blob: dcd553cc4ee89b52ec511f70361c4f7fd586d976 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * kernel/sched.c
3 *
4 * Kernel scheduler and related syscalls
5 *
6 * Copyright (C) 1991-2002 Linus Torvalds
7 *
8 * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and
9 * make semaphores SMP safe
10 * 1998-11-19 Implemented schedule_timeout() and related stuff
11 * by Andrea Arcangeli
12 * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar:
13 * hybrid priority-list and round-robin design with
14 * an array-switch method of distributing timeslices
15 * and per-CPU runqueues. Cleanups and useful suggestions
16 * by Davide Libenzi, preemptible kernel bits by Robert Love.
17 * 2003-09-03 Interactivity tuning by Con Kolivas.
18 * 2004-04-02 Scheduler domains code by Nick Piggin
Ingo Molnarc31f2e82007-07-09 18:52:01 +020019 * 2007-04-15 Work begun on replacing all interactivity tuning with a
20 * fair scheduling design by Con Kolivas.
21 * 2007-05-05 Load balancing (smp-nice) and other improvements
22 * by Peter Williams
23 * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith
24 * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri
Ingo Molnarb9131762008-01-25 21:08:19 +010025 * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins,
26 * Thomas Gleixner, Mike Kravetz
Linus Torvalds1da177e2005-04-16 15:20:36 -070027 */
28
29#include <linux/mm.h>
30#include <linux/module.h>
31#include <linux/nmi.h>
32#include <linux/init.h>
Ingo Molnardff06c12007-07-09 18:52:00 +020033#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/highmem.h>
35#include <linux/smp_lock.h>
36#include <asm/mmu_context.h>
37#include <linux/interrupt.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080038#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/completion.h>
40#include <linux/kernel_stat.h>
Ingo Molnar9a11b49a2006-07-03 00:24:33 -070041#include <linux/debug_locks.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/security.h>
43#include <linux/notifier.h>
44#include <linux/profile.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080045#include <linux/freezer.h>
akpm@osdl.org198e2f12006-01-12 01:05:30 -080046#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/blkdev.h>
48#include <linux/delay.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070049#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/smp.h>
51#include <linux/threads.h>
52#include <linux/timer.h>
53#include <linux/rcupdate.h>
54#include <linux/cpu.h>
55#include <linux/cpuset.h>
56#include <linux/percpu.h>
57#include <linux/kthread.h>
58#include <linux/seq_file.h>
Nick Piggine692ab52007-07-26 13:40:43 +020059#include <linux/sysctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include <linux/syscalls.h>
61#include <linux/times.h>
Jay Lan8f0ab512006-09-30 23:28:59 -070062#include <linux/tsacct_kern.h>
bibo maoc6fd91f2006-03-26 01:38:20 -080063#include <linux/kprobes.h>
Shailabh Nagar0ff92242006-07-14 00:24:37 -070064#include <linux/delayacct.h>
Eric Dumazet5517d862007-05-08 00:32:57 -070065#include <linux/reciprocal_div.h>
Ingo Molnardff06c12007-07-09 18:52:00 +020066#include <linux/unistd.h>
Jens Axboef5ff8422007-09-21 09:19:54 +020067#include <linux/pagemap.h>
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +010068#include <linux/hrtimer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
Eric Dumazet5517d862007-05-08 00:32:57 -070070#include <asm/tlb.h>
Satyam Sharma838225b2007-10-24 18:23:50 +020071#include <asm/irq_regs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
73/*
Alexey Dobriyanb035b6d2007-02-10 01:45:10 -080074 * Scheduler clock - returns current time in nanosec units.
75 * This is default implementation.
76 * Architectures and sub-architectures can override this.
77 */
78unsigned long long __attribute__((weak)) sched_clock(void)
79{
Eric Dumazetd6322fa2007-11-09 22:39:38 +010080 return (unsigned long long)jiffies * (NSEC_PER_SEC / HZ);
Alexey Dobriyanb035b6d2007-02-10 01:45:10 -080081}
82
83/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 * Convert user-nice values [ -20 ... 0 ... 19 ]
85 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
86 * and back.
87 */
88#define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
89#define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
90#define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
91
92/*
93 * 'User priority' is the nice value converted to something we
94 * can work with better when scaling various scheduler parameters,
95 * it's a [ 0 ... 39 ] range.
96 */
97#define USER_PRIO(p) ((p)-MAX_RT_PRIO)
98#define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
99#define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
100
101/*
Ingo Molnard7876a02008-01-25 21:08:19 +0100102 * Helpers for converting nanosecond timing to jiffy resolution
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 */
Eric Dumazetd6322fa2007-11-09 22:39:38 +0100104#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200106#define NICE_0_LOAD SCHED_LOAD_SCALE
107#define NICE_0_SHIFT SCHED_LOAD_SHIFT
108
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109/*
110 * These are the 'tuning knobs' of the scheduler:
111 *
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +0200112 * default timeslice is 100 msecs (used only for SCHED_RR tasks).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 * Timeslices get refilled after they expire.
114 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115#define DEF_TIMESLICE (100 * HZ / 1000)
Peter Williams2dd73a42006-06-27 02:54:34 -0700116
Eric Dumazet5517d862007-05-08 00:32:57 -0700117#ifdef CONFIG_SMP
118/*
119 * Divide a load by a sched group cpu_power : (load / sg->__cpu_power)
120 * Since cpu_power is a 'constant', we can use a reciprocal divide.
121 */
122static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load)
123{
124 return reciprocal_divide(load, sg->reciprocal_cpu_power);
125}
126
127/*
128 * Each time a sched group cpu_power is changed,
129 * we must compute its reciprocal value
130 */
131static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val)
132{
133 sg->__cpu_power += val;
134 sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power);
135}
136#endif
137
Ingo Molnare05606d2007-07-09 18:51:59 +0200138static inline int rt_policy(int policy)
139{
140 if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR))
141 return 1;
142 return 0;
143}
144
145static inline int task_has_rt_policy(struct task_struct *p)
146{
147 return rt_policy(p->policy);
148}
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150/*
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200151 * This is the priority-queue data structure of the RT scheduling class:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 */
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200153struct rt_prio_array {
154 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
155 struct list_head queue[MAX_RT_PRIO];
156};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100158#ifdef CONFIG_GROUP_SCHED
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200159
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -0700160#include <linux/cgroup.h>
161
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200162struct cfs_rq;
163
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100164static LIST_HEAD(task_groups);
165
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200166/* task group related information */
Ingo Molnar4cf86d72007-10-15 17:00:14 +0200167struct task_group {
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100168#ifdef CONFIG_CGROUP_SCHED
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -0700169 struct cgroup_subsys_state css;
170#endif
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100171
172#ifdef CONFIG_FAIR_GROUP_SCHED
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200173 /* schedulable entities of this group on each cpu */
174 struct sched_entity **se;
175 /* runqueue "owned" by this group on each cpu */
176 struct cfs_rq **cfs_rq;
177 unsigned long shares;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100178#endif
179
180#ifdef CONFIG_RT_GROUP_SCHED
181 struct sched_rt_entity **rt_se;
182 struct rt_rq **rt_rq;
183
184 u64 rt_runtime;
185#endif
Srivatsa Vaddagiri6b2d7702008-01-25 21:08:00 +0100186
Srivatsa Vaddagiriae8393e2007-10-29 21:18:11 +0100187 struct rcu_head rcu;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100188 struct list_head list;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200189};
190
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100191#ifdef CONFIG_FAIR_GROUP_SCHED
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200192/* Default task group's sched entity on each cpu */
193static DEFINE_PER_CPU(struct sched_entity, init_sched_entity);
194/* Default task group's cfs_rq on each cpu */
195static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp;
196
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100197static struct sched_entity *init_sched_entity_p[NR_CPUS];
198static struct cfs_rq *init_cfs_rq_p[NR_CPUS];
199#endif
200
201#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100202static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity);
203static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp;
204
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100205static struct sched_rt_entity *init_sched_rt_entity_p[NR_CPUS];
206static struct rt_rq *init_rt_rq_p[NR_CPUS];
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100207#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100208
Peter Zijlstra8ed36992008-02-13 15:45:39 +0100209/* task_group_lock serializes add/remove of task groups and also changes to
Srivatsa Vaddagiriec2c5072008-01-25 21:07:59 +0100210 * a task group's cpu shares.
211 */
Peter Zijlstra8ed36992008-02-13 15:45:39 +0100212static DEFINE_SPINLOCK(task_group_lock);
Srivatsa Vaddagiriec2c5072008-01-25 21:07:59 +0100213
Srivatsa Vaddagiria1835612008-01-25 21:08:00 +0100214/* doms_cur_mutex serializes access to doms_cur[] array */
215static DEFINE_MUTEX(doms_cur_mutex);
216
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100217#ifdef CONFIG_FAIR_GROUP_SCHED
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100218#ifdef CONFIG_USER_SCHED
Ingo Molnar0eab9142008-01-25 21:08:19 +0100219# define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD)
Srivatsa Vaddagiri24e377a2007-10-15 17:00:09 +0200220#else
Srivatsa Vaddagiri93f992c2008-01-25 21:07:59 +0100221# define INIT_TASK_GROUP_LOAD NICE_0_LOAD
Srivatsa Vaddagiri24e377a2007-10-15 17:00:09 +0200222#endif
223
Srivatsa Vaddagiri93f992c2008-01-25 21:07:59 +0100224static int init_task_group_load = INIT_TASK_GROUP_LOAD;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100225#endif
226
227/* Default task group.
228 * Every task in system belong to this group at bootup.
229 */
230struct task_group init_task_group = {
231#ifdef CONFIG_FAIR_GROUP_SCHED
232 .se = init_sched_entity_p,
233 .cfs_rq = init_cfs_rq_p,
234#endif
235
236#ifdef CONFIG_RT_GROUP_SCHED
237 .rt_se = init_sched_rt_entity_p,
238 .rt_rq = init_rt_rq_p,
239#endif
240};
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200241
242/* return group to which a task belongs */
Ingo Molnar4cf86d72007-10-15 17:00:14 +0200243static inline struct task_group *task_group(struct task_struct *p)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200244{
Ingo Molnar4cf86d72007-10-15 17:00:14 +0200245 struct task_group *tg;
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +0200246
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100247#ifdef CONFIG_USER_SCHED
Srivatsa Vaddagiri24e377a2007-10-15 17:00:09 +0200248 tg = p->user->tg;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100249#elif defined(CONFIG_CGROUP_SCHED)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -0700250 tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id),
251 struct task_group, css);
Srivatsa Vaddagiri24e377a2007-10-15 17:00:09 +0200252#else
Ingo Molnar41a2d6c2007-12-05 15:46:09 +0100253 tg = &init_task_group;
Srivatsa Vaddagiri24e377a2007-10-15 17:00:09 +0200254#endif
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +0200255 return tg;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200256}
257
258/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100259static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200260{
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100261#ifdef CONFIG_FAIR_GROUP_SCHED
Dmitry Adamushkoce96b5a2007-11-15 20:57:40 +0100262 p->se.cfs_rq = task_group(p)->cfs_rq[cpu];
263 p->se.parent = task_group(p)->se[cpu];
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100264#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100265
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100266#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100267 p->rt.rt_rq = task_group(p)->rt_rq[cpu];
268 p->rt.parent = task_group(p)->rt_se[cpu];
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100269#endif
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200270}
271
Srivatsa Vaddagiria1835612008-01-25 21:08:00 +0100272static inline void lock_doms_cur(void)
273{
274 mutex_lock(&doms_cur_mutex);
275}
276
277static inline void unlock_doms_cur(void)
278{
279 mutex_unlock(&doms_cur_mutex);
280}
281
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200282#else
283
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100284static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
Srivatsa Vaddagiria1835612008-01-25 21:08:00 +0100285static inline void lock_doms_cur(void) { }
286static inline void unlock_doms_cur(void) { }
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200287
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100288#endif /* CONFIG_GROUP_SCHED */
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200289
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200290/* CFS-related fields in a runqueue */
291struct cfs_rq {
292 struct load_weight load;
293 unsigned long nr_running;
294
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200295 u64 exec_clock;
Ingo Molnare9acbff2007-10-15 17:00:04 +0200296 u64 min_vruntime;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200297
298 struct rb_root tasks_timeline;
299 struct rb_node *rb_leftmost;
300 struct rb_node *rb_load_balance_curr;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200301 /* 'curr' points to currently running entity on this cfs_rq.
302 * It is set to NULL otherwise (i.e when none are currently running).
303 */
304 struct sched_entity *curr;
Peter Zijlstraddc97292007-10-15 17:00:10 +0200305
306 unsigned long nr_spread_over;
307
Ingo Molnar62160e3f2007-10-15 17:00:03 +0200308#ifdef CONFIG_FAIR_GROUP_SCHED
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200309 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
310
Ingo Molnar41a2d6c2007-12-05 15:46:09 +0100311 /*
312 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200313 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
314 * (like users, containers etc.)
315 *
316 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
317 * list is used during load balance.
318 */
Ingo Molnar41a2d6c2007-12-05 15:46:09 +0100319 struct list_head leaf_cfs_rq_list;
320 struct task_group *tg; /* group that "owns" this runqueue */
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200321#endif
322};
323
324/* Real-Time classes' related field in a runqueue: */
325struct rt_rq {
326 struct rt_prio_array active;
Steven Rostedt63489e42008-01-25 21:08:03 +0100327 unsigned long rt_nr_running;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100328#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100329 int highest_prio; /* highest queued rt task prio */
330#endif
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100331#ifdef CONFIG_SMP
Gregory Haskins73fe6aa2008-01-25 21:08:07 +0100332 unsigned long rt_nr_migratory;
Gregory Haskinsa22d7fc2008-01-25 21:08:12 +0100333 int overloaded;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100334#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100335 int rt_throttled;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100336 u64 rt_time;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100337
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100338#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +0100339 unsigned long rt_nr_boosted;
340
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100341 struct rq *rq;
342 struct list_head leaf_rt_rq_list;
343 struct task_group *tg;
344 struct sched_rt_entity *rt_se;
345#endif
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200346};
347
Gregory Haskins57d885f2008-01-25 21:08:18 +0100348#ifdef CONFIG_SMP
349
350/*
351 * We add the notion of a root-domain which will be used to define per-domain
Ingo Molnar0eab9142008-01-25 21:08:19 +0100352 * variables. Each exclusive cpuset essentially defines an island domain by
353 * fully partitioning the member cpus from any other cpuset. Whenever a new
Gregory Haskins57d885f2008-01-25 21:08:18 +0100354 * exclusive cpuset is created, we also create and attach a new root-domain
355 * object.
356 *
Gregory Haskins57d885f2008-01-25 21:08:18 +0100357 */
358struct root_domain {
359 atomic_t refcount;
360 cpumask_t span;
361 cpumask_t online;
Gregory Haskins637f5082008-01-25 21:08:18 +0100362
Ingo Molnar0eab9142008-01-25 21:08:19 +0100363 /*
Gregory Haskins637f5082008-01-25 21:08:18 +0100364 * The "RT overload" flag: it gets set if a CPU has more than
365 * one runnable RT task.
366 */
367 cpumask_t rto_mask;
Ingo Molnar0eab9142008-01-25 21:08:19 +0100368 atomic_t rto_count;
Gregory Haskins57d885f2008-01-25 21:08:18 +0100369};
370
Gregory Haskinsdc938522008-01-25 21:08:26 +0100371/*
372 * By default the system creates a single root-domain with all cpus as
373 * members (mimicking the global state we have today).
374 */
Gregory Haskins57d885f2008-01-25 21:08:18 +0100375static struct root_domain def_root_domain;
376
377#endif
378
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200379/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 * This is the main, per-CPU runqueue data structure.
381 *
382 * Locking rule: those places that want to lock multiple runqueues
383 * (such as the load balancing or the thread migration code), lock
384 * acquire operations must be ordered by ascending &runqueue.
385 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700386struct rq {
Ingo Molnard8016492007-10-18 21:32:55 +0200387 /* runqueue lock: */
388 spinlock_t lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
390 /*
391 * nr_running and cpu_load should be in the same cacheline because
392 * remote CPUs use both these fields when doing load calculation.
393 */
394 unsigned long nr_running;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200395 #define CPU_LOAD_IDX_MAX 5
396 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
Siddha, Suresh Bbdecea32007-05-08 00:32:48 -0700397 unsigned char idle_at_tick;
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -0700398#ifdef CONFIG_NO_HZ
399 unsigned char in_nohz_recently;
400#endif
Ingo Molnard8016492007-10-18 21:32:55 +0200401 /* capture load from *all* tasks on this cpu: */
402 struct load_weight load;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200403 unsigned long nr_load_updates;
404 u64 nr_switches;
405
406 struct cfs_rq cfs;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100407 struct rt_rq rt;
408 u64 rt_period_expire;
Peter Zijlstra48d5e252008-01-25 21:08:31 +0100409 int rt_throttled;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100410
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200411#ifdef CONFIG_FAIR_GROUP_SCHED
Ingo Molnard8016492007-10-18 21:32:55 +0200412 /* list of leaf cfs_rq on this cpu: */
413 struct list_head leaf_cfs_rq_list;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100414#endif
415#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100416 struct list_head leaf_rt_rq_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
419 /*
420 * This is part of a global counter where only the total sum
421 * over all CPUs matters. A task can increase this counter on
422 * one CPU and if it got migrated afterwards it may decrease
423 * it on another CPU. Always updated under the runqueue lock:
424 */
425 unsigned long nr_uninterruptible;
426
Ingo Molnar36c8b582006-07-03 00:25:41 -0700427 struct task_struct *curr, *idle;
Christoph Lameterc9819f42006-12-10 02:20:25 -0800428 unsigned long next_balance;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 struct mm_struct *prev_mm;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200430
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200431 u64 clock, prev_clock_raw;
432 s64 clock_max_delta;
433
Guillaume Chazaraincc203d22008-01-25 21:08:34 +0100434 unsigned int clock_warps, clock_overflows, clock_underflows;
Ingo Molnar2aa44d02007-08-23 15:18:02 +0200435 u64 idle_clock;
436 unsigned int clock_deep_idle_events;
Ingo Molnar529c7722007-08-10 23:05:11 +0200437 u64 tick_timestamp;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200438
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 atomic_t nr_iowait;
440
441#ifdef CONFIG_SMP
Ingo Molnar0eab9142008-01-25 21:08:19 +0100442 struct root_domain *rd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 struct sched_domain *sd;
444
445 /* For active balancing */
446 int active_balance;
447 int push_cpu;
Ingo Molnard8016492007-10-18 21:32:55 +0200448 /* cpu of this runqueue: */
449 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Ingo Molnar36c8b582006-07-03 00:25:41 -0700451 struct task_struct *migration_thread;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 struct list_head migration_queue;
453#endif
454
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100455#ifdef CONFIG_SCHED_HRTICK
456 unsigned long hrtick_flags;
457 ktime_t hrtick_expire;
458 struct hrtimer hrtick_timer;
459#endif
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461#ifdef CONFIG_SCHEDSTATS
462 /* latency stats */
463 struct sched_info rq_sched_info;
464
465 /* sys_sched_yield() stats */
Ken Chen480b9432007-10-18 21:32:56 +0200466 unsigned int yld_exp_empty;
467 unsigned int yld_act_empty;
468 unsigned int yld_both_empty;
469 unsigned int yld_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
471 /* schedule() stats */
Ken Chen480b9432007-10-18 21:32:56 +0200472 unsigned int sched_switch;
473 unsigned int sched_count;
474 unsigned int sched_goidle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
476 /* try_to_wake_up() stats */
Ken Chen480b9432007-10-18 21:32:56 +0200477 unsigned int ttwu_count;
478 unsigned int ttwu_local;
Ingo Molnarb8efb562007-10-15 17:00:10 +0200479
480 /* BKL stats */
Ken Chen480b9432007-10-18 21:32:56 +0200481 unsigned int bkl_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482#endif
Ingo Molnarfcb99372006-07-03 00:25:10 -0700483 struct lock_class_key rq_lock_key;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484};
485
Fenghua Yuf34e3b62007-07-19 01:48:13 -0700486static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
Ingo Molnardd41f592007-07-09 18:51:59 +0200488static inline void check_preempt_curr(struct rq *rq, struct task_struct *p)
489{
490 rq->curr->sched_class->check_preempt_curr(rq, p);
491}
492
Christoph Lameter0a2966b2006-09-25 23:30:51 -0700493static inline int cpu_of(struct rq *rq)
494{
495#ifdef CONFIG_SMP
496 return rq->cpu;
497#else
498 return 0;
499#endif
500}
501
Nick Piggin674311d2005-06-25 14:57:27 -0700502/*
Ingo Molnarb04a0f42007-08-09 11:16:46 +0200503 * Update the per-runqueue clock, as finegrained as the platform can give
504 * us, but without assuming monotonicity, etc.:
Ingo Molnar20d315d2007-07-09 18:51:58 +0200505 */
Ingo Molnarb04a0f42007-08-09 11:16:46 +0200506static void __update_rq_clock(struct rq *rq)
Ingo Molnar20d315d2007-07-09 18:51:58 +0200507{
508 u64 prev_raw = rq->prev_clock_raw;
509 u64 now = sched_clock();
510 s64 delta = now - prev_raw;
511 u64 clock = rq->clock;
512
Ingo Molnarb04a0f42007-08-09 11:16:46 +0200513#ifdef CONFIG_SCHED_DEBUG
514 WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
515#endif
Ingo Molnar20d315d2007-07-09 18:51:58 +0200516 /*
517 * Protect against sched_clock() occasionally going backwards:
518 */
519 if (unlikely(delta < 0)) {
520 clock++;
521 rq->clock_warps++;
522 } else {
523 /*
524 * Catch too large forward jumps too:
525 */
Ingo Molnar529c7722007-08-10 23:05:11 +0200526 if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) {
527 if (clock < rq->tick_timestamp + TICK_NSEC)
528 clock = rq->tick_timestamp + TICK_NSEC;
529 else
530 clock++;
Ingo Molnar20d315d2007-07-09 18:51:58 +0200531 rq->clock_overflows++;
532 } else {
533 if (unlikely(delta > rq->clock_max_delta))
534 rq->clock_max_delta = delta;
535 clock += delta;
536 }
537 }
538
539 rq->prev_clock_raw = now;
540 rq->clock = clock;
Ingo Molnar20d315d2007-07-09 18:51:58 +0200541}
542
Ingo Molnarb04a0f42007-08-09 11:16:46 +0200543static void update_rq_clock(struct rq *rq)
Ingo Molnar20d315d2007-07-09 18:51:58 +0200544{
Ingo Molnarb04a0f42007-08-09 11:16:46 +0200545 if (likely(smp_processor_id() == cpu_of(rq)))
546 __update_rq_clock(rq);
547}
Ingo Molnar20d315d2007-07-09 18:51:58 +0200548
Ingo Molnar20d315d2007-07-09 18:51:58 +0200549/*
Nick Piggin674311d2005-06-25 14:57:27 -0700550 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -0700551 * See detach_destroy_domains: synchronize_sched for details.
Nick Piggin674311d2005-06-25 14:57:27 -0700552 *
553 * The domain tree of any CPU may only be accessed from within
554 * preempt-disabled sections.
555 */
Ingo Molnar48f24c42006-07-03 00:25:40 -0700556#define for_each_domain(cpu, __sd) \
557 for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
559#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
560#define this_rq() (&__get_cpu_var(runqueues))
561#define task_rq(p) cpu_rq(task_cpu(p))
562#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
563
Peter Zijlstra48d5e252008-01-25 21:08:31 +0100564unsigned long rt_needs_cpu(int cpu)
565{
566 struct rq *rq = cpu_rq(cpu);
567 u64 delta;
568
569 if (!rq->rt_throttled)
570 return 0;
571
572 if (rq->clock > rq->rt_period_expire)
573 return 1;
574
575 delta = rq->rt_period_expire - rq->clock;
576 do_div(delta, NSEC_PER_SEC / HZ);
577
578 return (unsigned long)delta;
579}
580
Ingo Molnare436d802007-07-19 21:28:35 +0200581/*
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200582 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
583 */
584#ifdef CONFIG_SCHED_DEBUG
585# define const_debug __read_mostly
586#else
587# define const_debug static const
588#endif
589
590/*
591 * Debugging: various feature bits
592 */
593enum {
Ingo Molnarbbdba7c2007-10-15 17:00:06 +0200594 SCHED_FEAT_NEW_FAIR_SLEEPERS = 1,
Ingo Molnar96126332007-11-15 20:57:40 +0100595 SCHED_FEAT_WAKEUP_PREEMPT = 2,
596 SCHED_FEAT_START_DEBIT = 4,
Ingo Molnar41a2d6c2007-12-05 15:46:09 +0100597 SCHED_FEAT_TREE_AVG = 8,
598 SCHED_FEAT_APPROX_AVG = 16,
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100599 SCHED_FEAT_HRTICK = 32,
600 SCHED_FEAT_DOUBLE_TICK = 64,
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200601};
602
603const_debug unsigned int sysctl_sched_features =
Ingo Molnar8401f772007-10-18 21:32:55 +0200604 SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 |
Ingo Molnar96126332007-11-15 20:57:40 +0100605 SCHED_FEAT_WAKEUP_PREEMPT * 1 |
Ingo Molnar8401f772007-10-18 21:32:55 +0200606 SCHED_FEAT_START_DEBIT * 1 |
607 SCHED_FEAT_TREE_AVG * 0 |
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100608 SCHED_FEAT_APPROX_AVG * 0 |
609 SCHED_FEAT_HRTICK * 1 |
610 SCHED_FEAT_DOUBLE_TICK * 0;
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200611
612#define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x)
613
614/*
Peter Zijlstrab82d9fd2007-11-09 22:39:39 +0100615 * Number of tasks to iterate in a single balance run.
616 * Limited because this is done with IRQs disabled.
617 */
618const_debug unsigned int sysctl_sched_nr_migrate = 32;
619
620/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100621 * period over which we measure -rt task cpu usage in us.
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100622 * default: 1s
623 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100624unsigned int sysctl_sched_rt_period = 1000000;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100625
Ingo Molnar6892b752008-02-13 14:02:36 +0100626static __read_mostly int scheduler_running;
627
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100628/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100629 * part of the period that we allow rt tasks to run in us.
630 * default: 0.95s
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100631 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100632int sysctl_sched_rt_runtime = 950000;
633
634/*
635 * single value that denotes runtime == period, ie unlimited time.
636 */
637#define RUNTIME_INF ((u64)~0ULL)
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100638
639/*
Ingo Molnare436d802007-07-19 21:28:35 +0200640 * For kernel-internal use: high-speed (but slightly incorrect) per-cpu
641 * clock constructed from sched_clock():
642 */
643unsigned long long cpu_clock(int cpu)
644{
Ingo Molnare436d802007-07-19 21:28:35 +0200645 unsigned long long now;
646 unsigned long flags;
Ingo Molnarb04a0f42007-08-09 11:16:46 +0200647 struct rq *rq;
Ingo Molnare436d802007-07-19 21:28:35 +0200648
Ingo Molnar8ced5f62007-12-07 19:02:47 +0100649 /*
650 * Only call sched_clock() if the scheduler has already been
651 * initialized (some code might call cpu_clock() very early):
652 */
Ingo Molnar6892b752008-02-13 14:02:36 +0100653 if (unlikely(!scheduler_running))
654 return 0;
655
656 local_irq_save(flags);
657 rq = cpu_rq(cpu);
658 update_rq_clock(rq);
Ingo Molnarb04a0f42007-08-09 11:16:46 +0200659 now = rq->clock;
Ingo Molnar2cd4d0e2007-07-26 13:40:43 +0200660 local_irq_restore(flags);
Ingo Molnare436d802007-07-19 21:28:35 +0200661
662 return now;
663}
Paul E. McKenneya58f6f22007-10-15 17:00:14 +0200664EXPORT_SYMBOL_GPL(cpu_clock);
Ingo Molnare436d802007-07-19 21:28:35 +0200665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666#ifndef prepare_arch_switch
Nick Piggin4866cde2005-06-25 14:57:23 -0700667# define prepare_arch_switch(next) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668#endif
Nick Piggin4866cde2005-06-25 14:57:23 -0700669#ifndef finish_arch_switch
670# define finish_arch_switch(prev) do { } while (0)
671#endif
672
Dmitry Adamushko051a1d12007-12-18 15:21:13 +0100673static inline int task_current(struct rq *rq, struct task_struct *p)
674{
675 return rq->curr == p;
676}
677
Nick Piggin4866cde2005-06-25 14:57:23 -0700678#ifndef __ARCH_WANT_UNLOCKED_CTXSW
Ingo Molnar70b97a72006-07-03 00:25:42 -0700679static inline int task_running(struct rq *rq, struct task_struct *p)
Nick Piggin4866cde2005-06-25 14:57:23 -0700680{
Dmitry Adamushko051a1d12007-12-18 15:21:13 +0100681 return task_current(rq, p);
Nick Piggin4866cde2005-06-25 14:57:23 -0700682}
683
Ingo Molnar70b97a72006-07-03 00:25:42 -0700684static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
Nick Piggin4866cde2005-06-25 14:57:23 -0700685{
686}
687
Ingo Molnar70b97a72006-07-03 00:25:42 -0700688static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
Nick Piggin4866cde2005-06-25 14:57:23 -0700689{
Ingo Molnarda04c032005-09-13 11:17:59 +0200690#ifdef CONFIG_DEBUG_SPINLOCK
691 /* this is a valid case when another task releases the spinlock */
692 rq->lock.owner = current;
693#endif
Ingo Molnar8a25d5d2006-07-03 00:24:54 -0700694 /*
695 * If we are tracking spinlock dependencies then we have to
696 * fix up the runqueue lock - which gets 'carried over' from
697 * prev into current:
698 */
699 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
700
Nick Piggin4866cde2005-06-25 14:57:23 -0700701 spin_unlock_irq(&rq->lock);
702}
703
704#else /* __ARCH_WANT_UNLOCKED_CTXSW */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700705static inline int task_running(struct rq *rq, struct task_struct *p)
Nick Piggin4866cde2005-06-25 14:57:23 -0700706{
707#ifdef CONFIG_SMP
708 return p->oncpu;
709#else
Dmitry Adamushko051a1d12007-12-18 15:21:13 +0100710 return task_current(rq, p);
Nick Piggin4866cde2005-06-25 14:57:23 -0700711#endif
712}
713
Ingo Molnar70b97a72006-07-03 00:25:42 -0700714static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
Nick Piggin4866cde2005-06-25 14:57:23 -0700715{
716#ifdef CONFIG_SMP
717 /*
718 * We can optimise this out completely for !SMP, because the
719 * SMP rebalancing from interrupt is the only thing that cares
720 * here.
721 */
722 next->oncpu = 1;
723#endif
724#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
725 spin_unlock_irq(&rq->lock);
726#else
727 spin_unlock(&rq->lock);
728#endif
729}
730
Ingo Molnar70b97a72006-07-03 00:25:42 -0700731static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
Nick Piggin4866cde2005-06-25 14:57:23 -0700732{
733#ifdef CONFIG_SMP
734 /*
735 * After ->oncpu is cleared, the task can be moved to a different CPU.
736 * We must ensure this doesn't happen until the switch is completely
737 * finished.
738 */
739 smp_wmb();
740 prev->oncpu = 0;
741#endif
742#ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
743 local_irq_enable();
744#endif
745}
746#endif /* __ARCH_WANT_UNLOCKED_CTXSW */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
748/*
Ingo Molnarb29739f2006-06-27 02:54:51 -0700749 * __task_rq_lock - lock the runqueue a given task resides on.
750 * Must be called interrupts disabled.
751 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700752static inline struct rq *__task_rq_lock(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -0700753 __acquires(rq->lock)
754{
Andi Kleen3a5c3592007-10-15 17:00:14 +0200755 for (;;) {
756 struct rq *rq = task_rq(p);
757 spin_lock(&rq->lock);
758 if (likely(rq == task_rq(p)))
759 return rq;
Ingo Molnarb29739f2006-06-27 02:54:51 -0700760 spin_unlock(&rq->lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -0700761 }
Ingo Molnarb29739f2006-06-27 02:54:51 -0700762}
763
764/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 * task_rq_lock - lock the runqueue a given task resides on and disable
Ingo Molnar41a2d6c2007-12-05 15:46:09 +0100766 * interrupts. Note the ordering: we can safely lookup the task_rq without
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 * explicitly disabling preemption.
768 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700769static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 __acquires(rq->lock)
771{
Ingo Molnar70b97a72006-07-03 00:25:42 -0700772 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
Andi Kleen3a5c3592007-10-15 17:00:14 +0200774 for (;;) {
775 local_irq_save(*flags);
776 rq = task_rq(p);
777 spin_lock(&rq->lock);
778 if (likely(rq == task_rq(p)))
779 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 spin_unlock_irqrestore(&rq->lock, *flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782}
783
Alexey Dobriyana9957442007-10-15 17:00:13 +0200784static void __task_rq_unlock(struct rq *rq)
Ingo Molnarb29739f2006-06-27 02:54:51 -0700785 __releases(rq->lock)
786{
787 spin_unlock(&rq->lock);
788}
789
Ingo Molnar70b97a72006-07-03 00:25:42 -0700790static inline void task_rq_unlock(struct rq *rq, unsigned long *flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 __releases(rq->lock)
792{
793 spin_unlock_irqrestore(&rq->lock, *flags);
794}
795
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796/*
Robert P. J. Daycc2a73b2006-12-10 02:20:00 -0800797 * this_rq_lock - lock this runqueue and disable interrupts.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 */
Alexey Dobriyana9957442007-10-15 17:00:13 +0200799static struct rq *this_rq_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 __acquires(rq->lock)
801{
Ingo Molnar70b97a72006-07-03 00:25:42 -0700802 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
804 local_irq_disable();
805 rq = this_rq();
806 spin_lock(&rq->lock);
807
808 return rq;
809}
810
Ingo Molnarc24d20d2007-07-09 18:51:59 +0200811/*
Ingo Molnar2aa44d02007-08-23 15:18:02 +0200812 * We are going deep-idle (irqs are disabled):
Ingo Molnar1b9f19c2007-07-09 18:51:59 +0200813 */
Ingo Molnar2aa44d02007-08-23 15:18:02 +0200814void sched_clock_idle_sleep_event(void)
Ingo Molnar1b9f19c2007-07-09 18:51:59 +0200815{
Ingo Molnar2aa44d02007-08-23 15:18:02 +0200816 struct rq *rq = cpu_rq(smp_processor_id());
Ingo Molnar1b9f19c2007-07-09 18:51:59 +0200817
Ingo Molnar2aa44d02007-08-23 15:18:02 +0200818 spin_lock(&rq->lock);
819 __update_rq_clock(rq);
820 spin_unlock(&rq->lock);
821 rq->clock_deep_idle_events++;
Ingo Molnar1b9f19c2007-07-09 18:51:59 +0200822}
Ingo Molnar2aa44d02007-08-23 15:18:02 +0200823EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event);
824
825/*
826 * We just idled delta nanoseconds (called with irqs disabled):
827 */
828void sched_clock_idle_wakeup_event(u64 delta_ns)
829{
830 struct rq *rq = cpu_rq(smp_processor_id());
831 u64 now = sched_clock();
832
833 rq->idle_clock += delta_ns;
834 /*
835 * Override the previous timestamp and ignore all
836 * sched_clock() deltas that occured while we idled,
837 * and use the PM-provided delta_ns to advance the
838 * rq clock:
839 */
840 spin_lock(&rq->lock);
841 rq->prev_clock_raw = now;
842 rq->clock += delta_ns;
843 spin_unlock(&rq->lock);
Guillaume Chazarain782daee2008-01-25 21:08:33 +0100844 touch_softlockup_watchdog();
Ingo Molnar2aa44d02007-08-23 15:18:02 +0200845}
846EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event);
Ingo Molnar1b9f19c2007-07-09 18:51:59 +0200847
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100848static void __resched_task(struct task_struct *p, int tif_bit);
849
850static inline void resched_task(struct task_struct *p)
851{
852 __resched_task(p, TIF_NEED_RESCHED);
853}
854
855#ifdef CONFIG_SCHED_HRTICK
856/*
857 * Use HR-timers to deliver accurate preemption points.
858 *
859 * Its all a bit involved since we cannot program an hrt while holding the
860 * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a
861 * reschedule event.
862 *
863 * When we get rescheduled we reprogram the hrtick_timer outside of the
864 * rq->lock.
865 */
866static inline void resched_hrt(struct task_struct *p)
867{
868 __resched_task(p, TIF_HRTICK_RESCHED);
869}
870
871static inline void resched_rq(struct rq *rq)
872{
873 unsigned long flags;
874
875 spin_lock_irqsave(&rq->lock, flags);
876 resched_task(rq->curr);
877 spin_unlock_irqrestore(&rq->lock, flags);
878}
879
880enum {
881 HRTICK_SET, /* re-programm hrtick_timer */
882 HRTICK_RESET, /* not a new slice */
883};
884
885/*
886 * Use hrtick when:
887 * - enabled by features
888 * - hrtimer is actually high res
889 */
890static inline int hrtick_enabled(struct rq *rq)
891{
892 if (!sched_feat(HRTICK))
893 return 0;
894 return hrtimer_is_hres_active(&rq->hrtick_timer);
895}
896
897/*
898 * Called to set the hrtick timer state.
899 *
900 * called with rq->lock held and irqs disabled
901 */
902static void hrtick_start(struct rq *rq, u64 delay, int reset)
903{
904 assert_spin_locked(&rq->lock);
905
906 /*
907 * preempt at: now + delay
908 */
909 rq->hrtick_expire =
910 ktime_add_ns(rq->hrtick_timer.base->get_time(), delay);
911 /*
912 * indicate we need to program the timer
913 */
914 __set_bit(HRTICK_SET, &rq->hrtick_flags);
915 if (reset)
916 __set_bit(HRTICK_RESET, &rq->hrtick_flags);
917
918 /*
919 * New slices are called from the schedule path and don't need a
920 * forced reschedule.
921 */
922 if (reset)
923 resched_hrt(rq->curr);
924}
925
926static void hrtick_clear(struct rq *rq)
927{
928 if (hrtimer_active(&rq->hrtick_timer))
929 hrtimer_cancel(&rq->hrtick_timer);
930}
931
932/*
933 * Update the timer from the possible pending state.
934 */
935static void hrtick_set(struct rq *rq)
936{
937 ktime_t time;
938 int set, reset;
939 unsigned long flags;
940
941 WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
942
943 spin_lock_irqsave(&rq->lock, flags);
944 set = __test_and_clear_bit(HRTICK_SET, &rq->hrtick_flags);
945 reset = __test_and_clear_bit(HRTICK_RESET, &rq->hrtick_flags);
946 time = rq->hrtick_expire;
947 clear_thread_flag(TIF_HRTICK_RESCHED);
948 spin_unlock_irqrestore(&rq->lock, flags);
949
950 if (set) {
951 hrtimer_start(&rq->hrtick_timer, time, HRTIMER_MODE_ABS);
952 if (reset && !hrtimer_active(&rq->hrtick_timer))
953 resched_rq(rq);
954 } else
955 hrtick_clear(rq);
956}
957
958/*
959 * High-resolution timer tick.
960 * Runs from hardirq context with interrupts disabled.
961 */
962static enum hrtimer_restart hrtick(struct hrtimer *timer)
963{
964 struct rq *rq = container_of(timer, struct rq, hrtick_timer);
965
966 WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
967
968 spin_lock(&rq->lock);
969 __update_rq_clock(rq);
970 rq->curr->sched_class->task_tick(rq, rq->curr, 1);
971 spin_unlock(&rq->lock);
972
973 return HRTIMER_NORESTART;
974}
975
976static inline void init_rq_hrtick(struct rq *rq)
977{
978 rq->hrtick_flags = 0;
979 hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
980 rq->hrtick_timer.function = hrtick;
981 rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ;
982}
983
984void hrtick_resched(void)
985{
986 struct rq *rq;
987 unsigned long flags;
988
989 if (!test_thread_flag(TIF_HRTICK_RESCHED))
990 return;
991
992 local_irq_save(flags);
993 rq = cpu_rq(smp_processor_id());
994 hrtick_set(rq);
995 local_irq_restore(flags);
996}
997#else
998static inline void hrtick_clear(struct rq *rq)
999{
1000}
1001
1002static inline void hrtick_set(struct rq *rq)
1003{
1004}
1005
1006static inline void init_rq_hrtick(struct rq *rq)
1007{
1008}
1009
1010void hrtick_resched(void)
1011{
1012}
1013#endif
1014
Ingo Molnar1b9f19c2007-07-09 18:51:59 +02001015/*
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001016 * resched_task - mark a task 'to be rescheduled now'.
1017 *
1018 * On UP this means the setting of the need_resched flag, on SMP it
1019 * might also involve a cross-CPU call to trigger the scheduler on
1020 * the target CPU.
1021 */
1022#ifdef CONFIG_SMP
1023
1024#ifndef tsk_is_polling
1025#define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG)
1026#endif
1027
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001028static void __resched_task(struct task_struct *p, int tif_bit)
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001029{
1030 int cpu;
1031
1032 assert_spin_locked(&task_rq(p)->lock);
1033
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001034 if (unlikely(test_tsk_thread_flag(p, tif_bit)))
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001035 return;
1036
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001037 set_tsk_thread_flag(p, tif_bit);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001038
1039 cpu = task_cpu(p);
1040 if (cpu == smp_processor_id())
1041 return;
1042
1043 /* NEED_RESCHED must be visible before we test polling */
1044 smp_mb();
1045 if (!tsk_is_polling(p))
1046 smp_send_reschedule(cpu);
1047}
1048
1049static void resched_cpu(int cpu)
1050{
1051 struct rq *rq = cpu_rq(cpu);
1052 unsigned long flags;
1053
1054 if (!spin_trylock_irqsave(&rq->lock, flags))
1055 return;
1056 resched_task(cpu_curr(cpu));
1057 spin_unlock_irqrestore(&rq->lock, flags);
1058}
1059#else
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001060static void __resched_task(struct task_struct *p, int tif_bit)
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001061{
1062 assert_spin_locked(&task_rq(p)->lock);
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001063 set_tsk_thread_flag(p, tif_bit);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001064}
1065#endif
1066
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001067#if BITS_PER_LONG == 32
1068# define WMULT_CONST (~0UL)
1069#else
1070# define WMULT_CONST (1UL << 32)
1071#endif
1072
1073#define WMULT_SHIFT 32
1074
Ingo Molnar194081e2007-08-09 11:16:51 +02001075/*
1076 * Shift right and round:
1077 */
Ingo Molnarcf2ab462007-09-05 14:32:49 +02001078#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
Ingo Molnar194081e2007-08-09 11:16:51 +02001079
Ingo Molnarcb1c4fc2007-08-02 17:41:40 +02001080static unsigned long
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001081calc_delta_mine(unsigned long delta_exec, unsigned long weight,
1082 struct load_weight *lw)
1083{
1084 u64 tmp;
1085
1086 if (unlikely(!lw->inv_weight))
Ingo Molnar194081e2007-08-09 11:16:51 +02001087 lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1;
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001088
1089 tmp = (u64)delta_exec * weight;
1090 /*
1091 * Check whether we'd overflow the 64-bit multiplication:
1092 */
Ingo Molnar194081e2007-08-09 11:16:51 +02001093 if (unlikely(tmp > WMULT_CONST))
Ingo Molnarcf2ab462007-09-05 14:32:49 +02001094 tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
Ingo Molnar194081e2007-08-09 11:16:51 +02001095 WMULT_SHIFT/2);
1096 else
Ingo Molnarcf2ab462007-09-05 14:32:49 +02001097 tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001098
Ingo Molnarecf691d2007-08-02 17:41:40 +02001099 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001100}
1101
1102static inline unsigned long
1103calc_delta_fair(unsigned long delta_exec, struct load_weight *lw)
1104{
1105 return calc_delta_mine(delta_exec, NICE_0_LOAD, lw);
1106}
1107
Ingo Molnar10919852007-10-15 17:00:04 +02001108static inline void update_load_add(struct load_weight *lw, unsigned long inc)
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001109{
1110 lw->weight += inc;
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001111}
1112
Ingo Molnar10919852007-10-15 17:00:04 +02001113static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001114{
1115 lw->weight -= dec;
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001116}
1117
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118/*
Peter Williams2dd73a42006-06-27 02:54:34 -07001119 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1120 * of tasks with abnormal "nice" values across CPUs the contribution that
1121 * each task makes to its run queue's load is weighted according to its
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01001122 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
Peter Williams2dd73a42006-06-27 02:54:34 -07001123 * scaled version of the new time slice allocation that they receive on time
1124 * slice expiry etc.
