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

src/hotspot/share/cds/archiveUtils.cpp
730 строк27 KB
David Simms
8389219: Implement JEP 401: Value Objects (Preview)
31 июл 2026, 03:45
31 июл 2026, 03:45cc278db
/*
 * Copyright (c) 2019, 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/aotCompressedPointers.hpp"
#include "cds/aotLogging.hpp"
#include "cds/aotMetaspace.hpp"
#include "cds/archiveBuilder.hpp"
#include "cds/archiveUtils.hpp"
#include "cds/cdsConfig.hpp"
#include "cds/classListParser.hpp"
#include "cds/classListWriter.hpp"
#include "cds/dumpAllocStats.hpp"
#include "cds/dynamicArchive.hpp"
#include "cds/filemap.hpp"
#include "cds/heapShared.hpp"
#include "cds/lambdaProxyClassDictionary.hpp"
#include "classfile/systemDictionaryShared.hpp"
#include "classfile/vmClasses.hpp"
#include "interpreter/bootstrapInfo.hpp"
#include "memory/metaspaceUtils.hpp"
#include "memory/resourceArea.hpp"
#include "oops/compressedOops.inline.hpp"
#include "oops/klass.inline.hpp"
#include "runtime/arguments.hpp"
#include "utilities/bitMap.inline.hpp"
#include "utilities/debug.hpp"
#include "utilities/formatBuffer.hpp"
#include "utilities/globalDefinitions.hpp"
#include "utilities/rbTree.inline.hpp"
#include "utilities/spinYield.hpp"

CHeapBitMap* ArchivePtrMarker::_ptrmap = nullptr;
CHeapBitMap* ArchivePtrMarker::_rw_ptrmap = nullptr;
CHeapBitMap* ArchivePtrMarker::_ro_ptrmap = nullptr;
VirtualSpace* ArchivePtrMarker::_vs;

bool ArchivePtrMarker::_compacted;

void ArchivePtrMarker::initialize(CHeapBitMap* ptrmap, VirtualSpace* vs) {
  assert(_ptrmap == nullptr, "initialize only once");
  assert(_rw_ptrmap == nullptr, "initialize only once");
  assert(_ro_ptrmap == nullptr, "initialize only once");
  _vs = vs;
  _compacted = false;
  _ptrmap = ptrmap;

  // Use this as initial guesstimate. We should need less space in the
  // archive, but if we're wrong the bitmap will be expanded automatically.
  size_t estimated_archive_size = MetaspaceGC::capacity_until_GC();
  // But set it smaller in debug builds so we always test the expansion code.
  // (Default archive is about 12MB).
  DEBUG_ONLY(estimated_archive_size = 6 * M);

  // We need one bit per pointer in the archive.
  _ptrmap->initialize(estimated_archive_size / sizeof(intptr_t));
}

void ArchivePtrMarker::initialize_rw_ro_maps(CHeapBitMap* rw_ptrmap, CHeapBitMap* ro_ptrmap) {
  address* buff_bottom = (address*)ArchiveBuilder::current()->buffer_bottom();
  address* rw_bottom   = (address*)ArchiveBuilder::current()->rw_region()->base();
  address* ro_bottom   = (address*)ArchiveBuilder::current()->ro_region()->base();

  // The bit in _ptrmap that cover the very first word in the rw/ro regions.
  size_t rw_start = rw_bottom - buff_bottom;
  size_t ro_start = ro_bottom - buff_bottom;

  // The number of bits used by the rw/ro ptrmaps. We might have lots of zero
  // bits at the bottom and top of rw/ro ptrmaps, but these zeros will be
  // removed by FileMapInfo::write_bitmap_region().
  size_t rw_size = ArchiveBuilder::current()->rw_region()->used() / sizeof(address);
  size_t ro_size = ArchiveBuilder::current()->ro_region()->used() / sizeof(address);

  // The last (exclusive) bit in _ptrmap that covers the rw/ro regions.
  // Note: _ptrmap is dynamically expanded only when an actual pointer is written, so
  // it may not be as large as we want.
  size_t rw_end = MIN2<size_t>(rw_start + rw_size, _ptrmap->size());
  size_t ro_end = MIN2<size_t>(ro_start + ro_size, _ptrmap->size());

  rw_ptrmap->initialize(rw_size);
  ro_ptrmap->initialize(ro_size);

  for (size_t rw_bit = rw_start; rw_bit < rw_end; rw_bit++) {
    rw_ptrmap->at_put(rw_bit - rw_start, _ptrmap->at(rw_bit));
  }

  for(size_t ro_bit = ro_start; ro_bit < ro_end; ro_bit++) {
    ro_ptrmap->at_put(ro_bit - ro_start, _ptrmap->at(ro_bit));
  }

  _rw_ptrmap = rw_ptrmap;
  _ro_ptrmap = ro_ptrmap;
}

void ArchivePtrMarker::mark_pointer(address* ptr_loc) {
  assert(_ptrmap != nullptr, "not initialized");
  assert(!_compacted, "cannot mark anymore");

  if (ptr_base() <= ptr_loc && ptr_loc < ptr_end()) {
    address value = *ptr_loc;
    if (value != nullptr) {
      // We don't want any pointer that points to very bottom of the AOT metaspace, otherwise
      // when AOTMetaspace::default_base_address()==0, we can't distinguish between a pointer
      // to nothing (null) vs a pointer to an objects that happens to be at the very bottom
      // of the AOT metaspace.
      //
      // This should never happen because the protection zone prevents any valid objects from
      // being allocated at the bottom of the AOT metaspace.
      assert(AOTMetaspace::protection_zone_size() > 0, "must be");
      assert(ArchiveBuilder::current()->any_to_offset(value) > 0, "cannot point to bottom of AOT metaspace");

      assert(uintx(ptr_loc) % sizeof(intptr_t) == 0, "pointers must be stored in aligned addresses");
      size_t idx = ptr_loc - ptr_base();
      if (_ptrmap->size() <= idx) {
        _ptrmap->resize((idx + 1) * 2);
      }
      assert(idx < _ptrmap->size(), "must be");
      _ptrmap->set_bit(idx);
    }
  }
}

void ArchivePtrMarker::clear_pointer(address* ptr_loc) {
  assert(_ptrmap != nullptr, "not initialized");
  assert(!_compacted, "cannot clear anymore");

  assert(ptr_base() <= ptr_loc && ptr_loc < ptr_end(), "must be");
  assert(uintx(ptr_loc) % sizeof(intptr_t) == 0, "pointers must be stored in aligned addresses");
  size_t idx = ptr_loc - ptr_base();
  assert(idx < _ptrmap->size(), "cannot clear pointers that have not been marked");
  _ptrmap->clear_bit(idx);
}

class ArchivePtrBitmapCleaner: public BitMapClosure {
  CHeapBitMap* _ptrmap;
  address* _ptr_base;
  address  _relocatable_base;
  address  _relocatable_end;
  size_t   _max_non_null_offset;

public:
  ArchivePtrBitmapCleaner(CHeapBitMap* ptrmap, address* ptr_base, address relocatable_base, address relocatable_end) :
    _ptrmap(ptrmap), _ptr_base(ptr_base),
    _relocatable_base(relocatable_base), _relocatable_end(relocatable_end), _max_non_null_offset(0) {}

  bool do_bit(size_t offset) {
    address* ptr_loc = _ptr_base + offset;
    address  ptr_value = *ptr_loc;
    if (ptr_value != nullptr) {
      assert(_relocatable_base <= ptr_value && ptr_value < _relocatable_end, "do not point to arbitrary locations!");
      if (_max_non_null_offset < offset) {
        _max_non_null_offset = offset;
      }
    } else {
      _ptrmap->clear_bit(offset);
      DEBUG_ONLY(log_trace(aot, reloc)("Clearing pointer [" PTR_FORMAT  "] -> null @ %9zu", p2i(ptr_loc), offset));
    }

    return true;
  }

  size_t max_non_null_offset() const { return _max_non_null_offset; }
};

void ArchivePtrMarker::compact(address relocatable_base, address relocatable_end) {
  assert(!_compacted, "cannot compact again");
  ArchivePtrBitmapCleaner cleaner(_ptrmap, ptr_base(), relocatable_base, relocatable_end);
  _ptrmap->iterate(&cleaner);
  compact(cleaner.max_non_null_offset());
}

void ArchivePtrMarker::compact(size_t max_non_null_offset) {
  assert(!_compacted, "cannot compact again");
  _ptrmap->resize(max_non_null_offset + 1);
  _compacted = true;
}

char* DumpRegion::expand_top_to(char* newtop) {
  assert(is_allocatable(), "must be initialized and not packed");
  assert(newtop >= _top, "must not grow backwards");
  if (newtop > _end) {
    ArchiveBuilder::current()->report_out_of_space(_name, newtop - _top);
    ShouldNotReachHere();
  }

  commit_to(newtop);
  _top = newtop;

  if (ArchiveBuilder::is_active() && ArchiveBuilder::current()->is_in_buffer_space(_base)) {
    uintx delta = ArchiveBuilder::current()->buffer_to_offset((address)(newtop-1));
    if (delta > AOTCompressedPointers::MaxMetadataOffsetBytes) {
      // This is just a sanity check and should not appear in any real world usage. This
      // happens only if you allocate more than 2GB of shared objects and would require
      // millions of shared classes.
      aot_log_error(aot)("Out of memory in the %s: Please reduce the number of shared classes.", CDSConfig::type_of_archive_being_written());
      AOTMetaspace::unrecoverable_writing_error();
    }
  }

  return _top;
}

void DumpRegion::commit_to(char* newtop) {
  assert(CDSConfig::is_dumping_archive(), "sanity");
  char* base = _rs->base();
  size_t need_committed_size = newtop - base;
  size_t has_committed_size = _vs->committed_size();
  if (need_committed_size < has_committed_size) {
    return;
  }

  size_t min_bytes = need_committed_size - has_committed_size;
  size_t preferred_bytes = 1 * M;
  size_t uncommitted = _vs->reserved_size() - has_committed_size;

  size_t commit = MAX2(min_bytes, preferred_bytes);
  commit = MIN2(commit, uncommitted);
  assert(commit <= uncommitted, "sanity");

  if (!_vs->expand_by(commit, false)) {
    aot_log_error(aot)("Failed to expand shared space to %zu bytes",
                    need_committed_size);
    AOTMetaspace::unrecoverable_writing_error();
  }

  const char* which;
  if (_rs->base() == (char*)AOTMetaspace::symbol_rs_base()) {
    which = "symbol";
  } else {
    which = "shared";
  }
  log_debug(aot)("Expanding %s spaces by %7zu bytes [total %9zu bytes ending at %p]",
                 which, commit, _vs->actual_committed_size(), _vs->high());
}

// Basic allocation. Any alignment gaps will be wasted.
char* DumpRegion::allocate(size_t num_bytes, size_t alignment) {
  // Always align to at least minimum alignment
  alignment = MAX2(SharedSpaceObjectAlignment, alignment);
  char* p = (char*)align_up(_top, alignment);
  char* newtop = p + align_up(num_bytes, SharedSpaceObjectAlignment);
  expand_top_to(newtop);
  memset(p, 0, newtop - p);
  return p;
}

class DumpRegion::AllocGap {
  size_t _gap_bytes;   // size of this gap in bytes
  char* _gap_bottom;   // must be SharedSpaceObjectAlignment aligned
public:
  size_t gap_bytes() const { return _gap_bytes; }
  char* gap_bottom() const { return _gap_bottom; }

  AllocGap(size_t bytes, char* bottom) : _gap_bytes(bytes), _gap_bottom(bottom) {
    precond(is_aligned(gap_bytes(), SharedSpaceObjectAlignment));
    precond(is_aligned(gap_bottom(), SharedSpaceObjectAlignment));
  }
};

struct DumpRegion::AllocGapCmp {
  static RBTreeOrdering cmp(AllocGap a, AllocGap b) {
    RBTreeOrdering order = rbtree_primitive_cmp(a.gap_bytes(), b.gap_bytes());
    if (order == RBTreeOrdering::EQ) {
      order = rbtree_primitive_cmp(a.gap_bottom(), b.gap_bottom());
    }
    return order;
  }
};

struct Empty {};
using AllocGapNode = RBNode<DumpRegion::AllocGap, Empty>;

class DumpRegion::AllocGapTree : public RBTreeCHeap<AllocGap, Empty, AllocGapCmp, mtClassShared> {
public:
  size_t add_gap(char* gap_bottom, char* gap_top) {
    precond(gap_bottom < gap_top);
    size_t gap_bytes = pointer_delta(gap_top, gap_bottom, 1);
    precond(gap_bytes > 0);

    _total_gap_bytes += gap_bytes;

