ART虚拟机的对象分配过程

最新推荐文章于 2023-04-03 23:38:55 发布

Invoker123

最新推荐文章于 2023-04-03 23:38:55 发布

阅读量1.8k

点赞数

分类专栏： ART

本文链接：https://blog.csdn.net/invoker123/article/details/80218605

版权

ART 专栏收录该内容

8 篇文章 0 订阅

订阅专栏

在art内部对线程Thread的实现中，会对函数跳转表进行初始化，这些函数跳转表的初始化在不同的平台有不同的实现，通常是记录一些关键的汇编函数的入口。Thread类有个struct tls_ptr_sized_values结构体成员，struct tls_ptr_sized_values的interpreter_entrypoints，jni_entrypoints，portable_entrypoints和quick_entrypoints用来分类保存函数跳转入口。这些函数跳转入口通过访问线程Thread对应的偏移量进入。例如下面的一段本地代码，就是将Thread类+292的偏移地址保存到lr，再用blx跳转过去，实际上就是进入pAllocObject函数。

      0x00001240: f8d9e124      ldr.w   lr, [r9, #292]  ; pAllocObject
      0x00001244: 1c06          mov     r6, r0
      0x00001246: 1c29          mov     r1, r5
      0x00001248: 2007          movs    r0, #7
      0x0000124a: 47f0          blx     lr

/art/runtime/thread.cc

void Thread::InitTlsEntryPoints() {
  // Insert a placeholder so we can easily tell if we call an unimplemented entry point.
  uintptr_t* begin = reinterpret_cast<uintptr_t*>(&tlsPtr_.interpreter_entrypoints);
  uintptr_t* end = reinterpret_cast<uintptr_t*>(reinterpret_cast<uint8_t*>(begin) +
                                                sizeof(tlsPtr_.quick_entrypoints));
  for (uintptr_t* it = begin; it != end; ++it) {
    *it = reinterpret_cast<uintptr_t>(UnimplementedEntryPoint);
  }
  InitEntryPoints(&tlsPtr_.interpreter_entrypoints, &tlsPtr_.jni_entrypoints,
                  &tlsPtr_.portable_entrypoints, &tlsPtr_.quick_entrypoints);
}

我们只关注alloc部分，即ResetQuickAllocEntryPoints函数。

/art/runtime/arch/arm/entrypoints_init_arm.cc

void InitEntryPoints(InterpreterEntryPoints* ipoints, JniEntryPoints* jpoints,
                     PortableEntryPoints* ppoints, QuickEntryPoints* qpoints) {
  // Interpreter
  ipoints->pInterpreterToInterpreterBridge = artInterpreterToInterpreterBridge;
  ipoints->pInterpreterToCompiledCodeBridge = artInterpreterToCompiledCodeBridge;

  // JNI
  jpoints->pDlsymLookup = art_jni_dlsym_lookup_stub;

  // Portable
  ppoints->pPortableResolutionTrampoline = art_portable_resolution_trampoline;
  ppoints->pPortableToInterpreterBridge = art_portable_to_interpreter_bridge;

  // Alloc
  ResetQuickAllocEntryPoints(qpoints);

  // Cast
  qpoints->pInstanceofNonTrivial = artIsAssignableFromCode;
  qpoints->pCheckCast = art_quick_check_cast;

  // DexCache
  qpoints->pInitializeStaticStorage = art_quick_initialize_static_storage;
  qpoints->pInitializeTypeAndVerifyAccess = art_quick_initialize_type_and_verify_access;
  qpoints->pInitializeType = art_quick_initialize_type;
  qpoints->pResolveString = art_quick_resolve_string;

  // Field
  qpoints->pSet32Instance = art_quick_set32_instance;
  qpoints->pSet32Static = art_quick_set32_static;
  qpoints->pSet64Instance = art_quick_set64_instance;
  qpoints->pSet64Static = art_quick_set64_static;
  qpoints->pSetObjInstance = art_quick_set_obj_instance;
  qpoints->pSetObjStatic = art_quick_set_obj_static;
  qpoints->pGet32Instance = art_quick_get32_instance;
  qpoints->pGet64Instance = art_quick_get64_instance;
  qpoints->pGetObjInstance = art_quick_get_obj_instance;
  qpoints->pGet32Static = art_quick_get32_static;
  qpoints->pGet64Static = art_quick_get64_static;
  qpoints->pGetObjStatic = art_quick_get_obj_static;

  // Array
  qpoints->pAputObjectWithNullAndBoundCheck = art_quick_aput_obj_with_null_and_bound_check;
  qpoints->pAputObjectWithBoundCheck = art_quick_aput_obj_with_bound_check;
  qpoints->pAputObject = art_quick_aput_obj;
  qpoints->pHandleFillArrayData = art_quick_handle_fill_data;

  // JNI
  qpoints->pJniMethodStart = JniMethodStart;
  qpoints->pJniMethodStartSynchronized = JniMethodStartSynchronized;
  qpoints->pJniMethodEnd = JniMethodEnd;
  qpoints->pJniMethodEndSynchronized = JniMethodEndSynchronized;
  qpoints->pJniMethodEndWithReference = JniMethodEndWithReference;
  qpoints->pJniMethodEndWithReferenceSynchronized = JniMethodEndWithReferenceSynchronized;
  qpoints->pQuickGenericJniTrampoline = art_quick_generic_jni_trampoline;

  // Locks
  qpoints->pLockObject = art_quick_lock_object;
  qpoints->pUnlockObject = art_quick_unlock_object;

  // Math
  qpoints->pCmpgDouble = CmpgDouble;
  qpoints->pCmpgFloat = CmpgFloat;
  qpoints->pCmplDouble = CmplDouble;
  qpoints->pCmplFloat = CmplFloat;
  qpoints->pFmod = fmod;
  qpoints->pL2d = __aeabi_l2d;
  qpoints->pFmodf = fmodf;
  qpoints->pL2f = __aeabi_l2f;
  qpoints->pD2iz = __aeabi_d2iz;
  qpoints->pF2iz = __aeabi_f2iz;
  qpoints->pIdivmod = __aeabi_idivmod;
  qpoints->pD2l = art_d2l;
  qpoints->pF2l = art_f2l;
  qpoints->pLdiv = __aeabi_ldivmod;
  qpoints->pLmod = __aeabi_ldivmod;  // result returned in r2:r3
  qpoints->pLmul = art_quick_mul_long;
  qpoints->pShlLong = art_quick_shl_long;
  qpoints->pShrLong = art_quick_shr_long;
  qpoints->pUshrLong = art_quick_ushr_long;

  // Intrinsics
  qpoints->pIndexOf = art_quick_indexof;
  qpoints->pStringCompareTo = art_quick_string_compareto;
  qpoints->pMemcpy = memcpy;

  // Invocation
  qpoints->pQuickImtConflictTrampoline = art_quick_imt_conflict_trampoline;
  qpoints->pQuickResolutionTrampoline = art_quick_resolution_trampoline;
  qpoints->pQuickToInterpreterBridge = art_quick_to_interpreter_bridge;
  qpoints->pInvokeDirectTrampolineWithAccessCheck = art_quick_invoke_direct_trampoline_with_access_check;
  qpoints->pInvokeInterfaceTrampolineWithAccessCheck = art_quick_invoke_interface_trampoline_with_access_check;
  qpoints->pInvokeStaticTrampolineWithAccessCheck = art_quick_invoke_static_trampoline_with_access_check;
  qpoints->pInvokeSuperTrampolineWithAccessCheck = art_quick_invoke_super_trampoline_with_access_check;
  qpoints->pInvokeVirtualTrampolineWithAccessCheck = art_quick_invoke_virtual_trampoline_with_access_check;