1125 */
1126
Ingo Molnardd41f592007-07-09 18:51:59 +02001127#define WEIGHT_IDLEPRIO 2
1128#define WMULT_IDLEPRIO (1 << 31)
1129
1130/*
1131 * Nice levels are multiplicative, with a gentle 10% change for every
1132 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1133 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1134 * that remained on nice 0.
1135 *
1136 * The "10% effect" is relative and cumulative: from _any_ nice level,
1137 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
Ingo Molnarf9153ee2007-07-16 09:46:30 +02001138 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1139 * If a task goes up by ~10% and another task goes down by ~10% then
1140 * the relative distance between them is ~25%.)
Ingo Molnardd41f592007-07-09 18:51:59 +02001141 */
1142static const int prio_to_weight[40] = {
Ingo Molnar254753d2007-08-09 11:16:51 +02001143 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1144 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1145 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1146 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1147 /* 0 */ 1024, 820, 655, 526, 423,
1148 /* 5 */ 335, 272, 215, 172, 137,
1149 /* 10 */ 110, 87, 70, 56, 45,
1150 /* 15 */ 36, 29, 23, 18, 15,
Ingo Molnardd41f592007-07-09 18:51:59 +02001151};
1152
Ingo Molnar5714d2d2007-07-16 09:46:31 +02001153/*
1154 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1155 *
1156 * In cases where the weight does not change often, we can use the
1157 * precalculated inverse to speed up arithmetics by turning divisions
1158 * into multiplications:
1159 */
Ingo Molnardd41f592007-07-09 18:51:59 +02001160static const u32 prio_to_wmult[40] = {
Ingo Molnar254753d2007-08-09 11:16:51 +02001161 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1162 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1163 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1164 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1165 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1166 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1167 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1168 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
Ingo Molnardd41f592007-07-09 18:51:59 +02001169};
Peter Williams2dd73a42006-06-27 02:54:34 -07001170
Ingo Molnardd41f592007-07-09 18:51:59 +02001171static void activate_task(struct rq *rq, struct task_struct *p, int wakeup);
1172
1173/*
1174 * runqueue iterator, to support SMP load-balancing between different
1175 * scheduling classes, without having to expose their internal data
1176 * structures to the load-balancing proper:
1177 */
1178struct rq_iterator {
1179 void *arg;
1180 struct task_struct *(*start)(void *);
1181 struct task_struct *(*next)(void *);
1182};
1183
Peter Williamse1d14842007-10-24 18:23:51 +02001184#ifdef CONFIG_SMP
1185static unsigned long
1186balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
1187 unsigned long max_load_move, struct sched_domain *sd,
1188 enum cpu_idle_type idle, int *all_pinned,
1189 int *this_best_prio, struct rq_iterator *iterator);
1190
1191static int
1192iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest,
1193 struct sched_domain *sd, enum cpu_idle_type idle,
1194 struct rq_iterator *iterator);
Peter Williamse1d14842007-10-24 18:23:51 +02001195#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02001196
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01001197#ifdef CONFIG_CGROUP_CPUACCT
1198static void cpuacct_charge(struct task_struct *tsk, u64 cputime);
1199#else
1200static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
1201#endif
1202
Gregory Haskinse7693a32008-01-25 21:08:09 +01001203#ifdef CONFIG_SMP
1204static unsigned long source_load(int cpu, int type);
1205static unsigned long target_load(int cpu, int type);
1206static unsigned long cpu_avg_load_per_task(int cpu);
1207static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd);
1208#endif /* CONFIG_SMP */
1209
Ingo Molnardd41f592007-07-09 18:51:59 +02001210#include "sched_stats.h"
Ingo Molnardd41f592007-07-09 18:51:59 +02001211#include "sched_idletask.c"
Ingo Molnar5522d5d2007-10-15 17:00:12 +02001212#include "sched_fair.c"
1213#include "sched_rt.c"
Ingo Molnardd41f592007-07-09 18:51:59 +02001214#ifdef CONFIG_SCHED_DEBUG
1215# include "sched_debug.c"
1216#endif
1217
1218#define sched_class_highest (&rt_sched_class)
1219
Peter Zijlstra62fb1852008-02-25 17:34:02 +01001220static inline void inc_load(struct rq *rq, const struct task_struct *p)
Ingo Molnar9c217242007-08-02 17:41:40 +02001221{
Peter Zijlstra62fb1852008-02-25 17:34:02 +01001222 update_load_add(&rq->load, p->se.load.weight);
Ingo Molnar9c217242007-08-02 17:41:40 +02001223}
1224
Peter Zijlstra62fb1852008-02-25 17:34:02 +01001225static inline void dec_load(struct rq *rq, const struct task_struct *p)
1226{
1227 update_load_sub(&rq->load, p->se.load.weight);
1228}
1229
1230static void inc_nr_running(struct task_struct *p, struct rq *rq)
1231{
1232 rq->nr_running++;
1233 inc_load(rq, p);
1234}
1235
1236static void dec_nr_running(struct task_struct *p, struct rq *rq)
Ingo Molnar9c217242007-08-02 17:41:40 +02001237{
1238 rq->nr_running--;
Peter Zijlstra62fb1852008-02-25 17:34:02 +01001239 dec_load(rq, p);
Ingo Molnar9c217242007-08-02 17:41:40 +02001240}
1241
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001242static void set_load_weight(struct task_struct *p)
1243{
1244 if (task_has_rt_policy(p)) {
Ingo Molnardd41f592007-07-09 18:51:59 +02001245 p->se.load.weight = prio_to_weight[0] * 2;
1246 p->se.load.inv_weight = prio_to_wmult[0] >> 1;
1247 return;
1248 }
1249
1250 /*
1251 * SCHED_IDLE tasks get minimal weight:
1252 */
1253 if (p->policy == SCHED_IDLE) {
1254 p->se.load.weight = WEIGHT_IDLEPRIO;
1255 p->se.load.inv_weight = WMULT_IDLEPRIO;
1256 return;
1257 }
1258
1259 p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO];
1260 p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO];
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001261}
1262
Ingo Molnar8159f872007-08-09 11:16:49 +02001263static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup)
Ingo Molnar71f8bd42007-07-09 18:51:59 +02001264{
1265 sched_info_queued(p);
Ingo Molnarfd390f62007-08-09 11:16:48 +02001266 p->sched_class->enqueue_task(rq, p, wakeup);
Ingo Molnardd41f592007-07-09 18:51:59 +02001267 p->se.on_rq = 1;
1268}
1269
Ingo Molnar69be72c2007-08-09 11:16:49 +02001270static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep)
Ingo Molnardd41f592007-07-09 18:51:59 +02001271{
Ingo Molnarf02231e2007-08-09 11:16:48 +02001272 p->sched_class->dequeue_task(rq, p, sleep);
Ingo Molnardd41f592007-07-09 18:51:59 +02001273 p->se.on_rq = 0;
Ingo Molnar71f8bd42007-07-09 18:51:59 +02001274}
1275
1276/*
Ingo Molnardd41f592007-07-09 18:51:59 +02001277 * __normal_prio - return the priority that is based on the static prio
Ingo Molnar71f8bd42007-07-09 18:51:59 +02001278 */
Ingo Molnar14531182007-07-09 18:51:59 +02001279static inline int __normal_prio(struct task_struct *p)
1280{
Ingo Molnardd41f592007-07-09 18:51:59 +02001281 return p->static_prio;
Ingo Molnar14531182007-07-09 18:51:59 +02001282}
1283
1284/*
Ingo Molnarb29739f2006-06-27 02:54:51 -07001285 * Calculate the expected normal priority: i.e. priority
1286 * without taking RT-inheritance into account. Might be
1287 * boosted by interactivity modifiers. Changes upon fork,
1288 * setprio syscalls, and whenever the interactivity
1289 * estimator recalculates.
1290 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001291static inline int normal_prio(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -07001292{
1293 int prio;
1294
Ingo Molnare05606d2007-07-09 18:51:59 +02001295 if (task_has_rt_policy(p))
Ingo Molnarb29739f2006-06-27 02:54:51 -07001296 prio = MAX_RT_PRIO-1 - p->rt_priority;
1297 else
1298 prio = __normal_prio(p);
1299 return prio;
1300}
1301
1302/*
1303 * Calculate the current priority, i.e. the priority
1304 * taken into account by the scheduler. This value might
1305 * be boosted by RT tasks, or might be boosted by
1306 * interactivity modifiers. Will be RT if the task got
1307 * RT-boosted. If not then it returns p->normal_prio.
1308 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001309static int effective_prio(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -07001310{
1311 p->normal_prio = normal_prio(p);
1312 /*
1313 * If we are RT tasks or we were boosted to RT priority,
1314 * keep the priority unchanged. Otherwise, update priority
1315 * to the normal priority:
1316 */
1317 if (!rt_prio(p->prio))
1318 return p->normal_prio;
1319 return p->prio;
1320}
1321
1322/*
Ingo Molnardd41f592007-07-09 18:51:59 +02001323 * activate_task - move a task to the runqueue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 */
Ingo Molnardd41f592007-07-09 18:51:59 +02001325static void activate_task(struct rq *rq, struct task_struct *p, int wakeup)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326{
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05001327 if (task_contributes_to_load(p))
Ingo Molnardd41f592007-07-09 18:51:59 +02001328 rq->nr_uninterruptible--;
1329
Ingo Molnar8159f872007-08-09 11:16:49 +02001330 enqueue_task(rq, p, wakeup);
Peter Zijlstra62fb1852008-02-25 17:34:02 +01001331 inc_nr_running(p, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332}
1333
1334/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 * deactivate_task - remove a task from the runqueue.
1336 */
Ingo Molnar2e1cb742007-08-09 11:16:49 +02001337static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338{
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05001339 if (task_contributes_to_load(p))
Ingo Molnardd41f592007-07-09 18:51:59 +02001340 rq->nr_uninterruptible++;
1341
Ingo Molnar69be72c2007-08-09 11:16:49 +02001342 dequeue_task(rq, p, sleep);
Peter Zijlstra62fb1852008-02-25 17:34:02 +01001343 dec_nr_running(p, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344}
1345
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346/**
1347 * task_curr - is this task currently executing on a CPU?
1348 * @p: the task in question.
1349 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001350inline int task_curr(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351{
1352 return cpu_curr(task_cpu(p)) == p;
1353}
1354
Peter Williams2dd73a42006-06-27 02:54:34 -07001355/* Used instead of source_load when we know the type == 0 */
1356unsigned long weighted_cpuload(const int cpu)
1357{
Dmitry Adamushko495eca42007-10-15 17:00:06 +02001358 return cpu_rq(cpu)->load.weight;
Ingo Molnardd41f592007-07-09 18:51:59 +02001359}
1360
1361static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1362{
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001363 set_task_rq(p, cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02001364#ifdef CONFIG_SMP
Dmitry Adamushkoce96b5a2007-11-15 20:57:40 +01001365 /*
1366 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1367 * successfuly executed on another CPU. We must ensure that updates of
1368 * per-task data have been completed by this moment.
1369 */
1370 smp_wmb();
Ingo Molnardd41f592007-07-09 18:51:59 +02001371 task_thread_info(p)->cpu = cpu;
Ingo Molnardd41f592007-07-09 18:51:59 +02001372#endif
Peter Williams2dd73a42006-06-27 02:54:34 -07001373}
1374
Steven Rostedtcb469842008-01-25 21:08:22 +01001375static inline void check_class_changed(struct rq *rq, struct task_struct *p,
1376 const struct sched_class *prev_class,
1377 int oldprio, int running)
1378{
1379 if (prev_class != p->sched_class) {
1380 if (prev_class->switched_from)
1381 prev_class->switched_from(rq, p, running);
1382 p->sched_class->switched_to(rq, p, running);
1383 } else
1384 p->sched_class->prio_changed(rq, p, oldprio, running);
1385}
1386
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387#ifdef CONFIG_SMP
Ingo Molnarc65cc872007-07-09 18:51:58 +02001388
Ingo Molnarcc367732007-10-15 17:00:18 +02001389/*
1390 * Is this task likely cache-hot:
1391 */
Gregory Haskinse7693a32008-01-25 21:08:09 +01001392static int
Ingo Molnarcc367732007-10-15 17:00:18 +02001393task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
1394{
1395 s64 delta;
1396
1397 if (p->sched_class != &fair_sched_class)
1398 return 0;
1399
Ingo Molnar6bc16652007-10-15 17:00:18 +02001400 if (sysctl_sched_migration_cost == -1)
1401 return 1;
1402 if (sysctl_sched_migration_cost == 0)
1403 return 0;
1404
Ingo Molnarcc367732007-10-15 17:00:18 +02001405 delta = now - p->se.exec_start;
1406
1407 return delta < (s64)sysctl_sched_migration_cost;
1408}
1409
1410
Ingo Molnardd41f592007-07-09 18:51:59 +02001411void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
Ingo Molnarc65cc872007-07-09 18:51:58 +02001412{
Ingo Molnardd41f592007-07-09 18:51:59 +02001413 int old_cpu = task_cpu(p);
1414 struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu);
Srivatsa Vaddagiri2830cf82007-10-15 17:00:12 +02001415 struct cfs_rq *old_cfsrq = task_cfs_rq(p),
1416 *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu);
Ingo Molnarbbdba7c2007-10-15 17:00:06 +02001417 u64 clock_offset;
Ingo Molnardd41f592007-07-09 18:51:59 +02001418
1419 clock_offset = old_rq->clock - new_rq->clock;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02001420
1421#ifdef CONFIG_SCHEDSTATS
1422 if (p->se.wait_start)
1423 p->se.wait_start -= clock_offset;
Ingo Molnardd41f592007-07-09 18:51:59 +02001424 if (p->se.sleep_start)
1425 p->se.sleep_start -= clock_offset;
1426 if (p->se.block_start)
1427 p->se.block_start -= clock_offset;
Ingo Molnarcc367732007-10-15 17:00:18 +02001428 if (old_cpu != new_cpu) {
1429 schedstat_inc(p, se.nr_migrations);
1430 if (task_hot(p, old_rq->clock, NULL))
1431 schedstat_inc(p, se.nr_forced2_migrations);
1432 }
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02001433#endif
Srivatsa Vaddagiri2830cf82007-10-15 17:00:12 +02001434 p->se.vruntime -= old_cfsrq->min_vruntime -
1435 new_cfsrq->min_vruntime;
Ingo Molnardd41f592007-07-09 18:51:59 +02001436
1437 __set_task_cpu(p, new_cpu);
Ingo Molnarc65cc872007-07-09 18:51:58 +02001438}
1439
Ingo Molnar70b97a72006-07-03 00:25:42 -07001440struct migration_req {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 struct list_head list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
Ingo Molnar36c8b582006-07-03 00:25:41 -07001443 struct task_struct *task;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 int dest_cpu;
1445
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 struct completion done;
Ingo Molnar70b97a72006-07-03 00:25:42 -07001447};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
1449/*
1450 * The task's runqueue lock must be held.
1451 * Returns true if you have to wait for migration thread.
1452 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001453static int
Ingo Molnar70b97a72006-07-03 00:25:42 -07001454migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455{
Ingo Molnar70b97a72006-07-03 00:25:42 -07001456 struct rq *rq = task_rq(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
1458 /*
1459 * If the task is not on a runqueue (and not running), then
1460 * it is sufficient to simply update the task's cpu field.
1461 */
Ingo Molnardd41f592007-07-09 18:51:59 +02001462 if (!p->se.on_rq && !task_running(rq, p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 set_task_cpu(p, dest_cpu);
1464 return 0;
1465 }
1466
1467 init_completion(&req->done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 req->task = p;
1469 req->dest_cpu = dest_cpu;
1470 list_add(&req->list, &rq->migration_queue);
Ingo Molnar48f24c42006-07-03 00:25:40 -07001471
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 return 1;
1473}
1474
1475/*
1476 * wait_task_inactive - wait for a thread to unschedule.
1477 *
1478 * The caller must ensure that the task *will* unschedule sometime soon,
1479 * else this function might spin for a *long* time. This function can't
1480 * be called with interrupts off, or it may introduce deadlock with
1481 * smp_call_function() if an IPI is sent by the same process we are
1482 * waiting to become inactive.
1483 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001484void wait_task_inactive(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485{
1486 unsigned long flags;
Ingo Molnardd41f592007-07-09 18:51:59 +02001487 int running, on_rq;
Ingo Molnar70b97a72006-07-03 00:25:42 -07001488 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
Andi Kleen3a5c3592007-10-15 17:00:14 +02001490 for (;;) {
1491 /*
1492 * We do the initial early heuristics without holding
1493 * any task-queue locks at all. We'll only try to get
1494 * the runqueue lock when things look like they will
1495 * work out!
1496 */
1497 rq = task_rq(p);
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07001498
Andi Kleen3a5c3592007-10-15 17:00:14 +02001499 /*
1500 * If the task is actively running on another CPU
1501 * still, just relax and busy-wait without holding
1502 * any locks.
1503 *
1504 * NOTE! Since we don't hold any locks, it's not
1505 * even sure that "rq" stays as the right runqueue!
1506 * But we don't care, since "task_running()" will
1507 * return false if the runqueue has changed and p
1508 * is actually now running somewhere else!
1509 */
1510 while (task_running(rq, p))
1511 cpu_relax();
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07001512
Andi Kleen3a5c3592007-10-15 17:00:14 +02001513 /*
1514 * Ok, time to look more closely! We need the rq
1515 * lock now, to be *sure*. If we're wrong, we'll
1516 * just go back and repeat.
1517 */
1518 rq = task_rq_lock(p, &flags);
1519 running = task_running(rq, p);
1520 on_rq = p->se.on_rq;
1521 task_rq_unlock(rq, &flags);
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07001522
Andi Kleen3a5c3592007-10-15 17:00:14 +02001523 /*
1524 * Was it really running after all now that we
1525 * checked with the proper locks actually held?
1526 *
1527 * Oops. Go back and try again..
1528 */
1529 if (unlikely(running)) {
1530 cpu_relax();
1531 continue;
1532 }
1533
1534 /*
1535 * It's not enough that it's not actively running,
1536 * it must be off the runqueue _entirely_, and not
1537 * preempted!
1538 *
1539 * So if it wa still runnable (but just not actively
1540 * running right now), it's preempted, and we should
1541 * yield - it could be a while.
1542 */
1543 if (unlikely(on_rq)) {
1544 schedule_timeout_uninterruptible(1);
1545 continue;
1546 }
1547
1548 /*
1549 * Ahh, all good. It wasn't running, and it wasn't
1550 * runnable, which means that it will never become
1551 * running in the future either. We're all done!
1552 */
1553 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555}
1556
1557/***
1558 * kick_process - kick a running thread to enter/exit the kernel
1559 * @p: the to-be-kicked thread
1560 *
1561 * Cause a process which is running on another CPU to enter
1562 * kernel-mode, without any delay. (to get signals handled.)
1563 *
1564 * NOTE: this function doesnt have to take the runqueue lock,
1565 * because all it wants to ensure is that the remote task enters
1566 * the kernel. If the IPI races and the task has been migrated
1567 * to another CPU then no harm is done and the purpose has been
1568 * achieved as well.
1569 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001570void kick_process(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571{
1572 int cpu;
1573
1574 preempt_disable();
1575 cpu = task_cpu(p);
1576 if ((cpu != smp_processor_id()) && task_curr(p))
1577 smp_send_reschedule(cpu);
1578 preempt_enable();
1579}
1580
1581/*
Peter Williams2dd73a42006-06-27 02:54:34 -07001582 * Return a low guess at the load of a migration-source cpu weighted
1583 * according to the scheduling class and "nice" value.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 *
1585 * We want to under-estimate the load of migration sources, to
1586 * balance conservatively.
1587 */
Alexey Dobriyana9957442007-10-15 17:00:13 +02001588static unsigned long source_load(int cpu, int type)
Con Kolivasb9104722005-11-08 21:38:55 -08001589{
Ingo Molnar70b97a72006-07-03 00:25:42 -07001590 struct rq *rq = cpu_rq(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02001591 unsigned long total = weighted_cpuload(cpu);
Nick Piggina2000572006-02-10 01:51:02 -08001592
Peter Williams2dd73a42006-06-27 02:54:34 -07001593 if (type == 0)
Ingo Molnardd41f592007-07-09 18:51:59 +02001594 return total;
Peter Williams2dd73a42006-06-27 02:54:34 -07001595
Ingo Molnardd41f592007-07-09 18:51:59 +02001596 return min(rq->cpu_load[type-1], total);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597}
1598
1599/*
Peter Williams2dd73a42006-06-27 02:54:34 -07001600 * Return a high guess at the load of a migration-target cpu weighted
1601 * according to the scheduling class and "nice" value.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 */
Alexey Dobriyana9957442007-10-15 17:00:13 +02001603static unsigned long target_load(int cpu, int type)
Con Kolivasb9104722005-11-08 21:38:55 -08001604{
Ingo Molnar70b97a72006-07-03 00:25:42 -07001605 struct rq *rq = cpu_rq(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02001606 unsigned long total = weighted_cpuload(cpu);
Nick Piggina2000572006-02-10 01:51:02 -08001607
Peter Williams2dd73a42006-06-27 02:54:34 -07001608 if (type == 0)
Ingo Molnardd41f592007-07-09 18:51:59 +02001609 return total;
Peter Williams2dd73a42006-06-27 02:54:34 -07001610
Ingo Molnardd41f592007-07-09 18:51:59 +02001611 return max(rq->cpu_load[type-1], total);
Peter Williams2dd73a42006-06-27 02:54:34 -07001612}
1613
1614/*
1615 * Return the average load per task on the cpu's run queue
1616 */
Gregory Haskinse7693a32008-01-25 21:08:09 +01001617static unsigned long cpu_avg_load_per_task(int cpu)
Peter Williams2dd73a42006-06-27 02:54:34 -07001618{
Ingo Molnar70b97a72006-07-03 00:25:42 -07001619 struct rq *rq = cpu_rq(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02001620 unsigned long total = weighted_cpuload(cpu);
Peter Williams2dd73a42006-06-27 02:54:34 -07001621 unsigned long n = rq->nr_running;
1622
Ingo Molnardd41f592007-07-09 18:51:59 +02001623 return n ? total / n : SCHED_LOAD_SCALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624}
1625
Nick Piggin147cbb42005-06-25 14:57:19 -07001626/*
1627 * find_idlest_group finds and returns the least busy CPU group within the
1628 * domain.
1629 */
1630static struct sched_group *
1631find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu)
1632{
1633 struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups;
1634 unsigned long min_load = ULONG_MAX, this_load = 0;
1635 int load_idx = sd->forkexec_idx;
1636 int imbalance = 100 + (sd->imbalance_pct-100)/2;
1637
1638 do {
1639 unsigned long load, avg_load;
1640 int local_group;
1641 int i;
1642
M.Baris Demirayda5a5522005-09-10 00:26:09 -07001643 /* Skip over this group if it has no CPUs allowed */
1644 if (!cpus_intersects(group->cpumask, p->cpus_allowed))
Andi Kleen3a5c3592007-10-15 17:00:14 +02001645 continue;
M.Baris Demirayda5a5522005-09-10 00:26:09 -07001646
Nick Piggin147cbb42005-06-25 14:57:19 -07001647 local_group = cpu_isset(this_cpu, group->cpumask);
Nick Piggin147cbb42005-06-25 14:57:19 -07001648
1649 /* Tally up the load of all CPUs in the group */
1650 avg_load = 0;
1651
1652 for_each_cpu_mask(i, group->cpumask) {
1653 /* Bias balancing toward cpus of our domain */
1654 if (local_group)
1655 load = source_load(i, load_idx);
1656 else
1657 load = target_load(i, load_idx);
1658
1659 avg_load += load;
1660 }
1661
1662 /* Adjust by relative CPU power of the group */
Eric Dumazet5517d862007-05-08 00:32:57 -07001663 avg_load = sg_div_cpu_power(group,
1664 avg_load * SCHED_LOAD_SCALE);
Nick Piggin147cbb42005-06-25 14:57:19 -07001665
1666 if (local_group) {
1667 this_load = avg_load;
1668 this = group;
1669 } else if (avg_load < min_load) {
1670 min_load = avg_load;
1671 idlest = group;
1672 }
Andi Kleen3a5c3592007-10-15 17:00:14 +02001673 } while (group = group->next, group != sd->groups);
Nick Piggin147cbb42005-06-25 14:57:19 -07001674
1675 if (!idlest || 100*this_load < imbalance*min_load)
1676 return NULL;
1677 return idlest;
1678}
1679
1680/*
Satoru Takeuchi0feaece2006-10-03 01:14:10 -07001681 * find_idlest_cpu - find the idlest cpu among the cpus in group.
Nick Piggin147cbb42005-06-25 14:57:19 -07001682 */
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07001683static int
1684find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
Nick Piggin147cbb42005-06-25 14:57:19 -07001685{
M.Baris Demirayda5a5522005-09-10 00:26:09 -07001686 cpumask_t tmp;
Nick Piggin147cbb42005-06-25 14:57:19 -07001687 unsigned long load, min_load = ULONG_MAX;
1688 int idlest = -1;
1689 int i;
1690
M.Baris Demirayda5a5522005-09-10 00:26:09 -07001691 /* Traverse only the allowed CPUs */
1692 cpus_and(tmp, group->cpumask, p->cpus_allowed);
1693
1694 for_each_cpu_mask(i, tmp) {
Peter Williams2dd73a42006-06-27 02:54:34 -07001695 load = weighted_cpuload(i);
Nick Piggin147cbb42005-06-25 14:57:19 -07001696
1697 if (load < min_load || (load == min_load && i == this_cpu)) {
1698 min_load = load;
1699 idlest = i;
1700 }
1701 }
1702
1703 return idlest;
1704}
1705
Nick Piggin476d1392005-06-25 14:57:29 -07001706/*
1707 * sched_balance_self: balance the current task (running on cpu) in domains
1708 * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and
1709 * SD_BALANCE_EXEC.
1710 *
1711 * Balance, ie. select the least loaded group.
1712 *
1713 * Returns the target CPU number, or the same CPU if no balancing is needed.
1714 *
1715 * preempt must be disabled.
1716 */
1717static int sched_balance_self(int cpu, int flag)
1718{
1719 struct task_struct *t = current;
1720 struct sched_domain *tmp, *sd = NULL;
Nick Piggin147cbb42005-06-25 14:57:19 -07001721
Chen, Kenneth Wc96d1452006-06-27 02:54:28 -07001722 for_each_domain(cpu, tmp) {
Ingo Molnar9761eea2007-07-09 18:52:00 +02001723 /*
1724 * If power savings logic is enabled for a domain, stop there.
1725 */
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07001726 if (tmp->flags & SD_POWERSAVINGS_BALANCE)
1727 break;
Nick Piggin476d1392005-06-25 14:57:29 -07001728 if (tmp->flags & flag)
1729 sd = tmp;
Chen, Kenneth Wc96d1452006-06-27 02:54:28 -07001730 }
Nick Piggin476d1392005-06-25 14:57:29 -07001731
1732 while (sd) {
1733 cpumask_t span;
1734 struct sched_group *group;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07001735 int new_cpu, weight;
1736
1737 if (!(sd->flags & flag)) {
1738 sd = sd->child;
1739 continue;
1740 }
Nick Piggin476d1392005-06-25 14:57:29 -07001741
1742 span = sd->span;
1743 group = find_idlest_group(sd, t, cpu);
Siddha, Suresh B1a848872006-10-03 01:14:08 -07001744 if (!group) {
1745 sd = sd->child;
1746 continue;
1747 }
Nick Piggin476d1392005-06-25 14:57:29 -07001748
M.Baris Demirayda5a5522005-09-10 00:26:09 -07001749 new_cpu = find_idlest_cpu(group, t, cpu);
Siddha, Suresh B1a848872006-10-03 01:14:08 -07001750 if (new_cpu == -1 || new_cpu == cpu) {
1751 /* Now try balancing at a lower domain level of cpu */
1752 sd = sd->child;
1753 continue;
1754 }
Nick Piggin476d1392005-06-25 14:57:29 -07001755
Siddha, Suresh B1a848872006-10-03 01:14:08 -07001756 /* Now try balancing at a lower domain level of new_cpu */
Nick Piggin476d1392005-06-25 14:57:29 -07001757 cpu = new_cpu;
Nick Piggin476d1392005-06-25 14:57:29 -07001758 sd = NULL;
1759 weight = cpus_weight(span);
1760 for_each_domain(cpu, tmp) {
1761 if (weight <= cpus_weight(tmp->span))
1762 break;
1763 if (tmp->flags & flag)
1764 sd = tmp;
1765 }
1766 /* while loop will break here if sd == NULL */
1767 }
1768
1769 return cpu;
1770}
1771
1772#endif /* CONFIG_SMP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774/***
1775 * try_to_wake_up - wake up a thread
1776 * @p: the to-be-woken-up thread
1777 * @state: the mask of task states that can be woken
1778 * @sync: do a synchronous wakeup?
1779 *
1780 * Put it on the run-queue if it's not already there. The "current"
1781 * thread is always on the run-queue (except when the actual
1782 * re-schedule is in progress), and as such you're allowed to do
1783 * the simpler "current->state = TASK_RUNNING" to mark yourself
1784 * runnable without the overhead of this.
1785 *
1786 * returns failure only if the task is already active.
1787 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001788static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789{
Ingo Molnarcc367732007-10-15 17:00:18 +02001790 int cpu, orig_cpu, this_cpu, success = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 unsigned long flags;
1792 long old_state;
Ingo Molnar70b97a72006-07-03 00:25:42 -07001793 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794
Linus Torvalds04e2f172008-02-23 18:05:03 -08001795 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 rq = task_rq_lock(p, &flags);
1797 old_state = p->state;
1798 if (!(old_state & state))
1799 goto out;
1800
Ingo Molnardd41f592007-07-09 18:51:59 +02001801 if (p->se.on_rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 goto out_running;
1803
1804 cpu = task_cpu(p);
Ingo Molnarcc367732007-10-15 17:00:18 +02001805 orig_cpu = cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 this_cpu = smp_processor_id();
1807
1808#ifdef CONFIG_SMP
1809 if (unlikely(task_running(rq, p)))
1810 goto out_activate;
1811
Dmitry Adamushko5d2f5a62008-01-25 21:08:21 +01001812 cpu = p->sched_class->select_task_rq(p, sync);
1813 if (cpu != orig_cpu) {
1814 set_task_cpu(p, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 task_rq_unlock(rq, &flags);
1816 /* might preempt at this point */
1817 rq = task_rq_lock(p, &flags);
1818 old_state = p->state;
1819 if (!(old_state & state))
1820 goto out;
Ingo Molnardd41f592007-07-09 18:51:59 +02001821 if (p->se.on_rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 goto out_running;
1823
1824 this_cpu = smp_processor_id();
1825 cpu = task_cpu(p);
1826 }
1827
Gregory Haskinse7693a32008-01-25 21:08:09 +01001828#ifdef CONFIG_SCHEDSTATS
1829 schedstat_inc(rq, ttwu_count);
1830 if (cpu == this_cpu)
1831 schedstat_inc(rq, ttwu_local);
1832 else {
1833 struct sched_domain *sd;
1834 for_each_domain(this_cpu, sd) {
1835 if (cpu_isset(cpu, sd->span)) {
1836 schedstat_inc(sd, ttwu_wake_remote);
1837 break;
1838 }
1839 }
1840 }
Gregory Haskinse7693a32008-01-25 21:08:09 +01001841#endif
1842
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843out_activate:
1844#endif /* CONFIG_SMP */
Ingo Molnarcc367732007-10-15 17:00:18 +02001845 schedstat_inc(p, se.nr_wakeups);
1846 if (sync)
1847 schedstat_inc(p, se.nr_wakeups_sync);
1848 if (orig_cpu != cpu)
1849 schedstat_inc(p, se.nr_wakeups_migrate);
1850 if (cpu == this_cpu)
1851 schedstat_inc(p, se.nr_wakeups_local);
1852 else
1853 schedstat_inc(p, se.nr_wakeups_remote);
Ingo Molnar2daa3572007-08-09 11:16:51 +02001854 update_rq_clock(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02001855 activate_task(rq, p, 1);
Ingo Molnar9c63d9c2007-10-15 17:00:20 +02001856 check_preempt_curr(rq, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 success = 1;
1858
1859out_running:
1860 p->state = TASK_RUNNING;
Steven Rostedt9a897c52008-01-25 21:08:22 +01001861#ifdef CONFIG_SMP
1862 if (p->sched_class->task_wake_up)
1863 p->sched_class->task_wake_up(rq, p);
1864#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865out:
1866 task_rq_unlock(rq, &flags);
1867
1868 return success;
1869}
1870
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08001871int wake_up_process(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872{
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05001873 return try_to_wake_up(p, TASK_ALL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875EXPORT_SYMBOL(wake_up_process);
1876
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08001877int wake_up_state(struct task_struct *p, unsigned int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878{
1879 return try_to_wake_up(p, state, 0);
1880}
1881
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882/*
1883 * Perform scheduler related setup for a newly forked process p.
1884 * p is forked by current.
Ingo Molnardd41f592007-07-09 18:51:59 +02001885 *
1886 * __sched_fork() is basic setup used by init_idle() too:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 */
Ingo Molnardd41f592007-07-09 18:51:59 +02001888static void __sched_fork(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889{
Ingo Molnardd41f592007-07-09 18:51:59 +02001890 p->se.exec_start = 0;
1891 p->se.sum_exec_runtime = 0;
Ingo Molnarf6cf8912007-08-28 12:53:24 +02001892 p->se.prev_sum_exec_runtime = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02001893
1894#ifdef CONFIG_SCHEDSTATS
1895 p->se.wait_start = 0;
Ingo Molnardd41f592007-07-09 18:51:59 +02001896 p->se.sum_sleep_runtime = 0;
1897 p->se.sleep_start = 0;
Ingo Molnardd41f592007-07-09 18:51:59 +02001898 p->se.block_start = 0;
1899 p->se.sleep_max = 0;
1900 p->se.block_max = 0;
1901 p->se.exec_max = 0;
Ingo Molnareba1ed42007-10-15 17:00:02 +02001902 p->se.slice_max = 0;
Ingo Molnardd41f592007-07-09 18:51:59 +02001903 p->se.wait_max = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02001904#endif
Nick Piggin476d1392005-06-25 14:57:29 -07001905
Peter Zijlstrafa717062008-01-25 21:08:27 +01001906 INIT_LIST_HEAD(&p->rt.run_list);
Ingo Molnardd41f592007-07-09 18:51:59 +02001907 p->se.on_rq = 0;
Nick Piggin476d1392005-06-25 14:57:29 -07001908
Avi Kivitye107be32007-07-26 13:40:43 +02001909#ifdef CONFIG_PREEMPT_NOTIFIERS
1910 INIT_HLIST_HEAD(&p->preempt_notifiers);
1911#endif
1912
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 /*
1914 * We mark the process as running here, but have not actually
1915 * inserted it onto the runqueue yet. This guarantees that
1916 * nobody will actually run it, and a signal or other external
1917 * event cannot wake it up and insert it on the runqueue either.
1918 */
1919 p->state = TASK_RUNNING;
Ingo Molnardd41f592007-07-09 18:51:59 +02001920}
1921
1922/*
1923 * fork()/clone()-time setup:
1924 */
1925void sched_fork(struct task_struct *p, int clone_flags)
1926{
1927 int cpu = get_cpu();
1928
1929 __sched_fork(p);
1930
1931#ifdef CONFIG_SMP
1932 cpu = sched_balance_self(cpu, SD_BALANCE_FORK);
1933#endif
Ingo Molnar02e4bac2007-10-15 17:00:11 +02001934 set_task_cpu(p, cpu);
Ingo Molnarb29739f2006-06-27 02:54:51 -07001935
1936 /*
1937 * Make sure we do not leak PI boosting priority to the child:
1938 */
1939 p->prio = current->normal_prio;
Hiroshi Shimamoto2ddbf952007-10-15 17:00:11 +02001940 if (!rt_prio(p->prio))
1941 p->sched_class = &fair_sched_class;
Ingo Molnarb29739f2006-06-27 02:54:51 -07001942
Chandra Seetharaman52f17b62006-07-14 00:24:38 -07001943#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
Ingo Molnardd41f592007-07-09 18:51:59 +02001944 if (likely(sched_info_on()))
Chandra Seetharaman52f17b62006-07-14 00:24:38 -07001945 memset(&p->sched_info, 0, sizeof(p->sched_info));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946#endif
Chen, Kenneth Wd6077cb2006-02-14 13:53:10 -08001947#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
Nick Piggin4866cde2005-06-25 14:57:23 -07001948 p->oncpu = 0;
1949#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950#ifdef CONFIG_PREEMPT
Nick Piggin4866cde2005-06-25 14:57:23 -07001951 /* Want to start with kernel preemption disabled. */
Al Viroa1261f52005-11-13 16:06:55 -08001952 task_thread_info(p)->preempt_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953#endif
Nick Piggin476d1392005-06-25 14:57:29 -07001954 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955}
1956
1957/*
1958 * wake_up_new_task - wake up a newly created task for the first time.
1959 *
1960 * This function will do some initial scheduler statistics housekeeping
1961 * that must be done for every newly created context, then puts the task
1962 * on the runqueue and wakes it.
1963 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08001964void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965{
1966 unsigned long flags;
Ingo Molnardd41f592007-07-09 18:51:59 +02001967 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968
1969 rq = task_rq_lock(p, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 BUG_ON(p->state != TASK_RUNNING);
Ingo Molnara8e504d2007-08-09 11:16:47 +02001971 update_rq_clock(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972
1973 p->prio = effective_prio(p);
1974
Srivatsa Vaddagirib9dca1e2007-10-17 16:55:11 +02001975 if (!p->sched_class->task_new || !current->se.on_rq) {
Ingo Molnardd41f592007-07-09 18:51:59 +02001976 activate_task(rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 /*
Ingo Molnardd41f592007-07-09 18:51:59 +02001979 * Let the scheduling class do new task startup
1980 * management (if any):
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 */
Ingo Molnaree0827d2007-08-09 11:16:49 +02001982 p->sched_class->task_new(rq, p);
Peter Zijlstra62fb1852008-02-25 17:34:02 +01001983 inc_nr_running(p, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 }
Ingo Molnardd41f592007-07-09 18:51:59 +02001985 check_preempt_curr(rq, p);
Steven Rostedt9a897c52008-01-25 21:08:22 +01001986#ifdef CONFIG_SMP
1987 if (p->sched_class->task_wake_up)
1988 p->sched_class->task_wake_up(rq, p);
1989#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02001990 task_rq_unlock(rq, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991}
1992
Avi Kivitye107be32007-07-26 13:40:43 +02001993#ifdef CONFIG_PREEMPT_NOTIFIERS
1994
1995/**
Randy Dunlap421cee22007-07-31 00:37:50 -07001996 * preempt_notifier_register - tell me when current is being being preempted & rescheduled
1997 * @notifier: notifier struct to register
Avi Kivitye107be32007-07-26 13:40:43 +02001998 */
1999void preempt_notifier_register(struct preempt_notifier *notifier)
2000{
2001 hlist_add_head(&notifier->link, &current->preempt_notifiers);
2002}
2003EXPORT_SYMBOL_GPL(preempt_notifier_register);
2004
2005/**
2006 * preempt_notifier_unregister - no longer interested in preemption notifications
Randy Dunlap421cee22007-07-31 00:37:50 -07002007 * @notifier: notifier struct to unregister
Avi Kivitye107be32007-07-26 13:40:43 +02002008 *
2009 * This is safe to call from within a preemption notifier.
2010 */
2011void preempt_notifier_unregister(struct preempt_notifier *notifier)
2012{
2013 hlist_del(&notifier->link);
2014}
2015EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
2016
2017static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2018{
2019 struct preempt_notifier *notifier;
2020 struct hlist_node *node;
2021
2022 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2023 notifier->ops->sched_in(notifier, raw_smp_processor_id());
2024}
2025
2026static void
2027fire_sched_out_preempt_notifiers(struct task_struct *curr,
2028 struct task_struct *next)
2029{
2030 struct preempt_notifier *notifier;
2031 struct hlist_node *node;
2032
2033 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2034 notifier->ops->sched_out(notifier, next);
2035}
2036
2037#else
2038
2039static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2040{
2041}
2042
2043static void
2044fire_sched_out_preempt_notifiers(struct task_struct *curr,
2045 struct task_struct *next)
2046{
2047}
2048
2049#endif
2050
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051/**
Nick Piggin4866cde2005-06-25 14:57:23 -07002052 * prepare_task_switch - prepare to switch tasks
2053 * @rq: the runqueue preparing to switch
Randy Dunlap421cee22007-07-31 00:37:50 -07002054 * @prev: the current task that is being switched out
Nick Piggin4866cde2005-06-25 14:57:23 -07002055 * @next: the task we are going to switch to.
2056 *
2057 * This is called with the rq lock held and interrupts off. It must
2058 * be paired with a subsequent finish_task_switch after the context
2059 * switch.
2060 *
2061 * prepare_task_switch sets up locking and calls architecture specific
2062 * hooks.
2063 */
Avi Kivitye107be32007-07-26 13:40:43 +02002064static inline void
2065prepare_task_switch(struct rq *rq, struct task_struct *prev,
2066 struct task_struct *next)
Nick Piggin4866cde2005-06-25 14:57:23 -07002067{
Avi Kivitye107be32007-07-26 13:40:43 +02002068 fire_sched_out_preempt_notifiers(prev, next);
Nick Piggin4866cde2005-06-25 14:57:23 -07002069 prepare_lock_switch(rq, next);
2070 prepare_arch_switch(next);
2071}
2072
2073/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 * finish_task_switch - clean up after a task-switch
Jeff Garzik344baba2005-09-07 01:15:17 -04002075 * @rq: runqueue associated with task-switch
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 * @prev: the thread we just switched away from.
2077 *
Nick Piggin4866cde2005-06-25 14:57:23 -07002078 * finish_task_switch must be called after the context switch, paired
2079 * with a prepare_task_switch call before the context switch.
2080 * finish_task_switch will reconcile locking set up by prepare_task_switch,
2081 * and do any other architecture-specific cleanup actions.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 *
2083 * Note that we may have delayed dropping an mm in context_switch(). If
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01002084 * so, we finish that here outside of the runqueue lock. (Doing it
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 * with the lock held can cause deadlocks; see schedule() for
2086 * details.)
