Add thermal control support for SONiC (sonic-net#3949)

Junchao-Mellanox · abdosi · commit 109a13cc0313 · 2020-04-30T22:39:17.000-07:00
diff --git a/device/mellanox/x86_64-mlnx_lssn2700-r0/thermal_policy.json b/device/mellanox/x86_64-mlnx_lssn2700-r0/thermal_policy.json
@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json
diff --git a/device/mellanox/x86_64-mlnx_msn2010-r0/thermal_policy.json b/device/mellanox/x86_64-mlnx_msn2010-r0/thermal_policy.json
@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json
diff --git a/device/mellanox/x86_64-mlnx_msn2100-r0/thermal_policy.json b/device/mellanox/x86_64-mlnx_msn2100-r0/thermal_policy.json
@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json
diff --git a/device/mellanox/x86_64-mlnx_msn2410-r0/thermal_policy.json b/device/mellanox/x86_64-mlnx_msn2410-r0/thermal_policy.json
@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json
diff --git a/device/mellanox/x86_64-mlnx_msn2700-r0/thermal_policy.json b/device/mellanox/x86_64-mlnx_msn2700-r0/thermal_policy.json
@@ -0,0 +1,72 @@
+{   
+    "thermal_control_algorithm": {
+        "run_at_boot_up": "false",
+        "fan_speed_when_suspend": "60"
+    },
+    "info_types": [
+        {
+            "type": "fan_info"
+        },
+        {
+            "type": "psu_info"
+        },
+        {
+            "type": "chassis_info"
+        }
+    ],
+    "policies": [
+        {
+            "name": "any fan absence",
+            "conditions": [
+                {
+                    "type": "fan.any.absence"
+                }
+            ],
+            "actions": [
+                {
+                    "type": "thermal_control.control",
+                    "status": "false"
+                },
+                {
+                    "type": "fan.all.set_speed",
+                    "speed": "100"
+                }
+            ]
+        },
+        {
+            "name": "any psu absence",
+            "conditions": [
+                {
+                    "type": "psu.any.absence"
+                }
+            ],
+            "actions": [
+                {
+                    "type": "thermal_control.control",
+                    "status": "false"
+                },
+                {
+                    "type": "fan.all.set_speed",
+                    "speed": "100"
+                }
+            ]
+        },
+        {
+            "name": "all fan and psu presence",
+            "conditions": [
+                {
+                    "type": "fan.all.presence"
+                },
+                {
+                    "type": "psu.all.presence"
+                }
+            ],
+            "actions": [
+                {
+                    "type": "fan.all.set_speed",
+                    "speed": "60"
+                }
+            ]
+        }
+    ]
+}
diff --git a/device/mellanox/x86_64-mlnx_msn2700_simx-r0/thermal_policy.json b/device/mellanox/x86_64-mlnx_msn2700_simx-r0/thermal_policy.json
@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json
diff --git a/device/mellanox/x86_64-mlnx_msn2740-r0/thermal_policy.json b/device/mellanox/x86_64-mlnx_msn2740-r0/thermal_policy.json
@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json
diff --git a/device/mellanox/x86_64-mlnx_msn3700-r0/thermal_policy.json b/device/mellanox/x86_64-mlnx_msn3700-r0/thermal_policy.json
@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json
diff --git a/device/mellanox/x86_64-mlnx_msn3700_simx-r0/thermal_policy.json b/device/mellanox/x86_64-mlnx_msn3700_simx-r0/thermal_policy.json
@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json
diff --git a/device/mellanox/x86_64-mlnx_msn3700c-r0/thermal_policy.json b/device/mellanox/x86_64-mlnx_msn3700c-r0/thermal_policy.json
@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json
diff --git a/device/mellanox/x86_64-mlnx_msn3800-r0/thermal_policy.json b/device/mellanox/x86_64-mlnx_msn3800-r0/thermal_policy.json
@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json
diff --git a/dockers/docker-platform-monitor/docker-pmon.supervisord.conf.j2 b/dockers/docker-platform-monitor/docker-pmon.supervisord.conf.j2
@@ -91,3 +91,14 @@ stdout_logfile=syslog
 stderr_logfile=syslog
 startsecs=0
 {% endif %}
+
+{% if not skip_thermalctld %}
+[program:thermalctld]
+command=/usr/bin/thermalctld
+priority=9
+autostart=false
+autorestart=true
+stdout_logfile=syslog
+stderr_logfile=syslog
+startsecs=0
+{% endif %}
diff --git a/dockers/docker-platform-monitor/start.sh.j2 b/dockers/docker-platform-monitor/start.sh.j2
@@ -75,3 +75,7 @@ supervisorctl start psud
 supervisorctl start syseepromd
 {% endif %}
 
