thermal: Adjust T6 thermal model

[hawkw]

Our current thermal model for the T6, which sets a target temperature of 70° C, a critical threshold of 80° C, and a power-down threshold of 85° C, is extremely conservative. According to Chelsio, the T_j Typical for the part is 100° C, so we are running with thermal parameters that will shut down the whole system long before the part even reaches the vendor's typical temperature. While I imagine that keeping the T6 well below T_j Typical is probably good for the long-term reliability of the part, the thermal loop is far too twitchy about powering down the whole system if the T6 gets a bit hot, so we can afford to be less careful with it.

This commit updates the thermal mdoel for T6 to set the target temperature to 95° C, the critical threshold to 100° C, and the power-down threshold to 115° C (which is Chelsio's T_j Max value for the part). This is still a bit more conservative than setting the target temperature to 100° C, but I feel like we can afford to be a bit conservative, if it improves long-term reliability --- this still gives us 25° C of additional headroom before the thermal loop gets scared, relative to the old thermal model.

Closes #2431

[mkeeter]

I was paranoid that this would tip us from T6-limited to CPU-limited, with unpredictable consequences for the thermal loop. However, doing a quick audit of rack2, many machines are already in the CPU-limited regime, so I think this is fine.

thermal_sweep.sh

Terrible bash script:

#!/bin/bash
pilot -rrack2 sp ls | egrep "(gimlet|cosmo)" | while read -r line ; do
        serial=$(echo $line|awk '{print $2}')
        type=$(echo $line|awk '{print $3}')
        ip=$(echo $line|awk '{print $5}')
        cpu=$(faux-mgs --interface=rack2sw0tp0 --discovery-addr="[$ip]:11111" inventory 2>/dev/null | grep "CPU temperature" | awk '{print $1}')
        t6=$(faux-mgs --interface=rack2sw0tp0 --discovery-addr="[$ip]:11111" inventory 2>/dev/null | grep "T6 temp" | awk '{print $1}')
        echo $serial $type $ip
        cpu_temp=$(faux-mgs --interface=rack2sw0tp0 --discovery-addr="[$ip]:11111" component-details $cpu 2>/dev/null)
        t6_temp=$(faux-mgs --interface=rack2sw0tp0 --discovery-addr="[$ip]:11111" component-details $t6 2>/dev/null)
        echo $cpu $t6
        echo $cpu_temp
        echo $t6_temp
        echo ""
done

matt@castle ~ $ ./thermal_sweep.sh
BRM42220006 gimlet fe80::aa40:25ff:fe04:181
P0/SBTSI U491
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(80.0) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(63.4375) })

BRM42220017 gimlet fe80::aa40:25ff:fe04:182
P0/SBTSI U491
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(80.625) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(60.5625) })

BRM42220051 gimlet fe80::aa40:25ff:fe04:185
P0/SBTSI U491
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(80.125) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(64.5) })

BRM42220018 gimlet fe80::aa40:25ff:fe04:1c1
P0/SBTSI U491
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(71.375) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(70.4375) })

BRM42220014 gimlet fe80::aa40:25ff:fe04:342
P0/SBTSI U491
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(79.625) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(63.75) })

BRM42220031 gimlet fe80::aa40:25ff:fe04:343
P0/SBTSI U491
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(78.125) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(69.5) })

BRM44220010 gimlet fe80::aa40:25ff:fe04:344
P0/SBTSI U491
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(75.25) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(69.9375) })

BRM44220005 gimlet fe80::aa40:25ff:fe04:347
P0/SBTSI U491
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(80.375) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(68.1875) })

BRM42220057 gimlet fe80::aa40:25ff:fe04:383
P0/SBTSI U491
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(80.25) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(66.9375) })

BRM42220016 gimlet fe80::aa40:25ff:fe04:385
P0/SBTSI U491
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(80.375) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(65.375) })

BRM42220009 gimlet fe80::aa40:25ff:fe04:3c4
P0/SBTSI U491
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(76.5) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(62.5) })

BRM44220011 gimlet fe80::aa40:25ff:fe04:3c5
P0/SBTSI U491
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(79.625) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(69.625) })

BRM13250012 cosmo fe80::aa40:25ff:fe04:402
U1/SBTSI U53
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(63.5) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(61.3125) })

BRM27230045 gimlet fe80::aa40:25ff:fe04:6c6
P0/SBTSI U491
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(80.75) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(69.5) })

BRM22250001 cosmo fe80::aa40:25ff:fe04:c86
U1/SBTSI U53
Measurement(Measurement { name: "CPU", kind: Temperature, value: Ok(61.75) })
Measurement(Measurement { name: "t6", kind: Temperature, value: Ok(58.9375) })