Merge branch 'sched-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip

* 'sched-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (46 commits)
  sched: Add comments to find_busiest_group() function
  sched: Refactor the power savings balance code
  sched: Optimize the !power_savings_balance during fbg()
  sched: Create a helper function to calculate imbalance
  sched: Create helper to calculate small_imbalance in fbg()
  sched: Create a helper function to calculate sched_domain stats for fbg()
  sched: Define structure to store the sched_domain statistics for fbg()
  sched: Create a helper function to calculate sched_group stats for fbg()
  sched: Define structure to store the sched_group statistics for fbg()
  sched: Fix indentations in find_busiest_group() using gotos
  sched: Simple helper functions for find_busiest_group()
  sched: remove unused fields from struct rq
  sched: jiffies not printed per CPU
  sched: small optimisation of can_migrate_task()
  sched: fix typos in documentation
  sched: add avg_overlap decay
  x86, sched_clock(): mark variables read-mostly
  sched: optimize ttwu vs group scheduling
  sched: TIF_NEED_RESCHED -> need_reshed() cleanup
  sched: don't rebalance if attached on NULL domain
  ...

20 files changed, 1262 insertions, 642 deletions

diff --git a/Documentation/scheduler/00-INDEX b/Documentation/scheduler/00-INDEX
index aabcc3a089b..3c00c9c3219 100644
--- a/Documentation/scheduler/00-INDEX
+++ b/Documentation/scheduler/00-INDEX
@@ -2,8 +2,6 @@
 	- this file.
 sched-arch.txt
 	- CPU Scheduler implementation hints for architecture specific code.
-sched-coding.txt
-	- reference for various scheduler-related methods in the O(1) scheduler.
 sched-design-CFS.txt
 	- goals, design and implementation of the Complete Fair Scheduler.
 sched-domains.txt
diff --git a/Documentation/scheduler/sched-coding.txt b/Documentation/scheduler/sched-coding.txt
deleted file mode 100644
index cbd8db752ac..00000000000
--- a/Documentation/scheduler/sched-coding.txt
+++ /dev/null
@@ -1,126 +0,0 @@
-     Reference for various scheduler-related methods in the O(1) scheduler
-		Robert Love <rml@tech9.net>, MontaVista Software
-
-
-Note most of these methods are local to kernel/sched.c - this is by design.
-The scheduler is meant to be self-contained and abstracted away.  This document
-is primarily for understanding the scheduler, not interfacing to it.  Some of
-the discussed interfaces, however, are general process/scheduling methods.
-They are typically defined in include/linux/sched.h.
-
-
-Main Scheduling Methods
------------------------
-
-void load_balance(runqueue_t *this_rq, int idle)
-	Attempts to pull tasks from one cpu to another to balance cpu usage,
-	if needed.  This method is called explicitly if the runqueues are
-	imbalanced or periodically by the timer tick.  Prior to calling,
-	the current runqueue must be locked and interrupts disabled.
-
-void schedule()
-	The main scheduling function.  Upon return, the highest priority
-	process will be active.
-
-
-Locking
--------
-
-Each runqueue has its own lock, rq->lock.  When multiple runqueues need
-to be locked, lock acquires must be ordered by ascending &runqueue value.
-
-A specific runqueue is locked via
-
-	task_rq_lock(task_t pid, unsigned long *flags)
-
-which disables preemption, disables interrupts, and locks the runqueue pid is
-running on.  Likewise,
-
-	task_rq_unlock(task_t pid, unsigned long *flags)
-
-unlocks the runqueue pid is running on, restores interrupts to their previous
-state, and reenables preemption.
-
-The routines
-
-	double_rq_lock(runqueue_t *rq1, runqueue_t *rq2)
-
-and
-
-	double_rq_unlock(runqueue_t *rq1, runqueue_t *rq2)
-
-safely lock and unlock, respectively, the two specified runqueues.  They do
-not, however, disable and restore interrupts.  Users are required to do so
-manually before and after calls.
-
-
-Values
-------
-
-MAX_PRIO
-	The maximum priority of the system, stored in the task as task->prio.
-	Lower priorities are higher.  Normal (non-RT) priorities range from
-	MAX_RT_PRIO to (MAX_PRIO - 1).
-MAX_RT_PRIO
-	The maximum real-time priority of the system.  Valid RT priorities
-	range from 0 to (MAX_RT_PRIO - 1).
-MAX_USER_RT_PRIO
-	The maximum real-time priority that is exported to user-space.  Should
-	always be equal to or less than MAX_RT_PRIO.  Setting it less allows
-	kernel threads to have higher priorities than any user-space task.
-MIN_TIMESLICE
-MAX_TIMESLICE
-	Respectively, the minimum and maximum timeslices (quanta) of a process.
-
-Data
-----
-
-struct runqueue
-	The main per-CPU runqueue data structure.
-struct task_struct
-	The main per-process data structure.
-
-
-General Methods
----------------
-
-cpu_rq(cpu)
-	Returns the runqueue of the specified cpu.
-this_rq()
-	Returns the runqueue of the current cpu.
-task_rq(pid)
-	Returns the runqueue which holds the specified pid.
-cpu_curr(cpu)
-	Returns the task currently running on the given cpu.
-rt_task(pid)
-	Returns true if pid is real-time, false if not.
-
-
-Process Control Methods
------------------------
-
-void set_user_nice(task_t *p, long nice)
-	Sets the "nice" value of task p to the given value.
-int setscheduler(pid_t pid, int policy, struct sched_param *param)
-	Sets the scheduling policy and parameters for the given pid.
-int set_cpus_allowed(task_t *p, unsigned long new_mask)
-	Sets a given task's CPU affinity and migrates it to a proper cpu.
-	Callers must have a valid reference to the task and assure the
-	task not exit prematurely.  No locks can be held during the call.
-set_task_state(tsk, state_value)
-	Sets the given task's state to the given value.
-set_current_state(state_value)
-	Sets the current task's state to the given value.
-void set_tsk_need_resched(struct task_struct *tsk)
-	Sets need_resched in the given task.
-void clear_tsk_need_resched(struct task_struct *tsk)
-	Clears need_resched in the given task.
-void set_need_resched()
-	Sets need_resched in the current task.
-void clear_need_resched()
-	Clears need_resched in the current task.
-int need_resched()
-	Returns true if need_resched is set in the current task, false
-	otherwise.
-yield()
-	Place the current process at the end of the runqueue and call schedule.
diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
index 24ff26a38ad..5fff00c70de 100644
--- a/arch/x86/kernel/cpu/intel.c
+++ b/arch/x86/kernel/cpu/intel.c
@@ -4,6 +4,7 @@
 #include <linux/string.h>
 #include <linux/bitops.h>
 #include <linux/smp.h>
+#include <linux/sched.h>
 #include <linux/thread_info.h>
 #include <linux/module.h>
 
