Reland: Create libsampler as V8 sampler library.

This patch does five things:

1. Extracts sampler as libsampler to provide sampling functionality support.
2. Makes SampleStack virtual so embedders can override the behaviour of sample collecting.
3. Removes sampler.[h|cc].
4. Moves sampling thread into log.cc as workaround to keep the --prof functionality.
5. Creates SamplerManager to manage the relationship between samplers and threads.

The reason we port hashmap.h is that in debug mode, STL containers are using
mutexes from a mutex pool, which may lead to deadlock when using asynchronously
signal handler.

Currently libsampler is used in V8 temporarily.

BUG=v8:4789
LOG=n

Committed: https://crrev.com/06cc9b7c176a6223971deaa9fbcafe1a05058c7b
Cr-Commit-Position: refs/heads/master@{#36527}

Review-Url: https://codereview.chromium.org/1922303002
Cr-Commit-Position: refs/heads/master@{#36532}

BUILD.gn

@@ -105,6 +105,11 @@ config("libplatform_config") {
   include_dirs = [ "include" ]
 }
 
+# This config should be applied to code using the libsampler.
+config("libsampler_config") {
+  include_dirs = [ "include" ]
+}
+
 # This config should only be applied to code using V8 and not any V8 code
 # itself.
 config("external_config") {
@@ -1329,8 +1334,6 @@ v8_source_set("v8_base") {
     "src/profiler/profile-generator-inl.h",
     "src/profiler/profile-generator.cc",
     "src/profiler/profile-generator.h",
-    "src/profiler/sampler.cc",
-    "src/profiler/sampler.h",
     "src/profiler/sampling-heap-profiler.cc",
     "src/profiler/sampling-heap-profiler.h",
     "src/profiler/strings-storage.cc",
@@ -1816,6 +1819,7 @@ v8_source_set("v8_base") {
   defines = []
   deps = [
     ":v8_libbase",
+    ":v8_libsampler",
   ]
 
   if (is_win) {
@@ -1970,6 +1974,23 @@ v8_source_set("v8_libplatform") {
   ]
 }
 
+v8_source_set("v8_libsampler") {
+  sources = [
+    "src/libsampler/hashmap.h",
+    "src/libsampler/utils.h",
+    "src/libsampler/v8-sampler.cc",
+    "src/libsampler/v8-sampler.h",
+  ]
+
+  configs = [ ":internal_config_base" ]
+
+  public_configs = [ ":libsampler_config" ]
+
+  deps = [
+    ":v8_libbase",
+  ]
+}
+
 v8_source_set("fuzzer_support") {
   visibility = [ ":*" ]  # Only targets in this file can depend on this.
 

include/v8.h

@@ -6331,6 +6331,12 @@ class V8_EXPORT Isolate {
    */
   void VisitWeakHandles(PersistentHandleVisitor* visitor);
 
+  /**
+   * Check if this isolate is in use.
+   * True if at least one thread Enter'ed this isolate.
+   */
+  bool IsInUse();
+
  private:
   template <class K, class V, class Traits>
   friend class PersistentValueMapBase;

src/api.cc

@@ -7793,6 +7793,12 @@ void Isolate::VisitExternalResources(ExternalResourceVisitor* visitor) {
 }
 
 
+bool Isolate::IsInUse() {
+  i::Isolate* isolate = reinterpret_cast<i::Isolate*>(this);
+  return isolate->IsInUse();
+}
+
+
 class VisitorAdapter : public i::ObjectVisitor {
  public:
   explicit VisitorAdapter(PersistentHandleVisitor* visitor)

src/isolate.cc

@@ -27,10 +27,10 @@
 #include "src/ic/stub-cache.h"
 #include "src/interpreter/interpreter.h"
 #include "src/isolate-inl.h"
+#include "src/libsampler/v8-sampler.h"
 #include "src/log.h"
 #include "src/messages.h"
 #include "src/profiler/cpu-profiler.h"
-#include "src/profiler/sampler.h"
 #include "src/prototype.h"
 #include "src/regexp/regexp-stack.h"
 #include "src/runtime-profiler.h"
@@ -2006,7 +2006,7 @@ void Isolate::Deinit() {
   }
 
   // We must stop the logger before we tear down other components.
-  Sampler* sampler = logger_->sampler();
+  sampler::Sampler* sampler = logger_->sampler();
   if (sampler && sampler->IsActive()) sampler->Stop();
 
   delete deoptimizer_data_;

src/libsampler/DEPS

+include_rules = [
+  "+include",
+  "-src",
+  "+src/base",
+  "+src/libsampler",
+]
\ No newline at end of file

src/libsampler/hashmap.h

+// Copyright 2012 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+// This file is ported from src/hashmap.h
+
+#ifndef V8_LIBSAMPLER_HASHMAP_H_
+#define V8_LIBSAMPLER_HASHMAP_H_
+
+#include "src/base/bits.h"
+#include "src/base/logging.h"
+#include "src/libsampler/utils.h"
+
+namespace v8 {
+namespace sampler {
+
+class HashMapImpl {
+ public:
+  typedef bool (*MatchFun) (void* key1, void* key2);
+
+  // The default capacity.
+  static const uint32_t kDefaultHashMapCapacity = 8;
+
+  // initial_capacity is the size of the initial hash map;
+  // it must be a power of 2 (and thus must not be 0).
+  HashMapImpl(MatchFun match,
+              uint32_t capacity = kDefaultHashMapCapacity);
+
+  ~HashMapImpl();
+
+  // HashMap entries are (key, value, hash) triplets.
+  // Some clients may not need to use the value slot
+  // (e.g. implementers of sets, where the key is the value).
+  struct Entry {
+    void* key;
+    void* value;
+    uint32_t hash;  // The full hash value for key
+    int order;  // If you never remove entries this is the insertion order.
+  };
+
+  // If an entry with matching key is found, returns that entry.
+  // Otherwise, NULL is returned.
+  Entry* Lookup(void* key, uint32_t hash) const;
+
+  // If an entry with matching key is found, returns that entry.
+  // If no matching entry is found, a new entry is inserted with
+  // corresponding key, key hash, and NULL value.
+  Entry* LookupOrInsert(void* key, uint32_t hash);
+
+  // Removes the entry with matching key.
+  // It returns the value of the deleted entry
+  // or null if there is no value for such key.
+  void* Remove(void* key, uint32_t hash);
+
+  // Empties the hash map (occupancy() == 0).
+  void Clear();
+
+  // The number of (non-empty) entries in the table.
+  uint32_t occupancy() const { return occupancy_; }
+
+  // The capacity of the table. The implementation
+  // makes sure that occupancy is at most 80% of
+  // the table capacity.
+  uint32_t capacity() const { return capacity_; }
+
+  // Iteration
+  //
+  // for (Entry* p = map.Start(); p != NULL; p = map.Next(p)) {
+  //   ...
+  // }
+  //
+  // If entries are inserted during iteration, the effect of
+  // calling Next() is undefined.
+  Entry* Start() const;
+  Entry* Next(Entry* p) const;
+
+  // Some match functions defined for convenience.
+  static bool PointersMatch(void* key1, void* key2) {
+    return key1 == key2;
+  }
+
+ private:
+  MatchFun match_;
+  Entry* map_;
+  uint32_t capacity_;
+  uint32_t occupancy_;
+
+  Entry* map_end() const { return map_ + capacity_; }
+  Entry* Probe(void* key, uint32_t hash) const;
+  void Initialize(uint32_t capacity);
+  void Resize();
+};
+
+typedef HashMapImpl HashMap;
+
+HashMapImpl::HashMapImpl(MatchFun match, uint32_t initial_capacity) {
+  match_ = match;
+  Initialize(initial_capacity);
+}
+
+
+HashMapImpl::~HashMapImpl() {
+  Malloced::Delete(map_);
+}
+
+
+HashMapImpl::Entry* HashMapImpl::Lookup(void* key, uint32_t hash) const {
+  Entry* p = Probe(key, hash);
+  return p->key != NULL ? p : NULL;
+}
+
+
+HashMapImpl::Entry* HashMapImpl::LookupOrInsert(void* key, uint32_t hash) {
+  // Find a matching entry.
+  Entry* p = Probe(key, hash);
+  if (p->key != NULL) {
+    return p;
+  }
+
+  // No entry found; insert one.
+  p->key = key;
+  p->value = NULL;
+  p->hash = hash;
+  p->order = occupancy_;
+  occupancy_++;
+
+  // Grow the map if we reached >= 80% occupancy.
+  if (occupancy_ + occupancy_ / 4 >= capacity_) {
+    Resize();
+    p = Probe(key, hash);
+  }
+
+  return p;
+}
+
+
+void* HashMapImpl::Remove(void* key, uint32_t hash) {
+  // Lookup the entry for the key to remove.
+  Entry* p = Probe(key, hash);
+  if (p->key == NULL) {
+    // Key not found nothing to remove.
+    return NULL;
+  }
+
+  void* value = p->value;
+  // To remove an entry we need to ensure that it does not create an empty
+  // entry that will cause the search for another entry to stop too soon. If all
+  // the entries between the entry to remove and the next empty slot have their
+  // initial position inside this interval, clearing the entry to remove will
+  // not break the search. If, while searching for the next empty entry, an
+  // entry is encountered which does not have its initial position between the
+  // entry to remove and the position looked at, then this entry can be moved to
+  // the place of the entry to remove without breaking the search for it. The
+  // entry made vacant by this move is now the entry to remove and the process
+  // starts over.
+  // Algorithm from http://en.wikipedia.org/wiki/Open_addressing.
+
+  // This guarantees loop termination as there is at least one empty entry so
+  // eventually the removed entry will have an empty entry after it.
+  DCHECK(occupancy_ < capacity_);
+
+  // p is the candidate entry to clear. q is used to scan forwards.
+  Entry* q = p;  // Start at the entry to remove.
+  while (true) {
+    // Move q to the next entry.
+    q = q + 1;
+    if (q == map_end()) {
+      q = map_;
+    }
+
+    // All entries between p and q have their initial position between p and q
+    // and the entry p can be cleared without breaking the search for these
+    // entries.
+    if (q->key == NULL) {
+      break;
+    }
+
+    // Find the initial position for the entry at position q.
+    Entry* r = map_ + (q->hash & (capacity_ - 1));
+
+    // If the entry at position q has its initial position outside the range
+    // between p and q it can be moved forward to position p and will still be
+    // found. There is now a new candidate entry for clearing.
+    if ((q > p && (r <= p || r > q)) ||
+        (q < p && (r <= p && r > q))) {
+      *p = *q;
+      p = q;
+    }
+  }
+
+  // Clear the entry which is allowed to en emptied.
+  p->key = NULL;
+  occupancy_--;
+  return value;
+}
+
+
+void HashMapImpl::Clear() {
+  // Mark all entries as empty.
+  const Entry* end = map_end();
+  for (Entry* p = map_; p < end; p++) {
+    p->key = NULL;
+  }
+  occupancy_ = 0;
+}
+
+
+HashMapImpl::Entry* HashMapImpl::Start() const {
+  return Next(map_ - 1);
+}
+
+
+HashMapImpl::Entry* HashMapImpl::Next(Entry* p) const {
+  const Entry* end = map_end();
+  DCHECK(map_ - 1 <= p && p < end);
+  for (p++; p < end; p++) {
+    if (p->key != NULL) {
+      return p;
+    }
+  }
+  return NULL;
+}
+
+
+HashMapImpl::Entry* HashMapImpl::Probe(void* key, uint32_t hash) const {
+  DCHECK(key != NULL);
+
+  DCHECK(base::bits::IsPowerOfTwo32(capacity_));
+  Entry* p = map_ + (hash & (capacity_ - 1));
+  const Entry* end = map_end();
+  DCHECK(map_ <= p && p < end);
+
+  DCHECK(occupancy_ < capacity_);  // Guarantees loop termination.
+  while (p->key != NULL && (hash != p->hash || !match_(key, p->key))) {
+    p++;
+    if (p >= end) {
+      p = map_;
+    }
+  }
+
+  return p;
+}
+
+
+void HashMapImpl::Initialize(uint32_t capacity) {
+  DCHECK(base::bits::IsPowerOfTwo32(capacity));
+  map_ = reinterpret_cast<Entry*>(Malloced::New(capacity * sizeof(Entry)));
+  CHECK(map_ != NULL);
+  capacity_ = capacity;
+  Clear();
+}
+
+
+void HashMapImpl::Resize() {
+  Entry* map = map_;
+  uint32_t n = occupancy_;
+
+  // Allocate larger map.
+  Initialize(capacity_ * 2);
+
+  // Rehash all current entries.
+  for (Entry* p = map; n > 0; p++) {
+    if (p->key != NULL) {
+      Entry* entry = LookupOrInsert(p->key, p->hash);
+      entry->value = p->value;
+      entry->order = p->order;
+      n--;
+    }
+  }
+
+  // Delete old map.
+  Malloced::Delete(map);
+}
+
+}  // namespace sampler
+}  // namespace v8
+
+#endif  // V8_LIBSAMPLER_HASHMAP_H_

