Re: [PATCH 2/4] sched: set new prio after checking schedule policy

[Valentin Schneider <valentin.schneider@arm.com>]

On 24/04/20 05:36, Hillf Danton wrote:
> Set newprio the same way as normal_prio() does after checking schedule
> policy and to MAX_PRIO -1 by default.
>
> What is also added is boundary checks for RT and fair priorities.
>
> Cc: Mike Galbraith <efault@gmx.de>
> Cc: Steven Rostedt <rostedt@goodmis.org>
> Cc: Mel Gorman <mgorman@techsingularity.net>
> Cc: Vincent Guittot <vincent.guittot@linaro.org>
> Cc: Phil Auld <pauld@redhat.com>
> Cc: Valentin Schneider <valentin.schneider@arm.com>
> Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
> Cc: Mark Rutland <mark.rutland@arm.com>
> Signed-off-by: Hillf Danton <hdanton@sina.com>
> ---
>
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -4768,8 +4768,7 @@ static int __sched_setscheduler(struct t
>                               const struct sched_attr *attr,
>                               bool user, bool pi)
>  {
> -	int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 :
> -		      MAX_RT_PRIO - 1 - attr->sched_priority;
> +	int newprio;
>       int retval, oldprio, oldpolicy = -1, queued, running;
>       int new_effective_prio, policy = attr->sched_policy;
>       const struct sched_class *prev_class;
> @@ -4800,6 +4799,26 @@ recheck:
>               return -EINVAL;
>
>       /*
> +	 * compute newprio after checking policy, see normal_prio();
> +	 * it's used in pi boost below
> +	 */
> +	if (dl_policy(policy)) {
> +		newprio = MAX_DL_PRIO - 1;
> +	}
> +	else if (rt_policy(policy)) {
> +		if (attr->sched_priority > MAX_RT_PRIO - 1)
> +			return -EINVAL;
> +		newprio = MAX_RT_PRIO - 1 - attr->sched_priority;
> +	}
> +	else if (fair_policy(policy)) {
> +		if (attr->sched_nice < MIN_NICE ||
> +		    attr->sched_nice > MAX_NICE)
> +			return -EINVAL;


We can't hit this with the syscall route, since we (silently) clamp those
values in sched_copy_attr(). setpriority() does the same. There's this
comment in sched_copy_attr() that asks whether we should clamp or return an
error; seems like the current consensus is on clamping, but then we might
want to get rid of that comment :)

> +		newprio = NICE_TO_PRIO(attr->sched_nice);

This is new, however AFAICT it doesn't change anything for CFS (or about to
be) tasks since what matters is calling check_class_changed() further
down.

> +	} else
> +		newprio = MAX_PRIO - 1;
> +
> +	/*
>        * Allow unprivileged RT tasks to decrease priority:
>        */
>       if (user && !capable(CAP_SYS_NICE)) {