sched: Remove reciprocal for cpu_power
authorPeter Zijlstra <a.p.zijlstra@chello.nl>
Tue, 1 Sep 2009 08:34:39 +0000 (10:34 +0200)
committerIngo Molnar <mingo@elte.hu>
Fri, 4 Sep 2009 08:09:56 +0000 (10:09 +0200)
Its a source of fail, also, now that cpu_power is dynamical,
its a waste of time.

before:
<idle>-0   [000]   132.877936: find_busiest_group: avg_load: 0 group_load: 8241 power: 1

after:
bash-1689  [001]   137.862151: find_busiest_group: avg_load: 10636288 group_load: 10387 power: 1

[ v2: build fix from From: Andreas Herrmann ]

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Tested-by: Andreas Herrmann <andreas.herrmann3@amd.com>
Acked-by: Andreas Herrmann <andreas.herrmann3@amd.com>
Acked-by: Gautham R Shenoy <ego@in.ibm.com>
Cc: Balbir Singh <balbir@in.ibm.com>
LKML-Reference: <20090901083826.425896304@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
include/linux/sched.h
kernel/sched.c

index c67ddf309c84732d69fc413ab791507769a9a85e..3b7f43e3b7368909b15de8b7a77c8181742a8fb7 100644 (file)
@@ -860,15 +860,9 @@ struct sched_group {
 
        /*
         * CPU power of this group, SCHED_LOAD_SCALE being max power for a
-        * single CPU. This is read only (except for setup, hotplug CPU).
-        * Note : Never change cpu_power without recompute its reciprocal
+        * single CPU.
         */
-       unsigned int __cpu_power;
-       /*
-        * reciprocal value of cpu_power to avoid expensive divides
-        * (see include/linux/reciprocal_div.h)
-        */
-       u32 reciprocal_cpu_power;
+       unsigned int cpu_power;
 
        /*
         * The CPUs this group covers.
index e1ebf9b00f5d4af776c131f5101d5b04c1e445b5..b537853468503cbf7bce77715665c7ce54fc6622 100644 (file)
@@ -64,7 +64,6 @@
 #include <linux/tsacct_kern.h>
 #include <linux/kprobes.h>
 #include <linux/delayacct.h>
-#include <linux/reciprocal_div.h>
 #include <linux/unistd.h>
 #include <linux/pagemap.h>
 #include <linux/hrtimer.h>
  */
 #define RUNTIME_INF    ((u64)~0ULL)
 
-#ifdef CONFIG_SMP
-
 static void double_rq_lock(struct rq *rq1, struct rq *rq2);
 
-/*
- * Divide a load by a sched group cpu_power : (load / sg->__cpu_power)
- * Since cpu_power is a 'constant', we can use a reciprocal divide.
- */
-static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load)
-{
-       return reciprocal_divide(load, sg->reciprocal_cpu_power);
-}
-
-/*
- * Each time a sched group cpu_power is changed,
- * we must compute its reciprocal value
- */
-static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val)
-{
-       sg->__cpu_power += val;
-       sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power);
-}
-#endif
-
 static inline int rt_policy(int policy)
 {
        if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR))
@@ -2335,8 +2312,7 @@ find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu)
                }
 
                /* Adjust by relative CPU power of the group */
-               avg_load = sg_div_cpu_power(group,
-                               avg_load * SCHED_LOAD_SCALE);
+               avg_load = (avg_load * SCHED_LOAD_SCALE) / group->cpu_power;
 
                if (local_group) {
                        this_load = avg_load;
@@ -3768,7 +3744,6 @@ static void update_cpu_power(struct sched_domain *sd, int cpu)
        unsigned long weight = cpumask_weight(sched_domain_span(sd));
        unsigned long power = SCHED_LOAD_SCALE;
        struct sched_group *sdg = sd->groups;
-       unsigned long old = sdg->__cpu_power;
 
        /* here we could scale based on cpufreq */
 
@@ -3783,33 +3758,26 @@ static void update_cpu_power(struct sched_domain *sd, int cpu)
        if (!power)
                power = 1;
 