@@ -56,11 +57,16 @@ static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
 
 	/*
 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
-	 * with P/T states and does not stop in deep C-states
+	 * with P/T states and does not stop in deep C-states.
+	 *
+	 * It is also reliable across cores and sockets. (but not across
+	 * cabinets - we turn it off in that case explicitly.)
 	 */
 	if (c->x86_power & (1 << 8)) {
 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
+		set_cpu_cap(c, X86_FEATURE_TSC_RELIABLE);
+		sched_clock_stable = 1;
 	}
 
 }
diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
index b8e7aaf7ef7..08afa1579e6 100644
--- a/arch/x86/kernel/tsc.c
+++ b/arch/x86/kernel/tsc.c
@@ -17,20 +17,21 @@
 #include <asm/delay.h>
 #include <asm/hypervisor.h>
 
-unsigned int cpu_khz;           /* TSC clocks / usec, not used here */
+unsigned int __read_mostly cpu_khz;	/* TSC clocks / usec, not used here */
 EXPORT_SYMBOL(cpu_khz);
-unsigned int tsc_khz;
+
+unsigned int __read_mostly tsc_khz;
 EXPORT_SYMBOL(tsc_khz);
 
 /*
  * TSC can be unstable due to cpufreq or due to unsynced TSCs
  */
-static int tsc_unstable;
+static int __read_mostly tsc_unstable;
 
