Re: [Xen-devel] [PATCH v11]xen: sched: convert RTDS from time to event driven model

["Chen, Tianyang" <tiche@xxxxxxxxxxxxxx>]

On 03/17/2016 02:27 PM, Dario Faggioli wrote:
> > Finally, when waking up a vcpu, it is now enough to tickle the
> > various
> > CPUs appropriately, like all other schedulers also do.
> >
> > Signed-off-by: Tianyang Chen <tiche@xxxxxxxxxxxxxx>
> > Signed-off-by: Meng Xu <mengxu@xxxxxxxxxxxxx>
> > Signed-off-by: Dagaen Golomb <dgolomb@xxxxxxxxxxxxxx>
> You seem to have taken care of all my remaining comments to v10, so the
> patch is:
>
> Reviewed-by: Dario Faggioli <dario.faggioli@xxxxxxxxxx>
>
> (which you could have put there yourself, as I said. :-))
>
> Thanks for all the good work.
:-)

> If I can already point you to something else that can be improved
> further in this scheduler, _completely_ independently from this, and
> hence in a fully new patch (series), can you have a look at whether we
> really need set_bit(), test_and_clear(), etc, instead of __set_bit(),
> __test_and_clear(), etc.?
>
> I'm asking because I'm under the impression that the latter are enough,
> and if that is the case, we should go for them, as they're more
> efficient.
Sure. __set_bit() and __test_and_clear() are not atomic and can be 
reordered. They need to be protected by locks. So, it looks like where 
RTDS is using them already have spin_locks in place and re-ordering 
inside the critical region doesn't really mess anything up.
rt_schedule() sets: __RTDS_scheduled, __RTDS_delay_runq_add
burn_budget() sets: __RTDS_depleted
rt_wake() sets: __RTDS_delay_runq_add

rt_vcpu_sleep() clears: __RTDS_delayed_runq_add
rt_context_saved() clears: __RTDS_scheduled, __RTDS_delayed_runq_add
repl_timer_handler() clears: __RTDS_depleted

burn_budget() is inside of rt_schedule() but they are using different 
flags of different vcpus. So it's enough to just use __set_bit(), 
__clear_bit() and __test_and_clear().
I'm assuming the new patch should be based on the last patch I sent? 
Because replenishment handler is not presented in the old RTDS code?
Thanks,
Tianyang Chen


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel