runtime/jit/jit_code_cache.cc

/*
 * Copyright 2014 The Android Open Source Project
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include "jit_code_cache.h"

#include <sstream>

#include "art_method-inl.h"
#include "base/enums.h"
#include "base/stl_util.h"
#include "base/systrace.h"
#include "base/time_utils.h"
#include "debugger_interface.h"
#include "entrypoints/runtime_asm_entrypoints.h"
#include "gc/accounting/bitmap-inl.h"
#include "gc/scoped_gc_critical_section.h"
#include "jit/jit.h"
#include "jit/profiling_info.h"
#include "linear_alloc.h"
#include "mem_map.h"
#include "oat_file-inl.h"
#include "scoped_thread_state_change-inl.h"
#include "thread_list.h"

namespace art {
namespace jit {

static constexpr int kProtAll = PROT_READ | PROT_WRITE | PROT_EXEC;
static constexpr int kProtData = PROT_READ | PROT_WRITE;
static constexpr int kProtCode = PROT_READ | PROT_EXEC;

static constexpr size_t kCodeSizeLogThreshold = 50 * KB;
static constexpr size_t kStackMapSizeLogThreshold = 50 * KB;

#define CHECKED_MPROTECT(memory, size, prot)                \
  do {                                                      \
    int rc = mprotect(memory, size, prot);                  \
    if (UNLIKELY(rc != 0)) {                                \
      errno = rc;                                           \
      PLOG(FATAL) << "Failed to mprotect jit code cache";   \
    }                                                       \
  } while (false)                                           \

JitCodeCache* JitCodeCache::Create(size_t initial_capacity,
                                   size_t max_capacity,
                                   bool generate_debug_info,
                                   std::string* error_msg) {
  ScopedTrace trace(__PRETTY_FUNCTION__);
  CHECK_GE(max_capacity, initial_capacity);

  // Generating debug information is mostly for using the 'perf' tool, which does
  // not work with ashmem.
  bool use_ashmem = !generate_debug_info;
  // With 'perf', we want a 1-1 mapping between an address and a method.
  bool garbage_collect_code = !generate_debug_info;

  // We need to have 32 bit offsets from method headers in code cache which point to things
  // in the data cache. If the maps are more than 4G apart, having multiple maps wouldn't work.
  // Ensure we're below 1 GB to be safe.
  if (max_capacity > 1 * GB) {
    std::ostringstream oss;
    oss << "Maxium code cache capacity is limited to 1 GB, "
        << PrettySize(max_capacity) << " is too big";
    *error_msg = oss.str();
    return nullptr;
  }

  std::string error_str;
  // Map name specific for android_os_Debug.cpp accounting.
  // Map in low 4gb to simplify accessing root tables for x86_64.
  // We could do PC-relative addressing to avoid this problem, but that
  // would require reserving code and data area before submitting, which
  // means more windows for the code memory to be RWX.
  MemMap* data_map = MemMap::MapAnonymous(
      "data-code-cache", nullptr,
      max_capacity,
      kProtAll,
      /* low_4gb */ true,
      /* reuse */ false,
      &error_str,
      use_ashmem);
  if (data_map == nullptr) {
    std::ostringstream oss;
    oss << "Failed to create read write execute cache: " << error_str << " size=" << max_capacity;
    *error_msg = oss.str();
    return nullptr;
  }

  // Align both capacities to page size, as that's the unit mspaces use.
  initial_capacity = RoundDown(initial_capacity, 2 * kPageSize);
  max_capacity = RoundDown(max_capacity, 2 * kPageSize);

  // Data cache is 1 / 2 of the map.
  // TODO: Make this variable?
  size_t data_size = max_capacity / 2;
  size_t code_size = max_capacity - data_size;
  DCHECK_EQ(code_size + data_size, max_capacity);
  uint8_t* divider = data_map->Begin() + data_size;

  MemMap* code_map =
      data_map->RemapAtEnd(divider, "jit-code-cache", kProtAll, &error_str, use_ashmem);
  if (code_map == nullptr) {
    std::ostringstream oss;
    oss << "Failed to create read write execute cache: " << error_str << " size=" << max_capacity;
    *error_msg = oss.str();
    return nullptr;
  }
  DCHECK_EQ(code_map->Begin(), divider);
  data_size = initial_capacity / 2;
  code_size = initial_capacity - data_size;
  DCHECK_EQ(code_size + data_size, initial_capacity);
  return new JitCodeCache(
      code_map, data_map, code_size, data_size, max_capacity, garbage_collect_code);
}

JitCodeCache::JitCodeCache(MemMap* code_map,
                           MemMap* data_map,
                           size_t initial_code_capacity,
                           size_t initial_data_capacity,
                           size_t max_capacity,
                           bool garbage_collect_code)
    : lock_("Jit code cache", kJitCodeCacheLock),
      lock_cond_("Jit code cache condition variable", lock_),
      collection_in_progress_(false),
      code_map_(code_map),
      data_map_(data_map),
      max_capacity_(max_capacity),
      current_capacity_(initial_code_capacity + initial_data_capacity),
      code_end_(initial_code_capacity),
      data_end_(initial_data_capacity),
      last_collection_increased_code_cache_(false),
      last_update_time_ns_(0),
      garbage_collect_code_(garbage_collect_code),
      used_memory_for_data_(0),
      used_memory_for_code_(0),
      number_of_compilations_(0),
      number_of_osr_compilations_(0),
      number_of_deoptimizations_(0),
      number_of_collections_(0),
      histogram_stack_map_memory_use_("Memory used for stack maps", 16),
      histogram_code_memory_use_("Memory used for compiled code", 16),
      histogram_profiling_info_memory_use_("Memory used for profiling info", 16),
      is_weak_access_enabled_(true),
      inline_cache_cond_("Jit inline cache condition variable", lock_) {

  DCHECK_GE(max_capacity, initial_code_capacity + initial_data_capacity);
  code_mspace_ = create_mspace_with_base(code_map_->Begin(), code_end_, false /*locked*/);
  data_mspace_ = create_mspace_with_base(data_map_->Begin(), data_end_, false /*locked*/);

  if (code_mspace_ == nullptr || data_mspace_ == nullptr) {
    PLOG(FATAL) << "create_mspace_with_base failed";
  }