+{% if not skip_thermalctld %}
+supervisorctl start thermalctld
+{% endif %}
+
diff --git a/platform/mellanox/hw-management/0002-hw-management.sh-Disable-thermal-policy-running-in-h.patch b/platform/mellanox/hw-management/0002-hw-management.sh-Disable-thermal-policy-running-in-h.patch
@@ -0,0 +1,31 @@
+From 76b02916794be2e2558fcff1d11609a594f633d7 Mon Sep 17 00:00:00 2001
+From: Stephen Sun <stephens@mellanox.com>
+Date: Fri, 14 Feb 2020 13:48:00 +0800
+Subject: [PATCH] Disable thermal policy running in hw-mgmt service SONiC
+ thermal control algorithm has been supported.
+
+Signed-off-by: Stephen Sun <stephens@mellanox.com>
+---
+ usr/usr/bin/hw-management.sh | 6 +++++-
+ 1 file changed, 5 insertions(+), 1 deletion(-)
+
+diff --git a/usr/usr/bin/hw-management.sh b/usr/usr/bin/hw-management.sh
+index 2cdbfb2..48b41d5 100755
+--- a/usr/usr/bin/hw-management.sh
++++ b/usr/usr/bin/hw-management.sh
+@@ -799,7 +799,11 @@ do_start()
+ 	#disabled for leopard chipless bringup.
+ 	echo 1 > $config_path/suspend
+ 
+-	$THERMAL_CONTROL $thermal_type $max_tachos $max_psus&
++#
++#	Disable thermal control algorithm in hw-management service
++#	because there has already been that in SONiC
++#
++#	$THERMAL_CONTROL $thermal_type $max_tachos $max_psus&
+ }
+ 
+ do_stop()
+-- 
+1.9.1
+
diff --git a/platform/mellanox/mlnx-platform-api.mk b/platform/mellanox/mlnx-platform-api.mk
@@ -3,6 +3,7 @@
 SONIC_PLATFORM_API_PY2 = mlnx_platform_api-1.0-py2-none-any.whl
 $(SONIC_PLATFORM_API_PY2)_SRC_PATH = $(PLATFORM_PATH)/mlnx-platform-api
 $(SONIC_PLATFORM_API_PY2)_PYTHON_VERSION = 2
+$(SONIC_PLATFORM_API_PY2)_DEPENDS = $(SONIC_PLATFORM_COMMON_PY2) $(SONIC_DAEMON_BASE_PY2) $(SONIC_CONFIG_ENGINE)
 SONIC_PYTHON_WHEELS += $(SONIC_PLATFORM_API_PY2)
 
 export mlnx_platform_api_py2_wheel_path="$(addprefix $(PYTHON_WHEELS_PATH)/,$(SONIC_PLATFORM_API_PY2))"
diff --git a/platform/mellanox/mlnx-platform-api/.gitignore b/platform/mellanox/mlnx-platform-api/.gitignore
@@ -0,0 +1,2 @@
+*.pyc
+.cache/
diff --git a/platform/mellanox/mlnx-platform-api/pytest.ini b/platform/mellanox/mlnx-platform-api/pytest.ini
@@ -0,0 +1,3 @@
+[pytest]
+filterwarnings =
+    ignore::DeprecationWarning
diff --git a/platform/mellanox/mlnx-platform-api/setup.cfg b/platform/mellanox/mlnx-platform-api/setup.cfg
@@ -0,0 +1,2 @@
+[aliases]
+test=pytest
diff --git a/platform/mellanox/mlnx-platform-api/setup.py b/platform/mellanox/mlnx-platform-api/setup.py
@@ -12,6 +12,14 @@
     maintainer_email='kevinw@mellanox.com',
     packages=[
         'sonic_platform',
+        'tests'
+    ],
+    setup_requires= [
+        'pytest-runner'
+    ],
+    tests_require = [
+        'pytest',
+        'mock>=2.0.0'
     ],
     classifiers=[
         'Development Status :: 3 - Alpha',
@@ -26,5 +34,6 @@
         'Topic :: Utilities',
     ],
     keywords='sonic SONiC platform PLATFORM',
+    test_suite='setup.get_test_suite'
 )
 
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/__init__.py b/platform/mellanox/mlnx-platform-api/sonic_platform/__init__.py
@@ -1,2 +1,2 @@
 __all__ = ["platform", "chassis"]
-from sonic_platform import *
+from sonic_platform import *
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/chassis.py b/platform/mellanox/mlnx-platform-api/sonic_platform/chassis.py
@@ -442,3 +442,8 @@ def get_change_event(self, timeout=0):
             return True, {'sfp':port_dict}
         else:
             return True, {'sfp':{}}
+
+    def get_thermal_manager(self):
+        from .thermal_manager import ThermalManager
+        return ThermalManager
+
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/fan.py b/platform/mellanox/mlnx-platform-api/sonic_platform/fan.py
@@ -38,17 +38,22 @@ def __init__(self, fan_index, drawer_index = 1, psu_fan = False):
             self.fan_speed_set_path = "fan{}_speed_set".format(self.index)
             self.fan_presence_path = "fan{}_status".format(self.drawer_index)
             self.fan_max_speed_path = "fan{}_max".format(self.index)
+            self._name = "fan{}".format(fan_index + 1)
         else:
             self.fan_speed_get_path = "psu{}_fan1_speed_get".format(self.index)
             self.fan_presence_path = "psu{}_fan1_speed_get".format(self.index)
-            self.fan_max_speed_path = "psu{}_max".format(self.index)
+            self._name = 'psu_{}_fan_{}'.format(self.index, fan_index)
+            self.fan_max_speed_path = None
         self.fan_status_path = "fan{}_fault".format(self.index)
         self.fan_green_led_path = "led_fan{}_green".format(self.drawer_index)
         self.fan_red_led_path = "led_fan{}_red".format(self.drawer_index)
         self.fan_orange_led_path = "led_fan{}_orange".format(self.drawer_index)
         self.fan_pwm_path = "pwm1"
         self.fan_led_cap_path = "led_fan{}_capability".format(self.drawer_index)
 
