17.4 Java堆空间内存分配

分配Java堆内存前，我们先通过两图来了解下C堆、Java堆、内核空间、native本地空间的关系。

1、从图17-1来看，Java堆的分配其实就是从Java进程运行时堆中选中一块内存区域来映射

2、从图17-2，可以看中各内存空间的关系，当然实际的内存区域比这个复杂的多，这里只是概括说明

图17-1

图17-2

17.4.1 genCollectedHeap.cpp

17.4.1.1 GenCollectedHeap::initialize

jint GenCollectedHeap::initialize() {
  // 这一步只是对c2编译器开通使用时，做一些参数赋值操作，这里就不展开讲
  CollectedHeap::pre_initialize();

  // 这里获取分代数_n_gens，就是2
  int i;
  _n_gens = gen_policy()->number_of_generations();

  // 保证2个值相等wordSize和HeapWordSize分别是在操作系统和Java堆中代表一个字word占用内存的大小，这两个值必然相同，否则出错
  guarantee(HeapWordSize == wordSize, "HeapWordSize must equal wordSize");

  // Java堆的对齐值，这个在`章节17.2.1.1`中有介绍
  size_t gen_alignment = Generation::GenGrain;
 // 获取分代对象数组，这个在`章节17.2.1.1`中有介绍，数组元素就2个，索引0元素表示年轻代，索引1元素表示老年代
  _gen_specs = gen_policy()->generations();

  // 分别遍历新生代和老年代，并设置各自分代的空间大小（初始值和最大值），同时确保内存对齐
  for (i = 0; i < _n_gens; i++) {
    _gen_specs[i]->align(gen_alignment);
  }

  // 下面才是给Java堆分配空间

  char* heap_address;
  size_t total_reserved = 0;
  int n_covered_regions = 0;
  ReservedSpace heap_rs;
  // 这是最外层Java堆的内存对齐值
  size_t heap_alignment = collector_policy()->heap_alignment();
  // 分配java堆内存，看`章节17.4.1.2`
  heap_address = allocate(heap_alignment, &total_reserved,
                          &n_covered_regions, &heap_rs);

  if (!heap_rs.is_reserved()) {
    vm_shutdown_during_initialization(
      "Could not reserve enough space for object heap");
    return JNI_ENOMEM;
  }
  // 将分配的Java堆内存，用 MemRegion 内存区域对象管理起来
  _reserved = MemRegion((HeapWord*)heap_rs.base(),
                        (HeapWord*)(heap_rs.base() + heap_rs.size()));

  // 参数赋值
  _reserved.set_word_size(0);
  _reserved.set_start((HeapWord*)heap_rs.base()); // Java堆内存的首地址
  size_t actual_heap_size = heap_rs.size(); // Java堆内存大小
    // Java堆内存的限制地址，也就是不能超过这条线
  _reserved.set_end((HeapWord*)(heap_rs.base() + actual_heap_size)); 
  // 接下来就是把Java堆按分代算法，分成4个区域（新生代、S1、S2、老年代）管理起来，这块逻辑，我们放`第十八章`讲
  _rem_set = collector_policy()->create_rem_set(_reserved, n_covered_regions);
  set_barrier_set(rem_set()->bs());

  _gch = this;

  for (i = 0; i < _n_gens; i++) {
    ReservedSpace this_rs = heap_rs.first_part(_gen_specs[i]->max_size(), false, false);
    _gens[i] = _gen_specs[i]->init(this_rs, i, rem_set());
    heap_rs = heap_rs.last_part(_gen_specs[i]->max_size());
  }
  clear_incremental_collection_failed();

#if INCLUDE_ALL_GCS
  // If we are running CMS, create the collector responsible
  // for collecting the CMS generations.
  if (collector_policy()->is_concurrent_mark_sweep_policy()) {
    bool success = create_cms_collector();
    if (!success) return JNI_ENOMEM;
  }
#endif // INCLUDE_ALL_GCS

  return JNI_OK;
}

17.4.1.2 GenCollectedHeap::allocate

char* GenCollectedHeap::allocate(size_t alignment,
                                 size_t* _total_reserved,
                                 int* _n_covered_regions,
                                 ReservedSpace* heap_rs){
  const char overflow_msg[] = "The size of the object heap + VM data exceeds "
    "the maximum representable size";

