【内核调度、负载均衡】【find_energy_efficient_cpu】

struct energy_env

struct energy_env {
	/* Utilization to move */
	struct task_struct	*p;
	unsigned long		util_delta;//Task­­ utilization的估计值
	unsigned long		util_delta_boosted;//boost Task­­ utilization的值

	/* Mask of CPUs candidates to evaluate */
	cpumask_t		cpus_mask;

	/* CPU candidates to evaluate */
	struct eenv_cpu *cpu;
	int eenv_cpu_count;

#ifdef DEBUG_EENV_DECISIONS
	/* pointer to the memory block reserved
	 * for debug on this CPU - there will be
	 * sizeof(struct _eenv_debug) *
	 *  (EAS_CPU_CNT * EAS_EENV_DEBUG_LEVELS)
	 * bytes allocated here.
	 */
	struct _eenv_debug *debug;
#endif
	/*//最节能的CPU的索引
	 * Index (into energy_env::cpu) of the morst energy efficient CPU for
	 * the specified energy_env::task//
	 */
	int	next_idx;
	int	max_cpu_count;

	/* Support data */
	struct sched_group	*sg_top;
	struct sched_group	*sg_cap;
	struct sched_group	*sg;
};


struct _eenv_debug {
	unsigned long cap;
	unsigned long norm_util;
	unsigned long cap_energy;
	unsigned long idle_energy;
	unsigned long this_energy;
	unsigned long this_busy_energy;
	unsigned long this_idle_energy;
	cpumask_t group_cpumask;
	unsigned long cpu_util[1];
};

struct eenv_cpu

struct eenv_cpu {
	/* CPU ID, must be in cpus_mask */
	int     cpu_id;

	/*
	 * Index (into sched_group_energy::cap_states) of the OPP the
	 * CPU needs to run at if the task is placed on it.
	 * This includes the both active and blocked load, due to
	 * other tasks on this CPU,  as well as the task's own
	 * utilization.
	*/
	int     cap_idx;
	int     cap;

	/* Estimated system energy */
	unsigned long energy;

	/* Estimated energy variation wrt EAS_CPU_PRV */
	long nrg_delta;

#ifdef DEBUG_EENV_DECISIONS
	struct _eenv_debug *debug;
	int debug_idx;
#endif /* DEBUG_EENV_DECISIONS */
};

reset_eenv

reset_eenv实际上就是复位eenv结构体，同时保存eenv->eenv_cpu_count = cpu_count;、

static inline void reset_eenv(struct energy_env *eenv)
{
	int cpu_count;
	struct eenv_cpu *cpu;
#ifdef DEBUG_EENV_DECISIONS
	struct _eenv_debug *debug;
	int cpu_idx;
	debug = eenv->debug;
#endif

	cpu_count = eenv->eenv_cpu_count;
	cpu = eenv->cpu;
	memset(eenv, 0, sizeof(struct energy_env));
	eenv->cpu = cpu;
	memset(eenv->cpu, 0, sizeof(struct eenv_cpu)*cpu_count);
	eenv->eenv_cpu_count = cpu_count;

#ifdef DEBUG_EENV_DECISIONS
	memset(debug, 0, eenv_debug_size());
	eenv->debug = debug;
	for (cpu_idx = 0; cpu_idx < eenv->eenv_cpu_count; cpu_idx++)
		eenv->cpu[cpu_idx].debug = eenv_debug_percpu_debug_env_ptr(debug, cpu_idx);
#endif
}
/* given a pointer to the per-cpu global copy of _eenv_debug, get
 * a pointer to the specified _eenv_cpu debug env.
 */
static inline struct _eenv_debug *eenv_debug_percpu_debug_env_ptr(struct _eenv_debug *base, int cpu_idx)
{
	char *ptr = (char *)base;
	ptr += (cpu_idx * eenv_debug_size_per_cpu_entry());
	return (struct _eenv_debug *)ptr;
}
static inline int eenv_debug_size_per_cpu_entry(void)
{
	/* each cpu struct has an array of _eenv_debug structs
	 * which have an array of unsigned longs at the end -
	 * the allocation should be extended so that there are
	 * at least 'num_possible_cpus' entries in the array.
	 */
	return EAS_EENV_DEBUG_LEVELS * eenv_debug_size_per_dbg_entry();
}

static inline int eenv_debug_size_per_dbg_entry(void)
{
	return sizeof(struct _eenv_debug) + (sizeof(unsigned long) * num_possible_cpus());
}

get_eenv

eenv->util_delta = task_util_est(p);
eenv->util_delta_boosted = boosted_task_util(p);

获取估算的使用率和boost的使用率

/*
 * get_eenv - reset the eenv struct cached for this CPU
 *
 * When the eenv is returned, it is configured to do
 * energy calculations for the maximum number of CPUs
 * the task can be placed on. The prev_cpu entry is
 * filled in here. Callers are responsible for adding
 * other CPU candidates up to eenv->max_cpu_count.
 */
/*
当eenv返回时，
  它被配置为针对可以放置任务的最大CPU数量进行能量计算。
prev_cpu条目在此处填写。 调用者负责将其他CPU候选者添加到eenv-> max_cpu_count。
*/
static inline struct energy_env *get_eenv(struct task_struct *p, int prev_cpu)
{
	struct energy_env *eenv;
	cpumask_t cpumask_possible_cpus;
	int cpu = smp_processor_id();
	int i;

	eenv = &(per_cpu(eenv_cache, cpu));
	reset_eenv(eenv);

	/* populate eenv */
	eenv->p = p;
	/* use boosted task util for capacity selection
	 * during energy calculation, but unboosted task
	 * util for group utilization calculations
	 */
	eenv->util_delta = task_util_est(p);
	eenv->util_delta_boosted = boosted_task_util(p);

	cpumask_and(&cpumask_possible_cpus, &p->cpus_allowed, cpu_online_mask);
	eenv->max_cpu_count = cpumask_weight(&cpumask_possible_cpus);

	for (i=0; i < eenv->max_cpu_count; i++)
		eenv->cpu[i].cpu_id = -1;
	eenv->cpu[EAS_CPU_PRV].cpu_id = prev_cpu;
	eenv->next_idx = EAS_CPU_PRV;

	return eenv;
}
#define EAS_CPU_PRV	0

wake_wide

/*
 * Detect M:N waker/wakee relationships via a switching-frequency heuristic.
 *
 * A waker of many should wake a different task than the one last awakened
 * at a frequency roughly N times higher than one of its wakees.
 *
 * In order to determine whether we should let the load spread vs consolidating
 * to shared cache, we look for a minimum 'flip' frequency of llc_size in one
 * partner, and a factor of lls_size higher frequency in the other.
 *
 * With both conditions met, we can be relatively sure that the relationship is
 * non-monogamous, with partner count exceeding socket size.
 *
 * Waker/wakee being client/server, worker/dispatcher, interrupt source or
 * whatever is irrelevant, spread criteria is apparent partner count exceeds
 * socket size.
 */
 //wakee的数量
static int wake_wide(struct task_struct *p, int sibling_count_hint)
{
	unsigned int master = current->wakee_flips;
	unsigned int slave = p->wakee_flips;
	int llc_size = this_cpu_read(sd_llc_size);

	if (sibling_count_hint >= llc_size)
		return 1;

	if (master < slave)
		swap(master, slave);
	if (slave < llc_size || master < slave * llc_size)
		return 0;
	return 1;
}

wake_cap

/*
 * Disable WAKE_AFFINE in the case where task @p doesn't fit in the
 * capacity of either the waking CPU