1 /*
   2  * Copyright (c) 2005, 2024, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2023, Alibaba Group Holding Limited. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "classfile/classLoaderData.inline.hpp"
  28 #include "classfile/classLoaderDataGraph.hpp"
  29 #include "classfile/javaClasses.inline.hpp"
  30 #include "classfile/symbolTable.hpp"
  31 #include "classfile/vmClasses.hpp"
  32 #include "classfile/vmSymbols.hpp"
  33 #include "gc/shared/gcLocker.hpp"
  34 #include "gc/shared/gcVMOperations.hpp"
  35 #include "gc/shared/workerThread.hpp"
  36 #include "jfr/jfrEvents.hpp"
  37 #include "jvm.h"
  38 #include "memory/allocation.inline.hpp"
  39 #include "memory/resourceArea.hpp"
  40 #include "memory/universe.hpp"
  41 #include "oops/fieldStreams.inline.hpp"
  42 #include "oops/klass.inline.hpp"
  43 #include "oops/objArrayKlass.hpp"
  44 #include "oops/objArrayOop.inline.hpp"
  45 #include "oops/flatArrayKlass.hpp"
  46 #include "oops/flatArrayOop.inline.hpp"
  47 #include "oops/oop.inline.hpp"
  48 #include "oops/typeArrayOop.inline.hpp"
  49 #include "runtime/continuationWrapper.inline.hpp"
  50 #include "runtime/fieldDescriptor.inline.hpp"
  51 #include "runtime/frame.inline.hpp"
  52 #include "runtime/handles.inline.hpp"
  53 #include "runtime/javaCalls.hpp"
  54 #include "runtime/javaThread.inline.hpp"
  55 #include "runtime/jniHandles.hpp"
  56 #include "runtime/os.hpp"
  57 #include "runtime/threads.hpp"
  58 #include "runtime/threadSMR.hpp"
  59 #include "runtime/vframe.hpp"
  60 #include "runtime/vmOperations.hpp"
  61 #include "runtime/vmThread.hpp"
  62 #include "runtime/timerTrace.hpp"
  63 #include "services/heapDumper.hpp"
  64 #include "services/heapDumperCompression.hpp"
  65 #include "services/threadService.hpp"
  66 #include "utilities/checkedCast.hpp"
  67 #include "utilities/macros.hpp"
  68 #include "utilities/ostream.hpp"
  69 #ifdef LINUX
  70 #include "os_linux.hpp"
  71 #endif
  72 
  73 /*
  74  * HPROF binary format - description copied from:
  75  *   src/share/demo/jvmti/hprof/hprof_io.c
  76  *
  77  *
  78  *  header    "JAVA PROFILE 1.0.2" (0-terminated)
  79  *
  80  *  u4        size of identifiers. Identifiers are used to represent
  81  *            UTF8 strings, objects, stack traces, etc. They usually
  82  *            have the same size as host pointers.
  83  * u4         high word
  84  * u4         low word    number of milliseconds since 0:00 GMT, 1/1/70
  85  * [record]*  a sequence of records.
  86  *
  87  *
  88  * Record format:
  89  *
  90  * u1         a TAG denoting the type of the record
  91  * u4         number of *microseconds* since the time stamp in the
  92  *            header. (wraps around in a little more than an hour)
  93  * u4         number of bytes *remaining* in the record. Note that
  94  *            this number excludes the tag and the length field itself.
  95  * [u1]*      BODY of the record (a sequence of bytes)
  96  *
  97  *
  98  * The following TAGs are supported:
  99  *
 100  * TAG           BODY       notes
 101  *----------------------------------------------------------
 102  * HPROF_UTF8               a UTF8-encoded name
 103  *
 104  *               id         name ID
 105  *               [u1]*      UTF8 characters (no trailing zero)
 106  *
 107  * HPROF_LOAD_CLASS         a newly loaded class
 108  *
 109  *                u4        class serial number (> 0)
 110  *                id        class object ID
 111  *                u4        stack trace serial number
 112  *                id        class name ID
 113  *
 114  * HPROF_UNLOAD_CLASS       an unloading class
 115  *
 116  *                u4        class serial_number
 117  *
 118  * HPROF_FRAME              a Java stack frame
 119  *
 120  *                id        stack frame ID
 121  *                id        method name ID
 122  *                id        method signature ID
 123  *                id        source file name ID
 124  *                u4        class serial number
 125  *                i4        line number. >0: normal
 126  *                                       -1: unknown
 127  *                                       -2: compiled method
 128  *                                       -3: native method
 129  *
 130  * HPROF_TRACE              a Java stack trace
 131  *
 132  *               u4         stack trace serial number
 133  *               u4         thread serial number
 134  *               u4         number of frames
 135  *               [id]*      stack frame IDs
 136  *
 137  *
 138  * HPROF_ALLOC_SITES        a set of heap allocation sites, obtained after GC
 139  *
 140  *               u2         flags 0x0001: incremental vs. complete
 141  *                                0x0002: sorted by allocation vs. live
 142  *                                0x0004: whether to force a GC
 143  *               u4         cutoff ratio
 144  *               u4         total live bytes
 145  *               u4         total live instances
 146  *               u8         total bytes allocated
 147  *               u8         total instances allocated
 148  *               u4         number of sites that follow
 149  *               [u1        is_array: 0:  normal object
 150  *                                    2:  object array
 151  *                                    4:  boolean array
 152  *                                    5:  char array
 153  *                                    6:  float array
 154  *                                    7:  double array
 155  *                                    8:  byte array
 156  *                                    9:  short array
 157  *                                    10: int array
 158  *                                    11: long array
 159  *                u4        class serial number (may be zero during startup)
 160  *                u4        stack trace serial number
 161  *                u4        number of bytes alive
 162  *                u4        number of instances alive
 163  *                u4        number of bytes allocated
 164  *                u4]*      number of instance allocated
 165  *
 166  * HPROF_START_THREAD       a newly started thread.
 167  *
 168  *               u4         thread serial number (> 0)
 169  *               id         thread object ID
 170  *               u4         stack trace serial number
 171  *               id         thread name ID
 172  *               id         thread group name ID
 173  *               id         thread group parent name ID
 174  *
 175  * HPROF_END_THREAD         a terminating thread.
 176  *
 177  *               u4         thread serial number
 178  *
 179  * HPROF_HEAP_SUMMARY       heap summary
 180  *
 181  *               u4         total live bytes
 182  *               u4         total live instances
 183  *               u8         total bytes allocated
 184  *               u8         total instances allocated
 185  *
 186  * HPROF_HEAP_DUMP          denote a heap dump
 187  *
 188  *               [heap dump sub-records]*
 189  *
 190  *                          There are four kinds of heap dump sub-records:
 191  *
 192  *               u1         sub-record type
 193  *
 194  *               HPROF_GC_ROOT_UNKNOWN         unknown root
 195  *
 196  *                          id         object ID
 197  *
 198  *               HPROF_GC_ROOT_THREAD_OBJ      thread object
 199  *
 200  *                          id         thread object ID  (may be 0 for a
 201  *                                     thread newly attached through JNI)
 202  *                          u4         thread sequence number
 203  *                          u4         stack trace sequence number
 204  *
 205  *               HPROF_GC_ROOT_JNI_GLOBAL      JNI global ref root
 206  *
 207  *                          id         object ID
 208  *                          id         JNI global ref ID
 209  *
 210  *               HPROF_GC_ROOT_JNI_LOCAL       JNI local ref
 211  *
 212  *                          id         object ID
 213  *                          u4         thread serial number
 214  *                          u4         frame # in stack trace (-1 for empty)
 215  *
 216  *               HPROF_GC_ROOT_JAVA_FRAME      Java stack frame
 217  *
 218  *                          id         object ID
 219  *                          u4         thread serial number
 220  *                          u4         frame # in stack trace (-1 for empty)
 221  *
 222  *               HPROF_GC_ROOT_NATIVE_STACK    Native stack
 223  *
 224  *                          id         object ID
 225  *                          u4         thread serial number
 226  *
 227  *               HPROF_GC_ROOT_STICKY_CLASS    System class
 228  *
 229  *                          id         object ID
 230  *
 231  *               HPROF_GC_ROOT_THREAD_BLOCK    Reference from thread block
 232  *
 233  *                          id         object ID
 234  *                          u4         thread serial number
 235  *
 236  *               HPROF_GC_ROOT_MONITOR_USED    Busy monitor
 237  *
 238  *                          id         object ID
 239  *
 240  *               HPROF_GC_CLASS_DUMP           dump of a class object
 241  *
 242  *                          id         class object ID
 243  *                          u4         stack trace serial number
 244  *                          id         super class object ID
 245  *                          id         class loader object ID
 246  *                          id         signers object ID
 247  *                          id         protection domain object ID
 248  *                          id         reserved
 249  *                          id         reserved
 250  *
 251  *                          u4         instance size (in bytes)
 252  *
 253  *                          u2         size of constant pool
 254  *                          [u2,       constant pool index,
 255  *                           ty,       type
 256  *                                     2:  object
 257  *                                     4:  boolean
 258  *                                     5:  char
 259  *                                     6:  float
 260  *                                     7:  double
 261  *                                     8:  byte
 262  *                                     9:  short
 263  *                                     10: int
 264  *                                     11: long
 265  *                           vl]*      and value
 266  *
 267  *                          u2         number of static fields
 268  *                          [id,       static field name,
 269  *                           ty,       type,
 270  *                           vl]*      and value
 271  *
 272  *                          u2         number of inst. fields (not inc. super)
 273  *                          [id,       instance field name,
 274  *                           ty]*      type
 275  *
 276  *               HPROF_GC_INSTANCE_DUMP        dump of a normal object
 277  *
 278  *                          id         object ID
 279  *                          u4         stack trace serial number
 280  *                          id         class object ID
 281  *                          u4         number of bytes that follow
 282  *                          [vl]*      instance field values (class, followed
 283  *                                     by super, super's super ...)
 284  *
 285  *               HPROF_GC_OBJ_ARRAY_DUMP       dump of an object array
 286  *
 287  *                          id         array object ID
 288  *                          u4         stack trace serial number
 289  *                          u4         number of elements
 290  *                          id         array class ID
 291  *                          [id]*      elements
 292  *
 293  *               HPROF_GC_PRIM_ARRAY_DUMP      dump of a primitive array
 294  *
 295  *                          id         array object ID
 296  *                          u4         stack trace serial number
 297  *                          u4         number of elements
 298  *                          u1         element type
 299  *                                     4:  boolean array
 300  *                                     5:  char array
 301  *                                     6:  float array
 302  *                                     7:  double array
 303  *                                     8:  byte array
 304  *                                     9:  short array
 305  *                                     10: int array
 306  *                                     11: long array
 307  *                          [u1]*      elements
 308  *
 309  * HPROF_CPU_SAMPLES        a set of sample traces of running threads
 310  *
 311  *                u4        total number of samples
 312  *                u4        # of traces
 313  *               [u4        # of samples
 314  *                u4]*      stack trace serial number
 315  *
 316  * HPROF_CONTROL_SETTINGS   the settings of on/off switches
 317  *
 318  *                u4        0x00000001: alloc traces on/off
 319  *                          0x00000002: cpu sampling on/off
 320  *                u2        stack trace depth
 321  *
 322  * HPROF_FLAT_ARRAYS        list of flat arrays
 323  *
 324  *               [flat array sub-records]*
 325  *
 326  *               HPROF_FLAT_ARRAY      flat array
 327  *
 328  *                          id         array object ID (dumped as HPROF_GC_PRIM_ARRAY_DUMP)
 329  *                          id         element class ID (dumped by HPROF_GC_CLASS_DUMP)
 330  *
 331  * HPROF_INLINED_FIELDS     decribes inlined fields
 332  *
 333  *               [class with inlined fields sub-records]*
 334  *
 335  *               HPROF_CLASS_WITH_INLINED_FIELDS
 336  *
 337  *                          id         class ID (dumped as HPROF_GC_CLASS_DUMP)
 338  *
 339  *                          u2         number of instance inlined fields (not including super)
 340  *                          [u2,       inlined field index,
 341  *                           u2,       synthetic field count,
 342  *                           id,       original field name,
 343  *                           id]*      inlined field class ID (dumped by HPROF_GC_CLASS_DUMP)
 344  *
 345  * When the header is "JAVA PROFILE 1.0.2" a heap dump can optionally
 346  * be generated as a sequence of heap dump segments. This sequence is
 347  * terminated by an end record. The additional tags allowed by format
 348  * "JAVA PROFILE 1.0.2" are:
 349  *
 350  * HPROF_HEAP_DUMP_SEGMENT  denote a heap dump segment
 351  *
 352  *               [heap dump sub-records]*
 353  *               The same sub-record types allowed by HPROF_HEAP_DUMP
 354  *
 355  * HPROF_HEAP_DUMP_END      denotes the end of a heap dump
 356  *
 357  */
 358 
 359 
 360 // HPROF tags
 361 
 362 enum hprofTag : u1 {
 363   // top-level records
 364   HPROF_UTF8                    = 0x01,
 365   HPROF_LOAD_CLASS              = 0x02,
 366   HPROF_UNLOAD_CLASS            = 0x03,
 367   HPROF_FRAME                   = 0x04,
 368   HPROF_TRACE                   = 0x05,
 369   HPROF_ALLOC_SITES             = 0x06,
 370   HPROF_HEAP_SUMMARY            = 0x07,
 371   HPROF_START_THREAD            = 0x0A,
 372   HPROF_END_THREAD              = 0x0B,
 373   HPROF_HEAP_DUMP               = 0x0C,
 374   HPROF_CPU_SAMPLES             = 0x0D,
 375   HPROF_CONTROL_SETTINGS        = 0x0E,
 376 
 377   // 1.0.2 record types
 378   HPROF_HEAP_DUMP_SEGMENT       = 0x1C,
 379   HPROF_HEAP_DUMP_END           = 0x2C,
 380 
 381   // inlined object support
 382   HPROF_FLAT_ARRAYS             = 0x12,
 383   HPROF_INLINED_FIELDS          = 0x13,
 384   // inlined object subrecords
 385   HPROF_FLAT_ARRAY                  = 0x01,
 386   HPROF_CLASS_WITH_INLINED_FIELDS   = 0x01,
 387 
 388   // field types
 389   HPROF_ARRAY_OBJECT            = 0x01,
 390   HPROF_NORMAL_OBJECT           = 0x02,
 391   HPROF_BOOLEAN                 = 0x04,
 392   HPROF_CHAR                    = 0x05,
 393   HPROF_FLOAT                   = 0x06,
 394   HPROF_DOUBLE                  = 0x07,
 395   HPROF_BYTE                    = 0x08,
 396   HPROF_SHORT                   = 0x09,
 397   HPROF_INT                     = 0x0A,
 398   HPROF_LONG                    = 0x0B,
 399 
 400   // data-dump sub-records
 401   HPROF_GC_ROOT_UNKNOWN         = 0xFF,
 402   HPROF_GC_ROOT_JNI_GLOBAL      = 0x01,
 403   HPROF_GC_ROOT_JNI_LOCAL       = 0x02,
 404   HPROF_GC_ROOT_JAVA_FRAME      = 0x03,
 405   HPROF_GC_ROOT_NATIVE_STACK    = 0x04,
 406   HPROF_GC_ROOT_STICKY_CLASS    = 0x05,
 407   HPROF_GC_ROOT_THREAD_BLOCK    = 0x06,
 408   HPROF_GC_ROOT_MONITOR_USED    = 0x07,
 409   HPROF_GC_ROOT_THREAD_OBJ      = 0x08,
 410   HPROF_GC_CLASS_DUMP           = 0x20,
 411   HPROF_GC_INSTANCE_DUMP        = 0x21,
 412   HPROF_GC_OBJ_ARRAY_DUMP       = 0x22,
 413   HPROF_GC_PRIM_ARRAY_DUMP      = 0x23
 414 };
 415 
 416 // Default stack trace ID (used for dummy HPROF_TRACE record)
 417 enum {
 418   STACK_TRACE_ID = 1,
 419   INITIAL_CLASS_COUNT = 200
 420 };
 421 
 422 
 423 class AbstractDumpWriter;
 424 
 425 class InlinedObjects {
 426 
 427   struct ClassInlinedFields {
 428     const Klass *klass;
 429     uintx base_index;   // base index of the inlined field names (1st field has index base_index+1).
 430     ClassInlinedFields(const Klass *klass = nullptr, uintx base_index = 0) : klass(klass), base_index(base_index) {}
 431 
 432     // For GrowableArray::find_sorted().
 433     static int compare(const ClassInlinedFields& a, const ClassInlinedFields& b) {
 434       return a.klass - b.klass;
 435     }
 436     // For GrowableArray::sort().
