Re: [patch 02/38] CKRM e18: Processor Delay Accounting

[Gerrit Huizenga]

On Thu, 23 Jun 2005 11:37:32 +0200, Ingo Molnar wrote:
> 
> * Ingo Molnar <mingo@xxxxxxx> wrote:
> 
> > 
> > * Gerrit Huizenga <gh@xxxxxxxxxx> wrote:
> > 
> > > +#ifdef CONFIG_DELAY_ACCT
> > > +int task_running_sys(struct task_struct *p)
> > > +{
> > > +	return task_is_running(p);
> > > +}
> > > +EXPORT_SYMBOL_GPL(task_running_sys);
> > > +#endif
> > 
> > why is this function defined, and why is it exported?

This was exported so it could be used in the classification engine
which is a loadable module which determines how and when tasks join
classes.  There are two classification engines - a basic one (RBCE)
and a more advanced one (CRBCE).  The latter allows a user to set some
basic rules on how newly created tasks will join a class.

> this:
> 
> +#define task_is_running(p)     (this_rq() == task_rq(p))
> 
> is totally bogus. What you are checking is not whether 'the task is 
> running', but it is a complex way of doing p->thread_info->cpu == 
> smp_processor_id(). This, combined with:
> 
> +               if (pdata == NULL)
> +                       /* some wierdo race condition .. simply ignore */
> +                       continue;
> +               if (thread->state == TASK_RUNNING) {
> +                       if (task_running_sys(thread)) {
> +                               atomic_inc((atomic_t *) &
> +                                          (PSAMPLE(pdata)->cpu_running));
> +                               run++;
> +                       } else {
> +                               atomic_inc((atomic_t *) &
> +                                          (PSAMPLE(pdata)->cpu_waiting));
> +                               wait++;
> +                       }
> +               }
> 
> yields completely incorrect code, and bogus data. If your goal is to 
> sample executing-on-cpu vs. on-runqueue-waiting-to-run states then you 
> should use the already existing task_curr(p) call.
> 
> 	Ingo

I'll clean this up later this afternoon and regen a patch.

thanks!

gerrit
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/