Re: [Qemu-devel] [RFC v2 0/5] arm AioContext with its own timer stuff

[Alex Bligh]

--On 29 July 2013 15:36:03 +0200 Stefan Hajnoczi <address@hidden> 
wrote:

> > I should say PingFan has probably paid more attention to thread safety
> > than me, as my work was intended to be applied before AioContexts were
> > used by multiple threads.
>
> Exactly.  That's why I think these patches should be based on your
> series - they add the thread safety on top.

OK. I think I'm pretty much done with my series and have incorporated
everyone's comments to date (save a couple I pushed back on of
a minor nature and heard no response).

The only other thing I was thinking of was rather than each QemuTimerList
holding an AioContext pointer so it knows it can use aio_notify, providing
a callback for notify (this is on clock enable). I think that's probably
polishing the ball though.

On the thread issue, for my own education, if an AioContext's timers
only ever get called by that AioContext (and thus in the same thread),
which is the case in PATCHv4, I don't think there is a thread safety
beyond manipulation of the list of QemuTimerLists associated with
each QEMUClock. That's only ever modified when the AioContext is
created or deleted. This assumes that timers are always set from
the same thread they are run (i.e. there is only one thread
walking the QEMUTimerList).

--
Alex Bligh