[201911][thermal control] Backport feature from master branch

device/mellanox/x86_64-mlnx_lssn2700-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn2010-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn2100-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn2410-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn2700-r0/thermal_policy.json

@@ -0,0 +1,80 @@
+{   
+    "thermal_control_algorithm": {
+        "run_at_boot_up": "true",
+        "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": "fan.all.set_speed",
+                    "speed": "100"
+                }
+            ]
+        },
+        {
+            "name": "any psu absence",
+            "conditions": [
+                {
+                    "type": "psu.any.absence"
+                }
+            ],
+            "actions": [
+                {
+                    "type": "fan.all.set_speed",
+                    "speed": "100"
+                }
+            ]
+        },
+        {
+            "name": "any fan broken",
+            "conditions": [
+                {
+                    "type": "fan.any.fault"
+                }
+            ],
+            "actions": [
+                {
+                    "type": "fan.all.set_speed",
+                    "speed": "100"
+                }
+            ]
+        },
+        {
+            "name": "all fan and psu presence",
+            "conditions": [
+                {
+                    "type": "fan.all.presence"
+                },
+                {
+                    "type": "psu.all.presence"
+                },
+                {
+                    "type": "fan.all.good"
+                }
+            ],
+            "actions": [
+                {
+                    "type": "thermal.recover"
+                }
+            ]
+        }
+    ]
+}

device/mellanox/x86_64-mlnx_msn2700_simx-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn2740-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn3420-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn3700-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn3700_simx-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn3700c-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn3800-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn4600c-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn4700-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

dockers/docker-platform-monitor/Dockerfile.j2

@@ -18,7 +18,8 @@ RUN apt-get update &&   \
         rrdtool         \
         python-smbus    \
         ethtool         \
-        dmidecode &&    \
+        dmidecode       \
+        i2c-tools &&    \
     pip install enum34
 
 {% if docker_platform_monitor_debs.strip() -%}

dockers/docker-platform-monitor/docker-pmon.supervisord.conf.j2

@@ -91,3 +91,14 @@ stdout_logfile=syslog
 stderr_logfile=syslog
 startsecs=10
 {% 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 %}

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 %}
+

platform/mellanox/hw-management/0001-hw-management.sh-Disable-thermal-policy-running-in-h.patch

@@ -0,0 +1,27 @@
+From 3512488c981eb81d51ce92cb3573721e36861f56 Mon Sep 17 00:00:00 2001
+From: Junchao Chen <junchao@mellanox.com>
+Date: Fri, 29 May 2020 10:38:53 +0300
+Subject: [PATCH] Disable hw-management thermal control service
+
+---
+ usr/usr/bin/hw-management.sh | 4 +++-
+ 1 file changed, 3 insertions(+), 1 deletion(-)
+
+diff --git a/usr/usr/bin/hw-management.sh b/usr/usr/bin/hw-management.sh
+index 65e5d39..0d1c4a1 100755
+--- a/usr/usr/bin/hw-management.sh
++++ b/usr/usr/bin/hw-management.sh
+@@ -832,7 +832,9 @@ do_start()
+ 	if [ -f $config_path/max_tachos ]; then
+ 		max_tachos=$(<$config_path/max_tachos)
+ 	fi
+-	$THERMAL_CONTROL $thermal_type $max_tachos $max_psus&
++	# Disable hw-management thermal control because 
++	# SONiC already implement it
++	#$THERMAL_CONTROL $thermal_type $max_tachos $max_psus&
+ }
+
+ do_stop()
+-- 
+1.9.1
+

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))"

platform/mellanox/mlnx-platform-api/.gitignore

@@ -0,0 +1,2 @@
+*.pyc
+.cache/

platform/mellanox/mlnx-platform-api/pytest.ini

@@ -0,0 +1,3 @@
+[pytest]
+filterwarnings =
+    ignore::DeprecationWarning

platform/mellanox/mlnx-platform-api/setup.cfg

@@ -0,0 +1,2 @@
+[aliases]
+test=pytest

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'
 )
 

platform/mellanox/mlnx-platform-api/sonic_platform/__init__.py

@@ -1,2 +1,2 @@
 __all__ = ["platform", "chassis"]
-from sonic_platform import *
+from sonic_platform import *

platform/mellanox/mlnx-platform-api/sonic_platform/chassis.py

@@ -15,6 +15,7 @@
     from sonic_daemon_base.daemon_base import Logger
     from os import listdir
     from os.path import isfile, join
+    from glob import glob
     import sys
     import io
     import re
@@ -28,12 +29,17 @@
 MLNX_NUM_PSU = 2
 
 GET_HWSKU_CMD = "sonic-cfggen -d -v DEVICE_METADATA.localhost.hwsku"
+GET_PLATFORM_CMD = "sonic-cfggen -d -v DEVICE_METADATA.localhost.platform"
 
 EEPROM_CACHE_ROOT = '/var/cache/sonic/decode-syseeprom'
 EEPROM_CACHE_FILE = 'syseeprom_cache'
 
 HWMGMT_SYSTEM_ROOT = '/var/run/hw-management/system/'
 
+MST_DEVICE_NAME_PATTERN = '/dev/mst/mt[0-9]*_pciconf0'
+MST_DEVICE_RE_PATTERN = '/dev/mst/mt([0-9]*)_pciconf0'
+SPECTRUM1_CHIP_ID = '52100'
+
 #reboot cause related definitions
 REBOOT_CAUSE_ROOT = HWMGMT_SYSTEM_ROOT
 
@@ -55,6 +61,7 @@ def __init__(self):
 
         # Initialize SKU name
         self.sku_name = self._get_sku_name()
+        self.platform_name = self._get_platform_name()
         mi = get_machine_info()
         if mi is not None:
             self.name = mi['onie_platform']
@@ -93,11 +100,21 @@ def initialize_fan(self):
         num_of_fan, num_of_drawer = self._extract_num_of_fans_and_fan_drawers()
         multi_rotor_in_drawer = num_of_fan > num_of_drawer
 
+        # Fan's direction isn't supported on spectrum 1 devices for now
+        mst_dev_list = glob(MST_DEVICE_NAME_PATTERN)
+        if not mst_dev_list:
+            raise RuntimeError("Can't get chip type due to {} not found".format(MST_DEVICE_NAME_PATTERN))
+        m = re.search(MST_DEVICE_RE_PATTERN, mst_dev_list[0])
+        if m.group(1) == SPECTRUM1_CHIP_ID:
+            has_fan_dir = False
+        else:
+            has_fan_dir = True
+
         for index in range(num_of_fan):
             if multi_rotor_in_drawer:
-                fan = Fan(index, index/2)
+                fan = Fan(has_fan_dir, index, index/2, False, self.platform_name)
             else:
-                fan = Fan(index, index)
+                fan = Fan(has_fan_dir, index, index, False, self.platform_name)
             self._fan_list.append(fan)
 
 
@@ -230,6 +247,12 @@ def _get_sku_name(self):
         return out.rstrip('\n')
 
 
+    def _get_platform_name(self):
+        p = subprocess.Popen(GET_PLATFORM_CMD, shell=True, stdout=subprocess.PIPE)
+        out, err = p.communicate()
+        return out.rstrip('\n')
+
+
     def _get_port_position_tuple_by_sku_name(self):
         position_tuple = port_position_tuple_list[hwsku_dict_port[self.sku_name]]
         return position_tuple
@@ -442,3 +465,8 @@ 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
+