  // Thread
  qpoints->pTestSuspend = art_quick_test_suspend;

  // Throws
  qpoints->pDeliverException = art_quick_deliver_exception;
  qpoints->pThrowArrayBounds = art_quick_throw_array_bounds;
  qpoints->pThrowDivZero = art_quick_throw_div_zero;
  qpoints->pThrowNoSuchMethod = art_quick_throw_no_such_method;
  qpoints->pThrowNullPointer = art_quick_throw_null_pointer_exception;
  qpoints->pThrowStackOverflow = art_quick_throw_stack_overflow;
};

entry_points_allocator是一个静态变量，被初始化为kAllocatorTypeDlMalloc,表示将会使用DlMalloc的分配器入口。可以在其他地方调用SetQuickAllocEntryPointsAllocator改变entry_points_allocator的值，从而改变线程的分配器入口。

/art/runtime/entrypoints/quick/quick_alloc_entrypoints.cc

void ResetQuickAllocEntryPoints(QuickEntryPoints* qpoints) {
  switch (entry_points_allocator) {
#if !defined(__APPLE__) || !defined(__LP64__)
    case gc::kAllocatorTypeDlMalloc: {
      SetQuickAllocEntryPoints_dlmalloc(qpoints, entry_points_instrumented);
      break;
    }
    case gc::kAllocatorTypeRosAlloc: {
      SetQuickAllocEntryPoints_rosalloc(qpoints, entry_points_instrumented);
      break;
    }
    case gc::kAllocatorTypeBumpPointer: {
      CHECK(kMovingCollector);
      SetQuickAllocEntryPoints_bump_pointer(qpoints, entry_points_instrumented);
      break;
    }
    case gc::kAllocatorTypeTLAB: {
      CHECK(kMovingCollector);
      SetQuickAllocEntryPoints_tlab(qpoints, entry_points_instrumented);
      break;
    }
#endif
    default: {
      LOG(FATAL) << "Unimplemented";
    }
  }
}

SetQuickAllocEntryPointsAllocator会在ChangeAllocator（修改分配器）时被调用，ChangeAllocator会在ChangeCollector（修改垃圾收集方式）时被调用。
根据ChangeCollector传入的CollectorType参数不同，设置不同的分配器。

/art/runtime/gc/heap.cc

void Heap::ChangeCollector(CollectorType collector_type) {
  // TODO: Only do this with all mutators suspended to avoid races.
  if (collector_type != collector_type_) {
    if (collector_type == kCollectorTypeMC) {
      // Don't allow mark compact unless support is compiled in.
      CHECK(kMarkCompactSupport);
    }
    collector_type_ = collector_type;
    gc_plan_.clear();
    switch (collector_type_) {
      case kCollectorTypeCC:  // Fall-through.
      case kCollectorTypeMC:  // Fall-through.
      case kCollectorTypeSS:  // Fall-through.
      case kCollectorTypeGSS: {
        gc_plan_.push_back(collector::kGcTypeFull);
        if (use_tlab_) {
          ChangeAllocator(kAllocatorTypeTLAB);
        } else {
          ChangeAllocator(kAllocatorTypeBumpPointer);
        }
        break;
      }
      case kCollectorTypeMS: {
        gc_plan_.push_back(collector::kGcTypeSticky);
        gc_plan_.push_back(collector::kGcTypePartial);
        gc_plan_.push_back(collector::kGcTypeFull);
        ChangeAllocator(kUseRosAlloc ? kAllocatorTypeRosAlloc : kAllocatorTypeDlMalloc);
        break;
      }
      case kCollectorTypeCMS: {
        gc_plan_.push_back(collector::kGcTypeSticky);
        gc_plan_.push_back(collector::kGcTypePartial);
        gc_plan_.push_back(collector::kGcTypeFull);
        ChangeAllocator(kUseRosAlloc ? kAllocatorTypeRosAlloc : kAllocatorTypeDlMalloc);
        break;
      }
      default: {
        LOG(FATAL) << "Unimplemented";
      }
    }
    if (IsGcConcurrent()) {
      concurrent_start_bytes_ =
          std::max(max_allowed_footprint_, kMinConcurrentRemainingBytes) - kMinConcurrentRemainingBytes;
    } else {
      concurrent_start_bytes_ = std::numeric_limits<size_t>::max();
    }
  }
}

ChangeCollector会在三处地方被调用。1.在Heap堆创建时，会调用ChangeCollector将当前的垃圾收集方式collector_type_修改为desired_collector_type_，desired_collector_type_与foreground_collector_type_相同，均为kCollectorTypeCMS，由代码可以看到，由于kUseRosAlloc为true，这时使用的分配器为kAllocatorTypeRosAlloc。2.发生在垃圾收集方式切换的时候。3.调用Heap::PreZygoteFork时，将当前的垃圾收集方式collector_type_修改为foreground_collector_type_，这时候使用的是kAllocatorTypeRosAlloc。
垃圾收集方式切换一般发生在应用程序的前后台状态发生变化时。前台状态，即是用户可以感知到的状态，用枚举变量kProcessStateJankPerceptible表示；后台状态，即是用户无法直接感知到的状态，用枚举变量kProcessStateJankImperceptible表示。

/art/runtime/gc/heap.h

// The process state passed in from the activity manager, used to determine when to do trimming
// and compaction.
enum ProcessState {
  kProcessStateJankPerceptible = 0,
  kProcessStateJankImperceptible = 1,
};

在ActivityThread里面有个更新应用进程状态的函数：updateProcessState。首先判断进程状态是否发生了变化，如果有变化，根据状态值是否小于ActivityManager.PROCESS_STATE_IMPORTANT_FOREGROUND来判断是否是前后台状态的变化。若发生了前后台状态变化，则需要进行垃圾收集方式的切换。

/frameworks/base/core/java/android/app/ActivityThread.java

        public void updateProcessState(int processState, boolean fromIpc) {
            synchronized (this) {
                if (mLastProcessState != processState) {
                    mLastProcessState = processState;
                    // Update Dalvik state based on ActivityManager.PROCESS_STATE_* constants.
                    final int DALVIK_PROCESS_STATE_JANK_PERCEPTIBLE = 0;
                    final int DALVIK_PROCESS_STATE_JANK_IMPERCEPTIBLE = 1;
                    int dalvikProcessState = DALVIK_PROCESS_STATE_JANK_IMPERCEPTIBLE;
                    // TODO: Tune this since things like gmail sync are important background but not jank perceptible.
                    if (processState <= ActivityManager.PROCESS_STATE_IMPORTANT_FOREGROUND) {
                        dalvikProcessState = DALVIK_PROCESS_STATE_JANK_PERCEPTIBLE;
                    }
                    VMRuntime.getRuntime().updateProcessState(dalvikProcessState);
                    if (false) {
                        Slog.i(TAG, "******************* PROCESS STATE CHANGED TO: " + processState
                                + (fromIpc ? " (from ipc": ""));
                    }
                }
            }
        }

