[GIT PULL] higher scheduler load resolution

From: Ingo Molnar
Date: Mon May 23 2011 - 15:24:44 EST

Linus,

Please pull the latest sched-core-for-linus git tree from:

git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip.git sched-core-for-linus

This is something that came a bit late so i kept separate and under observation
- but i think it's good and ready for v2.6.40. This change allows better
balancing for larger amounts of cgroups (on 64-bit systems) - we were running
out of resolution.

Thanks,

Ingo

------------------>
Nikhil Rao (3):
sched: Cleanup set_load_weight()
sched: Introduce SCHED_POWER_SCALE to scale cpu_power calculations
sched: Increase SCHED_LOAD_SCALE resolution


include/linux/sched.h | 34 ++++++++++++++++++++++++++-----
kernel/sched.c | 39 +++++++++++++++++++++++++-----------
kernel/sched_fair.c | 52 +++++++++++++++++++++++++-----------------------
3 files changed, 82 insertions(+), 43 deletions(-)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 12211e1..c34a718 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -788,17 +788,39 @@ enum cpu_idle_type {
};

/*
- * sched-domains (multiprocessor balancing) declarations:
+ * Increase resolution of nice-level calculations for 64-bit architectures.
+ * The extra resolution improves shares distribution and load balancing of
+ * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
+ * hierarchies, especially on larger systems. This is not a user-visible change
+ * and does not change the user-interface for setting shares/weights.
+ *
+ * We increase resolution only if we have enough bits to allow this increased
+ * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution
+ * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the
+ * increased costs.
*/
+#if BITS_PER_LONG > 32
+# define SCHED_LOAD_RESOLUTION 10
+# define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION)
+# define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION)
+#else
+# define SCHED_LOAD_RESOLUTION 0
+# define scale_load(w) (w)
+# define scale_load_down(w) (w)
+#endif

-/*
- * Increase resolution of nice-level calculations:
- */
-#define SCHED_LOAD_SHIFT 10
+#define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION)
#define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT)

-#define SCHED_LOAD_SCALE_FUZZ SCHED_LOAD_SCALE
+/*
+ * Increase resolution of cpu_power calculations
+ */
+#define SCHED_POWER_SHIFT 10
+#define SCHED_POWER_SCALE (1L << SCHED_POWER_SHIFT)

+/*
+ * sched-domains (multiprocessor balancing) declarations:
+ */
#ifdef CONFIG_SMP
#define SD_LOAD_BALANCE 0x0001 /* Do load balancing on this domain. */
#define SD_BALANCE_NEWIDLE 0x0002 /* Balance when about to become idle */
diff --git a/kernel/sched.c b/kernel/sched.c
index c62acf4..bb504e1 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -293,7 +293,7 @@ static DEFINE_SPINLOCK(task_group_lock);
* limitation from this.)
*/
#define MIN_SHARES 2
-#define MAX_SHARES (1UL << 18)
+#define MAX_SHARES (1UL << (18 + SCHED_LOAD_RESOLUTION))

static int root_task_group_load = ROOT_TASK_GROUP_LOAD;
#endif
@@ -1330,13 +1330,25 @@ calc_delta_mine(unsigned long delta_exec, unsigned long weight,
{
u64 tmp;

- tmp = (u64)delta_exec * weight;
+ /*
+ * weight can be less than 2^SCHED_LOAD_RESOLUTION for task group sched
+ * entities since MIN_SHARES = 2. Treat weight as 1 if less than
+ * 2^SCHED_LOAD_RESOLUTION.
+ */
+ if (likely(weight > (1UL << SCHED_LOAD_RESOLUTION)))
+ tmp = (u64)delta_exec * scale_load_down(weight);
+ else
+ tmp = (u64)delta_exec;

if (!lw->inv_weight) {
- if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST))
+ unsigned long w = scale_load_down(lw->weight);
+
+ if (BITS_PER_LONG > 32 && unlikely(w >= WMULT_CONST))
lw->inv_weight = 1;
+ else if (unlikely(!w))
+ lw->inv_weight = WMULT_CONST;
else
- lw->inv_weight = WMULT_CONST / lw->weight;
+ lw->inv_weight = WMULT_CONST / w;
}

/*
@@ -1778,17 +1790,20 @@ static void dec_nr_running(struct rq *rq)

static void set_load_weight(struct task_struct *p)
{
+ int prio = p->static_prio - MAX_RT_PRIO;
+ struct load_weight *load = &p->se.load;
+
/*
* SCHED_IDLE tasks get minimal weight:
*/
if (p->policy == SCHED_IDLE) {
- p->se.load.weight = WEIGHT_IDLEPRIO;
- p->se.load.inv_weight = WMULT_IDLEPRIO;
+ load->weight = scale_load(WEIGHT_IDLEPRIO);
+ load->inv_weight = WMULT_IDLEPRIO;
return;
}