    AllocGap gap(gap_bytes, gap_bottom); // constructor checks alignment
    AllocGapNode* node = allocate_node(gap, Empty{});
    insert(gap, node);

    log_trace(aot, alloc)("adding a gap of %zu bytes @ %p (total = %zu, used = %zu) in %zu blocks",
                          gap_bytes, gap_bottom, _total_gap_bytes, _total_gap_bytes_used, size());
    return gap_bytes;
  }

  char* allocate_from_gap(size_t num_bytes) {
    // The gaps are sorted in ascending order of their sizes. When two gaps have the same
    // size, the one with a lower gap_bottom comes first.
    //
    // Find the first gap that's big enough, with the lowest gap_bottom.
    AllocGap target(num_bytes, nullptr);
    AllocGapNode* node = closest_ge(target);
    if (node == nullptr) {
      return nullptr; // Didn't find any usable gap.
    }

    size_t gap_bytes = node->key().gap_bytes();
    char* gap_bottom = node->key().gap_bottom();
    char* result = gap_bottom;
    precond(is_aligned(result, SharedSpaceObjectAlignment));

    remove(node);

    _total_gap_bytes_used += num_bytes;
    _total_gap_allocs++;
    DEBUG_ONLY(node = nullptr); // Don't use it anymore!

    precond(gap_bytes >= num_bytes);
    if (gap_bytes > num_bytes) {
      AllocGap gap(gap_bytes - num_bytes, gap_bottom + num_bytes); // constructor checks alignment
      AllocGapNode* new_node = allocate_node(gap, Empty{});
      insert(gap, new_node);
    }
    size_t unfilled_bytes = _total_gap_bytes - _total_gap_bytes_used;
    log_trace(aot, alloc)("%zu bytes @ %p in a gap of %zu bytes (used gaps %zu times, remain gap = %zu bytes in %zu blocks)",
                          num_bytes, result, gap_bytes, _total_gap_allocs, unfilled_bytes, size());
    return result;
  }
};

size_t DumpRegion::_total_gap_bytes = 0; // All the gaps that have ever been created
size_t DumpRegion::_total_gap_bytes_used = 0; // All the gaps that have been used
size_t DumpRegion::_total_gap_allocs = 0;
DumpRegion::AllocGapTree DumpRegion::_gap_tree;

// Alignment gaps happen only for the RW space. Collect the gaps into the _gap_tree so they can be
// used for future small object allocation.
char* DumpRegion::allocate_metaspace_obj(size_t num_bytes, address src, MetaspaceClosureType type, bool read_only, DumpAllocStats* stats) {
  num_bytes = align_up(num_bytes, SharedSpaceObjectAlignment);
  size_t alignment = SharedSpaceObjectAlignment; // alignment for the dest pointer
  bool is_class = (type == MetaspaceClosureType::ClassType);
  bool is_instance_class = is_class && ((Klass*)src)->is_instance_klass();

#ifdef _LP64
  // More strict alignments needed for Klass objects
  if (is_class) {
    size_t klass_alignment = checked_cast<size_t>(nth_bit(ArchiveBuilder::precomputed_narrow_klass_shift()));
    alignment = MAX2(alignment, klass_alignment);
    precond(is_aligned(alignment, SharedSpaceObjectAlignment));
  }
#endif

  if (alignment == SharedSpaceObjectAlignment && type != MetaspaceClosureType::SymbolType) {
    // The addresses of Symbols must be in the same order as they are in ArchiveBuilder::SourceObjList.
    // If we put them in gaps, their order will change.
    //
    // We have enough small objects that all gaps are usually filled.
    char* p = _gap_tree.allocate_from_gap(num_bytes);
    if (p != nullptr) {
      // Already memset to 0 when adding the gap
      stats->record(type, checked_cast<int>(num_bytes), /*read_only=*/false); // all gaps are from RW space (for classes)
      return p;
    }
  }

  // Reserve space for a pointer directly in front of the buffered InstanceKlass, so
  // we can do a quick lookup from InstanceKlass* -> RunTimeClassInfo*
  // without building another hashtable. See RunTimeClassInfo::get_for()
  // in systemDictionaryShared.cpp.
  const size_t RuntimeClassInfoPtrSize = is_instance_class ? sizeof(address) : 0;

  if (is_class && !is_aligned(top() + RuntimeClassInfoPtrSize, alignment)) {
    // We need to add a gap to align the buffered Klass. Save the gap for future small allocations.
    assert(read_only == false, "only gaps in RW region are reusable");
    char* gap_bottom = top();
    char* gap_top = align_up(gap_bottom + RuntimeClassInfoPtrSize, alignment) - RuntimeClassInfoPtrSize;
    size_t gap_bytes = _gap_tree.add_gap(gap_bottom, gap_top);
    allocate(gap_bytes);
  }