 /* native_sched_clock() is called before tsc_init(), so
    we must start with the TSC soft disabled to prevent
    erroneous rdtsc usage on !cpu_has_tsc processors */
-static int tsc_disabled = -1;
+static int __read_mostly tsc_disabled = -1;
 
 static int tsc_clocksource_reliable;
 /*
diff --git a/include/linux/init_task.h b/include/linux/init_task.h
index e752d973fa2..af1de95e711 100644
--- a/include/linux/init_task.h
+++ b/include/linux/init_task.h
@@ -147,6 +147,7 @@ extern struct cred init_cred;
 		.nr_cpus_allowed = NR_CPUS,				\
 	},								\
 	.tasks		= LIST_HEAD_INIT(tsk.tasks),			\
+	.pushable_tasks = PLIST_NODE_INIT(tsk.pushable_tasks, MAX_PRIO), \
 	.ptraced	= LIST_HEAD_INIT(tsk.ptraced),			\
 	.ptrace_entry	= LIST_HEAD_INIT(tsk.ptrace_entry),		\
 	.real_parent	= &tsk,						\
diff --git a/include/linux/latencytop.h b/include/linux/latencytop.h
index 901c2d6377a..b0e99898527 100644
--- a/include/linux/latencytop.h
+++ b/include/linux/latencytop.h
@@ -9,6 +9,7 @@
 #ifndef _INCLUDE_GUARD_LATENCYTOP_H_
 #define _INCLUDE_GUARD_LATENCYTOP_H_
 
+#include <linux/compiler.h>
 #ifdef CONFIG_LATENCYTOP
 
 #define LT_SAVECOUNT		32
@@ -24,7 +25,14 @@ struct latency_record {
 
 struct task_struct;
 
-void account_scheduler_latency(struct task_struct *task, int usecs, int inter);
+extern int latencytop_enabled;
+void __account_scheduler_latency(struct task_struct *task, int usecs, int inter);
+static inline void
+account_scheduler_latency(struct task_struct *task, int usecs, int inter)
+{
+	if (unlikely(latencytop_enabled))
+		__account_scheduler_latency(task, usecs, inter);
+}
 
 void clear_all_latency_tracing(struct task_struct *p);
 
diff --git a/include/linux/plist.h b/include/linux/plist.h
index 85de2f05587..45926d77d6a 100644
--- a/include/linux/plist.h
+++ b/include/linux/plist.h
@@ -96,6 +96,10 @@ struct plist_node {
 # define PLIST_HEAD_LOCK_INIT(_lock)
 #endif
 
+#define _PLIST_HEAD_INIT(head)				\
+	.prio_list = LIST_HEAD_INIT((head).prio_list),	\
+	.node_list = LIST_HEAD_INIT((head).node_list)
+
 /**
  * PLIST_HEAD_INIT - static struct plist_head initializer
  * @head:	struct plist_head variable name
@@ -103,8 +107,7 @@ struct plist_node {
  */
 #define PLIST_HEAD_INIT(head, _lock)			\
 {							\
-	.prio_list = LIST_HEAD_INIT((head).prio_list),	\
-	.node_list = LIST_HEAD_INIT((head).node_list),	\
+        _PLIST_HEAD_INIT(head),                         \
 	PLIST_HEAD_LOCK_INIT(&(_lock))			\
 }
 
@@ -116,7 +119,7 @@ struct plist_node {
 #define PLIST_NODE_INIT(node, __prio)			\
 {							\
 	.prio  = (__prio),				\
-	.plist = PLIST_HEAD_INIT((node).plist, NULL),	\
+	.plist = { _PLIST_HEAD_INIT((node).plist) }, 	\
 }
 
 /**
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 2c36f62e754..ff904b0606d 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -998,6 +998,7 @@ struct sched_class {
 			      struct rq *busiest, struct sched_domain *sd,
 			      enum cpu_idle_type idle);
 	void (*pre_schedule) (struct rq *this_rq, struct task_struct *task);
+	int (*needs_post_schedule) (struct rq *this_rq);
 	void (*post_schedule) (struct rq *this_rq);
 	void (*task_wake_up) (struct rq *this_rq, struct task_struct *task);
 
@@ -1052,6 +1053,10 @@ struct sched_entity {
 	u64			last_wakeup;
 	u64			avg_overlap;
 