  SetFootprintLimit(current_capacity_);

  CHECKED_MPROTECT(code_map_->Begin(), code_map_->Size(), kProtCode);
  CHECKED_MPROTECT(data_map_->Begin(), data_map_->Size(), kProtData);

  VLOG(jit) << "Created jit code cache: initial data size="
            << PrettySize(initial_data_capacity)
            << ", initial code size="
            << PrettySize(initial_code_capacity);
}

bool JitCodeCache::ContainsPc(const void* ptr) const {
  return code_map_->Begin() <= ptr && ptr < code_map_->End();
}

bool JitCodeCache::ContainsMethod(ArtMethod* method) {
  MutexLock mu(Thread::Current(), lock_);
  for (auto& it : method_code_map_) {
    if (it.second == method) {
      return true;
    }
  }
  return false;
}

class ScopedCodeCacheWrite : ScopedTrace {
 public:
  explicit ScopedCodeCacheWrite(MemMap* code_map)
      : ScopedTrace("ScopedCodeCacheWrite"),
        code_map_(code_map) {
    ScopedTrace trace("mprotect all");
    CHECKED_MPROTECT(code_map_->Begin(), code_map_->Size(), kProtAll);
  }
  ~ScopedCodeCacheWrite() {
    ScopedTrace trace("mprotect code");
    CHECKED_MPROTECT(code_map_->Begin(), code_map_->Size(), kProtCode);
  }
 private:
  MemMap* const code_map_;

  DISALLOW_COPY_AND_ASSIGN(ScopedCodeCacheWrite);
};

uint8_t* JitCodeCache::CommitCode(Thread* self,
                                  ArtMethod* method,
                                  uint8_t* stack_map,
                                  uint8_t* roots_data,
                                  size_t frame_size_in_bytes,
                                  size_t core_spill_mask,
                                  size_t fp_spill_mask,
                                  const uint8_t* code,
                                  size_t code_size,
                                  bool osr,
                                  Handle<mirror::ObjectArray<mirror::Object>> roots) {
  uint8_t* result = CommitCodeInternal(self,
                                       method,
                                       stack_map,
                                       roots_data,
                                       frame_size_in_bytes,
                                       core_spill_mask,
                                       fp_spill_mask,
                                       code,
                                       code_size,
                                       osr,
                                       roots);
  if (result == nullptr) {
    // Retry.
    GarbageCollectCache(self);
    result = CommitCodeInternal(self,
                                method,
                                stack_map,
                                roots_data,
                                frame_size_in_bytes,
                                core_spill_mask,
                                fp_spill_mask,
                                code,
                                code_size,
                                osr,
                                roots);
  }
  return result;
}

bool JitCodeCache::WaitForPotentialCollectionToComplete(Thread* self) {
  bool in_collection = false;
  while (collection_in_progress_) {
    in_collection = true;
    lock_cond_.Wait(self);
  }
  return in_collection;
}

static uintptr_t FromCodeToAllocation(const void* code) {
  size_t alignment = GetInstructionSetAlignment(kRuntimeISA);
  return reinterpret_cast<uintptr_t>(code) - RoundUp(sizeof(OatQuickMethodHeader), alignment);
}

static uint32_t ComputeRootTableSize(uint32_t number_of_roots) {
  return sizeof(uint32_t) + number_of_roots * sizeof(GcRoot<mirror::Object>);
}

static uint32_t GetNumberOfRoots(const uint8_t* stack_map) {
  // The length of the table is stored just before the stack map (and therefore at the end of
  // the table itself), in order to be able to fetch it from a `stack_map` pointer.
  return reinterpret_cast<const uint32_t*>(stack_map)[-1];
}

static void FillRootTableLength(uint8_t* roots_data, uint32_t length) {
  // Store the length of the table at the end. This will allow fetching it from a `stack_map`
  // pointer.
  reinterpret_cast<uint32_t*>(roots_data)[length] = length;
}

static void FillRootTable(uint8_t* roots_data, Handle<mirror::ObjectArray<mirror::Object>> roots)
    REQUIRES_SHARED(Locks::mutator_lock_) {
  GcRoot<mirror::Object>* gc_roots = reinterpret_cast<GcRoot<mirror::Object>*>(roots_data);
  const uint32_t length = roots->GetLength();
  // Put all roots in `roots_data`.
  for (uint32_t i = 0; i < length; ++i) {
    ObjPtr<mirror::Object> object = roots->Get(i);
    if (kIsDebugBuild) {
      // Ensure the string is strongly interned. b/32995596
      CHECK(object->IsString());
      ObjPtr<mirror::String> str = reinterpret_cast<mirror::String*>(object.Ptr());
      ClassLinker* class_linker = Runtime::Current()->GetClassLinker();
      CHECK(class_linker->GetInternTable()->LookupStrong(Thread::Current(), str) != nullptr);
    }
    gc_roots[i] = GcRoot<mirror::Object>(object);
  }
  FillRootTableLength(roots_data, length);
}

static uint8_t* GetRootTable(const void* code_ptr, uint32_t* number_of_roots = nullptr) {
  OatQuickMethodHeader* method_header = OatQuickMethodHeader::FromCodePointer(code_ptr);
  uint8_t* data = method_header->GetOptimizedCodeInfoPtr();
  uint32_t roots = GetNumberOfRoots(data);
  if (number_of_roots != nullptr) {
    *number_of_roots = roots;
  }
  return data - ComputeRootTableSize(roots);
}