  char* oldtop = top();
  if (is_instance_class) {
    SystemDictionaryShared::validate_before_archiving((InstanceKlass*)src);
    allocate(RuntimeClassInfoPtrSize);
  }

  precond(is_aligned(top(), alignment));
  char* result = allocate(num_bytes);
  log_trace(aot, alloc)("%zu bytes @ %p", num_bytes, result);
  stats->record(type, pointer_delta_as_int(top(), oldtop), read_only); // includes RuntimeClassInfoPtrSize for classes

  return result;
}

// Usually we have no gaps left.
void DumpRegion::report_gaps(DumpAllocStats* stats) {
  _gap_tree.visit_in_order([&](const AllocGapNode* node) {
        stats->record_gap(checked_cast<int>(node->key().gap_bytes()));
        return true;
      });

  double unfilled_percent = 0.0;
  size_t unfilled_bytes = _total_gap_bytes - _total_gap_bytes_used;
  if (_gap_tree.size() > 0) {
    unfilled_percent = percent_of(unfilled_bytes, _total_gap_bytes);
    if (unfilled_percent > 5.0) {
      // We have a limited number of small objects, so some small gaps may remain
      // unfilled. If more than 5% of the gaps are unfilled, this likely indicates
      // a systematic error that should be investigated. Otherwise, do not warn to
      // avoid noise.
      log_warning(aot)("Unexpected %zu gaps (%zu bytes, %.2f%%) for Klass alignment",
                       _gap_tree.size(), _total_gap_bytes, unfilled_percent);
    }
  }
  if (_total_gap_allocs > 0) {
    log_info(aot)("Allocated %zu objects of %zu bytes in gaps (remain = %zu bytes, %.2f%%)",
                  _total_gap_allocs, _total_gap_bytes_used, unfilled_bytes, unfilled_percent);
  }
}

void DumpRegion::append_intptr_t(intptr_t n, bool need_to_mark) {
  assert(is_aligned(_top, sizeof(intptr_t)), "bad alignment");
  intptr_t *p = (intptr_t*)_top;
  char* newtop = _top + sizeof(intptr_t);
  expand_top_to(newtop);
  *p = n;
  if (need_to_mark) {
    ArchivePtrMarker::mark_pointer(p);
  }
}

void DumpRegion::print(size_t total_bytes) const {
  char* base = used() > 0 ? ArchiveBuilder::current()->to_requested(_base) : nullptr;
  log_debug(aot)("%s space: %9zu [ %4.1f%% of total] out of %9zu bytes [%5.1f%% used] at " INTPTR_FORMAT,
                 _name, used(), percent_of(used(), total_bytes), reserved(), percent_of(used(), reserved()),
                 p2i(base));
}

void DumpRegion::print_out_of_space_msg(const char* failing_region, size_t needed_bytes) {
  aot_log_error(aot)("[%-8s] " PTR_FORMAT " - " PTR_FORMAT " capacity =%9d, allocated =%9d",
                 _name, p2i(_base), p2i(_top), int(_end - _base), int(_top - _base));
  if (strcmp(_name, failing_region) == 0) {
    aot_log_error(aot)(" required = %d", int(needed_bytes));
  }
}

void DumpRegion::init(ReservedSpace* rs, VirtualSpace* vs) {
  _rs = rs;
  _vs = vs;
  // Start with 0 committed bytes. The memory will be committed as needed.
  if (!_vs->initialize(*_rs, 0)) {
    fatal("Unable to allocate memory for shared space");
  }
  _base = _top = _rs->base();
  _end = _rs->end();
}

void DumpRegion::pack(DumpRegion* next) {
  if (!is_packed()) {
    _end = (char*)align_up(_top, AOTMetaspace::core_region_alignment());
    _is_packed = true;
  }
  _end = (char*)align_up(_top, AOTMetaspace::core_region_alignment());
  _is_packed = true;
  if (next != nullptr) {
    next->_rs = _rs;
    next->_vs = _vs;
    next->_base = next->_top = this->_end;
    next->_end = _rs->end();
  }
}

void WriteClosure::do_ptr(void** p) {
  address ptr = *(address*)p;
  AOTCompressedPointers::narrowPtr narrowp = AOTCompressedPointers::encode(ptr);
  _dump_region->append_intptr_t(checked_cast<intptr_t>(narrowp), false);
}

