blob: 3925a1bbf5ddfac43271aec828bb260a1892111b [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 Molnar9a11b492006-07-03 00:24:33 -070041#include <linux/debug_locks.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020042#include <linux/perf_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/security.h>
44#include <linux/notifier.h>
45#include <linux/profile.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080046#include <linux/freezer.h>
akpm@osdl.org198e2f12006-01-12 01:05:30 -080047#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/blkdev.h>
49#include <linux/delay.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070050#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/smp.h>
52#include <linux/threads.h>
53#include <linux/timer.h>
54#include <linux/rcupdate.h>
55#include <linux/cpu.h>
56#include <linux/cpuset.h>
57#include <linux/percpu.h>
Alexey Dobriyanb5aadf72008-10-06 13:23:43 +040058#include <linux/proc_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#include <linux/seq_file.h>
Tejun Heo969c7922010-05-06 18:49:21 +020060#include <linux/stop_machine.h>
Nick Piggine692ab52007-07-26 13:40:43 +020061#include <linux/sysctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070062#include <linux/syscalls.h>
63#include <linux/times.h>
Jay Lan8f0ab512006-09-30 23:28:59 -070064#include <linux/tsacct_kern.h>
bibo maoc6fd91f2006-03-26 01:38:20 -080065#include <linux/kprobes.h>
Shailabh Nagar0ff92242006-07-14 00:24:37 -070066#include <linux/delayacct.h>
Ingo Molnardff06c12007-07-09 18:52:00 +020067#include <linux/unistd.h>
Jens Axboef5ff8422007-09-21 09:19:54 +020068#include <linux/pagemap.h>
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +010069#include <linux/hrtimer.h>
Reynes Philippe30914a52008-03-17 16:19:05 -070070#include <linux/tick.h>
Peter Zijlstraf00b45c2008-04-19 19:45:00 +020071#include <linux/debugfs.h>
72#include <linux/ctype.h>
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +020073#include <linux/ftrace.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090074#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
Eric Dumazet5517d862007-05-08 00:32:57 -070076#include <asm/tlb.h>
Satyam Sharma838225b2007-10-24 18:23:50 +020077#include <asm/irq_regs.h>
Gerald Schaefer335d7af2010-11-22 15:47:36 +010078#include <asm/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
Gregory Haskins6e0534f2008-05-12 21:21:01 +020080#include "sched_cpupri.h"
Tejun Heo21aa9af2010-06-08 21:40:37 +020081#include "workqueue_sched.h"
Mike Galbraith5091faa2010-11-30 14:18:03 +010082#include "sched_autogroup.h"
Gregory Haskins6e0534f2008-05-12 21:21:01 +020083
Steven Rostedta8d154b2009-04-10 09:36:00 -040084#define CREATE_TRACE_POINTS
Steven Rostedtad8d75f2009-04-14 19:39:12 -040085#include <trace/events/sched.h>
Steven Rostedta8d154b2009-04-10 09:36:00 -040086
Linus Torvalds1da177e2005-04-16 15:20:36 -070087/*
88 * Convert user-nice values [ -20 ... 0 ... 19 ]
89 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
90 * and back.
91 */
92#define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
93#define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
94#define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
95
96/*
97 * 'User priority' is the nice value converted to something we
98 * can work with better when scaling various scheduler parameters,
99 * it's a [ 0 ... 39 ] range.
100 */
101#define USER_PRIO(p) ((p)-MAX_RT_PRIO)
102#define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
103#define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
104
105/*
Ingo Molnard7876a02008-01-25 21:08:19 +0100106 * Helpers for converting nanosecond timing to jiffy resolution
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 */
Eric Dumazetd6322fa2007-11-09 22:39:38 +0100108#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200110#define NICE_0_LOAD SCHED_LOAD_SCALE
111#define NICE_0_SHIFT SCHED_LOAD_SHIFT
112
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113/*
114 * These are the 'tuning knobs' of the scheduler:
115 *
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +0200116 * default timeslice is 100 msecs (used only for SCHED_RR tasks).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 * Timeslices get refilled after they expire.
118 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119#define DEF_TIMESLICE (100 * HZ / 1000)
Peter Williams2dd73a42006-06-27 02:54:34 -0700120
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200121/*
122 * single value that denotes runtime == period, ie unlimited time.
123 */
124#define RUNTIME_INF ((u64)~0ULL)
125
Ingo Molnare05606d2007-07-09 18:51:59 +0200126static inline int rt_policy(int policy)
127{
Roel Kluin3f33a7c2008-05-13 23:44:11 +0200128 if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR))
Ingo Molnare05606d2007-07-09 18:51:59 +0200129 return 1;
130 return 0;
131}
132
133static inline int task_has_rt_policy(struct task_struct *p)
134{
135 return rt_policy(p->policy);
136}
137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138/*
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200139 * This is the priority-queue data structure of the RT scheduling class:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 */
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200141struct rt_prio_array {
142 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
143 struct list_head queue[MAX_RT_PRIO];
144};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200146struct rt_bandwidth {
Ingo Molnarea736ed2008-03-25 13:51:45 +0100147 /* nests inside the rq lock: */
Thomas Gleixner0986b112009-11-17 15:32:06 +0100148 raw_spinlock_t rt_runtime_lock;
Ingo Molnarea736ed2008-03-25 13:51:45 +0100149 ktime_t rt_period;
150 u64 rt_runtime;
151 struct hrtimer rt_period_timer;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200152};
153
154static struct rt_bandwidth def_rt_bandwidth;
155
156static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
157
158static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
159{
160 struct rt_bandwidth *rt_b =
161 container_of(timer, struct rt_bandwidth, rt_period_timer);
162 ktime_t now;
163 int overrun;
164 int idle = 0;
165
166 for (;;) {
167 now = hrtimer_cb_get_time(timer);
168 overrun = hrtimer_forward(timer, now, rt_b->rt_period);
169
170 if (!overrun)
171 break;
172
173 idle = do_sched_rt_period_timer(rt_b, overrun);
174 }
175
176 return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
177}
178
179static
180void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
181{
182 rt_b->rt_period = ns_to_ktime(period);
183 rt_b->rt_runtime = runtime;
184
Thomas Gleixner0986b112009-11-17 15:32:06 +0100185 raw_spin_lock_init(&rt_b->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200186
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200187 hrtimer_init(&rt_b->rt_period_timer,
188 CLOCK_MONOTONIC, HRTIMER_MODE_REL);
189 rt_b->rt_period_timer.function = sched_rt_period_timer;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200190}
191
Krzysztof Heltc8bfff62008-09-05 23:46:19 +0200192static inline int rt_bandwidth_enabled(void)
193{
194 return sysctl_sched_rt_runtime >= 0;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200195}
196
197static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
198{
199 ktime_t now;
200
Hiroshi Shimamotocac64d02009-02-25 09:59:26 -0800201 if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200202 return;
203
204 if (hrtimer_active(&rt_b->rt_period_timer))
205 return;
206
Thomas Gleixner0986b112009-11-17 15:32:06 +0100207 raw_spin_lock(&rt_b->rt_runtime_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200208 for (;;) {
Peter Zijlstra7f1e2ca2009-03-13 12:21:27 +0100209 unsigned long delta;
210 ktime_t soft, hard;
211
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200212 if (hrtimer_active(&rt_b->rt_period_timer))
213 break;
214
215 now = hrtimer_cb_get_time(&rt_b->rt_period_timer);
216 hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period);
Peter Zijlstra7f1e2ca2009-03-13 12:21:27 +0100217
218 soft = hrtimer_get_softexpires(&rt_b->rt_period_timer);
219 hard = hrtimer_get_expires(&rt_b->rt_period_timer);
220 delta = ktime_to_ns(ktime_sub(hard, soft));
221 __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta,
Arun R Bharadwaj5c333862009-04-16 12:14:37 +0530222 HRTIMER_MODE_ABS_PINNED, 0);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200223 }
Thomas Gleixner0986b112009-11-17 15:32:06 +0100224 raw_spin_unlock(&rt_b->rt_runtime_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200225}
226
227#ifdef CONFIG_RT_GROUP_SCHED
228static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
229{
230 hrtimer_cancel(&rt_b->rt_period_timer);
231}
232#endif
233
Heiko Carstens712555e2008-04-28 11:33:07 +0200234/*
235 * sched_domains_mutex serializes calls to arch_init_sched_domains,
236 * detach_destroy_domains and partition_sched_domains.
237 */
238static DEFINE_MUTEX(sched_domains_mutex);
239
Dhaval Giani7c941432010-01-20 13:26:18 +0100240#ifdef CONFIG_CGROUP_SCHED
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200241
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -0700242#include <linux/cgroup.h>
243
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200244struct cfs_rq;
245
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100246static LIST_HEAD(task_groups);
247
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200248/* task group related information */
Ingo Molnar4cf86d72007-10-15 17:00:14 +0200249struct task_group {
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -0700250 struct cgroup_subsys_state css;
Arun R Bharadwaj6c415b92008-12-01 20:49:05 +0530251
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100252#ifdef CONFIG_FAIR_GROUP_SCHED
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200253 /* schedulable entities of this group on each cpu */
254 struct sched_entity **se;
255 /* runqueue "owned" by this group on each cpu */
256 struct cfs_rq **cfs_rq;
257 unsigned long shares;
Peter Zijlstra2069dd72010-11-15 15:47:00 -0800258
259 atomic_t load_weight;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100260#endif
261
262#ifdef CONFIG_RT_GROUP_SCHED
263 struct sched_rt_entity **rt_se;
264 struct rt_rq **rt_rq;
265
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200266 struct rt_bandwidth rt_bandwidth;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100267#endif
Srivatsa Vaddagiri6b2d7702008-01-25 21:08:00 +0100268
Srivatsa Vaddagiriae8393e2007-10-29 21:18:11 +0100269 struct rcu_head rcu;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100270 struct list_head list;
Peter Zijlstraf473aa52008-04-19 19:45:00 +0200271
272 struct task_group *parent;
273 struct list_head siblings;
274 struct list_head children;
Mike Galbraith5091faa2010-11-30 14:18:03 +0100275
276#ifdef CONFIG_SCHED_AUTOGROUP
277 struct autogroup *autogroup;
278#endif
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200279};
280
Peter Zijlstraeff766a2008-04-19 19:45:00 +0200281#define root_task_group init_task_group
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100282
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -0800283/* task_group_lock serializes the addition/removal of task groups */
Peter Zijlstra8ed36992008-02-13 15:45:39 +0100284static DEFINE_SPINLOCK(task_group_lock);
Srivatsa Vaddagiriec2c5072008-01-25 21:07:59 +0100285
Cyrill Gorcunove9036b32009-10-26 22:24:14 +0300286#ifdef CONFIG_FAIR_GROUP_SCHED
287
Srivatsa Vaddagiri93f992c2008-01-25 21:07:59 +0100288# define INIT_TASK_GROUP_LOAD NICE_0_LOAD
Srivatsa Vaddagiri24e377a2007-10-15 17:00:09 +0200289
Miao Xiecb4ad1f2008-04-28 12:54:56 +0800290/*
Lai Jiangshan2e084782008-06-12 16:42:58 +0800291 * A weight of 0 or 1 can cause arithmetics problems.
292 * A weight of a cfs_rq is the sum of weights of which entities
293 * are queued on this cfs_rq, so a weight of a entity should not be
294 * too large, so as the shares value of a task group.
Miao Xiecb4ad1f2008-04-28 12:54:56 +0800295 * (The default weight is 1024 - so there's no practical
296 * limitation from this.)
297 */
Peter Zijlstra18d95a22008-04-19 19:45:00 +0200298#define MIN_SHARES 2
Lai Jiangshan2e084782008-06-12 16:42:58 +0800299#define MAX_SHARES (1UL << 18)
Peter Zijlstra18d95a22008-04-19 19:45:00 +0200300
Srivatsa Vaddagiri93f992c2008-01-25 21:07:59 +0100301static int init_task_group_load = INIT_TASK_GROUP_LOAD;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100302#endif
303
304/* Default task group.
305 * Every task in system belong to this group at bootup.
306 */
Mike Travis434d53b2008-04-04 18:11:04 -0700307struct task_group init_task_group;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200308
Dhaval Giani7c941432010-01-20 13:26:18 +0100309#endif /* CONFIG_CGROUP_SCHED */
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200310
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200311/* CFS-related fields in a runqueue */
312struct cfs_rq {
313 struct load_weight load;
314 unsigned long nr_running;
315
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200316 u64 exec_clock;
Ingo Molnare9acbff2007-10-15 17:00:04 +0200317 u64 min_vruntime;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200318
319 struct rb_root tasks_timeline;
320 struct rb_node *rb_leftmost;
Peter Zijlstra4a55bd52008-04-19 19:45:00 +0200321
322 struct list_head tasks;
323 struct list_head *balance_iterator;
324
325 /*
326 * 'curr' points to currently running entity on this cfs_rq.
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200327 * It is set to NULL otherwise (i.e when none are currently running).
328 */
Peter Zijlstra47932412008-11-04 21:25:09 +0100329 struct sched_entity *curr, *next, *last;
Peter Zijlstraddc97292007-10-15 17:00:10 +0200330
Peter Zijlstra5ac5c4d2008-11-10 10:46:32 +0100331 unsigned int nr_spread_over;
Peter Zijlstraddc97292007-10-15 17:00:10 +0200332
Ingo Molnar62160e32007-10-15 17:00:03 +0200333#ifdef CONFIG_FAIR_GROUP_SCHED
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200334 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
335
Ingo Molnar41a2d6c2007-12-05 15:46:09 +0100336 /*
337 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200338 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
339 * (like users, containers etc.)
340 *
341 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
342 * list is used during load balance.
343 */
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -0800344 int on_list;
Ingo Molnar41a2d6c2007-12-05 15:46:09 +0100345 struct list_head leaf_cfs_rq_list;
346 struct task_group *tg; /* group that "owns" this runqueue */
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200347
348#ifdef CONFIG_SMP
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200349 /*
Peter Zijlstrac8cba852008-06-27 13:41:23 +0200350 * the part of load.weight contributed by tasks
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200351 */
Peter Zijlstrac8cba852008-06-27 13:41:23 +0200352 unsigned long task_weight;
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200353
Peter Zijlstrac8cba852008-06-27 13:41:23 +0200354 /*
355 * h_load = weight * f(tg)
356 *
357 * Where f(tg) is the recursive weight fraction assigned to
358 * this group.
359 */
360 unsigned long h_load;
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200361
Paul Turner3b3d1902010-11-15 15:47:08 -0800362 /*
363 * Maintaining per-cpu shares distribution for group scheduling
364 *
365 * load_stamp is the last time we updated the load average
366 * load_last is the last time we updated the load average and saw load
367 * load_unacc_exec_time is currently unaccounted execution time
368 */
Peter Zijlstra2069dd72010-11-15 15:47:00 -0800369 u64 load_avg;
370 u64 load_period;
Paul Turner3b3d1902010-11-15 15:47:08 -0800371 u64 load_stamp, load_last, load_unacc_exec_time;
Peter Zijlstraf1d239f2008-06-27 13:41:38 +0200372
Peter Zijlstra2069dd72010-11-15 15:47:00 -0800373 unsigned long load_contribution;
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200374#endif
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200375#endif
376};
377
378/* Real-Time classes' related field in a runqueue: */
379struct rt_rq {
380 struct rt_prio_array active;
Steven Rostedt63489e42008-01-25 21:08:03 +0100381 unsigned long rt_nr_running;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100382#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
Gregory Haskinse864c492008-12-29 09:39:49 -0500383 struct {
384 int curr; /* highest queued rt task prio */
Gregory Haskins398a1532009-01-14 09:10:04 -0500385#ifdef CONFIG_SMP
Gregory Haskinse864c492008-12-29 09:39:49 -0500386 int next; /* next highest */
Gregory Haskins398a1532009-01-14 09:10:04 -0500387#endif
Gregory Haskinse864c492008-12-29 09:39:49 -0500388 } highest_prio;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100389#endif
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100390#ifdef CONFIG_SMP
Gregory Haskins73fe6aa2008-01-25 21:08:07 +0100391 unsigned long rt_nr_migratory;
Peter Zijlstraa1ba4d82009-04-01 18:40:15 +0200392 unsigned long rt_nr_total;
Gregory Haskinsa22d7fc2008-01-25 21:08:12 +0100393 int overloaded;
Gregory Haskins917b6272008-12-29 09:39:53 -0500394 struct plist_head pushable_tasks;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100395#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100396 int rt_throttled;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100397 u64 rt_time;
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200398 u64 rt_runtime;
Ingo Molnarea736ed2008-03-25 13:51:45 +0100399 /* Nests inside the rq lock: */
Thomas Gleixner0986b112009-11-17 15:32:06 +0100400 raw_spinlock_t rt_runtime_lock;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100401
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100402#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +0100403 unsigned long rt_nr_boosted;
404
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100405 struct rq *rq;
406 struct list_head leaf_rt_rq_list;
407 struct task_group *tg;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100408#endif
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200409};
410
Gregory Haskins57d885f2008-01-25 21:08:18 +0100411#ifdef CONFIG_SMP
412
413/*
414 * We add the notion of a root-domain which will be used to define per-domain
Ingo Molnar0eab9142008-01-25 21:08:19 +0100415 * variables. Each exclusive cpuset essentially defines an island domain by
416 * fully partitioning the member cpus from any other cpuset. Whenever a new
Gregory Haskins57d885f2008-01-25 21:08:18 +0100417 * exclusive cpuset is created, we also create and attach a new root-domain
418 * object.
419 *
Gregory Haskins57d885f2008-01-25 21:08:18 +0100420 */
421struct root_domain {
422 atomic_t refcount;
Rusty Russellc6c49272008-11-25 02:35:05 +1030423 cpumask_var_t span;
424 cpumask_var_t online;
Gregory Haskins637f5082008-01-25 21:08:18 +0100425
Ingo Molnar0eab9142008-01-25 21:08:19 +0100426 /*
Gregory Haskins637f5082008-01-25 21:08:18 +0100427 * The "RT overload" flag: it gets set if a CPU has more than
428 * one runnable RT task.
429 */
Rusty Russellc6c49272008-11-25 02:35:05 +1030430 cpumask_var_t rto_mask;
Ingo Molnar0eab9142008-01-25 21:08:19 +0100431 atomic_t rto_count;
Gregory Haskins6e0534f2008-05-12 21:21:01 +0200432 struct cpupri cpupri;
Gregory Haskins57d885f2008-01-25 21:08:18 +0100433};
434
Gregory Haskinsdc938522008-01-25 21:08:26 +0100435/*
436 * By default the system creates a single root-domain with all cpus as
437 * members (mimicking the global state we have today).
438 */
Gregory Haskins57d885f2008-01-25 21:08:18 +0100439static struct root_domain def_root_domain;
440
Christian Dietriched2d3722010-09-06 16:37:05 +0200441#endif /* CONFIG_SMP */
Gregory Haskins57d885f2008-01-25 21:08:18 +0100442
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200443/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 * This is the main, per-CPU runqueue data structure.
445 *
446 * Locking rule: those places that want to lock multiple runqueues
447 * (such as the load balancing or the thread migration code), lock
448 * acquire operations must be ordered by ascending &runqueue.
449 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700450struct rq {
Ingo Molnard8016492007-10-18 21:32:55 +0200451 /* runqueue lock: */
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100452 raw_spinlock_t lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
454 /*
455 * nr_running and cpu_load should be in the same cacheline because
456 * remote CPUs use both these fields when doing load calculation.
457 */
458 unsigned long nr_running;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200459 #define CPU_LOAD_IDX_MAX 5
460 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -0700461 unsigned long last_load_update_tick;
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -0700462#ifdef CONFIG_NO_HZ
Mike Galbraith39c0cbe2010-03-11 17:17:13 +0100463 u64 nohz_stamp;
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -0700464 unsigned char nohz_balance_kick;
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -0700465#endif
Mike Galbraitha64692a2010-03-11 17:16:20 +0100466 unsigned int skip_clock_update;
467
Ingo Molnard8016492007-10-18 21:32:55 +0200468 /* capture load from *all* tasks on this cpu: */
469 struct load_weight load;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200470 unsigned long nr_load_updates;
471 u64 nr_switches;
472
473 struct cfs_rq cfs;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100474 struct rt_rq rt;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100475
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200476#ifdef CONFIG_FAIR_GROUP_SCHED
Ingo Molnard8016492007-10-18 21:32:55 +0200477 /* list of leaf cfs_rq on this cpu: */
478 struct list_head leaf_cfs_rq_list;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100479#endif
480#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100481 struct list_head leaf_rt_rq_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
484 /*
485 * This is part of a global counter where only the total sum
486 * over all CPUs matters. A task can increase this counter on
487 * one CPU and if it got migrated afterwards it may decrease
488 * it on another CPU. Always updated under the runqueue lock:
489 */
490 unsigned long nr_uninterruptible;
491
Peter Zijlstra34f971f2010-09-22 13:53:15 +0200492 struct task_struct *curr, *idle, *stop;
Christoph Lameterc9819f42006-12-10 02:20:25 -0800493 unsigned long next_balance;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 struct mm_struct *prev_mm;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200495
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200496 u64 clock;
Venkatesh Pallipadi305e6832010-10-04 17:03:21 -0700497 u64 clock_task;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 atomic_t nr_iowait;
500
501#ifdef CONFIG_SMP
Ingo Molnar0eab9142008-01-25 21:08:19 +0100502 struct root_domain *rd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 struct sched_domain *sd;
504
Peter Zijlstrae51fd5e2010-05-31 12:37:30 +0200505 unsigned long cpu_power;
506
Henrik Austada0a522c2009-02-13 20:35:45 +0100507 unsigned char idle_at_tick;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 /* For active balancing */
Gregory Haskins3f029d32009-07-29 11:08:47 -0400509 int post_schedule;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 int active_balance;
511 int push_cpu;
Tejun Heo969c7922010-05-06 18:49:21 +0200512 struct cpu_stop_work active_balance_work;
Ingo Molnard8016492007-10-18 21:32:55 +0200513 /* cpu of this runqueue: */
514 int cpu;
Gregory Haskins1f11eb62008-06-04 15:04:05 -0400515 int online;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
Peter Zijlstraa8a51d52008-06-27 13:41:26 +0200517 unsigned long avg_load_per_task;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Peter Zijlstrae9e92502009-09-01 10:34:37 +0200519 u64 rt_avg;
520 u64 age_stamp;
Mike Galbraith1b9508f2009-11-04 17:53:50 +0100521 u64 idle_stamp;
522 u64 avg_idle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523#endif
524
Venkatesh Pallipadiaa483802010-10-04 17:03:22 -0700525#ifdef CONFIG_IRQ_TIME_ACCOUNTING
526 u64 prev_irq_time;
527#endif
528
Thomas Gleixnerdce48a82009-04-11 10:43:41 +0200529 /* calc_load related fields */
530 unsigned long calc_load_update;
531 long calc_load_active;
532
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100533#ifdef CONFIG_SCHED_HRTICK
Peter Zijlstra31656512008-07-18 18:01:23 +0200534#ifdef CONFIG_SMP
535 int hrtick_csd_pending;
536 struct call_single_data hrtick_csd;
537#endif
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100538 struct hrtimer hrtick_timer;
539#endif
540
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541#ifdef CONFIG_SCHEDSTATS
542 /* latency stats */
543 struct sched_info rq_sched_info;
Ken Chen9c2c4802008-12-16 23:41:22 -0800544 unsigned long long rq_cpu_time;
545 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
547 /* sys_sched_yield() stats */
Ken Chen480b9432007-10-18 21:32:56 +0200548 unsigned int yld_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
550 /* schedule() stats */
Ken Chen480b9432007-10-18 21:32:56 +0200551 unsigned int sched_switch;
552 unsigned int sched_count;
553 unsigned int sched_goidle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
555 /* try_to_wake_up() stats */
Ken Chen480b9432007-10-18 21:32:56 +0200556 unsigned int ttwu_count;
557 unsigned int ttwu_local;
Ingo Molnarb8efb562007-10-15 17:00:10 +0200558
559 /* BKL stats */
Ken Chen480b9432007-10-18 21:32:56 +0200560 unsigned int bkl_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561#endif
562};
563
Fenghua Yuf34e3b62007-07-19 01:48:13 -0700564static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
Mike Galbraitha64692a2010-03-11 17:16:20 +0100566
Peter Zijlstra1e5a7402010-10-31 12:37:04 +0100567static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
Ingo Molnardd41f592007-07-09 18:51:59 +0200568
Christoph Lameter0a2966b2006-09-25 23:30:51 -0700569static inline int cpu_of(struct rq *rq)
570{
571#ifdef CONFIG_SMP
572 return rq->cpu;
573#else
574 return 0;
575#endif
576}
577
Paul E. McKenney497f0ab2010-02-22 17:04:51 -0800578#define rcu_dereference_check_sched_domain(p) \
Paul E. McKenneyd11c5632010-02-22 17:04:50 -0800579 rcu_dereference_check((p), \
580 rcu_read_lock_sched_held() || \
581 lockdep_is_held(&sched_domains_mutex))
582
Ingo Molnar20d315d2007-07-09 18:51:58 +0200583/*
Nick Piggin674311d2005-06-25 14:57:27 -0700584 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -0700585 * See detach_destroy_domains: synchronize_sched for details.
Nick Piggin674311d2005-06-25 14:57:27 -0700586 *
587 * The domain tree of any CPU may only be accessed from within
588 * preempt-disabled sections.
589 */
Ingo Molnar48f24c42006-07-03 00:25:40 -0700590#define for_each_domain(cpu, __sd) \
Paul E. McKenney497f0ab2010-02-22 17:04:51 -0800591 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
593#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
594#define this_rq() (&__get_cpu_var(runqueues))
595#define task_rq(p) cpu_rq(task_cpu(p))
596#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
Hitoshi Mitake54d35f22009-06-29 14:44:57 +0900597#define raw_rq() (&__raw_get_cpu_var(runqueues))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200599#ifdef CONFIG_CGROUP_SCHED
600
601/*
602 * Return the group to which this tasks belongs.
603 *
604 * We use task_subsys_state_check() and extend the RCU verification
605 * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach()
606 * holds that lock for each task it moves into the cgroup. Therefore
607 * by holding that lock, we pin the task to the current cgroup.
608 */
609static inline struct task_group *task_group(struct task_struct *p)
610{
Mike Galbraith5091faa2010-11-30 14:18:03 +0100611 struct task_group *tg;
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200612 struct cgroup_subsys_state *css;
613
614 css = task_subsys_state_check(p, cpu_cgroup_subsys_id,
615 lockdep_is_held(&task_rq(p)->lock));
Mike Galbraith5091faa2010-11-30 14:18:03 +0100616 tg = container_of(css, struct task_group, css);
617
618 return autogroup_task_group(p, tg);
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200619}
620
621/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
622static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
623{
624#ifdef CONFIG_FAIR_GROUP_SCHED
625 p->se.cfs_rq = task_group(p)->cfs_rq[cpu];
626 p->se.parent = task_group(p)->se[cpu];
627#endif
628
629#ifdef CONFIG_RT_GROUP_SCHED
630 p->rt.rt_rq = task_group(p)->rt_rq[cpu];
631 p->rt.parent = task_group(p)->rt_se[cpu];
632#endif
633}
634
635#else /* CONFIG_CGROUP_SCHED */
636
637static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
638static inline struct task_group *task_group(struct task_struct *p)
639{
640 return NULL;
641}
642
643#endif /* CONFIG_CGROUP_SCHED */
644
Venkatesh Pallipadi305e6832010-10-04 17:03:21 -0700645static u64 irq_time_cpu(int cpu);
Venkatesh Pallipadiaa483802010-10-04 17:03:22 -0700646static void sched_irq_time_avg_update(struct rq *rq, u64 irq_time);
Venkatesh Pallipadi305e6832010-10-04 17:03:21 -0700647
Ingo Molnaraa9c4c02008-12-17 14:10:57 +0100648inline void update_rq_clock(struct rq *rq)
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200649{
Venkatesh Pallipadi305e6832010-10-04 17:03:21 -0700650 if (!rq->skip_clock_update) {
651 int cpu = cpu_of(rq);
652 u64 irq_time;
653
654 rq->clock = sched_clock_cpu(cpu);
655 irq_time = irq_time_cpu(cpu);
656 if (rq->clock - irq_time > rq->clock_task)
657 rq->clock_task = rq->clock - irq_time;
Venkatesh Pallipadiaa483802010-10-04 17:03:22 -0700658
659 sched_irq_time_avg_update(rq, irq_time);
Venkatesh Pallipadi305e6832010-10-04 17:03:21 -0700660 }
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200661}
662
Ingo Molnare436d802007-07-19 21:28:35 +0200663/*
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200664 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
665 */
666#ifdef CONFIG_SCHED_DEBUG
667# define const_debug __read_mostly
668#else
669# define const_debug static const
670#endif
671
Ingo Molnar017730c2008-05-12 21:20:52 +0200672/**
673 * runqueue_is_locked
Randy Dunlape17b38b2009-10-11 19:12:00 -0700674 * @cpu: the processor in question.
Ingo Molnar017730c2008-05-12 21:20:52 +0200675 *
676 * Returns true if the current cpu runqueue is locked.
677 * This interface allows printk to be called with the runqueue lock
678 * held and know whether or not it is OK to wake up the klogd.
