diff --git a/tensorflow/core/common_runtime/direct_session.cc b/tensorflow/core/common_runtime/direct_session.cc
index a1bbcde94bd..3a49f6f3232 100644
--- a/tensorflow/core/common_runtime/direct_session.cc
+++ b/tensorflow/core/common_runtime/direct_session.cc
@@ -183,7 +183,7 @@ class DirectSessionFactory : public SessionFactory {
 
     // Must do this before the CPU allocator is created.
     if (options.config.graph_options().build_cost_model() > 0) {
-      EnableCPUAllocatorFullStats(true);
+      EnableCPUAllocatorFullStats();
     }
     std::vector<std::unique_ptr<Device>> devices;
     TF_RETURN_IF_ERROR(DeviceFactory::AddDevices(
diff --git a/tensorflow/core/framework/allocator.cc b/tensorflow/core/framework/allocator.cc
index d20f779c8da..d032276d0f7 100644
--- a/tensorflow/core/framework/allocator.cc
+++ b/tensorflow/core/framework/allocator.cc
@@ -56,9 +56,7 @@ Allocator::~Allocator() {}
 // If true, cpu allocator collects full stats.
 static bool cpu_allocator_collect_full_stats = false;
 
-void EnableCPUAllocatorFullStats(bool enable) {
-  cpu_allocator_collect_full_stats = enable;
-}
+void EnableCPUAllocatorFullStats() { cpu_allocator_collect_full_stats = true; }
 bool CPUAllocatorFullStatsEnabled() { return cpu_allocator_collect_full_stats; }
 
 string AllocatorAttributes::DebugString() const {
diff --git a/tensorflow/core/framework/allocator.h b/tensorflow/core/framework/allocator.h
index 7b8eba0fda9..f7402f7b293 100644
--- a/tensorflow/core/framework/allocator.h
+++ b/tensorflow/core/framework/allocator.h
@@ -410,14 +410,17 @@ Allocator* cpu_allocator_base();
 // call it directly.
 Allocator* cpu_allocator(int numa_node = port::kNUMANoAffinity);
 
-// If 'enable' is true, the default CPU allocator implementation will collect
-// AllocatorStats. By default, it's disabled.
-void EnableCPUAllocatorStats(bool enable);
+// Enables AllocatorStats in the default CPU allocator implementation.  By
+// default, it's disabled.
+void EnableCPUAllocatorStats();
+// Disables AllocatorStats in the default CPU allocator implementation.  By
+// default, it's disabled.
+void DisableCPUAllocatorStats();
 bool CPUAllocatorStatsEnabled();
 
-// If 'enable' is true, the default CPU allocator implementation will collect
-// full statistics. By default, it's disabled.
-void EnableCPUAllocatorFullStats(bool enable);
+// Enables full statistics collection in the default CPU allocator
+// implementation.  By default, it's disabled.
+void EnableCPUAllocatorFullStats();
 bool CPUAllocatorFullStatsEnabled();
 
 // An object that does the underlying suballoc/free of memory for a higher-level
diff --git a/tensorflow/core/framework/allocator_test.cc b/tensorflow/core/framework/allocator_test.cc
index 3caab02eeba..0ac3da1a19c 100644
--- a/tensorflow/core/framework/allocator_test.cc
+++ b/tensorflow/core/framework/allocator_test.cc
@@ -133,7 +133,7 @@ TEST(AllocatorAttributesDeathTest, MergeDifferentScopeIds) {
 }
 
 TEST(CPUAllocatorTest, Simple) {
-  EnableCPUAllocatorStats(true);
+  EnableCPUAllocatorStats();
   Allocator* a = cpu_allocator();
   std::vector<void*> ptrs;
   for (int s = 1; s < 1024; s++) {
@@ -162,7 +162,7 @@ TEST(CPUAllocatorTest, Simple) {
              1048576 * sizeof(double));
   a->ClearStats();
   CheckStats(a, 0, 0, 0, 0);
-  EnableCPUAllocatorStats(false);
+  DisableCPUAllocatorStats();
 }
 
 // Define a struct that we will use to observe behavior in the unit tests
@@ -227,13 +227,13 @@ static void BM_Allocation(int iters, int arg) {
   std::vector<int> sizes = {256, 4096, 16384, 524288, 512, 1048576};
   int size_index = 0;
 
