socmaxtemp question...

[pottendo]

Hi,
on a RPi4 I tried to monitor/control the temperature when throttling should happen.
To test this cmdline.txt;

console=serial0,115200 socmaxtemp=60

The code:

    unsigned tmask = SystemStateUnderVoltageOccurred | SystemStateFrequencyCappingOccurred |
					 SystemStateThrottlingOccurred | SystemStateSoftTempLimitOccurred;
	CPUThrottle.RegisterSystemThrottledHandler(tmask, monitorhandler, nullptr);
	// CPUThrottle.DumpStatus(true);
	log("Maximum temp set to %d, dynamic adaption%spossible, curret freq = %dMHz (max=%dMHz)", 
			CPUThrottle.GetMaxTemperature(), 
			CPUThrottle.IsDynamic() ? " " : " not ",
			CPUThrottle.GetClockRate() / 1000000L, 
			CPUThrottle.GetMaxClockRate() / 1000000L);

	CMachineInfo *mi = CMachineInfo::Get();
	if (mi)
	{
		TMachineModel model = mi->GetMachineModel();
		if (((model == MachineModel4B) ||
			 (model == MachineModel400) ||
			 (model == MachineModelCM4)) &&
			(CPUThrottle.SetSpeed(CPUSpeedMaximum, true) != CPUSpeedUnknown))
		{
			log("maxed freq to %dMHz", CPUThrottle.GetClockRate() / 1000000L);
		}
	}

which results in:

00:00:02.23 pottendo-log:: Maximum temp set to 85, dynamic adaption possible, curret freq = 600MHz (max=1500MHz)
00:00:02.24 pottendo-log:: maxed freq to 1500MHz

So it seems that socmaxtemp=... in cmdline.txt has no effect - CPUThrottle.GetMaxTemperature() still delivers 85C, so my handler is not called as I've not reached 85 degrees.

What am I doing wrong?
thx, pottendo

[rsta2]

You have to call CPUThrottle.Update() in your application's main loop from time to time (at least all 4 seconds).

[pottendo]

Ah, OK.
Im calling SetOnTemperature(void) already. Didn't read the docs further down... silly me!
Is the Core significant? My Monitor runs on Core3 - which probably could be changed.
Thx pottendo

[rsta2]

SetOnTemperature() does not call the registered callback. This is only done by Update(). I assume that it can be called from Core 3, but the callback will be called on this core too then.

[pottendo]

Hi,
somehow it works:
here the monitor code, which is launched on core 3:

void CKernel::run_tempmonitor(void)
{
    unsigned tmask = SystemStateUnderVoltageOccurred | SystemStateFrequencyCappingOccurred |
					 SystemStateThrottlingOccurred | SystemStateSoftTempLimitOccurred;
	CPUThrottle.RegisterSystemThrottledHandler(tmask, monitorhandler, nullptr);
	CPUThrottle.Update();

	while (true) {
		if (CPUThrottle.SetOnTemperature() == false)
			log("temperature monitor failed...");
		MsDelay(5 * 1000);
		if (!CPUThrottle.Update())
			log("CPUThrottle Update failed");
		log("Temp %dC (max=%dC), dynamic adaption%spossible, current freq = %dMHz (max=%dMHz)", 
			CPUThrottle.GetTemperature(),
			CPUThrottle.GetMaxTemperature(), 
			CPUThrottle.IsDynamic() ? " " : " not ",
			CPUThrottle.GetClockRate() / 1000000L, 
			CPUThrottle.GetMaxClockRate() / 1000000L);
	}
}

and the output:

00:00:37.49 pottendo-log:: Temp 55C (max=85C), dynamic adaption possible, current freq = 1500MHz (max=1500MHz)
00:00:42.50 pottendo-log:: Temp 54C (max=85C), dynamic adaption possible, current freq = 1500MHz (max=1500MHz)
00:00:47.51 pottendo-log:: Temp 55C (max=85C), dynamic adaption possible, current freq = 600MHz (max=1500MHz)
00:00:52.51 pottendo-log:: Temp 52C (max=85C), dynamic adaption possible, current freq = 600MHz (max=1500MHz)
00:00:57.52 pottendo-log:: Temp 54C (max=85C), dynamic adaption possible, current freq = 600MHz (max=1500MHz)
00:01:02.53 pottendo-log:: Temp 52C (max=85C), dynamic adaption possible, current freq = 1500MHz (max=1500MHz)
00:01:07.54 pottendo-log:: Temp 53C (max=85C), dynamic adaption possible, current freq = 1500MHz (max=1500MHz)

As one can see the throttling happens. I've forced it with maxsoctemp=55 in cmdline.txt.

However, the registered callback is not called in my case.
Also I had expected CPUThrottle.GetMaxTemperature() would give you the supervised upper threshold - in my case 55 from cmdline.txt. But this may have been a wrong assumption.

Anyway, problem solved in principle as the throttling shows the desired effect (temp goes down) -> consequently my app fails then due to too slow responses, but that's a different story... ;-)

thx, pottendo

PS:
Here's the code of the monitor - spiced with some logs to see if it's called.

void monitorhandler(TSystemThrottledState CurrentState, void *pParam)
{
	Kernel.log("%s - state = 0x%04x", __FUNCTION__, CurrentState);
	if (CurrentState & SystemStateUnderVoltageOccurred) {
		Kernel.log("%s: undervoltage occured...", __FUNCTION__);
	}
	if (CurrentState & SystemStateFrequencyCappingOccurred) {
		Kernel.log("%s: frequency capping occured...", __FUNCTION__);
	}
	if (CurrentState & SystemStateThrottlingOccurred) {
		Kernel.log("%s: throttling occured to %dMHz", __FUNCTION__, CPUThrottle.GetClockRate() / 1000000L);
	}
	if (CurrentState & SystemStateSoftTempLimitOccurred) {
		Kernel.log("%s: softtemplimit occured...", __FUNCTION__);
	}
}

[rsta2]

The CPU throttle callback is triggered from the firmware, when one of the conditions occurs, which can be selected with the mask parameter. The SystemStateSoftTempLimitOccurred condition uses the the maximum SoC temperature here, which is reported by the firmware (85°C). This is different from the value, which is set with socmaxtemp=, which is enforced by Circle CCPUThrottle itself, not by the firmware. So when the callback is not called, no one of the system throttled conditions has been triggered. I have to admit, that I haven't tested this for a while. Changes to the firmware API sometimes occur. You don't need to call CPUThrottle.SetOnTemperature(), if you call CPUThrottle.Update(). Both do nearly the same, but Update() also checks the system throttled state and ensures, that it is not called too frequently. If Update() is called, when the last call to it was later than 4 seconds ago, the call is simply ignored. So you can call Update() all the time, without the need to care about the timing.

[pottendo]

Thx for the clarification, pottendo