src/hotspot/share/cds/aotMappedHeapWriter.cpp - master - githubmirror/panama-vector

src/hotspot/share/cds/aotMappedHeapWriter.cpp
974 строки39 KB
David Simms
8389219: Implement JEP 401: Value Objects (Preview)
31 июл 2026, 03:45
31 июл 2026, 03:45cc278db
/*
 * Copyright (c) 2023, 2026, Oracle and/or its affiliates. All rights reserved.
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 * or visit www.oracle.com if you need additional information or have any
 * questions.
 *
 */

#include "cds/aotMappedHeap.hpp"
#include "cds/aotMappedHeapWriter.hpp"
#include "cds/aotReferenceObjSupport.hpp"
#include "cds/cdsConfig.hpp"
#include "cds/filemap.hpp"
#include "cds/heapShared.inline.hpp"
#include "cds/regeneratedClasses.hpp"
#include "classfile/javaClasses.hpp"
#include "classfile/modules.hpp"
#include "classfile/systemDictionary.hpp"
#include "gc/shared/collectedHeap.hpp"
#include "memory/allocation.inline.hpp"
#include "memory/iterator.inline.hpp"
#include "memory/oopFactory.hpp"
#include "memory/universe.hpp"
#include "oops/compressedOops.hpp"
#include "oops/objArrayOop.inline.hpp"
#include "oops/oop.inline.hpp"
#include "oops/oopHandle.inline.hpp"
#include "oops/typeArrayKlass.hpp"
#include "oops/typeArrayOop.hpp"
#include "runtime/arguments.hpp"
#include "runtime/java.hpp"
#include "runtime/mutexLocker.hpp"
#include "utilities/bitMap.inline.hpp"
#if INCLUDE_G1GC
#include "gc/g1/g1CollectedHeap.hpp"
#include "gc/g1/g1HeapRegion.hpp"
#endif

#if INCLUDE_CDS_JAVA_HEAP

GrowableArrayCHeap<u1, mtClassShared>* AOTMappedHeapWriter::_buffer = nullptr;

bool AOTMappedHeapWriter::_is_writing_deterministic_heap = false;
size_t AOTMappedHeapWriter::_buffer_used;

// Heap root segments
HeapRootSegments AOTMappedHeapWriter::_heap_root_segments;

address AOTMappedHeapWriter::_requested_bottom;
address AOTMappedHeapWriter::_requested_top;

static size_t _num_strings = 0;
static size_t _string_bytes = 0;
static size_t _num_packages = 0;
static size_t _num_protection_domains = 0;

GrowableArrayCHeap<AOTMappedHeapWriter::NativePointerInfo, mtClassShared>* AOTMappedHeapWriter::_native_pointers;
GrowableArrayCHeap<oop, mtClassShared>* AOTMappedHeapWriter::_source_objs;
GrowableArrayCHeap<AOTMappedHeapWriter::HeapObjOrder, mtClassShared>* AOTMappedHeapWriter::_source_objs_order;

AOTMappedHeapWriter::BufferOffsetToSourceObjectTable*
AOTMappedHeapWriter::_buffer_offset_to_source_obj_table = nullptr;

typedef HashTable<
      size_t,    // offset of a filler from AOTMappedHeapWriter::buffer_bottom()
      size_t,    // size of this filler (in bytes)
      127,       // prime number
      AnyObj::C_HEAP,
      mtClassShared> FillersTable;
static FillersTable* _fillers;
static int _num_native_ptrs = 0;

void AOTMappedHeapWriter::init() {
  if (CDSConfig::is_dumping_heap()) {
    Universe::heap()->collect(GCCause::_java_lang_system_gc);

    _buffer_offset_to_source_obj_table = new (mtClassShared) BufferOffsetToSourceObjectTable(/*size (prime)*/36137, /*max size*/1 * M);
    _fillers = new (mtClassShared) FillersTable();
    _requested_bottom = nullptr;
    _requested_top = nullptr;

    _native_pointers = new GrowableArrayCHeap<NativePointerInfo, mtClassShared>(2048);
    _source_objs = new GrowableArrayCHeap<oop, mtClassShared>(10000);

    guarantee(MIN_GC_REGION_ALIGNMENT <= G1HeapRegion::min_region_size_in_words() * HeapWordSize, "must be");

    if (CDSConfig::old_cds_flags_used()) {
      // With the old CDS workflow, we can guatantee determninistic output: given
      // the same classlist file, we can generate the same static CDS archive.
      // To ensure determinism, we always use the same compressed oop encoding
      // (zero-based, no shift). See set_requested_address_range().
      _is_writing_deterministic_heap = true;
    } else {
      // Determninistic output is not supported by the new AOT workflow, so
      // we don't force the (zero-based, no shift) encoding. This way, it is more
      // likely that we can avoid oop relocation in the production run.
      _is_writing_deterministic_heap = false;
    }
  }
}