src/libsampler/utils.h

+// Copyright 2016 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef V8_LIBSAMPLER_UTILS_H_
+#define V8_LIBSAMPLER_UTILS_H_
+
+#include "include/v8.h"
+
+namespace v8 {
+namespace sampler {
+
+class Malloced {
+ public:
+  static void* New(size_t size) {
+    return malloc(size);
+  }
+
+  static void Delete(void* p) {
+    free(p);
+  }
+};
+
+}  // namespace sampler
+}  // namespace v8
+
+#endif  // V8_LIBSAMPLER_UTILS_H_

src/profiler/sampler.cc → src/libsampler/v8-sampler.cc

-// Copyright 2013 the V8 project authors. All rights reserved.
+// Copyright 2016 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
-#include "src/profiler/sampler.h"
+#include "src/libsampler/v8-sampler.h"
 
 #if V8_OS_POSIX && !V8_OS_CYGWIN
 
@@ -42,12 +42,13 @@
 
 #endif
 
+#include <algorithm>
+#include <vector>
+#include <map>
+
 #include "src/base/atomic-utils.h"
 #include "src/base/platform/platform.h"
-#include "src/profiler/cpu-profiler-inl.h"
-#include "src/profiler/tick-sample.h"
-#include "src/simulator.h"
-#include "src/v8threads.h"
+#include "src/libsampler/hashmap.h"
 
 
 #if V8_OS_ANDROID && !defined(__BIONIC_HAVE_UCONTEXT_T)
@@ -155,26 +156,14 @@ enum { REG_RBP = 10, REG_RSP = 15, REG_RIP = 16 };
 
 
 namespace v8 {
-namespace internal {
+namespace sampler {
 
 namespace {
 
-class PlatformDataCommon : public Malloced {
- public:
-  PlatformDataCommon() : profiled_thread_id_(ThreadId::Current()) {}
-  ThreadId profiled_thread_id() { return profiled_thread_id_; }
-
- protected:
-  ~PlatformDataCommon() {}
-
- private:
-  ThreadId profiled_thread_id_;
-};
-
-
-typedef List<Sampler*> SamplerList;
-
 #if defined(USE_SIGNALS)
+typedef std::vector<Sampler*> SamplerList;
+typedef SamplerList::iterator SamplerListIterator;
+
 class AtomicGuard {
  public:
   explicit AtomicGuard(base::AtomicValue<int>* atomic, bool is_block = true)
@@ -216,13 +205,14 @@ uint32_t ThreadHash(pthread_t thread_id) {
   return static_cast<uint32_t>(thread_id);
 #endif
 }
+
 #endif  // USE_SIGNALS
 
 }  // namespace
 
 #if defined(USE_SIGNALS)
 
-class Sampler::PlatformData : public PlatformDataCommon {
+class Sampler::PlatformData {
  public:
   PlatformData() : vm_tid_(pthread_self()) {}
   pthread_t vm_tid() const { return vm_tid_; }
@@ -231,13 +221,85 @@ class Sampler::PlatformData : public PlatformDataCommon {
   pthread_t vm_tid_;
 };
 
+
+class SamplerManager {
+ public:
+  static void AddSampler(Sampler* sampler) {
+    AtomicGuard atomic_guard(&samplers_access_counter_);
+    DCHECK(sampler->IsActive() || !sampler->IsRegistered());
+    // Add sampler into map if needed.
+    pthread_t thread_id = sampler->platform_data()->vm_tid();
+    HashMap::Entry* entry =
+        sampler_map_.Pointer()->LookupOrInsert(ThreadKey(thread_id),
+                                               ThreadHash(thread_id));
+    DCHECK(entry != NULL);
+    if (entry->value == NULL) {
+      SamplerList* samplers = new SamplerList();
+      samplers->push_back(sampler);
+      entry->value = samplers;
+    } else {
+      SamplerList* samplers = reinterpret_cast<SamplerList*>(entry->value);
+      bool exists = false;
+      for (SamplerListIterator iter = samplers->begin();
+           iter != samplers->end(); ++iter) {
+        if (*iter == sampler) {
+          exists = true;
+          break;
+        }
+      }
+      if (!exists) {
+        samplers->push_back(sampler);
+      }
+    }
+  }
+
+  static void RemoveSampler(Sampler* sampler) {
+    AtomicGuard atomic_guard(&samplers_access_counter_);
+    DCHECK(sampler->IsActive() || sampler->IsRegistered());
+    // Remove sampler from map.
+    pthread_t thread_id = sampler->platform_data()->vm_tid();
+    void* thread_key = ThreadKey(thread_id);
+    uint32_t thread_hash = ThreadHash(thread_id);
+    HashMap::Entry* entry = sampler_map_.Get().Lookup(thread_key, thread_hash);
+    DCHECK(entry != NULL);
+    SamplerList* samplers = reinterpret_cast<SamplerList*>(entry->value);
+    for (SamplerListIterator iter = samplers->begin(); iter != samplers->end();
+         ++iter) {
+      if (*iter == sampler) {
+        samplers->erase(iter);
+        break;
+      }
+    }
+    if (samplers->empty()) {
+      sampler_map_.Pointer()->Remove(thread_key, thread_hash);
+      delete samplers;
+    }
+  }
+
+ private:
+  struct HashMapCreateTrait {
+    static void Construct(HashMap* allocated_ptr) {
+      new (allocated_ptr) HashMap(HashMap::PointersMatch);
+    }
+  };
+  friend class SignalHandler;
+  static base::LazyInstance<HashMap, HashMapCreateTrait>::type
+      sampler_map_;
+  static base::AtomicValue<int> samplers_access_counter_;
+};
+
+base::LazyInstance<HashMap, SamplerManager::HashMapCreateTrait>::type
+    SamplerManager::sampler_map_ = LAZY_INSTANCE_INITIALIZER;
+base::AtomicValue<int> SamplerManager::samplers_access_counter_(0);
+
+
 #elif V8_OS_WIN || V8_OS_CYGWIN
 
 // ----------------------------------------------------------------------------
 // Win32 profiler support. On Cygwin we use the same sampler implementation as
 // on Win32.
 
-class Sampler::PlatformData : public PlatformDataCommon {
+class Sampler::PlatformData {
  public:
   // Get a handle to the calling thread. This is the thread that we are
   // going to profile. We need to make a copy of the handle because we are
@@ -263,12 +325,11 @@ class Sampler::PlatformData : public PlatformDataCommon {
  private:
   HANDLE profiled_thread_;
 };
-#endif
+#endif  // USE_SIGNALS
 
 
 #if defined(USE_SIGNALS)
-
-class SignalHandler : public AllStatic {
+class SignalHandler {
  public:
   static void SetUp() { if (!mutex_) mutex_ = new base::Mutex(); }
   static void TearDown() { delete mutex_; mutex_ = NULL; }
@@ -284,13 +345,10 @@ class SignalHandler : public AllStatic {
   }
 
   static bool Installed() {
+    base::LockGuard<base::Mutex> lock_guard(mutex_);
     return signal_handler_installed_;
   }
 
-#if !V8_OS_NACL
-  static void CollectSample(void* context, Sampler* sampler);
-#endif
-
  private:
   static void Install() {
 #if !V8_OS_NACL
@@ -304,7 +362,7 @@ class SignalHandler : public AllStatic {
 #endif
     signal_handler_installed_ =
         (sigaction(SIGPROF, &sa, &old_signal_handler_) == 0);
-#endif
+#endif  // !V8_OS_NACL
   }
 
   static void Restore() {
@@ -317,6 +375,7 @@ class SignalHandler : public AllStatic {
   }
 
 #if !V8_OS_NACL
+  static void FillRegisterState(void* context, RegisterState* regs);
   static void HandleProfilerSignal(int signal, siginfo_t* info, void* context);
 #endif
   // Protects the process wide state below.
@@ -335,26 +394,36 @@ bool SignalHandler::signal_handler_installed_ = false;
 
 // As Native Client does not support signal handling, profiling is disabled.
 #if !V8_OS_NACL
-void SignalHandler::CollectSample(void* context, Sampler* sampler) {
-  if (sampler == NULL || (!sampler->IsProfiling() &&
-                          !sampler->IsRegistered())) {
-    return;
-  }
-  Isolate* isolate = sampler->isolate();
+void SignalHandler::HandleProfilerSignal(int signal, siginfo_t* info,
+                                         void* context) {
+  USE(info);
+  if (signal != SIGPROF) return;
+  AtomicGuard atomic_guard(&SamplerManager::samplers_access_counter_, false);
+  if (!atomic_guard.is_success()) return;
+  pthread_t thread_id = pthread_self();
+  HashMap::Entry* entry =
+      SamplerManager::sampler_map_.Pointer()->Lookup(ThreadKey(thread_id),
+                                                     ThreadHash(thread_id));
+  if (entry == NULL) return;
+  SamplerList* samplers = reinterpret_cast<SamplerList*>(entry->value);
 
-  // We require a fully initialized and entered isolate.
-  if (isolate == NULL || !isolate->IsInUse()) return;
+  v8::RegisterState state;
+  FillRegisterState(context, &state);
 
-  if (v8::Locker::IsActive() &&
-      !isolate->thread_manager()->IsLockedByCurrentThread()) {
-    return;
-  }
+  for (int i = 0; i < samplers->size(); ++i) {
+    Sampler* sampler = (*samplers)[i];
+    Isolate* isolate = sampler->isolate();
 
-  v8::RegisterState state;
+    // We require a fully initialized and entered isolate.
+    if (isolate == NULL || !isolate->IsInUse()) return;
 
