Re: [PATCH RFC 3/8] sched/cpufreq_schedutil: make worker kthread be SCHED_DEADLINE

[Peter Zijlstra]

On Tue, May 23, 2017 at 09:53:46AM +0100, Juri Lelli wrote:
> diff --git a/include/uapi/linux/sched.h b/include/uapi/linux/sched.h
> index e2a6c7b3510b..72723859ef74 100644
> --- a/include/uapi/linux/sched.h
> +++ b/include/uapi/linux/sched.h
> @@ -48,5 +48,6 @@
>   */
>  #define SCHED_FLAG_RESET_ON_FORK	0x01
>  #define SCHED_FLAG_RECLAIM		0x02
> +#define SCHED_FLAG_SPECIAL		0x04
>  
>  #endif /* _UAPI_LINUX_SCHED_H */
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 7fc2011c3ce7..ba57e2ef9aef 100644

> @@ -4205,7 +4212,9 @@ static int __sched_setscheduler(struct task_struct *p,
>  	}
>  
>  	if (attr->sched_flags &
> -		~(SCHED_FLAG_RESET_ON_FORK | SCHED_FLAG_RECLAIM))
> +		~(SCHED_FLAG_RESET_ON_FORK |
> +		  SCHED_FLAG_RECLAIM |
> +		  SCHED_FLAG_SPECIAL))
>  		return -EINVAL;
>  
>  	/*

Could we pretty please not expose this gruesome hack to userspace?

So if you stick it in attr->sched_flags, use a high bit and don't put it
in a uapi header. Also make the flags check explicitly fail on it when
@user. Such that only _nocheck() (and thus kernel) callers have access
to it.

Also, there's not nearly enough warnings and other derisory comments
near it.