void JitCodeCache::SweepRootTables(IsMarkedVisitor* visitor) {
  MutexLock mu(Thread::Current(), lock_);
  for (const auto& entry : method_code_map_) {
    uint32_t number_of_roots = 0;
    uint8_t* roots_data = GetRootTable(entry.first, &number_of_roots);
    GcRoot<mirror::Object>* roots = reinterpret_cast<GcRoot<mirror::Object>*>(roots_data);
    for (uint32_t i = 0; i < number_of_roots; ++i) {
      // This does not need a read barrier because this is called by GC.
      mirror::Object* object = roots[i].Read<kWithoutReadBarrier>();
      DCHECK(object != nullptr);
      mirror::Object* new_object = visitor->IsMarked(object);
      // We know the string is marked because it's a strongly-interned string that
      // is always alive. The IsMarked implementation of the CMS collector returns
      // null for newly allocated objects, but we know those haven't moved. Therefore,
      // only update the entry if we get a different non-null string.
      // TODO: Do not use IsMarked for j.l.Class, and adjust once we move this method
      // out of the weak access/creation pause. b/32167580
      if (new_object != nullptr && new_object != object) {
        DCHECK(new_object->IsString());
        roots[i] = GcRoot<mirror::Object>(new_object);
      }
    }
  }
  // Walk over inline caches to clear entries containing unloaded classes.
  for (ProfilingInfo* info : profiling_infos_) {
    for (size_t i = 0; i < info->number_of_inline_caches_; ++i) {
      InlineCache* cache = &info->cache_[i];
      for (size_t j = 0; j < InlineCache::kIndividualCacheSize; ++j) {
        // This does not need a read barrier because this is called by GC.
        mirror::Class* cls = cache->classes_[j].Read<kWithoutReadBarrier>();
        if (cls != nullptr) {
          // Look at the classloader of the class to know if it has been
          // unloaded.
          // This does not need a read barrier because this is called by GC.
          mirror::Object* class_loader =
              cls->GetClassLoader<kDefaultVerifyFlags, kWithoutReadBarrier>();
          if (class_loader == nullptr || visitor->IsMarked(class_loader) != nullptr) {
            // The class loader is live, update the entry if the class has moved.
            mirror::Class* new_cls = down_cast<mirror::Class*>(visitor->IsMarked(cls));
            // Note that new_object can be null for CMS and newly allocated objects.
            if (new_cls != nullptr && new_cls != cls) {
              cache->classes_[j] = GcRoot<mirror::Class>(new_cls);
            }
          } else {
            // The class loader is not live, clear the entry.
            cache->classes_[j] = GcRoot<mirror::Class>(nullptr);
          }
        }
      }
    }
  }
}

void JitCodeCache::FreeCode(const void* code_ptr, ArtMethod* method ATTRIBUTE_UNUSED) {
  uintptr_t allocation = FromCodeToAllocation(code_ptr);
  // Notify native debugger that we are about to remove the code.
  // It does nothing if we are not using native debugger.
  DeleteJITCodeEntryForAddress(reinterpret_cast<uintptr_t>(code_ptr));
  FreeData(GetRootTable(code_ptr));
  FreeCode(reinterpret_cast<uint8_t*>(allocation));
}

void JitCodeCache::RemoveMethodsIn(Thread* self, const LinearAlloc& alloc) {
  ScopedTrace trace(__PRETTY_FUNCTION__);
  MutexLock mu(self, lock_);
  // We do not check if a code cache GC is in progress, as this method comes
  // with the classlinker_classes_lock_ held, and suspending ourselves could
  // lead to a deadlock.
  {
    ScopedCodeCacheWrite scc(code_map_.get());
    for (auto it = method_code_map_.begin(); it != method_code_map_.end();) {
      if (alloc.ContainsUnsafe(it->second)) {
        FreeCode(it->first, it->second);
        it = method_code_map_.erase(it);
      } else {
        ++it;
      }
    }
  }
  for (auto it = osr_code_map_.begin(); it != osr_code_map_.end();) {
    if (alloc.ContainsUnsafe(it->first)) {
      // Note that the code has already been removed in the loop above.
      it = osr_code_map_.erase(it);
    } else {
      ++it;
    }
  }
  for (auto it = profiling_infos_.begin(); it != profiling_infos_.end();) {
    ProfilingInfo* info = *it;
    if (alloc.ContainsUnsafe(info->GetMethod())) {
      info->GetMethod()->SetProfilingInfo(nullptr);
      FreeData(reinterpret_cast<uint8_t*>(info));
      it = profiling_infos_.erase(it);
    } else {
      ++it;
    }
  }
}

bool JitCodeCache::IsWeakAccessEnabled(Thread* self) const {
  return kUseReadBarrier
      ? self->GetWeakRefAccessEnabled()
      : is_weak_access_enabled_.LoadSequentiallyConsistent();
}

void JitCodeCache::WaitUntilInlineCacheAccessible(Thread* self) {
  if (IsWeakAccessEnabled(self)) {
    return;
  }
  ScopedThreadSuspension sts(self, kWaitingWeakGcRootRead);
  MutexLock mu(self, lock_);
  while (!IsWeakAccessEnabled(self)) {
    inline_cache_cond_.Wait(self);
  }
}

void JitCodeCache::BroadcastForInlineCacheAccess() {
  Thread* self = Thread::Current();
  MutexLock mu(self, lock_);
  inline_cache_cond_.Broadcast(self);
}

void JitCodeCache::AllowInlineCacheAccess() {
  DCHECK(!kUseReadBarrier);
  is_weak_access_enabled_.StoreSequentiallyConsistent(true);
  BroadcastForInlineCacheAccess();
}

void JitCodeCache::DisallowInlineCacheAccess() {
  DCHECK(!kUseReadBarrier);
  is_weak_access_enabled_.StoreSequentiallyConsistent(false);
}

void JitCodeCache::CopyInlineCacheInto(const InlineCache& ic,
                                       Handle<mirror::ObjectArray<mirror::Class>> array) {
  WaitUntilInlineCacheAccessible(Thread::Current());
  // Note that we don't need to lock `lock_` here, the compiler calling
  // this method has already ensured the inline cache will not be deleted.
  for (size_t in_cache = 0, in_array = 0;
       in_cache < InlineCache::kIndividualCacheSize;
       ++in_cache) {
    mirror::Class* object = ic.classes_[in_cache].Read();
    if (object != nullptr) {
      array->Set(in_array++, object);
    }
  }
}

uint8_t* JitCodeCache::CommitCodeInternal(Thread* self,
                                          ArtMethod* method,
                                          uint8_t* stack_map,
                                          uint8_t* roots_data,
                                          size_t frame_size_in_bytes,
                                          size_t core_spill_mask,
                                          size_t fp_spill_mask,
                                          const uint8_t* code,
                                          size_t code_size,
                                          bool osr,
                                          Handle<mirror::ObjectArray<mirror::Object>> roots) {
  DCHECK(stack_map != nullptr);
  size_t alignment = GetInstructionSetAlignment(kRuntimeISA);
  // Ensure the header ends up at expected instruction alignment.
  size_t header_size = RoundUp(sizeof(OatQuickMethodHeader), alignment);
  size_t total_size = header_size + code_size;
  const uint32_t num_roots = roots->GetLength();

