[Mellanox] Add test case for CPU thermal control algorithm for SN4800 (…

…sonic-net#5309)

Add test case for PR sonic-net/sonic-buildimage#10202

- How did you do it?
Mock changing CPU temperature and verify cooling level is updated by algorithm accordingly

- How did you verify/test it?
Run the new test and passed.

- Any platform specific information?
Mellanox/Nvidia SN4800 platform

tests/platform_tests/mellanox/mellanox_thermal_control_test_helper.py

@@ -1154,3 +1154,29 @@ def deinit(self):
 
     def mock_psu_status(self, psu_index, status):
         self.mock_helper.mock_thermal_value(self.PSU_PRESENCE.format(psu_index), '1' if status else '0')
+
+
+@mocker('CpuThermalMocker')
+class CpuThermalMocker(object):
+    LOW_THRESHOLD = 80000
+    HIGH_THRESHOLD = 95000
+    MIN_COOLING_STATE = 2
+    MAX_COOLING_STATE = 10
+    CPU_COOLING_STATE_FILE = '/var/run/hw-management/thermal/cooling2_cur_state'
+    CPU_PACK_TEMP_FILE = '/var/run/hw-management/thermal/cpu_pack'
+
+    def __init__(self, dut):
+        self.mock_helper = MockerHelper(dut)
+
+    def deinit(self):
+        """
+        Destructor of CpuThermalMocker.
+        :return:
+        """
+        self.mock_helper.deinit()
+
+    def mock_cpu_pack_temperature(self, temperature):
+        self.mock_helper.mock_value(self.CPU_PACK_TEMP_FILE, temperature)
+
+    def get_cpu_cooling_state(self):
+        return int(self.mock_helper.read_value(self.CPU_COOLING_STATE_FILE))

tests/platform_tests/mellanox/test_thermal_control.py

@@ -2,6 +2,7 @@
 import operator
 import pytest
 import random
+from tests.common.helpers.assertions import pytest_assert, pytest_require
 from tests.common.mellanox_data import get_platform_data
 from tests.common.utilities import wait_until
 from tests.platform_tests.thermal_control_test_helper import *
@@ -159,6 +160,39 @@ def test_psu_absence_policy(duthosts, rand_one_dut_hostname, mocker_factory):
     assert check_fan_speed(duthost, MAX_PWM), 'Fan speed is not turn to {}'.format(MAX_PWM)
 
 
+@pytest.mark.disable_loganalyzer
+def test_cpu_thermal_control(rand_selected_dut, mocker_factory):
+    duthost = rand_selected_dut
+    dut_platform = duthost.facts["platform"]
+    pytest_require(dut_platform == "x86_64-nvidia_sn4800-r0", 'This test case is only for platform x86_64-nvidia_sn4800-r0, skipping...')
+    mocker = mocker_factory(duthost, 'CpuThermalMocker')
+
+    temp_step = 1000
+    # Mock CPU temperature is lower than low threshold
+    mocker.mock_cpu_pack_temperature(mocker.LOW_THRESHOLD - temp_step)
+    wait_result = wait_until(10, 3, 0, check_cpu_cooling_state, mocker, mocker.MIN_COOLING_STATE)
+    pytest_assert(wait_result, 
+                  'CPU cooling state is not MIN when temperature is below low threshold')
+
+    # Mock CPU temperature is raising
+    mocker.mock_cpu_pack_temperature(mocker.LOW_THRESHOLD)
+    wait_result = wait_until(10, 3, 0, check_cpu_cooling_state, mocker, mocker.MIN_COOLING_STATE + 1)
+    pytest_assert(wait_result, 
+                  'CPU cooling state is not increasing when temperature is rasing')
+
+    # Mock CPU temperature is larger than high threshold
+    mocker.mock_cpu_pack_temperature(mocker.HIGH_THRESHOLD + temp_step)
+    wait_result = wait_until(10, 3, 0, check_cpu_cooling_state, mocker, mocker.MAX_COOLING_STATE)
+    pytest_assert(wait_result, 
+                  'CPU cooling state is not MAX increasing when temperature is beyond high threshold')
+
+    # Mock CPU temperature is decreasing
+    mocker.mock_cpu_pack_temperature(mocker.HIGH_THRESHOLD)
+    wait_result = wait_until(10, 3, 0, check_cpu_cooling_state, mocker, mocker.MAX_COOLING_STATE - 1)
+    pytest_assert(wait_result, 
+                  'CPU cooling state is not decreasing when temperature is decreasing')
+
+
 def _check_psu_fan_speed_in_range(actual_speed, max_speed, cooling_level):
     expect_speed = max_speed * cooling_level / 10.0
     logger.info('Expect speed: {}, actual speed: {}'.format(expect_speed, actual_speed))
@@ -270,3 +304,9 @@ def check_fan_speed(duthost, expect_value):
             logging.error('For file {}, Expect speed {}, but actual is {}'.format(file, expect_value, actual_speed))
             return False
     return True
+
+
+def check_cpu_cooling_state(mocker, expect_value):
+    actual_value = mocker.get_cpu_cooling_state()
+    logging.debug('Expect cpu cooling value is {}, actual value is {}'.format(expect_value, actual_value))
+    return actual_value == expect_value

tests/platform_tests/thermal_control_test_helper.py

@@ -111,7 +111,7 @@ def _create_mocker(dut, mocker_name):
         platform = dut.facts['platform']
         mocker_object = None
 
-        if 'mlnx' in platform:
+        if 'mlnx' in platform or 'nvidia' in platform:
             from tests.platform_tests.mellanox import mellanox_thermal_control_test_helper
             mocker_type = BaseMocker.get_mocker_type(mocker_name)
             if mocker_type: