Re: [RFC] sched/deadline: Prevent rt_time growth to infinity

[Juri Lelli <juri.lelli@gmail.com>]

On Fri, 21 Feb 2014 16:09:25 +0400
Kirill Tkhai <tkhai@yandex.ru> wrote:

> 
> 
> 21.02.2014, 15:39, "Kirill Tkhai" <tkhai@yandex.ru>:
> > 21.02.2014, 14:37, "Peter Zijlstra" <peterz@infradead.org>:
> >
> >>  On Thu, Feb 20, 2014 at 02:16:00AM +0400, Kirill Tkhai wrote:
> >>>   Since deadline tasks share rt bandwidth, we must care about
> >>>   bandwidth timer set. Otherwise rt_time may grow up to infinity
> >>>   in update_curr_dl(), if there are no other available RT tasks
> >>>   on top level bandwidth.
> >>>
> >>>   I'm going to decide the problem the way below. Almost untested
> >>>   because of I skipped almost all of recent patches which haveto be applied from lkml.
> >>>
> >>>   Please say, if I skipped anything in idea. Maybe better put
> >>>   start_top_rt_bandwidth() into set_curr_task_dl()?
> >>  How about we only increment rt_time when there's an RT bandwidth timer
> >>  active?
> >
> > This case RT and DL may eat all the time:
> >
> > --------------  time ------------------>
> > |RT's working |DL's working|
> > ----------------------------
> > |rt_runtime   |            |
> > ----------------------------
> > |        rt_period         |
> >
> > Or at least more, than it's allowed.
> >
> > It looks like, if we want to limit time of high priority classes
> > execution, we have to set the timer anyway.
> 
> Oh, above is confusing. Sorry.
> 
> I mean one RT task and
> 
> -----------------  time -------------------------->
> |DL's working    |RT's working| |DL's working    |
> --------------------------------------------------
> |                |rt_runtime  | |                |
> --------------------------------------------------
> |                |   rt_period  |                |
> 
> In this case FAIR receives less ratio, than (rt_period-rt_runtime)/rt_period.
> 

DL tasks won't be allowed to run in this situation, as their bw exceedes
rt_runtime/rt_period.

Thanks,

- Juri