe_fill_gc_region_gap(size_t required_byte_size);
200   static size_t filler_array_byte_size(int length);
201   static int filler_array_length(size_t fill_bytes);
202   static HeapWord* init_filler_array_at_buffer_top(int array_length, size_t fill_bytes);
203 
204   static void set_requested_address(ArchiveHeapInfo* info);
205   static void relocate_embedded_oops(GrowableArrayCHeap<oop, mtClassShared>* roots, ArchiveHeapInfo* info);
206   static void compute_ptrmap(ArchiveHeapInfo *info);
207   static bool is_in_requested_range(oop o);
208   static oop requested_obj_from_buffer_offset(size_t offset);
209 
210   static oop load_oop_from_buffer(oop* buffered_addr);
211   static oop load_oop_from_buffer(narrowOop* buffered_addr);
212   inline static void store_oop_in_buffer(oop* buffered_addr, oop requested_obj);
213   inline static void store_oop_in_buffer(narrowOop* buffered_addr, oop requested_obj);
214 
215   template <typename T> static oop load_source_oop_from_buffer(T* buffered_addr);
216   template <typename T> static void store_requested_oop_in_buffer(T* buffered_addr, oop request_oop);
217 
218   template <typename T> static T* requested_addr_to_buffered_addr(T* p);
219   template <typename T> static void relocate_field_in_buffer(T* field_addr_in_buffer, CHeapBitMap* oopmap);
220   template <typename T> static void mark_oop_pointer(T* buffered_addr, CHeapBitMap* oopmap);
221 
222   static void update_header_for_requested_obj(oop requested_obj, oop src_obj, Klass* src_klass);
223 
224   static int compare_objs_by_oop_fields(HeapObjOrder* a, HeapObjOrder* b);
225   static void sort_source_objs();
226 
227 public:
228   static void init() NOT_CDS_JAVA_HEAP_RETURN;
229   static void add_source_obj(oop src_obj);
230   static bool is_too_large_to_archive(size_t size);
231   static bool is_too_large_to_archive(oop obj);
232   static bool is_string_too_large_to_archive(oop string);
233   static void write(GrowableArrayCHeap<oop, mtClassShared>*, ArchiveHeapInfo* heap_info);
234   static address requested_address();  // requested address of the lowest achived heap object
235   static size_t get_filler_size_at(address buffered_addr);
236 
237   static void mark_native_pointer(oop src_obj, int offset);
238   static bool is_marked_as_native_pointer(ArchiveHeapInfo* heap_info, oop src_obj, int field_offset);
239   static oop source_obj_to_requested_obj(oop src_obj);
240   static oop buffered_addr_to_source_obj(address buffered_addr);
241   static address buffered_addr_to_requested_addr(address buffered_addr);
242 
243   // Archived heap object headers carry pre-computed narrow Klass ids calculated with the
244   // following scheme:
245   // 1) the encoding base must be the mapping start address.
246   // 2) shift must be large enough to result in an encoding range that covers the runtime Klass range.
247   //    That Klass range is defined by CDS archive size and runtime class space size. Luckily, the maximum
248   //    size can be predicted: archive size is assumed to be <1G, class space size capped at 3G, and at
249   //    runtime we put both regions adjacent to each other. Therefore, runtime Klass range size < 4G.
250   //    Since nKlass itself is 32 bit, our encoding range len is 4G, and since we set the base directly
251   //    at mapping start, these 4G are enough. Therefore, we don't need to shift at all (shift=0).
252   static constexpr int precomputed_narrow_klass_shift = 0;
253 
254 };
255 #endif // INCLUDE_CDS_JAVA_HEAP
256 #endif // SHARE_CDS_ARCHIVEHEAPWRITER_HPP