当需要切换到前台状态时，使用RequestCollectorTransition切换垃圾收集方式为foreground_collector_type_；当需要切换到后台状态时，使用RequestCollectorTransition切换垃圾收集方式为
background_collector_type_。

/art/runtime/gc/heap.cc

void Heap::UpdateProcessState(ProcessState process_state) {
  if (process_state_ != process_state) {
    process_state_ = process_state;
    for (size_t i = 1; i <= kCollectorTransitionStressIterations; ++i) {
      // Start at index 1 to avoid "is always false" warning.
      // Have iteration 1 always transition the collector.
      TransitionCollector((((i & 1) == 1) == (process_state_ == kProcessStateJankPerceptible))
                          ? foreground_collector_type_ : background_collector_type_);
      usleep(kCollectorTransitionStressWait);
    }
    if (process_state_ == kProcessStateJankPerceptible) {
      // Transition back to foreground right away to prevent jank.
      RequestCollectorTransition(foreground_collector_type_, 0);
    } else {
      // Don't delay for debug builds since we may want to stress test the GC.
      // If background_collector_type_ is kCollectorTypeHomogeneousSpaceCompact then we have
      // special handling which does a homogenous space compaction once but then doesn't transition
      // the collector.
      RequestCollectorTransition(background_collector_type_,
                                 kIsDebugBuild ? 0 : kCollectorTransitionWait);
    }
  }
}

RequestCollectorTransition就是将desired_collector_type_设置为入参表示的垃圾收集方式，并且调用java层Daemons类的requestHeapTrim进行堆裁剪。

/art/runtime/gc/heap.cc

void Heap::RequestCollectorTransition(CollectorType desired_collector_type, uint64_t delta_time) {
  Thread* self = Thread::Current();
  {
    MutexLock mu(self, *heap_trim_request_lock_);
    if (desired_collector_type_ == desired_collector_type) {
      return;
    }
    heap_transition_or_trim_target_time_ =
        std::max(heap_transition_or_trim_target_time_, NanoTime() + delta_time);
    desired_collector_type_ = desired_collector_type;
  }
  SignalHeapTrimDaemon(self);
}

&emsp；由log可以看到，RequestCollectorTransition主要是在foreground_collector_type_和background_collector_type_之间进行切换，并改变desired_collector_type_的值。在《ART虚拟机堆的创建过程》中提到，Heap堆创建时，
foreground_collector_type_为kCollectorTypeCMS,
background_collector_type_为kCollectorTypeHomogeneousSpaceCompact,
desired_collector_type_跟foreground_collector_type_相同。

$ logcat -v time|grep RequestCollectorTransition                               
01-01 20:00:15.875 I/art     ( 2543): RequestCollectorTransition:CollectorTypeCMS-->CollectorTypeHomogeneousSpaceCompact
01-01 20:00:15.942 I/art     ( 1856): RequestCollectorTransition:CollectorTypeCMS-->CollectorTypeHomogeneousSpaceCompact
01-01 20:00:16.055 I/art     ( 1856): RequestCollectorTransition:CollectorTypeHomogeneousSpaceCompact-->CollectorTypeCMS
01-01 20:00:16.067 I/art     ( 1856): RequestCollectorTransition:CollectorTypeCMS-->CollectorTypeHomogeneousSpaceCompact
01-01 20:00:16.154 I/art     ( 2564): RequestCollectorTransition:CollectorTypeCMS-->CollectorTypeHomogeneousSpaceCompact
01-01 20:00:16.365 I/art     ( 2585): RequestCollectorTransition:CollectorTypeCMS-->CollectorTypeHomogeneousSpaceCompact
01-01 20:00:16.369 I/art     ( 1856): RequestCollectorTransition:CollectorTypeHomogeneousSpaceCompact-->CollectorTypeCMS
01-01 20:00:16.511 I/art     ( 1856): RequestCollectorTransition:CollectorTypeCMS-->CollectorTypeHomogeneousSpaceCompact
01-01 20:00:16.608 I/art     ( 2616): RequestCollectorTransition:CollectorTypeCMS-->CollectorTypeHomogeneousSpaceCompact
01-01 20:00:16.694 I/art     ( 2636): RequestCollectorTransition:CollectorTypeCMS-->CollectorTypeHomogeneousSpaceCompact
01-01 20:00:16.839 I/art     ( 2655): RequestCollectorTransition:CollectorTypeCMS-->CollectorTypeHomogeneousSpaceCompact
01-01 20:00:16.972 I/art     ( 2679): RequestCollectorTransition:CollectorTypeCMS-->CollectorTypeHomogeneousSpaceCompact
01-01 20:00:17.153 I/art     ( 2704): RequestCollectorTransition:CollectorTypeCMS-->CollectorTypeHomogeneousSpaceCompact
01-01 20:00:17.674 I/art     ( 2753): RequestCollectorTransition:CollectorTypeCMS-->CollectorTypeHomogeneousSpaceCompact
01-01 20:01:15.182 I/art     ( 2801): RequestCollectorTransition:CollectorTypeCMS-->CollectorTypeHomogeneousSpaceCompact

最终会调用到Heap::DoPendingTransitionOrTrim函数。
如果是切换到background_collector_type_，该线程会睡眠5秒，这个时候使用的仍然是foreground_collector_type_。对于切换到后台状态的进程，我们并不关心暂停时间，所以CareAboutPauseTimes返回false，所以5秒后将会调用PerformHomogeneousSpaceCompact进行一次同构空间压缩，然后返回，不需要进行额外的堆裁剪。
如果是切换到foreground_collector_type_，则调用TransitionCollector进行垃圾收集方式的切换，切换完后调用Heap::Trim()进行堆裁剪。

/art/runtime/gc/heap.cc

void Heap::DoPendingTransitionOrTrim() {
  Thread* self = Thread::Current();
  CollectorType desired_collector_type;
  // Wait until we reach the desired transition time.
  while (true) {
    uint64_t wait_time;
    {
      MutexLock mu(self, *heap_trim_request_lock_);
      desired_collector_type = desired_collector_type_;
      uint64_t current_time = NanoTime();
      if (current_time >= heap_transition_or_trim_target_time_) {
        break;
      }
      wait_time = heap_transition_or_trim_target_time_ - current_time;
    }
    ScopedThreadStateChange tsc(self, kSleeping);
    usleep(wait_time / 1000);  // Usleep takes microseconds.
  }
  // Launch homogeneous space compaction if it is desired.
  if (desired_collector_type == kCollectorTypeHomogeneousSpaceCompact) {
    if (!CareAboutPauseTimes()) {
      PerformHomogeneousSpaceCompact();
    }
    // No need to Trim(). Homogeneous space compaction may free more virtual and physical memory.
    desired_collector_type = collector_type_;
    return;
  }
  // Transition the collector if the desired collector type is not the same as the current
  // collector type.
  TransitionCollector(desired_collector_type);
  if (!CareAboutPauseTimes()) {
    // Deflate the monitors, this can cause a pause but shouldn't matter since we don't care
    // about pauses.
    Runtime* runtime = Runtime::Current();
    runtime->GetThreadList()->SuspendAll();
    uint64_t start_time = NanoTime();
    size_t count = runtime->GetMonitorList()->DeflateMonitors();
    VLOG(heap) << "Deflating " << count << " monitors took "
        << PrettyDuration(NanoTime() - start_time);
    runtime->GetThreadList()->ResumeAll();
  }
  // Do a heap trim if it is needed.
  Trim();
}

