Re: [RFC, PATCH 1/2] sched: change the fairness model of the CFS group scheduler

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Mon, 2008-02-25 at 19:46 +0530, Dhaval Giani wrote:
> This patch allows tasks and groups to exist in the same cfs_rq. With this
> change the CFS group scheduling follows a 1/(M+N) model from a 1/(1+N)
> fairness model where M tasks and N groups exist at the cfs_rq level.

Good. You know I agree with this direction.

> Signed-off-by: Dhaval Giani <dhaval@linux.vnet.ibm.com>
> Signed-off-by: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
> ---
>  kernel/sched.c      |   46 +++++++++++++++++++++
>  kernel/sched_fair.c |  113 +++++++++++++++++++++++++++++++++++++++++-----------
>  2 files changed, 137 insertions(+), 22 deletions(-)
> 
> Index: linux-2.6.25-rc2/kernel/sched.c
> ===================================================================
> --- linux-2.6.25-rc2.orig/kernel/sched.c
> +++ linux-2.6.25-rc2/kernel/sched.c
> @@ -224,10 +224,13 @@ struct task_group {
>  };
>  
>  #ifdef CONFIG_FAIR_GROUP_SCHED
> +
> +#ifdef CONFIG_USER_SCHED
>  /* Default task group's sched entity on each cpu */
>  static DEFINE_PER_CPU(struct sched_entity, init_sched_entity);
>  /* Default task group's cfs_rq on each cpu */
>  static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp;
> +#endif
>  
>  static struct sched_entity *init_sched_entity_p[NR_CPUS];
>  static struct cfs_rq *init_cfs_rq_p[NR_CPUS];
> @@ -7163,6 +7166,10 @@ static void init_tg_cfs_entry(struct rq 
>  		list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list);
>  
>  	tg->se[cpu] = se;
> +	/* se could be NULL for init_task_group */
> +	if (!se)
> +		return;
> +
>  	se->cfs_rq = &rq->cfs;
>  	se->my_q = cfs_rq;
>  	se->load.weight = tg->shares;
> @@ -7217,11 +7224,46 @@ void __init sched_init(void)
>  #ifdef CONFIG_FAIR_GROUP_SCHED
>  		init_task_group.shares = init_task_group_load;
>  		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
> +#ifdef CONFIG_CGROUP_SCHED
> +		/*
> +		 * How much cpu bandwidth does init_task_group get?
> +		 *
> +		 * In case of task-groups formed thr' the cgroup filesystem, it
> +		 * gets 100% of the cpu resources in the system. This overall
> +		 * system cpu resource is divided among the tasks of
> +		 * init_task_group and its child task-groups in a fair manner,
> +		 * based on each entity's (task or task-group's) weight
> +		 * (se->load.weight).
> +		 *
> +		 * In other words, if init_task_group has 10 tasks of weight
> +		 * 1024) and two child groups A0 and A1 (of weight 1024 each),
> +		 * then A0's share of the cpu resource is:
> +		 *
> +		 * 	A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
> +		 *
> +		 * We achieve this by letting init_task_group's tasks sit
> +		 * directly in rq->cfs (i.e init_task_group->se[] = NULL).
> +		 */
> +		init_tg_cfs_entry(rq, &init_task_group, &rq->cfs, NULL, i, 1);
> +		init_tg_rt_entry(rq, &init_task_group, &rq->rt, NULL, i, 1);

That seems to agree with that.

> +#elif defined CONFIG_USER_SCHED
> +		/*
> +		 * In case of task-groups formed thr' the user id of tasks,
> +		 * init_task_group represents tasks belonging to root user.
> +		 * Hence it forms a sibling of all subsequent groups formed.
> +		 * In this case, init_task_group gets only a fraction of overall
> +		 * system cpu resource, based on the weight assigned to root
> +		 * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished
> +		 * by letting tasks of init_task_group sit in a separate cfs_rq
> +		 * (init_cfs_rq) and having one entity represent this group of
> +		 * tasks in rq->cfs (i.e init_task_group->se[] != NULL).
> +		 */
>  		init_tg_cfs_entry(rq, &init_task_group,
>  				&per_cpu(init_cfs_rq, i),
>  				&per_cpu(init_sched_entity, i), i, 1);

But I fail to parse this lengthy comment. What does it do:

    init_group
   /     |    \
uid-0 uid-1000 uid-n

or does it blend uid-0 into the init_group?

