Make sure that idle scavenges are just performed when enough objects are allocated in new space.

src/heap/gc-idle-time-handler.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.
 
 #include "src/heap/gc-idle-time-handler.h"
 #include "src/heap/gc-tracer.h"
 #include "src/utils.h"
 
 namespace v8 {
 namespace internal {
@@ skipping 104 matching lines @@
     size_t new_space_allocation_throughput_in_bytes_per_ms) {
   size_t new_space_allocation_limit =
       kMaxScheduledIdleTime * scavenge_speed_in_bytes_per_ms;
 
   // If the limit is larger than the new space size, then scavenging used to be
   // really fast. We can take advantage of the whole new space.
   if (new_space_allocation_limit > new_space_size) {
     new_space_allocation_limit = new_space_size;
   }
 
-  // We do not know the allocation throughput before the first Scavenge.
-  // TODO(hpayer): Estimate allocation throughput before the first Scavenge.
+  // We do not know the allocation throughput before the first scavenge.
+  // TODO(hpayer): Estimate allocation throughput before the first scavenge.
   if (new_space_allocation_throughput_in_bytes_per_ms == 0) {
     new_space_allocation_limit =
         static_cast<size_t>(new_space_size * kConservativeTimeRatio);
   } else {
     // We have to trigger scavenge before we reach the end of new space.
     size_t adjust_limit = new_space_allocation_throughput_in_bytes_per_ms *
                           kTimeUntilNextIdleEvent;
-    if (adjust_limit > new_space_allocation_limit)
+    if (adjust_limit > new_space_allocation_limit) {
       new_space_allocation_limit = 0;
-    else
+    } else {
       new_space_allocation_limit -= adjust_limit;
+    }
+  }
+
+  // The allocated new space limit to trigger a scavange has to be at least
+  // kMinimumNewSpaceSizeToPerformScavenge.
+  if (new_space_allocation_limit < kMinimumNewSpaceSizeToPerformScavenge) {
+    new_space_allocation_limit = kMinimumNewSpaceSizeToPerformScavenge;
   }
 
   if (scavenge_speed_in_bytes_per_ms == 0) {
     scavenge_speed_in_bytes_per_ms = kInitialConservativeScavengeSpeed;
   }
 
   if (new_space_allocation_limit <= used_new_space_size) {
     if (used_new_space_size / scavenge_speed_in_bytes_per_ms <=
         idle_time_in_ms) {
       return true;
@@ skipping 90 matching lines @@
   }
 }
 
 
 // The following logic is implemented by the controller:
 // (1) If we don't have any idle time, do nothing, unless a context was
 // disposed, incremental marking is stopped, and the heap is small. Then do
 // a full GC.
 // (2) If the context disposal rate is high and we cannot perform a full GC,
 // we do nothing until the context disposal rate becomes lower.
-// (3) If the new space is almost full and we can affort a Scavenge or if the
-// next Scavenge will very likely take long, then a Scavenge is performed.
+// (3) If the new space is almost full and we can affort a scavenge or if the
+// next scavenge will very likely take long, then a scavenge is performed.
 // (4) If there is currently no MarkCompact idle round going on, we start a
 // new idle round if enough garbage was created. Otherwise we do not perform
 // garbage collection to keep system utilization low.
 // (5) If incremental marking is done, we perform a full garbage collection
 // if  we are allowed to still do full garbage collections during this idle
 // round or if we are not allowed to start incremental marking. Otherwise we
 // do not perform garbage collection to keep system utilization low.
 // (6) If sweeping is in progress and we received a large enough idle time
 // request, we finalize sweeping here.
 // (7) If incremental marking is in progress, we perform a marking step. Note,
@@ skipping 116 matching lines @@
       }
       if (mutator_gcs > idle_mark_compacts_) {
         return kReduceLatency;
       }
       break;
   }
   return mode_;
 }
 }
 }