-#if defined(USE_SIMULATOR)
-  if (!SimulatorHelper::FillRegisters(isolate, &state)) return;
-#else
+    if (v8::Locker::IsActive() && !Locker::IsLocked(isolate)) return;
+
+    sampler->SampleStack(state);
+  }
+}
+
+void SignalHandler::FillRegisterState(void* context, RegisterState* state) {
   // Extracting the sample from the context is extremely machine dependent.
   ucontext_t* ucontext = reinterpret_cast<ucontext_t*>(context);
 #if !(V8_OS_OPENBSD || (V8_OS_LINUX && (V8_HOST_ARCH_PPC || V8_HOST_ARCH_S390)))
@@ -362,381 +431,187 @@ void SignalHandler::CollectSample(void* context, Sampler* sampler) {
 #endif
 #if V8_OS_LINUX
 #if V8_HOST_ARCH_IA32
-  state.pc = reinterpret_cast<Address>(mcontext.gregs[REG_EIP]);
-  state.sp = reinterpret_cast<Address>(mcontext.gregs[REG_ESP]);
-  state.fp = reinterpret_cast<Address>(mcontext.gregs[REG_EBP]);
+  state->pc = reinterpret_cast<void*>(mcontext.gregs[REG_EIP]);
+  state->sp = reinterpret_cast<void*>(mcontext.gregs[REG_ESP]);
+  state->fp = reinterpret_cast<void*>(mcontext.gregs[REG_EBP]);
 #elif V8_HOST_ARCH_X64
-  state.pc = reinterpret_cast<Address>(mcontext.gregs[REG_RIP]);
-  state.sp = reinterpret_cast<Address>(mcontext.gregs[REG_RSP]);
-  state.fp = reinterpret_cast<Address>(mcontext.gregs[REG_RBP]);
+  state->pc = reinterpret_cast<void*>(mcontext.gregs[REG_RIP]);
+  state->sp = reinterpret_cast<void*>(mcontext.gregs[REG_RSP]);
+  state->fp = reinterpret_cast<void*>(mcontext.gregs[REG_RBP]);
 #elif V8_HOST_ARCH_ARM
 #if V8_LIBC_GLIBC && !V8_GLIBC_PREREQ(2, 4)
   // Old GLibc ARM versions used a gregs[] array to access the register
   // values from mcontext_t.
-  state.pc = reinterpret_cast<Address>(mcontext.gregs[R15]);
-  state.sp = reinterpret_cast<Address>(mcontext.gregs[R13]);
-  state.fp = reinterpret_cast<Address>(mcontext.gregs[R11]);
+  state->pc = reinterpret_cast<void*>(mcontext.gregs[R15]);
+  state->sp = reinterpret_cast<void*>(mcontext.gregs[R13]);
+  state->fp = reinterpret_cast<void*>(mcontext.gregs[R11]);
 #else
-  state.pc = reinterpret_cast<Address>(mcontext.arm_pc);
-  state.sp = reinterpret_cast<Address>(mcontext.arm_sp);
-  state.fp = reinterpret_cast<Address>(mcontext.arm_fp);
+  state->pc = reinterpret_cast<void*>(mcontext.arm_pc);
+  state->sp = reinterpret_cast<void*>(mcontext.arm_sp);
+  state->fp = reinterpret_cast<void*>(mcontext.arm_fp);
 #endif  // V8_LIBC_GLIBC && !V8_GLIBC_PREREQ(2, 4)
 #elif V8_HOST_ARCH_ARM64
-  state.pc = reinterpret_cast<Address>(mcontext.pc);
-  state.sp = reinterpret_cast<Address>(mcontext.sp);
+  state->pc = reinterpret_cast<void*>(mcontext.pc);
+  state->sp = reinterpret_cast<void*>(mcontext.sp);
   // FP is an alias for x29.
-  state.fp = reinterpret_cast<Address>(mcontext.regs[29]);
+  state->fp = reinterpret_cast<void*>(mcontext.regs[29]);
 #elif V8_HOST_ARCH_MIPS
-  state.pc = reinterpret_cast<Address>(mcontext.pc);
-  state.sp = reinterpret_cast<Address>(mcontext.gregs[29]);
-  state.fp = reinterpret_cast<Address>(mcontext.gregs[30]);
+  state->pc = reinterpret_cast<void*>(mcontext.pc);
+  state->sp = reinterpret_cast<void*>(mcontext.gregs[29]);
+  state->fp = reinterpret_cast<void*>(mcontext.gregs[30]);
 #elif V8_HOST_ARCH_MIPS64
-  state.pc = reinterpret_cast<Address>(mcontext.pc);
-  state.sp = reinterpret_cast<Address>(mcontext.gregs[29]);
-  state.fp = reinterpret_cast<Address>(mcontext.gregs[30]);
+  state->pc = reinterpret_cast<void*>(mcontext.pc);
+  state->sp = reinterpret_cast<void*>(mcontext.gregs[29]);
+  state->fp = reinterpret_cast<void*>(mcontext.gregs[30]);
 #elif V8_HOST_ARCH_PPC
-  state.pc = reinterpret_cast<Address>(ucontext->uc_mcontext.regs->nip);
-  state.sp = reinterpret_cast<Address>(ucontext->uc_mcontext.regs->gpr[PT_R1]);
-  state.fp = reinterpret_cast<Address>(ucontext->uc_mcontext.regs->gpr[PT_R31]);
+  state->pc = reinterpret_cast<void*>(ucontext->uc_mcontext.regs->nip);
+  state->sp =
+      reinterpret_cast<void*>(ucontext->uc_mcontext.regs->gpr[PT_R1]);
+  state->fp =
+      reinterpret_cast<void*>(ucontext->uc_mcontext.regs->gpr[PT_R31]);
 #elif V8_HOST_ARCH_S390
 #if V8_TARGET_ARCH_32_BIT
   // 31-bit target will have bit 0 (MSB) of the PSW set to denote addressing
   // mode.  This bit needs to be masked out to resolve actual address.
-  state.pc =
-      reinterpret_cast<Address>(ucontext->uc_mcontext.psw.addr & 0x7FFFFFFF);
+  state->pc =
+      reinterpret_cast<void*>(ucontext->uc_mcontext.psw.addr & 0x7FFFFFFF);
 #else
-  state.pc = reinterpret_cast<Address>(ucontext->uc_mcontext.psw.addr);
+  state->pc = reinterpret_cast<void*>(ucontext->uc_mcontext.psw.addr);
 #endif  // V8_TARGET_ARCH_32_BIT
-  state.sp = reinterpret_cast<Address>(ucontext->uc_mcontext.gregs[15]);
-  state.fp = reinterpret_cast<Address>(ucontext->uc_mcontext.gregs[11]);
+  state->sp = reinterpret_cast<void*>(ucontext->uc_mcontext.gregs[15]);
+  state->fp = reinterpret_cast<void*>(ucontext->uc_mcontext.gregs[11]);
 #endif  // V8_HOST_ARCH_*
 #elif V8_OS_MACOSX
 #if V8_HOST_ARCH_X64
 #if __DARWIN_UNIX03
-  state.pc = reinterpret_cast<Address>(mcontext->__ss.__rip);
-  state.sp = reinterpret_cast<Address>(mcontext->__ss.__rsp);
-  state.fp = reinterpret_cast<Address>(mcontext->__ss.__rbp);
+  state->pc = reinterpret_cast<void*>(mcontext->__ss.__rip);
+  state->sp = reinterpret_cast<void*>(mcontext->__ss.__rsp);
+  state->fp = reinterpret_cast<void*>(mcontext->__ss.__rbp);
 #else  // !__DARWIN_UNIX03
-  state.pc = reinterpret_cast<Address>(mcontext->ss.rip);
-  state.sp = reinterpret_cast<Address>(mcontext->ss.rsp);
-  state.fp = reinterpret_cast<Address>(mcontext->ss.rbp);
+  state->pc = reinterpret_cast<void*>(mcontext->ss.rip);
+  state->sp = reinterpret_cast<void*>(mcontext->ss.rsp);
+  state->fp = reinterpret_cast<void*>(mcontext->ss.rbp);
 #endif  // __DARWIN_UNIX03
 #elif V8_HOST_ARCH_IA32
 #if __DARWIN_UNIX03
-  state.pc = reinterpret_cast<Address>(mcontext->__ss.__eip);
-  state.sp = reinterpret_cast<Address>(mcontext->__ss.__esp);
-  state.fp = reinterpret_cast<Address>(mcontext->__ss.__ebp);
+  state->pc = reinterpret_cast<void*>(mcontext->__ss.__eip);
+  state->sp = reinterpret_cast<void*>(mcontext->__ss.__esp);
+  state->fp = reinterpret_cast<void*>(mcontext->__ss.__ebp);
 #else  // !__DARWIN_UNIX03
-  state.pc = reinterpret_cast<Address>(mcontext->ss.eip);
-  state.sp = reinterpret_cast<Address>(mcontext->ss.esp);
-  state.fp = reinterpret_cast<Address>(mcontext->ss.ebp);
+  state->pc = reinterpret_cast<void*>(mcontext->ss.eip);
+  state->sp = reinterpret_cast<void*>(mcontext->ss.esp);
+  state->fp = reinterpret_cast<void*>(mcontext->ss.ebp);
 #endif  // __DARWIN_UNIX03
 #endif  // V8_HOST_ARCH_IA32
 #elif V8_OS_FREEBSD
 #if V8_HOST_ARCH_IA32
-  state.pc = reinterpret_cast<Address>(mcontext.mc_eip);
-  state.sp = reinterpret_cast<Address>(mcontext.mc_esp);
-  state.fp = reinterpret_cast<Address>(mcontext.mc_ebp);
+  state->pc = reinterpret_cast<void*>(mcontext.mc_eip);
+  state->sp = reinterpret_cast<void*>(mcontext.mc_esp);
+  state->fp = reinterpret_cast<void*>(mcontext.mc_ebp);
 #elif V8_HOST_ARCH_X64
-  state.pc = reinterpret_cast<Address>(mcontext.mc_rip);
-  state.sp = reinterpret_cast<Address>(mcontext.mc_rsp);
-  state.fp = reinterpret_cast<Address>(mcontext.mc_rbp);
+  state->pc = reinterpret_cast<void*>(mcontext.mc_rip);
+  state->sp = reinterpret_cast<void*>(mcontext.mc_rsp);
+  state->fp = reinterpret_cast<void*>(mcontext.mc_rbp);
 #elif V8_HOST_ARCH_ARM
-  state.pc = reinterpret_cast<Address>(mcontext.mc_r15);
-  state.sp = reinterpret_cast<Address>(mcontext.mc_r13);
-  state.fp = reinterpret_cast<Address>(mcontext.mc_r11);
+  state->pc = reinterpret_cast<void*>(mcontext.mc_r15);
+  state->sp = reinterpret_cast<void*>(mcontext.mc_r13);
+  state->fp = reinterpret_cast<void*>(mcontext.mc_r11);
 #endif  // V8_HOST_ARCH_*
 #elif V8_OS_NETBSD
 #if V8_HOST_ARCH_IA32
-  state.pc = reinterpret_cast<Address>(mcontext.__gregs[_REG_EIP]);
-  state.sp = reinterpret_cast<Address>(mcontext.__gregs[_REG_ESP]);
-  state.fp = reinterpret_cast<Address>(mcontext.__gregs[_REG_EBP]);
+  state->pc = reinterpret_cast<void*>(mcontext.__gregs[_REG_EIP]);
+  state->sp = reinterpret_cast<void*>(mcontext.__gregs[_REG_ESP]);
+  state->fp = reinterpret_cast<void*>(mcontext.__gregs[_REG_EBP]);
 #elif V8_HOST_ARCH_X64
-  state.pc = reinterpret_cast<Address>(mcontext.__gregs[_REG_RIP]);
-  state.sp = reinterpret_cast<Address>(mcontext.__gregs[_REG_RSP]);
-  state.fp = reinterpret_cast<Address>(mcontext.__gregs[_REG_RBP]);
+  state->pc = reinterpret_cast<void*>(mcontext.__gregs[_REG_RIP]);
+  state->sp = reinterpret_cast<void*>(mcontext.__gregs[_REG_RSP]);
+  state->fp = reinterpret_cast<void*>(mcontext.__gregs[_REG_RBP]);
 #endif  // V8_HOST_ARCH_*
 #elif V8_OS_OPENBSD
 #if V8_HOST_ARCH_IA32
-  state.pc = reinterpret_cast<Address>(ucontext->sc_eip);
-  state.sp = reinterpret_cast<Address>(ucontext->sc_esp);
-  state.fp = reinterpret_cast<Address>(ucontext->sc_ebp);
+  state->pc = reinterpret_cast<void*>(ucontext->sc_eip);
+  state->sp = reinterpret_cast<void*>(ucontext->sc_esp);
+  state->fp = reinterpret_cast<void*>(ucontext->sc_ebp);
 #elif V8_HOST_ARCH_X64
-  state.pc = reinterpret_cast<Address>(ucontext->sc_rip);
-  state.sp = reinterpret_cast<Address>(ucontext->sc_rsp);
-  state.fp = reinterpret_cast<Address>(ucontext->sc_rbp);
+  state->pc = reinterpret_cast<void*>(ucontext->sc_rip);
+  state->sp = reinterpret_cast<void*>(ucontext->sc_rsp);
+  state->fp = reinterpret_cast<void*>(ucontext->sc_rbp);
 #endif  // V8_HOST_ARCH_*
 #elif V8_OS_SOLARIS
-  state.pc = reinterpret_cast<Address>(mcontext.gregs[REG_PC]);
-  state.sp = reinterpret_cast<Address>(mcontext.gregs[REG_SP]);
-  state.fp = reinterpret_cast<Address>(mcontext.gregs[REG_FP]);
+  state->pc = reinterpret_cast<void*>(mcontext.gregs[REG_PC]);
+  state->sp = reinterpret_cast<void*>(mcontext.gregs[REG_SP]);
+  state->fp = reinterpret_cast<void*>(mcontext.gregs[REG_FP]);
 #elif V8_OS_QNX
 #if V8_HOST_ARCH_IA32
-  state.pc = reinterpret_cast<Address>(mcontext.cpu.eip);
-  state.sp = reinterpret_cast<Address>(mcontext.cpu.esp);
-  state.fp = reinterpret_cast<Address>(mcontext.cpu.ebp);
+  state->pc = reinterpret_cast<void*>(mcontext.cpu.eip);
+  state->sp = reinterpret_cast<void*>(mcontext.cpu.esp);
+  state->fp = reinterpret_cast<void*>(mcontext.cpu.ebp);
 #elif V8_HOST_ARCH_ARM
-  state.pc = reinterpret_cast<Address>(mcontext.cpu.gpr[ARM_REG_PC]);
-  state.sp = reinterpret_cast<Address>(mcontext.cpu.gpr[ARM_REG_SP]);
-  state.fp = reinterpret_cast<Address>(mcontext.cpu.gpr[ARM_REG_FP]);
+  state->pc = reinterpret_cast<void*>(mcontext.cpu.gpr[ARM_REG_PC]);
+  state->sp = reinterpret_cast<void*>(mcontext.cpu.gpr[ARM_REG_SP]);
+  state->fp = reinterpret_cast<void*>(mcontext.cpu.gpr[ARM_REG_FP]);
 #endif  // V8_HOST_ARCH_*
 #elif V8_OS_AIX
-  state.pc = reinterpret_cast<Address>(mcontext.jmp_context.iar);
-  state.sp = reinterpret_cast<Address>(mcontext.jmp_context.gpr[1]);
-  state.fp = reinterpret_cast<Address>(mcontext.jmp_context.gpr[31]);
+  state->pc = reinterpret_cast<void*>(mcontext.jmp_context.iar);
+  state->sp = reinterpret_cast<void*>(mcontext.jmp_context.gpr[1]);
+  state->fp = reinterpret_cast<void*>(mcontext.jmp_context.gpr[31]);
 #endif  // V8_OS_AIX
-#endif  // USE_SIMULATOR
-  sampler->SampleStack(state);
 }
-#endif  // V8_OS_NACL
-
-#endif  // USE_SIGNALS
-
-
-class SamplerThread : public base::Thread {
- public:
-  static const int kSamplerThreadStackSize = 64 * KB;
-
-  explicit SamplerThread(int interval)
-      : Thread(base::Thread::Options("SamplerThread", kSamplerThreadStackSize)),
-        interval_(interval) {}
-
-  static void SetUp() { if (!mutex_) mutex_ = new base::Mutex(); }
-  static void TearDown() { delete mutex_; mutex_ = NULL; }
-
-  static void AddActiveSampler(Sampler* sampler) {
-    bool need_to_start = false;
-    base::LockGuard<base::Mutex> lock_guard(mutex_);
-    if (instance_ == NULL) {
-      // Start a thread that will send SIGPROF signal to VM threads,
-      // when CPU profiling will be enabled.
-      instance_ = new SamplerThread(sampler->interval());
-      need_to_start = true;
-    }
-
-    DCHECK(sampler->IsActive());
-    DCHECK(instance_->interval_ == sampler->interval());
-
-#if defined(USE_SIGNALS)
-    AddSampler(sampler);
-#else
-    DCHECK(!instance_->active_samplers_.Contains(sampler));
-    instance_->active_samplers_.Add(sampler);
-#endif  // USE_SIGNALS
-
-    if (need_to_start) instance_->StartSynchronously();
-  }
-
-  static void RemoveSampler(Sampler* sampler) {
-    SamplerThread* instance_to_remove = NULL;
-    {
-      base::LockGuard<base::Mutex> lock_guard(mutex_);
-
-      DCHECK(sampler->IsActive() || sampler->IsRegistered());
-#if defined(USE_SIGNALS)
-      {
-        AtomicGuard atomic_guard(&sampler_list_access_counter_);
-        // Remove sampler from map.
-        pthread_t thread_id = sampler->platform_data()->vm_tid();
-        void* thread_key = ThreadKey(thread_id);
-        uint32_t thread_hash = ThreadHash(thread_id);
-        HashMap::Entry* entry =
-            thread_id_to_samplers_.Get().Lookup(thread_key, thread_hash);
-        DCHECK(entry != NULL);
-        SamplerList* samplers = reinterpret_cast<SamplerList*>(entry->value);
-        samplers->RemoveElement(sampler);
-        if (samplers->is_empty()) {
-          thread_id_to_samplers_.Pointer()->Remove(thread_key, thread_hash);
-          delete samplers;
-        }
-        if (thread_id_to_samplers_.Get().occupancy() == 0) {
-          instance_to_remove = instance_;
-          instance_ = NULL;
-        }
-      }
-#else
-      bool removed = instance_->active_samplers_.RemoveElement(sampler);
-      DCHECK(removed);
-      USE(removed);
-
-      // We cannot delete the instance immediately as we need to Join() the
-      // thread but we are holding mutex_ and the thread may try to acquire it.
-      if (instance_->active_samplers_.is_empty()) {
-        instance_to_remove = instance_;
-        instance_ = NULL;
-      }
-#endif  // USE_SIGNALS
-    }
 
