From: Nikhil Rao <ncrao@google.com>
To: Ingo Molnar <mingo@elte.hu>,
Peter Zijlstra <peterz@infradead.org>,
Mike Galbraith <efault@gmx.de>
Cc: linux-kernel@vger.kernel.org,
"Nikunj A. Dadhania" <nikunj@linux.vnet.ibm.com>,
Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>,
Stephan Barwolf <stephan.baerwolf@tu-ilmenau.de>,
Nikhil Rao <ncrao@google.com>
Subject: [PATCH v1 01/19] sched: introduce SCHED_POWER_SCALE to scale cpu_power calculations
Date: Sun, 1 May 2011 18:18:59 -0700 [thread overview]
Message-ID: <1304299157-25769-2-git-send-email-ncrao@google.com> (raw)
In-Reply-To: <1304299157-25769-1-git-send-email-ncrao@google.com>
SCHED_LOAD_SCALE is used to increase nice resolution and to scale cpu_power
calculations in the scheduler. This patch introduces SCHED_POWER_SCALE and
converts all uses of SCHED_LOAD_SCALE for scaling cpu_power to use
SCHED_POWER_SCALE instead.
This is a preparatory patch for increasing the resolution of SCHED_LOAD_SCALE,
and there is no need to increase resolution for cpu_power calculations.
Signed-off-by: Nikhil Rao <ncrao@google.com>
---
include/linux/sched.h | 13 +++++++----
kernel/sched.c | 11 ++++-----
kernel/sched_fair.c | 52 +++++++++++++++++++++++++-----------------------
3 files changed, 40 insertions(+), 36 deletions(-)
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 18d63ce..8d1ff2b 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -792,17 +792,20 @@ enum cpu_idle_type {
};
/*
- * sched-domains (multiprocessor balancing) declarations:
- */
-
-/*
* Increase resolution of nice-level calculations:
*/
#define SCHED_LOAD_SHIFT 10
#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 312f8b9..f4b4679 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -1302,8 +1302,7 @@ static void sched_avg_update(struct rq *rq)
* delta *= weight / lw
*/
static unsigned long
-calc_delta_mine(unsigned long delta_exec, unsigned long weight,
- struct load_weight *lw)
+calc_delta_mine(unsigned long delta_exec, u64 weight, struct load_weight *lw)
{
u64 tmp;
@@ -6468,7 +6467,7 @@ 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);
}
@@ -7176,7 +7175,7 @@ static void init_sched_groups_power(int cpu, struct sched_domain *sd)
sd->groups->cpu_power = 0;
if (!child) {
- power = SCHED_LOAD_SCALE;
+ power = SCHED_POWER_SCALE;
weight = cpumask_weight(sched_domain_span(sd));
/*
* SMT siblings share the power of a single core.
@@ -7187,7 +7186,7 @@ static void init_sched_groups_power(int cpu, struct sched_domain *sd)
if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) {
power *= sd->smt_gain;
power /= weight;
- power >>= SCHED_LOAD_SHIFT;
+ power >>= SCHED_POWER_SHIFT;
}
sd->groups->cpu_power += power;
return;
@@ -8224,7 +8223,7 @@ 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;
diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index 6fa833a..1a9340c 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -1557,7 +1557,7 @@ 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;
@@ -1692,7 +1692,7 @@ select_task_rq_fair(struct rq *rq, struct task_struct *p, int sd_flag, int wake_
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;
@@ -2534,7 +2534,7 @@ 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)
@@ -2571,10 +2571,10 @@ 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);
}
@@ -2582,7 +2582,7 @@ 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) {
@@ -2591,7 +2591,7 @@ 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;
@@ -2601,10 +2601,10 @@ 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;
@@ -2646,7 +2646,7 @@ 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->level != SD_LV_SIBLING)
return 0;
@@ -2734,7 +2734,7 @@ 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
@@ -2751,7 +2751,8 @@ 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;
@@ -2925,7 +2926,7 @@ 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;
}
@@ -2954,7 +2955,7 @@ 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 >=
@@ -2973,10 +2974,10 @@ 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 *
@@ -2984,15 +2985,15 @@ 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)
@@ -3034,7 +3035,7 @@ 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;
}
@@ -3054,7 +3055,7 @@ 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
@@ -3123,7 +3124,7 @@ 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
@@ -3202,7 +3203,8 @@ 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)
@@ -3227,7 +3229,7 @@ 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;
--
1.7.3.1
next prev parent reply other threads:[~2011-05-02 1:23 UTC|newest]
Thread overview: 34+ messages / expand[flat|nested] mbox.gz Atom feed top
2011-05-02 1:18 [PATCH v1 00/19] Increase resolution of load weights Nikhil Rao
2011-05-02 1:18 ` Nikhil Rao [this message]
2011-05-02 1:19 ` [PATCH v1 02/19] sched: increase SCHED_LOAD_SCALE resolution Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 03/19] sched: use u64 for load_weight fields Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 04/19] sched: update cpu_load to be u64 Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 05/19] sched: update this_cpu_load() to return u64 value Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 06/19] sched: update source_load(), target_load() and weighted_cpuload() to use u64 Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 07/19] sched: update find_idlest_cpu() to use u64 for load Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 08/19] sched: update find_idlest_group() to use u64 Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 09/19] sched: update division in cpu_avg_load_per_task to use div_u64 Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 10/19] sched: update wake_affine path to use u64, s64 for weights Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 11/19] sched: update update_sg_lb_stats() to use u64 Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 12/19] sched: Update update_sd_lb_stats() " Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 13/19] sched: update f_b_g() to use u64 for weights Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 14/19] sched: change type of imbalance to be u64 Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 15/19] sched: update h_load to use u64 Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 16/19] sched: update move_task() and helper functions to use u64 for weights Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 17/19] sched: update f_b_q() to use u64 for weighted cpuload Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 18/19] sched: update shares distribution to use u64 Nikhil Rao
2011-05-02 1:19 ` [PATCH v1 19/19] sched: convert atomic ops in shares update to use atomic64_t ops Nikhil Rao
2011-05-02 6:14 ` [PATCH v1 00/19] Increase resolution of load weights Ingo Molnar
2011-05-04 0:58 ` Nikhil Rao
2011-05-04 1:07 ` Nikhil Rao
2011-05-04 11:13 ` Ingo Molnar
2011-05-06 1:29 ` Nikhil Rao
2011-05-06 6:59 ` Ingo Molnar
2011-05-11 0:14 ` Nikhil Rao
2011-05-11 6:59 ` Ingo Molnar
2011-05-12 8:56 ` Nikhil Rao
2011-05-12 10:55 ` Ingo Molnar
2011-05-12 18:44 ` Nikhil Rao
2011-05-12 9:08 ` Peter Zijlstra
2011-05-12 17:30 ` Nikhil Rao
2011-05-13 7:19 ` Peter Zijlstra
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1304299157-25769-2-git-send-email-ncrao@google.com \
--to=ncrao@google.com \
--cc=efault@gmx.de \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@elte.hu \
--cc=nikunj@linux.vnet.ibm.com \
--cc=peterz@infradead.org \
--cc=stephan.baerwolf@tu-ilmenau.de \
--cc=vatsa@linux.vnet.ibm.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox