diff --git a/nova/compute/resource_tracker.py b/nova/compute/resource_tracker.py
index 4fdefa3c7f60..a753488a464a 100644
--- a/nova/compute/resource_tracker.py
+++ b/nova/compute/resource_tracker.py
@@ -982,7 +982,7 @@ class ResourceTracker(object):
         # set some initial values, reserve room for host/hypervisor:
         cn.local_gb_used = CONF.reserved_host_disk_mb / 1024
         cn.memory_mb_used = CONF.reserved_host_memory_mb
-        cn.vcpus_used = 0
+        cn.vcpus_used = CONF.reserved_host_cpus
         cn.free_ram_mb = (cn.memory_mb - cn.memory_mb_used)
         cn.free_disk_gb = (cn.local_gb - cn.local_gb_used)
         cn.current_workload = 0
diff --git a/nova/conf/compute.py b/nova/conf/compute.py
index 3075de700f99..355416c2cf8b 100644
--- a/nova/conf/compute.py
+++ b/nova/conf/compute.py
@@ -393,6 +393,20 @@ Possible values:
 
 * Any positive integer representing amount of memory in MB to reserve
   for the host.
+"""),
+    cfg.IntOpt('reserved_host_cpus',
+        default=0,
+        min=0,
+        help="""
+Number of physical CPUs to reserve for the host. The host resources usage is
+reported back to the scheduler continuously from nova-compute running on the
+compute node. To prevent the host CPU from being considered as available,
+this option is used to reserve random pCPU(s) for the host.
+
+Possible values:
+
+* Any positive integer representing number of physical CPUs to reserve
+  for the host.
 """),
 ]
 
diff --git a/nova/scheduler/client/report.py b/nova/scheduler/client/report.py
index bf414fc3984c..87def6493df6 100644
--- a/nova/scheduler/client/report.py
+++ b/nova/scheduler/client/report.py
@@ -108,7 +108,7 @@ def _compute_node_to_inventory_dict(compute_node):
     if compute_node.vcpus > 0:
         result[VCPU] = {
             'total': compute_node.vcpus,
-            'reserved': 0,
+            'reserved': CONF.reserved_host_cpus,
             'min_unit': 1,
             'max_unit': compute_node.vcpus,
             'step_size': 1,
diff --git a/nova/tests/unit/compute/test_resource_tracker.py b/nova/tests/unit/compute/test_resource_tracker.py
index 5be3f99a53c3..b5fdd02b7388 100644
--- a/nova/tests/unit/compute/test_resource_tracker.py
+++ b/nova/tests/unit/compute/test_resource_tracker.py
@@ -453,7 +453,8 @@ class BaseTestCase(test.NoDBTestCase):
         self.rt = None
         self.flags(my_ip='1.1.1.1',
                    reserved_host_disk_mb=0,
-                   reserved_host_memory_mb=0)
+                   reserved_host_memory_mb=0,
+                   reserved_host_cpus=0)
 
     def _setup_rt(self, virt_resources=_VIRT_DRIVER_AVAIL_RESOURCES,
                   estimate_overhead=overhead_zero):
@@ -564,11 +565,12 @@ class TestUpdateAvailableResources(BaseTestCase):
     @mock.patch('nova.objects.ComputeNode.get_by_host_and_nodename')
     @mock.patch('nova.objects.MigrationList.get_in_progress_by_host_and_node')
     @mock.patch('nova.objects.InstanceList.get_by_host_and_node')
-    def test_no_instances_no_migrations_reserved_disk_and_ram(
+    def test_no_instances_no_migrations_reserved_disk_ram_and_cpu(
             self, get_mock, migr_mock, get_cn_mock, pci_mock,
             instance_pci_mock):
         self.flags(reserved_host_disk_mb=1024,
-                   reserved_host_memory_mb=512)
+                   reserved_host_memory_mb=512,
+                   reserved_host_cpus=1)
         self._setup_rt()
 
         get_mock.return_value = []
@@ -585,7 +587,7 @@ class TestUpdateAvailableResources(BaseTestCase):
             'local_gb': 6,
             'free_ram_mb': 0,  # 512MB avail - 512MB reserved
             'memory_mb_used': 512,  # 0MB used + 512MB reserved
-            'vcpus_used': 0,
+            'vcpus_used': 1,
             'local_gb_used': 1,  # 0GB used + 1 GB reserved
             'memory_mb': 512,
             'current_workload': 0,
diff --git a/nova/tests/unit/scheduler/client/test_report.py b/nova/tests/unit/scheduler/client/test_report.py
index 38a7f5e4244c..3dfcad3b802a 100644
--- a/nova/tests/unit/scheduler/client/test_report.py
+++ b/nova/tests/unit/scheduler/client/test_report.py
@@ -592,13 +592,14 @@ class TestComputeNodeToInventoryDict(test.NoDBTestCase):
 
         self.flags(reserved_host_memory_mb=1000)
         self.flags(reserved_host_disk_mb=200)
+        self.flags(reserved_host_cpus=1)
 
         result = report._compute_node_to_inventory_dict(compute_node)
 
         expected = {
             'VCPU': {
                 'total': compute_node.vcpus,
-                'reserved': 0,
+                'reserved': CONF.reserved_host_cpus,
                 'min_unit': 1,
                 'max_unit': compute_node.vcpus,
                 'step_size': 1,
@@ -652,7 +653,7 @@ class TestInventory(SchedulerReportClientTestCase):
         expected_inv_data = {
             'VCPU': {
                 'total': 8,
-                'reserved': 0,
+                'reserved': CONF.reserved_host_cpus,
                 'min_unit': 1,
                 'max_unit': 8,
                 'step_size': 1,
@@ -897,7 +898,7 @@ There was a conflict when trying to complete your request.
             'inventories': {
                 'VCPU': {
                     'total': 8,
-                    'reserved': 0,
+                    'reserved': CONF.reserved_host_cpus,
                     'min_unit': 1,
                     'max_unit': compute_node.vcpus,
                     'step_size': 1,
@@ -937,7 +938,7 @@ There was a conflict when trying to complete your request.
             'inventories': {
                 'VCPU': {
                     'total': 8,
-                    'reserved': 0,
+                    'reserved': CONF.reserved_host_cpus,
                     'min_unit': 1,
                     'max_unit': compute_node.vcpus,
                     'step_size': 1,
diff --git a/releasenotes/notes/reserved_host_cpus-e7de4aa9b89bd947.yaml b/releasenotes/notes/reserved_host_cpus-e7de4aa9b89bd947.yaml
new file mode 100644
index 000000000000..fb530955810a
--- /dev/null
+++ b/releasenotes/notes/reserved_host_cpus-e7de4aa9b89bd947.yaml
@@ -0,0 +1,6 @@
+---
+features:
+  - A new configuration option ``reserved_host_cpus`` has been added for
+    compute services. It helps operators to provide how many physical CPUs they
+    would like to reserve for the hypervisor separately from what the instances
+    use.