Re: [Qemu-devel] [PATCH 1/3] colo-compare: reconstruct the mutex lock usage

[Jason Wang <jasowang@redhat.com>]

On 2017年02月14日 15:33, Hailiang Zhang wrote:
> On 2017/2/14 12:08, Jason Wang wrote:
>>
>>
>> On 2017年02月14日 10:32, Hailiang Zhang wrote:
>>>>>
>>>>
>>>
>>> Hi Jason,
>>>
>>>> I think the better way is notify the comparing thread and let it do 
>>>> the
>>>> releasing. You probably need similar mechanism to notify from 
>>>> comparing
>>>> thread to checkpoint thread.
>>>>
>>>
>>> It seems that there is no available APIs in glib to notify a thread 
>>> which
>>> are run coroutine to do something (idle source ?). What about using
>>> anonymous pipe
>>> as the GPollFD to communicate between colo comparing thread and colo
>>> thread ?
>>>
>>> Any ideas ?
>>>
>>> Hailiang
>>
>> Haven't thought this deeply. But I think you can try event notifier
>> first which was designed for such kind of notification.
>>
>
> Hmm, It is quite same with what i said above except it uses the
> return value of eventfd as the GPollFD.
> The things here are more complex especially if there are more than
> one vNIC, which each is corresponding to one compare thread.
>
> How about using g_source_attach(), g_source_attach(), g_source_remove(),
> g_source_set_callback() and g_source_set_priority() to control the 
> process
> of releasing packets dynamically ? Which we call g_source_attach()
> to add the coroutine of releasing packets when do checkpoint, and remove
> it once finishing the checkpoint process.

I'm not sure I get the idea, maybe you can post patch for early review.

>
> About colo comparing thread notifies colo thread, i think we can use
> qemu_cond_wait(),qemu_cond_broadcast(), just like what we do in
> pause_all_vcpus().

Yes, agree.

Thanks

>
> Thanks.
> Hailiang
>
>> Thanks
>>
>> .
>>
>
>