2087 */
Alexey Dobriyana9957442007-10-15 17:00:13 +02002088static void finish_task_switch(struct rq *rq, struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 __releases(rq->lock)
2090{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 struct mm_struct *mm = rq->prev_mm;
Oleg Nesterov55a101f2006-09-29 02:01:10 -07002092 long prev_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
2094 rq->prev_mm = NULL;
2095
2096 /*
2097 * A task struct has one reference for the use as "current".
Oleg Nesterovc394cc92006-09-29 02:01:11 -07002098 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
Oleg Nesterov55a101f2006-09-29 02:01:10 -07002099 * schedule one last time. The schedule call will never return, and
2100 * the scheduled task must drop that reference.
Oleg Nesterovc394cc92006-09-29 02:01:11 -07002101 * The test for TASK_DEAD must occur while the runqueue locks are
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 * still held, otherwise prev could be scheduled on another cpu, die
2103 * there before we look at prev->state, and then the reference would
2104 * be dropped twice.
2105 * Manfred Spraul <manfred@colorfullife.com>
2106 */
Oleg Nesterov55a101f2006-09-29 02:01:10 -07002107 prev_state = prev->state;
Nick Piggin4866cde2005-06-25 14:57:23 -07002108 finish_arch_switch(prev);
2109 finish_lock_switch(rq, prev);
Steven Rostedt9a897c52008-01-25 21:08:22 +01002110#ifdef CONFIG_SMP
2111 if (current->sched_class->post_schedule)
2112 current->sched_class->post_schedule(rq);
2113#endif
Steven Rostedte8fa1362008-01-25 21:08:05 +01002114
Avi Kivitye107be32007-07-26 13:40:43 +02002115 fire_sched_in_preempt_notifiers(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 if (mm)
2117 mmdrop(mm);
Oleg Nesterovc394cc92006-09-29 02:01:11 -07002118 if (unlikely(prev_state == TASK_DEAD)) {
bibo maoc6fd91f2006-03-26 01:38:20 -08002119 /*
2120 * Remove function-return probe instances associated with this
2121 * task and put them back on the free list.
Ingo Molnar9761eea2007-07-09 18:52:00 +02002122 */
bibo maoc6fd91f2006-03-26 01:38:20 -08002123 kprobe_flush_task(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 put_task_struct(prev);
bibo maoc6fd91f2006-03-26 01:38:20 -08002125 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126}
2127
2128/**
2129 * schedule_tail - first thing a freshly forked thread must call.
2130 * @prev: the thread we just switched away from.
2131 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07002132asmlinkage void schedule_tail(struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 __releases(rq->lock)
2134{
Ingo Molnar70b97a72006-07-03 00:25:42 -07002135 struct rq *rq = this_rq();
2136
Nick Piggin4866cde2005-06-25 14:57:23 -07002137 finish_task_switch(rq, prev);
2138#ifdef __ARCH_WANT_UNLOCKED_CTXSW
2139 /* In this case, finish_task_switch does not reenable preemption */
2140 preempt_enable();
2141#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 if (current->set_child_tid)
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002143 put_user(task_pid_vnr(current), current->set_child_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144}
2145
2146/*
2147 * context_switch - switch to the new MM and the new
2148 * thread's register state.
2149 */
Ingo Molnardd41f592007-07-09 18:51:59 +02002150static inline void
Ingo Molnar70b97a72006-07-03 00:25:42 -07002151context_switch(struct rq *rq, struct task_struct *prev,
Ingo Molnar36c8b582006-07-03 00:25:41 -07002152 struct task_struct *next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153{
Ingo Molnardd41f592007-07-09 18:51:59 +02002154 struct mm_struct *mm, *oldmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155
Avi Kivitye107be32007-07-26 13:40:43 +02002156 prepare_task_switch(rq, prev, next);
Ingo Molnardd41f592007-07-09 18:51:59 +02002157 mm = next->mm;
2158 oldmm = prev->active_mm;
Zachary Amsden9226d122007-02-13 13:26:21 +01002159 /*
2160 * For paravirt, this is coupled with an exit in switch_to to
2161 * combine the page table reload and the switch backend into
2162 * one hypercall.
2163 */
2164 arch_enter_lazy_cpu_mode();
2165
Ingo Molnardd41f592007-07-09 18:51:59 +02002166 if (unlikely(!mm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 next->active_mm = oldmm;
2168 atomic_inc(&oldmm->mm_count);
2169 enter_lazy_tlb(oldmm, next);
2170 } else
2171 switch_mm(oldmm, mm, next);
2172
Ingo Molnardd41f592007-07-09 18:51:59 +02002173 if (unlikely(!prev->mm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 prev->active_mm = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 rq->prev_mm = oldmm;
2176 }
Ingo Molnar3a5f5e42006-07-14 00:24:27 -07002177 /*
2178 * Since the runqueue lock will be released by the next
2179 * task (which is an invalid locking op but in the case
2180 * of the scheduler it's an obvious special-case), so we
2181 * do an early lockdep release here:
2182 */
2183#ifndef __ARCH_WANT_UNLOCKED_CTXSW
Ingo Molnar8a25d5d2006-07-03 00:24:54 -07002184 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
Ingo Molnar3a5f5e42006-07-14 00:24:27 -07002185#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186
2187 /* Here we just switch the register state and the stack. */
2188 switch_to(prev, next, prev);
2189
Ingo Molnardd41f592007-07-09 18:51:59 +02002190 barrier();
2191 /*
2192 * this_rq must be evaluated again because prev may have moved
2193 * CPUs since it called schedule(), thus the 'rq' on its stack
2194 * frame will be invalid.
2195 */
2196 finish_task_switch(this_rq(), prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197}
2198
2199/*
2200 * nr_running, nr_uninterruptible and nr_context_switches:
2201 *
2202 * externally visible scheduler statistics: current number of runnable
2203 * threads, current number of uninterruptible-sleeping threads, total
2204 * number of context switches performed since bootup.
2205 */
2206unsigned long nr_running(void)
2207{
2208 unsigned long i, sum = 0;
2209
2210 for_each_online_cpu(i)
2211 sum += cpu_rq(i)->nr_running;
2212
2213 return sum;
2214}
2215
2216unsigned long nr_uninterruptible(void)
2217{
2218 unsigned long i, sum = 0;
2219
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002220 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 sum += cpu_rq(i)->nr_uninterruptible;
2222
2223 /*
2224 * Since we read the counters lockless, it might be slightly
2225 * inaccurate. Do not allow it to go below zero though:
2226 */
2227 if (unlikely((long)sum < 0))
2228 sum = 0;
2229
2230 return sum;
2231}
2232
2233unsigned long long nr_context_switches(void)
2234{
Steven Rostedtcc94abf2006-06-27 02:54:31 -07002235 int i;
2236 unsigned long long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002238 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 sum += cpu_rq(i)->nr_switches;
2240
2241 return sum;
2242}
2243
2244unsigned long nr_iowait(void)
2245{
2246 unsigned long i, sum = 0;
2247
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002248 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 sum += atomic_read(&cpu_rq(i)->nr_iowait);
2250
2251 return sum;
2252}
2253
Jack Steinerdb1b1fe2006-03-31 02:31:21 -08002254unsigned long nr_active(void)
2255{
2256 unsigned long i, running = 0, uninterruptible = 0;
2257
2258 for_each_online_cpu(i) {
2259 running += cpu_rq(i)->nr_running;
2260 uninterruptible += cpu_rq(i)->nr_uninterruptible;
2261 }
2262
2263 if (unlikely((long)uninterruptible < 0))
2264 uninterruptible = 0;
2265
2266 return running + uninterruptible;
2267}
2268
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269/*
Ingo Molnardd41f592007-07-09 18:51:59 +02002270 * Update rq->cpu_load[] statistics. This function is usually called every
2271 * scheduler tick (TICK_NSEC).
Ingo Molnar48f24c42006-07-03 00:25:40 -07002272 */
Ingo Molnardd41f592007-07-09 18:51:59 +02002273static void update_cpu_load(struct rq *this_rq)
Ingo Molnar48f24c42006-07-03 00:25:40 -07002274{
Dmitry Adamushko495eca42007-10-15 17:00:06 +02002275 unsigned long this_load = this_rq->load.weight;
Ingo Molnardd41f592007-07-09 18:51:59 +02002276 int i, scale;
2277
2278 this_rq->nr_load_updates++;
Ingo Molnardd41f592007-07-09 18:51:59 +02002279
2280 /* Update our load: */
2281 for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
2282 unsigned long old_load, new_load;
2283
2284 /* scale is effectively 1 << i now, and >> i divides by scale */
2285
2286 old_load = this_rq->cpu_load[i];
2287 new_load = this_load;
Ingo Molnara25707f2007-10-15 17:00:03 +02002288 /*
2289 * Round up the averaging division if load is increasing. This
2290 * prevents us from getting stuck on 9 if the load is 10, for
2291 * example.
2292 */
2293 if (new_load > old_load)
2294 new_load += scale-1;
Ingo Molnardd41f592007-07-09 18:51:59 +02002295 this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i;
2296 }
Ingo Molnar48f24c42006-07-03 00:25:40 -07002297}
2298
Ingo Molnardd41f592007-07-09 18:51:59 +02002299#ifdef CONFIG_SMP
2300
Ingo Molnar48f24c42006-07-03 00:25:40 -07002301/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 * double_rq_lock - safely lock two runqueues
2303 *
2304 * Note this does not disable interrupts like task_rq_lock,
2305 * you need to do so manually before calling.
2306 */
Ingo Molnar70b97a72006-07-03 00:25:42 -07002307static void double_rq_lock(struct rq *rq1, struct rq *rq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 __acquires(rq1->lock)
2309 __acquires(rq2->lock)
2310{
Kirill Korotaev054b9102006-12-10 02:20:11 -08002311 BUG_ON(!irqs_disabled());
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 if (rq1 == rq2) {
2313 spin_lock(&rq1->lock);
2314 __acquire(rq2->lock); /* Fake it out ;) */
2315 } else {
Chen, Kenneth Wc96d1452006-06-27 02:54:28 -07002316 if (rq1 < rq2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 spin_lock(&rq1->lock);
2318 spin_lock(&rq2->lock);
2319 } else {
2320 spin_lock(&rq2->lock);
2321 spin_lock(&rq1->lock);
2322 }
2323 }
Ingo Molnar6e82a3b2007-08-09 11:16:51 +02002324 update_rq_clock(rq1);
2325 update_rq_clock(rq2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326}
2327
2328/*
2329 * double_rq_unlock - safely unlock two runqueues
2330 *
2331 * Note this does not restore interrupts like task_rq_unlock,
2332 * you need to do so manually after calling.
2333 */
Ingo Molnar70b97a72006-07-03 00:25:42 -07002334static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 __releases(rq1->lock)
2336 __releases(rq2->lock)
2337{
2338 spin_unlock(&rq1->lock);
2339 if (rq1 != rq2)
2340 spin_unlock(&rq2->lock);
2341 else
2342 __release(rq2->lock);
2343}
2344
2345/*
2346 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2347 */
Steven Rostedte8fa1362008-01-25 21:08:05 +01002348static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 __releases(this_rq->lock)
2350 __acquires(busiest->lock)
2351 __acquires(this_rq->lock)
2352{
Steven Rostedte8fa1362008-01-25 21:08:05 +01002353 int ret = 0;
2354
Kirill Korotaev054b9102006-12-10 02:20:11 -08002355 if (unlikely(!irqs_disabled())) {
2356 /* printk() doesn't work good under rq->lock */
2357 spin_unlock(&this_rq->lock);
2358 BUG_ON(1);
2359 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 if (unlikely(!spin_trylock(&busiest->lock))) {
Chen, Kenneth Wc96d1452006-06-27 02:54:28 -07002361 if (busiest < this_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 spin_unlock(&this_rq->lock);
2363 spin_lock(&busiest->lock);
2364 spin_lock(&this_rq->lock);
Steven Rostedte8fa1362008-01-25 21:08:05 +01002365 ret = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 } else
2367 spin_lock(&busiest->lock);
2368 }
Steven Rostedte8fa1362008-01-25 21:08:05 +01002369 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370}
2371
2372/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 * If dest_cpu is allowed for this process, migrate the task to it.
2374 * This is accomplished by forcing the cpu_allowed mask to only
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01002375 * allow dest_cpu, which will force the cpu onto dest_cpu. Then
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 * the cpu_allowed mask is restored.
2377 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07002378static void sched_migrate_task(struct task_struct *p, int dest_cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379{
Ingo Molnar70b97a72006-07-03 00:25:42 -07002380 struct migration_req req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07002382 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383
2384 rq = task_rq_lock(p, &flags);
2385 if (!cpu_isset(dest_cpu, p->cpus_allowed)
2386 || unlikely(cpu_is_offline(dest_cpu)))
2387 goto out;
2388
2389 /* force the process onto the specified CPU */
2390 if (migrate_task(p, dest_cpu, &req)) {
2391 /* Need to wait for migration thread (might exit: take ref). */
2392 struct task_struct *mt = rq->migration_thread;
Ingo Molnar36c8b582006-07-03 00:25:41 -07002393
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 get_task_struct(mt);
2395 task_rq_unlock(rq, &flags);
2396 wake_up_process(mt);
2397 put_task_struct(mt);
2398 wait_for_completion(&req.done);
Ingo Molnar36c8b582006-07-03 00:25:41 -07002399
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 return;
2401 }
2402out:
2403 task_rq_unlock(rq, &flags);
2404}
2405
2406/*
Nick Piggin476d1392005-06-25 14:57:29 -07002407 * sched_exec - execve() is a valuable balancing opportunity, because at
2408 * this point the task has the smallest effective memory and cache footprint.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 */
2410void sched_exec(void)
2411{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 int new_cpu, this_cpu = get_cpu();
Nick Piggin476d1392005-06-25 14:57:29 -07002413 new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 put_cpu();
Nick Piggin476d1392005-06-25 14:57:29 -07002415 if (new_cpu != this_cpu)
2416 sched_migrate_task(current, new_cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417}
2418
2419/*
2420 * pull_task - move a task from a remote runqueue to the local runqueue.
2421 * Both runqueues must be locked.
2422 */
Ingo Molnardd41f592007-07-09 18:51:59 +02002423static void pull_task(struct rq *src_rq, struct task_struct *p,
2424 struct rq *this_rq, int this_cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425{
Ingo Molnar2e1cb742007-08-09 11:16:49 +02002426 deactivate_task(src_rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 set_task_cpu(p, this_cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02002428 activate_task(this_rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 /*
2430 * Note that idle threads have a prio of MAX_PRIO, for this test
2431 * to be always true for them.
2432 */
Ingo Molnardd41f592007-07-09 18:51:59 +02002433 check_preempt_curr(this_rq, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434}
2435
2436/*
2437 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
2438 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08002439static
Ingo Molnar70b97a72006-07-03 00:25:42 -07002440int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu,
Ingo Molnard15bcfd2007-07-09 18:51:57 +02002441 struct sched_domain *sd, enum cpu_idle_type idle,
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07002442 int *all_pinned)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443{
2444 /*
2445 * We do not migrate tasks that are:
2446 * 1) running (obviously), or
2447 * 2) cannot be migrated to this CPU due to cpus_allowed, or
2448 * 3) are cache-hot on their current CPU.
2449 */
Ingo Molnarcc367732007-10-15 17:00:18 +02002450 if (!cpu_isset(this_cpu, p->cpus_allowed)) {
2451 schedstat_inc(p, se.nr_failed_migrations_affine);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 return 0;
Ingo Molnarcc367732007-10-15 17:00:18 +02002453 }
Nick Piggin81026792005-06-25 14:57:07 -07002454 *all_pinned = 0;
2455
Ingo Molnarcc367732007-10-15 17:00:18 +02002456 if (task_running(rq, p)) {
2457 schedstat_inc(p, se.nr_failed_migrations_running);
Nick Piggin81026792005-06-25 14:57:07 -07002458 return 0;
Ingo Molnarcc367732007-10-15 17:00:18 +02002459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460
Ingo Molnarda84d962007-10-15 17:00:18 +02002461 /*
2462 * Aggressive migration if:
2463 * 1) task is cache cold, or
2464 * 2) too many balance attempts have failed.
2465 */
2466
Ingo Molnar6bc16652007-10-15 17:00:18 +02002467 if (!task_hot(p, rq->clock, sd) ||
2468 sd->nr_balance_failed > sd->cache_nice_tries) {
Ingo Molnarda84d962007-10-15 17:00:18 +02002469#ifdef CONFIG_SCHEDSTATS
Ingo Molnarcc367732007-10-15 17:00:18 +02002470 if (task_hot(p, rq->clock, sd)) {
Ingo Molnarda84d962007-10-15 17:00:18 +02002471 schedstat_inc(sd, lb_hot_gained[idle]);
Ingo Molnarcc367732007-10-15 17:00:18 +02002472 schedstat_inc(p, se.nr_forced_migrations);
2473 }
Ingo Molnarda84d962007-10-15 17:00:18 +02002474#endif
2475 return 1;
2476 }
2477
Ingo Molnarcc367732007-10-15 17:00:18 +02002478 if (task_hot(p, rq->clock, sd)) {
2479 schedstat_inc(p, se.nr_failed_migrations_hot);
Ingo Molnarda84d962007-10-15 17:00:18 +02002480 return 0;
Ingo Molnarcc367732007-10-15 17:00:18 +02002481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 return 1;
2483}
2484
Peter Williamse1d14842007-10-24 18:23:51 +02002485static unsigned long
2486balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
2487 unsigned long max_load_move, struct sched_domain *sd,
2488 enum cpu_idle_type idle, int *all_pinned,
2489 int *this_best_prio, struct rq_iterator *iterator)
Ingo Molnardd41f592007-07-09 18:51:59 +02002490{
Peter Zijlstrab82d9fd2007-11-09 22:39:39 +01002491 int loops = 0, pulled = 0, pinned = 0, skip_for_load;
Ingo Molnardd41f592007-07-09 18:51:59 +02002492 struct task_struct *p;
2493 long rem_load_move = max_load_move;
2494
Peter Williamse1d14842007-10-24 18:23:51 +02002495 if (max_load_move == 0)
Ingo Molnardd41f592007-07-09 18:51:59 +02002496 goto out;
2497
2498 pinned = 1;
2499
2500 /*
2501 * Start the load-balancing iterator:
2502 */
2503 p = iterator->start(iterator->arg);
2504next:
Peter Zijlstrab82d9fd2007-11-09 22:39:39 +01002505 if (!p || loops++ > sysctl_sched_nr_migrate)
Ingo Molnardd41f592007-07-09 18:51:59 +02002506 goto out;
2507 /*
Peter Zijlstrab82d9fd2007-11-09 22:39:39 +01002508 * To help distribute high priority tasks across CPUs we don't
Ingo Molnardd41f592007-07-09 18:51:59 +02002509 * skip a task if it will be the highest priority task (i.e. smallest
2510 * prio value) on its new queue regardless of its load weight
2511 */
2512 skip_for_load = (p->se.load.weight >> 1) > rem_load_move +
2513 SCHED_LOAD_SCALE_FUZZ;
Peter Williamsa4ac01c2007-08-09 11:16:46 +02002514 if ((skip_for_load && p->prio >= *this_best_prio) ||
Ingo Molnardd41f592007-07-09 18:51:59 +02002515 !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) {
Ingo Molnardd41f592007-07-09 18:51:59 +02002516 p = iterator->next(iterator->arg);
2517 goto next;
2518 }
2519
2520 pull_task(busiest, p, this_rq, this_cpu);
2521 pulled++;
2522 rem_load_move -= p->se.load.weight;
2523
2524 /*
Peter Zijlstrab82d9fd2007-11-09 22:39:39 +01002525 * We only want to steal up to the prescribed amount of weighted load.
Ingo Molnardd41f592007-07-09 18:51:59 +02002526 */
Peter Williamse1d14842007-10-24 18:23:51 +02002527 if (rem_load_move > 0) {
Peter Williamsa4ac01c2007-08-09 11:16:46 +02002528 if (p->prio < *this_best_prio)
2529 *this_best_prio = p->prio;
Ingo Molnardd41f592007-07-09 18:51:59 +02002530 p = iterator->next(iterator->arg);
2531 goto next;
2532 }
2533out:
2534 /*
Peter Williamse1d14842007-10-24 18:23:51 +02002535 * Right now, this is one of only two places pull_task() is called,
Ingo Molnardd41f592007-07-09 18:51:59 +02002536 * so we can safely collect pull_task() stats here rather than
2537 * inside pull_task().
2538 */
2539 schedstat_add(sd, lb_gained[idle], pulled);
2540
2541 if (all_pinned)
2542 *all_pinned = pinned;
Peter Williamse1d14842007-10-24 18:23:51 +02002543
2544 return max_load_move - rem_load_move;
Ingo Molnardd41f592007-07-09 18:51:59 +02002545}
Ingo Molnar48f24c42006-07-03 00:25:40 -07002546
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547/*
Peter Williams43010652007-08-09 11:16:46 +02002548 * move_tasks tries to move up to max_load_move weighted load from busiest to
2549 * this_rq, as part of a balancing operation within domain "sd".
2550 * Returns 1 if successful and 0 otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 *
2552 * Called with both runqueues locked.
2553 */
Ingo Molnar70b97a72006-07-03 00:25:42 -07002554static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
Peter Williams43010652007-08-09 11:16:46 +02002555 unsigned long max_load_move,
Ingo Molnard15bcfd2007-07-09 18:51:57 +02002556 struct sched_domain *sd, enum cpu_idle_type idle,
Peter Williams2dd73a42006-06-27 02:54:34 -07002557 int *all_pinned)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558{
Ingo Molnar5522d5d2007-10-15 17:00:12 +02002559 const struct sched_class *class = sched_class_highest;
Peter Williams43010652007-08-09 11:16:46 +02002560 unsigned long total_load_moved = 0;
Peter Williamsa4ac01c2007-08-09 11:16:46 +02002561 int this_best_prio = this_rq->curr->prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562
Ingo Molnardd41f592007-07-09 18:51:59 +02002563 do {
Peter Williams43010652007-08-09 11:16:46 +02002564 total_load_moved +=
2565 class->load_balance(this_rq, this_cpu, busiest,
Peter Williamse1d14842007-10-24 18:23:51 +02002566 max_load_move - total_load_moved,
Peter Williamsa4ac01c2007-08-09 11:16:46 +02002567 sd, idle, all_pinned, &this_best_prio);
Ingo Molnardd41f592007-07-09 18:51:59 +02002568 class = class->next;
Peter Williams43010652007-08-09 11:16:46 +02002569 } while (class && max_load_move > total_load_moved);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
Peter Williams43010652007-08-09 11:16:46 +02002571 return total_load_moved > 0;
2572}
2573
Peter Williamse1d14842007-10-24 18:23:51 +02002574static int
2575iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest,
2576 struct sched_domain *sd, enum cpu_idle_type idle,
2577 struct rq_iterator *iterator)
2578{
2579 struct task_struct *p = iterator->start(iterator->arg);
2580 int pinned = 0;
2581
2582 while (p) {
2583 if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) {
2584 pull_task(busiest, p, this_rq, this_cpu);
2585 /*
2586 * Right now, this is only the second place pull_task()
2587 * is called, so we can safely collect pull_task()
2588 * stats here rather than inside pull_task().
2589 */
2590 schedstat_inc(sd, lb_gained[idle]);
2591
2592 return 1;
2593 }
2594 p = iterator->next(iterator->arg);
2595 }
2596
2597 return 0;
2598}
2599
Peter Williams43010652007-08-09 11:16:46 +02002600/*
2601 * move_one_task tries to move exactly one task from busiest to this_rq, as
2602 * part of active balancing operations within "domain".
2603 * Returns 1 if successful and 0 otherwise.
2604 *
2605 * Called with both runqueues locked.
2606 */
2607static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest,
2608 struct sched_domain *sd, enum cpu_idle_type idle)
2609{
Ingo Molnar5522d5d2007-10-15 17:00:12 +02002610 const struct sched_class *class;
Peter Williams43010652007-08-09 11:16:46 +02002611
2612 for (class = sched_class_highest; class; class = class->next)
Peter Williamse1d14842007-10-24 18:23:51 +02002613 if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle))
Peter Williams43010652007-08-09 11:16:46 +02002614 return 1;
2615
2616 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617}
2618
2619/*
2620 * find_busiest_group finds and returns the busiest CPU group within the
Ingo Molnar48f24c42006-07-03 00:25:40 -07002621 * domain. It calculates and returns the amount of weighted load which
2622 * should be moved to restore balance via the imbalance parameter.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 */
2624static struct sched_group *
2625find_busiest_group(struct sched_domain *sd, int this_cpu,
Ingo Molnardd41f592007-07-09 18:51:59 +02002626 unsigned long *imbalance, enum cpu_idle_type idle,
2627 int *sd_idle, cpumask_t *cpus, int *balance)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628{
2629 struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups;
2630 unsigned long max_load, avg_load, total_load, this_load, total_pwr;
Siddha, Suresh B0c117f12005-09-10 00:26:21 -07002631 unsigned long max_pull;
Peter Williams2dd73a42006-06-27 02:54:34 -07002632 unsigned long busiest_load_per_task, busiest_nr_running;
2633 unsigned long this_load_per_task, this_nr_running;
Ken Chen908a7c12007-10-17 16:55:11 +02002634 int load_idx, group_imb = 0;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002635#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
2636 int power_savings_balance = 1;
2637 unsigned long leader_nr_running = 0, min_load_per_task = 0;
2638 unsigned long min_nr_running = ULONG_MAX;
2639 struct sched_group *group_min = NULL, *group_leader = NULL;
2640#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641
2642 max_load = this_load = total_load = total_pwr = 0;
Peter Williams2dd73a42006-06-27 02:54:34 -07002643 busiest_load_per_task = busiest_nr_running = 0;
2644 this_load_per_task = this_nr_running = 0;
Ingo Molnard15bcfd2007-07-09 18:51:57 +02002645 if (idle == CPU_NOT_IDLE)
Nick Piggin78979862005-06-25 14:57:13 -07002646 load_idx = sd->busy_idx;
Ingo Molnard15bcfd2007-07-09 18:51:57 +02002647 else if (idle == CPU_NEWLY_IDLE)
Nick Piggin78979862005-06-25 14:57:13 -07002648 load_idx = sd->newidle_idx;
2649 else
2650 load_idx = sd->idle_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651
2652 do {
Ken Chen908a7c12007-10-17 16:55:11 +02002653 unsigned long load, group_capacity, max_cpu_load, min_cpu_load;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 int local_group;
2655 int i;
Ken Chen908a7c12007-10-17 16:55:11 +02002656 int __group_imb = 0;
Siddha, Suresh B783609c2006-12-10 02:20:33 -08002657 unsigned int balance_cpu = -1, first_idle_cpu = 0;
Peter Williams2dd73a42006-06-27 02:54:34 -07002658 unsigned long sum_nr_running, sum_weighted_load;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659
2660 local_group = cpu_isset(this_cpu, group->cpumask);
2661
Siddha, Suresh B783609c2006-12-10 02:20:33 -08002662 if (local_group)
2663 balance_cpu = first_cpu(group->cpumask);
2664
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 /* Tally up the load of all CPUs in the group */
Peter Williams2dd73a42006-06-27 02:54:34 -07002666 sum_weighted_load = sum_nr_running = avg_load = 0;
Ken Chen908a7c12007-10-17 16:55:11 +02002667 max_cpu_load = 0;
2668 min_cpu_load = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669
2670 for_each_cpu_mask(i, group->cpumask) {
Christoph Lameter0a2966b2006-09-25 23:30:51 -07002671 struct rq *rq;
2672
2673 if (!cpu_isset(i, *cpus))
2674 continue;
2675
2676 rq = cpu_rq(i);
Peter Williams2dd73a42006-06-27 02:54:34 -07002677
Suresh Siddha9439aab2007-07-19 21:28:35 +02002678 if (*sd_idle && rq->nr_running)
Nick Piggin5969fe02005-09-10 00:26:19 -07002679 *sd_idle = 0;
2680
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 /* Bias balancing toward cpus of our domain */
Siddha, Suresh B783609c2006-12-10 02:20:33 -08002682 if (local_group) {
2683 if (idle_cpu(i) && !first_idle_cpu) {
2684 first_idle_cpu = 1;
2685 balance_cpu = i;
2686 }
2687
Nick Piggina2000572006-02-10 01:51:02 -08002688 load = target_load(i, load_idx);
Ken Chen908a7c12007-10-17 16:55:11 +02002689 } else {
Nick Piggina2000572006-02-10 01:51:02 -08002690 load = source_load(i, load_idx);
Ken Chen908a7c12007-10-17 16:55:11 +02002691 if (load > max_cpu_load)
2692 max_cpu_load = load;
2693 if (min_cpu_load > load)
2694 min_cpu_load = load;
2695 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696
2697 avg_load += load;
Peter Williams2dd73a42006-06-27 02:54:34 -07002698 sum_nr_running += rq->nr_running;
Ingo Molnardd41f592007-07-09 18:51:59 +02002699 sum_weighted_load += weighted_cpuload(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 }
2701
Siddha, Suresh B783609c2006-12-10 02:20:33 -08002702 /*
2703 * First idle cpu or the first cpu(busiest) in this sched group
2704 * is eligible for doing load balancing at this and above
Suresh Siddha9439aab2007-07-19 21:28:35 +02002705 * domains. In the newly idle case, we will allow all the cpu's
2706 * to do the newly idle load balance.
Siddha, Suresh B783609c2006-12-10 02:20:33 -08002707 */
Suresh Siddha9439aab2007-07-19 21:28:35 +02002708 if (idle != CPU_NEWLY_IDLE && local_group &&
2709 balance_cpu != this_cpu && balance) {
Siddha, Suresh B783609c2006-12-10 02:20:33 -08002710 *balance = 0;
2711 goto ret;
2712 }
2713
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714 total_load += avg_load;
Eric Dumazet5517d862007-05-08 00:32:57 -07002715 total_pwr += group->__cpu_power;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716
2717 /* Adjust by relative CPU power of the group */
Eric Dumazet5517d862007-05-08 00:32:57 -07002718 avg_load = sg_div_cpu_power(group,
2719 avg_load * SCHED_LOAD_SCALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720
Ken Chen908a7c12007-10-17 16:55:11 +02002721 if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE)
2722 __group_imb = 1;
2723
Eric Dumazet5517d862007-05-08 00:32:57 -07002724 group_capacity = group->__cpu_power / SCHED_LOAD_SCALE;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002725
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 if (local_group) {
2727 this_load = avg_load;
2728 this = group;
Peter Williams2dd73a42006-06-27 02:54:34 -07002729 this_nr_running = sum_nr_running;
2730 this_load_per_task = sum_weighted_load;
2731 } else if (avg_load > max_load &&
Ken Chen908a7c12007-10-17 16:55:11 +02002732 (sum_nr_running > group_capacity || __group_imb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 max_load = avg_load;
2734 busiest = group;
Peter Williams2dd73a42006-06-27 02:54:34 -07002735 busiest_nr_running = sum_nr_running;
2736 busiest_load_per_task = sum_weighted_load;
Ken Chen908a7c12007-10-17 16:55:11 +02002737 group_imb = __group_imb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 }
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002739
2740#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
2741 /*
2742 * Busy processors will not participate in power savings
2743 * balance.
2744 */
Ingo Molnardd41f592007-07-09 18:51:59 +02002745 if (idle == CPU_NOT_IDLE ||
2746 !(sd->flags & SD_POWERSAVINGS_BALANCE))
2747 goto group_next;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002748
2749 /*
2750 * If the local group is idle or completely loaded
2751 * no need to do power savings balance at this domain
2752 */
2753 if (local_group && (this_nr_running >= group_capacity ||
2754 !this_nr_running))
2755 power_savings_balance = 0;
2756
Ingo Molnardd41f592007-07-09 18:51:59 +02002757 /*
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002758 * If a group is already running at full capacity or idle,
2759 * don't include that group in power savings calculations
Ingo Molnardd41f592007-07-09 18:51:59 +02002760 */
2761 if (!power_savings_balance || sum_nr_running >= group_capacity
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002762 || !sum_nr_running)
Ingo Molnardd41f592007-07-09 18:51:59 +02002763 goto group_next;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002764
Ingo Molnardd41f592007-07-09 18:51:59 +02002765 /*
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002766 * Calculate the group which has the least non-idle load.
Ingo Molnardd41f592007-07-09 18:51:59 +02002767 * This is the group from where we need to pick up the load
2768 * for saving power
2769 */
2770 if ((sum_nr_running < min_nr_running) ||
2771 (sum_nr_running == min_nr_running &&
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002772 first_cpu(group->cpumask) <
2773 first_cpu(group_min->cpumask))) {
Ingo Molnardd41f592007-07-09 18:51:59 +02002774 group_min = group;
2775 min_nr_running = sum_nr_running;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002776 min_load_per_task = sum_weighted_load /
2777 sum_nr_running;
Ingo Molnardd41f592007-07-09 18:51:59 +02002778 }
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002779
Ingo Molnardd41f592007-07-09 18:51:59 +02002780 /*
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002781 * Calculate the group which is almost near its
Ingo Molnardd41f592007-07-09 18:51:59 +02002782 * capacity but still has some space to pick up some load
2783 * from other group and save more power
2784 */
2785 if (sum_nr_running <= group_capacity - 1) {
2786 if (sum_nr_running > leader_nr_running ||
2787 (sum_nr_running == leader_nr_running &&
2788 first_cpu(group->cpumask) >
2789 first_cpu(group_leader->cpumask))) {
2790 group_leader = group;
2791 leader_nr_running = sum_nr_running;
2792 }
Ingo Molnar48f24c42006-07-03 00:25:40 -07002793 }
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002794group_next:
2795#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 group = group->next;
2797 } while (group != sd->groups);
2798
Peter Williams2dd73a42006-06-27 02:54:34 -07002799 if (!busiest || this_load >= max_load || busiest_nr_running == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 goto out_balanced;
2801
2802 avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr;
2803
2804 if (this_load >= avg_load ||
2805 100*max_load <= sd->imbalance_pct*this_load)
2806 goto out_balanced;
2807
Peter Williams2dd73a42006-06-27 02:54:34 -07002808 busiest_load_per_task /= busiest_nr_running;
Ken Chen908a7c12007-10-17 16:55:11 +02002809 if (group_imb)
2810 busiest_load_per_task = min(busiest_load_per_task, avg_load);
2811
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 /*
2813 * We're trying to get all the cpus to the average_load, so we don't
2814 * want to push ourselves above the average load, nor do we wish to
2815 * reduce the max loaded cpu below the average load, as either of these
2816 * actions would just result in more rebalancing later, and ping-pong
2817 * tasks around. Thus we look for the minimum possible imbalance.
2818 * Negative imbalances (*we* are more loaded than anyone else) will
2819 * be counted as no imbalance for these purposes -- we can't fix that
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01002820 * by pulling tasks to us. Be careful of negative numbers as they'll
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 * appear as very large values with unsigned longs.
2822 */
Peter Williams2dd73a42006-06-27 02:54:34 -07002823 if (max_load <= busiest_load_per_task)
2824 goto out_balanced;
2825
2826 /*
2827 * In the presence of smp nice balancing, certain scenarios can have
2828 * max load less than avg load(as we skip the groups at or below
2829 * its cpu_power, while calculating max_load..)
2830 */
2831 if (max_load < avg_load) {
2832 *imbalance = 0;
2833 goto small_imbalance;
2834 }
Siddha, Suresh B0c117f12005-09-10 00:26:21 -07002835
2836 /* Don't want to pull so many tasks that a group would go idle */
Peter Williams2dd73a42006-06-27 02:54:34 -07002837 max_pull = min(max_load - avg_load, max_load - busiest_load_per_task);
Siddha, Suresh B0c117f12005-09-10 00:26:21 -07002838
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 /* How much load to actually move to equalise the imbalance */
Eric Dumazet5517d862007-05-08 00:32:57 -07002840 *imbalance = min(max_pull * busiest->__cpu_power,
2841 (avg_load - this_load) * this->__cpu_power)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842 / SCHED_LOAD_SCALE;
2843
Peter Williams2dd73a42006-06-27 02:54:34 -07002844 /*
2845 * if *imbalance is less than the average load per runnable task
2846 * there is no gaurantee that any tasks will be moved so we'll have
2847 * a think about bumping its value to force at least one task to be
2848 * moved
2849 */
Suresh Siddha7fd0d2d2007-09-05 14:32:48 +02002850 if (*imbalance < busiest_load_per_task) {
Ingo Molnar48f24c42006-07-03 00:25:40 -07002851 unsigned long tmp, pwr_now, pwr_move;
Peter Williams2dd73a42006-06-27 02:54:34 -07002852 unsigned int imbn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853
Peter Williams2dd73a42006-06-27 02:54:34 -07002854small_imbalance:
2855 pwr_move = pwr_now = 0;
2856 imbn = 2;
2857 if (this_nr_running) {
2858 this_load_per_task /= this_nr_running;
2859 if (busiest_load_per_task > this_load_per_task)
2860 imbn = 1;
2861 } else
2862 this_load_per_task = SCHED_LOAD_SCALE;
2863
Ingo Molnardd41f592007-07-09 18:51:59 +02002864 if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >=
2865 busiest_load_per_task * imbn) {
Peter Williams2dd73a42006-06-27 02:54:34 -07002866 *imbalance = busiest_load_per_task;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 return busiest;
2868 }
2869
2870 /*
2871 * OK, we don't have enough imbalance to justify moving tasks,
2872 * however we may be able to increase total CPU power used by
2873 * moving them.
2874 */
2875
Eric Dumazet5517d862007-05-08 00:32:57 -07002876 pwr_now += busiest->__cpu_power *
2877 min(busiest_load_per_task, max_load);
2878 pwr_now += this->__cpu_power *
2879 min(this_load_per_task, this_load);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 pwr_now /= SCHED_LOAD_SCALE;
2881
2882 /* Amount of load we'd subtract */
Eric Dumazet5517d862007-05-08 00:32:57 -07002883 tmp = sg_div_cpu_power(busiest,
2884 busiest_load_per_task * SCHED_LOAD_SCALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 if (max_load > tmp)
Eric Dumazet5517d862007-05-08 00:32:57 -07002886 pwr_move += busiest->__cpu_power *
Peter Williams2dd73a42006-06-27 02:54:34 -07002887 min(busiest_load_per_task, max_load - tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888
2889 /* Amount of load we'd add */
Eric Dumazet5517d862007-05-08 00:32:57 -07002890 if (max_load * busiest->__cpu_power <
Miguel Ojeda Sandonis33859f72006-12-10 02:20:38 -08002891 busiest_load_per_task * SCHED_LOAD_SCALE)
Eric Dumazet5517d862007-05-08 00:32:57 -07002892 tmp = sg_div_cpu_power(this,
2893 max_load * busiest->__cpu_power);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 else
Eric Dumazet5517d862007-05-08 00:32:57 -07002895 tmp = sg_div_cpu_power(this,
2896 busiest_load_per_task * SCHED_LOAD_SCALE);
2897 pwr_move += this->__cpu_power *
2898 min(this_load_per_task, this_load + tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 pwr_move /= SCHED_LOAD_SCALE;
2900
2901 /* Move if we gain throughput */
Suresh Siddha7fd0d2d2007-09-05 14:32:48 +02002902 if (pwr_move > pwr_now)
2903 *imbalance = busiest_load_per_task;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 }
2905
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 return busiest;
2907
2908out_balanced:
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002909#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
Ingo Molnard15bcfd2007-07-09 18:51:57 +02002910 if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE))
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002911 goto ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002913 if (this == group_leader && group_leader != group_min) {
2914 *imbalance = min_load_per_task;
2915 return group_min;
2916 }
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07002917#endif
Siddha, Suresh B783609c2006-12-10 02:20:33 -08002918ret:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919 *imbalance = 0;
2920 return NULL;
2921}
2922
2923/*
2924 * find_busiest_queue - find the busiest runqueue among the cpus in group.
2925 */
Ingo Molnar70b97a72006-07-03 00:25:42 -07002926static struct rq *
Ingo Molnard15bcfd2007-07-09 18:51:57 +02002927find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle,
Christoph Lameter0a2966b2006-09-25 23:30:51 -07002928 unsigned long imbalance, cpumask_t *cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929{
Ingo Molnar70b97a72006-07-03 00:25:42 -07002930 struct rq *busiest = NULL, *rq;
Peter Williams2dd73a42006-06-27 02:54:34 -07002931 unsigned long max_load = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 int i;
2933
2934 for_each_cpu_mask(i, group->cpumask) {
Ingo Molnardd41f592007-07-09 18:51:59 +02002935 unsigned long wl;
Christoph Lameter0a2966b2006-09-25 23:30:51 -07002936
2937 if (!cpu_isset(i, *cpus))
2938 continue;
2939
Ingo Molnar48f24c42006-07-03 00:25:40 -07002940 rq = cpu_rq(i);
Ingo Molnardd41f592007-07-09 18:51:59 +02002941 wl = weighted_cpuload(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942
Ingo Molnardd41f592007-07-09 18:51:59 +02002943 if (rq->nr_running == 1 && wl > imbalance)
Peter Williams2dd73a42006-06-27 02:54:34 -07002944 continue;
2945
Ingo Molnardd41f592007-07-09 18:51:59 +02002946 if (wl > max_load) {
2947 max_load = wl;
Ingo Molnar48f24c42006-07-03 00:25:40 -07002948 busiest = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 }
2950 }
2951
2952 return busiest;
2953}
2954
2955/*
Nick Piggin77391d72005-06-25 14:57:30 -07002956 * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but
2957 * so long as it is large enough.
2958 */
2959#define MAX_PINNED_INTERVAL 512
2960
2961/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 * Check this_cpu to ensure it is balanced within domain. Attempt to move
2963 * tasks if there is an imbalance.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964 */
Ingo Molnar70b97a72006-07-03 00:25:42 -07002965static int load_balance(int this_cpu, struct rq *this_rq,
Ingo Molnard15bcfd2007-07-09 18:51:57 +02002966 struct sched_domain *sd, enum cpu_idle_type idle,
Siddha, Suresh B783609c2006-12-10 02:20:33 -08002967 int *balance)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968{
Peter Williams43010652007-08-09 11:16:46 +02002969 int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970 struct sched_group *group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971 unsigned long imbalance;
Ingo Molnar70b97a72006-07-03 00:25:42 -07002972 struct rq *busiest;
Christoph Lameter0a2966b2006-09-25 23:30:51 -07002973 cpumask_t cpus = CPU_MASK_ALL;
Christoph Lameterfe2eea32006-12-10 02:20:21 -08002974 unsigned long flags;
Nick Piggin5969fe02005-09-10 00:26:19 -07002975
Siddha, Suresh B89c47102006-10-03 01:14:09 -07002976 /*
2977 * When power savings policy is enabled for the parent domain, idle
2978 * sibling can pick up load irrespective of busy siblings. In this case,
Ingo Molnardd41f592007-07-09 18:51:59 +02002979 * let the state of idle sibling percolate up as CPU_IDLE, instead of
Ingo Molnard15bcfd2007-07-09 18:51:57 +02002980 * portraying it as CPU_NOT_IDLE.