-  if (arg) EnableCPUAllocatorStats(true);
+  if (arg) EnableCPUAllocatorStats();
   while (--iters > 0) {
     int bytes = sizes[size_index++ % sizes.size()];
     void* p = a->AllocateRaw(1, bytes);
     a->DeallocateRaw(p);
   }
-  if (arg) EnableCPUAllocatorStats(false);
+  if (arg) DisableCPUAllocatorStats();
 }
 BENCHMARK(BM_Allocation)->Arg(0)->Arg(1);
 
diff --git a/tensorflow/core/framework/cpu_allocator_impl.cc b/tensorflow/core/framework/cpu_allocator_impl.cc
index 814233074fb..511cfce8ab5 100644
--- a/tensorflow/core/framework/cpu_allocator_impl.cc
+++ b/tensorflow/core/framework/cpu_allocator_impl.cc
@@ -29,9 +29,8 @@ namespace tensorflow {
 // If true, cpu allocator collects more stats.
 static bool cpu_allocator_collect_stats = false;
 
-void EnableCPUAllocatorStats(bool enable) {
-  cpu_allocator_collect_stats = enable;
-}
+void EnableCPUAllocatorStats() { cpu_allocator_collect_stats = true; }
+void DisableCPUAllocatorStats() { cpu_allocator_collect_stats = false; }
 bool CPUAllocatorStatsEnabled() { return cpu_allocator_collect_stats; }
 
 static const int kMaxTotalAllocationWarnings = 1;
diff --git a/tensorflow/core/grappler/clusters/cluster.h b/tensorflow/core/grappler/clusters/cluster.h
index 0f415cf0392..d391e15de1a 100644
--- a/tensorflow/core/grappler/clusters/cluster.h
+++ b/tensorflow/core/grappler/clusters/cluster.h
@@ -103,9 +103,9 @@ class Cluster {
   // superset of the devices listed in GetDevices/GetDeviceNames().
   virtual const DeviceSet* GetDeviceSet() const { return nullptr; }
 
-  // Enables collecting the allocator stats. Call with enable=true must be made
-  // before Provision().
-  virtual Status EnablePeakMemoryStats(bool enable) {
+  // Enables collecting the allocator stats. If called, must be called before
+  // Provision().
+  virtual Status EnablePeakMemoryStats() {
     return errors::Unimplemented(strings ::StrCat(
         "Peak Memory Stats are not supported on ", type(), " clusters"));
   }
diff --git a/tensorflow/core/grappler/clusters/single_machine.cc b/tensorflow/core/grappler/clusters/single_machine.cc
index c44b74efcdc..678daed02e4 100644
--- a/tensorflow/core/grappler/clusters/single_machine.cc
+++ b/tensorflow/core/grappler/clusters/single_machine.cc
@@ -202,9 +202,9 @@ Status SingleMachine::Run(const GraphDef& graph_def,
   return Status::OK();
 }
 
-Status SingleMachine::EnablePeakMemoryStats(bool enable) {
-  EnableCPUAllocatorStats(enable);
-  cpu_allocator_stats_enabled_ = enable;
+Status SingleMachine::EnablePeakMemoryStats() {
+  EnableCPUAllocatorStats();
+  cpu_allocator_stats_enabled_ = true;
   // No need to enable GPU allocator stats since its stats are always collected.
   return Status::OK();
 }
diff --git a/tensorflow/core/grappler/clusters/single_machine.h b/tensorflow/core/grappler/clusters/single_machine.h
index 9e085d161b6..48f56940ec4 100644
--- a/tensorflow/core/grappler/clusters/single_machine.h
+++ b/tensorflow/core/grappler/clusters/single_machine.h
@@ -45,7 +45,7 @@ class SingleMachine : public Cluster {
 
   const DeviceSet* GetDeviceSet() const override { return device_set_.get(); }
 
-  Status EnablePeakMemoryStats(bool enable) override;
+  Status EnablePeakMemoryStats() override;
 
   // It requires EnableAllocatorStats(true) be called before Provision().
   Status GetPeakMemoryUsage(
diff --git a/tensorflow/core/grappler/clusters/single_machine_test.cc b/tensorflow/core/grappler/clusters/single_machine_test.cc
index a85e4e17748..d0d525e0222 100644
--- a/tensorflow/core/grappler/clusters/single_machine_test.cc
+++ b/tensorflow/core/grappler/clusters/single_machine_test.cc
@@ -51,7 +51,7 @@ class SingleMachineTest : public ::testing::Test {
 #endif
     cluster_.reset(
         new SingleMachine(timeout_s, 3 /* num_cpu_cores */, 0 /* num_gpus */));
-    TF_CHECK_OK(cluster_->EnablePeakMemoryStats(true));
+    TF_CHECK_OK(cluster_->EnablePeakMemoryStats());
     TF_CHECK_OK(cluster_->Provision());
   }
 
diff --git a/tensorflow/core/kernels/hexagon/hexagon_graph_execution_test.cc b/tensorflow/core/kernels/hexagon/hexagon_graph_execution_test.cc
index 7a6924e2ebf..461fb7deb78 100644
--- a/tensorflow/core/kernels/hexagon/hexagon_graph_execution_test.cc
+++ b/tensorflow/core/kernels/hexagon/hexagon_graph_execution_test.cc
@@ -414,7 +414,7 @@ TEST(GraphTransferer,
 
