On Mi, 2018-10-03 at 02:12 +0300, Grazvydas Ignotas wrote:
Some more tests:
- disabling LPAE has no effect, same hang after few minutes
- attempting to use 2GB MLO/u-boot on new 4GB board makes it very
unstable (crashes or hangs early while booting the kernel)
Ah, thanks for the info. It's interesting that LPAE doesn't have an effect, as it had one with our first 4GB RAM board tests, afair. But that was with a different kernel and DT files.
The 2GB MLO / u-boot has different timings, so it seems the timings certainly has an effect.
Hm, but would idle / power saving trigger when the system is doing something? Because when doing my test runs, I always run a script that spits out the voltages, temperatures and remaining temperature left every 10 seconds.
And what does it do in between those 10s? If the system has no work for any periods, even as short as 1ms, it's called idle and power saving mechanisms start.
Oh, okay. Didn't know powersaving is THAT aggressive. That's a good idea then!
BTW: When I ran the 4GB unit from the SD Card that Nikolaus gave me (with QuantumSTEP installed), it didn't fully freeze. While the Heartbeat of the LEDs stopped and it looked at first glance it had frozen, it was still running incredibly slow, as the touchscreen still reacted.
I didn't have it connected to serial out, so I have no idea if this was still reacting or not. But at least some part of i2c stopped working, hence the frozen LEDs.
Here's a video of that (don't mind the German comments ;)) https://cloud.openpandora.org/index.php/s/mfGobZfeFRyKS6P
I'm not sure whether it even really ever froze on QuantumStep. And this "half-freeze" usually happened after 5 - 10 minutes, so the same time it freezes with out Linux.
Not sure if this is the same or even related, but I thought I'd mention it.
- Boot the kernel directly into a shell (without loading any other drivers except for the ones directly included in the kernel).
Test whether this runs more stable and then modprobe the remaining drivers one after the other. After all, Tony was running the memtest similar as well, and that didn't fail.
Yeah that's something to try. Perhaps just move /lib/modules somewhere so they don't autoload (not sure if they go to the ramdisk currently).
Shouldn't it be enough booting into /bin/bash as initrd?