Initial implementation of GetStackSample sampling profiler API.

test/cctest/test-sampler-api.cc

+// Copyright 2014 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 the sampling API in include/v8.h
+
+#include <string>
+#include "include/v8.h"
+#include "src/simulator.h"
+#include "src/utils.h"
+#include "src/v8.h"
+#include "test/cctest/cctest.h"
+
+#if V8_OS_POSIX && !V8_OS_CYGWIN && !V8_OS_MACOSX
+#include <ucontext.h>
+
+#elif V8_OS_WIN || V8_OS_CYGWIN  // V8_OS_POSIX && !V8_OS_CYGWIN
+#include "src/base/win32-headers.h"
+
+#endif  // V8_OS_WIN || V8_OS_CYGWIN
+
+using v8::Local;

[yurys, 2014/09/23 14:57:53]

style nit: we don't usually declare such using in other places in v8, so I'd
rather be consistent with the rest of the codebase.

[alph, 2014/09/24 13:12:18]

Done.

+using v8::internal::Address;
+using v8::internal::Isolate;
+
+namespace {
+
+class Sample {
+ public:
+  enum { kFramesLimit = 255 };
+
+  Sample() {}
+
+  typedef const void* const* const_iterator;
+  const_iterator begin() const { return data_.start(); }
+  const_iterator end() const { return &data_[data_.length()]; }
+
+  int size() const { return data_.length(); }
+  v8::internal::Vector<void*>& data() { return data_; }
+
+ private:
+  v8::internal::EmbeddedVector<void*, kFramesLimit> data_;
+};
+
+
+// The Sample which CollectSample fills up
+Sample* sample;
+
+// The isolate used in the test
+v8::Isolate* isolate;
+
+// Forward declaration
+// (platform specific implementation at the bottom of this file)
+void FillRegisterState(v8::RegisterState* state);
+
+// The JavaScript calls this function when on full stack depth.
+void CollectSample(const v8::FunctionCallbackInfo<v8::Value>& args) {
+  v8::RegisterState state;
+  FillRegisterState(&state);
+  v8::SampleInfo info;
+  isolate->GetStackSample(state, sample->data().start(),
+                          static_cast<size_t>(sample->size()), &info);
+  size_t frames_count = info.frames_count;
+  CHECK_LE(frames_count, static_cast<size_t>(sample->size()));
+  sample->data().Truncate(static_cast<int>(frames_count));
+}
+
+
+// A JavaScript function which takes stack depth
+// (minimum value 2) as an argument.
+// When at the bottom of the recursion,
+// the JavaScript code calls into C++ test code,
+// waiting for the sampler to take a sample.
+static const char* test_function =
+    "function func(depth) {"
+    "  if (depth == 2) CollectSample();"
+    "  else return func(depth - 1);"
+    "}";
+
+}  // namespace
+
+
+#define SAMPLER_API_TESTS_BOOTSTRAP()                                       \

[yurys, 2014/09/23 14:57:53]

style: I'd use a class rather than a macro here.

[alph, 2014/09/24 13:12:18]

Done.

+  Sample test_sample;                                                       \
+  sample = &test_sample;                                                    \
+  isolate = CcTest::isolate();                                              \
+  v8::HandleScope scope(isolate);                                           \
+  v8::Handle<v8::ObjectTemplate> global = v8::ObjectTemplate::New(isolate); \
+  global->Set(v8::String::NewFromUtf8(isolate, "CollectSample"),            \
+              v8::FunctionTemplate::New(isolate, CollectSample));           \
+  LocalContext env(isolate, NULL, global)
+
+
+TEST(StackDepthIsConsistent) {
+  SAMPLER_API_TESTS_BOOTSTRAP();
+
+  std::string source(test_function);
+  source.append("func(8);");
+  v8::Script::Compile(v8::String::NewFromUtf8(isolate, source.c_str()))->Run();
+
+  CHECK_EQ(8, sample->size());
+}
+
+
+TEST(StackDepthDoesNotExceedMaxValue) {
+  SAMPLER_API_TESTS_BOOTSTRAP();
+
+  std::string source(test_function);
+  source.append("func(300);");
+  v8::Script::Compile(v8::String::NewFromUtf8(isolate, source.c_str()))->Run();
+
+  int MAX_SIZE = Sample::kFramesLimit;

[yurys, 2014/09/23 14:57:53]

inline this

[alph, 2014/09/24 13:12:18]

Done.

