[PATCH] sched: mc/smt power savings sched policy

sysfs entries 'sched_mc_power_savings' and 'sched_smt_power_savings' in
/sys/devices/system/cpu/ control the MC/SMT power savings policy for the
scheduler.

Based on the values (1-enable, 0-disable) for these controls, sched groups
cpu power will be determined for different domains.  When power savings
policy is enabled and under light load conditions, scheduler will minimize
the physical packages/cpu cores carrying the load and thus conserving
power(with a perf impact based on the workload characteristics...  see OLS
2005 CMP kernel scheduler paper for more details..)

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Con Kolivas <kernel@kolivas.org>
Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

11 files changed, 262 insertions, 33 deletions

diff --git a/arch/i386/kernel/smpboot.c b/arch/i386/kernel/smpboot.c
index ab5275beddf785..89e7315e539c49 100644
--- a/arch/i386/kernel/smpboot.c
+++ b/arch/i386/kernel/smpboot.c
@@ -448,10 +448,12 @@ cpumask_t cpu_coregroup_map(int cpu)
 	struct cpuinfo_x86 *c = cpu_data + cpu;
 	/*
 	 * For perf, we return last level cache shared map.
-	 * TBD: when power saving sched policy is added, we will return
-	 *      cpu_core_map when power saving policy is enabled
+	 * And for power savings, we return cpu_core_map
 	 */
-	return c->llc_shared_map;
+	if (sched_mc_power_savings || sched_smt_power_savings)
+		return cpu_core_map[cpu];
+	else
+		return c->llc_shared_map;
 }
 
 /* representing cpus for which sibling maps can be computed */
diff --git a/arch/x86_64/kernel/smpboot.c b/arch/x86_64/kernel/smpboot.c
index 4e9755179ecf57..540c0ccbcccc89 100644
--- a/arch/x86_64/kernel/smpboot.c
+++ b/arch/x86_64/kernel/smpboot.c
@@ -455,10 +455,12 @@ cpumask_t cpu_coregroup_map(int cpu)
 	struct cpuinfo_x86 *c = cpu_data + cpu;
 	/*
 	 * For perf, we return last level cache shared map.
-	 * TBD: when power saving sched policy is added, we will return
-	 *      cpu_core_map when power saving policy is enabled
+	 * And for power savings, we return cpu_core_map
 	 */
-	return c->llc_shared_map;
+	if (sched_mc_power_savings || sched_smt_power_savings)
+		return cpu_core_map[cpu];
+	else
+		return c->llc_shared_map;
 }
 
 /* representing cpus for which sibling maps can be computed */
diff --git a/drivers/base/cpu.c b/drivers/base/cpu.c
index 3972d8ac978687..4bef76a2f3f2a7 100644
--- a/drivers/base/cpu.c
+++ b/drivers/base/cpu.c
@@ -143,5 +143,13 @@ EXPORT_SYMBOL_GPL(get_cpu_sysdev);
 
 int __init cpu_dev_init(void)
 {
-	return sysdev_class_register(&cpu_sysdev_class);
+	int err;
+
+	err = sysdev_class_register(&cpu_sysdev_class);
+#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
+	if (!err)
+		err = sched_create_sysfs_power_savings_entries(&cpu_sysdev_class);
+#endif
+
+	return err;
 }
diff --git a/include/asm-i386/topology.h b/include/asm-i386/topology.h
index aa4185ee81fb2b..6adbd9b1ae8811 100644
--- a/include/asm-i386/topology.h
+++ b/include/asm-i386/topology.h
@@ -112,4 +112,9 @@ extern unsigned long node_remap_size[];
 
 extern cpumask_t cpu_coregroup_map(int cpu);
 
+#ifdef CONFIG_SMP
+#define mc_capable()	(boot_cpu_data.x86_max_cores > 1)
+#define smt_capable()	(smp_num_siblings > 1)
+#endif
+
 #endif /* _ASM_I386_TOPOLOGY_H */
diff --git a/include/asm-ia64/topology.h b/include/asm-ia64/topology.h
index 616b5ed2aa7277..937c212575239f 100644
--- a/include/asm-ia64/topology.h
+++ b/include/asm-ia64/topology.h
@@ -112,6 +112,7 @@ void build_cpu_to_node_map(void);
 #define topology_core_id(cpu)			(cpu_data(cpu)->core_id)
 #define topology_core_siblings(cpu)		(cpu_core_map[cpu])
 #define topology_thread_siblings(cpu)		(cpu_sibling_map[cpu])
+#define smt_capable() 				(smp_num_siblings > 1)
 #endif
 