  // Now figure out the total size.
  size_t total_reserved = 0;
  int n_covered_regions = 0;
  const size_t pageSize = UseLargePages ?
      os::large_page_size() : os::vm_page_size();

  assert(alignment % pageSize == 0, "Must be");
  // 遍历_gen_specs，求得新生代和老年代的分配大小
  for (int i = 0; i < _n_gens; i++) {
    total_reserved += _gen_specs[i]->max_size();
    if (total_reserved < _gen_specs[i]->max_size()) {
      vm_exit_during_initialization(overflow_msg);
    }
    n_covered_regions += _gen_specs[i]->n_covered_regions();  // 最终为2
  }
  assert(total_reserved % alignment == 0,
         err_msg("Gen size; total_reserved=" SIZE_FORMAT ", alignment="
                 SIZE_FORMAT, total_reserved, alignment));

  // Needed until the cardtable is fixed to have the right number
  // of covered regions.
  n_covered_regions += 2;  // 再加2，就是4，也就是把堆最终分成4个区（新生代、S1、S2、老年代）

  *_total_reserved = total_reserved;
  *_n_covered_regions = n_covered_regions;
  // 分配内存，实现细节看`章节17.4.2`
  *heap_rs = Universe::reserve_heap(total_reserved, alignment);
  return heap_rs->base();
}

17.4.2 universe.cpp

17.4.2.1 Universe::reserve_heap

ReservedSpace Universe::reserve_heap(size_t heap_size, size_t alignment) {
  assert(alignment <= Arguments::conservative_max_heap_alignment(),
      err_msg("actual alignment " SIZE_FORMAT " must be within maximum heap alignment " SIZE_FORMAT,
          alignment, Arguments::conservative_max_heap_alignment()));
  // 通过内存对齐，得到要分配的空间大小
  size_t total_reserved = align_size_up(heap_size, alignment);
  assert(!UseCompressedOops || (total_reserved <= (OopEncodingHeapMax - os::vm_page_size())),
      "heap size is too big for compressed oops");
  // 大页时考虑，本系列文章中不考虑大而情况，忽略
  bool use_large_pages = UseLargePages && is_size_aligned(alignment, os::large_page_size());
  assert(!UseLargePages
      || UseParallelGC
      || use_large_pages, "Wrong alignment to use large pages");
  // 取出Java堆的基址base的值，32位机器时，就是0，实现细节看`章节17.4.2.2`
  char* addr = Universe::preferred_heap_base(total_reserved, alignment, Universe::UnscaledNarrowOop);
  // 创建一个ReservedHeapSpace对象，该对象就是用来保留连续内存地址范围空间的数据结构，实现细节看`章节17.4.3`
  ReservedHeapSpace total_rs(total_reserved, alignment, use_large_pages, addr);

  if (UseCompressedOops) {
    if (addr != NULL && !total_rs.is_reserved()) {
      // Failed to reserve at specified address - the requested memory
      // region is taken already, for example, by 'java' launcher.
      // Try again to reserver heap higher.
      addr = Universe::preferred_heap_base(total_reserved, alignment, Universe::ZeroBasedNarrowOop);

      ReservedHeapSpace total_rs0(total_reserved, alignment,
          use_large_pages, addr);

      if (addr != NULL && !total_rs0.is_reserved()) {
        // Failed to reserve at specified address again - give up.
        addr = Universe::preferred_heap_base(total_reserved, alignment, Universe::HeapBasedNarrowOop);
        assert(addr == NULL, "");

        ReservedHeapSpace total_rs1(total_reserved, alignment,
            use_large_pages, addr);
        total_rs = total_rs1;
      } else {
        total_rs = total_rs0;
      }
    }
  }

  if (!total_rs.is_reserved()) {
    vm_exit_during_initialization(err_msg("Could not reserve enough space for " SIZE_FORMAT "KB object heap", total_reserved/K));
    return total_rs;
  }

  if (UseCompressedOops) {
    // Universe::initialize_heap() will reset this to NULL if unscaled
    // or zero-based narrow oops are actually used.
    address base = (address)(total_rs.base() - os::vm_page_size());
    Universe::set_narrow_oop_base(base);
  }
  // 返回total_rs
  return total_rs;
}

