AM5728: CPU reset during high temperature testing

Hi Koichiro Tashiro,

You can read the temperature of the SoC from Linux using:

cat /sys/class/thermal/thermal_zone*/temp

You can see the names of various zones using:

cat /sys/class/thermal/thermal_zone*/type

55C is too less for reset.

If you look at the thresholds here: arch/arm/boot/dts/am57xx-industrial-grade.dtsi

&cpu_crit {
temperature = <105000>; /* milliCelsius */
};

So reset should happen in the vicinity of 105C.

Can you please look at the temperatures when you are seeing the resets using the commands
i suggested?

Best Regards,
Keerthy

Hi Keerthy,

Customer tried these commands, but could not get expected responses.

cat /sys/class/thermal/thermal_zone*/temp
cat: /sys/class/thermal/thermal_zone0/temp: Invalid argument
cat: /sys/class/thermal/thermal_zone1/temp: Invalid argument
cat: /sys/class/thermal/thermal_zone2/temp: Invalid argument
cat: /sys/class/thermal/thermal_zone3/temp: Invalid argument
cat: /sys/class/thermal/thermal_zone4/temp: Invalid argument

cat /sys/class/thermal/thermal_zone*/type
cpu_thermal
gpu_thermal
core_thermal
dspeve_thermal
iva_thermal

It seems something is different.
What should be checked?

Thanks and regards,
Koichiro Tashiro

Hi Koichiro,

It seems like there is some failure with the thermal driver probe. I believe i know the reason as RT
Linux disables the CPUFREQ or the dynamic voltage frequency scaling for A15. The CPUFREQ enabling
is a must of thermal monitoring. Can you try the same on Non-RT version?

Latest SDK: software-dl.ti.com/.../index_FDS.html

Best Regards,
Keerthy

Hi Keerthy,

Customer manually checked configuration.
Alert is set as 90C.
Crit is set as 105C.

--------------------------------------------------
root@HX-SDC:/proc/device-tree# for file in `find -name temperature`; do echo $file; hexdump -Cv $file; done
./thermal-zones/dspeve_thermal/trips/dspeve_crit/temperature
00000000  00 01 9a 28                                       |...(|
00000004
./thermal-zones/gpu_thermal/trips/gpu_crit/temperature
00000000  00 01 9a 28                                       |...(|
00000004
./thermal-zones/cpu_thermal/trips/cpu_alert/temperature
00000000  00 01 5f 90                                       |.._.|
00000004
./thermal-zones/cpu_thermal/trips/cpu_crit/temperature
00000000  00 01 9a 28                                       |...(|
00000004
./thermal-zones/iva_thermal/trips/iva_crit/temperature
00000000  00 01 9a 28                                       |...(|
00000004
./thermal-zones/core_thermal/trips/core_crit/temperature
00000000  00 01 9a 28                                       |...(|
00000004
root@HX-SDC:/proc/device-tree# echo $((0x19a28))
105000
root@HX-SDC:/proc/device-tree# echo $((0x15f90))
90000
--------------------------------------------------

Customer also read temperature register values manually.
Just before the Reset, temperature values are below:
MPU：120.8
GPU：114.0
CORE:114.4
Register dump is also attached.
register dump.xlsx
I thought the reset is triggered at 105C, but it seems internal temperature is much higher.
Is this expected?

Thanks and regards,
Koichiro Tashiro

Hi Koichiro Tashiro,

I assumed we are still talking about RT-Linux here right? With RT_Linux there is no thermal
support as the cooling Agent CPUFREQ is disabled. Could you please check that CPUFREQ support
is in place?

Best Regards,
Keerthy

Hi Keerthy,

Yes, we are talking about RT-Linux as changing environment to Non-RT requires a lot of work for customer.
Could you explain a bit more about CPUFREQ support?
- What is CPUFREQ?
- How to check CPUFREQ is enabled or disabled?
- Any documentation available?

Thanks and regards,
Koichiro Tashiro

Hi Koichiro Tashiro,

CPUFREQ is the DVFS (Dynamic Voltage Frequency Scaling) for the CPU. CPU has multiple OPPs(Operating Power Points). OPPs are
nothing but Voltage/Freqeuncy Pair. Based on the system load one can use either high OPP or Lower OPPs that gives the required
power to sustain the load.

CONFIG_ARM_TI_CPUFREQ=y

The Above tells you if the CPUFREQ for TI SoC am57xx is enabled.

Documentation: Documentation/devicetree/bindings/cpufreq/ti-cpufreq.txt

The CONFIG_ARM_TI_CPUFREQ should be disabled on RT Kernel as the frequency changes will make it
difficult to preserve the Real-Time nature of the kernel.

Let me know if all of your questions are answered.

Best Regards,
Keerthy

Hi Keerthy,

Thanks for your explanation. I understood followings:
- With CPUFREQ enabled, the kernel can manage OPPs based on the required load.
- With CPUFREQ enabled, the kernel can handle OPPs to reduce power if internal temperature goes higher than Alert or Crit temp.
- Without CPUFREQ, above features are disabled, so temperature goes up continuously and results in thermal reset.
- For RT-Linux, CPUFREQ cannot be used to preserved real time nature.

Now think about my customer’s original problem, it seems the internal temperature is higher than Crit temp. just before the reset.
I think this is due to no OPP management done in RT-kernel without CPUFREQ.
Is it possible to enable CPUFREQ in RT-kernel just to check it workaround the thermal reset customer observes?

And, what is the practical solution to avoid thermal reset in RT-kernel environment?

Thanks and regards,
Koichiro Tashiro

Koichiro Tashiro said:

Is it possible to enable CPUFREQ in RT-kernel just to check it workaround the thermal reset customer observes?

Can you try the defconfig in the non-RT kernel which has the CPUFREQ enabled?

Koichiro Tashiro said:

And, what is the practical solution to avoid thermal reset in RT-kernel environment?

This will need some external cooling. I can loop in hardware experts to comment on this. Let me know.

Best Regards,
Keerthy

Hi Keerthy,

I have a question in Thermal Management registers.
According to TRM(SPRUHZ6L) section 18.4.6.2.3 “Thermal Shutdown Comparators”,
it says;
"The values of the five low and high TSHUT thresholds are fix and can be neither overridden nor read by
software. The value for the high TSHUT threshold is 123°C (assuming ±2°C temperature sensor accuracy)
and for the low TSHUT threshold it is 105°C."

On the other hand, CTRL_CORE_BANDGAP_TSHUT_xxx regsiter description(ex. Table 18-167) says TSHUT_HOT and TSHUT_COLD can be overridden by SW.
Which description is correct?

Thanks and regards,
Koichiro Tashiro

Hi Keerthy,

Thanks for your reply.
Customer tested and confirmed CTRL_CORE_BANDGAP_TSHUT_xxx override works.

Thanks and regards,
Koichiro Tashiro