On Mon, Jul 29, 2013 at 9:11 PM, Paolo Bonzini <pbonz...@redhat.com> wrote:
> Il 29/07/2013 10:20, liu ping fan ha scritto:
>>> > Another issue is that deadline computation is not using the AioContext's
>>> > timer lists.
>>> >
>> Sorry, can not catch the meaning. When AioContext has its own timer
>> lists, it will have its own deadline(for ppoll timeout). So the
>> computation should be based on AioContext's timer lists, right?
>
> Associating the main loop's timer lists to other AioContexts is really
Main loop's timers will only be associated with the qemu_aio_context,
not with other AioContext. (When creating timer, we have bound it with
a specified AioContext, so it will only run on that AioContext)
> really wrong... It is _already_ wrong to run timers in the main
> AioContext's aio_poll (because timers may not be ready to run from an
Sorry, but except that timer cb will call aio_poll, any other reason ?
> arbitrary aio_poll), it is even worse to run them in other threads
> without taking the BQL.
>
Can be fixed by some method to sync the dedicated thread with vcpus?
> What exactly is the purpose of this series?
>
Enable timers to run on the dedicated threads(using aio as the
mini-loop), currently for the following devices,
virtio-blk-dataplane for throttling
hpet for the best-effort resolution
Thanks and regards,
Pingfan
> Paolo