// For AOTMappedHeapWriter::narrow_oop_{mode, base, shift}(), see comments
// in AOTMappedHeapWriter::set_requested_address_range(),
CompressedOops::Mode AOTMappedHeapWriter::narrow_oop_mode() {
  if (is_writing_deterministic_heap()) {
    return CompressedOops::UnscaledNarrowOop;
  } else {
    return CompressedOops::mode();
  }
}

address AOTMappedHeapWriter::narrow_oop_base() {
  if (is_writing_deterministic_heap()) {
    return nullptr;
  } else {
    return CompressedOops::base();
  }
}

int AOTMappedHeapWriter::narrow_oop_shift() {
  if (is_writing_deterministic_heap()) {
    return 0;
  } else {
    return CompressedOops::shift();
  }
}

void AOTMappedHeapWriter::delete_tables_with_raw_oops() {
  delete _source_objs;
  _source_objs = nullptr;
}

void AOTMappedHeapWriter::add_source_obj(oop src_obj) {
  _source_objs->append(src_obj);
}

void AOTMappedHeapWriter::write(GrowableArrayCHeap<oop, mtClassShared>* roots,
                                AOTMappedHeapInfo* heap_info) {
  assert(CDSConfig::is_dumping_heap(), "sanity");
  allocate_buffer();
  copy_source_objs_to_buffer(roots);
  set_requested_address_range(heap_info);
  relocate_embedded_oops(roots, heap_info);
}

bool AOTMappedHeapWriter::is_too_large_to_archive(oop o) {
  return is_too_large_to_archive(o->size());
}

bool AOTMappedHeapWriter::is_string_too_large_to_archive(oop string) {
  typeArrayOop value = java_lang_String::value_no_keepalive(string);
  return is_too_large_to_archive(value);
}

bool AOTMappedHeapWriter::is_too_large_to_archive(size_t size) {
  assert(size > 0, "no zero-size object");
  assert(size * HeapWordSize > size, "no overflow");
  static_assert(MIN_GC_REGION_ALIGNMENT > 0, "must be positive");

  size_t byte_size = size * HeapWordSize;
  if (byte_size > size_t(MIN_GC_REGION_ALIGNMENT)) {
    return true;
  } else {
    return false;
  }
}

// Various lookup functions between source_obj, buffered_obj and requested_obj
bool AOTMappedHeapWriter::is_in_requested_range(oop o) {
  assert(_requested_bottom != nullptr, "do not call before _requested_bottom is initialized");
  address a = cast_from_oop<address>(o);
  return (_requested_bottom <= a && a < _requested_top);
}

oop AOTMappedHeapWriter::requested_obj_from_buffer_offset(size_t offset) {
  oop req_obj = cast_to_oop(_requested_bottom + offset);
  assert(is_in_requested_range(req_obj), "must be");
  return req_obj;
}

oop AOTMappedHeapWriter::source_obj_to_requested_obj(oop src_obj) {
  assert(CDSConfig::is_dumping_heap(), "dump-time only");
  HeapShared::CachedOopInfo* p = HeapShared::get_cached_oop_info(src_obj);
  if (p != nullptr) {
    return requested_obj_from_buffer_offset(p->buffer_offset());
  } else {
    return nullptr;
  }
}

oop AOTMappedHeapWriter::buffered_addr_to_source_obj(address buffered_addr) {
  OopHandle* oh = _buffer_offset_to_source_obj_table->get(buffered_address_to_offset(buffered_addr));
  if (oh != nullptr) {
    return oh->resolve();
  } else {
    return nullptr;
  }
}

Klass* AOTMappedHeapWriter::real_klass_of_buffered_oop(address buffered_addr) {
  oop p = buffered_addr_to_source_obj(buffered_addr);
  if (p != nullptr) {
    return p->klass();
  } else if (get_filler_size_at(buffered_addr) > 0) {
    return Universe::fillerArrayKlass();
  } else {
    // This is one of the root segments
    return Universe::objectArrayKlass();
  }
}

size_t AOTMappedHeapWriter::size_of_buffered_oop(address buffered_addr) {
  oop p = buffered_addr_to_source_obj(buffered_addr);
  if (p != nullptr) {
    return p->size();
  }

  size_t nbytes = get_filler_size_at(buffered_addr);
  if (nbytes > 0) {
    assert((nbytes % BytesPerWord) == 0, "should be aligned");
    return nbytes / BytesPerWord;
  }

  address hrs = buffer_bottom();
  for (size_t seg_idx = 0; seg_idx < _heap_root_segments.count(); seg_idx++) {
    nbytes = _heap_root_segments.size_in_bytes(seg_idx);
    if (hrs == buffered_addr) {
      assert((nbytes % BytesPerWord) == 0, "should be aligned");
      return nbytes / BytesPerWord;
    }
    hrs += nbytes;
  }

  ShouldNotReachHere();
  return 0;
}

address AOTMappedHeapWriter::buffered_addr_to_requested_addr(address buffered_addr) {
  return _requested_bottom + buffered_address_to_offset(buffered_addr);
}

address AOTMappedHeapWriter::requested_address() {
  assert(_buffer != nullptr, "must be initialized");
  return _requested_bottom;
}

void AOTMappedHeapWriter::allocate_buffer() {
  int initial_buffer_size = 100000;
  _buffer = new GrowableArrayCHeap<u1, mtClassShared>(initial_buffer_size);
  _buffer_used = 0;
  ensure_buffer_space(1); // so that buffer_bottom() works
}

void AOTMappedHeapWriter::ensure_buffer_space(size_t min_bytes) {
  // We usually have very small heaps. If we get a huge one it's probably caused by a bug.
  guarantee(min_bytes <= max_jint, "we dont support archiving more than 2G of objects");
  _buffer->at_grow(to_array_index(min_bytes));
}

objArrayOop AOTMappedHeapWriter::allocate_root_segment(size_t offset, int element_count) {
  HeapWord* mem = offset_to_buffered_address<HeapWord *>(offset);
  memset(mem, 0, refArrayOopDesc::object_size(element_count));