 437     static int compare(ClassInlinedFields* a, ClassInlinedFields* b) {
 438       return compare(*a, *b);
 439     }
 440   };
 441 
 442   uintx _min_string_id;
 443   uintx _max_string_id;
 444 
 445   GrowableArray<ClassInlinedFields> *_inlined_field_map;
 446 
 447   // counters for classes with inlined fields and for the fields
 448   int _classes_count;
 449   int _inlined_fields_count;
 450 
 451   static InlinedObjects *_instance;
 452 
 453   static void inlined_field_names_callback(InlinedObjects* _this, const Klass *klass, uintx base_index, int count);
 454 
 455   GrowableArray<oop> *_flat_arrays;
 456 
 457 public:
 458   InlinedObjects()
 459     : _min_string_id(0), _max_string_id(0),
 460     _inlined_field_map(nullptr),
 461     _classes_count(0), _inlined_fields_count(0),
 462     _flat_arrays(nullptr) {
 463   }
 464 
 465   static InlinedObjects* get_instance() {
 466     return _instance;
 467   }
 468 
 469   void init();
 470   void release();
 471 
 472   void dump_inlined_field_names(AbstractDumpWriter *writer);
 473 
 474   uintx get_base_index_for(Klass* k);
 475   uintx get_next_string_id(uintx id);
 476 
 477   void dump_classed_with_inlined_fields(AbstractDumpWriter* writer);
 478 
 479   void add_flat_array(oop array);
 480   void dump_flat_arrays(AbstractDumpWriter* writer);
 481 
 482 };
 483 
 484 InlinedObjects *InlinedObjects::_instance = nullptr;
 485 
 486 
 487 // Supports I/O operations for a dump
 488 // Base class for dump and parallel dump
 489 class AbstractDumpWriter : public CHeapObj<mtInternal> {
 490  protected:
 491   enum {
 492     io_buffer_max_size = 1*M,
 493     dump_segment_header_size = 9
 494   };
 495 
 496   char* _buffer;    // internal buffer
 497   size_t _size;
 498   size_t _pos;
 499 
 500   bool _in_dump_segment; // Are we currently in a dump segment?
 501   bool _is_huge_sub_record; // Are we writing a sub-record larger than the buffer size?
 502   DEBUG_ONLY(size_t _sub_record_left;) // The bytes not written for the current sub-record.
 503   DEBUG_ONLY(bool _sub_record_ended;) // True if we have called the end_sub_record().
 504 
 505   char* buffer() const                          { return _buffer; }
 506   size_t buffer_size() const                    { return _size; }
 507   void set_position(size_t pos)                 { _pos = pos; }
 508 
 509   // Can be called if we have enough room in the buffer.
 510   void write_fast(const void* s, size_t len);
 511 
 512   // Returns true if we have enough room in the buffer for 'len' bytes.
 513   bool can_write_fast(size_t len);
 514 
 515   void write_address(address a);
 516 
 517  public:
 518   AbstractDumpWriter() :
 519     _buffer(nullptr),
 520     _size(io_buffer_max_size),
 521     _pos(0),
 522     _in_dump_segment(false) { }
 523 
 524   // Total number of bytes written to the disk
 525   virtual julong bytes_written() const = 0;
 526   // Return non-null if error occurred
 527   virtual char const* error() const = 0;
 528 
 529   size_t position() const                       { return _pos; }
 530   // writer functions
 531   virtual void write_raw(const void* s, size_t len);
 532   void write_u1(u1 x);
 533   void write_u2(u2 x);
 534   void write_u4(u4 x);
 535   void write_u8(u8 x);
 536   void write_objectID(oop o);
 537   void write_rootID(oop* p);
 538   void write_symbolID(Symbol* o);
 539   void write_classID(Klass* k);
 540   void write_id(u4 x);
 541 
 542   // Start a new sub-record. Starts a new heap dump segment if needed.
 543   void start_sub_record(u1 tag, u4 len);
 544   // Ends the current sub-record.
 545   void end_sub_record();
 546   // Finishes the current dump segment if not already finished.
 547   void finish_dump_segment();
 548   // Flush internal buffer to persistent storage
 549   virtual void flush() = 0;
 550 };
 551 
 552 void AbstractDumpWriter::write_fast(const void* s, size_t len) {
 553   assert(!_in_dump_segment || (_sub_record_left >= len), "sub-record too large");
 554   assert(buffer_size() - position() >= len, "Must fit");
 555   debug_only(_sub_record_left -= len);
 556   memcpy(buffer() + position(), s, len);
 557   set_position(position() + len);
 558 }
 559 
 560 bool AbstractDumpWriter::can_write_fast(size_t len) {
 561   return buffer_size() - position() >= len;
 562 }
 563 
 564 // write raw bytes
 565 void AbstractDumpWriter::write_raw(const void* s, size_t len) {
 566   assert(!_in_dump_segment || (_sub_record_left >= len), "sub-record too large");
 567   debug_only(_sub_record_left -= len);
 568 
 569   // flush buffer to make room.
 570   while (len > buffer_size() - position()) {
 571     assert(!_in_dump_segment || _is_huge_sub_record,
 572            "Cannot overflow in non-huge sub-record.");
 573     size_t to_write = buffer_size() - position();
 574     memcpy(buffer() + position(), s, to_write);
 575     s = (void*) ((char*) s + to_write);
 576     len -= to_write;
 577     set_position(position() + to_write);
 578     flush();
 579   }
 580 
 581   memcpy(buffer() + position(), s, len);
 582   set_position(position() + len);
 583 }
 584 
 585 // Makes sure we inline the fast write into the write_u* functions. This is a big speedup.
 586 #define WRITE_KNOWN_TYPE(p, len) do { if (can_write_fast((len))) write_fast((p), (len)); \
 587                                       else write_raw((p), (len)); } while (0)
 588 
 589 void AbstractDumpWriter::write_u1(u1 x) {
 590   WRITE_KNOWN_TYPE(&x, 1);
 591 }
 592 
 593 void AbstractDumpWriter::write_u2(u2 x) {
 594   u2 v;
 595   Bytes::put_Java_u2((address)&v, x);
 596   WRITE_KNOWN_TYPE(&v, 2);
 597 }
 598 
 599 void AbstractDumpWriter::write_u4(u4 x) {
 600   u4 v;
 601   Bytes::put_Java_u4((address)&v, x);
 602   WRITE_KNOWN_TYPE(&v, 4);
 603 }
 604 
 605 void AbstractDumpWriter::write_u8(u8 x) {
 606   u8 v;
 607   Bytes::put_Java_u8((address)&v, x);
 608   WRITE_KNOWN_TYPE(&v, 8);
 609 }
 610 
 611 void AbstractDumpWriter::write_address(address a) {
 612 #ifdef _LP64
 613   write_u8((u8)a);
 614 #else
 615   write_u4((u4)a);
 616 #endif
 617 }
 618 
 619 void AbstractDumpWriter::write_objectID(oop o) {
 620   write_address(cast_from_oop<address>(o));
 621 }
 622 
 623 void AbstractDumpWriter::write_rootID(oop* p) {
 624   write_address((address)p);
 625 }
 626 
 627 void AbstractDumpWriter::write_symbolID(Symbol* s) {
 628   write_address((address)((uintptr_t)s));
 629 }
 630 
 631 void AbstractDumpWriter::write_id(u4 x) {
 632 #ifdef _LP64
 633   write_u8((u8) x);
 634 #else
 635   write_u4(x);
 636 #endif
 637 }
 638 
 639 // We use java mirror as the class ID
 640 void AbstractDumpWriter::write_classID(Klass* k) {
 641   write_objectID(k->java_mirror());
 642 }
 643 
 644 void AbstractDumpWriter::finish_dump_segment() {
 645   if (_in_dump_segment) {
 646     assert(_sub_record_left == 0, "Last sub-record not written completely");
 647     assert(_sub_record_ended, "sub-record must have ended");
 648 
 649     // Fix up the dump segment length if we haven't written a huge sub-record last
 650     // (in which case the segment length was already set to the correct value initially).
 651     if (!_is_huge_sub_record) {
 652       assert(position() > dump_segment_header_size, "Dump segment should have some content");
 653       Bytes::put_Java_u4((address) (buffer() + 5),
 654                          (u4) (position() - dump_segment_header_size));
 655     } else {
 656       // Finish process huge sub record
 657       // Set _is_huge_sub_record to false so the parallel dump writer can flush data to file.
 658       _is_huge_sub_record = false;
 659     }
 660 
 661     _in_dump_segment = false;
 662     flush();
 663   }
 664 }
 665 
 666 void AbstractDumpWriter::start_sub_record(u1 tag, u4 len) {
 667   if (!_in_dump_segment) {
 668     if (position() > 0) {
 669       flush();
 670     }
 671 
 672     assert(position() == 0 && buffer_size() > dump_segment_header_size, "Must be at the start");
 673 
 674     write_u1(HPROF_HEAP_DUMP_SEGMENT);
 675     write_u4(0); // timestamp
 676     // Will be fixed up later if we add more sub-records.  If this is a huge sub-record,
 677     // this is already the correct length, since we don't add more sub-records.
 678     write_u4(len);
 679     assert(Bytes::get_Java_u4((address)(buffer() + 5)) == len, "Inconsistent size!");
 680     _in_dump_segment = true;
 681     _is_huge_sub_record = len > buffer_size() - dump_segment_header_size;
 682   } else if (_is_huge_sub_record || (len > buffer_size() - position())) {
 683     // This object will not fit in completely or the last sub-record was huge.
 684     // Finish the current segment and try again.
 685     finish_dump_segment();
 686     start_sub_record(tag, len);
 687 
 688     return;
 689   }
 690 
 691   debug_only(_sub_record_left = len);
 692   debug_only(_sub_record_ended = false);
 693 
 694   write_u1(tag);
 695 }
 696 
 697 void AbstractDumpWriter::end_sub_record() {
 698   assert(_in_dump_segment, "must be in dump segment");
 699   assert(_sub_record_left == 0, "sub-record not written completely");
 700   assert(!_sub_record_ended, "Must not have ended yet");
 701   debug_only(_sub_record_ended = true);
 702 }
 703 
 704 // Supports I/O operations for a dump
 705 
 706 class DumpWriter : public AbstractDumpWriter {
 707 private:
 708   FileWriter* _writer;
 709   AbstractCompressor* _compressor;
 710   size_t _bytes_written;
 711   char* _error;
 712   // Compression support
 713   char* _out_buffer;
 714   size_t _out_size;
 715   size_t _out_pos;
 716   char* _tmp_buffer;
 717   size_t _tmp_size;
 718 
 719 private:
 720   void do_compress();
 721 
 722 public:
 723   DumpWriter(const char* path, bool overwrite, AbstractCompressor* compressor);
 724   ~DumpWriter();
 725   julong bytes_written() const override        { return (julong) _bytes_written; }
 726   char const* error() const override           { return _error; }
 727   void set_error(const char* error)            { _error = (char*)error; }
 728   bool has_error() const                       { return _error != nullptr; }
 729   const char* get_file_path() const            { return _writer->get_file_path(); }
 730   AbstractCompressor* compressor()             { return _compressor; }
 731   bool is_overwrite() const                    { return _writer->is_overwrite(); }
 732 
 733   void flush() override;
 734 
 735 private:
 736   // internals for DumpMerger
 737   friend class DumpMerger;
 738   void set_bytes_written(julong bytes_written) { _bytes_written = bytes_written; }
 739   int get_fd() const                           { return _writer->get_fd(); }
 740   void set_compressor(AbstractCompressor* p)   { _compressor = p; }
 741 };
 742 
 743 DumpWriter::DumpWriter(const char* path, bool overwrite, AbstractCompressor* compressor) :
 744   AbstractDumpWriter(),
 745   _writer(new (std::nothrow) FileWriter(path, overwrite)),
 746   _compressor(compressor),
 747   _bytes_written(0),
 748   _error(nullptr),
 749   _out_buffer(nullptr),
 750   _out_size(0),
 751   _out_pos(0),
 752   _tmp_buffer(nullptr),
 753   _tmp_size(0) {
 754   _error = (char*)_writer->open_writer();
 755   if (_error == nullptr) {
 756     _buffer = (char*)os::malloc(io_buffer_max_size, mtInternal);
 757     if (compressor != nullptr) {
 758       _error = (char*)_compressor->init(io_buffer_max_size, &_out_size, &_tmp_size);
 759       if (_error == nullptr) {
 760         if (_out_size > 0) {
 761           _out_buffer = (char*)os::malloc(_out_size, mtInternal);
 762         }
 763         if (_tmp_size > 0) {
 764           _tmp_buffer = (char*)os::malloc(_tmp_size, mtInternal);
 765         }
 766       }
 767     }
 768   }
 769   // initialize internal buffer
 770   _pos = 0;
 771   _size = io_buffer_max_size;
 772 }
 773 
 774 DumpWriter::~DumpWriter(){
 775   if (_buffer != nullptr) {
 776     os::free(_buffer);
 777   }
 778   if (_out_buffer != nullptr) {
 779     os::free(_out_buffer);
 780   }
 781   if (_tmp_buffer != nullptr) {
 782     os::free(_tmp_buffer);
 783   }
 784   if (_writer != nullptr) {
 785     delete _writer;
 786   }
 787   _bytes_written = -1;
 788 }
 789 
 790 // flush any buffered bytes to the file
 791 void DumpWriter::flush() {
 792   if (_pos <= 0) {
 793     return;
 794   }
 795   if (has_error()) {
 796     _pos = 0;
 797     return;
 798   }
 799   char* result = nullptr;
 800   if (_compressor == nullptr) {
 801     result = (char*)_writer->write_buf(_buffer, _pos);
 802     _bytes_written += _pos;
 803   } else {
 804     do_compress();
 805     if (!has_error()) {
 806       result = (char*)_writer->write_buf(_out_buffer, _out_pos);
 807       _bytes_written += _out_pos;
 808     }
 809   }
 810   _pos = 0; // reset pos to make internal buffer available
 811 
 812   if (result != nullptr) {
 813     set_error(result);
 814   }
 815 }
 816 
 817 void DumpWriter::do_compress() {
 818   const char* msg = _compressor->compress(_buffer, _pos, _out_buffer, _out_size,
 819                                           _tmp_buffer, _tmp_size, &_out_pos);
 820 
 821   if (msg != nullptr) {
 822     set_error(msg);
 823   }
 824 }
 825 
 826 class DumperClassCacheTable;
 827 class DumperClassCacheTableEntry;
 828 
 829 // Support class with a collection of functions used when dumping the heap
 830 class DumperSupport : AllStatic {
 831  public:
 832 
 833   // write a header of the given type
 834   static void write_header(AbstractDumpWriter* writer, hprofTag tag, u4 len);
 835 
 836   // returns hprof tag for the given type signature
 837   static hprofTag sig2tag(Symbol* sig);
 838   // returns hprof tag for the given basic type
 839   static hprofTag type2tag(BasicType type);
 840   // Returns the size of the data to write.
 841   static u4 sig2size(Symbol* sig);
 842 
 843   // calculates the total size of the all fields of the given class.
 844   static u4 instance_size(InstanceKlass* ik, DumperClassCacheTableEntry* class_cache_entry = nullptr);
 845 
 846   // dump a jfloat
 847   static void dump_float(AbstractDumpWriter* writer, jfloat f);
 848   // dump a jdouble
 849   static void dump_double(AbstractDumpWriter* writer, jdouble d);
 850   // dumps the raw value of the given field
 851   static void dump_field_value(AbstractDumpWriter* writer, char type, oop obj, int offset);
 852   // returns the size of the static fields; also counts the static fields
 853   static u4 get_static_fields_size(InstanceKlass* ik, u2& field_count);
 854   // dumps static fields of the given class
 855   static void dump_static_fields(AbstractDumpWriter* writer, Klass* k);
 856   // dump the raw values of the instance fields of the given identity or inlined object;
 857   // for identity objects offset is 0 and 'klass' is o->klass(),
 858   // for inlined objects offset is the offset in the holder object, 'klass' is inlined object class
 859   static void dump_instance_fields(AbstractDumpWriter* writer, oop o, int offset, DumperClassCacheTable* class_cache, DumperClassCacheTableEntry* class_cache_entry);
 860   // dump the raw values of the instance fields of the given inlined object;
 861   // dump_instance_fields wrapper for inlined objects
 862   static void dump_inlined_object_fields(AbstractDumpWriter* writer, oop o, int offset, DumperClassCacheTable* class_cache, DumperClassCacheTableEntry* class_cache_entry);
 863 
 864   // get the count of the instance fields for a given class
 865   static u2 get_instance_fields_count(InstanceKlass* ik);
 866   // dumps the definition of the instance fields for a given class
 867   static void dump_instance_field_descriptors(AbstractDumpWriter* writer, InstanceKlass* k, uintx *inlined_fields_index = nullptr);
 868   // creates HPROF_GC_INSTANCE_DUMP record for the given object
 869   static void dump_instance(AbstractDumpWriter* writer, oop o, DumperClassCacheTable* class_cache);
 870   // creates HPROF_GC_CLASS_DUMP record for the given instance class
 871   static void dump_instance_class(AbstractDumpWriter* writer, Klass* k);
 872   // creates HPROF_GC_CLASS_DUMP record for a given array class
 873   static void dump_array_class(AbstractDumpWriter* writer, Klass* k);
 874 
 875   // creates HPROF_GC_OBJ_ARRAY_DUMP record for the given object array
 876   static void dump_object_array(AbstractDumpWriter* writer, objArrayOop array);
 877   // creates HPROF_GC_PRIM_ARRAY_DUMP record for the given flat array
 878   static void dump_flat_array(AbstractDumpWriter* writer, flatArrayOop array, DumperClassCacheTable* class_cache);
 879   // creates HPROF_GC_PRIM_ARRAY_DUMP record for the given type array
 880   static void dump_prim_array(AbstractDumpWriter* writer, typeArrayOop array);
 881   // create HPROF_FRAME record for the given method and bci
 882   static void dump_stack_frame(AbstractDumpWriter* writer, int frame_serial_num, int class_serial_num, Method* m, int bci);
 883 
 884   // check if we need to truncate an array
 885   static int calculate_array_max_length(AbstractDumpWriter* writer, arrayOop array, short header_size);
 886   // extended version to dump flat arrays as primitive arrays;
 887   // type_size specifies size of the inlined objects.
