--- old/src/share/vm/gc/shared/collectedHeap.cpp	2017-06-12 10:25:10.870262634 -0700
+++ new/src/share/vm/gc/shared/collectedHeap.cpp	2017-06-12 10:25:10.750263101 -0700
@@ -38,6 +38,7 @@
 #include "memory/resourceArea.hpp"
 #include "oops/instanceMirrorKlass.hpp"
 #include "oops/oop.inline.hpp"
+#include "runtime/heapMonitoring.hpp"
 #include "runtime/init.hpp"
 #include "runtime/thread.inline.hpp"
 #include "services/heapDumper.hpp"
@@ -296,6 +297,27 @@
 #endif
 
 HeapWord* CollectedHeap::allocate_from_tlab_slow(Klass* klass, Thread* thread, size_t size) {
+  // We can come here for three reasons:
+  //  - We either really did fill the tlab.
+  //  - We pretended to everyone we did and we want to sample.
+  //  - Both of the above reasons are true at the same time.
+  if (HeapMonitoring::enabled()) {
+    if (thread->tlab().should_sample()) {
+      // The tlab could still have space after this sample.
+      thread->tlab().set_back_actual_end();
+
+      // Try to allocate again: it could work now.
+      HeapWord* obj = thread->tlab().allocate(size);
+      if (obj != NULL) {
+        // Object got allocated, sample it now.
+        HeapMonitoring::object_alloc_do_sample(thread,
+                                               reinterpret_cast<oopDesc*>(obj),
+                                               size);
+        thread->tlab().pick_next_sample();
+        return obj;
+      }
+    }
+  }
 
   // Retain tlab and allocate object in shared space if
   // the amount free in the tlab is too large to discard.
@@ -336,6 +358,16 @@
 #endif // ASSERT
   }
   thread->tlab().fill(obj, obj + size, new_tlab_size);
+
+  if (HeapMonitoring::enabled()) {
+    if (thread->tlab().should_sample()) {
+      HeapMonitoring::object_alloc_do_sample(thread,
+                                             reinterpret_cast<oopDesc*>(obj),
+                                             size);
+    }
+    // Always pick a next sample here.
+    thread->tlab().pick_next_sample();
+  }
   return obj;
 }