 #include <asm-generic/topology.h>
diff --git a/include/asm-powerpc/topology.h b/include/asm-powerpc/topology.h
index 92f3e5507d224c..bbc3844b086fca 100644
--- a/include/asm-powerpc/topology.h
+++ b/include/asm-powerpc/topology.h
@@ -93,5 +93,10 @@ static inline void sysfs_remove_device_from_node(struct sys_device *dev,
 
 #endif /* CONFIG_NUMA */
 
+#ifdef CONFIG_SMP
+#include <asm/cputable.h>
+#define smt_capable() 		(cpu_has_feature(CPU_FTR_SMT))
+#endif
+
 #endif /* __KERNEL__ */
 #endif	/* _ASM_POWERPC_TOPOLOGY_H */
diff --git a/include/asm-sparc64/topology.h b/include/asm-sparc64/topology.h
index 0e234e201bd6b5..98a6c613589dcb 100644
--- a/include/asm-sparc64/topology.h
+++ b/include/asm-sparc64/topology.h
@@ -1,6 +1,9 @@
 #ifndef _ASM_SPARC64_TOPOLOGY_H
 #define _ASM_SPARC64_TOPOLOGY_H
 
+#include <asm/spitfire.h>
+#define smt_capable()	(tlb_type == hypervisor)
+
 #include <asm-generic/topology.h>
 
 #endif /* _ASM_SPARC64_TOPOLOGY_H */
diff --git a/include/asm-x86_64/topology.h b/include/asm-x86_64/topology.h
index c4e46e7fa7ba78..6e7a2e976b04fc 100644
--- a/include/asm-x86_64/topology.h
+++ b/include/asm-x86_64/topology.h
@@ -59,6 +59,8 @@ extern int __node_distance(int, int);
 #define topology_core_id(cpu)			(cpu_data[cpu].cpu_core_id)
 #define topology_core_siblings(cpu)		(cpu_core_map[cpu])
 #define topology_thread_siblings(cpu)		(cpu_sibling_map[cpu])
+#define mc_capable()			(boot_cpu_data.x86_max_cores > 1)
+#define smt_capable() 			(smp_num_siblings > 1)
 #endif
 
 #include <asm-generic/topology.h>
diff --git a/include/linux/sched.h b/include/linux/sched.h
index ab8ffc54423a3e..0bc81a151e508e 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -570,6 +570,11 @@ enum idle_type
 #define SD_WAKE_AFFINE		32	/* Wake task to waking CPU */
 #define SD_WAKE_BALANCE		64	/* Perform balancing at task wakeup */
 #define SD_SHARE_CPUPOWER	128	/* Domain members share cpu power */
+#define SD_POWERSAVINGS_BALANCE	256	/* Balance for power savings */
+
+#define BALANCE_FOR_POWER	((sched_mc_power_savings || sched_smt_power_savings) \
+				 ? SD_POWERSAVINGS_BALANCE : 0)
+
 
 struct sched_group {
 	struct sched_group *next;	/* Must be a circular list */
@@ -1412,6 +1417,11 @@ static inline void arch_pick_mmap_layout(struct mm_struct *mm)
 extern long sched_setaffinity(pid_t pid, cpumask_t new_mask);
 extern long sched_getaffinity(pid_t pid, cpumask_t *mask);
 
+#include <linux/sysdev.h>
+extern int sched_mc_power_savings, sched_smt_power_savings;
+extern struct sysdev_attribute attr_sched_mc_power_savings, attr_sched_smt_power_savings;
+extern int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls);
+
 extern void normalize_rt_tasks(void);
 