-       if (power != old) {
-               sdg->__cpu_power = power;
-               sdg->reciprocal_cpu_power = reciprocal_value(power);
-       }
+       sdg->cpu_power = power;
 }
 
 static void update_group_power(struct sched_domain *sd, int cpu)
 {
        struct sched_domain *child = sd->child;
        struct sched_group *group, *sdg = sd->groups;
-       unsigned long power = sdg->__cpu_power;
 
        if (!child) {
                update_cpu_power(sd, cpu);
                return;
        }
 
-       sdg->__cpu_power = 0;
+       sdg->cpu_power = 0;
 
        group = child->groups;
        do {
-               sdg->__cpu_power += group->__cpu_power;
+               sdg->cpu_power += group->cpu_power;
                group = group->next;
        } while (group != child->groups);
-
-       if (power != sdg->__cpu_power)
-               sdg->reciprocal_cpu_power = reciprocal_value(sdg->__cpu_power);
 }
 
 /**
@@ -3889,8 +3857,7 @@ static inline void update_sg_lb_stats(struct sched_domain *sd,
        }
 
        /* Adjust by relative CPU power of the group */
-       sgs->avg_load = sg_div_cpu_power(group,
-                       sgs->group_load * SCHED_LOAD_SCALE);
+       sgs->avg_load = (sgs->group_load * SCHED_LOAD_SCALE) / group->cpu_power;
 
 
        /*
@@ -3902,14 +3869,14 @@ static inline void update_sg_lb_stats(struct sched_domain *sd,
         *      normalized nr_running number somewhere that negates
         *      the hierarchy?
         */
-       avg_load_per_task = sg_div_cpu_power(group,
-                       sum_avg_load_per_task * SCHED_LOAD_SCALE);
+       avg_load_per_task = (sum_avg_load_per_task * SCHED_LOAD_SCALE) /
+               group->cpu_power;
 
        if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task)
                sgs->group_imb = 1;
 
        sgs->group_capacity =
-               DIV_ROUND_CLOSEST(group->__cpu_power, SCHED_LOAD_SCALE);
+               DIV_ROUND_CLOSEST(group->cpu_power, SCHED_LOAD_SCALE);
 }
 
 /**
@@ -3951,7 +3918,7 @@ static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu,
                        return;
 
                sds->total_load += sgs.group_load;
-               sds->total_pwr += group->__cpu_power;
+               sds->total_pwr += group->cpu_power;
 
                /*
                 * In case the child domain prefers tasks go to siblings
@@ -4016,28 +3983,28 @@ static inline void fix_small_imbalance(struct sd_lb_stats *sds,
         * moving them.
         */
 
-       pwr_now += sds->busiest->__cpu_power *
+       pwr_now += sds->busiest->cpu_power *
                        min(sds->busiest_load_per_task, sds->max_load);
-       pwr_now += sds->this->__cpu_power *
+       pwr_now += sds->this->cpu_power *
                        min(sds->this_load_per_task, sds->this_load);
        pwr_now /= SCHED_LOAD_SCALE;
 
        /* Amount of load we'd subtract */
-       tmp = sg_div_cpu_power(sds->busiest,
-                       sds->busiest_load_per_task * SCHED_LOAD_SCALE);
+       tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) /
+               sds->busiest->cpu_power;
        if (sds->max_load > tmp)
-               pwr_move += sds->busiest->__cpu_power *
+               pwr_move += sds->busiest->cpu_power *
                        min(sds->busiest_load_per_task, sds->max_load - tmp);
 
        /* Amount of load we'd add */
-       if (sds->max_load * sds->busiest->__cpu_power <
+       if (sds->max_load * sds->busiest->cpu_power <
                sds->busiest_load_per_task * SCHED_LOAD_SCALE)
-               tmp = sg_div_cpu_power(sds->this,
-                       sds->max_load * sds->busiest->__cpu_power);
+               tmp = (sds->max_load * sds->busiest->cpu_power) /
+                       sds->this->cpu_power;
        else
-               tmp = sg_div_cpu_power(sds->this,
-                       sds->busiest_load_per_task * SCHED_LOAD_SCALE);
-       pwr_move += sds->this->__cpu_power *
+               tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) /
+                       sds->this->cpu_power;
+       pwr_move += sds->this->cpu_power *
                        min(sds->this_load_per_task, sds->this_load + tmp);
        pwr_move /= SCHED_LOAD_SCALE;
 