Siddha, Suresh B89c47102006-10-03 01:14:09 -07002981 */
Ingo Molnard15bcfd2007-07-09 18:51:57 +02002982 if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER &&
Siddha, Suresh B89c47102006-10-03 01:14:09 -07002983 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
Nick Piggin5969fe02005-09-10 00:26:19 -07002984 sd_idle = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985
Ingo Molnar2d723762007-10-15 17:00:12 +02002986 schedstat_inc(sd, lb_count[idle]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987
Christoph Lameter0a2966b2006-09-25 23:30:51 -07002988redo:
2989 group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle,
Siddha, Suresh B783609c2006-12-10 02:20:33 -08002990 &cpus, balance);
2991
Chen, Kenneth W06066712006-12-10 02:20:35 -08002992 if (*balance == 0)
Siddha, Suresh B783609c2006-12-10 02:20:33 -08002993 goto out_balanced;
Siddha, Suresh B783609c2006-12-10 02:20:33 -08002994
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 if (!group) {
2996 schedstat_inc(sd, lb_nobusyg[idle]);
2997 goto out_balanced;
2998 }
2999
Christoph Lameter0a2966b2006-09-25 23:30:51 -07003000 busiest = find_busiest_queue(group, idle, imbalance, &cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 if (!busiest) {
3002 schedstat_inc(sd, lb_nobusyq[idle]);
3003 goto out_balanced;
3004 }
3005
Nick Piggindb935db2005-06-25 14:57:11 -07003006 BUG_ON(busiest == this_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007
3008 schedstat_add(sd, lb_imbalance[idle], imbalance);
3009
Peter Williams43010652007-08-09 11:16:46 +02003010 ld_moved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 if (busiest->nr_running > 1) {
3012 /*
3013 * Attempt to move tasks. If find_busiest_group has found
3014 * an imbalance but busiest->nr_running <= 1, the group is
Peter Williams43010652007-08-09 11:16:46 +02003015 * still unbalanced. ld_moved simply stays zero, so it is
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016 * correctly treated as an imbalance.
3017 */
Christoph Lameterfe2eea32006-12-10 02:20:21 -08003018 local_irq_save(flags);
Nick Piggine17224b2005-09-10 00:26:18 -07003019 double_rq_lock(this_rq, busiest);
Peter Williams43010652007-08-09 11:16:46 +02003020 ld_moved = move_tasks(this_rq, this_cpu, busiest,
Ingo Molnar48f24c42006-07-03 00:25:40 -07003021 imbalance, sd, idle, &all_pinned);
Nick Piggine17224b2005-09-10 00:26:18 -07003022 double_rq_unlock(this_rq, busiest);
Christoph Lameterfe2eea32006-12-10 02:20:21 -08003023 local_irq_restore(flags);
Nick Piggin81026792005-06-25 14:57:07 -07003024
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003025 /*
3026 * some other cpu did the load balance for us.
3027 */
Peter Williams43010652007-08-09 11:16:46 +02003028 if (ld_moved && this_cpu != smp_processor_id())
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003029 resched_cpu(this_cpu);
3030
Nick Piggin81026792005-06-25 14:57:07 -07003031 /* All tasks on this runqueue were pinned by CPU affinity */
Christoph Lameter0a2966b2006-09-25 23:30:51 -07003032 if (unlikely(all_pinned)) {
3033 cpu_clear(cpu_of(busiest), cpus);
3034 if (!cpus_empty(cpus))
3035 goto redo;
Nick Piggin81026792005-06-25 14:57:07 -07003036 goto out_balanced;
Christoph Lameter0a2966b2006-09-25 23:30:51 -07003037 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 }
Nick Piggin81026792005-06-25 14:57:07 -07003039
Peter Williams43010652007-08-09 11:16:46 +02003040 if (!ld_moved) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 schedstat_inc(sd, lb_failed[idle]);
3042 sd->nr_balance_failed++;
3043
3044 if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045
Christoph Lameterfe2eea32006-12-10 02:20:21 -08003046 spin_lock_irqsave(&busiest->lock, flags);
Siddha, Suresh Bfa3b6dd2005-09-10 00:26:21 -07003047
3048 /* don't kick the migration_thread, if the curr
3049 * task on busiest cpu can't be moved to this_cpu
3050 */
3051 if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) {
Christoph Lameterfe2eea32006-12-10 02:20:21 -08003052 spin_unlock_irqrestore(&busiest->lock, flags);
Siddha, Suresh Bfa3b6dd2005-09-10 00:26:21 -07003053 all_pinned = 1;
3054 goto out_one_pinned;
3055 }
3056
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057 if (!busiest->active_balance) {
3058 busiest->active_balance = 1;
3059 busiest->push_cpu = this_cpu;
Nick Piggin81026792005-06-25 14:57:07 -07003060 active_balance = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 }
Christoph Lameterfe2eea32006-12-10 02:20:21 -08003062 spin_unlock_irqrestore(&busiest->lock, flags);
Nick Piggin81026792005-06-25 14:57:07 -07003063 if (active_balance)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 wake_up_process(busiest->migration_thread);
3065
3066 /*
3067 * We've kicked active balancing, reset the failure
3068 * counter.
3069 */
Nick Piggin39507452005-06-25 14:57:09 -07003070 sd->nr_balance_failed = sd->cache_nice_tries+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071 }
Nick Piggin81026792005-06-25 14:57:07 -07003072 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 sd->nr_balance_failed = 0;
3074
Nick Piggin81026792005-06-25 14:57:07 -07003075 if (likely(!active_balance)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076 /* We were unbalanced, so reset the balancing interval */
3077 sd->balance_interval = sd->min_interval;
Nick Piggin81026792005-06-25 14:57:07 -07003078 } else {
3079 /*
3080 * If we've begun active balancing, start to back off. This
3081 * case may not be covered by the all_pinned logic if there
3082 * is only 1 task on the busy runqueue (because we don't call
3083 * move_tasks).
3084 */
3085 if (sd->balance_interval < sd->max_interval)
3086 sd->balance_interval *= 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087 }
3088
Peter Williams43010652007-08-09 11:16:46 +02003089 if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
Siddha, Suresh B89c47102006-10-03 01:14:09 -07003090 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
Nick Piggin5969fe02005-09-10 00:26:19 -07003091 return -1;
Peter Williams43010652007-08-09 11:16:46 +02003092 return ld_moved;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093
3094out_balanced:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 schedstat_inc(sd, lb_balanced[idle]);
3096
Nick Piggin16cfb1c2005-06-25 14:57:08 -07003097 sd->nr_balance_failed = 0;
Siddha, Suresh Bfa3b6dd2005-09-10 00:26:21 -07003098
3099out_one_pinned:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100 /* tune up the balancing interval */
Nick Piggin77391d72005-06-25 14:57:30 -07003101 if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) ||
3102 (sd->balance_interval < sd->max_interval))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103 sd->balance_interval *= 2;
3104
Ingo Molnar48f24c42006-07-03 00:25:40 -07003105 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
Siddha, Suresh B89c47102006-10-03 01:14:09 -07003106 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
Nick Piggin5969fe02005-09-10 00:26:19 -07003107 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 return 0;
3109}
3110
3111/*
3112 * Check this_cpu to ensure it is balanced within domain. Attempt to move
3113 * tasks if there is an imbalance.
3114 *
Ingo Molnard15bcfd2007-07-09 18:51:57 +02003115 * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116 * this_rq is locked.
3117 */
Ingo Molnar48f24c42006-07-03 00:25:40 -07003118static int
Ingo Molnar70b97a72006-07-03 00:25:42 -07003119load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120{
3121 struct sched_group *group;
Ingo Molnar70b97a72006-07-03 00:25:42 -07003122 struct rq *busiest = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 unsigned long imbalance;
Peter Williams43010652007-08-09 11:16:46 +02003124 int ld_moved = 0;
Nick Piggin5969fe02005-09-10 00:26:19 -07003125 int sd_idle = 0;
Suresh Siddha969bb4e2007-07-19 21:28:35 +02003126 int all_pinned = 0;
Christoph Lameter0a2966b2006-09-25 23:30:51 -07003127 cpumask_t cpus = CPU_MASK_ALL;
Nick Piggin5969fe02005-09-10 00:26:19 -07003128
Siddha, Suresh B89c47102006-10-03 01:14:09 -07003129 /*
3130 * When power savings policy is enabled for the parent domain, idle
3131 * sibling can pick up load irrespective of busy siblings. In this case,
3132 * let the state of idle sibling percolate up as IDLE, instead of
Ingo Molnard15bcfd2007-07-09 18:51:57 +02003133 * portraying it as CPU_NOT_IDLE.
Siddha, Suresh B89c47102006-10-03 01:14:09 -07003134 */
3135 if (sd->flags & SD_SHARE_CPUPOWER &&
3136 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
Nick Piggin5969fe02005-09-10 00:26:19 -07003137 sd_idle = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138
Ingo Molnar2d723762007-10-15 17:00:12 +02003139 schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]);
Christoph Lameter0a2966b2006-09-25 23:30:51 -07003140redo:
Ingo Molnard15bcfd2007-07-09 18:51:57 +02003141 group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE,
Siddha, Suresh B783609c2006-12-10 02:20:33 -08003142 &sd_idle, &cpus, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143 if (!group) {
Ingo Molnard15bcfd2007-07-09 18:51:57 +02003144 schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]);
Nick Piggin16cfb1c2005-06-25 14:57:08 -07003145 goto out_balanced;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146 }
3147
Ingo Molnard15bcfd2007-07-09 18:51:57 +02003148 busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance,
Christoph Lameter0a2966b2006-09-25 23:30:51 -07003149 &cpus);
Nick Piggindb935db2005-06-25 14:57:11 -07003150 if (!busiest) {
Ingo Molnard15bcfd2007-07-09 18:51:57 +02003151 schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]);
Nick Piggin16cfb1c2005-06-25 14:57:08 -07003152 goto out_balanced;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153 }
3154
Nick Piggindb935db2005-06-25 14:57:11 -07003155 BUG_ON(busiest == this_rq);
3156
Ingo Molnard15bcfd2007-07-09 18:51:57 +02003157 schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance);
Nick Piggind6d5cfa2005-09-10 00:26:16 -07003158
Peter Williams43010652007-08-09 11:16:46 +02003159 ld_moved = 0;
Nick Piggind6d5cfa2005-09-10 00:26:16 -07003160 if (busiest->nr_running > 1) {
3161 /* Attempt to move tasks */
3162 double_lock_balance(this_rq, busiest);
Ingo Molnar6e82a3b2007-08-09 11:16:51 +02003163 /* this_rq->clock is already updated */
3164 update_rq_clock(busiest);
Peter Williams43010652007-08-09 11:16:46 +02003165 ld_moved = move_tasks(this_rq, this_cpu, busiest,
Suresh Siddha969bb4e2007-07-19 21:28:35 +02003166 imbalance, sd, CPU_NEWLY_IDLE,
3167 &all_pinned);
Nick Piggind6d5cfa2005-09-10 00:26:16 -07003168 spin_unlock(&busiest->lock);
Christoph Lameter0a2966b2006-09-25 23:30:51 -07003169
Suresh Siddha969bb4e2007-07-19 21:28:35 +02003170 if (unlikely(all_pinned)) {
Christoph Lameter0a2966b2006-09-25 23:30:51 -07003171 cpu_clear(cpu_of(busiest), cpus);
3172 if (!cpus_empty(cpus))
3173 goto redo;
3174 }
Nick Piggind6d5cfa2005-09-10 00:26:16 -07003175 }
3176
Peter Williams43010652007-08-09 11:16:46 +02003177 if (!ld_moved) {
Ingo Molnard15bcfd2007-07-09 18:51:57 +02003178 schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]);
Siddha, Suresh B89c47102006-10-03 01:14:09 -07003179 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
3180 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
Nick Piggin5969fe02005-09-10 00:26:19 -07003181 return -1;
3182 } else
Nick Piggin16cfb1c2005-06-25 14:57:08 -07003183 sd->nr_balance_failed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184
Peter Williams43010652007-08-09 11:16:46 +02003185 return ld_moved;
Nick Piggin16cfb1c2005-06-25 14:57:08 -07003186
3187out_balanced:
Ingo Molnard15bcfd2007-07-09 18:51:57 +02003188 schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]);
Ingo Molnar48f24c42006-07-03 00:25:40 -07003189 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
Siddha, Suresh B89c47102006-10-03 01:14:09 -07003190 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
Nick Piggin5969fe02005-09-10 00:26:19 -07003191 return -1;
Nick Piggin16cfb1c2005-06-25 14:57:08 -07003192 sd->nr_balance_failed = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07003193
Nick Piggin16cfb1c2005-06-25 14:57:08 -07003194 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195}
3196
3197/*
3198 * idle_balance is called by schedule() if this_cpu is about to become
3199 * idle. Attempts to pull tasks from other CPUs.
3200 */
Ingo Molnar70b97a72006-07-03 00:25:42 -07003201static void idle_balance(int this_cpu, struct rq *this_rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202{
3203 struct sched_domain *sd;
Ingo Molnardd41f592007-07-09 18:51:59 +02003204 int pulled_task = -1;
3205 unsigned long next_balance = jiffies + HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206
3207 for_each_domain(this_cpu, sd) {
Christoph Lameter92c4ca52007-06-23 17:16:33 -07003208 unsigned long interval;
3209
3210 if (!(sd->flags & SD_LOAD_BALANCE))
3211 continue;
3212
3213 if (sd->flags & SD_BALANCE_NEWIDLE)
Ingo Molnar48f24c42006-07-03 00:25:40 -07003214 /* If we've pulled tasks over stop searching: */
Christoph Lameter1bd77f22006-12-10 02:20:27 -08003215 pulled_task = load_balance_newidle(this_cpu,
Christoph Lameter92c4ca52007-06-23 17:16:33 -07003216 this_rq, sd);
3217
3218 interval = msecs_to_jiffies(sd->balance_interval);
3219 if (time_after(next_balance, sd->last_balance + interval))
3220 next_balance = sd->last_balance + interval;
3221 if (pulled_task)
3222 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223 }
Ingo Molnardd41f592007-07-09 18:51:59 +02003224 if (pulled_task || time_after(jiffies, this_rq->next_balance)) {
Christoph Lameter1bd77f22006-12-10 02:20:27 -08003225 /*
3226 * We are going idle. next_balance may be set based on
3227 * a busy processor. So reset next_balance.
3228 */
3229 this_rq->next_balance = next_balance;
Ingo Molnardd41f592007-07-09 18:51:59 +02003230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231}
3232
3233/*
3234 * active_load_balance is run by migration threads. It pushes running tasks
3235 * off the busiest CPU onto idle CPUs. It requires at least 1 task to be
3236 * running on each physical CPU where possible, and avoids physical /
3237 * logical imbalances.
3238 *
3239 * Called with busiest_rq locked.
3240 */
Ingo Molnar70b97a72006-07-03 00:25:42 -07003241static void active_load_balance(struct rq *busiest_rq, int busiest_cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242{
Nick Piggin39507452005-06-25 14:57:09 -07003243 int target_cpu = busiest_rq->push_cpu;
Ingo Molnar70b97a72006-07-03 00:25:42 -07003244 struct sched_domain *sd;
3245 struct rq *target_rq;
Nick Piggin39507452005-06-25 14:57:09 -07003246
Ingo Molnar48f24c42006-07-03 00:25:40 -07003247 /* Is there any task to move? */
Nick Piggin39507452005-06-25 14:57:09 -07003248 if (busiest_rq->nr_running <= 1)
Nick Piggin39507452005-06-25 14:57:09 -07003249 return;
3250
3251 target_rq = cpu_rq(target_cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252
3253 /*
Nick Piggin39507452005-06-25 14:57:09 -07003254 * This condition is "impossible", if it occurs
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003255 * we need to fix it. Originally reported by
Nick Piggin39507452005-06-25 14:57:09 -07003256 * Bjorn Helgaas on a 128-cpu setup.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257 */
Nick Piggin39507452005-06-25 14:57:09 -07003258 BUG_ON(busiest_rq == target_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259
Nick Piggin39507452005-06-25 14:57:09 -07003260 /* move a task from busiest_rq to target_rq */
3261 double_lock_balance(busiest_rq, target_rq);
Ingo Molnar6e82a3b2007-08-09 11:16:51 +02003262 update_rq_clock(busiest_rq);
3263 update_rq_clock(target_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264
Nick Piggin39507452005-06-25 14:57:09 -07003265 /* Search for an sd spanning us and the target CPU. */
Chen, Kenneth Wc96d1452006-06-27 02:54:28 -07003266 for_each_domain(target_cpu, sd) {
Nick Piggin39507452005-06-25 14:57:09 -07003267 if ((sd->flags & SD_LOAD_BALANCE) &&
Ingo Molnar48f24c42006-07-03 00:25:40 -07003268 cpu_isset(busiest_cpu, sd->span))
Nick Piggin39507452005-06-25 14:57:09 -07003269 break;
Chen, Kenneth Wc96d1452006-06-27 02:54:28 -07003270 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271
Ingo Molnar48f24c42006-07-03 00:25:40 -07003272 if (likely(sd)) {
Ingo Molnar2d723762007-10-15 17:00:12 +02003273 schedstat_inc(sd, alb_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274
Peter Williams43010652007-08-09 11:16:46 +02003275 if (move_one_task(target_rq, target_cpu, busiest_rq,
3276 sd, CPU_IDLE))
Ingo Molnar48f24c42006-07-03 00:25:40 -07003277 schedstat_inc(sd, alb_pushed);
3278 else
3279 schedstat_inc(sd, alb_failed);
3280 }
Nick Piggin39507452005-06-25 14:57:09 -07003281 spin_unlock(&target_rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282}
3283
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003284#ifdef CONFIG_NO_HZ
3285static struct {
3286 atomic_t load_balancer;
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003287 cpumask_t cpu_mask;
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003288} nohz ____cacheline_aligned = {
3289 .load_balancer = ATOMIC_INIT(-1),
3290 .cpu_mask = CPU_MASK_NONE,
3291};
3292
Christoph Lameter7835b982006-12-10 02:20:22 -08003293/*
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003294 * This routine will try to nominate the ilb (idle load balancing)
3295 * owner among the cpus whose ticks are stopped. ilb owner will do the idle
3296 * load balancing on behalf of all those cpus. If all the cpus in the system
3297 * go into this tickless mode, then there will be no ilb owner (as there is
3298 * no need for one) and all the cpus will sleep till the next wakeup event
3299 * arrives...
Christoph Lameter7835b982006-12-10 02:20:22 -08003300 *
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003301 * For the ilb owner, tick is not stopped. And this tick will be used
3302 * for idle load balancing. ilb owner will still be part of
3303 * nohz.cpu_mask..
3304 *
3305 * While stopping the tick, this cpu will become the ilb owner if there
3306 * is no other owner. And will be the owner till that cpu becomes busy
3307 * or if all cpus in the system stop their ticks at which point
3308 * there is no need for ilb owner.
3309 *
3310 * When the ilb owner becomes busy, it nominates another owner, during the
3311 * next busy scheduler_tick()
3312 */
3313int select_nohz_load_balancer(int stop_tick)
3314{
3315 int cpu = smp_processor_id();
3316
3317 if (stop_tick) {
3318 cpu_set(cpu, nohz.cpu_mask);
3319 cpu_rq(cpu)->in_nohz_recently = 1;
3320
3321 /*
3322 * If we are going offline and still the leader, give up!
3323 */
3324 if (cpu_is_offline(cpu) &&
3325 atomic_read(&nohz.load_balancer) == cpu) {
3326 if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu)
3327 BUG();
3328 return 0;
3329 }
3330
3331 /* time for ilb owner also to sleep */
3332 if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) {
3333 if (atomic_read(&nohz.load_balancer) == cpu)
3334 atomic_set(&nohz.load_balancer, -1);
3335 return 0;
3336 }
3337
3338 if (atomic_read(&nohz.load_balancer) == -1) {
3339 /* make me the ilb owner */
3340 if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1)
3341 return 1;
3342 } else if (atomic_read(&nohz.load_balancer) == cpu)
3343 return 1;
3344 } else {
3345 if (!cpu_isset(cpu, nohz.cpu_mask))
3346 return 0;
3347
3348 cpu_clear(cpu, nohz.cpu_mask);
3349
3350 if (atomic_read(&nohz.load_balancer) == cpu)
3351 if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu)
3352 BUG();
3353 }
3354 return 0;
3355}
3356#endif
3357
3358static DEFINE_SPINLOCK(balancing);
3359
3360/*
Christoph Lameter7835b982006-12-10 02:20:22 -08003361 * It checks each scheduling domain to see if it is due to be balanced,
3362 * and initiates a balancing operation if so.
3363 *
3364 * Balancing parameters are set up in arch_init_sched_domains.
3365 */
Alexey Dobriyana9957442007-10-15 17:00:13 +02003366static void rebalance_domains(int cpu, enum cpu_idle_type idle)
Christoph Lameter7835b982006-12-10 02:20:22 -08003367{
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003368 int balance = 1;
3369 struct rq *rq = cpu_rq(cpu);
Christoph Lameter7835b982006-12-10 02:20:22 -08003370 unsigned long interval;
3371 struct sched_domain *sd;
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003372 /* Earliest time when we have to do rebalance again */
Christoph Lameterc9819f42006-12-10 02:20:25 -08003373 unsigned long next_balance = jiffies + 60*HZ;
Suresh Siddhaf549da82007-08-23 15:18:02 +02003374 int update_next_balance = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003376 for_each_domain(cpu, sd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377 if (!(sd->flags & SD_LOAD_BALANCE))
3378 continue;
3379
3380 interval = sd->balance_interval;
Ingo Molnard15bcfd2007-07-09 18:51:57 +02003381 if (idle != CPU_IDLE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382 interval *= sd->busy_factor;
3383
3384 /* scale ms to jiffies */
3385 interval = msecs_to_jiffies(interval);
3386 if (unlikely(!interval))
3387 interval = 1;
Ingo Molnardd41f592007-07-09 18:51:59 +02003388 if (interval > HZ*NR_CPUS/10)
3389 interval = HZ*NR_CPUS/10;
3390
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391
Christoph Lameter08c183f2006-12-10 02:20:29 -08003392 if (sd->flags & SD_SERIALIZE) {
3393 if (!spin_trylock(&balancing))
3394 goto out;
3395 }
3396
Christoph Lameterc9819f42006-12-10 02:20:25 -08003397 if (time_after_eq(jiffies, sd->last_balance + interval)) {
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003398 if (load_balance(cpu, rq, sd, idle, &balance)) {
Siddha, Suresh Bfa3b6dd2005-09-10 00:26:21 -07003399 /*
3400 * We've pulled tasks over so either we're no
Nick Piggin5969fe02005-09-10 00:26:19 -07003401 * longer idle, or one of our SMT siblings is
3402 * not idle.
3403 */
Ingo Molnard15bcfd2007-07-09 18:51:57 +02003404 idle = CPU_NOT_IDLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405 }
Christoph Lameter1bd77f22006-12-10 02:20:27 -08003406 sd->last_balance = jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407 }
Christoph Lameter08c183f2006-12-10 02:20:29 -08003408 if (sd->flags & SD_SERIALIZE)
3409 spin_unlock(&balancing);
3410out:
Suresh Siddhaf549da82007-08-23 15:18:02 +02003411 if (time_after(next_balance, sd->last_balance + interval)) {
Christoph Lameterc9819f42006-12-10 02:20:25 -08003412 next_balance = sd->last_balance + interval;
Suresh Siddhaf549da82007-08-23 15:18:02 +02003413 update_next_balance = 1;
3414 }
Siddha, Suresh B783609c2006-12-10 02:20:33 -08003415
3416 /*
3417 * Stop the load balance at this level. There is another
3418 * CPU in our sched group which is doing load balancing more
3419 * actively.
3420 */
3421 if (!balance)
3422 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423 }
Suresh Siddhaf549da82007-08-23 15:18:02 +02003424
3425 /*
3426 * next_balance will be updated only when there is a need.
3427 * When the cpu is attached to null domain for ex, it will not be
3428 * updated.
3429 */
3430 if (likely(update_next_balance))
3431 rq->next_balance = next_balance;
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003432}
3433
3434/*
3435 * run_rebalance_domains is triggered when needed from the scheduler tick.
3436 * In CONFIG_NO_HZ case, the idle load balance owner will do the
3437 * rebalancing for all the cpus for whom scheduler ticks are stopped.
3438 */
3439static void run_rebalance_domains(struct softirq_action *h)
3440{
Ingo Molnardd41f592007-07-09 18:51:59 +02003441 int this_cpu = smp_processor_id();
3442 struct rq *this_rq = cpu_rq(this_cpu);
3443 enum cpu_idle_type idle = this_rq->idle_at_tick ?
3444 CPU_IDLE : CPU_NOT_IDLE;
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003445
Ingo Molnardd41f592007-07-09 18:51:59 +02003446 rebalance_domains(this_cpu, idle);
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003447
3448#ifdef CONFIG_NO_HZ
3449 /*
3450 * If this cpu is the owner for idle load balancing, then do the
3451 * balancing on behalf of the other idle cpus whose ticks are
3452 * stopped.
3453 */
Ingo Molnardd41f592007-07-09 18:51:59 +02003454 if (this_rq->idle_at_tick &&
3455 atomic_read(&nohz.load_balancer) == this_cpu) {
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003456 cpumask_t cpus = nohz.cpu_mask;
3457 struct rq *rq;
3458 int balance_cpu;
3459
Ingo Molnardd41f592007-07-09 18:51:59 +02003460 cpu_clear(this_cpu, cpus);
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003461 for_each_cpu_mask(balance_cpu, cpus) {
3462 /*
3463 * If this cpu gets work to do, stop the load balancing
3464 * work being done for other cpus. Next load
3465 * balancing owner will pick it up.
3466 */
3467 if (need_resched())
3468 break;
3469
Oleg Nesterovde0cf892007-08-12 18:08:19 +02003470 rebalance_domains(balance_cpu, CPU_IDLE);
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003471
3472 rq = cpu_rq(balance_cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02003473 if (time_after(this_rq->next_balance, rq->next_balance))
3474 this_rq->next_balance = rq->next_balance;
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003475 }
3476 }
3477#endif
3478}
3479
3480/*
3481 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
3482 *
3483 * In case of CONFIG_NO_HZ, this is the place where we nominate a new
3484 * idle load balancing owner or decide to stop the periodic load balancing,
3485 * if the whole system is idle.
3486 */
Ingo Molnardd41f592007-07-09 18:51:59 +02003487static inline void trigger_load_balance(struct rq *rq, int cpu)
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003488{
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -07003489#ifdef CONFIG_NO_HZ
3490 /*
3491 * If we were in the nohz mode recently and busy at the current
3492 * scheduler tick, then check if we need to nominate new idle
3493 * load balancer.
3494 */
3495 if (rq->in_nohz_recently && !rq->idle_at_tick) {
3496 rq->in_nohz_recently = 0;
3497
3498 if (atomic_read(&nohz.load_balancer) == cpu) {
3499 cpu_clear(cpu, nohz.cpu_mask);
3500 atomic_set(&nohz.load_balancer, -1);
3501 }
3502
3503 if (atomic_read(&nohz.load_balancer) == -1) {
3504 /*
3505 * simple selection for now: Nominate the
3506 * first cpu in the nohz list to be the next
3507 * ilb owner.
3508 *
3509 * TBD: Traverse the sched domains and nominate
3510 * the nearest cpu in the nohz.cpu_mask.
3511 */
3512 int ilb = first_cpu(nohz.cpu_mask);
3513
3514 if (ilb != NR_CPUS)
3515 resched_cpu(ilb);
3516 }
3517 }
3518
3519 /*
3520 * If this cpu is idle and doing idle load balancing for all the
3521 * cpus with ticks stopped, is it time for that to stop?
3522 */
3523 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu &&
3524 cpus_weight(nohz.cpu_mask) == num_online_cpus()) {
3525 resched_cpu(cpu);
3526 return;
3527 }
3528
3529 /*
3530 * If this cpu is idle and the idle load balancing is done by
3531 * someone else, then no need raise the SCHED_SOFTIRQ
3532 */
3533 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu &&
3534 cpu_isset(cpu, nohz.cpu_mask))
3535 return;
3536#endif
3537 if (time_after_eq(jiffies, rq->next_balance))
3538 raise_softirq(SCHED_SOFTIRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003539}
Ingo Molnardd41f592007-07-09 18:51:59 +02003540
3541#else /* CONFIG_SMP */
3542
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543/*
3544 * on UP we do not need to balance between CPUs:
3545 */
Ingo Molnar70b97a72006-07-03 00:25:42 -07003546static inline void idle_balance(int cpu, struct rq *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547{
3548}
Ingo Molnardd41f592007-07-09 18:51:59 +02003549
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550#endif
3551
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552DEFINE_PER_CPU(struct kernel_stat, kstat);
3553
3554EXPORT_PER_CPU_SYMBOL(kstat);
3555
3556/*
Ingo Molnar41b86e92007-07-09 18:51:58 +02003557 * Return p->sum_exec_runtime plus any more ns on the sched_clock
3558 * that have not yet been banked in case the task is currently running.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003559 */
Ingo Molnar41b86e92007-07-09 18:51:58 +02003560unsigned long long task_sched_runtime(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 unsigned long flags;
Ingo Molnar41b86e92007-07-09 18:51:58 +02003563 u64 ns, delta_exec;
3564 struct rq *rq;
Ingo Molnar48f24c42006-07-03 00:25:40 -07003565
Ingo Molnar41b86e92007-07-09 18:51:58 +02003566 rq = task_rq_lock(p, &flags);
3567 ns = p->se.sum_exec_runtime;
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01003568 if (task_current(rq, p)) {
Ingo Molnara8e504d2007-08-09 11:16:47 +02003569 update_rq_clock(rq);
3570 delta_exec = rq->clock - p->se.exec_start;
Ingo Molnar41b86e92007-07-09 18:51:58 +02003571 if ((s64)delta_exec > 0)
3572 ns += delta_exec;
3573 }
3574 task_rq_unlock(rq, &flags);
Ingo Molnar48f24c42006-07-03 00:25:40 -07003575
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576 return ns;
3577}
3578
3579/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580 * Account user cpu time to a process.
3581 * @p: the process that the cpu time gets accounted to
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582 * @cputime: the cpu time spent in user space since the last update
3583 */
3584void account_user_time(struct task_struct *p, cputime_t cputime)
3585{
3586 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3587 cputime64_t tmp;
3588
3589 p->utime = cputime_add(p->utime, cputime);
3590
3591 /* Add user time to cpustat. */
3592 tmp = cputime_to_cputime64(cputime);
3593 if (TASK_NICE(p) > 0)
3594 cpustat->nice = cputime64_add(cpustat->nice, tmp);
3595 else
3596 cpustat->user = cputime64_add(cpustat->user, tmp);
3597}
3598
3599/*
Laurent Vivier94886b82007-10-15 17:00:19 +02003600 * Account guest cpu time to a process.
3601 * @p: the process that the cpu time gets accounted to
3602 * @cputime: the cpu time spent in virtual machine since the last update
3603 */
Adrian Bunkf7402e02007-10-29 21:18:10 +01003604static void account_guest_time(struct task_struct *p, cputime_t cputime)
Laurent Vivier94886b82007-10-15 17:00:19 +02003605{
3606 cputime64_t tmp;
3607 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3608
3609 tmp = cputime_to_cputime64(cputime);
3610
3611 p->utime = cputime_add(p->utime, cputime);
3612 p->gtime = cputime_add(p->gtime, cputime);
3613
3614 cpustat->user = cputime64_add(cpustat->user, tmp);
3615 cpustat->guest = cputime64_add(cpustat->guest, tmp);
3616}
3617
3618/*
Michael Neulingc66f08b2007-10-18 03:06:34 -07003619 * Account scaled user cpu time to a process.
3620 * @p: the process that the cpu time gets accounted to
3621 * @cputime: the cpu time spent in user space since the last update
3622 */
3623void account_user_time_scaled(struct task_struct *p, cputime_t cputime)
3624{
3625 p->utimescaled = cputime_add(p->utimescaled, cputime);
3626}
3627
3628/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 * Account system cpu time to a process.
3630 * @p: the process that the cpu time gets accounted to
3631 * @hardirq_offset: the offset to subtract from hardirq_count()
3632 * @cputime: the cpu time spent in kernel space since the last update
3633 */
3634void account_system_time(struct task_struct *p, int hardirq_offset,
3635 cputime_t cputime)
3636{
3637 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
Ingo Molnar70b97a72006-07-03 00:25:42 -07003638 struct rq *rq = this_rq();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639 cputime64_t tmp;
3640
Christian Borntraeger97783852007-11-15 20:57:39 +01003641 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0))
3642 return account_guest_time(p, cputime);
Laurent Vivier94886b82007-10-15 17:00:19 +02003643
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644 p->stime = cputime_add(p->stime, cputime);
3645
3646 /* Add system time to cpustat. */
3647 tmp = cputime_to_cputime64(cputime);
3648 if (hardirq_count() - hardirq_offset)
3649 cpustat->irq = cputime64_add(cpustat->irq, tmp);
3650 else if (softirq_count())
3651 cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
Andrew Mortoncfb52852007-11-14 16:59:45 -08003652 else if (p != rq->idle)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003653 cpustat->system = cputime64_add(cpustat->system, tmp);
Andrew Mortoncfb52852007-11-14 16:59:45 -08003654 else if (atomic_read(&rq->nr_iowait) > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655 cpustat->iowait = cputime64_add(cpustat->iowait, tmp);
3656 else
3657 cpustat->idle = cputime64_add(cpustat->idle, tmp);
3658 /* Account for system time used */
3659 acct_update_integrals(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660}
3661
3662/*
Michael Neulingc66f08b2007-10-18 03:06:34 -07003663 * Account scaled system cpu time to a process.
3664 * @p: the process that the cpu time gets accounted to
3665 * @hardirq_offset: the offset to subtract from hardirq_count()
3666 * @cputime: the cpu time spent in kernel space since the last update
3667 */
3668void account_system_time_scaled(struct task_struct *p, cputime_t cputime)
3669{
3670 p->stimescaled = cputime_add(p->stimescaled, cputime);
3671}
3672
3673/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674 * Account for involuntary wait time.
3675 * @p: the process from which the cpu time has been stolen
3676 * @steal: the cpu time spent in involuntary wait
3677 */
3678void account_steal_time(struct task_struct *p, cputime_t steal)
3679{
3680 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3681 cputime64_t tmp = cputime_to_cputime64(steal);
Ingo Molnar70b97a72006-07-03 00:25:42 -07003682 struct rq *rq = this_rq();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003683
3684 if (p == rq->idle) {
3685 p->stime = cputime_add(p->stime, steal);
3686 if (atomic_read(&rq->nr_iowait) > 0)
3687 cpustat->iowait = cputime64_add(cpustat->iowait, tmp);
3688 else
3689 cpustat->idle = cputime64_add(cpustat->idle, tmp);
Andrew Mortoncfb52852007-11-14 16:59:45 -08003690 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691 cpustat->steal = cputime64_add(cpustat->steal, tmp);
3692}
3693
Christoph Lameter7835b982006-12-10 02:20:22 -08003694/*
3695 * This function gets called by the timer code, with HZ frequency.
3696 * We call it with interrupts disabled.
3697 *
3698 * It also gets called by the fork code, when changing the parent's
3699 * timeslices.
3700 */
3701void scheduler_tick(void)
3702{
Christoph Lameter7835b982006-12-10 02:20:22 -08003703 int cpu = smp_processor_id();
3704 struct rq *rq = cpu_rq(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02003705 struct task_struct *curr = rq->curr;
Ingo Molnar529c7722007-08-10 23:05:11 +02003706 u64 next_tick = rq->tick_timestamp + TICK_NSEC;
Christoph Lameter7835b982006-12-10 02:20:22 -08003707
Ingo Molnardd41f592007-07-09 18:51:59 +02003708 spin_lock(&rq->lock);
Ingo Molnar546fe3c2007-08-09 11:16:51 +02003709 __update_rq_clock(rq);
Ingo Molnar529c7722007-08-10 23:05:11 +02003710 /*
3711 * Let rq->clock advance by at least TICK_NSEC:
3712 */
Guillaume Chazaraincc203d22008-01-25 21:08:34 +01003713 if (unlikely(rq->clock < next_tick)) {
Ingo Molnar529c7722007-08-10 23:05:11 +02003714 rq->clock = next_tick;
Guillaume Chazaraincc203d22008-01-25 21:08:34 +01003715 rq->clock_underflows++;
3716 }
Ingo Molnar529c7722007-08-10 23:05:11 +02003717 rq->tick_timestamp = rq->clock;
Ingo Molnarf1a438d2007-08-09 11:16:45 +02003718 update_cpu_load(rq);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01003719 curr->sched_class->task_tick(rq, curr, 0);
3720 update_sched_rt_period(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02003721 spin_unlock(&rq->lock);
3722
Christoph Lametere418e1c2006-12-10 02:20:23 -08003723#ifdef CONFIG_SMP
Ingo Molnardd41f592007-07-09 18:51:59 +02003724 rq->idle_at_tick = idle_cpu(cpu);
3725 trigger_load_balance(rq, cpu);
Christoph Lametere418e1c2006-12-10 02:20:23 -08003726#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003727}
3728
Linus Torvalds1da177e2005-04-16 15:20:36 -07003729#if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT)
3730
Srinivasa Ds43627582008-02-23 15:24:04 -08003731void __kprobes add_preempt_count(int val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003732{
3733 /*
3734 * Underflow?
3735 */
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07003736 if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
3737 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738 preempt_count() += val;
3739 /*
3740 * Spinlock count overflowing soon?
3741 */
Miguel Ojeda Sandonis33859f72006-12-10 02:20:38 -08003742 DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
3743 PREEMPT_MASK - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003744}
3745EXPORT_SYMBOL(add_preempt_count);
3746
Srinivasa Ds43627582008-02-23 15:24:04 -08003747void __kprobes sub_preempt_count(int val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748{
3749 /*
3750 * Underflow?
3751 */
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07003752 if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3753 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754 /*
3755 * Is the spinlock portion underflowing?
3756 */
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07003757 if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
3758 !(preempt_count() & PREEMPT_MASK)))
3759 return;
3760
Linus Torvalds1da177e2005-04-16 15:20:36 -07003761 preempt_count() -= val;
3762}
3763EXPORT_SYMBOL(sub_preempt_count);
3764
3765#endif
3766
3767/*
Ingo Molnardd41f592007-07-09 18:51:59 +02003768 * Print scheduling while atomic bug:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769 */
Ingo Molnardd41f592007-07-09 18:51:59 +02003770static noinline void __schedule_bug(struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771{
Satyam Sharma838225b2007-10-24 18:23:50 +02003772 struct pt_regs *regs = get_irq_regs();
3773
3774 printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
3775 prev->comm, prev->pid, preempt_count());
3776
Ingo Molnardd41f592007-07-09 18:51:59 +02003777 debug_show_held_locks(prev);
3778 if (irqs_disabled())
3779 print_irqtrace_events(prev);
Satyam Sharma838225b2007-10-24 18:23:50 +02003780
3781 if (regs)
3782 show_regs(regs);
3783 else
3784 dump_stack();
Ingo Molnardd41f592007-07-09 18:51:59 +02003785}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786
Ingo Molnardd41f592007-07-09 18:51:59 +02003787/*
3788 * Various schedule()-time debugging checks and statistics:
3789 */
3790static inline void schedule_debug(struct task_struct *prev)
3791{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792 /*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003793 * Test if we are atomic. Since do_exit() needs to call into
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794 * schedule() atomically, we ignore that path for now.
3795 * Otherwise, whine if we are scheduling when we should not be.
3796 */
Ingo Molnardd41f592007-07-09 18:51:59 +02003797 if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state))
3798 __schedule_bug(prev);
3799
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800 profile_hit(SCHED_PROFILING, __builtin_return_address(0));
3801
Ingo Molnar2d723762007-10-15 17:00:12 +02003802 schedstat_inc(this_rq(), sched_count);
Ingo Molnarb8efb562007-10-15 17:00:10 +02003803#ifdef CONFIG_SCHEDSTATS
3804 if (unlikely(prev->lock_depth >= 0)) {
Ingo Molnar2d723762007-10-15 17:00:12 +02003805 schedstat_inc(this_rq(), bkl_count);
3806 schedstat_inc(prev, sched_info.bkl_count);
Ingo Molnarb8efb562007-10-15 17:00:10 +02003807 }
3808#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02003809}
3810
3811/*
3812 * Pick up the highest-prio task:
3813 */
3814static inline struct task_struct *
Ingo Molnarff95f3d2007-08-09 11:16:49 +02003815pick_next_task(struct rq *rq, struct task_struct *prev)
Ingo Molnardd41f592007-07-09 18:51:59 +02003816{
Ingo Molnar5522d5d2007-10-15 17:00:12 +02003817 const struct sched_class *class;
Ingo Molnardd41f592007-07-09 18:51:59 +02003818 struct task_struct *p;
3819
3820 /*
3821 * Optimization: we know that if all tasks are in
3822 * the fair class we can call that function directly:
3823 */
3824 if (likely(rq->nr_running == rq->cfs.nr_running)) {
Ingo Molnarfb8d4722007-08-09 11:16:48 +02003825 p = fair_sched_class.pick_next_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02003826 if (likely(p))
3827 return p;
3828 }
3829
3830 class = sched_class_highest;
3831 for ( ; ; ) {
Ingo Molnarfb8d4722007-08-09 11:16:48 +02003832 p = class->pick_next_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02003833 if (p)
3834 return p;
3835 /*
3836 * Will never be NULL as the idle class always
3837 * returns a non-NULL p:
3838 */
3839 class = class->next;
3840 }
3841}
3842
3843/*
3844 * schedule() is the main scheduler function.
3845 */
3846asmlinkage void __sched schedule(void)
3847{
3848 struct task_struct *prev, *next;
Harvey Harrison67ca7bd2008-02-15 09:56:36 -08003849 unsigned long *switch_count;
Ingo Molnardd41f592007-07-09 18:51:59 +02003850 struct rq *rq;
Ingo Molnardd41f592007-07-09 18:51:59 +02003851 int cpu;
3852
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853need_resched:
3854 preempt_disable();
Ingo Molnardd41f592007-07-09 18:51:59 +02003855 cpu = smp_processor_id();
3856 rq = cpu_rq(cpu);
3857 rcu_qsctr_inc(cpu);
3858 prev = rq->curr;
3859 switch_count = &prev->nivcsw;
3860
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861 release_kernel_lock(prev);
3862need_resched_nonpreemptible:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863
Ingo Molnardd41f592007-07-09 18:51:59 +02003864 schedule_debug(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003865
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003866 hrtick_clear(rq);
3867
Ingo Molnar1e819952007-10-15 17:00:13 +02003868 /*
3869 * Do the rq-clock update outside the rq lock:
3870 */
3871 local_irq_disable();
Ingo Molnarc1b3da32007-08-09 11:16:47 +02003872 __update_rq_clock(rq);
Ingo Molnar1e819952007-10-15 17:00:13 +02003873 spin_lock(&rq->lock);
3874 clear_tsk_need_resched(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875
Ingo Molnardd41f592007-07-09 18:51:59 +02003876 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
3877 if (unlikely((prev->state & TASK_INTERRUPTIBLE) &&
3878 unlikely(signal_pending(prev)))) {
3879 prev->state = TASK_RUNNING;
3880 } else {
Ingo Molnar2e1cb742007-08-09 11:16:49 +02003881 deactivate_task(rq, prev, 1);
Ingo Molnardd41f592007-07-09 18:51:59 +02003882 }
3883 switch_count = &prev->nvcsw;
3884 }
3885
Steven Rostedt9a897c52008-01-25 21:08:22 +01003886#ifdef CONFIG_SMP
3887 if (prev->sched_class->pre_schedule)
3888 prev->sched_class->pre_schedule(rq, prev);
3889#endif
Steven Rostedtf65eda42008-01-25 21:08:07 +01003890
Ingo Molnardd41f592007-07-09 18:51:59 +02003891 if (unlikely(!rq->nr_running))
3892 idle_balance(cpu, rq);
3893
Ingo Molnar31ee5292007-08-09 11:16:49 +02003894 prev->sched_class->put_prev_task(rq, prev);
Ingo Molnarff95f3d2007-08-09 11:16:49 +02003895 next = pick_next_task(rq, prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896
3897 sched_info_switch(prev, next);
Ingo Molnardd41f592007-07-09 18:51:59 +02003898
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899 if (likely(prev != next)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003900 rq->nr_switches++;
3901 rq->curr = next;
3902 ++*switch_count;
3903
Ingo Molnardd41f592007-07-09 18:51:59 +02003904 context_switch(rq, prev, next); /* unlocks the rq */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003905 /*
3906 * the context switch might have flipped the stack from under
3907 * us, hence refresh the local variables.