- p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO];
- p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO];
+ load->weight = scale_load(prio_to_weight[prio]);
+ load->inv_weight = prio_to_wmult[prio];
}

static void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
@@ -6527,7 +6542,7 @@ 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) {
+ if (group->cpu_power != SCHED_POWER_SCALE) {
printk(KERN_CONT " (cpu_power = %d)",
group->cpu_power);
}
@@ -7902,7 +7917,7 @@ void __init sched_init(void)
#ifdef CONFIG_SMP
rq->sd = NULL;
rq->rd = NULL;
- rq->cpu_power = SCHED_LOAD_SCALE;
+ rq->cpu_power = SCHED_POWER_SCALE;
rq->post_schedule = 0;
rq->active_balance = 0;
rq->next_balance = jiffies;
@@ -8806,14 +8821,14 @@ cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp,
static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype,
u64 shareval)
{
- return sched_group_set_shares(cgroup_tg(cgrp), shareval);
+ return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval));
}

static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft)
{
struct task_group *tg = cgroup_tg(cgrp);

- return (u64) tg->shares;
+ return (u64) scale_load_down(tg->shares);
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index 37f2262..e32a9b7 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -1584,7 +1584,7 @@ find_idlest_group(struct sched_domain *sd, struct task_struct *p,
}

/* Adjust by relative CPU power of the group */
- avg_load = (avg_load * SCHED_LOAD_SCALE) / group->cpu_power;
+ avg_load = (avg_load * SCHED_POWER_SCALE) / group->cpu_power;

if (local_group) {
this_load = avg_load;
@@ -1722,7 +1722,7 @@ select_task_rq_fair(struct task_struct *p, int sd_flag, int wake_flags)
nr_running += cpu_rq(i)->cfs.nr_running;
}

- capacity = DIV_ROUND_CLOSEST(power, SCHED_LOAD_SCALE);
+ capacity = DIV_ROUND_CLOSEST(power, SCHED_POWER_SCALE);

if (tmp->flags & SD_POWERSAVINGS_BALANCE)
nr_running /= 2;
@@ -2570,7 +2570,7 @@ static inline int check_power_save_busiest_group(struct sd_lb_stats *sds,

unsigned long default_scale_freq_power(struct sched_domain *sd, int cpu)
{
- return SCHED_LOAD_SCALE;
+ return SCHED_POWER_SCALE;
}

unsigned long __weak arch_scale_freq_power(struct sched_domain *sd, int cpu)
@@ -2607,10 +2607,10 @@ unsigned long scale_rt_power(int cpu)
available = total - rq->rt_avg;
}

- if (unlikely((s64)total < SCHED_LOAD_SCALE))
- total = SCHED_LOAD_SCALE;
+ if (unlikely((s64)total < SCHED_POWER_SCALE))
+ total = SCHED_POWER_SCALE;

- total >>= SCHED_LOAD_SHIFT;
+ total >>= SCHED_POWER_SHIFT;

return div_u64(available, total);
}
@@ -2618,7 +2618,7 @@ unsigned long scale_rt_power(int cpu)
static void update_cpu_power(struct sched_domain *sd, int cpu)
{
unsigned long weight = sd->span_weight;
- unsigned long power = SCHED_LOAD_SCALE;
+ unsigned long power = SCHED_POWER_SCALE;
struct sched_group *sdg = sd->groups;

if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) {
@@ -2627,7 +2627,7 @@ static void update_cpu_power(struct sched_domain *sd, int cpu)
else
power *= default_scale_smt_power(sd, cpu);

- power >>= SCHED_LOAD_SHIFT;
+ power >>= SCHED_POWER_SHIFT;
}

sdg->cpu_power_orig = power;
@@ -2637,10 +2637,10 @@ static void update_cpu_power(struct sched_domain *sd, int cpu)
else
power *= default_scale_freq_power(sd, cpu);

- power >>= SCHED_LOAD_SHIFT;
+ power >>= SCHED_POWER_SHIFT;

power *= scale_rt_power(cpu);
- power >>= SCHED_LOAD_SHIFT;
+ power >>= SCHED_POWER_SHIFT;

if (!power)
power = 1;
@@ -2682,7 +2682,7 @@ static inline int
fix_small_capacity(struct sched_domain *sd, struct sched_group *group)
{
/*
- * Only siblings can have significantly less than SCHED_LOAD_SCALE
+ * Only siblings can have significantly less than SCHED_POWER_SCALE
*/
if (!(sd->flags & SD_SHARE_CPUPOWER))
return 0;
@@ -2770,7 +2770,7 @@ static inline void update_sg_lb_stats(struct sched_domain *sd,
}