+  CHECK_EQ(MAX_SIZE, sample->size());
+}
+
+
+namespace {
+std::vector<v8::JitCodeEvent> inner_funcs;

[yurys, 2014/09/23 14:57:53]

JitCodeEvent contains Handle<Script>, is it a global or local handle? In the
latter case it may be unsafe to put it into a global vector like this as
corresponding HandleScope may be destroyed before the handle.

[alph, 2014/09/24 13:12:18]

Done.

+std::vector<v8::JitCodeEvent> outer_funcs;
+
+void TestJitCodeEventHandler(const v8::JitCodeEvent* event) {
+  if (event->type != v8::JitCodeEvent::CODE_ADDED) return;

[yurys, 2014/09/23 14:57:53]

What about v8::JitCodeEvent::CODE_MOVE? How can we be sure that GC won't move
the code during the test?

[alph, 2014/09/24 13:12:18]

Done.

+  std::string name(event->name.str, event->name.len);
+  if (name.find("test_sampler_api_inner") != std::string::npos)
+    inner_funcs.push_back(*event);
+  if (name.find("test_sampler_api_outer") != std::string::npos)
+    outer_funcs.push_back(*event);
+}
+
+
+// Note: The arguments.callee stuff is there so that the
+//       functions are not optimized away.
+static const char* test_script =
+    "function test_sampler_api_inner() {"
+    "  CollectSample();"
+    "  return arguments.callee.toString();"
+    "}"
+    "function test_sampler_api_outer() {"
+    "  return test_sampler_api_inner() + arguments.callee.toString();"
+    "}"
+    "test_sampler_api_outer();";
+}
+
+
+// The captured sample should have three pc values.
+// They should fall in the range where the compiled code

[yurys, 2014/09/23 14:57:53]

typo: ... where the compiled code resides.

[alph, 2014/09/24 13:12:18]

Done.

+// The expected stack is:
+// bottom of stack [{anon script}, outer, inner] top of stack
+//                              ^      ^       ^
+// sample.stack indices         2      1       0
+TEST(StackFramesConsistent) {
+  SAMPLER_API_TESTS_BOOTSTRAP();
+
+  isolate->SetJitCodeEventHandler(v8::kJitCodeEventDefault,
+                                  TestJitCodeEventHandler);
+  v8::Script::Compile(v8::String::NewFromUtf8(isolate, test_script))->Run();
+
+  CHECK_EQ(3, sample->size());
+
+  bool stack_top_is_inner = false;
+  bool below_inner_is_outer = false;
+
+  for (unsigned i = 0; i < inner_funcs.size(); i++) {
+    void* start_addr = inner_funcs[i].code_start;
+    void* end_addr = reinterpret_cast<void*>(
+        (int64_t)inner_funcs[i].code_start + inner_funcs[i].code_len);

[yurys, 2014/09/23 14:57:53]

static_cast<intptr_t>(inner_funcs[i].code_start) as we don't use c-style cast.

[alph, 2014/09/24 13:12:18]

Done.

+    if ((*sample->begin() >= start_addr) && (*sample->begin() < end_addr))
+      stack_top_is_inner = true;
+  }
+
+  for (unsigned i = 0; i < outer_funcs.size(); i++) {
+    void* start_addr = outer_funcs[i].code_start;
+    void* end_addr = reinterpret_cast<void*>(
+        (int64_t)outer_funcs[i].code_start + outer_funcs[i].code_len);
+    if ((*(sample->begin() + 1) >= start_addr) &&
+        (*(sample->begin() + 1) < end_addr))
+      below_inner_is_outer = true;
+  }
+
+  CHECK(stack_top_is_inner);
+  CHECK(below_inner_is_outer);
+}
+
+
+namespace {
+
+#if defined(USE_SIMULATOR)
+class SimulatorHelper {

[yurys, 2014/09/23 14:57:53]

I don't think we need to pull all these platform-specific code here. Since we
always call GetSample after exiting into the runtime we can pass address of some
C-function, fp and sp values as the register state, they should be ignored by
the stack iterator as it should take top frame from Isolate::c_entry_fp().

[alph, 2014/09/24 13:12:19]

Done.