+	u64			start_runtime;
+	u64			avg_wakeup;
+	u64			nr_migrations;
+
 #ifdef CONFIG_SCHEDSTATS
 	u64			wait_start;
 	u64			wait_max;
@@ -1067,7 +1072,6 @@ struct sched_entity {
 	u64			exec_max;
 	u64			slice_max;
 
-	u64			nr_migrations;
 	u64			nr_migrations_cold;
 	u64			nr_failed_migrations_affine;
 	u64			nr_failed_migrations_running;
@@ -1164,6 +1168,7 @@ struct task_struct {
 #endif
 
 	struct list_head tasks;
+	struct plist_node pushable_tasks;
 
 	struct mm_struct *mm, *active_mm;
 
@@ -1675,6 +1680,16 @@ static inline int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask)
 	return set_cpus_allowed_ptr(p, &new_mask);
 }
 
+/*
+ * Architectures can set this to 1 if they have specified
+ * CONFIG_HAVE_UNSTABLE_SCHED_CLOCK in their arch Kconfig,
+ * but then during bootup it turns out that sched_clock()
+ * is reliable after all:
+ */
+#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
+extern int sched_clock_stable;
+#endif
+
 extern unsigned long long sched_clock(void);
 
 extern void sched_clock_init(void);
diff --git a/init/Kconfig b/init/Kconfig
index 6a5c5fed66c..68699137b14 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -966,7 +966,6 @@ config SLABINFO
 
 config RT_MUTEXES
 	boolean
-	select PLIST
 
 config BASE_SMALL
 	int
diff --git a/kernel/latencytop.c b/kernel/latencytop.c
index 449db466bdb..ca07c5c0c91 100644
--- a/kernel/latencytop.c
+++ b/kernel/latencytop.c
@@ -9,6 +9,44 @@
  * as published by the Free Software Foundation; version 2
  * of the License.
  */
+
+/*
+ * CONFIG_LATENCYTOP enables a kernel latency tracking infrastructure that is
+ * used by the "latencytop" userspace tool. The latency that is tracked is not
+ * the 'traditional' interrupt latency (which is primarily caused by something
+ * else consuming CPU), but instead, it is the latency an application encounters
+ * because the kernel sleeps on its behalf for various reasons.
+ *
+ * This code tracks 2 levels of statistics:
+ * 1) System level latency
+ * 2) Per process latency
+ *
+ * The latency is stored in fixed sized data structures in an accumulated form;
+ * if the "same" latency cause is hit twice, this will be tracked as one entry
+ * in the data structure. Both the count, total accumulated latency and maximum
+ * latency are tracked in this data structure. When the fixed size structure is
+ * full, no new causes are tracked until the buffer is flushed by writing to
+ * the /proc file; the userspace tool does this on a regular basis.
+ *
+ * A latency cause is identified by a stringified backtrace at the point that
+ * the scheduler gets invoked. The userland tool will use this string to
+ * identify the cause of the latency in human readable form.
+ *
+ * The information is exported via /proc/latency_stats and /proc/<pid>/latency.
+ * These files look like this:
+ *
+ * Latency Top version : v0.1
+ * 70 59433 4897 i915_irq_wait drm_ioctl vfs_ioctl do_vfs_ioctl sys_ioctl
+ * |    |    |    |
+ * |    |    |    +----> the stringified backtrace
+ * |    |    +---------> The maximum latency for this entry in microseconds
+ * |    +--------------> The accumulated latency for this entry (microseconds)
+ * +-------------------> The number of times this entry is hit
+ *
+ * (note: the average latency is the accumulated latency divided by the number
+ * of times)
+ */
+
 #include <linux/latencytop.h>
 #include <linux/kallsyms.h>
 #include <linux/seq_file.h>
@@ -72,7 +110,7 @@ account_global_scheduler_latency(struct task_struct *tsk, struct latency_record
 				firstnonnull = i;
 			continue;
 		}
-		for (q = 0 ; q < LT_BACKTRACEDEPTH ; q++) {
+		for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 			unsigned long record = lat->backtrace[q];
 
 			if (latency_record[i].backtrace[q] != record) {
@@ -101,31 +139,52 @@ account_global_scheduler_latency(struct task_struct *tsk, struct latency_record
 	memcpy(&latency_record[i], lat, sizeof(struct latency_record));
 }
 
-static inline void store_stacktrace(struct task_struct *tsk, struct latency_record *lat)
+/*
+ * Iterator to store a backtrace into a latency record entry
+ */
+static inline void store_stacktrace(struct task_struct *tsk,
+					struct latency_record *lat)
 {
 	struct stack_trace trace;
 
 	memset(&trace, 0, sizeof(trace));
 	trace.max_entries = LT_BACKTRACEDEPTH;
 	trace.entries = &lat->backtrace[0];
-	trace.skip = 0;
 	save_stack_trace_tsk(tsk, &trace);
 }
 
+/**
+ * __account_scheduler_latency - record an occured latency
+ * @tsk - the task struct of the task hitting the latency
+ * @usecs - the duration of the latency in microseconds
+ * @inter - 1 if the sleep was interruptible, 0 if uninterruptible
+ *
+ * This function is the main entry point for recording latency entries
+ * as called by the scheduler.
+ *
+ * This function has a few special cases to deal with normal 'non-latency'
+ * sleeps: specifically, interruptible sleep longer than 5 msec is skipped
+ * since this usually is caused by waiting for events via select() and co.
+ *
+ * Negative latencies (caused by time going backwards) are also explicitly
+ * skipped.
+ */
 void __sched
-account_scheduler_latency(struct task_struct *tsk, int usecs, int inter)
+__account_scheduler_latency(struct task_struct *tsk, int usecs, int inter)
 {
 	unsigned long flags;
 	int i, q;
 	struct latency_record lat;
 
-	if (!latencytop_enabled)
-		return;
-
 	/* Long interruptible waits are generally user requested... */
 	if (inter && usecs > 5000)
 		return;
 
+	/* Negative sleeps are time going backwards */
+	/* Zero-time sleeps are non-interesting */
+	if (usecs <= 0)
+		return;
+
 	memset(&lat, 0, sizeof(lat));
 	lat.count = 1;
 	lat.time = usecs;
@@ -143,12 +202,12 @@ account_scheduler_latency(struct task_struct *tsk, int usecs, int inter)
 	if (tsk->latency_record_count >= LT_SAVECOUNT)
 		goto out_unlock;
 
-	for (i = 0; i < LT_SAVECOUNT ; i++) {
+	for (i = 0; i < LT_SAVECOUNT; i++) {
 		struct latency_record *mylat;
 		int same = 1;
 
 		mylat = &tsk->latency_record[i];
-		for (q = 0 ; q < LT_BACKTRACEDEPTH ; q++) {
+		for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 			unsigned long record = lat.backtrace[q];
 
 			if (mylat->backtrace[q] != record) {
@@ -186,7 +245,7 @@ static int lstats_show(struct seq_file *m, void *v)
 	for (i = 0; i < MAXLR; i++) {
 		if (latency_record[i].backtrace[0]) {
 			int q;
-			seq_printf(m, "%i %li %li ",
+			seq_printf(m, "%i %lu %lu ",
 				latency_record[i].count,
 				latency_record[i].time,
 				latency_record[i].max);
@@ -223,7 +282,7 @@ static int lstats_open(struct inode *inode, struct file *filp)
 	return single_open(filp, lstats_show, NULL);
 }
 
-static struct file_operations lstats_fops = {
+static const struct file_operations lstats_fops = {
 	.open		= lstats_open,
 	.read		= seq_read,
 	.write		= lstats_write,
@@ -236,4 +295,4 @@ static int __init init_lstats_procfs(void)
 	proc_create("latency_stats", 0644, NULL, &lstats_fops);
 	return 0;
 }
-__initcall(init_lstats_procfs);
+device_initcall(init_lstats_procfs);
diff --git a/kernel/sched.c b/kernel/sched.c
index 8e2558c2ba6..9f8506d68fd 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -331,6 +331,13 @@ static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp;
  */
 static DEFINE_SPINLOCK(task_group_lock);
 
+#ifdef CONFIG_SMP
+static int root_task_group_empty(void)
+{
+	return list_empty(&root_task_group.children);
+}
+#endif
+
 #ifdef CONFIG_FAIR_GROUP_SCHED
 #ifdef CONFIG_USER_SCHED
 # define INIT_TASK_GROUP_LOAD	(2*NICE_0_LOAD)
@@ -391,6 +398,13 @@ static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
 