  OatQuickMethodHeader* method_header = nullptr;
  uint8_t* code_ptr = nullptr;
  uint8_t* memory = nullptr;
  {
    ScopedThreadSuspension sts(self, kSuspended);
    MutexLock mu(self, lock_);
    WaitForPotentialCollectionToComplete(self);
    {
      ScopedCodeCacheWrite scc(code_map_.get());
      memory = AllocateCode(total_size);
      if (memory == nullptr) {
        // Fill root table length so that ClearData works correctly in case of failure. Otherwise
        // the length will be 0 and cause incorrect DCHECK failure.
        FillRootTableLength(roots_data, num_roots);
        return nullptr;
      }
      code_ptr = memory + header_size;

      std::copy(code, code + code_size, code_ptr);
      method_header = OatQuickMethodHeader::FromCodePointer(code_ptr);
      new (method_header) OatQuickMethodHeader(
          code_ptr - stack_map,
          frame_size_in_bytes,
          core_spill_mask,
          fp_spill_mask,
          code_size);
      // Flush caches before we remove write permission because on some ARMv8 hardware,
      // flushing caches require write permissions.
      //
      // For reference, here are kernel patches discussing about this issue:
      // https://android.googlesource.com/kernel/msm/%2B/0e7f7bcc3fc87489cda5aa6aff8ce40eed912279
      // https://patchwork.kernel.org/patch/9047921/
      FlushInstructionCache(reinterpret_cast<char*>(code_ptr),
                            reinterpret_cast<char*>(code_ptr + code_size));
    }

    number_of_compilations_++;
  }
  // We need to update the entry point in the runnable state for the instrumentation.
  {
    MutexLock mu(self, lock_);
    method_code_map_.Put(code_ptr, method);
    // Fill the root table before updating the entry point.
    FillRootTable(roots_data, roots);
    if (osr) {
      number_of_osr_compilations_++;
      osr_code_map_.Put(method, code_ptr);
    } else {
      Runtime::Current()->GetInstrumentation()->UpdateMethodsCode(
          method, method_header->GetEntryPoint());
    }
    if (collection_in_progress_) {
      // We need to update the live bitmap if there is a GC to ensure it sees this new
      // code.
      GetLiveBitmap()->AtomicTestAndSet(FromCodeToAllocation(code_ptr));
    }
    last_update_time_ns_.StoreRelease(NanoTime());
    VLOG(jit)
        << "JIT added (osr=" << std::boolalpha << osr << std::noboolalpha << ") "
        << ArtMethod::PrettyMethod(method) << "@" << method
        << " ccache_size=" << PrettySize(CodeCacheSizeLocked()) << ": "
        << " dcache_size=" << PrettySize(DataCacheSizeLocked()) << ": "
        << reinterpret_cast<const void*>(method_header->GetEntryPoint()) << ","
        << reinterpret_cast<const void*>(method_header->GetEntryPoint() + method_header->code_size_);
    histogram_code_memory_use_.AddValue(code_size);
    if (code_size > kCodeSizeLogThreshold) {
      LOG(INFO) << "JIT allocated "
                << PrettySize(code_size)
                << " for compiled code of "
                << ArtMethod::PrettyMethod(method);
    }
  }

  return reinterpret_cast<uint8_t*>(method_header);
}

size_t JitCodeCache::CodeCacheSize() {
  MutexLock mu(Thread::Current(), lock_);
  return CodeCacheSizeLocked();
}

size_t JitCodeCache::CodeCacheSizeLocked() {
  return used_memory_for_code_;
}

size_t JitCodeCache::DataCacheSize() {
  MutexLock mu(Thread::Current(), lock_);
  return DataCacheSizeLocked();
}

size_t JitCodeCache::DataCacheSizeLocked() {
  return used_memory_for_data_;
}

static const uint8_t* FromStackMapToRoots(const uint8_t* stack_map_data) {
  return stack_map_data - ComputeRootTableSize(GetNumberOfRoots(stack_map_data));
}

void JitCodeCache::ClearData(Thread* self,
                             uint8_t* stack_map_data,
                             uint8_t* roots_data) {
  DCHECK_EQ(FromStackMapToRoots(stack_map_data), roots_data);
  MutexLock mu(self, lock_);
  FreeData(reinterpret_cast<uint8_t*>(roots_data));
}

void JitCodeCache::ReserveData(Thread* self,
                               size_t stack_map_size,
                               size_t number_of_roots,
                               ArtMethod* method,
                               uint8_t** stack_map_data,
                               uint8_t** roots_data) {
  size_t table_size = ComputeRootTableSize(number_of_roots);
  size_t size = RoundUp(stack_map_size + table_size, sizeof(void*));
  uint8_t* result = nullptr;

  {
    ScopedThreadSuspension sts(self, kSuspended);
    MutexLock mu(self, lock_);
    WaitForPotentialCollectionToComplete(self);
    result = AllocateData(size);
  }

  if (result == nullptr) {
    // Retry.
    GarbageCollectCache(self);
    ScopedThreadSuspension sts(self, kSuspended);
    MutexLock mu(self, lock_);
    WaitForPotentialCollectionToComplete(self);
    result = AllocateData(size);
  }