-    if (!instance_to_remove) return;
-    instance_to_remove->Join();
-    delete instance_to_remove;
-  }
-
-  // Unlike AddActiveSampler, this method only adds a sampler,
-  // but won't start the sampler thread.
-  static void RegisterSampler(Sampler* sampler) {
-    base::LockGuard<base::Mutex> lock_guard(mutex_);
-#if defined(USE_SIGNALS)
-    AddSampler(sampler);
-#endif  // USE_SIGNALS
-  }
-
-  // Implement Thread::Run().
-  virtual void Run() {
-    while (true) {
-      {
-        base::LockGuard<base::Mutex> lock_guard(mutex_);
-#if defined(USE_SIGNALS)
-        if (thread_id_to_samplers_.Get().occupancy() == 0) break;
-        if (SignalHandler::Installed()) {
-          for (HashMap::Entry *p = thread_id_to_samplers_.Get().Start();
-               p != NULL; p = thread_id_to_samplers_.Get().Next(p)) {
-#if V8_OS_AIX && V8_TARGET_ARCH_PPC64
-            // on AIX64, cannot cast (void *) to pthread_t which is
-            // of type unsigned int (4bytes)
-            pthread_t thread_id = reinterpret_cast<intptr_t>(p->key);
-#else
-            pthread_t thread_id = reinterpret_cast<pthread_t>(p->key);
-#endif
-            pthread_kill(thread_id, SIGPROF);
-          }
-        }
-#else
-        if (active_samplers_.is_empty()) break;
-        // When CPU profiling is enabled both JavaScript and C++ code is
-        // profiled. We must not suspend.
-        for (int i = 0; i < active_samplers_.length(); ++i) {
-          Sampler* sampler = active_samplers_.at(i);
-          if (!sampler->IsProfiling()) continue;
-          sampler->DoSample();
-        }
-#endif  // USE_SIGNALS
-      }
-      base::OS::Sleep(base::TimeDelta::FromMilliseconds(interval_));
-    }
-  }
-
- private:
-  // Protects the process wide state below.
-  static base::Mutex* mutex_;
-  static SamplerThread* instance_;
-
-  const int interval_;
+#endif  // !V8_OS_NACL
 