 888   static int calculate_array_max_length(AbstractDumpWriter* writer, arrayOop array, int type_size, short header_size);
 889 
 890   // fixes up the current dump record and writes HPROF_HEAP_DUMP_END record
 891   static void end_of_dump(AbstractDumpWriter* writer);
 892 
 893   static oop mask_dormant_archived_object(oop o, oop ref_obj) {
 894     if (o != nullptr && o->klass()->java_mirror_no_keepalive() == nullptr) {
 895       // Ignore this object since the corresponding java mirror is not loaded.
 896       // Might be a dormant archive object.
 897       report_dormant_archived_object(o, ref_obj);
 898       return nullptr;
 899     } else {
 900       return o;
 901     }
 902   }
 903 
 904   // helper methods for inlined fields.
 905   static bool is_inlined_field(const fieldDescriptor& fld) {
 906     return fld.is_flat();
 907   }
 908   static InlineKlass* get_inlined_field_klass(const fieldDescriptor& fld) {
 909     assert(is_inlined_field(fld), "must be inlined field");
 910     InstanceKlass* holder_klass = fld.field_holder();
 911     return InlineKlass::cast(holder_klass->get_inline_type_field_klass(fld.index()));
 912   }
 913 
 914   static void report_dormant_archived_object(oop o, oop ref_obj) {
 915     if (log_is_enabled(Trace, cds, heap)) {
 916       ResourceMark rm;
 917       if (ref_obj != nullptr) {
 918         log_trace(cds, heap)("skipped dormant archived object " INTPTR_FORMAT " (%s) referenced by " INTPTR_FORMAT " (%s)",
 919                   p2i(o), o->klass()->external_name(),
 920                   p2i(ref_obj), ref_obj->klass()->external_name());
 921       } else {
 922         log_trace(cds, heap)("skipped dormant archived object " INTPTR_FORMAT " (%s)",
 923                   p2i(o), o->klass()->external_name());
 924       }
 925     }
 926   }
 927 };
 928 
 929 // Hash table of klasses to the klass metadata. This should greatly improve the
 930 // hash dumping performance. This hash table is supposed to be used by a single
 931 // thread only.
 932 //
 933 class DumperClassCacheTableEntry : public CHeapObj<mtServiceability> {
 934   friend class DumperClassCacheTable;
 935 private:
 936   GrowableArray<char> _sigs_start;
 937   GrowableArray<int> _offsets;
 938   GrowableArray<InlineKlass*> _inline_klasses;
 939   u4 _instance_size;
 940   int _entries;
 941 
 942 public:
 943   DumperClassCacheTableEntry() : _instance_size(0), _entries(0) {};
 944 
 945   int field_count()             { return _entries; }
 946   char sig_start(int field_idx) { return _sigs_start.at(field_idx); }
 947   void push_sig_start_inlined() { _sigs_start.push('Q'); }
 948   bool is_inlined(int field_idx){ return _sigs_start.at(field_idx) == 'Q'; }
 949   InlineKlass* inline_klass(int field_idx) { assert(is_inlined(field_idx), "Not inlined"); return _inline_klasses.at(field_idx); }
 950   int offset(int field_idx)     { return _offsets.at(field_idx); }
 951   u4 instance_size()            { return _instance_size; }
 952 };
 953 
 954 class DumperClassCacheTable {
 955 private:
 956   // ResourceHashtable SIZE is specified at compile time so we
 957   // use 1031 which is the first prime after 1024.
 958   static constexpr size_t TABLE_SIZE = 1031;
 959 
 960   // Maintain the cache for N classes. This limits memory footprint
 961   // impact, regardless of how many classes we have in the dump.
 962   // This also improves look up performance by keeping the statically
 963   // sized table from overloading.
 964   static constexpr int CACHE_TOP = 256;
 965 
 966   typedef ResourceHashtable<InstanceKlass*, DumperClassCacheTableEntry*,
 967                             TABLE_SIZE, AnyObj::C_HEAP, mtServiceability> PtrTable;
 968   PtrTable* _ptrs;
 969 
 970   // Single-slot cache to handle the major case of objects of the same
 971   // class back-to-back, e.g. from T[].
 972   InstanceKlass* _last_ik;
 973   DumperClassCacheTableEntry* _last_entry;
 974 
 975   void unlink_all(PtrTable* table) {
 976     class CleanupEntry: StackObj {
 977     public:
 978       bool do_entry(InstanceKlass*& key, DumperClassCacheTableEntry*& entry) {
 979         delete entry;
 980         return true;
 981       }
 982     } cleanup;
 983     table->unlink(&cleanup);
 984   }
 985 
 986 public:
 987   DumperClassCacheTableEntry* lookup_or_create(InstanceKlass* ik) {
 988     if (_last_ik == ik) {
 989       return _last_entry;
 990     }
 991 
 992     DumperClassCacheTableEntry* entry;
 993     DumperClassCacheTableEntry** from_cache = _ptrs->get(ik);
 994     if (from_cache == nullptr) {
 995       entry = new DumperClassCacheTableEntry();
 996       for (HierarchicalFieldStream<JavaFieldStream> fld(ik); !fld.done(); fld.next()) {
 997         if (!fld.access_flags().is_static()) {
 998           InlineKlass* inlineKlass = nullptr;
 999           if (DumperSupport::is_inlined_field(fld.field_descriptor())) {
1000             inlineKlass = DumperSupport::get_inlined_field_klass(fld.field_descriptor());
1001             entry->push_sig_start_inlined();
1002             entry->_instance_size += DumperSupport::instance_size(inlineKlass);
1003           } else {
1004             Symbol* sig = fld.signature();
1005             entry->_sigs_start.push(sig->char_at(0));
1006             entry->_instance_size += DumperSupport::sig2size(sig);
1007           }
1008           entry->_inline_klasses.push(inlineKlass);
1009           entry->_offsets.push(fld.offset());
1010           entry->_entries++;
1011         }
1012       }
1013 
1014       if (_ptrs->number_of_entries() >= CACHE_TOP) {
1015         // We do not track the individual hit rates for table entries.
1016         // Purge the entire table, and let the cache catch up with new
1017         // distribution.
1018         unlink_all(_ptrs);
1019       }
1020 
1021       _ptrs->put(ik, entry);
1022     } else {
1023       entry = *from_cache;
1024     }
1025 
1026     // Remember for single-slot cache.
1027     _last_ik = ik;
1028     _last_entry = entry;
1029 
1030     return entry;
1031   }
1032 
1033   DumperClassCacheTable() : _ptrs(new (mtServiceability) PtrTable), _last_ik(nullptr), _last_entry(nullptr) {}
1034 
1035   ~DumperClassCacheTable() {
1036     unlink_all(_ptrs);
1037     delete _ptrs;
1038   }
1039 };
1040 
1041 // write a header of the given type
1042 void DumperSupport:: write_header(AbstractDumpWriter* writer, hprofTag tag, u4 len) {
1043   writer->write_u1(tag);
1044   writer->write_u4(0);                  // current ticks
1045   writer->write_u4(len);
1046 }
1047 
1048 // returns hprof tag for the given type signature
1049 hprofTag DumperSupport::sig2tag(Symbol* sig) {
1050   switch (sig->char_at(0)) {
1051     case JVM_SIGNATURE_CLASS    : return HPROF_NORMAL_OBJECT;
1052     case JVM_SIGNATURE_ARRAY    : return HPROF_NORMAL_OBJECT;
1053     case JVM_SIGNATURE_BYTE     : return HPROF_BYTE;
1054     case JVM_SIGNATURE_CHAR     : return HPROF_CHAR;
1055     case JVM_SIGNATURE_FLOAT    : return HPROF_FLOAT;
1056     case JVM_SIGNATURE_DOUBLE   : return HPROF_DOUBLE;
1057     case JVM_SIGNATURE_INT      : return HPROF_INT;
1058     case JVM_SIGNATURE_LONG     : return HPROF_LONG;
1059     case JVM_SIGNATURE_SHORT    : return HPROF_SHORT;
1060     case JVM_SIGNATURE_BOOLEAN  : return HPROF_BOOLEAN;
1061     default : ShouldNotReachHere(); /* to shut up compiler */ return HPROF_BYTE;
1062   }
1063 }
1064 
1065 hprofTag DumperSupport::type2tag(BasicType type) {
1066   switch (type) {
1067     case T_BYTE     : return HPROF_BYTE;
1068     case T_CHAR     : return HPROF_CHAR;
1069     case T_FLOAT    : return HPROF_FLOAT;
1070     case T_DOUBLE   : return HPROF_DOUBLE;
1071     case T_INT      : return HPROF_INT;
1072     case T_LONG     : return HPROF_LONG;
1073     case T_SHORT    : return HPROF_SHORT;
1074     case T_BOOLEAN  : return HPROF_BOOLEAN;
1075     default : ShouldNotReachHere(); /* to shut up compiler */ return HPROF_BYTE;
1076   }
1077 }
1078 
1079 u4 DumperSupport::sig2size(Symbol* sig) {
1080   switch (sig->char_at(0)) {
1081     case JVM_SIGNATURE_CLASS:
1082     case JVM_SIGNATURE_ARRAY: return sizeof(address);
1083     case JVM_SIGNATURE_BOOLEAN:
1084     case JVM_SIGNATURE_BYTE: return 1;
1085     case JVM_SIGNATURE_SHORT:
1086     case JVM_SIGNATURE_CHAR: return 2;
1087     case JVM_SIGNATURE_INT:
1088     case JVM_SIGNATURE_FLOAT: return 4;
1089     case JVM_SIGNATURE_LONG:
1090     case JVM_SIGNATURE_DOUBLE: return 8;
1091     default: ShouldNotReachHere(); /* to shut up compiler */ return 0;
1092   }
1093 }
1094 
1095 template<typename T, typename F> T bit_cast(F from) { // replace with the real thing when we can use c++20
1096   T to;
1097   static_assert(sizeof(to) == sizeof(from), "must be of the same size");
1098   memcpy(&to, &from, sizeof(to));
1099   return to;
1100 }
1101 
1102 // dump a jfloat
1103 void DumperSupport::dump_float(AbstractDumpWriter* writer, jfloat f) {
1104   if (g_isnan(f)) {
1105     writer->write_u4(0x7fc00000); // collapsing NaNs
1106   } else {
1107     writer->write_u4(bit_cast<u4>(f));
1108   }
1109 }
1110 
1111 // dump a jdouble
1112 void DumperSupport::dump_double(AbstractDumpWriter* writer, jdouble d) {
1113   if (g_isnan(d)) {
1114     writer->write_u8(0x7ff80000ull << 32); // collapsing NaNs
1115   } else {
1116     writer->write_u8(bit_cast<u8>(d));
1117   }
1118 }
1119 
1120 
1121 // dumps the raw value of the given field
1122 void DumperSupport::dump_field_value(AbstractDumpWriter* writer, char type, oop obj, int offset) {
1123   switch (type) {
1124     case JVM_SIGNATURE_CLASS :
1125     case JVM_SIGNATURE_ARRAY : {
1126       oop o = obj->obj_field_access<ON_UNKNOWN_OOP_REF | AS_NO_KEEPALIVE>(offset);
1127       o = mask_dormant_archived_object(o, obj);
1128       assert(oopDesc::is_oop_or_null(o), "Expected an oop or nullptr at " PTR_FORMAT, p2i(o));
1129       writer->write_objectID(o);
1130       break;
1131     }
1132     case JVM_SIGNATURE_BYTE : {
1133       jbyte b = obj->byte_field(offset);
1134       writer->write_u1(b);
1135       break;
1136     }
1137     case JVM_SIGNATURE_CHAR : {
1138       jchar c = obj->char_field(offset);
1139       writer->write_u2(c);
1140       break;
1141     }
1142     case JVM_SIGNATURE_SHORT : {
1143       jshort s = obj->short_field(offset);
1144       writer->write_u2(s);
1145       break;
1146     }
1147     case JVM_SIGNATURE_FLOAT : {
1148       jfloat f = obj->float_field(offset);
1149       dump_float(writer, f);
1150       break;
1151     }
1152     case JVM_SIGNATURE_DOUBLE : {
1153       jdouble d = obj->double_field(offset);
1154       dump_double(writer, d);
1155       break;
1156     }
1157     case JVM_SIGNATURE_INT : {
1158       jint i = obj->int_field(offset);
1159       writer->write_u4(i);
1160       break;
1161     }
1162     case JVM_SIGNATURE_LONG : {
1163       jlong l = obj->long_field(offset);
1164       writer->write_u8(l);
1165       break;
1166     }
1167     case JVM_SIGNATURE_BOOLEAN : {
1168       jboolean b = obj->bool_field(offset);
1169       writer->write_u1(b);
1170       break;
1171     }
1172     default : {
1173       ShouldNotReachHere();
1174       break;
1175     }
1176   }
1177 }
1178 
1179 // calculates the total size of the all fields of the given class.
1180 u4 DumperSupport::instance_size(InstanceKlass* ik, DumperClassCacheTableEntry* class_cache_entry) {
1181   if (class_cache_entry != nullptr) {
1182     return class_cache_entry->instance_size();
1183   } else {
1184     u4 size = 0;
1185     for (HierarchicalFieldStream<JavaFieldStream> fld(ik); !fld.done(); fld.next()) {
1186       if (!fld.access_flags().is_static()) {
1187         if (is_inlined_field(fld.field_descriptor())) {
1188           size += instance_size(get_inlined_field_klass(fld.field_descriptor()));
1189         } else {
1190           size += sig2size(fld.signature());
1191         }
1192       }
1193     }
1194     return size;
1195   }
1196 }
1197 
1198 u4 DumperSupport::get_static_fields_size(InstanceKlass* ik, u2& field_count) {
1199   field_count = 0;
1200   u4 size = 0;
1201 
1202   for (JavaFieldStream fldc(ik); !fldc.done(); fldc.next()) {
1203     if (fldc.access_flags().is_static()) {
1204       assert(!is_inlined_field(fldc.field_descriptor()), "static fields cannot be inlined");
1205 
1206       field_count++;
1207       size += sig2size(fldc.signature());
1208     }
1209   }
1210 
1211   // Add in resolved_references which is referenced by the cpCache
1212   // The resolved_references is an array per InstanceKlass holding the
1213   // strings and other oops resolved from the constant pool.
1214   oop resolved_references = ik->constants()->resolved_references_or_null();
1215   if (resolved_references != nullptr) {
1216     field_count++;
1217     size += sizeof(address);
1218 
1219     // Add in the resolved_references of the used previous versions of the class
1220     // in the case of RedefineClasses
1221     InstanceKlass* prev = ik->previous_versions();
1222     while (prev != nullptr && prev->constants()->resolved_references_or_null() != nullptr) {
1223       field_count++;
1224       size += sizeof(address);
1225       prev = prev->previous_versions();
1226     }
1227   }
1228 
1229   // We write the value itself plus a name and a one byte type tag per field.
1230   return checked_cast<u4>(size + field_count * (sizeof(address) + 1));
1231 }
1232 
1233 // dumps static fields of the given class
1234 void DumperSupport::dump_static_fields(AbstractDumpWriter* writer, Klass* k) {
1235   InstanceKlass* ik = InstanceKlass::cast(k);
1236 
1237   // dump the field descriptors and raw values
1238   for (JavaFieldStream fld(ik); !fld.done(); fld.next()) {
1239     if (fld.access_flags().is_static()) {
1240       assert(!is_inlined_field(fld.field_descriptor()), "static fields cannot be inlined");
1241 
1242       Symbol* sig = fld.signature();
1243 
1244       writer->write_symbolID(fld.name());   // name
1245       writer->write_u1(sig2tag(sig));       // type
1246 
1247       // value
1248       dump_field_value(writer, sig->char_at(0), ik->java_mirror(), fld.offset());
1249     }
1250   }
1251 
1252   // Add resolved_references for each class that has them
1253   oop resolved_references = ik->constants()->resolved_references_or_null();
1254   if (resolved_references != nullptr) {
1255     writer->write_symbolID(vmSymbols::resolved_references_name());  // name
1256     writer->write_u1(sig2tag(vmSymbols::object_array_signature())); // type
1257     writer->write_objectID(resolved_references);
1258 
1259     // Also write any previous versions
1260     InstanceKlass* prev = ik->previous_versions();
1261     while (prev != nullptr && prev->constants()->resolved_references_or_null() != nullptr) {
1262       writer->write_symbolID(vmSymbols::resolved_references_name());  // name
1263       writer->write_u1(sig2tag(vmSymbols::object_array_signature())); // type
1264       writer->write_objectID(prev->constants()->resolved_references());
1265       prev = prev->previous_versions();
1266     }
1267   }
1268 }
1269 
1270 // dump the raw values of the instance fields of the given identity or inlined object;
1271 // for identity objects offset is 0 and 'klass' is o->klass(),
1272 // for inlined objects offset is the offset in the holder object, 'klass' is inlined object class.
1273 void DumperSupport::dump_instance_fields(AbstractDumpWriter* writer, oop o, int offset, DumperClassCacheTable* class_cache, DumperClassCacheTableEntry* class_cache_entry) {
1274   assert(class_cache_entry != nullptr, "Pre-condition: must be provided");
1275   for (int idx = 0; idx < class_cache_entry->field_count(); idx++) {
1276     if (class_cache_entry->is_inlined(idx)) {
1277       InlineKlass* field_klass = class_cache_entry->inline_klass(idx);
1278       int fields_offset = offset + (class_cache_entry->offset(idx) - field_klass->first_field_offset());
1279       DumperClassCacheTableEntry* inline_class_cache_entry = class_cache->lookup_or_create(field_klass);
1280       dump_inlined_object_fields(writer, o, fields_offset, class_cache, inline_class_cache_entry);
1281     } else {
1282       dump_field_value(writer, class_cache_entry->sig_start(idx), o, class_cache_entry->offset(idx));
1283     }
1284   }
1285 }
1286 
1287 void DumperSupport::dump_inlined_object_fields(AbstractDumpWriter* writer, oop o, int offset, DumperClassCacheTable* class_cache, DumperClassCacheTableEntry* class_cache_entry) {
1288   // the object is inlined, so all its fields are stored without headers.