  // The initialization code is copied from MemAllocator::finish and ObjArrayAllocator::initialize.
  if (UseCompactObjectHeaders) {
    oopDesc::release_set_mark(mem, Universe::objectArrayKlass()->prototype_header());
  } else {
    assert(!Arguments::is_valhalla_enabled() || Universe::objectArrayKlass()->prototype_header() == markWord::prototype(), "should be the same");
    oopDesc::set_mark(mem, markWord::prototype());
    oopDesc::release_set_klass(mem, Universe::objectArrayKlass());
  }
  arrayOopDesc::set_length(mem, element_count);
  return objArrayOop(cast_to_oop(mem));
}

void AOTMappedHeapWriter::root_segment_at_put(objArrayOop segment, int index, oop root) {
  // Do not use arrayOop->obj_at_put(i, o) as arrayOop is outside the real heap!
  if (UseCompressedOops) {
    *segment->obj_at_addr<narrowOop>(index) = CompressedOops::encode(root);
  } else {
    *segment->obj_at_addr<oop>(index) = root;
  }
}

void AOTMappedHeapWriter::copy_roots_to_buffer(GrowableArrayCHeap<oop, mtClassShared>* roots) {
  // Depending on the number of classes we are archiving, a single roots array may be
  // larger than MIN_GC_REGION_ALIGNMENT. Roots are allocated first in the buffer, which
  // allows us to chop the large array into a series of "segments". Current layout
  // starts with zero or more segments exactly fitting MIN_GC_REGION_ALIGNMENT, and end
  // with a single segment that may be smaller than MIN_GC_REGION_ALIGNMENT.
  // This is simple and efficient. We do not need filler objects anywhere between the segments,
  // or immediately after the last segment. This allows starting the object dump immediately
  // after the roots.

  assert((_buffer_used % MIN_GC_REGION_ALIGNMENT) == 0,
         "Pre-condition: Roots start at aligned boundary: %zu", _buffer_used);

  int max_elem_count = ((MIN_GC_REGION_ALIGNMENT - arrayOopDesc::header_size_in_bytes()) / heapOopSize);
  assert(refArrayOopDesc::object_size(max_elem_count)*HeapWordSize == MIN_GC_REGION_ALIGNMENT,
         "Should match exactly");

  HeapRootSegments segments(_buffer_used,
                            roots->length(),
                            MIN_GC_REGION_ALIGNMENT,
                            max_elem_count);

  int root_index = 0;
  for (size_t seg_idx = 0; seg_idx < segments.count(); seg_idx++) {
    int size_elems = segments.size_in_elems(seg_idx);
    size_t size_bytes = segments.size_in_bytes(seg_idx);

    size_t oop_offset = _buffer_used;
    _buffer_used = oop_offset + size_bytes;
    ensure_buffer_space(_buffer_used);

    assert((oop_offset % MIN_GC_REGION_ALIGNMENT) == 0,
           "Roots segment %zu start is not aligned: %zu",
           segments.count(), oop_offset);

    objArrayOop seg_oop = allocate_root_segment(oop_offset, size_elems);
    for (int i = 0; i < size_elems; i++) {
      root_segment_at_put(seg_oop, i, roots->at(root_index++));
    }

    log_info(aot, heap)("archived obj root segment [%d] = %zu bytes, obj = " PTR_FORMAT,
                        size_elems, size_bytes, p2i(seg_oop));
  }

  assert(root_index == roots->length(), "Post-condition: All roots are handled");

  _heap_root_segments = segments;
}

// The goal is to sort the objects in increasing order of:
// - objects that have only oop pointers
// - objects that have both native and oop pointers
// - objects that have only native pointers
// - objects that have no pointers
static int oop_sorting_rank(oop o) {
  bool has_oop_ptr, has_native_ptr;
  HeapShared::get_pointer_info(o, has_oop_ptr, has_native_ptr);

  if (has_oop_ptr) {
    if (!has_native_ptr) {
      return 0;
    } else {
      return 1;
    }
  } else {
    if (has_native_ptr) {
      return 2;
    } else {
      return 3;
    }
  }
}

int AOTMappedHeapWriter::compare_objs_by_oop_fields(HeapObjOrder* a, HeapObjOrder* b) {
  int rank_a = a->_rank;
  int rank_b = b->_rank;

  if (rank_a != rank_b) {
    return rank_a - rank_b;
  } else {
    // If they are the same rank, sort them by their position in the _source_objs array
    return a->_index - b->_index;
  }
}

void AOTMappedHeapWriter::sort_source_objs() {
  log_info(aot)("sorting heap objects");
  int len = _source_objs->length();
  _source_objs_order = new GrowableArrayCHeap<HeapObjOrder, mtClassShared>(len);

  for (int i = 0; i < len; i++) {
    oop o = _source_objs->at(i);
    int rank = oop_sorting_rank(o);
    HeapObjOrder os = {i, rank};
    _source_objs_order->append(os);
  }
  log_info(aot)("computed ranks");
  _source_objs_order->sort(compare_objs_by_oop_fields);
  log_info(aot)("sorting heap objects done");
}

void AOTMappedHeapWriter::copy_source_objs_to_buffer(GrowableArrayCHeap<oop, mtClassShared>* roots) {
  // There could be multiple root segments, which we want to be aligned by region.
  // Putting them ahead of objects makes sure we waste no space.
  copy_roots_to_buffer(roots);

  sort_source_objs();
  for (int i = 0; i < _source_objs_order->length(); i++) {
    int src_obj_index = _source_objs_order->at(i)._index;
    oop src_obj = _source_objs->at(src_obj_index);
    HeapShared::CachedOopInfo* info = HeapShared::get_cached_oop_info(src_obj);
    assert(info != nullptr, "must be");
    size_t buffer_offset = copy_one_source_obj_to_buffer(src_obj);
    info->set_buffer_offset(buffer_offset);
    assert(buffer_offset <= 0x7fffffff, "sanity");