 #ifdef CONFIG_PM
diff --git a/include/linux/topology.h b/include/linux/topology.h
index a305ae2e44b6dd..ec1eca85290ade 100644
--- a/include/linux/topology.h
+++ b/include/linux/topology.h
@@ -134,7 +134,8 @@
 	.flags			= SD_LOAD_BALANCE	\
 				| SD_BALANCE_NEWIDLE	\
 				| SD_BALANCE_EXEC	\
-				| SD_WAKE_AFFINE,	\
+				| SD_WAKE_AFFINE	\
+				| BALANCE_FOR_POWER,	\
 	.last_balance		= jiffies,		\
 	.balance_interval	= 1,			\
 	.nr_balance_failed	= 0,			\
diff --git a/kernel/sched.c b/kernel/sched.c
index 122b75584a138d..54fa282657cc09 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -1162,6 +1162,11 @@ static int sched_balance_self(int cpu, int flag)
 	struct sched_domain *tmp, *sd = NULL;
 
 	for_each_domain(cpu, tmp) {
+ 		/*
+ 	 	 * If power savings logic is enabled for a domain, stop there.
+ 	 	 */
+		if (tmp->flags & SD_POWERSAVINGS_BALANCE)
+			break;
 		if (tmp->flags & flag)
 			sd = tmp;
 	}
@@ -2082,6 +2087,12 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
 	unsigned long busiest_load_per_task, busiest_nr_running;
 	unsigned long this_load_per_task, this_nr_running;
 	int load_idx;
+#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
+	int power_savings_balance = 1;
+	unsigned long leader_nr_running = 0, min_load_per_task = 0;
+	unsigned long min_nr_running = ULONG_MAX;
+	struct sched_group *group_min = NULL, *group_leader = NULL;
+#endif
 
 	max_load = this_load = total_load = total_pwr = 0;
 	busiest_load_per_task = busiest_nr_running = 0;
@@ -2094,7 +2105,7 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
 		load_idx = sd->idle_idx;
 
 	do {
-		unsigned long load;
+		unsigned long load, group_capacity;
 		int local_group;
 		int i;
 		unsigned long sum_nr_running, sum_weighted_load;
@@ -2127,18 +2138,76 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
 		/* Adjust by relative CPU power of the group */
 		avg_load = (avg_load * SCHED_LOAD_SCALE) / group->cpu_power;
 
+		group_capacity = group->cpu_power / SCHED_LOAD_SCALE;
+
 		if (local_group) {
 			this_load = avg_load;
 			this = group;
 			this_nr_running = sum_nr_running;
 			this_load_per_task = sum_weighted_load;
 		} else if (avg_load > max_load &&
-			   sum_nr_running > group->cpu_power / SCHED_LOAD_SCALE) {
+			   sum_nr_running > group_capacity) {
 			max_load = avg_load;
 			busiest = group;
 			busiest_nr_running = sum_nr_running;
 			busiest_load_per_task = sum_weighted_load;
 		}
+
+#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
+		/*
+		 * Busy processors will not participate in power savings
+		 * balance.
+		 */
+ 		if (idle == NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE))
+ 			goto group_next;
+
+		/*
+		 * If the local group is idle or completely loaded
+		 * no need to do power savings balance at this domain
+		 */
+		if (local_group && (this_nr_running >= group_capacity ||
+				    !this_nr_running))
+			power_savings_balance = 0;
+
+ 		/*
+		 * If a group is already running at full capacity or idle,
+		 * don't include that group in power savings calculations
+ 		 */
+ 		if (!power_savings_balance || sum_nr_running >= group_capacity
+		    || !sum_nr_running)
+ 			goto group_next;
+
+ 		/*
+		 * Calculate the group which has the least non-idle load.
+ 		 * This is the group from where we need to pick up the load
+ 		 * for saving power
+ 		 */
+ 		if ((sum_nr_running < min_nr_running) ||
+ 		    (sum_nr_running == min_nr_running &&
+		     first_cpu(group->cpumask) <
+		     first_cpu(group_min->cpumask))) {
+ 			group_min = group;
+ 			min_nr_running = sum_nr_running;
+			min_load_per_task = sum_weighted_load /
+						sum_nr_running;
+ 		}
+
+ 		/*
+		 * Calculate the group which is almost near its
+ 		 * capacity but still has some space to pick up some load
+ 		 * from other group and save more power
+ 		 */
+ 		if (sum_nr_running <= group_capacity - 1)
+ 			if (sum_nr_running > leader_nr_running ||
+ 			    (sum_nr_running == leader_nr_running &&
+ 			     first_cpu(group->cpumask) >
+ 			      first_cpu(group_leader->cpumask))) {
+ 				group_leader = group;
+ 				leader_nr_running = sum_nr_running;
+ 			}
+
+group_next:
+#endif
 		group = group->next;
 	} while (group != sd->groups);
 
