[Spice-devel] Current qemu-master hangs when used with qxl + linux guest

[Hans de Goede]

Hi,

On 10/08/2013 04:30 PM, Daniel P. Berrange wrote:
> On Tue, Oct 08, 2013 at 04:27:38PM +0200, Hans de Goede wrote:
>> Hi All,
>>
>> I'm having this weird problem with qemu master + spice/qxl using
>> guests. As soon as the guest starts Xorg, I get the following message
>> from qemu:
>>
>> main-loop: WARNING: I/O thread spun for 1000 iterations
>>
>> And from then on the guest hangs and qemu consumes 100% cpu. The qemu
>> console still works, and I can quit qemu that way.
>>
>> Doing ctrl+c + a thread apply all bt in qemy shows one cpu thread waiting
>> for the iothread-lock, and all other threads waiting in poll.
>>
>> This happens both with non kms guests (tried RHEL-6.5, older Fedoras)
>> as well as with kms guests (tried a fully up2date F-19).
>>
>> Since I've not seen any similar reports, I assume it is something
>> with my setup ...
>>
>> I've tried changing various things:
>> -removing the spice agent channel
>> -changing the number of virtual cpus (tried 1 and 2 virtual cpus)
>> -upgrading spice-server to the latest git master
>>
>> But all to no avail.
>>
>> This is with qemu-master build from source on a fully up2date
>> F-20 system, using the F-20 seabios files.
>>
>> If someone has any clever ideas I'll happily try debugging this further.
>
> Does the QEMU build in F20 work correctly ?

Ah, yes it does, good dea.

> If so that'd give you a
> starting point from which to 'git bisect' the problem.

Yep, note I asked for a "clever" idea, iow not the big hammer of bisecting :)

But since 1.6.0 in F-20 worked it was not such a large bisect, so I went
ahead and bisected anyways, which pointed me to commit 7b595f35d8 :
"aio / timers: Convert mainloop to use timeout"

After careful review of that commit I had a hunch what might be wrong, and
it turned out to be right. So after this mail I'm going to send a patch
fixing this. If you want to know the details see the patch, titled:
"main-loop: Don't lock starve io-threads when main_loop_tlg has pending events"

Regards,

Hans