Actual LTS Kernel (4.9) on Colibri T20

Hello,

i running the latest LTS Kernel from the Yocto-Project (4.9.46) on my Colibri T20.
Most of the system is running fine until now.

But if i activate the display controller and output an image (or framebuffer console) on the RGB interface the system freezes after a few minutes. No more communication over serial or network or any further interaction with the system is possible.
Without an activated display controller the system runs stable for days without any problems.

There is no information about a kernel panic or oops on the serial line. So i have no idea where to start to find the problem.
Are there any knowen issues with the kernel drivers?
Do you have an idea how to investigate this problem?

Greetings
Jochen Jägers

What exact kind of user space are you running? I believe I remember having seen this when mixing and matching incompatible stuff.

I’m running a yocto pyro userspace and tried the following things for the graphics output.

  1. just a kernel framebuffer console. No special drivers or libaries in userspace
  2. XServer (1.19.1) with fbdev driver
  3. XServer (1.19.1) with opentegra driver, libdrm and mesa from the grate project ( grate-driver · GitHub )
  4. added libgrate from GitHub - grate-driver/grate: Open source Tegra2+ 2D/3D user-space developer tools and repeated version 3

The result is the same in every version. The system freezes after a few minutes

I did some testing on my own and figured something got broken towards 4.9 which I am now in the process of bisecting. I will report back once I figured out what exactly is going wrong. For now it looks like 4.4.95 runs stable and anyway would be a really nice candidate due to its extended LTS status.

Having attempted a multi-level bisection points towards the following first bad commit:

cpuidle: menu: Fix menu_select() for CPUIDLE_DRIVER_STATE_START == 0

Unfortunately as that file has gotten further edits it is not trivial to just revert it.

However I found out that it indeed has to do with CPU idle which may be disabled via the following Linux kernel configuration option: CONFIG_CPU_IDLE. After disabling that one also LTS 4.9.59 as well as latest 4.14-rc7 run stable overnight.

I will enquire on the Tegra mailing list about this as well. Maybe somebody there knows what exactly may be causing this.

Thanks for your effort. Is there any way I can support you to find the problem?

Thanks for your effort.

You are very welcome.

Is there any way I can support you to find the problem?

As I do not know your level of expertise this is hard to say. You may have a look at the feedback from Dmitry on the mailing list. I will try a few more of his and my suggestions once I get to it however that may take a few more days.

Are there any news on this?

No, unfortunately not.

Hi @Jochen,

Would it be possible to share some details on u-boot version and/or kernel config.
I’m not really familiar with yocto, but if you could provide me with a link to kernel sources used there, that would be usefull as well.
I just try to boot 4.9.x (git repo) but unfortunately it doesn’t boot at all.

Thanks in advance.

Kr,
Bart.

Could you please ask a new question rather than hijacking this one? Thanks!