@@ -2247,7 +2316,16 @@ small_imbalance:
 	return busiest;
 
 out_balanced:
+#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
+	if (idle == NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE))
+		goto ret;
 
+	if (this == group_leader && group_leader != group_min) {
+		*imbalance = min_load_per_task;
+		return group_min;
+	}
+ret:
+#endif
 	*imbalance = 0;
 	return NULL;
 }
@@ -2300,7 +2378,8 @@ static int load_balance(int this_cpu, runqueue_t *this_rq,
 	int active_balance = 0;
 	int sd_idle = 0;
 
-	if (idle != NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER)
+	if (idle != NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER &&
+	    !sched_smt_power_savings)
 		sd_idle = 1;
 
 	schedstat_inc(sd, lb_cnt[idle]);
@@ -2389,7 +2468,8 @@ static int load_balance(int this_cpu, runqueue_t *this_rq,
 			sd->balance_interval *= 2;
 	}
 
-	if (!nr_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER)
+	if (!nr_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
+	    !sched_smt_power_savings)
 		return -1;
 	return nr_moved;
 
@@ -2404,7 +2484,7 @@ out_one_pinned:
 			(sd->balance_interval < sd->max_interval))
 		sd->balance_interval *= 2;
 
-	if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER)
+	if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && !sched_smt_power_savings)
 		return -1;
 	return 0;
 }
@@ -2425,7 +2505,7 @@ static int load_balance_newidle(int this_cpu, runqueue_t *this_rq,
 	int nr_moved = 0;
 	int sd_idle = 0;
 
-	if (sd->flags & SD_SHARE_CPUPOWER)
+	if (sd->flags & SD_SHARE_CPUPOWER && !sched_smt_power_savings)
 		sd_idle = 1;
 
 	schedstat_inc(sd, lb_cnt[NEWLY_IDLE]);
@@ -2466,7 +2546,7 @@ static int load_balance_newidle(int this_cpu, runqueue_t *this_rq,
 
 out_balanced:
 	schedstat_inc(sd, lb_balanced[NEWLY_IDLE]);
-	if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER)
+	if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && !sched_smt_power_savings)
 		return -1;
 	sd->nr_balance_failed = 0;
 	return 0;
@@ -5732,6 +5812,7 @@ static cpumask_t sched_domain_node_span(int node)
 }
 #endif
 
+int sched_smt_power_savings = 0, sched_mc_power_savings = 0;
 /*
  * At the moment, CONFIG_SCHED_SMT is never defined, but leave it in so we
  * can switch it on easily if needed.
@@ -6113,37 +6194,72 @@ static int build_sched_domains(const cpumask_t *cpu_map)
 #endif
 
 	/* Calculate CPU power for physical packages and nodes */
+#ifdef CONFIG_SCHED_SMT
 	for_each_cpu_mask(i, *cpu_map) {
-		int power;
 		struct sched_domain *sd;
-#ifdef CONFIG_SCHED_SMT
 		sd = &per_cpu(cpu_domains, i);
-		power = SCHED_LOAD_SCALE;
-		sd->groups->cpu_power = power;
+		sd->groups->cpu_power = SCHED_LOAD_SCALE;
+	}
 #endif
 #ifdef CONFIG_SCHED_MC
+	for_each_cpu_mask(i, *cpu_map) {
+		int power;
+		struct sched_domain *sd;
 		sd = &per_cpu(core_domains, i);
-		power = SCHED_LOAD_SCALE + (cpus_weight(sd->groups->cpumask)-1)
+		if (sched_smt_power_savings)
+			power = SCHED_LOAD_SCALE * cpus_weight(sd->groups->cpumask);
+		else
+			power = SCHED_LOAD_SCALE + (cpus_weight(sd->groups->cpumask)-1)
 					    * SCHED_LOAD_SCALE / 10;
 		sd->groups->cpu_power = power;
+	}
+#endif
 
+	for_each_cpu_mask(i, *cpu_map) {
+		struct sched_domain *sd;
+#ifdef CONFIG_SCHED_MC
 		sd = &per_cpu(phys_domains, i);
+		if (i != first_cpu(sd->groups->cpumask))
+			continue;
 
