raise thermal trip points for GPU, DDR, LMH and CPU core emerg

[elkoled]

The thermal trip points are raised to ensure that, under normal operating conditions, these thresholds are never reached.

GPU (gpu_trip0): 95°C to 110°C
DDR/PoP (pop_trip): 95°C to 110°C
LMH DCVS (silver/gold): 95°C to 110°C
CPU (per-core emerg): 110°C to 115°C
LMH LIMITS_TEMP_DEFAULT: 75°C to 90°C

master	new	HW	Triggers	Margin to Tjmax (125°C)
95°C	110°C	GPU, DDR/PoP, LMH	GPU throttling, CPU throttling	15°C
110°C	115°C	CPU	Per-core CPU offline	10°C

Validation with modified AGNOS:

paste <(cat /sys/class/thermal/thermal_zone*/type) <(cat /sys/class/thermal/thermal_zone*/trip_point_0_temp) | grep -E "step|lmh"
gpu-virt-max-step	110000
silv-virt-max-step	120000 # unchanged, has no effect on throttling
gold-virt-max-step	120000 # unchanged, has no effect on throttling
pop-mem-step	110000
cpu0-silver-step	115000
cpu1-silver-step	115000
cpu2-silver-step	115000
cpu3-silver-step	115000
cpu0-gold-step	115000
cpu1-gold-step	115000
cpu2-gold-step	115000
cpu3-gold-step	115000
lmh-dcvs-01	110000
lmh-dcvs-00	110000

[adeebshihadeh]

Did you confirm there's no throttling under this new threshold? We should also confirm there's no hysteresis, i.e. <110C never throttles.

[elkoled]

I am running the oven test tonight, will post the freq/temp plots here after.

Absolute hysteresis values are unchanged:
GPU/PoP: 0°C
LMH: 30°C - 110°C active, 80°C reset

[adeebshihadeh]

Kk, we need to make sure whatever temp that we pick, there is never any throttling at or below it.

[elkoled]

[elkoled]

[elkoled]

Ran mici in oven from 75°C to 120°C with openpilot onroad, compared stock kernel vs kernel raised limits.
10 Hz sysfs sampling of all CPU/GPU frequencies, temperatures, and SOM power.

GPU limit is spot on, stock throttles at 95°C, raised limits hold until 110°C.
CPU freq reports static 1,689 MHz over full temp range in both configs, using SOM power instead.
Stock kernel seems to be throttling at over 107°C which is 12°C over LMH.
Raised kernel limits increase it by about 4°C to 111°C.

Device shuts down at just below 120°C CPU/GPU temp

ToDo: Validate LMH behavior and Hysteresis using cooldown ramp.