+    def get_name(self):
+        return self._name
+
     def get_status(self):
         """
         Retrieves the operational status of fan
@@ -123,7 +128,11 @@ def get_speed(self):
                 speed_in_rpm = int(fan_curr_speed.read())
         except (ValueError, IOError):
             speed_in_rpm = 0
-        
+
+        if self.fan_max_speed_path is None:
+            # in case of max speed unsupported, we just return speed in unit of RPM.
+            return speed_in_rpm
+
         max_speed_in_rpm = self._get_max_speed_in_rpm()
         speed = 100*speed_in_rpm/max_speed_in_rpm
 
@@ -136,11 +145,10 @@ def get_target_speed(self):
         Returns:
             int: percentage of the max fan speed
         """
-        speed = 0
-
         if self.is_psu_fan:
             # Not like system fan, psu fan speed can not be modified, so target speed is N/A 
-            return speed
+            return self.get_speed()
+
         try:
             with open(os.path.join(FAN_PATH, self.fan_speed_set_path), 'r') as fan_pwm:
                 pwm = int(fan_pwm.read())
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/platform.py b/platform/mellanox/mlnx-platform-api/sonic_platform/platform.py
@@ -24,6 +24,7 @@ def __init__(self):
             self._chassis.initialize_psu()
             self._chassis.initialize_fan()
             self._chassis.initialize_eeprom()
+            self._chassis.initialize_thermals()
 
     def _is_host(self):
         """
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/psu.py b/platform/mellanox/mlnx-platform-api/sonic_platform/psu.py
@@ -60,6 +60,7 @@ def __init__(self, psu_index, sku):
         psu_oper_status = "thermal/psu{}_pwr_status".format(self.index)
         #psu_oper_status should always be present for all SKUs
         self.psu_oper_status = os.path.join(self.psu_path, psu_oper_status)
+        self._name = "PSU{}".format(psu_index + 1)
 
         if sku in hwsku_dict_psu:
             filemap = psu_profile_list[hwsku_dict_psu[sku]]
@@ -92,7 +93,10 @@ def __init__(self, psu_index, sku):
 
         fan = Fan(psu_index, psu_index, True)
         if fan.get_presence():
-            self._fan = fan
+            self._fan_list.append(fan)
+
+    def get_name(self):
+        return self._name
 
     def _read_generic_file(self, filename, len):
         """
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/thermal.py b/platform/mellanox/mlnx-platform-api/sonic_platform/thermal.py
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/thermal_actions.py b/platform/mellanox/mlnx-platform-api/sonic_platform/thermal_actions.py
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/thermal_conditions.py b/platform/mellanox/mlnx-platform-api/sonic_platform/thermal_conditions.py
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/thermal_infos.py b/platform/mellanox/mlnx-platform-api/sonic_platform/thermal_infos.py
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/thermal_manager.py b/platform/mellanox/mlnx-platform-api/sonic_platform/thermal_manager.py
diff --git a/platform/mellanox/mlnx-platform-api/tests/__init__.py b/platform/mellanox/mlnx-platform-api/tests/__init__.py
diff --git a/platform/mellanox/mlnx-platform-api/tests/mock_platform.py b/platform/mellanox/mlnx-platform-api/tests/mock_platform.py
diff --git a/platform/mellanox/mlnx-platform-api/tests/test_thermal_policy.py b/platform/mellanox/mlnx-platform-api/tests/test_thermal_policy.py
diff --git a/platform/mellanox/mlnx-platform-api/tests/thermal_policy.json b/platform/mellanox/mlnx-platform-api/tests/thermal_policy.json
diff --git a/rules/docker-platform-monitor.mk b/rules/docker-platform-monitor.mk
diff --git a/rules/sonic-thermalctld.mk b/rules/sonic-thermalctld.mk
diff --git a/src/sonic-daemon-base/sonic_daemon_base/task_base.py b/src/sonic-daemon-base/sonic_daemon_base/task_base.py

Original file line number	Diff line number	Diff line change
`@@ -0,0 +1 @@`
	`1`	`+../x86_64-mlnx_msn2700-r0/thermal_policy.json`