1289   dump_instance_fields(writer, o, offset, class_cache, class_cache_entry);
1290 }
1291 
1292 // gets the count of the instance fields for a given class
1293 u2 DumperSupport::get_instance_fields_count(InstanceKlass* ik) {
1294   u2 field_count = 0;
1295 
1296   for (JavaFieldStream fldc(ik); !fldc.done(); fldc.next()) {
1297     if (!fldc.access_flags().is_static()) {
1298       if (is_inlined_field(fldc.field_descriptor())) {
1299         // add "synthetic" fields for inlined fields.
1300         field_count += get_instance_fields_count(get_inlined_field_klass(fldc.field_descriptor()));
1301       } else {
1302         field_count++;
1303       }
1304     }
1305   }
1306 
1307   return field_count;
1308 }
1309 
1310 // dumps the definition of the instance fields for a given class
1311 // inlined_fields_id is not-nullptr for inlined fields (to get synthetic field name IDs
1312 // by using InlinedObjects::get_next_string_id()).
1313 void DumperSupport::dump_instance_field_descriptors(AbstractDumpWriter* writer, InstanceKlass* ik, uintx* inlined_fields_id) {
1314   // inlined_fields_id != nullptr means ik is a class of inlined field.
1315   // Inlined field id pointer for this class; lazyly initialized
1316   // if the class has inlined field(s) and the caller didn't provide inlined_fields_id.
1317   uintx *this_klass_inlined_fields_id = inlined_fields_id;
1318   uintx inlined_id = 0;
1319 
1320   // dump the field descriptors
1321   for (JavaFieldStream fld(ik); !fld.done(); fld.next()) {
1322     if (!fld.access_flags().is_static()) {
1323       if (is_inlined_field(fld.field_descriptor())) {
1324         // dump "synthetic" fields for inlined fields.
1325         if (this_klass_inlined_fields_id == nullptr) {
1326           inlined_id = InlinedObjects::get_instance()->get_base_index_for(ik);
1327           this_klass_inlined_fields_id = &inlined_id;
1328         }
1329         dump_instance_field_descriptors(writer, get_inlined_field_klass(fld.field_descriptor()), this_klass_inlined_fields_id);
1330       } else {
1331         Symbol* sig = fld.signature();
1332         Symbol* name = nullptr;
1333         // Use inlined_fields_id provided by caller.
1334         if (inlined_fields_id != nullptr) {
1335           uintx name_id = InlinedObjects::get_instance()->get_next_string_id(*inlined_fields_id);
1336 
1337           // name_id == 0 is returned on error. use original field signature.
1338           if (name_id != 0) {
1339             *inlined_fields_id = name_id;
1340             name = reinterpret_cast<Symbol*>(name_id);
1341           }
1342         }
1343         if (name == nullptr) {
1344           name = fld.name();
1345         }
1346 
1347         writer->write_symbolID(name);         // name
1348         writer->write_u1(sig2tag(sig));       // type
1349       }
1350     }
1351   }
1352 }
1353 
1354 // creates HPROF_GC_INSTANCE_DUMP record for the given object
1355 void DumperSupport::dump_instance(AbstractDumpWriter* writer, oop o, DumperClassCacheTable* class_cache) {
1356   InstanceKlass* ik = InstanceKlass::cast(o->klass());
1357 
1358   DumperClassCacheTableEntry* cache_entry = class_cache->lookup_or_create(ik);
1359 
1360   u4 is = instance_size(ik, cache_entry);
1361   u4 size = 1 + sizeof(address) + 4 + sizeof(address) + 4 + is;
1362 
1363   writer->start_sub_record(HPROF_GC_INSTANCE_DUMP, size);
1364   writer->write_objectID(o);
1365   writer->write_u4(STACK_TRACE_ID);
1366 
1367   // class ID
1368   writer->write_classID(ik);
1369 
1370   // number of bytes that follow
1371   writer->write_u4(is);
1372 
1373   // field values
1374   dump_instance_fields(writer, o, 0, class_cache, cache_entry);
1375 
1376   writer->end_sub_record();
1377 }
1378 
1379 // creates HPROF_GC_CLASS_DUMP record for the given instance class
1380 void DumperSupport::dump_instance_class(AbstractDumpWriter* writer, Klass* k) {
1381   InstanceKlass* ik = InstanceKlass::cast(k);
1382 
1383   // We can safepoint and do a heap dump at a point where we have a Klass,
1384   // but no java mirror class has been setup for it. So we need to check
1385   // that the class is at least loaded, to avoid crash from a null mirror.
1386   if (!ik->is_loaded()) {
1387     return;
1388   }
1389 
1390   u2 static_fields_count = 0;
1391   u4 static_size = get_static_fields_size(ik, static_fields_count);
1392   u2 instance_fields_count = get_instance_fields_count(ik);
1393   u4 instance_fields_size = instance_fields_count * (sizeof(address) + 1);
1394   u4 size = checked_cast<u4>(1 + sizeof(address) + 4 + 6 * sizeof(address) + 4 + 2 + 2 + static_size + 2 + instance_fields_size);
1395 
1396   writer->start_sub_record(HPROF_GC_CLASS_DUMP, size);
1397 
1398   // class ID
1399   writer->write_classID(ik);
1400   writer->write_u4(STACK_TRACE_ID);
1401 
1402   // super class ID
1403   InstanceKlass* java_super = ik->java_super();
1404   if (java_super == nullptr) {
1405     writer->write_objectID(oop(nullptr));
1406   } else {
1407     writer->write_classID(java_super);
1408   }
1409 
1410   writer->write_objectID(ik->class_loader());
1411   writer->write_objectID(ik->signers());
1412   writer->write_objectID(ik->protection_domain());
1413 
1414   // reserved
1415   writer->write_objectID(oop(nullptr));
1416   writer->write_objectID(oop(nullptr));
1417 
1418   // instance size
1419   writer->write_u4(HeapWordSize * ik->size_helper());
1420 
1421   // size of constant pool - ignored by HAT 1.1
1422   writer->write_u2(0);
1423 
1424   // static fields
1425   writer->write_u2(static_fields_count);
1426   dump_static_fields(writer, ik);
1427 
1428   // description of instance fields
1429   writer->write_u2(instance_fields_count);
1430   dump_instance_field_descriptors(writer, ik);
1431 
1432   writer->end_sub_record();
1433 }
1434 
1435 // creates HPROF_GC_CLASS_DUMP record for the given array class
1436 void DumperSupport::dump_array_class(AbstractDumpWriter* writer, Klass* k) {
1437   InstanceKlass* ik = nullptr; // bottom class for object arrays, null for primitive type arrays
1438   if (k->is_objArray_klass()) {
1439     Klass *bk = ObjArrayKlass::cast(k)->bottom_klass();
1440     assert(bk != nullptr, "checking");
1441     if (bk->is_instance_klass()) {
1442       ik = InstanceKlass::cast(bk);
1443     }
1444   }
1445 
1446   u4 size = 1 + sizeof(address) + 4 + 6 * sizeof(address) + 4 + 2 + 2 + 2;
1447   writer->start_sub_record(HPROF_GC_CLASS_DUMP, size);
1448   writer->write_classID(k);
1449   writer->write_u4(STACK_TRACE_ID);
1450 
1451   // super class of array classes is java.lang.Object
1452   InstanceKlass* java_super = k->java_super();
1453   assert(java_super != nullptr, "checking");
1454   writer->write_classID(java_super);
1455 
1456   writer->write_objectID(ik == nullptr ? oop(nullptr) : ik->class_loader());
1457   writer->write_objectID(ik == nullptr ? oop(nullptr) : ik->signers());
1458   writer->write_objectID(ik == nullptr ? oop(nullptr) : ik->protection_domain());
1459 
1460   writer->write_objectID(oop(nullptr));    // reserved
1461   writer->write_objectID(oop(nullptr));
1462   writer->write_u4(0);             // instance size
1463   writer->write_u2(0);             // constant pool
1464   writer->write_u2(0);             // static fields
1465   writer->write_u2(0);             // instance fields
1466 
1467   writer->end_sub_record();
1468 
1469 }
1470 
1471 // Hprof uses an u4 as record length field,
1472 // which means we need to truncate arrays that are too long.
1473 int DumperSupport::calculate_array_max_length(AbstractDumpWriter* writer, arrayOop array, int type_size, short header_size) {
1474   int length = array->length();
1475 
1476   size_t length_in_bytes = (size_t)length * type_size;
1477   uint max_bytes = max_juint - header_size;
1478 
1479   if (length_in_bytes > max_bytes) {
1480     length = max_bytes / type_size;
1481     length_in_bytes = (size_t)length * type_size;
1482 
1483     BasicType type = ArrayKlass::cast(array->klass())->element_type();
1484     warning("cannot dump array of type %s[] with length %d; truncating to length %d",
1485             type2name_tab[type], array->length(), length);
1486   }
1487   return length;
1488 }
1489 
1490 int DumperSupport::calculate_array_max_length(AbstractDumpWriter* writer, arrayOop array, short header_size) {
1491   BasicType type = ArrayKlass::cast(array->klass())->element_type();
1492   assert((type >= T_BOOLEAN && type <= T_OBJECT) || type == T_PRIMITIVE_OBJECT, "invalid array element type");
1493   int type_size;
1494   if (type == T_OBJECT) {
1495     type_size = sizeof(address);
1496   } else if (type == T_PRIMITIVE_OBJECT) {
1497       // TODO: FIXME
1498       fatal("Not supported yet"); // FIXME: JDK-8325678
1499   } else {
1500     type_size = type2aelembytes(type);
1501   }
1502 
1503   return calculate_array_max_length(writer, array, type_size, header_size);
1504 }
1505 
1506 // creates HPROF_GC_OBJ_ARRAY_DUMP record for the given object array
1507 void DumperSupport::dump_object_array(AbstractDumpWriter* writer, objArrayOop array) {
1508   // sizeof(u1) + 2 * sizeof(u4) + sizeof(objectID) + sizeof(classID)
1509   short header_size = 1 + 2 * 4 + 2 * sizeof(address);
1510   int length = calculate_array_max_length(writer, array, header_size);
1511   u4 size = checked_cast<u4>(header_size + length * sizeof(address));
1512 
1513   writer->start_sub_record(HPROF_GC_OBJ_ARRAY_DUMP, size);
1514   writer->write_objectID(array);
1515   writer->write_u4(STACK_TRACE_ID);
1516   writer->write_u4(length);
1517 
1518   // array class ID
1519   writer->write_classID(array->klass());
1520 
1521   // [id]* elements
1522   for (int index = 0; index < length; index++) {
1523     oop o = array->obj_at(index);
1524     o = mask_dormant_archived_object(o, array);
1525     writer->write_objectID(o);
1526   }
1527 
1528   writer->end_sub_record();
1529 }
1530 
1531 // creates HPROF_GC_PRIM_ARRAY_DUMP record for the given flat array
1532 void DumperSupport::dump_flat_array(AbstractDumpWriter* writer, flatArrayOop array, DumperClassCacheTable* class_cache) {
1533   FlatArrayKlass* array_klass = FlatArrayKlass::cast(array->klass());
1534   InlineKlass* element_klass = array_klass->element_klass();
1535   int element_size = instance_size(element_klass);
1536   /*                          id         array object ID
1537    *                          u4         stack trace serial number
1538    *                          u4         number of elements
1539    *                          u1         element type
1540    */
1541   short header_size = 1 + sizeof(address) + 2 * 4 + 1;
1542 
1543   // TODO: use T_SHORT/T_INT/T_LONG if needed to avoid truncation
1544   BasicType type = T_BYTE;
1545   int type_size = type2aelembytes(type);
1546   int length = calculate_array_max_length(writer, array, element_size, header_size);
1547   u4 length_in_bytes = (u4)(length * element_size);
1548   u4 size = header_size + length_in_bytes;
1549 
1550   writer->start_sub_record(HPROF_GC_PRIM_ARRAY_DUMP, size);
1551   writer->write_objectID(array);
1552   writer->write_u4(STACK_TRACE_ID);
1553   // TODO: round up array length for T_SHORT/T_INT/T_LONG
1554   writer->write_u4(length * element_size);
1555   writer->write_u1(type2tag(type));
1556 
1557   for (int index = 0; index < length; index++) {
1558     // need offset in the holder to read inlined object. calculate it from flatArrayOop::value_at_addr()
1559     int offset = (int)((address)array->value_at_addr(index, array_klass->layout_helper())
1560                   - cast_from_oop<address>(array));
1561     DumperClassCacheTableEntry* class_cache_entry = class_cache->lookup_or_create(element_klass);
1562     dump_inlined_object_fields(writer, array, offset, class_cache, class_cache_entry);
1563   }
1564 
1565   // TODO: write padding bytes for T_SHORT/T_INT/T_LONG
1566 
1567   InlinedObjects::get_instance()->add_flat_array(array);
1568 
1569   writer->end_sub_record();
1570 }
1571 
1572 #define WRITE_ARRAY(Array, Type, Size, Length) \
1573   for (int i = 0; i < Length; i++) { writer->write_##Size((Size)Array->Type##_at(i)); }
1574 
1575 // creates HPROF_GC_PRIM_ARRAY_DUMP record for the given type array
1576 void DumperSupport::dump_prim_array(AbstractDumpWriter* writer, typeArrayOop array) {
1577   BasicType type = TypeArrayKlass::cast(array->klass())->element_type();
1578   // 2 * sizeof(u1) + 2 * sizeof(u4) + sizeof(objectID)
1579   short header_size = 2 * 1 + 2 * 4 + sizeof(address);
1580 
1581   int length = calculate_array_max_length(writer, array, header_size);
1582   int type_size = type2aelembytes(type);
1583   u4 length_in_bytes = (u4)length * type_size;
1584   u4 size = header_size + length_in_bytes;
1585 
1586   writer->start_sub_record(HPROF_GC_PRIM_ARRAY_DUMP, size);
1587   writer->write_objectID(array);
1588   writer->write_u4(STACK_TRACE_ID);
1589   writer->write_u4(length);
1590   writer->write_u1(type2tag(type));
1591 
1592   // nothing to copy
1593   if (length == 0) {
1594     writer->end_sub_record();
1595     return;
1596   }
1597 
1598   // If the byte ordering is big endian then we can copy most types directly
1599 
1600   switch (type) {
1601     case T_INT : {
1602       if (Endian::is_Java_byte_ordering_different()) {
1603         WRITE_ARRAY(array, int, u4, length);
1604       } else {
1605         writer->write_raw(array->int_at_addr(0), length_in_bytes);
1606       }
1607       break;
1608     }
1609     case T_BYTE : {
1610       writer->write_raw(array->byte_at_addr(0), length_in_bytes);
1611       break;
1612     }
1613     case T_CHAR : {
1614       if (Endian::is_Java_byte_ordering_different()) {
1615         WRITE_ARRAY(array, char, u2, length);
1616       } else {
1617         writer->write_raw(array->char_at_addr(0), length_in_bytes);
1618       }
1619       break;
1620     }
1621     case T_SHORT : {
1622       if (Endian::is_Java_byte_ordering_different()) {
1623         WRITE_ARRAY(array, short, u2, length);
1624       } else {
1625         writer->write_raw(array->short_at_addr(0), length_in_bytes);
1626       }
1627       break;
1628     }
1629     case T_BOOLEAN : {
1630       if (Endian::is_Java_byte_ordering_different()) {
1631         WRITE_ARRAY(array, bool, u1, length);
1632       } else {
1633         writer->write_raw(array->bool_at_addr(0), length_in_bytes);
1634       }
1635       break;
1636     }
1637     case T_LONG : {
1638       if (Endian::is_Java_byte_ordering_different()) {
1639         WRITE_ARRAY(array, long, u8, length);
1640       } else {
1641         writer->write_raw(array->long_at_addr(0), length_in_bytes);
1642       }
1643       break;
1644     }
1645 
1646     // handle float/doubles in a special value to ensure than NaNs are
1647     // written correctly. TO DO: Check if we can avoid this on processors that
1648     // use IEEE 754.
1649 
1650     case T_FLOAT : {
1651       for (int i = 0; i < length; i++) {
1652         dump_float(writer, array->float_at(i));
1653       }
1654       break;
1655     }
1656     case T_DOUBLE : {
1657       for (int i = 0; i < length; i++) {
1658         dump_double(writer, array->double_at(i));
1659       }
1660       break;
1661     }
1662     default : ShouldNotReachHere();
1663   }
1664 
1665   writer->end_sub_record();
1666 }
1667 
1668 // create a HPROF_FRAME record of the given Method* and bci
1669 void DumperSupport::dump_stack_frame(AbstractDumpWriter* writer,
1670                                      int frame_serial_num,
1671                                      int class_serial_num,
1672                                      Method* m,
1673                                      int bci) {
1674   int line_number;
1675   if (m->is_native()) {
1676     line_number = -3;  // native frame
1677   } else {
1678     line_number = m->line_number_from_bci(bci);
1679   }
1680 
1681   write_header(writer, HPROF_FRAME, 4*oopSize + 2*sizeof(u4));
1682   writer->write_id(frame_serial_num);               // frame serial number
1683   writer->write_symbolID(m->name());                // method's name
1684   writer->write_symbolID(m->signature());           // method's signature
1685 
1686   assert(m->method_holder()->is_instance_klass(), "not InstanceKlass");
1687   writer->write_symbolID(m->method_holder()->source_file_name());  // source file name
1688   writer->write_u4(class_serial_num);               // class serial number
1689   writer->write_u4((u4) line_number);               // line number
1690 }
1691 
1692 
1693 class InlinedFieldNameDumper : public LockedClassesDo {
1694 public:
1695   typedef void (*Callback)(InlinedObjects *owner, const Klass *klass, uintx base_index, int count);
1696 
1697 private:
1698   AbstractDumpWriter* _writer;
1699   InlinedObjects *_owner;
1700   Callback       _callback;
1701   uintx _index;
1702 
1703   void dump_inlined_field_names(GrowableArray<Symbol*>* super_names, Symbol* field_name, InlineKlass* klass) {
1704     super_names->push(field_name);
1705     for (HierarchicalFieldStream<JavaFieldStream> fld(klass); !fld.done(); fld.next()) {
1706       if (!fld.access_flags().is_static()) {
1707         if (DumperSupport::is_inlined_field(fld.field_descriptor())) {
1708           dump_inlined_field_names(super_names, fld.name(), DumperSupport::get_inlined_field_klass(fld.field_descriptor()));
1709         } else {
1710           // get next string ID.
1711           uintx next_index = _owner->get_next_string_id(_index);
1712           if (next_index == 0) {
1713             // something went wrong (overflow?)
1714             // stop generation; the rest of inlined objects will have original field names.
1715             return;
1716           }
1717           _index = next_index;
1718 
1719           // Calculate length.
1720           int len = fld.name()->utf8_length();
1721           for (GrowableArrayIterator<Symbol*> it = super_names->begin(); it != super_names->end(); ++it) {
1722             len += (*it)->utf8_length() + 1;    // +1 for ".".
1723           }
1724 
1725           DumperSupport::write_header(_writer, HPROF_UTF8, oopSize + len);
1726           _writer->write_symbolID(reinterpret_cast<Symbol*>(_index));
1727           // Write the string value.
1728           // 1) super_names.
1729           for (GrowableArrayIterator<Symbol*> it = super_names->begin(); it != super_names->end(); ++it) {
1730             _writer->write_raw((*it)->bytes(), (*it)->utf8_length());
1731             _writer->write_u1('.');
1732           }
1733           // 2) field name.
1734           _writer->write_raw(fld.name()->bytes(), fld.name()->utf8_length());
1735         }
1736       }
1737     }
1738     super_names->pop();
1739   }
1740 
1741   void dump_inlined_field_names(Symbol* field_name, InlineKlass* field_klass) {
1742     GrowableArray<Symbol*> super_names(4, mtServiceability);
1743     dump_inlined_field_names(&super_names, field_name, field_klass);
1744   }
1745 
1746 public:
1747   InlinedFieldNameDumper(AbstractDumpWriter* writer, InlinedObjects* owner, Callback callback)
1748     : _writer(writer), _owner(owner), _callback(callback), _index(0)  {
1749   }
1750 
1751   void do_klass(Klass* k) {
1752     if (!k->is_instance_klass()) {
1753       return;
1754     }
1755     InstanceKlass* ik = InstanceKlass::cast(k);
1756     // if (ik->has_inline_type_fields()) {
1757     //   return;
1758     // }
1759 
1760     uintx base_index = _index;
1761     int count = 0;
1762 
1763     for (HierarchicalFieldStream<JavaFieldStream> fld(ik); !fld.done(); fld.next()) {
1764       if (!fld.access_flags().is_static()) {
1765         if (DumperSupport::is_inlined_field(fld.field_descriptor())) {
1766           dump_inlined_field_names(fld.name(), DumperSupport::get_inlined_field_klass(fld.field_descriptor()));
1767           count++;
1768         }
1769       }
1770     }
1771 
1772     if (count != 0) {
1773       _callback(_owner, k, base_index, count);
1774     }
1775   }
1776 };
1777 
1778 class InlinedFieldsDumper : public LockedClassesDo {
1779 private:
1780   AbstractDumpWriter* _writer;
1781 
1782 public:
1783   InlinedFieldsDumper(AbstractDumpWriter* writer) : _writer(writer) {}
1784 
1785   void do_klass(Klass* k) {
1786     if (!k->is_instance_klass()) {
1787       return;
1788     }
1789     InstanceKlass* ik = InstanceKlass::cast(k);
1790     // if (ik->has_inline_type_fields()) {
1791     //   return;
1792     // }
1793 
1794     // We can be at a point where java mirror does not exist yet.
1795     // So we need to check that the class is at least loaded, to avoid crash from a null mirror.
1796     if (!ik->is_loaded()) {
1797       return;
1798     }
1799 
1800     u2 inlined_count = 0;
1801     for (HierarchicalFieldStream<JavaFieldStream> fld(ik); !fld.done(); fld.next()) {
1802       if (!fld.access_flags().is_static()) {
1803         if (DumperSupport::is_inlined_field(fld.field_descriptor())) {
1804           inlined_count++;
1805         }
1806       }
1807     }
1808     if (inlined_count != 0) {
1809       _writer->write_u1(HPROF_CLASS_WITH_INLINED_FIELDS);
1810 
1811       // class ID
1812       _writer->write_classID(ik);
1813       // number of inlined fields
1814       _writer->write_u2(inlined_count);
1815       u2 index = 0;
1816       for (HierarchicalFieldStream<JavaFieldStream> fld(ik); !fld.done(); fld.next()) {
1817         if (!fld.access_flags().is_static()) {
1818           if (DumperSupport::is_inlined_field(fld.field_descriptor())) {
1819             // inlined field index
1820             _writer->write_u2(index);
1821             // synthetic field count
1822             u2 field_count = DumperSupport::get_instance_fields_count(DumperSupport::get_inlined_field_klass(fld.field_descriptor()));
1823             _writer->write_u2(field_count);
1824             // original field name
1825             _writer->write_symbolID(fld.name());
1826             // inlined field class ID
1827             _writer->write_classID(DumperSupport::get_inlined_field_klass(fld.field_descriptor()));
1828 
1829             index += field_count;
1830           } else {
1831             index++;
1832           }
1833         }
1834       }
1835     }
1836   }
1837 };
1838 
1839 
1840 void InlinedObjects::init() {
1841   _instance = this;
1842 
1843   struct Closure : public SymbolClosure {
1844     uintx _min_id = max_uintx;
1845     uintx _max_id = 0;
1846     Closure() : _min_id(max_uintx), _max_id(0) {}
1847 
1848     void do_symbol(Symbol** p) {
1849       uintx val = reinterpret_cast<uintx>(*p);
1850       if (val < _min_id) {
1851         _min_id = val;
1852       }
1853       if (val > _max_id) {
1854         _max_id = val;
1855       }
1856     }
1857   } closure;
1858 
1859   SymbolTable::symbols_do(&closure);
1860 
1861   _min_string_id = closure._min_id;
1862   _max_string_id = closure._max_id;
1863 }
1864 
1865 void InlinedObjects::release() {
1866   _instance = nullptr;
1867 
1868   if (_inlined_field_map != nullptr) {
1869     delete _inlined_field_map;
1870     _inlined_field_map = nullptr;
1871   }
1872   if (_flat_arrays != nullptr) {
1873     delete _flat_arrays;
1874     _flat_arrays = nullptr;
1875   }
1876 }
1877 
1878 void InlinedObjects::inlined_field_names_callback(InlinedObjects* _this, const Klass* klass, uintx base_index, int count) {
1879   if (_this->_inlined_field_map == nullptr) {
1880     _this->_inlined_field_map = new (mtServiceability) GrowableArray<ClassInlinedFields>(100, mtServiceability);
1881   }
1882   _this->_inlined_field_map->append(ClassInlinedFields(klass, base_index));
1883 
1884   // counters for dumping classes with inlined fields
1885   _this->_classes_count++;
1886   _this->_inlined_fields_count += count;
1887 }
1888 
1889 void InlinedObjects::dump_inlined_field_names(AbstractDumpWriter* writer) {
1890   InlinedFieldNameDumper nameDumper(writer, this, inlined_field_names_callback);
1891   ClassLoaderDataGraph::classes_do(&nameDumper);
1892 
1893   if (_inlined_field_map != nullptr) {
1894     // prepare the map for  get_base_index_for().
1895     _inlined_field_map->sort(ClassInlinedFields::compare);
1896   }
1897 }
1898 
1899 uintx InlinedObjects::get_base_index_for(Klass* k) {
1900   if (_inlined_field_map != nullptr) {
1901     bool found = false;
1902     int idx = _inlined_field_map->find_sorted<ClassInlinedFields, ClassInlinedFields::compare>(ClassInlinedFields(k, 0), found);
1903     if (found) {
1904         return _inlined_field_map->at(idx).base_index;
1905     }
1906   }
1907 
1908   // return max_uintx, so get_next_string_id returns 0.
1909   return max_uintx;
1910 }
1911 
1912 uintx InlinedObjects::get_next_string_id(uintx id) {
1913   if (++id == _min_string_id) {
1914     return _max_string_id + 1;
1915   }
1916   return id;
1917 }
1918 
1919 void InlinedObjects::dump_classed_with_inlined_fields(AbstractDumpWriter* writer) {
1920   if (_classes_count != 0) {
1921     // Record for each class contains tag(u1), class ID and count(u2)
1922     // for each inlined field index(u2), synthetic fields count(u2), original field name and class ID
1923     int size = _classes_count * (1 + sizeof(address) + 2)
1924              + _inlined_fields_count * (2 + 2 + sizeof(address) + sizeof(address));
1925     DumperSupport::write_header(writer, HPROF_INLINED_FIELDS, (u4)size);
1926 
1927     InlinedFieldsDumper dumper(writer);
1928     ClassLoaderDataGraph::classes_do(&dumper);
1929   }
1930 }
1931 
1932 void InlinedObjects::add_flat_array(oop array) {
1933   if (_flat_arrays == nullptr) {
1934     _flat_arrays = new (mtServiceability) GrowableArray<oop>(100, mtServiceability);
1935   }
1936   _flat_arrays->append(array);
1937 }
1938 
1939 void InlinedObjects::dump_flat_arrays(AbstractDumpWriter* writer) {
1940   if (_flat_arrays != nullptr) {
1941     // For each flat array the record contains tag (u1), object ID and class ID.
1942     int size = _flat_arrays->length() * (1 + sizeof(address) + sizeof(address));
1943 
1944     DumperSupport::write_header(writer, HPROF_FLAT_ARRAYS, (u4)size);
1945     for (GrowableArrayIterator<oop> it = _flat_arrays->begin(); it != _flat_arrays->end(); ++it) {
1946       flatArrayOop array = flatArrayOop(*it);
1947       FlatArrayKlass* array_klass = FlatArrayKlass::cast(array->klass());
1948       InlineKlass* element_klass = array_klass->element_klass();
1949       writer->write_u1(HPROF_FLAT_ARRAY);
1950       writer->write_objectID(array);
1951       writer->write_classID(element_klass);
1952     }
1953   }
1954 }
1955 
1956 
1957 // Support class used to generate HPROF_UTF8 records from the entries in the
1958 // SymbolTable.
1959 
1960 class SymbolTableDumper : public SymbolClosure {
1961  private:
1962   AbstractDumpWriter* _writer;
1963   AbstractDumpWriter* writer() const                { return _writer; }
1964  public:
1965   SymbolTableDumper(AbstractDumpWriter* writer)     { _writer = writer; }
1966   void do_symbol(Symbol** p);
1967 };
1968 
1969 void SymbolTableDumper::do_symbol(Symbol** p) {
1970   ResourceMark rm;
1971   Symbol* sym = *p;
1972   int len = sym->utf8_length();
1973   if (len > 0) {
1974     char* s = sym->as_utf8();
1975     DumperSupport::write_header(writer(), HPROF_UTF8, oopSize + len);
1976     writer()->write_symbolID(sym);
1977     writer()->write_raw(s, len);
1978   }
1979 }
1980 
1981 // Support class used to generate HPROF_GC_CLASS_DUMP records
1982 
1983 class ClassDumper : public KlassClosure {
1984  private:
1985   AbstractDumpWriter* _writer;
1986   AbstractDumpWriter* writer() const { return _writer; }
1987 
1988  public:
1989   ClassDumper(AbstractDumpWriter* writer) : _writer(writer) {}
1990 
1991   void do_klass(Klass* k) {
1992     if (k->is_instance_klass()) {
1993       DumperSupport::dump_instance_class(writer(), k);
1994     } else {
1995       DumperSupport::dump_array_class(writer(), k);
1996     }
1997   }
1998 };
1999 
2000 // Support class used to generate HPROF_GC_ROOT_JNI_LOCAL records
2001 
2002 class JNILocalsDumper : public OopClosure {
2003  private:
2004   AbstractDumpWriter* _writer;
2005   u4 _thread_serial_num;
2006   int _frame_num;
2007   AbstractDumpWriter* writer() const                { return _writer; }
2008  public:
2009   JNILocalsDumper(AbstractDumpWriter* writer, u4 thread_serial_num) {
2010     _writer = writer;
2011     _thread_serial_num = thread_serial_num;
2012     _frame_num = -1;  // default - empty stack
2013   }
2014   void set_frame_number(int n) { _frame_num = n; }
2015   void do_oop(oop* obj_p);
2016   void do_oop(narrowOop* obj_p) { ShouldNotReachHere(); }
2017 };
2018 
2019 void JNILocalsDumper::do_oop(oop* obj_p) {
2020   // ignore null handles
2021   oop o = *obj_p;
2022   if (o != nullptr) {
2023     u4 size = 1 + sizeof(address) + 4 + 4;
2024     writer()->start_sub_record(HPROF_GC_ROOT_JNI_LOCAL, size);
2025     writer()->write_objectID(o);
2026     writer()->write_u4(_thread_serial_num);
2027     writer()->write_u4((u4)_frame_num);
2028     writer()->end_sub_record();
2029   }
2030 }
2031 
2032 
2033 // Support class used to generate HPROF_GC_ROOT_JNI_GLOBAL records
2034 
2035 class JNIGlobalsDumper : public OopClosure {
2036  private:
2037   AbstractDumpWriter* _writer;
2038   AbstractDumpWriter* writer() const                { return _writer; }
2039 
2040  public:
2041   JNIGlobalsDumper(AbstractDumpWriter* writer) {
2042     _writer = writer;
2043   }
2044   void do_oop(oop* obj_p);
2045   void do_oop(narrowOop* obj_p) { ShouldNotReachHere(); }
2046 };
2047 
2048 void JNIGlobalsDumper::do_oop(oop* obj_p) {
2049   oop o = NativeAccess<AS_NO_KEEPALIVE>::oop_load(obj_p);
2050 
2051   // ignore these
2052   if (o == nullptr) return;
2053   // we ignore global ref to symbols and other internal objects
2054   if (o->is_instance() || o->is_objArray() || o->is_typeArray()) {
2055     u4 size = 1 + 2 * sizeof(address);
2056     writer()->start_sub_record(HPROF_GC_ROOT_JNI_GLOBAL, size);
2057     writer()->write_objectID(o);
2058     writer()->write_rootID(obj_p);      // global ref ID
2059     writer()->end_sub_record();
2060   }
2061 };
2062 
2063 // Support class used to generate HPROF_GC_ROOT_STICKY_CLASS records
2064 
2065 class StickyClassDumper : public KlassClosure {
2066  private:
2067   AbstractDumpWriter* _writer;
2068   AbstractDumpWriter* writer() const                { return _writer; }
2069  public:
2070   StickyClassDumper(AbstractDumpWriter* writer) {
2071     _writer = writer;
2072   }
2073   void do_klass(Klass* k) {
2074     if (k->is_instance_klass()) {
2075       InstanceKlass* ik = InstanceKlass::cast(k);
2076       u4 size = 1 + sizeof(address);
2077       writer()->start_sub_record(HPROF_GC_ROOT_STICKY_CLASS, size);
2078       writer()->write_classID(ik);
2079       writer()->end_sub_record();
2080     }
2081   }
2082 };
2083 
2084 // Support class used to generate HPROF_GC_ROOT_JAVA_FRAME records.
2085 
2086 class JavaStackRefDumper : public StackObj {
2087 private:
2088   AbstractDumpWriter* _writer;
2089   u4 _thread_serial_num;
2090   int _frame_num;
2091   AbstractDumpWriter* writer() const { return _writer; }
2092 public:
2093   JavaStackRefDumper(AbstractDumpWriter* writer, u4 thread_serial_num)
2094       : _writer(writer), _thread_serial_num(thread_serial_num), _frame_num(-1) // default - empty stack
2095   {
2096   }
2097 
2098   void set_frame_number(int n) { _frame_num = n; }
2099 
2100   void dump_java_stack_refs(StackValueCollection* values);
2101 };
2102 
2103 void JavaStackRefDumper::dump_java_stack_refs(StackValueCollection* values) {
2104   for (int index = 0; index < values->size(); index++) {
2105     if (values->at(index)->type() == T_OBJECT) {
2106       oop o = values->obj_at(index)();
2107       if (o != nullptr) {
2108         u4 size = 1 + sizeof(address) + 4 + 4;
2109         writer()->start_sub_record(HPROF_GC_ROOT_JAVA_FRAME, size);
2110         writer()->write_objectID(o);
2111         writer()->write_u4(_thread_serial_num);
2112         writer()->write_u4((u4)_frame_num);
2113         writer()->end_sub_record();
2114       }
2115     }
2116   }
2117 }
2118 
2119 // Class to collect, store and dump thread-related data:
2120 // - HPROF_TRACE and HPROF_FRAME records;
2121 // - HPROF_GC_ROOT_THREAD_OBJ/HPROF_GC_ROOT_JAVA_FRAME/HPROF_GC_ROOT_JNI_LOCAL subrecords.
