[DellEMC S6000]: Move fan control to thermalctld from fancontrol.sh (#7239)
Why I did it To implement fan control using thermalctld in DellEMC S6000 platform Requires: Azure/sonic-linux-kernel#241 How I did it Add thermal policies in 'thermal_policy.json' Implemented thermal_manager.py and the necessary modules to perform fan control via thermalctld Removed fancontrol.sh How to verify it Verified that the fan speeds are set based on the fan and temperature status. Logs: S6000_fan_control_test_logs.txt
This commit is contained in:
Arun Saravanan Balachandran
GitHub
parent
969cea07aa
commit
6d07efa890
64
device/dell/x86_64-dell_s6000_s1220-r0/thermal_policy.json
Normal file
64
device/dell/x86_64-dell_s6000_s1220-r0/thermal_policy.json
Normal file
@ -0,0 +1,64 @@
|
||||
{
|
||||
"thermal_control_algorithm": {
|
||||
"run_at_boot_up": "true",
|
||||
"fan_speed_when_suspend": "80"
|
||||
},
|
||||
"info_types": [
|
||||
{
|
||||
"type": "chassis_info"
|
||||
},
|
||||
{
|
||||
"type": "fandrawer_info"
|
||||
},
|
||||
{
|
||||
"type": "psu_fan_info"
|
||||
}
|
||||
],
|
||||
"policies": [
|
||||
{
|
||||
"name": "over temperature",
|
||||
"conditions" : [
|
||||
{
|
||||
"type": "chassis.over_temperature"
|
||||
}
|
||||
],
|
||||
"actions": [
|
||||
{
|
||||
"type": "chassis.thermal_shutdown"
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"name": "any fandrawer fault",
|
||||
"conditions": [
|
||||
{
|
||||
"type": "fandrawer.any.fault"
|
||||
}
|
||||
],
|
||||
"actions": [
|
||||
{
|
||||
"type": "fan.all.set_max_speed"
|
||||
},
|
||||
{
|
||||
"type": "fandrawer.fault.set_status_led"
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"name": "all fandrawer normal",
|
||||
"conditions": [
|
||||
{
|
||||
"type": "fandrawer.all.normal"
|
||||
}
|
||||
],
|
||||
"actions": [
|
||||
{
|
||||
"type": "fan.all.set_thermal_level_speed"
|
||||
},
|
||||
{
|
||||
"type": "fandrawer.normal.set_status_led"
|
||||
}
|
||||
]
|
||||
}
|
||||
]
|
||||
}
|
||||
@ -1,10 +1,8 @@
|
||||
s6000/scripts/s6000_platform.sh usr/local/bin
|
||||
s6000/scripts/reset-qsfp usr/local/bin
|
||||
s6000/scripts/set-fan-speed usr/local/bin
|
||||
s6000/scripts/fancontrol.sh usr/local/bin
|
||||
s6000/scripts/hw-management-generate-dump.sh usr/bin
|
||||
s6000/systemd/platform-modules-s6000.service etc/systemd/system
|
||||
s6000/systemd/fancontrol.service etc/systemd/system
|
||||
common/io_rd_wr.py usr/local/bin
|
||||
common/actions.sh usr/share/sonic/device/x86_64-dell_s6000_s1220-r0
|
||||
s6000/scripts/platform_reboot_override usr/share/sonic/device/x86_64-dell_s6000_s1220-r0
|
||||
|
||||
@ -5,8 +5,4 @@ depmod -a
|
||||
systemctl enable platform-modules-s6000.service
|
||||
systemctl start platform-modules-s6000.service
|
||||
|
||||
systemctl enable fancontrol.service
|
||||
systemctl start fancontrol.service
|
||||
|
||||
|
||||
#DEBHELPER#
|
||||
|
||||
@ -1,253 +0,0 @@
|
||||
#!/bin/bash
|
||||
trap 'cleanup' 0 1 2 3 6 9 10 11 13 15
|
||||
|
||||
LEVEL=99
|
||||
INTERVAL=5
|
||||
FAULTY_FANTRAY1=1
|
||||
FAULTY_FANTRAY2=1
|
||||
FAULTY_FANTRAY3=1
|
||||
|
||||
# FAN RPM Speed
|
||||
IDLE=7000
|
||||
LEVEL1=10000
|
||||
LEVEL2=13000
|
||||
LEVEL3=16000
|
||||
LEVEL4=19000
|
||||
LEVEL5=19000
|
||||
|
||||
LRR_FILE="/host/reboot-cause/reboot-cause.txt"
|
||||
I2C_ADAPTER="/sys/class/i2c-adapter/i2c-2/i2c-11"
|
||||
|
||||
SENSOR1="$I2C_ADAPTER/11-004c/hwmon/hwmon*/temp1_input"
|
||||
SENSOR2="$I2C_ADAPTER/11-004d/hwmon/hwmon*/temp1_input"
|
||||
SENSOR3="$I2C_ADAPTER/11-004e/hwmon/hwmon*/temp1_input"
|
||||
|
||||
SENSOR1_MAX="$I2C_ADAPTER/11-004c/hwmon/hwmon*/temp1_max"
|
||||
SENSOR2_MAX="$I2C_ADAPTER/11-004d/hwmon/hwmon*/temp1_max"
|
||||
SENSOR3_MAX="$I2C_ADAPTER/11-004e/hwmon/hwmon*/temp1_max"
|
||||
|
||||
SENSOR1_MAX_VAL=$(cat $SENSOR1_MAX)
|
||||
SENSOR2_MAX_VAL=$(cat $SENSOR2_MAX)
|
||||
SENSOR3_MAX_VAL=$(cat $SENSOR3_MAX)
|
||||
|
||||
# Reducing by 63 to differentiate this temperature settings
|
||||
# from pmon sensors configuration settings
|
||||
SENSOR1_NEW_MAX=$(expr `echo $SENSOR1_MAX_VAL` + 5000 - 63)
|
||||
SENSOR2_NEW_MAX=$(expr `echo $SENSOR2_MAX_VAL` + 5000 - 63)
|
||||
SENSOR3_NEW_MAX=$(expr `echo $SENSOR3_MAX_VAL` + 5000 - 63)
|
||||
|
||||
# Three fan trays with each contains two separate fans
|
||||
# fan1-fan4 fan2-fan5 fan3-fan6
|
||||
FANTRAY1_FAN1=$I2C_ADAPTER/11-0029/fan1_target
|
||||
FANTRAY1_FAN2=$I2C_ADAPTER/11-0029/fan2_target
|
||||
FANTRAY2_FAN1=$I2C_ADAPTER/11-0029/fan3_target
|
||||
FANTRAY2_FAN2=$I2C_ADAPTER/11-0029/fan4_target
|
||||
FANTRAY3_FAN1=$I2C_ADAPTER/11-002a/fan1_target
|
||||
FANTRAY3_FAN2=$I2C_ADAPTER/11-002a/fan2_target
|
||||
|
||||
FANTRAY1_FAN1_RPM=$I2C_ADAPTER/11-0029/fan1_input
|
||||
FANTRAY1_FAN2_RPM=$I2C_ADAPTER/11-0029/fan2_input
|
||||
FANTRAY2_FAN1_RPM=$I2C_ADAPTER/11-0029/fan3_input
|
||||
FANTRAY2_FAN2_RPM=$I2C_ADAPTER/11-0029/fan4_input
|
||||
FANTRAY3_FAN1_RPM=$I2C_ADAPTER/11-002a/fan1_input
|
||||
FANTRAY3_FAN2_RPM=$I2C_ADAPTER/11-002a/fan2_input
|
||||
|
||||
function check_module
|
||||
{
|
||||
MODULE=$1
|
||||
lsmod | grep "$MODULE" > /dev/null
|
||||
ret=$?
|
||||
if [[ $ret = "1" ]]; then
|
||||
echo "$MODULE is not loaded!"
|
||||
exit 1
|
||||
fi
|
||||
}
|
||||
|
||||
function cleanup
|
||||
{
|
||||
echo $SENSOR1_MAX_VAL > $SENSOR1_MAX
|
||||
echo $SENSOR2_MAX_VAL > $SENSOR2_MAX
|
||||
echo $SENSOR3_MAX_VAL > $SENSOR3_MAX
|
||||
exit 1
|
||||
}
|
||||
|
||||
function check_faulty_fan
|
||||
{
|
||||
|
||||
# Assume fans in FanTray spins less than 1000 RPM is faulty.
|
||||
# To Maintain temperature assign max speed 16200 RPM to all other fans.
|
||||
# This RPM speed handle temperature upto 75C degrees
|
||||
|
||||
fan1=$(cat $FANTRAY1_FAN1_RPM)
|
||||
fan2=$(cat $FANTRAY1_FAN2_RPM)
|
||||
fan3=$(cat $FANTRAY2_FAN1_RPM)
|
||||
fan4=$(cat $FANTRAY2_FAN2_RPM)
|
||||
fan5=$(cat $FANTRAY3_FAN1_RPM)
|
||||
fan6=$(cat $FANTRAY3_FAN2_RPM)
|
||||
|
||||
# FanTray1
|
||||
if [ "$fan1" -le "1000" ] || [ "$fan2" -le "1000" ]; then
|
||||
|
||||
# First time detecting failure
|
||||
if [ $FAULTY_FANTRAY1 -lt "2" ]; then
|
||||
FAULTY_FANTRAY1=2
|
||||
/usr/local/bin/set-fan-speed 16200 2 > /dev/null
|
||||
logger "Faulty Fans in Fantray1 $fan1 $fan2 Please check."
|
||||
fi
|
||||
|
||||
elif [ "$fan1" -ge "1000" ] || [ "$fan2" -ge "1000" ]; then
|
||||
FAULTY_FANTRAY1=0
|
||||
fi
|
||||
|
||||
|
||||
# FanTray2
|
||||
if [ "$fan3" -le "1000" ] || [ "$fan4" -le "1000" ]; then
|
||||
|
||||
# First time detecting failure
|
||||
if [ $FAULTY_FANTRAY2 -lt "2" ]; then
|
||||
|
||||
FAULTY_FANTRAY2=2
|
||||
/usr/local/bin/set-fan-speed 16200 2 > /dev/null
|
||||
logger "Faulty Fans in FanTray2: $fan3 $fan4. Please check."
|
||||
fi
|
||||
|
||||
elif [ "$fan3" -ge "1000" ] || [ "$fan4" -ge "1000" ]; then
|
||||
FAULTY_FANTRAY2=0
|
||||
fi
|
||||
|
||||
# FanTray3
|
||||
if [ "$fan5" -le "1000" ] || [ "$fan6" -le "1000" ]; then
|
||||
|
||||
# First time detecting failure
|
||||
if [ $FAULTY_FANTRAY3 -lt "2" ]; then
|
||||
|
||||
FAULTY_FANTRAY3=2
|
||||
/usr/local/bin/set-fan-speed 16200 2 > /dev/null
|
||||
logger "FanTray3 Fans are Faulty.. $fan5 $fan6. Please check."
|
||||
fi
|
||||
|
||||
elif [ "$fan5" -ge "1000" ] || [ "$fan6" -ge "1000" ]; then
|
||||
|
||||
FAULTY_FANTRAY3=0
|
||||
fi
|
||||
|
||||
}
|
||||
|
||||
function update_fan_speed
|
||||
{
|
||||
local fan_speed=$1
|
||||
|
||||
echo $fan_speed > $FANTRAY1_FAN1
|
||||
echo $fan_speed > $FANTRAY1_FAN2
|
||||
echo $fan_speed > $FANTRAY2_FAN1
|
||||
echo $fan_speed > $FANTRAY2_FAN2
|
||||
echo $fan_speed > $FANTRAY3_FAN1
|
||||
echo $fan_speed > $FANTRAY3_FAN2
|
||||
|
||||
}
|
||||
|
||||
function monitor_temp_sensors
|
||||
{
|
||||
SENSOR1_CUR_MAX_VAL=$(cat $SENSOR1_MAX)
|
||||
SENSOR2_CUR_MAX_VAL=$(cat $SENSOR2_MAX)
|
||||
SENSOR3_CUR_MAX_VAL=$(cat $SENSOR3_MAX)
|
||||
if [ "$SENSOR1_CUR_MAX_VAL" -ne "$SENSOR1_NEW_MAX" ]
|
||||
then
|
||||
SENSOR1_NEW_MAX=$(expr `echo $SENSOR1_CUR_MAX_VAL` + 5000 - 63)
|
||||
SENSOR1_MAX_VAL=$SENSOR1_CUR_MAX_VAL
|
||||
echo $SENSOR1_NEW_MAX > $SENSOR1_MAX
|
||||
fi
|
||||
if [ "$SENSOR2_CUR_MAX_VAL" -ne "$SENSOR2_NEW_MAX" ]
|
||||
then
|
||||
SENSOR2_NEW_MAX=$(expr `echo $SENSOR2_CUR_MAX_VAL` + 5000 - 63)
|
||||
SENSOR2_MAX_VAL=$SENSOR2_CUR_MAX_VAL
|
||||
echo $SENSOR2_NEW_MAX > $SENSOR2_MAX
|
||||
fi
|
||||
if [ "$SENSOR3_CUR_MAX_VAL" -ne "$SENSOR3_NEW_MAX" ]
|
||||
then
|
||||
SENSOR3_NEW_MAX=$(expr `echo $SENSOR3_CUR_MAX_VAL` + 5000 - 63)
|
||||
SENSOR3_MAX_VAL=$SENSOR3_CUR_MAX_VAL
|
||||
echo $SENSOR3_NEW_MAX > $SENSOR3_MAX
|
||||
fi
|
||||
|
||||
# go through all temp sensor outputs
|
||||
sensor1=$(expr `echo $(cat $SENSOR1)` / 1000)
|
||||
sensor2=$(expr `echo $(cat $SENSOR2)` / 1000)
|
||||
sensor3=$(expr `echo $(cat $SENSOR3)` / 1000)
|
||||
# All sensors output in 1000's
|
||||
s1=$(cat $SENSOR1)
|
||||
s2=$(cat $SENSOR2)
|
||||
s3=$(cat $SENSOR3)
|
||||
|
||||
if [ "$s1" -ge "$SENSOR1_MAX_VAL" ] || [ "$s2" -ge "$SENSOR2_MAX_VAL" ] || [ "$s3" -ge "$SENSOR3_MAX_VAL" ]
|
||||
then
|
||||
# Thermal trip is about to happen
|
||||
echo "Thermal Overload $sensor1 $sensor2 $sensor3" > $LRR_FILE
|
||||
logger "Shutting down due to over temperature ($sensor1 degree, $sensor2 degree, $sensor3 degree)"
|
||||
sync
|
||||
sleep 1 # Give time to send logger message to server
|
||||
# Assigning the original max values back in sensors
|
||||
echo $SENSOR1_MAX_VAL > $SENSOR1_MAX
|
||||
echo $SENSOR2_MAX_VAL > $SENSOR2_MAX
|
||||
echo $SENSOR3_MAX_VAL > $SENSOR3_MAX
|
||||
|
||||
### Not Reached ###
|
||||
# In case if HW fails to shutdown
|
||||
/sbin/shutdown -P now
|
||||
fi
|
||||
sum=$(($sensor1 + $sensor2 + $sensor3))
|
||||
sensor_temp=$(($sum/3))
|
||||
|
||||
if [ "$sensor_temp" -le "25" ] && [ "$LEVEL" -ne "0" ]
