Bug 1081795

Ok guys, the volume isn't the problem at all.

I see the volume jump to 100% for a short time with ac97 (just let pavucontrol run with the applications tab open and watch the slider).  It jumps back before the actual playback starts.  With intel-hda the volume jump isn't present in the first place.

Nevertheless we have the sound artifact issue with both ac97 and intel-hda.  And it isn't just loud, it is simply noise.  It sounds that way if you try to playback a memory buffer which has random data instead of audio samples.  So most likely someone got the sound buffer management wrong here.

It doesn't look like qemu is at fault here though.  If I cut off spice from the sound path, by starting qemu with "QEMU_AUDIO_DRV=pa" environment variable set (audio goes "qemu -> pulse" instead of "qemu -> spice-server -> spice-client -> pulse" then) the sound artifact goes away.

Assigning back to spice for further investigation.

moving to 7.2

Is this a rhel only issue? I can't reproduce with f21 both with ac97 and hda.
Even the patch from https://bugzilla.redhat.com/show_bug.cgi?id=1012868 doesn't seem to change anything. Perhaps the recent codec/sound work changed something.
thanks

rhel7 client to f21 hosts: glitch is audible
f21 client to f21 hosts: no glitch*

* an awful volume jump may occur on Fedora client however, hopefully addressed in recent patch series

BTW: when the agent is connected, the migration between f21 hosts fail - the destination host segfaults (no backtrace ATM), the client disconnects and the VM keeps running in the source host. In the linux guest, I needed to switch to text console in order to make migration succeed...

(In reply to David Jaša from comment #8)
> rhel7 client to f21 hosts: glitch is audible
> f21 client to f21 hosts: no glitch*
> 
> * an awful volume jump may occur on Fedora client however, hopefully
> addressed in recent patch series

Can you confirm if this bug is still reproducible with patches from https://bugzilla.redhat.com/show_bug.cgi?id=1012868 ?

> 
> BTW: when the agent is connected, the migration between f21 hosts fail - the
> destination host segfaults (no backtrace ATM), the client disconnects and
> the VM keeps running in the source host. In the linux guest, I needed to
> switch to text console in order to make migration succeed...

If you reproduce, please open a bug for it.

(In reply to Victor Toso from comment #9)
> (In reply to David Jaša from comment #8)
> > rhel7 client to f21 hosts: glitch is audible
> > f21 client to f21 hosts: no glitch*
> > 
> > * an awful volume jump may occur on Fedora client however, hopefully
> > addressed in recent patch series
> 
> Can you confirm if this bug is still reproducible with patches from
> https://bugzilla.redhat.com/show_bug.cgi?id=1012868 ?

I can't reproduce the issue in current 7.2 builds that don't include the spice-server patch (new patches are included in the rebased spice-gtk)
spice-gtk3-0.26-3.el7.x86_64
spice-server-0.12.4-9.el7.x86_64
qemu-kvm-1.5.3-93.el7.x86_64

I we won't hit the issue again further in the release cycle, we can close as CURRENTRELEASE.

> 
> > 
> > BTW: when the agent is connected, the migration between f21 hosts fail - the
> > destination host segfaults (no backtrace ATM), the client disconnects and
> > the VM keeps running in the source host. In the linux guest, I needed to
> > switch to text console in order to make migration succeed...
> 
> If you reproduce, please open a bug for it.

David just tested this and we can confirm that this is fixed in current release.

We were able to trigger the glitch in 6.5 client even with newer spice-server. But 6.7 and 7.2 clients does not trigger the glitch in any situation.
Most likely this was fixed in spice-gtk between 6.5 and 6.7.

Thanks David.