Fix raspberry pi zero cpu variant recognition

[lstolcman]

The PR fixes a corner-case problem in which images for raspberry pi 1 or raspberry pi zero single board computers are downloaded for a wrong architecture, resulting in unability to run containers.

Also, PR adds some prints to improve the debugging.

The problem may be easily reproduced when using a rpi 1 or rpi zero with podman:

current version:

> podman run --log-level=info --syslog busybox echo ok
INFO[0000] podman filtering at log level info
INFO[0000] Setting parallel job count to 4
Resolved "busybox" as an alias (/home/pi/.cache/containers/short-name-aliases.conf)
Trying to pull docker.io/library/busybox:latest...
Getting image source signatures
Copying blob 0983f321071f done
Copying config 87837ce2bf done
Writing manifest to image destination
Storing signatures
INFO[0011] Running conmon under slice user.slice and unitName libpod-conmon-148722857d0739cc36a346b4f035448abb0736ea3b10c04132735da82e96b49d.scope
INFO[0012] Got Conmon PID as 21404
INFO[0012] Received shutdown.Stop(), terminating!        PID=21384
➜  podman image ls
REPOSITORY                 TAG         IMAGE ID      CREATED       SIZE
docker.io/library/busybox  latest      87837ce2bf82  3 weeks ago   2.66 MB

expected output:

➜  podman run --log-level=info --syslog busybox echo ok
INFO[0000] ./podman filtering at log level info
INFO[0000] Setting parallel job count to 4
Resolved "busybox" as an alias (/home/pi/.cache/containers/short-name-aliases.conf)
Trying to pull docker.io/library/busybox:latest...
Getting image source signatures
Copying blob 60966b1a2cb7 done
Copying config 9f28bca8fa done
Writing manifest to image destination
WARNING: image platform (linux/arm/v6) does not match the expected platform (linux/arm)
INFO[0009] Received shutdown.Stop(), terminating!        PID=21440
INFO[0010] Failed to add conmon to cgroupfs sandbox cgroup: creating cgroup path conmon: open /sys/fs/cgroup/cgroup.subtree_control: permission denied
INFO[0010] Got Conmon PID as 21459
ok
➜  bin podman image ls
REPOSITORY                 TAG         IMAGE ID      CREATED       SIZE
docker.io/library/busybox  latest      9f28bca8fad0  3 weeks ago   1.91 MB

As we can see, image for arm v7 is downloaded (87837ce2bf82) instead of image for arm v6 (9f28bca8fad0)

I would also expect that explicit notice of the error been returned to the console when running failed (like in docker), however it is not a case here. I believe it is another issue, to be submitted directly to podman repo.
The error 139 is reported in syslog though: conmon fba43d5acde494cff782 <ninfo>: container 2072 exited with status 139

references:

• Add RPi1/RPi0 workaround containerd/containerd#4530
• https://www.raspberrypi.org/forums/viewtopic.php?t=12614
• Add multi-arch image support for ARM moby/moby#36121 (comment)

[mtrmac]

Thanks!

I know very little about ARM, still, looking at the available documentation, this doesn’t seem to be a very precise fix. Ultimately actually reading the CPU registers directly seems more accurate… but that would require an ARM expert to discuss what the relevant values for the “v6”/“v7” ABI.

Meh, being consistent with containerd seems a good enough of an improvement for now.

[mtrmac]

Non-blocking:

Plausibly getCPUInfo could parse everything at once and return a map[key]value. That would remove some of the error paths….

[mtrmac]

I would also expect that explicit notice of the error been returned to the console when running failed (like in docker), however it is not a case here. I believe it is another issue, to be submitted directly to podman repo.

Yes, please file that in https://github.com/containers/podman/issues .

[rhatdan]

@lstolcman are you still interested in this PR?

[lstolcman]

@rhatdan yes I am. Since I do not have time at the moment to apply the fixes, feel free to do it yourself (or anyone else interested), or I will do once I have some more free time

[lstolcman]

@rhatdan @mtrmac please review again

[rhatdan]

LGTM

[mtrmac]

Thanks!

Code LGTM. The commit messages do not always match the commit contents; it would probably be simplest to rebase + squash this into one commit now.

[mtrmac]

Thanks again!