    OopHandle handle(Universe::vm_global(), src_obj);
    _buffer_offset_to_source_obj_table->put_when_absent(buffer_offset, handle);
    _buffer_offset_to_source_obj_table->maybe_grow();

    if (java_lang_Module::is_instance(src_obj)) {
      Modules::check_archived_module_oop(src_obj);
    }
  }

  log_info(aot)("Size of heap region = %zu bytes, %d objects, %d roots, %d native ptrs",
                _buffer_used, _source_objs->length() + 1, roots->length(), _num_native_ptrs);
  log_info(aot)("   strings            = %8zu (%zu bytes)", _num_strings, _string_bytes);
  log_info(aot)("   packages           = %8zu", _num_packages);
  log_info(aot)("   protection domains = %8zu", _num_protection_domains);
}

size_t AOTMappedHeapWriter::filler_array_byte_size(int length) {
  size_t byte_size = refArrayOopDesc::object_size(length) * HeapWordSize;
  return byte_size;
}

int AOTMappedHeapWriter::filler_array_length(size_t fill_bytes) {
  assert(is_object_aligned(fill_bytes), "must be");
  size_t elemSize = (UseCompressedOops ? sizeof(narrowOop) : sizeof(oop));

  int initial_length = to_array_length(fill_bytes / elemSize);
  for (int length = initial_length; length >= 0; length --) {
    size_t array_byte_size = filler_array_byte_size(length);
    if (array_byte_size == fill_bytes) {
      return length;
    }
  }

  ShouldNotReachHere();
  return -1;
}

HeapWord* AOTMappedHeapWriter::init_filler_array_at_buffer_top(int array_length, size_t fill_bytes) {
  Klass* oak = Universe::objectArrayKlass(); // already relocated to point to archived klass
  HeapWord* mem = offset_to_buffered_address<HeapWord*>(_buffer_used);
  memset(mem, 0, fill_bytes);
  narrowKlass nk = ArchiveBuilder::current()->get_requested_narrow_klass(oak);
  if (UseCompactObjectHeaders) {
    oopDesc::release_set_mark(mem, markWord::prototype().set_narrow_klass(nk));
  } else {
    assert(!Arguments::is_valhalla_enabled() || Universe::objectArrayKlass()->prototype_header() == markWord::prototype(), "should be the same");
    oopDesc::set_mark(mem, markWord::prototype());
    cast_to_oop(mem)->set_narrow_klass(nk);
  }
  arrayOopDesc::set_length(mem, array_length);
  return mem;
}

void AOTMappedHeapWriter::maybe_fill_gc_region_gap(size_t required_byte_size) {
  // We fill only with arrays (so we don't need to use a single HeapWord filler if the
  // leftover space is smaller than a zero-sized array object). Therefore, we need to
  // make sure there's enough space of min_filler_byte_size in the current region after
  // required_byte_size has been allocated. If not, fill the remainder of the current
  // region.
  size_t min_filler_byte_size = filler_array_byte_size(0);
  size_t new_used = _buffer_used + required_byte_size + min_filler_byte_size;

  const size_t cur_min_region_bottom = align_down(_buffer_used, MIN_GC_REGION_ALIGNMENT);
  const size_t next_min_region_bottom = align_down(new_used, MIN_GC_REGION_ALIGNMENT);

  if (cur_min_region_bottom != next_min_region_bottom) {
    // Make sure that no objects span across MIN_GC_REGION_ALIGNMENT. This way
    // we can map the region in any region-based collector.
    assert(next_min_region_bottom > cur_min_region_bottom, "must be");
    assert(next_min_region_bottom - cur_min_region_bottom == MIN_GC_REGION_ALIGNMENT,
           "no buffered object can be larger than %d bytes",  MIN_GC_REGION_ALIGNMENT);

    const size_t filler_end = next_min_region_bottom;
    const size_t fill_bytes = filler_end - _buffer_used;
    assert(fill_bytes > 0, "must be");
    ensure_buffer_space(filler_end);

    int array_length = filler_array_length(fill_bytes);
    log_info(aot, heap)("Inserting filler obj array of %d elements (%zu bytes total) @ buffer offset %zu",
                        array_length, fill_bytes, _buffer_used);
    HeapWord* filler = init_filler_array_at_buffer_top(array_length, fill_bytes);
    _buffer_used = filler_end;
    _fillers->put(buffered_address_to_offset((address)filler), fill_bytes);
  }
}

size_t AOTMappedHeapWriter::get_filler_size_at(address buffered_addr) {
  size_t* p = _fillers->get(buffered_address_to_offset(buffered_addr));
  if (p != nullptr) {
    assert(*p > 0, "filler must be larger than zero bytes");
    return *p;
  } else {
    return 0; // buffered_addr is not a filler
  }
}

template <typename T>
void update_buffered_object_field(address buffered_obj, int field_offset, T value) {
  T* field_addr = cast_to_oop(buffered_obj)->field_addr<T>(field_offset);
  *field_addr = value;
}

void AOTMappedHeapWriter::update_stats(oop src_obj) {
  if (java_lang_String::is_instance(src_obj)) {
    _num_strings ++;
    _string_bytes += src_obj->size() * HeapWordSize;
    _string_bytes += java_lang_String::value(src_obj)->size() * HeapWordSize;
  } else {
    Klass* k = src_obj->klass();
    Symbol* name = k->name();
    if (name->equals("java/lang/NamedPackage") || name->equals("java/lang/Package")) {
      _num_packages ++;
    } else if (name->equals("java/security/ProtectionDomain")) {
      _num_protection_domains ++;
    }
  }
}

size_t AOTMappedHeapWriter::copy_one_source_obj_to_buffer(oop src_obj) {
  update_stats(src_obj);

  assert(!is_too_large_to_archive(src_obj), "already checked");
  size_t byte_size = src_obj->size() * HeapWordSize;
  assert(byte_size > 0, "no zero-size objects");

