Re: [Qemu-devel] [PATCH] reenable vm_clock when resuming all vcpus

[Eduardo Habkost]

On Tue, Nov 08, 2011 at 11:24:30AM -0600, Anthony Liguori wrote:
> On 11/03/2011 09:45 PM, Wen Congyang wrote:
> >We disable vm_clock when pausing all vcpus, but we forget to
> >reenable it when resuming all vcpus. It will cause that the
> >guest can not be rebooted.
> >
> >Signed-off-by: Wen Congyang<address@hidden>
> 
> Applied.  Thanks.

FYI, I found out that this patch solved an issue we were having in our
Autotest runs: https://bugs.launchpad.net/qemu/+bug/886255

But the reason it fixed the issue are a mystery to me. The symptoms
were:

- A RHEL 6.1 guest was installed
- After booting, sshd created the ssh keys normally
- Guest was rebooted (I was not seeing the "guest can't be rebooted"
  issue mentioned on the description of this patch)
- After rebooting, the ssh key files on /etc/ssh had size 0 for some
  unknown reason
- sshd failed to re-create the keys because of the empty key files, and
  failed to start
  - The immediate reason was because the (empty) public key file has
    selinux context that can't be written by ssh-keygen, but that only
    happened because the files become empty.

For some reason, the problem stopped happening after this commit. I have
no idea why it fixed the issue. I don't know if it fixed the actual bug
that caused the problem, or the bug is timing-dependent and for some
reason this committed just changed the timing conditions and the bug is
still hidden in qemu.

The bug was 100% reproducible on the machine I was testing, previously,
and after this commit it didn't happen anymore. I double-checked by
re-testing this commit (47113ab6), and the previous one (f67ab77a)
twice, after bisecting.

(If anybody has some possible explanation on how this patch could have
any impact on the above symptoms, or want to investigate further, please
let me know. I think I will stop debugging this, move on, and finally
stress-test my qemu_fclose() series, now.)

-- 
Eduardo