/* Adjust by relative CPU power of the group */
- sgs->avg_load = (sgs->group_load * SCHED_LOAD_SCALE) / group->cpu_power;
+ sgs->avg_load = (sgs->group_load*SCHED_POWER_SCALE) / group->cpu_power;

/*
* Consider the group unbalanced when the imbalance is larger
@@ -2787,7 +2787,8 @@ static inline void update_sg_lb_stats(struct sched_domain *sd,
if ((max_cpu_load - min_cpu_load) >= avg_load_per_task && max_nr_running > 1)
sgs->group_imb = 1;

- sgs->group_capacity = DIV_ROUND_CLOSEST(group->cpu_power, SCHED_LOAD_SCALE);
+ sgs->group_capacity = DIV_ROUND_CLOSEST(group->cpu_power,
+ SCHED_POWER_SCALE);
if (!sgs->group_capacity)
sgs->group_capacity = fix_small_capacity(sd, group);
sgs->group_weight = group->group_weight;
@@ -2961,7 +2962,7 @@ static int check_asym_packing(struct sched_domain *sd,
return 0;

*imbalance = DIV_ROUND_CLOSEST(sds->max_load * sds->busiest->cpu_power,
- SCHED_LOAD_SCALE);
+ SCHED_POWER_SCALE);
return 1;
}

@@ -2990,7 +2991,7 @@ static inline void fix_small_imbalance(struct sd_lb_stats *sds,
cpu_avg_load_per_task(this_cpu);

scaled_busy_load_per_task = sds->busiest_load_per_task
- * SCHED_LOAD_SCALE;
+ * SCHED_POWER_SCALE;
scaled_busy_load_per_task /= sds->busiest->cpu_power;

if (sds->max_load - sds->this_load + scaled_busy_load_per_task >=
@@ -3009,10 +3010,10 @@ static inline void fix_small_imbalance(struct sd_lb_stats *sds,
min(sds->busiest_load_per_task, sds->max_load);
pwr_now += sds->this->cpu_power *
min(sds->this_load_per_task, sds->this_load);
- pwr_now /= SCHED_LOAD_SCALE;
+ pwr_now /= SCHED_POWER_SCALE;

/* Amount of load we'd subtract */
- tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) /
+ tmp = (sds->busiest_load_per_task * SCHED_POWER_SCALE) /
sds->busiest->cpu_power;
if (sds->max_load > tmp)
pwr_move += sds->busiest->cpu_power *
@@ -3020,15 +3021,15 @@ static inline void fix_small_imbalance(struct sd_lb_stats *sds,

/* Amount of load we'd add */
if (sds->max_load * sds->busiest->cpu_power <
- sds->busiest_load_per_task * SCHED_LOAD_SCALE)
+ sds->busiest_load_per_task * SCHED_POWER_SCALE)
tmp = (sds->max_load * sds->busiest->cpu_power) /
sds->this->cpu_power;
else
- tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) /
+ tmp = (sds->busiest_load_per_task * SCHED_POWER_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;
+ pwr_move /= SCHED_POWER_SCALE;

/* Move if we gain throughput */
if (pwr_move > pwr_now)
@@ -3070,7 +3071,7 @@ static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu,
load_above_capacity = (sds->busiest_nr_running -
sds->busiest_group_capacity);

- load_above_capacity *= (SCHED_LOAD_SCALE * SCHED_LOAD_SCALE);
+ load_above_capacity *= (SCHED_LOAD_SCALE * SCHED_POWER_SCALE);

load_above_capacity /= sds->busiest->cpu_power;
}
@@ -3090,7 +3091,7 @@ static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu,
/* 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)
- / SCHED_LOAD_SCALE;
+ / SCHED_POWER_SCALE;

/*
* if *imbalance is less than the average load per runnable task
@@ -3159,7 +3160,7 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
if (!sds.busiest || sds.busiest_nr_running == 0)
goto out_balanced;

- sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr;
+ sds.avg_load = (SCHED_POWER_SCALE * sds.total_load) / sds.total_pwr;

/*
* If the busiest group is imbalanced the below checks don't
@@ -3238,7 +3239,8 @@ find_busiest_queue(struct sched_domain *sd, struct sched_group *group,

for_each_cpu(i, sched_group_cpus(group)) {
unsigned long power = power_of(i);
- unsigned long capacity = DIV_ROUND_CLOSEST(power, SCHED_LOAD_SCALE);
+ unsigned long capacity = DIV_ROUND_CLOSEST(power,
+ SCHED_POWER_SCALE);
unsigned long wl;

if (!capacity)
@@ -3263,7 +3265,7 @@ find_busiest_queue(struct sched_domain *sd, struct sched_group *group,
* the load can be moved away from the cpu that is potentially
* running at a lower capacity.
*/
- wl = (wl * SCHED_LOAD_SCALE) / power;
+ wl = (wl * SCHED_POWER_SCALE) / power;

if (wl > max_load) {
max_load = wl;
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/