  // For region-based collectors such as G1, the archive heap may be mapped into
  // multiple regions. We need to make sure that we don't have an object that can possible
  // span across two regions.
  maybe_fill_gc_region_gap(byte_size);

  size_t new_used = _buffer_used + byte_size;
  assert(new_used > _buffer_used, "no wrap around");

  size_t cur_min_region_bottom = align_down(_buffer_used, MIN_GC_REGION_ALIGNMENT);
  size_t next_min_region_bottom = align_down(new_used, MIN_GC_REGION_ALIGNMENT);
  assert(cur_min_region_bottom == next_min_region_bottom, "no object should cross minimal GC region boundaries");

  ensure_buffer_space(new_used);

  address from = cast_from_oop<address>(src_obj);
  address to = offset_to_buffered_address<address>(_buffer_used);
  assert(is_object_aligned(_buffer_used), "sanity");
  assert(is_object_aligned(byte_size), "sanity");
  memcpy(to, from, byte_size);

  // These native pointers will be restored explicitly at run time.
  if (java_lang_Module::is_instance(src_obj)) {
    update_buffered_object_field<ModuleEntry*>(to, java_lang_Module::module_entry_offset(), nullptr);
  } else if (java_lang_ClassLoader::is_instance(src_obj)) {
#ifdef ASSERT
    // We only archive these loaders
    if (src_obj != SystemDictionary::java_platform_loader() &&
        src_obj != SystemDictionary::java_system_loader()) {
      assert(src_obj->klass()->name()->equals("jdk/internal/loader/ClassLoaders$BootClassLoader"), "must be");
    }
#endif
    update_buffered_object_field<ClassLoaderData*>(to, java_lang_ClassLoader::loader_data_offset(), nullptr);
  }

  size_t buffered_obj_offset = _buffer_used;
  _buffer_used = new_used;

  return buffered_obj_offset;
}

// Set the range [_requested_bottom, _requested_top), the requested address range of all
// the archived heap objects in the production run.
//
// (1) UseCompressedOops == true && !is_writing_deterministic_heap()
//
//     The archived objects are stored using the COOPS encoding of the assembly phase.
//     We pick a range within the heap used by the assembly phase.
//
//     In the production run, if different COOPS encodings are used:
//         - The heap contents needs to be relocated.
//
// (2) UseCompressedOops == true && is_writing_deterministic_heap()
//
//     We always use zero-based, zero-shift encoding. _requested_top is aligned to 0x10000000.
//
// (3) UseCompressedOops == false:
//
//     In the production run, the heap range is usually picked (randomly) by the OS, so we
//     will almost always need to perform relocation, regardless of how we pick the requested
//     address range.
//
//     So we just hard code it to NOCOOPS_REQUESTED_BASE.
//
void AOTMappedHeapWriter::set_requested_address_range(AOTMappedHeapInfo* info) {
  assert(!info->is_used(), "only set once");

  size_t heap_region_byte_size = _buffer_used;
  assert(heap_region_byte_size > 0, "must archived at least one object!");

  if (UseCompressedOops) {
    if (is_writing_deterministic_heap()) {
      // Pick a heap range so that requested addresses can be encoded with zero-base/no shift.
      // We align the requested bottom to at least 1 MB: if the production run uses G1 with a small
      // heap (e.g., -Xmx256m), it's likely that we can map the archived objects at the
      // requested location to avoid relocation.
      //
      // For other collectors or larger heaps, relocation is unavoidable, but is usually
      // quite cheap. If you really want to avoid relocation, use the AOT workflow instead.
      address heap_end = (address)0x100000000;
      size_t alignment = MAX2(MIN_GC_REGION_ALIGNMENT, 1024 * 1024);
      if (align_up(heap_region_byte_size, alignment) >= (size_t)heap_end) {
        log_error(aot, heap)("cached heap space is too large: %zu bytes", heap_region_byte_size);
        AOTMetaspace::unrecoverable_writing_error();
      }
      _requested_bottom = align_down(heap_end - heap_region_byte_size, alignment);
    } else if (UseG1GC) {
      // For G1, pick the range at the top of the current heap. If the exact same heap sizes
      // are used in the production run, it's likely that we can map the archived objects
      // at the requested location to avoid relocation.
      address heap_end = (address)G1CollectedHeap::heap()->reserved().end();
      log_info(aot, heap)("Heap end = %p", heap_end);
      _requested_bottom = align_down(heap_end - heap_region_byte_size, G1HeapRegion::GrainBytes);
      _requested_bottom = align_down(_requested_bottom, MIN_GC_REGION_ALIGNMENT);
      assert(is_aligned(_requested_bottom, G1HeapRegion::GrainBytes), "sanity");
    } else {
      _requested_bottom = align_up(CompressedOops::begin(), MIN_GC_REGION_ALIGNMENT);
    }
  } else {
    // We always write the objects as if the heap started at this address. This
    // makes the contents of the archive heap deterministic.
    //
    // Note that at runtime, the heap address is selected by the OS, so the archive
    // heap will not be mapped at 0x10000000, and the contents need to be patched.
    _requested_bottom = align_up((address)NOCOOPS_REQUESTED_BASE, MIN_GC_REGION_ALIGNMENT);
  }

  assert(is_aligned(_requested_bottom, MIN_GC_REGION_ALIGNMENT), "sanity");