|
||||
then
|
||||
# Set Fan Speed to 7000 RPM"
|
||||
LEVEL=0
|
||||
update_fan_speed $IDLE
|
||||
logger "Adjusted FAN Speed to $IDLE RPM against $sensor_temp Temperature"
|
||||
|
||||
elif [ "$sensor_temp" -ge "26" ] && [ "$sensor_temp" -le "44" ] && [ "$LEVEL" -ne "1" ]
|
||||
then
|
||||
# Set Fan Speed to 10000 RPM"
|
||||
LEVEL=1
|
||||
update_fan_speed $LEVEL1
|
||||
logger "Adjusted FAN Speed to $IDLE RPM against $sensor_temp Temperature"
|
||||
|
||||
elif [ "$sensor_temp" -ge "45" ] && [ "$sensor_temp" -le "59" ] && [ "$LEVEL" -ne "2" ]
|
||||
then
|
||||
# Set Fan Speed to 13000 RPM"
|
||||
LEVEL=2
|
||||
update_fan_speed $LEVEL2
|
||||
logger "Adjusted FAN Speed to $IDLE RPM against $sensor_temp Temperature"
|
||||
|
||||
elif [ "$sensor_temp" -ge "60" ] && [ "$sensor_temp" -le "79" ] && [ "$LEVEL" -ne "3" ]
|
||||
then
|
||||
# Set Fan Speed to 16000 RPM"
|
||||
LEVEL=3
|
||||
update_fan_speed $LEVEL3
|
||||
logger "Adjusted FAN Speed to $IDLE RPM against $sensor_temp Temperature"
|
||||
|
||||
elif [ "$sensor_temp" -ge "80" ] && [ "$LEVEL" -ne "4" ]
|
||||
then
|
||||
# Set Fan Speed to 19000 RPM"
|
||||
LEVEL=4
|
||||
update_fan_speed $LEVEL4
|
||||
logger "Adjusted FAN Speed to $IDLE RPM against $sensor_temp Temperature"
|
||||
fi
|
||||
|
||||
# Check for faulty fan
|
||||
check_faulty_fan
|
||||
|
||||
}
|
||||
|
||||
# Check drivers for sysfs attributes
|
||||
check_module "dell_s6000_platform"
|
||||
check_module "max6620"
|
||||
|
||||
# main loop calling the main function at specified intervals
|
||||
while true
|
||||
do
|
||||
monitor_temp_sensors
|
||||
# Sleep while still handling signals
|
||||
sleep $INTERVAL &
|
||||
wait
|
||||
done
|
||||
@ -125,6 +125,26 @@ switch_board_qsfp_lpmode() {
|
||||
echo $value > /sys/bus/platform/devices/dell-s6000-cpld.0/qsfp_lpmode
|
||||
}
|
||||
|
||||
set_max6620_dynamic_reg() {
|
||||
DEVICE1=/sys/class/i2c-adapter/i2c-11/11-0029/hwmon/hwmon*/fan1_div
|
||||
DEVICE2=/sys/class/i2c-adapter/i2c-11/11-002a/hwmon/hwmon*/fan1_div
|
||||
|
||||
# Retry three times
|
||||
for count in `seq 1 3`; do
|
||||
if [ -w $DEVICE1 -o -w $DEVICE2 ]; then
|
||||
for i in `seq 1 4`; do
|
||||
echo $1 > /sys/class/i2c-adapter/i2c-11/11-0029/hwmon/hwmon*/fan${i}_div
|
||||
done
|
||||
for i in `seq 1 2`; do
|
||||
echo $1 > /sys/class/i2c-adapter/i2c-11/11-002a/hwmon/hwmon*/fan${i}_div
|
||||
done
|
||||
return
|
||||
fi
|
||||
# Sleep for 3 seconds to wait for device tree to be ready
|
||||
sleep 3
|
||||
done
|
||||
}
|
||||
|
||||
install_python_api_package() {
|
||||
device="/usr/share/sonic/device"
|
||||
platform=$(/usr/local/bin/sonic-cfggen -H -v DEVICE_METADATA.localhost.platform)
|
||||
@ -154,7 +174,7 @@ if [[ "$1" == "init" ]]; then
|
||||
#Use 1 for PCIe Gen1, 2 for PCIe Gen2
|
||||
change_pcie_speed 1
|
||||
add_i2c_devices
|
||||
|
||||
set_max6620_dynamic_reg 4
|
||||
/usr/local/bin/set-fan-speed 15000
|
||||
switch_board_qsfp_lpmode "disable"
|
||||
/usr/local/bin/reset-qsfp
|
||||
|
||||
@ -21,12 +21,12 @@ PSU_FAN2=/sys/class/i2c-adapter/i2c-1/1-0059/fan1_target
|
||||
|
||||
# Three fan trays with each contains two separate fans
|
||||
# fan1-fan4 fan2-fan5 fan3-fan6
|
||||
FAN1=/sys/class/i2c-adapter/i2c-11/11-0029/fan1_target
|
||||
FAN2=/sys/class/i2c-adapter/i2c-11/11-0029/fan2_target
|
||||
FAN3=/sys/class/i2c-adapter/i2c-11/11-0029/fan3_target
|
||||
FAN4=/sys/class/i2c-adapter/i2c-11/11-0029/fan4_target
|
||||
FAN5=/sys/class/i2c-adapter/i2c-11/11-002a/fan1_target
|
||||
FAN6=/sys/class/i2c-adapter/i2c-11/11-002a/fan2_target
|
||||
FAN1=/sys/class/i2c-adapter/i2c-11/11-0029/hwmon/hwmon*/fan1_target
|
||||
FAN2=/sys/class/i2c-adapter/i2c-11/11-0029/hwmon/hwmon*/fan2_target
|
||||
FAN3=/sys/class/i2c-adapter/i2c-11/11-0029/hwmon/hwmon*/fan3_target
|
||||
FAN4=/sys/class/i2c-adapter/i2c-11/11-0029/hwmon/hwmon*/fan4_target
|
||||
FAN5=/sys/class/i2c-adapter/i2c-11/11-002a/hwmon/hwmon*/fan1_target
|
||||
FAN6=/sys/class/i2c-adapter/i2c-11/11-002a/hwmon/hwmon*/fan2_target
|
||||
|
||||
speed=$1
|
||||
logger -t platform-modules "Trying to set fan speed to $speed"
|
||||
|
||||
@ -10,6 +10,7 @@
|
||||
try:
|
||||
import os
|
||||
import time
|
||||
import subprocess
|
||||
import struct
|
||||
from sonic_platform_base.chassis_base import ChassisBase
|
||||
from sonic_platform.sfp import Sfp
|
||||
@ -27,6 +28,14 @@ MAX_S6000_PSU = 2
|
||||
MAX_S6000_THERMAL = 6
|
||||
MAX_S6000_COMPONENT = 4
|
||||
|
||||
HYST_RANGE = 5
|
||||
LEVEL0_THRESHOLD = 25
|
||||
LEVEL1_THRESHOLD = 30
|
||||
LEVEL2_THRESHOLD = 45
|
||||
LEVEL3_THRESHOLD = 60
|
||||
LEVEL4_THRESHOLD = 80
|
||||
LEVEL5_THRESHOLD = 85
