Fan speed should not be 100% when PSU is powered off

tests/platform_tests/mellanox/mellanox_thermal_control_test_helper.py

@@ -1136,3 +1136,21 @@ def deinit(self):
         :return:
         """
         self.mock_helper.deinit()
+
+
+@mocker('PsuMocker')
+class PsuMocker(object):
+    PSU_PRESENCE = 'psu{}_status'
+
+    def __init__(self, dut):
+        self.mock_helper = MockerHelper(dut)
+
+    def deinit(self):
+        """
+        Destructor of MinTableMocker.
+        :return:
+        """
+        self.mock_helper.deinit()
+
+    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')

tests/platform_tests/mellanox/test_thermal_control.py

@@ -20,11 +20,14 @@
 THERMAL_CONTROL_TEST_WAIT_TIME = 75
 THERMAL_CONTROL_TEST_CHECK_INTERVAL = 5
 
+THERMAL_PATH = '/run/hw-management/thermal/'
 COOLING_CUR_STATE_PATH = '/run/hw-management/thermal/cooling_cur_state'
 COOLING_CUR_STATE_THRESHOLD = 7
 PSU_PRESENCE_PATH = '/run/hw-management/thermal/psu{}_status'
 PSU_SPEED_PATH = '/run/hw-management/thermal/psu{}_fan1_speed_get'
 PSU_MAX_SPEED_PATH = '/run/hw-management/config/psu_fan_max'
+PWM_PATH = '/run/hw-management/thermal/pwm1'
+MAX_PWM = 255
 PSU_SPEED_TOLERANCE = 0.25
 
 MAX_COOLING_LEVEL = 10
@@ -133,6 +136,29 @@ def test_set_psu_fan_speed(duthosts, rand_one_dut_hostname, mocker_factory):
             assert False, 'Wait for PSU fan speed change to normal failed'
 
 
+@pytest.mark.disable_loganalyzer
+def test_psu_absence_policy(duthosts, rand_one_dut_hostname, mocker_factory):
+    duthost = duthosts[rand_one_dut_hostname]
+    platform_data = get_platform_data(duthost)
+    hot_swappable = platform_data['psus']['hot_swappable']
+    if not hot_swappable:
+        pytest.skip('The platform {} does not support this test case.'.format(duthost.facts["platform"]))
+
+    psu_num = platform_data['psus']['number']
+    psu_mocker = mocker_factory(duthost, 'PsuMocker')
+    psu_index = random.randint(1, psu_num)
+    psu_mocker.mock_psu_status(psu_index, False)
+    wait_result = wait_until(THERMAL_CONTROL_TEST_WAIT_TIME,
+                             THERMAL_CONTROL_TEST_CHECK_INTERVAL,
+                             0,
+                             check_pwm,
+                             duthost,
+                             MAX_PWM,
+                             operator.eq)
+    assert wait_result, 'PSU is absent, but PWM value is not turned to {}'.format(MAX_PWM)
+    assert check_fan_speed(duthost, MAX_PWM), 'Fan speed is not turn to {}'.format(MAX_PWM)
+
+
 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))
@@ -208,3 +234,39 @@ def check_psu_fan_speed(duthost, psu_num, psu_max_speed, op):
 def check_cooling_level_larger_than_minimum(duthost, expect_minimum_cooling_level):
     actual_cooling_level = get_cooling_cur_state(duthost)
     return actual_cooling_level >= expect_minimum_cooling_level
+
+
+def get_pwm_value(dut):
+    cmd_output = dut.command('cat {}'.format(PWM_PATH))
+    try:
+        pwm_value = int(cmd_output['stdout'])
+        logger.info('PWM is {}'.format(pwm_value))
+        return pwm_value
+    except Exception as e:
+        assert False, 'Bad content in {} - {}'.format(PWM_PATH, e)
+
+
+def check_pwm(duthost, expect_value, op):
+    """Check if FAN PWM value is the expect value
+
+    Args:
+        duthost (object): DUT host object
+        expect_value (int): Expect PWM value
+        op (object): Operator eq or ne
+
+    Returns:
+        boolean: True if the pwm value is expected
+    """
+    pwm_value = get_pwm_value(duthost)
+    return op(pwm_value, expect_value)
+
+
+def check_fan_speed(duthost, expect_value):
+    get_fan_speed_sysfs_cmd = 'ls {}fan*_speed_set'.format(THERMAL_PATH)
+    file_list = duthost.shell(get_fan_speed_sysfs_cmd)['stdout'].splitlines()
+    for file in file_list:
+        actual_speed = int(duthost.shell('cat {}'.format(file))['stdout'].strip())
+        if actual_speed != expect_value:
+            logging.error('For file {}, Expect speed {}, but actual is {}'.format(file, expect_value, actual_speed))
+            return False
+    return True

tests/platform_tests/test_platform_info.py

@@ -335,97 +335,6 @@ def check_thermal_control_load_invalid_file(duthost, file_name):
             restart_thermal_control_daemon(duthost)
 