   GraphTransferer gt;
   gt.EnableStrictCheckMode(false);
-  profile_utils::CpuUtils::EnableClockCycleProfiling(true);
+  profile_utils::CpuUtils::EnableClockCycleProfiling();
   ClockCycleProfiler prof;
   prof.Start();
   Status status = gt.LoadGraphFromProtoFile(
@@ -447,7 +447,7 @@ TEST(GraphTransferer,
 
   GraphTransferer gt;
   gt.EnableStrictCheckMode(false);
-  profile_utils::CpuUtils::EnableClockCycleProfiling(true);
+  profile_utils::CpuUtils::EnableClockCycleProfiling();
   ClockCycleProfiler prof;
   prof.Start();
   Status status = gt.LoadGraphFromProtoFile(
@@ -481,7 +481,7 @@ TEST(GraphTransferer,
 
   GraphTransferer gt;
   gt.EnableStrictCheckMode(false);
-  profile_utils::CpuUtils::EnableClockCycleProfiling(true);
+  profile_utils::CpuUtils::EnableClockCycleProfiling();
   ClockCycleProfiler prof;
   prof.Start();
   Status status = gt.LoadGraphFromProtoFile(
@@ -540,7 +540,7 @@ TEST(GraphTransferer, DISABLED_RunInceptionV3OnHexagonExampleWithFusedGraph) {
 
 TEST(GraphTransferer, DISABLED_CheckShapeInferencePerformance) {
   CheckHexagonControllerVersion();
-  profile_utils::CpuUtils::EnableClockCycleProfiling(true);
+  profile_utils::CpuUtils::EnableClockCycleProfiling();
 
   const IRemoteFusedGraphOpsDefinitions* ops_definitions =
       &HexagonOpsDefinitions::getInstance();
diff --git a/tensorflow/core/platform/profile_utils/android_armv7a_cpu_utils_helper.cc b/tensorflow/core/platform/profile_utils/android_armv7a_cpu_utils_helper.cc
index 6dc1826d93b..d61a036181d 100644
--- a/tensorflow/core/platform/profile_utils/android_armv7a_cpu_utils_helper.cc
+++ b/tensorflow/core/platform/profile_utils/android_armv7a_cpu_utils_helper.cc
@@ -54,12 +54,11 @@ uint64 AndroidArmV7ACpuUtilsHelper::GetCurrentClockCycle() {
   return static_cast<uint64>(count);
 }
 
-void AndroidArmV7ACpuUtilsHelper::EnableClockCycleProfiling(const bool enable) {
+void AndroidArmV7ACpuUtilsHelper::EnableClockCycleProfiling() {
   if (!is_initialized_) {
     // Initialize here to avoid unnecessary initialization
     InitializeInternal();
   }
-  if (enable) {
     const int64 cpu0_scaling_min = ReadCpuFrequencyFile(0, "scaling_min");
     const int64 cpu0_scaling_max = ReadCpuFrequencyFile(0, "scaling_max");
     if (cpu0_scaling_max != cpu0_scaling_min) {
@@ -69,9 +68,14 @@ void AndroidArmV7ACpuUtilsHelper::EnableClockCycleProfiling(const bool enable) {
     }
     ResetClockCycle();
     ioctl(fd_, PERF_EVENT_IOC_ENABLE, 0);
-  } else {
-    ioctl(fd_, PERF_EVENT_IOC_DISABLE, 0);
+}
+
+void AndroidArmV7ACpuUtilsHelper::DisableClockCycleProfiling() {
+  if (!is_initialized_) {
+    // Initialize here to avoid unnecessary initialization
+    InitializeInternal();
   }
+  ioctl(fd_, PERF_EVENT_IOC_DISABLE, 0);
 }
 
 int64 AndroidArmV7ACpuUtilsHelper::CalculateCpuFrequency() {
diff --git a/tensorflow/core/platform/profile_utils/android_armv7a_cpu_utils_helper.h b/tensorflow/core/platform/profile_utils/android_armv7a_cpu_utils_helper.h
index 2d94736c978..66bc0fd5928 100644
--- a/tensorflow/core/platform/profile_utils/android_armv7a_cpu_utils_helper.h
+++ b/tensorflow/core/platform/profile_utils/android_armv7a_cpu_utils_helper.h
@@ -36,7 +36,8 @@ class AndroidArmV7ACpuUtilsHelper : public ICpuUtilsHelper {
   AndroidArmV7ACpuUtilsHelper() = default;
   void ResetClockCycle() final;
   uint64 GetCurrentClockCycle() final;
-  void EnableClockCycleProfiling(bool enable) final;
+  void EnableClockCycleProfiling() final;
+  void DisableClockCycleProfiling() final;
   int64 CalculateCpuFrequency() final;
 
  private:
diff --git a/tensorflow/core/platform/profile_utils/cpu_utils.cc b/tensorflow/core/platform/profile_utils/cpu_utils.cc
index b22123a804a..7cd1c4de88f 100644
--- a/tensorflow/core/platform/profile_utils/cpu_utils.cc
+++ b/tensorflow/core/platform/profile_utils/cpu_utils.cc
@@ -58,8 +58,12 @@ static ICpuUtilsHelper* cpu_utils_helper_instance_ = nullptr;
   GetCpuUtilsHelperSingletonInstance().ResetClockCycle();
 }
 
-/* static */ void CpuUtils::EnableClockCycleProfiling(const bool enable) {
-  GetCpuUtilsHelperSingletonInstance().EnableClockCycleProfiling(enable);
+/* static */ void CpuUtils::EnableClockCycleProfiling() {
+  GetCpuUtilsHelperSingletonInstance().EnableClockCycleProfiling();
+}
+
+/* static */ void CpuUtils::DisableClockCycleProfiling() {
+  GetCpuUtilsHelperSingletonInstance().DisableClockCycleProfiling();
 }
 
 /* static */ std::chrono::duration<double> CpuUtils::ConvertClockCycleToTime(
diff --git a/tensorflow/core/platform/profile_utils/cpu_utils.h b/tensorflow/core/platform/profile_utils/cpu_utils.h
index d26f28478a5..1132c485f90 100644
--- a/tensorflow/core/platform/profile_utils/cpu_utils.h
+++ b/tensorflow/core/platform/profile_utils/cpu_utils.h
@@ -138,9 +138,10 @@ class CpuUtils {
   // clock cycle counters from overflowing on some platforms.
   static void ResetClockCycle();
 
-  // Enable clock cycle profile
+  // Enable/Disable clock cycle profile
   // You can enable / disable profile if it's supported by the platform
-  static void EnableClockCycleProfiling(bool enable);
+  static void EnableClockCycleProfiling();
+  static void DisableClockCycleProfiling();
 
   // Return chrono::duration per each clock
   static std::chrono::duration<double> ConvertClockCycleToTime(
@@ -152,7 +153,8 @@ class CpuUtils {
     DefaultCpuUtilsHelper() = default;
     void ResetClockCycle() final {}
     uint64 GetCurrentClockCycle() final { return DUMMY_CYCLE_CLOCK; }
-    void EnableClockCycleProfiling(bool /* enable */) final {}
+    void EnableClockCycleProfiling() final {}
+    void DisableClockCycleProfiling() final {}
     int64 CalculateCpuFrequency() final { return INVALID_FREQUENCY; }
 
    private:
diff --git a/tensorflow/core/platform/profile_utils/cpu_utils_test.cc b/tensorflow/core/platform/profile_utils/cpu_utils_test.cc
index eb8161fbfd5..a18561a1156 100644
--- a/tensorflow/core/platform/profile_utils/cpu_utils_test.cc
+++ b/tensorflow/core/platform/profile_utils/cpu_utils_test.cc
@@ -26,7 +26,7 @@ static constexpr bool DBG = false;
 
 class CpuUtilsTest : public ::testing::Test {
  protected:
-  void SetUp() override { CpuUtils::EnableClockCycleProfiling(true); }
+  void SetUp() override { CpuUtils::EnableClockCycleProfiling(); }
 };
 
 TEST_F(CpuUtilsTest, SetUpTestCase) {}
diff --git a/tensorflow/core/platform/profile_utils/i_cpu_utils_helper.h b/tensorflow/core/platform/profile_utils/i_cpu_utils_helper.h
index cab7618a70a..bd63ffd0e85 100644
--- a/tensorflow/core/platform/profile_utils/i_cpu_utils_helper.h
+++ b/tensorflow/core/platform/profile_utils/i_cpu_utils_helper.h
@@ -35,9 +35,10 @@ class ICpuUtilsHelper {
   virtual void ResetClockCycle() = 0;
   // Return current clock cycle.
   virtual uint64 GetCurrentClockCycle() = 0;
-  // Enable clock cycle profile
+  // Enable/Disable clock cycle profile
   // You can enable / disable profile if it's supported by the platform
-  virtual void EnableClockCycleProfiling(bool enable) = 0;
+  virtual void EnableClockCycleProfiling() = 0;
+  virtual void DisableClockCycleProfiling() = 0;
   // Return cpu frequency.
   // CAVEAT: as this method may read file and/or call system calls,
   // this call is supposed to be slow.