void ReadClosure::do_ptr(void** p) {
  assert(*p == nullptr, "initializing previous initialized pointer.");
  u4 narrowp = checked_cast<u4>(nextPtr());
  *p = AOTCompressedPointers::decode<void*>(cast_from_u4(narrowp), _base_address);
}

void ReadClosure::do_u4(u4* p) {
  intptr_t obj = nextPtr();
  *p = (u4)(uintx(obj));
}

void ReadClosure::do_int(int* p) {
  intptr_t obj = nextPtr();
  *p = (int)(intx(obj));
}

void ReadClosure::do_bool(bool* p) {
  intptr_t obj = nextPtr();
  *p = (bool)(uintx(obj));
}

void ReadClosure::do_tag(int tag) {
  int old_tag;
  old_tag = (int)(intptr_t)nextPtr();
  // do_int(&old_tag);
  assert(tag == old_tag, "tag doesn't match (%d, expected %d)", old_tag, tag);
  FileMapInfo::assert_mark(tag == old_tag);
}

void ArchiveUtils::log_to_classlist(BootstrapInfo* bootstrap_specifier, TRAPS) {
  if (ClassListWriter::is_enabled()) {
    if (LambdaProxyClassDictionary::is_supported_invokedynamic(bootstrap_specifier)) {
      const constantPoolHandle& pool = bootstrap_specifier->pool();
      if (SystemDictionaryShared::is_builtin_loader(pool->pool_holder()->class_loader_data())) {
        // Currently lambda proxy classes are supported only for the built-in loaders.
        ResourceMark rm(THREAD);
        int pool_index = bootstrap_specifier->bss_index();
        ClassListWriter w;
        w.stream()->print("%s %s", ClassListParser::lambda_proxy_tag(), pool->pool_holder()->name()->as_C_string());
        CDSIndyInfo cii;
        ClassListParser::populate_cds_indy_info(pool, pool_index, &cii, CHECK);
        GrowableArray<const char*>* indy_items = cii.items();
        for (int i = 0; i < indy_items->length(); i++) {
          w.stream()->print(" %s", indy_items->at(i));
        }
        w.stream()->cr();
      }
    }
  }
}

bool ArchiveUtils::has_aot_initialized_mirror(InstanceKlass* src_ik) {
  if (!ArchiveBuilder::current()->has_been_archived(src_ik)) {
    return false;
  }
  return ArchiveBuilder::current()->get_buffered_addr(src_ik)->has_aot_initialized_mirror();
}

size_t HeapRootSegments::size_in_bytes(size_t seg_idx) {
  assert(seg_idx < _count, "In range");
  return refArrayOopDesc::object_size(size_in_elems(seg_idx)) * HeapWordSize;
}

int HeapRootSegments::size_in_elems(size_t seg_idx) {
  assert(seg_idx < _count, "In range");
  if (seg_idx != _count - 1) {
    return _max_size_in_elems;
  } else {
    // Last slice, leftover
    return _roots_count % _max_size_in_elems;
  }
}

size_t HeapRootSegments::segment_offset(size_t seg_idx) {
  assert(seg_idx < _count, "In range");
  return _base_offset + seg_idx * _max_size_in_bytes;
}

ArchiveWorkers::ArchiveWorkers() :
        _end_semaphore(0),
        _num_workers(max_workers()),
        _started_workers(0),
        _finish_tokens(0),
        _state(UNUSED),
        _task(nullptr) {}

ArchiveWorkers::~ArchiveWorkers() {
  assert(AtomicAccess::load(&_state) != WORKING, "Should not be working");
}

int ArchiveWorkers::max_workers() {
  // The pool is used for short-lived bursty tasks. We do not want to spend
  // too much time creating and waking up threads unnecessarily. Plus, we do
  // not want to overwhelm large machines. This is why we want to be very
  // conservative about the number of workers actually needed.
  return MAX2(0, log2i_graceful(os::active_processor_count()));
}

bool ArchiveWorkers::is_parallel() {
  return _num_workers > 0;
}

void ArchiveWorkers::start_worker_if_needed() {
  while (true) {
    int cur = AtomicAccess::load(&_started_workers);
    if (cur >= _num_workers) {
      return;
    }
    if (AtomicAccess::cmpxchg(&_started_workers, cur, cur + 1, memory_order_relaxed) == cur) {
      new ArchiveWorkerThread(this);
      return;
    }
  }
}

