block nbd0: Unexpected reply

[Mike Holmes]

http://validation.linaro.org/scheduler/job/80934/log_file#L_67_526 block nbd0: Unexpected reply (ec24dda0)
http://validation.linaro.org/scheduler/job/80934/log_file#L_19_530 /dev/nbd0 disconn[ 779.818827] block nbd0: Unexpected reply (ec727da0)
http://validation.linaro.org/scheduler/job/80934/log_file#L_67_517 mount: special device /dev/nbd0p2 does not exist

There appear to be a number of failures related to kvm tests.

[Mike Holmes (mike-holmes)]

This also occurs with the Ubuntu filesystem, http://validation.linaro.org/scheduler/job/81145/log_file#L_37_4116

[Kim Phillips (kim-phillips)]

qemu 1.6.50 (at least on OE) is an experimental version - this should be switched to qemu v1.6.1.

Other than that, a few upstream commits have been identified that fix issues with nbd disconnect.

[Mike Holmes (mike-holmes)]

Still occurring http://validation.linaro.org/scheduler/job/84007/log_file#L_67_523

[Maxim Uvarov (maxim-uvarov)]

Might be also ndb-server crash on host, like this:
http://comments.gmane.org/gmane.linux.drivers.nbd.general/706

[Kim Phillips (kim-phillips)]

This bug has two components:

(a) the unreliable connect, which occurs only sometimes, and more likely in -rt, as Mike's last log reference points out.
(b) the samesake, "block nbd0: Unexpected reply", that's reliable (occurs every time), but has no apparent consequence to the kvm test (ie., the disconnect still occurs successfully).

So I personally consider (a) more important than (b), since when (a) occurs, files are not available for the remaining kvm test script, and the test eventually fails.

I got (a) to reliably fail on my local arndale board. I updated qemu to ToT, i.e., commit 1680d485777ecf436d724631ea8722cc0c66990e (in the v1.6.90 range). After the update, nbd always succeeds to connect.

So, in investigating how to update meta-linaro, I see the following commit:

commit 119e9e32d10cf8c5b89bb421b36aa6fca6167935
Author: Riku Voipio <email address hidden>
Date: Thu Sep 12 15:14:32 2013 +0300

    qemu: move to git head

    Set up a qemu recipe that follows qemu head and builds for both aarch64
    and armv7.

yet can't find the reason why qemu isn't being automatically updated north of 1.6.50 for armv7 runs.

Meanwhile, it appears Peter Maydell's tree is being used for aarch64.

Can someone more bitbake savvy take a look at why meta-linaro isn't building qemu past 1.6.50? Is it not being run, or did commit 3edcdd64a82bda2d2585fadb89705e6a7c341000 "qemu_git: fix native build" somehow break it?

Thanks,

Kim

[Kim Phillips (kim-phillips)]

Kim to confirm with Anders that qemu has been updated to between v1.7-rc1 and rc2, which fixes the connect problem (fatal).

putting the disconnect problem on the back burner for now since it's innocuous.

[Kim Phillips (kim-phillips)]

qemu has been updated, no connection problems have been seen since, degrading importance due to disconnect not being fatal.

[Kim Phillips (kim-phillips)]

decision has been made to not pursue fixing the non-fatal disconnect (a virtio storage mechanism should really replace qcow file usage anyway, mount via loopback).