>  
>  #endif
> +#endif /* CONFIG_FAIR_GROUP_SCHED */
>  #ifdef CONFIG_RT_GROUP_SCHED
>  		init_task_group.rt_runtime =
>  			sysctl_sched_rt_runtime * NSEC_PER_USEC;
> @@ -7435,6 +7477,10 @@ static int rebalance_shares(struct sched
>  		unsigned long total_load = 0, total_shares;
>  		struct task_group *tg = cfs_rq->tg;
>  
> +		/* Skip this group if there is no associated group entity */
> +		if (unlikely(!tg->se[this_cpu]))
> +			continue;

I'm not directly seeing where tg->se[cpu] == NULL comes from..

>  		/* Gather total task load of this group across cpus */
>  		for_each_cpu_mask(i, sdspan)
>  			total_load += tg->cfs_rq[i]->load.weight;
> Index: linux-2.6.25-rc2/kernel/sched_fair.c
> ===================================================================
> --- linux-2.6.25-rc2.orig/kernel/sched_fair.c
> +++ linux-2.6.25-rc2/kernel/sched_fair.c
> @@ -732,6 +732,21 @@ static inline struct sched_entity *paren
>  	return se->parent;
>  }
>  
> +/* return the cpu load contributed by a given group on a given cpu */
> +static inline unsigned long group_cpu_load(struct task_group *tg, int cpu)
> +{
> +	struct sched_entity *se = tg->se[cpu], *top_se;
> +	struct cfs_rq *cfs_rq = tg->cfs_rq[cpu];
> +
> +	if (unlikely(!se))
> +		return cfs_rq->load.weight;
> +
> +	for_each_sched_entity(se)
> +		top_se = se;
> +
> +	return top_se->load.weight;
> +}
> +
>  #define GROUP_IMBALANCE_PCT	20
>  
>  #else	/* CONFIG_FAIR_GROUP_SCHED */
> @@ -1073,6 +1088,17 @@ out_set_cpu:
>  }
>  #endif /* CONFIG_SMP */
>  
> +/* return depth at which a sched entity is present in the hierarchy */
> +static inline int depth_se(struct sched_entity *se)
> +{
> +	int depth = 0;
> +
> +	for_each_sched_entity(se)
> +		depth++;
> +
> +	return depth;
> +}
> +
>  
>  /*
>   * Preempt the current task with a newly woken task if needed:
> @@ -1083,6 +1109,7 @@ static void check_preempt_wakeup(struct 
>  	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
>  	struct sched_entity *se = &curr->se, *pse = &p->se;
>  	unsigned long gran;
> +	int se_depth, pse_depth;
>  
>  	if (unlikely(rt_prio(p->prio))) {
>  		update_rq_clock(rq);
> @@ -1100,6 +1127,27 @@ static void check_preempt_wakeup(struct 
>  	if (!sched_feat(WAKEUP_PREEMPT))
>  		return;
>  
> +	/*
> +	 * preemption test can be made between sibling entities who are in the
> +	 * same cfs_rq i.e who have a common parent. Walk up the hierarchy of
> +	 * both tasks untill we find their ancestors who are siblings of common
> +	 * parent.
> +	 */
> +
> +	/* First walk up until both entities are at same depth */
> +	se_depth = depth_se(se);
> +	pse_depth = depth_se(pse);
> +
> +	while (se_depth > pse_depth) {
> +		se_depth--;
> +		se = parent_entity(se);
> +	}
> +
> +	while (pse_depth > se_depth) {
> +		pse_depth--;
> +		pse = parent_entity(pse);
> +	}

Sad, but needed.. for now..

