Tegra2 freeze and heat problems with kernel 4.19

Hello,

I refer to the problems mentioned in the post LTS Kernel RGB Freeze including the mailing list issue Re: T20 Cpuidle Freeze and wanted to know if there is any news on that?

The problem still persists on our hardware. After the system has booted up, it completely freezes after a certain time and can no longer be operated when graphics output is activated (not only with rgb output but also with HDMI/DVI output as well).

We’ve done a litte more research on that and found out that the problem persists from kernel 4.9 until 4.19. We could locate the issue is coming from switching from sleep state 0 to sleep state 1 in the menu govenor. Disabling going into state 1 in the menu driver prevents the system from freezing. In connection with the mail from Dmitry Osipenko from the mailing list, I assume that switching the sleep mode on the Colibri is not working properly yet, cause he could not confirm the issue with tegra2 on an Acer A500 tablet.

Furthermore the the cpu is getting very hot in idle state (~46°C), even with going into sleep mode 1 (before system freeze), but i think that is caused by the problem i mentioned above.

I could cause this issue with the following configurations:

  • Kernel-Mainline: 4.9, 4.19.6
  • Toradex Kernel from toradex_4.19.y branch
  • Config: standard tegra_defconfig
  • Drivers: tegra nand driver with activated ubi and ubifs driver and graphics driver
  • Hardware: Colibri developer board, Iris Carrier Board with device tree from linux-toradex git repositoy

My question is now, if someone can confirm the problem and possibly has an idea to solve the problem?

Greetings Sebastian Lehmann

I could cause this issue with the following configurations:

Kernel-Mainline: 4.9, 4.19.6

Toradex Kernel from toradex_4.19.y branch

Please note that this branch is only meant for Torizon which currently exclusively runs on i.MX 6 and 7 SoCs.

Config: standard tegra_defconfig

Drivers: tegra nand driver with activated ubi and ubifs driver and graphics driver

Hardware: Colibri developer board,

I guess you mean the Colibri Evaluation Board, right?

Iris Carrier Board with device tree from linux-toradex git repositoy

I’m not quite sure as to what exact device tree you are referring to here.

My question is now, if someone can confirm the problem and possibly has an idea to solve the problem?

I can confirm that the CPU idle freeze still happens for me as well. So far I just disabled that feature and otherwise had quite good results with mainline on the Colibri T20.

I guess you mean the Colibri Evaluation Board, right?

sry, yes i meant the Colibri Evaluation Board

I’m not quite sure as to what exact device tree you are referring to here.

I used the device tree from the toradex_4.19.y branch. For proper working in mainline kernel I also patched tegra20.dtsi and tegra20-colibri.dtsi.

We’ve also found that scaling the CPU frequency, unlike with kernel 3.1.10, has no impact on power consumption. Could there be a problem with the clock / cpu settings or drivers?

So far I just disabled that feature and otherwise had quite good results with mainline on the Colibri T20.

The actueal problem is the waste heat of the CPU or can 46 ° C with switched on graphics output in idle mode be considered suitable?

I previously forgot to comment on the following:

I assume that switching the sleep mode on the Colibri is not working properly yet, cause he could not confirm the issue with tegra2 on an Acer A500 tablet.

What you are talking about is just CPU idle not to be confused with any suspend states. I am not aware that anything special being hardware related would need to be done for that.

Furthermore the the cpu is getting very hot in idle state (~46°C), even with going into sleep mode 1 (before system freeze), but i think that is caused by the problem i mentioned above.

I would not consider 46 degree celsium to be particularly hot. The SoC can easily sustain more than double.

I used the device tree from the toradex_4.19.y branch.

As mentioned before that branch has absolutely no official meaning beyond being used for Torizon on i.MX 6 and 7 SoCs. Do not rely on it!

For proper working in mainline kernel I also patched tegra20.dtsi and tegra20-colibri.dtsi.

I am not aware of any such being needed at all resp. you may rather want to look at the upcoming 4.20 rework here, here or here instead. What exact changes are you talking about?

We’ve also found that scaling the CPU frequency, unlike with kernel 3.1.10, has no impact on power consumption.

I don’t think 4.19 does support CPU frequency scaling on T20 at all.

Could there be a problem with the clock / cpu settings or drivers?

As I do not know what exactly you modified this is tricky to say. However, I would rather guess the issue being somewhat related to your configuration.

The actueal problem is the waste heat of the CPU or can 46 ° C with switched on graphics output in idle mode be considered suitable?

Suitable or not depends entirely on your requirements but definitely not an issue per se.

did you ever find the reason for the CPU idle freeze?

No, not really but later Linux kernel 5.4 would now come with full CPU and memory DVFS support for T20 as well (T30 was added just very recently in 5.5). However, we have not really validated any of this on any Toradex hardware be it Colibri T20 or Apalis/Colibri T30 as of yet.