17.4.2.2 Universe::preferred_heap_base

char* Universe::preferred_heap_base(size_t heap_size, size_t alignment, NARROW_OOP_MODE mode) {
  assert(is_size_aligned((size_t)OopEncodingHeapMax, alignment), "Must be");
  assert(is_size_aligned((size_t)UnscaledOopHeapMax, alignment), "Must be");
  assert(is_size_aligned(heap_size, alignment), "Must be");

  // HeapBaseMinAddress 是操作系统明确设定的堆内存的最低地址限制，默认设置的是2*G，这里按alignment对齐，把HeapBaseMinAddress的值按alignment对齐后，作为堆内存的最低地址
  uintx heap_base_min_address_aligned = align_size_up(HeapBaseMinAddress, alignment);

  size_t base = 0;
#ifdef _LP64  // 下面是对64位机器及使用压缩指针时的实现，我们只讲32位的，这块逻辑略过
  if (UseCompressedOops) {
    assert(mode == UnscaledNarrowOop  ||
           mode == ZeroBasedNarrowOop ||
           mode == HeapBasedNarrowOop, "mode is invalid");
    const size_t total_size = heap_size + heap_base_min_address_aligned;
    // Return specified base for the first request.
    if (!FLAG_IS_DEFAULT(HeapBaseMinAddress) && (mode == UnscaledNarrowOop)) {
      base = heap_base_min_address_aligned;

    // If the total size is small enough to allow UnscaledNarrowOop then
    // just use UnscaledNarrowOop.
    } else if ((total_size <= OopEncodingHeapMax) && (mode != HeapBasedNarrowOop)) {
      if ((total_size <= UnscaledOopHeapMax) && (mode == UnscaledNarrowOop) &&
          (Universe::narrow_oop_shift() == 0)) {
        // Use 32-bits oops without encoding and
        // place heap's top on the 4Gb boundary
        base = (UnscaledOopHeapMax - heap_size);
      } else {
        // Can't reserve with NarrowOopShift == 0
        Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes);

        if (mode == UnscaledNarrowOop ||
            mode == ZeroBasedNarrowOop && total_size <= UnscaledOopHeapMax) {

          // Use zero based compressed oops with encoding and
          // place heap's top on the 32Gb boundary in case
          // total_size > 4Gb or failed to reserve below 4Gb.
          uint64_t heap_top = OopEncodingHeapMax;

          // For small heaps, save some space for compressed class pointer
          // space so it can be decoded with no base.
          if (UseCompressedClassPointers && !UseSharedSpaces &&
              OopEncodingHeapMax <= 32*G) {

            uint64_t class_space = align_size_up(CompressedClassSpaceSize, alignment);
            assert(is_size_aligned((size_t)OopEncodingHeapMax-class_space,
                   alignment), "difference must be aligned too");
            uint64_t new_top = OopEncodingHeapMax-class_space;

            if (total_size <= new_top) {
              heap_top = new_top;
            }
          }

          // Align base to the adjusted top of the heap
          base = heap_top - heap_size;
        }
      }
    } else {
      // UnscaledNarrowOop encoding didn't work, and no base was found for ZeroBasedOops or
      // HeapBasedNarrowOop encoding was requested.  So, can't reserve below 32Gb.
      Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes);
    }

    // Set narrow_oop_base and narrow_oop_use_implicit_null_checks
    // used in ReservedHeapSpace() constructors.
    // The final values will be set in initialize_heap() below.
    if ((base != 0) && ((base + heap_size) <= OopEncodingHeapMax)) {
      // Use zero based compressed oops
      Universe::set_narrow_oop_base(NULL);
      // Don't need guard page for implicit checks in indexed
      // addressing mode with zero based Compressed Oops.
      Universe::set_narrow_oop_use_implicit_null_checks(true);
    } else {
      // Set to a non-NULL value so the ReservedSpace ctor computes
      // the correct no-access prefix.
      // The final value will be set in initialize_heap() below.
      Universe::set_narrow_oop_base((address)UnscaledOopHeapMax);
#if defined(_WIN64) || defined(AIX)
      if (UseLargePages) {
        // Cannot allocate guard pages for implicit checks in indexed
        // addressing mode when large pages are specified on windows.
        Universe::set_narrow_oop_use_implicit_null_checks(false);
      }
#endif //  _WIN64
    }
  }
#endif

  assert(is_ptr_aligned((char*)base, alignment), "Must be");
  // 最终返回base,在32位机器时，虚拟机就是返回0
  return (char*)base; // also return NULL (don't care) for 32-bit VM
}

