Temporary commit to evaluate perf impact of prototype CPU profiler

base/profiler/cpu_profiler_win.cc

Index: base/profiler/cpu_profiler_win.cc
diff --git a/base/profiler/cpu_profiler_win.cc b/base/profiler/cpu_profiler_win.cc
new file mode 100644
index 0000000000000000000000000000000000000000..aa1947ad004af740c13ec8825c8a71a80c8c3280
--- /dev/null
+++ b/base/profiler/cpu_profiler_win.cc
@@ -0,0 +1,178 @@
+// Copyright 2015 The Chromium 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 "base/profiler/cpu_profiler.h"
+
+#include <Tlhelp32.h>
+#include <dbghelp.h>
+#include <stddef.h>
+#include <windows.h>
+
+#include "base/debug/stack_trace.h"
+#include "base/threading/platform_thread.h"
+#include "base/threading/thread_id_name_manager.h"
+#include "base/time/time.h"
+
+namespace base {
+
+namespace {
+
+const DWORD kSuspendFailed = static_cast<DWORD>(-1);
+
+int StackTrace64(CONTEXT* context, int stack_depth,
+                 StackTraceEntry* stack_trace) {
+#ifdef _WIN64
+  IMAGEHLP_SYMBOL64 sym;
+  sym.SizeOfStruct  = sizeof(sym);
+  sym.MaxNameLength = 0;
+
+  for (int i = 0; i < stack_depth; ++i, ++stack_trace) {
+    // Try to look up unwind metadata for the current function.
+    ULONG64 image_base;
+    PRUNTIME_FUNCTION runtime_function =
+        RtlLookupFunctionEntry(context->Rip, &image_base, nullptr);
+
+    stack_trace->rsp = context->Rsp;
+    stack_trace->rip = context->Rip;
+    stack_trace->module = nullptr;
+
+    if (runtime_function) {
+      HMODULE module = NULL;
+      if (GetModuleHandleEx(GET_MODULE_HANDLE_EX_FLAG_FROM_ADDRESS,

[cpu_(ooo_6.6-7.5), 2015/02/05 01:03:26]

I think or-ing both should work GET_MODULE_HANDLE_EX_UNCHANGED_REFCOUNT |
GET_MODULE_HANDLE_EX_FLAG_FROM_ADDRESS

does it not?

[Mike Wittman, 2015/02/05 02:08:48]

It seems that specifying GET_MODULE_HANDLE_EX_UNCHANGED_REFCOUNT either with or
without GET_MODULE_HANDLE_EX_FLAG_FROM_ADDRESS results in deadlocking the UI
thread on startup, typically within a memory allocation or deallocation.

I'll move the module lookup outside of the SuspendThread/ResumeThread section to
avoid this entirely.

+              reinterpret_cast<LPCTSTR>(context->Rip), &module)) {
+        stack_trace->module = module;
+        FreeLibrary(module);
+      }
+
+      KNONVOLATILE_CONTEXT_POINTERS nvcontext = {0};
+      PVOID handler_data;

[cpu_(ooo_6.6-7.5), 2015/02/05 01:03:27]

pvoid --> void*

[Mike Wittman, 2015/02/05 02:08:48]

Done.

+      ULONG64 establisher_frame;
+      RtlVirtualUnwind(0, image_base, context->Rip, runtime_function, context,
+          &handler_data, &establisher_frame, &nvcontext);
+    } else {
+      // If we don't have a RUNTIME_FUNCTION, then we've encountered
+      // a leaf function.  Adjust the stack appropriately.
+      context->Rip = *reinterpret_cast<PDWORD64>(context->Rsp);
+      context->Rsp += 8;
+    }
+
+    if (!context->Rip)
+      return i;
+  }
+  return stack_depth;
+#else
+  return 0;
+#endif
+}
+
+int SampleThread(HANDLE thread_handle, int max_stack_size,
+                 StackTraceEntry* stack) {
+  if (!thread_handle)

[cpu_(ooo_6.6-7.5), 2015/02/05 01:03:26]

can we not fire the timer when you don't have a thread handle?

[Mike Wittman, 2015/02/05 02:08:47]

We should always have a thread handle on x64 since we now get it from the main
thread on construction of CpuProfiler. On x86 we never instantiate the
SamplingThread.

Removed the conditional.

