[Bug 1935617] Re: systemd autopkgtest broken on ppc64el with qemu 6.0

[Christian Ehrhardt ]

Since the bare metal case reports a memory issue on the crashes, let us
give it a bit more memory - with 2G it works finally.

Instead of the other tests in launchpad/canonistack this uses KVM mode
via "-machine accel=kvm -enable-kvm -cpu host". Switching that instead
to "-machine accel=tcg" it fails the same.

But I found what consumed all the memory, the guest is (by default)
started with the same CPU set as the host which means -smp 156.
Disabling SMT in general on the host and starting the guest with 4 cpus
allows to reduce the memory to 1G (but 512M still fails). Some other
environment detail seems to bump the memory demand up.

As a cross check I can it on canonistack with 2G but it still fails with
the segfaults.

With the adapted config it worked in qemu 5.2 in KVM and TCG mode.

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to systemd in Ubuntu.
https://bugs.launchpad.net/bugs/1935617

Title:
  systemd autopkgtest broken on ppc64el with qemu 6.0

Status in qemu package in Ubuntu:
  New
Status in systemd package in Ubuntu:
  New

Bug description:
  I'm not sure yet if this is flaky or a real issue, but I'm filing it
  to avoid multiple people analyzing the same.

  The Qemu 6.0 upload https://launchpad.net/ubuntu/+source/qemu/1:6.0+dfsg-1~ubuntu2 triggers a test failure like
  https://autopkgtest.ubuntu.com/results/autopkgtest-impish/impish/ppc64el/s/systemd/20210708_223311_e3bbb@/log.gz

  I have tested the new qemu on ppc64 and it worked fine for device emulation and migration cases.
  But this is suspicious.
  Of the last tests exactly and only those with the new qemu failed.

  
  impish
    ppc64el
      tests-in-lxd                   (F  5% f  0% S  0% B  0% => P 95%/) F.........F.............................
      systemd-fsckd                  (F  0% f  0% S 100% B  0% => P  0%/) SSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSS
      upstream-1                     (F 15% f  0% S  0% B  0% => P 85%/) F..FFFF....F............................
      upstream-2                     (F 12% f  0% S  0% B  0% => P 87%/) F..FFFF.................................

  
  For an insight in flakyness/reproducibility I've retriggered the missing qemu and the non-qemu cases a few times. If those reproduce all-bad vs all-good again this would further indicate a real issue.

  Unfortunately the ppc maas seems down right now and canonistack also
  isn't too nice this week - overall that inhibits the testing a bit :-/

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/qemu/+bug/1935617/+subscriptions