2122 class ThreadDumper : public CHeapObj<mtInternal> {
2123 public:
2124   enum class ThreadType { Platform, MountedVirtual, UnmountedVirtual };
2125 
2126 private:
2127   ThreadType _thread_type;
2128   JavaThread* _java_thread;
2129   oop _thread_oop;
2130 
2131   GrowableArray<StackFrameInfo*>* _frames;
2132   // non-null if the thread is OOM thread
2133   Method* _oome_constructor;
2134   int _thread_serial_num;
2135   int _start_frame_serial_num;
2136 
2137   vframe* get_top_frame() const;
2138 
2139 public:
2140   static bool should_dump_pthread(JavaThread* thread) {
2141     return thread->threadObj() != nullptr && !thread->is_exiting() && !thread->is_hidden_from_external_view();
2142   }
2143 
2144   static bool should_dump_vthread(oop vt) {
2145     return java_lang_VirtualThread::state(vt) != java_lang_VirtualThread::NEW
2146         && java_lang_VirtualThread::state(vt) != java_lang_VirtualThread::TERMINATED;
2147   }
2148 
2149   static bool is_vthread_mounted(oop vt) {
2150     // The code should be consistent with the "mounted virtual thread" case
2151     // (VM_HeapDumper::dump_stack_traces(), ThreadDumper::get_top_frame()).
2152     // I.e. virtual thread is mounted if its carrierThread is not null
2153     // and is_vthread_mounted() for the carrier thread returns true.
2154     oop carrier_thread = java_lang_VirtualThread::carrier_thread(vt);
2155     if (carrier_thread == nullptr) {
2156       return false;
2157     }
2158     JavaThread* java_thread = java_lang_Thread::thread(carrier_thread);
2159     return java_thread->is_vthread_mounted();
2160   }
2161 
2162   ThreadDumper(ThreadType thread_type, JavaThread* java_thread, oop thread_oop);
2163 
2164   // affects frame_count
2165   void add_oom_frame(Method* oome_constructor) {
2166     assert(_start_frame_serial_num == 0, "add_oom_frame cannot be called after init_serial_nums");
2167     _oome_constructor = oome_constructor;
2168   }
2169 
2170   void init_serial_nums(volatile int* thread_counter, volatile int* frame_counter) {
2171     assert(_start_frame_serial_num == 0, "already initialized");
2172     _thread_serial_num = Atomic::fetch_then_add(thread_counter, 1);
2173     _start_frame_serial_num = Atomic::fetch_then_add(frame_counter, frame_count());
2174   }
2175 
2176   bool oom_thread() const {
2177     return _oome_constructor != nullptr;
2178   }
2179 
2180   int frame_count() const {
2181     return _frames->length() + (oom_thread() ? 1 : 0);
2182   }
2183 
2184   u4 thread_serial_num() const {
2185     return (u4)_thread_serial_num;
2186   }
2187 
2188   u4 stack_trace_serial_num() const {
2189     return (u4)(_thread_serial_num + STACK_TRACE_ID);
2190   }
2191 
2192   // writes HPROF_TRACE and HPROF_FRAME records
2193   // returns number of dumped frames
2194   void dump_stack_traces(AbstractDumpWriter* writer, GrowableArray<Klass*>* klass_map);
2195 
2196   // writes HPROF_GC_ROOT_THREAD_OBJ subrecord
2197   void dump_thread_obj(AbstractDumpWriter* writer);
2198 
2199   // Walk the stack of the thread.
2200   // Dumps a HPROF_GC_ROOT_JAVA_FRAME subrecord for each local
2201   // Dumps a HPROF_GC_ROOT_JNI_LOCAL subrecord for each JNI local
2202   void dump_stack_refs(AbstractDumpWriter* writer);
2203 
2204 };
2205 
2206 ThreadDumper::ThreadDumper(ThreadType thread_type, JavaThread* java_thread, oop thread_oop)
2207     : _thread_type(thread_type), _java_thread(java_thread), _thread_oop(thread_oop),
2208       _oome_constructor(nullptr),
2209       _thread_serial_num(0), _start_frame_serial_num(0)
2210 {
2211   // sanity checks
2212   if (_thread_type == ThreadType::UnmountedVirtual) {
2213     assert(_java_thread == nullptr, "sanity");
2214     assert(_thread_oop != nullptr, "sanity");
2215   } else {
2216     assert(_java_thread != nullptr, "sanity");
2217     assert(_thread_oop != nullptr, "sanity");
2218   }
2219 
2220   _frames = new (mtServiceability) GrowableArray<StackFrameInfo*>(10, mtServiceability);
2221   bool stop_at_vthread_entry = _thread_type == ThreadType::MountedVirtual;
2222 
2223   // vframes are resource allocated
2224   Thread* current_thread = Thread::current();
2225   ResourceMark rm(current_thread);
2226   HandleMark hm(current_thread);
2227 
2228   for (vframe* vf = get_top_frame(); vf != nullptr; vf = vf->sender()) {
2229     if (stop_at_vthread_entry && vf->is_vthread_entry()) {
2230       break;
2231     }
2232     if (vf->is_java_frame()) {
2233       javaVFrame* jvf = javaVFrame::cast(vf);
2234       _frames->append(new StackFrameInfo(jvf, false));
2235     } else {
2236       // ignore non-Java frames
2237     }
2238   }
2239 }
2240 
2241 void ThreadDumper::dump_stack_traces(AbstractDumpWriter* writer, GrowableArray<Klass*>* klass_map) {
2242   assert(_thread_serial_num != 0 && _start_frame_serial_num != 0, "serial_nums are not initialized");
2243 
2244   // write HPROF_FRAME records for this thread's stack trace
2245   int depth = _frames->length();
2246   int frame_serial_num = _start_frame_serial_num;
2247 
2248   if (oom_thread()) {
2249     // OOM thread
2250     // write fake frame that makes it look like the thread, which caused OOME,
2251     // is in the OutOfMemoryError zero-parameter constructor
2252     int oome_serial_num = klass_map->find(_oome_constructor->method_holder());
2253     // the class serial number starts from 1
2254     assert(oome_serial_num > 0, "OutOfMemoryError class not found");
2255     DumperSupport::dump_stack_frame(writer, ++frame_serial_num, oome_serial_num, _oome_constructor, 0);
2256     depth++;
2257   }
2258 
2259   for (int j = 0; j < _frames->length(); j++) {
2260     StackFrameInfo* frame = _frames->at(j);
2261     Method* m = frame->method();
2262     int class_serial_num = klass_map->find(m->method_holder());
2263     // the class serial number starts from 1
2264     assert(class_serial_num > 0, "class not found");
2265     DumperSupport::dump_stack_frame(writer, ++frame_serial_num, class_serial_num, m, frame->bci());
2266   }
2267 
2268   // write HPROF_TRACE record for the thread
2269   DumperSupport::write_header(writer, HPROF_TRACE, checked_cast<u4>(3 * sizeof(u4) + depth * oopSize));
2270   writer->write_u4(stack_trace_serial_num());   // stack trace serial number
2271   writer->write_u4(thread_serial_num());        // thread serial number
2272   writer->write_u4((u4)depth);                  // frame count (including oom frame)
2273   for (int j = 1; j <= depth; j++) {
2274     writer->write_id(_start_frame_serial_num + j);
2275   }
2276 }
2277 
2278 void ThreadDumper::dump_thread_obj(AbstractDumpWriter * writer) {
2279   assert(_thread_serial_num != 0 && _start_frame_serial_num != 0, "serial_num is not initialized");
2280 
2281   u4 size = 1 + sizeof(address) + 4 + 4;
2282   writer->start_sub_record(HPROF_GC_ROOT_THREAD_OBJ, size);
2283   writer->write_objectID(_thread_oop);
2284   writer->write_u4(thread_serial_num());      // thread serial number
2285   writer->write_u4(stack_trace_serial_num()); // stack trace serial number
2286   writer->end_sub_record();
2287 }
2288 
2289 void ThreadDumper::dump_stack_refs(AbstractDumpWriter * writer) {
2290   assert(_thread_serial_num != 0 && _start_frame_serial_num != 0, "serial_num is not initialized");
2291 
2292   JNILocalsDumper blk(writer, thread_serial_num());
2293   if (_thread_type == ThreadType::Platform) {
2294     if (!_java_thread->has_last_Java_frame()) {
2295       // no last java frame but there may be JNI locals
2296       _java_thread->active_handles()->oops_do(&blk);
2297       return;
2298     }
2299   }
2300 
2301   JavaStackRefDumper java_ref_dumper(writer, thread_serial_num());
2302 
2303   // vframes are resource allocated
2304   Thread* current_thread = Thread::current();
2305   ResourceMark rm(current_thread);
2306   HandleMark hm(current_thread);
2307 
2308   bool stopAtVthreadEntry = _thread_type == ThreadType::MountedVirtual;
2309   frame* last_entry_frame = nullptr;
2310   bool is_top_frame = true;
2311   int depth = 0;
2312   if (oom_thread()) {
2313     depth++;
2314   }
2315 
2316   for (vframe* vf = get_top_frame(); vf != nullptr; vf = vf->sender()) {
2317     if (stopAtVthreadEntry && vf->is_vthread_entry()) {
2318       break;
2319     }
2320 
2321     if (vf->is_java_frame()) {
2322       javaVFrame* jvf = javaVFrame::cast(vf);
2323       if (!(jvf->method()->is_native())) {
2324         java_ref_dumper.set_frame_number(depth);
2325         java_ref_dumper.dump_java_stack_refs(jvf->locals());
2326         java_ref_dumper.dump_java_stack_refs(jvf->expressions());
2327       } else {
2328         // native frame
2329         blk.set_frame_number(depth);
2330         if (is_top_frame) {
2331           // JNI locals for the top frame.
2332           assert(_java_thread != nullptr, "impossible for unmounted vthread");
2333           _java_thread->active_handles()->oops_do(&blk);
2334         } else {
2335           if (last_entry_frame != nullptr) {
2336             // JNI locals for the entry frame
2337             assert(last_entry_frame->is_entry_frame(), "checking");
2338             last_entry_frame->entry_frame_call_wrapper()->handles()->oops_do(&blk);
2339           }
2340         }
2341       }
2342       last_entry_frame = nullptr;
2343       // increment only for Java frames
2344       depth++;
2345     } else {
2346       // externalVFrame - for an entry frame then we report the JNI locals
2347       // when we find the corresponding javaVFrame
2348       frame* fr = vf->frame_pointer();
2349       assert(fr != nullptr, "sanity check");
2350       if (fr->is_entry_frame()) {
2351         last_entry_frame = fr;
2352       }
2353     }
2354   is_top_frame = false;
2355   }
2356   assert(depth == frame_count(), "total number of Java frames not matched");
2357 }
2358 
2359 vframe* ThreadDumper::get_top_frame() const {
2360   if (_thread_type == ThreadType::UnmountedVirtual) {
2361     ContinuationWrapper cont(java_lang_VirtualThread::continuation(_thread_oop));
2362     if (cont.is_empty()) {
2363       return nullptr;
2364     }
2365     assert(!cont.is_mounted(), "sanity check");
2366     stackChunkOop chunk = cont.last_nonempty_chunk();
2367     if (chunk == nullptr || chunk->is_empty()) {
2368       return nullptr;
2369     }
2370 
2371     RegisterMap reg_map(cont.continuation(), RegisterMap::UpdateMap::include);
2372     frame fr = chunk->top_frame(&reg_map);
2373     vframe* vf = vframe::new_vframe(&fr, &reg_map, nullptr); // don't need JavaThread
2374     return vf;
2375   }
2376 
2377   RegisterMap reg_map(_java_thread,
2378       RegisterMap::UpdateMap::include,
2379       RegisterMap::ProcessFrames::include,
2380       RegisterMap::WalkContinuation::skip);
2381   switch (_thread_type) {
2382   case ThreadType::Platform:
2383     if (!_java_thread->has_last_Java_frame()) {
2384       return nullptr;
2385     }
2386     return _java_thread->is_vthread_mounted()
2387         ? _java_thread->carrier_last_java_vframe(&reg_map)
2388         : _java_thread->platform_thread_last_java_vframe(&reg_map);
2389 
2390   case ThreadType::MountedVirtual:
2391     return _java_thread->last_java_vframe(&reg_map);
2392 
2393   default: // make compilers happy
2394       break;
2395   }
2396   ShouldNotReachHere();
2397   return nullptr;
2398 }
2399 
2400 // Callback to dump thread-related data for unmounted virtual threads;
2401 // implemented by VM_HeapDumper.
2402 class UnmountedVThreadDumper {
2403  public:
2404   virtual void dump_vthread(oop vt, AbstractDumpWriter* segment_writer) = 0;
2405 };
2406 
2407 // Support class used when iterating over the heap.
2408 class HeapObjectDumper : public ObjectClosure {
2409  private:
2410   AbstractDumpWriter* _writer;
2411   AbstractDumpWriter* writer()                  { return _writer; }
2412   UnmountedVThreadDumper* _vthread_dumper;
2413 
2414   DumperClassCacheTable _class_cache;
2415 
2416  public:
2417   HeapObjectDumper(AbstractDumpWriter* writer, UnmountedVThreadDumper* vthread_dumper)
2418     : _writer(writer), _vthread_dumper(vthread_dumper) {}
2419 
2420   // called for each object in the heap
2421   void do_object(oop o);
2422 };
2423 
2424 void HeapObjectDumper::do_object(oop o) {
2425   // skip classes as these emitted as HPROF_GC_CLASS_DUMP records
2426   if (o->klass() == vmClasses::Class_klass()) {
2427     if (!java_lang_Class::is_primitive(o)) {
2428       return;
2429     }
2430   }
2431 
2432   if (DumperSupport::mask_dormant_archived_object(o, nullptr) == nullptr) {
2433     return;
2434   }
2435 
2436   if (o->is_instance()) {
2437     // create a HPROF_GC_INSTANCE record for each object
2438     DumperSupport::dump_instance(writer(), o, &_class_cache);
2439     // If we encounter an unmounted virtual thread it needs to be dumped explicitly
2440     // (mounted virtual threads are dumped with their carriers).
2441     if (java_lang_VirtualThread::is_instance(o)
2442         && ThreadDumper::should_dump_vthread(o) && !ThreadDumper::is_vthread_mounted(o)) {
2443       _vthread_dumper->dump_vthread(o, writer());
2444     }
2445   } else if (o->is_objArray()) {
2446     // create a HPROF_GC_OBJ_ARRAY_DUMP record for each object array
2447     DumperSupport::dump_object_array(writer(), objArrayOop(o));
2448   } else if (o->is_flatArray()) {
2449     DumperSupport::dump_flat_array(writer(), flatArrayOop(o), &_class_cache);
2450   } else if (o->is_typeArray()) {
2451     // create a HPROF_GC_PRIM_ARRAY_DUMP record for each type array
2452     DumperSupport::dump_prim_array(writer(), typeArrayOop(o));
2453   }
2454 }
2455 
2456 // The dumper controller for parallel heap dump
2457 class DumperController : public CHeapObj<mtInternal> {
2458  private:
2459    Monitor* _lock;
2460    Mutex* _global_writer_lock;
2461 
2462    const uint   _dumper_number;
2463    uint   _complete_number;
2464 
2465    bool   _started; // VM dumper started and acquired global writer lock
2466 
2467  public:
2468    DumperController(uint number) :
2469      // _lock and _global_writer_lock are used for synchronization between GC worker threads inside safepoint,
2470      // so we lock with _no_safepoint_check_flag.
2471      // signal_start() acquires _lock when global writer is locked,
2472      // its rank must be less than _global_writer_lock rank.
2473      _lock(new (std::nothrow) PaddedMonitor(Mutex::nosafepoint - 1, "DumperController_lock")),
2474      _global_writer_lock(new (std::nothrow) Mutex(Mutex::nosafepoint, "DumpWriter_lock")),
2475      _dumper_number(number),
2476      _complete_number(0),
2477      _started(false)
2478    {}
2479 
2480    ~DumperController() {
2481      delete _lock;
2482      delete _global_writer_lock;
2483    }
2484 
2485    // parallel (non VM) dumpers must wait until VM dumper acquires global writer lock
2486    void wait_for_start_signal() {
2487      MonitorLocker ml(_lock, Mutex::_no_safepoint_check_flag);
2488      while (_started == false) {
2489        ml.wait();
2490      }
2491    }
2492 
2493    void signal_start() {
2494      MonitorLocker ml(_lock, Mutex::_no_safepoint_check_flag);
2495      _started = true;
2496      ml.notify_all();
2497    }
2498 
2499    void lock_global_writer() {
2500      _global_writer_lock->lock_without_safepoint_check();
2501    }
2502 
2503    void unlock_global_writer() {
2504      _global_writer_lock->unlock();
2505    }
2506 
2507    void dumper_complete(DumpWriter* local_writer, DumpWriter* global_writer) {
2508      MonitorLocker ml(_lock, Mutex::_no_safepoint_check_flag);
2509      _complete_number++;
2510      // propagate local error to global if any
2511      if (local_writer->has_error()) {
2512        global_writer->set_error(local_writer->error());
2513      }
2514      ml.notify();
2515    }
2516 
2517    void wait_all_dumpers_complete() {
2518      MonitorLocker ml(_lock, Mutex::_no_safepoint_check_flag);
2519      while (_complete_number != _dumper_number) {
2520         ml.wait();
2521      }
2522    }
2523 };
2524 
2525 // DumpMerger merges separate dump files into a complete one
2526 class DumpMerger : public StackObj {
2527 private:
2528   DumpWriter* _writer;
2529   InlinedObjects*  _inlined_objects;
2530   const char* _path;
2531   bool _has_error;
2532   int _dump_seq;
2533 
2534 private:
2535   void merge_file(const char* path);
2536   void merge_done();
2537   void set_error(const char* msg);
2538 
2539 public:
2540   DumpMerger(const char* path, DumpWriter* writer, InlinedObjects* inlined_objects, int dump_seq) :
2541     _writer(writer),
2542     _inlined_objects(inlined_objects),
2543     _path(path),
2544     _has_error(_writer->has_error()),
2545     _dump_seq(dump_seq) {}
2546 
2547   void do_merge();
2548 
2549   // returns path for the parallel DumpWriter (resource allocated)
2550   static char* get_writer_path(const char* base_path, int seq);
2551 
2552 };
2553 
2554 char* DumpMerger::get_writer_path(const char* base_path, int seq) {
2555   // approximate required buffer size
2556   size_t buf_size = strlen(base_path)
2557                     + 2                 // ".p"
2558                     + 10                // number (that's enough for 2^32 parallel dumpers)
2559                     + 1;                // '\0'
2560 
2561   char* path = NEW_RESOURCE_ARRAY(char, buf_size);
2562   memset(path, 0, buf_size);
2563 
2564   os::snprintf(path, buf_size, "%s.p%d", base_path, seq);
2565 
2566   return path;
2567 }
2568 
2569 
2570 void DumpMerger::merge_done() {
2571   // Writes the HPROF_HEAP_DUMP_END record.
