Il 11/08/2013 10:29, Alex Bligh ha scritto: > Paolo, > > --On 11 August 2013 09:53:38 +0200 Paolo Bonzini <pbonz...@redhat.com> > wrote: > >> There is actually a disadvantage of moving TimerListGroup to AioContext. >> The disadvantage is that GSources can only work with millisecond >> resolution. Thus you would need anyway some special casing of the >> "timer AioContext" to get the deadline in nanoseconds. > > We also need to special case the notifier as it needs to qemu_notify() > rather than aio_notify().
No, not really, because qemu_notify_event() is exactly the same aio_notify(main_loop_context). You do not need to special case the notifier even if TimerListGroup remains separate. >> So let's keep the TimerListGroup for now. > > OK - do you want me to wrap it in a struct? Other than that I think I've > done all the comments in v8. Happy to do that with v10 if there are > other comments on v9. No, it's okay. Unless you want to remove the callback and use aio_notify everywhere. > I note no one has yet commented on the changes to the icount stuff > where a timeout is apparently arbitrarily capped at 2^31 ns (about > 2.1 seconds) in PATCHv9 19/31 - attached below. That's the area > I'm worried about as I'm not sure I understood the code. Yes, that's ok. I'm more worried about the thread safety, but it's not a problem yet. Paolo