  MutexLock mu(self, lock_);
  histogram_stack_map_memory_use_.AddValue(size);
  if (size > kStackMapSizeLogThreshold) {
    LOG(INFO) << "JIT allocated "
              << PrettySize(size)
              << " for stack maps of "
              << ArtMethod::PrettyMethod(method);
  }
  *roots_data = result;
  *stack_map_data = result + table_size;
}

class MarkCodeVisitor FINAL : public StackVisitor {
 public:
  MarkCodeVisitor(Thread* thread_in, JitCodeCache* code_cache_in)
      : StackVisitor(thread_in, nullptr, StackVisitor::StackWalkKind::kSkipInlinedFrames),
        code_cache_(code_cache_in),
        bitmap_(code_cache_->GetLiveBitmap()) {}

  bool VisitFrame() OVERRIDE REQUIRES_SHARED(Locks::mutator_lock_) {
    const OatQuickMethodHeader* method_header = GetCurrentOatQuickMethodHeader();
    if (method_header == nullptr) {
      return true;
    }
    const void* code = method_header->GetCode();
    if (code_cache_->ContainsPc(code)) {
      // Use the atomic set version, as multiple threads are executing this code.
      bitmap_->AtomicTestAndSet(FromCodeToAllocation(code));
    }
    return true;
  }

 private:
  JitCodeCache* const code_cache_;
  CodeCacheBitmap* const bitmap_;
};

class MarkCodeClosure FINAL : public Closure {
 public:
  MarkCodeClosure(JitCodeCache* code_cache, Barrier* barrier)
      : code_cache_(code_cache), barrier_(barrier) {}

  void Run(Thread* thread) OVERRIDE REQUIRES_SHARED(Locks::mutator_lock_) {
    ScopedTrace trace(__PRETTY_FUNCTION__);
    DCHECK(thread == Thread::Current() || thread->IsSuspended());
    MarkCodeVisitor visitor(thread, code_cache_);
    visitor.WalkStack();
    if (kIsDebugBuild) {
      // The stack walking code queries the side instrumentation stack if it
      // sees an instrumentation exit pc, so the JIT code of methods in that stack
      // must have been seen. We sanity check this below.
      for (const instrumentation::InstrumentationStackFrame& frame
              : *thread->GetInstrumentationStack()) {
        // The 'method_' in InstrumentationStackFrame is the one that has return_pc_ in
        // its stack frame, it is not the method owning return_pc_. We just pass null to
        // LookupMethodHeader: the method is only checked against in debug builds.
        OatQuickMethodHeader* method_header =
            code_cache_->LookupMethodHeader(frame.return_pc_, nullptr);
        if (method_header != nullptr) {
          const void* code = method_header->GetCode();
          CHECK(code_cache_->GetLiveBitmap()->Test(FromCodeToAllocation(code)));
        }
      }
    }
    barrier_->Pass(Thread::Current());
  }

 private:
  JitCodeCache* const code_cache_;
  Barrier* const barrier_;
};

void JitCodeCache::NotifyCollectionDone(Thread* self) {
  collection_in_progress_ = false;
  lock_cond_.Broadcast(self);
}

void JitCodeCache::SetFootprintLimit(size_t new_footprint) {
  size_t per_space_footprint = new_footprint / 2;
  DCHECK(IsAlignedParam(per_space_footprint, kPageSize));
  DCHECK_EQ(per_space_footprint * 2, new_footprint);
  mspace_set_footprint_limit(data_mspace_, per_space_footprint);
  {
    ScopedCodeCacheWrite scc(code_map_.get());
    mspace_set_footprint_limit(code_mspace_, per_space_footprint);
  }
}

bool JitCodeCache::IncreaseCodeCacheCapacity() {
  if (current_capacity_ == max_capacity_) {
    return false;
  }

  // Double the capacity if we're below 1MB, or increase it by 1MB if
  // we're above.
  if (current_capacity_ < 1 * MB) {
    current_capacity_ *= 2;
  } else {
    current_capacity_ += 1 * MB;
  }
  if (current_capacity_ > max_capacity_) {
    current_capacity_ = max_capacity_;
  }

  if (!kIsDebugBuild || VLOG_IS_ON(jit)) {
    LOG(INFO) << "Increasing code cache capacity to " << PrettySize(current_capacity_);
  }

  SetFootprintLimit(current_capacity_);

  return true;
}

void JitCodeCache::MarkCompiledCodeOnThreadStacks(Thread* self) {
  Barrier barrier(0);
  size_t threads_running_checkpoint = 0;
  MarkCodeClosure closure(this, &barrier);
  threads_running_checkpoint = Runtime::Current()->GetThreadList()->RunCheckpoint(&closure);
  // Now that we have run our checkpoint, move to a suspended state and wait
  // for other threads to run the checkpoint.
  ScopedThreadSuspension sts(self, kSuspended);
  if (threads_running_checkpoint != 0) {
    barrier.Increment(self, threads_running_checkpoint);
  }
}

bool JitCodeCache::ShouldDoFullCollection() {
  if (current_capacity_ == max_capacity_) {
    // Always do a full collection when the code cache is full.
    return true;
  } else if (current_capacity_ < kReservedCapacity) {
    // Always do partial collection when the code cache size is below the reserved
    // capacity.
    return false;
  } else if (last_collection_increased_code_cache_) {
    // This time do a full collection.
    return true;
  } else {
    // This time do a partial collection.
    return false;
  }
}

void JitCodeCache::GarbageCollectCache(Thread* self) {
  ScopedTrace trace(__FUNCTION__);
  if (!garbage_collect_code_) {
    MutexLock mu(self, lock_);
    IncreaseCodeCacheCapacity();
    return;
  }

  // Wait for an existing collection, or let everyone know we are starting one.
  {
    ScopedThreadSuspension sts(self, kSuspended);
    MutexLock mu(self, lock_);
    if (WaitForPotentialCollectionToComplete(self)) {
      return;
    } else {
      number_of_collections_++;
      live_bitmap_.reset(CodeCacheBitmap::Create(
          "code-cache-bitmap",
          reinterpret_cast<uintptr_t>(code_map_->Begin()),
          reinterpret_cast<uintptr_t>(code_map_->Begin() + current_capacity_ / 2)));
      collection_in_progress_ = true;
    }
  }