@@ -4072,8 +4039,8 @@ static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu,
                        sds->max_load - sds->busiest_load_per_task);
 
        /* How much load to actually move to equalise the imbalance */
-       *imbalance = min(max_pull * sds->busiest->__cpu_power,
-               (sds->avg_load - sds->this_load) * sds->this->__cpu_power)
+       *imbalance = min(max_pull * sds->busiest->cpu_power,
+               (sds->avg_load - sds->this_load) * sds->this->cpu_power)
                        / SCHED_LOAD_SCALE;
 
        /*
@@ -4208,7 +4175,7 @@ static unsigned long power_of(int cpu)
        if (!group)
                return SCHED_LOAD_SCALE;
 
-       return group->__cpu_power;
+       return group->cpu_power;
 }
 
 /*
@@ -7922,7 +7889,7 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
                        break;
                }
 
-               if (!group->__cpu_power) {
+               if (!group->cpu_power) {
                        printk(KERN_CONT "\n");
                        printk(KERN_ERR "ERROR: domain->cpu_power not "
                                        "set\n");
@@ -7946,9 +7913,9 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
                cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
 
                printk(KERN_CONT " %s", str);
-               if (group->__cpu_power != SCHED_LOAD_SCALE) {
-                       printk(KERN_CONT " (__cpu_power = %d)",
-                               group->__cpu_power);
+               if (group->cpu_power != SCHED_LOAD_SCALE) {
+                       printk(KERN_CONT " (cpu_power = %d)",
+                               group->cpu_power);
                }
 
                group = group->next;
@@ -8233,7 +8200,7 @@ init_sched_build_groups(const struct cpumask *span,
                        continue;
 
                cpumask_clear(sched_group_cpus(sg));
-               sg->__cpu_power = 0;
+               sg->cpu_power = 0;
 
                for_each_cpu(j, span) {
                        if (group_fn(j, cpu_map, NULL, tmpmask) != group)
@@ -8491,7 +8458,7 @@ static void init_numa_sched_groups_power(struct sched_group *group_head)
                                continue;
                        }
 
-                       sg_inc_cpu_power(sg, sd->groups->__cpu_power);
+                       sg->cpu_power += sd->groups->cpu_power;
                }
                sg = sg->next;
        } while (sg != group_head);
@@ -8528,7 +8495,7 @@ static int build_numa_sched_groups(struct s_data *d,
                sd->groups = sg;
        }
 
-       sg->__cpu_power = 0;
+       sg->cpu_power = 0;
        cpumask_copy(sched_group_cpus(sg), d->nodemask);
        sg->next = sg;
        cpumask_or(d->covered, d->covered, d->nodemask);
@@ -8551,7 +8518,7 @@ static int build_numa_sched_groups(struct s_data *d,
                               "Can not alloc domain group for node %d\n", j);
                        return -ENOMEM;
                }
-               sg->__cpu_power = 0;
+               sg->cpu_power = 0;
                cpumask_copy(sched_group_cpus(sg), d->tmpmask);
                sg->next = prev->next;
                cpumask_or(d->covered, d->covered, d->tmpmask);
@@ -8629,7 +8596,7 @@ static void init_sched_groups_power(int cpu, struct sched_domain *sd)
 
        child = sd->child;
 
-       sd->groups->__cpu_power = 0;
+       sd->groups->cpu_power = 0;
 
        if (!child) {
                power = SCHED_LOAD_SCALE;
@@ -8645,7 +8612,7 @@ static void init_sched_groups_power(int cpu, struct sched_domain *sd)
                        power /= weight;
                        power >>= SCHED_LOAD_SHIFT;
                }
-               sg_inc_cpu_power(sd->groups, power);
+               sd->groups->cpu_power += power;
                return;
        }
 
@@ -8654,7 +8621,7 @@ static void init_sched_groups_power(int cpu, struct sched_domain *sd)
         */
        group = child->groups;
        do {
-               sg_inc_cpu_power(sd->groups, group->__cpu_power);
+               sd->groups->cpu_power += group->cpu_power;
                group = group->next;
        } while (group != child->groups);
 }