- 		/*
- 		 * This has to be < 2 * SCHED_LOAD_SCALE
- 		 * Lets keep it SCHED_LOAD_SCALE, so that
- 		 * while calculating NUMA group's cpu_power
- 		 * we can simply do
- 		 *  numa_group->cpu_power += phys_group->cpu_power;
- 		 *
- 		 * See "only add power once for each physical pkg"
- 		 * comment below
- 		 */
- 		sd->groups->cpu_power = SCHED_LOAD_SCALE;
+		sd->groups->cpu_power = 0;
+		if (sched_mc_power_savings || sched_smt_power_savings) {
+			int j;
+
+ 			for_each_cpu_mask(j, sd->groups->cpumask) {
+				struct sched_domain *sd1;
+ 				sd1 = &per_cpu(core_domains, j);
+ 				/*
+ 			 	 * for each core we will add once
+ 				 * to the group in physical domain
+ 			 	 */
+  	 			if (j != first_cpu(sd1->groups->cpumask))
+ 					continue;
+
+ 				if (sched_smt_power_savings)
+   					sd->groups->cpu_power += sd1->groups->cpu_power;
+ 				else
+   					sd->groups->cpu_power += SCHED_LOAD_SCALE;
+   			}
+ 		} else
+ 			/*
+ 			 * This has to be < 2 * SCHED_LOAD_SCALE
+ 			 * Lets keep it SCHED_LOAD_SCALE, so that
+ 			 * while calculating NUMA group's cpu_power
+ 			 * we can simply do
+ 			 *  numa_group->cpu_power += phys_group->cpu_power;
+ 			 *
+ 			 * See "only add power once for each physical pkg"
+ 			 * comment below
+ 			 */
+ 			sd->groups->cpu_power = SCHED_LOAD_SCALE;
 #else
+		int power;
 		sd = &per_cpu(phys_domains, i);
-		power = SCHED_LOAD_SCALE + SCHED_LOAD_SCALE *
-				(cpus_weight(sd->groups->cpumask)-1) / 10;
+		if (sched_smt_power_savings)
+			power = SCHED_LOAD_SCALE * cpus_weight(sd->groups->cpumask);
+		else
+			power = SCHED_LOAD_SCALE;
 		sd->groups->cpu_power = power;
 #endif
 	}
@@ -6244,6 +6360,80 @@ int partition_sched_domains(cpumask_t *partition1, cpumask_t *partition2)
 	return err;
 }
 
+#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
+int arch_reinit_sched_domains(void)
+{
+	int err;
+
+	lock_cpu_hotplug();
+	detach_destroy_domains(&cpu_online_map);
+	err = arch_init_sched_domains(&cpu_online_map);
+	unlock_cpu_hotplug();
+
+	return err;
+}
+
+static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt)
+{
+	int ret;
+
+	if (buf[0] != '0' && buf[0] != '1')
+		return -EINVAL;
+
+	if (smt)
+		sched_smt_power_savings = (buf[0] == '1');
+	else
+		sched_mc_power_savings = (buf[0] == '1');
+
+	ret = arch_reinit_sched_domains();
+
+	return ret ? ret : count;
+}
+
+int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls)
+{
+	int err = 0;
+#ifdef CONFIG_SCHED_SMT
+	if (smt_capable())
+		err = sysfs_create_file(&cls->kset.kobj,
+					&attr_sched_smt_power_savings.attr);
+#endif
+#ifdef CONFIG_SCHED_MC
+	if (!err && mc_capable())
+		err = sysfs_create_file(&cls->kset.kobj,
+					&attr_sched_mc_power_savings.attr);
+#endif
+	return err;
+}
+#endif
+
+#ifdef CONFIG_SCHED_MC
+static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page)
+{
+	return sprintf(page, "%u\n", sched_mc_power_savings);
+}
+static ssize_t sched_mc_power_savings_store(struct sys_device *dev, const char *buf, size_t count)
+{
+	return sched_power_savings_store(buf, count, 0);
+}
+SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show,
+	    sched_mc_power_savings_store);
+#endif
+
+#ifdef CONFIG_SCHED_SMT
+static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page)
+{
+	return sprintf(page, "%u\n", sched_smt_power_savings);
+}
+static ssize_t sched_smt_power_savings_store(struct sys_device *dev, const char *buf, size_t count)
+{
+	return sched_power_savings_store(buf, count, 1);
+}
+SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show,
+	    sched_smt_power_savings_store);
+#endif
+
+
 #ifdef CONFIG_HOTPLUG_CPU
 /*
  * Force a reinitialization of the sched domains hierarchy.  The domains