|
||||
|
||||
|
||||
class Chassis(ChassisBase):
|
||||
"""
|
||||
@ -42,6 +51,11 @@ class Chassis(ChassisBase):
|
||||
reset_reason_dict[0x6] = ChassisBase.REBOOT_CAUSE_NON_HARDWARE
|
||||
reset_reason_dict[0x7] = ChassisBase.REBOOT_CAUSE_THERMAL_OVERLOAD_OTHER
|
||||
|
||||
_num_monitor_thermals = 3
|
||||
_monitor_thermal_list = []
|
||||
_is_fan_control_enabled = False
|
||||
_fan_control_initialised = False
|
||||
|
||||
def __init__(self):
|
||||
ChassisBase.__init__(self)
|
||||
self.status_led_reg = "system_led"
|
||||
@ -142,20 +156,12 @@ class Chassis(ChassisBase):
|
||||
return
|
||||
os.close(fd)
|
||||
|
||||
def _get_thermal_reset(self):
|
||||
reset_file = "/host/reboot-cause/reboot-cause.txt"
|
||||
if (not os.path.isfile(reset_file)):
|
||||
return False
|
||||
try:
|
||||
with open(reset_file, 'r') as fd:
|
||||
rv = fd.read()
|
||||
except Exception as error:
|
||||
return False
|
||||
def _init_fan_control(self):
|
||||
|
||||
if "Thermal Overload" in rv:
|
||||
return True
|
||||
|
||||
return False
|
||||
if not self._fan_control_initialised:
|
||||
for i in range(self._num_monitor_thermals):
|
||||
self._monitor_thermal_list.append(Thermal(i))
|
||||
self._fan_control_initialised = True
|
||||
|
||||
def get_name(self):
|
||||
"""
|
||||
@ -245,9 +251,6 @@ class Chassis(ChassisBase):
|
||||
# NVRAM. Only Warmboot and Coldboot reason are supported here.
|
||||
# Since it does not support any hardware reason, we return
|
||||
# non_hardware as default
|
||||
if self._get_thermal_reset() == True:
|
||||
self._nvram_write(0x49, 0x7)
|
||||
|
||||
lrr = self._get_cpld_register('last_reboot_reason')
|
||||
if (lrr != 'ERR'):
|
||||
reset_reason = int(lrr, base=16)
|
||||
@ -382,3 +385,77 @@ class Chassis(ChassisBase):
|
||||
return status_led
|
||||
else:
|
||||
return None
|
||||
|
||||
def get_thermal_manager(self):
|
||||
"""
|
||||
Retrieves thermal manager class on this chassis
|
||||
|
||||
Returns:
|
||||
A class derived from ThermalManagerBase representing the
|
||||
specified thermal manager
|
||||
"""
|
||||
from .thermal_manager import ThermalManager
|
||||
return ThermalManager
|
||||
|
||||
def set_fan_control_status(self, enable):
|
||||
|
||||
if enable and not self._is_fan_control_enabled:
|
||||
self._init_fan_control()
|
||||
for thermal in self._monitor_thermal_list:
|
||||
thermal.set_high_threshold(LEVEL5_THRESHOLD, force=True)
|
||||
self._is_fan_control_enabled = True
|
||||
elif not enable and self._is_fan_control_enabled:
|
||||
for thermal in self._monitor_thermal_list:
|
||||
thermal.set_high_threshold(LEVEL4_THRESHOLD, force=True)
|
||||
self._is_fan_control_enabled = False
|
||||
|
||||
def get_monitor_thermals(self):
|
||||
return self._monitor_thermal_list
|
||||
|
||||
def thermal_shutdown(self):
|
||||
# Update reboot cause
|
||||
self._nvram_write(0x49, 0x7)
|
||||
|
||||
subprocess.call('sync')
|
||||
time.sleep(1)
|
||||
for thermal in self._monitor_thermal_list:
|
||||
thermal.set_high_threshold(LEVEL4_THRESHOLD, force=True)
|
||||
|
||||
@staticmethod
|
||||
def get_system_thermal_level(curr_thermal_level, system_temperature):
|
||||
|
||||
def get_level_in_hystersis(curr_level, level1, level2):
|
||||
if curr_level != level1 and curr_level != level2:
|
||||
return level1 if abs(curr_level - level1) < abs(curr_level - level2) else level2
|
||||
else:
|
||||
return curr_level
|
||||
|
||||
if system_temperature < LEVEL0_THRESHOLD:
|
||||
curr_thermal_level = 0
|
||||
elif LEVEL0_THRESHOLD <= system_temperature < LEVEL1_THRESHOLD:
|
||||
curr_thermal_level = get_level_in_hystersis(curr_thermal_level, 0, 1)
|
||||
elif LEVEL1_THRESHOLD <= system_temperature <= (LEVEL2_THRESHOLD - HYST_RANGE):
|
||||
curr_thermal_level = 1
|
||||
elif (LEVEL2_THRESHOLD - HYST_RANGE) < system_temperature < LEVEL2_THRESHOLD:
|
||||
curr_thermal_level = get_level_in_hystersis(curr_thermal_level, 1, 2)
|
||||
elif LEVEL2_THRESHOLD <= system_temperature <= (LEVEL3_THRESHOLD - HYST_RANGE):
|
||||
curr_thermal_level = 2
|
||||
elif (LEVEL3_THRESHOLD - HYST_RANGE) < system_temperature < LEVEL3_THRESHOLD:
|
||||
curr_thermal_level = get_level_in_hystersis(curr_thermal_level, 2, 3)
|
||||
elif LEVEL3_THRESHOLD <= system_temperature < LEVEL4_THRESHOLD:
|
||||
curr_thermal_level = 3
|
||||
else:
|
||||
curr_thermal_level = 4
|
||||
|
||||
return curr_thermal_level
|
||||
|
||||
@staticmethod
|
||||
def is_over_temperature(temperature_list):
|
||||
|
||||
over_temperature = False
|
||||
for temperature in temperature_list:
|
||||
if temperature > LEVEL4_THRESHOLD:
|
||||
over_temperature = True
|
||||
break
|
||||
|
||||
