Orange Pi Plus2E out of memory errors?

[sheffield_nick]

This is with standard Armbian Debian GNU/Linux 8 (jessie) 3.4.112-sun8i at the stock 1.296GHz CPU speed, and there is a fan+heatsink to stop overheating. I've disabled swap.

 

# memtester 239 5

is OK, while

# memtester 240 5

crashes out almost immediately:

[10933.657662] Out of memory: Kill process 17061 (memtester) score 89 or sacrifice child
[10933.657673] Killed process 17061 (memtester) total-vm:246992kB, anon-rss:245380kB, file-rss:28kB

If I enable a small 32MB swap file, memtester seems to work fine up to 1200MB (still 750MB reported free, and no swap used) and crashes out for any larger values?? I'm getting the same out of memory problems trying to run the hpcc (Linpack) benchmark - very small values of N work, but anything larger gets killed with out of memory.
 

I never got these errors using Raspbian/Debian Jessie 8.4 on the Raspberry Pi Zero/3 or NanoPC-T3 boards (all with swap disabled).

 

Any suggestions much appreciated! Thanks

[tkaiser]

If I enable a small 32MB swap file

 

Armbian uses 128MB for a reason. It doesn't hurt (since it won't be used anyway) and it prevents the issues you're running into. Please keep in mind that the oop-killer trouble you're running into is not related to the distro you use but the kernel. Armbian's legacy kernel sources for H3 are a smelly 3.4.39 Allwinner Android kernel we put +100 patches on top while we're waiting for mainline kernel being ready.

 

So either waste huge amounts of your time debugging Android kernel behaviour or simply stay with Armbian defaults (enabling monitoring is easy: It's just an 'sudo armbianmonitor -r' to install RPi-Monitor so you can have an eye on '0 bytes swap used' even if 128MB are defined)

[sheffield_nick]

Fantastic - many thanks for the help! I'll give that a go.

[sheffield_nick]

Increasing the swap to 128MB hasn't helped - memtester is still being killed off with "out of memory"

 

Armbian uses 128MB for a reason.

[jernej]

I also get instant crash on my OpenELEC fork for opiplus2e. sheffield_nick, are you using desktop version of Armbian, e.g. with graphical interface? There are some fixes for A20 legacy kernel for 2GB memory, but they are all related to video and gpu driver. If you confirm this, it is worth looking into...

[sheffield_nick]

No, no desktop here. Logging into a shell via ssh.

[sheffield_nick]

Still getting errors, even with a 128MB swap file. Just tried linpack with settings that use around 1100MB of RAM, and I got these 5 errors from 5 *different* Plus2E boards. None were overheating.

 

Do you think it could be a kernel bug, or perhaps running the memory too fast? Is there a tool I can use to check all 2GB of the Plus2E memory?

[orangepiplus2e:02404] *** Process received signal ***
[orangepiplus2e:02404] Signal: Segmentation fault (11)
[orangepiplus2e:02404] Signal code: Address not mapped (1)
[orangepiplus2e:02404] Failing at address: 0xc
[orangepiplus2e:02404] *** End of error message ***

[orangepiplus2e:02615] *** Process received signal ***
[orangepiplus2e:02615] Signal: Segmentation fault (11)
[orangepiplus2e:02615] Signal code: Address not mapped (1)
[orangepiplus2e:02615] Failing at address: 0xeddee65c
[orangepiplus2e:02615] *** End of error message ***

[orangepiplus2e:02329] *** Process received signal ***
[orangepiplus2e:02329] Signal: Bus error (7)
[orangepiplus2e:02329] Signal code: Invalid address alignment (1)
[orangepiplus2e:02329] Failing at address: 0xa014b64a
[orangepiplus2e:02329] *** End of error message ***

[orangepiplus2e:02326] *** Process received signal ***
[orangepiplus2e:02326] Signal: Segmentation fault (11)
[orangepiplus2e:02326] Signal code: Address not mapped (1)
[orangepiplus2e:02326] Failing at address: 0x41bb9110
[orangepiplus2e:02326] *** End of error message ***

mpirun noticed that process rank 2 with PID 2353 on node orangepiplus2e exited on signal 4 (Illegal instruction).