Bug 1346415 - Collect native stacks at the same time as pseudostacks on nightly, r=mconley

toolkit/components/telemetry/ThreadHangStats.h

@@ -73,9 +73,17 @@ public:
   HangStack(HangStack&& aOther)
     : mImpl(mozilla::Move(aOther.mImpl))
     , mBuffer(mozilla::Move(aOther.mBuffer))
+    , mNativeFrames(mozilla::Move(aOther.mNativeFrames))
   {
   }
 
+  HangStack& operator=(HangStack&& aOther) {
+    mImpl = mozilla::Move(aOther.mImpl);
+    mBuffer = mozilla::Move(aOther.mBuffer);
+    mNativeFrames = mozilla::Move(aOther.mNativeFrames);
+    return *this;
+  }
+
   bool operator==(const HangStack& aOther) const {
     for (size_t i = 0; i < length(); i++) {
       if (!IsSameAsEntry(operator[](i), aOther[i])) {
@@ -231,6 +239,7 @@ private:
 public:
   TimeHistogram mActivity;
   mozilla::Vector<HangHistogram, 4> mHangs;
+  uint32_t mNativeStackCnt;
 
   explicit ThreadHangStats(const char* aName)
     : mName(aName)
@@ -240,6 +249,7 @@ public:
     : mName(mozilla::Move(aOther.mName))
     , mActivity(mozilla::Move(aOther.mActivity))
     , mHangs(mozilla::Move(aOther.mHangs))
+    , mNativeStackCnt(0)
   {
   }
   const char* GetName() const {

xpcom/threads/BackgroundHangMonitor.cpp

@@ -32,6 +32,16 @@
 // It can be scaled back again in the future
 #define BHR_BETA_MOD 1;
 
+// This variable controls the maximum number of native hang stacks which may be
+// attached to a ping. This is due to how large native stacks can be. We want to
+// reduce the chance of a ping being discarded due to it exceeding the maximum
+// ping size.
+//
+// NOTE: 300 native hang stacks would, by a rough estimation based on stacks
+// collected from nightly on March 21, 2017, take up approximately 168kb of
+// space.
+static const uint32_t kMaximumNativeHangStacks = 300;
+
 // Maximum depth of the call stack in the reported thread hangs. This value represents
 // the 99.9th percentile of the thread hangs stack depths reported by Telemetry.
 static const size_t kMaxThreadHangStackDepth = 30;
@@ -180,6 +190,8 @@ public:
   ThreadStackHelper mStackHelper;
   // Stack of current hang
   Telemetry::HangStack mHangStack;
+  // Native stack of current hang
+  Telemetry::HangStack mNativeHangStack;
   // Statistics for telemetry
   Telemetry::ThreadHangStats mStats;
   // Annotations for the current hang
@@ -337,6 +349,16 @@ BackgroundHangManager::RunMonitorThread()
           currentThread->mHanging = true;
           currentThread->mAnnotations =
             currentThread->mAnnotators.GatherAnnotations();
+
+#ifdef NIGHTLY_BUILD
+          // NOTE: In nightly builds of firefox we want to collect native stacks
+          // for all hangs, not just permahangs.
+          currentThread->mStats.mNativeStackCnt += 1;
+          if (currentThread->mStats.mNativeStackCnt <= kMaximumNativeHangStacks) {
+            currentThread->mStackHelper.GetNativeStack(
+              currentThread->mNativeHangStack);
+          }
+#endif
         }
       } else {
         if (MOZ_LIKELY(interval != currentThread->mHangStart)) {
@@ -452,6 +474,7 @@ BackgroundHangThread::ReportHang(PRIntervalTime aHangTime)
   }
 
   Telemetry::HangHistogram newHistogram(Move(mHangStack));
+  newHistogram.GetNativeStack() = Move(mNativeHangStack);
   for (Telemetry::HangHistogram* oldHistogram = mStats.mHangs.begin();
        oldHistogram != mStats.mHangs.end(); oldHistogram++) {
     if (newHistogram == *oldHistogram) {
@@ -477,7 +500,10 @@ BackgroundHangThread::ReportPermaHang()
   Telemetry::HangHistogram& hang = ReportHang(mMaxTimeout);
   Telemetry::HangStack& stack = hang.GetNativeStack();
   if (stack.empty()) {
-    mStackHelper.GetNativeStack(stack);
+    mStats.mNativeStackCnt += 1;
+    if (mStats.mNativeStackCnt <= kMaximumNativeHangStacks) {
+      mStackHelper.GetNativeStack(stack);
+    }
   }
 }