在for循环里面，首先是等待当前正在执行的gc完成，然后当原来的垃圾收集方式和切换过去的垃圾收集方式均为MovingGc或MovingGc未被禁用时，跳出for循环。
然后用ThreadList::SuspendAll阻塞当前线程外的其他线程。这里传入到TransitionCollector的参数为kCollectorTypeCMS，当前的垃圾收集方式collector_type_不是MovingGc，所以”case kCollectorTypeCMS:”部分的内容不会被执行，直接进入到ChangeCollector函数。最后恢复之前被阻塞的线程，做一些收尾工作。
在ChangeCollector函数中，将往gc_plan_添加三种gc类型：kGcTypeSticky，kGcTypePartial和kGcTypeFull，且将分配器设置为kAllocatorTypeRosAlloc。
现在可以知道，由于Heap堆创建，Zygote fork出新进程和应用进程从后台状态切换到前台状态时都会调用ChangeCollector，而且入参为kCollectorTypeCMS，所以大部分情况下entry_points_allocator这个值为kAllocatorTypeRosAlloc。

/art/runtime/gc/heap.cc

void Heap::TransitionCollector(CollectorType collector_type) {
  if (collector_type == collector_type_) {
    return;
  }
  VLOG(heap) << "TransitionCollector: " << static_cast<int>(collector_type_)
             << " -> " << static_cast<int>(collector_type);
  uint64_t start_time = NanoTime();
  uint32_t before_allocated = num_bytes_allocated_.LoadSequentiallyConsistent();
  Runtime* const runtime = Runtime::Current();
  ThreadList* const tl = runtime->GetThreadList();
  Thread* const self = Thread::Current();
  ScopedThreadStateChange tsc(self, kWaitingPerformingGc);
  Locks::mutator_lock_->AssertNotHeld(self);
  // Busy wait until we can GC (StartGC can fail if we have a non-zero
  // compacting_gc_disable_count_, this should rarely occurs).
  for (;;) {
    {
      ScopedThreadStateChange tsc(self, kWaitingForGcToComplete);
      MutexLock mu(self, *gc_complete_lock_);
      // Ensure there is only one GC at a time.
      WaitForGcToCompleteLocked(kGcCauseCollectorTransition, self);
      // Currently we only need a heap transition if we switch from a moving collector to a
      // non-moving one, or visa versa.
      const bool copying_transition = IsMovingGc(collector_type_) != IsMovingGc(collector_type);
      // If someone else beat us to it and changed the collector before we could, exit.
      // This is safe to do before the suspend all since we set the collector_type_running_ before
      // we exit the loop. If another thread attempts to do the heap transition before we exit,
      // then it would get blocked on WaitForGcToCompleteLocked.
      if (collector_type == collector_type_) {
        return;
      }
      // GC can be disabled if someone has a used GetPrimitiveArrayCritical but not yet released.
      if (!copying_transition || disable_moving_gc_count_ == 0) {
        // TODO: Not hard code in semi-space collector?
        collector_type_running_ = copying_transition ? kCollectorTypeSS : collector_type;
        break;
      }
    }
    usleep(1000);
  }
  if (runtime->IsShuttingDown(self)) {
    // Don't allow heap transitions to happen if the runtime is shutting down since these can
    // cause objects to get finalized.
    FinishGC(self, collector::kGcTypeNone);
    return;
  }
  tl->SuspendAll();
  switch (collector_type) {
    case kCollectorTypeSS: {
      if (!IsMovingGc(collector_type_)) {
        // Create the bump pointer space from the backup space.
        CHECK(main_space_backup_ != nullptr);
        std::unique_ptr<MemMap> mem_map(main_space_backup_->ReleaseMemMap());
        // We are transitioning from non moving GC -> moving GC, since we copied from the bump
        // pointer space last transition it will be protected.
        CHECK(mem_map != nullptr);
        mem_map->Protect(PROT_READ | PROT_WRITE);
        bump_pointer_space_ = space::BumpPointerSpace::CreateFromMemMap("Bump pointer space",
                                                                        mem_map.release());
        AddSpace(bump_pointer_space_);
        Compact(bump_pointer_space_, main_space_, kGcCauseCollectorTransition);
        // Use the now empty main space mem map for the bump pointer temp space.
        mem_map.reset(main_space_->ReleaseMemMap());
        // Unset the pointers just in case.
        if (dlmalloc_space_ == main_space_) {
          dlmalloc_space_ = nullptr;
        } else if (rosalloc_space_ == main_space_) {
          rosalloc_space_ = nullptr;
        }
        // Remove the main space so that we don't try to trim it, this doens't work for debug
        // builds since RosAlloc attempts to read the magic number from a protected page.
        RemoveSpace(main_space_);
        RemoveRememberedSet(main_space_);
        delete main_space_;  // Delete the space since it has been removed.
        main_space_ = nullptr;
        RemoveRememberedSet(main_space_backup_.get());
        main_space_backup_.reset(nullptr);  // Deletes the space.
        temp_space_ = space::BumpPointerSpace::CreateFromMemMap("Bump pointer space 2",
                                                                mem_map.release());
        AddSpace(temp_space_);
      }
      break;
    }
    case kCollectorTypeMS:
      // Fall through.
    case kCollectorTypeCMS: {
      if (IsMovingGc(collector_type_)) {
        CHECK(temp_space_ != nullptr);
        std::unique_ptr<MemMap> mem_map(temp_space_->ReleaseMemMap());
        RemoveSpace(temp_space_);
        temp_space_ = nullptr;
        mem_map->Protect(PROT_READ | PROT_WRITE);
        CreateMainMallocSpace(mem_map.get(), kDefaultInitialSize,
                              std::min(mem_map->Size(), growth_limit_), mem_map->Size());
        mem_map.release();
        // Compact to the main space from the bump pointer space, don't need to swap semispaces.
        AddSpace(main_space_);
        Compact(main_space_, bump_pointer_space_, kGcCauseCollectorTransition);
        mem_map.reset(bump_pointer_space_->ReleaseMemMap());
        RemoveSpace(bump_pointer_space_);
        bump_pointer_space_ = nullptr;
        const char* name = kUseRosAlloc ? kRosAllocSpaceName[1] : kDlMallocSpaceName[1];
        // Temporarily unprotect the backup mem map so rosalloc can write the debug magic number.
        if (kIsDebugBuild && kUseRosAlloc) {
          mem_map->Protect(PROT_READ | PROT_WRITE);
        }
        main_space_backup_.reset(CreateMallocSpaceFromMemMap(
            mem_map.get(), kDefaultInitialSize, std::min(mem_map->Size(), growth_limit_),
            mem_map->Size(), name, true));
        if (kIsDebugBuild && kUseRosAlloc) {
          mem_map->Protect(PROT_NONE);
        }
        mem_map.release();
      }
      break;
    }
    default: {
      LOG(FATAL) << "Attempted to transition to invalid collector type "
                 << static_cast<size_t>(collector_type);
      break;
    }
  }
  ChangeCollector(collector_type);
  tl->ResumeAll();
  // Can't call into java code with all threads suspended.
  reference_processor_.EnqueueClearedReferences(self);
  uint64_t duration = NanoTime() - start_time;
  GrowForUtilization(semi_space_collector_);
  FinishGC(self, collector::kGcTypeFull);
  int32_t after_allocated = num_bytes_allocated_.LoadSequentiallyConsistent();
  int32_t delta_allocated = before_allocated - after_allocated;
  std::string saved_str;
  if (delta_allocated >= 0) {
    saved_str = " saved at least " + PrettySize(delta_allocated);
  } else {
    saved_str = " expanded " + PrettySize(-delta_allocated);
  }
  VLOG(heap) << "Heap transition to " << process_state_ << " took "
      << PrettyDuration(duration) << saved_str;
}

回到alloc跳转函数的初始化中。
/art/runtime/entrypoints/quick/quick_alloc_entrypoints.cc

    case gc::kAllocatorTypeRosAlloc: {
      SetQuickAllocEntryPoints_rosalloc(qpoints, entry_points_instrumented);
      break;
    }

GENERATE_ENTRYPOINTS这个宏定义声明了一些汇编函数，并且将一些分配器入口设置为这些汇编函数。当我们需要为对象分配空间时，就会根据对象的类型选择合适的汇编入口。

/art/runtime/entrypoints/quick/quick_alloc_entrypoints.cc

#define GENERATE_ENTRYPOINTS(suffix) \
extern "C" void* art_quick_alloc_array##suffix(uint32_t, void*, int32_t); \
extern "C" void* art_quick_alloc_array_resolved##suffix(void* klass, void*, int32_t); \
extern "C" void* art_quick_alloc_array_with_access_check##suffix(uint32_t, void*, int32_t); \
extern "C" void* art_quick_alloc_object##suffix(uint32_t type_idx, void* method); \
extern "C" void* art_quick_alloc_object_resolved##suffix(void* klass, void* method); \
extern "C" void* art_quick_alloc_object_initialized##suffix(void* klass, void* method); \
extern "C" void* art_quick_alloc_object_with_access_check##suffix(uint32_t type_idx, void* method); \
extern "C" void* art_quick_check_and_alloc_array##suffix(uint32_t, void*, int32_t); \
extern "C" void* art_quick_check_and_alloc_array_with_access_check##suffix(uint32_t, void*, int32_t); \
extern "C" void* art_quick_alloc_array##suffix##_instrumented(uint32_t, void*, int32_t); \
extern "C" void* art_quick_alloc_array_resolved##suffix##_instrumented(void* klass, void*, int32_t); \
extern "C" void* art_quick_alloc_array_with_access_check##suffix##_instrumented(uint32_t, void*, int32_t); \
extern "C" void* art_quick_alloc_object##suffix##_instrumented(uint32_t type_idx, void* method); \
extern "C" void* art_quick_alloc_object_resolved##suffix##_instrumented(void* klass, void* method); \
extern "C" void* art_quick_alloc_object_initialized##suffix##_instrumented(void* klass, void* method); \
extern "C" void* art_quick_alloc_object_with_access_check##suffix##_instrumented(uint32_t type_idx, void* method); \
extern "C" void* art_quick_check_and_alloc_array##suffix##_instrumented(uint32_t, void*, int32_t); \
extern "C" void* art_quick_check_and_alloc_array_with_access_check##suffix##_instrumented(uint32_t, void*, int32_t); \
void SetQuickAllocEntryPoints##suffix(QuickEntryPoints* qpoints, bool instrumented) { \
  if (instrumented) { \
    qpoints->pAllocArray = art_quick_alloc_array##suffix##_instrumented; \
    qpoints->pAllocArrayResolved = art_quick_alloc_array_resolved##suffix##_instrumented; \
    qpoints->pAllocArrayWithAccessCheck = art_quick_alloc_array_with_access_check##suffix##_instrumented; \
    qpoints->pAllocObject = art_quick_alloc_object##suffix##_instrumented; \
    qpoints->pAllocObjectResolved = art_quick_alloc_object_resolved##suffix##_instrumented; \
    qpoints->pAllocObjectInitialized = art_quick_alloc_object_initialized##suffix##_instrumented; \
    qpoints->pAllocObjectWithAccessCheck = art_quick_alloc_object_with_access_check##suffix##_instrumented; \
    qpoints->pCheckAndAllocArray = art_quick_check_and_alloc_array##suffix##_instrumented; \
    qpoints->pCheckAndAllocArrayWithAccessCheck = art_quick_check_and_alloc_array_with_access_check##suffix##_instrumented; \
  } else { \
    qpoints->pAllocArray = art_quick_alloc_array##suffix; \
    qpoints->pAllocArrayResolved = art_quick_alloc_array_resolved##suffix; \
    qpoints->pAllocArrayWithAccessCheck = art_quick_alloc_array_with_access_check##suffix; \
    qpoints->pAllocObject = art_quick_alloc_object##suffix; \
    qpoints->pAllocObjectResolved = art_quick_alloc_object_resolved##suffix; \
    qpoints->pAllocObjectInitialized = art_quick_alloc_object_initialized##suffix; \
    qpoints->pAllocObjectWithAccessCheck = art_quick_alloc_object_with_access_check##suffix; \
    qpoints->pCheckAndAllocArray = art_quick_check_and_alloc_array##suffix; \
    qpoints->pCheckAndAllocArrayWithAccessCheck = art_quick_check_and_alloc_array_with_access_check##suffix; \
  } \
}

选取一个简单的函数：pAllocObject。看看它的汇编实现。可见art_quick_alloc_object_rosalloc实际上是使用bl指令跳转到C函数artAllocObjectFromCodeRosAlloc。参数type_idx描述的是要分配的对象的类型，通过寄存器r0传递，参数method描述的是当前调用的类方法，通过寄存器r1传递。

/art/runtime/arch/quick_alloc_entrypoints.S

.macro GENERATE_ALLOC_ENTRYPOINTS c_suffix, cxx_suffix
// Called by managed code to allocate an object.
TWO_ARG_DOWNCALL art_quick_alloc_object\c_suffix, artAllocObjectFromCode\cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO
// Called by managed code to allocate an object of a resolved class.

/art/runtime/arch/arm/quick_entrypoints_arm.S

// Macro to facilitate adding new allocation entrypoints.
.macro TWO_ARG_DOWNCALL name, entrypoint, return
    .extern \entrypoint
ENTRY \name
    SETUP_REF_ONLY_CALLEE_SAVE_FRAME  @ save callee saves in case of GC
    mov    r2, r9                     @ pass Thread::Current
    mov    r3, sp                     @ pass SP
    bl     \entrypoint     @ (uint32_t type_idx, Method* method, Thread*, SP)
    RESTORE_REF_ONLY_CALLEE_SAVE_FRAME
    \return
    DELIVER_PENDING_EXCEPTION
END \name
.endm

GENERATE_ENTRYPOINTS_FOR_ALLOCATOR 生成了artAllocObjectFromCodeRosAlloc函数的代码。

/art/runtime/entrypoints/quick/quick_alloc_entrypoints.cc

#define GENERATE_ENTRYPOINTS_FOR_ALLOCATOR_INST(suffix, suffix2, instrumented_bool, allocator_type) \
extern "C" mirror::Object* artAllocObjectFromCode ##suffix##suffix2( \
    uint32_t type_idx, mirror::ArtMethod* method, Thread* self, \
    StackReference<mirror::ArtMethod>* sp) \
    SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) { \
  FinishCalleeSaveFrameSetup(self, sp, Runtime::kRefsOnly); \
  return AllocObjectFromCode<false, instrumented_bool>(type_idx, method, self, allocator_type); \
} \

/art/runtime/entrypoints/quick/quick_alloc_entrypoints.cc

#define GENERATE_ENTRYPOINTS_FOR_ALLOCATOR(suffix, allocator_type) \
    GENERATE_ENTRYPOINTS_FOR_ALLOCATOR_INST(suffix, Instrumented, true, allocator_type) \
    GENERATE_ENTRYPOINTS_FOR_ALLOCATOR_INST(suffix, , false, allocator_type)

CheckObjectAlloc根据入参type_idx和method查找对应的DexCache是否已经解析过这个类了，若没有，则使用ResolveType进行解析,此外，还要确保这个类进行初始化。如果是使用ResolveType对类进行解析的，则slow_path为true;若在DexCache查找到解析记录的，则slow_path为false。CheckObjectAlloc返回要分配的对象对应的class。
对于slow_path为true的情况，使用的分配器是Heap堆使用的分配器；对于slow_path为false的情况,则使用入参对应的分配器。

/art/runtime/entrypoints/entrypoint_utils-inl.h

template <bool kAccessCheck, bool kInstrumented>
static inline mirror::Object* AllocObjectFromCode(uint32_t type_idx,
                                                  mirror::ArtMethod* method,
                                                  Thread* self,
                                                  gc::AllocatorType allocator_type) {
  bool slow_path = false;
  mirror::Class* klass = CheckObjectAlloc<kAccessCheck>(type_idx, method, self, &slow_path);
  if (UNLIKELY(slow_path)) {
    if (klass == nullptr) {
      return nullptr;
    }
    return klass->Alloc<kInstrumented>(self, Runtime::Current()->GetHeap()->GetCurrentAllocator());
  }
  DCHECK(klass != nullptr);
  return klass->Alloc<kInstrumented>(self, allocator_type);
}

add_finalizer表示这个类是否重写了finalize()函数，如果这个类重写了finalize()函数，则调用Heap::AddFinalizerReference调用Java代码，以分配出来的Object为参数生成一个FinalizerReference，所有的FinalizerReference将构成一个链表，当gc需要回收这个Object时，会先执行对应的FinalizerReference的finalize()函数，执行完后，再判断是否还需要回收。

/art/runtime/mirror/class-inl.h

template<bool kIsInstrumented, bool kCheckAddFinalizer>
inline Object* Class::Alloc(Thread* self, gc::AllocatorType allocator_type) {
  CheckObjectAlloc();
  gc::Heap* heap = Runtime::Current()->GetHeap();
  const bool add_finalizer = kCheckAddFinalizer && IsFinalizable();
  if (!kCheckAddFinalizer) {
    DCHECK(!IsFinalizable());
  }
  mirror::Object* obj =
      heap->AllocObjectWithAllocator<kIsInstrumented, false>(self, this, this->object_size_,
                                                             allocator_type, VoidFunctor());
  if (add_finalizer && LIKELY(obj != nullptr)) {
    heap->AddFinalizerReference(self, &obj);
  }
  return obj;
}

现在假定AllocObjectWithAllocator的分配器类型入参为kAllocatorTypeRosAlloc。
首先是调用TryToAllocate进行分配，如果不成功，就尝试先gc后再分配，即是调用AllocateInternalWithGc函数，如果调用AllocateInternalWithGc还是没有分配成功而且这个过程中堆的当前分配器类型从参数allocator表示的分配器类型变成了其他的分配器类型，则调用Heap类的AllocObject进行分配，AllocObject会重新调用到AllocObjectWithAllocator，这个AllocObjectWithAllocator和上一次的AllocObjectWithAllocator不同，因为第一次调用时模板参数kCheckLargeObject为false,这次调用时模板参数kCheckLargeObject为true，还有作为入参的分配器类型为Heap堆当前使用的分配器类型。也就是意味着这次的AllocObjectWithAllocator会走分配大对象的流程。ShouldAllocLargeObject检查要分配的空间是否大于12kb且是一个原始类型数组，如果是，则调用AllocLargeObject分配大对象空间，分配器类型为kAllocatorTypeLOS。若分配失败，则清掉可能抛出的OOM异常，继续往下走，重复AllocObjectWithAllocator的流程，直到分配成功或者抛出OOM异常。

/art/runtime/gc/heap-inl.h

template <bool kInstrumented, bool kCheckLargeObject, typename PreFenceVisitor>
inline mirror::Object* Heap::AllocObjectWithAllocator(Thread* self, mirror::Class* klass,
                                                      size_t byte_count, AllocatorType allocator,
                                                      const PreFenceVisitor& pre_fence_visitor) {
  if (kIsDebugBuild) {
    CheckPreconditionsForAllocObject(klass, byte_count);
    // Since allocation can cause a GC which will need to SuspendAll, make sure all allocations are
    // done in the runnable state where suspension is expected.
    CHECK_EQ(self->GetState(), kRunnable);
    self->AssertThreadSuspensionIsAllowable();
  }
  // Need to check that we arent the large object allocator since the large object allocation code
  // path this function. If we didn't check we would have an infinite loop.
  mirror::Object* obj;
  if (kCheckLargeObject && UNLIKELY(ShouldAllocLargeObject(klass, byte_count))) {
    obj = AllocLargeObject<kInstrumented, PreFenceVisitor>(self, &klass, byte_count,
                                                           pre_fence_visitor);
    if (obj != nullptr) {
      return obj;
    } else {
      // There should be an OOM exception, since we are retrying, clear it.
      self->ClearException();
    }
    // If the large object allocation failed, try to use the normal spaces (main space,
    // non moving space). This can happen if there is significant virtual address space
    // fragmentation.
  }
  AllocationTimer alloc_timer(this, &obj);
  size_t bytes_allocated;
  size_t usable_size;
  size_t new_num_bytes_allocated = 0;
  if (allocator == kAllocatorTypeTLAB) {
    byte_count = RoundUp(byte_count, space::BumpPointerSpace::kAlignment);
  }
  // If we have a thread local allocation we don't need to update bytes allocated.
  if (allocator == kAllocatorTypeTLAB && byte_count <= self->TlabSize()) {
    obj = self->AllocTlab(byte_count);
    DCHECK(obj != nullptr) << "AllocTlab can't fail";
    obj->SetClass(klass);
    if (kUseBakerOrBrooksReadBarrier) {
      if (kUseBrooksReadBarrier) {
        obj->SetReadBarrierPointer(obj);
      }
      obj->AssertReadBarrierPointer();
    }
    bytes_allocated = byte_count;
    usable_size = bytes_allocated;
    pre_fence_visitor(obj, usable_size);
    QuasiAtomic::ThreadFenceForConstructor();
  } else {
    obj = TryToAllocate<kInstrumented, false>(self, allocator, byte_count, &bytes_allocated,
                                              &usable_size);
    if (UNLIKELY(obj == nullptr)) {
      bool is_current_allocator = allocator == GetCurrentAllocator();
      obj = AllocateInternalWithGc(self, allocator, byte_count, &bytes_allocated, &usable_size,
                                   &klass);
      if (obj == nullptr) {
        bool after_is_current_allocator = allocator == GetCurrentAllocator();
        // If there is a pending exception, fail the allocation right away since the next one
        // could cause OOM and abort the runtime.
        if (!self->IsExceptionPending() && is_current_allocator && !after_is_current_allocator) {
          // If the allocator changed, we need to restart the allocation.
          return AllocObject<kInstrumented>(self, klass, byte_count, pre_fence_visitor);
        }
        return nullptr;
      }
    }
    DCHECK_GT(bytes_allocated, 0u);
    DCHECK_GT(usable_size, 0u);
    obj->SetClass(klass);
    if (kUseBakerOrBrooksReadBarrier) {
      if (kUseBrooksReadBarrier) {
        obj->SetReadBarrierPointer(obj);
      }
      obj->AssertReadBarrierPointer();
    }
    if (collector::SemiSpace::kUseRememberedSet && UNLIKELY(allocator == kAllocatorTypeNonMoving)) {
      // (Note this if statement will be constant folded away for the
      // fast-path quick entry points.) Because SetClass() has no write
      // barrier, if a non-moving space allocation, we need a write
      // barrier as the class pointer may point to the bump pointer
      // space (where the class pointer is an "old-to-young" reference,
      // though rare) under the GSS collector with the remembered set
      // enabled. We don't need this for kAllocatorTypeRosAlloc/DlMalloc
      // cases because we don't directly allocate into the main alloc
      // space (besides promotions) under the SS/GSS collector.
      WriteBarrierField(obj, mirror::Object::ClassOffset(), klass);
    }
    pre_fence_visitor(obj, usable_size);
    new_num_bytes_allocated =
        static_cast<size_t>(num_bytes_allocated_.FetchAndAddSequentiallyConsistent(bytes_allocated))
        + bytes_allocated;
  }
  if (kIsDebugBuild && Runtime::Current()->IsStarted()) {
    CHECK_LE(obj->SizeOf(), usable_size);
  }
  // TODO: Deprecate.
  if (kInstrumented) {
    if (Runtime::Current()->HasStatsEnabled()) {
      RuntimeStats* thread_stats = self->GetStats();
      ++thread_stats->allocated_objects;
      thread_stats->allocated_bytes += bytes_allocated;
      RuntimeStats* global_stats = Runtime::Current()->GetStats();
      ++global_stats->allocated_objects;
      global_stats->allocated_bytes += bytes_allocated;
    }
  } else {
    DCHECK(!Runtime::Current()->HasStatsEnabled());
  }
  if (AllocatorHasAllocationStack(allocator)) {
    PushOnAllocationStack(self, &obj);
  }
  if (kInstrumented) {
    if (Dbg::IsAllocTrackingEnabled()) {
      Dbg::RecordAllocation(klass, bytes_allocated);
    }
  } else {
    DCHECK(!Dbg::IsAllocTrackingEnabled());
  }
  // IsConcurrentGc() isn't known at compile time so we can optimize by not checking it for
  // the BumpPointer or TLAB allocators. This is nice since it allows the entire if statement to be
  // optimized out. And for the other allocators, AllocatorMayHaveConcurrentGC is a constant since
  // the allocator_type should be constant propagated.
  if (AllocatorMayHaveConcurrentGC(allocator) && IsGcConcurrent()) {
    CheckConcurrentGC(self, new_num_bytes_allocated, &obj);
  }
  VerifyObject(obj);
  self->VerifyStack();
  return obj;
}

