Re: [PATCH v4 3/4] sched: Allow sched_{get,set}attr to change latency_nice of the task

[Parth Shah <parth@linux.ibm.com>]

On 2/25/20 12:24 PM, Pavan Kondeti wrote:
> On Mon, Feb 24, 2020 at 02:29:17PM +0530, Parth Shah wrote:
> 
> [...]
> 
>> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
>> index 65b6c00d6dac..e1dc536d4ca3 100644
>> --- a/kernel/sched/core.c
>> +++ b/kernel/sched/core.c
>> @@ -4723,6 +4723,8 @@ static void __setscheduler_params(struct task_struct *p,
>>  	p->rt_priority = attr->sched_priority;
>>  	p->normal_prio = normal_prio(p);
>>  	set_load_weight(p, true);
>> +
>> +	p->latency_nice = attr->sched_latency_nice;
>>  }
> 
> We don't want this when SCHED_FLAG_LATENCY_NICE is not set in
> attr->flags.
> 
> The user may pass SCHED_FLAG_KEEP_PARAMS | SCHED_FLAG_LATENCY_NICE to
> change only latency nice value. So we have to update latency_nice
> outside __setscheduler_params(), I think.


AFAICT, passing SCHED_FLAG_KEEP_PARAMS with any other flag will prevent us
from changing the latency_nice value because of the below code flow:

__sched_setscheduler()
	__setscheduler()
		if (attr->sched_flags & SCHED_FLAG_KEEP_PARAMS) return;
		__setscheduler_params()

whereas, one thing we still can do is add if condition when setting the
value, i.e.,

@@ -4724,7 +4724,8 @@ static void __setscheduler_params(struct task_struct *p,
        p->normal_prio = normal_prio(p);
        set_load_weight(p, true);

-       p->latency_nice = attr->sched_latency_nice;
+       if (attr->sched_flags & SCHED_FLAG_LATENCY_NICE)
+               p->latency_nice = attr->sched_latency_nice;
 }


> 
>>  
>>  /* Actually do priority change: must hold pi & rq lock. */
>> @@ -4880,6 +4882,13 @@ static int __sched_setscheduler(struct task_struct *p,
>>  			return retval;
>>  	}
>>  
>> +	if (attr->sched_flags & SCHED_FLAG_LATENCY_NICE) {
>> +		if (attr->sched_latency_nice > MAX_LATENCY_NICE)
>> +			return -EINVAL;
>> +		if (attr->sched_latency_nice < MIN_LATENCY_NICE)
>> +			return -EINVAL;
>> +	}
>> +
>>  	if (pi)
>>  		cpuset_read_lock();
>>  
>> @@ -4914,6 +4923,9 @@ static int __sched_setscheduler(struct task_struct *p,
>>  			goto change;
>>  		if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)
>>  			goto change;
>> +		if (attr->sched_flags & SCHED_FLAG_LATENCY_NICE &&
>> +		    attr->sched_latency_nice != p->latency_nice)
>> +			goto change;
>>  
>>  		p->sched_reset_on_fork = reset_on_fork;
>>  		retval = 0;
>> @@ -5162,6 +5174,9 @@ static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *a
>>  	    size < SCHED_ATTR_SIZE_VER1)
>>  		return -EINVAL;
>>  
>> +	if ((attr->sched_flags & SCHED_FLAG_LATENCY_NICE) &&
>> +	    size < SCHED_ATTR_SIZE_VER2)
>> +		return -EINVAL;
>>  	/*
>>  	 * XXX: Do we want to be lenient like existing syscalls; or do we want
>>  	 * to be strict and return an error on out-of-bounds values?
>> @@ -5391,6 +5406,8 @@ SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
>>  	else
>>  		kattr.sched_nice = task_nice(p);
>>  
>> +	kattr.sched_latency_nice = p->latency_nice;
>> +
> 
> Can you consider printing latency_nice value in proc_sched_show_task() in this
> patch/series?

Sure, I will add it.


Thanks,
Parth