Re: [PATCH v2] sched: Enabled schedstat when schedstat tracepoints are enabled

From: Steven Rostedt
Date: Thu Apr 13 2017 - 10:14:25 EST


On Thu, 13 Apr 2017 04:05:48 -0700
Srikar Dronamraju <srikar@xxxxxxxxxxxxxxxxxx> wrote:

> > +#if defined(CONFIG_TRACING) && defined(CONFIG_SCHEDSTATS)
> > +int schedstat_tracepoint_reg(void);
> > +void schedstat_tracepoint_unreg(void);
> > +#else
> > +static inline int schedstat_tracepoint_reg(void) { return 0; }
> > +static inline void schedstat_tracepoint_unreg(void) { }
> > +#endif
> > +
> >
>
>
> > +#ifdef CONFIG_TRACING
>
> Shouldn't this be also
> > +#if defined(CONFIG_TRACING) && defined(CONFIG_SCHEDSTATS) ????

This code is within a #ifdef CONFIG_SCHEDSTATS block. That would be
redundant putting it here too.

-- Steve

>
> If CONFIG_TRACING is defined but CONFIG_SCHEDSTATS is not then build
> should complain about duplicate schedstat_tracepoint_reg(). No?
>
> > +static int schedstat_tracepoint_ref;
> > +static bool schedstat_save_state;
> > +/*
> > + * schedstat_tracepoint_reg() and unreg() are called by the tracepoint
> > + * regfunc/unregfunc functions. They are protected by the tracepoint mutex.
> > + * See kernel/tracepoint.c:tracepoint_add_func().
> > + *
> > + * The modifications to schedstat_tracepoint_ref and schedstat_save_state
> > + * are only done under that mutex, and do not need further protection.
> > + */
> > +int schedstat_tracepoint_reg(void)
> > +{
> > + if (!schedstat_tracepoint_ref) {
> > + schedstat_save_state = schedstat_enabled();
> > + if (!schedstat_save_state)
> > + set_schedstats(true);
> > + }
> > + schedstat_tracepoint_ref++;
> > + return 0;
> > +}
> > +
> > +void schedstat_tracepoint_unreg(void)
> > +{
> > + schedstat_tracepoint_ref--;
> > + if (schedstat_tracepoint_ref || schedstat_save_state)
> > + return;
> > + set_schedstats(false);
> > +}
> > +#endif