Add UMA stats for hard-fault counts for Windows 7 and greater.

components/startup_metric_utils/startup_metric_utils.cc

Index: components/startup_metric_utils/startup_metric_utils.cc
diff --git a/components/startup_metric_utils/startup_metric_utils.cc b/components/startup_metric_utils/startup_metric_utils.cc
index 642a0c488823b675c2468a2098e0277227823212..dd108fba2f132f2ebadeb04cbf9002ca1e299a0a 100644
--- a/components/startup_metric_utils/startup_metric_utils.cc
+++ b/components/startup_metric_utils/startup_metric_utils.cc
@@ -16,6 +16,10 @@
 #include "base/sys_info.h"
 #include "base/time/time.h"
 
+#if defined(OS_WIN)
+#include <winternl.h>
+#endif
+
 namespace {
 
 // Mark as volatile to defensively make sure usage is thread-safe.
@@ -40,6 +44,148 @@ base::Lock* GetSubsystemStartupTimeHashLock() {
   return slow_startup_time_hash_lock;
 }
 
+// The struct used to return system process information via the NT internal
+// QuerySystemInformation call. This is partially documented at
+// http://goo.gl/Ja9MrH and fully documented at http://goo.gl/QJ70rn
+struct SYSTEM_PROCESS_INFORMATION_EX {
+  ULONG NextEntryOffset;
+  ULONG NumberOfThreads;
+  LARGE_INTEGER WorkingSetPrivateSize;
+  ULONG HardFaultCount;
+  BYTE Reserved1[36];
+  PVOID Reserved2[3];
+  ULONG UniqueProcessId;
+  PVOID Reserved3;
+  ULONG HandleCount;
+  BYTE Reserved4[4];
+  PVOID Reserved5[11];
+  SIZE_T PeakPagefileUsage;
+  SIZE_T PrivatePageCount;
+  LARGE_INTEGER Reserved6[6];
+  // Array of SYSTEM_THREAD_INFORMATION structs follows.
+};
+static_assert(
+    sizeof(SYSTEM_PROCESS_INFORMATION_EX) == 184,

[Alexei Svitkine (slow), 2015/04/14 16:53:51]

Is the size the same between 32 and 64 bit?

[chrisha, 2015/04/14 17:09:07]

Good catch. Not the same size, but has the same layout on 64-bit. Removed the
static_assert, as it was mostly just for me when I was writing this to ensure
that I had the right definition.

+    "The definition of SYSTEM_PROCESS_INFORMATION_EX does not match the "
+    "expected size.");
+
+// The signature of the NtQuerySystemInformation function.
+typedef NTSTATUS (WINAPI *NtQuerySystemInformationPtr)(
+    SYSTEM_INFORMATION_CLASS, PVOID, ULONG, PULONG);
+
+// Gets the hard fault count of the current process, returning it via
+// |hard_fault_count|. Returns true on success, false otherwise. Also returns
+// whether or not the system call was even possible for the current OS version
+// via |has_os_support|.
+bool GetHardFaultCountForCurrentProcess(uint32* hard_fault_count,
+                                        bool* has_os_support) {
+  DCHECK(hard_fault_count);
+  DCHECK(has_os_support);
+
+  *has_os_support = false;
+  OSVERSIONINFOEX version_info = {};
+  version_info.dwOSVersionInfoSize = sizeof(OSVERSIONINFOEX);
+  if (!::GetVersionEx(reinterpret_cast<LPOSVERSIONINFOW>(&version_info)))

[Alexei Svitkine (slow), 2015/04/14 16:53:51]

Is this inside a file that's Windows-specific? If not, you probably need more
ifdefs. (or move to a separate file)