679 */
Andrew Morton89f19f02009-09-19 11:55:44 -0700680int runqueue_is_locked(int cpu)
Ingo Molnar017730c2008-05-12 21:20:52 +0200681{
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100682 return raw_spin_is_locked(&cpu_rq(cpu)->lock);
Ingo Molnar017730c2008-05-12 21:20:52 +0200683}
684
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200685/*
686 * Debugging: various feature bits
687 */
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200688
689#define SCHED_FEAT(name, enabled) \
690 __SCHED_FEAT_##name ,
691
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200692enum {
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200693#include "sched_features.h"
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200694};
695
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200696#undef SCHED_FEAT
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200697
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200698#define SCHED_FEAT(name, enabled) \
699 (1UL << __SCHED_FEAT_##name) * enabled |
700
701const_debug unsigned int sysctl_sched_features =
702#include "sched_features.h"
703 0;
704
705#undef SCHED_FEAT
706
707#ifdef CONFIG_SCHED_DEBUG
708#define SCHED_FEAT(name, enabled) \
709 #name ,
710
Harvey Harrison983ed7a2008-04-24 18:17:55 -0700711static __read_mostly char *sched_feat_names[] = {
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200712#include "sched_features.h"
713 NULL
714};
715
716#undef SCHED_FEAT
717
Li Zefan34f3a812008-10-30 15:23:32 +0800718static int sched_feat_show(struct seq_file *m, void *v)
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200719{
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200720 int i;
721
722 for (i = 0; sched_feat_names[i]; i++) {
Li Zefan34f3a812008-10-30 15:23:32 +0800723 if (!(sysctl_sched_features & (1UL << i)))
724 seq_puts(m, "NO_");
725 seq_printf(m, "%s ", sched_feat_names[i]);
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200726 }
Li Zefan34f3a812008-10-30 15:23:32 +0800727 seq_puts(m, "\n");
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200728
Li Zefan34f3a812008-10-30 15:23:32 +0800729 return 0;
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200730}
731
732static ssize_t
733sched_feat_write(struct file *filp, const char __user *ubuf,
734 size_t cnt, loff_t *ppos)
735{
736 char buf[64];
Mathieu Desnoyers77401912010-09-13 17:47:00 -0400737 char *cmp;
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200738 int neg = 0;
739 int i;
740
741 if (cnt > 63)
742 cnt = 63;
743
744 if (copy_from_user(&buf, ubuf, cnt))
745 return -EFAULT;
746
747 buf[cnt] = 0;
Mathieu Desnoyers77401912010-09-13 17:47:00 -0400748 cmp = strstrip(buf);
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200749
Ingo Molnarc24b7c52008-04-18 10:55:34 +0200750 if (strncmp(buf, "NO_", 3) == 0) {
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200751 neg = 1;
752 cmp += 3;
753 }
754
755 for (i = 0; sched_feat_names[i]; i++) {
Mathieu Desnoyers77401912010-09-13 17:47:00 -0400756 if (strcmp(cmp, sched_feat_names[i]) == 0) {
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200757 if (neg)
758 sysctl_sched_features &= ~(1UL << i);
759 else
760 sysctl_sched_features |= (1UL << i);
761 break;
762 }
763 }
764
765 if (!sched_feat_names[i])
766 return -EINVAL;
767
Jan Blunck42994722009-11-20 17:40:37 +0100768 *ppos += cnt;
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200769
770 return cnt;
771}
772
Li Zefan34f3a812008-10-30 15:23:32 +0800773static int sched_feat_open(struct inode *inode, struct file *filp)
774{
775 return single_open(filp, sched_feat_show, NULL);
776}
777
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700778static const struct file_operations sched_feat_fops = {
Li Zefan34f3a812008-10-30 15:23:32 +0800779 .open = sched_feat_open,
780 .write = sched_feat_write,
781 .read = seq_read,
782 .llseek = seq_lseek,
783 .release = single_release,
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200784};
785
786static __init int sched_init_debug(void)
787{
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200788 debugfs_create_file("sched_features", 0644, NULL, NULL,
789 &sched_feat_fops);
790
791 return 0;
792}
793late_initcall(sched_init_debug);
794
795#endif
796
797#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200798
799/*
Peter Zijlstrab82d9fd2007-11-09 22:39:39 +0100800 * Number of tasks to iterate in a single balance run.
801 * Limited because this is done with IRQs disabled.
802 */
803const_debug unsigned int sysctl_sched_nr_migrate = 32;
804
805/*
Peter Zijlstrae9e92502009-09-01 10:34:37 +0200806 * period over which we average the RT time consumption, measured
807 * in ms.
808 *
809 * default: 1s
810 */
811const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC;
812
813/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100814 * period over which we measure -rt task cpu usage in us.
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100815 * default: 1s
816 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100817unsigned int sysctl_sched_rt_period = 1000000;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100818
Ingo Molnar6892b752008-02-13 14:02:36 +0100819static __read_mostly int scheduler_running;
820
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100821/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100822 * part of the period that we allow rt tasks to run in us.
823 * default: 0.95s
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100824 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100825int sysctl_sched_rt_runtime = 950000;
826
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200827static inline u64 global_rt_period(void)
828{
829 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
830}
831
832static inline u64 global_rt_runtime(void)
833{
roel kluine26873b2008-07-22 16:51:15 -0400834 if (sysctl_sched_rt_runtime < 0)
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200835 return RUNTIME_INF;
836
837 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
838}
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100839
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840#ifndef prepare_arch_switch
Nick Piggin4866cde2005-06-25 14:57:23 -0700841# define prepare_arch_switch(next) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842#endif
Nick Piggin4866cde2005-06-25 14:57:23 -0700843#ifndef finish_arch_switch
844# define finish_arch_switch(prev) do { } while (0)
845#endif
846
Dmitry Adamushko051a1d12007-12-18 15:21:13 +0100847static inline int task_current(struct rq *rq, struct task_struct *p)
848{
849 return rq->curr == p;
850}
851
Nick Piggin4866cde2005-06-25 14:57:23 -0700852#ifndef __ARCH_WANT_UNLOCKED_CTXSW
Ingo Molnar70b97a72006-07-03 00:25:42 -0700853static inline int task_running(struct rq *rq, struct task_struct *p)
Nick Piggin4866cde2005-06-25 14:57:23 -0700854{
Dmitry Adamushko051a1d12007-12-18 15:21:13 +0100855 return task_current(rq, p);
Nick Piggin4866cde2005-06-25 14:57:23 -0700856}
857
Ingo Molnar70b97a72006-07-03 00:25:42 -0700858static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
Nick Piggin4866cde2005-06-25 14:57:23 -0700859{
860}
861
Ingo Molnar70b97a72006-07-03 00:25:42 -0700862static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
Nick Piggin4866cde2005-06-25 14:57:23 -0700863{
Ingo Molnarda04c032005-09-13 11:17:59 +0200864#ifdef CONFIG_DEBUG_SPINLOCK
865 /* this is a valid case when another task releases the spinlock */
866 rq->lock.owner = current;
867#endif
Ingo Molnar8a25d5d2006-07-03 00:24:54 -0700868 /*
869 * If we are tracking spinlock dependencies then we have to
870 * fix up the runqueue lock - which gets 'carried over' from
871 * prev into current:
872 */
873 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
874
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100875 raw_spin_unlock_irq(&rq->lock);
Nick Piggin4866cde2005-06-25 14:57:23 -0700876}
877
878#else /* __ARCH_WANT_UNLOCKED_CTXSW */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700879static inline int task_running(struct rq *rq, struct task_struct *p)
Nick Piggin4866cde2005-06-25 14:57:23 -0700880{
881#ifdef CONFIG_SMP
882 return p->oncpu;
883#else
Dmitry Adamushko051a1d12007-12-18 15:21:13 +0100884 return task_current(rq, p);
Nick Piggin4866cde2005-06-25 14:57:23 -0700885#endif
886}
887
Ingo Molnar70b97a72006-07-03 00:25:42 -0700888static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
Nick Piggin4866cde2005-06-25 14:57:23 -0700889{
890#ifdef CONFIG_SMP
891 /*
892 * We can optimise this out completely for !SMP, because the
893 * SMP rebalancing from interrupt is the only thing that cares
894 * here.
895 */
896 next->oncpu = 1;
897#endif
898#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100899 raw_spin_unlock_irq(&rq->lock);
Nick Piggin4866cde2005-06-25 14:57:23 -0700900#else
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100901 raw_spin_unlock(&rq->lock);
Nick Piggin4866cde2005-06-25 14:57:23 -0700902#endif
903}
904
Ingo Molnar70b97a72006-07-03 00:25:42 -0700905static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
Nick Piggin4866cde2005-06-25 14:57:23 -0700906{
907#ifdef CONFIG_SMP
908 /*
909 * After ->oncpu is cleared, the task can be moved to a different CPU.
910 * We must ensure this doesn't happen until the switch is completely
911 * finished.
912 */
913 smp_wmb();
914 prev->oncpu = 0;
915#endif
916#ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
917 local_irq_enable();
918#endif
919}
920#endif /* __ARCH_WANT_UNLOCKED_CTXSW */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
922/*
Peter Zijlstra65cc8e42010-03-25 21:05:16 +0100923 * Check whether the task is waking, we use this to synchronize ->cpus_allowed
924 * against ttwu().
Peter Zijlstra0970d292010-02-15 14:45:54 +0100925 */
926static inline int task_is_waking(struct task_struct *p)
927{
Peter Zijlstra0017d732010-03-24 18:34:10 +0100928 return unlikely(p->state == TASK_WAKING);
Peter Zijlstra0970d292010-02-15 14:45:54 +0100929}
930
931/*
Ingo Molnarb29739f2006-06-27 02:54:51 -0700932 * __task_rq_lock - lock the runqueue a given task resides on.
933 * Must be called interrupts disabled.
934 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700935static inline struct rq *__task_rq_lock(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -0700936 __acquires(rq->lock)
937{
Peter Zijlstra0970d292010-02-15 14:45:54 +0100938 struct rq *rq;
939
Andi Kleen3a5c3592007-10-15 17:00:14 +0200940 for (;;) {
Peter Zijlstra0970d292010-02-15 14:45:54 +0100941 rq = task_rq(p);
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100942 raw_spin_lock(&rq->lock);
Peter Zijlstra65cc8e42010-03-25 21:05:16 +0100943 if (likely(rq == task_rq(p)))
Andi Kleen3a5c3592007-10-15 17:00:14 +0200944 return rq;
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100945 raw_spin_unlock(&rq->lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -0700946 }
Ingo Molnarb29739f2006-06-27 02:54:51 -0700947}
948
949/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 * task_rq_lock - lock the runqueue a given task resides on and disable
Ingo Molnar41a2d6c2007-12-05 15:46:09 +0100951 * interrupts. Note the ordering: we can safely lookup the task_rq without
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 * explicitly disabling preemption.
953 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700954static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 __acquires(rq->lock)
956{
Ingo Molnar70b97a72006-07-03 00:25:42 -0700957 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
Andi Kleen3a5c3592007-10-15 17:00:14 +0200959 for (;;) {
960 local_irq_save(*flags);
961 rq = task_rq(p);
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100962 raw_spin_lock(&rq->lock);
Peter Zijlstra65cc8e42010-03-25 21:05:16 +0100963 if (likely(rq == task_rq(p)))
Andi Kleen3a5c3592007-10-15 17:00:14 +0200964 return rq;
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100965 raw_spin_unlock_irqrestore(&rq->lock, *flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967}
968
Alexey Dobriyana9957442007-10-15 17:00:13 +0200969static void __task_rq_unlock(struct rq *rq)
Ingo Molnarb29739f2006-06-27 02:54:51 -0700970 __releases(rq->lock)
971{
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100972 raw_spin_unlock(&rq->lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -0700973}
974
Ingo Molnar70b97a72006-07-03 00:25:42 -0700975static inline void task_rq_unlock(struct rq *rq, unsigned long *flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 __releases(rq->lock)
977{
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100978 raw_spin_unlock_irqrestore(&rq->lock, *flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979}
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981/*
Robert P. J. Daycc2a73b2006-12-10 02:20:00 -0800982 * this_rq_lock - lock this runqueue and disable interrupts.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 */
Alexey Dobriyana9957442007-10-15 17:00:13 +0200984static struct rq *this_rq_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 __acquires(rq->lock)
986{
Ingo Molnar70b97a72006-07-03 00:25:42 -0700987 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
989 local_irq_disable();
990 rq = this_rq();
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100991 raw_spin_lock(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
993 return rq;
994}
995
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100996#ifdef CONFIG_SCHED_HRTICK
997/*
998 * Use HR-timers to deliver accurate preemption points.
999 *
1000 * Its all a bit involved since we cannot program an hrt while holding the
1001 * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a
1002 * reschedule event.
1003 *
1004 * When we get rescheduled we reprogram the hrtick_timer outside of the
1005 * rq->lock.
1006 */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001007
1008/*
1009 * Use hrtick when:
1010 * - enabled by features
1011 * - hrtimer is actually high res
1012 */
1013static inline int hrtick_enabled(struct rq *rq)
1014{
1015 if (!sched_feat(HRTICK))
1016 return 0;
Ingo Molnarba420592008-07-20 11:02:06 +02001017 if (!cpu_active(cpu_of(rq)))
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001018 return 0;
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001019 return hrtimer_is_hres_active(&rq->hrtick_timer);
1020}
1021
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001022static void hrtick_clear(struct rq *rq)
1023{
1024 if (hrtimer_active(&rq->hrtick_timer))
1025 hrtimer_cancel(&rq->hrtick_timer);
1026}
1027
1028/*
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001029 * High-resolution timer tick.
1030 * Runs from hardirq context with interrupts disabled.
1031 */
1032static enum hrtimer_restart hrtick(struct hrtimer *timer)
1033{
1034 struct rq *rq = container_of(timer, struct rq, hrtick_timer);
1035
1036 WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
1037
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001038 raw_spin_lock(&rq->lock);
Peter Zijlstra3e51f332008-05-03 18:29:28 +02001039 update_rq_clock(rq);
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001040 rq->curr->sched_class->task_tick(rq, rq->curr, 1);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001041 raw_spin_unlock(&rq->lock);
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001042
1043 return HRTIMER_NORESTART;
1044}
1045
Rabin Vincent95e904c2008-05-11 05:55:33 +05301046#ifdef CONFIG_SMP
Peter Zijlstra31656512008-07-18 18:01:23 +02001047/*
1048 * called from hardirq (IPI) context
1049 */
1050static void __hrtick_start(void *arg)
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001051{
Peter Zijlstra31656512008-07-18 18:01:23 +02001052 struct rq *rq = arg;
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001053
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001054 raw_spin_lock(&rq->lock);
Peter Zijlstra31656512008-07-18 18:01:23 +02001055 hrtimer_restart(&rq->hrtick_timer);
1056 rq->hrtick_csd_pending = 0;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001057 raw_spin_unlock(&rq->lock);
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001058}
1059
Peter Zijlstra31656512008-07-18 18:01:23 +02001060/*
1061 * Called to set the hrtick timer state.
1062 *
1063 * called with rq->lock held and irqs disabled
1064 */
1065static void hrtick_start(struct rq *rq, u64 delay)
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001066{
Peter Zijlstra31656512008-07-18 18:01:23 +02001067 struct hrtimer *timer = &rq->hrtick_timer;
1068 ktime_t time = ktime_add_ns(timer->base->get_time(), delay);
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001069
Arjan van de Vencc584b22008-09-01 15:02:30 -07001070 hrtimer_set_expires(timer, time);
Peter Zijlstra31656512008-07-18 18:01:23 +02001071
1072 if (rq == this_rq()) {
1073 hrtimer_restart(timer);
1074 } else if (!rq->hrtick_csd_pending) {
Peter Zijlstra6e275632009-02-25 13:59:48 +01001075 __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0);
Peter Zijlstra31656512008-07-18 18:01:23 +02001076 rq->hrtick_csd_pending = 1;
1077 }
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001078}
1079
1080static int
1081hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu)
1082{
1083 int cpu = (int)(long)hcpu;
1084
1085 switch (action) {
1086 case CPU_UP_CANCELED:
1087 case CPU_UP_CANCELED_FROZEN:
1088 case CPU_DOWN_PREPARE:
1089 case CPU_DOWN_PREPARE_FROZEN:
1090 case CPU_DEAD:
1091 case CPU_DEAD_FROZEN:
Peter Zijlstra31656512008-07-18 18:01:23 +02001092 hrtick_clear(cpu_rq(cpu));
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001093 return NOTIFY_OK;
1094 }
1095
1096 return NOTIFY_DONE;
1097}
1098
Rakib Mullickfa748202008-09-22 14:55:45 -07001099static __init void init_hrtick(void)
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001100{
1101 hotcpu_notifier(hotplug_hrtick, 0);
1102}
Peter Zijlstra31656512008-07-18 18:01:23 +02001103#else
1104/*
1105 * Called to set the hrtick timer state.
1106 *
1107 * called with rq->lock held and irqs disabled
1108 */
1109static void hrtick_start(struct rq *rq, u64 delay)
1110{
Peter Zijlstra7f1e2ca2009-03-13 12:21:27 +01001111 __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0,
Arun R Bharadwaj5c333862009-04-16 12:14:37 +05301112 HRTIMER_MODE_REL_PINNED, 0);
Peter Zijlstra31656512008-07-18 18:01:23 +02001113}
1114
Andrew Morton006c75f2008-09-22 14:55:46 -07001115static inline void init_hrtick(void)
Peter Zijlstra31656512008-07-18 18:01:23 +02001116{
1117}
Rabin Vincent95e904c2008-05-11 05:55:33 +05301118#endif /* CONFIG_SMP */
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001119
1120static void init_rq_hrtick(struct rq *rq)
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001121{
Peter Zijlstra31656512008-07-18 18:01:23 +02001122#ifdef CONFIG_SMP
1123 rq->hrtick_csd_pending = 0;
1124
1125 rq->hrtick_csd.flags = 0;
1126 rq->hrtick_csd.func = __hrtick_start;
1127 rq->hrtick_csd.info = rq;
1128#endif
1129
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001130 hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1131 rq->hrtick_timer.function = hrtick;
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001132}
Andrew Morton006c75f2008-09-22 14:55:46 -07001133#else /* CONFIG_SCHED_HRTICK */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001134static inline void hrtick_clear(struct rq *rq)
1135{
1136}
1137
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001138static inline void init_rq_hrtick(struct rq *rq)
1139{
1140}
1141
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001142static inline void init_hrtick(void)
1143{
1144}
Andrew Morton006c75f2008-09-22 14:55:46 -07001145#endif /* CONFIG_SCHED_HRTICK */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001146
Ingo Molnar1b9f19c2007-07-09 18:51:59 +02001147/*
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001148 * resched_task - mark a task 'to be rescheduled now'.
1149 *
1150 * On UP this means the setting of the need_resched flag, on SMP it
1151 * might also involve a cross-CPU call to trigger the scheduler on
1152 * the target CPU.
1153 */
1154#ifdef CONFIG_SMP
1155
1156#ifndef tsk_is_polling
1157#define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG)
1158#endif
1159
Peter Zijlstra31656512008-07-18 18:01:23 +02001160static void resched_task(struct task_struct *p)
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001161{
1162 int cpu;
1163
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001164 assert_raw_spin_locked(&task_rq(p)->lock);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001165
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08001166 if (test_tsk_need_resched(p))
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001167 return;
1168
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08001169 set_tsk_need_resched(p);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001170
1171 cpu = task_cpu(p);
1172 if (cpu == smp_processor_id())
1173 return;
1174
1175 /* NEED_RESCHED must be visible before we test polling */
1176 smp_mb();
1177 if (!tsk_is_polling(p))
1178 smp_send_reschedule(cpu);
1179}
1180
1181static void resched_cpu(int cpu)
1182{
1183 struct rq *rq = cpu_rq(cpu);
1184 unsigned long flags;
1185
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001186 if (!raw_spin_trylock_irqsave(&rq->lock, flags))
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001187 return;
1188 resched_task(cpu_curr(cpu));
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001189 raw_spin_unlock_irqrestore(&rq->lock, flags);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001190}
Thomas Gleixner06d83082008-03-22 09:20:24 +01001191
1192#ifdef CONFIG_NO_HZ
1193/*
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -07001194 * In the semi idle case, use the nearest busy cpu for migrating timers
1195 * from an idle cpu. This is good for power-savings.
1196 *
1197 * We don't do similar optimization for completely idle system, as
1198 * selecting an idle cpu will add more delays to the timers than intended
1199 * (as that cpu's timer base may not be uptodate wrt jiffies etc).
1200 */
1201int get_nohz_timer_target(void)
1202{
1203 int cpu = smp_processor_id();
1204 int i;
1205 struct sched_domain *sd;
1206
1207 for_each_domain(cpu, sd) {
1208 for_each_cpu(i, sched_domain_span(sd))
1209 if (!idle_cpu(i))
1210 return i;
1211 }
1212 return cpu;
1213}
1214/*
Thomas Gleixner06d83082008-03-22 09:20:24 +01001215 * When add_timer_on() enqueues a timer into the timer wheel of an
1216 * idle CPU then this timer might expire before the next timer event
1217 * which is scheduled to wake up that CPU. In case of a completely
1218 * idle system the next event might even be infinite time into the
1219 * future. wake_up_idle_cpu() ensures that the CPU is woken up and
1220 * leaves the inner idle loop so the newly added timer is taken into
1221 * account when the CPU goes back to idle and evaluates the timer
1222 * wheel for the next timer event.
1223 */
1224void wake_up_idle_cpu(int cpu)
1225{
1226 struct rq *rq = cpu_rq(cpu);
1227
1228 if (cpu == smp_processor_id())
1229 return;
1230
1231 /*
1232 * This is safe, as this function is called with the timer
1233 * wheel base lock of (cpu) held. When the CPU is on the way
1234 * to idle and has not yet set rq->curr to idle then it will
1235 * be serialized on the timer wheel base lock and take the new
1236 * timer into account automatically.
1237 */
1238 if (rq->curr != rq->idle)
1239 return;
1240
1241 /*
1242 * We can set TIF_RESCHED on the idle task of the other CPU
1243 * lockless. The worst case is that the other CPU runs the
1244 * idle task through an additional NOOP schedule()
1245 */
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08001246 set_tsk_need_resched(rq->idle);
Thomas Gleixner06d83082008-03-22 09:20:24 +01001247
1248 /* NEED_RESCHED must be visible before we test polling */
1249 smp_mb();
1250 if (!tsk_is_polling(rq->idle))
1251 smp_send_reschedule(cpu);
1252}
Mike Galbraith39c0cbe2010-03-11 17:17:13 +01001253
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02001254#endif /* CONFIG_NO_HZ */
Thomas Gleixner06d83082008-03-22 09:20:24 +01001255
Peter Zijlstrae9e92502009-09-01 10:34:37 +02001256static u64 sched_avg_period(void)
1257{
1258 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1259}
1260
1261static void sched_avg_update(struct rq *rq)
1262{
1263 s64 period = sched_avg_period();
1264
1265 while ((s64)(rq->clock - rq->age_stamp) > period) {
Will Deacon0d98bb22010-05-24 12:11:43 -07001266 /*
1267 * Inline assembly required to prevent the compiler
1268 * optimising this loop into a divmod call.
1269 * See __iter_div_u64_rem() for another example of this.
1270 */
1271 asm("" : "+rm" (rq->age_stamp));
Peter Zijlstrae9e92502009-09-01 10:34:37 +02001272 rq->age_stamp += period;
1273 rq->rt_avg /= 2;
1274 }
1275}
1276
1277static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1278{
1279 rq->rt_avg += rt_delta;
1280 sched_avg_update(rq);
1281}
1282
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02001283#else /* !CONFIG_SMP */
Peter Zijlstra31656512008-07-18 18:01:23 +02001284static void resched_task(struct task_struct *p)
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001285{
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001286 assert_raw_spin_locked(&task_rq(p)->lock);
Peter Zijlstra31656512008-07-18 18:01:23 +02001287 set_tsk_need_resched(p);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001288}
Peter Zijlstrae9e92502009-09-01 10:34:37 +02001289
1290static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1291{
1292}
Suresh Siddhada2b71e2010-08-23 13:42:51 -07001293
1294static void sched_avg_update(struct rq *rq)
1295{
1296}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02001297#endif /* CONFIG_SMP */
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001298
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001299#if BITS_PER_LONG == 32
1300# define WMULT_CONST (~0UL)
1301#else
1302# define WMULT_CONST (1UL << 32)
1303#endif
1304
1305#define WMULT_SHIFT 32
1306
Ingo Molnar194081e2007-08-09 11:16:51 +02001307/*
1308 * Shift right and round:
1309 */
Ingo Molnarcf2ab462007-09-05 14:32:49 +02001310#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
Ingo Molnar194081e2007-08-09 11:16:51 +02001311
Peter Zijlstraa7be37a2008-06-27 13:41:11 +02001312/*
1313 * delta *= weight / lw
1314 */
Ingo Molnarcb1c4fc2007-08-02 17:41:40 +02001315static unsigned long
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001316calc_delta_mine(unsigned long delta_exec, unsigned long weight,
1317 struct load_weight *lw)
1318{
1319 u64 tmp;
1320
Lai Jiangshan7a232e02008-06-12 16:43:07 +08001321 if (!lw->inv_weight) {
1322 if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST))
1323 lw->inv_weight = 1;
1324 else
1325 lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2)
1326 / (lw->weight+1);
1327 }
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001328
1329 tmp = (u64)delta_exec * weight;
1330 /*
1331 * Check whether we'd overflow the 64-bit multiplication:
1332 */
Ingo Molnar194081e2007-08-09 11:16:51 +02001333 if (unlikely(tmp > WMULT_CONST))
Ingo Molnarcf2ab462007-09-05 14:32:49 +02001334 tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
Ingo Molnar194081e2007-08-09 11:16:51 +02001335 WMULT_SHIFT/2);
1336 else
Ingo Molnarcf2ab462007-09-05 14:32:49 +02001337 tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001338
Ingo Molnarecf691d2007-08-02 17:41:40 +02001339 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001340}
1341
Ingo Molnar10919852007-10-15 17:00:04 +02001342static inline void update_load_add(struct load_weight *lw, unsigned long inc)
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001343{
1344 lw->weight += inc;
Ingo Molnare89996a2008-03-14 23:48:28 +01001345 lw->inv_weight = 0;
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001346}
1347
Ingo Molnar10919852007-10-15 17:00:04 +02001348static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001349{
1350 lw->weight -= dec;
Ingo Molnare89996a2008-03-14 23:48:28 +01001351 lw->inv_weight = 0;
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001352}
1353
Peter Zijlstra2069dd72010-11-15 15:47:00 -08001354static inline void update_load_set(struct load_weight *lw, unsigned long w)
1355{
1356 lw->weight = w;
1357 lw->inv_weight = 0;
1358}
1359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360/*
Peter Williams2dd73a42006-06-27 02:54:34 -07001361 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1362 * of tasks with abnormal "nice" values across CPUs the contribution that
1363 * each task makes to its run queue's load is weighted according to its
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01001364 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
Peter Williams2dd73a42006-06-27 02:54:34 -07001365 * scaled version of the new time slice allocation that they receive on time
1366 * slice expiry etc.
1367 */
1368
Peter Zijlstracce7ade2009-01-15 14:53:37 +01001369#define WEIGHT_IDLEPRIO 3
1370#define WMULT_IDLEPRIO 1431655765
Ingo Molnardd41f592007-07-09 18:51:59 +02001371
1372/*
1373 * Nice levels are multiplicative, with a gentle 10% change for every
1374 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1375 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1376 * that remained on nice 0.
1377 *
1378 * The "10% effect" is relative and cumulative: from _any_ nice level,
1379 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
Ingo Molnarf9153ee2007-07-16 09:46:30 +02001380 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1381 * If a task goes up by ~10% and another task goes down by ~10% then
1382 * the relative distance between them is ~25%.)
Ingo Molnardd41f592007-07-09 18:51:59 +02001383 */
1384static const int prio_to_weight[40] = {
Ingo Molnar254753d2007-08-09 11:16:51 +02001385 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1386 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1387 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1388 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1389 /* 0 */ 1024, 820, 655, 526, 423,
1390 /* 5 */ 335, 272, 215, 172, 137,
1391 /* 10 */ 110, 87, 70, 56, 45,
1392 /* 15 */ 36, 29, 23, 18, 15,
Ingo Molnardd41f592007-07-09 18:51:59 +02001393};
1394
Ingo Molnar5714d2d2007-07-16 09:46:31 +02001395/*
1396 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1397 *
1398 * In cases where the weight does not change often, we can use the
1399 * precalculated inverse to speed up arithmetics by turning divisions
1400 * into multiplications:
1401 */
Ingo Molnardd41f592007-07-09 18:51:59 +02001402static const u32 prio_to_wmult[40] = {
Ingo Molnar254753d2007-08-09 11:16:51 +02001403 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1404 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1405 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1406 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1407 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1408 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1409 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1410 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
Ingo Molnardd41f592007-07-09 18:51:59 +02001411};
Peter Williams2dd73a42006-06-27 02:54:34 -07001412
Bharata B Raoef12fef2009-03-31 10:02:22 +05301413/* Time spent by the tasks of the cpu accounting group executing in ... */
1414enum cpuacct_stat_index {
1415 CPUACCT_STAT_USER, /* ... user mode */
1416 CPUACCT_STAT_SYSTEM, /* ... kernel mode */
1417
1418 CPUACCT_STAT_NSTATS,
1419};
1420
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01001421#ifdef CONFIG_CGROUP_CPUACCT
1422static void cpuacct_charge(struct task_struct *tsk, u64 cputime);
Bharata B Raoef12fef2009-03-31 10:02:22 +05301423static void cpuacct_update_stats(struct task_struct *tsk,
1424 enum cpuacct_stat_index idx, cputime_t val);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01001425#else
1426static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
Bharata B Raoef12fef2009-03-31 10:02:22 +05301427static inline void cpuacct_update_stats(struct task_struct *tsk,
1428 enum cpuacct_stat_index idx, cputime_t val) {}
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01001429#endif
1430
Peter Zijlstra18d95a22008-04-19 19:45:00 +02001431static inline void inc_cpu_load(struct rq *rq, unsigned long load)
1432{
1433 update_load_add(&rq->load, load);
1434}
1435
1436static inline void dec_cpu_load(struct rq *rq, unsigned long load)
1437{
1438 update_load_sub(&rq->load, load);
1439}
1440
Ingo Molnar7940ca32008-08-19 13:40:47 +02001441#if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED)
Peter Zijlstraeb755802008-08-19 12:33:05 +02001442typedef int (*tg_visitor)(struct task_group *, void *);
1443
1444/*
1445 * Iterate the full tree, calling @down when first entering a node and @up when
1446 * leaving it for the final time.
1447 */
1448static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
1449{
1450 struct task_group *parent, *child;
1451 int ret;
1452
1453 rcu_read_lock();
1454 parent = &root_task_group;
1455down:
1456 ret = (*down)(parent, data);
1457 if (ret)
1458 goto out_unlock;
1459 list_for_each_entry_rcu(child, &parent->children, siblings) {
1460 parent = child;
1461 goto down;
1462
1463up:
1464 continue;
1465 }
1466 ret = (*up)(parent, data);
1467 if (ret)
1468 goto out_unlock;
1469
1470 child = parent;
1471 parent = parent->parent;
1472 if (parent)
1473 goto up;
1474out_unlock:
1475 rcu_read_unlock();
1476
1477 return ret;
1478}
1479
1480static int tg_nop(struct task_group *tg, void *data)
1481{
1482 return 0;
1483}
1484#endif
1485
Gregory Haskinse7693a32008-01-25 21:08:09 +01001486#ifdef CONFIG_SMP
Peter Zijlstraf5f08f32009-09-10 13:35:28 +02001487/* Used instead of source_load when we know the type == 0 */
1488static unsigned long weighted_cpuload(const int cpu)
1489{
1490 return cpu_rq(cpu)->load.weight;
1491}
1492
1493/*
1494 * Return a low guess at the load of a migration-source cpu weighted
1495 * according to the scheduling class and "nice" value.
1496 *
1497 * We want to under-estimate the load of migration sources, to
1498 * balance conservatively.
1499 */
1500static unsigned long source_load(int cpu, int type)
1501{
1502 struct rq *rq = cpu_rq(cpu);
1503 unsigned long total = weighted_cpuload(cpu);
1504
1505 if (type == 0 || !sched_feat(LB_BIAS))
1506 return total;
1507
1508 return min(rq->cpu_load[type-1], total);
1509}
1510
1511/*
1512 * Return a high guess at the load of a migration-target cpu weighted
1513 * according to the scheduling class and "nice" value.
1514 */
1515static unsigned long target_load(int cpu, int type)
1516{
1517 struct rq *rq = cpu_rq(cpu);
1518 unsigned long total = weighted_cpuload(cpu);
1519
1520 if (type == 0 || !sched_feat(LB_BIAS))
1521 return total;
1522
1523 return max(rq->cpu_load[type-1], total);
1524}
1525
Peter Zijlstraae154be2009-09-10 14:40:57 +02001526static unsigned long power_of(int cpu)
1527{
Peter Zijlstrae51fd5e2010-05-31 12:37:30 +02001528 return cpu_rq(cpu)->cpu_power;
Peter Zijlstraae154be2009-09-10 14:40:57 +02001529}
1530
Gregory Haskinse7693a32008-01-25 21:08:09 +01001531static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001532
Peter Zijlstraa8a51d52008-06-27 13:41:26 +02001533static unsigned long cpu_avg_load_per_task(int cpu)
1534{
1535 struct rq *rq = cpu_rq(cpu);
Ingo Molnaraf6d5962008-11-29 20:45:15 +01001536 unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
Peter Zijlstraa8a51d52008-06-27 13:41:26 +02001537
Steven Rostedt4cd42622008-11-26 21:04:24 -05001538 if (nr_running)
1539 rq->avg_load_per_task = rq->load.weight / nr_running;
Balbir Singha2d47772008-11-12 16:19:00 +05301540 else
1541 rq->avg_load_per_task = 0;
Peter Zijlstraa8a51d52008-06-27 13:41:26 +02001542
1543 return rq->avg_load_per_task;
1544}
1545
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001546#ifdef CONFIG_FAIR_GROUP_SCHED
1547
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001548/*
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001549 * Compute the cpu's hierarchical load factor for each task group.
1550 * This needs to be done in a top-down fashion because the load of a child
1551 * group is a fraction of its parents load.
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001552 */
Peter Zijlstraeb755802008-08-19 12:33:05 +02001553static int tg_load_down(struct task_group *tg, void *data)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001554{
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001555 unsigned long load;
Peter Zijlstraeb755802008-08-19 12:33:05 +02001556 long cpu = (long)data;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001557
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001558 if (!tg->parent) {
1559 load = cpu_rq(cpu)->load.weight;
1560 } else {
1561 load = tg->parent->cfs_rq[cpu]->h_load;
Peter Zijlstra2069dd72010-11-15 15:47:00 -08001562 load *= tg->se[cpu]->load.weight;
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001563 load /= tg->parent->cfs_rq[cpu]->load.weight + 1;
1564 }
1565
1566 tg->cfs_rq[cpu]->h_load = load;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001567
Peter Zijlstraeb755802008-08-19 12:33:05 +02001568 return 0;
Peter Zijlstra4d8d5952008-06-27 13:41:19 +02001569}
1570
Peter Zijlstraeb755802008-08-19 12:33:05 +02001571static void update_h_load(long cpu)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001572{
Peter Zijlstraeb755802008-08-19 12:33:05 +02001573 walk_tg_tree(tg_load_down, tg_nop, (void *)cpu);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001574}
1575
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001576#endif
1577
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001578#ifdef CONFIG_PREEMPT
1579
Peter Zijlstrab78bb862009-09-15 14:23:18 +02001580static void double_rq_lock(struct rq *rq1, struct rq *rq2);
1581
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001582/*
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001583 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1584 * way at the expense of forcing extra atomic operations in all
1585 * invocations. This assures that the double_lock is acquired using the
1586 * same underlying policy as the spinlock_t on this architecture, which
1587 * reduces latency compared to the unfair variant below. However, it
1588 * also adds more overhead and therefore may reduce throughput.
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001589 */
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001590static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1591 __releases(this_rq->lock)
1592 __acquires(busiest->lock)
1593 __acquires(this_rq->lock)
1594{
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001595 raw_spin_unlock(&this_rq->lock);
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001596 double_rq_lock(this_rq, busiest);
1597
1598 return 1;
1599}
1600
1601#else
1602/*
1603 * Unfair double_lock_balance: Optimizes throughput at the expense of
1604 * latency by eliminating extra atomic operations when the locks are
1605 * already in proper order on entry. This favors lower cpu-ids and will
1606 * grant the double lock to lower cpus over higher ids under contention,
1607 * regardless of entry order into the function.
1608 */
1609static int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001610 __releases(this_rq->lock)
1611 __acquires(busiest->lock)
1612 __acquires(this_rq->lock)
1613{
1614 int ret = 0;
1615
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001616 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001617 if (busiest < this_rq) {
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001618 raw_spin_unlock(&this_rq->lock);
1619 raw_spin_lock(&busiest->lock);
1620 raw_spin_lock_nested(&this_rq->lock,
1621 SINGLE_DEPTH_NESTING);
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001622 ret = 1;
1623 } else
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001624 raw_spin_lock_nested(&busiest->lock,
1625 SINGLE_DEPTH_NESTING);
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001626 }
1627 return ret;
1628}
1629
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001630#endif /* CONFIG_PREEMPT */
1631
1632/*
1633 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1634 */
1635static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1636{
1637 if (unlikely(!irqs_disabled())) {
1638 /* printk() doesn't work good under rq->lock */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001639 raw_spin_unlock(&this_rq->lock);
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001640 BUG_ON(1);
1641 }
1642
1643 return _double_lock_balance(this_rq, busiest);
1644}
1645
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001646static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1647 __releases(busiest->lock)
1648{
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001649 raw_spin_unlock(&busiest->lock);
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001650 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1651}
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001652
1653/*
1654 * double_rq_lock - safely lock two runqueues
1655 *
1656 * Note this does not disable interrupts like task_rq_lock,
1657 * you need to do so manually before calling.
1658 */
1659static void double_rq_lock(struct rq *rq1, struct rq *rq2)
1660 __acquires(rq1->lock)
1661 __acquires(rq2->lock)
1662{
1663 BUG_ON(!irqs_disabled());
1664 if (rq1 == rq2) {
1665 raw_spin_lock(&rq1->lock);
1666 __acquire(rq2->lock); /* Fake it out ;) */
1667 } else {
1668 if (rq1 < rq2) {
1669 raw_spin_lock(&rq1->lock);
1670 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1671 } else {
1672 raw_spin_lock(&rq2->lock);
1673 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1674 }
1675 }
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001676}
1677
1678/*
1679 * double_rq_unlock - safely unlock two runqueues
1680 *
1681 * Note this does not restore interrupts like task_rq_unlock,
1682 * you need to do so manually after calling.
1683 */
1684static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1685 __releases(rq1->lock)
1686 __releases(rq2->lock)
1687{
1688 raw_spin_unlock(&rq1->lock);
1689 if (rq1 != rq2)
1690 raw_spin_unlock(&rq2->lock);
1691 else
1692 __release(rq2->lock);
1693}
1694
Peter Zijlstra18d95a22008-04-19 19:45:00 +02001695#endif
Peter Zijlstra18d95a22008-04-19 19:45:00 +02001696
Peter Zijlstra74f51872010-04-22 21:50:19 +02001697static void calc_load_account_idle(struct rq *this_rq);
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01001698static void update_sysctl(void);
Christian Ehrhardtacb4a842009-11-30 12:16:48 +01001699static int get_update_sysctl_factor(void);
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07001700static void update_cpu_load(struct rq *this_rq);
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02001701
Peter Zijlstracd29fe62009-11-27 17:32:46 +01001702static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1703{
1704 set_task_rq(p, cpu);
1705#ifdef CONFIG_SMP
1706 /*
1707 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1708 * successfuly executed on another CPU. We must ensure that updates of
1709 * per-task data have been completed by this moment.
1710 */
1711 smp_wmb();
1712 task_thread_info(p)->cpu = cpu;
1713#endif
1714}
Peter Zijlstra18d95a22008-04-19 19:45:00 +02001715
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001716static const struct sched_class rt_sched_class;
Ingo Molnardd41f592007-07-09 18:51:59 +02001717
Peter Zijlstra34f971f2010-09-22 13:53:15 +02001718#define sched_class_highest (&stop_sched_class)
Gregory Haskins1f11eb62008-06-04 15:04:05 -04001719#define for_each_class(class) \
1720 for (class = sched_class_highest; class; class = class->next)
Ingo Molnardd41f592007-07-09 18:51:59 +02001721
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001722#include "sched_stats.h"
1723
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001724static void inc_nr_running(struct rq *rq)
Ingo Molnar6363ca52008-05-29 11:28:57 +02001725{
1726 rq->nr_running++;
Ingo Molnar6363ca52008-05-29 11:28:57 +02001727}
1728
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001729static void dec_nr_running(struct rq *rq)
Ingo Molnar9c217242007-08-02 17:41:40 +02001730{
1731 rq->nr_running--;
Ingo Molnar9c217242007-08-02 17:41:40 +02001732}
1733
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001734static void set_load_weight(struct task_struct *p)
1735{
Ingo Molnardd41f592007-07-09 18:51:59 +02001736 /*
1737 * SCHED_IDLE tasks get minimal weight:
1738 */
1739 if (p->policy == SCHED_IDLE) {
1740 p->se.load.weight = WEIGHT_IDLEPRIO;
1741 p->se.load.inv_weight = WMULT_IDLEPRIO;
1742 return;
1743 }
1744
1745 p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO];
1746 p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO];
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001747}
1748
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001749static void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
Gregory Haskins2087a1a2008-06-27 14:30:00 -06001750{
Mike Galbraitha64692a2010-03-11 17:16:20 +01001751 update_rq_clock(rq);
Ingo Molnar71f8bd42007-07-09 18:51:59 +02001752 sched_info_queued(p);
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001753 p->sched_class->enqueue_task(rq, p, flags);
Ingo Molnardd41f592007-07-09 18:51:59 +02001754 p->se.on_rq = 1;
1755}
1756
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001757static void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
Ingo Molnardd41f592007-07-09 18:51:59 +02001758{
Mike Galbraitha64692a2010-03-11 17:16:20 +01001759 update_rq_clock(rq);
Ankita Garg46ac22b2008-07-01 14:30:06 +05301760 sched_info_dequeued(p);
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001761 p->sched_class->dequeue_task(rq, p, flags);
Ingo Molnardd41f592007-07-09 18:51:59 +02001762 p->se.on_rq = 0;
Ingo Molnar71f8bd42007-07-09 18:51:59 +02001763}
1764
1765/*
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001766 * activate_task - move a task to the runqueue.
1767 */
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001768static void activate_task(struct rq *rq, struct task_struct *p, int flags)
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001769{
1770 if (task_contributes_to_load(p))
1771 rq->nr_uninterruptible--;
1772
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001773 enqueue_task(rq, p, flags);
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001774 inc_nr_running(rq);
1775}
1776
1777/*
1778 * deactivate_task - remove a task from the runqueue.
1779 */
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001780static void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001781{
1782 if (task_contributes_to_load(p))
1783 rq->nr_uninterruptible++;
1784
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001785 dequeue_task(rq, p, flags);
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001786 dec_nr_running(rq);
1787}
1788
Venkatesh Pallipadib52bfee2010-10-04 17:03:19 -07001789#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1790
Venkatesh Pallipadi305e6832010-10-04 17:03:21 -07001791/*
1792 * There are no locks covering percpu hardirq/softirq time.
1793 * They are only modified in account_system_vtime, on corresponding CPU
1794 * with interrupts disabled. So, writes are safe.
1795 * They are read and saved off onto struct rq in update_rq_clock().
1796 * This may result in other CPU reading this CPU's irq time and can
1797 * race with irq/account_system_vtime on this CPU. We would either get old
1798 * or new value (or semi updated value on 32 bit) with a side effect of
1799 * accounting a slice of irq time to wrong task when irq is in progress
1800 * while we read rq->clock. That is a worthy compromise in place of having
1801 * locks on each irq in account_system_time.
1802 */
Venkatesh Pallipadib52bfee2010-10-04 17:03:19 -07001803static DEFINE_PER_CPU(u64, cpu_hardirq_time);
1804static DEFINE_PER_CPU(u64, cpu_softirq_time);
1805
1806static DEFINE_PER_CPU(u64, irq_start_time);
1807static int sched_clock_irqtime;
1808
1809void enable_sched_clock_irqtime(void)
1810{
1811 sched_clock_irqtime = 1;
1812}
1813
1814void disable_sched_clock_irqtime(void)
1815{
1816 sched_clock_irqtime = 0;
1817}
1818
Venkatesh Pallipadi305e6832010-10-04 17:03:21 -07001819static u64 irq_time_cpu(int cpu)
1820{
1821 if (!sched_clock_irqtime)
1822 return 0;
1823
1824 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1825}
1826
Venkatesh Pallipadib52bfee2010-10-04 17:03:19 -07001827void account_system_vtime(struct task_struct *curr)
1828{
1829 unsigned long flags;
1830 int cpu;
1831 u64 now, delta;
1832
1833 if (!sched_clock_irqtime)
1834 return;
1835
1836 local_irq_save(flags);
1837
Venkatesh Pallipadib52bfee2010-10-04 17:03:19 -07001838 cpu = smp_processor_id();
Venkatesh Pallipadid267f872010-10-04 17:03:23 -07001839 now = sched_clock_cpu(cpu);
Venkatesh Pallipadib52bfee2010-10-04 17:03:19 -07001840 delta = now - per_cpu(irq_start_time, cpu);
1841 per_cpu(irq_start_time, cpu) = now;
1842 /*
1843 * We do not account for softirq time from ksoftirqd here.
1844 * We want to continue accounting softirq time to ksoftirqd thread
1845 * in that case, so as not to confuse scheduler with a special task
1846 * that do not consume any time, but still wants to run.
1847 */
1848 if (hardirq_count())
1849 per_cpu(cpu_hardirq_time, cpu) += delta;
1850 else if (in_serving_softirq() && !(curr->flags & PF_KSOFTIRQD))
1851 per_cpu(cpu_softirq_time, cpu) += delta;
1852
1853 local_irq_restore(flags);
1854}
Ingo Molnarb7dadc32010-10-18 20:00:37 +02001855EXPORT_SYMBOL_GPL(account_system_vtime);
Venkatesh Pallipadib52bfee2010-10-04 17:03:19 -07001856
Venkatesh Pallipadiaa483802010-10-04 17:03:22 -07001857static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time)
1858{
1859 if (sched_clock_irqtime && sched_feat(NONIRQ_POWER)) {
1860 u64 delta_irq = curr_irq_time - rq->prev_irq_time;
1861 rq->prev_irq_time = curr_irq_time;
1862 sched_rt_avg_update(rq, delta_irq);
1863 }
1864}
1865
Venkatesh Pallipadi305e6832010-10-04 17:03:21 -07001866#else
1867
1868static u64 irq_time_cpu(int cpu)
1869{
1870 return 0;
1871}
1872
Venkatesh Pallipadiaa483802010-10-04 17:03:22 -07001873static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time) { }
1874
Venkatesh Pallipadib52bfee2010-10-04 17:03:19 -07001875#endif
1876
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001877#include "sched_idletask.c"
1878#include "sched_fair.c"
1879#include "sched_rt.c"
Mike Galbraith5091faa2010-11-30 14:18:03 +01001880#include "sched_autogroup.c"
Peter Zijlstra34f971f2010-09-22 13:53:15 +02001881#include "sched_stoptask.c"
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001882#ifdef CONFIG_SCHED_DEBUG
1883# include "sched_debug.c"
1884#endif
1885
Peter Zijlstra34f971f2010-09-22 13:53:15 +02001886void sched_set_stop_task(int cpu, struct task_struct *stop)
1887{
1888 struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
1889 struct task_struct *old_stop = cpu_rq(cpu)->stop;
1890
1891 if (stop) {
1892 /*
1893 * Make it appear like a SCHED_FIFO task, its something
1894 * userspace knows about and won't get confused about.
1895 *
1896 * Also, it will make PI more or less work without too
1897 * much confusion -- but then, stop work should not
1898 * rely on PI working anyway.
1899 */
1900 sched_setscheduler_nocheck(stop, SCHED_FIFO, &param);
1901
1902 stop->sched_class = &stop_sched_class;
1903 }
1904
1905 cpu_rq(cpu)->stop = stop;
1906
1907 if (old_stop) {
1908 /*
1909 * Reset it back to a normal scheduling class so that
1910 * it can die in pieces.
1911 */
1912 old_stop->sched_class = &rt_sched_class;
1913 }
1914}
1915
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001916/*
Ingo Molnardd41f592007-07-09 18:51:59 +02001917 * __normal_prio - return the priority that is based on the static prio
Ingo Molnar71f8bd42007-07-09 18:51:59 +02001918 */
Ingo Molnar14531182007-07-09 18:51:59 +02001919static inline int __normal_prio(struct task_struct *p)
1920{
Ingo Molnardd41f592007-07-09 18:51:59 +02001921 return p->static_prio;
Ingo Molnar14531182007-07-09 18:51:59 +02001922}
1923
1924/*
Ingo Molnarb29739f2006-06-27 02:54:51 -07001925 * Calculate the expected normal priority: i.e. priority
1926 * without taking RT-inheritance into account. Might be
1927 * boosted by interactivity modifiers. Changes upon fork,
1928 * setprio syscalls, and whenever the interactivity
1929 * estimator recalculates.
1930 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001931static inline int normal_prio(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -07001932{
1933 int prio;
1934
Ingo Molnare05606d2007-07-09 18:51:59 +02001935 if (task_has_rt_policy(p))
Ingo Molnarb29739f2006-06-27 02:54:51 -07001936 prio = MAX_RT_PRIO-1 - p->rt_priority;
1937 else
1938 prio = __normal_prio(p);
1939 return prio;
1940}
1941
1942/*
1943 * Calculate the current priority, i.e. the priority
1944 * taken into account by the scheduler. This value might
1945 * be boosted by RT tasks, or might be boosted by
1946 * interactivity modifiers. Will be RT if the task got
1947 * RT-boosted. If not then it returns p->normal_prio.
1948 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001949static int effective_prio(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -07001950{
1951 p->normal_prio = normal_prio(p);
1952 /*
1953 * If we are RT tasks or we were boosted to RT priority,
1954 * keep the priority unchanged. Otherwise, update priority
1955 * to the normal priority:
1956 */
1957 if (!rt_prio(p->prio))
1958 return p->normal_prio;
1959 return p->prio;
1960}
1961
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962/**
1963 * task_curr - is this task currently executing on a CPU?
1964 * @p: the task in question.
1965 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001966inline int task_curr(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967{
1968 return cpu_curr(task_cpu(p)) == p;
1969}
1970
Steven Rostedtcb469842008-01-25 21:08:22 +01001971static inline void check_class_changed(struct rq *rq, struct task_struct *p,
1972 const struct sched_class *prev_class,
1973 int oldprio, int running)
1974{
1975 if (prev_class != p->sched_class) {
1976 if (prev_class->switched_from)
1977 prev_class->switched_from(rq, p, running);
1978 p->sched_class->switched_to(rq, p, running);
1979 } else
1980 p->sched_class->prio_changed(rq, p, oldprio, running);
1981}
1982
Peter Zijlstra1e5a7402010-10-31 12:37:04 +01001983static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
1984{
1985 const struct sched_class *class;
1986
1987 if (p->sched_class == rq->curr->sched_class) {
1988 rq->curr->sched_class->check_preempt_curr(rq, p, flags);
1989 } else {
1990 for_each_class(class) {
1991 if (class == rq->curr->sched_class)
1992 break;
1993 if (class == p->sched_class) {
1994 resched_task(rq->curr);
1995 break;
1996 }
1997 }
1998 }
1999
2000 /*
2001 * A queue event has occurred, and we're going to schedule. In
2002 * this case, we can save a useless back to back clock update.
2003 */
2004 if (test_tsk_need_resched(rq->curr))
2005 rq->skip_clock_update = 1;
2006}
2007
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008#ifdef CONFIG_SMP
Ingo Molnarcc367732007-10-15 17:00:18 +02002009/*
2010 * Is this task likely cache-hot:
2011 */
Gregory Haskinse7693a32008-01-25 21:08:09 +01002012static int
Ingo Molnarcc367732007-10-15 17:00:18 +02002013task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
2014{
2015 s64 delta;
2016
Peter Zijlstrae6c8fba2009-12-16 18:04:33 +01002017 if (p->sched_class != &fair_sched_class)
2018 return 0;
2019
Nikhil Raoef8002f2010-10-13 12:09:35 -07002020 if (unlikely(p->policy == SCHED_IDLE))
2021 return 0;
2022
Ingo Molnarf540a602008-03-15 17:10:34 +01002023 /*
2024 * Buddy candidates are cache hot:
2025 */
Mike Galbraithf685cea2009-10-23 23:09:22 +02002026 if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running &&
Peter Zijlstra47932412008-11-04 21:25:09 +01002027 (&p->se == cfs_rq_of(&p->se)->next ||
2028 &p->se == cfs_rq_of(&p->se)->last))
Ingo Molnarf540a602008-03-15 17:10:34 +01002029 return 1;
2030
Ingo Molnar6bc16652007-10-15 17:00:18 +02002031 if (sysctl_sched_migration_cost == -1)
2032 return 1;
2033 if (sysctl_sched_migration_cost == 0)
2034 return 0;
2035
Ingo Molnarcc367732007-10-15 17:00:18 +02002036 delta = now - p->se.exec_start;
2037
2038 return delta < (s64)sysctl_sched_migration_cost;
2039}
2040
Ingo Molnardd41f592007-07-09 18:51:59 +02002041void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
Ingo Molnarc65cc872007-07-09 18:51:58 +02002042{
Peter Zijlstrae2912002009-12-16 18:04:36 +01002043#ifdef CONFIG_SCHED_DEBUG
2044 /*
2045 * We should never call set_task_cpu() on a blocked task,
2046 * ttwu() will sort out the placement.
2047 */
Peter Zijlstra077614e2009-12-17 13:16:31 +01002048 WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
2049 !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE));
Peter Zijlstrae2912002009-12-16 18:04:36 +01002050#endif
2051
Mathieu Desnoyersde1d7282009-05-05 16:49:59 +08002052 trace_sched_migrate_task(p, new_cpu);
Peter Zijlstracbc34ed2008-12-10 08:08:22 +01002053
Peter Zijlstra0c697742009-12-22 15:43:19 +01002054 if (task_cpu(p) != new_cpu) {
2055 p->se.nr_migrations++;
2056 perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0);
2057 }
Ingo Molnardd41f592007-07-09 18:51:59 +02002058
2059 __set_task_cpu(p, new_cpu);
Ingo Molnarc65cc872007-07-09 18:51:58 +02002060}
2061
Tejun Heo969c7922010-05-06 18:49:21 +02002062struct migration_arg {
Ingo Molnar36c8b582006-07-03 00:25:41 -07002063 struct task_struct *task;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 int dest_cpu;
Ingo Molnar70b97a72006-07-03 00:25:42 -07002065};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066
Tejun Heo969c7922010-05-06 18:49:21 +02002067static int migration_cpu_stop(void *data);
2068
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069/*
2070 * The task's runqueue lock must be held.
2071 * Returns true if you have to wait for migration thread.
2072 */
Nikanth Karthikesanb7a2b392010-11-26 12:37:09 +05302073static bool migrate_task(struct task_struct *p, struct rq *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 /*
2076 * If the task is not on a runqueue (and not running), then
Peter Zijlstrae2912002009-12-16 18:04:36 +01002077 * the next wake-up will properly place the task.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 */
Tejun Heo969c7922010-05-06 18:49:21 +02002079 return p->se.on_rq || task_running(rq, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080}
2081
2082/*
2083 * wait_task_inactive - wait for a thread to unschedule.
2084 *
Roland McGrath85ba2d82008-07-25 19:45:58 -07002085 * If @match_state is nonzero, it's the @p->state value just checked and
2086 * not expected to change. If it changes, i.e. @p might have woken up,
2087 * then return zero. When we succeed in waiting for @p to be off its CPU,
2088 * we return a positive number (its total switch count). If a second call
2089 * a short while later returns the same number, the caller can be sure that
2090 * @p has remained unscheduled the whole time.
2091 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 * The caller must ensure that the task *will* unschedule sometime soon,
2093 * else this function might spin for a *long* time. This function can't
2094 * be called with interrupts off, or it may introduce deadlock with
2095 * smp_call_function() if an IPI is sent by the same process we are
2096 * waiting to become inactive.
2097 */
Roland McGrath85ba2d82008-07-25 19:45:58 -07002098unsigned long wait_task_inactive(struct task_struct *p, long match_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099{
2100 unsigned long flags;
Ingo Molnardd41f592007-07-09 18:51:59 +02002101 int running, on_rq;
Roland McGrath85ba2d82008-07-25 19:45:58 -07002102 unsigned long ncsw;
Ingo Molnar70b97a72006-07-03 00:25:42 -07002103 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
Andi Kleen3a5c3592007-10-15 17:00:14 +02002105 for (;;) {
2106 /*
2107 * We do the initial early heuristics without holding
2108 * any task-queue locks at all. We'll only try to get
2109 * the runqueue lock when things look like they will
2110 * work out!
2111 */
2112 rq = task_rq(p);
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07002113
Andi Kleen3a5c3592007-10-15 17:00:14 +02002114 /*
2115 * If the task is actively running on another CPU
2116 * still, just relax and busy-wait without holding
2117 * any locks.
2118 *
2119 * NOTE! Since we don't hold any locks, it's not
2120 * even sure that "rq" stays as the right runqueue!
2121 * But we don't care, since "task_running()" will
2122 * return false if the runqueue has changed and p
2123 * is actually now running somewhere else!
2124 */
Roland McGrath85ba2d82008-07-25 19:45:58 -07002125 while (task_running(rq, p)) {
2126 if (match_state && unlikely(p->state != match_state))
2127 return 0;
Andi Kleen3a5c3592007-10-15 17:00:14 +02002128 cpu_relax();
Roland McGrath85ba2d82008-07-25 19:45:58 -07002129 }
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07002130
Andi Kleen3a5c3592007-10-15 17:00:14 +02002131 /*
2132 * Ok, time to look more closely! We need the rq
2133 * lock now, to be *sure*. If we're wrong, we'll
2134 * just go back and repeat.
2135 */
2136 rq = task_rq_lock(p, &flags);
Peter Zijlstra27a9da62010-05-04 20:36:56 +02002137 trace_sched_wait_task(p);
Andi Kleen3a5c3592007-10-15 17:00:14 +02002138 running = task_running(rq, p);
2139 on_rq = p->se.on_rq;
Roland McGrath85ba2d82008-07-25 19:45:58 -07002140 ncsw = 0;
Oleg Nesterovf31e11d2008-08-20 16:54:44 -07002141 if (!match_state || p->state == match_state)
Oleg Nesterov93dcf552008-08-20 16:54:44 -07002142 ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
Andi Kleen3a5c3592007-10-15 17:00:14 +02002143 task_rq_unlock(rq, &flags);
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07002144
Andi Kleen3a5c3592007-10-15 17:00:14 +02002145 /*
Roland McGrath85ba2d82008-07-25 19:45:58 -07002146 * If it changed from the expected state, bail out now.
2147 */
2148 if (unlikely(!ncsw))
2149 break;
2150
2151 /*
Andi Kleen3a5c3592007-10-15 17:00:14 +02002152 * Was it really running after all now that we
2153 * checked with the proper locks actually held?
2154 *
2155 * Oops. Go back and try again..
2156 */
2157 if (unlikely(running)) {
2158 cpu_relax();
2159 continue;
2160 }
2161
2162 /*
2163 * It's not enough that it's not actively running,
2164 * it must be off the runqueue _entirely_, and not
2165 * preempted!
2166 *
Luis Henriques80dd99b2009-03-16 19:58:09 +00002167 * So if it was still runnable (but just not actively
Andi Kleen3a5c3592007-10-15 17:00:14 +02002168 * running right now), it's preempted, and we should
2169 * yield - it could be a while.
2170 */
2171 if (unlikely(on_rq)) {
2172 schedule_timeout_uninterruptible(1);
2173 continue;
2174 }
2175
2176 /*
2177 * Ahh, all good. It wasn't running, and it wasn't
2178 * runnable, which means that it will never become
2179 * running in the future either. We're all done!
