Re: [Qemu-devel] [PATCH] [RFC] aio/timers: Drop alarm timers; introduce QEMUClock to AioContext; run timers in aio_poll

[Stefan Hajnoczi]

On Fri, Jul 19, 2013 at 06:26:23PM +0100, Alex Bligh wrote:
> [ This is a patch for RFC purposes only. It is compile tested on Linux x86_64 
> only
> and passes make check (or rather did before make check started dying in the
> boot order test - different bug). I'd like to know whether I'm going in
> the right direction ]

Looks promising.

> We no longer need alarm timers to trigger QEMUTimer as we'll be polling
> them in aio_poll.
> 
> Remove static declaration from qemu_new_clock and introduce qemu_free_clock.
> 
> Maintain a list of QEMUClocks.
> 
> Introduce qemu_clock_deadline_ns and qemu_clock_deadine_all_ns which 
> calculate how
> long aio_poll etc. should wait, plus (for the time being) a conversion to 
> milliseconds.
> 
> Make qemu_run_timers return a bool to indicate progress.
> 
> Add QEMUClock to AioContext.
> 
> Run timers attached to clock in aio_poll

Too many logical changes for a single patch :).  Please split this into
a series.

> @@ -52,6 +44,8 @@ struct QEMUClock {
>  
>      int type;
>      bool enabled;
> +
> +    QLIST_ENTRY(QEMUClock) list;

Please avoid global state.  AioContext should be usable from multiple
threads, this list would require synchronization.

main-loop.c should call functions to run timers and calculate the
nearest deadline on vm_clock/host_clock/rt_clock.  It knows about them,
they are defined in <qemu/timer.h>.

That way no list is needed.