systemd 233-8ubuntu1 ADT test failure with linux 4.12.0-4.5

[Seth Forshee]

Testing failed on:
    ppc64el: https://objectstorage.prodstack4-5.canonical.com/v1/AUTH_77e2ada1e7a84929a74ba3b87153c0ac/autopkgtest-artful-canonical-kernel-team-unstable/artful/ppc64el/s/systemd/20170627_083238_11cda@/log.gz

[Dimitri John Ledkov (xnox)]

========== TEST-05-RLIMITS ==========
make: Entering directory '/tmp/autopkgtest.ipRf3i/build.uhX/systemd-233/test/TEST-05-RLIMITS'
TEST CLEANUP: Resource limits-related tests
TEST SETUP: Resource limits-related tests
0+0 records in
0+0 records out
0 bytes copied, 0.000211387 s, 0.0 kB/s
Checking that no-one is using this disk right now ... OK

Disk /dev/loop4: 400 MiB, 419430400 bytes, 819200 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes

>>> Created a new DOS disklabel with disk identifier 0xa1dd3fbe.
/dev/loop4p1: Created a new partition 1 of type 'Linux' and of size 390 MiB.
/dev/loop4p2: Created a new partition 2 of type 'Linux' and of size 9 MiB.
/dev/loop4p3: Done.

New situation:

Device Boot Start End Sectors Size Id Type
/dev/loop4p1 2048 800767 798720 390M 83 Linux
/dev/loop4p2 800768 819199 18432 9M 83 Linux

The partition table has been altered.
Calling ioctl() to re-read partition table.
Syncing disks.
The file /dev/loop4p1 does not exist and no size was specified.
F: Failed to mkfs -t ext3
Makefile:4: recipe for target 'setup' failed

[Dimitri John Ledkov (xnox)]

The test doesn't even get past the setup function.

The setup function is quite basic. Create an image, partition the image, create filesystem on the image.

https://git.launchpad.net/~ubuntu-core-dev/+git/systemd/tree/test/test-functions?h=ubuntu-artful#n333

It appears that the kernel has failed to re-read the partition table, as /dev/loop4p1 is expected to exist.

Is sfdisk not sufficient to re-read the parition table?! i wonder if between calling sfdisk and mkfs, and udevadm wait for idle state is needed.

[Dimitri John Ledkov (xnox)]

(As in, there is a race that mkfs was attempted before /dev/loop4p1 device was created, however i thought return from ioctl to re-read parition table returns only after the p* devices have been created by the kernel)

dmesg from this machine would be interesting.

[Seth Forshee (sforshee)]

By the time BLKRRPART returns the partition device nodes should have been created in devtmpfs, so there shouldn't be any need to wait for things to settle. I'm not sure where a race could exist.

Yet it's odd that this failed only for ppc since all the code which deals with this is generic, not platform-specific. It would be nice if we had dmesg, unfortunately we do not.

I'm re-running the test just to see if it happens consistently. We have a ppc test machine, however it's currently occupied running tests. Once it's free I can do some testing there.

[Seth Forshee (sforshee)]

I failed to follow up.

There were no failures when re-running the test, and no errors with 4.12.0-5.6, so marking this invalid.