-#if defined(USE_SIGNALS)
-  struct HashMapCreateTrait {
-    static void Construct(HashMap* allocated_ptr) {
-      new (allocated_ptr) HashMap(HashMap::PointersMatch);
-    }
-  };
-  friend class SignalHandler;
-  static base::LazyInstance<HashMap, HashMapCreateTrait>::type
-      thread_id_to_samplers_;
-  static base::AtomicValue<int> sampler_list_access_counter_;
-  static void AddSampler(Sampler* sampler) {
-    AtomicGuard atomic_guard(&sampler_list_access_counter_);
-    // Add sampler into map if needed.
-    pthread_t thread_id = sampler->platform_data()->vm_tid();
-    HashMap::Entry *entry =
-        thread_id_to_samplers_.Pointer()->LookupOrInsert(ThreadKey(thread_id),
-                                                         ThreadHash(thread_id));
-    if (entry->value == NULL) {
-      SamplerList* samplers = new SamplerList();
-      samplers->Add(sampler);
-      entry->value = samplers;
-    } else {
-      SamplerList* samplers = reinterpret_cast<SamplerList*>(entry->value);
-      if (!samplers->Contains(sampler)) {
-        samplers->Add(sampler);
-      }
-    }
-  }
-#else
-  SamplerList active_samplers_;
 #endif  // USE_SIGNALS
 
-  DISALLOW_COPY_AND_ASSIGN(SamplerThread);
-};
-
-
-base::Mutex* SamplerThread::mutex_ = NULL;
-SamplerThread* SamplerThread::instance_ = NULL;
-#if defined(USE_SIGNALS)
-base::LazyInstance<HashMap, SamplerThread::HashMapCreateTrait>::type
-    SamplerThread::thread_id_to_samplers_ = LAZY_INSTANCE_INITIALIZER;
-base::AtomicValue<int> SamplerThread::sampler_list_access_counter_(0);
-
-// As Native Client does not support signal handling, profiling is disabled.
-#if !V8_OS_NACL
-void SignalHandler::HandleProfilerSignal(int signal, siginfo_t* info,
-                                         void* context) {
-  USE(info);
-  if (signal != SIGPROF) return;
-  AtomicGuard atomic_guard(&SamplerThread::sampler_list_access_counter_, false);
-  if (!atomic_guard.is_success()) return;
-  pthread_t thread_id = pthread_self();
-  HashMap::Entry* entry =
-      SamplerThread::thread_id_to_samplers_.Pointer()->Lookup(
-          ThreadKey(thread_id), ThreadHash(thread_id));
-  if (entry == NULL)
-    return;
-  SamplerList* samplers = reinterpret_cast<SamplerList*>(entry->value);
-  for (int i = 0; i < samplers->length(); ++i) {
-    Sampler* sampler = samplers->at(i);
-    CollectSample(context, sampler);
-  }
-}
-#endif  // !V8_OS_NACL
-#endif  // USE_SIGNALs
-
 
 void Sampler::SetUp() {
 #if defined(USE_SIGNALS)
   SignalHandler::SetUp();
 #endif
-  SamplerThread::SetUp();
 }
 
 
 void Sampler::TearDown() {
-  SamplerThread::TearDown();
 #if defined(USE_SIGNALS)
   SignalHandler::TearDown();
 #endif
 }
 
-Sampler::Sampler(Isolate* isolate, int interval)
-    : isolate_(isolate),
-      interval_(interval),
+Sampler::Sampler(Isolate* isolate)
+    : is_counting_samples_(false),
+      js_sample_count_(0),
+      external_sample_count_(0),
+      isolate_(isolate),
       profiling_(false),
       has_processing_thread_(false),
       active_(false),
-      registered_(false),
-      is_counting_samples_(false),
-      js_sample_count_(0),
-      external_sample_count_(0) {
+      registered_(false) {
   data_ = new PlatformData;
 }
 
 Sampler::~Sampler() {
   DCHECK(!IsActive());
+#if defined(USE_SIGNALS)
   if (IsRegistered()) {
-    SamplerThread::RemoveSampler(this);
+    SamplerManager::RemoveSampler(this);
   }
+#endif
   delete data_;
 }
 
 void Sampler::Start() {
   DCHECK(!IsActive());
   SetActive(true);
-  SamplerThread::AddActiveSampler(this);
+#if defined(USE_SIGNALS)
+  SamplerManager::AddSampler(this);
+#endif
 }
 
 
 void Sampler::Stop() {
+#if defined(USE_SIGNALS)
+  SamplerManager::RemoveSampler(this);
+#endif
   DCHECK(IsActive());
-  SamplerThread::RemoveSampler(this);
   SetActive(false);
   SetRegistered(false);
 }
@@ -758,28 +633,12 @@ void Sampler::DecreaseProfilingDepth() {
 }
 
 
-void Sampler::SampleStack(const v8::RegisterState& state) {
-  TickSample* sample = isolate_->cpu_profiler()->StartTickSample();
-  TickSample sample_obj;
-  if (sample == NULL) sample = &sample_obj;
-  sample->Init(isolate_, state, TickSample::kIncludeCEntryFrame, true);
-  if (is_counting_samples_ && !sample->timestamp.IsNull()) {
-    if (sample->state == JS) ++js_sample_count_;
-    if (sample->state == EXTERNAL) ++external_sample_count_;
-  }
-  Tick(sample);
-  if (sample != &sample_obj) {
-    isolate_->cpu_profiler()->FinishTickSample();
-  }
-}
-
-
 #if defined(USE_SIGNALS)
 
 void Sampler::DoSample() {
   if (!SignalHandler::Installed()) return;
   if (!IsActive() && !IsRegistered()) {
-    SamplerThread::RegisterSampler(this);
+    SamplerManager::AddSampler(this);
     SetRegistered(true);
   }
   pthread_kill(platform_data()->vm_tid(), SIGPROF);
@@ -800,22 +659,15 @@ void Sampler::DoSample() {
   context.ContextFlags = CONTEXT_FULL;
   if (GetThreadContext(profiled_thread, &context) != 0) {
     v8::RegisterState state;
-#if defined(USE_SIMULATOR)
-    if (!SimulatorHelper::FillRegisters(isolate(), &state)) {
-      ResumeThread(profiled_thread);
-      return;
-    }
-#else
 #if V8_HOST_ARCH_X64
-    state.pc = reinterpret_cast<Address>(context.Rip);
-    state.sp = reinterpret_cast<Address>(context.Rsp);
-    state.fp = reinterpret_cast<Address>(context.Rbp);
+    state.pc = reinterpret_cast<void*>(context.Rip);
+    state.sp = reinterpret_cast<void*>(context.Rsp);
+    state.fp = reinterpret_cast<void*>(context.Rbp);
 #else
-    state.pc = reinterpret_cast<Address>(context.Eip);
-    state.sp = reinterpret_cast<Address>(context.Esp);
-    state.fp = reinterpret_cast<Address>(context.Ebp);
+    state.pc = reinterpret_cast<void*>(context.Eip);
+    state.sp = reinterpret_cast<void*>(context.Esp);
+    state.fp = reinterpret_cast<void*>(context.Ebp);
 #endif
-#endif  // USE_SIMULATOR
     SampleStack(state);
   }
   ResumeThread(profiled_thread);
@@ -823,6 +675,5 @@ void Sampler::DoSample() {
 
 #endif  // USE_SIGNALS
 
-
-}  // namespace internal
+}  // namespace sampler
 }  // namespace v8

src/profiler/sampler.h → src/libsampler/v8-sampler.h

-// Copyright 2013 the V8 project authors. All rights reserved.
+// Copyright 2016 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
-#ifndef V8_PROFILER_SAMPLER_H_
-#define V8_PROFILER_SAMPLER_H_
+#ifndef V8_LIBSAMPLER_SAMPLER_H_
+#define V8_LIBSAMPLER_SAMPLER_H_
 
 #include "include/v8.h"
 
@@ -11,10 +11,7 @@
 #include "src/base/macros.h"
 
 namespace v8 {
-namespace internal {
-
-class Isolate;
-struct TickSample;
+namespace sampler {
 
 // ----------------------------------------------------------------------------
 // Sampler
@@ -25,19 +22,23 @@ struct TickSample;
 
 class Sampler {
  public:
+  static const int kMaxFramesCountLog2 = 8;
+  static const unsigned kMaxFramesCount = (1u << kMaxFramesCountLog2) - 1;
+
   // Initializes the Sampler support. Called once at VM startup.
   static void SetUp();
   static void TearDown();
 
   // Initialize sampler.
-  Sampler(Isolate* isolate, int interval);
+  explicit Sampler(Isolate* isolate);
   virtual ~Sampler();
 
   Isolate* isolate() const { return isolate_; }
-  int interval() const { return interval_; }
 
   // Performs stack sampling.
-  void SampleStack(const v8::RegisterState& regs);
+  // Clients should override this method in order to do something on samples,
+  // for example buffer samples in a queue.
+  virtual void SampleStack(const v8::RegisterState& regs) = 0;
 