2180 */
2181 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 }
Roland McGrath85ba2d82008-07-25 19:45:58 -07002183
2184 return ncsw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185}
2186
2187/***
2188 * kick_process - kick a running thread to enter/exit the kernel
2189 * @p: the to-be-kicked thread
2190 *
2191 * Cause a process which is running on another CPU to enter
2192 * kernel-mode, without any delay. (to get signals handled.)
2193 *
2194 * NOTE: this function doesnt have to take the runqueue lock,
2195 * because all it wants to ensure is that the remote task enters
2196 * the kernel. If the IPI races and the task has been migrated
2197 * to another CPU then no harm is done and the purpose has been
2198 * achieved as well.
2199 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07002200void kick_process(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201{
2202 int cpu;
2203
2204 preempt_disable();
2205 cpu = task_cpu(p);
2206 if ((cpu != smp_processor_id()) && task_curr(p))
2207 smp_send_reschedule(cpu);
2208 preempt_enable();
2209}
Rusty Russellb43e3522009-06-12 22:27:00 -06002210EXPORT_SYMBOL_GPL(kick_process);
Nick Piggin476d1392005-06-25 14:57:29 -07002211#endif /* CONFIG_SMP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212
Thomas Gleixner0793a612008-12-04 20:12:29 +01002213/**
2214 * task_oncpu_function_call - call a function on the cpu on which a task runs
2215 * @p: the task to evaluate
2216 * @func: the function to be called
2217 * @info: the function call argument
2218 *
2219 * Calls the function @func when the task is currently running. This might
2220 * be on the current CPU, which just calls the function directly
2221 */
2222void task_oncpu_function_call(struct task_struct *p,
2223 void (*func) (void *info), void *info)
2224{
2225 int cpu;
2226
2227 preempt_disable();
2228 cpu = task_cpu(p);
2229 if (task_curr(p))
2230 smp_call_function_single(cpu, func, info, 1);
2231 preempt_enable();
2232}
2233
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002234#ifdef CONFIG_SMP
Oleg Nesterov30da6882010-03-15 10:10:19 +01002235/*
2236 * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held.
2237 */
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01002238static int select_fallback_rq(int cpu, struct task_struct *p)
2239{
2240 int dest_cpu;
2241 const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu));
2242
2243 /* Look for allowed, online CPU in same node. */
2244 for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask)
2245 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
2246 return dest_cpu;
2247
2248 /* Any allowed, online CPU? */
2249 dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask);
2250 if (dest_cpu < nr_cpu_ids)
2251 return dest_cpu;
2252
2253 /* No more Mr. Nice Guy. */
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01002254 dest_cpu = cpuset_cpus_allowed_fallback(p);
2255 /*
2256 * Don't tell them about moving exiting tasks or
2257 * kernel threads (both mm NULL), since they never
2258 * leave kernel.
2259 */
2260 if (p->mm && printk_ratelimit()) {
2261 printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n",
2262 task_pid_nr(p), p->comm, cpu);
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01002263 }
2264
2265 return dest_cpu;
2266}
2267
Peter Zijlstrae2912002009-12-16 18:04:36 +01002268/*
Oleg Nesterov30da6882010-03-15 10:10:19 +01002269 * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable.
Peter Zijlstrae2912002009-12-16 18:04:36 +01002270 */
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002271static inline
Peter Zijlstra0017d732010-03-24 18:34:10 +01002272int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags)
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002273{
Peter Zijlstra0017d732010-03-24 18:34:10 +01002274 int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags);
Peter Zijlstrae2912002009-12-16 18:04:36 +01002275
2276 /*
2277 * In order not to call set_task_cpu() on a blocking task we need
2278 * to rely on ttwu() to place the task on a valid ->cpus_allowed
2279 * cpu.
2280 *
2281 * Since this is common to all placement strategies, this lives here.
2282 *
2283 * [ this allows ->select_task() to simply return task_cpu(p) and
2284 * not worry about this generic constraint ]
2285 */
2286 if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) ||
Peter Zijlstra70f11202009-12-20 17:36:27 +01002287 !cpu_online(cpu)))
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01002288 cpu = select_fallback_rq(task_cpu(p), p);
Peter Zijlstrae2912002009-12-16 18:04:36 +01002289
2290 return cpu;
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002291}
Mike Galbraith09a40af2010-04-15 07:29:59 +02002292
2293static void update_avg(u64 *avg, u64 sample)
2294{
2295 s64 diff = sample - *avg;
2296 *avg += diff >> 3;
2297}
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002298#endif
2299
Tejun Heo9ed38112009-12-03 15:08:03 +09002300static inline void ttwu_activate(struct task_struct *p, struct rq *rq,
2301 bool is_sync, bool is_migrate, bool is_local,
2302 unsigned long en_flags)
2303{
2304 schedstat_inc(p, se.statistics.nr_wakeups);
2305 if (is_sync)
2306 schedstat_inc(p, se.statistics.nr_wakeups_sync);
2307 if (is_migrate)
2308 schedstat_inc(p, se.statistics.nr_wakeups_migrate);
2309 if (is_local)
2310 schedstat_inc(p, se.statistics.nr_wakeups_local);
2311 else
2312 schedstat_inc(p, se.statistics.nr_wakeups_remote);
2313
2314 activate_task(rq, p, en_flags);
2315}
2316
2317static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq,
2318 int wake_flags, bool success)
2319{
2320 trace_sched_wakeup(p, success);
2321 check_preempt_curr(rq, p, wake_flags);
2322
2323 p->state = TASK_RUNNING;
2324#ifdef CONFIG_SMP
2325 if (p->sched_class->task_woken)
2326 p->sched_class->task_woken(rq, p);
2327
2328 if (unlikely(rq->idle_stamp)) {
2329 u64 delta = rq->clock - rq->idle_stamp;
2330 u64 max = 2*sysctl_sched_migration_cost;
2331
2332 if (delta > max)
2333 rq->avg_idle = max;
2334 else
2335 update_avg(&rq->avg_idle, delta);
2336 rq->idle_stamp = 0;
2337 }
2338#endif
Tejun Heo21aa9af2010-06-08 21:40:37 +02002339 /* if a worker is waking up, notify workqueue */
2340 if ((p->flags & PF_WQ_WORKER) && success)
2341 wq_worker_waking_up(p, cpu_of(rq));
Tejun Heo9ed38112009-12-03 15:08:03 +09002342}
2343
2344/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 * try_to_wake_up - wake up a thread
Tejun Heo9ed38112009-12-03 15:08:03 +09002346 * @p: the thread to be awakened
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 * @state: the mask of task states that can be woken
Tejun Heo9ed38112009-12-03 15:08:03 +09002348 * @wake_flags: wake modifier flags (WF_*)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 *
2350 * Put it on the run-queue if it's not already there. The "current"
2351 * thread is always on the run-queue (except when the actual
2352 * re-schedule is in progress), and as such you're allowed to do
2353 * the simpler "current->state = TASK_RUNNING" to mark yourself
2354 * runnable without the overhead of this.
2355 *
Tejun Heo9ed38112009-12-03 15:08:03 +09002356 * Returns %true if @p was woken up, %false if it was already running
2357 * or @state didn't match @p's state.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 */
Peter Zijlstra7d478722009-09-14 19:55:44 +02002359static int try_to_wake_up(struct task_struct *p, unsigned int state,
2360 int wake_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361{
Ingo Molnarcc367732007-10-15 17:00:18 +02002362 int cpu, orig_cpu, this_cpu, success = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 unsigned long flags;
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01002364 unsigned long en_flags = ENQUEUE_WAKEUP;
Dan Carpenterab3b3aa2010-03-06 14:17:52 +03002365 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002367 this_cpu = get_cpu();
Peter Zijlstra2398f2c2008-06-27 13:41:35 +02002368
Linus Torvalds04e2f172008-02-23 18:05:03 -08002369 smp_wmb();
Dan Carpenterab3b3aa2010-03-06 14:17:52 +03002370 rq = task_rq_lock(p, &flags);
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002371 if (!(p->state & state))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 goto out;
2373
Ingo Molnardd41f592007-07-09 18:51:59 +02002374 if (p->se.on_rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 goto out_running;
2376
2377 cpu = task_cpu(p);
Ingo Molnarcc367732007-10-15 17:00:18 +02002378 orig_cpu = cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379
2380#ifdef CONFIG_SMP
2381 if (unlikely(task_running(rq, p)))
2382 goto out_activate;
2383
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002384 /*
2385 * In order to handle concurrent wakeups and release the rq->lock
2386 * we put the task in TASK_WAKING state.
Ingo Molnareb240732009-09-16 21:09:13 +02002387 *
2388 * First fix up the nr_uninterruptible count:
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002389 */
Peter Zijlstracc87f762010-03-26 12:22:14 +01002390 if (task_contributes_to_load(p)) {
2391 if (likely(cpu_online(orig_cpu)))
2392 rq->nr_uninterruptible--;
2393 else
2394 this_rq()->nr_uninterruptible--;
2395 }
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002396 p->state = TASK_WAKING;
Peter Zijlstraefbbd052009-12-16 18:04:40 +01002397
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01002398 if (p->sched_class->task_waking) {
Peter Zijlstraefbbd052009-12-16 18:04:40 +01002399 p->sched_class->task_waking(rq, p);
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01002400 en_flags |= ENQUEUE_WAKING;
Peter Zijlstra0970d292010-02-15 14:45:54 +01002401 }
Peter Zijlstraab19cb22009-11-27 15:44:43 +01002402
Peter Zijlstra0017d732010-03-24 18:34:10 +01002403 cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags);
2404 if (cpu != orig_cpu)
Mike Galbraithf5dc3752009-10-09 08:35:03 +02002405 set_task_cpu(p, cpu);
Peter Zijlstra0017d732010-03-24 18:34:10 +01002406 __task_rq_unlock(rq);
Peter Zijlstraab19cb22009-11-27 15:44:43 +01002407
Peter Zijlstra0970d292010-02-15 14:45:54 +01002408 rq = cpu_rq(cpu);
2409 raw_spin_lock(&rq->lock);
Mike Galbraithf5dc3752009-10-09 08:35:03 +02002410
Peter Zijlstra0970d292010-02-15 14:45:54 +01002411 /*
2412 * We migrated the task without holding either rq->lock, however
2413 * since the task is not on the task list itself, nobody else
2414 * will try and migrate the task, hence the rq should match the
2415 * cpu we just moved it to.
2416 */
2417 WARN_ON(task_cpu(p) != cpu);
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002418 WARN_ON(p->state != TASK_WAKING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419
Gregory Haskinse7693a32008-01-25 21:08:09 +01002420#ifdef CONFIG_SCHEDSTATS
2421 schedstat_inc(rq, ttwu_count);
2422 if (cpu == this_cpu)
2423 schedstat_inc(rq, ttwu_local);
2424 else {
2425 struct sched_domain *sd;
2426 for_each_domain(this_cpu, sd) {
Rusty Russell758b2cd2008-11-25 02:35:04 +10302427 if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
Gregory Haskinse7693a32008-01-25 21:08:09 +01002428 schedstat_inc(sd, ttwu_wake_remote);
2429 break;
2430 }
2431 }
2432 }
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02002433#endif /* CONFIG_SCHEDSTATS */
Gregory Haskinse7693a32008-01-25 21:08:09 +01002434
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435out_activate:
2436#endif /* CONFIG_SMP */
Tejun Heo9ed38112009-12-03 15:08:03 +09002437 ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu,
2438 cpu == this_cpu, en_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 success = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440out_running:
Tejun Heo9ed38112009-12-03 15:08:03 +09002441 ttwu_post_activation(p, rq, wake_flags, success);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442out:
2443 task_rq_unlock(rq, &flags);
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002444 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445
2446 return success;
2447}
2448
David Howells50fa6102009-04-28 15:01:38 +01002449/**
Tejun Heo21aa9af2010-06-08 21:40:37 +02002450 * try_to_wake_up_local - try to wake up a local task with rq lock held
2451 * @p: the thread to be awakened
2452 *
2453 * Put @p on the run-queue if it's not alredy there. The caller must
2454 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2455 * the current task. this_rq() stays locked over invocation.
2456 */
2457static void try_to_wake_up_local(struct task_struct *p)
2458{
2459 struct rq *rq = task_rq(p);
2460 bool success = false;
2461
2462 BUG_ON(rq != this_rq());
2463 BUG_ON(p == current);
2464 lockdep_assert_held(&rq->lock);
2465
2466 if (!(p->state & TASK_NORMAL))
2467 return;
2468
2469 if (!p->se.on_rq) {
2470 if (likely(!task_running(rq, p))) {
2471 schedstat_inc(rq, ttwu_count);
2472 schedstat_inc(rq, ttwu_local);
2473 }
2474 ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP);
2475 success = true;
2476 }
2477 ttwu_post_activation(p, rq, 0, success);
2478}
2479
2480/**
David Howells50fa6102009-04-28 15:01:38 +01002481 * wake_up_process - Wake up a specific process
2482 * @p: The process to be woken up.
2483 *
2484 * Attempt to wake up the nominated process and move it to the set of runnable
2485 * processes. Returns 1 if the process was woken up, 0 if it was already
2486 * running.
2487 *
2488 * It may be assumed that this function implies a write memory barrier before
2489 * changing the task state if and only if any tasks are woken up.
2490 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08002491int wake_up_process(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492{
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05002493 return try_to_wake_up(p, TASK_ALL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495EXPORT_SYMBOL(wake_up_process);
2496
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08002497int wake_up_state(struct task_struct *p, unsigned int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498{
2499 return try_to_wake_up(p, state, 0);
2500}
2501
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502/*
2503 * Perform scheduler related setup for a newly forked process p.
2504 * p is forked by current.
Ingo Molnardd41f592007-07-09 18:51:59 +02002505 *
2506 * __sched_fork() is basic setup used by init_idle() too:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 */
Ingo Molnardd41f592007-07-09 18:51:59 +02002508static void __sched_fork(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509{
Ingo Molnardd41f592007-07-09 18:51:59 +02002510 p->se.exec_start = 0;
2511 p->se.sum_exec_runtime = 0;
Ingo Molnarf6cf8912007-08-28 12:53:24 +02002512 p->se.prev_sum_exec_runtime = 0;
Ingo Molnar6c594c22008-12-14 12:34:15 +01002513 p->se.nr_migrations = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02002514
2515#ifdef CONFIG_SCHEDSTATS
Lucas De Marchi41acab82010-03-10 23:37:45 -03002516 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02002517#endif
Nick Piggin476d1392005-06-25 14:57:29 -07002518
Peter Zijlstrafa717062008-01-25 21:08:27 +01002519 INIT_LIST_HEAD(&p->rt.run_list);
Ingo Molnardd41f592007-07-09 18:51:59 +02002520 p->se.on_rq = 0;
Peter Zijlstra4a55bd52008-04-19 19:45:00 +02002521 INIT_LIST_HEAD(&p->se.group_node);
Nick Piggin476d1392005-06-25 14:57:29 -07002522
Avi Kivitye107be32007-07-26 13:40:43 +02002523#ifdef CONFIG_PREEMPT_NOTIFIERS
2524 INIT_HLIST_HEAD(&p->preempt_notifiers);
2525#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02002526}
2527
2528/*
2529 * fork()/clone()-time setup:
2530 */
2531void sched_fork(struct task_struct *p, int clone_flags)
2532{
2533 int cpu = get_cpu();
2534
2535 __sched_fork(p);
Peter Zijlstra06b83b52009-12-16 18:04:35 +01002536 /*
Peter Zijlstra0017d732010-03-24 18:34:10 +01002537 * We mark the process as running here. This guarantees that
Peter Zijlstra06b83b52009-12-16 18:04:35 +01002538 * nobody will actually run it, and a signal or other external
2539 * event cannot wake it up and insert it on the runqueue either.
2540 */
Peter Zijlstra0017d732010-03-24 18:34:10 +01002541 p->state = TASK_RUNNING;
Ingo Molnardd41f592007-07-09 18:51:59 +02002542
Ingo Molnarb29739f2006-06-27 02:54:51 -07002543 /*
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02002544 * Revert to default priority/policy on fork if requested.
2545 */
2546 if (unlikely(p->sched_reset_on_fork)) {
Peter Williamsf83f9ac2009-09-24 06:47:10 +00002547 if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) {
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02002548 p->policy = SCHED_NORMAL;
Peter Williamsf83f9ac2009-09-24 06:47:10 +00002549 p->normal_prio = p->static_prio;
2550 }
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02002551
Mike Galbraith6c697bd2009-06-17 10:48:02 +02002552 if (PRIO_TO_NICE(p->static_prio) < 0) {
2553 p->static_prio = NICE_TO_PRIO(0);
Peter Williamsf83f9ac2009-09-24 06:47:10 +00002554 p->normal_prio = p->static_prio;
Mike Galbraith6c697bd2009-06-17 10:48:02 +02002555 set_load_weight(p);
2556 }
2557
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02002558 /*
2559 * We don't need the reset flag anymore after the fork. It has
2560 * fulfilled its duty:
2561 */
2562 p->sched_reset_on_fork = 0;
2563 }
Lennart Poetteringca94c442009-06-15 17:17:47 +02002564
Peter Williamsf83f9ac2009-09-24 06:47:10 +00002565 /*
2566 * Make sure we do not leak PI boosting priority to the child.
2567 */
2568 p->prio = current->normal_prio;
2569
Hiroshi Shimamoto2ddbf952007-10-15 17:00:11 +02002570 if (!rt_prio(p->prio))
2571 p->sched_class = &fair_sched_class;
Ingo Molnarb29739f2006-06-27 02:54:51 -07002572
Peter Zijlstracd29fe62009-11-27 17:32:46 +01002573 if (p->sched_class->task_fork)
2574 p->sched_class->task_fork(p);
2575
Peter Zijlstra86951592010-06-22 11:44:53 +02002576 /*
2577 * The child is not yet in the pid-hash so no cgroup attach races,
2578 * and the cgroup is pinned to this child due to cgroup_fork()
2579 * is ran before sched_fork().
2580 *
2581 * Silence PROVE_RCU.
2582 */
2583 rcu_read_lock();
Peter Zijlstra5f3edc12009-09-10 13:42:00 +02002584 set_task_cpu(p, cpu);
Peter Zijlstra86951592010-06-22 11:44:53 +02002585 rcu_read_unlock();
Peter Zijlstra5f3edc12009-09-10 13:42:00 +02002586
Chandra Seetharaman52f17b62006-07-14 00:24:38 -07002587#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
Ingo Molnardd41f592007-07-09 18:51:59 +02002588 if (likely(sched_info_on()))
Chandra Seetharaman52f17b62006-07-14 00:24:38 -07002589 memset(&p->sched_info, 0, sizeof(p->sched_info));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590#endif
Chen, Kenneth Wd6077cb2006-02-14 13:53:10 -08002591#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
Nick Piggin4866cde2005-06-25 14:57:23 -07002592 p->oncpu = 0;
2593#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594#ifdef CONFIG_PREEMPT
Nick Piggin4866cde2005-06-25 14:57:23 -07002595 /* Want to start with kernel preemption disabled. */
Al Viroa1261f52005-11-13 16:06:55 -08002596 task_thread_info(p)->preempt_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597#endif
Dario Faggioli806c09a2010-11-30 19:51:33 +01002598#ifdef CONFIG_SMP
Gregory Haskins917b6272008-12-29 09:39:53 -05002599 plist_node_init(&p->pushable_tasks, MAX_PRIO);
Dario Faggioli806c09a2010-11-30 19:51:33 +01002600#endif
Gregory Haskins917b6272008-12-29 09:39:53 -05002601
Nick Piggin476d1392005-06-25 14:57:29 -07002602 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603}
2604
2605/*
2606 * wake_up_new_task - wake up a newly created task for the first time.
2607 *
2608 * This function will do some initial scheduler statistics housekeeping
2609 * that must be done for every newly created context, then puts the task
2610 * on the runqueue and wakes it.
2611 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08002612void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613{
2614 unsigned long flags;
Ingo Molnardd41f592007-07-09 18:51:59 +02002615 struct rq *rq;
Andrew Mortonc8906922010-03-11 14:08:43 -08002616 int cpu __maybe_unused = get_cpu();
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002617
2618#ifdef CONFIG_SMP
Peter Zijlstra0017d732010-03-24 18:34:10 +01002619 rq = task_rq_lock(p, &flags);
2620 p->state = TASK_WAKING;
2621
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002622 /*
2623 * Fork balancing, do it here and not earlier because:
2624 * - cpus_allowed can change in the fork path
2625 * - any previously selected cpu might disappear through hotplug
2626 *
Peter Zijlstra0017d732010-03-24 18:34:10 +01002627 * We set TASK_WAKING so that select_task_rq() can drop rq->lock
2628 * without people poking at ->cpus_allowed.
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002629 */
Peter Zijlstra0017d732010-03-24 18:34:10 +01002630 cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0);
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002631 set_task_cpu(p, cpu);
Peter Zijlstra0017d732010-03-24 18:34:10 +01002632
2633 p->state = TASK_RUNNING;
2634 task_rq_unlock(rq, &flags);
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002635#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636
Peter Zijlstra0017d732010-03-24 18:34:10 +01002637 rq = task_rq_lock(p, &flags);
Peter Zijlstracd29fe62009-11-27 17:32:46 +01002638 activate_task(rq, p, 0);
Peter Zijlstra27a9da62010-05-04 20:36:56 +02002639 trace_sched_wakeup_new(p, 1);
Peter Zijlstraa7558e02009-09-14 20:02:34 +02002640 check_preempt_curr(rq, p, WF_FORK);
Steven Rostedt9a897c52008-01-25 21:08:22 +01002641#ifdef CONFIG_SMP
Peter Zijlstraefbbd052009-12-16 18:04:40 +01002642 if (p->sched_class->task_woken)
2643 p->sched_class->task_woken(rq, p);
Steven Rostedt9a897c52008-01-25 21:08:22 +01002644#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02002645 task_rq_unlock(rq, &flags);
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002646 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647}
2648
Avi Kivitye107be32007-07-26 13:40:43 +02002649#ifdef CONFIG_PREEMPT_NOTIFIERS
2650
2651/**
Luis Henriques80dd99b2009-03-16 19:58:09 +00002652 * preempt_notifier_register - tell me when current is being preempted & rescheduled
Randy Dunlap421cee22007-07-31 00:37:50 -07002653 * @notifier: notifier struct to register
Avi Kivitye107be32007-07-26 13:40:43 +02002654 */
2655void preempt_notifier_register(struct preempt_notifier *notifier)
2656{
2657 hlist_add_head(&notifier->link, &current->preempt_notifiers);
2658}
2659EXPORT_SYMBOL_GPL(preempt_notifier_register);
2660
2661/**
2662 * preempt_notifier_unregister - no longer interested in preemption notifications
Randy Dunlap421cee22007-07-31 00:37:50 -07002663 * @notifier: notifier struct to unregister
Avi Kivitye107be32007-07-26 13:40:43 +02002664 *
2665 * This is safe to call from within a preemption notifier.
2666 */
2667void preempt_notifier_unregister(struct preempt_notifier *notifier)
2668{
2669 hlist_del(&notifier->link);
2670}
2671EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
2672
2673static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2674{
2675 struct preempt_notifier *notifier;
2676 struct hlist_node *node;
2677
2678 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2679 notifier->ops->sched_in(notifier, raw_smp_processor_id());
2680}
2681
2682static void
2683fire_sched_out_preempt_notifiers(struct task_struct *curr,
2684 struct task_struct *next)
2685{
2686 struct preempt_notifier *notifier;
2687 struct hlist_node *node;
2688
2689 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2690 notifier->ops->sched_out(notifier, next);
2691}
2692
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02002693#else /* !CONFIG_PREEMPT_NOTIFIERS */
Avi Kivitye107be32007-07-26 13:40:43 +02002694
2695static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2696{
2697}
2698
2699static void
2700fire_sched_out_preempt_notifiers(struct task_struct *curr,
2701 struct task_struct *next)
2702{
2703}
2704
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02002705#endif /* CONFIG_PREEMPT_NOTIFIERS */
Avi Kivitye107be32007-07-26 13:40:43 +02002706
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707/**
Nick Piggin4866cde2005-06-25 14:57:23 -07002708 * prepare_task_switch - prepare to switch tasks
2709 * @rq: the runqueue preparing to switch
Randy Dunlap421cee22007-07-31 00:37:50 -07002710 * @prev: the current task that is being switched out
Nick Piggin4866cde2005-06-25 14:57:23 -07002711 * @next: the task we are going to switch to.
2712 *
2713 * This is called with the rq lock held and interrupts off. It must
2714 * be paired with a subsequent finish_task_switch after the context
2715 * switch.
2716 *
2717 * prepare_task_switch sets up locking and calls architecture specific
2718 * hooks.
2719 */
Avi Kivitye107be32007-07-26 13:40:43 +02002720static inline void
2721prepare_task_switch(struct rq *rq, struct task_struct *prev,
2722 struct task_struct *next)
Nick Piggin4866cde2005-06-25 14:57:23 -07002723{
Avi Kivitye107be32007-07-26 13:40:43 +02002724 fire_sched_out_preempt_notifiers(prev, next);
Nick Piggin4866cde2005-06-25 14:57:23 -07002725 prepare_lock_switch(rq, next);
2726 prepare_arch_switch(next);
2727}
2728
2729/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 * finish_task_switch - clean up after a task-switch
Jeff Garzik344baba2005-09-07 01:15:17 -04002731 * @rq: runqueue associated with task-switch
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 * @prev: the thread we just switched away from.
2733 *
Nick Piggin4866cde2005-06-25 14:57:23 -07002734 * finish_task_switch must be called after the context switch, paired
2735 * with a prepare_task_switch call before the context switch.
2736 * finish_task_switch will reconcile locking set up by prepare_task_switch,
2737 * and do any other architecture-specific cleanup actions.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 *
2739 * Note that we may have delayed dropping an mm in context_switch(). If
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01002740 * so, we finish that here outside of the runqueue lock. (Doing it
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741 * with the lock held can cause deadlocks; see schedule() for
2742 * details.)
2743 */
Alexey Dobriyana9957442007-10-15 17:00:13 +02002744static void finish_task_switch(struct rq *rq, struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745 __releases(rq->lock)
2746{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 struct mm_struct *mm = rq->prev_mm;
Oleg Nesterov55a101f2006-09-29 02:01:10 -07002748 long prev_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749
2750 rq->prev_mm = NULL;
2751
2752 /*
2753 * A task struct has one reference for the use as "current".
Oleg Nesterovc394cc92006-09-29 02:01:11 -07002754 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
Oleg Nesterov55a101f2006-09-29 02:01:10 -07002755 * schedule one last time. The schedule call will never return, and
2756 * the scheduled task must drop that reference.
Oleg Nesterovc394cc92006-09-29 02:01:11 -07002757 * The test for TASK_DEAD must occur while the runqueue locks are
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 * still held, otherwise prev could be scheduled on another cpu, die
2759 * there before we look at prev->state, and then the reference would
2760 * be dropped twice.
2761 * Manfred Spraul <manfred@colorfullife.com>
2762 */
Oleg Nesterov55a101f2006-09-29 02:01:10 -07002763 prev_state = prev->state;
Nick Piggin4866cde2005-06-25 14:57:23 -07002764 finish_arch_switch(prev);
Jamie Iles8381f652010-01-08 15:27:33 +00002765#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2766 local_irq_disable();
2767#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
Peter Zijlstra49f47432009-12-27 11:51:52 +01002768 perf_event_task_sched_in(current);
Jamie Iles8381f652010-01-08 15:27:33 +00002769#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2770 local_irq_enable();
2771#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
Nick Piggin4866cde2005-06-25 14:57:23 -07002772 finish_lock_switch(rq, prev);
Steven Rostedte8fa1362008-01-25 21:08:05 +01002773
Avi Kivitye107be32007-07-26 13:40:43 +02002774 fire_sched_in_preempt_notifiers(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 if (mm)
2776 mmdrop(mm);
Oleg Nesterovc394cc92006-09-29 02:01:11 -07002777 if (unlikely(prev_state == TASK_DEAD)) {
bibo maoc6fd91f2006-03-26 01:38:20 -08002778 /*
2779 * Remove function-return probe instances associated with this
2780 * task and put them back on the free list.
Ingo Molnar9761eea2007-07-09 18:52:00 +02002781 */
bibo maoc6fd91f2006-03-26 01:38:20 -08002782 kprobe_flush_task(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 put_task_struct(prev);
bibo maoc6fd91f2006-03-26 01:38:20 -08002784 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785}
2786
Gregory Haskins3f029d32009-07-29 11:08:47 -04002787#ifdef CONFIG_SMP
2788
2789/* assumes rq->lock is held */
2790static inline void pre_schedule(struct rq *rq, struct task_struct *prev)
2791{
2792 if (prev->sched_class->pre_schedule)
2793 prev->sched_class->pre_schedule(rq, prev);
2794}
2795
2796/* rq->lock is NOT held, but preemption is disabled */
2797static inline void post_schedule(struct rq *rq)
2798{
2799 if (rq->post_schedule) {
2800 unsigned long flags;
2801
Thomas Gleixner05fa7852009-11-17 14:28:38 +01002802 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins3f029d32009-07-29 11:08:47 -04002803 if (rq->curr->sched_class->post_schedule)
2804 rq->curr->sched_class->post_schedule(rq);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01002805 raw_spin_unlock_irqrestore(&rq->lock, flags);
Gregory Haskins3f029d32009-07-29 11:08:47 -04002806
2807 rq->post_schedule = 0;
2808 }
2809}
2810
2811#else
2812
2813static inline void pre_schedule(struct rq *rq, struct task_struct *p)
2814{
2815}
2816
2817static inline void post_schedule(struct rq *rq)
2818{
2819}
2820
2821#endif
2822
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823/**
2824 * schedule_tail - first thing a freshly forked thread must call.
2825 * @prev: the thread we just switched away from.