  TimingLogger logger("JIT code cache timing logger", true, VLOG_IS_ON(jit));
  {
    TimingLogger::ScopedTiming st("Code cache collection", &logger);

    bool do_full_collection = false;
    {
      MutexLock mu(self, lock_);
      do_full_collection = ShouldDoFullCollection();
    }

    if (!kIsDebugBuild || VLOG_IS_ON(jit)) {
      LOG(INFO) << "Do "
                << (do_full_collection ? "full" : "partial")
                << " code cache collection, code="
                << PrettySize(CodeCacheSize())
                << ", data=" << PrettySize(DataCacheSize());
    }

    DoCollection(self, /* collect_profiling_info */ do_full_collection);

    if (!kIsDebugBuild || VLOG_IS_ON(jit)) {
      LOG(INFO) << "After code cache collection, code="
                << PrettySize(CodeCacheSize())
                << ", data=" << PrettySize(DataCacheSize());
    }

    {
      MutexLock mu(self, lock_);

      // Increase the code cache only when we do partial collections.
      // TODO: base this strategy on how full the code cache is?
      if (do_full_collection) {
        last_collection_increased_code_cache_ = false;
      } else {
        last_collection_increased_code_cache_ = true;
        IncreaseCodeCacheCapacity();
      }

      bool next_collection_will_be_full = ShouldDoFullCollection();

      // Start polling the liveness of compiled code to prepare for the next full collection.
      if (next_collection_will_be_full) {
        // Save the entry point of methods we have compiled, and update the entry
        // point of those methods to the interpreter. If the method is invoked, the
        // interpreter will update its entry point to the compiled code and call it.
        for (ProfilingInfo* info : profiling_infos_) {
          const void* entry_point = info->GetMethod()->GetEntryPointFromQuickCompiledCode();
          if (ContainsPc(entry_point)) {
            info->SetSavedEntryPoint(entry_point);
            Runtime::Current()->GetInstrumentation()->UpdateMethodsCode(
                info->GetMethod(), GetQuickToInterpreterBridge());
          }
        }

        DCHECK(CheckLiveCompiledCodeHasProfilingInfo());
      }
      live_bitmap_.reset(nullptr);
      NotifyCollectionDone(self);
    }
  }
  Runtime::Current()->GetJit()->AddTimingLogger(logger);
}

void JitCodeCache::RemoveUnmarkedCode(Thread* self) {
  ScopedTrace trace(__FUNCTION__);
  MutexLock mu(self, lock_);
  ScopedCodeCacheWrite scc(code_map_.get());
  // Iterate over all compiled code and remove entries that are not marked.
  for (auto it = method_code_map_.begin(); it != method_code_map_.end();) {
    const void* code_ptr = it->first;
    ArtMethod* method = it->second;
    uintptr_t allocation = FromCodeToAllocation(code_ptr);
    if (GetLiveBitmap()->Test(allocation)) {
      ++it;
    } else {
      FreeCode(code_ptr, method);
      it = method_code_map_.erase(it);
    }
  }
}

void JitCodeCache::DoCollection(Thread* self, bool collect_profiling_info) {
  ScopedTrace trace(__FUNCTION__);
  {
    MutexLock mu(self, lock_);
    if (collect_profiling_info) {
      // Clear the profiling info of methods that do not have compiled code as entrypoint.
      // Also remove the saved entry point from the ProfilingInfo objects.
      for (ProfilingInfo* info : profiling_infos_) {
        const void* ptr = info->GetMethod()->GetEntryPointFromQuickCompiledCode();
        if (!ContainsPc(ptr) && !info->IsInUseByCompiler()) {
          info->GetMethod()->SetProfilingInfo(nullptr);
        }

        if (info->GetSavedEntryPoint() != nullptr) {
          info->SetSavedEntryPoint(nullptr);
          // We are going to move this method back to interpreter. Clear the counter now to
          // give it a chance to be hot again.
          info->GetMethod()->ClearCounter();
        }
      }
    } else if (kIsDebugBuild) {
      // Sanity check that the profiling infos do not have a dangling entry point.
      for (ProfilingInfo* info : profiling_infos_) {
        DCHECK(info->GetSavedEntryPoint() == nullptr);
      }
    }

    // Mark compiled code that are entrypoints of ArtMethods. Compiled code that is not
    // an entry point is either:
    // - an osr compiled code, that will be removed if not in a thread call stack.
    // - discarded compiled code, that will be removed if not in a thread call stack.
    for (const auto& it : method_code_map_) {
      ArtMethod* method = it.second;
      const void* code_ptr = it.first;
      const OatQuickMethodHeader* method_header = OatQuickMethodHeader::FromCodePointer(code_ptr);
      if (method_header->GetEntryPoint() == method->GetEntryPointFromQuickCompiledCode()) {
        GetLiveBitmap()->AtomicTestAndSet(FromCodeToAllocation(code_ptr));
      }
    }

    // Empty osr method map, as osr compiled code will be deleted (except the ones
    // on thread stacks).
    osr_code_map_.clear();
  }

  // Run a checkpoint on all threads to mark the JIT compiled code they are running.
  MarkCompiledCodeOnThreadStacks(self);

  // At this point, mutator threads are still running, and entrypoints of methods can
  // change. We do know they cannot change to a code cache entry that is not marked,
  // therefore we can safely remove those entries.
  RemoveUnmarkedCode(self);

  if (collect_profiling_info) {
    ScopedThreadSuspension sts(self, kSuspended);
    MutexLock mu(self, lock_);
    // Free all profiling infos of methods not compiled nor being compiled.
    auto profiling_kept_end = std::remove_if(profiling_infos_.begin(), profiling_infos_.end(),
      [this] (ProfilingInfo* info) NO_THREAD_SAFETY_ANALYSIS {
        const void* ptr = info->GetMethod()->GetEntryPointFromQuickCompiledCode();
        // We have previously cleared the ProfilingInfo pointer in the ArtMethod in the hope
        // that the compiled code would not get revived. As mutator threads run concurrently,
        // they may have revived the compiled code, and now we are in the situation where
        // a method has compiled code but no ProfilingInfo.
        // We make sure compiled methods have a ProfilingInfo object. It is needed for
        // code cache collection.
        if (ContainsPc(ptr) &&
            info->GetMethod()->GetProfilingInfo(kRuntimePointerSize) == nullptr) {
          info->GetMethod()->SetProfilingInfo(info);
        } else if (info->GetMethod()->GetProfilingInfo(kRuntimePointerSize) != info) {
          // No need for this ProfilingInfo object anymore.
          FreeData(reinterpret_cast<uint8_t*>(info));
          return true;
        }
        return false;
      });
    profiling_infos_.erase(profiling_kept_end, profiling_infos_.end());
    DCHECK(CheckLiveCompiledCodeHasProfilingInfo());
  }
}