3908 */
3909 cpu = smp_processor_id();
3910 rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911 } else
3912 spin_unlock_irq(&rq->lock);
3913
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003914 hrtick_set(rq);
3915
3916 if (unlikely(reacquire_kernel_lock(current) < 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003917 goto need_resched_nonpreemptible;
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003918
Linus Torvalds1da177e2005-04-16 15:20:36 -07003919 preempt_enable_no_resched();
3920 if (unlikely(test_thread_flag(TIF_NEED_RESCHED)))
3921 goto need_resched;
3922}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923EXPORT_SYMBOL(schedule);
3924
3925#ifdef CONFIG_PREEMPT
3926/*
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003927 * this is the entry point to schedule() from in-kernel preemption
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003928 * off of preempt_enable. Kernel preemptions off return from interrupt
Linus Torvalds1da177e2005-04-16 15:20:36 -07003929 * occur there and call schedule directly.
3930 */
3931asmlinkage void __sched preempt_schedule(void)
3932{
3933 struct thread_info *ti = current_thread_info();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003934 struct task_struct *task = current;
3935 int saved_lock_depth;
Ingo Molnar6478d882008-01-25 21:08:33 +01003936
Linus Torvalds1da177e2005-04-16 15:20:36 -07003937 /*
3938 * If there is a non-zero preempt_count or interrupts are disabled,
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003939 * we do not want to preempt the current task. Just return..
Linus Torvalds1da177e2005-04-16 15:20:36 -07003940 */
Nick Pigginbeed33a2006-10-11 01:21:52 -07003941 if (likely(ti->preempt_count || irqs_disabled()))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003942 return;
3943
Andi Kleen3a5c3592007-10-15 17:00:14 +02003944 do {
3945 add_preempt_count(PREEMPT_ACTIVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003946
Andi Kleen3a5c3592007-10-15 17:00:14 +02003947 /*
3948 * We keep the big kernel semaphore locked, but we
3949 * clear ->lock_depth so that schedule() doesnt
3950 * auto-release the semaphore:
3951 */
Andi Kleen3a5c3592007-10-15 17:00:14 +02003952 saved_lock_depth = task->lock_depth;
3953 task->lock_depth = -1;
Andi Kleen3a5c3592007-10-15 17:00:14 +02003954 schedule();
Andi Kleen3a5c3592007-10-15 17:00:14 +02003955 task->lock_depth = saved_lock_depth;
Andi Kleen3a5c3592007-10-15 17:00:14 +02003956 sub_preempt_count(PREEMPT_ACTIVE);
3957
3958 /*
3959 * Check again in case we missed a preemption opportunity
3960 * between schedule and now.
3961 */
3962 barrier();
3963 } while (unlikely(test_thread_flag(TIF_NEED_RESCHED)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003964}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965EXPORT_SYMBOL(preempt_schedule);
3966
3967/*
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003968 * this is the entry point to schedule() from kernel preemption
Linus Torvalds1da177e2005-04-16 15:20:36 -07003969 * off of irq context.
3970 * Note, that this is called and return with irqs disabled. This will
3971 * protect us against recursive calling from irq.
3972 */
3973asmlinkage void __sched preempt_schedule_irq(void)
3974{
3975 struct thread_info *ti = current_thread_info();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976 struct task_struct *task = current;
3977 int saved_lock_depth;
Ingo Molnar6478d882008-01-25 21:08:33 +01003978
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003979 /* Catch callers which need to be fixed */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003980 BUG_ON(ti->preempt_count || !irqs_disabled());
3981
Andi Kleen3a5c3592007-10-15 17:00:14 +02003982 do {
3983 add_preempt_count(PREEMPT_ACTIVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984
Andi Kleen3a5c3592007-10-15 17:00:14 +02003985 /*
3986 * We keep the big kernel semaphore locked, but we
3987 * clear ->lock_depth so that schedule() doesnt
3988 * auto-release the semaphore:
3989 */
Andi Kleen3a5c3592007-10-15 17:00:14 +02003990 saved_lock_depth = task->lock_depth;
3991 task->lock_depth = -1;
Andi Kleen3a5c3592007-10-15 17:00:14 +02003992 local_irq_enable();
3993 schedule();
3994 local_irq_disable();
Andi Kleen3a5c3592007-10-15 17:00:14 +02003995 task->lock_depth = saved_lock_depth;
Andi Kleen3a5c3592007-10-15 17:00:14 +02003996 sub_preempt_count(PREEMPT_ACTIVE);
3997
3998 /*
3999 * Check again in case we missed a preemption opportunity
4000 * between schedule and now.
4001 */
4002 barrier();
4003 } while (unlikely(test_thread_flag(TIF_NEED_RESCHED)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004004}
4005
4006#endif /* CONFIG_PREEMPT */
4007
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004008int default_wake_function(wait_queue_t *curr, unsigned mode, int sync,
4009 void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004010{
Ingo Molnar48f24c42006-07-03 00:25:40 -07004011 return try_to_wake_up(curr->private, mode, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004012}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004013EXPORT_SYMBOL(default_wake_function);
4014
4015/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01004016 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
4017 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018 * number) then we wake all the non-exclusive tasks and one exclusive task.
4019 *
4020 * There are circumstances in which we can try to wake a task which has already
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01004021 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022 * zero in this (rare) case, and we handle it by continuing to scan the queue.
4023 */
4024static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
4025 int nr_exclusive, int sync, void *key)
4026{
Matthias Kaehlcke2e458742007-10-15 17:00:02 +02004027 wait_queue_t *curr, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028
Matthias Kaehlcke2e458742007-10-15 17:00:02 +02004029 list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
Ingo Molnar48f24c42006-07-03 00:25:40 -07004030 unsigned flags = curr->flags;
4031
Linus Torvalds1da177e2005-04-16 15:20:36 -07004032 if (curr->func(curr, mode, sync, key) &&
Ingo Molnar48f24c42006-07-03 00:25:40 -07004033 (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004034 break;
4035 }
4036}
4037
4038/**
4039 * __wake_up - wake up threads blocked on a waitqueue.
4040 * @q: the waitqueue
4041 * @mode: which threads
4042 * @nr_exclusive: how many wake-one or wake-many threads to wake up
Martin Waitz67be2dd2005-05-01 08:59:26 -07004043 * @key: is directly passed to the wakeup function
Linus Torvalds1da177e2005-04-16 15:20:36 -07004044 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08004045void __wake_up(wait_queue_head_t *q, unsigned int mode,
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004046 int nr_exclusive, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004047{
4048 unsigned long flags;
4049
4050 spin_lock_irqsave(&q->lock, flags);
4051 __wake_up_common(q, mode, nr_exclusive, 0, key);
4052 spin_unlock_irqrestore(&q->lock, flags);
4053}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004054EXPORT_SYMBOL(__wake_up);
4055
4056/*
4057 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
4058 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08004059void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060{
4061 __wake_up_common(q, mode, 1, 0, NULL);
4062}
4063
4064/**
Martin Waitz67be2dd2005-05-01 08:59:26 -07004065 * __wake_up_sync - wake up threads blocked on a waitqueue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004066 * @q: the waitqueue
4067 * @mode: which threads
4068 * @nr_exclusive: how many wake-one or wake-many threads to wake up
4069 *
4070 * The sync wakeup differs that the waker knows that it will schedule
4071 * away soon, so while the target thread will be woken up, it will not
4072 * be migrated to another CPU - ie. the two threads are 'synchronized'
4073 * with each other. This can prevent needless bouncing between CPUs.
4074 *
4075 * On UP it can prevent extra preemption.
4076 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08004077void
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004078__wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004079{
4080 unsigned long flags;
4081 int sync = 1;
4082
4083 if (unlikely(!q))
4084 return;
4085
4086 if (unlikely(!nr_exclusive))
4087 sync = 0;
4088
4089 spin_lock_irqsave(&q->lock, flags);
4090 __wake_up_common(q, mode, nr_exclusive, sync, NULL);
4091 spin_unlock_irqrestore(&q->lock, flags);
4092}
4093EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
4094
Ingo Molnarb15136e2007-10-24 18:23:48 +02004095void complete(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004096{
4097 unsigned long flags;
4098
4099 spin_lock_irqsave(&x->wait.lock, flags);
4100 x->done++;
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05004101 __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004102 spin_unlock_irqrestore(&x->wait.lock, flags);
4103}
4104EXPORT_SYMBOL(complete);
4105
Ingo Molnarb15136e2007-10-24 18:23:48 +02004106void complete_all(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004107{
4108 unsigned long flags;
4109
4110 spin_lock_irqsave(&x->wait.lock, flags);
4111 x->done += UINT_MAX/2;
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05004112 __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004113 spin_unlock_irqrestore(&x->wait.lock, flags);
4114}
4115EXPORT_SYMBOL(complete_all);
4116
Andi Kleen8cbbe862007-10-15 17:00:14 +02004117static inline long __sched
4118do_wait_for_common(struct completion *x, long timeout, int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004119{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004120 if (!x->done) {
4121 DECLARE_WAITQUEUE(wait, current);
4122
4123 wait.flags |= WQ_FLAG_EXCLUSIVE;
4124 __add_wait_queue_tail(&x->wait, &wait);
4125 do {
Matthew Wilcox009e5772007-12-06 12:29:54 -05004126 if ((state == TASK_INTERRUPTIBLE &&
4127 signal_pending(current)) ||
4128 (state == TASK_KILLABLE &&
4129 fatal_signal_pending(current))) {
Andi Kleen8cbbe862007-10-15 17:00:14 +02004130 __remove_wait_queue(&x->wait, &wait);
4131 return -ERESTARTSYS;
4132 }
4133 __set_current_state(state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004134 spin_unlock_irq(&x->wait.lock);
Andi Kleen8cbbe862007-10-15 17:00:14 +02004135 timeout = schedule_timeout(timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004136 spin_lock_irq(&x->wait.lock);
Andi Kleen8cbbe862007-10-15 17:00:14 +02004137 if (!timeout) {
4138 __remove_wait_queue(&x->wait, &wait);
4139 return timeout;
4140 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004141 } while (!x->done);
4142 __remove_wait_queue(&x->wait, &wait);
4143 }
4144 x->done--;
Andi Kleen8cbbe862007-10-15 17:00:14 +02004145 return timeout;
4146}
4147
4148static long __sched
4149wait_for_common(struct completion *x, long timeout, int state)
4150{
4151 might_sleep();
4152
4153 spin_lock_irq(&x->wait.lock);
4154 timeout = do_wait_for_common(x, timeout, state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004155 spin_unlock_irq(&x->wait.lock);
Andi Kleen8cbbe862007-10-15 17:00:14 +02004156 return timeout;
4157}
4158
Ingo Molnarb15136e2007-10-24 18:23:48 +02004159void __sched wait_for_completion(struct completion *x)
Andi Kleen8cbbe862007-10-15 17:00:14 +02004160{
4161 wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162}
4163EXPORT_SYMBOL(wait_for_completion);
4164
Ingo Molnarb15136e2007-10-24 18:23:48 +02004165unsigned long __sched
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166wait_for_completion_timeout(struct completion *x, unsigned long timeout)
4167{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004168 return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004169}
4170EXPORT_SYMBOL(wait_for_completion_timeout);
4171
Andi Kleen8cbbe862007-10-15 17:00:14 +02004172int __sched wait_for_completion_interruptible(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004173{
Andi Kleen51e97992007-10-18 21:32:55 +02004174 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
4175 if (t == -ERESTARTSYS)
4176 return t;
4177 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004178}
4179EXPORT_SYMBOL(wait_for_completion_interruptible);
4180
Ingo Molnarb15136e2007-10-24 18:23:48 +02004181unsigned long __sched
Linus Torvalds1da177e2005-04-16 15:20:36 -07004182wait_for_completion_interruptible_timeout(struct completion *x,
4183 unsigned long timeout)
4184{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004185 return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004186}
4187EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
4188
Matthew Wilcox009e5772007-12-06 12:29:54 -05004189int __sched wait_for_completion_killable(struct completion *x)
4190{
4191 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
4192 if (t == -ERESTARTSYS)
4193 return t;
4194 return 0;
4195}
4196EXPORT_SYMBOL(wait_for_completion_killable);
4197
Andi Kleen8cbbe862007-10-15 17:00:14 +02004198static long __sched
4199sleep_on_common(wait_queue_head_t *q, int state, long timeout)
Ingo Molnar0fec1712007-07-09 18:52:01 +02004200{
4201 unsigned long flags;
4202 wait_queue_t wait;
4203
4204 init_waitqueue_entry(&wait, current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004205
Andi Kleen8cbbe862007-10-15 17:00:14 +02004206 __set_current_state(state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004207
Andi Kleen8cbbe862007-10-15 17:00:14 +02004208 spin_lock_irqsave(&q->lock, flags);
4209 __add_wait_queue(q, &wait);
4210 spin_unlock(&q->lock);
4211 timeout = schedule_timeout(timeout);
4212 spin_lock_irq(&q->lock);
4213 __remove_wait_queue(q, &wait);
4214 spin_unlock_irqrestore(&q->lock, flags);
4215
4216 return timeout;
4217}
4218
4219void __sched interruptible_sleep_on(wait_queue_head_t *q)
4220{
4221 sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004222}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004223EXPORT_SYMBOL(interruptible_sleep_on);
4224
Ingo Molnar0fec1712007-07-09 18:52:01 +02004225long __sched
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004226interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004227{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004228 return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004229}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004230EXPORT_SYMBOL(interruptible_sleep_on_timeout);
4231
Ingo Molnar0fec1712007-07-09 18:52:01 +02004232void __sched sleep_on(wait_queue_head_t *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004233{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004234 sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004235}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004236EXPORT_SYMBOL(sleep_on);
4237
Ingo Molnar0fec1712007-07-09 18:52:01 +02004238long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004239{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004240 return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004241}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004242EXPORT_SYMBOL(sleep_on_timeout);
4243
Ingo Molnarb29739f2006-06-27 02:54:51 -07004244#ifdef CONFIG_RT_MUTEXES
4245
4246/*
4247 * rt_mutex_setprio - set the current priority of a task
4248 * @p: task
4249 * @prio: prio value (kernel-internal form)
4250 *
4251 * This function changes the 'effective' priority of a task. It does
4252 * not touch ->normal_prio like __setscheduler().
4253 *
4254 * Used by the rt_mutex code to implement priority inheritance logic.
4255 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004256void rt_mutex_setprio(struct task_struct *p, int prio)
Ingo Molnarb29739f2006-06-27 02:54:51 -07004257{
4258 unsigned long flags;
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004259 int oldprio, on_rq, running;
Ingo Molnar70b97a72006-07-03 00:25:42 -07004260 struct rq *rq;
Steven Rostedtcb469842008-01-25 21:08:22 +01004261 const struct sched_class *prev_class = p->sched_class;
Ingo Molnarb29739f2006-06-27 02:54:51 -07004262
4263 BUG_ON(prio < 0 || prio > MAX_PRIO);
4264
4265 rq = task_rq_lock(p, &flags);
Ingo Molnara8e504d2007-08-09 11:16:47 +02004266 update_rq_clock(rq);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004267
Andrew Mortond5f9f942007-05-08 20:27:06 -07004268 oldprio = p->prio;
Ingo Molnardd41f592007-07-09 18:51:59 +02004269 on_rq = p->se.on_rq;
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01004270 running = task_current(rq, p);
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004271 if (on_rq) {
Ingo Molnar69be72c2007-08-09 11:16:49 +02004272 dequeue_task(rq, p, 0);
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004273 if (running)
4274 p->sched_class->put_prev_task(rq, p);
4275 }
Ingo Molnardd41f592007-07-09 18:51:59 +02004276
4277 if (rt_prio(prio))
4278 p->sched_class = &rt_sched_class;
4279 else
4280 p->sched_class = &fair_sched_class;
4281
Ingo Molnarb29739f2006-06-27 02:54:51 -07004282 p->prio = prio;
4283
Ingo Molnardd41f592007-07-09 18:51:59 +02004284 if (on_rq) {
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004285 if (running)
4286 p->sched_class->set_curr_task(rq);
Steven Rostedtcb469842008-01-25 21:08:22 +01004287
Ingo Molnar8159f872007-08-09 11:16:49 +02004288 enqueue_task(rq, p, 0);
Steven Rostedtcb469842008-01-25 21:08:22 +01004289
4290 check_class_changed(rq, p, prev_class, oldprio, running);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004291 }
4292 task_rq_unlock(rq, &flags);
4293}
4294
4295#endif
4296
Ingo Molnar36c8b582006-07-03 00:25:41 -07004297void set_user_nice(struct task_struct *p, long nice)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004298{
Ingo Molnardd41f592007-07-09 18:51:59 +02004299 int old_prio, delta, on_rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004300 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07004301 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004302
4303 if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
4304 return;
4305 /*
4306 * We have to be careful, if called from sys_setpriority(),
4307 * the task might be in the middle of scheduling on another CPU.
4308 */
4309 rq = task_rq_lock(p, &flags);
Ingo Molnara8e504d2007-08-09 11:16:47 +02004310 update_rq_clock(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004311 /*
4312 * The RT priorities are set via sched_setscheduler(), but we still
4313 * allow the 'normal' nice value to be set - but as expected
4314 * it wont have any effect on scheduling until the task is
Ingo Molnardd41f592007-07-09 18:51:59 +02004315 * SCHED_FIFO/SCHED_RR:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004316 */
Ingo Molnare05606d2007-07-09 18:51:59 +02004317 if (task_has_rt_policy(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004318 p->static_prio = NICE_TO_PRIO(nice);
4319 goto out_unlock;
4320 }
Ingo Molnardd41f592007-07-09 18:51:59 +02004321 on_rq = p->se.on_rq;
Peter Zijlstra62fb1852008-02-25 17:34:02 +01004322 if (on_rq) {
Ingo Molnar69be72c2007-08-09 11:16:49 +02004323 dequeue_task(rq, p, 0);
Peter Zijlstra62fb1852008-02-25 17:34:02 +01004324 dec_load(rq, p);
4325 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004326
Linus Torvalds1da177e2005-04-16 15:20:36 -07004327 p->static_prio = NICE_TO_PRIO(nice);
Peter Williams2dd73a42006-06-27 02:54:34 -07004328 set_load_weight(p);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004329 old_prio = p->prio;
4330 p->prio = effective_prio(p);
4331 delta = p->prio - old_prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004332
Ingo Molnardd41f592007-07-09 18:51:59 +02004333 if (on_rq) {
Ingo Molnar8159f872007-08-09 11:16:49 +02004334 enqueue_task(rq, p, 0);
Peter Zijlstra62fb1852008-02-25 17:34:02 +01004335 inc_load(rq, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004336 /*
Andrew Mortond5f9f942007-05-08 20:27:06 -07004337 * If the task increased its priority or is running and
4338 * lowered its priority, then reschedule its CPU:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004339 */
Andrew Mortond5f9f942007-05-08 20:27:06 -07004340 if (delta < 0 || (delta > 0 && task_running(rq, p)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004341 resched_task(rq->curr);
4342 }
4343out_unlock:
4344 task_rq_unlock(rq, &flags);
4345}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004346EXPORT_SYMBOL(set_user_nice);
4347
Matt Mackalle43379f2005-05-01 08:59:00 -07004348/*
4349 * can_nice - check if a task can reduce its nice value
4350 * @p: task
4351 * @nice: nice value
4352 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004353int can_nice(const struct task_struct *p, const int nice)
Matt Mackalle43379f2005-05-01 08:59:00 -07004354{
Matt Mackall024f4742005-08-18 11:24:19 -07004355 /* convert nice value [19,-20] to rlimit style value [1,40] */
4356 int nice_rlim = 20 - nice;
Ingo Molnar48f24c42006-07-03 00:25:40 -07004357
Matt Mackalle43379f2005-05-01 08:59:00 -07004358 return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur ||
4359 capable(CAP_SYS_NICE));
4360}
4361
Linus Torvalds1da177e2005-04-16 15:20:36 -07004362#ifdef __ARCH_WANT_SYS_NICE
4363
4364/*
4365 * sys_nice - change the priority of the current process.
4366 * @increment: priority increment
4367 *
4368 * sys_setpriority is a more generic, but much slower function that
4369 * does similar things.
4370 */
4371asmlinkage long sys_nice(int increment)
4372{
Ingo Molnar48f24c42006-07-03 00:25:40 -07004373 long nice, retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004374
4375 /*
4376 * Setpriority might change our priority at the same moment.
4377 * We don't have to worry. Conceptually one call occurs first
4378 * and we have a single winner.
4379 */
Matt Mackalle43379f2005-05-01 08:59:00 -07004380 if (increment < -40)
4381 increment = -40;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004382 if (increment > 40)
4383 increment = 40;
4384
4385 nice = PRIO_TO_NICE(current->static_prio) + increment;
4386 if (nice < -20)
4387 nice = -20;
4388 if (nice > 19)
4389 nice = 19;
4390
Matt Mackalle43379f2005-05-01 08:59:00 -07004391 if (increment < 0 && !can_nice(current, nice))
4392 return -EPERM;
4393
Linus Torvalds1da177e2005-04-16 15:20:36 -07004394 retval = security_task_setnice(current, nice);
4395 if (retval)
4396 return retval;
4397
4398 set_user_nice(current, nice);
4399 return 0;
4400}
4401
4402#endif
4403
4404/**
4405 * task_prio - return the priority value of a given task.
4406 * @p: the task in question.
4407 *
4408 * This is the priority value as seen by users in /proc.
4409 * RT tasks are offset by -200. Normal tasks are centered
4410 * around 0, value goes from -16 to +15.
4411 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004412int task_prio(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004413{
4414 return p->prio - MAX_RT_PRIO;
4415}
4416
4417/**
4418 * task_nice - return the nice value of a given task.
4419 * @p: the task in question.
4420 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004421int task_nice(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004422{
4423 return TASK_NICE(p);
4424}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004425EXPORT_SYMBOL_GPL(task_nice);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004426
4427/**
4428 * idle_cpu - is a given cpu idle currently?
4429 * @cpu: the processor in question.
4430 */
4431int idle_cpu(int cpu)
4432{
4433 return cpu_curr(cpu) == cpu_rq(cpu)->idle;
4434}
4435
Linus Torvalds1da177e2005-04-16 15:20:36 -07004436/**
4437 * idle_task - return the idle task for a given cpu.
4438 * @cpu: the processor in question.
4439 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004440struct task_struct *idle_task(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004441{
4442 return cpu_rq(cpu)->idle;
4443}
4444
4445/**
4446 * find_process_by_pid - find a process with a matching PID value.
4447 * @pid: the pid in question.
4448 */
Alexey Dobriyana9957442007-10-15 17:00:13 +02004449static struct task_struct *find_process_by_pid(pid_t pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004450{
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07004451 return pid ? find_task_by_vpid(pid) : current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004452}
4453
4454/* Actually do priority change: must hold rq lock. */
Ingo Molnardd41f592007-07-09 18:51:59 +02004455static void
4456__setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004457{
Ingo Molnardd41f592007-07-09 18:51:59 +02004458 BUG_ON(p->se.on_rq);
Ingo Molnar48f24c42006-07-03 00:25:40 -07004459
Linus Torvalds1da177e2005-04-16 15:20:36 -07004460 p->policy = policy;
Ingo Molnardd41f592007-07-09 18:51:59 +02004461 switch (p->policy) {
4462 case SCHED_NORMAL:
4463 case SCHED_BATCH:
4464 case SCHED_IDLE:
4465 p->sched_class = &fair_sched_class;
4466 break;
4467 case SCHED_FIFO:
4468 case SCHED_RR:
4469 p->sched_class = &rt_sched_class;
4470 break;
4471 }
4472
Linus Torvalds1da177e2005-04-16 15:20:36 -07004473 p->rt_priority = prio;
Ingo Molnarb29739f2006-06-27 02:54:51 -07004474 p->normal_prio = normal_prio(p);
4475 /* we are holding p->pi_lock already */
4476 p->prio = rt_mutex_getprio(p);
Peter Williams2dd73a42006-06-27 02:54:34 -07004477 set_load_weight(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004478}
4479
4480/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08004481 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004482 * @p: the task in question.
4483 * @policy: new policy.
4484 * @param: structure containing the new RT priority.
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004485 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08004486 * NOTE that the task may be already dead.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004487 */
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004488int sched_setscheduler(struct task_struct *p, int policy,
4489 struct sched_param *param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004490{
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004491 int retval, oldprio, oldpolicy = -1, on_rq, running;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004492 unsigned long flags;
Steven Rostedtcb469842008-01-25 21:08:22 +01004493 const struct sched_class *prev_class = p->sched_class;
Ingo Molnar70b97a72006-07-03 00:25:42 -07004494 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004495
Steven Rostedt66e53932006-06-27 02:54:44 -07004496 /* may grab non-irq protected spin_locks */
4497 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07004498recheck:
4499 /* double check policy once rq lock held */
4500 if (policy < 0)
4501 policy = oldpolicy = p->policy;
4502 else if (policy != SCHED_FIFO && policy != SCHED_RR &&
Ingo Molnardd41f592007-07-09 18:51:59 +02004503 policy != SCHED_NORMAL && policy != SCHED_BATCH &&
4504 policy != SCHED_IDLE)
Ingo Molnarb0a94992006-01-14 13:20:41 -08004505 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004506 /*
4507 * Valid priorities for SCHED_FIFO and SCHED_RR are
Ingo Molnardd41f592007-07-09 18:51:59 +02004508 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
4509 * SCHED_BATCH and SCHED_IDLE is 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004510 */
4511 if (param->sched_priority < 0 ||
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004512 (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
Steven Rostedtd46523e2005-07-25 16:28:39 -04004513 (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004514 return -EINVAL;
Ingo Molnare05606d2007-07-09 18:51:59 +02004515 if (rt_policy(policy) != (param->sched_priority != 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004516 return -EINVAL;
4517
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004518 /*
4519 * Allow unprivileged RT tasks to decrease priority:
4520 */
4521 if (!capable(CAP_SYS_NICE)) {
Ingo Molnare05606d2007-07-09 18:51:59 +02004522 if (rt_policy(policy)) {
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07004523 unsigned long rlim_rtprio;
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004524
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07004525 if (!lock_task_sighand(p, &flags))
4526 return -ESRCH;
4527 rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur;
4528 unlock_task_sighand(p, &flags);
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004529
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07004530 /* can't set/change the rt policy */
4531 if (policy != p->policy && !rlim_rtprio)
4532 return -EPERM;
4533
4534 /* can't increase priority */
4535 if (param->sched_priority > p->rt_priority &&
4536 param->sched_priority > rlim_rtprio)
4537 return -EPERM;
4538 }
Ingo Molnardd41f592007-07-09 18:51:59 +02004539 /*
4540 * Like positive nice levels, dont allow tasks to
4541 * move out of SCHED_IDLE either:
4542 */
4543 if (p->policy == SCHED_IDLE && policy != SCHED_IDLE)
4544 return -EPERM;
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07004545
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004546 /* can't change other user's priorities */
4547 if ((current->euid != p->euid) &&
4548 (current->euid != p->uid))
4549 return -EPERM;
4550 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004551
Peter Zijlstrab68aa232008-02-13 15:45:40 +01004552#ifdef CONFIG_RT_GROUP_SCHED
4553 /*
4554 * Do not allow realtime tasks into groups that have no runtime
4555 * assigned.
4556 */
4557 if (rt_policy(policy) && task_group(p)->rt_runtime == 0)
4558 return -EPERM;
4559#endif
4560
Linus Torvalds1da177e2005-04-16 15:20:36 -07004561 retval = security_task_setscheduler(p, policy, param);
4562 if (retval)
4563 return retval;
4564 /*
Ingo Molnarb29739f2006-06-27 02:54:51 -07004565 * make sure no PI-waiters arrive (or leave) while we are
4566 * changing the priority of the task:
4567 */
4568 spin_lock_irqsave(&p->pi_lock, flags);
4569 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004570 * To be able to change p->policy safely, the apropriate
4571 * runqueue lock must be held.
4572 */
Ingo Molnarb29739f2006-06-27 02:54:51 -07004573 rq = __task_rq_lock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004574 /* recheck policy now with rq lock held */
4575 if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
4576 policy = oldpolicy = -1;
Ingo Molnarb29739f2006-06-27 02:54:51 -07004577 __task_rq_unlock(rq);
4578 spin_unlock_irqrestore(&p->pi_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004579 goto recheck;
4580 }
Ingo Molnar2daa3572007-08-09 11:16:51 +02004581 update_rq_clock(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02004582 on_rq = p->se.on_rq;
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01004583 running = task_current(rq, p);
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004584 if (on_rq) {
Ingo Molnar2e1cb742007-08-09 11:16:49 +02004585 deactivate_task(rq, p, 0);
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004586 if (running)
4587 p->sched_class->put_prev_task(rq, p);
4588 }
Dmitry Adamushkof6b53202007-10-15 17:00:08 +02004589
Linus Torvalds1da177e2005-04-16 15:20:36 -07004590 oldprio = p->prio;
Ingo Molnardd41f592007-07-09 18:51:59 +02004591 __setscheduler(rq, p, policy, param->sched_priority);
Dmitry Adamushkof6b53202007-10-15 17:00:08 +02004592
Ingo Molnardd41f592007-07-09 18:51:59 +02004593 if (on_rq) {
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004594 if (running)
4595 p->sched_class->set_curr_task(rq);
Steven Rostedtcb469842008-01-25 21:08:22 +01004596
Ingo Molnardd41f592007-07-09 18:51:59 +02004597 activate_task(rq, p, 0);
Steven Rostedtcb469842008-01-25 21:08:22 +01004598
4599 check_class_changed(rq, p, prev_class, oldprio, running);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004600 }
Ingo Molnarb29739f2006-06-27 02:54:51 -07004601 __task_rq_unlock(rq);
4602 spin_unlock_irqrestore(&p->pi_lock, flags);
4603
Thomas Gleixner95e02ca2006-06-27 02:55:02 -07004604 rt_mutex_adjust_pi(p);
4605
Linus Torvalds1da177e2005-04-16 15:20:36 -07004606 return 0;
4607}
4608EXPORT_SYMBOL_GPL(sched_setscheduler);
4609
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004610static int
4611do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004612{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004613 struct sched_param lparam;
4614 struct task_struct *p;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004615 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004616
4617 if (!param || pid < 0)
4618 return -EINVAL;
4619 if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
4620 return -EFAULT;
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004621
4622 rcu_read_lock();
4623 retval = -ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004624 p = find_process_by_pid(pid);
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004625 if (p != NULL)
4626 retval = sched_setscheduler(p, policy, &lparam);
4627 rcu_read_unlock();
Ingo Molnar36c8b582006-07-03 00:25:41 -07004628
Linus Torvalds1da177e2005-04-16 15:20:36 -07004629 return retval;
4630}
4631
4632/**
4633 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
4634 * @pid: the pid in question.
4635 * @policy: new policy.
4636 * @param: structure containing the new RT priority.
4637 */
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01004638asmlinkage long
4639sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004640{
Jason Baronc21761f2006-01-18 17:43:03 -08004641 /* negative values for policy are not valid */
4642 if (policy < 0)
4643 return -EINVAL;
4644
Linus Torvalds1da177e2005-04-16 15:20:36 -07004645 return do_sched_setscheduler(pid, policy, param);
4646}
4647
4648/**
4649 * sys_sched_setparam - set/change the RT priority of a thread
4650 * @pid: the pid in question.
4651 * @param: structure containing the new RT priority.
4652 */
4653asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param)
4654{
4655 return do_sched_setscheduler(pid, -1, param);
4656}
4657
4658/**
4659 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
4660 * @pid: the pid in question.
4661 */
4662asmlinkage long sys_sched_getscheduler(pid_t pid)
4663{
Ingo Molnar36c8b582006-07-03 00:25:41 -07004664 struct task_struct *p;
Andi Kleen3a5c3592007-10-15 17:00:14 +02004665 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004666
4667 if (pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02004668 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004669
4670 retval = -ESRCH;
4671 read_lock(&tasklist_lock);
4672 p = find_process_by_pid(pid);
4673 if (p) {
4674 retval = security_task_getscheduler(p);
4675 if (!retval)
4676 retval = p->policy;
4677 }
4678 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004679 return retval;
4680}
4681
4682/**
4683 * sys_sched_getscheduler - get the RT priority of a thread
4684 * @pid: the pid in question.
4685 * @param: structure containing the RT priority.
4686 */
4687asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param)
4688{
4689 struct sched_param lp;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004690 struct task_struct *p;
Andi Kleen3a5c3592007-10-15 17:00:14 +02004691 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004692
4693 if (!param || pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02004694 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004695
4696 read_lock(&tasklist_lock);
4697 p = find_process_by_pid(pid);
4698 retval = -ESRCH;
4699 if (!p)
4700 goto out_unlock;
4701
4702 retval = security_task_getscheduler(p);
4703 if (retval)
4704 goto out_unlock;
4705
4706 lp.sched_priority = p->rt_priority;
4707 read_unlock(&tasklist_lock);
4708
4709 /*
4710 * This one might sleep, we cannot do it with a spinlock held ...
4711 */
4712 retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
4713
Linus Torvalds1da177e2005-04-16 15:20:36 -07004714 return retval;
4715
4716out_unlock:
4717 read_unlock(&tasklist_lock);
4718 return retval;
4719}
4720
4721long sched_setaffinity(pid_t pid, cpumask_t new_mask)
4722{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004723 cpumask_t cpus_allowed;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004724 struct task_struct *p;
4725 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004726
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004727 get_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004728 read_lock(&tasklist_lock);
4729
4730 p = find_process_by_pid(pid);
4731 if (!p) {
4732 read_unlock(&tasklist_lock);
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004733 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004734 return -ESRCH;
4735 }
4736
4737 /*
4738 * It is not safe to call set_cpus_allowed with the
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01004739 * tasklist_lock held. We will bump the task_struct's
Linus Torvalds1da177e2005-04-16 15:20:36 -07004740 * usage count and then drop tasklist_lock.
4741 */
4742 get_task_struct(p);
4743 read_unlock(&tasklist_lock);
4744
4745 retval = -EPERM;
4746 if ((current->euid != p->euid) && (current->euid != p->uid) &&
4747 !capable(CAP_SYS_NICE))
4748 goto out_unlock;
4749
David Quigleye7834f82006-06-23 02:03:59 -07004750 retval = security_task_setscheduler(p, 0, NULL);
4751 if (retval)
4752 goto out_unlock;
4753
Linus Torvalds1da177e2005-04-16 15:20:36 -07004754 cpus_allowed = cpuset_cpus_allowed(p);
4755 cpus_and(new_mask, new_mask, cpus_allowed);
Paul Menage8707d8b2007-10-18 23:40:22 -07004756 again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004757 retval = set_cpus_allowed(p, new_mask);
4758
Paul Menage8707d8b2007-10-18 23:40:22 -07004759 if (!retval) {
4760 cpus_allowed = cpuset_cpus_allowed(p);
4761 if (!cpus_subset(new_mask, cpus_allowed)) {
4762 /*
4763 * We must have raced with a concurrent cpuset
4764 * update. Just reset the cpus_allowed to the
4765 * cpuset's cpus_allowed
4766 */
4767 new_mask = cpus_allowed;
4768 goto again;
4769 }
4770 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004771out_unlock:
4772 put_task_struct(p);
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004773 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004774 return retval;
4775}
4776
4777static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4778 cpumask_t *new_mask)
4779{
4780 if (len < sizeof(cpumask_t)) {
4781 memset(new_mask, 0, sizeof(cpumask_t));
4782 } else if (len > sizeof(cpumask_t)) {
4783 len = sizeof(cpumask_t);
4784 }
4785 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
4786}
4787
4788/**
4789 * sys_sched_setaffinity - set the cpu affinity of a process
4790 * @pid: pid of the process
4791 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4792 * @user_mask_ptr: user-space pointer to the new cpu mask
4793 */
4794asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len,
4795 unsigned long __user *user_mask_ptr)
4796{
4797 cpumask_t new_mask;
4798 int retval;
4799
4800 retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask);
4801 if (retval)
4802 return retval;
4803
4804 return sched_setaffinity(pid, new_mask);
4805}
4806
4807/*
4808 * Represents all cpu's present in the system
4809 * In systems capable of hotplug, this map could dynamically grow
4810 * as new cpu's are detected in the system via any platform specific
4811 * method, such as ACPI for e.g.
4812 */
4813
Andi Kleen4cef0c62006-01-11 22:44:57 +01004814cpumask_t cpu_present_map __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004815EXPORT_SYMBOL(cpu_present_map);
4816
4817#ifndef CONFIG_SMP
Andi Kleen4cef0c62006-01-11 22:44:57 +01004818cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL;
Greg Bankse16b38f2006-10-02 02:17:40 -07004819EXPORT_SYMBOL(cpu_online_map);
4820
Andi Kleen4cef0c62006-01-11 22:44:57 +01004821cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL;
Greg Bankse16b38f2006-10-02 02:17:40 -07004822EXPORT_SYMBOL(cpu_possible_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004823#endif
4824
4825long sched_getaffinity(pid_t pid, cpumask_t *mask)
4826{
Ingo Molnar36c8b582006-07-03 00:25:41 -07004827 struct task_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004828 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004829
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004830 get_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004831 read_lock(&tasklist_lock);
4832
4833 retval = -ESRCH;
4834 p = find_process_by_pid(pid);
4835 if (!p)
4836 goto out_unlock;
4837
David Quigleye7834f82006-06-23 02:03:59 -07004838 retval = security_task_getscheduler(p);
4839 if (retval)
4840 goto out_unlock;
4841
Jack Steiner2f7016d2006-02-01 03:05:18 -08004842 cpus_and(*mask, p->cpus_allowed, cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004843
4844out_unlock:
4845 read_unlock(&tasklist_lock);
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004846 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004847
Ulrich Drepper9531b622007-08-09 11:16:46 +02004848 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004849}
4850
4851/**
4852 * sys_sched_getaffinity - get the cpu affinity of a process
4853 * @pid: pid of the process
4854 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4855 * @user_mask_ptr: user-space pointer to hold the current cpu mask
4856 */
4857asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len,
4858 unsigned long __user *user_mask_ptr)
4859{
4860 int ret;
4861 cpumask_t mask;
4862
4863 if (len < sizeof(cpumask_t))
4864 return -EINVAL;
4865
4866 ret = sched_getaffinity(pid, &mask);
4867 if (ret < 0)
4868 return ret;
4869
4870 if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t)))
4871 return -EFAULT;
4872
4873 return sizeof(cpumask_t);
4874}
4875
4876/**
4877 * sys_sched_yield - yield the current processor to other threads.
4878 *
Ingo Molnardd41f592007-07-09 18:51:59 +02004879 * This function yields the current CPU to other tasks. If there are no
4880 * other threads running on this CPU then this function will return.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004881 */
4882asmlinkage long sys_sched_yield(void)
4883{
Ingo Molnar70b97a72006-07-03 00:25:42 -07004884 struct rq *rq = this_rq_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004885
Ingo Molnar2d723762007-10-15 17:00:12 +02004886 schedstat_inc(rq, yld_count);
Dmitry Adamushko4530d7a2007-10-15 17:00:08 +02004887 current->sched_class->yield_task(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004888
4889 /*
4890 * Since we are going to call schedule() anyway, there's
4891 * no need to preempt or enable interrupts:
4892 */
4893 __release(rq->lock);
Ingo Molnar8a25d5d2006-07-03 00:24:54 -07004894 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004895 _raw_spin_unlock(&rq->lock);
4896 preempt_enable_no_resched();
4897
4898 schedule();
4899
4900 return 0;
4901}
4902
Andrew Mortone7b38402006-06-30 01:56:00 -07004903static void __cond_resched(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004904{
Ingo Molnar8e0a43d2006-06-23 02:05:23 -07004905#ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
4906 __might_sleep(__FILE__, __LINE__);
4907#endif
Ingo Molnar5bbcfd92005-07-07 17:57:04 -07004908 /*
4909 * The BKS might be reacquired before we have dropped
4910 * PREEMPT_ACTIVE, which could trigger a second
4911 * cond_resched() call.
4912 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004913 do {
4914 add_preempt_count(PREEMPT_ACTIVE);
4915 schedule();
4916 sub_preempt_count(PREEMPT_ACTIVE);
4917 } while (need_resched());
4918}
4919
Herbert Xu02b67cc32008-01-25 21:08:28 +01004920#if !defined(CONFIG_PREEMPT) || defined(CONFIG_PREEMPT_VOLUNTARY)
4921int __sched _cond_resched(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004922{
Ingo Molnar94142322006-12-29 16:48:13 -08004923 if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) &&
4924 system_state == SYSTEM_RUNNING) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004925 __cond_resched();
4926 return 1;
4927 }
4928 return 0;
4929}
Herbert Xu02b67cc32008-01-25 21:08:28 +01004930EXPORT_SYMBOL(_cond_resched);
4931#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004932
4933/*
4934 * cond_resched_lock() - if a reschedule is pending, drop the given lock,
4935 * call schedule, and on return reacquire the lock.
4936 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01004937 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
Linus Torvalds1da177e2005-04-16 15:20:36 -07004938 * operations here to prevent schedule() from being called twice (once via
4939 * spin_unlock(), once by hand).
4940 */
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004941int cond_resched_lock(spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004942{
Nick Piggin95c354f2008-01-30 13:31:20 +01004943 int resched = need_resched() && system_state == SYSTEM_RUNNING;
Jan Kara6df3cec2005-06-13 15:52:32 -07004944 int ret = 0;
4945
Nick Piggin95c354f2008-01-30 13:31:20 +01004946 if (spin_needbreak(lock) || resched) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004947 spin_unlock(lock);
Nick Piggin95c354f2008-01-30 13:31:20 +01004948 if (resched && need_resched())
4949 __cond_resched();
4950 else
4951 cpu_relax();
Jan Kara6df3cec2005-06-13 15:52:32 -07004952 ret = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004953 spin_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004954 }
Jan Kara6df3cec2005-06-13 15:52:32 -07004955 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004956}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004957EXPORT_SYMBOL(cond_resched_lock);
4958
4959int __sched cond_resched_softirq(void)
4960{
4961 BUG_ON(!in_softirq());
4962
Ingo Molnar94142322006-12-29 16:48:13 -08004963 if (need_resched() && system_state == SYSTEM_RUNNING) {
Thomas Gleixner98d825672007-05-23 13:58:18 -07004964 local_bh_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004965 __cond_resched();
4966 local_bh_disable();
4967 return 1;
4968 }
4969 return 0;
4970}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004971EXPORT_SYMBOL(cond_resched_softirq);
4972
Linus Torvalds1da177e2005-04-16 15:20:36 -07004973/**
4974 * yield - yield the current processor to other threads.