17.4.3 virtualspace.cpp

17.4.3.1 ReservedHeapSpace::ReservedHeapSpace

ReservedHeapSpace::ReservedHeapSpace(size_t size, size_t alignment,
                                     bool large, char* requested_address) :
  /* 先调用父类构造函数
  */
  ReservedSpace(size, alignment, large,
                requested_address,
                (UseCompressedOops && (Universe::narrow_oop_base() != NULL) &&
                 Universe::narrow_oop_use_implicit_null_checks()) ?
                  lcm(os::vm_page_size(), alignment) : 0) {
  if (base() != NULL) {
    MemTracker::record_virtual_memory_type((address)base(), mtJavaHeap);
  }

  // Only reserved space for the java heap should have a noaccess_prefix
  // if using compressed oops.
  protect_noaccess_prefix(size);
}

17.4.3.2 ReservedSpace::ReservedSpace

ReservedSpace::ReservedSpace(size_t size, size_t alignment,
                             bool large,
                             char* requested_address,
                             const size_t noaccess_prefix) {
  initialize(size+noaccess_prefix, alignment, large, requested_address,
             noaccess_prefix, false);
}

17.4.3.3 ReservedSpace::initialize

入口函数： ReservedHeapSpace total_rs(total_reserved, alignment, use_large_pages, addr);

参数：

total_reserved 对应 size：空间大小

alignment 对应 alignment：内存对齐值

use_large_pages 对应 large：这里不考虑大页，就设置为false

addr 对应 requested_address：32位时，addr为0

noaccess_prefix 为 0

executable 为 false

void ReservedSpace::initialize(size_t size, size_t alignment, bool large,
                               char* requested_address,
                               const size_t noaccess_prefix,
                               bool executable) {
  // 看源码得知，这里就是取page size（页大小），没什么逻辑
  const size_t granularity = os::vm_allocation_granularity();
  // 断言检验
  assert((size & (granularity - 1)) == 0,
         "size not aligned to os::vm_allocation_granularity()");
  assert((alignment & (granularity - 1)) == 0,
         "alignment not aligned to os::vm_allocation_granularity()");
  assert(alignment == 0 || is_power_of_2((intptr_t)alignment),
         "not a power of 2");
  // 取二者最大值对齐
  alignment = MAX2(alignment, (size_t)os::vm_page_size());

  // Assert that if noaccess_prefix is used, it is the same as alignment.
  assert(noaccess_prefix == 0 ||
         noaccess_prefix == alignment, "noaccess prefix wrong");

  _base = NULL;
  _size = 0;
  _special = false;
  _executable = executable;
  _alignment = 0;
  _noaccess_prefix = 0;
  if (size == 0) {
    return;
  }

  // 不存在大页，special 为 false
  bool special = large && !os::can_commit_large_page_memory();
  char* base = NULL;
  // 32位机器时 requested_address == 0，这条线也不会走
  if (requested_address != 0) {
    requested_address -= noaccess_prefix; // adjust requested address
    assert(requested_address != NULL, "huge noaccess prefix?");
  }
  // special为false,这个if不会走
  if (special) {

    base = os::reserve_memory_special(size, alignment, requested_address, executable);

    if (base != NULL) {
      if (failed_to_reserve_as_requested(base, requested_address, size, true)) {
        // OS ignored requested address. Try different address.
        return;
      }
      // Check alignment constraints.
      assert((uintptr_t) base % alignment == 0,
             err_msg("Large pages returned a non-aligned address, base: "
                 PTR_FORMAT " alignment: " PTR_FORMAT,
                 base, (void*)(uintptr_t)alignment));
      _special = true;
    } else {
      // failed; try to reserve regular memory below
      if (UseLargePages && (!FLAG_IS_DEFAULT(UseLargePages) ||
                            !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {
        if (PrintCompressedOopsMode) {
          tty->cr();
          tty->print_cr("Reserve regular memory without large pages.");
        }
      }
    }
  }

