Re: [Xen-devel] [PATCH] xl: check for meaningful combination of sedf config file parameters

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Wed, 2012-06-06 at 12:05 +0100, Dario Faggioli wrote:
> On Wed, 2012-06-06 at 11:35 +0100, Ian Campbell wrote:
> > > diff --git a/tools/libxl/xl_cmdimpl.c b/tools/libxl/xl_cmdimpl.c
> > > --- a/tools/libxl/xl_cmdimpl.c
> > > +++ b/tools/libxl/xl_cmdimpl.c
> > > @@ -561,6 +561,7 @@ static void parse_config_data(const char
> > >      long l;
> > >      XLU_Config *config;
> > >      XLU_ConfigList *cpus, *vbds, *nics, *pcis, *cvfbs, *cpuids;
> > > +    int opt_w = 0, opt_p = 0, opt_s = 0;
> > 
> > These names don't make much sense in this context.
> > 
> Yeah, I agree... It's just I needed something :-/
> 
> > Perhaps you can just check each interesting option against the
> > corresponding LIBXL_DOAIN_SCHED_PARAM_DEFAULT? 
> >
> Mmm... I was mistakenly thinking these default values not to be there
> yet, but I now see it. Yes, I guess I can do that.
> 
> > That might make some long
> > lines. Perhaps pulling this out into a separate valid_sched_params()
> > would help with that?
> >
> Maybe, but what to put here depends on your thought on the below...
> 
> > >      if (!xlu_cfg_get_long (config, "latency", &l, 0))
> > >          b_info->sched_params.latency = l;
> > >      if (!xlu_cfg_get_long (config, "extratime", &l, 0))
> > >          b_info->sched_params.extratime = l;
> > > +    /* The sedf scheduler needs some more consistency checking */
> > > +    if (opt_w && (opt_p || opt_s)) {
> > > +        fprintf(stderr, "Either specify a weight OR a period and 
> > > slice\n");
> > 
> > Does this constrain you from setting valid combinations of credit*
> > parameters? I think not since period and slice are SEDF specific.
> > 
> I'd say not at all, for the exact reason you're suggesting. Then, if you
> ask what happens if you boot with sched=credit and then try to specify
> both a cpu_weight and a period, then yes, it will kick you out.
> 
> The whole point is, period and slice are only meaningful for sedf so, if
> you are using them, I take it like you meant to be using sedf, and thus
> asking for a cpu_weight at the same time is wrong.
> 
> Of course, one can think at it the other way around (scheduler is
> credit, so cpu_weight is fine and period and slice should be ignored).
> If that is better, I can add a libxl_is_the_scheduler_credit? kind of
> check to that if...

Lets keep it simple for now and go with the "don't do that" answer --
i.e. reject as invalid setting weight and period regardless of the
actual scheduler in use.

Ian.


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