return over_temperature
|
||||
|
||||
@ -16,9 +16,14 @@ try:
|
||||
except ImportError as e:
|
||||
raise ImportError(str(e) + "- required module not found")
|
||||
|
||||
|
||||
MAX_S6000_PSU_FAN_SPEED = 18000
|
||||
MAX_S6000_FAN_SPEED = 19000
|
||||
MAX_S6000_FAN_TARGET_SPEED = 18900
|
||||
|
||||
# Each element corresponds to required speed (in RPM)
|
||||
# for a given system thermal level
|
||||
THERMAL_LEVEL_PSU_FAN_SPEED = (7200, 10800, 14400, 16200, 18000)
|
||||
THERMAL_LEVEL_FAN_SPEED = (7000, 10000, 13000, 16000, 19000)
|
||||
|
||||
|
||||
class Fan(FanBase):
|
||||
@ -34,6 +39,7 @@ class Fan(FanBase):
|
||||
def __init__(self, fantray_index=1, fan_index=1,
|
||||
psu_index=1, psu_fan=False, dependency=None):
|
||||
FanBase.__init__(self)
|
||||
self._target_speed = None
|
||||
self.is_psu_fan = psu_fan
|
||||
self.is_driver_initialized = True
|
||||
|
||||
@ -42,11 +48,18 @@ class Fan(FanBase):
|
||||
self.fantray_index = fantray_index
|
||||
self.index = fan_index
|
||||
self.dependency = dependency
|
||||
self.get_fan_speed_reg = self.I2C_DIR +\
|
||||
"{}/fan{}_input".format(*self.FAN_DEV_MAPPING[fantray_index][fan_index])
|
||||
self.set_fan_speed_reg = self.I2C_DIR +\
|
||||
"{}/fan{}_target".format(*self.FAN_DEV_MAPPING[fantray_index][fan_index])
|
||||
|
||||
hwmon_dir = self.I2C_DIR +\
|
||||
"{}/hwmon/".format(self.FAN_DEV_MAPPING[fantray_index][fan_index][0])
|
||||
hwmon_node = os.listdir(hwmon_dir)[0]
|
||||
self.fan_status_reg = hwmon_dir + hwmon_node +\
|
||||
"/fan{}_alarm".format(self.FAN_DEV_MAPPING[fantray_index][fan_index][1])
|
||||
self.get_fan_speed_reg = hwmon_dir + hwmon_node +\
|
||||
"/fan{}_input".format(self.FAN_DEV_MAPPING[fantray_index][fan_index][1])
|
||||
self.set_fan_speed_reg = hwmon_dir + hwmon_node +\
|
||||
"/fan{}_target".format(self.FAN_DEV_MAPPING[fantray_index][fan_index][1])
|
||||
self.max_fan_speed = MAX_S6000_FAN_SPEED
|
||||
self.thermal_level_to_speed = THERMAL_LEVEL_FAN_SPEED
|
||||
else:
|
||||
self.psu_index = psu_index
|
||||
self.index = 1
|
||||
@ -64,6 +77,7 @@ class Fan(FanBase):
|
||||
|
||||
self.get_fan_speed_reg = hwmon_dir + hwmon_node + '/fan1_input'
|
||||
self.max_fan_speed = MAX_S6000_PSU_FAN_SPEED
|
||||
self.thermal_level_to_speed = THERMAL_LEVEL_PSU_FAN_SPEED
|
||||
|
||||
def _get_i2c_register(self, reg_file):
|
||||
# On successful read, returns the value read from given
|
||||
@ -114,6 +128,31 @@ class Fan(FanBase):
|
||||
self.get_fan_speed_reg = fan_speed_reg[0]
|
||||
self.is_driver_initialized = True
|
||||
|
||||
def _get_speed_to_percentage(self, speed):
|
||||
speed_percent = (100 * speed) // self.max_fan_speed
|
||||
return speed_percent if speed_percent <= 100 else 100
|
||||
|
||||
def _get_target_speed_rpm(self):
|
||||
target_speed_rpm = self._get_i2c_register(self.set_fan_speed_reg)
|
||||
if (target_speed_rpm != 'ERR') and self.get_presence():
|
||||
target_speed_rpm = int(target_speed_rpm, 10)
|
||||
else:
|
||||
target_speed_rpm = 0
|
||||
|
||||
return target_speed_rpm
|
||||
|
||||
def _set_speed_rpm(self, speed):
|
||||
if not self.is_psu_fan:
|
||||
if speed > MAX_S6000_FAN_TARGET_SPEED:
|
||||
speed = MAX_S6000_FAN_TARGET_SPEED
|
||||
self._target_speed = speed
|
||||
|
||||
rv = self._set_i2c_register(self.set_fan_speed_reg, speed)
|
||||
if (rv != 'ERR'):
|
||||
return True
|
||||
else:
|
||||
return False
|
||||
|
||||
def get_name(self):
|
||||
"""
|
||||
Retrieves the name of the Fan
|
||||
@ -159,10 +198,17 @@ class Fan(FanBase):
|
||||
bool: True if Fan is operating properly, False if not
|
||||
"""
|
||||
status = False
|
||||
fan_speed = self._get_i2c_register(self.get_fan_speed_reg)
|
||||
if (fan_speed != 'ERR'):
|
||||
if (int(fan_speed) > 1000):
|
||||
status = True
|
||||
if self.is_psu_fan:
|
||||
fan_speed = self._get_i2c_register(self.get_fan_speed_reg)
|
||||
if (fan_speed != 'ERR'):
|
||||
if (int(fan_speed) > 1000):
|
||||
status = True
|
||||
else:
|
||||
fan_status = self._get_i2c_register(self.fan_status_reg)
|
||||
if (fan_status != 'ERR'):
|
||||
fan_status = int(fan_status, 10)
|
||||
if ~fan_status & 0b1:
|
||||
status = True
|
||||
|
||||
return status
|
||||
|
||||
@ -214,8 +260,7 @@ class Fan(FanBase):
|
||||
"""
|
||||
fan_speed = self._get_i2c_register(self.get_fan_speed_reg)
|
||||
if (fan_speed != 'ERR') and self.get_presence():
|
||||
speed_in_rpm = int(fan_speed, 10)
|
||||
speed = (100 * speed_in_rpm)//self.max_fan_speed
|
||||
speed = self._get_speed_to_percentage(int(fan_speed, 10))
|
||||
else:
|
||||
speed = 0
|
||||
|
||||
@ -247,11 +292,7 @@ class Fan(FanBase):
|
||||
bool: True if set success, False if fail.
|
||||
"""
|
||||
fan_set = (speed * self.max_fan_speed) // 100
|
||||
rv = self._set_i2c_register(self.set_fan_speed_reg, fan_set)
|
||||
if (rv != 'ERR'):
|
||||
return True
|
||||
else:
|
||||
return False
|
||||
return self._set_speed_rpm(fan_set)
|
||||
|
||||
def set_status_led(self, color):
|
||||
"""
|
||||
@ -284,5 +325,26 @@ class Fan(FanBase):
|
||||
An integer, the percentage of full fan speed, in the range 0
|
||||
(off) to 100 (full speed)
|
||||
"""
|
||||
# Fan speeds are controlled by fancontrol.sh
|
||||
return self.get_speed()
|
||||
target_speed_rpm = self._get_target_speed_rpm()
|
||||
|
||||
if not self.is_psu_fan and self._target_speed:
|
||||
# Handle max6620 driver approximation
|
||||
max6620_conv_factor = (60 * 8192 * 4) / 2
|
||||
expected_speed_rpm = max6620_conv_factor // (max6620_conv_factor // self._target_speed)
|
||||
|
||||
if expected_speed_rpm == target_speed_rpm:
|
||||
if self._target_speed >= MAX_S6000_FAN_TARGET_SPEED:
|
||||
return 100
|
||||
else:
|
||||
return self._get_speed_to_percentage(self._target_speed)
|
||||
|
||||
return self._get_speed_to_percentage(target_speed_rpm)
|
||||
|
||||
def set_speed_for_thermal_level(self, thermal_level):
|
||||
|
||||
req_speed_rpm = self.thermal_level_to_speed[thermal_level]
|
||||
req_speed = self._get_speed_to_percentage(req_speed_rpm)
|
||||
target_speed = self.get_target_speed()
|
||||
|
||||
if req_speed != target_speed:
|
||||
self._set_speed_rpm(req_speed_rpm)
|
||||
|
||||
@ -101,13 +101,27 @@ class Thermal(ThermalBase):
|
||||
try:
|
||||
with open(sysfs_file, 'r') as fd:
|
||||
rv = fd.read()
|
||||
except:
|
||||
except Exception:
|
||||
rv = 'ERR'
|
||||
|
||||
rv = rv.rstrip('\r\n')
|
||||
rv = rv.lstrip(" ")
|
||||
return rv
|
||||
|
||||
def _write_sysfs_file(self, sysfs_file, value):
|
||||
rv = 'ERR'
|
||||
|
||||
if (not os.path.isfile(sysfs_file)):
|
||||
return rv
|
||||
|
||||
try:
|
||||
with open(sysfs_file, 'w') as fd:
|
||||
rv = fd.write(str(value))
|
||||
except Exception as e:
|
||||
rv = 'ERR'
|
||||
|
||||
return rv
|
||||
|
||||
def _get_sysfs_path(self):
|
||||
temperature_path = glob.glob(self.thermal_temperature_file)
|
||||
high_threshold_path = glob.glob(self.thermal_high_threshold_file)
|
||||
@ -243,18 +257,28 @@ class Thermal(ThermalBase):
|
||||
|
||||
return thermal_low_threshold / 1000.0
|
||||
|
||||
def set_high_threshold(self, temperature):
|
||||
def set_high_threshold(self, temperature, force=False):
|
||||
"""
|
||||
Sets the high threshold temperature of thermal
|
||||
|
||||
Args :
|
||||
temperature: A float number up to nearest thousandth of one
|
||||
degree Celsius, e.g. 30.125
|
||||
force (optional): A boolean, True if set threshold. Only to
|
||||
be used via thermal Manager.
|
||||
Returns:
|
||||
A boolean, True if threshold is set successfully, False if
|
||||
not
|
||||
"""
|
||||
# Thermal threshold values are pre-defined based on HW.
|
||||
# Only to be used by Thermal Manager
|
||||
if force and self.index <= 3 and (80 <= temperature <=85):
|
||||
high_threshold = temperature * 1000
|
||||
result = self._write_sysfs_file(self.thermal_high_threshold_file,
|
||||
high_threshold)
|
||||
if result != 'ERR':
|
||||
return True
|
||||
|
||||
return False
|
||||
|
||||
def set_low_threshold(self, temperature):
|
||||
|
||||
@ -0,0 +1,105 @@
|
||||
from sonic_platform_base.sonic_thermal_control.thermal_action_base import ThermalPolicyActionBase
|
||||
from sonic_platform_base.sonic_thermal_control.thermal_json_object import thermal_json_object
|
||||
from sonic_py_common.logger import Logger
|
||||
|
||||
logger = Logger()
|
||||
|
||||
class ThermalPolicyAction(ThermalPolicyActionBase):
|
||||
|
||||
@staticmethod
|
||||
def get_chassis_info(thermal_info_dict):
|
||||
from .thermal_info import ChassisInfo
|
||||
|
||||
chassis_info = thermal_info_dict.get(ChassisInfo.INFO_NAME)
|
||||
return chassis_info if isinstance(chassis_info, ChassisInfo) else None
|
||||
|
||||
@staticmethod
|
||||
def get_fandrawer_info(thermal_info_dict):
|
||||
from .thermal_info import FanDrawerInfo
|
||||
|
||||
fandrawer_info = thermal_info_dict.get(FanDrawerInfo.INFO_NAME)
|
||||
return fandrawer_info if isinstance(fandrawer_info, FanDrawerInfo) else None
|
||||
|
||||
@staticmethod
|
||||
def get_psu_fan_info(thermal_info_dict):
|
||||
from .thermal_info import PsuFanInfo
|
||||
|
||||
psu_fan_info = thermal_info_dict.get(PsuFanInfo.INFO_NAME)
|
||||
return psu_fan_info if isinstance(psu_fan_info, PsuFanInfo) else None
|
||||
|
||||
@thermal_json_object('fandrawer.fault.set_status_led')
|
||||
class SetFanDrawerFaultStatusLed(ThermalPolicyAction):
|
||||
|
||||
def execute(self, thermal_info_dict):
|
||||
fandrawer_info = self.get_fandrawer_info(thermal_info_dict)
|
||||
if fandrawer_info and fandrawer_info.is_status_changed:
|
||||
for fandrawer in fandrawer_info.fault_fandrawers:
|
||||
if fandrawer.get_status_led() != 'amber':
|
||||
fandrawer.set_status_led('amber')
|
||||
|
||||
|
||||
@thermal_json_object('fandrawer.normal.set_status_led')
|
||||
class SetFanDrawerNormalStatusLed(ThermalPolicyAction):
|
||||
|
||||
def execute(self, thermal_info_dict):
|
||||
fandrawer_info = self.get_fandrawer_info(thermal_info_dict)
|
||||
if fandrawer_info and fandrawer_info.is_status_changed:
|
||||
for fandrawer in fandrawer_info.non_fault_fandrawers:
|
||||
if fandrawer.get_status_led() != 'green':
|
||||
fandrawer.set_status_led('green')
|
||||
|
||||
|
||||
@thermal_json_object('fan.all.set_max_speed')
|
||||
class SetAllFanMaxSpeedAction(ThermalPolicyAction):
|
||||
def execute(self, thermal_info_dict):
|
||||
fandrawer_info = self.get_fandrawer_info(thermal_info_dict)
|
||||
psu_fan_info = self.get_psu_fan_info(thermal_info_dict)
|
||||
|
||||
if fandrawer_info:
|
||||
if fandrawer_info.is_status_changed and fandrawer_info.is_new_fault:
|
||||
logger.log_warning("Fandrawer fault detected. Setting all fans to maximum speed")
|
||||
|
||||
for fan in fandrawer_info.non_fault_fans:
|
||||
fan.set_speed(100)
|
||||
|
||||
if psu_fan_info:
|
||||