2572   if (!_has_error) {
2573     DumperSupport::end_of_dump(_writer);
2574     _inlined_objects->dump_flat_arrays(_writer);
2575     _writer->flush();
2576     _inlined_objects->release();
2577   }
2578   _dump_seq = 0; //reset
2579 }
2580 
2581 void DumpMerger::set_error(const char* msg) {
2582   assert(msg != nullptr, "sanity check");
2583   log_error(heapdump)("%s (file: %s)", msg, _path);
2584   _writer->set_error(msg);
2585   _has_error = true;
2586 }
2587 
2588 #ifdef LINUX
2589 // Merge segmented heap files via sendfile, it's more efficient than the
2590 // read+write combination, which would require transferring data to and from
2591 // user space.
2592 void DumpMerger::merge_file(const char* path) {
2593   TraceTime timer("Merge segmented heap file directly", TRACETIME_LOG(Info, heapdump));
2594 
2595   int segment_fd = os::open(path, O_RDONLY, 0);
2596   if (segment_fd == -1) {
2597     set_error("Can not open segmented heap file during merging");
2598     return;
2599   }
2600 
2601   struct stat st;
2602   if (os::stat(path, &st) != 0) {
2603     ::close(segment_fd);
2604     set_error("Can not get segmented heap file size during merging");
2605     return;
2606   }
2607 
2608   // A successful call to sendfile may write fewer bytes than requested; the
2609   // caller should be prepared to retry the call if there were unsent bytes.
2610   jlong offset = 0;
2611   while (offset < st.st_size) {
2612     int ret = os::Linux::sendfile(_writer->get_fd(), segment_fd, &offset, st.st_size);
2613     if (ret == -1) {
2614       ::close(segment_fd);
2615       set_error("Failed to merge segmented heap file");
2616       return;
2617     }
2618   }
2619 
2620   // As sendfile variant does not call the write method of the global writer,
2621   // bytes_written is also incorrect for this variant, we need to explicitly
2622   // accumulate bytes_written for the global writer in this case
2623   julong accum = _writer->bytes_written() + st.st_size;
2624   _writer->set_bytes_written(accum);
2625   ::close(segment_fd);
2626 }
2627 #else
2628 // Generic implementation using read+write
2629 void DumpMerger::merge_file(const char* path) {
2630   TraceTime timer("Merge segmented heap file", TRACETIME_LOG(Info, heapdump));
2631 
2632   fileStream segment_fs(path, "rb");
2633   if (!segment_fs.is_open()) {
2634     set_error("Can not open segmented heap file during merging");
2635     return;
2636   }
2637 
2638   jlong total = 0;
2639   size_t cnt = 0;
2640 
2641   // Use _writer buffer for reading.
2642   while ((cnt = segment_fs.read(_writer->buffer(), 1, _writer->buffer_size())) != 0) {
2643     _writer->set_position(cnt);
2644     _writer->flush();
2645     total += cnt;
2646   }
2647 
2648   if (segment_fs.fileSize() != total) {
2649     set_error("Merged heap dump is incomplete");
2650   }
2651 }
2652 #endif
2653 
2654 void DumpMerger::do_merge() {
2655   TraceTime timer("Merge heap files complete", TRACETIME_LOG(Info, heapdump));
2656 
2657   // Since contents in segmented heap file were already zipped, we don't need to zip
2658   // them again during merging.
2659   AbstractCompressor* saved_compressor = _writer->compressor();
2660   _writer->set_compressor(nullptr);
2661 
2662   // Merge the content of the remaining files into base file. Regardless of whether
2663   // the merge process is successful or not, these segmented files will be deleted.
2664   for (int i = 0; i < _dump_seq; i++) {
2665     ResourceMark rm;
2666     const char* path = get_writer_path(_path, i);
2667     if (!_has_error) {
2668       merge_file(path);
2669     }
2670     // Delete selected segmented heap file nevertheless
2671     if (remove(path) != 0) {
2672       log_info(heapdump)("Removal of segment file (%d) failed (%d)", i, errno);
2673     }
2674   }
2675 
2676   // restore compressor for further use
2677   _writer->set_compressor(saved_compressor);
2678   merge_done();
2679 }
2680 
2681 // The VM operation that performs the heap dump
2682 class VM_HeapDumper : public VM_GC_Operation, public WorkerTask, public UnmountedVThreadDumper {
2683  private:
2684   static VM_HeapDumper*   _global_dumper;
2685   static DumpWriter*      _global_writer;
2686   DumpWriter*             _local_writer;
2687   JavaThread*             _oome_thread;
2688   Method*                 _oome_constructor;
2689   bool                    _gc_before_heap_dump;
2690   GrowableArray<Klass*>*  _klass_map;
2691 
2692   ThreadDumper**          _thread_dumpers; // platform, carrier and mounted virtual threads
2693   int                     _thread_dumpers_count;
2694   volatile int            _thread_serial_num;
2695   volatile int            _frame_serial_num;
2696 
2697   volatile int            _dump_seq;
2698 
2699   // Inlined object support.
2700   InlinedObjects          _inlined_objects;
2701 
2702   // parallel heap dump support
2703   uint                    _num_dumper_threads;
2704   DumperController*       _dumper_controller;
2705   ParallelObjectIterator* _poi;
2706 
2707   // Dumper id of VMDumper thread.
2708   static const int VMDumperId = 0;
2709   // VM dumper dumps both heap and non-heap data, other dumpers dump heap-only data.
2710   static bool is_vm_dumper(int dumper_id) { return dumper_id == VMDumperId; }
2711   // the 1st dumper calling get_next_dumper_id becomes VM dumper
2712   int get_next_dumper_id() {
2713     return Atomic::fetch_then_add(&_dump_seq, 1);
2714   }
2715 
2716   // accessors and setters
2717   static VM_HeapDumper* dumper()         {  assert(_global_dumper != nullptr, "Error"); return _global_dumper; }
2718   static DumpWriter* writer()            {  assert(_global_writer != nullptr, "Error"); return _global_writer; }
2719 
2720   void set_global_dumper() {
2721     assert(_global_dumper == nullptr, "Error");
2722     _global_dumper = this;
2723   }
2724   void set_global_writer() {
2725     assert(_global_writer == nullptr, "Error");
2726     _global_writer = _local_writer;
2727   }
2728   void clear_global_dumper() { _global_dumper = nullptr; }
2729   void clear_global_writer() { _global_writer = nullptr; }
2730 
2731   bool skip_operation() const;
2732 
2733   // writes a HPROF_LOAD_CLASS record to global writer
2734   static void do_load_class(Klass* k);
2735 
2736   // HPROF_GC_ROOT_THREAD_OBJ records for platform and mounted virtual threads
2737   void dump_threads(AbstractDumpWriter* writer);
2738 
2739   void add_class_serial_number(Klass* k, int serial_num) {
2740     _klass_map->at_put_grow(serial_num, k);
2741   }
2742 
2743   bool is_oom_thread(JavaThread* thread) const {
2744     return thread == _oome_thread && _oome_constructor != nullptr;
2745   }
2746 
2747   // HPROF_TRACE and HPROF_FRAME records for platform and mounted virtual threads
2748   void dump_stack_traces(AbstractDumpWriter* writer);
2749 
2750  public:
2751   VM_HeapDumper(DumpWriter* writer, bool gc_before_heap_dump, bool oome, uint num_dump_threads) :
2752     VM_GC_Operation(0 /* total collections,      dummy, ignored */,
2753                     GCCause::_heap_dump /* GC Cause */,
2754                     0 /* total full collections, dummy, ignored */,
2755                     gc_before_heap_dump),
2756     WorkerTask("dump heap") {
2757     _local_writer = writer;
2758     _gc_before_heap_dump = gc_before_heap_dump;
2759     _klass_map = new (mtServiceability) GrowableArray<Klass*>(INITIAL_CLASS_COUNT, mtServiceability);
2760 
2761     _thread_dumpers = nullptr;
2762     _thread_dumpers_count = 0;
2763     _thread_serial_num = 1;
2764     _frame_serial_num = 1;
2765 
2766     _dump_seq = VMDumperId;
2767     _num_dumper_threads = num_dump_threads;
2768     _dumper_controller = nullptr;
2769     _poi = nullptr;
2770     if (oome) {
2771       assert(!Thread::current()->is_VM_thread(), "Dump from OutOfMemoryError cannot be called by the VMThread");
2772       // get OutOfMemoryError zero-parameter constructor
2773       InstanceKlass* oome_ik = vmClasses::OutOfMemoryError_klass();
2774       _oome_constructor = oome_ik->find_method(vmSymbols::object_initializer_name(),
2775                                                           vmSymbols::void_method_signature());
2776       // get thread throwing OOME when generating the heap dump at OOME
2777       _oome_thread = JavaThread::current();
2778     } else {
2779       _oome_thread = nullptr;
2780       _oome_constructor = nullptr;
2781     }
2782   }
2783 
2784   ~VM_HeapDumper() {
2785     if (_thread_dumpers != nullptr) {
2786       for (int i = 0; i < _thread_dumpers_count; i++) {
2787         delete _thread_dumpers[i];
2788       }
2789       FREE_C_HEAP_ARRAY(ThreadDumper*, _thread_dumpers);
2790     }
2791 
2792     if (_dumper_controller != nullptr) {
2793       delete _dumper_controller;
2794       _dumper_controller = nullptr;
2795     }
2796     delete _klass_map;
2797   }
2798   int dump_seq()           { return _dump_seq; }
2799   bool is_parallel_dump()  { return _num_dumper_threads > 1; }
2800   void prepare_parallel_dump(WorkerThreads* workers);
2801 
2802   InlinedObjects* inlined_objects() { return &_inlined_objects; }
2803 
2804   VMOp_Type type() const { return VMOp_HeapDumper; }
2805   virtual bool doit_prologue();
2806   void doit();
2807   void work(uint worker_id);
2808 
2809   // UnmountedVThreadDumper implementation
2810   void dump_vthread(oop vt, AbstractDumpWriter* segment_writer);
2811 };
2812 
2813 VM_HeapDumper* VM_HeapDumper::_global_dumper = nullptr;
2814 DumpWriter*    VM_HeapDumper::_global_writer = nullptr;
2815 
2816 bool VM_HeapDumper::skip_operation() const {
2817   return false;
2818 }
2819 
2820 // fixes up the current dump record and writes HPROF_HEAP_DUMP_END record
2821 void DumperSupport::end_of_dump(AbstractDumpWriter* writer) {
2822   writer->finish_dump_segment();
2823 
2824   writer->write_u1(HPROF_HEAP_DUMP_END);
2825   writer->write_u4(0);
2826   writer->write_u4(0);
2827 }
2828 
2829 // writes a HPROF_LOAD_CLASS record for the class
2830 void VM_HeapDumper::do_load_class(Klass* k) {
2831   static u4 class_serial_num = 0;
2832 
2833   // len of HPROF_LOAD_CLASS record
2834   u4 remaining = 2*oopSize + 2*sizeof(u4);
2835 
2836   DumperSupport::write_header(writer(), HPROF_LOAD_CLASS, remaining);
2837 
2838   // class serial number is just a number
2839   writer()->write_u4(++class_serial_num);
2840 
2841   // class ID
2842   writer()->write_classID(k);
2843 
2844   // add the Klass* and class serial number pair
2845   dumper()->add_class_serial_number(k, class_serial_num);
2846 
2847   writer()->write_u4(STACK_TRACE_ID);
2848 
2849   // class name ID
2850   Symbol* name = k->name();
2851   writer()->write_symbolID(name);
2852 }
2853 
2854 // Write a HPROF_GC_ROOT_THREAD_OBJ record for platform/carrier and mounted virtual threads.
2855 // Then walk the stack so that locals and JNI locals are dumped.
2856 void VM_HeapDumper::dump_threads(AbstractDumpWriter* writer) {
2857   for (int i = 0; i < _thread_dumpers_count; i++) {
2858     _thread_dumpers[i]->dump_thread_obj(writer);
2859     _thread_dumpers[i]->dump_stack_refs(writer);
2860   }
2861 }
2862 
2863 bool VM_HeapDumper::doit_prologue() {
2864   if (_gc_before_heap_dump && UseZGC) {
2865     // ZGC cannot perform a synchronous GC cycle from within the VM thread.
2866     // So ZCollectedHeap::collect_as_vm_thread() is a noop. To respect the
2867     // _gc_before_heap_dump flag a synchronous GC cycle is performed from
2868     // the caller thread in the prologue.
2869     Universe::heap()->collect(GCCause::_heap_dump);
2870   }
2871   return VM_GC_Operation::doit_prologue();
2872 }
2873 
2874 void VM_HeapDumper::prepare_parallel_dump(WorkerThreads* workers) {
2875   uint num_active_workers = workers != nullptr ? workers->active_workers() : 0;
2876   uint num_requested_dump_threads = _num_dumper_threads;
2877   // check if we can dump in parallel based on requested and active threads
2878   if (num_active_workers <= 1 || num_requested_dump_threads <= 1) {
2879     _num_dumper_threads = 1;
2880   } else {
2881     _num_dumper_threads = clamp(num_requested_dump_threads, 2U, num_active_workers);
2882   }
2883   _dumper_controller = new (std::nothrow) DumperController(_num_dumper_threads);
2884   bool can_parallel = _num_dumper_threads > 1;
2885   log_info(heapdump)("Requested dump threads %u, active dump threads %u, "
2886                      "actual dump threads %u, parallelism %s",
2887                      num_requested_dump_threads, num_active_workers,
2888                      _num_dumper_threads, can_parallel ? "true" : "false");
2889 }
2890 
2891 // The VM operation that dumps the heap. The dump consists of the following
2892 // records:
2893 //
2894 //  HPROF_HEADER
2895 //  [HPROF_UTF8]*
2896 //  [HPROF_LOAD_CLASS]*
2897 //  [[HPROF_FRAME]*|HPROF_TRACE]*
2898 //  [HPROF_GC_CLASS_DUMP]*
2899 //  [HPROF_HEAP_DUMP_SEGMENT]*
2900 //  HPROF_HEAP_DUMP_END
2901 //
2902 // The HPROF_TRACE records represent the stack traces where the heap dump
2903 // is generated and a "dummy trace" record which does not include
2904 // any frames. The dummy trace record is used to be referenced as the
2905 // unknown object alloc site.
2906 //
2907 // Each HPROF_HEAP_DUMP_SEGMENT record has a length followed by sub-records.
2908 // To allow the heap dump be generated in a single pass we remember the position
2909 // of the dump length and fix it up after all sub-records have been written.
2910 // To generate the sub-records we iterate over the heap, writing
2911 // HPROF_GC_INSTANCE_DUMP, HPROF_GC_OBJ_ARRAY_DUMP, and HPROF_GC_PRIM_ARRAY_DUMP
2912 // records as we go. Once that is done we write records for some of the GC
2913 // roots.
2914 
2915 void VM_HeapDumper::doit() {
2916 
2917   CollectedHeap* ch = Universe::heap();
2918 
2919   ch->ensure_parsability(false); // must happen, even if collection does
2920                                  // not happen (e.g. due to GCLocker)
2921 
2922   if (_gc_before_heap_dump) {
2923     if (GCLocker::is_active()) {
2924       warning("GC locker is held; pre-heapdump GC was skipped");
2925     } else {
2926       ch->collect_as_vm_thread(GCCause::_heap_dump);
2927     }
2928   }
2929 
2930   // At this point we should be the only dumper active, so
2931   // the following should be safe.
2932   set_global_dumper();
2933   set_global_writer();
2934 
2935   WorkerThreads* workers = ch->safepoint_workers();
2936   prepare_parallel_dump(workers);
2937 
2938   if (!is_parallel_dump()) {
2939     work(VMDumperId);
2940   } else {
2941     ParallelObjectIterator poi(_num_dumper_threads);
2942     _poi = &poi;
2943     workers->run_task(this, _num_dumper_threads);
2944     _poi = nullptr;
2945   }
2946 
2947   // Now we clear the global variables, so that a future dumper can run.
2948   clear_global_dumper();
2949   clear_global_writer();
2950 }
2951 
2952 void VM_HeapDumper::work(uint worker_id) {
2953   // VM Dumper works on all non-heap data dumping and part of heap iteration.