  if (base == NULL) {
    if (requested_address != 0) {
      base = os::attempt_reserve_memory_at(size, requested_address);
      if (failed_to_reserve_as_requested(base, requested_address, size, false)) {
        // OS ignored requested address. Try different address.
        base = NULL;
      }
    } else {
      // 这一步就是通过系统调用mmap映射一块size大小的内存，Java堆内存就是mmap映射出来的
      base = os::reserve_memory(size, NULL, alignment);
    }
    // 映射失败，直接退出函数，分配Java堆内存失败
    if (base == NULL) return;

    // 验证对齐，为啥要验证呢，因为base是mmap映射后返回的内存首地址，这个地址是os自己的规则选取的一个地址，不一定能按照alignment对齐，所以这一定要验证
    if ((((size_t)base + noaccess_prefix) & (alignment - 1)) != 0) {
      // base没有对齐，只能释放刚才mmap映射的内存，然后重试
      if (!os::release_memory(base, size)) fatal("os::release_memory failed");
      // 确保对齐
      size = align_size_up(size, alignment);
      // 再次mmap映射内存，返回的base同样有上面一样的不对齐问题，所以这个函数中包含了手动对齐操作，细节看`章节17.4.3.4`
      base = os::reserve_memory_aligned(size, alignment);

      if (requested_address != 0 &&
          failed_to_reserve_as_requested(base, requested_address, size, false)) {
        // As a result of the alignment constraints, the allocated base differs
        // from the requested address. Return back to the caller who can
        // take remedial action (like try again without a requested address).
        assert(_base == NULL, "should be");
        return;
      }
    }
  }
  // Done
  _base = base;  // 最终拿到了Java堆的首地址
  _size = size;  // 最终拿到了Java堆的大小
  _alignment = alignment;  // 对齐值
  _noaccess_prefix = noaccess_prefix;  // 0

  // 断言判断
  assert(noaccess_prefix == 0 ||
         noaccess_prefix == _alignment, "noaccess prefix wrong");

  assert(markOopDesc::encode_pointer_as_mark(_base)->decode_pointer() == _base,
         "area must be distinguisable from marks for mark-sweep");
  assert(markOopDesc::encode_pointer_as_mark(&_base[size])->decode_pointer() == &_base[size],
         "area must be distinguisable from marks for mark-sweep");
}

17.4.3.4 os_posix.cpp->os::reserve_memory_aligned

char* os::reserve_memory_aligned(size_t size, size_t alignment) {
  assert((alignment & (os::vm_allocation_granularity() - 1)) == 0,
      "Alignment must be a multiple of allocation granularity (page size)");
  assert((size & (alignment -1)) == 0, "size must be 'alignment' aligned");

  size_t extra_size = size + alignment;
  assert(extra_size >= size, "overflow, size is too large to allow alignment");
  // mmap映射一块内存区域，返回首地址
  char* extra_base = os::reserve_memory(extra_size, NULL, alignment);

  if (extra_base == NULL) {
    return NULL;
  }

  // 手动对齐
  char* aligned_base = (char*) align_size_up((uintptr_t) extra_base, alignment);

  // [  |                                       |  ]
  // ^ extra_base
  //    ^ extra_base + begin_offset == aligned_base
  //     extra_base + begin_offset + size       ^
  //                       extra_base + extra_size ^
  // |<>| == begin_offset
  //                              end_offset == |<>|
  // 用对齐后的地址-mmap的首地址，得出与首地址的偏移值
  size_t begin_offset = aligned_base - extra_base;
  // 结束地址对齐后的偏移
  size_t end_offset = (extra_base + extra_size) - (aligned_base + size);
  // begin_offset > 0，表示确实有偏移，那就把extra_base到偏移的这部分释放掉，因为有新的首地址了
  if (begin_offset > 0) {
      os::release_memory(extra_base, begin_offset);
  }
  // end_offset > 0，表示确实有偏移，那就把end_offset偏移的这部分释放掉，因为有新的限制地址了
  if (end_offset > 0) {
      os::release_memory(extra_base + begin_offset + size, end_offset);
  }
  // 返回首地址
  return aligned_base;
}