for fan in psu_fan_info.present_fans:
|
||||
fan.set_speed(100)
|
||||
|
||||
|
||||
@thermal_json_object('fan.all.set_thermal_level_speed')
|
||||
class SetAllFanThermalLevelSpeedAction(ThermalPolicyAction):
|
||||
def execute(self, thermal_info_dict):
|
||||
|
||||
chassis_info = self.get_chassis_info(thermal_info_dict)
|
||||
fandrawer_info = self.get_fandrawer_info(thermal_info_dict)
|
||||
psu_fan_info = self.get_psu_fan_info(thermal_info_dict)
|
||||
|
||||
if chassis_info:
|
||||
if chassis_info.is_status_changed:
|
||||
if chassis_info.initial_run:
|
||||
logger.log_notice("System thermal level is at LEVEL{}".format(chassis_info.system_thermal_level))
|
||||
else:
|
||||
logger.log_notice("System thermal level changed to LEVEL{}".format(chassis_info.system_thermal_level))
|
||||
|
||||
if fandrawer_info:
|
||||
if fandrawer_info.is_status_changed and not chassis_info.initial_run:
|
||||
logger.log_notice("All fandrawers back to normal")
|
||||
|
||||
for fan in fandrawer_info.non_fault_fans:
|
||||
fan.set_speed_for_thermal_level(chassis_info.system_thermal_level)
|
||||
|
||||
if psu_fan_info:
|
||||
for fan in psu_fan_info.present_fans:
|
||||
fan.set_speed_for_thermal_level(chassis_info.system_thermal_level)
|
||||
|
||||
|
||||
@thermal_json_object('chassis.thermal_shutdown')
|
||||
class ThermalShutdownAction(ThermalPolicyAction):
|
||||
def execute(self, thermal_info_dict):
|
||||
|
||||
chassis_info = self.get_chassis_info(thermal_info_dict)
|
||||
if chassis_info:
|
||||
logger.log_warning("Shutting down due to over temperature - "
|
||||
+ ",".join("{} C".format(i) for i in chassis_info.temperature_list))
|
||||
chassis_info.chassis.thermal_shutdown()
|
||||
@ -0,0 +1,35 @@
|
||||
from sonic_platform_base.sonic_thermal_control.thermal_condition_base import ThermalPolicyConditionBase
|
||||
from sonic_platform_base.sonic_thermal_control.thermal_json_object import thermal_json_object
|
||||
|
||||
|
||||
class FanDrawerCondition(ThermalPolicyConditionBase):
|
||||
|
||||
@staticmethod
|
||||
def get_fandrawer_info(thermal_info_dict):
|
||||
from .thermal_info import FanDrawerInfo
|
||||
|
||||
fandrawer_info = thermal_info_dict.get(FanDrawerInfo.INFO_NAME)
|
||||
return fandrawer_info if isinstance(fandrawer_info, FanDrawerInfo) else None
|
||||
|
||||
|
||||
@thermal_json_object('fandrawer.any.fault')
|
||||
class AnyFanDrawerAbsentOrFaultCondition(FanDrawerCondition):
|
||||
def is_match(self, thermal_info_dict):
|
||||
fandrawer_info = self.get_fandrawer_info(thermal_info_dict)
|
||||
return fandrawer_info.fault if fandrawer_info else False
|
||||
|
||||
@thermal_json_object('fandrawer.all.normal')
|
||||
class AllFanDrawerGoodCondition(FanDrawerCondition):
|
||||
def is_match(self, thermal_info_dict):
|
||||
fandrawer_info = self.get_fandrawer_info(thermal_info_dict)
|
||||
return not fandrawer_info.fault if fandrawer_info else False
|
||||
|
||||
@thermal_json_object('chassis.over_temperature')
|
||||
class OverTemperatureCondition(ThermalPolicyConditionBase):
|
||||
|
||||
@staticmethod
|
||||
def is_match(thermal_info_dict):
|
||||
from .thermal_info import ChassisInfo
|
||||
|
||||
chassis_info = thermal_info_dict.get(ChassisInfo.INFO_NAME)
|
||||
return chassis_info.is_over_temperature if isinstance(chassis_info, ChassisInfo) else False
|
||||
@ -0,0 +1,189 @@
|
||||
from sonic_platform_base.sonic_thermal_control.thermal_info_base import ThermalPolicyInfoBase
|
||||
from sonic_platform_base.sonic_thermal_control.thermal_json_object import thermal_json_object
|
||||
|
||||
|
||||
@thermal_json_object('chassis_info')
|
||||
class ChassisInfo(ThermalPolicyInfoBase):
|
||||
"""
|
||||
Chassis information needed by thermal policy
|
||||
"""
|
||||
INFO_NAME = 'chassis_info'
|
||||
|
||||
def __init__(self):
|
||||
self._chassis = None
|
||||
self._temperature_list = []
|
||||
self._thermal_list = []
|
||||
self._system_thermal_level = 0
|
||||
self._initial_run = False
|
||||
self._is_over_temperature = False
|
||||
self._is_status_changed = False
|
||||
|
||||
def collect(self, chassis):
|
||||
"""
|
||||
Collect platform chassis.
|
||||
:param chassis: The chassis object
|
||||
:return:
|
||||
"""
|
||||
self._initial_run = False
|
||||
self._is_status_changed = False
|
||||
self._temperature_list = []
|
||||
|
||||
if not self._chassis:
|
||||
self._initial_run = True
|
||||
self._chassis = chassis
|
||||
self._thermal_list = chassis.get_monitor_thermals()
|
||||
|
||||
for thermal in self._thermal_list:
|
||||
self._temperature_list.append(thermal.get_temperature())
|
||||
|
||||
system_temperature = sum(self._temperature_list) / len(self._temperature_list)
|
||||
curr_level = chassis.get_system_thermal_level(self._system_thermal_level,
|
||||
system_temperature)
|
||||
if curr_level != self._system_thermal_level:
|
||||
self._is_status_changed = True
|
||||
self._system_thermal_level = curr_level
|
||||
|
||||
self._is_over_temperature = chassis.is_over_temperature(self._temperature_list)
|
||||
|
||||
@property
|
||||
def chassis(self):
|
||||
return self._chassis
|
||||
|
||||
@property
|
||||
def initial_run(self):
|
||||
return self._initial_run
|
||||
|
||||
@property
|
||||
def is_over_temperature(self):
|
||||
return self._is_over_temperature
|
||||
|
||||
@property
|
||||
def is_status_changed(self):
|
||||
return self._is_status_changed
|
||||
|
||||
@property
|
||||
def system_thermal_level(self):
|
||||
return self._system_thermal_level
|
||||
|
||||
@property
|
||||
def temperature_list(self):
|
||||
return self._temperature_list
|
||||
|
||||
|
||||
@thermal_json_object('fandrawer_info')
|
||||
class FanDrawerInfo(ThermalPolicyInfoBase):
|
||||
|
||||
INFO_NAME = 'fandrawer_info'
|
||||
|
||||
def __init__(self):
|
||||
self._fault_fandrawers = set()
|
||||
self._present_fandrawers = set()
|
||||
self._fault_fans = set()
|
||||
self._present_fans = set()
|
||||
self._fault = False
|
||||
self._is_new_fault = False
|
||||
self._is_status_changed = False
|
||||
|
||||
def collect(self, chassis):
|
||||
"""
|
||||
Collect fan information for thermal policy.