 #else
 
+#ifdef CONFIG_SMP
+static int root_task_group_empty(void)
+{
+	return 1;
+}
+#endif
+
 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
 static inline struct task_group *task_group(struct task_struct *p)
 {
@@ -467,11 +481,17 @@ struct rt_rq {
 	struct rt_prio_array active;
 	unsigned long rt_nr_running;
 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
-	int highest_prio; /* highest queued rt task prio */
+	struct {
+		int curr; /* highest queued rt task prio */
+#ifdef CONFIG_SMP
+		int next; /* next highest */
+#endif
+	} highest_prio;
 #endif
 #ifdef CONFIG_SMP
 	unsigned long rt_nr_migratory;
 	int overloaded;
+	struct plist_head pushable_tasks;
 #endif
 	int rt_throttled;
 	u64 rt_time;
@@ -549,7 +569,6 @@ struct rq {
 	unsigned long nr_running;
 	#define CPU_LOAD_IDX_MAX 5
 	unsigned long cpu_load[CPU_LOAD_IDX_MAX];
-	unsigned char idle_at_tick;
 #ifdef CONFIG_NO_HZ
 	unsigned long last_tick_seen;
 	unsigned char in_nohz_recently;
@@ -590,6 +609,7 @@ struct rq {
 	struct root_domain *rd;
 	struct sched_domain *sd;
 
+	unsigned char idle_at_tick;
 	/* For active balancing */
 	int active_balance;
 	int push_cpu;
@@ -618,9 +638,6 @@ struct rq {
 	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
 
 	/* sys_sched_yield() stats */
-	unsigned int yld_exp_empty;
-	unsigned int yld_act_empty;
-	unsigned int yld_both_empty;
 	unsigned int yld_count;
 
 	/* schedule() stats */
@@ -1183,10 +1200,10 @@ static void resched_task(struct task_struct *p)
 
 	assert_spin_locked(&task_rq(p)->lock);
 
-	if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED)))
+	if (test_tsk_need_resched(p))
 		return;
 
-	set_tsk_thread_flag(p, TIF_NEED_RESCHED);
+	set_tsk_need_resched(p);
 
 	cpu = task_cpu(p);
 	if (cpu == smp_processor_id())
@@ -1242,7 +1259,7 @@ void wake_up_idle_cpu(int cpu)
 	 * lockless. The worst case is that the other CPU runs the
 	 * idle task through an additional NOOP schedule()
 	 */
-	set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED);
+	set_tsk_need_resched(rq->idle);
 
 	/* NEED_RESCHED must be visible before we test polling */
 	smp_mb();
@@ -1610,21 +1627,42 @@ static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd)
 
 #endif
 
+#ifdef CONFIG_PREEMPT
+
 /*
- * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
+ * fair double_lock_balance: Safely acquires both rq->locks in a fair
+ * way at the expense of forcing extra atomic operations in all
+ * invocations.  This assures that the double_lock is acquired using the
+ * same underlying policy as the spinlock_t on this architecture, which
+ * reduces latency compared to the unfair variant below.  However, it
+ * also adds more overhead and therefore may reduce throughput.
  */
-static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
+static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
+	__releases(this_rq->lock)
+	__acquires(busiest->lock)
+	__acquires(this_rq->lock)
+{
+	spin_unlock(&this_rq->lock);
+	double_rq_lock(this_rq, busiest);
+
+	return 1;
+}
+
+#else
+/*
+ * Unfair double_lock_balance: Optimizes throughput at the expense of
+ * latency by eliminating extra atomic operations when the locks are
+ * already in proper order on entry.  This favors lower cpu-ids and will
+ * grant the double lock to lower cpus over higher ids under contention,
+ * regardless of entry order into the function.
+ */
+static int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
 	__releases(this_rq->lock)
 	__acquires(busiest->lock)
 	__acquires(this_rq->lock)
 {
 	int ret = 0;
 
-	if (unlikely(!irqs_disabled())) {
-		/* printk() doesn't work good under rq->lock */
-		spin_unlock(&this_rq->lock);
-		BUG_ON(1);
-	}
 	if (unlikely(!spin_trylock(&busiest->lock))) {
 		if (busiest < this_rq) {
 			spin_unlock(&this_rq->lock);
@@ -1637,6 +1675,22 @@ static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
 	return ret;
 }
 