+    return 0;
+
+  if (::SuspendThread(thread_handle) == kSuspendFailed) {
+    LOG(ERROR) << "SuspendThread failed: " << GetLastError();
+    return 0;
+  }
+
+  CONTEXT thread_context = {0};
+  thread_context.ContextFlags = CONTEXT_ALL;
+  if (!::GetThreadContext(thread_handle, &thread_context)) {
+    LOG(ERROR) << "GetThreadContext failed: " << GetLastError();
+  }
+
+  int stack_depth = StackTrace64(&thread_context, max_stack_size, stack);
+
+  if (::ResumeThread(thread_handle) == kSuspendFailed) {
+    LOG(ERROR) << "ResumeThread failed: " << GetLastError();
+  }
+  return stack_depth;
+}
+
+void GetNames(StackTraceEntry* stack_trace, int stack_depth) {
+  HANDLE process = GetCurrentProcess();
+  DWORD64 displacement = 0;
+  DWORD line_displacement = 0;
+  char buffer[sizeof(SYMBOL_INFO) + MAX_SYM_NAME * sizeof(wchar_t)];
+  PSYMBOL_INFO symbol_info = reinterpret_cast<PSYMBOL_INFO>(buffer);

[cpu_(ooo_6.6-7.5), 2015/02/05 01:03:27]

psymbol_info -->symbol_info*

[Mike Wittman, 2015/02/05 02:08:48]

Done.

+  symbol_info->SizeOfStruct = sizeof(SYMBOL_INFO);
+  symbol_info->MaxNameLen = MAX_SYM_NAME;
+  std::string file_name;
+
+  IMAGEHLP_LINE64 line;
+  line.SizeOfStruct = sizeof(IMAGEHLP_LINE64);
+
+  for (int i = 0; i < stack_depth; ++i, ++stack_trace) {
+    DWORD64 address = stack_trace->rip;
+    if (!SymFromAddr(process, address, &displacement, symbol_info)) {
+      wcscpy(reinterpret_cast<wchar_t*>(symbol_info->Name), L"failed");

[cpu_(ooo_6.6-7.5), 2015/02/05 01:03:26]

use the safe versions wcscpy_s

[Mike Wittman, 2015/02/05 02:08:48]

Done.

+    }
+    if (!SymGetLineFromAddr64(process, address, &line_displacement, &line)) {
+      line.LineNumber = 0;
+      file_name.clear();
+    } else {
+      file_name.assign(line.FileName);
+    }
+  }
+}
+
+}
+
+CpuProfiler::CpuProfiler()
+    : main_thread_(0),
+      main_thread_stack_depth_(0) {
+#ifdef _WIN64
+  // Record the main thread's handle.
+  BOOL result = ::DuplicateHandle(::GetCurrentProcess(), ::GetCurrentThread(),
+                                  ::GetCurrentProcess(), &main_thread_, 0,
+                                  FALSE, DUPLICATE_SAME_ACCESS);
+  DCHECK(result) << "DuplicateHandle failed";
+
+  // Initialization
+  if (RtlVirtualUnwind == NULL && RtlLookupFunctionEntry == NULL) {
+    const HMODULE nt_dll_handle = ::GetModuleHandle(L"ntdll.dll");
+    reinterpret_cast<void*&>(RtlVirtualUnwind) =
+        ::GetProcAddress(nt_dll_handle, "RtlVirtualUnwind");
+    reinterpret_cast<void*&>(RtlLookupFunctionEntry) =
+        ::GetProcAddress(nt_dll_handle, "RtlLookupFunctionEntry");
+  }
+#endif
+}
+
+CpuProfiler::~CpuProfiler() {
+  CloseHandle(main_thread_);
+}
+
+// static
+bool CpuProfiler::IsPlatformSupported() {
+#ifdef _WIN64
+  return true;
+#else
+  return false;
+#endif
+}
+
+void CpuProfiler::OnTimer() {
+  main_thread_stack_depth_ = SampleThread(
+      main_thread_, arraysize(main_thread_stack_), main_thread_stack_);
+  ProcessStack(main_thread_stack_, main_thread_stack_depth_);
+
+  modules_.clear();
+  main_thread_stack_depth_ = 0;
+}
+
+void CpuProfiler::ProcessStack(StackTraceEntry* stack, int stack_depth) {
+  for (int i = 0; i < stack_depth; ++i, ++stack) {
+    HMODULE module = stack->module;
+    if (!module || (modules_.find(module) != modules_.end()))
+      continue;
+
+    wchar_t module_name[MAX_PATH];
+    if (GetModuleFileName(module, module_name, arraysize(module_name))) {
+      modules_.insert(std::pair<HMODULE, base::string16>(module, module_name));
+    }
+  }
+}
+
+}  // namespace base