Yocto version

General Discussions
1

Hi,

I’m working on a Raspberry 3 Compute Module based project. I’m having issues integrating Mender based on dunfell. What is/are the current supported version(s) I can use?

With kind regards,
Matthijs ter Woord

2

Hi @mterwood

While we have not explicitly tested the CM on dunfell (at least I have not), I would expect it to work since it is supposed to be binary compatible with the RPI3. I suppose it’s possible the drivers are a bit different based on what is included in the CM and is what is part of the base board. Can you elaborate on what issues you are having?

Drew

3

Based on Raspberry Pi 3 Model B/B+, mender was validated to work with Dunfell.
Can you double-check your steps with the ones described in the link above?

4

Somehow I feel like I’m losing sanity here.

I have a local.conf file, with a custom layer. This layer contains, amongst other small stuff, an image definition. When I build that image, it takes a while and the build fails. Retrying then (maybe once or twice more) results in a successful build. Am I missing something?

5

What I now did, was clean the tmp, sstate and out directories. Then build the image. It fails on building binutils.

Retry, and everything builds fine. Is this something others see as well?

I collected logs of both build attempts, if anyone is interested.

6

Sounds to me like a race condition of some kind. If you can share the failing build log that may be instructive.
Drew

7

See the following gist for log files. First one is from the failing attempt, second one from the succeeding, which was executed exactly the same way.

8

Hmm. I see the following in your failure log.

{standard input}: Error: open CFI at the end of file; missing .cfi_endproc directive
/yocto/out/TWCB_COMBINED/work/cortexa7t2hf-neon-vfpv4-poky-linux-gnueabi/binutils/2.34-r0/recipe-sysroot/usr/include/c++/9.3.0/bits/stl_vector.h:1195:4: note: parameter passing for argument of type '__gnu_cxx::__normal_iterator<long long unsigned int*, std::vector<long long unsigned int> >' changed in GCC 7.1
 1195 |    _M_realloc_insert(end(), __x);
      |    ^~~~~~~~~~~~~~~~~
arm-poky-linux-gnueabi-g++: fatal error: Killed signal terminated program cc1plus
compilation terminated.
Makefile:1137: recipe for target 'powerpc.o' failed
make[4]: *** [powerpc.o] Error 1
make[4]: *** Waiting for unfinished jobs....

I googled the error messages and it seems related to a memory issue, out of memory or something. Can you try again with less parallelism?

What is your build host? Is it a native install or a VM? How much RAM? How many CPUs?

Drew

9

Build host is a VM on HyperV. Has 12 virtual CPU’s (out of 12 cores/threads of an i7-6800k).

Runs on a RAID5 disk setup, with 8GB of ram. Wil increase that to 16 and try a full build again. Will keep you posted.

I really hope this is the issue, will keep you posted. (I googled now and see the hints towards out of memory. If that’s the case, I’m feeling stupid)

10

I seem to recall seeing somewhere a recommendation of 1GB per core so hopefully, that is your issue.

11

Looking at the VM’s host (instead of connecting via SSH), I now see a lot of “Out of memory: killed process” messages. I think we’re close here. :slight_smile:

12

That solved it. Complete build within 2 hours, without any errors or retries. Thanks for googling for me. Will try to do myself next time…