+    return false;
+
+  // The hard fault counter is only available as of Windows 7 (version 6.1).
+  if (version_info.dwMajorVersion < 6)
+    return false;
+  if (version_info.dwMajorVersion == 6 && version_info.dwMinorVersion < 1)
+    return false;
+
+  // At this point the OS supports the required system call.
+  *has_os_support = true;
+
+  // Get the function pointer.
+  NtQuerySystemInformationPtr query_sys_info =
+      reinterpret_cast<NtQuerySystemInformationPtr>(
+          ::GetProcAddress(GetModuleHandle(L"ntdll.dll"),
+                           "NtQuerySystemInformation"));
+  if (query_sys_info == nullptr)
+    return false;
+
+  // The output of this system call depends on the number of threads and
+  // processes on the entire system, and this can change between calls. Retry
+  // a small handful of times growing the buffer along the way.
+  std::vector<uint8> buffer(1024);
+  for (size_t tries = 0; tries < 3; ++tries) {
+    ULONG return_length = 0;
+    NTSTATUS status = query_sys_info(
+        SystemProcessInformation,
+        buffer.data(),
+        buffer.size(),
+        &return_length);
+    // Insufficient space in the buffer.
+    if (return_length > buffer.size()) {
+      buffer.resize(return_length);
+      continue;
+    }
+    if (status == 0 && return_length <= buffer.size())
+      break;
+    return false;
+  }
+  // The only way out to get here is if the system call was successful.
+
+  // Look for the struct housing information for the current process.
+  DWORD proc_id = ::GetCurrentProcessId();
+  size_t index = 0;
+  while (index < buffer.size()) {
+    SYSTEM_PROCESS_INFORMATION_EX* proc_info =
+        reinterpret_cast<SYSTEM_PROCESS_INFORMATION_EX*>(buffer.data() + index);
+    if (proc_info->UniqueProcessId == proc_id) {
+      *hard_fault_count = proc_info->HardFaultCount;
+      return true;
+    }
+    // The list ends with NextEntryOffset is zero. This also prevents busy
+    // looping if the data is in fact invalid.
+    if (proc_info->NextEntryOffset == 0)
+      return false;
+    index += proc_info->NextEntryOffset;
+  }
+
+  return false;
+}
+
+// On Windows, records the number of hard-faults that have occurred in the
+// current chrome.exe process since it was started.
+// crbug.com/476923
+// TODO(chrisha): If this proves useful, use it to split startup stats in two.
+void RecordHardFaultHistogram(bool is_first_run) {
+  uint32 hard_fault_count = 0;
+  bool has_os_support = false;
+  bool success = GetHardFaultCountForCurrentProcess(
+      &hard_fault_count, &has_os_support);
+
+  // Log whether or not the system call was successful, assuming the OS was
+  // detected to support it.
+  if (has_os_support) {
+    UMA_HISTOGRAM_BOOLEAN(
+        "Startup.BrowserMessageLoopStartHardFaultCount.Success",
+        success);
+  }
+
+  // Don't log a histogram value if unable to get the hard fault count.
+  if (!success)
+    return;
+
+  // Hard fault counts are expected to be in the thousands range,
+  // corresponding to faulting in ~10s of MBs of code ~10s of KBs at a time.
+  // (Observed to vary from 1000 to 10000 on various test machines and
+  // platforms.)
+  if (is_first_run) {
+    UMA_HISTOGRAM_CUSTOM_COUNTS(
+        "Startup.BrowserMessageLoopStartHardFaultCount.FirstRun",
+        hard_fault_count,
+        0, 20000, 50);
+  } else {
+    UMA_HISTOGRAM_CUSTOM_COUNTS(
+        "Startup.BrowserMessageLoopStartHardFaultCount",
+        hard_fault_count,
+        0, 20000, 50);
+  }
+}
+
 // Record time of main entry so it can be read from Telemetry performance
 // tests.
 // TODO(jeremy): Remove once crbug.com/317481 is fixed.
@@ -115,6 +261,7 @@ const base::Time MainEntryStartTime() {
 }
 
 void OnBrowserStartupComplete(bool is_first_run) {
+  RecordHardFaultHistogram(is_first_run);
   RecordMainEntryTimeHistogram();
 
   // Bail if uptime < 7 minutes, to filter out cases where Chrome may have been