4975 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08004976 * This is a shortcut for kernel-space yielding - it marks the
Linus Torvalds1da177e2005-04-16 15:20:36 -07004977 * thread runnable and calls sys_sched_yield().
4978 */
4979void __sched yield(void)
4980{
4981 set_current_state(TASK_RUNNING);
4982 sys_sched_yield();
4983}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004984EXPORT_SYMBOL(yield);
4985
4986/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01004987 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
Linus Torvalds1da177e2005-04-16 15:20:36 -07004988 * that process accounting knows that this is a task in IO wait state.
4989 *
4990 * But don't do that if it is a deliberate, throttling IO wait (this task
4991 * has set its backing_dev_info: the queue against which it should throttle)
4992 */
4993void __sched io_schedule(void)
4994{
Ingo Molnar70b97a72006-07-03 00:25:42 -07004995 struct rq *rq = &__raw_get_cpu_var(runqueues);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004996
Shailabh Nagar0ff92242006-07-14 00:24:37 -07004997 delayacct_blkio_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004998 atomic_inc(&rq->nr_iowait);
4999 schedule();
5000 atomic_dec(&rq->nr_iowait);
Shailabh Nagar0ff92242006-07-14 00:24:37 -07005001 delayacct_blkio_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005002}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005003EXPORT_SYMBOL(io_schedule);
5004
5005long __sched io_schedule_timeout(long timeout)
5006{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005007 struct rq *rq = &__raw_get_cpu_var(runqueues);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005008 long ret;
5009
Shailabh Nagar0ff92242006-07-14 00:24:37 -07005010 delayacct_blkio_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005011 atomic_inc(&rq->nr_iowait);
5012 ret = schedule_timeout(timeout);
5013 atomic_dec(&rq->nr_iowait);
Shailabh Nagar0ff92242006-07-14 00:24:37 -07005014 delayacct_blkio_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005015 return ret;
5016}
5017
5018/**
5019 * sys_sched_get_priority_max - return maximum RT priority.
5020 * @policy: scheduling class.
5021 *
5022 * this syscall returns the maximum rt_priority that can be used
5023 * by a given scheduling class.
5024 */
5025asmlinkage long sys_sched_get_priority_max(int policy)
5026{
5027 int ret = -EINVAL;
5028
5029 switch (policy) {
5030 case SCHED_FIFO:
5031 case SCHED_RR:
5032 ret = MAX_USER_RT_PRIO-1;
5033 break;
5034 case SCHED_NORMAL:
Ingo Molnarb0a94992006-01-14 13:20:41 -08005035 case SCHED_BATCH:
Ingo Molnardd41f592007-07-09 18:51:59 +02005036 case SCHED_IDLE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005037 ret = 0;
5038 break;
5039 }
5040 return ret;
5041}
5042
5043/**
5044 * sys_sched_get_priority_min - return minimum RT priority.
5045 * @policy: scheduling class.
5046 *
5047 * this syscall returns the minimum rt_priority that can be used
5048 * by a given scheduling class.
5049 */
5050asmlinkage long sys_sched_get_priority_min(int policy)
5051{
5052 int ret = -EINVAL;
5053
5054 switch (policy) {
5055 case SCHED_FIFO:
5056 case SCHED_RR:
5057 ret = 1;
5058 break;
5059 case SCHED_NORMAL:
Ingo Molnarb0a94992006-01-14 13:20:41 -08005060 case SCHED_BATCH:
Ingo Molnardd41f592007-07-09 18:51:59 +02005061 case SCHED_IDLE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005062 ret = 0;
5063 }
5064 return ret;
5065}
5066
5067/**
5068 * sys_sched_rr_get_interval - return the default timeslice of a process.
5069 * @pid: pid of the process.
5070 * @interval: userspace pointer to the timeslice value.
5071 *
5072 * this syscall writes the default timeslice value of a given process
5073 * into the user-space timespec buffer. A value of '0' means infinity.
5074 */
5075asmlinkage
5076long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval)
5077{
Ingo Molnar36c8b582006-07-03 00:25:41 -07005078 struct task_struct *p;
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005079 unsigned int time_slice;
Andi Kleen3a5c3592007-10-15 17:00:14 +02005080 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005081 struct timespec t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005082
5083 if (pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02005084 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005085
5086 retval = -ESRCH;
5087 read_lock(&tasklist_lock);
5088 p = find_process_by_pid(pid);
5089 if (!p)
5090 goto out_unlock;
5091
5092 retval = security_task_getscheduler(p);
5093 if (retval)
5094 goto out_unlock;
5095
Ingo Molnar77034932007-12-04 17:04:39 +01005096 /*
5097 * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER
5098 * tasks that are on an otherwise idle runqueue:
5099 */
5100 time_slice = 0;
5101 if (p->policy == SCHED_RR) {
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005102 time_slice = DEF_TIMESLICE;
Ingo Molnar77034932007-12-04 17:04:39 +01005103 } else {
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005104 struct sched_entity *se = &p->se;
5105 unsigned long flags;
5106 struct rq *rq;
5107
5108 rq = task_rq_lock(p, &flags);
Ingo Molnar77034932007-12-04 17:04:39 +01005109 if (rq->cfs.load.weight)
5110 time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se));
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005111 task_rq_unlock(rq, &flags);
5112 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005113 read_unlock(&tasklist_lock);
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005114 jiffies_to_timespec(time_slice, &t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005115 retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005116 return retval;
Andi Kleen3a5c3592007-10-15 17:00:14 +02005117
Linus Torvalds1da177e2005-04-16 15:20:36 -07005118out_unlock:
5119 read_unlock(&tasklist_lock);
5120 return retval;
5121}
5122
Andreas Mohr2ed6e342006-07-10 04:43:52 -07005123static const char stat_nam[] = "RSDTtZX";
Ingo Molnar36c8b582006-07-03 00:25:41 -07005124
Ingo Molnar82a1fcb2008-01-25 21:08:02 +01005125void sched_show_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005126{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005127 unsigned long free = 0;
Ingo Molnar36c8b582006-07-03 00:25:41 -07005128 unsigned state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005129
Linus Torvalds1da177e2005-04-16 15:20:36 -07005130 state = p->state ? __ffs(p->state) + 1 : 0;
Ingo Molnarcc4ea792007-10-18 21:32:56 +02005131 printk(KERN_INFO "%-13.13s %c", p->comm,
Andreas Mohr2ed6e342006-07-10 04:43:52 -07005132 state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
Ingo Molnar4bd77322007-07-11 21:21:47 +02005133#if BITS_PER_LONG == 32
Linus Torvalds1da177e2005-04-16 15:20:36 -07005134 if (state == TASK_RUNNING)
Ingo Molnarcc4ea792007-10-18 21:32:56 +02005135 printk(KERN_CONT " running ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005136 else
Ingo Molnarcc4ea792007-10-18 21:32:56 +02005137 printk(KERN_CONT " %08lx ", thread_saved_pc(p));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005138#else
5139 if (state == TASK_RUNNING)
Ingo Molnarcc4ea792007-10-18 21:32:56 +02005140 printk(KERN_CONT " running task ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005141 else
Ingo Molnarcc4ea792007-10-18 21:32:56 +02005142 printk(KERN_CONT " %016lx ", thread_saved_pc(p));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005143#endif
5144#ifdef CONFIG_DEBUG_STACK_USAGE
5145 {
Al Viro10ebffd2005-11-13 16:06:56 -08005146 unsigned long *n = end_of_stack(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005147 while (!*n)
5148 n++;
Al Viro10ebffd2005-11-13 16:06:56 -08005149 free = (unsigned long)n - (unsigned long)end_of_stack(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005150 }
5151#endif
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07005152 printk(KERN_CONT "%5lu %5d %6d\n", free,
Roland McGrathfcfd50a2008-01-09 00:03:23 -08005153 task_pid_nr(p), task_pid_nr(p->real_parent));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005154
Nick Piggin5fb5e6d2008-01-25 21:08:34 +01005155 show_stack(p, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005156}
5157
Ingo Molnare59e2ae2006-12-06 20:35:59 -08005158void show_state_filter(unsigned long state_filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005159{
Ingo Molnar36c8b582006-07-03 00:25:41 -07005160 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005161
Ingo Molnar4bd77322007-07-11 21:21:47 +02005162#if BITS_PER_LONG == 32
5163 printk(KERN_INFO
5164 " task PC stack pid father\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005165#else
Ingo Molnar4bd77322007-07-11 21:21:47 +02005166 printk(KERN_INFO
5167 " task PC stack pid father\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005168#endif
5169 read_lock(&tasklist_lock);
5170 do_each_thread(g, p) {
5171 /*
5172 * reset the NMI-timeout, listing all files on a slow
5173 * console might take alot of time:
5174 */
5175 touch_nmi_watchdog();
Ingo Molnar39bc89f2007-04-25 20:50:03 -07005176 if (!state_filter || (p->state & state_filter))
Ingo Molnar82a1fcb2008-01-25 21:08:02 +01005177 sched_show_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005178 } while_each_thread(g, p);
5179
Jeremy Fitzhardinge04c91672007-05-08 00:28:05 -07005180 touch_all_softlockup_watchdogs();
5181
Ingo Molnardd41f592007-07-09 18:51:59 +02005182#ifdef CONFIG_SCHED_DEBUG
5183 sysrq_sched_debug_show();
5184#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005185 read_unlock(&tasklist_lock);
Ingo Molnare59e2ae2006-12-06 20:35:59 -08005186 /*
5187 * Only show locks if all tasks are dumped:
5188 */
5189 if (state_filter == -1)
5190 debug_show_all_locks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005191}
5192
Ingo Molnar1df21052007-07-09 18:51:58 +02005193void __cpuinit init_idle_bootup_task(struct task_struct *idle)
5194{
Ingo Molnardd41f592007-07-09 18:51:59 +02005195 idle->sched_class = &idle_sched_class;
Ingo Molnar1df21052007-07-09 18:51:58 +02005196}
5197
Ingo Molnarf340c0d2005-06-28 16:40:42 +02005198/**
5199 * init_idle - set up an idle thread for a given CPU
5200 * @idle: task in question
5201 * @cpu: cpu the idle task belongs to
5202 *
5203 * NOTE: this function does not set the idle thread's NEED_RESCHED
5204 * flag, to make booting more robust.
5205 */
Nick Piggin5c1e1762006-10-03 01:14:04 -07005206void __cpuinit init_idle(struct task_struct *idle, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005207{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005208 struct rq *rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005209 unsigned long flags;
5210
Ingo Molnardd41f592007-07-09 18:51:59 +02005211 __sched_fork(idle);
5212 idle->se.exec_start = sched_clock();
5213
Ingo Molnarb29739f2006-06-27 02:54:51 -07005214 idle->prio = idle->normal_prio = MAX_PRIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005215 idle->cpus_allowed = cpumask_of_cpu(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02005216 __set_task_cpu(idle, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005217
5218 spin_lock_irqsave(&rq->lock, flags);
5219 rq->curr = rq->idle = idle;
Nick Piggin4866cde2005-06-25 14:57:23 -07005220#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
5221 idle->oncpu = 1;
5222#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005223 spin_unlock_irqrestore(&rq->lock, flags);
5224
5225 /* Set the preempt count _outside_ the spinlocks! */
Al Viroa1261f52005-11-13 16:06:55 -08005226 task_thread_info(idle)->preempt_count = 0;
Ingo Molnar6478d882008-01-25 21:08:33 +01005227
Ingo Molnardd41f592007-07-09 18:51:59 +02005228 /*
5229 * The idle tasks have their own, simple scheduling class:
5230 */
5231 idle->sched_class = &idle_sched_class;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005232}
5233
5234/*
5235 * In a system that switches off the HZ timer nohz_cpu_mask
5236 * indicates which cpus entered this state. This is used
5237 * in the rcu update to wait only for active cpus. For system
5238 * which do not switch off the HZ timer nohz_cpu_mask should
5239 * always be CPU_MASK_NONE.
5240 */
5241cpumask_t nohz_cpu_mask = CPU_MASK_NONE;
5242
Ingo Molnar19978ca2007-11-09 22:39:38 +01005243/*
5244 * Increase the granularity value when there are more CPUs,
5245 * because with more CPUs the 'effective latency' as visible
5246 * to users decreases. But the relationship is not linear,
5247 * so pick a second-best guess by going with the log2 of the
5248 * number of CPUs.
5249 *
5250 * This idea comes from the SD scheduler of Con Kolivas:
5251 */
5252static inline void sched_init_granularity(void)
5253{
5254 unsigned int factor = 1 + ilog2(num_online_cpus());
5255 const unsigned long limit = 200000000;
5256
5257 sysctl_sched_min_granularity *= factor;
5258 if (sysctl_sched_min_granularity > limit)
5259 sysctl_sched_min_granularity = limit;
5260
5261 sysctl_sched_latency *= factor;
5262 if (sysctl_sched_latency > limit)
5263 sysctl_sched_latency = limit;
5264
5265 sysctl_sched_wakeup_granularity *= factor;
5266 sysctl_sched_batch_wakeup_granularity *= factor;
5267}
5268
Linus Torvalds1da177e2005-04-16 15:20:36 -07005269#ifdef CONFIG_SMP
5270/*
5271 * This is how migration works:
5272 *
Ingo Molnar70b97a72006-07-03 00:25:42 -07005273 * 1) we queue a struct migration_req structure in the source CPU's
Linus Torvalds1da177e2005-04-16 15:20:36 -07005274 * runqueue and wake up that CPU's migration thread.
5275 * 2) we down() the locked semaphore => thread blocks.
5276 * 3) migration thread wakes up (implicitly it forces the migrated
5277 * thread off the CPU)
5278 * 4) it gets the migration request and checks whether the migrated
5279 * task is still in the wrong runqueue.
5280 * 5) if it's in the wrong runqueue then the migration thread removes
5281 * it and puts it into the right queue.
5282 * 6) migration thread up()s the semaphore.
5283 * 7) we wake up and the migration is done.
5284 */
5285
5286/*
5287 * Change a given task's CPU affinity. Migrate the thread to a
5288 * proper CPU and schedule it away if the CPU it's executing on
5289 * is removed from the allowed bitmask.
5290 *
5291 * NOTE: the caller must have a valid reference to the task, the
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005292 * task must not exit() & deallocate itself prematurely. The
Linus Torvalds1da177e2005-04-16 15:20:36 -07005293 * call is not atomic; no spinlocks may be held.
5294 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07005295int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005296{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005297 struct migration_req req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005298 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07005299 struct rq *rq;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005300 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005301
5302 rq = task_rq_lock(p, &flags);
5303 if (!cpus_intersects(new_mask, cpu_online_map)) {
5304 ret = -EINVAL;
5305 goto out;
5306 }
5307
Gregory Haskins73fe6aa2008-01-25 21:08:07 +01005308 if (p->sched_class->set_cpus_allowed)
5309 p->sched_class->set_cpus_allowed(p, &new_mask);
5310 else {
Ingo Molnar0eab9142008-01-25 21:08:19 +01005311 p->cpus_allowed = new_mask;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01005312 p->rt.nr_cpus_allowed = cpus_weight(new_mask);
Gregory Haskins73fe6aa2008-01-25 21:08:07 +01005313 }
5314
Linus Torvalds1da177e2005-04-16 15:20:36 -07005315 /* Can the task run on the task's current CPU? If so, we're done */
5316 if (cpu_isset(task_cpu(p), new_mask))
5317 goto out;
5318
5319 if (migrate_task(p, any_online_cpu(new_mask), &req)) {
5320 /* Need help from migration thread: drop lock and wait. */
5321 task_rq_unlock(rq, &flags);
5322 wake_up_process(rq->migration_thread);
5323 wait_for_completion(&req.done);
5324 tlb_migrate_finish(p->mm);
5325 return 0;
5326 }
5327out:
5328 task_rq_unlock(rq, &flags);
Ingo Molnar48f24c42006-07-03 00:25:40 -07005329
Linus Torvalds1da177e2005-04-16 15:20:36 -07005330 return ret;
5331}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005332EXPORT_SYMBOL_GPL(set_cpus_allowed);
5333
5334/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005335 * Move (not current) task off this cpu, onto dest cpu. We're doing
Linus Torvalds1da177e2005-04-16 15:20:36 -07005336 * this because either it can't run here any more (set_cpus_allowed()
5337 * away from this CPU, or CPU going down), or because we're
5338 * attempting to rebalance this task on exec (sched_exec).
5339 *
5340 * So we race with normal scheduler movements, but that's OK, as long
5341 * as the task is no longer on this CPU.
Kirill Korotaevefc30812006-06-27 02:54:32 -07005342 *
5343 * Returns non-zero if task was successfully migrated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005344 */
Kirill Korotaevefc30812006-06-27 02:54:32 -07005345static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005346{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005347 struct rq *rq_dest, *rq_src;
Ingo Molnardd41f592007-07-09 18:51:59 +02005348 int ret = 0, on_rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005349
5350 if (unlikely(cpu_is_offline(dest_cpu)))
Kirill Korotaevefc30812006-06-27 02:54:32 -07005351 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005352
5353 rq_src = cpu_rq(src_cpu);
5354 rq_dest = cpu_rq(dest_cpu);
5355
5356 double_rq_lock(rq_src, rq_dest);
5357 /* Already moved. */
5358 if (task_cpu(p) != src_cpu)
5359 goto out;
5360 /* Affinity changed (again). */
5361 if (!cpu_isset(dest_cpu, p->cpus_allowed))
5362 goto out;
5363
Ingo Molnardd41f592007-07-09 18:51:59 +02005364 on_rq = p->se.on_rq;
Ingo Molnar6e82a3b2007-08-09 11:16:51 +02005365 if (on_rq)
Ingo Molnar2e1cb742007-08-09 11:16:49 +02005366 deactivate_task(rq_src, p, 0);
Ingo Molnar6e82a3b2007-08-09 11:16:51 +02005367
Linus Torvalds1da177e2005-04-16 15:20:36 -07005368 set_task_cpu(p, dest_cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02005369 if (on_rq) {
5370 activate_task(rq_dest, p, 0);
5371 check_preempt_curr(rq_dest, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005372 }
Kirill Korotaevefc30812006-06-27 02:54:32 -07005373 ret = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005374out:
5375 double_rq_unlock(rq_src, rq_dest);
Kirill Korotaevefc30812006-06-27 02:54:32 -07005376 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005377}
5378
5379/*
5380 * migration_thread - this is a highprio system thread that performs
5381 * thread migration by bumping thread off CPU then 'pushing' onto
5382 * another runqueue.
5383 */
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07005384static int migration_thread(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005385{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005386 int cpu = (long)data;
Ingo Molnar70b97a72006-07-03 00:25:42 -07005387 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005388
5389 rq = cpu_rq(cpu);
5390 BUG_ON(rq->migration_thread != current);
5391
5392 set_current_state(TASK_INTERRUPTIBLE);
5393 while (!kthread_should_stop()) {
Ingo Molnar70b97a72006-07-03 00:25:42 -07005394 struct migration_req *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005395 struct list_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005396
Linus Torvalds1da177e2005-04-16 15:20:36 -07005397 spin_lock_irq(&rq->lock);
5398
5399 if (cpu_is_offline(cpu)) {
5400 spin_unlock_irq(&rq->lock);
5401 goto wait_to_die;
5402 }
5403
5404 if (rq->active_balance) {
5405 active_load_balance(rq, cpu);
5406 rq->active_balance = 0;
5407 }
5408
5409 head = &rq->migration_queue;
5410
5411 if (list_empty(head)) {
5412 spin_unlock_irq(&rq->lock);
5413 schedule();
5414 set_current_state(TASK_INTERRUPTIBLE);
5415 continue;
5416 }
Ingo Molnar70b97a72006-07-03 00:25:42 -07005417 req = list_entry(head->next, struct migration_req, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005418 list_del_init(head->next);
5419
Nick Piggin674311d2005-06-25 14:57:27 -07005420 spin_unlock(&rq->lock);
5421 __migrate_task(req->task, cpu, req->dest_cpu);
5422 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005423
5424 complete(&req->done);
5425 }
5426 __set_current_state(TASK_RUNNING);
5427 return 0;
5428
5429wait_to_die:
5430 /* Wait for kthread_stop */
5431 set_current_state(TASK_INTERRUPTIBLE);
5432 while (!kthread_should_stop()) {
5433 schedule();
5434 set_current_state(TASK_INTERRUPTIBLE);
5435 }
5436 __set_current_state(TASK_RUNNING);
5437 return 0;
5438}
5439
5440#ifdef CONFIG_HOTPLUG_CPU
Oleg Nesterovf7b4cdd2007-10-16 23:30:56 -07005441
5442static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu)
5443{
5444 int ret;
5445
5446 local_irq_disable();
5447 ret = __migrate_task(p, src_cpu, dest_cpu);
5448 local_irq_enable();
5449 return ret;
5450}
5451
Kirill Korotaev054b9102006-12-10 02:20:11 -08005452/*
Robert P. J. Day3a4fa0a2007-10-19 23:10:43 +02005453 * Figure out where task on dead CPU should go, use force if necessary.
Kirill Korotaev054b9102006-12-10 02:20:11 -08005454 * NOTE: interrupts should be disabled by the caller
5455 */
Ingo Molnar48f24c42006-07-03 00:25:40 -07005456static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005457{
Kirill Korotaevefc30812006-06-27 02:54:32 -07005458 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005459 cpumask_t mask;
Ingo Molnar70b97a72006-07-03 00:25:42 -07005460 struct rq *rq;
5461 int dest_cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005462
Andi Kleen3a5c3592007-10-15 17:00:14 +02005463 do {
5464 /* On same node? */
5465 mask = node_to_cpumask(cpu_to_node(dead_cpu));
5466 cpus_and(mask, mask, p->cpus_allowed);
5467 dest_cpu = any_online_cpu(mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005468
Andi Kleen3a5c3592007-10-15 17:00:14 +02005469 /* On any allowed CPU? */
5470 if (dest_cpu == NR_CPUS)
5471 dest_cpu = any_online_cpu(p->cpus_allowed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005472
Andi Kleen3a5c3592007-10-15 17:00:14 +02005473 /* No more Mr. Nice Guy. */
5474 if (dest_cpu == NR_CPUS) {
Cliff Wickman470fd642007-10-18 23:40:46 -07005475 cpumask_t cpus_allowed = cpuset_cpus_allowed_locked(p);
5476 /*
5477 * Try to stay on the same cpuset, where the
5478 * current cpuset may be a subset of all cpus.
5479 * The cpuset_cpus_allowed_locked() variant of
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005480 * cpuset_cpus_allowed() will not block. It must be
Cliff Wickman470fd642007-10-18 23:40:46 -07005481 * called within calls to cpuset_lock/cpuset_unlock.
5482 */
Andi Kleen3a5c3592007-10-15 17:00:14 +02005483 rq = task_rq_lock(p, &flags);
Cliff Wickman470fd642007-10-18 23:40:46 -07005484 p->cpus_allowed = cpus_allowed;
Andi Kleen3a5c3592007-10-15 17:00:14 +02005485 dest_cpu = any_online_cpu(p->cpus_allowed);
5486 task_rq_unlock(rq, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005487
Andi Kleen3a5c3592007-10-15 17:00:14 +02005488 /*
5489 * Don't tell them about moving exiting tasks or
5490 * kernel threads (both mm NULL), since they never
5491 * leave kernel.
5492 */
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005493 if (p->mm && printk_ratelimit()) {
Andi Kleen3a5c3592007-10-15 17:00:14 +02005494 printk(KERN_INFO "process %d (%s) no "
5495 "longer affine to cpu%d\n",
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005496 task_pid_nr(p), p->comm, dead_cpu);
5497 }
Andi Kleen3a5c3592007-10-15 17:00:14 +02005498 }
Oleg Nesterovf7b4cdd2007-10-16 23:30:56 -07005499 } while (!__migrate_task_irq(p, dead_cpu, dest_cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005500}
5501
5502/*
5503 * While a dead CPU has no uninterruptible tasks queued at this point,
5504 * it might still have a nonzero ->nr_uninterruptible counter, because
5505 * for performance reasons the counter is not stricly tracking tasks to
5506 * their home CPUs. So we just add the counter to another CPU's counter,
5507 * to keep the global sum constant after CPU-down:
5508 */
Ingo Molnar70b97a72006-07-03 00:25:42 -07005509static void migrate_nr_uninterruptible(struct rq *rq_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005510{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005511 struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005512 unsigned long flags;
5513
5514 local_irq_save(flags);
5515 double_rq_lock(rq_src, rq_dest);
5516 rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible;
5517 rq_src->nr_uninterruptible = 0;
5518 double_rq_unlock(rq_src, rq_dest);
5519 local_irq_restore(flags);
5520}
5521
5522/* Run through task list and migrate tasks from the dead cpu. */
5523static void migrate_live_tasks(int src_cpu)
5524{
Ingo Molnar48f24c42006-07-03 00:25:40 -07005525 struct task_struct *p, *t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005526
Oleg Nesterovf7b4cdd2007-10-16 23:30:56 -07005527 read_lock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005528
Ingo Molnar48f24c42006-07-03 00:25:40 -07005529 do_each_thread(t, p) {
5530 if (p == current)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005531 continue;
5532
Ingo Molnar48f24c42006-07-03 00:25:40 -07005533 if (task_cpu(p) == src_cpu)
5534 move_task_off_dead_cpu(src_cpu, p);
5535 } while_each_thread(t, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005536
Oleg Nesterovf7b4cdd2007-10-16 23:30:56 -07005537 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005538}
5539
Ingo Molnardd41f592007-07-09 18:51:59 +02005540/*
5541 * Schedules idle task to be the next runnable task on current CPU.
Dmitry Adamushko94bc9a72007-11-15 20:57:40 +01005542 * It does so by boosting its priority to highest possible.
5543 * Used by CPU offline code.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005544 */
5545void sched_idle_next(void)
5546{
Ingo Molnar48f24c42006-07-03 00:25:40 -07005547 int this_cpu = smp_processor_id();
Ingo Molnar70b97a72006-07-03 00:25:42 -07005548 struct rq *rq = cpu_rq(this_cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005549 struct task_struct *p = rq->idle;
5550 unsigned long flags;
5551
5552 /* cpu has to be offline */
Ingo Molnar48f24c42006-07-03 00:25:40 -07005553 BUG_ON(cpu_online(this_cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005554
Ingo Molnar48f24c42006-07-03 00:25:40 -07005555 /*
5556 * Strictly not necessary since rest of the CPUs are stopped by now
5557 * and interrupts disabled on the current cpu.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005558 */
5559 spin_lock_irqsave(&rq->lock, flags);
5560
Ingo Molnardd41f592007-07-09 18:51:59 +02005561 __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1);
Ingo Molnar48f24c42006-07-03 00:25:40 -07005562
Dmitry Adamushko94bc9a72007-11-15 20:57:40 +01005563 update_rq_clock(rq);
5564 activate_task(rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005565
5566 spin_unlock_irqrestore(&rq->lock, flags);
5567}
5568
Ingo Molnar48f24c42006-07-03 00:25:40 -07005569/*
5570 * Ensures that the idle task is using init_mm right before its cpu goes
Linus Torvalds1da177e2005-04-16 15:20:36 -07005571 * offline.
5572 */
5573void idle_task_exit(void)
5574{
5575 struct mm_struct *mm = current->active_mm;
5576
5577 BUG_ON(cpu_online(smp_processor_id()));
5578
5579 if (mm != &init_mm)
5580 switch_mm(mm, &init_mm, current);
5581 mmdrop(mm);
5582}
5583
Kirill Korotaev054b9102006-12-10 02:20:11 -08005584/* called under rq->lock with disabled interrupts */
Ingo Molnar36c8b582006-07-03 00:25:41 -07005585static void migrate_dead(unsigned int dead_cpu, struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005586{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005587 struct rq *rq = cpu_rq(dead_cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005588
5589 /* Must be exiting, otherwise would be on tasklist. */
Eugene Teo270f7222007-10-18 23:40:38 -07005590 BUG_ON(!p->exit_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005591
5592 /* Cannot have done final schedule yet: would have vanished. */
Oleg Nesterovc394cc92006-09-29 02:01:11 -07005593 BUG_ON(p->state == TASK_DEAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005594
Ingo Molnar48f24c42006-07-03 00:25:40 -07005595 get_task_struct(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005596
5597 /*
5598 * Drop lock around migration; if someone else moves it,
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005599 * that's OK. No task can be added to this CPU, so iteration is
Linus Torvalds1da177e2005-04-16 15:20:36 -07005600 * fine.
5601 */
Oleg Nesterovf7b4cdd2007-10-16 23:30:56 -07005602 spin_unlock_irq(&rq->lock);
Ingo Molnar48f24c42006-07-03 00:25:40 -07005603 move_task_off_dead_cpu(dead_cpu, p);
Oleg Nesterovf7b4cdd2007-10-16 23:30:56 -07005604 spin_lock_irq(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005605
Ingo Molnar48f24c42006-07-03 00:25:40 -07005606 put_task_struct(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005607}
5608
5609/* release_task() removes task from tasklist, so we won't find dead tasks. */
5610static void migrate_dead_tasks(unsigned int dead_cpu)
5611{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005612 struct rq *rq = cpu_rq(dead_cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02005613 struct task_struct *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005614
Ingo Molnardd41f592007-07-09 18:51:59 +02005615 for ( ; ; ) {
5616 if (!rq->nr_running)
5617 break;
Ingo Molnara8e504d2007-08-09 11:16:47 +02005618 update_rq_clock(rq);
Ingo Molnarff95f3d2007-08-09 11:16:49 +02005619 next = pick_next_task(rq, rq->curr);
Ingo Molnardd41f592007-07-09 18:51:59 +02005620 if (!next)
5621 break;
5622 migrate_dead(dead_cpu, next);
Nick Piggine692ab52007-07-26 13:40:43 +02005623
Linus Torvalds1da177e2005-04-16 15:20:36 -07005624 }
5625}
5626#endif /* CONFIG_HOTPLUG_CPU */
5627
Nick Piggine692ab52007-07-26 13:40:43 +02005628#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
5629
5630static struct ctl_table sd_ctl_dir[] = {
Alexey Dobriyane0361852007-08-09 11:16:46 +02005631 {
5632 .procname = "sched_domain",
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005633 .mode = 0555,
Alexey Dobriyane0361852007-08-09 11:16:46 +02005634 },
Ingo Molnar38605ca2007-10-29 21:18:11 +01005635 {0, },
Nick Piggine692ab52007-07-26 13:40:43 +02005636};
5637
5638static struct ctl_table sd_ctl_root[] = {
Alexey Dobriyane0361852007-08-09 11:16:46 +02005639 {
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005640 .ctl_name = CTL_KERN,
Alexey Dobriyane0361852007-08-09 11:16:46 +02005641 .procname = "kernel",
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005642 .mode = 0555,
Alexey Dobriyane0361852007-08-09 11:16:46 +02005643 .child = sd_ctl_dir,
5644 },
Ingo Molnar38605ca2007-10-29 21:18:11 +01005645 {0, },
Nick Piggine692ab52007-07-26 13:40:43 +02005646};
5647
5648static struct ctl_table *sd_alloc_ctl_entry(int n)
5649{
5650 struct ctl_table *entry =
Milton Miller5cf9f062007-10-15 17:00:19 +02005651 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
Nick Piggine692ab52007-07-26 13:40:43 +02005652
Nick Piggine692ab52007-07-26 13:40:43 +02005653 return entry;
5654}
5655
Milton Miller6382bc92007-10-15 17:00:19 +02005656static void sd_free_ctl_entry(struct ctl_table **tablep)
5657{
Milton Millercd7900762007-10-17 16:55:11 +02005658 struct ctl_table *entry;
Milton Miller6382bc92007-10-15 17:00:19 +02005659
Milton Millercd7900762007-10-17 16:55:11 +02005660 /*
5661 * In the intermediate directories, both the child directory and
5662 * procname are dynamically allocated and could fail but the mode
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005663 * will always be set. In the lowest directory the names are
Milton Millercd7900762007-10-17 16:55:11 +02005664 * static strings and all have proc handlers.
5665 */
5666 for (entry = *tablep; entry->mode; entry++) {
Milton Miller6382bc92007-10-15 17:00:19 +02005667 if (entry->child)
5668 sd_free_ctl_entry(&entry->child);
Milton Millercd7900762007-10-17 16:55:11 +02005669 if (entry->proc_handler == NULL)
5670 kfree(entry->procname);
5671 }
Milton Miller6382bc92007-10-15 17:00:19 +02005672
5673 kfree(*tablep);
5674 *tablep = NULL;
5675}
5676
Nick Piggine692ab52007-07-26 13:40:43 +02005677static void
Alexey Dobriyane0361852007-08-09 11:16:46 +02005678set_table_entry(struct ctl_table *entry,
Nick Piggine692ab52007-07-26 13:40:43 +02005679 const char *procname, void *data, int maxlen,
5680 mode_t mode, proc_handler *proc_handler)
5681{
Nick Piggine692ab52007-07-26 13:40:43 +02005682 entry->procname = procname;
5683 entry->data = data;
5684 entry->maxlen = maxlen;
5685 entry->mode = mode;
5686 entry->proc_handler = proc_handler;
5687}
5688
5689static struct ctl_table *
5690sd_alloc_ctl_domain_table(struct sched_domain *sd)
5691{
Zou Nan haiace8b3d2007-10-15 17:00:14 +02005692 struct ctl_table *table = sd_alloc_ctl_entry(12);
Nick Piggine692ab52007-07-26 13:40:43 +02005693
Milton Millerad1cdc12007-10-15 17:00:19 +02005694 if (table == NULL)
5695 return NULL;
5696
Alexey Dobriyane0361852007-08-09 11:16:46 +02005697 set_table_entry(&table[0], "min_interval", &sd->min_interval,
Nick Piggine692ab52007-07-26 13:40:43 +02005698 sizeof(long), 0644, proc_doulongvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005699 set_table_entry(&table[1], "max_interval", &sd->max_interval,
Nick Piggine692ab52007-07-26 13:40:43 +02005700 sizeof(long), 0644, proc_doulongvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005701 set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005702 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005703 set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005704 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005705 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005706 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005707 set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005708 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005709 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005710 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005711 set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
Nick Piggine692ab52007-07-26 13:40:43 +02005712 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005713 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
Nick Piggine692ab52007-07-26 13:40:43 +02005714 sizeof(int), 0644, proc_dointvec_minmax);
Zou Nan haiace8b3d2007-10-15 17:00:14 +02005715 set_table_entry(&table[9], "cache_nice_tries",
Nick Piggine692ab52007-07-26 13:40:43 +02005716 &sd->cache_nice_tries,
5717 sizeof(int), 0644, proc_dointvec_minmax);
Zou Nan haiace8b3d2007-10-15 17:00:14 +02005718 set_table_entry(&table[10], "flags", &sd->flags,
Nick Piggine692ab52007-07-26 13:40:43 +02005719 sizeof(int), 0644, proc_dointvec_minmax);
Milton Miller6323469f2007-10-15 17:00:19 +02005720 /* &table[11] is terminator */
Nick Piggine692ab52007-07-26 13:40:43 +02005721
5722 return table;
5723}
5724
Ingo Molnar9a4e7152007-11-28 15:52:56 +01005725static ctl_table *sd_alloc_ctl_cpu_table(int cpu)
Nick Piggine692ab52007-07-26 13:40:43 +02005726{
5727 struct ctl_table *entry, *table;
5728 struct sched_domain *sd;
5729 int domain_num = 0, i;
5730 char buf[32];
5731
5732 for_each_domain(cpu, sd)
5733 domain_num++;
5734 entry = table = sd_alloc_ctl_entry(domain_num + 1);
Milton Millerad1cdc12007-10-15 17:00:19 +02005735 if (table == NULL)
5736 return NULL;
Nick Piggine692ab52007-07-26 13:40:43 +02005737
5738 i = 0;
5739 for_each_domain(cpu, sd) {
5740 snprintf(buf, 32, "domain%d", i);
Nick Piggine692ab52007-07-26 13:40:43 +02005741 entry->procname = kstrdup(buf, GFP_KERNEL);
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005742 entry->mode = 0555;
Nick Piggine692ab52007-07-26 13:40:43 +02005743 entry->child = sd_alloc_ctl_domain_table(sd);
5744 entry++;
5745 i++;
5746 }
5747 return table;
5748}
5749
5750static struct ctl_table_header *sd_sysctl_header;
Milton Miller6382bc92007-10-15 17:00:19 +02005751static void register_sched_domain_sysctl(void)
Nick Piggine692ab52007-07-26 13:40:43 +02005752{
5753 int i, cpu_num = num_online_cpus();
5754 struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
5755 char buf[32];
5756
Milton Miller73785472007-10-24 18:23:48 +02005757 WARN_ON(sd_ctl_dir[0].child);
5758 sd_ctl_dir[0].child = entry;
5759
Milton Millerad1cdc12007-10-15 17:00:19 +02005760 if (entry == NULL)
5761 return;
5762
Milton Miller97b6ea72007-10-15 17:00:19 +02005763 for_each_online_cpu(i) {
Nick Piggine692ab52007-07-26 13:40:43 +02005764 snprintf(buf, 32, "cpu%d", i);
Nick Piggine692ab52007-07-26 13:40:43 +02005765 entry->procname = kstrdup(buf, GFP_KERNEL);
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005766 entry->mode = 0555;
Nick Piggine692ab52007-07-26 13:40:43 +02005767 entry->child = sd_alloc_ctl_cpu_table(i);
Milton Miller97b6ea72007-10-15 17:00:19 +02005768 entry++;
Nick Piggine692ab52007-07-26 13:40:43 +02005769 }
Milton Miller73785472007-10-24 18:23:48 +02005770
5771 WARN_ON(sd_sysctl_header);
Nick Piggine692ab52007-07-26 13:40:43 +02005772 sd_sysctl_header = register_sysctl_table(sd_ctl_root);
5773}
Milton Miller6382bc92007-10-15 17:00:19 +02005774
Milton Miller73785472007-10-24 18:23:48 +02005775/* may be called multiple times per register */
Milton Miller6382bc92007-10-15 17:00:19 +02005776static void unregister_sched_domain_sysctl(void)
5777{
Milton Miller73785472007-10-24 18:23:48 +02005778 if (sd_sysctl_header)
5779 unregister_sysctl_table(sd_sysctl_header);
Milton Miller6382bc92007-10-15 17:00:19 +02005780 sd_sysctl_header = NULL;
Milton Miller73785472007-10-24 18:23:48 +02005781 if (sd_ctl_dir[0].child)
5782 sd_free_ctl_entry(&sd_ctl_dir[0].child);
Milton Miller6382bc92007-10-15 17:00:19 +02005783}
Nick Piggine692ab52007-07-26 13:40:43 +02005784#else
Milton Miller6382bc92007-10-15 17:00:19 +02005785static void register_sched_domain_sysctl(void)
5786{
5787}
5788static void unregister_sched_domain_sysctl(void)
Nick Piggine692ab52007-07-26 13:40:43 +02005789{
5790}
5791#endif
5792
Linus Torvalds1da177e2005-04-16 15:20:36 -07005793/*
5794 * migration_call - callback that gets triggered when a CPU is added.
5795 * Here we can start up the necessary migration thread for the new CPU.
5796 */
Ingo Molnar48f24c42006-07-03 00:25:40 -07005797static int __cpuinit
5798migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005799{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005800 struct task_struct *p;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005801 int cpu = (long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005802 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07005803 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005804
5805 switch (action) {
Gautham R Shenoy5be93612007-05-09 02:34:04 -07005806
Linus Torvalds1da177e2005-04-16 15:20:36 -07005807 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005808 case CPU_UP_PREPARE_FROZEN:
Ingo Molnardd41f592007-07-09 18:51:59 +02005809 p = kthread_create(migration_thread, hcpu, "migration/%d", cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005810 if (IS_ERR(p))
5811 return NOTIFY_BAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005812 kthread_bind(p, cpu);
5813 /* Must be high prio: stop_machine expects to yield to it. */
5814 rq = task_rq_lock(p, &flags);
Ingo Molnardd41f592007-07-09 18:51:59 +02005815 __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005816 task_rq_unlock(rq, &flags);
5817 cpu_rq(cpu)->migration_thread = p;
5818 break;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005819
Linus Torvalds1da177e2005-04-16 15:20:36 -07005820 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005821 case CPU_ONLINE_FROZEN:
Robert P. J. Day3a4fa0a2007-10-19 23:10:43 +02005822 /* Strictly unnecessary, as first user will wake it. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005823 wake_up_process(cpu_rq(cpu)->migration_thread);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005824
5825 /* Update our root-domain */
5826 rq = cpu_rq(cpu);
5827 spin_lock_irqsave(&rq->lock, flags);
5828 if (rq->rd) {
5829 BUG_ON(!cpu_isset(cpu, rq->rd->span));
5830 cpu_set(cpu, rq->rd->online);
5831 }
5832 spin_unlock_irqrestore(&rq->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005833 break;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005834
Linus Torvalds1da177e2005-04-16 15:20:36 -07005835#ifdef CONFIG_HOTPLUG_CPU
5836 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005837 case CPU_UP_CANCELED_FROZEN:
Heiko Carstensfc75cdf2006-06-25 05:49:10 -07005838 if (!cpu_rq(cpu)->migration_thread)
5839 break;
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005840 /* Unbind it from offline cpu so it can run. Fall thru. */
Heiko Carstensa4c4af72005-11-07 00:58:38 -08005841 kthread_bind(cpu_rq(cpu)->migration_thread,
5842 any_online_cpu(cpu_online_map));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005843 kthread_stop(cpu_rq(cpu)->migration_thread);
5844 cpu_rq(cpu)->migration_thread = NULL;
5845 break;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005846
Linus Torvalds1da177e2005-04-16 15:20:36 -07005847 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005848 case CPU_DEAD_FROZEN:
Cliff Wickman470fd642007-10-18 23:40:46 -07005849 cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005850 migrate_live_tasks(cpu);
5851 rq = cpu_rq(cpu);
5852 kthread_stop(rq->migration_thread);
5853 rq->migration_thread = NULL;
5854 /* Idle task back to normal (off runqueue, low prio) */
Oleg Nesterovd2da2722007-10-16 23:30:56 -07005855 spin_lock_irq(&rq->lock);
Ingo Molnara8e504d2007-08-09 11:16:47 +02005856 update_rq_clock(rq);
Ingo Molnar2e1cb742007-08-09 11:16:49 +02005857 deactivate_task(rq, rq->idle, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005858 rq->idle->static_prio = MAX_PRIO;
Ingo Molnardd41f592007-07-09 18:51:59 +02005859 __setscheduler(rq, rq->idle, SCHED_NORMAL, 0);
5860 rq->idle->sched_class = &idle_sched_class;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005861 migrate_dead_tasks(cpu);
Oleg Nesterovd2da2722007-10-16 23:30:56 -07005862 spin_unlock_irq(&rq->lock);
Cliff Wickman470fd642007-10-18 23:40:46 -07005863 cpuset_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005864 migrate_nr_uninterruptible(rq);
5865 BUG_ON(rq->nr_running != 0);
5866
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005867 /*
5868 * No need to migrate the tasks: it was best-effort if
5869 * they didn't take sched_hotcpu_mutex. Just wake up
5870 * the requestors.