// The size of a thread-local allocation stack in the number of references.
static constexpr size_t kThreadLocalAllocationStackSize = 128;

inline void Heap::PushOnAllocationStack(Thread* self, mirror::Object** obj) {
  if (kUseThreadLocalAllocationStack) {
    if (UNLIKELY(!self->PushOnThreadLocalAllocationStack(*obj))) {
      PushOnThreadLocalAllocationStackWithInternalGC(self, obj);
    }
  } else if (UNLIKELY(!allocation_stack_->AtomicPushBack(*obj))) {
    PushOnAllocationStackWithInternalGC(self, obj);
  }
}

template <bool kInstrumented, typename PreFenceVisitor>
inline mirror::Object* Heap::AllocLargeObject(Thread* self, mirror::Class** klass,
                                              size_t byte_count,
                                              const PreFenceVisitor& pre_fence_visitor) {
  // Save and restore the class in case it moves.
  StackHandleScope<1> hs(self);
  auto klass_wrapper = hs.NewHandleWrapper(klass);
  return AllocObjectWithAllocator<kInstrumented, false, PreFenceVisitor>(self, *klass, byte_count,
                                                                         kAllocatorTypeLOS,
                                                                         pre_fence_visitor);
}

可以看到Heap::TryToAllocate是在rosalloc_space_上调用AllocNonvirtual进行分配的。具体的分配原理以后再讲。

/art/runtime/gc/heap-inl.h

template <const bool kInstrumented, const bool kGrow>
inline mirror::Object* Heap::TryToAllocate(Thread* self, AllocatorType allocator_type,
                                           size_t alloc_size, size_t* bytes_allocated,
                                           size_t* usable_size) {
  if (allocator_type != kAllocatorTypeTLAB &&
      UNLIKELY(IsOutOfMemoryOnAllocation<kGrow>(allocator_type, alloc_size))) {
    return nullptr;
  }
  mirror::Object* ret;
  switch (allocator_type) {
  ...
      case kAllocatorTypeRosAlloc: {
      if (kInstrumented && UNLIKELY(running_on_valgrind_)) {
        // If running on valgrind, we should be using the instrumented path.
        ret = rosalloc_space_->Alloc(self, alloc_size, bytes_allocated, usable_size);
      } else {
        DCHECK(!running_on_valgrind_);
        ret = rosalloc_space_->AllocNonvirtual(self, alloc_size, bytes_allocated, usable_size);
      }
      break;
    }

AllocateInternalWithGc首先是等待当前正在执行的gc完成。当等待的gc完成以后，调用TryToAllocate进行一次分配，因为刚刚结束的gc已经释放了一部分空间。若分配失败，则调用CollectGarbageInternal进行一次GcType为next_gc_type_，GcCause为kGcCauseForAlloc的gc。gc完毕后，调用TryToAllocate进行一次分配。如果还是分配失败，对于gc_plan_里面的每一种不同的GcType，力度从小到大进行一次GcCause为kGcCauseForAlloc的gc，然后再进行分配操作。因为力度越大的gc对程序造成的卡顿影响越大，所以要循序渐进。
在使用了所有的GcType进行gc以后，如果还是分配失败，再进行一次TryToAllocate的操作，这个TryToAllocate的模板参数kGrow为true，只会在分配方式为kAllocatorTypeTLAB时起作用。其作用是当分配的总内存空间超过了max_allowed_footprint_设定的阈值时，TryToAllocate会返回null，如果kGrow为true，就会调整max_allowed_footprint_为已分配内存和这次分配所需空间的总和。
如果这样还是分配失败，就进行一次清除力度最大（GcType为kGcTypeFull），允许堆增长并且会清除软引用的gc，然后再进行分配。如果还是分配不到，针对kAllocatorTypeRosAlloc和kAllocatorTypeDlMalloc的分配方式执行一次同构空间压缩PerformHomogeneousSpaceCompact，然后在进行一次分配。针对kAllocatorTypeNonMoving的分配方式，可能的原因是堆空间尚未用满但是non_moving_space_已经满了，这时将non_moving_space_指定为main_space_，将main_space_当作non_moving_space_来使用，然后调用TryToAllocate进行分配。
如果还是分配不到，只能抛出OOM异常了。