2954   int dumper_id = get_next_dumper_id();
2955 
2956   if (is_vm_dumper(dumper_id)) {
2957     // lock global writer, it will be unlocked after VM Dumper finishes with non-heap data
2958     _dumper_controller->lock_global_writer();
2959     _dumper_controller->signal_start();
2960   } else {
2961     _dumper_controller->wait_for_start_signal();
2962   }
2963 
2964   if (is_vm_dumper(dumper_id)) {
2965     TraceTime timer("Dump non-objects", TRACETIME_LOG(Info, heapdump));
2966     // Write the file header - we always use 1.0.2
2967     const char* header = "JAVA PROFILE 1.0.2";
2968 
2969     // header is few bytes long - no chance to overflow int
2970     writer()->write_raw(header, strlen(header) + 1); // NUL terminated
2971     writer()->write_u4(oopSize);
2972     // timestamp is current time in ms
2973     writer()->write_u8(os::javaTimeMillis());
2974     // HPROF_UTF8 records
2975     SymbolTableDumper sym_dumper(writer());
2976     SymbolTable::symbols_do(&sym_dumper);
2977 
2978     // HPROF_UTF8 records for inlined field names.
2979     inlined_objects()->init();
2980     inlined_objects()->dump_inlined_field_names(writer());
2981 
2982     // HPROF_INLINED_FIELDS
2983     inlined_objects()->dump_classed_with_inlined_fields(writer());
2984 
2985     // write HPROF_LOAD_CLASS records
2986     {
2987       LockedClassesDo locked_load_classes(&do_load_class);
2988       ClassLoaderDataGraph::classes_do(&locked_load_classes);
2989     }
2990 
2991     // write HPROF_FRAME and HPROF_TRACE records
2992     // this must be called after _klass_map is built when iterating the classes above.
2993     dump_stack_traces(writer());
2994 
2995     // unlock global writer, so parallel dumpers can dump stack traces of unmounted virtual threads
2996     _dumper_controller->unlock_global_writer();
2997   }
2998 
2999   // HPROF_HEAP_DUMP/HPROF_HEAP_DUMP_SEGMENT starts here
3000 
3001   ResourceMark rm;
3002   // share global compressor, local DumpWriter is not responsible for its life cycle
3003   DumpWriter segment_writer(DumpMerger::get_writer_path(writer()->get_file_path(), dumper_id),
3004                             writer()->is_overwrite(), writer()->compressor());
3005   if (!segment_writer.has_error()) {
3006     if (is_vm_dumper(dumper_id)) {
3007       // dump some non-heap subrecords to heap dump segment
3008       TraceTime timer("Dump non-objects (part 2)", TRACETIME_LOG(Info, heapdump));
3009       // Writes HPROF_GC_CLASS_DUMP records
3010       ClassDumper class_dumper(&segment_writer);
3011       ClassLoaderDataGraph::classes_do(&class_dumper);
3012 
3013       // HPROF_GC_ROOT_THREAD_OBJ + frames + jni locals
3014       dump_threads(&segment_writer);
3015 
3016       // HPROF_GC_ROOT_JNI_GLOBAL
3017       JNIGlobalsDumper jni_dumper(&segment_writer);
3018       JNIHandles::oops_do(&jni_dumper);
3019       // technically not jni roots, but global roots
3020       // for things like preallocated throwable backtraces
3021       Universe::vm_global()->oops_do(&jni_dumper);
3022       // HPROF_GC_ROOT_STICKY_CLASS
3023       // These should be classes in the null class loader data, and not all classes
3024       // if !ClassUnloading
3025       StickyClassDumper stiky_class_dumper(&segment_writer);
3026       ClassLoaderData::the_null_class_loader_data()->classes_do(&stiky_class_dumper);
3027     }
3028 
3029     // Heap iteration.
3030     // writes HPROF_GC_INSTANCE_DUMP records.
3031     // After each sub-record is written check_segment_length will be invoked
3032     // to check if the current segment exceeds a threshold. If so, a new
3033     // segment is started.
3034     // The HPROF_GC_CLASS_DUMP and HPROF_GC_INSTANCE_DUMP are the vast bulk
3035     // of the heap dump.
3036 
3037     TraceTime timer(is_parallel_dump() ? "Dump heap objects in parallel" : "Dump heap objects", TRACETIME_LOG(Info, heapdump));
3038     HeapObjectDumper obj_dumper(&segment_writer, this);
3039     if (!is_parallel_dump()) {
3040       Universe::heap()->object_iterate(&obj_dumper);
3041     } else {
3042       // == Parallel dump
3043       _poi->object_iterate(&obj_dumper, worker_id);
3044     }
3045 
3046     segment_writer.finish_dump_segment();
3047     segment_writer.flush();
3048   }
3049 
3050   _dumper_controller->dumper_complete(&segment_writer, writer());
3051 
3052   if (is_vm_dumper(dumper_id)) {
3053     _dumper_controller->wait_all_dumpers_complete();
3054 
3055     // flush global writer
3056     writer()->flush();
3057 
3058     // At this point, all fragments of the heapdump have been written to separate files.
3059     // We need to merge them into a complete heapdump and write HPROF_HEAP_DUMP_END at that time.
3060   }
3061 }
3062 
3063 void VM_HeapDumper::dump_stack_traces(AbstractDumpWriter* writer) {
3064   // write a HPROF_TRACE record without any frames to be referenced as object alloc sites
3065   DumperSupport::write_header(writer, HPROF_TRACE, 3 * sizeof(u4));
3066   writer->write_u4((u4)STACK_TRACE_ID);
3067   writer->write_u4(0);                    // thread number
3068   writer->write_u4(0);                    // frame count
3069 
3070   // max number if every platform thread is carrier with mounted virtual thread
3071   _thread_dumpers = NEW_C_HEAP_ARRAY(ThreadDumper*, Threads::number_of_threads() * 2, mtInternal);
3072 
3073   for (JavaThreadIteratorWithHandle jtiwh; JavaThread * thread = jtiwh.next(); ) {
3074     if (ThreadDumper::should_dump_pthread(thread)) {
3075       bool add_oom_frame = is_oom_thread(thread);
3076 
3077       oop mounted_vt = thread->is_vthread_mounted() ? thread->vthread() : nullptr;
3078       if (mounted_vt != nullptr && !ThreadDumper::should_dump_vthread(mounted_vt)) {
3079         mounted_vt = nullptr;
3080       }
3081 
3082       // mounted vthread (if any)
3083       if (mounted_vt != nullptr) {
3084         ThreadDumper* thread_dumper = new ThreadDumper(ThreadDumper::ThreadType::MountedVirtual, thread, mounted_vt);
3085         _thread_dumpers[_thread_dumpers_count++] = thread_dumper;
3086         if (add_oom_frame) {
3087           thread_dumper->add_oom_frame(_oome_constructor);
3088           // we add oom frame to the VT stack, don't add it to the carrier thread stack
3089           add_oom_frame = false;
3090         }
3091         thread_dumper->init_serial_nums(&_thread_serial_num, &_frame_serial_num);
3092         thread_dumper->dump_stack_traces(writer, _klass_map);
3093       }
3094 
3095       // platform or carrier thread
3096       ThreadDumper* thread_dumper = new ThreadDumper(ThreadDumper::ThreadType::Platform, thread, thread->threadObj());
3097       _thread_dumpers[_thread_dumpers_count++] = thread_dumper;
3098       if (add_oom_frame) {
3099         thread_dumper->add_oom_frame(_oome_constructor);
3100       }
3101       thread_dumper->init_serial_nums(&_thread_serial_num, &_frame_serial_num);
3102       thread_dumper->dump_stack_traces(writer, _klass_map);
3103     }
3104   }
3105 }
3106 
3107 void VM_HeapDumper::dump_vthread(oop vt, AbstractDumpWriter* segment_writer) {
3108   // unmounted vthread has no JavaThread
3109   ThreadDumper thread_dumper(ThreadDumper::ThreadType::UnmountedVirtual, nullptr, vt);
3110   thread_dumper.init_serial_nums(&_thread_serial_num, &_frame_serial_num);
3111 
3112   // write HPROF_TRACE/HPROF_FRAME records to global writer
3113   _dumper_controller->lock_global_writer();
3114   thread_dumper.dump_stack_traces(writer(), _klass_map);
3115   _dumper_controller->unlock_global_writer();
3116 
3117   // write HPROF_GC_ROOT_THREAD_OBJ/HPROF_GC_ROOT_JAVA_FRAME/HPROF_GC_ROOT_JNI_LOCAL subrecord
3118   // to segment writer
3119   thread_dumper.dump_thread_obj(segment_writer);
3120   thread_dumper.dump_stack_refs(segment_writer);
3121 }
3122 
3123 // dump the heap to given path.
3124 int HeapDumper::dump(const char* path, outputStream* out, int compression, bool overwrite, uint num_dump_threads) {
3125   assert(path != nullptr && strlen(path) > 0, "path missing");
3126 
3127   // print message in interactive case
3128   if (out != nullptr) {
3129     out->print_cr("Dumping heap to %s ...", path);
3130     timer()->start();
3131   }
3132 
3133   if (_oome && num_dump_threads > 1) {
3134     // Each additional parallel writer requires several MB of internal memory
3135     // (DumpWriter buffer, DumperClassCacheTable, GZipCompressor buffers).
3136     // For the OOM handling we may already be limited in memory.
3137     // Lets ensure we have at least 20MB per thread.
3138     julong max_threads = os::free_memory() / (20 * M);
3139     if (num_dump_threads > max_threads) {
3140       num_dump_threads = MAX2<uint>(1, (uint)max_threads);
3141     }
3142   }
3143 
3144   // create JFR event
3145   EventHeapDump event;
3146 
3147   AbstractCompressor* compressor = nullptr;
3148 
3149   if (compression > 0) {
3150     compressor = new (std::nothrow) GZipCompressor(compression);
3151 
3152     if (compressor == nullptr) {
3153       set_error("Could not allocate gzip compressor");
3154       return -1;
3155     }
3156   }
3157 
3158   DumpWriter writer(path, overwrite, compressor);
3159 
3160   if (writer.error() != nullptr) {
3161     set_error(writer.error());
3162     if (out != nullptr) {
3163       out->print_cr("Unable to create %s: %s", path,
3164         (error() != nullptr) ? error() : "reason unknown");
3165     }
3166     return -1;
3167   }
3168 
3169   // generate the segmented heap dump into separate files
3170   VM_HeapDumper dumper(&writer, _gc_before_heap_dump, _oome, num_dump_threads);
3171   VMThread::execute(&dumper);
3172 
3173   // record any error that the writer may have encountered
3174   set_error(writer.error());
3175 
3176   // Heap dump process is done in two phases
3177   //
3178   // Phase 1: Concurrent threads directly write heap data to multiple heap files.
3179   //          This is done by VM_HeapDumper, which is performed within safepoint.
3180   //
3181   // Phase 2: Merge multiple heap files into one complete heap dump file.
3182   //          This is done by DumpMerger, which is performed outside safepoint
3183 
3184   DumpMerger merger(path, &writer, dumper.inlined_objects(), dumper.dump_seq());
3185   // Perform heapdump file merge operation in the current thread prevents us
3186   // from occupying the VM Thread, which in turn affects the occurrence of
3187   // GC and other VM operations.
3188   merger.do_merge();
3189   if (writer.error() != nullptr) {
3190     set_error(writer.error());
3191   }
3192 
3193   // emit JFR event
3194   if (error() == nullptr) {
3195     event.set_destination(path);
3196     event.set_gcBeforeDump(_gc_before_heap_dump);
3197     event.set_size(writer.bytes_written());
3198     event.set_onOutOfMemoryError(_oome);
3199     event.set_overwrite(overwrite);
3200     event.set_compression(compression);
3201     event.commit();
3202   } else {
3203     log_debug(cds, heap)("Error %s while dumping heap", error());
3204   }
3205 
3206   // print message in interactive case
3207   if (out != nullptr) {
3208     timer()->stop();
3209     if (error() == nullptr) {
3210       out->print_cr("Heap dump file created [" JULONG_FORMAT " bytes in %3.3f secs]",
3211                     writer.bytes_written(), timer()->seconds());
3212     } else {
3213       out->print_cr("Dump file is incomplete: %s", writer.error());
3214     }
3215   }
3216 
3217   if (compressor != nullptr) {
3218     delete compressor;
3219   }
3220   return (writer.error() == nullptr) ? 0 : -1;
3221 }
3222 
3223 // stop timer (if still active), and free any error string we might be holding
3224 HeapDumper::~HeapDumper() {
3225   if (timer()->is_active()) {
3226     timer()->stop();
3227   }
3228   set_error(nullptr);
3229 }
3230 
3231 
3232 // returns the error string (resource allocated), or null
3233 char* HeapDumper::error_as_C_string() const {
3234   if (error() != nullptr) {
3235     char* str = NEW_RESOURCE_ARRAY(char, strlen(error())+1);
3236     strcpy(str, error());
3237     return str;
3238   } else {
3239     return nullptr;
3240   }
3241 }
3242 
3243 // set the error string
3244 void HeapDumper::set_error(char const* error) {
3245   if (_error != nullptr) {
3246     os::free(_error);
3247   }
3248   if (error == nullptr) {
3249     _error = nullptr;
3250   } else {
3251     _error = os::strdup(error);
3252     assert(_error != nullptr, "allocation failure");
3253   }
3254 }
3255 
3256 // Called by out-of-memory error reporting by a single Java thread
3257 // outside of a JVM safepoint
3258 void HeapDumper::dump_heap_from_oome() {
3259   HeapDumper::dump_heap(true);
3260 }
3261 
3262 // Called by error reporting by a single Java thread outside of a JVM safepoint,
3263 // or by heap dumping by the VM thread during a (GC) safepoint. Thus, these various
3264 // callers are strictly serialized and guaranteed not to interfere below. For more
3265 // general use, however, this method will need modification to prevent
3266 // inteference when updating the static variables base_path and dump_file_seq below.
3267 void HeapDumper::dump_heap() {
3268   HeapDumper::dump_heap(false);
3269 }
3270 
3271 void HeapDumper::dump_heap(bool oome) {
3272   static char base_path[JVM_MAXPATHLEN] = {'\0'};
3273   static uint dump_file_seq = 0;
3274   char* my_path;
3275   const int max_digit_chars = 20;
3276 
3277   const char* dump_file_name = "java_pid";
3278   const char* dump_file_ext  = HeapDumpGzipLevel > 0 ? ".hprof.gz" : ".hprof";
3279 
3280   // The dump file defaults to java_pid<pid>.hprof in the current working
3281   // directory. HeapDumpPath=<file> can be used to specify an alternative
3282   // dump file name or a directory where dump file is created.
3283   if (dump_file_seq == 0) { // first time in, we initialize base_path
3284     // Calculate potentially longest base path and check if we have enough
3285     // allocated statically.
3286     const size_t total_length =
3287                       (HeapDumpPath == nullptr ? 0 : strlen(HeapDumpPath)) +
3288                       strlen(os::file_separator()) + max_digit_chars +
3289                       strlen(dump_file_name) + strlen(dump_file_ext) + 1;
3290     if (total_length > sizeof(base_path)) {
3291       warning("Cannot create heap dump file.  HeapDumpPath is too long.");
3292       return;
3293     }
3294 
3295     bool use_default_filename = true;
3296     if (HeapDumpPath == nullptr || HeapDumpPath[0] == '\0') {
3297       // HeapDumpPath=<file> not specified
3298     } else {
3299       strcpy(base_path, HeapDumpPath);
3300       // check if the path is a directory (must exist)
3301       DIR* dir = os::opendir(base_path);
3302       if (dir == nullptr) {
3303         use_default_filename = false;
3304       } else {
3305         // HeapDumpPath specified a directory. We append a file separator
3306         // (if needed).
3307         os::closedir(dir);
3308         size_t fs_len = strlen(os::file_separator());
3309         if (strlen(base_path) >= fs_len) {
3310           char* end = base_path;
3311           end += (strlen(base_path) - fs_len);
3312           if (strcmp(end, os::file_separator()) != 0) {
3313             strcat(base_path, os::file_separator());
3314           }
3315         }
3316       }
3317     }
3318     // If HeapDumpPath wasn't a file name then we append the default name
3319     if (use_default_filename) {
3320       const size_t dlen = strlen(base_path);  // if heap dump dir specified
3321       jio_snprintf(&base_path[dlen], sizeof(base_path)-dlen, "%s%d%s",
3322                    dump_file_name, os::current_process_id(), dump_file_ext);
3323     }
3324     const size_t len = strlen(base_path) + 1;
3325     my_path = (char*)os::malloc(len, mtInternal);
3326     if (my_path == nullptr) {
3327       warning("Cannot create heap dump file.  Out of system memory.");
3328       return;
3329     }
3330     strncpy(my_path, base_path, len);
3331   } else {
3332     // Append a sequence number id for dumps following the first
3333     const size_t len = strlen(base_path) + max_digit_chars + 2; // for '.' and \0
3334     my_path = (char*)os::malloc(len, mtInternal);
3335     if (my_path == nullptr) {
3336       warning("Cannot create heap dump file.  Out of system memory.");
3337       return;
3338     }
3339     jio_snprintf(my_path, len, "%s.%d", base_path, dump_file_seq);
3340   }
3341   dump_file_seq++;   // increment seq number for next time we dump
3342 
3343   HeapDumper dumper(false /* no GC before heap dump */,
3344                     oome  /* pass along out-of-memory-error flag */);
3345   dumper.dump(my_path, tty, HeapDumpGzipLevel);
3346   os::free(my_path);
3347 }