>  	while (!is_same_group(se, pse)) {
>  		se = parent_entity(se);
>  		pse = parent_entity(pse);
> @@ -1166,13 +1214,22 @@ static void put_prev_task_fair(struct rq
>  static struct task_struct *
>  __load_balance_iterator(struct cfs_rq *cfs_rq, struct rb_node *curr)
>  {
> -	struct task_struct *p;
> +	struct task_struct *p = NULL;
> +	struct sched_entity *se;
>  
>  	if (!curr)
>  		return NULL;
>  
> -	p = rb_entry(curr, struct task_struct, se.run_node);
> -	cfs_rq->rb_load_balance_curr = rb_next(curr);
> +	/* Skip over entities that are not tasks */
> +	do {
> +		se = rb_entry(curr, struct sched_entity, run_node);
> +		curr = rb_next(curr);
> +	} while (curr && !entity_is_task(se));
> +
> +	cfs_rq->rb_load_balance_curr = curr;
> +
> +	if (entity_is_task(se))
> +		p = task_of(se);
>  
>  	return p;
>  }
> @@ -1210,21 +1267,28 @@ load_balance_fair(struct rq *this_rq, in
>  		struct cfs_rq *this_cfs_rq = busy_cfs_rq->tg->cfs_rq[this_cpu];
>  		unsigned long maxload, task_load, group_weight;
>  		unsigned long thisload, per_task_load;
> -		struct sched_entity *se = busy_cfs_rq->tg->se[busiest->cpu];
> +		struct task_group *tg = busy_cfs_rq->tg;
> +		struct sched_entity *se = tg->se[busiest->cpu],
> +				    *this_se = tg->se[this_cpu];
>  
>  		task_load = busy_cfs_rq->load.weight;
> -		group_weight = se->load.weight;
> +		if (!task_load)
> +			continue;
> +
> +		group_weight = group_cpu_load(tg, busiest->cpu);
>  
>  		/*
> -		 * 'group_weight' is contributed by tasks of total weight
> +		 * 'group_weight' is contributed by entities of total weight
>  		 * 'task_load'. To move 'rem_load_move' worth of weight only,
>  		 * we need to move a maximum task load of:
>  		 *
>  		 * 	maxload = (remload / group_weight) * task_load;
>  		 */
>  		maxload = (rem_load_move * task_load) / group_weight;
> +		if (!maxload)
> +			continue;
>  
> -		if (!maxload || !task_load)
> +		if (!busy_cfs_rq->nr_running)
>  			continue;
>  
>  		per_task_load = task_load / busy_cfs_rq->nr_running;
> @@ -1253,23 +1317,28 @@ load_balance_fair(struct rq *this_rq, in
>  					       &cfs_rq_iterator);
>  
>  #ifdef CONFIG_FAIR_GROUP_SCHED
> -		/*
> -		 * load_moved holds the task load that was moved. The
> -		 * effective (group) weight moved would be:
> -		 * 	load_moved_eff = load_moved/task_load * group_weight;
> -		 */
> -		load_moved = (group_weight * load_moved) / task_load;
> -
>  		/* Adjust shares on both cpus to reflect load_moved */
> -		group_weight -= load_moved;
> -		set_se_shares(se, group_weight);
> +		if (likely(se)) {
> +			unsigned long load_moved_eff;
> +			unsigned long se_shares;
>  
> -		se = busy_cfs_rq->tg->se[this_cpu];
> -		if (!thisload)
> -			group_weight = load_moved;
> -		else
> -			group_weight = se->load.weight + load_moved;
> -		set_se_shares(se, group_weight);
> +			/*
> +			 * load_moved holds the task load that was moved. The
> +			 * effective (group) weight moved would be:
> +			 *	load_moved_eff = load_moved/task_load *
> +			 *					group_weight;
> +			 */
> +			load_moved_eff = (se->load.weight *
> +						 load_moved) / task_load;
> +
> +			set_se_shares(se, se->load.weight - load_moved_eff);
> +			if (!thisload)
> +				se_shares = load_moved_eff;
> +			else
> +				se_shares = this_se->load.weight +
> +							load_moved_eff;
> +			set_se_shares(this_se, se_shares);
> +		}
>  #endif
>  
>  		rem_load_move -= load_moved;

Didn't pay too much attention to the lb part as you said that needs
work.