+ public:
+  inline bool Init(v8::Isolate* isolate) {
+    simulator_ = reinterpret_cast<v8::internal::Isolate*>(isolate)
+                     ->thread_local_top()
+                     ->simulator_;
+    // Check if there is active simulator.
+    return simulator_ != NULL;
+  }
+
+  inline void FillRegisters(v8::RegisterState* state) {
+#if V8_TARGET_ARCH_ARM
+    state->pc = reinterpret_cast<Address>(simulator_->get_pc());
+    state->sp = reinterpret_cast<Address>(
+        simulator_->get_register(v8::internal::Simulator::sp));
+    state->fp = reinterpret_cast<Address>(
+        simulator_->get_register(v8::internal::Simulator::r11));
+#elif V8_TARGET_ARCH_ARM64
+    if (simulator_->sp() == 0 || simulator_->fp() == 0) {
+      // 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.
+      return;
+    }
+    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
+    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_MIPS64
+    state->pc = reinterpret_cast<Address>(simulator_->get_pc());
+    state->sp = reinterpret_cast<Address>(
+        simulator_->get_register(v8::internal::Simulator::sp));
+    state->fp = reinterpret_cast<Address>(
+        simulator_->get_register(v8::internal::Simulator::fp));
+#endif
+  }
+
+ private:
+  v8::internal::Simulator* simulator_;
+};
+#endif  // USE_SIMULATOR
+
+// Platform specific implementation of FillRegisterState
+void FillRegisterState(v8::RegisterState* state) {
+#if defined(USE_SIMULATOR)
+  SimulatorHelper helper;
+  if (!helper.Init(isolate)) return;
+  helper.FillRegisters(state);
+  // 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.
+  if (state->sp == 0 || state->fp == 0) return;
+
+#elif V8_OS_MACOSX
+  // TODO(alph): MacOSX doesn't support getcontext, so do nothing
+  // at the moment. Consider enabling the test by obtaining the
+  // context via POSIX signals.
+  return;
+
+#elif V8_OS_POSIX && !V8_OS_CYGWIN  // defined(USE_SIMULATOR)
+  ucontext_t ucontext;
+  getcontext(&ucontext);
+#if !V8_OS_OPENBSD
+  mcontext_t& mcontext = ucontext.uc_mcontext;
+#endif
+#if V8_OS_LINUX
+#if V8_HOST_ARCH_IA32
+  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<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 defined(__GLIBC__) && !defined(__UCLIBC__) && \
+    (__GLIBC__ < 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ <= 3))
+  // Old GLibc ARM versions used a gregs[] array to access the register
+  // values from mcontext_t.
+  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<void*>(mcontext.arm_pc);
+  state->sp = reinterpret_cast<void*>(mcontext.arm_sp);
+  state->fp = reinterpret_cast<void*>(mcontext.arm_fp);
+#endif  // defined(__GLIBC__) && !defined(__UCLIBC__) &&
+// (__GLIBC__ < 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ <= 3))
+#elif V8_HOST_ARCH_ARM64
+  state->pc = reinterpret_cast<void*>(mcontext.pc);
+  state->sp = reinterpret_cast<void*>(mcontext.sp);
+  // FP is an alias for x29.
+  state->fp = reinterpret_cast<void*>(mcontext.regs[29]);
+#elif V8_HOST_ARCH_MIPS
+  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<void*>(mcontext.pc);
+  state->sp = reinterpret_cast<void*>(mcontext.gregs[29]);
+  state->fp = reinterpret_cast<void*>(mcontext.gregs[30]);
+#endif  // V8_HOST_ARCH_*
+#elif V8_OS_MACOSX
+#if V8_HOST_ARCH_X64
+#if __DARWIN_UNIX03
+  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<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<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<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<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<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<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<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<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<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<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<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<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<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_*
+#endif  // V8_OS_QNX
+
+#elif V8_OS_WIN || V8_OS_CYGWIN  // V8_OS_POSIX && !V8_OS_CYGWIN
+  CONTEXT context;
+  memset(&context, 0, sizeof(context));
+  context.ContextFlags = CONTEXT_FULL;
+  GetThreadContext(OpenThread(THREAD_GET_CONTEXT | THREAD_SUSPEND_RESUME |
+                                  THREAD_QUERY_INFORMATION,
+                              false, GetCurrentThreadId()),
+                   &context);
+#if V8_HOST_ARCH_X64
+  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<void*>(context.Eip);
+  state->sp = reinterpret_cast<void*>(context.Esp);
+  state->fp = reinterpret_cast<void*>(context.Ebp);
+#endif  // V8_HOST_ARCH_X64
+#endif  // V8_OS_WIN || V8_OS_CYGWIN
+}
+
+}  // namespace