Make Isolate::GetStackSample API support simulator

src/profiler/sampler.cc

 // Copyright 2013 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"
 
 #if V8_OS_POSIX && !V8_OS_CYGWIN
 
 #define USE_SIGNALS
 
@@ skipping 30 matching lines @@
 #include "src/base/win32-headers.h"
 
 #endif
 
 #include "src/atomic-utils.h"
 #include "src/base/platform/platform.h"
 #include "src/flags.h"
 #include "src/frames-inl.h"
 #include "src/log.h"
 #include "src/profiler/cpu-profiler-inl.h"
-#include "src/simulator.h"
 #include "src/v8threads.h"
 #include "src/vm-state-inl.h"
 
 
 #if V8_OS_ANDROID && !defined(__BIONIC_HAVE_UCONTEXT_T)
 
 // Not all versions of Android's C library provide ucontext_t.
 // Detect this and provide custom but compatible definitions. Note that these
 // follow the GLibc naming convention to access register values from
 // mcontext_t.
@@ skipping 263 matching lines @@
 
   HANDLE profiled_thread() { return profiled_thread_; }
 
  private:
   HANDLE profiled_thread_;
 };
 #endif
 
 
 #if defined(USE_SIMULATOR)
-class SimulatorHelper {
- public:
-  inline bool Init(Isolate* isolate) {
-    simulator_ = isolate->thread_local_top()->simulator_;
-    // Check if there is active simulator.
-    return simulator_ != NULL;
+bool SimulatorHelper::FillRegisters(Isolate* isolate,
+                                    v8::RegisterState* state) {
+  Simulator *simulator = isolate->thread_local_top()->simulator_;
+  // Check if there is active simulator.
+  if (simulator == NULL) return false;
+#if V8_TARGET_ARCH_ARM
+  if (!simulator->has_bad_pc()) {

[alph, 2016/04/29 20:59:57]

not sure how did it work before. let's just return false for all archs if
has_bad_pc. Otherwise we'd just return an uninitialized pc and it's impossible
to detect in the client.

[lpy, 2016/04/29 21:43:10]

an uninitialized register state has a pc value 0, let's keep it unless we know
for sure that it is correct if we return false for all archs.

+    state->pc = reinterpret_cast<Address>(simulator->get_pc());
   }
-
-  inline void FillRegisters(v8::RegisterState* state) {
-#if V8_TARGET_ARCH_ARM
-    if (!simulator_->has_bad_pc()) {
-      state->pc = reinterpret_cast<Address>(simulator_->get_pc());
-    }
-    state->sp = reinterpret_cast<Address>(simulator_->get_register(
-        Simulator::sp));
-    state->fp = reinterpret_cast<Address>(simulator_->get_register(
-        Simulator::r11));
+  state->sp = reinterpret_cast<Address>(simulator->get_register(

[alph, 2016/04/29 20:59:57]

nit: should fit one line. run git cl format.

[lpy, 2016/04/29 21:43:10]

Done.

+      Simulator::sp));
+  state->fp = reinterpret_cast<Address>(simulator->get_register(
+      Simulator::r11));
 #elif V8_TARGET_ARCH_ARM64
-    if (simulator_->sp() == 0 || simulator_->fp() == 0) {
-      // It's possible that the simulator is interrupted while it is updating
-      // the sp or fp register. ARM64 simulator does this in two steps:
-      // first setting it to zero and then setting it to a new value.
-      // Bailout if sp/fp doesn't contain the new value.
-      //
-      // FIXME: The above doesn't really solve the issue.
-      // If a 64-bit target is executed on a 32-bit host even the final
-      // write is non-atomic, so it might obtain a half of the result.
-      // Moreover as long as the register set code uses memcpy (as of now),
-      // it is not guaranteed to be atomic even when both host and target
-      // are of same bitness.
-      return;
-    }
-    state->pc = reinterpret_cast<Address>(simulator_->pc());
-    state->sp = reinterpret_cast<Address>(simulator_->sp());
-    state->fp = reinterpret_cast<Address>(simulator_->fp());
+  if (simulator->sp() == 0 || simulator->fp() == 0) {
+    // It's possible that the simulator is interrupted while it is updating
+    // the sp or fp register. ARM64 simulator does this in two steps:
+    // first setting it to zero and then setting it to a new value.
+    // Bailout if sp/fp doesn't contain the new value.
+    //
+    // FIXME: The above doesn't really solve the issue.
+    // If a 64-bit target is executed on a 32-bit host even the final
+    // write is non-atomic, so it might obtain a half of the result.
+    // Moreover as long as the register set code uses memcpy (as of now),
+    // it is not guaranteed to be atomic even when both host and target
+    // are of same bitness.
+    return false;
+  }
+  state->pc = reinterpret_cast<Address>(simulator->pc());
+  state->sp = reinterpret_cast<Address>(simulator->sp());
+  state->fp = reinterpret_cast<Address>(simulator->fp());
 #elif V8_TARGET_ARCH_MIPS || V8_TARGET_ARCH_MIPS64
-    if (!simulator_->has_bad_pc()) {
-      state->pc = reinterpret_cast<Address>(simulator_->get_pc());
-    }
-    state->sp = reinterpret_cast<Address>(simulator_->get_register(
-        Simulator::sp));
-    state->fp = reinterpret_cast<Address>(simulator_->get_register(
-        Simulator::fp));
+  if (!simulator->has_bad_pc()) {
+    state->pc = reinterpret_cast<Address>(simulator->get_pc());
+  }
+  state->sp = reinterpret_cast<Address>(simulator->get_register(Simulator::sp));
+  state->fp = reinterpret_cast<Address>(simulator->get_register(Simulator::fp));
 #elif V8_TARGET_ARCH_PPC
-    if (!simulator_->has_bad_pc()) {
-      state->pc = reinterpret_cast<Address>(simulator_->get_pc());
-    }
-    state->sp =
-        reinterpret_cast<Address>(simulator_->get_register(Simulator::sp));
-    state->fp =
-        reinterpret_cast<Address>(simulator_->get_register(Simulator::fp));
+  if (!simulator->has_bad_pc()) {
+    state->pc = reinterpret_cast<Address>(simulator->get_pc());
+  }
+  state->sp = reinterpret_cast<Address>(simulator->get_register(Simulator::sp));
+  state->fp = reinterpret_cast<Address>(simulator->get_register(Simulator::fp));
 #elif V8_TARGET_ARCH_S390
-    if (!simulator_->has_bad_pc()) {
-      state->pc = reinterpret_cast<Address>(simulator_->get_pc());
-    }
-    state->sp =
-        reinterpret_cast<Address>(simulator_->get_register(Simulator::sp));
-    state->fp =
-        reinterpret_cast<Address>(simulator_->get_register(Simulator::fp));
+  if (!simulator->has_bad_pc()) {
+    state->pc = reinterpret_cast<Address>(simulator->get_pc());
+  }
+  state->sp = reinterpret_cast<Address>(simulator->get_register(Simulator::sp));
+  state->fp = reinterpret_cast<Address>(simulator->get_register(Simulator::fp));
 #endif
-  }
-
- private:
-  Simulator* simulator_;
-};
+  // It possible that the simulator is interrupted while it is updating
+  // the sp or fp register. ARM64 simulator does this in two steps:
+  // first setting it to zero and then setting it to the new value.
+  // Bailout if sp/fp doesn't contain the new value.
+  if (state->sp == 0 || state->fp == 0) return false;