+#endif /* CONFIG_PREEMPT */
+
+/*
+ * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
+ */
+static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
+{
+	if (unlikely(!irqs_disabled())) {
+		/* printk() doesn't work good under rq->lock */
+		spin_unlock(&this_rq->lock);
+		BUG_ON(1);
+	}
+
+	return _double_lock_balance(this_rq, busiest);
+}
+
 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
 	__releases(busiest->lock)
 {
@@ -1705,6 +1759,9 @@ static void update_avg(u64 *avg, u64 sample)
 
 static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup)
 {
+	if (wakeup)
+		p->se.start_runtime = p->se.sum_exec_runtime;
+
 	sched_info_queued(p);
 	p->sched_class->enqueue_task(rq, p, wakeup);
 	p->se.on_rq = 1;
@@ -1712,10 +1769,15 @@ static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup)
 
 static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep)
 {
-	if (sleep && p->se.last_wakeup) {
-		update_avg(&p->se.avg_overlap,
-			   p->se.sum_exec_runtime - p->se.last_wakeup);
-		p->se.last_wakeup = 0;
+	if (sleep) {
+		if (p->se.last_wakeup) {
+			update_avg(&p->se.avg_overlap,
+				p->se.sum_exec_runtime - p->se.last_wakeup);
+			p->se.last_wakeup = 0;
+		} else {
+			update_avg(&p->se.avg_wakeup,
+				sysctl_sched_wakeup_granularity);
+		}
 	}
 
 	sched_info_dequeued(p);
@@ -2017,7 +2079,7 @@ unsigned long wait_task_inactive(struct task_struct *p, long match_state)
 		 * it must be off the runqueue _entirely_, and not
 		 * preempted!
 		 *
-		 * So if it wa still runnable (but just not actively
+		 * So if it was still runnable (but just not actively
 		 * running right now), it's preempted, and we should
 		 * yield - it could be a while.
 		 */
@@ -2267,7 +2329,7 @@ static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync)
 		sync = 0;
 
 #ifdef CONFIG_SMP
-	if (sched_feat(LB_WAKEUP_UPDATE)) {
+	if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) {
 		struct sched_domain *sd;
 
 		this_cpu = raw_smp_processor_id();
@@ -2345,6 +2407,22 @@ out_activate:
 	activate_task(rq, p, 1);
 	success = 1;
 
+	/*
+	 * Only attribute actual wakeups done by this task.
+	 */
+	if (!in_interrupt()) {
+		struct sched_entity *se = &current->se;
+		u64 sample = se->sum_exec_runtime;
+
+		if (se->last_wakeup)
+			sample -= se->last_wakeup;
+		else
+			sample -= se->start_runtime;
+		update_avg(&se->avg_wakeup, sample);
+
+		se->last_wakeup = se->sum_exec_runtime;
+	}
+
 out_running:
 	trace_sched_wakeup(rq, p, success);
 	check_preempt_curr(rq, p, sync);
@@ -2355,8 +2433,6 @@ out_running:
 		p->sched_class->task_wake_up(rq, p);
 #endif
 out:
-	current->se.last_wakeup = current->se.sum_exec_runtime;
-
 	task_rq_unlock(rq, &flags);
 
 	return success;
@@ -2386,6 +2462,8 @@ static void __sched_fork(struct task_struct *p)
 	p->se.prev_sum_exec_runtime	= 0;
 	p->se.last_wakeup		= 0;
 	p->se.avg_overlap		= 0;
+	p->se.start_runtime		= 0;
+	p->se.avg_wakeup		= sysctl_sched_wakeup_granularity;
 
 #ifdef CONFIG_SCHEDSTATS
 	p->se.wait_start		= 0;
@@ -2448,6 +2526,8 @@ void sched_fork(struct task_struct *p, int clone_flags)
 	/* Want to start with kernel preemption disabled. */
 	task_thread_info(p)->preempt_count = 1;
 #endif
+	plist_node_init(&p->pushable_tasks, MAX_PRIO);
+
 	put_cpu();
 }
 
@@ -2491,7 +2571,7 @@ void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
 #ifdef CONFIG_PREEMPT_NOTIFIERS
 