/art/runtime/gc/heap.cc

mirror::Object* Heap::AllocateInternalWithGc(Thread* self, AllocatorType allocator,
                                             size_t alloc_size, size_t* bytes_allocated,
                                             size_t* usable_size,
                                             mirror::Class** klass) {
  bool was_default_allocator = allocator == GetCurrentAllocator();
  // Make sure there is no pending exception since we may need to throw an OOME.
  self->AssertNoPendingException();
  DCHECK(klass != nullptr);
  StackHandleScope<1> hs(self);
  HandleWrapper<mirror::Class> h(hs.NewHandleWrapper(klass));
  klass = nullptr;  // Invalidate for safety.
  // The allocation failed. If the GC is running, block until it completes, and then retry the
  // allocation.
  collector::GcType last_gc = WaitForGcToComplete(kGcCauseForAlloc, self);
  if (last_gc != collector::kGcTypeNone) {
    // If we were the default allocator but the allocator changed while we were suspended,
    // abort the allocation.
    if (was_default_allocator && allocator != GetCurrentAllocator()) {
      return nullptr;
    }
    // A GC was in progress and we blocked, retry allocation now that memory has been freed.
    mirror::Object* ptr = TryToAllocate<true, false>(self, allocator, alloc_size, bytes_allocated,
                                                     usable_size);
    if (ptr != nullptr) {
      return ptr;
    }
  }

  collector::GcType tried_type = next_gc_type_;
  const bool gc_ran =
      CollectGarbageInternal(tried_type, kGcCauseForAlloc, false) != collector::kGcTypeNone;
  if (was_default_allocator && allocator != GetCurrentAllocator()) {
    return nullptr;
  }
  if (gc_ran) {
    mirror::Object* ptr = TryToAllocate<true, false>(self, allocator, alloc_size, bytes_allocated,
                                                     usable_size);
    if (ptr != nullptr) {
      return ptr;
    }
  }

  // Loop through our different Gc types and try to Gc until we get enough free memory.
  for (collector::GcType gc_type : gc_plan_) {
    if (gc_type == tried_type) {
      continue;
    }
    // Attempt to run the collector, if we succeed, re-try the allocation.
    const bool gc_ran =
        CollectGarbageInternal(gc_type, kGcCauseForAlloc, false) != collector::kGcTypeNone;
    if (was_default_allocator && allocator != GetCurrentAllocator()) {
      return nullptr;
    }
    if (gc_ran) {
      // Did we free sufficient memory for the allocation to succeed?
      mirror::Object* ptr = TryToAllocate<true, false>(self, allocator, alloc_size, bytes_allocated,
                                                       usable_size);
      if (ptr != nullptr) {
        return ptr;
      }
    }
  }
  // Allocations have failed after GCs;  this is an exceptional state.
  // Try harder, growing the heap if necessary.
  mirror::Object* ptr = TryToAllocate<true, true>(self, allocator, alloc_size, bytes_allocated,
                                                  usable_size);
  if (ptr != nullptr) {
    return ptr;
  }
  // Most allocations should have succeeded by now, so the heap is really full, really fragmented,
  // or the requested size is really big. Do another GC, collecting SoftReferences this time. The
  // VM spec requires that all SoftReferences have been collected and cleared before throwing
  // OOME.
  VLOG(gc) << "Forcing collection of SoftReferences for " << PrettySize(alloc_size)
           << " allocation";
  // TODO: Run finalization, but this may cause more allocations to occur.
  // We don't need a WaitForGcToComplete here either.
  DCHECK(!gc_plan_.empty());
  CollectGarbageInternal(gc_plan_.back(), kGcCauseForAlloc, true);
  if (was_default_allocator && allocator != GetCurrentAllocator()) {
    return nullptr;
  }
  ptr = TryToAllocate<true, true>(self, allocator, alloc_size, bytes_allocated, usable_size);
  if (ptr == nullptr) {
    const uint64_t current_time = NanoTime();
    switch (allocator) {
      case kAllocatorTypeRosAlloc:
        // Fall-through.
      case kAllocatorTypeDlMalloc: {
        if (use_homogeneous_space_compaction_for_oom_ &&
            current_time - last_time_homogeneous_space_compaction_by_oom_ >
            min_interval_homogeneous_space_compaction_by_oom_) {
          last_time_homogeneous_space_compaction_by_oom_ = current_time;
          HomogeneousSpaceCompactResult result = PerformHomogeneousSpaceCompact();
          switch (result) {
            case HomogeneousSpaceCompactResult::kSuccess:
              // If the allocation succeeded, we delayed an oom.
              ptr = TryToAllocate<true, true>(self, allocator, alloc_size, bytes_allocated,
                                              usable_size);
              if (ptr != nullptr) {
                count_delayed_oom_++;
              }
              break;
            case HomogeneousSpaceCompactResult::kErrorReject:
              // Reject due to disabled moving GC.
              break;
            case HomogeneousSpaceCompactResult::kErrorVMShuttingDown:
              // Throw OOM by default.
              break;
            default: {
              LOG(FATAL) << "Unimplemented homogeneous space compaction result "
                         << static_cast<size_t>(result);
            }
          }
          // Always print that we ran homogeneous space compation since this can cause jank.
          VLOG(heap) << "Ran heap homogeneous space compaction, "
                    << " requested defragmentation "
                    << count_requested_homogeneous_space_compaction_.LoadSequentiallyConsistent()
                    << " performed defragmentation "
                    << count_performed_homogeneous_space_compaction_.LoadSequentiallyConsistent()
                    << " ignored homogeneous space compaction "
                    << count_ignored_homogeneous_space_compaction_.LoadSequentiallyConsistent()
                    << " delayed count = "
                    << count_delayed_oom_.LoadSequentiallyConsistent();
        }
        break;
      }
      case kAllocatorTypeNonMoving: {
        // Try to transition the heap if the allocation failure was due to the space being full.
        if (!IsOutOfMemoryOnAllocation<false>(allocator, alloc_size)) {
          // If we aren't out of memory then the OOM was probably from the non moving space being
          // full. Attempt to disable compaction and turn the main space into a non moving space.
          DisableMovingGc();
          // If we are still a moving GC then something must have caused the transition to fail.
          if (IsMovingGc(collector_type_)) {
            MutexLock mu(self, *gc_complete_lock_);
            // If we couldn't disable moving GC, just throw OOME and return null.
            LOG(WARNING) << "Couldn't disable moving GC with disable GC count "
                         << disable_moving_gc_count_;
          } else {
            LOG(WARNING) << "Disabled moving GC due to the non moving space being full";
            ptr = TryToAllocate<true, true>(self, allocator, alloc_size, bytes_allocated,
                                            usable_size);
          }
        }
        break;
      }
      default: {
        // Do nothing for others allocators.
      }
    }
  }
  // If the allocation hasn't succeeded by this point, throw an OOM error.
  if (ptr == nullptr) {
    ThrowOutOfMemoryError(self, alloc_size, allocator);
  }
  return ptr;
}

结合《ART类和方法》所提到的内容，当在Java代码中new一个新对象时，首先会用findClass查找到这个类，在findClass过程中先会调用AllocClass为这个类分配空间，再在LoadClassMembers中为类的field和method在art内部对应的ArtField和ArtMethod分配空间。
findClass完成后，就会使用内存分配函数（例如pAllocObject入口等）进行对实例对象内存的分配，实例对象内存大小由Class类的object_size_成员决定。分配完毕后，才会执行类的构造函数。