5871 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005872 spin_lock_irq(&rq->lock);
5873 while (!list_empty(&rq->migration_queue)) {
Ingo Molnar70b97a72006-07-03 00:25:42 -07005874 struct migration_req *req;
5875
Linus Torvalds1da177e2005-04-16 15:20:36 -07005876 req = list_entry(rq->migration_queue.next,
Ingo Molnar70b97a72006-07-03 00:25:42 -07005877 struct migration_req, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005878 list_del_init(&req->list);
5879 complete(&req->done);
5880 }
5881 spin_unlock_irq(&rq->lock);
5882 break;
Gregory Haskins57d885f2008-01-25 21:08:18 +01005883
5884 case CPU_DOWN_PREPARE:
5885 /* Update our root-domain */
5886 rq = cpu_rq(cpu);
5887 spin_lock_irqsave(&rq->lock, flags);
5888 if (rq->rd) {
5889 BUG_ON(!cpu_isset(cpu, rq->rd->span));
5890 cpu_clear(cpu, rq->rd->online);
5891 }
5892 spin_unlock_irqrestore(&rq->lock, flags);
5893 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005894#endif
5895 }
5896 return NOTIFY_OK;
5897}
5898
5899/* Register at highest priority so that task migration (migrate_all_tasks)
5900 * happens before everything else.
5901 */
Chandra Seetharaman26c21432006-06-27 02:54:10 -07005902static struct notifier_block __cpuinitdata migration_notifier = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005903 .notifier_call = migration_call,
5904 .priority = 10
5905};
5906
Adrian Bunke6fe6642007-11-09 22:39:39 +01005907void __init migration_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005908{
5909 void *cpu = (void *)(long)smp_processor_id();
Akinobu Mita07dccf32006-09-29 02:00:22 -07005910 int err;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005911
5912 /* Start one for the boot CPU: */
Akinobu Mita07dccf32006-09-29 02:00:22 -07005913 err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
5914 BUG_ON(err == NOTIFY_BAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005915 migration_call(&migration_notifier, CPU_ONLINE, cpu);
5916 register_cpu_notifier(&migration_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005917}
5918#endif
5919
5920#ifdef CONFIG_SMP
Christoph Lameter476f3532007-05-06 14:48:58 -07005921
5922/* Number of possible processor ids */
5923int nr_cpu_ids __read_mostly = NR_CPUS;
5924EXPORT_SYMBOL(nr_cpu_ids);
5925
Ingo Molnar3e9830d2007-10-15 17:00:13 +02005926#ifdef CONFIG_SCHED_DEBUG
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005927
5928static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level)
5929{
5930 struct sched_group *group = sd->groups;
5931 cpumask_t groupmask;
5932 char str[NR_CPUS];
5933
5934 cpumask_scnprintf(str, NR_CPUS, sd->span);
5935 cpus_clear(groupmask);
5936
5937 printk(KERN_DEBUG "%*s domain %d: ", level, "", level);
5938
5939 if (!(sd->flags & SD_LOAD_BALANCE)) {
5940 printk("does not load-balance\n");
5941 if (sd->parent)
5942 printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
5943 " has parent");
5944 return -1;
5945 }
5946
5947 printk(KERN_CONT "span %s\n", str);
5948
5949 if (!cpu_isset(cpu, sd->span)) {
5950 printk(KERN_ERR "ERROR: domain->span does not contain "
5951 "CPU%d\n", cpu);
5952 }
5953 if (!cpu_isset(cpu, group->cpumask)) {
5954 printk(KERN_ERR "ERROR: domain->groups does not contain"
5955 " CPU%d\n", cpu);
5956 }
5957
5958 printk(KERN_DEBUG "%*s groups:", level + 1, "");
5959 do {
5960 if (!group) {
5961 printk("\n");
5962 printk(KERN_ERR "ERROR: group is NULL\n");
5963 break;
5964 }
5965
5966 if (!group->__cpu_power) {
5967 printk(KERN_CONT "\n");
5968 printk(KERN_ERR "ERROR: domain->cpu_power not "
5969 "set\n");
5970 break;
5971 }
5972
5973 if (!cpus_weight(group->cpumask)) {
5974 printk(KERN_CONT "\n");
5975 printk(KERN_ERR "ERROR: empty group\n");
5976 break;
5977 }
5978
5979 if (cpus_intersects(groupmask, group->cpumask)) {
5980 printk(KERN_CONT "\n");
5981 printk(KERN_ERR "ERROR: repeated CPUs\n");
5982 break;
5983 }
5984
5985 cpus_or(groupmask, groupmask, group->cpumask);
5986
5987 cpumask_scnprintf(str, NR_CPUS, group->cpumask);
5988 printk(KERN_CONT " %s", str);
5989
5990 group = group->next;
5991 } while (group != sd->groups);
5992 printk(KERN_CONT "\n");
5993
5994 if (!cpus_equal(sd->span, groupmask))
5995 printk(KERN_ERR "ERROR: groups don't span domain->span\n");
5996
5997 if (sd->parent && !cpus_subset(groupmask, sd->parent->span))
5998 printk(KERN_ERR "ERROR: parent span is not a superset "
5999 "of domain->span\n");
6000 return 0;
6001}
6002
Linus Torvalds1da177e2005-04-16 15:20:36 -07006003static void sched_domain_debug(struct sched_domain *sd, int cpu)
6004{
6005 int level = 0;
6006
Nick Piggin41c7ce92005-06-25 14:57:24 -07006007 if (!sd) {
6008 printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
6009 return;
6010 }
6011
Linus Torvalds1da177e2005-04-16 15:20:36 -07006012 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
6013
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006014 for (;;) {
6015 if (sched_domain_debug_one(sd, cpu, level))
Linus Torvalds1da177e2005-04-16 15:20:36 -07006016 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006017 level++;
6018 sd = sd->parent;
Miguel Ojeda Sandonis33859f72006-12-10 02:20:38 -08006019 if (!sd)
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006020 break;
6021 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006022}
6023#else
Ingo Molnar48f24c42006-07-03 00:25:40 -07006024# define sched_domain_debug(sd, cpu) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006025#endif
6026
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006027static int sd_degenerate(struct sched_domain *sd)
Suresh Siddha245af2c2005-06-25 14:57:25 -07006028{
6029 if (cpus_weight(sd->span) == 1)
6030 return 1;
6031
6032 /* Following flags need at least 2 groups */
6033 if (sd->flags & (SD_LOAD_BALANCE |
6034 SD_BALANCE_NEWIDLE |
6035 SD_BALANCE_FORK |
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006036 SD_BALANCE_EXEC |
6037 SD_SHARE_CPUPOWER |
6038 SD_SHARE_PKG_RESOURCES)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006039 if (sd->groups != sd->groups->next)
6040 return 0;
6041 }
6042
6043 /* Following flags don't use groups */
6044 if (sd->flags & (SD_WAKE_IDLE |
6045 SD_WAKE_AFFINE |
6046 SD_WAKE_BALANCE))
6047 return 0;
6048
6049 return 1;
6050}
6051
Ingo Molnar48f24c42006-07-03 00:25:40 -07006052static int
6053sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
Suresh Siddha245af2c2005-06-25 14:57:25 -07006054{
6055 unsigned long cflags = sd->flags, pflags = parent->flags;
6056
6057 if (sd_degenerate(parent))
6058 return 1;
6059
6060 if (!cpus_equal(sd->span, parent->span))
6061 return 0;
6062
6063 /* Does parent contain flags not in child? */
6064 /* WAKE_BALANCE is a subset of WAKE_AFFINE */
6065 if (cflags & SD_WAKE_AFFINE)
6066 pflags &= ~SD_WAKE_BALANCE;
6067 /* Flags needing groups don't count if only 1 group in parent */
6068 if (parent->groups == parent->groups->next) {
6069 pflags &= ~(SD_LOAD_BALANCE |
6070 SD_BALANCE_NEWIDLE |
6071 SD_BALANCE_FORK |
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006072 SD_BALANCE_EXEC |
6073 SD_SHARE_CPUPOWER |
6074 SD_SHARE_PKG_RESOURCES);
Suresh Siddha245af2c2005-06-25 14:57:25 -07006075 }
6076 if (~cflags & pflags)
6077 return 0;
6078
6079 return 1;
6080}
6081
Gregory Haskins57d885f2008-01-25 21:08:18 +01006082static void rq_attach_root(struct rq *rq, struct root_domain *rd)
6083{
6084 unsigned long flags;
6085 const struct sched_class *class;
6086
6087 spin_lock_irqsave(&rq->lock, flags);
6088
6089 if (rq->rd) {
6090 struct root_domain *old_rd = rq->rd;
6091
Ingo Molnar0eab9142008-01-25 21:08:19 +01006092 for (class = sched_class_highest; class; class = class->next) {
Gregory Haskins57d885f2008-01-25 21:08:18 +01006093 if (class->leave_domain)
6094 class->leave_domain(rq);
Ingo Molnar0eab9142008-01-25 21:08:19 +01006095 }
Gregory Haskins57d885f2008-01-25 21:08:18 +01006096
Gregory Haskinsdc938522008-01-25 21:08:26 +01006097 cpu_clear(rq->cpu, old_rd->span);
6098 cpu_clear(rq->cpu, old_rd->online);
6099
Gregory Haskins57d885f2008-01-25 21:08:18 +01006100 if (atomic_dec_and_test(&old_rd->refcount))
6101 kfree(old_rd);
6102 }
6103
6104 atomic_inc(&rd->refcount);
6105 rq->rd = rd;
6106
Gregory Haskinsdc938522008-01-25 21:08:26 +01006107 cpu_set(rq->cpu, rd->span);
6108 if (cpu_isset(rq->cpu, cpu_online_map))
6109 cpu_set(rq->cpu, rd->online);
6110
Ingo Molnar0eab9142008-01-25 21:08:19 +01006111 for (class = sched_class_highest; class; class = class->next) {
Gregory Haskins57d885f2008-01-25 21:08:18 +01006112 if (class->join_domain)
6113 class->join_domain(rq);
Ingo Molnar0eab9142008-01-25 21:08:19 +01006114 }
Gregory Haskins57d885f2008-01-25 21:08:18 +01006115
6116 spin_unlock_irqrestore(&rq->lock, flags);
6117}
6118
Gregory Haskinsdc938522008-01-25 21:08:26 +01006119static void init_rootdomain(struct root_domain *rd)
Gregory Haskins57d885f2008-01-25 21:08:18 +01006120{
6121 memset(rd, 0, sizeof(*rd));
6122
Gregory Haskinsdc938522008-01-25 21:08:26 +01006123 cpus_clear(rd->span);
6124 cpus_clear(rd->online);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006125}
6126
6127static void init_defrootdomain(void)
6128{
Gregory Haskinsdc938522008-01-25 21:08:26 +01006129 init_rootdomain(&def_root_domain);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006130 atomic_set(&def_root_domain.refcount, 1);
6131}
6132
Gregory Haskinsdc938522008-01-25 21:08:26 +01006133static struct root_domain *alloc_rootdomain(void)
Gregory Haskins57d885f2008-01-25 21:08:18 +01006134{
6135 struct root_domain *rd;
6136
6137 rd = kmalloc(sizeof(*rd), GFP_KERNEL);
6138 if (!rd)
6139 return NULL;
6140
Gregory Haskinsdc938522008-01-25 21:08:26 +01006141 init_rootdomain(rd);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006142
6143 return rd;
6144}
6145
Linus Torvalds1da177e2005-04-16 15:20:36 -07006146/*
Ingo Molnar0eab9142008-01-25 21:08:19 +01006147 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
Linus Torvalds1da177e2005-04-16 15:20:36 -07006148 * hold the hotplug lock.
6149 */
Ingo Molnar0eab9142008-01-25 21:08:19 +01006150static void
6151cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006152{
Ingo Molnar70b97a72006-07-03 00:25:42 -07006153 struct rq *rq = cpu_rq(cpu);
Suresh Siddha245af2c2005-06-25 14:57:25 -07006154 struct sched_domain *tmp;
6155
6156 /* Remove the sched domains which do not contribute to scheduling. */
6157 for (tmp = sd; tmp; tmp = tmp->parent) {
6158 struct sched_domain *parent = tmp->parent;
6159 if (!parent)
6160 break;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006161 if (sd_parent_degenerate(tmp, parent)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006162 tmp->parent = parent->parent;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006163 if (parent->parent)
6164 parent->parent->child = tmp;
6165 }
Suresh Siddha245af2c2005-06-25 14:57:25 -07006166 }
6167
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006168 if (sd && sd_degenerate(sd)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006169 sd = sd->parent;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006170 if (sd)
6171 sd->child = NULL;
6172 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006173
6174 sched_domain_debug(sd, cpu);
6175
Gregory Haskins57d885f2008-01-25 21:08:18 +01006176 rq_attach_root(rq, rd);
Nick Piggin674311d2005-06-25 14:57:27 -07006177 rcu_assign_pointer(rq->sd, sd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006178}
6179
6180/* cpus with isolated domains */
Tim Chen67af63a2006-12-22 01:07:50 -08006181static cpumask_t cpu_isolated_map = CPU_MASK_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006182
6183/* Setup the mask of cpus configured for isolated domains */
6184static int __init isolated_cpu_setup(char *str)
6185{
6186 int ints[NR_CPUS], i;
6187
6188 str = get_options(str, ARRAY_SIZE(ints), ints);
6189 cpus_clear(cpu_isolated_map);
6190 for (i = 1; i <= ints[0]; i++)
6191 if (ints[i] < NR_CPUS)
6192 cpu_set(ints[i], cpu_isolated_map);
6193 return 1;
6194}
6195
Ingo Molnar8927f492007-10-15 17:00:13 +02006196__setup("isolcpus=", isolated_cpu_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006197
6198/*
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006199 * init_sched_build_groups takes the cpumask we wish to span, and a pointer
6200 * to a function which identifies what group(along with sched group) a CPU
6201 * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS
6202 * (due to the fact that we keep track of groups covered with a cpumask_t).
Linus Torvalds1da177e2005-04-16 15:20:36 -07006203 *
6204 * init_sched_build_groups will build a circular linked list of the groups
6205 * covered by the given span, and will set each group's ->cpumask correctly,
6206 * and ->cpu_power to 0.
6207 */
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006208static void
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006209init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map,
6210 int (*group_fn)(int cpu, const cpumask_t *cpu_map,
6211 struct sched_group **sg))
Linus Torvalds1da177e2005-04-16 15:20:36 -07006212{
6213 struct sched_group *first = NULL, *last = NULL;
6214 cpumask_t covered = CPU_MASK_NONE;
6215 int i;
6216
6217 for_each_cpu_mask(i, span) {
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006218 struct sched_group *sg;
6219 int group = group_fn(i, cpu_map, &sg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006220 int j;
6221
6222 if (cpu_isset(i, covered))
6223 continue;
6224
6225 sg->cpumask = CPU_MASK_NONE;
Eric Dumazet5517d862007-05-08 00:32:57 -07006226 sg->__cpu_power = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006227
6228 for_each_cpu_mask(j, span) {
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006229 if (group_fn(j, cpu_map, NULL) != group)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006230 continue;
6231
6232 cpu_set(j, covered);
6233 cpu_set(j, sg->cpumask);
6234 }
6235 if (!first)
6236 first = sg;
6237 if (last)
6238 last->next = sg;
6239 last = sg;
6240 }
6241 last->next = first;
6242}
6243
John Hawkes9c1cfda2005-09-06 15:18:14 -07006244#define SD_NODES_PER_DOMAIN 16
Linus Torvalds1da177e2005-04-16 15:20:36 -07006245
John Hawkes9c1cfda2005-09-06 15:18:14 -07006246#ifdef CONFIG_NUMA
akpm@osdl.org198e2f12006-01-12 01:05:30 -08006247
John Hawkes9c1cfda2005-09-06 15:18:14 -07006248/**
6249 * find_next_best_node - find the next node to include in a sched_domain
6250 * @node: node whose sched_domain we're building
6251 * @used_nodes: nodes already in the sched_domain
6252 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006253 * Find the next node to include in a given scheduling domain. Simply
John Hawkes9c1cfda2005-09-06 15:18:14 -07006254 * finds the closest node not already in the @used_nodes map.
6255 *
6256 * Should use nodemask_t.
6257 */
6258static int find_next_best_node(int node, unsigned long *used_nodes)
6259{
6260 int i, n, val, min_val, best_node = 0;
6261
6262 min_val = INT_MAX;
6263
6264 for (i = 0; i < MAX_NUMNODES; i++) {
6265 /* Start at @node */
6266 n = (node + i) % MAX_NUMNODES;
6267
6268 if (!nr_cpus_node(n))
6269 continue;
6270
6271 /* Skip already used nodes */
6272 if (test_bit(n, used_nodes))
6273 continue;
6274
6275 /* Simple min distance search */
6276 val = node_distance(node, n);
6277
6278 if (val < min_val) {
6279 min_val = val;
6280 best_node = n;
6281 }
6282 }
6283
6284 set_bit(best_node, used_nodes);
6285 return best_node;
6286}
6287
6288/**
6289 * sched_domain_node_span - get a cpumask for a node's sched_domain
6290 * @node: node whose cpumask we're constructing
6291 * @size: number of nodes to include in this span
6292 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006293 * Given a node, construct a good cpumask for its sched_domain to span. It
John Hawkes9c1cfda2005-09-06 15:18:14 -07006294 * should be one that prevents unnecessary balancing, but also spreads tasks
6295 * out optimally.
6296 */
6297static cpumask_t sched_domain_node_span(int node)
6298{
John Hawkes9c1cfda2005-09-06 15:18:14 -07006299 DECLARE_BITMAP(used_nodes, MAX_NUMNODES);
Ingo Molnar48f24c42006-07-03 00:25:40 -07006300 cpumask_t span, nodemask;
6301 int i;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006302
6303 cpus_clear(span);
6304 bitmap_zero(used_nodes, MAX_NUMNODES);
6305
6306 nodemask = node_to_cpumask(node);
6307 cpus_or(span, span, nodemask);
6308 set_bit(node, used_nodes);
6309
6310 for (i = 1; i < SD_NODES_PER_DOMAIN; i++) {
6311 int next_node = find_next_best_node(node, used_nodes);
Ingo Molnar48f24c42006-07-03 00:25:40 -07006312
John Hawkes9c1cfda2005-09-06 15:18:14 -07006313 nodemask = node_to_cpumask(next_node);
6314 cpus_or(span, span, nodemask);
6315 }
6316
6317 return span;
6318}
6319#endif
6320
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006321int sched_smt_power_savings = 0, sched_mc_power_savings = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07006322
John Hawkes9c1cfda2005-09-06 15:18:14 -07006323/*
Ingo Molnar48f24c42006-07-03 00:25:40 -07006324 * SMT sched-domains:
John Hawkes9c1cfda2005-09-06 15:18:14 -07006325 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006326#ifdef CONFIG_SCHED_SMT
6327static DEFINE_PER_CPU(struct sched_domain, cpu_domains);
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006328static DEFINE_PER_CPU(struct sched_group, sched_group_cpus);
Ingo Molnar48f24c42006-07-03 00:25:40 -07006329
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006330static int
6331cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006332{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006333 if (sg)
6334 *sg = &per_cpu(sched_group_cpus, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006335 return cpu;
6336}
6337#endif
6338
Ingo Molnar48f24c42006-07-03 00:25:40 -07006339/*
6340 * multi-core sched-domains:
6341 */
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006342#ifdef CONFIG_SCHED_MC
6343static DEFINE_PER_CPU(struct sched_domain, core_domains);
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006344static DEFINE_PER_CPU(struct sched_group, sched_group_core);
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006345#endif
6346
6347#if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT)
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006348static int
6349cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg)
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006350{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006351 int group;
Mike Travisd5a74302007-10-16 01:24:05 -07006352 cpumask_t mask = per_cpu(cpu_sibling_map, cpu);
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006353 cpus_and(mask, mask, *cpu_map);
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006354 group = first_cpu(mask);
6355 if (sg)
6356 *sg = &per_cpu(sched_group_core, group);
6357 return group;
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006358}
6359#elif defined(CONFIG_SCHED_MC)
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006360static int
6361cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg)
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006362{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006363 if (sg)
6364 *sg = &per_cpu(sched_group_core, cpu);
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006365 return cpu;
6366}
6367#endif
6368
Linus Torvalds1da177e2005-04-16 15:20:36 -07006369static DEFINE_PER_CPU(struct sched_domain, phys_domains);
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006370static DEFINE_PER_CPU(struct sched_group, sched_group_phys);
Ingo Molnar48f24c42006-07-03 00:25:40 -07006371
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006372static int
6373cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006374{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006375 int group;
Ingo Molnar48f24c42006-07-03 00:25:40 -07006376#ifdef CONFIG_SCHED_MC
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006377 cpumask_t mask = cpu_coregroup_map(cpu);
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006378 cpus_and(mask, mask, *cpu_map);
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006379 group = first_cpu(mask);
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006380#elif defined(CONFIG_SCHED_SMT)
Mike Travisd5a74302007-10-16 01:24:05 -07006381 cpumask_t mask = per_cpu(cpu_sibling_map, cpu);
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006382 cpus_and(mask, mask, *cpu_map);
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006383 group = first_cpu(mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006384#else
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006385 group = cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006386#endif
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006387 if (sg)
6388 *sg = &per_cpu(sched_group_phys, group);
6389 return group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006390}
6391
6392#ifdef CONFIG_NUMA
John Hawkes9c1cfda2005-09-06 15:18:14 -07006393/*
6394 * The init_sched_build_groups can't handle what we want to do with node
6395 * groups, so roll our own. Now each node has its own list of groups which
6396 * gets dynamically allocated.
6397 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006398static DEFINE_PER_CPU(struct sched_domain, node_domains);
John Hawkesd1b55132005-09-06 15:18:14 -07006399static struct sched_group **sched_group_nodes_bycpu[NR_CPUS];
John Hawkes9c1cfda2005-09-06 15:18:14 -07006400
6401static DEFINE_PER_CPU(struct sched_domain, allnodes_domains);
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006402static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006403
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006404static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map,
6405 struct sched_group **sg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006406{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006407 cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu));
6408 int group;
6409
6410 cpus_and(nodemask, nodemask, *cpu_map);
6411 group = first_cpu(nodemask);
6412
6413 if (sg)
6414 *sg = &per_cpu(sched_group_allnodes, group);
6415 return group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006416}
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006417
Siddha, Suresh B08069032006-03-27 01:15:23 -08006418static void init_numa_sched_groups_power(struct sched_group *group_head)
6419{
6420 struct sched_group *sg = group_head;
6421 int j;
6422
6423 if (!sg)
6424 return;
Andi Kleen3a5c3592007-10-15 17:00:14 +02006425 do {
6426 for_each_cpu_mask(j, sg->cpumask) {
6427 struct sched_domain *sd;
Siddha, Suresh B08069032006-03-27 01:15:23 -08006428
Andi Kleen3a5c3592007-10-15 17:00:14 +02006429 sd = &per_cpu(phys_domains, j);
6430 if (j != first_cpu(sd->groups->cpumask)) {
6431 /*
6432 * Only add "power" once for each
6433 * physical package.
6434 */
6435 continue;
6436 }
6437
6438 sg_inc_cpu_power(sg, sd->groups->__cpu_power);
Siddha, Suresh B08069032006-03-27 01:15:23 -08006439 }
Andi Kleen3a5c3592007-10-15 17:00:14 +02006440 sg = sg->next;
6441 } while (sg != group_head);
Siddha, Suresh B08069032006-03-27 01:15:23 -08006442}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006443#endif
6444
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006445#ifdef CONFIG_NUMA
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006446/* Free memory allocated for various sched_group structures */
6447static void free_sched_groups(const cpumask_t *cpu_map)
6448{
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006449 int cpu, i;
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006450
6451 for_each_cpu_mask(cpu, *cpu_map) {
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006452 struct sched_group **sched_group_nodes
6453 = sched_group_nodes_bycpu[cpu];
6454
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006455 if (!sched_group_nodes)
6456 continue;
6457
6458 for (i = 0; i < MAX_NUMNODES; i++) {
6459 cpumask_t nodemask = node_to_cpumask(i);
6460 struct sched_group *oldsg, *sg = sched_group_nodes[i];
6461
6462 cpus_and(nodemask, nodemask, *cpu_map);
6463 if (cpus_empty(nodemask))
6464 continue;
6465
6466 if (sg == NULL)
6467 continue;
6468 sg = sg->next;
6469next_sg:
6470 oldsg = sg;
6471 sg = sg->next;
6472 kfree(oldsg);
6473 if (oldsg != sched_group_nodes[i])
6474 goto next_sg;
6475 }
6476 kfree(sched_group_nodes);
6477 sched_group_nodes_bycpu[cpu] = NULL;
6478 }
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006479}
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006480#else
6481static void free_sched_groups(const cpumask_t *cpu_map)
6482{
6483}
6484#endif
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006485
Linus Torvalds1da177e2005-04-16 15:20:36 -07006486/*
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006487 * Initialize sched groups cpu_power.
6488 *
6489 * cpu_power indicates the capacity of sched group, which is used while
6490 * distributing the load between different sched groups in a sched domain.
6491 * Typically cpu_power for all the groups in a sched domain will be same unless
6492 * there are asymmetries in the topology. If there are asymmetries, group
6493 * having more cpu_power will pickup more load compared to the group having
6494 * less cpu_power.
6495 *
6496 * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents
6497 * the maximum number of tasks a group can handle in the presence of other idle
6498 * or lightly loaded groups in the same sched domain.
6499 */
6500static void init_sched_groups_power(int cpu, struct sched_domain *sd)
6501{
6502 struct sched_domain *child;
6503 struct sched_group *group;
6504
6505 WARN_ON(!sd || !sd->groups);
6506
6507 if (cpu != first_cpu(sd->groups->cpumask))
6508 return;
6509
6510 child = sd->child;
6511
Eric Dumazet5517d862007-05-08 00:32:57 -07006512 sd->groups->__cpu_power = 0;
6513
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006514 /*
6515 * For perf policy, if the groups in child domain share resources
6516 * (for example cores sharing some portions of the cache hierarchy
6517 * or SMT), then set this domain groups cpu_power such that each group
6518 * can handle only one task, when there are other idle groups in the
6519 * same sched domain.
6520 */
6521 if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) &&
6522 (child->flags &
6523 (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) {
Eric Dumazet5517d862007-05-08 00:32:57 -07006524 sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE);
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006525 return;
6526 }
6527
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006528 /*
6529 * add cpu_power of each child group to this groups cpu_power
6530 */
6531 group = child->groups;
6532 do {
Eric Dumazet5517d862007-05-08 00:32:57 -07006533 sg_inc_cpu_power(sd->groups, group->__cpu_power);
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006534 group = group->next;
6535 } while (group != child->groups);
6536}
6537
6538/*
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006539 * Build sched domains for a given set of cpus and attach the sched domains
6540 * to the individual cpus
Linus Torvalds1da177e2005-04-16 15:20:36 -07006541 */
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006542static int build_sched_domains(const cpumask_t *cpu_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006543{
6544 int i;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006545 struct root_domain *rd;
John Hawkesd1b55132005-09-06 15:18:14 -07006546#ifdef CONFIG_NUMA
6547 struct sched_group **sched_group_nodes = NULL;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006548 int sd_allnodes = 0;
John Hawkesd1b55132005-09-06 15:18:14 -07006549
6550 /*
6551 * Allocate the per-node list of sched groups
6552 */
Milton Miller5cf9f062007-10-15 17:00:19 +02006553 sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *),
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006554 GFP_KERNEL);
John Hawkesd1b55132005-09-06 15:18:14 -07006555 if (!sched_group_nodes) {
6556 printk(KERN_WARNING "Can not alloc sched group node list\n");
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006557 return -ENOMEM;
John Hawkesd1b55132005-09-06 15:18:14 -07006558 }
6559 sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes;
6560#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07006561
Gregory Haskinsdc938522008-01-25 21:08:26 +01006562 rd = alloc_rootdomain();
Gregory Haskins57d885f2008-01-25 21:08:18 +01006563 if (!rd) {
6564 printk(KERN_WARNING "Cannot alloc root domain\n");
6565 return -ENOMEM;
6566 }
6567
Linus Torvalds1da177e2005-04-16 15:20:36 -07006568 /*
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006569 * Set up domains for cpus specified by the cpu_map.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006570 */
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006571 for_each_cpu_mask(i, *cpu_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006572 struct sched_domain *sd = NULL, *p;
6573 cpumask_t nodemask = node_to_cpumask(cpu_to_node(i));
6574
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006575 cpus_and(nodemask, nodemask, *cpu_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006576
6577#ifdef CONFIG_NUMA
Ingo Molnardd41f592007-07-09 18:51:59 +02006578 if (cpus_weight(*cpu_map) >
6579 SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) {
John Hawkes9c1cfda2005-09-06 15:18:14 -07006580 sd = &per_cpu(allnodes_domains, i);
6581 *sd = SD_ALLNODES_INIT;
6582 sd->span = *cpu_map;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006583 cpu_to_allnodes_group(i, cpu_map, &sd->groups);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006584 p = sd;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006585 sd_allnodes = 1;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006586 } else
6587 p = NULL;
6588
Linus Torvalds1da177e2005-04-16 15:20:36 -07006589 sd = &per_cpu(node_domains, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006590 *sd = SD_NODE_INIT;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006591 sd->span = sched_domain_node_span(cpu_to_node(i));
6592 sd->parent = p;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006593 if (p)
6594 p->child = sd;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006595 cpus_and(sd->span, sd->span, *cpu_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006596#endif
6597
6598 p = sd;
6599 sd = &per_cpu(phys_domains, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006600 *sd = SD_CPU_INIT;
6601 sd->span = nodemask;
6602 sd->parent = p;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006603 if (p)
6604 p->child = sd;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006605 cpu_to_phys_group(i, cpu_map, &sd->groups);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006606
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006607#ifdef CONFIG_SCHED_MC
6608 p = sd;
6609 sd = &per_cpu(core_domains, i);
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006610 *sd = SD_MC_INIT;
6611 sd->span = cpu_coregroup_map(i);
6612 cpus_and(sd->span, sd->span, *cpu_map);
6613 sd->parent = p;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006614 p->child = sd;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006615 cpu_to_core_group(i, cpu_map, &sd->groups);
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006616#endif
6617
Linus Torvalds1da177e2005-04-16 15:20:36 -07006618#ifdef CONFIG_SCHED_SMT
6619 p = sd;
6620 sd = &per_cpu(cpu_domains, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006621 *sd = SD_SIBLING_INIT;
Mike Travisd5a74302007-10-16 01:24:05 -07006622 sd->span = per_cpu(cpu_sibling_map, i);
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006623 cpus_and(sd->span, sd->span, *cpu_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006624 sd->parent = p;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006625 p->child = sd;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006626 cpu_to_cpu_group(i, cpu_map, &sd->groups);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006627#endif
6628 }
6629
6630#ifdef CONFIG_SCHED_SMT
6631 /* Set up CPU (sibling) groups */
John Hawkes9c1cfda2005-09-06 15:18:14 -07006632 for_each_cpu_mask(i, *cpu_map) {
Mike Travisd5a74302007-10-16 01:24:05 -07006633 cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i);
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006634 cpus_and(this_sibling_map, this_sibling_map, *cpu_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006635 if (i != first_cpu(this_sibling_map))
6636 continue;
6637
Ingo Molnardd41f592007-07-09 18:51:59 +02006638 init_sched_build_groups(this_sibling_map, cpu_map,
6639 &cpu_to_cpu_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006640 }
6641#endif
6642
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006643#ifdef CONFIG_SCHED_MC
6644 /* Set up multi-core groups */
6645 for_each_cpu_mask(i, *cpu_map) {
6646 cpumask_t this_core_map = cpu_coregroup_map(i);
6647 cpus_and(this_core_map, this_core_map, *cpu_map);
6648 if (i != first_cpu(this_core_map))
6649 continue;
Ingo Molnardd41f592007-07-09 18:51:59 +02006650 init_sched_build_groups(this_core_map, cpu_map,
6651 &cpu_to_core_group);
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006652 }
6653#endif
6654
Linus Torvalds1da177e2005-04-16 15:20:36 -07006655 /* Set up physical groups */
6656 for (i = 0; i < MAX_NUMNODES; i++) {
6657 cpumask_t nodemask = node_to_cpumask(i);
6658
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006659 cpus_and(nodemask, nodemask, *cpu_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006660 if (cpus_empty(nodemask))
6661 continue;
6662
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006663 init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006664 }
6665
6666#ifdef CONFIG_NUMA
6667 /* Set up node groups */
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006668 if (sd_allnodes)
Ingo Molnardd41f592007-07-09 18:51:59 +02006669 init_sched_build_groups(*cpu_map, cpu_map,
6670 &cpu_to_allnodes_group);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006671
6672 for (i = 0; i < MAX_NUMNODES; i++) {
6673 /* Set up node groups */
6674 struct sched_group *sg, *prev;
6675 cpumask_t nodemask = node_to_cpumask(i);
6676 cpumask_t domainspan;
6677 cpumask_t covered = CPU_MASK_NONE;
6678 int j;
6679
6680 cpus_and(nodemask, nodemask, *cpu_map);
John Hawkesd1b55132005-09-06 15:18:14 -07006681 if (cpus_empty(nodemask)) {
6682 sched_group_nodes[i] = NULL;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006683 continue;
John Hawkesd1b55132005-09-06 15:18:14 -07006684 }
John Hawkes9c1cfda2005-09-06 15:18:14 -07006685
6686 domainspan = sched_domain_node_span(i);
6687 cpus_and(domainspan, domainspan, *cpu_map);
6688
Srivatsa Vaddagiri15f0b672006-06-27 02:54:40 -07006689 sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i);
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006690 if (!sg) {
6691 printk(KERN_WARNING "Can not alloc domain group for "
6692 "node %d\n", i);
6693 goto error;
6694 }
John Hawkes9c1cfda2005-09-06 15:18:14 -07006695 sched_group_nodes[i] = sg;
6696 for_each_cpu_mask(j, nodemask) {
6697 struct sched_domain *sd;
Ingo Molnar9761eea2007-07-09 18:52:00 +02006698
John Hawkes9c1cfda2005-09-06 15:18:14 -07006699 sd = &per_cpu(node_domains, j);
6700 sd->groups = sg;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006701 }
Eric Dumazet5517d862007-05-08 00:32:57 -07006702 sg->__cpu_power = 0;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006703 sg->cpumask = nodemask;
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006704 sg->next = sg;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006705 cpus_or(covered, covered, nodemask);
6706 prev = sg;
6707
6708 for (j = 0; j < MAX_NUMNODES; j++) {
6709 cpumask_t tmp, notcovered;
6710 int n = (i + j) % MAX_NUMNODES;
6711
6712 cpus_complement(notcovered, covered);
6713 cpus_and(tmp, notcovered, *cpu_map);
6714 cpus_and(tmp, tmp, domainspan);
6715 if (cpus_empty(tmp))
6716 break;
6717
6718 nodemask = node_to_cpumask(n);
6719 cpus_and(tmp, tmp, nodemask);
6720 if (cpus_empty(tmp))
6721 continue;
6722
Srivatsa Vaddagiri15f0b672006-06-27 02:54:40 -07006723 sg = kmalloc_node(sizeof(struct sched_group),
6724 GFP_KERNEL, i);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006725 if (!sg) {
6726 printk(KERN_WARNING
6727 "Can not alloc domain group for node %d\n", j);
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006728 goto error;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006729 }
Eric Dumazet5517d862007-05-08 00:32:57 -07006730 sg->__cpu_power = 0;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006731 sg->cpumask = tmp;
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006732 sg->next = prev->next;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006733 cpus_or(covered, covered, tmp);
6734 prev->next = sg;
6735 prev = sg;
6736 }
John Hawkes9c1cfda2005-09-06 15:18:14 -07006737 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006738#endif
6739
6740 /* Calculate CPU power for physical packages and nodes */
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006741#ifdef CONFIG_SCHED_SMT
6742 for_each_cpu_mask(i, *cpu_map) {
Ingo Molnardd41f592007-07-09 18:51:59 +02006743 struct sched_domain *sd = &per_cpu(cpu_domains, i);
6744
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006745 init_sched_groups_power(i, sd);
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006746 }
6747#endif
6748#ifdef CONFIG_SCHED_MC
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006749 for_each_cpu_mask(i, *cpu_map) {
Ingo Molnardd41f592007-07-09 18:51:59 +02006750 struct sched_domain *sd = &per_cpu(core_domains, i);
6751
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006752 init_sched_groups_power(i, sd);
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006753 }
6754#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07006755
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006756 for_each_cpu_mask(i, *cpu_map) {
Ingo Molnardd41f592007-07-09 18:51:59 +02006757 struct sched_domain *sd = &per_cpu(phys_domains, i);
6758
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006759 init_sched_groups_power(i, sd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006760 }
6761
John Hawkes9c1cfda2005-09-06 15:18:14 -07006762#ifdef CONFIG_NUMA
Siddha, Suresh B08069032006-03-27 01:15:23 -08006763 for (i = 0; i < MAX_NUMNODES; i++)
6764 init_numa_sched_groups_power(sched_group_nodes[i]);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006765
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006766 if (sd_allnodes) {
6767 struct sched_group *sg;
Siddha, Suresh Bf712c0c2006-07-30 03:02:59 -07006768
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006769 cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg);
Siddha, Suresh Bf712c0c2006-07-30 03:02:59 -07006770 init_numa_sched_groups_power(sg);
6771 }
John Hawkes9c1cfda2005-09-06 15:18:14 -07006772#endif
6773
Linus Torvalds1da177e2005-04-16 15:20:36 -07006774 /* Attach the domains */
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006775 for_each_cpu_mask(i, *cpu_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006776 struct sched_domain *sd;
6777#ifdef CONFIG_SCHED_SMT
6778 sd = &per_cpu(cpu_domains, i);
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006779#elif defined(CONFIG_SCHED_MC)
6780 sd = &per_cpu(core_domains, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006781#else
6782 sd = &per_cpu(phys_domains, i);
6783#endif
Gregory Haskins57d885f2008-01-25 21:08:18 +01006784 cpu_attach_domain(sd, rd, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006785 }
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006786
6787 return 0;
6788
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006789#ifdef CONFIG_NUMA
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006790error:
6791 free_sched_groups(cpu_map);
6792 return -ENOMEM;
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006793#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07006794}
Paul Jackson029190c2007-10-18 23:40:20 -07006795
6796static cpumask_t *doms_cur; /* current sched domains */
6797static int ndoms_cur; /* number of sched domains in 'doms_cur' */
6798
6799/*
6800 * Special case: If a kmalloc of a doms_cur partition (array of
6801 * cpumask_t) fails, then fallback to a single sched domain,
6802 * as determined by the single cpumask_t fallback_doms.
6803 */
6804static cpumask_t fallback_doms;
6805
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006806/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006807 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
Paul Jackson029190c2007-10-18 23:40:20 -07006808 * For now this just excludes isolated cpus, but could be used to
6809 * exclude other special cases in the future.
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006810 */
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006811static int arch_init_sched_domains(const cpumask_t *cpu_map)
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006812{
Milton Miller73785472007-10-24 18:23:48 +02006813 int err;
6814
Paul Jackson029190c2007-10-18 23:40:20 -07006815 ndoms_cur = 1;
6816 doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL);
6817 if (!doms_cur)
6818 doms_cur = &fallback_doms;
6819 cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map);
Milton Miller73785472007-10-24 18:23:48 +02006820 err = build_sched_domains(doms_cur);
Milton Miller6382bc92007-10-15 17:00:19 +02006821 register_sched_domain_sysctl();
Milton Miller73785472007-10-24 18:23:48 +02006822
6823 return err;
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006824}
6825
6826static void arch_destroy_sched_domains(const cpumask_t *cpu_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006827{
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006828 free_sched_groups(cpu_map);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006829}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006830
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006831/*
6832 * Detach sched domains from a group of cpus specified in cpu_map
6833 * These cpus will now be attached to the NULL domain
6834 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08006835static void detach_destroy_domains(const cpumask_t *cpu_map)
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006836{
6837 int i;
6838
Milton Miller6382bc92007-10-15 17:00:19 +02006839 unregister_sched_domain_sysctl();
6840
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006841 for_each_cpu_mask(i, *cpu_map)
Gregory Haskins57d885f2008-01-25 21:08:18 +01006842 cpu_attach_domain(NULL, &def_root_domain, i);
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006843 synchronize_sched();
6844 arch_destroy_sched_domains(cpu_map);
6845}
6846
Paul Jackson029190c2007-10-18 23:40:20 -07006847/*
6848 * Partition sched domains as specified by the 'ndoms_new'
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006849 * cpumasks in the array doms_new[] of cpumasks. This compares
Paul Jackson029190c2007-10-18 23:40:20 -07006850 * doms_new[] to the current sched domain partitioning, doms_cur[].
6851 * It destroys each deleted domain and builds each new domain.
6852 *
6853 * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'.
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006854 * The masks don't intersect (don't overlap.) We should setup one
6855 * sched domain for each mask. CPUs not in any of the cpumasks will
6856 * not be load balanced. If the same cpumask appears both in the
Paul Jackson029190c2007-10-18 23:40:20 -07006857 * current 'doms_cur' domains and in the new 'doms_new', we can leave
6858 * it as it is.
6859 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006860 * The passed in 'doms_new' should be kmalloc'd. This routine takes
6861 * ownership of it and will kfree it when done with it. If the caller
Paul Jackson029190c2007-10-18 23:40:20 -07006862 * failed the kmalloc call, then it can pass in doms_new == NULL,
6863 * and partition_sched_domains() will fallback to the single partition
6864 * 'fallback_doms'.