  _requested_top = _requested_bottom + _buffer_used;

  info->set_buffer_region(MemRegion(offset_to_buffered_address<HeapWord*>(0),
                                    offset_to_buffered_address<HeapWord*>(_buffer_used)));
  info->set_root_segments(_heap_root_segments);
}

// Oop relocation

template <typename T> T* AOTMappedHeapWriter::requested_addr_to_buffered_addr(T* p) {
  assert(is_in_requested_range(cast_to_oop(p)), "must be");

  address addr = address(p);
  assert(addr >= _requested_bottom, "must be");
  size_t offset = addr - _requested_bottom;
  return offset_to_buffered_address<T*>(offset);
}

template <typename T> oop AOTMappedHeapWriter::load_source_oop_from_buffer(T* buffered_addr) {
  oop o = load_oop_from_buffer(buffered_addr);
  assert(!in_buffer(cast_from_oop<address>(o)), "must point to source oop");
  return o;
}

template <typename T> void AOTMappedHeapWriter::store_requested_oop_in_buffer(T* buffered_addr,
                                                                                   oop request_oop) {
  assert(request_oop == nullptr || is_in_requested_range(request_oop), "must be");
  store_oop_in_buffer(buffered_addr, request_oop);
}

inline void AOTMappedHeapWriter::store_oop_in_buffer(oop* buffered_addr, oop requested_obj) {
  *buffered_addr = requested_obj;
}

inline void AOTMappedHeapWriter::store_oop_in_buffer(narrowOop* buffered_addr, oop requested_obj) {
  narrowOop val = CompressedOops::encode(requested_obj);
  *buffered_addr = val;
}

oop AOTMappedHeapWriter::load_oop_from_buffer(oop* buffered_addr) {
  return *buffered_addr;
}

oop AOTMappedHeapWriter::load_oop_from_buffer(narrowOop* buffered_addr) {
  return CompressedOops::decode(*buffered_addr);
}

template <typename T> void AOTMappedHeapWriter::relocate_field_in_buffer(T* field_addr_in_buffer, oop source_referent, CHeapBitMap* oopmap) {
  oop request_referent = source_obj_to_requested_obj(source_referent);
  if (UseCompressedOops && is_writing_deterministic_heap()) {
    // We use zero-based, 0-shift encoding, so the narrowOop is just the lower
    // 32 bits of request_referent
    intptr_t addr = cast_from_oop<intptr_t>(request_referent);
    *((narrowOop*)field_addr_in_buffer) = CompressedOops::narrow_oop_cast(addr);
  } else {
    store_requested_oop_in_buffer<T>(field_addr_in_buffer, request_referent);
  }
  if (request_referent != nullptr) {
    mark_oop_pointer<T>(field_addr_in_buffer, oopmap);
  }
}

template <typename T> void AOTMappedHeapWriter::mark_oop_pointer(T* buffered_addr, CHeapBitMap* oopmap) {
  T* request_p = (T*)(buffered_addr_to_requested_addr((address)buffered_addr));
  address requested_region_bottom;

  assert(request_p >= (T*)_requested_bottom, "sanity");
  assert(request_p <  (T*)_requested_top, "sanity");
  requested_region_bottom = _requested_bottom;

  // Mark the pointer in the oopmap
  T* region_bottom = (T*)requested_region_bottom;
  assert(request_p >= region_bottom, "must be");
  BitMap::idx_t idx = request_p - region_bottom;
  assert(idx < oopmap->size(), "overflow");
  oopmap->set_bit(idx);
}

void AOTMappedHeapWriter::update_header_for_requested_obj(oop requested_obj, oop src_obj,  Klass* src_klass) {
  narrowKlass nk = ArchiveBuilder::current()->get_requested_narrow_klass(src_klass);
  address buffered_addr = requested_addr_to_buffered_addr(cast_from_oop<address>(requested_obj));

  oop fake_oop = cast_to_oop(buffered_addr);
  if (UseCompactObjectHeaders) {
    markWord prototype_header = src_klass->prototype_header().set_narrow_klass(nk);
    fake_oop->set_mark(prototype_header);
  } else {
    fake_oop->set_narrow_klass(nk);
  }

  if (src_obj == nullptr) {
    return;
  }
  // We need to retain the identity_hash, because it may have been used by some hashtables
  // in the shared heap.
  if (!src_obj->fast_no_hash_check() && (!(Arguments::is_valhalla_enabled() && src_obj->mark().is_inline_type()))) {
    intptr_t src_hash = src_obj->identity_hash();
    if (UseCompactObjectHeaders) {
      fake_oop->set_mark(fake_oop->mark().copy_set_hash(src_hash));
    } else if (Arguments::is_valhalla_enabled()) {
      fake_oop->set_mark(src_klass->prototype_header().copy_set_hash(src_hash));
    } else {
      fake_oop->set_mark(markWord::prototype().copy_set_hash(src_hash));
    }
    assert(fake_oop->mark().is_unlocked(), "sanity");