void ArchiveWorkers::run_task(ArchiveWorkerTask* task) {
  assert(AtomicAccess::load(&_state) == UNUSED, "Should be unused yet");
  assert(AtomicAccess::load(&_task) == nullptr, "Should not have running tasks");
  AtomicAccess::store(&_state, WORKING);

  if (is_parallel()) {
    run_task_multi(task);
  } else {
    run_task_single(task);
  }

  assert(AtomicAccess::load(&_state) == WORKING, "Should be working");
  AtomicAccess::store(&_state, SHUTDOWN);
}

void ArchiveWorkers::run_task_single(ArchiveWorkerTask* task) {
  // Single thread needs no chunking.
  task->configure_max_chunks(1);

  // Execute the task ourselves, as there are no workers.
  task->work(0, 1);
}

void ArchiveWorkers::run_task_multi(ArchiveWorkerTask* task) {
  // Multiple threads can work with multiple chunks.
  task->configure_max_chunks(_num_workers * CHUNKS_PER_WORKER);

  // Set up the run and publish the task. Issue one additional finish token
  // to cover the semaphore shutdown path, see below.
  AtomicAccess::store(&_finish_tokens, _num_workers + 1);
  AtomicAccess::release_store(&_task, task);

  // Kick off pool startup by starting a single worker, and proceed
  // immediately to executing the task locally.
  start_worker_if_needed();

  // Execute the task ourselves, while workers are catching up.
  // This allows us to hide parts of task handoff latency.
  task->run();

  // Done executing task locally, wait for any remaining workers to complete.
  // Once all workers report, we can proceed to termination. To do this safely,
  // we need to make sure every worker has left. A spin-wait alone would suffice,
  // but we do not want to burn cycles on it. A semaphore alone would not be safe,
  // since workers can still be inside it as we proceed from wait here. So we block
  // on semaphore first, and then spin-wait for all workers to terminate.
  _end_semaphore.wait();
  SpinYield spin;
  while (AtomicAccess::load(&_finish_tokens) != 0) {
    spin.wait();
  }

  OrderAccess::fence();

  assert(AtomicAccess::load(&_finish_tokens) == 0, "All tokens are consumed");
}

void ArchiveWorkers::run_as_worker() {
  assert(is_parallel(), "Should be in parallel mode");

  ArchiveWorkerTask* task = AtomicAccess::load_acquire(&_task);
  task->run();

  // All work done in threads should be visible to caller.
  OrderAccess::fence();

  // Signal the pool the work is complete, and we are exiting.
  // Worker cannot do anything else with the pool after this.
  if (AtomicAccess::sub(&_finish_tokens, 1, memory_order_relaxed) == 1) {
    // Last worker leaving. Notify the pool it can unblock to spin-wait.
    // Then consume the last token and leave.
    _end_semaphore.signal();
    int last = AtomicAccess::sub(&_finish_tokens, 1, memory_order_relaxed);
    assert(last == 0, "Should be");
  }
}

void ArchiveWorkerTask::run() {
  while (true) {
    int chunk = AtomicAccess::load(&_chunk);
    if (chunk >= _max_chunks) {
      return;
    }
    if (AtomicAccess::cmpxchg(&_chunk, chunk, chunk + 1, memory_order_relaxed) == chunk) {
      assert(0 <= chunk && chunk < _max_chunks, "Sanity");
      work(chunk, _max_chunks);
    }
  }
}

void ArchiveWorkerTask::configure_max_chunks(int max_chunks) {
  if (_max_chunks == 0) {
    _max_chunks = max_chunks;
  }
}

ArchiveWorkerThread::ArchiveWorkerThread(ArchiveWorkers* pool) : NamedThread(), _pool(pool) {
  set_name("ArchiveWorkerThread");
  if (os::create_thread(this, os::os_thread)) {
    os::start_thread(this);
  } else {
    vm_exit_during_initialization("Unable to create archive worker",
                                  os::native_thread_creation_failed_msg());
  }
}

void ArchiveWorkerThread::run() {
  // Avalanche startup: each worker starts two others.
  _pool->start_worker_if_needed();
  _pool->start_worker_if_needed();

  // Set ourselves up.
  os::set_priority(this, NearMaxPriority);

  // Work.
  _pool->run_as_worker();
}

void ArchiveWorkerThread::post_run() {
  this->NamedThread::post_run();
  delete this;
}