[alph, 2016/04/29 20:59:57]

this seems to be already done at line 348. We probably want to do it for all
archs, not just ARM64. as e.g. MIPS might have the same issue.

[lpy, 2016/04/29 21:43:11]

Yes, line 348 is for ARM64 only, let's delete the one above instead.

+  return true;
+}
 #endif  // USE_SIMULATOR
 
 
 #if defined(USE_SIGNALS)
 
 class SignalHandler : public AllStatic {
  public:
   static void SetUp() { if (!mutex_) mutex_ = new base::Mutex(); }
   static void TearDown() { delete mutex_; mutex_ = NULL; }
 
@@ skipping 70 matching lines @@
   if (isolate == NULL || !isolate->IsInUse()) return;
 
   if (v8::Locker::IsActive() &&
       !isolate->thread_manager()->IsLockedByCurrentThread()) {
     return;
   }
 
   v8::RegisterState state;
 
 #if defined(USE_SIMULATOR)
-  SimulatorHelper helper;
-  if (!helper.Init(isolate)) return;
-  helper.FillRegisters(&state);
-  // It possible that the simulator is interrupted while it is updating
-  // the sp or fp register. ARM64 simulator does this in two steps:
-  // first setting it to zero and then setting it to the new value.
-  // Bailout if sp/fp doesn't contain the new value.
-  if (state.sp == 0 || state.fp == 0) return;
+  if (!SimulatorHelper::FillRegisters(isolate, &state)) return;
 #else
   // 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)))
   mcontext_t& mcontext = ucontext->uc_mcontext;
 #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]);
@@ skipping 508 matching lines @@
   }
   pthread_kill(platform_data()->vm_tid(), SIGPROF);
 }
 
 #elif V8_OS_WIN || V8_OS_CYGWIN
 
 void Sampler::DoSample() {
   HANDLE profiled_thread = platform_data()->profiled_thread();
   if (profiled_thread == NULL) return;
 
-#if defined(USE_SIMULATOR)
-  SimulatorHelper helper;
-  if (!helper.Init(isolate())) return;
-#endif
-
   const DWORD kSuspendFailed = static_cast<DWORD>(-1);
   if (SuspendThread(profiled_thread) == kSuspendFailed) return;
 
   // Context used for sampling the register state of the profiled thread.
   CONTEXT context;
   memset(&context, 0, sizeof(context));
   context.ContextFlags = CONTEXT_FULL;
   if (GetThreadContext(profiled_thread, &context) != 0) {
     v8::RegisterState state;
 #if defined(USE_SIMULATOR)
-    helper.FillRegisters(&state);
+    if (!SimulatorHelper::FillRegisters(isolate(), &state)) return;

[alph, 2016/04/29 20:59:57]

you cannot return without resuming the thread.

[lpy, 2016/04/29 21:43:10]

Done.

 #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);
 #else
     state.pc = reinterpret_cast<Address>(context.Eip);
     state.sp = reinterpret_cast<Address>(context.Esp);
     state.fp = reinterpret_cast<Address>(context.Ebp);
 #endif
 #endif  // USE_SIMULATOR
     SampleStack(state);
   }
   ResumeThread(profiled_thread);
 }
 
 #endif  // USE_SIGNALS
 
 
 }  // namespace internal
 }  // namespace v8