 /**
- * preempt_notifier_register - tell me when current is being being preempted & rescheduled
+ * preempt_notifier_register - tell me when current is being preempted & rescheduled
  * @notifier: notifier struct to register
  */
 void preempt_notifier_register(struct preempt_notifier *notifier)
@@ -2588,6 +2668,12 @@ static void finish_task_switch(struct rq *rq, struct task_struct *prev)
 {
 	struct mm_struct *mm = rq->prev_mm;
 	long prev_state;
+#ifdef CONFIG_SMP
+	int post_schedule = 0;
+
+	if (current->sched_class->needs_post_schedule)
+		post_schedule = current->sched_class->needs_post_schedule(rq);
+#endif
 
 	rq->prev_mm = NULL;
 
@@ -2606,7 +2692,7 @@ static void finish_task_switch(struct rq *rq, struct task_struct *prev)
 	finish_arch_switch(prev);
 	finish_lock_switch(rq, prev);
 #ifdef CONFIG_SMP
-	if (current->sched_class->post_schedule)
+	if (post_schedule)
 		current->sched_class->post_schedule(rq);
 #endif
 
@@ -2913,6 +2999,7 @@ int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu,
 		     struct sched_domain *sd, enum cpu_idle_type idle,
 		     int *all_pinned)
 {
+	int tsk_cache_hot = 0;
 	/*
 	 * We do not migrate tasks that are:
 	 * 1) running (obviously), or
@@ -2936,10 +3023,11 @@ int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu,
 	 * 2) too many balance attempts have failed.
 	 */
 
-	if (!task_hot(p, rq->clock, sd) ||
-			sd->nr_balance_failed > sd->cache_nice_tries) {
+	tsk_cache_hot = task_hot(p, rq->clock, sd);
+	if (!tsk_cache_hot ||
+		sd->nr_balance_failed > sd->cache_nice_tries) {
 #ifdef CONFIG_SCHEDSTATS
-		if (task_hot(p, rq->clock, sd)) {
+		if (tsk_cache_hot) {
 			schedstat_inc(sd, lb_hot_gained[idle]);
 			schedstat_inc(p, se.nr_forced_migrations);
 		}
@@ -2947,7 +3035,7 @@ int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu,
 		return 1;
 	}
 
-	if (task_hot(p, rq->clock, sd)) {
+	if (tsk_cache_hot) {
 		schedstat_inc(p, se.nr_failed_migrations_hot);
 		return 0;
 	}
@@ -2987,6 +3075,16 @@ next:
 	pulled++;
 	rem_load_move -= p->se.load.weight;
 
+#ifdef CONFIG_PREEMPT
+	/*
+	 * NEWIDLE balancing is a source of latency, so preemptible kernels
+	 * will stop after the first task is pulled to minimize the critical
+	 * section.
+	 */
+	if (idle == CPU_NEWLY_IDLE)
+		goto out;
+#endif
+
 	/*
 	 * We only want to steal up to the prescribed amount of weighted load.
 	 */
@@ -3033,9 +3131,15 @@ static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
 				sd, idle, all_pinned, &this_best_prio);
 		class = class->next;
 
+#ifdef CONFIG_PREEMPT
+		/*
+		 * NEWIDLE balancing is a source of latency, so preemptible
+		 * kernels will stop after the first task is pulled to minimize
+		 * the critical section.
+		 */
 		if (idle == CPU_NEWLY_IDLE && this_rq->nr_running)
 			break;
-
+#endif
 	} while (class && max_load_move > total_load_moved);
 
 	return total_load_moved > 0;
@@ -3085,246 +3189,479 @@ static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest,
 
 	return 0;
 }
+/********** Helpers for find_busiest_group ************************/
+/**
+ * sd_lb_stats - Structure to store the statistics of a sched_domain
+ * 		during load balancing.
+ */
+struct sd_lb_stats {
+	struct sched_group *busiest; /* Busiest group in this sd */
+	struct sched_group *this;  /* Local group in this sd */
+	unsigned long total_load;  /* Total load of all groups in sd */
+	unsigned long total_pwr;   /*	Total power of all groups in sd */
+	unsigned long avg_load;	   /* Average load across all groups in sd */
+