   // Start and stop sampler.
   void Start();
@@ -60,8 +61,7 @@ class Sampler {
   bool IsRegistered() const { return base::NoBarrier_Load(&registered_); }
 
   void DoSample();
-  // If true next sample must be initiated on the profiler event processor
-  // thread right after latest sample is processed.
+
   void SetHasProcessingThread(bool value) {
     base::NoBarrier_Store(&has_processing_thread_, value);
   }
@@ -79,30 +79,25 @@ class Sampler {
   PlatformData* platform_data() const { return data_; }
 
  protected:
-  // This method is called for each sampling period with the current
-  // program counter.
-  virtual void Tick(TickSample* sample) = 0;
+  // Counts stack samples taken in various VM states.
+  bool is_counting_samples_;
+  unsigned js_sample_count_;
+  unsigned external_sample_count_;
 
  private:
   void SetActive(bool value) { base::NoBarrier_Store(&active_, value); }
-
   void SetRegistered(bool value) { base::NoBarrier_Store(&registered_, value); }
 
   Isolate* isolate_;
-  const int interval_;
   base::Atomic32 profiling_;
   base::Atomic32 has_processing_thread_;
   base::Atomic32 active_;
   base::Atomic32 registered_;
   PlatformData* data_;  // Platform specific data.
-  // Counts stack samples taken in various VM states.
-  bool is_counting_samples_;
-  unsigned js_sample_count_;
-  unsigned external_sample_count_;
   DISALLOW_IMPLICIT_CONSTRUCTORS(Sampler);
 };
 
-}  // namespace internal
+}  // namespace sampler
 }  // namespace v8
 
-#endif  // V8_PROFILER_SAMPLER_H_
+#endif  // V8_LIBSAMPLER_SAMPLER_H_

src/log.cc

@@ -15,11 +15,12 @@
 #include "src/global-handles.h"
 #include "src/interpreter/bytecodes.h"
 #include "src/interpreter/interpreter.h"
+#include "src/libsampler/v8-sampler.h"
 #include "src/log-inl.h"
 #include "src/log-utils.h"
 #include "src/macro-assembler.h"
 #include "src/perf-jit.h"
-#include "src/profiler/cpu-profiler.h"
+#include "src/profiler/cpu-profiler-inl.h"
 #include "src/runtime-profiler.h"
 #include "src/string-stream.h"
 #include "src/vm-state-inl.h"
@@ -539,6 +540,31 @@ void JitLogger::EndCodePosInfoEvent(AbstractCode* code,
 }
 
 
+// TODO(lpy): Keeping sampling thread inside V8 is a workaround currently,
+// the reason is to reduce code duplication during migration to sampler library,
+// sampling thread, as well as the sampler, will be moved to D8 eventually.
+class SamplingThread : public base::Thread {
+ public:
+  static const int kSamplingThreadStackSize = 64 * KB;
+
+  SamplingThread(sampler::Sampler* sampler, int interval)
+      : base::Thread(base::Thread::Options("SamplingThread",
+                                           kSamplingThreadStackSize)),
+        sampler_(sampler),
+        interval_(interval) {}
+  void Run() override {
+    while (sampler_->IsProfiling()) {
+      sampler_->DoSample();
+      base::OS::Sleep(base::TimeDelta::FromMilliseconds(interval_));
+    }
+  }
+
+ private:
+  sampler::Sampler* sampler_;
+  const int interval_;
+};
+
+
 // The Profiler samples pc and sp values for the main thread.
 // Each sample is appended to a circular buffer.
 // An independent thread removes data and writes it to the log.
@@ -611,16 +637,16 @@ class Profiler: public base::Thread {
 // Ticker used to provide ticks to the profiler and the sliding state
 // window.
 //
-class Ticker: public Sampler {
+class Ticker: public sampler::Sampler {
  public:
   Ticker(Isolate* isolate, int interval):
-      Sampler(isolate, interval),
-      profiler_(NULL) {}
-
-  ~Ticker() { if (IsActive()) Stop(); }
+      sampler::Sampler(reinterpret_cast<v8::Isolate*>(isolate)),
+      profiler_(NULL),
+      sampling_thread_(new SamplingThread(this, interval)) {}
 
-  virtual void Tick(TickSample* sample) {
-    if (profiler_) profiler_->Insert(sample);
+  ~Ticker() {
+    if (IsActive()) Stop();
+    delete sampling_thread_;
   }
 
   void SetProfiler(Profiler* profiler) {
@@ -628,16 +654,40 @@ class Ticker: public Sampler {
     profiler_ = profiler;
     IncreaseProfilingDepth();
     if (!IsActive()) Start();
+    sampling_thread_->StartSynchronously();
   }
 
   void ClearProfiler() {
     profiler_ = NULL;
     if (IsActive()) Stop();
     DecreaseProfilingDepth();
+    sampling_thread_->Join();
+  }
+
+  void SampleStack(const v8::RegisterState& state) override {
+    v8::Isolate* v8_isolate = isolate();
+    Isolate* isolate = reinterpret_cast<Isolate*>(v8_isolate);
+#if defined(USE_SIMULATOR)
+    SimulatorHelper::FillRegisters(isolate,
+                                   const_cast<v8::RegisterState*>(&state));
+#endif
+    TickSample* sample = isolate->cpu_profiler()->StartTickSample();
+    TickSample sample_obj;
+    if (sample == NULL) sample = &sample_obj;
+    sample->Init(isolate, state, TickSample::kIncludeCEntryFrame, true);
+    if (is_counting_samples_ && !sample->timestamp.IsNull()) {
+      if (sample->state == JS) ++js_sample_count_;
+      if (sample->state == EXTERNAL) ++external_sample_count_;
+    }
+    if (profiler_) profiler_->Insert(sample);
+    if (sample != &sample_obj) {
+      isolate->cpu_profiler()->FinishTickSample();
+    }
   }
 
  private:
   Profiler* profiler_;
+  SamplingThread* sampling_thread_;
 };
 
 
@@ -1810,7 +1860,7 @@ void Logger::SetCodeEventHandler(uint32_t options,
 }
 
 
-Sampler* Logger::sampler() {
+sampler::Sampler* Logger::sampler() {
   return ticker_;
 }
 

src/log.h

@@ -19,6 +19,10 @@ namespace base {
 class Semaphore;
 }
 
+namespace sampler {
+class Sampler;
+}
+
 namespace internal {
 
 // Logger is used for collecting logging information from V8 during
@@ -141,7 +145,6 @@ class JitLogger;
 class PerfBasicLogger;
 class LowLevelLogger;
 class PerfJitLogger;
-class Sampler;
 
 class Logger {
  public:
@@ -161,7 +164,7 @@ class Logger {
   void SetCodeEventHandler(uint32_t options,
                            JitCodeEventHandler event_handler);
 
-  Sampler* sampler();
+  sampler::Sampler* sampler();
 
   // Frees resources acquired in SetUp.
   // When a temporary file is used for the log, returns its stream descriptor,

src/profiler/cpu-profiler.cc

@@ -21,7 +21,7 @@ static const int kProfilerStackSize = 64 * KB;
 
 
 ProfilerEventsProcessor::ProfilerEventsProcessor(ProfileGenerator* generator,
-                                                 Sampler* sampler,
+                                                 sampler::Sampler* sampler,
                                                  base::TimeDelta period)
     : Thread(Thread::Options("v8:ProfEvntProc", kProfilerStackSize)),
       generator_(generator),
@@ -566,7 +566,7 @@ void CpuProfiler::StartProcessorIfNotStarted() {
   saved_is_logging_ = logger->is_logging_;
   logger->is_logging_ = false;
   generator_ = new ProfileGenerator(profiles_);
-  Sampler* sampler = logger->sampler();
+  sampler::Sampler* sampler = logger->sampler();
   processor_ = new ProfilerEventsProcessor(
       generator_, sampler, sampling_interval_);
   is_profiling_ = true;
@@ -612,7 +612,8 @@ void CpuProfiler::StopProcessorIfLastProfile(const char* title) {
 
 void CpuProfiler::StopProcessor() {
   Logger* logger = isolate_->logger();
-  Sampler* sampler = reinterpret_cast<Sampler*>(logger->ticker_);
+  sampler::Sampler* sampler =
+      reinterpret_cast<sampler::Sampler*>(logger->ticker_);
   is_profiling_ = false;
   processor_->StopSynchronously();
   delete processor_;

src/profiler/cpu-profiler.h

@@ -10,9 +10,9 @@
 #include "src/base/atomicops.h"
 #include "src/base/platform/time.h"
 #include "src/compiler.h"
+#include "src/libsampler/v8-sampler.h"
 #include "src/locked-queue.h"
 #include "src/profiler/circular-queue.h"
-#include "src/profiler/sampler.h"
 #include "src/profiler/tick-sample.h"
 
 namespace v8 {
@@ -128,7 +128,7 @@ class CodeEventsContainer {
 class ProfilerEventsProcessor : public base::Thread {
  public:
   ProfilerEventsProcessor(ProfileGenerator* generator,
-                          Sampler* sampler,
+                          sampler::Sampler* sampler,
                           base::TimeDelta period);
   virtual ~ProfilerEventsProcessor();
 