|
||||
:param chassis: The chassis object.
|
||||
:return:
|
||||
"""
|
||||
fault = False
|
||||
self._is_new_fault = False
|
||||
self._is_status_changed = False
|
||||
for fandrawer in chassis.get_all_fan_drawers():
|
||||
presence = fandrawer.get_presence()
|
||||
|
||||
if presence and (fandrawer not in self._present_fandrawers):
|
||||
self._is_status_changed = True
|
||||
self._present_fandrawers.add(fandrawer)
|
||||
self._present_fans.update(fandrawer.get_all_fans())
|
||||
elif not presence and (fandrawer in self._present_fandrawers):
|
||||
self._is_status_changed = True
|
||||
self._present_fandrawers.discard(fandrawer)
|
||||
self._present_fans.difference_update(fandrawer.get_all_fans())
|
||||
|
||||
fan_fault = False
|
||||
for fan in fandrawer.get_all_fans():
|
||||
status = fan.get_status()
|
||||
fan_fault |= not status
|
||||
|
||||
if status and (fan in self._fault_fans):
|
||||
self._is_status_changed = True
|
||||
self._fault_fans.discard(fan)
|
||||
elif not status and (fan not in self._fault_fans):
|
||||
self._is_status_changed = True
|
||||
self._fault_fans.add(fan)
|
||||
|
||||
if self._is_status_changed:
|
||||
if fan_fault and (fandrawer not in self._fault_fandrawers):
|
||||
self._fault_fandrawers.add(fandrawer)
|
||||
elif not fan_fault:
|
||||
self._fault_fandrawers.discard(fandrawer)
|
||||
|
||||
if self._fault_fans or (chassis.get_num_fans() != len(self._present_fans)):
|
||||
fault = True
|
||||
|
||||
if self._is_status_changed:
|
||||
if fault and not self._fault:
|
||||
self._is_new_fault = True
|
||||
|
||||
self._fault = fault
|
||||
|
||||
@property
|
||||
def fault_fans(self):
|
||||
return self._present_fans.intersection(self._fault_fans)
|
||||
|
||||
@property
|
||||
def non_fault_fans(self):
|
||||
return self._present_fans.difference(self._fault_fans)
|
||||
|
||||
@property
|
||||
def fault_fandrawers(self):
|
||||
return self._present_fandrawers.intersection(self._fault_fandrawers)
|
||||
|
||||
@property
|
||||
def non_fault_fandrawers(self):
|
||||
return self._present_fandrawers.difference(self._fault_fandrawers)
|
||||
|
||||
@property
|
||||
def fault(self):
|
||||
return self._fault
|
||||
|
||||
@property
|
||||
def is_new_fault(self):
|
||||
return self._is_new_fault
|
||||
|
||||
@property
|
||||
def is_status_changed(self):
|
||||
return self._is_status_changed
|
||||
|
||||
|
||||
@thermal_json_object('psu_fan_info')
|
||||
class PsuFanInfo(ThermalPolicyInfoBase):
|
||||
|
||||
INFO_NAME = 'psu_fan_info'
|
||||
|
||||
def __init__(self):
|
||||
self._present_fans = set()
|
||||
|
||||
def collect(self, chassis):
|
||||
"""
|
||||
Collect fan information for thermal policy.
|
||||
:param chassis: The chassis object.
|
||||
:return:
|
||||
"""
|
||||
for psu in chassis.get_all_psus():
|
||||
for fan in psu.get_all_fans():
|
||||
presence = fan.get_presence()
|
||||
|
||||
if presence and (fan not in self._present_fans):
|
||||
self._present_fans.add(fan)
|
||||
elif not presence and (fan in self._present_fans):
|
||||
self._present_fans.discard(fan)
|
||||
|
||||
@property
|
||||
def present_fans(self):
|
||||
return self._present_fans
|
||||
@ -0,0 +1,63 @@
|
||||
from sonic_platform_base.sonic_thermal_control.thermal_manager_base import ThermalManagerBase
|
||||
|
||||
from .thermal_action import (
|
||||
SetAllFanMaxSpeedAction, SetAllFanThermalLevelSpeedAction,
|
||||
SetFanDrawerFaultStatusLed, SetFanDrawerNormalStatusLed,
|
||||
ThermalShutdownAction
|
||||
)
|
||||
from .thermal_condition import (
|
||||
AllFanDrawerGoodCondition, AnyFanDrawerAbsentOrFaultCondition,
|
||||
OverTemperatureCondition
|
||||
)
|
||||
from .thermal_info import ChassisInfo, FanDrawerInfo, PsuFanInfo
|
||||
|
||||
|
||||
class ThermalManager(ThermalManagerBase):
|
||||
|
||||
_chassis = None
|
||||
_fan_speed_default = 80
|
||||
|
||||
@classmethod
|
||||
def deinitialize(cls):
|
||||
"""
|
||||
Destroy thermal manager, including any vendor specific cleanup.
|
||||
:return:
|
||||
"""
|
||||
cls.stop_thermal_algorithm()
|
||||
|
||||
@classmethod
|
||||
def init_thermal_algorithm(cls, chassis):
|
||||
"""
|
||||
Initialize thermal algorithm according to policy file.
|
||||
:param chassis: The chassis object.
|
||||
:return:
|
||||
"""
|
||||
if cls._chassis is None:
|
||||
cls._chassis = chassis
|
||||
|
||||
cls.start_thermal_algorithm()
|
||||
|
||||
@classmethod
|
||||
def start_thermal_algorithm(cls):
|
||||
"""
|
||||
Start vendor specific thermal control algorithm.
|
||||
:return:
|
||||
"""
|
||||
if cls._chassis:
|
||||
cls._chassis.set_fan_control_status(True)
|
||||
|
||||
@classmethod
|
||||
def stop_thermal_algorithm(cls):
|
||||
"""
|
||||
Stop vendor specific thermal control algorithm.
|
||||
:return:
|
||||
"""
|
||||
if cls._chassis:
|
||||
cls._chassis.set_fan_control_status(False)
|
||||
|
||||
for fan in cls._chassis.get_all_fans():
|
||||
fan.set_speed(cls._fan_speed_default)
|
||||
|
||||
for psu in cls._chassis.get_all_psus():
|
||||
for fan in psu.get_all_fans():
|
||||
fan.set_speed(cls._fan_speed_default)
|
||||
@ -1,13 +0,0 @@
|
||||
[Unit]
|
||||
Description=Dell S6000 fan speed regulator
|
||||
After=platform-modules-s6000.service
|
||||
Before=pmon.service
|
||||
|
||||
[Service]
|
||||
ExecStart=-/usr/local/bin/fancontrol.sh
|
||||
Restart=always
|
||||
RestartSec=30
|
||||
|
||||
[Install]
|
||||
WantedBy=multi-user.target
|
||||
|
||||
Loading…
Reference in New Issue
Block a user