6865 *
6866 * Call with hotplug lock held
6867 */
6868void partition_sched_domains(int ndoms_new, cpumask_t *doms_new)
6869{
6870 int i, j;
6871
Srivatsa Vaddagiria1835612008-01-25 21:08:00 +01006872 lock_doms_cur();
6873
Milton Miller73785472007-10-24 18:23:48 +02006874 /* always unregister in case we don't destroy any domains */
6875 unregister_sched_domain_sysctl();
6876
Paul Jackson029190c2007-10-18 23:40:20 -07006877 if (doms_new == NULL) {
6878 ndoms_new = 1;
6879 doms_new = &fallback_doms;
6880 cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map);
6881 }
6882
6883 /* Destroy deleted domains */
6884 for (i = 0; i < ndoms_cur; i++) {
6885 for (j = 0; j < ndoms_new; j++) {
6886 if (cpus_equal(doms_cur[i], doms_new[j]))
6887 goto match1;
6888 }
6889 /* no match - a current sched domain not in new doms_new[] */
6890 detach_destroy_domains(doms_cur + i);
6891match1:
6892 ;
6893 }
6894
6895 /* Build new domains */
6896 for (i = 0; i < ndoms_new; i++) {
6897 for (j = 0; j < ndoms_cur; j++) {
6898 if (cpus_equal(doms_new[i], doms_cur[j]))
6899 goto match2;
6900 }
6901 /* no match - add a new doms_new */
6902 build_sched_domains(doms_new + i);
6903match2:
6904 ;
6905 }
6906
6907 /* Remember the new sched domains */
6908 if (doms_cur != &fallback_doms)
6909 kfree(doms_cur);
6910 doms_cur = doms_new;
6911 ndoms_cur = ndoms_new;
Milton Miller73785472007-10-24 18:23:48 +02006912
6913 register_sched_domain_sysctl();
Srivatsa Vaddagiria1835612008-01-25 21:08:00 +01006914
6915 unlock_doms_cur();
Paul Jackson029190c2007-10-18 23:40:20 -07006916}
6917
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006918#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
Adrian Bunk6707de002007-08-12 18:08:19 +02006919static int arch_reinit_sched_domains(void)
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006920{
6921 int err;
6922
Gautham R Shenoy95402b32008-01-25 21:08:02 +01006923 get_online_cpus();
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006924 detach_destroy_domains(&cpu_online_map);
6925 err = arch_init_sched_domains(&cpu_online_map);
Gautham R Shenoy95402b32008-01-25 21:08:02 +01006926 put_online_cpus();
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006927
6928 return err;
6929}
6930
6931static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt)
6932{
6933 int ret;
6934
6935 if (buf[0] != '0' && buf[0] != '1')
6936 return -EINVAL;
6937
6938 if (smt)
6939 sched_smt_power_savings = (buf[0] == '1');
6940 else
6941 sched_mc_power_savings = (buf[0] == '1');
6942
6943 ret = arch_reinit_sched_domains();
6944
6945 return ret ? ret : count;
6946}
6947
Adrian Bunk6707de002007-08-12 18:08:19 +02006948#ifdef CONFIG_SCHED_MC
6949static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page)
6950{
6951 return sprintf(page, "%u\n", sched_mc_power_savings);
6952}
6953static ssize_t sched_mc_power_savings_store(struct sys_device *dev,
6954 const char *buf, size_t count)
6955{
6956 return sched_power_savings_store(buf, count, 0);
6957}
6958static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show,
6959 sched_mc_power_savings_store);
6960#endif
6961
6962#ifdef CONFIG_SCHED_SMT
6963static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page)
6964{
6965 return sprintf(page, "%u\n", sched_smt_power_savings);
6966}
6967static ssize_t sched_smt_power_savings_store(struct sys_device *dev,
6968 const char *buf, size_t count)
6969{
6970 return sched_power_savings_store(buf, count, 1);
6971}
6972static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show,
6973 sched_smt_power_savings_store);
6974#endif
6975
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006976int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls)
6977{
6978 int err = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07006979
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006980#ifdef CONFIG_SCHED_SMT
6981 if (smt_capable())
6982 err = sysfs_create_file(&cls->kset.kobj,
6983 &attr_sched_smt_power_savings.attr);
6984#endif
6985#ifdef CONFIG_SCHED_MC
6986 if (!err && mc_capable())
6987 err = sysfs_create_file(&cls->kset.kobj,
6988 &attr_sched_mc_power_savings.attr);
6989#endif
6990 return err;
6991}
6992#endif
6993
Linus Torvalds1da177e2005-04-16 15:20:36 -07006994/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006995 * Force a reinitialization of the sched domains hierarchy. The domains
Linus Torvalds1da177e2005-04-16 15:20:36 -07006996 * and groups cannot be updated in place without racing with the balancing
Nick Piggin41c7ce92005-06-25 14:57:24 -07006997 * code, so we temporarily attach all running cpus to the NULL domain
Linus Torvalds1da177e2005-04-16 15:20:36 -07006998 * which will prevent rebalancing while the sched domains are recalculated.
6999 */
7000static int update_sched_domains(struct notifier_block *nfb,
7001 unsigned long action, void *hcpu)
7002{
Linus Torvalds1da177e2005-04-16 15:20:36 -07007003 switch (action) {
7004 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007005 case CPU_UP_PREPARE_FROZEN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07007006 case CPU_DOWN_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007007 case CPU_DOWN_PREPARE_FROZEN:
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007008 detach_destroy_domains(&cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007009 return NOTIFY_OK;
7010
7011 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007012 case CPU_UP_CANCELED_FROZEN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07007013 case CPU_DOWN_FAILED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007014 case CPU_DOWN_FAILED_FROZEN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07007015 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007016 case CPU_ONLINE_FROZEN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07007017 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007018 case CPU_DEAD_FROZEN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07007019 /*
7020 * Fall through and re-initialise the domains.
7021 */
7022 break;
7023 default:
7024 return NOTIFY_DONE;
7025 }
7026
7027 /* The hotplug lock is already held by cpu_up/cpu_down */
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007028 arch_init_sched_domains(&cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007029
7030 return NOTIFY_OK;
7031}
Linus Torvalds1da177e2005-04-16 15:20:36 -07007032
7033void __init sched_init_smp(void)
7034{
Nick Piggin5c1e1762006-10-03 01:14:04 -07007035 cpumask_t non_isolated_cpus;
7036
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007037 get_online_cpus();
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007038 arch_init_sched_domains(&cpu_online_map);
Nathan Lynche5e56732007-01-10 23:15:28 -08007039 cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map);
Nick Piggin5c1e1762006-10-03 01:14:04 -07007040 if (cpus_empty(non_isolated_cpus))
7041 cpu_set(smp_processor_id(), non_isolated_cpus);
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007042 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007043 /* XXX: Theoretical race here - CPU may be hotplugged now */
7044 hotcpu_notifier(update_sched_domains, 0);
Nick Piggin5c1e1762006-10-03 01:14:04 -07007045
7046 /* Move init over to a non-isolated CPU */
7047 if (set_cpus_allowed(current, non_isolated_cpus) < 0)
7048 BUG();
Ingo Molnar19978ca2007-11-09 22:39:38 +01007049 sched_init_granularity();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007050}
7051#else
7052void __init sched_init_smp(void)
7053{
Ingo Molnar19978ca2007-11-09 22:39:38 +01007054 sched_init_granularity();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007055}
7056#endif /* CONFIG_SMP */
7057
7058int in_sched_functions(unsigned long addr)
7059{
Linus Torvalds1da177e2005-04-16 15:20:36 -07007060 return in_lock_functions(addr) ||
7061 (addr >= (unsigned long)__sched_text_start
7062 && addr < (unsigned long)__sched_text_end);
7063}
7064
Alexey Dobriyana9957442007-10-15 17:00:13 +02007065static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq)
Ingo Molnardd41f592007-07-09 18:51:59 +02007066{
7067 cfs_rq->tasks_timeline = RB_ROOT;
Ingo Molnardd41f592007-07-09 18:51:59 +02007068#ifdef CONFIG_FAIR_GROUP_SCHED
7069 cfs_rq->rq = rq;
7070#endif
Peter Zijlstra67e9fb22007-10-15 17:00:10 +02007071 cfs_rq->min_vruntime = (u64)(-(1LL << 20));
Ingo Molnardd41f592007-07-09 18:51:59 +02007072}
7073
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007074static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq)
7075{
7076 struct rt_prio_array *array;
7077 int i;
7078
7079 array = &rt_rq->active;
7080 for (i = 0; i < MAX_RT_PRIO; i++) {
7081 INIT_LIST_HEAD(array->queue + i);
7082 __clear_bit(i, array->bitmap);
7083 }
7084 /* delimiter for bitsearch: */
7085 __set_bit(MAX_RT_PRIO, array->bitmap);
7086
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007087#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
Peter Zijlstra48d5e252008-01-25 21:08:31 +01007088 rt_rq->highest_prio = MAX_RT_PRIO;
7089#endif
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007090#ifdef CONFIG_SMP
7091 rt_rq->rt_nr_migratory = 0;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007092 rt_rq->overloaded = 0;
7093#endif
7094
7095 rt_rq->rt_time = 0;
7096 rt_rq->rt_throttled = 0;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007097
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007098#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +01007099 rt_rq->rt_nr_boosted = 0;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007100 rt_rq->rq = rq;
7101#endif
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007102}
7103
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007104#ifdef CONFIG_FAIR_GROUP_SCHED
7105static void init_tg_cfs_entry(struct rq *rq, struct task_group *tg,
7106 struct cfs_rq *cfs_rq, struct sched_entity *se,
7107 int cpu, int add)
7108{
7109 tg->cfs_rq[cpu] = cfs_rq;
7110 init_cfs_rq(cfs_rq, rq);
7111 cfs_rq->tg = tg;
7112 if (add)
7113 list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list);
7114
7115 tg->se[cpu] = se;
7116 se->cfs_rq = &rq->cfs;
7117 se->my_q = cfs_rq;
7118 se->load.weight = tg->shares;
7119 se->load.inv_weight = div64_64(1ULL<<32, se->load.weight);
7120 se->parent = NULL;
7121}
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007122#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007123
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007124#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007125static void init_tg_rt_entry(struct rq *rq, struct task_group *tg,
7126 struct rt_rq *rt_rq, struct sched_rt_entity *rt_se,
7127 int cpu, int add)
7128{
7129 tg->rt_rq[cpu] = rt_rq;
7130 init_rt_rq(rt_rq, rq);
7131 rt_rq->tg = tg;
7132 rt_rq->rt_se = rt_se;
7133 if (add)
7134 list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list);
7135
7136 tg->rt_se[cpu] = rt_se;
7137 rt_se->rt_rq = &rq->rt;
7138 rt_se->my_q = rt_rq;
7139 rt_se->parent = NULL;
7140 INIT_LIST_HEAD(&rt_se->run_list);
7141}
7142#endif
7143
Linus Torvalds1da177e2005-04-16 15:20:36 -07007144void __init sched_init(void)
7145{
Christoph Lameter476f3532007-05-06 14:48:58 -07007146 int highest_cpu = 0;
Ingo Molnardd41f592007-07-09 18:51:59 +02007147 int i, j;
7148
Gregory Haskins57d885f2008-01-25 21:08:18 +01007149#ifdef CONFIG_SMP
7150 init_defrootdomain();
7151#endif
7152
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007153#ifdef CONFIG_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007154 list_add(&init_task_group.list, &task_groups);
7155#endif
7156
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08007157 for_each_possible_cpu(i) {
Ingo Molnar70b97a72006-07-03 00:25:42 -07007158 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007159
7160 rq = cpu_rq(i);
7161 spin_lock_init(&rq->lock);
Ingo Molnarfcb99372006-07-03 00:25:10 -07007162 lockdep_set_class(&rq->lock, &rq->rq_lock_key);
Nick Piggin78979862005-06-25 14:57:13 -07007163 rq->nr_running = 0;
Ingo Molnardd41f592007-07-09 18:51:59 +02007164 rq->clock = 1;
7165 init_cfs_rq(&rq->cfs, rq);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007166 init_rt_rq(&rq->rt, rq);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007167#ifdef CONFIG_FAIR_GROUP_SCHED
7168 init_task_group.shares = init_task_group_load;
7169 INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
7170 init_tg_cfs_entry(rq, &init_task_group,
7171 &per_cpu(init_cfs_rq, i),
7172 &per_cpu(init_sched_entity, i), i, 1);
7173
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007174#endif
7175#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007176 init_task_group.rt_runtime =
7177 sysctl_sched_rt_runtime * NSEC_PER_USEC;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007178 INIT_LIST_HEAD(&rq->leaf_rt_rq_list);
7179 init_tg_rt_entry(rq, &init_task_group,
7180 &per_cpu(init_rt_rq, i),
7181 &per_cpu(init_sched_rt_entity, i), i, 1);
7182#endif
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007183 rq->rt_period_expire = 0;
Peter Zijlstra48d5e252008-01-25 21:08:31 +01007184 rq->rt_throttled = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007185
Ingo Molnardd41f592007-07-09 18:51:59 +02007186 for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
7187 rq->cpu_load[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007188#ifdef CONFIG_SMP
Nick Piggin41c7ce92005-06-25 14:57:24 -07007189 rq->sd = NULL;
Gregory Haskins57d885f2008-01-25 21:08:18 +01007190 rq->rd = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007191 rq->active_balance = 0;
Ingo Molnardd41f592007-07-09 18:51:59 +02007192 rq->next_balance = jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007193 rq->push_cpu = 0;
Christoph Lameter0a2966b2006-09-25 23:30:51 -07007194 rq->cpu = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007195 rq->migration_thread = NULL;
7196 INIT_LIST_HEAD(&rq->migration_queue);
Gregory Haskinsdc938522008-01-25 21:08:26 +01007197 rq_attach_root(rq, &def_root_domain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007198#endif
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01007199 init_rq_hrtick(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007200 atomic_set(&rq->nr_iowait, 0);
Christoph Lameter476f3532007-05-06 14:48:58 -07007201 highest_cpu = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007202 }
7203
Peter Williams2dd73a42006-06-27 02:54:34 -07007204 set_load_weight(&init_task);
Heiko Carstensb50f60c2006-07-30 03:03:52 -07007205
Avi Kivitye107be32007-07-26 13:40:43 +02007206#ifdef CONFIG_PREEMPT_NOTIFIERS
7207 INIT_HLIST_HEAD(&init_task.preempt_notifiers);
7208#endif
7209
Christoph Lameterc9819f42006-12-10 02:20:25 -08007210#ifdef CONFIG_SMP
Christoph Lameter476f3532007-05-06 14:48:58 -07007211 nr_cpu_ids = highest_cpu + 1;
Christoph Lameterc9819f42006-12-10 02:20:25 -08007212 open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL);
7213#endif
7214
Heiko Carstensb50f60c2006-07-30 03:03:52 -07007215#ifdef CONFIG_RT_MUTEXES
7216 plist_head_init(&init_task.pi_waiters, &init_task.pi_lock);
7217#endif
7218
Linus Torvalds1da177e2005-04-16 15:20:36 -07007219 /*
7220 * The boot idle thread does lazy MMU switching as well:
7221 */
7222 atomic_inc(&init_mm.mm_count);
7223 enter_lazy_tlb(&init_mm, current);
7224
7225 /*
7226 * Make us the idle thread. Technically, schedule() should not be
7227 * called from this thread, however somewhere below it might be,
7228 * but because we are the idle thread, we just pick up running again
7229 * when this runqueue becomes "idle".
7230 */
7231 init_idle(current, smp_processor_id());
Ingo Molnardd41f592007-07-09 18:51:59 +02007232 /*
7233 * During early bootup we pretend to be a normal task:
7234 */
7235 current->sched_class = &fair_sched_class;
Ingo Molnar6892b752008-02-13 14:02:36 +01007236
7237 scheduler_running = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007238}
7239
7240#ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
7241void __might_sleep(char *file, int line)
7242{
Ingo Molnar48f24c42006-07-03 00:25:40 -07007243#ifdef in_atomic
Linus Torvalds1da177e2005-04-16 15:20:36 -07007244 static unsigned long prev_jiffy; /* ratelimiting */
7245
7246 if ((in_atomic() || irqs_disabled()) &&
7247 system_state == SYSTEM_RUNNING && !oops_in_progress) {
7248 if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
7249 return;
7250 prev_jiffy = jiffies;
Ingo Molnar91368d72006-03-23 03:00:54 -08007251 printk(KERN_ERR "BUG: sleeping function called from invalid"
Linus Torvalds1da177e2005-04-16 15:20:36 -07007252 " context at %s:%d\n", file, line);
7253 printk("in_atomic():%d, irqs_disabled():%d\n",
7254 in_atomic(), irqs_disabled());
Peter Zijlstraa4c410f2006-12-06 20:37:21 -08007255 debug_show_held_locks(current);
Ingo Molnar3117df02006-12-13 00:34:43 -08007256 if (irqs_disabled())
7257 print_irqtrace_events(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007258 dump_stack();
7259 }
7260#endif
7261}
7262EXPORT_SYMBOL(__might_sleep);
7263#endif
7264
7265#ifdef CONFIG_MAGIC_SYSRQ
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02007266static void normalize_task(struct rq *rq, struct task_struct *p)
7267{
7268 int on_rq;
7269 update_rq_clock(rq);
7270 on_rq = p->se.on_rq;
7271 if (on_rq)
7272 deactivate_task(rq, p, 0);
7273 __setscheduler(rq, p, SCHED_NORMAL, 0);
7274 if (on_rq) {
7275 activate_task(rq, p, 0);
7276 resched_task(rq->curr);
7277 }
7278}
7279
Linus Torvalds1da177e2005-04-16 15:20:36 -07007280void normalize_rt_tasks(void)
7281{
Ingo Molnara0f98a12007-06-17 18:37:45 +02007282 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007283 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07007284 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007285
Peter Zijlstra4cf5d772008-02-13 15:45:39 +01007286 read_lock_irqsave(&tasklist_lock, flags);
Ingo Molnara0f98a12007-06-17 18:37:45 +02007287 do_each_thread(g, p) {
Ingo Molnar178be792007-10-15 17:00:18 +02007288 /*
7289 * Only normalize user tasks:
7290 */
7291 if (!p->mm)
7292 continue;
7293
Ingo Molnardd41f592007-07-09 18:51:59 +02007294 p->se.exec_start = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02007295#ifdef CONFIG_SCHEDSTATS
7296 p->se.wait_start = 0;
7297 p->se.sleep_start = 0;
Ingo Molnardd41f592007-07-09 18:51:59 +02007298 p->se.block_start = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02007299#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02007300 task_rq(p)->clock = 0;
7301
7302 if (!rt_task(p)) {
7303 /*
7304 * Renice negative nice level userspace
7305 * tasks back to 0:
7306 */
7307 if (TASK_NICE(p) < 0 && p->mm)
7308 set_user_nice(p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007309 continue;
Ingo Molnardd41f592007-07-09 18:51:59 +02007310 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007311
Peter Zijlstra4cf5d772008-02-13 15:45:39 +01007312 spin_lock(&p->pi_lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -07007313 rq = __task_rq_lock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007314
Ingo Molnar178be792007-10-15 17:00:18 +02007315 normalize_task(rq, p);
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02007316
Ingo Molnarb29739f2006-06-27 02:54:51 -07007317 __task_rq_unlock(rq);
Peter Zijlstra4cf5d772008-02-13 15:45:39 +01007318 spin_unlock(&p->pi_lock);
Ingo Molnara0f98a12007-06-17 18:37:45 +02007319 } while_each_thread(g, p);
7320
Peter Zijlstra4cf5d772008-02-13 15:45:39 +01007321 read_unlock_irqrestore(&tasklist_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007322}
7323
7324#endif /* CONFIG_MAGIC_SYSRQ */
Linus Torvalds1df5c102005-09-12 07:59:21 -07007325
7326#ifdef CONFIG_IA64
7327/*
7328 * These functions are only useful for the IA64 MCA handling.
7329 *
7330 * They can only be called when the whole system has been
7331 * stopped - every CPU needs to be quiescent, and no scheduling
7332 * activity can take place. Using them for anything else would
7333 * be a serious bug, and as a result, they aren't even visible
7334 * under any other configuration.
7335 */
7336
7337/**
7338 * curr_task - return the current task for a given cpu.
7339 * @cpu: the processor in question.
7340 *
7341 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
7342 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07007343struct task_struct *curr_task(int cpu)
Linus Torvalds1df5c102005-09-12 07:59:21 -07007344{
7345 return cpu_curr(cpu);
7346}
7347
7348/**
7349 * set_curr_task - set the current task for a given cpu.
7350 * @cpu: the processor in question.
7351 * @p: the task pointer to set.
7352 *
7353 * Description: This function must only be used when non-maskable interrupts
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007354 * are serviced on a separate stack. It allows the architecture to switch the
7355 * notion of the current task on a cpu in a non-blocking manner. This function
Linus Torvalds1df5c102005-09-12 07:59:21 -07007356 * must be called with all CPU's synchronized, and interrupts disabled, the
7357 * and caller must save the original value of the current task (see
7358 * curr_task() above) and restore that value before reenabling interrupts and
7359 * re-starting the system.
7360 *
7361 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
7362 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07007363void set_curr_task(int cpu, struct task_struct *p)
Linus Torvalds1df5c102005-09-12 07:59:21 -07007364{
7365 cpu_curr(cpu) = p;
7366}
7367
7368#endif
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007369
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007370#ifdef CONFIG_GROUP_SCHED
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007371
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007372#ifdef CONFIG_FAIR_GROUP_SCHED
7373static void free_fair_sched_group(struct task_group *tg)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007374{
7375 int i;
7376
7377 for_each_possible_cpu(i) {
7378 if (tg->cfs_rq)
7379 kfree(tg->cfs_rq[i]);
7380 if (tg->se)
7381 kfree(tg->se[i]);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007382 }
7383
7384 kfree(tg->cfs_rq);
7385 kfree(tg->se);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007386}
7387
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007388static int alloc_fair_sched_group(struct task_group *tg)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007389{
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007390 struct cfs_rq *cfs_rq;
7391 struct sched_entity *se;
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007392 struct rq *rq;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007393 int i;
7394
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007395 tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007396 if (!tg->cfs_rq)
7397 goto err;
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007398 tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007399 if (!tg->se)
7400 goto err;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007401
7402 tg->shares = NICE_0_LOAD;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007403
7404 for_each_possible_cpu(i) {
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007405 rq = cpu_rq(i);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007406
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007407 cfs_rq = kmalloc_node(sizeof(struct cfs_rq),
7408 GFP_KERNEL|__GFP_ZERO, cpu_to_node(i));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007409 if (!cfs_rq)
7410 goto err;
7411
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007412 se = kmalloc_node(sizeof(struct sched_entity),
7413 GFP_KERNEL|__GFP_ZERO, cpu_to_node(i));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007414 if (!se)
7415 goto err;
7416
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007417 init_tg_cfs_entry(rq, tg, cfs_rq, se, i, 0);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007418 }
7419
7420 return 1;
7421
7422 err:
7423 return 0;
7424}
7425
7426static inline void register_fair_sched_group(struct task_group *tg, int cpu)
7427{
7428 list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list,
7429 &cpu_rq(cpu)->leaf_cfs_rq_list);
7430}
7431
7432static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
7433{
7434 list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list);
7435}
7436#else
7437static inline void free_fair_sched_group(struct task_group *tg)
7438{
7439}
7440
7441static inline int alloc_fair_sched_group(struct task_group *tg)
7442{
7443 return 1;
7444}
7445
7446static inline void register_fair_sched_group(struct task_group *tg, int cpu)
7447{
7448}
7449
7450static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
7451{
7452}
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007453#endif
7454
7455#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007456static void free_rt_sched_group(struct task_group *tg)
7457{
7458 int i;
7459
7460 for_each_possible_cpu(i) {
7461 if (tg->rt_rq)
7462 kfree(tg->rt_rq[i]);
7463 if (tg->rt_se)
7464 kfree(tg->rt_se[i]);
7465 }
7466
7467 kfree(tg->rt_rq);
7468 kfree(tg->rt_se);
7469}
7470
7471static int alloc_rt_sched_group(struct task_group *tg)
7472{
7473 struct rt_rq *rt_rq;
7474 struct sched_rt_entity *rt_se;
7475 struct rq *rq;
7476 int i;
7477
7478 tg->rt_rq = kzalloc(sizeof(rt_rq) * NR_CPUS, GFP_KERNEL);
7479 if (!tg->rt_rq)
7480 goto err;
7481 tg->rt_se = kzalloc(sizeof(rt_se) * NR_CPUS, GFP_KERNEL);
7482 if (!tg->rt_se)
7483 goto err;
7484
7485 tg->rt_runtime = 0;
7486
7487 for_each_possible_cpu(i) {
7488 rq = cpu_rq(i);
7489
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007490 rt_rq = kmalloc_node(sizeof(struct rt_rq),
7491 GFP_KERNEL|__GFP_ZERO, cpu_to_node(i));
7492 if (!rt_rq)
7493 goto err;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007494
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007495 rt_se = kmalloc_node(sizeof(struct sched_rt_entity),
7496 GFP_KERNEL|__GFP_ZERO, cpu_to_node(i));
7497 if (!rt_se)
7498 goto err;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007499
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007500 init_tg_rt_entry(rq, tg, rt_rq, rt_se, i, 0);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007501 }
7502
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007503 return 1;
7504
7505 err:
7506 return 0;
7507}
7508
7509static inline void register_rt_sched_group(struct task_group *tg, int cpu)
7510{
7511 list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list,
7512 &cpu_rq(cpu)->leaf_rt_rq_list);
7513}
7514
7515static inline void unregister_rt_sched_group(struct task_group *tg, int cpu)
7516{
7517 list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list);
7518}
7519#else
7520static inline void free_rt_sched_group(struct task_group *tg)
7521{
7522}
7523
7524static inline int alloc_rt_sched_group(struct task_group *tg)
7525{
7526 return 1;
7527}
7528
7529static inline void register_rt_sched_group(struct task_group *tg, int cpu)
7530{
7531}
7532
7533static inline void unregister_rt_sched_group(struct task_group *tg, int cpu)
7534{
7535}
7536#endif
7537
7538static void free_sched_group(struct task_group *tg)
7539{
7540 free_fair_sched_group(tg);
7541 free_rt_sched_group(tg);
7542 kfree(tg);
7543}
7544
7545/* allocate runqueue etc for a new task group */
7546struct task_group *sched_create_group(void)
7547{
7548 struct task_group *tg;
7549 unsigned long flags;
7550 int i;
7551
7552 tg = kzalloc(sizeof(*tg), GFP_KERNEL);
7553 if (!tg)
7554 return ERR_PTR(-ENOMEM);
7555
7556 if (!alloc_fair_sched_group(tg))
7557 goto err;
7558
7559 if (!alloc_rt_sched_group(tg))
7560 goto err;
7561
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007562 spin_lock_irqsave(&task_group_lock, flags);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007563 for_each_possible_cpu(i) {
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007564 register_fair_sched_group(tg, i);
7565 register_rt_sched_group(tg, i);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007566 }
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007567 list_add_rcu(&tg->list, &task_groups);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007568 spin_unlock_irqrestore(&task_group_lock, flags);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007569
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007570 return tg;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007571
7572err:
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007573 free_sched_group(tg);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007574 return ERR_PTR(-ENOMEM);
7575}
7576
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007577/* rcu callback to free various structures associated with a task group */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007578static void free_sched_group_rcu(struct rcu_head *rhp)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007579{
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007580 /* now it should be safe to free those cfs_rqs */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007581 free_sched_group(container_of(rhp, struct task_group, rcu));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007582}
7583
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007584/* Destroy runqueue etc associated with a task group */
Ingo Molnar4cf86d72007-10-15 17:00:14 +02007585void sched_destroy_group(struct task_group *tg)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007586{
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007587 unsigned long flags;
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007588 int i;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007589
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007590 spin_lock_irqsave(&task_group_lock, flags);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007591 for_each_possible_cpu(i) {
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007592 unregister_fair_sched_group(tg, i);
7593 unregister_rt_sched_group(tg, i);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007594 }
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007595 list_del_rcu(&tg->list);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007596 spin_unlock_irqrestore(&task_group_lock, flags);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007597
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007598 /* wait for possible concurrent references to cfs_rqs complete */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007599 call_rcu(&tg->rcu, free_sched_group_rcu);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007600}
7601
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007602/* change task's runqueue when it moves between groups.
Ingo Molnar3a252012007-10-15 17:00:12 +02007603 * The caller of this function should have put the task in its new group
7604 * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to
7605 * reflect its new group.
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007606 */
7607void sched_move_task(struct task_struct *tsk)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007608{
7609 int on_rq, running;
7610 unsigned long flags;
7611 struct rq *rq;
7612
7613 rq = task_rq_lock(tsk, &flags);
7614
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007615 update_rq_clock(rq);
7616
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01007617 running = task_current(rq, tsk);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007618 on_rq = tsk->se.on_rq;
7619
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02007620 if (on_rq) {
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007621 dequeue_task(rq, tsk, 0);
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02007622 if (unlikely(running))
7623 tsk->sched_class->put_prev_task(rq, tsk);
7624 }
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007625
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007626 set_task_rq(tsk, task_cpu(tsk));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007627
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02007628 if (on_rq) {
7629 if (unlikely(running))
7630 tsk->sched_class->set_curr_task(rq);
Dmitry Adamushko7074bad2007-10-15 17:00:07 +02007631 enqueue_task(rq, tsk, 0);
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02007632 }
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007633
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007634 task_rq_unlock(rq, &flags);
7635}
7636
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007637#ifdef CONFIG_FAIR_GROUP_SCHED
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007638static void set_se_shares(struct sched_entity *se, unsigned long shares)
7639{
7640 struct cfs_rq *cfs_rq = se->cfs_rq;
7641 struct rq *rq = cfs_rq->rq;
7642 int on_rq;
7643
Peter Zijlstra62fb1852008-02-25 17:34:02 +01007644 spin_lock_irq(&rq->lock);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007645
7646 on_rq = se->on_rq;
Peter Zijlstra62fb1852008-02-25 17:34:02 +01007647 if (on_rq)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007648 dequeue_entity(cfs_rq, se, 0);
7649
7650 se->load.weight = shares;
7651 se->load.inv_weight = div64_64((1ULL<<32), shares);
7652
Peter Zijlstra62fb1852008-02-25 17:34:02 +01007653 if (on_rq)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007654 enqueue_entity(cfs_rq, se, 0);
Peter Zijlstra62fb1852008-02-25 17:34:02 +01007655
7656 spin_unlock_irq(&rq->lock);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007657}
7658
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007659static DEFINE_MUTEX(shares_mutex);
7660
Ingo Molnar4cf86d72007-10-15 17:00:14 +02007661int sched_group_set_shares(struct task_group *tg, unsigned long shares)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007662{
7663 int i;
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007664 unsigned long flags;
Ingo Molnarc61935f2008-01-22 11:24:58 +01007665
Peter Zijlstra62fb1852008-02-25 17:34:02 +01007666 /*
7667 * A weight of 0 or 1 can cause arithmetics problems.
7668 * (The default weight is 1024 - so there's no practical
7669 * limitation from this.)
7670 */
7671 if (shares < 2)
7672 shares = 2;
7673
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007674 mutex_lock(&shares_mutex);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007675 if (tg->shares == shares)
Dhaval Giani5cb350b2007-10-15 17:00:14 +02007676 goto done;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007677
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007678 spin_lock_irqsave(&task_group_lock, flags);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007679 for_each_possible_cpu(i)
7680 unregister_fair_sched_group(tg, i);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007681 spin_unlock_irqrestore(&task_group_lock, flags);
Srivatsa Vaddagiri6b2d7702008-01-25 21:08:00 +01007682
7683 /* wait for any ongoing reference to this group to finish */
7684 synchronize_sched();
7685
7686 /*
7687 * Now we are free to modify the group's share on each cpu
7688 * w/o tripping rebalance_share or load_balance_fair.
7689 */
7690 tg->shares = shares;
Peter Zijlstra62fb1852008-02-25 17:34:02 +01007691 for_each_possible_cpu(i)
Srivatsa Vaddagiri6b2d7702008-01-25 21:08:00 +01007692 set_se_shares(tg->se[i], shares);
Srivatsa Vaddagiri6b2d7702008-01-25 21:08:00 +01007693
7694 /*
7695 * Enable load balance activity on this group, by inserting it back on
7696 * each cpu's rq->leaf_cfs_rq_list.
7697 */
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007698 spin_lock_irqsave(&task_group_lock, flags);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007699 for_each_possible_cpu(i)
7700 register_fair_sched_group(tg, i);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007701 spin_unlock_irqrestore(&task_group_lock, flags);
Dhaval Giani5cb350b2007-10-15 17:00:14 +02007702done:
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007703 mutex_unlock(&shares_mutex);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007704 return 0;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007705}
7706
Dhaval Giani5cb350b2007-10-15 17:00:14 +02007707unsigned long sched_group_shares(struct task_group *tg)
7708{
7709 return tg->shares;
7710}
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007711#endif
Dhaval Giani5cb350b2007-10-15 17:00:14 +02007712
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007713#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007714/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007715 * Ensure that the real time constraints are schedulable.
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007716 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007717static DEFINE_MUTEX(rt_constraints_mutex);
7718
7719static unsigned long to_ratio(u64 period, u64 runtime)
7720{
7721 if (runtime == RUNTIME_INF)
7722 return 1ULL << 16;
7723
7724 runtime *= (1ULL << 16);
7725 div64_64(runtime, period);
7726 return runtime;
7727}
7728
7729static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007730{
7731 struct task_group *tgi;
7732 unsigned long total = 0;
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007733 unsigned long global_ratio =
7734 to_ratio(sysctl_sched_rt_period,
7735 sysctl_sched_rt_runtime < 0 ?
7736 RUNTIME_INF : sysctl_sched_rt_runtime);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007737
7738 rcu_read_lock();
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007739 list_for_each_entry_rcu(tgi, &task_groups, list) {
7740 if (tgi == tg)
7741 continue;
7742
7743 total += to_ratio(period, tgi->rt_runtime);
7744 }
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007745 rcu_read_unlock();
7746
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007747 return total + to_ratio(period, runtime) < global_ratio;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007748}
7749
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007750int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007751{
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007752 u64 rt_runtime, rt_period;
7753 int err = 0;
7754
7755 rt_period = sysctl_sched_rt_period * NSEC_PER_USEC;
7756 rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
7757 if (rt_runtime_us == -1)
7758 rt_runtime = rt_period;
7759
7760 mutex_lock(&rt_constraints_mutex);
7761 if (!__rt_schedulable(tg, rt_period, rt_runtime)) {
7762 err = -EINVAL;
7763 goto unlock;
7764 }
7765 if (rt_runtime_us == -1)
7766 rt_runtime = RUNTIME_INF;
7767 tg->rt_runtime = rt_runtime;
7768 unlock:
7769 mutex_unlock(&rt_constraints_mutex);
7770
7771 return err;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007772}
7773
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007774long sched_group_rt_runtime(struct task_group *tg)
7775{
7776 u64 rt_runtime_us;
7777
7778 if (tg->rt_runtime == RUNTIME_INF)
7779 return -1;
7780
7781 rt_runtime_us = tg->rt_runtime;
7782 do_div(rt_runtime_us, NSEC_PER_USEC);
7783 return rt_runtime_us;
7784}
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007785#endif
7786#endif /* CONFIG_GROUP_SCHED */
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007787
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007788#ifdef CONFIG_CGROUP_SCHED
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007789
7790/* return corresponding task_group object of a cgroup */
Paul Menage2b01dfe2007-10-24 18:23:50 +02007791static inline struct task_group *cgroup_tg(struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007792{
Paul Menage2b01dfe2007-10-24 18:23:50 +02007793 return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id),
7794 struct task_group, css);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007795}
7796
7797static struct cgroup_subsys_state *
Paul Menage2b01dfe2007-10-24 18:23:50 +02007798cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007799{
7800 struct task_group *tg;
7801
Paul Menage2b01dfe2007-10-24 18:23:50 +02007802 if (!cgrp->parent) {
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007803 /* This is early initialization for the top cgroup */
Paul Menage2b01dfe2007-10-24 18:23:50 +02007804 init_task_group.css.cgroup = cgrp;
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007805 return &init_task_group.css;
7806 }
7807
7808 /* we support only 1-level deep hierarchical scheduler atm */
Paul Menage2b01dfe2007-10-24 18:23:50 +02007809 if (cgrp->parent->parent)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007810 return ERR_PTR(-EINVAL);
7811
7812 tg = sched_create_group();
7813 if (IS_ERR(tg))
7814 return ERR_PTR(-ENOMEM);
7815
7816 /* Bind the cgroup to task_group object we just created */
Paul Menage2b01dfe2007-10-24 18:23:50 +02007817 tg->css.cgroup = cgrp;
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007818
7819 return &tg->css;
7820}
7821
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007822static void
7823cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007824{
Paul Menage2b01dfe2007-10-24 18:23:50 +02007825 struct task_group *tg = cgroup_tg(cgrp);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007826
7827 sched_destroy_group(tg);
7828}
7829
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007830static int
7831cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
7832 struct task_struct *tsk)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007833{
Peter Zijlstrab68aa232008-02-13 15:45:40 +01007834#ifdef CONFIG_RT_GROUP_SCHED
7835 /* Don't accept realtime tasks when there is no way for them to run */
7836 if (rt_task(tsk) && cgroup_tg(cgrp)->rt_runtime == 0)
7837 return -EINVAL;
7838#else
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007839 /* We don't support RT-tasks being in separate groups */
7840 if (tsk->sched_class != &fair_sched_class)
7841 return -EINVAL;
Peter Zijlstrab68aa232008-02-13 15:45:40 +01007842#endif
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007843
7844 return 0;
7845}
7846
7847static void
Paul Menage2b01dfe2007-10-24 18:23:50 +02007848cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007849 struct cgroup *old_cont, struct task_struct *tsk)
7850{
7851 sched_move_task(tsk);
7852}
7853
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007854#ifdef CONFIG_FAIR_GROUP_SCHED
Paul Menage2b01dfe2007-10-24 18:23:50 +02007855static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype,
7856 u64 shareval)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007857{
Paul Menage2b01dfe2007-10-24 18:23:50 +02007858 return sched_group_set_shares(cgroup_tg(cgrp), shareval);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007859}
7860
Paul Menage2b01dfe2007-10-24 18:23:50 +02007861static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007862{
Paul Menage2b01dfe2007-10-24 18:23:50 +02007863 struct task_group *tg = cgroup_tg(cgrp);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007864
7865 return (u64) tg->shares;
7866}
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007867#endif
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007868
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007869#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007870static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft,
7871 struct file *file,
7872 const char __user *userbuf,
7873 size_t nbytes, loff_t *unused_ppos)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007874{
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007875 char buffer[64];
7876 int retval = 0;
7877 s64 val;
7878 char *end;
7879
7880 if (!nbytes)
7881 return -EINVAL;
7882 if (nbytes >= sizeof(buffer))
7883 return -E2BIG;
7884 if (copy_from_user(buffer, userbuf, nbytes))
7885 return -EFAULT;
7886
7887 buffer[nbytes] = 0; /* nul-terminate */
7888
7889 /* strip newline if necessary */
7890 if (nbytes && (buffer[nbytes-1] == '\n'))
7891 buffer[nbytes-1] = 0;
7892 val = simple_strtoll(buffer, &end, 0);
7893 if (*end)
7894 return -EINVAL;
7895
7896 /* Pass to subsystem */
7897 retval = sched_group_set_rt_runtime(cgroup_tg(cgrp), val);
7898 if (!retval)
7899 retval = nbytes;
7900 return retval;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007901}
7902
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007903static ssize_t cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft,
7904 struct file *file,
7905 char __user *buf, size_t nbytes,
7906 loff_t *ppos)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007907{
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007908 char tmp[64];
7909 long val = sched_group_rt_runtime(cgroup_tg(cgrp));
7910 int len = sprintf(tmp, "%ld\n", val);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007911
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007912 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007913}
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007914#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007915
Paul Menagefe5c7cc2007-10-29 21:18:11 +01007916static struct cftype cpu_files[] = {
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007917#ifdef CONFIG_FAIR_GROUP_SCHED
Paul Menagefe5c7cc2007-10-29 21:18:11 +01007918 {
7919 .name = "shares",
7920 .read_uint = cpu_shares_read_uint,
7921 .write_uint = cpu_shares_write_uint,
7922 },
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007923#endif
7924#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007925 {
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007926 .name = "rt_runtime_us",
7927 .read = cpu_rt_runtime_read,
7928 .write = cpu_rt_runtime_write,
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007929 },
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007930#endif
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007931};
7932
7933static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont)
7934{
Paul Menagefe5c7cc2007-10-29 21:18:11 +01007935 return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files));
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007936}
7937
7938struct cgroup_subsys cpu_cgroup_subsys = {
Ingo Molnar38605ca2007-10-29 21:18:11 +01007939 .name = "cpu",
7940 .create = cpu_cgroup_create,
7941 .destroy = cpu_cgroup_destroy,
7942 .can_attach = cpu_cgroup_can_attach,
7943 .attach = cpu_cgroup_attach,
7944 .populate = cpu_cgroup_populate,
7945 .subsys_id = cpu_cgroup_subsys_id,
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007946 .early_init = 1,
7947};
7948
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007949#endif /* CONFIG_CGROUP_SCHED */
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007950
7951#ifdef CONFIG_CGROUP_CPUACCT
7952
7953/*
7954 * CPU accounting code for task groups.
7955 *
7956 * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
7957 * (balbir@in.ibm.com).
7958 */
7959
7960/* track cpu usage of a group of tasks */
7961struct cpuacct {
7962 struct cgroup_subsys_state css;
7963 /* cpuusage holds pointer to a u64-type object on every cpu */
7964 u64 *cpuusage;
7965};
7966
7967struct cgroup_subsys cpuacct_subsys;
7968
7969/* return cpu accounting group corresponding to this container */
7970static inline struct cpuacct *cgroup_ca(struct cgroup *cont)
7971{
7972 return container_of(cgroup_subsys_state(cont, cpuacct_subsys_id),
7973 struct cpuacct, css);
7974}
7975
7976/* return cpu accounting group to which this task belongs */
7977static inline struct cpuacct *task_ca(struct task_struct *tsk)
7978{
7979 return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
7980 struct cpuacct, css);
7981}
7982
7983/* create a new cpu accounting group */
7984static struct cgroup_subsys_state *cpuacct_create(
7985 struct cgroup_subsys *ss, struct cgroup *cont)
7986{
7987 struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
7988
7989 if (!ca)
7990 return ERR_PTR(-ENOMEM);
7991
7992 ca->cpuusage = alloc_percpu(u64);
7993 if (!ca->cpuusage) {
7994 kfree(ca);
7995 return ERR_PTR(-ENOMEM);
7996 }
7997
7998 return &ca->css;
7999}
8000
8001/* destroy an existing cpu accounting group */
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01008002static void
8003cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008004{
8005 struct cpuacct *ca = cgroup_ca(cont);
8006
8007 free_percpu(ca->cpuusage);
8008 kfree(ca);
8009}
8010
8011/* return total cpu usage (in nanoseconds) of a group */
8012static u64 cpuusage_read(struct cgroup *cont, struct cftype *cft)
8013{
8014 struct cpuacct *ca = cgroup_ca(cont);
8015 u64 totalcpuusage = 0;
8016 int i;
8017
8018 for_each_possible_cpu(i) {
8019 u64 *cpuusage = percpu_ptr(ca->cpuusage, i);
8020
8021 /*
8022 * Take rq->lock to make 64-bit addition safe on 32-bit
8023 * platforms.
8024 */
8025 spin_lock_irq(&cpu_rq(i)->lock);
8026 totalcpuusage += *cpuusage;
8027 spin_unlock_irq(&cpu_rq(i)->lock);
8028 }
8029
8030 return totalcpuusage;
8031}
8032
8033static struct cftype files[] = {
8034 {
8035 .name = "usage",
8036 .read_uint = cpuusage_read,
8037 },
8038};
8039
8040static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cont)
8041{
8042 return cgroup_add_files(cont, ss, files, ARRAY_SIZE(files));
8043}
8044
8045/*
8046 * charge this task's execution time to its accounting group.
8047 *
8048 * called with rq->lock held.
8049 */
8050static void cpuacct_charge(struct task_struct *tsk, u64 cputime)
8051{
8052 struct cpuacct *ca;
8053
8054 if (!cpuacct_subsys.active)
8055 return;
8056
8057 ca = task_ca(tsk);
8058 if (ca) {
8059 u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk));
8060
8061 *cpuusage += cputime;
8062 }
8063}
8064
8065struct cgroup_subsys cpuacct_subsys = {
8066 .name = "cpuacct",
8067 .create = cpuacct_create,
8068 .destroy = cpuacct_destroy,
8069 .populate = cpuacct_populate,
8070 .subsys_id = cpuacct_subsys_id,
8071};
8072#endif /* CONFIG_CGROUP_CPUACCT */