    DEBUG_ONLY(intptr_t archived_hash = fake_oop->identity_hash());
    assert(src_hash == archived_hash, "Different hash codes: original " INTPTR_FORMAT ", archived " INTPTR_FORMAT, src_hash, archived_hash);
  }
  // Strip age bits.
  fake_oop->set_mark(fake_oop->mark().set_age(0));
}

class AOTMappedHeapWriter::EmbeddedOopRelocator: public BasicOopIterateClosure {
  oop _src_obj;
  address _buffered_obj;
  CHeapBitMap* _oopmap;
  bool _is_java_lang_ref;
public:
  EmbeddedOopRelocator(oop src_obj, address buffered_obj, CHeapBitMap* oopmap) :
    _src_obj(src_obj), _buffered_obj(buffered_obj), _oopmap(oopmap)
  {
    _is_java_lang_ref = AOTReferenceObjSupport::check_if_ref_obj(src_obj);
  }

  void do_oop(narrowOop *p) { EmbeddedOopRelocator::do_oop_work(p); }
  void do_oop(      oop *p) { EmbeddedOopRelocator::do_oop_work(p); }

private:
  template <class T> void do_oop_work(T *p) {
    int field_offset = pointer_delta_as_int((char*)p, cast_from_oop<char*>(_src_obj));
    T* field_addr = (T*)(_buffered_obj + field_offset);
    oop referent = load_source_oop_from_buffer<T>(field_addr);
    referent = HeapShared::maybe_remap_referent(_is_java_lang_ref, field_offset, referent);
    AOTMappedHeapWriter::relocate_field_in_buffer<T>(field_addr, referent, _oopmap);
  }
};

static void log_bitmap_usage(const char* which, BitMap* bitmap, size_t total_bits) {
  // The whole heap is covered by total_bits, but there are only non-zero bits within [start ... end).
  size_t start = bitmap->find_first_set_bit(0);
  size_t end = bitmap->size();
  log_info(aot)("%s = %7zu ... %7zu (%3zu%% ... %3zu%% = %3zu%%)", which,
                start, end,
                start * 100 / total_bits,
                end * 100 / total_bits,
                (end - start) * 100 / total_bits);
}

// Update all oop fields embedded in the buffered objects
void AOTMappedHeapWriter::relocate_embedded_oops(GrowableArrayCHeap<oop, mtClassShared>* roots,
                                                      AOTMappedHeapInfo* heap_info) {
  size_t oopmap_unit = (UseCompressedOops ? sizeof(narrowOop) : sizeof(oop));
  size_t heap_region_byte_size = _buffer_used;
  heap_info->oopmap()->resize(heap_region_byte_size   / oopmap_unit);

  for (int i = 0; i < _source_objs_order->length(); i++) {
    int src_obj_index = _source_objs_order->at(i)._index;
    oop src_obj = _source_objs->at(src_obj_index);
    HeapShared::CachedOopInfo* info = HeapShared::get_cached_oop_info(src_obj);
    assert(info != nullptr, "must be");
    oop requested_obj = requested_obj_from_buffer_offset(info->buffer_offset());
    update_header_for_requested_obj(requested_obj, src_obj, src_obj->klass());
    address buffered_obj = offset_to_buffered_address<address>(info->buffer_offset());
    EmbeddedOopRelocator relocator(src_obj, buffered_obj, heap_info->oopmap());
    src_obj->oop_iterate(&relocator);
    mark_native_pointers(src_obj);
  };

  // Relocate HeapShared::roots(), which is created in copy_roots_to_buffer() and
  // doesn't have a corresponding src_obj, so we can't use EmbeddedOopRelocator on it.
  for (size_t seg_idx = 0; seg_idx < _heap_root_segments.count(); seg_idx++) {
    size_t seg_offset = _heap_root_segments.segment_offset(seg_idx);

    objArrayOop requested_obj = (objArrayOop)requested_obj_from_buffer_offset(seg_offset);
    update_header_for_requested_obj(requested_obj, nullptr, Universe::objectArrayKlass());
    address buffered_obj = offset_to_buffered_address<address>(seg_offset);
    int length = _heap_root_segments.size_in_elems(seg_idx);

    size_t elem_size = UseCompressedOops ? sizeof(narrowOop) : sizeof(oop);

    for (int i = 0; i < length; i++) {
      // There is no source object; these are native oops - load, translate and
      // write back
      size_t elem_offset = objArrayOopDesc::base_offset_in_bytes() + elem_size * i;
      HeapWord* elem_addr = (HeapWord*)(buffered_obj + elem_offset);
      oop obj = NativeAccess<>::oop_load(elem_addr);
      obj = HeapShared::maybe_remap_referent(false /* is_reference_field */, elem_offset, obj);
      if (UseCompressedOops) {
        relocate_field_in_buffer<narrowOop>((narrowOop*)elem_addr, obj, heap_info->oopmap());
      } else {
        relocate_field_in_buffer<oop>((oop*)elem_addr, obj, heap_info->oopmap());
      }
    }
  }

  compute_ptrmap(heap_info);

  size_t total_bytes = (size_t)_buffer->length();
  log_bitmap_usage("oopmap", heap_info->oopmap(), total_bytes / (UseCompressedOops ? sizeof(narrowOop) : sizeof(oop)));
  log_bitmap_usage("ptrmap", heap_info->ptrmap(), total_bytes / sizeof(address));
}