2826 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07002827asmlinkage void schedule_tail(struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 __releases(rq->lock)
2829{
Ingo Molnar70b97a72006-07-03 00:25:42 -07002830 struct rq *rq = this_rq();
2831
Nick Piggin4866cde2005-06-25 14:57:23 -07002832 finish_task_switch(rq, prev);
Steven Rostedtda19ab52009-07-29 00:21:22 -04002833
Gregory Haskins3f029d32009-07-29 11:08:47 -04002834 /*
2835 * FIXME: do we need to worry about rq being invalidated by the
2836 * task_switch?
2837 */
2838 post_schedule(rq);
Steven Rostedtda19ab52009-07-29 00:21:22 -04002839
Nick Piggin4866cde2005-06-25 14:57:23 -07002840#ifdef __ARCH_WANT_UNLOCKED_CTXSW
2841 /* In this case, finish_task_switch does not reenable preemption */
2842 preempt_enable();
2843#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 if (current->set_child_tid)
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002845 put_user(task_pid_vnr(current), current->set_child_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846}
2847
2848/*
2849 * context_switch - switch to the new MM and the new
2850 * thread's register state.
2851 */
Ingo Molnardd41f592007-07-09 18:51:59 +02002852static inline void
Ingo Molnar70b97a72006-07-03 00:25:42 -07002853context_switch(struct rq *rq, struct task_struct *prev,
Ingo Molnar36c8b582006-07-03 00:25:41 -07002854 struct task_struct *next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855{
Ingo Molnardd41f592007-07-09 18:51:59 +02002856 struct mm_struct *mm, *oldmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857
Avi Kivitye107be32007-07-26 13:40:43 +02002858 prepare_task_switch(rq, prev, next);
Peter Zijlstra27a9da62010-05-04 20:36:56 +02002859 trace_sched_switch(prev, next);
Ingo Molnardd41f592007-07-09 18:51:59 +02002860 mm = next->mm;
2861 oldmm = prev->active_mm;
Zachary Amsden9226d122007-02-13 13:26:21 +01002862 /*
2863 * For paravirt, this is coupled with an exit in switch_to to
2864 * combine the page table reload and the switch backend into
2865 * one hypercall.
2866 */
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -08002867 arch_start_context_switch(prev);
Zachary Amsden9226d122007-02-13 13:26:21 +01002868
Heiko Carstens31915ab2010-09-16 14:42:25 +02002869 if (!mm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 next->active_mm = oldmm;
2871 atomic_inc(&oldmm->mm_count);
2872 enter_lazy_tlb(oldmm, next);
2873 } else
2874 switch_mm(oldmm, mm, next);
2875
Heiko Carstens31915ab2010-09-16 14:42:25 +02002876 if (!prev->mm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 prev->active_mm = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 rq->prev_mm = oldmm;
2879 }
Ingo Molnar3a5f5e42006-07-14 00:24:27 -07002880 /*
2881 * Since the runqueue lock will be released by the next
2882 * task (which is an invalid locking op but in the case
2883 * of the scheduler it's an obvious special-case), so we
2884 * do an early lockdep release here:
2885 */
2886#ifndef __ARCH_WANT_UNLOCKED_CTXSW
Ingo Molnar8a25d5d2006-07-03 00:24:54 -07002887 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
Ingo Molnar3a5f5e42006-07-14 00:24:27 -07002888#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889
2890 /* Here we just switch the register state and the stack. */
2891 switch_to(prev, next, prev);
2892
Ingo Molnardd41f592007-07-09 18:51:59 +02002893 barrier();
2894 /*
2895 * this_rq must be evaluated again because prev may have moved
2896 * CPUs since it called schedule(), thus the 'rq' on its stack
2897 * frame will be invalid.
2898 */
2899 finish_task_switch(this_rq(), prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900}
2901
2902/*
2903 * nr_running, nr_uninterruptible and nr_context_switches:
2904 *
2905 * externally visible scheduler statistics: current number of runnable
2906 * threads, current number of uninterruptible-sleeping threads, total
2907 * number of context switches performed since bootup.
2908 */
2909unsigned long nr_running(void)
2910{
2911 unsigned long i, sum = 0;
2912
2913 for_each_online_cpu(i)
2914 sum += cpu_rq(i)->nr_running;
2915
2916 return sum;
2917}
2918
2919unsigned long nr_uninterruptible(void)
2920{
2921 unsigned long i, sum = 0;
2922
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002923 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 sum += cpu_rq(i)->nr_uninterruptible;
2925
2926 /*
2927 * Since we read the counters lockless, it might be slightly
2928 * inaccurate. Do not allow it to go below zero though:
2929 */
2930 if (unlikely((long)sum < 0))
2931 sum = 0;
2932
2933 return sum;
2934}
2935
2936unsigned long long nr_context_switches(void)
2937{
Steven Rostedtcc94abf2006-06-27 02:54:31 -07002938 int i;
2939 unsigned long long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002941 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942 sum += cpu_rq(i)->nr_switches;
2943
2944 return sum;
2945}
2946
2947unsigned long nr_iowait(void)
2948{
2949 unsigned long i, sum = 0;
2950
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002951 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952 sum += atomic_read(&cpu_rq(i)->nr_iowait);
2953
2954 return sum;
2955}
2956
Peter Zijlstra8c215bd2010-07-01 09:07:17 +02002957unsigned long nr_iowait_cpu(int cpu)
Arjan van de Ven69d25872009-09-21 17:04:08 -07002958{
Peter Zijlstra8c215bd2010-07-01 09:07:17 +02002959 struct rq *this = cpu_rq(cpu);
Arjan van de Ven69d25872009-09-21 17:04:08 -07002960 return atomic_read(&this->nr_iowait);
2961}
2962
2963unsigned long this_cpu_load(void)
2964{
2965 struct rq *this = this_rq();
2966 return this->cpu_load[0];
2967}
2968
2969
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002970/* Variables and functions for calc_load */
2971static atomic_long_t calc_load_tasks;
2972static unsigned long calc_load_update;
2973unsigned long avenrun[3];
2974EXPORT_SYMBOL(avenrun);
2975
Peter Zijlstra74f51872010-04-22 21:50:19 +02002976static long calc_load_fold_active(struct rq *this_rq)
2977{
2978 long nr_active, delta = 0;
2979
2980 nr_active = this_rq->nr_running;
2981 nr_active += (long) this_rq->nr_uninterruptible;
2982
2983 if (nr_active != this_rq->calc_load_active) {
2984 delta = nr_active - this_rq->calc_load_active;
2985 this_rq->calc_load_active = nr_active;
2986 }
2987
2988 return delta;
2989}
2990
2991#ifdef CONFIG_NO_HZ
2992/*
2993 * For NO_HZ we delay the active fold to the next LOAD_FREQ update.
2994 *
2995 * When making the ILB scale, we should try to pull this in as well.
2996 */
2997static atomic_long_t calc_load_tasks_idle;
2998
2999static void calc_load_account_idle(struct rq *this_rq)
3000{
3001 long delta;
3002
3003 delta = calc_load_fold_active(this_rq);
3004 if (delta)
3005 atomic_long_add(delta, &calc_load_tasks_idle);
3006}
3007
3008static long calc_load_fold_idle(void)
3009{
3010 long delta = 0;
3011
3012 /*
3013 * Its got a race, we don't care...
3014 */
3015 if (atomic_long_read(&calc_load_tasks_idle))
3016 delta = atomic_long_xchg(&calc_load_tasks_idle, 0);
3017
3018 return delta;
3019}
3020#else
3021static void calc_load_account_idle(struct rq *this_rq)
3022{
3023}
3024
3025static inline long calc_load_fold_idle(void)
3026{
3027 return 0;
3028}
3029#endif
3030
Thomas Gleixner2d024942009-05-02 20:08:52 +02003031/**
3032 * get_avenrun - get the load average array
3033 * @loads: pointer to dest load array
3034 * @offset: offset to add
3035 * @shift: shift count to shift the result left
3036 *
3037 * These values are estimates at best, so no need for locking.
3038 */
3039void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
3040{
3041 loads[0] = (avenrun[0] + offset) << shift;
3042 loads[1] = (avenrun[1] + offset) << shift;
3043 loads[2] = (avenrun[2] + offset) << shift;
3044}
3045
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003046static unsigned long
3047calc_load(unsigned long load, unsigned long exp, unsigned long active)
Jack Steinerdb1b1fe2006-03-31 02:31:21 -08003048{
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003049 load *= exp;
3050 load += active * (FIXED_1 - exp);
3051 return load >> FSHIFT;
3052}
Jack Steinerdb1b1fe2006-03-31 02:31:21 -08003053
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003054/*
3055 * calc_load - update the avenrun load estimates 10 ticks after the
3056 * CPUs have updated calc_load_tasks.
3057 */
3058void calc_global_load(void)
3059{
3060 unsigned long upd = calc_load_update + 10;
3061 long active;
3062
3063 if (time_before(jiffies, upd))
3064 return;
3065
3066 active = atomic_long_read(&calc_load_tasks);
3067 active = active > 0 ? active * FIXED_1 : 0;
3068
3069 avenrun[0] = calc_load(avenrun[0], EXP_1, active);
3070 avenrun[1] = calc_load(avenrun[1], EXP_5, active);
3071 avenrun[2] = calc_load(avenrun[2], EXP_15, active);
3072
3073 calc_load_update += LOAD_FREQ;
3074}
3075
3076/*
Peter Zijlstra74f51872010-04-22 21:50:19 +02003077 * Called from update_cpu_load() to periodically update this CPU's
3078 * active count.
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003079 */
3080static void calc_load_account_active(struct rq *this_rq)
3081{
Peter Zijlstra74f51872010-04-22 21:50:19 +02003082 long delta;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003083
Peter Zijlstra74f51872010-04-22 21:50:19 +02003084 if (time_before(jiffies, this_rq->calc_load_update))
3085 return;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003086
Peter Zijlstra74f51872010-04-22 21:50:19 +02003087 delta = calc_load_fold_active(this_rq);
3088 delta += calc_load_fold_idle();
3089 if (delta)
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003090 atomic_long_add(delta, &calc_load_tasks);
Peter Zijlstra74f51872010-04-22 21:50:19 +02003091
3092 this_rq->calc_load_update += LOAD_FREQ;
Jack Steinerdb1b1fe2006-03-31 02:31:21 -08003093}
3094
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095/*
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003096 * The exact cpuload at various idx values, calculated at every tick would be
3097 * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load
3098 *
3099 * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called
3100 * on nth tick when cpu may be busy, then we have:
3101 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3102 * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load
3103 *
3104 * decay_load_missed() below does efficient calculation of
3105 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3106 * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load
3107 *
3108 * The calculation is approximated on a 128 point scale.
3109 * degrade_zero_ticks is the number of ticks after which load at any
3110 * particular idx is approximated to be zero.
3111 * degrade_factor is a precomputed table, a row for each load idx.
3112 * Each column corresponds to degradation factor for a power of two ticks,
3113 * based on 128 point scale.
3114 * Example:
3115 * row 2, col 3 (=12) says that the degradation at load idx 2 after
3116 * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8).
3117 *
3118 * With this power of 2 load factors, we can degrade the load n times
3119 * by looking at 1 bits in n and doing as many mult/shift instead of
3120 * n mult/shifts needed by the exact degradation.
3121 */
3122#define DEGRADE_SHIFT 7
3123static const unsigned char
3124 degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128};
3125static const unsigned char
3126 degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = {
3127 {0, 0, 0, 0, 0, 0, 0, 0},
3128 {64, 32, 8, 0, 0, 0, 0, 0},
3129 {96, 72, 40, 12, 1, 0, 0},
3130 {112, 98, 75, 43, 15, 1, 0},
3131 {120, 112, 98, 76, 45, 16, 2} };
3132
3133/*
3134 * Update cpu_load for any missed ticks, due to tickless idle. The backlog
3135 * would be when CPU is idle and so we just decay the old load without
3136 * adding any new load.
3137 */
3138static unsigned long
3139decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
3140{
3141 int j = 0;
3142
3143 if (!missed_updates)
3144 return load;
3145
3146 if (missed_updates >= degrade_zero_ticks[idx])
3147 return 0;
3148
3149 if (idx == 1)
3150 return load >> missed_updates;
3151
3152 while (missed_updates) {
3153 if (missed_updates % 2)
3154 load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT;
3155
3156 missed_updates >>= 1;
3157 j++;
3158 }
3159 return load;
3160}
3161
3162/*
Ingo Molnardd41f592007-07-09 18:51:59 +02003163 * Update rq->cpu_load[] statistics. This function is usually called every
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003164 * scheduler tick (TICK_NSEC). With tickless idle this will not be called
3165 * every tick. We fix it up based on jiffies.
Ingo Molnar48f24c42006-07-03 00:25:40 -07003166 */
Ingo Molnardd41f592007-07-09 18:51:59 +02003167static void update_cpu_load(struct rq *this_rq)
Ingo Molnar48f24c42006-07-03 00:25:40 -07003168{
Dmitry Adamushko495eca42007-10-15 17:00:06 +02003169 unsigned long this_load = this_rq->load.weight;
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003170 unsigned long curr_jiffies = jiffies;
3171 unsigned long pending_updates;
Ingo Molnardd41f592007-07-09 18:51:59 +02003172 int i, scale;
3173
3174 this_rq->nr_load_updates++;
Ingo Molnardd41f592007-07-09 18:51:59 +02003175
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003176 /* Avoid repeated calls on same jiffy, when moving in and out of idle */
3177 if (curr_jiffies == this_rq->last_load_update_tick)
3178 return;
3179
3180 pending_updates = curr_jiffies - this_rq->last_load_update_tick;
3181 this_rq->last_load_update_tick = curr_jiffies;
3182
Ingo Molnardd41f592007-07-09 18:51:59 +02003183 /* Update our load: */
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003184 this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */
3185 for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
Ingo Molnardd41f592007-07-09 18:51:59 +02003186 unsigned long old_load, new_load;
3187
3188 /* scale is effectively 1 << i now, and >> i divides by scale */
3189
3190 old_load = this_rq->cpu_load[i];
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003191 old_load = decay_load_missed(old_load, pending_updates - 1, i);
Ingo Molnardd41f592007-07-09 18:51:59 +02003192 new_load = this_load;
Ingo Molnara25707f2007-10-15 17:00:03 +02003193 /*
3194 * Round up the averaging division if load is increasing. This
3195 * prevents us from getting stuck on 9 if the load is 10, for
3196 * example.
3197 */
3198 if (new_load > old_load)
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003199 new_load += scale - 1;
3200
3201 this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i;
Ingo Molnardd41f592007-07-09 18:51:59 +02003202 }
Suresh Siddhada2b71e2010-08-23 13:42:51 -07003203
3204 sched_avg_update(this_rq);
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003205}
3206
3207static void update_cpu_load_active(struct rq *this_rq)
3208{
3209 update_cpu_load(this_rq);
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003210
Peter Zijlstra74f51872010-04-22 21:50:19 +02003211 calc_load_account_active(this_rq);
Ingo Molnar48f24c42006-07-03 00:25:40 -07003212}
3213
Ingo Molnardd41f592007-07-09 18:51:59 +02003214#ifdef CONFIG_SMP
3215
Ingo Molnar48f24c42006-07-03 00:25:40 -07003216/*
Peter Zijlstra38022902009-12-16 18:04:37 +01003217 * sched_exec - execve() is a valuable balancing opportunity, because at
3218 * this point the task has the smallest effective memory and cache footprint.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003219 */
Peter Zijlstra38022902009-12-16 18:04:37 +01003220void sched_exec(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221{
Peter Zijlstra38022902009-12-16 18:04:37 +01003222 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07003224 struct rq *rq;
Peter Zijlstra0017d732010-03-24 18:34:10 +01003225 int dest_cpu;
Peter Zijlstra38022902009-12-16 18:04:37 +01003226
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227 rq = task_rq_lock(p, &flags);
Peter Zijlstra0017d732010-03-24 18:34:10 +01003228 dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0);
3229 if (dest_cpu == smp_processor_id())
3230 goto unlock;
Peter Zijlstra38022902009-12-16 18:04:37 +01003231
3232 /*
3233 * select_task_rq() can race against ->cpus_allowed
3234 */
Oleg Nesterov30da6882010-03-15 10:10:19 +01003235 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) &&
Nikanth Karthikesanb7a2b392010-11-26 12:37:09 +05303236 likely(cpu_active(dest_cpu)) && migrate_task(p, rq)) {
Tejun Heo969c7922010-05-06 18:49:21 +02003237 struct migration_arg arg = { p, dest_cpu };
Ingo Molnar36c8b582006-07-03 00:25:41 -07003238
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239 task_rq_unlock(rq, &flags);
Tejun Heo969c7922010-05-06 18:49:21 +02003240 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241 return;
3242 }
Peter Zijlstra0017d732010-03-24 18:34:10 +01003243unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244 task_rq_unlock(rq, &flags);
3245}
3246
Linus Torvalds1da177e2005-04-16 15:20:36 -07003247#endif
3248
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249DEFINE_PER_CPU(struct kernel_stat, kstat);
3250
3251EXPORT_PER_CPU_SYMBOL(kstat);
3252
3253/*
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003254 * Return any ns on the sched_clock that have not yet been accounted in
Frank Mayharf06febc2008-09-12 09:54:39 -07003255 * @p in case that task is currently running.
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003256 *
3257 * Called with task_rq_lock() held on @rq.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258 */
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003259static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq)
3260{
3261 u64 ns = 0;
3262
3263 if (task_current(rq, p)) {
3264 update_rq_clock(rq);
Venkatesh Pallipadi305e6832010-10-04 17:03:21 -07003265 ns = rq->clock_task - p->se.exec_start;
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003266 if ((s64)ns < 0)
3267 ns = 0;
3268 }
3269
3270 return ns;
3271}
3272
Frank Mayharbb34d922008-09-12 09:54:39 -07003273unsigned long long task_delta_exec(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275 unsigned long flags;
Ingo Molnar41b86e92007-07-09 18:51:58 +02003276 struct rq *rq;
Frank Mayharbb34d922008-09-12 09:54:39 -07003277 u64 ns = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07003278
Ingo Molnar41b86e92007-07-09 18:51:58 +02003279 rq = task_rq_lock(p, &flags);
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003280 ns = do_task_delta_exec(p, rq);
3281 task_rq_unlock(rq, &flags);
Ingo Molnar15084872008-09-30 08:28:17 +02003282
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003283 return ns;
3284}
Frank Mayharf06febc2008-09-12 09:54:39 -07003285
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003286/*
3287 * Return accounted runtime for the task.
3288 * In case the task is currently running, return the runtime plus current's
3289 * pending runtime that have not been accounted yet.
3290 */
3291unsigned long long task_sched_runtime(struct task_struct *p)
3292{
3293 unsigned long flags;
3294 struct rq *rq;
3295 u64 ns = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07003296
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003297 rq = task_rq_lock(p, &flags);
3298 ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq);
3299 task_rq_unlock(rq, &flags);
3300
3301 return ns;
3302}
3303
3304/*
3305 * Return sum_exec_runtime for the thread group.
3306 * In case the task is currently running, return the sum plus current's
3307 * pending runtime that have not been accounted yet.
3308 *
3309 * Note that the thread group might have other running tasks as well,
3310 * so the return value not includes other pending runtime that other
3311 * running tasks might have.
3312 */
3313unsigned long long thread_group_sched_runtime(struct task_struct *p)
3314{
3315 struct task_cputime totals;
3316 unsigned long flags;
3317 struct rq *rq;
3318 u64 ns;
3319
3320 rq = task_rq_lock(p, &flags);
3321 thread_group_cputime(p, &totals);
3322 ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323 task_rq_unlock(rq, &flags);
3324
3325 return ns;
3326}
3327
3328/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329 * Account user cpu time to a process.
3330 * @p: the process that the cpu time gets accounted to
Linus Torvalds1da177e2005-04-16 15:20:36 -07003331 * @cputime: the cpu time spent in user space since the last update
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003332 * @cputime_scaled: cputime scaled by cpu frequency
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333 */
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003334void account_user_time(struct task_struct *p, cputime_t cputime,
3335 cputime_t cputime_scaled)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003336{
3337 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3338 cputime64_t tmp;
3339
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003340 /* Add user time to process. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341 p->utime = cputime_add(p->utime, cputime);
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003342 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
Frank Mayharf06febc2008-09-12 09:54:39 -07003343 account_group_user_time(p, cputime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344
3345 /* Add user time to cpustat. */
3346 tmp = cputime_to_cputime64(cputime);
3347 if (TASK_NICE(p) > 0)
3348 cpustat->nice = cputime64_add(cpustat->nice, tmp);
3349 else
3350 cpustat->user = cputime64_add(cpustat->user, tmp);
Bharata B Raoef12fef2009-03-31 10:02:22 +05303351
3352 cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime);
Jonathan Lim49b5cf32008-07-25 01:48:40 -07003353 /* Account for user time used */
3354 acct_update_integrals(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355}
3356
3357/*
Laurent Vivier94886b82007-10-15 17:00:19 +02003358 * Account guest cpu time to a process.
3359 * @p: the process that the cpu time gets accounted to
3360 * @cputime: the cpu time spent in virtual machine since the last update
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003361 * @cputime_scaled: cputime scaled by cpu frequency
Laurent Vivier94886b82007-10-15 17:00:19 +02003362 */
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003363static void account_guest_time(struct task_struct *p, cputime_t cputime,
3364 cputime_t cputime_scaled)
Laurent Vivier94886b82007-10-15 17:00:19 +02003365{
3366 cputime64_t tmp;
3367 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3368
3369 tmp = cputime_to_cputime64(cputime);
3370
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003371 /* Add guest time to process. */
Laurent Vivier94886b82007-10-15 17:00:19 +02003372 p->utime = cputime_add(p->utime, cputime);
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003373 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
Frank Mayharf06febc2008-09-12 09:54:39 -07003374 account_group_user_time(p, cputime);
Laurent Vivier94886b82007-10-15 17:00:19 +02003375 p->gtime = cputime_add(p->gtime, cputime);
3376
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003377 /* Add guest time to cpustat. */
Ryota Ozakice0e7b22009-10-24 01:20:10 +09003378 if (TASK_NICE(p) > 0) {
3379 cpustat->nice = cputime64_add(cpustat->nice, tmp);
3380 cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp);
3381 } else {
3382 cpustat->user = cputime64_add(cpustat->user, tmp);
3383 cpustat->guest = cputime64_add(cpustat->guest, tmp);
3384 }
Laurent Vivier94886b82007-10-15 17:00:19 +02003385}
3386
3387/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388 * Account system cpu time to a process.
3389 * @p: the process that the cpu time gets accounted to
3390 * @hardirq_offset: the offset to subtract from hardirq_count()
3391 * @cputime: the cpu time spent in kernel space since the last update
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003392 * @cputime_scaled: cputime scaled by cpu frequency
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393 */
3394void account_system_time(struct task_struct *p, int hardirq_offset,
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003395 cputime_t cputime, cputime_t cputime_scaled)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396{
3397 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398 cputime64_t tmp;
3399
Harvey Harrison983ed7a2008-04-24 18:17:55 -07003400 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003401 account_guest_time(p, cputime, cputime_scaled);
Harvey Harrison983ed7a2008-04-24 18:17:55 -07003402 return;
3403 }
Laurent Vivier94886b82007-10-15 17:00:19 +02003404
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003405 /* Add system time to process. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406 p->stime = cputime_add(p->stime, cputime);
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003407 p->stimescaled = cputime_add(p->stimescaled, cputime_scaled);
Frank Mayharf06febc2008-09-12 09:54:39 -07003408 account_group_system_time(p, cputime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409
3410 /* Add system time to cpustat. */
3411 tmp = cputime_to_cputime64(cputime);
3412 if (hardirq_count() - hardirq_offset)
3413 cpustat->irq = cputime64_add(cpustat->irq, tmp);
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -07003414 else if (in_serving_softirq())
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415 cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003416 else
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003417 cpustat->system = cputime64_add(cpustat->system, tmp);
3418
Bharata B Raoef12fef2009-03-31 10:02:22 +05303419 cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime);
3420
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421 /* Account for system time used */
3422 acct_update_integrals(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423}
3424
3425/*
3426 * Account for involuntary wait time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427 * @steal: the cpu time spent in involuntary wait
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428 */
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003429void account_steal_time(cputime_t cputime)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003432 cputime64_t cputime64 = cputime_to_cputime64(cputime);
3433
3434 cpustat->steal = cputime64_add(cpustat->steal, cputime64);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435}
3436
Christoph Lameter7835b982006-12-10 02:20:22 -08003437/*
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003438 * Account for idle time.
3439 * @cputime: the cpu time spent in idle wait
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440 */
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003441void account_idle_time(cputime_t cputime)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442{
3443 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003444 cputime64_t cputime64 = cputime_to_cputime64(cputime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445 struct rq *rq = this_rq();
3446
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003447 if (atomic_read(&rq->nr_iowait) > 0)
3448 cpustat->iowait = cputime64_add(cpustat->iowait, cputime64);
3449 else
3450 cpustat->idle = cputime64_add(cpustat->idle, cputime64);
Christoph Lameter7835b982006-12-10 02:20:22 -08003451}
3452
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003453#ifndef CONFIG_VIRT_CPU_ACCOUNTING
3454
3455/*
3456 * Account a single tick of cpu time.
3457 * @p: the process that the cpu time gets accounted to
3458 * @user_tick: indicates if the tick is a user or a system tick
3459 */
3460void account_process_tick(struct task_struct *p, int user_tick)
3461{
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02003462 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003463 struct rq *rq = this_rq();
3464
3465 if (user_tick)
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02003466 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
Eric Dumazetf5f293a2009-04-29 14:44:49 +02003467 else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02003468 account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003469 one_jiffy_scaled);
3470 else
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02003471 account_idle_time(cputime_one_jiffy);
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003472}
3473
3474/*
3475 * Account multiple ticks of steal time.
3476 * @p: the process from which the cpu time has been stolen
3477 * @ticks: number of stolen ticks
3478 */
3479void account_steal_ticks(unsigned long ticks)
3480{
3481 account_steal_time(jiffies_to_cputime(ticks));
3482}
3483
3484/*
3485 * Account multiple ticks of idle time.
3486 * @ticks: number of stolen ticks
3487 */
3488void account_idle_ticks(unsigned long ticks)
3489{
3490 account_idle_time(jiffies_to_cputime(ticks));
3491}
3492
3493#endif
3494
Christoph Lameter7835b982006-12-10 02:20:22 -08003495/*
Balbir Singh49048622008-09-05 18:12:23 +02003496 * Use precise platform statistics if available:
3497 */
3498#ifdef CONFIG_VIRT_CPU_ACCOUNTING
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003499void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
Balbir Singh49048622008-09-05 18:12:23 +02003500{
Hidetoshi Setod99ca3b2009-12-02 17:26:47 +09003501 *ut = p->utime;
3502 *st = p->stime;
Balbir Singh49048622008-09-05 18:12:23 +02003503}
3504
Hidetoshi Seto0cf55e12009-12-02 17:28:07 +09003505void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
Balbir Singh49048622008-09-05 18:12:23 +02003506{
Hidetoshi Seto0cf55e12009-12-02 17:28:07 +09003507 struct task_cputime cputime;
3508
3509 thread_group_cputime(p, &cputime);
3510
3511 *ut = cputime.utime;
3512 *st = cputime.stime;
Balbir Singh49048622008-09-05 18:12:23 +02003513}
3514#else
Hidetoshi Seto761b1d22009-11-12 13:33:45 +09003515
3516#ifndef nsecs_to_cputime
Hidetoshi Setob7b20df2009-11-26 14:49:27 +09003517# define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs)
Hidetoshi Seto761b1d22009-11-12 13:33:45 +09003518#endif
3519
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003520void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
Balbir Singh49048622008-09-05 18:12:23 +02003521{
Hidetoshi Setod99ca3b2009-12-02 17:26:47 +09003522 cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime);
Balbir Singh49048622008-09-05 18:12:23 +02003523
3524 /*
3525 * Use CFS's precise accounting:
3526 */
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003527 rtime = nsecs_to_cputime(p->se.sum_exec_runtime);
Balbir Singh49048622008-09-05 18:12:23 +02003528
3529 if (total) {
Stanislaw Gruszkae75e8632010-09-14 16:35:14 +02003530 u64 temp = rtime;
Balbir Singh49048622008-09-05 18:12:23 +02003531
Stanislaw Gruszkae75e8632010-09-14 16:35:14 +02003532 temp *= utime;
Balbir Singh49048622008-09-05 18:12:23 +02003533 do_div(temp, total);
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003534 utime = (cputime_t)temp;
3535 } else
3536 utime = rtime;
Balbir Singh49048622008-09-05 18:12:23 +02003537
3538 /*
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003539 * Compare with previous values, to keep monotonicity:
Balbir Singh49048622008-09-05 18:12:23 +02003540 */
Hidetoshi Seto761b1d22009-11-12 13:33:45 +09003541 p->prev_utime = max(p->prev_utime, utime);
Hidetoshi Setod99ca3b2009-12-02 17:26:47 +09003542 p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime));
Balbir Singh49048622008-09-05 18:12:23 +02003543
Hidetoshi Setod99ca3b2009-12-02 17:26:47 +09003544 *ut = p->prev_utime;
3545 *st = p->prev_stime;
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003546}
Balbir Singh49048622008-09-05 18:12:23 +02003547
Hidetoshi Seto0cf55e12009-12-02 17:28:07 +09003548/*
3549 * Must be called with siglock held.
3550 */
3551void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3552{
3553 struct signal_struct *sig = p->signal;
3554 struct task_cputime cputime;
3555 cputime_t rtime, utime, total;
3556
3557 thread_group_cputime(p, &cputime);
3558
3559 total = cputime_add(cputime.utime, cputime.stime);
3560 rtime = nsecs_to_cputime(cputime.sum_exec_runtime);
3561
3562 if (total) {
Stanislaw Gruszkae75e8632010-09-14 16:35:14 +02003563 u64 temp = rtime;
Hidetoshi Seto0cf55e12009-12-02 17:28:07 +09003564
Stanislaw Gruszkae75e8632010-09-14 16:35:14 +02003565 temp *= cputime.utime;
Hidetoshi Seto0cf55e12009-12-02 17:28:07 +09003566 do_div(temp, total);
3567 utime = (cputime_t)temp;
3568 } else
3569 utime = rtime;
3570
3571 sig->prev_utime = max(sig->prev_utime, utime);
3572 sig->prev_stime = max(sig->prev_stime,
3573 cputime_sub(rtime, sig->prev_utime));
3574
3575 *ut = sig->prev_utime;
3576 *st = sig->prev_stime;
Balbir Singh49048622008-09-05 18:12:23 +02003577}
3578#endif
3579
Balbir Singh49048622008-09-05 18:12:23 +02003580/*
Christoph Lameter7835b982006-12-10 02:20:22 -08003581 * This function gets called by the timer code, with HZ frequency.