@@ -166,7 +166,7 @@ class ProfilerEventsProcessor : public base::Thread {
   SampleProcessingResult ProcessOneSample();
 
   ProfileGenerator* generator_;
-  Sampler* sampler_;
+  sampler::Sampler* sampler_;
   base::Atomic32 running_;
   const base::TimeDelta period_;  // Samples & code events processing period.
   LockedQueue<CodeEventsContainer> events_buffer_;

src/v8.cc

@@ -14,9 +14,9 @@
 #include "src/elements.h"
 #include "src/frames.h"
 #include "src/isolate.h"
+#include "src/libsampler/v8-sampler.h"
 #include "src/objects.h"
 #include "src/profiler/heap-profiler.h"
-#include "src/profiler/sampler.h"
 #include "src/runtime-profiler.h"
 #include "src/snapshot/natives.h"
 #include "src/snapshot/snapshot.h"
@@ -48,7 +48,7 @@ void V8::TearDown() {
   ExternalReference::TearDownMathExpData();
   RegisteredExtension::UnregisterAll();
   Isolate::GlobalTearDown();
-  Sampler::TearDown();
+  sampler::Sampler::TearDown();
   FlagList::ResetAllFlags();  // Frees memory held by string arguments.
 }
 
@@ -76,7 +76,7 @@ void V8::InitializeOncePerProcessImpl() {
 
   Isolate::InitializeOncePerProcess();
 
-  Sampler::SetUp();
+  sampler::Sampler::SetUp();
   CpuFeatures::Probe(false);
   ElementsAccessor::InitializeOncePerProcess();
   LOperand::SetUpCaches();

src/v8.gyp

@@ -381,6 +381,7 @@
       'type': 'static_library',
       'dependencies': [
         'v8_libbase',
+        'v8_libsampler',
       ],
       'variables': {
         'optimize': 'max',
@@ -1001,8 +1002,6 @@
         'profiler/profile-generator-inl.h',
         'profiler/profile-generator.cc',
         'profiler/profile-generator.h',
-        'profiler/sampler.cc',
-        'profiler/sampler.h',
         'profiler/sampling-heap-profiler.cc',
         'profiler/sampling-heap-profiler.h',
         'profiler/strings-storage.cc',
@@ -1939,6 +1938,38 @@
         ],
       },
     },
+    {
+      'target_name': 'v8_libsampler',
+      'type': 'static_library',
+      'variables': {
+        'optimize': 'max',
+      },
+      'dependencies': [
+        'v8_libbase',
+      ],
+      'include_dirs+': [
+        '..',
+        '../include',
+      ],
+      'sources': [
+        'libsampler/hashmap.h',
+        'libsampler/utils.h',
+        'libsampler/v8-sampler.cc',
+        'libsampler/v8-sampler.h'
+      ],
+      'conditions': [
+        ['want_separate_host_toolset==1', {
+          'toolsets': ['host', 'target'],
+        }, {
+          'toolsets': ['target'],
+        }],
+      ],
+      'direct_dependent_settings': {
+        'include_dirs': [
+          '../include',
+        ],
+      },
+    },
     {
       'target_name': 'natives_blob',
       'type': 'none',

test/cctest/cctest.gyp

@@ -108,6 +108,7 @@
         'heap/test-lab.cc',
         'heap/test-mark-compact.cc',
         'heap/test-spaces.cc',
+        'libsampler/test-sampler.cc',
         'print-extension.cc',
         'profiler-extension.cc',
         'test-accessors.cc',

test/cctest/libsampler/test-sampler.cc

+// Copyright 2016 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+// Tests of sampler functionalities.
+
+#include "src/libsampler/v8-sampler.h"
+
+#include "src/base/platform/platform.h"
+#include "test/cctest/cctest.h"
+
+
+namespace v8 {
+namespace sampler {
+
+namespace {
+
+class TestSamplingThread : public base::Thread {
+ public:
+  static const int kSamplerThreadStackSize = 64 * 1024;
+
+  explicit TestSamplingThread(Sampler* sampler)
+      : Thread(base::Thread::Options("TestSamplingThread",
+                                     kSamplerThreadStackSize)),
+        sampler_(sampler) {}
+
+  // Implement Thread::Run().
+  void Run() override {
+    while (sampler_->IsProfiling()) {
+      sampler_->DoSample();
+      base::OS::Sleep(base::TimeDelta::FromMilliseconds(1));
+    }
+  }
+
+ private:
+  Sampler* sampler_;
+};
+
+
+class TestSampler : public Sampler {
+ public:
+  explicit TestSampler(Isolate* isolate) : Sampler(isolate) {}
+
+  void SampleStack(const v8::RegisterState& regs) override {
+    void* frames[Sampler::kMaxFramesCount];
+    SampleInfo sample_info;
+    isolate()->GetStackSample(regs, reinterpret_cast<void**>(frames),
+                              Sampler::kMaxFramesCount, &sample_info);
+    if (is_counting_samples_) {
+      if (sample_info.vm_state == JS) ++js_sample_count_;
+      if (sample_info.vm_state == EXTERNAL) ++external_sample_count_;
+    }
+  }
+};
+
+
+class TestApiCallbacks {
+ public:
+  TestApiCallbacks() {}
+
+  static void Getter(v8::Local<v8::String> name,
+                     const v8::PropertyCallbackInfo<v8::Value>& info) {
+  }
+
+  static void Setter(v8::Local<v8::String> name,
+                     v8::Local<v8::Value> value,
+                     const v8::PropertyCallbackInfo<void>& info) {
+  }
+};
+
+
+static void RunSampler(v8::Local<v8::Context> env,
+                       v8::Local<v8::Function> function,
+                       v8::Local<v8::Value> argv[], int argc,
+                       unsigned min_js_samples = 0,
+                       unsigned min_external_samples = 0) {
+  Sampler::SetUp();
+  TestSampler* sampler = new TestSampler(env->GetIsolate());
+  TestSamplingThread* thread = new TestSamplingThread(sampler);
+  sampler->IncreaseProfilingDepth();
+  sampler->Start();
+  sampler->StartCountingSamples();
+  thread->StartSynchronously();
+  do {
+    function->Call(env, env->Global(), argc, argv).ToLocalChecked();
+  } while (sampler->js_sample_count() < min_js_samples ||
+           sampler->external_sample_count() < min_external_samples);
+  sampler->Stop();
+  sampler->DecreaseProfilingDepth();
+  thread->Join();
+  delete thread;
+  delete sampler;
+  Sampler::TearDown();
+}
+
+}  // namespace
+
+static const char* sampler_test_source = "function start(count) {\n"
+"  for (var i = 0; i < count; i++) {\n"
+"    var o = instance.foo;\n"
+"    instance.foo = o + 1;\n"
+"  }\n"
+"}\n";
+
+static v8::Local<v8::Function> GetFunction(v8::Local<v8::Context> env,
+                                           const char* name) {
+  return v8::Local<v8::Function>::Cast(
+      env->Global()->Get(env, v8_str(name)).ToLocalChecked());
+}
+
+
+TEST(LibSamplerCollectSample) {
+  LocalContext env;
+  v8::Isolate* isolate = env->GetIsolate();
+  v8::HandleScope scope(isolate);
+
+  v8::Local<v8::FunctionTemplate> func_template =
+      v8::FunctionTemplate::New(isolate);
+  v8::Local<v8::ObjectTemplate> instance_template =
+      func_template->InstanceTemplate();
+
+  TestApiCallbacks accessors;
+  v8::Local<v8::External> data =
+      v8::External::New(isolate, &accessors);
+  instance_template->SetAccessor(v8_str("foo"), &TestApiCallbacks::Getter,
+                                 &TestApiCallbacks::Setter, data);
+  v8::Local<v8::Function> func =
+      func_template->GetFunction(env.local()).ToLocalChecked();
+  v8::Local<v8::Object> instance =
+      func->NewInstance(env.local()).ToLocalChecked();
+  env->Global()->Set(env.local(), v8_str("instance"), instance).FromJust();
+
+  CompileRun(sampler_test_source);
+  v8::Local<v8::Function> function = GetFunction(env.local(), "start");
+
+  int32_t repeat_count = 100;
+  v8::Local<v8::Value> args[] = {v8::Integer::New(isolate, repeat_count)};
+  RunSampler(env.local(), function, args, arraysize(args), 100, 100);
+}
+
+}  // namespace sampler
+}  // namespace v8

test/cctest/test-cpu-profiler.cc

@@ -422,7 +422,7 @@ static v8::CpuProfile* RunProfiler(v8::Local<v8::Context> env,
   cpu_profiler->SetSamplingInterval(100);
   cpu_profiler->StartProfiling(profile_name, collect_samples);
 
-  i::Sampler* sampler =
+  v8::sampler::Sampler* sampler =
       reinterpret_cast<i::Isolate*>(env->GetIsolate())->logger()->sampler();
   sampler->StartCountingSamples();
   do {