[LITMUS^RT] request for review: patches for arbitrary deadlines

[Jonathan Herman]

We could avoid future API pain by passing in initialized rt_task
structs into sporadic_task_ns. This would also make it easier to add
in extra per-task fields for per-paper branches, which has been
handled awkwardly in the past. Besides, I've heard that 7 was a good
hard limit for number of parameters to a function. Now
sporadic_task_ns has 8.

> +	lt_t		rdeadline;

Why not just deadline? I doubt there will be confusion, and it looks a
tad cleaner.

> -	if (tp.period < tp.exec_cost)
> +	if (tp.rdeadline < tp.exec_cost)
>  	{
>  		printk(KERN_INFO "litmus: real-time task %d rejected "
> -		       "because wcet > period\n", pid);
> +		       "because wcet greater than relative deadline\n", pid);
>  		goto out_unlock;
>  	}

As far as I can tell, this means there is no longer a check that
tp.period >= tp.exec_cost. Am I wrong?

> +	BUG_ON(!t);
> +
> +	/* prepare next release */
> +	t->rt_param.job_params.release = start;
> +	t->rt_param.job_params.deadline  = start + get_rt_rdeadline(t);
> +	t->rt_param.job_params.exec_time = 0;
> +
> +	/* update job sequence number */
> +	t->rt_param.job_params.job_no++;
> +
> +	/* don't confuse Linux */
> +	t->rt.time_slice = 1;
> +

The following might be a lesser evil. Duplicated logic is always scary.

inline void do_release(struct task_struct *t, lt_t time)
{
	BUG_ON(!t);

	/* prepare next release */
	t->rt_param.job_params.release = time;
	t->rt_param.job_params.deadline  = time+ get_rt_rdeadline(t);
	t->rt_param.job_params.exec_time = 0;

	/* update job sequence number */
	t->rt_param.job_params.job_no++;

	/* don't confuse Linux */
	t->rt.time_slice = 1;
}

void release(struct task_struct *t)
{
        do_release(get_release(t) + get_rt_deadline(t));
}

void release_at(struct task_struct *t, lt_t time)
{
        do_release(t, time);
        set_rt_flags(t, RT_F_RUNNING);
}


Having the capabilities in this patch seems super reasonable to me.

On Mon, Jul 16, 2012 at 5:52 PM, Glenn Elliott <gelliott at cs.unc.edu> wrote:
> Hello All,
>
> Please find two patched attached for review that add support for
> arbitrary deadlines in Litmus.  One patch is for litmus and the other
> is for liblitmus.  Please read the patch commit messages for more
> information.
>
> I'm not exactly happy with the API changes for liblitmus because they
> are disruptive with legacy code (a relative deadline has been added).
> However, this seemed to be the correct change with respect to the API
> though.
>
> Patches are also available for view at rtsrv.unc.edu/cgit for UNC folks.
>
> Thanks,
> Glenn
>
> _______________________________________________
> litmus-dev mailing list
> litmus-dev at lists.litmus-rt.org
> https://lists.litmus-rt.org/listinfo/litmus-dev
>



-- 
Jonathan Herman
Department of Computer Science at UNC Chapel Hill