bool JitCodeCache::CheckLiveCompiledCodeHasProfilingInfo() {
  ScopedTrace trace(__FUNCTION__);
  // Check that methods we have compiled do have a ProfilingInfo object. We would
  // have memory leaks of compiled code otherwise.
  for (const auto& it : method_code_map_) {
    ArtMethod* method = it.second;
    if (method->GetProfilingInfo(kRuntimePointerSize) == nullptr) {
      const void* code_ptr = it.first;
      const OatQuickMethodHeader* method_header = OatQuickMethodHeader::FromCodePointer(code_ptr);
      if (method_header->GetEntryPoint() == method->GetEntryPointFromQuickCompiledCode()) {
        // If the code is not dead, then we have a problem. Note that this can even
        // happen just after a collection, as mutator threads are running in parallel
        // and could deoptimize an existing compiled code.
        return false;
      }
    }
  }
  return true;
}

OatQuickMethodHeader* JitCodeCache::LookupMethodHeader(uintptr_t pc, ArtMethod* method) {
  static_assert(kRuntimeISA != kThumb2, "kThumb2 cannot be a runtime ISA");
  if (kRuntimeISA == kArm) {
    // On Thumb-2, the pc is offset by one.
    --pc;
  }
  if (!ContainsPc(reinterpret_cast<const void*>(pc))) {
    return nullptr;
  }

  MutexLock mu(Thread::Current(), lock_);
  if (method_code_map_.empty()) {
    return nullptr;
  }
  auto it = method_code_map_.lower_bound(reinterpret_cast<const void*>(pc));
  --it;

  const void* code_ptr = it->first;
  OatQuickMethodHeader* method_header = OatQuickMethodHeader::FromCodePointer(code_ptr);
  if (!method_header->Contains(pc)) {
    return nullptr;
  }
  if (kIsDebugBuild && method != nullptr) {
    DCHECK_EQ(it->second, method)
        << ArtMethod::PrettyMethod(method) << " " << ArtMethod::PrettyMethod(it->second) << " "
        << std::hex << pc;
  }
  return method_header;
}

OatQuickMethodHeader* JitCodeCache::LookupOsrMethodHeader(ArtMethod* method) {
  MutexLock mu(Thread::Current(), lock_);
  auto it = osr_code_map_.find(method);
  if (it == osr_code_map_.end()) {
    return nullptr;
  }
  return OatQuickMethodHeader::FromCodePointer(it->second);
}

ProfilingInfo* JitCodeCache::AddProfilingInfo(Thread* self,
                                              ArtMethod* method,
                                              const std::vector<uint32_t>& entries,
                                              bool retry_allocation)
    // No thread safety analysis as we are using TryLock/Unlock explicitly.
    NO_THREAD_SAFETY_ANALYSIS {
  ProfilingInfo* info = nullptr;
  if (!retry_allocation) {
    // If we are allocating for the interpreter, just try to lock, to avoid
    // lock contention with the JIT.
    if (lock_.ExclusiveTryLock(self)) {
      info = AddProfilingInfoInternal(self, method, entries);
      lock_.ExclusiveUnlock(self);
    }
  } else {
    {
      MutexLock mu(self, lock_);
      info = AddProfilingInfoInternal(self, method, entries);
    }

    if (info == nullptr) {
      GarbageCollectCache(self);
      MutexLock mu(self, lock_);
      info = AddProfilingInfoInternal(self, method, entries);
    }
  }
  return info;
}

ProfilingInfo* JitCodeCache::AddProfilingInfoInternal(Thread* self ATTRIBUTE_UNUSED,
                                                      ArtMethod* method,
                                                      const std::vector<uint32_t>& entries) {
  size_t profile_info_size = RoundUp(
      sizeof(ProfilingInfo) + sizeof(InlineCache) * entries.size(),
      sizeof(void*));

  // Check whether some other thread has concurrently created it.
  ProfilingInfo* info = method->GetProfilingInfo(kRuntimePointerSize);
  if (info != nullptr) {
    return info;
  }

  uint8_t* data = AllocateData(profile_info_size);
  if (data == nullptr) {
    return nullptr;
  }
  info = new (data) ProfilingInfo(method, entries);

  // Make sure other threads see the data in the profiling info object before the
  // store in the ArtMethod's ProfilingInfo pointer.
  QuasiAtomic::ThreadFenceRelease();

  method->SetProfilingInfo(info);
  profiling_infos_.push_back(info);
  histogram_profiling_info_memory_use_.AddValue(profile_info_size);
  return info;
}

// NO_THREAD_SAFETY_ANALYSIS as this is called from mspace code, at which point the lock
// is already held.
void* JitCodeCache::MoreCore(const void* mspace, intptr_t increment) NO_THREAD_SAFETY_ANALYSIS {
  if (code_mspace_ == mspace) {
    size_t result = code_end_;
    code_end_ += increment;
    return reinterpret_cast<void*>(result + code_map_->Begin());
  } else {
    DCHECK_EQ(data_mspace_, mspace);
    size_t result = data_end_;
    data_end_ += increment;
    return reinterpret_cast<void*>(result + data_map_->Begin());
  }
}

void JitCodeCache::GetProfiledMethods(const std::set<std::string>& dex_base_locations,
                                      std::vector<MethodReference>& methods) {
  ScopedTrace trace(__FUNCTION__);
  MutexLock mu(Thread::Current(), lock_);
  for (const ProfilingInfo* info : profiling_infos_) {
    ArtMethod* method = info->GetMethod();
    const DexFile* dex_file = method->GetDexFile();
    if (ContainsElement(dex_base_locations, dex_file->GetBaseLocation())) {
      methods.emplace_back(dex_file,  method->GetDexMethodIndex());
    }
  }
}

uint64_t JitCodeCache::GetLastUpdateTimeNs() const {
  return last_update_time_ns_.LoadAcquire();
}

bool JitCodeCache::IsOsrCompiled(ArtMethod* method) {
  MutexLock mu(Thread::Current(), lock_);
  return osr_code_map_.find(method) != osr_code_map_.end();
}

bool JitCodeCache::NotifyCompilationOf(ArtMethod* method, Thread* self, bool osr) {
  if (!osr && ContainsPc(method->GetEntryPointFromQuickCompiledCode())) {
    return false;
  }

  MutexLock mu(self, lock_);
  if (osr && (osr_code_map_.find(method) != osr_code_map_.end())) {
    return false;
  }

  ProfilingInfo* info = method->GetProfilingInfo(kRuntimePointerSize);
  if (info == nullptr) {
    VLOG(jit) << method->PrettyMethod() << " needs a ProfilingInfo to be compiled";
    // Because the counter is not atomic, there are some rare cases where we may not
    // hit the threshold for creating the ProfilingInfo. Reset the counter now to
    // "correct" this.
    method->ClearCounter();
    return false;
  }

  if (info->IsMethodBeingCompiled(osr)) {
    return false;
  }

  info->SetIsMethodBeingCompiled(true, osr);
  return true;
}

ProfilingInfo* JitCodeCache::NotifyCompilerUse(ArtMethod* method, Thread* self) {
  MutexLock mu(self, lock_);
  ProfilingInfo* info = method->GetProfilingInfo(kRuntimePointerSize);
  if (info != nullptr) {
    info->IncrementInlineUse();
  }
  return info;
}

void JitCodeCache::DoneCompilerUse(ArtMethod* method, Thread* self) {
  MutexLock mu(self, lock_);
  ProfilingInfo* info = method->GetProfilingInfo(kRuntimePointerSize);
  DCHECK(info != nullptr);
  info->DecrementInlineUse();
}

void JitCodeCache::DoneCompiling(ArtMethod* method, Thread* self ATTRIBUTE_UNUSED, bool osr) {
  ProfilingInfo* info = method->GetProfilingInfo(kRuntimePointerSize);
  DCHECK(info->IsMethodBeingCompiled(osr));
  info->SetIsMethodBeingCompiled(false, osr);
}

size_t JitCodeCache::GetMemorySizeOfCodePointer(const void* ptr) {
  MutexLock mu(Thread::Current(), lock_);
  return mspace_usable_size(reinterpret_cast<const void*>(FromCodeToAllocation(ptr)));
}

void JitCodeCache::InvalidateCompiledCodeFor(ArtMethod* method,
                                             const OatQuickMethodHeader* header) {
  ProfilingInfo* profiling_info = method->GetProfilingInfo(kRuntimePointerSize);
  if ((profiling_info != nullptr) &&
      (profiling_info->GetSavedEntryPoint() == header->GetEntryPoint())) {
    // Prevent future uses of the compiled code.
    profiling_info->SetSavedEntryPoint(nullptr);
  }

  if (method->GetEntryPointFromQuickCompiledCode() == header->GetEntryPoint()) {
    // The entrypoint is the one to invalidate, so we just update
    // it to the interpreter entry point and clear the counter to get the method
    // Jitted again.
    Runtime::Current()->GetInstrumentation()->UpdateMethodsCode(
        method, GetQuickToInterpreterBridge());
    method->ClearCounter();
  } else {
    MutexLock mu(Thread::Current(), lock_);
    auto it = osr_code_map_.find(method);
    if (it != osr_code_map_.end() && OatQuickMethodHeader::FromCodePointer(it->second) == header) {
      // Remove the OSR method, to avoid using it again.
      osr_code_map_.erase(it);
    }
  }
  MutexLock mu(Thread::Current(), lock_);
  number_of_deoptimizations_++;
}

uint8_t* JitCodeCache::AllocateCode(size_t code_size) {
  size_t alignment = GetInstructionSetAlignment(kRuntimeISA);
  uint8_t* result = reinterpret_cast<uint8_t*>(
      mspace_memalign(code_mspace_, alignment, code_size));
  size_t header_size = RoundUp(sizeof(OatQuickMethodHeader), alignment);
  // Ensure the header ends up at expected instruction alignment.
  DCHECK_ALIGNED_PARAM(reinterpret_cast<uintptr_t>(result + header_size), alignment);
  used_memory_for_code_ += mspace_usable_size(result);
  return result;
}

void JitCodeCache::FreeCode(uint8_t* code) {
  used_memory_for_code_ -= mspace_usable_size(code);
  mspace_free(code_mspace_, code);
}

uint8_t* JitCodeCache::AllocateData(size_t data_size) {
  void* result = mspace_malloc(data_mspace_, data_size);
  used_memory_for_data_ += mspace_usable_size(result);
  return reinterpret_cast<uint8_t*>(result);
}

void JitCodeCache::FreeData(uint8_t* data) {
  used_memory_for_data_ -= mspace_usable_size(data);
  mspace_free(data_mspace_, data);
}

void JitCodeCache::Dump(std::ostream& os) {
  MutexLock mu(Thread::Current(), lock_);
  os << "Current JIT code cache size: " << PrettySize(used_memory_for_code_) << "\n"
     << "Current JIT data cache size: " << PrettySize(used_memory_for_data_) << "\n"
     << "Current JIT capacity: " << PrettySize(current_capacity_) << "\n"
     << "Current number of JIT code cache entries: " << method_code_map_.size() << "\n"
     << "Total number of JIT compilations: " << number_of_compilations_ << "\n"
     << "Total number of JIT compilations for on stack replacement: "
        << number_of_osr_compilations_ << "\n"
     << "Total number of deoptimizations: " << number_of_deoptimizations_ << "\n"
     << "Total number of JIT code cache collections: " << number_of_collections_ << std::endl;
  histogram_stack_map_memory_use_.PrintMemoryUse(os);
  histogram_code_memory_use_.PrintMemoryUse(os);
  histogram_profiling_info_memory_use_.PrintMemoryUse(os);
}

}  // namespace jit
}  // namespace art