void AOTMappedHeapWriter::mark_native_pointer(oop src_obj, int field_offset) {
  Metadata* ptr = src_obj->metadata_field_acquire(field_offset);
  if (ptr != nullptr) {
    NativePointerInfo info;
    info._src_obj = src_obj;
    info._field_offset = field_offset;
    _native_pointers->append(info);
    HeapShared::set_has_native_pointers(src_obj);
    _num_native_ptrs ++;
  }
}

void AOTMappedHeapWriter::mark_native_pointers(oop orig_obj) {
  HeapShared::do_metadata_offsets(orig_obj, [&](int offset) {
    mark_native_pointer(orig_obj, offset);
  });
}

void AOTMappedHeapWriter::compute_ptrmap(AOTMappedHeapInfo* heap_info) {
  int num_non_null_ptrs = 0;
  Metadata** bottom = (Metadata**) _requested_bottom;
  Metadata** top = (Metadata**) _requested_top; // exclusive
  heap_info->ptrmap()->resize(top - bottom);

  BitMap::idx_t max_idx = 32; // paranoid - don't make it too small
  for (int i = 0; i < _native_pointers->length(); i++) {
    NativePointerInfo info = _native_pointers->at(i);
    oop src_obj = info._src_obj;
    int field_offset = info._field_offset;
    HeapShared::CachedOopInfo* p = HeapShared::get_cached_oop_info(src_obj);
    // requested_field_addr = the address of this field in the requested space
    oop requested_obj = requested_obj_from_buffer_offset(p->buffer_offset());
    Metadata** requested_field_addr = (Metadata**)(cast_from_oop<address>(requested_obj) + field_offset);
    assert(bottom <= requested_field_addr && requested_field_addr < top, "range check");

    // Mark this field in the bitmap
    BitMap::idx_t idx = requested_field_addr - bottom;
    heap_info->ptrmap()->set_bit(idx);
    num_non_null_ptrs ++;
    max_idx = MAX2(max_idx, idx);

    // Set the native pointer to the requested address of the metadata (at runtime, the metadata will have
    // this address if the RO/RW regions are mapped at the default location).

    Metadata** buffered_field_addr = requested_addr_to_buffered_addr(requested_field_addr);
    Metadata* native_ptr = *buffered_field_addr;
    guarantee(native_ptr != nullptr, "sanity");

    native_ptr = RegeneratedClasses::maybe_get_regenerated_object(native_ptr);

    if (!ArchiveBuilder::current()->has_been_archived((address)native_ptr)) {
      ResourceMark rm;
      LogStreamHandle(Error, aot) log;
      log.print("Marking native pointer for oop %p (type = %s, offset = %d)",
                cast_from_oop<void*>(src_obj), src_obj->klass()->external_name(), field_offset);
      src_obj->print_on(&log);
      fatal("Metadata %p should have been archived", native_ptr);
    }

    address buffered_native_ptr = ArchiveBuilder::current()->get_buffered_addr((address)native_ptr);
    address requested_native_ptr = ArchiveBuilder::current()->to_requested(buffered_native_ptr);
    *buffered_field_addr = (Metadata*)requested_native_ptr;
  }

  heap_info->ptrmap()->resize(max_idx + 1);
  log_info(aot, heap)("calculate_ptrmap: marked %d non-null native pointers for heap region (%zu bits)",
                      num_non_null_ptrs, size_t(heap_info->ptrmap()->size()));
}

AOTMapLogger::OopDataIterator* AOTMappedHeapWriter::oop_iterator(AOTMappedHeapInfo* heap_info) {
  class MappedWriterOopIterator : public AOTMappedHeapOopIterator {
  public:
    MappedWriterOopIterator(address buffer_start,
                            address buffer_end,
                            address requested_base,
                            address requested_start,
                            int requested_shift,
                            size_t num_root_segments) :
      AOTMappedHeapOopIterator(buffer_start,
                               buffer_end,
                               requested_base,
                               requested_start,
                               requested_shift,
                               num_root_segments) {}

    AOTMapLogger::OopData capture(address buffered_addr) override {
      oopDesc* raw_oop = (oopDesc*)buffered_addr;
      size_t size = size_of_buffered_oop(buffered_addr);
      address requested_addr = buffered_addr_to_requested_addr(buffered_addr);
      intptr_t target_location = (intptr_t)requested_addr;
      uint64_t pd = (uint64_t)(pointer_delta(buffered_addr, _buffer_start, 1));
      uint32_t narrow_location = checked_cast<uint32_t>(_buffer_start_narrow_oop + (pd >> _requested_shift));
      Klass* klass = real_klass_of_buffered_oop(buffered_addr);

      return { buffered_addr,
               requested_addr,
               target_location,
               narrow_location,
               raw_oop,
               klass,
               size,
               false };
    }
  };

  MemRegion r = heap_info->buffer_region();
  address buffer_start = address(r.start());
  address buffer_end = address(r.end());

  address requested_base = UseCompressedOops ? AOTMappedHeapWriter::narrow_oop_base() : (address)AOTMappedHeapWriter::NOCOOPS_REQUESTED_BASE;
  address requested_start = UseCompressedOops ? AOTMappedHeapWriter::buffered_addr_to_requested_addr(buffer_start) : requested_base;
  int requested_shift = AOTMappedHeapWriter::narrow_oop_shift();

  return new MappedWriterOopIterator(buffer_start,
                                     buffer_end,
                                     requested_base,
                                     requested_start,
                                     requested_shift,
                                     heap_info->root_segments().count());
}

#endif // INCLUDE_CDS_JAVA_HEAP