 
-@pytest.mark.disable_loganalyzer
-@pytest.mark.parametrize('ignore_particular_error_log', [SKIP_ERROR_LOG_PSU_ABSENCE], indirect=True)
-def test_thermal_control_psu_absence(duthosts, enum_rand_one_per_hwsku_hostname, pdu_controller, mocker_factory, ignore_particular_error_log):
-    """
-    @summary: Turn off/on PSUs, check thermal control is working as expect.
-    """
-    duthost = duthosts[enum_rand_one_per_hwsku_hostname]
-    psu_num = get_psu_num(duthost)
-
-    pytest_require(psu_num >= 2, "At least 2 PSUs required for rest of the testing in this case")
-
-    logging.info("Create PDU controller for testing")
-    pdu_ctrl = pdu_controller
-
-    pytest_require(pdu_ctrl, "No PDU controller for %s, skip rest of the testing in this case" % duthost.hostname)
-
-    logging.info("To avoid DUT being shutdown, need to turn on PSUs that are not powered")
-    turn_all_outlets_on(pdu_ctrl)
-
-    logging.info("Initialize test results")
-    psu_test_results = {}
-    pytest_require(check_all_psu_on(duthost, psu_test_results), "Some PSU are still down, skip rest of the testing in this case")
-
-    with ThermalPolicyFileContext(duthost, THERMAL_POLICY_VALID_FILE):
-        fan_mocker = mocker_factory(duthost, 'FanStatusMocker')
-        pytest_require(fan_mocker, "No FanStatusMocker for %s, skip rest of the testing in this case" % duthost.facts['asic_type'])
-
-        restart_thermal_control_daemon(duthost)
-        logging.info('Wait and check all FAN speed turn to 60%...')
-        wait_result = wait_until(THERMAL_CONTROL_TEST_WAIT_TIME,
-                                 THERMAL_CONTROL_TEST_CHECK_INTERVAL,
-                                 0,
-                                 fan_mocker.check_all_fan_speed,
-                                 60)
-
-        pytest_require(wait_result, "FAN speed is not 60%, there might be abnormal in FAN/PSU, skip rest of the testing in this case")
-
-        check_thermal_algorithm_status(duthost, mocker_factory, False)
-
-        logging.info('Shutdown first PDU outlet and check thermal control result...')
-        all_outlet_status = pdu_ctrl.get_outlet_status()
-        pytest_require(all_outlet_status and len(all_outlet_status) >= 2, 'Skip the test, cannot get at least 2 outlet status: {}'.format(all_outlet_status))
-        outlet = all_outlet_status[0]
-        turn_off_outlet_and_check_thermal_control(duthost, pdu_ctrl, outlet, fan_mocker)
-        psu_test_results.clear()
-        pytest_require(check_all_psu_on(duthost, psu_test_results), "Some PSU are still down, skip rest of the testing in this case")
-
-        logging.info('Shutdown second PDU outlet and check thermal control result...')
-        outlet = all_outlet_status[1]
-        turn_off_outlet_and_check_thermal_control(duthost, pdu_ctrl, outlet, fan_mocker)
-        psu_test_results.clear()
-        pytest_require(check_all_psu_on(duthost, psu_test_results), "Some PSU are still down, skip rest of the testing in this case")
-
-        logging.info('Wait and check all FAN speed turn to 65%...')
-        pytest_assert(wait_until(THERMAL_CONTROL_TEST_WAIT_TIME,
-                                 THERMAL_CONTROL_TEST_CHECK_INTERVAL,
-                                 0,
-                                 fan_mocker.check_all_fan_speed,
-                                 65), 'FAN speed not change to 65% according to policy')
-
-
-def turn_off_outlet_and_check_thermal_control(dut, pdu_ctrl, outlet, mocker):
-    """
-    @summary: Turn off PSUs, check all FAN speed are set to 100% according to thermal
-              control policy file.
-    """
-    logging.info("Turn off outlet %s" % str(outlet["outlet_id"]))
-    pdu_ctrl.turn_off_outlet(outlet)
-    time.sleep(5)
-
-    psu_under_test = None
-    psu_line_pattern = get_dut_psu_line_pattern(dut)
-    cli_psu_status = dut.command(CMD_PLATFORM_PSUSTATUS)
-    for line in cli_psu_status["stdout_lines"][2:]:
-        psu_match = psu_line_pattern.match(line)
-        pytest_assert(psu_match, "Unexpected PSU status output")
-        if psu_match.group(2) != "OK":
-            psu_under_test = psu_match.group(1)
-
-    pytest_assert(psu_under_test is not None, "No PSU is turned off")
-    logging.info('Wait and check all FAN speed turn to 100%...')
-    pytest_assert(wait_until(THERMAL_CONTROL_TEST_WAIT_TIME,
-                             THERMAL_CONTROL_TEST_CHECK_INTERVAL,
-                             0,
-                             mocker.check_all_fan_speed,
-                             100), 'FAN speed not turn to 100% after PSU off')
-
-    pdu_ctrl.turn_on_outlet(outlet)
-    time.sleep(5)
-
-
 @pytest.mark.disable_loganalyzer
 def test_thermal_control_fan_status(duthosts, enum_rand_one_per_hwsku_hostname, mocker_factory):
     """