3582 * We call it with interrupts disabled.
3583 *
3584 * It also gets called by the fork code, when changing the parent's
3585 * timeslices.
3586 */
3587void scheduler_tick(void)
3588{
Christoph Lameter7835b982006-12-10 02:20:22 -08003589 int cpu = smp_processor_id();
3590 struct rq *rq = cpu_rq(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02003591 struct task_struct *curr = rq->curr;
Peter Zijlstra3e51f332008-05-03 18:29:28 +02003592
3593 sched_clock_tick();
Christoph Lameter7835b982006-12-10 02:20:22 -08003594
Thomas Gleixner05fa7852009-11-17 14:28:38 +01003595 raw_spin_lock(&rq->lock);
Peter Zijlstra3e51f332008-05-03 18:29:28 +02003596 update_rq_clock(rq);
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003597 update_cpu_load_active(rq);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01003598 curr->sched_class->task_tick(rq, curr, 0);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01003599 raw_spin_unlock(&rq->lock);
Ingo Molnardd41f592007-07-09 18:51:59 +02003600
Peter Zijlstrae9d2b062010-09-17 11:28:50 +02003601 perf_event_task_tick();
Peter Zijlstrae220d2d2009-05-23 18:28:55 +02003602
Christoph Lametere418e1c2006-12-10 02:20:23 -08003603#ifdef CONFIG_SMP
Ingo Molnardd41f592007-07-09 18:51:59 +02003604 rq->idle_at_tick = idle_cpu(cpu);
3605 trigger_load_balance(rq, cpu);
Christoph Lametere418e1c2006-12-10 02:20:23 -08003606#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003607}
3608
Lai Jiangshan132380a2009-04-02 14:18:25 +08003609notrace unsigned long get_parent_ip(unsigned long addr)
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003610{
3611 if (in_lock_functions(addr)) {
3612 addr = CALLER_ADDR2;
3613 if (in_lock_functions(addr))
3614 addr = CALLER_ADDR3;
3615 }
3616 return addr;
3617}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003618
Steven Rostedt7e49fcc2009-01-22 19:01:40 -05003619#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
3620 defined(CONFIG_PREEMPT_TRACER))
3621
Srinivasa Ds43627582008-02-23 15:24:04 -08003622void __kprobes add_preempt_count(int val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623{
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003624#ifdef CONFIG_DEBUG_PREEMPT
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625 /*
3626 * Underflow?
3627 */
Ingo Molnar9a11b492006-07-03 00:24:33 -07003628 if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
3629 return;
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003630#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631 preempt_count() += val;
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003632#ifdef CONFIG_DEBUG_PREEMPT
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633 /*
3634 * Spinlock count overflowing soon?
3635 */
Miguel Ojeda Sandonis33859f72006-12-10 02:20:38 -08003636 DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
3637 PREEMPT_MASK - 10);
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003638#endif
3639 if (preempt_count() == val)
3640 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641}
3642EXPORT_SYMBOL(add_preempt_count);
3643
Srinivasa Ds43627582008-02-23 15:24:04 -08003644void __kprobes sub_preempt_count(int val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645{
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003646#ifdef CONFIG_DEBUG_PREEMPT
Linus Torvalds1da177e2005-04-16 15:20:36 -07003647 /*
3648 * Underflow?
3649 */
Ingo Molnar01e3eb82009-01-12 13:00:50 +01003650 if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
Ingo Molnar9a11b492006-07-03 00:24:33 -07003651 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652 /*
3653 * Is the spinlock portion underflowing?
3654 */
Ingo Molnar9a11b492006-07-03 00:24:33 -07003655 if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
3656 !(preempt_count() & PREEMPT_MASK)))
3657 return;
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003658#endif
Ingo Molnar9a11b492006-07-03 00:24:33 -07003659
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003660 if (preempt_count() == val)
3661 trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003662 preempt_count() -= val;
3663}
3664EXPORT_SYMBOL(sub_preempt_count);
3665
3666#endif
3667
3668/*
Ingo Molnardd41f592007-07-09 18:51:59 +02003669 * Print scheduling while atomic bug:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670 */
Ingo Molnardd41f592007-07-09 18:51:59 +02003671static noinline void __schedule_bug(struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672{
Satyam Sharma838225b2007-10-24 18:23:50 +02003673 struct pt_regs *regs = get_irq_regs();
3674
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01003675 printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
3676 prev->comm, prev->pid, preempt_count());
Satyam Sharma838225b2007-10-24 18:23:50 +02003677
Ingo Molnardd41f592007-07-09 18:51:59 +02003678 debug_show_held_locks(prev);
Arjan van de Vene21f5b12008-05-23 09:05:58 -07003679 print_modules();
Ingo Molnardd41f592007-07-09 18:51:59 +02003680 if (irqs_disabled())
3681 print_irqtrace_events(prev);
Satyam Sharma838225b2007-10-24 18:23:50 +02003682
3683 if (regs)
3684 show_regs(regs);
3685 else
3686 dump_stack();
Ingo Molnardd41f592007-07-09 18:51:59 +02003687}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688
Ingo Molnardd41f592007-07-09 18:51:59 +02003689/*
3690 * Various schedule()-time debugging checks and statistics:
3691 */
3692static inline void schedule_debug(struct task_struct *prev)
3693{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694 /*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003695 * Test if we are atomic. Since do_exit() needs to call into
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696 * schedule() atomically, we ignore that path for now.
3697 * Otherwise, whine if we are scheduling when we should not be.
3698 */
Roel Kluin3f33a7c2008-05-13 23:44:11 +02003699 if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
Ingo Molnardd41f592007-07-09 18:51:59 +02003700 __schedule_bug(prev);
3701
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702 profile_hit(SCHED_PROFILING, __builtin_return_address(0));
3703
Ingo Molnar2d723762007-10-15 17:00:12 +02003704 schedstat_inc(this_rq(), sched_count);
Ingo Molnarb8efb562007-10-15 17:00:10 +02003705#ifdef CONFIG_SCHEDSTATS
3706 if (unlikely(prev->lock_depth >= 0)) {
Ingo Molnar2d723762007-10-15 17:00:12 +02003707 schedstat_inc(this_rq(), bkl_count);
3708 schedstat_inc(prev, sched_info.bkl_count);
Ingo Molnarb8efb562007-10-15 17:00:10 +02003709 }
3710#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02003711}
3712
Peter Zijlstra6cecd082009-11-30 13:00:37 +01003713static void put_prev_task(struct rq *rq, struct task_struct *prev)
Mike Galbraithdf1c99d2009-03-10 19:08:11 +01003714{
Mike Galbraitha64692a2010-03-11 17:16:20 +01003715 if (prev->se.on_rq)
3716 update_rq_clock(rq);
3717 rq->skip_clock_update = 0;
Peter Zijlstra6cecd082009-11-30 13:00:37 +01003718 prev->sched_class->put_prev_task(rq, prev);
Mike Galbraithdf1c99d2009-03-10 19:08:11 +01003719}
3720
Ingo Molnardd41f592007-07-09 18:51:59 +02003721/*
3722 * Pick up the highest-prio task:
3723 */
3724static inline struct task_struct *
Wang Chenb67802e2009-03-02 13:55:26 +08003725pick_next_task(struct rq *rq)
Ingo Molnardd41f592007-07-09 18:51:59 +02003726{
Ingo Molnar5522d5d2007-10-15 17:00:12 +02003727 const struct sched_class *class;
Ingo Molnardd41f592007-07-09 18:51:59 +02003728 struct task_struct *p;
3729
3730 /*
3731 * Optimization: we know that if all tasks are in
3732 * the fair class we can call that function directly:
3733 */
3734 if (likely(rq->nr_running == rq->cfs.nr_running)) {
Ingo Molnarfb8d4722007-08-09 11:16:48 +02003735 p = fair_sched_class.pick_next_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02003736 if (likely(p))
3737 return p;
3738 }
3739
Peter Zijlstra34f971f2010-09-22 13:53:15 +02003740 for_each_class(class) {
Ingo Molnarfb8d4722007-08-09 11:16:48 +02003741 p = class->pick_next_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02003742 if (p)
3743 return p;
Ingo Molnardd41f592007-07-09 18:51:59 +02003744 }
Peter Zijlstra34f971f2010-09-22 13:53:15 +02003745
3746 BUG(); /* the idle class will always have a runnable task */
Ingo Molnardd41f592007-07-09 18:51:59 +02003747}
3748
3749/*
3750 * schedule() is the main scheduler function.
3751 */
Peter Zijlstraff743342009-03-13 12:21:26 +01003752asmlinkage void __sched schedule(void)
Ingo Molnardd41f592007-07-09 18:51:59 +02003753{
3754 struct task_struct *prev, *next;
Harvey Harrison67ca7bd2008-02-15 09:56:36 -08003755 unsigned long *switch_count;
Ingo Molnardd41f592007-07-09 18:51:59 +02003756 struct rq *rq;
Peter Zijlstra31656512008-07-18 18:01:23 +02003757 int cpu;
Ingo Molnardd41f592007-07-09 18:51:59 +02003758
Peter Zijlstraff743342009-03-13 12:21:26 +01003759need_resched:
3760 preempt_disable();
Ingo Molnardd41f592007-07-09 18:51:59 +02003761 cpu = smp_processor_id();
3762 rq = cpu_rq(cpu);
Paul E. McKenney25502a62010-04-01 17:37:01 -07003763 rcu_note_context_switch(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02003764 prev = rq->curr;
Ingo Molnardd41f592007-07-09 18:51:59 +02003765
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766 release_kernel_lock(prev);
3767need_resched_nonpreemptible:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003768
Ingo Molnardd41f592007-07-09 18:51:59 +02003769 schedule_debug(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770
Peter Zijlstra31656512008-07-18 18:01:23 +02003771 if (sched_feat(HRTICK))
Mike Galbraithf333fdc2008-05-12 21:20:55 +02003772 hrtick_clear(rq);
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003773
Thomas Gleixner05fa7852009-11-17 14:28:38 +01003774 raw_spin_lock_irq(&rq->lock);
Ingo Molnar1e819952007-10-15 17:00:13 +02003775 clear_tsk_need_resched(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003776
Oleg Nesterov246d86b2010-05-19 14:57:11 +02003777 switch_count = &prev->nivcsw;
Ingo Molnardd41f592007-07-09 18:51:59 +02003778 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
Tejun Heo21aa9af2010-06-08 21:40:37 +02003779 if (unlikely(signal_pending_state(prev->state, prev))) {
Ingo Molnardd41f592007-07-09 18:51:59 +02003780 prev->state = TASK_RUNNING;
Tejun Heo21aa9af2010-06-08 21:40:37 +02003781 } else {
3782 /*
3783 * If a worker is going to sleep, notify and
3784 * ask workqueue whether it wants to wake up a
3785 * task to maintain concurrency. If so, wake
3786 * up the task.
3787 */
3788 if (prev->flags & PF_WQ_WORKER) {
3789 struct task_struct *to_wakeup;
3790
3791 to_wakeup = wq_worker_sleeping(prev, cpu);
3792 if (to_wakeup)
3793 try_to_wake_up_local(to_wakeup);
3794 }
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01003795 deactivate_task(rq, prev, DEQUEUE_SLEEP);
Tejun Heo21aa9af2010-06-08 21:40:37 +02003796 }
Ingo Molnardd41f592007-07-09 18:51:59 +02003797 switch_count = &prev->nvcsw;
3798 }
3799
Gregory Haskins3f029d32009-07-29 11:08:47 -04003800 pre_schedule(rq, prev);
Steven Rostedtf65eda42008-01-25 21:08:07 +01003801
Ingo Molnardd41f592007-07-09 18:51:59 +02003802 if (unlikely(!rq->nr_running))
3803 idle_balance(cpu, rq);
3804
Mike Galbraithdf1c99d2009-03-10 19:08:11 +01003805 put_prev_task(rq, prev);
Wang Chenb67802e2009-03-02 13:55:26 +08003806 next = pick_next_task(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808 if (likely(prev != next)) {
David Simner673a90a2008-04-29 10:08:59 +01003809 sched_info_switch(prev, next);
Peter Zijlstra49f47432009-12-27 11:51:52 +01003810 perf_event_task_sched_out(prev, next);
David Simner673a90a2008-04-29 10:08:59 +01003811
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812 rq->nr_switches++;
3813 rq->curr = next;
3814 ++*switch_count;
3815
Ingo Molnardd41f592007-07-09 18:51:59 +02003816 context_switch(rq, prev, next); /* unlocks the rq */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003817 /*
Oleg Nesterov246d86b2010-05-19 14:57:11 +02003818 * The context switch have flipped the stack from under us
3819 * and restored the local variables which were saved when
3820 * this task called schedule() in the past. prev == current
3821 * is still correct, but it can be moved to another cpu/rq.
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003822 */
3823 cpu = smp_processor_id();
3824 rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825 } else
Thomas Gleixner05fa7852009-11-17 14:28:38 +01003826 raw_spin_unlock_irq(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827
Gregory Haskins3f029d32009-07-29 11:08:47 -04003828 post_schedule(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003829
Oleg Nesterov246d86b2010-05-19 14:57:11 +02003830 if (unlikely(reacquire_kernel_lock(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003831 goto need_resched_nonpreemptible;
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003832
Linus Torvalds1da177e2005-04-16 15:20:36 -07003833 preempt_enable_no_resched();
Peter Zijlstraff743342009-03-13 12:21:26 +01003834 if (need_resched())
Linus Torvalds1da177e2005-04-16 15:20:36 -07003835 goto need_resched;
3836}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837EXPORT_SYMBOL(schedule);
3838
Frederic Weisbeckerc08f7822009-12-02 20:49:17 +01003839#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003840/*
3841 * Look out! "owner" is an entirely speculative pointer
3842 * access and not reliable.
3843 */
3844int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner)
3845{
3846 unsigned int cpu;
3847 struct rq *rq;
3848
3849 if (!sched_feat(OWNER_SPIN))
3850 return 0;
3851
3852#ifdef CONFIG_DEBUG_PAGEALLOC
3853 /*
3854 * Need to access the cpu field knowing that
3855 * DEBUG_PAGEALLOC could have unmapped it if
3856 * the mutex owner just released it and exited.
3857 */
3858 if (probe_kernel_address(&owner->cpu, cpu))
Benjamin Herrenschmidt4b402212010-04-16 23:20:00 +02003859 return 0;
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003860#else
3861 cpu = owner->cpu;
3862#endif
3863
3864 /*
3865 * Even if the access succeeded (likely case),
3866 * the cpu field may no longer be valid.
3867 */
3868 if (cpu >= nr_cpumask_bits)
Benjamin Herrenschmidt4b402212010-04-16 23:20:00 +02003869 return 0;
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003870
3871 /*
3872 * We need to validate that we can do a
3873 * get_cpu() and that we have the percpu area.
3874 */
3875 if (!cpu_online(cpu))
Benjamin Herrenschmidt4b402212010-04-16 23:20:00 +02003876 return 0;
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003877
3878 rq = cpu_rq(cpu);
3879
3880 for (;;) {
3881 /*
3882 * Owner changed, break to re-assess state.
3883 */
Tim Chen9d0f4dc2010-08-18 15:00:27 -07003884 if (lock->owner != owner) {
3885 /*
3886 * If the lock has switched to a different owner,
3887 * we likely have heavy contention. Return 0 to quit
3888 * optimistic spinning and not contend further:
3889 */
3890 if (lock->owner)
3891 return 0;
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003892 break;
Tim Chen9d0f4dc2010-08-18 15:00:27 -07003893 }
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003894
3895 /*
3896 * Is that owner really running on that cpu?
3897 */
3898 if (task_thread_info(rq->curr) != owner || need_resched())
3899 return 0;
3900
Gerald Schaefer335d7af2010-11-22 15:47:36 +01003901 arch_mutex_cpu_relax();
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003902 }
Benjamin Herrenschmidt4b402212010-04-16 23:20:00 +02003903
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003904 return 1;
3905}
3906#endif
3907
Linus Torvalds1da177e2005-04-16 15:20:36 -07003908#ifdef CONFIG_PREEMPT
3909/*
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003910 * this is the entry point to schedule() from in-kernel preemption
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003911 * off of preempt_enable. Kernel preemptions off return from interrupt
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912 * occur there and call schedule directly.
3913 */
Steven Rostedtd1f74e22010-06-02 21:52:29 -04003914asmlinkage void __sched notrace preempt_schedule(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003915{
3916 struct thread_info *ti = current_thread_info();
Ingo Molnar6478d882008-01-25 21:08:33 +01003917
Linus Torvalds1da177e2005-04-16 15:20:36 -07003918 /*
3919 * If there is a non-zero preempt_count or interrupts are disabled,
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003920 * we do not want to preempt the current task. Just return..
Linus Torvalds1da177e2005-04-16 15:20:36 -07003921 */
Nick Pigginbeed33a2006-10-11 01:21:52 -07003922 if (likely(ti->preempt_count || irqs_disabled()))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923 return;
3924
Andi Kleen3a5c3592007-10-15 17:00:14 +02003925 do {
Steven Rostedtd1f74e22010-06-02 21:52:29 -04003926 add_preempt_count_notrace(PREEMPT_ACTIVE);
Andi Kleen3a5c3592007-10-15 17:00:14 +02003927 schedule();
Steven Rostedtd1f74e22010-06-02 21:52:29 -04003928 sub_preempt_count_notrace(PREEMPT_ACTIVE);
Andi Kleen3a5c3592007-10-15 17:00:14 +02003929
3930 /*
3931 * Check again in case we missed a preemption opportunity
3932 * between schedule and now.
3933 */
3934 barrier();
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08003935 } while (need_resched());
Linus Torvalds1da177e2005-04-16 15:20:36 -07003936}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003937EXPORT_SYMBOL(preempt_schedule);
3938
3939/*
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003940 * this is the entry point to schedule() from kernel preemption
Linus Torvalds1da177e2005-04-16 15:20:36 -07003941 * off of irq context.
3942 * Note, that this is called and return with irqs disabled. This will
3943 * protect us against recursive calling from irq.
3944 */
3945asmlinkage void __sched preempt_schedule_irq(void)
3946{
3947 struct thread_info *ti = current_thread_info();
Ingo Molnar6478d882008-01-25 21:08:33 +01003948
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003949 /* Catch callers which need to be fixed */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950 BUG_ON(ti->preempt_count || !irqs_disabled());
3951
Andi Kleen3a5c3592007-10-15 17:00:14 +02003952 do {
3953 add_preempt_count(PREEMPT_ACTIVE);
Andi Kleen3a5c3592007-10-15 17:00:14 +02003954 local_irq_enable();
3955 schedule();
3956 local_irq_disable();
Andi Kleen3a5c3592007-10-15 17:00:14 +02003957 sub_preempt_count(PREEMPT_ACTIVE);
3958
3959 /*
3960 * Check again in case we missed a preemption opportunity
3961 * between schedule and now.
3962 */
3963 barrier();
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08003964 } while (need_resched());
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965}
3966
3967#endif /* CONFIG_PREEMPT */
3968
Peter Zijlstra63859d42009-09-15 19:14:42 +02003969int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07003970 void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003971{
Peter Zijlstra63859d42009-09-15 19:14:42 +02003972 return try_to_wake_up(curr->private, mode, wake_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003973}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974EXPORT_SYMBOL(default_wake_function);
3975
3976/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003977 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
3978 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979 * number) then we wake all the non-exclusive tasks and one exclusive task.
3980 *
3981 * There are circumstances in which we can try to wake a task which has already
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003982 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983 * zero in this (rare) case, and we handle it by continuing to scan the queue.
3984 */
Johannes Weiner78ddb082009-04-14 16:53:05 +02003985static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
Peter Zijlstra63859d42009-09-15 19:14:42 +02003986 int nr_exclusive, int wake_flags, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987{
Matthias Kaehlcke2e458742007-10-15 17:00:02 +02003988 wait_queue_t *curr, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989
Matthias Kaehlcke2e458742007-10-15 17:00:02 +02003990 list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
Ingo Molnar48f24c42006-07-03 00:25:40 -07003991 unsigned flags = curr->flags;
3992
Peter Zijlstra63859d42009-09-15 19:14:42 +02003993 if (curr->func(curr, mode, wake_flags, key) &&
Ingo Molnar48f24c42006-07-03 00:25:40 -07003994 (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 break;
3996 }
3997}
3998
3999/**
4000 * __wake_up - wake up threads blocked on a waitqueue.
4001 * @q: the waitqueue
4002 * @mode: which threads
4003 * @nr_exclusive: how many wake-one or wake-many threads to wake up
Martin Waitz67be2dd2005-05-01 08:59:26 -07004004 * @key: is directly passed to the wakeup function
David Howells50fa6102009-04-28 15:01:38 +01004005 *
4006 * It may be assumed that this function implies a write memory barrier before
4007 * changing the task state if and only if any tasks are woken up.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004008 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08004009void __wake_up(wait_queue_head_t *q, unsigned int mode,
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004010 int nr_exclusive, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004011{
4012 unsigned long flags;
4013
4014 spin_lock_irqsave(&q->lock, flags);
4015 __wake_up_common(q, mode, nr_exclusive, 0, key);
4016 spin_unlock_irqrestore(&q->lock, flags);
4017}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018EXPORT_SYMBOL(__wake_up);
4019
4020/*
4021 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
4022 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08004023void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004024{
4025 __wake_up_common(q, mode, 1, 0, NULL);
4026}
Michal Nazarewicz22c43c82010-05-05 12:53:11 +02004027EXPORT_SYMBOL_GPL(__wake_up_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028
Davide Libenzi4ede8162009-03-31 15:24:20 -07004029void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
4030{
4031 __wake_up_common(q, mode, 1, 0, key);
4032}
4033
Linus Torvalds1da177e2005-04-16 15:20:36 -07004034/**
Davide Libenzi4ede8162009-03-31 15:24:20 -07004035 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004036 * @q: the waitqueue
4037 * @mode: which threads
4038 * @nr_exclusive: how many wake-one or wake-many threads to wake up
Davide Libenzi4ede8162009-03-31 15:24:20 -07004039 * @key: opaque value to be passed to wakeup targets
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040 *
4041 * The sync wakeup differs that the waker knows that it will schedule
4042 * away soon, so while the target thread will be woken up, it will not
4043 * be migrated to another CPU - ie. the two threads are 'synchronized'
4044 * with each other. This can prevent needless bouncing between CPUs.
4045 *
4046 * On UP it can prevent extra preemption.
David Howells50fa6102009-04-28 15:01:38 +01004047 *
4048 * It may be assumed that this function implies a write memory barrier before
4049 * changing the task state if and only if any tasks are woken up.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004050 */
Davide Libenzi4ede8162009-03-31 15:24:20 -07004051void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
4052 int nr_exclusive, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004053{
4054 unsigned long flags;
Peter Zijlstra7d478722009-09-14 19:55:44 +02004055 int wake_flags = WF_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004056
4057 if (unlikely(!q))
4058 return;
4059
4060 if (unlikely(!nr_exclusive))
Peter Zijlstra7d478722009-09-14 19:55:44 +02004061 wake_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004062
4063 spin_lock_irqsave(&q->lock, flags);
Peter Zijlstra7d478722009-09-14 19:55:44 +02004064 __wake_up_common(q, mode, nr_exclusive, wake_flags, key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004065 spin_unlock_irqrestore(&q->lock, flags);
4066}
Davide Libenzi4ede8162009-03-31 15:24:20 -07004067EXPORT_SYMBOL_GPL(__wake_up_sync_key);
4068
4069/*
4070 * __wake_up_sync - see __wake_up_sync_key()
4071 */
4072void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
4073{
4074 __wake_up_sync_key(q, mode, nr_exclusive, NULL);
4075}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004076EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
4077
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004078/**
4079 * complete: - signals a single thread waiting on this completion
4080 * @x: holds the state of this particular completion
4081 *
4082 * This will wake up a single thread waiting on this completion. Threads will be
4083 * awakened in the same order in which they were queued.
4084 *
4085 * See also complete_all(), wait_for_completion() and related routines.
David Howells50fa6102009-04-28 15:01:38 +01004086 *
4087 * It may be assumed that this function implies a write memory barrier before
4088 * changing the task state if and only if any tasks are woken up.
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004089 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02004090void complete(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091{
4092 unsigned long flags;
4093
4094 spin_lock_irqsave(&x->wait.lock, flags);
4095 x->done++;
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05004096 __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004097 spin_unlock_irqrestore(&x->wait.lock, flags);
4098}
4099EXPORT_SYMBOL(complete);
4100
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004101/**
4102 * complete_all: - signals all threads waiting on this completion
4103 * @x: holds the state of this particular completion
4104 *
4105 * This will wake up all threads waiting on this particular completion event.
David Howells50fa6102009-04-28 15:01:38 +01004106 *
4107 * It may be assumed that this function implies a write memory barrier before
4108 * changing the task state if and only if any tasks are woken up.
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004109 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02004110void complete_all(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004111{
4112 unsigned long flags;
4113
4114 spin_lock_irqsave(&x->wait.lock, flags);
4115 x->done += UINT_MAX/2;
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05004116 __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004117 spin_unlock_irqrestore(&x->wait.lock, flags);
4118}
4119EXPORT_SYMBOL(complete_all);
4120
Andi Kleen8cbbe862007-10-15 17:00:14 +02004121static inline long __sched
4122do_wait_for_common(struct completion *x, long timeout, int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004123{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004124 if (!x->done) {
4125 DECLARE_WAITQUEUE(wait, current);
4126
Changli Gaoa93d2f12010-05-07 14:33:26 +08004127 __add_wait_queue_tail_exclusive(&x->wait, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004128 do {
Oleg Nesterov94d3d822008-08-20 16:54:41 -07004129 if (signal_pending_state(state, current)) {
Oleg Nesterovea71a542008-06-20 18:32:20 +04004130 timeout = -ERESTARTSYS;
4131 break;
Andi Kleen8cbbe862007-10-15 17:00:14 +02004132 }
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);
Oleg Nesterovea71a542008-06-20 18:32:20 +04004137 } while (!x->done && timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004138 __remove_wait_queue(&x->wait, &wait);
Oleg Nesterovea71a542008-06-20 18:32:20 +04004139 if (!x->done)
4140 return timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004141 }
4142 x->done--;
Oleg Nesterovea71a542008-06-20 18:32:20 +04004143 return timeout ?: 1;
Andi Kleen8cbbe862007-10-15 17:00:14 +02004144}
4145
4146static long __sched
4147wait_for_common(struct completion *x, long timeout, int state)
4148{
4149 might_sleep();
4150
4151 spin_lock_irq(&x->wait.lock);
4152 timeout = do_wait_for_common(x, timeout, state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004153 spin_unlock_irq(&x->wait.lock);
Andi Kleen8cbbe862007-10-15 17:00:14 +02004154 return timeout;
4155}
4156
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004157/**
4158 * wait_for_completion: - waits for completion of a task
4159 * @x: holds the state of this particular completion
4160 *
4161 * This waits to be signaled for completion of a specific task. It is NOT
4162 * interruptible and there is no timeout.
4163 *
4164 * See also similar routines (i.e. wait_for_completion_timeout()) with timeout
4165 * and interrupt capability. Also see complete().
4166 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02004167void __sched wait_for_completion(struct completion *x)
Andi Kleen8cbbe862007-10-15 17:00:14 +02004168{
4169 wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004170}
4171EXPORT_SYMBOL(wait_for_completion);
4172
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004173/**
4174 * wait_for_completion_timeout: - waits for completion of a task (w/timeout)
4175 * @x: holds the state of this particular completion
4176 * @timeout: timeout value in jiffies
4177 *
4178 * This waits for either a completion of a specific task to be signaled or for a
4179 * specified timeout to expire. The timeout is in jiffies. It is not
4180 * interruptible.
4181 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02004182unsigned long __sched
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183wait_for_completion_timeout(struct completion *x, unsigned long timeout)
4184{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004185 return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004186}
4187EXPORT_SYMBOL(wait_for_completion_timeout);
4188
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004189/**
4190 * wait_for_completion_interruptible: - waits for completion of a task (w/intr)
4191 * @x: holds the state of this particular completion
4192 *
4193 * This waits for completion of a specific task to be signaled. It is
4194 * interruptible.
4195 */
Andi Kleen8cbbe862007-10-15 17:00:14 +02004196int __sched wait_for_completion_interruptible(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004197{
Andi Kleen51e97992007-10-18 21:32:55 +02004198 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
4199 if (t == -ERESTARTSYS)
4200 return t;
4201 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004202}
4203EXPORT_SYMBOL(wait_for_completion_interruptible);
4204
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004205/**
4206 * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr))
4207 * @x: holds the state of this particular completion
4208 * @timeout: timeout value in jiffies
4209 *
4210 * This waits for either a completion of a specific task to be signaled or for a
4211 * specified timeout to expire. It is interruptible. The timeout is in jiffies.
4212 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02004213unsigned long __sched
Linus Torvalds1da177e2005-04-16 15:20:36 -07004214wait_for_completion_interruptible_timeout(struct completion *x,
4215 unsigned long timeout)
4216{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004217 return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004218}
4219EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
4220
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004221/**
4222 * wait_for_completion_killable: - waits for completion of a task (killable)
4223 * @x: holds the state of this particular completion
4224 *
4225 * This waits to be signaled for completion of a specific task. It can be
4226 * interrupted by a kill signal.
4227 */
Matthew Wilcox009e5772007-12-06 12:29:54 -05004228int __sched wait_for_completion_killable(struct completion *x)
4229{
4230 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
4231 if (t == -ERESTARTSYS)
4232 return t;
4233 return 0;
4234}
4235EXPORT_SYMBOL(wait_for_completion_killable);
4236
Dave Chinnerbe4de352008-08-15 00:40:44 -07004237/**
Sage Weil0aa12fb2010-05-29 09:12:30 -07004238 * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable))
4239 * @x: holds the state of this particular completion
4240 * @timeout: timeout value in jiffies
4241 *
4242 * This waits for either a completion of a specific task to be
4243 * signaled or for a specified timeout to expire. It can be
4244 * interrupted by a kill signal. The timeout is in jiffies.
4245 */
4246unsigned long __sched
4247wait_for_completion_killable_timeout(struct completion *x,
4248 unsigned long timeout)
4249{
4250 return wait_for_common(x, timeout, TASK_KILLABLE);
4251}
4252EXPORT_SYMBOL(wait_for_completion_killable_timeout);
4253
4254/**
Dave Chinnerbe4de352008-08-15 00:40:44 -07004255 * try_wait_for_completion - try to decrement a completion without blocking
4256 * @x: completion structure
4257 *
4258 * Returns: 0 if a decrement cannot be done without blocking
4259 * 1 if a decrement succeeded.
4260 *
4261 * If a completion is being used as a counting completion,
4262 * attempt to decrement the counter without blocking. This
4263 * enables us to avoid waiting if the resource the completion
4264 * is protecting is not available.
4265 */
4266bool try_wait_for_completion(struct completion *x)
4267{
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004268 unsigned long flags;
Dave Chinnerbe4de352008-08-15 00:40:44 -07004269 int ret = 1;
4270
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004271 spin_lock_irqsave(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07004272 if (!x->done)
4273 ret = 0;
4274 else
4275 x->done--;
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004276 spin_unlock_irqrestore(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07004277 return ret;
4278}
4279EXPORT_SYMBOL(try_wait_for_completion);
4280
4281/**
4282 * completion_done - Test to see if a completion has any waiters
4283 * @x: completion structure
4284 *
4285 * Returns: 0 if there are waiters (wait_for_completion() in progress)
4286 * 1 if there are no waiters.
4287 *
4288 */
4289bool completion_done(struct completion *x)
4290{
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004291 unsigned long flags;
Dave Chinnerbe4de352008-08-15 00:40:44 -07004292 int ret = 1;
4293
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004294 spin_lock_irqsave(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07004295 if (!x->done)
4296 ret = 0;
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004297 spin_unlock_irqrestore(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07004298 return ret;
4299}
4300EXPORT_SYMBOL(completion_done);
4301
Andi Kleen8cbbe862007-10-15 17:00:14 +02004302static long __sched
4303sleep_on_common(wait_queue_head_t *q, int state, long timeout)
Ingo Molnar0fec1712007-07-09 18:52:01 +02004304{
4305 unsigned long flags;
4306 wait_queue_t wait;
4307
4308 init_waitqueue_entry(&wait, current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004309
Andi Kleen8cbbe862007-10-15 17:00:14 +02004310 __set_current_state(state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004311
Andi Kleen8cbbe862007-10-15 17:00:14 +02004312 spin_lock_irqsave(&q->lock, flags);
4313 __add_wait_queue(q, &wait);
4314 spin_unlock(&q->lock);
4315 timeout = schedule_timeout(timeout);
4316 spin_lock_irq(&q->lock);
4317 __remove_wait_queue(q, &wait);
4318 spin_unlock_irqrestore(&q->lock, flags);
4319
4320 return timeout;
4321}
4322
4323void __sched interruptible_sleep_on(wait_queue_head_t *q)
4324{
4325 sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004326}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004327EXPORT_SYMBOL(interruptible_sleep_on);
4328
Ingo Molnar0fec1712007-07-09 18:52:01 +02004329long __sched
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004330interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004331{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004332 return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004333}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004334EXPORT_SYMBOL(interruptible_sleep_on_timeout);
4335
Ingo Molnar0fec1712007-07-09 18:52:01 +02004336void __sched sleep_on(wait_queue_head_t *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004337{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004338 sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004339}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004340EXPORT_SYMBOL(sleep_on);
4341
Ingo Molnar0fec1712007-07-09 18:52:01 +02004342long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004343{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004344 return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004345}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004346EXPORT_SYMBOL(sleep_on_timeout);
4347
Ingo Molnarb29739f2006-06-27 02:54:51 -07004348#ifdef CONFIG_RT_MUTEXES
4349
4350/*
4351 * rt_mutex_setprio - set the current priority of a task
4352 * @p: task
4353 * @prio: prio value (kernel-internal form)
4354 *
4355 * This function changes the 'effective' priority of a task. It does
4356 * not touch ->normal_prio like __setscheduler().
4357 *
4358 * Used by the rt_mutex code to implement priority inheritance logic.
4359 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004360void rt_mutex_setprio(struct task_struct *p, int prio)
Ingo Molnarb29739f2006-06-27 02:54:51 -07004361{
4362 unsigned long flags;
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004363 int oldprio, on_rq, running;
Ingo Molnar70b97a72006-07-03 00:25:42 -07004364 struct rq *rq;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01004365 const struct sched_class *prev_class;
Ingo Molnarb29739f2006-06-27 02:54:51 -07004366
4367 BUG_ON(prio < 0 || prio > MAX_PRIO);
4368
4369 rq = task_rq_lock(p, &flags);
4370
Steven Rostedta8027072010-09-20 15:13:34 -04004371 trace_sched_pi_setprio(p, prio);
Andrew Mortond5f9f942007-05-08 20:27:06 -07004372 oldprio = p->prio;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01004373 prev_class = p->sched_class;
Ingo Molnardd41f592007-07-09 18:51:59 +02004374 on_rq = p->se.on_rq;
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01004375 running = task_current(rq, p);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004376 if (on_rq)
Ingo Molnar69be72c2007-08-09 11:16:49 +02004377 dequeue_task(rq, p, 0);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004378 if (running)
4379 p->sched_class->put_prev_task(rq, p);
Ingo Molnardd41f592007-07-09 18:51:59 +02004380
4381 if (rt_prio(prio))
4382 p->sched_class = &rt_sched_class;
4383 else
4384 p->sched_class = &fair_sched_class;
4385
Ingo Molnarb29739f2006-06-27 02:54:51 -07004386 p->prio = prio;
4387
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004388 if (running)
4389 p->sched_class->set_curr_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02004390 if (on_rq) {
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01004391 enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0);
Steven Rostedtcb469842008-01-25 21:08:22 +01004392
4393 check_class_changed(rq, p, prev_class, oldprio, running);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004394 }
4395 task_rq_unlock(rq, &flags);
4396}
4397
4398#endif
4399
Ingo Molnar36c8b582006-07-03 00:25:41 -07004400void set_user_nice(struct task_struct *p, long nice)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004401{
Ingo Molnardd41f592007-07-09 18:51:59 +02004402 int old_prio, delta, on_rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004403 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07004404 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004405
4406 if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
4407 return;
4408 /*
4409 * We have to be careful, if called from sys_setpriority(),
4410 * the task might be in the middle of scheduling on another CPU.
4411 */
4412 rq = task_rq_lock(p, &flags);
4413 /*
4414 * The RT priorities are set via sched_setscheduler(), but we still
4415 * allow the 'normal' nice value to be set - but as expected
4416 * it wont have any effect on scheduling until the task is
Ingo Molnardd41f592007-07-09 18:51:59 +02004417 * SCHED_FIFO/SCHED_RR:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004418 */
Ingo Molnare05606d2007-07-09 18:51:59 +02004419 if (task_has_rt_policy(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420 p->static_prio = NICE_TO_PRIO(nice);
4421 goto out_unlock;
4422 }
Ingo Molnardd41f592007-07-09 18:51:59 +02004423 on_rq = p->se.on_rq;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02004424 if (on_rq)
Ingo Molnar69be72c2007-08-09 11:16:49 +02004425 dequeue_task(rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004426
Linus Torvalds1da177e2005-04-16 15:20:36 -07004427 p->static_prio = NICE_TO_PRIO(nice);
Peter Williams2dd73a42006-06-27 02:54:34 -07004428 set_load_weight(p);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004429 old_prio = p->prio;
4430 p->prio = effective_prio(p);
4431 delta = p->prio - old_prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004432
Ingo Molnardd41f592007-07-09 18:51:59 +02004433 if (on_rq) {
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01004434 enqueue_task(rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004435 /*
Andrew Mortond5f9f942007-05-08 20:27:06 -07004436 * If the task increased its priority or is running and
4437 * lowered its priority, then reschedule its CPU:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004438 */
Andrew Mortond5f9f942007-05-08 20:27:06 -07004439 if (delta < 0 || (delta > 0 && task_running(rq, p)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004440 resched_task(rq->curr);
4441 }
4442out_unlock:
4443 task_rq_unlock(rq, &flags);
4444}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004445EXPORT_SYMBOL(set_user_nice);
4446
Matt Mackalle43379f2005-05-01 08:59:00 -07004447/*
4448 * can_nice - check if a task can reduce its nice value
4449 * @p: task
4450 * @nice: nice value
4451 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004452int can_nice(const struct task_struct *p, const int nice)
Matt Mackalle43379f2005-05-01 08:59:00 -07004453{
Matt Mackall024f4742005-08-18 11:24:19 -07004454 /* convert nice value [19,-20] to rlimit style value [1,40] */
4455 int nice_rlim = 20 - nice;
Ingo Molnar48f24c42006-07-03 00:25:40 -07004456
Jiri Slaby78d7d402010-03-05 13:42:54 -08004457 return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
Matt Mackalle43379f2005-05-01 08:59:00 -07004458 capable(CAP_SYS_NICE));
4459}
4460
Linus Torvalds1da177e2005-04-16 15:20:36 -07004461#ifdef __ARCH_WANT_SYS_NICE
4462
4463/*
4464 * sys_nice - change the priority of the current process.
4465 * @increment: priority increment
4466 *
4467 * sys_setpriority is a more generic, but much slower function that
4468 * does similar things.
4469 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004470SYSCALL_DEFINE1(nice, int, increment)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004471{
Ingo Molnar48f24c42006-07-03 00:25:40 -07004472 long nice, retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004473
4474 /*
4475 * Setpriority might change our priority at the same moment.
4476 * We don't have to worry. Conceptually one call occurs first
4477 * and we have a single winner.
4478 */
Matt Mackalle43379f2005-05-01 08:59:00 -07004479 if (increment < -40)
4480 increment = -40;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004481 if (increment > 40)
4482 increment = 40;
4483
Américo Wang2b8f8362009-02-16 18:54:21 +08004484 nice = TASK_NICE(current) + increment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004485 if (nice < -20)
4486 nice = -20;
4487 if (nice > 19)
4488 nice = 19;
4489
Matt Mackalle43379f2005-05-01 08:59:00 -07004490 if (increment < 0 && !can_nice(current, nice))
4491 return -EPERM;
4492
Linus Torvalds1da177e2005-04-16 15:20:36 -07004493 retval = security_task_setnice(current, nice);
4494 if (retval)
4495 return retval;
4496
4497 set_user_nice(current, nice);
4498 return 0;
4499}
4500
4501#endif
4502
4503/**
4504 * task_prio - return the priority value of a given task.
4505 * @p: the task in question.
4506 *
4507 * This is the priority value as seen by users in /proc.
4508 * RT tasks are offset by -200. Normal tasks are centered
4509 * around 0, value goes from -16 to +15.
4510 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004511int task_prio(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004512{
4513 return p->prio - MAX_RT_PRIO;
4514}
4515
4516/**
4517 * task_nice - return the nice value of a given task.
4518 * @p: the task in question.
4519 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004520int task_nice(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004521{
4522 return TASK_NICE(p);
4523}
Pavel Roskin150d8be2008-03-05 16:56:37 -05004524EXPORT_SYMBOL(task_nice);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004525
4526/**
4527 * idle_cpu - is a given cpu idle currently?
4528 * @cpu: the processor in question.
4529 */
4530int idle_cpu(int cpu)
4531{
4532 return cpu_curr(cpu) == cpu_rq(cpu)->idle;
4533}
4534
Linus Torvalds1da177e2005-04-16 15:20:36 -07004535/**
4536 * idle_task - return the idle task for a given cpu.
4537 * @cpu: the processor in question.
4538 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004539struct task_struct *idle_task(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004540{
4541 return cpu_rq(cpu)->idle;
4542}
4543
4544/**
4545 * find_process_by_pid - find a process with a matching PID value.
4546 * @pid: the pid in question.
4547 */
Alexey Dobriyana9957442007-10-15 17:00:13 +02004548static struct task_struct *find_process_by_pid(pid_t pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004549{
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07004550 return pid ? find_task_by_vpid(pid) : current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004551}
4552
4553/* Actually do priority change: must hold rq lock. */
Ingo Molnardd41f592007-07-09 18:51:59 +02004554static void
4555__setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004556{
Ingo Molnardd41f592007-07-09 18:51:59 +02004557 BUG_ON(p->se.on_rq);
Ingo Molnar48f24c42006-07-03 00:25:40 -07004558
Linus Torvalds1da177e2005-04-16 15:20:36 -07004559 p->policy = policy;
4560 p->rt_priority = prio;
Ingo Molnarb29739f2006-06-27 02:54:51 -07004561 p->normal_prio = normal_prio(p);
4562 /* we are holding p->pi_lock already */
4563 p->prio = rt_mutex_getprio(p);
Peter Zijlstraffd44db2009-11-10 20:12:01 +01004564 if (rt_prio(p->prio))
4565 p->sched_class = &rt_sched_class;
4566 else
4567 p->sched_class = &fair_sched_class;
Peter Williams2dd73a42006-06-27 02:54:34 -07004568 set_load_weight(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004569}
4570
David Howellsc69e8d92008-11-14 10:39:19 +11004571/*
4572 * check the target process has a UID that matches the current process's
4573 */
4574static bool check_same_owner(struct task_struct *p)
4575{
4576 const struct cred *cred = current_cred(), *pcred;
4577 bool match;
4578
4579 rcu_read_lock();
4580 pcred = __task_cred(p);
4581 match = (cred->euid == pcred->euid ||
4582 cred->euid == pcred->uid);
4583 rcu_read_unlock();
4584 return match;
4585}
4586
Rusty Russell961ccdd2008-06-23 13:55:38 +10004587static int __sched_setscheduler(struct task_struct *p, int policy,
KOSAKI Motohirofe7de492010-10-20 16:01:12 -07004588 const struct sched_param *param, bool user)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004589{
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004590 int retval, oldprio, oldpolicy = -1, on_rq, running;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004591 unsigned long flags;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01004592 const struct sched_class *prev_class;
Ingo Molnar70b97a72006-07-03 00:25:42 -07004593 struct rq *rq;
Lennart Poetteringca94c442009-06-15 17:17:47 +02004594 int reset_on_fork;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004595
Steven Rostedt66e53932006-06-27 02:54:44 -07004596 /* may grab non-irq protected spin_locks */
4597 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07004598recheck:
4599 /* double check policy once rq lock held */
Lennart Poetteringca94c442009-06-15 17:17:47 +02004600 if (policy < 0) {
4601 reset_on_fork = p->sched_reset_on_fork;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004602 policy = oldpolicy = p->policy;
Lennart Poetteringca94c442009-06-15 17:17:47 +02004603 } else {
4604 reset_on_fork = !!(policy & SCHED_RESET_ON_FORK);
4605 policy &= ~SCHED_RESET_ON_FORK;
4606
4607 if (policy != SCHED_FIFO && policy != SCHED_RR &&
4608 policy != SCHED_NORMAL && policy != SCHED_BATCH &&
4609 policy != SCHED_IDLE)
4610 return -EINVAL;
4611 }
4612
Linus Torvalds1da177e2005-04-16 15:20:36 -07004613 /*
4614 * Valid priorities for SCHED_FIFO and SCHED_RR are
Ingo Molnardd41f592007-07-09 18:51:59 +02004615 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
4616 * SCHED_BATCH and SCHED_IDLE is 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004617 */
4618 if (param->sched_priority < 0 ||
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004619 (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
Steven Rostedtd46523e2005-07-25 16:28:39 -04004620 (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004621 return -EINVAL;
Ingo Molnare05606d2007-07-09 18:51:59 +02004622 if (rt_policy(policy) != (param->sched_priority != 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004623 return -EINVAL;
4624
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004625 /*
4626 * Allow unprivileged RT tasks to decrease priority:
4627 */
Rusty Russell961ccdd2008-06-23 13:55:38 +10004628 if (user && !capable(CAP_SYS_NICE)) {
Ingo Molnare05606d2007-07-09 18:51:59 +02004629 if (rt_policy(policy)) {
Oleg Nesterova44702e2010-06-11 01:09:44 +02004630 unsigned long rlim_rtprio =
4631 task_rlimit(p, RLIMIT_RTPRIO);
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004632
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07004633 /* can't set/change the rt policy */
4634 if (policy != p->policy && !rlim_rtprio)
4635 return -EPERM;
4636
4637 /* can't increase priority */
4638 if (param->sched_priority > p->rt_priority &&
4639 param->sched_priority > rlim_rtprio)
4640 return -EPERM;
4641 }
Ingo Molnardd41f592007-07-09 18:51:59 +02004642 /*
4643 * Like positive nice levels, dont allow tasks to
4644 * move out of SCHED_IDLE either:
4645 */
4646 if (p->policy == SCHED_IDLE && policy != SCHED_IDLE)
4647 return -EPERM;
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07004648
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004649 /* can't change other user's priorities */
David Howellsc69e8d92008-11-14 10:39:19 +11004650 if (!check_same_owner(p))
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004651 return -EPERM;
Lennart Poetteringca94c442009-06-15 17:17:47 +02004652
4653 /* Normal users shall not reset the sched_reset_on_fork flag */
4654 if (p->sched_reset_on_fork && !reset_on_fork)
4655 return -EPERM;
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004656 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004657
Jeremy Fitzhardinge725aad22008-08-03 09:33:03 -07004658 if (user) {
KOSAKI Motohirob0ae1982010-10-15 04:21:18 +09004659 retval = security_task_setscheduler(p);
Jeremy Fitzhardinge725aad22008-08-03 09:33:03 -07004660 if (retval)
4661 return retval;
4662 }
4663
Linus Torvalds1da177e2005-04-16 15:20:36 -07004664 /*
Ingo Molnarb29739f2006-06-27 02:54:51 -07004665 * make sure no PI-waiters arrive (or leave) while we are
4666 * changing the priority of the task:
4667 */
Thomas Gleixner1d615482009-11-17 14:54:03 +01004668 raw_spin_lock_irqsave(&p->pi_lock, flags);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004669 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004670 * To be able to change p->policy safely, the apropriate
4671 * runqueue lock must be held.
4672 */
Ingo Molnarb29739f2006-06-27 02:54:51 -07004673 rq = __task_rq_lock(p);
Peter Zijlstradc61b1d2010-06-08 11:40:42 +02004674
Peter Zijlstra34f971f2010-09-22 13:53:15 +02004675 /*
4676 * Changing the policy of the stop threads its a very bad idea
4677 */
4678 if (p == rq->stop) {
4679 __task_rq_unlock(rq);
4680 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4681 return -EINVAL;
4682 }
4683
Peter Zijlstradc61b1d2010-06-08 11:40:42 +02004684#ifdef CONFIG_RT_GROUP_SCHED
4685 if (user) {
4686 /*
4687 * Do not allow realtime tasks into groups that have no runtime
4688 * assigned.
4689 */
4690 if (rt_bandwidth_enabled() && rt_policy(policy) &&
4691 task_group(p)->rt_bandwidth.rt_runtime == 0) {
4692 __task_rq_unlock(rq);
4693 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4694 return -EPERM;
4695 }
4696 }
4697#endif
4698
Linus Torvalds1da177e2005-04-16 15:20:36 -07004699 /* recheck policy now with rq lock held */
4700 if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
4701 policy = oldpolicy = -1;
Ingo Molnarb29739f2006-06-27 02:54:51 -07004702 __task_rq_unlock(rq);
Thomas Gleixner1d615482009-11-17 14:54:03 +01004703 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004704 goto recheck;
4705 }
Ingo Molnardd41f592007-07-09 18:51:59 +02004706 on_rq = p->se.on_rq;
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01004707 running = task_current(rq, p);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004708 if (on_rq)
Ingo Molnar2e1cb742007-08-09 11:16:49 +02004709 deactivate_task(rq, p, 0);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004710 if (running)
4711 p->sched_class->put_prev_task(rq, p);
Dmitry Adamushkof6b53202007-10-15 17:00:08 +02004712
Lennart Poetteringca94c442009-06-15 17:17:47 +02004713 p->sched_reset_on_fork = reset_on_fork;
4714
Linus Torvalds1da177e2005-04-16 15:20:36 -07004715 oldprio = p->prio;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01004716 prev_class = p->sched_class;
Ingo Molnardd41f592007-07-09 18:51:59 +02004717 __setscheduler(rq, p, policy, param->sched_priority);
Dmitry Adamushkof6b53202007-10-15 17:00:08 +02004718
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004719 if (running)
4720 p->sched_class->set_curr_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02004721 if (on_rq) {
4722 activate_task(rq, p, 0);
Steven Rostedtcb469842008-01-25 21:08:22 +01004723
4724 check_class_changed(rq, p, prev_class, oldprio, running);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004725 }
Ingo Molnarb29739f2006-06-27 02:54:51 -07004726 __task_rq_unlock(rq);
Thomas Gleixner1d615482009-11-17 14:54:03 +01004727 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004728
Thomas Gleixner95e02ca2006-06-27 02:55:02 -07004729 rt_mutex_adjust_pi(p);
4730
Linus Torvalds1da177e2005-04-16 15:20:36 -07004731 return 0;
4732}
Rusty Russell961ccdd2008-06-23 13:55:38 +10004733
4734/**
4735 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
4736 * @p: the task in question.
4737 * @policy: new policy.
4738 * @param: structure containing the new RT priority.
4739 *
4740 * NOTE that the task may be already dead.
4741 */
4742int sched_setscheduler(struct task_struct *p, int policy,
KOSAKI Motohirofe7de492010-10-20 16:01:12 -07004743 const struct sched_param *param)
Rusty Russell961ccdd2008-06-23 13:55:38 +10004744{
4745 return __sched_setscheduler(p, policy, param, true);
4746}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004747EXPORT_SYMBOL_GPL(sched_setscheduler);
4748
Rusty Russell961ccdd2008-06-23 13:55:38 +10004749/**
4750 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
4751 * @p: the task in question.
4752 * @policy: new policy.
4753 * @param: structure containing the new RT priority.
4754 *
4755 * Just like sched_setscheduler, only don't bother checking if the
4756 * current context has permission. For example, this is needed in
4757 * stop_machine(): we create temporary high priority worker threads,
4758 * but our caller might not have that capability.
4759 */
4760int sched_setscheduler_nocheck(struct task_struct *p, int policy,
KOSAKI Motohirofe7de492010-10-20 16:01:12 -07004761 const struct sched_param *param)
Rusty Russell961ccdd2008-06-23 13:55:38 +10004762{
4763 return __sched_setscheduler(p, policy, param, false);
4764}
4765
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004766static int
4767do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004768{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004769 struct sched_param lparam;
4770 struct task_struct *p;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004771 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004772
4773 if (!param || pid < 0)
4774 return -EINVAL;
4775 if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
4776 return -EFAULT;
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004777
4778 rcu_read_lock();
4779 retval = -ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004780 p = find_process_by_pid(pid);
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004781 if (p != NULL)
4782 retval = sched_setscheduler(p, policy, &lparam);
4783 rcu_read_unlock();
Ingo Molnar36c8b582006-07-03 00:25:41 -07004784
Linus Torvalds1da177e2005-04-16 15:20:36 -07004785 return retval;
4786}
4787
4788/**
4789 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
4790 * @pid: the pid in question.
4791 * @policy: new policy.
4792 * @param: structure containing the new RT priority.
4793 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004794SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
4795 struct sched_param __user *, param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004796{
Jason Baronc21761f2006-01-18 17:43:03 -08004797 /* negative values for policy are not valid */
4798 if (policy < 0)
4799 return -EINVAL;
4800
Linus Torvalds1da177e2005-04-16 15:20:36 -07004801 return do_sched_setscheduler(pid, policy, param);
4802}
4803
4804/**
4805 * sys_sched_setparam - set/change the RT priority of a thread
4806 * @pid: the pid in question.
4807 * @param: structure containing the new RT priority.
4808 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004809SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004810{
4811 return do_sched_setscheduler(pid, -1, param);
4812}
4813
4814/**
4815 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
4816 * @pid: the pid in question.
4817 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004818SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004819{
Ingo Molnar36c8b582006-07-03 00:25:41 -07004820 struct task_struct *p;
Andi Kleen3a5c3592007-10-15 17:00:14 +02004821 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004822
4823 if (pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02004824 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004825
4826 retval = -ESRCH;
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004827 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004828 p = find_process_by_pid(pid);
4829 if (p) {
4830 retval = security_task_getscheduler(p);
4831 if (!retval)
Lennart Poetteringca94c442009-06-15 17:17:47 +02004832 retval = p->policy
4833 | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004834 }
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004835 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004836 return retval;
4837}
4838
4839/**
Lennart Poetteringca94c442009-06-15 17:17:47 +02004840 * sys_sched_getparam - get the RT priority of a thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07004841 * @pid: the pid in question.
4842 * @param: structure containing the RT priority.
4843 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004844SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004845{
4846 struct sched_param lp;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004847 struct task_struct *p;
Andi Kleen3a5c3592007-10-15 17:00:14 +02004848 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004849
4850 if (!param || pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02004851 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004852
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004853 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004854 p = find_process_by_pid(pid);
4855 retval = -ESRCH;
4856 if (!p)
4857 goto out_unlock;
4858
4859 retval = security_task_getscheduler(p);
4860 if (retval)
4861 goto out_unlock;
4862
4863 lp.sched_priority = p->rt_priority;
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004864 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004865
4866 /*
4867 * This one might sleep, we cannot do it with a spinlock held ...
4868 */
4869 retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
4870
Linus Torvalds1da177e2005-04-16 15:20:36 -07004871 return retval;
4872
4873out_unlock:
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004874 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004875 return retval;
4876}
4877
Rusty Russell96f874e2008-11-25 02:35:14 +10304878long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004879{
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304880 cpumask_var_t cpus_allowed, new_mask;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004881 struct task_struct *p;
4882 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004883
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004884 get_online_cpus();
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004885 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004886
4887 p = find_process_by_pid(pid);
4888 if (!p) {
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004889 rcu_read_unlock();
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004890 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004891 return -ESRCH;
4892 }
4893
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004894 /* Prevent p going away */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004895 get_task_struct(p);
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004896 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004897
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304898 if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
4899 retval = -ENOMEM;
4900 goto out_put_task;
4901 }
4902 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
4903 retval = -ENOMEM;
4904 goto out_free_cpus_allowed;
4905 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004906 retval = -EPERM;
David Howellsc69e8d92008-11-14 10:39:19 +11004907 if (!check_same_owner(p) && !capable(CAP_SYS_NICE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004908 goto out_unlock;
4909
KOSAKI Motohirob0ae1982010-10-15 04:21:18 +09004910 retval = security_task_setscheduler(p);
David Quigleye7834f82006-06-23 02:03:59 -07004911 if (retval)
4912 goto out_unlock;
4913
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304914 cpuset_cpus_allowed(p, cpus_allowed);
4915 cpumask_and(new_mask, in_mask, cpus_allowed);
Peter Zijlstra49246272010-10-17 21:46:10 +02004916again:
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304917 retval = set_cpus_allowed_ptr(p, new_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004918
Paul Menage8707d8b2007-10-18 23:40:22 -07004919 if (!retval) {
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304920 cpuset_cpus_allowed(p, cpus_allowed);
4921 if (!cpumask_subset(new_mask, cpus_allowed)) {
Paul Menage8707d8b2007-10-18 23:40:22 -07004922 /*
4923 * We must have raced with a concurrent cpuset
4924 * update. Just reset the cpus_allowed to the
4925 * cpuset's cpus_allowed
4926 */
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304927 cpumask_copy(new_mask, cpus_allowed);
Paul Menage8707d8b2007-10-18 23:40:22 -07004928 goto again;
4929 }
4930 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004931out_unlock:
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304932 free_cpumask_var(new_mask);
4933out_free_cpus_allowed:
4934 free_cpumask_var(cpus_allowed);
4935out_put_task:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004936 put_task_struct(p);
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004937 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004938 return retval;
4939}
4940
4941static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
Rusty Russell96f874e2008-11-25 02:35:14 +10304942 struct cpumask *new_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004943{
Rusty Russell96f874e2008-11-25 02:35:14 +10304944 if (len < cpumask_size())
4945 cpumask_clear(new_mask);
4946 else if (len > cpumask_size())
4947 len = cpumask_size();
4948
Linus Torvalds1da177e2005-04-16 15:20:36 -07004949 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
4950}
4951
4952/**
4953 * sys_sched_setaffinity - set the cpu affinity of a process
4954 * @pid: pid of the process
4955 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4956 * @user_mask_ptr: user-space pointer to the new cpu mask
4957 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004958SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
4959 unsigned long __user *, user_mask_ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004960{
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304961 cpumask_var_t new_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004962 int retval;
4963
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304964 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
4965 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004966
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304967 retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
4968 if (retval == 0)
4969 retval = sched_setaffinity(pid, new_mask);
4970 free_cpumask_var(new_mask);
4971 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004972}
4973
Rusty Russell96f874e2008-11-25 02:35:14 +10304974long sched_getaffinity(pid_t pid, struct cpumask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004975{
Ingo Molnar36c8b582006-07-03 00:25:41 -07004976 struct task_struct *p;
Thomas Gleixner31605682009-12-08 20:24:16 +00004977 unsigned long flags;
4978 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004979 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004980
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004981 get_online_cpus();
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004982 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004983
4984 retval = -ESRCH;
4985 p = find_process_by_pid(pid);
4986 if (!p)
4987 goto out_unlock;
4988
David Quigleye7834f82006-06-23 02:03:59 -07004989 retval = security_task_getscheduler(p);
4990 if (retval)
4991 goto out_unlock;
4992
Thomas Gleixner31605682009-12-08 20:24:16 +00004993 rq = task_rq_lock(p, &flags);
Rusty Russell96f874e2008-11-25 02:35:14 +10304994 cpumask_and(mask, &p->cpus_allowed, cpu_online_mask);
Thomas Gleixner31605682009-12-08 20:24:16 +00004995 task_rq_unlock(rq, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004996
4997out_unlock:
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004998 rcu_read_unlock();
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004999 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005000
Ulrich Drepper9531b622007-08-09 11:16:46 +02005001 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005002}
5003
5004/**
5005 * sys_sched_getaffinity - get the cpu affinity of a process
5006 * @pid: pid of the process
5007 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
5008 * @user_mask_ptr: user-space pointer to hold the current cpu mask
5009 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01005010SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
5011 unsigned long __user *, user_mask_ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005012{
5013 int ret;
Rusty Russellf17c8602008-11-25 02:35:11 +10305014 cpumask_var_t mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005015
Anton Blanchard84fba5e2010-04-06 17:02:19 +10005016 if ((len * BITS_PER_BYTE) < nr_cpu_ids)
KOSAKI Motohirocd3d8032010-03-12 16:15:36 +09005017 return -EINVAL;
5018 if (len & (sizeof(unsigned long)-1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005019 return -EINVAL;
5020
Rusty Russellf17c8602008-11-25 02:35:11 +10305021 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
5022 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005023
Rusty Russellf17c8602008-11-25 02:35:11 +10305024 ret = sched_getaffinity(pid, mask);
5025 if (ret == 0) {
KOSAKI Motohiro8bc037f2010-03-17 09:36:58 +09005026 size_t retlen = min_t(size_t, len, cpumask_size());
KOSAKI Motohirocd3d8032010-03-12 16:15:36 +09005027
5028 if (copy_to_user(user_mask_ptr, mask, retlen))
Rusty Russellf17c8602008-11-25 02:35:11 +10305029 ret = -EFAULT;
5030 else
KOSAKI Motohirocd3d8032010-03-12 16:15:36 +09005031 ret = retlen;
Rusty Russellf17c8602008-11-25 02:35:11 +10305032 }
5033 free_cpumask_var(mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005034
Rusty Russellf17c8602008-11-25 02:35:11 +10305035 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005036}
5037
5038/**
5039 * sys_sched_yield - yield the current processor to other threads.
5040 *
Ingo Molnardd41f592007-07-09 18:51:59 +02005041 * This function yields the current CPU to other tasks. If there are no
5042 * other threads running on this CPU then this function will return.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005043 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01005044SYSCALL_DEFINE0(sched_yield)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005045{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005046 struct rq *rq = this_rq_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005047
Ingo Molnar2d723762007-10-15 17:00:12 +02005048 schedstat_inc(rq, yld_count);
Dmitry Adamushko4530d7a2007-10-15 17:00:08 +02005049 current->sched_class->yield_task(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005050
5051 /*
5052 * Since we are going to call schedule() anyway, there's
5053 * no need to preempt or enable interrupts:
5054 */
5055 __release(rq->lock);
Ingo Molnar8a25d5d2006-07-03 00:24:54 -07005056 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
Thomas Gleixner9828ea92009-12-03 20:55:53 +01005057 do_raw_spin_unlock(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005058 preempt_enable_no_resched();
5059
5060 schedule();
5061
5062 return 0;
5063}
5064
Peter Zijlstrad86ee482009-07-10 14:57:57 +02005065static inline int should_resched(void)
5066{
5067 return need_resched() && !(preempt_count() & PREEMPT_ACTIVE);
5068}
5069
Andrew Mortone7b38402006-06-30 01:56:00 -07005070static void __cond_resched(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005071{
Frederic Weisbeckere7aaaa62009-07-16 15:44:29 +02005072 add_preempt_count(PREEMPT_ACTIVE);
5073 schedule();
5074 sub_preempt_count(PREEMPT_ACTIVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005075}
5076
Herbert Xu02b67cc2008-01-25 21:08:28 +01005077int __sched _cond_resched(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005078{
Peter Zijlstrad86ee482009-07-10 14:57:57 +02005079 if (should_resched()) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005080 __cond_resched();
5081 return 1;
5082 }
5083 return 0;
5084}
Herbert Xu02b67cc2008-01-25 21:08:28 +01005085EXPORT_SYMBOL(_cond_resched);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005086
5087/*
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02005088 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005089 * call schedule, and on return reacquire the lock.
5090 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005091 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
Linus Torvalds1da177e2005-04-16 15:20:36 -07005092 * operations here to prevent schedule() from being called twice (once via
5093 * spin_unlock(), once by hand).
5094 */
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02005095int __cond_resched_lock(spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005096{
Peter Zijlstrad86ee482009-07-10 14:57:57 +02005097 int resched = should_resched();
Jan Kara6df3cec2005-06-13 15:52:32 -07005098 int ret = 0;
5099
Peter Zijlstraf607c662009-07-20 19:16:29 +02005100 lockdep_assert_held(lock);
5101
Nick Piggin95c354f2008-01-30 13:31:20 +01005102 if (spin_needbreak(lock) || resched) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005103 spin_unlock(lock);
Peter Zijlstrad86ee482009-07-10 14:57:57 +02005104 if (resched)
Nick Piggin95c354f2008-01-30 13:31:20 +01005105 __cond_resched();
5106 else
5107 cpu_relax();
Jan Kara6df3cec2005-06-13 15:52:32 -07005108 ret = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005109 spin_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005110 }
Jan Kara6df3cec2005-06-13 15:52:32 -07005111 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005112}
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02005113EXPORT_SYMBOL(__cond_resched_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005114
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02005115int __sched __cond_resched_softirq(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005116{
5117 BUG_ON(!in_softirq());
5118
Peter Zijlstrad86ee482009-07-10 14:57:57 +02005119 if (should_resched()) {
Thomas Gleixner98d82562007-05-23 13:58:18 -07005120 local_bh_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005121 __cond_resched();
5122 local_bh_disable();
5123 return 1;
5124 }
5125 return 0;
5126}
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02005127EXPORT_SYMBOL(__cond_resched_softirq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005128
Linus Torvalds1da177e2005-04-16 15:20:36 -07005129/**
5130 * yield - yield the current processor to other threads.
5131 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08005132 * This is a shortcut for kernel-space yielding - it marks the
Linus Torvalds1da177e2005-04-16 15:20:36 -07005133 * thread runnable and calls sys_sched_yield().
5134 */
5135void __sched yield(void)
5136{
5137 set_current_state(TASK_RUNNING);
5138 sys_sched_yield();
5139}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005140EXPORT_SYMBOL(yield);
5141
5142/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005143 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
Linus Torvalds1da177e2005-04-16 15:20:36 -07005144 * that process accounting knows that this is a task in IO wait state.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005145 */
5146void __sched io_schedule(void)
5147{
Hitoshi Mitake54d35f22009-06-29 14:44:57 +09005148 struct rq *rq = raw_rq();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005149
Shailabh Nagar0ff92242006-07-14 00:24:37 -07005150 delayacct_blkio_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005151 atomic_inc(&rq->nr_iowait);
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07005152 current->in_iowait = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005153 schedule();
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07005154 current->in_iowait = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005155 atomic_dec(&rq->nr_iowait);
Shailabh Nagar0ff92242006-07-14 00:24:37 -07005156 delayacct_blkio_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005157}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005158EXPORT_SYMBOL(io_schedule);
5159
5160long __sched io_schedule_timeout(long timeout)
5161{
Hitoshi Mitake54d35f22009-06-29 14:44:57 +09005162 struct rq *rq = raw_rq();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005163 long ret;
5164
Shailabh Nagar0ff92242006-07-14 00:24:37 -07005165 delayacct_blkio_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005166 atomic_inc(&rq->nr_iowait);
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07005167 current->in_iowait = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005168 ret = schedule_timeout(timeout);
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07005169 current->in_iowait = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005170 atomic_dec(&rq->nr_iowait);
Shailabh Nagar0ff92242006-07-14 00:24:37 -07005171 delayacct_blkio_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005172 return ret;
5173}
5174
5175/**
5176 * sys_sched_get_priority_max - return maximum RT priority.
5177 * @policy: scheduling class.
5178 *
5179 * this syscall returns the maximum rt_priority that can be used
5180 * by a given scheduling class.
5181 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01005182SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005183{
5184 int ret = -EINVAL;
5185
5186 switch (policy) {
5187 case SCHED_FIFO:
5188 case SCHED_RR:
5189 ret = MAX_USER_RT_PRIO-1;
5190 break;
5191 case SCHED_NORMAL:
Ingo Molnarb0a94992006-01-14 13:20:41 -08005192 case SCHED_BATCH:
Ingo Molnardd41f592007-07-09 18:51:59 +02005193 case SCHED_IDLE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005194 ret = 0;
5195 break;
5196 }
5197 return ret;
5198}
5199
5200/**
5201 * sys_sched_get_priority_min - return minimum RT priority.
5202 * @policy: scheduling class.
5203 *
5204 * this syscall returns the minimum rt_priority that can be used
5205 * by a given scheduling class.
5206 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01005207SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005208{
5209 int ret = -EINVAL;
5210
5211 switch (policy) {
5212 case SCHED_FIFO:
5213 case SCHED_RR:
5214 ret = 1;
5215 break;
5216 case SCHED_NORMAL:
Ingo Molnarb0a94992006-01-14 13:20:41 -08005217 case SCHED_BATCH:
Ingo Molnardd41f592007-07-09 18:51:59 +02005218 case SCHED_IDLE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005219 ret = 0;
5220 }
5221 return ret;
5222}
5223
5224/**
5225 * sys_sched_rr_get_interval - return the default timeslice of a process.
5226 * @pid: pid of the process.
5227 * @interval: userspace pointer to the timeslice value.
5228 *
5229 * this syscall writes the default timeslice value of a given process
5230 * into the user-space timespec buffer. A value of '0' means infinity.
5231 */
Heiko Carstens17da2bd2009-01-14 14:14:10 +01005232SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
Heiko Carstens754fe8d2009-01-14 14:14:09 +01005233 struct timespec __user *, interval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005234{
Ingo Molnar36c8b582006-07-03 00:25:41 -07005235 struct task_struct *p;
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005236 unsigned int time_slice;
Thomas Gleixnerdba091b2009-12-09 09:32:03 +01005237 unsigned long flags;
5238 struct rq *rq;
Andi Kleen3a5c3592007-10-15 17:00:14 +02005239 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005240 struct timespec t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005241
5242 if (pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02005243 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005244
5245 retval = -ESRCH;
Thomas Gleixner1a551ae2009-12-09 10:15:11 +00005246 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005247 p = find_process_by_pid(pid);
5248 if (!p)
5249 goto out_unlock;
5250
5251 retval = security_task_getscheduler(p);
5252 if (retval)
5253 goto out_unlock;
5254
Thomas Gleixnerdba091b2009-12-09 09:32:03 +01005255 rq = task_rq_lock(p, &flags);
5256 time_slice = p->sched_class->get_rr_interval(rq, p);
5257 task_rq_unlock(rq, &flags);
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005258
Thomas Gleixner1a551ae2009-12-09 10:15:11 +00005259 rcu_read_unlock();
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005260 jiffies_to_timespec(time_slice, &t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005261 retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005262 return retval;
Andi Kleen3a5c3592007-10-15 17:00:14 +02005263
Linus Torvalds1da177e2005-04-16 15:20:36 -07005264out_unlock:
Thomas Gleixner1a551ae2009-12-09 10:15:11 +00005265 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005266 return retval;
5267}
5268
Steven Rostedt7c731e02008-05-12 21:20:41 +02005269static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
Ingo Molnar36c8b582006-07-03 00:25:41 -07005270
Ingo Molnar82a1fcb2008-01-25 21:08:02 +01005271void sched_show_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005272{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005273 unsigned long free = 0;
Ingo Molnar36c8b582006-07-03 00:25:41 -07005274 unsigned state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005275
Linus Torvalds1da177e2005-04-16 15:20:36 -07005276 state = p->state ? __ffs(p->state) + 1 : 0;
Erik Gilling28d06862010-11-19 18:08:51 -08005277 printk(KERN_INFO "%-15.15s %c", p->comm,
Andreas Mohr2ed6e342006-07-10 04:43:52 -07005278 state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
Ingo Molnar4bd77322007-07-11 21:21:47 +02005279#if BITS_PER_LONG == 32
Linus Torvalds1da177e2005-04-16 15:20:36 -07005280 if (state == TASK_RUNNING)
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005281 printk(KERN_CONT " running ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005282 else
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005283 printk(KERN_CONT " %08lx ", thread_saved_pc(p));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005284#else
5285 if (state == TASK_RUNNING)
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005286 printk(KERN_CONT " running task ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005287 else
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005288 printk(KERN_CONT " %016lx ", thread_saved_pc(p));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005289#endif
5290#ifdef CONFIG_DEBUG_STACK_USAGE
Eric Sandeen7c9f8862008-04-22 16:38:23 -05005291 free = stack_not_used(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005292#endif
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005293 printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
David Rientjesaa47b7e2009-05-04 01:38:05 -07005294 task_pid_nr(p), task_pid_nr(p->real_parent),
5295 (unsigned long)task_thread_info(p)->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005296
Nick Piggin5fb5e6d2008-01-25 21:08:34 +01005297 show_stack(p, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005298}
5299
Ingo Molnare59e2ae2006-12-06 20:35:59 -08005300void show_state_filter(unsigned long state_filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005301{
Ingo Molnar36c8b582006-07-03 00:25:41 -07005302 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005303
Ingo Molnar4bd77322007-07-11 21:21:47 +02005304#if BITS_PER_LONG == 32
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005305 printk(KERN_INFO
5306 " task PC stack pid father\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005307#else
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005308 printk(KERN_INFO
5309 " task PC stack pid father\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005310#endif
5311 read_lock(&tasklist_lock);
5312 do_each_thread(g, p) {
5313 /*
5314 * reset the NMI-timeout, listing all files on a slow
5315 * console might take alot of time:
5316 */
5317 touch_nmi_watchdog();
Ingo Molnar39bc89f2007-04-25 20:50:03 -07005318 if (!state_filter || (p->state & state_filter))
Ingo Molnar82a1fcb2008-01-25 21:08:02 +01005319 sched_show_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005320 } while_each_thread(g, p);
5321
Jeremy Fitzhardinge04c91672007-05-08 00:28:05 -07005322 touch_all_softlockup_watchdogs();
5323
Ingo Molnardd41f592007-07-09 18:51:59 +02005324#ifdef CONFIG_SCHED_DEBUG
5325 sysrq_sched_debug_show();
5326#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005327 read_unlock(&tasklist_lock);
Ingo Molnare59e2ae2006-12-06 20:35:59 -08005328 /*
5329 * Only show locks if all tasks are dumped:
5330 */
Shmulik Ladkani93335a22009-11-25 15:23:41 +02005331 if (!state_filter)
Ingo Molnare59e2ae2006-12-06 20:35:59 -08005332 debug_show_all_locks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005333}
5334
Ingo Molnar1df21052007-07-09 18:51:58 +02005335void __cpuinit init_idle_bootup_task(struct task_struct *idle)
5336{
Ingo Molnardd41f592007-07-09 18:51:59 +02005337 idle->sched_class = &idle_sched_class;
Ingo Molnar1df21052007-07-09 18:51:58 +02005338}
5339
Ingo Molnarf340c0d2005-06-28 16:40:42 +02005340/**
5341 * init_idle - set up an idle thread for a given CPU
5342 * @idle: task in question
5343 * @cpu: cpu the idle task belongs to
5344 *
5345 * NOTE: this function does not set the idle thread's NEED_RESCHED
5346 * flag, to make booting more robust.
5347 */
Nick Piggin5c1e1762006-10-03 01:14:04 -07005348void __cpuinit init_idle(struct task_struct *idle, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005349{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005350 struct rq *rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005351 unsigned long flags;
5352
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005353 raw_spin_lock_irqsave(&rq->lock, flags);
Ingo Molnar5cbd54e2008-11-12 20:05:50 +01005354
Ingo Molnardd41f592007-07-09 18:51:59 +02005355 __sched_fork(idle);
Peter Zijlstra06b83b52009-12-16 18:04:35 +01005356 idle->state = TASK_RUNNING;
Ingo Molnardd41f592007-07-09 18:51:59 +02005357 idle->se.exec_start = sched_clock();
5358
Rusty Russell96f874e2008-11-25 02:35:14 +10305359 cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu));
Peter Zijlstra6506cf6c2010-09-16 17:50:31 +02005360 /*
5361 * We're having a chicken and egg problem, even though we are
5362 * holding rq->lock, the cpu isn't yet set to this cpu so the
5363 * lockdep check in task_group() will fail.
5364 *
5365 * Similar case to sched_fork(). / Alternatively we could
5366 * use task_rq_lock() here and obtain the other rq->lock.
5367 *
5368 * Silence PROVE_RCU
5369 */
5370 rcu_read_lock();
Ingo Molnardd41f592007-07-09 18:51:59 +02005371 __set_task_cpu(idle, cpu);
Peter Zijlstra6506cf6c2010-09-16 17:50:31 +02005372 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005373
Linus Torvalds1da177e2005-04-16 15:20:36 -07005374 rq->curr = rq->idle = idle;
Nick Piggin4866cde2005-06-25 14:57:23 -07005375#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
5376 idle->oncpu = 1;
5377#endif
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005378 raw_spin_unlock_irqrestore(&rq->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005379
5380 /* Set the preempt count _outside_ the spinlocks! */
Linus Torvalds8e3e0762008-05-10 20:58:02 -07005381#if defined(CONFIG_PREEMPT)
5382 task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0);
5383#else
Al Viroa1261f52005-11-13 16:06:55 -08005384 task_thread_info(idle)->preempt_count = 0;
Linus Torvalds8e3e0762008-05-10 20:58:02 -07005385#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02005386 /*
5387 * The idle tasks have their own, simple scheduling class:
5388 */
5389 idle->sched_class = &idle_sched_class;
Frederic Weisbeckerfb526072008-11-25 21:07:04 +01005390 ftrace_graph_init_task(idle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005391}
5392
5393/*
5394 * In a system that switches off the HZ timer nohz_cpu_mask
5395 * indicates which cpus entered this state. This is used
5396 * in the rcu update to wait only for active cpus. For system
5397 * which do not switch off the HZ timer nohz_cpu_mask should
Rusty Russell6a7b3dc2008-11-25 02:35:04 +10305398 * always be CPU_BITS_NONE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005399 */
Rusty Russell6a7b3dc2008-11-25 02:35:04 +10305400cpumask_var_t nohz_cpu_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005401
Ingo Molnar19978ca2007-11-09 22:39:38 +01005402/*
5403 * Increase the granularity value when there are more CPUs,
5404 * because with more CPUs the 'effective latency' as visible
5405 * to users decreases. But the relationship is not linear,
5406 * so pick a second-best guess by going with the log2 of the
5407 * number of CPUs.
5408 *
5409 * This idea comes from the SD scheduler of Con Kolivas:
5410 */
Christian Ehrhardtacb4a842009-11-30 12:16:48 +01005411static int get_update_sysctl_factor(void)
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01005412{
Mike Galbraith4ca3ef72009-12-10 09:25:53 +01005413 unsigned int cpus = min_t(int, num_online_cpus(), 8);
Christian Ehrhardt1983a922009-11-30 12:16:47 +01005414 unsigned int factor;
5415
5416 switch (sysctl_sched_tunable_scaling) {
5417 case SCHED_TUNABLESCALING_NONE:
5418 factor = 1;
5419 break;
5420 case SCHED_TUNABLESCALING_LINEAR:
5421 factor = cpus;
5422 break;
5423 case SCHED_TUNABLESCALING_LOG:
5424 default:
5425 factor = 1 + ilog2(cpus);
5426 break;
5427 }
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01005428
Christian Ehrhardtacb4a842009-11-30 12:16:48 +01005429 return factor;
5430}
5431
5432static void update_sysctl(void)
5433{
5434 unsigned int factor = get_update_sysctl_factor();
5435
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01005436#define SET_SYSCTL(name) \
5437 (sysctl_##name = (factor) * normalized_sysctl_##name)
5438 SET_SYSCTL(sched_min_granularity);
5439 SET_SYSCTL(sched_latency);
5440 SET_SYSCTL(sched_wakeup_granularity);
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01005441#undef SET_SYSCTL
5442}
5443
Ingo Molnar19978ca2007-11-09 22:39:38 +01005444static inline void sched_init_granularity(void)
5445{
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01005446 update_sysctl();
Ingo Molnar19978ca2007-11-09 22:39:38 +01005447}
5448
Linus Torvalds1da177e2005-04-16 15:20:36 -07005449#ifdef CONFIG_SMP
5450/*
5451 * This is how migration works:
5452 *
Tejun Heo969c7922010-05-06 18:49:21 +02005453 * 1) we invoke migration_cpu_stop() on the target CPU using
5454 * stop_one_cpu().
5455 * 2) stopper starts to run (implicitly forcing the migrated thread
5456 * off the CPU)
5457 * 3) it checks whether the migrated task is still in the wrong runqueue.
5458 * 4) if it's in the wrong runqueue then the migration thread removes
Linus Torvalds1da177e2005-04-16 15:20:36 -07005459 * it and puts it into the right queue.
Tejun Heo969c7922010-05-06 18:49:21 +02005460 * 5) stopper completes and stop_one_cpu() returns and the migration
5461 * is done.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005462 */
5463
5464/*
5465 * Change a given task's CPU affinity. Migrate the thread to a
5466 * proper CPU and schedule it away if the CPU it's executing on
5467 * is removed from the allowed bitmask.
5468 *
5469 * NOTE: the caller must have a valid reference to the task, the
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005470 * task must not exit() & deallocate itself prematurely. The
Linus Torvalds1da177e2005-04-16 15:20:36 -07005471 * call is not atomic; no spinlocks may be held.
5472 */
Rusty Russell96f874e2008-11-25 02:35:14 +10305473int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005474{
5475 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07005476 struct rq *rq;
Tejun Heo969c7922010-05-06 18:49:21 +02005477 unsigned int dest_cpu;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005478 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005479
Peter Zijlstra65cc8e42010-03-25 21:05:16 +01005480 /*
5481 * Serialize against TASK_WAKING so that ttwu() and wunt() can
5482 * drop the rq->lock and still rely on ->cpus_allowed.
5483 */
5484again:
5485 while (task_is_waking(p))
5486 cpu_relax();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005487 rq = task_rq_lock(p, &flags);
Peter Zijlstra65cc8e42010-03-25 21:05:16 +01005488 if (task_is_waking(p)) {
5489 task_rq_unlock(rq, &flags);
5490 goto again;
5491 }
Peter Zijlstrae2912002009-12-16 18:04:36 +01005492
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01005493 if (!cpumask_intersects(new_mask, cpu_active_mask)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005494 ret = -EINVAL;
5495 goto out;
5496 }
5497
David Rientjes9985b0b2008-06-05 12:57:11 -07005498 if (unlikely((p->flags & PF_THREAD_BOUND) && p != current &&
Rusty Russell96f874e2008-11-25 02:35:14 +10305499 !cpumask_equal(&p->cpus_allowed, new_mask))) {
David Rientjes9985b0b2008-06-05 12:57:11 -07005500 ret = -EINVAL;
5501 goto out;
5502 }
5503
Gregory Haskins73fe6aa2008-01-25 21:08:07 +01005504 if (p->sched_class->set_cpus_allowed)
Mike Traviscd8ba7c2008-03-26 14:23:49 -07005505 p->sched_class->set_cpus_allowed(p, new_mask);
Gregory Haskins73fe6aa2008-01-25 21:08:07 +01005506 else {
Rusty Russell96f874e2008-11-25 02:35:14 +10305507 cpumask_copy(&p->cpus_allowed, new_mask);
5508 p->rt.nr_cpus_allowed = cpumask_weight(new_mask);
Gregory Haskins73fe6aa2008-01-25 21:08:07 +01005509 }
5510
Linus Torvalds1da177e2005-04-16 15:20:36 -07005511 /* Can the task run on the task's current CPU? If so, we're done */
Rusty Russell96f874e2008-11-25 02:35:14 +10305512 if (cpumask_test_cpu(task_cpu(p), new_mask))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005513 goto out;
5514
Tejun Heo969c7922010-05-06 18:49:21 +02005515 dest_cpu = cpumask_any_and(cpu_active_mask, new_mask);
Nikanth Karthikesanb7a2b392010-11-26 12:37:09 +05305516 if (migrate_task(p, rq)) {
Tejun Heo969c7922010-05-06 18:49:21 +02005517 struct migration_arg arg = { p, dest_cpu };
Linus Torvalds1da177e2005-04-16 15:20:36 -07005518 /* Need help from migration thread: drop lock and wait. */
5519 task_rq_unlock(rq, &flags);
Tejun Heo969c7922010-05-06 18:49:21 +02005520 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005521 tlb_migrate_finish(p->mm);
5522 return 0;
5523 }
5524out:
5525 task_rq_unlock(rq, &flags);
Ingo Molnar48f24c42006-07-03 00:25:40 -07005526
Linus Torvalds1da177e2005-04-16 15:20:36 -07005527 return ret;
5528}
Mike Traviscd8ba7c2008-03-26 14:23:49 -07005529EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005530
5531/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005532 * Move (not current) task off this cpu, onto dest cpu. We're doing
Linus Torvalds1da177e2005-04-16 15:20:36 -07005533 * this because either it can't run here any more (set_cpus_allowed()
5534 * away from this CPU, or CPU going down), or because we're
5535 * attempting to rebalance this task on exec (sched_exec).
5536 *
5537 * So we race with normal scheduler movements, but that's OK, as long
5538 * as the task is no longer on this CPU.
Kirill Korotaevefc30812006-06-27 02:54:32 -07005539 *
5540 * Returns non-zero if task was successfully migrated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005541 */
Kirill Korotaevefc30812006-06-27 02:54:32 -07005542static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005543{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005544 struct rq *rq_dest, *rq_src;
Peter Zijlstrae2912002009-12-16 18:04:36 +01005545 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005546
Max Krasnyanskye761b772008-07-15 04:43:49 -07005547 if (unlikely(!cpu_active(dest_cpu)))
Kirill Korotaevefc30812006-06-27 02:54:32 -07005548 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005549
5550 rq_src = cpu_rq(src_cpu);
5551 rq_dest = cpu_rq(dest_cpu);
5552
5553 double_rq_lock(rq_src, rq_dest);
5554 /* Already moved. */
5555 if (task_cpu(p) != src_cpu)
Linus Torvaldsb1e38732008-07-10 11:25:03 -07005556 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005557 /* Affinity changed (again). */
Rusty Russell96f874e2008-11-25 02:35:14 +10305558 if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
Linus Torvaldsb1e38732008-07-10 11:25:03 -07005559 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005560
Peter Zijlstrae2912002009-12-16 18:04:36 +01005561 /*
5562 * If we're not on a rq, the next wake-up will ensure we're
5563 * placed properly.
5564 */
5565 if (p->se.on_rq) {
Ingo Molnar2e1cb742007-08-09 11:16:49 +02005566 deactivate_task(rq_src, p, 0);
Peter Zijlstrae2912002009-12-16 18:04:36 +01005567 set_task_cpu(p, dest_cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02005568 activate_task(rq_dest, p, 0);
Peter Zijlstra15afe092008-09-20 23:38:02 +02005569 check_preempt_curr(rq_dest, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005570 }
Linus Torvaldsb1e38732008-07-10 11:25:03 -07005571done:
Kirill Korotaevefc30812006-06-27 02:54:32 -07005572 ret = 1;
Linus Torvaldsb1e38732008-07-10 11:25:03 -07005573fail:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005574 double_rq_unlock(rq_src, rq_dest);
Kirill Korotaevefc30812006-06-27 02:54:32 -07005575 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005576}
5577
5578/*
Tejun Heo969c7922010-05-06 18:49:21 +02005579 * migration_cpu_stop - this will be executed by a highprio stopper thread
5580 * and performs thread migration by bumping thread off CPU then
5581 * 'pushing' onto another runqueue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005582 */
Tejun Heo969c7922010-05-06 18:49:21 +02005583static int migration_cpu_stop(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005584{
Tejun Heo969c7922010-05-06 18:49:21 +02005585 struct migration_arg *arg = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005586
Tejun Heo969c7922010-05-06 18:49:21 +02005587 /*
5588 * The original target cpu might have gone down and we might
5589 * be on another cpu but it doesn't matter.
5590 */
5591 local_irq_disable();
5592 __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu);
5593 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005594 return 0;
5595}
5596
5597#ifdef CONFIG_HOTPLUG_CPU
Linus Torvalds1da177e2005-04-16 15:20:36 -07005598
Ingo Molnar48f24c42006-07-03 00:25:40 -07005599/*
5600 * Ensures that the idle task is using init_mm right before its cpu goes
Linus Torvalds1da177e2005-04-16 15:20:36 -07005601 * offline.
5602 */
5603void idle_task_exit(void)
5604{
5605 struct mm_struct *mm = current->active_mm;
5606
5607 BUG_ON(cpu_online(smp_processor_id()));
5608
5609 if (mm != &init_mm)
5610 switch_mm(mm, &init_mm, current);
5611 mmdrop(mm);
5612}
5613
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01005614/*
5615 * While a dead CPU has no uninterruptible tasks queued at this point,
5616 * it might still have a nonzero ->nr_uninterruptible counter, because
5617 * for performance reasons the counter is not stricly tracking tasks to
5618 * their home CPUs. So we just add the counter to another CPU's counter,
5619 * to keep the global sum constant after CPU-down:
5620 */
5621static void migrate_nr_uninterruptible(struct rq *rq_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005622{
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01005623 struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005624
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01005625 rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible;
5626 rq_src->nr_uninterruptible = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005627}
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02005628
5629/*
5630 * remove the tasks which were accounted by rq from calc_load_tasks.
5631 */
5632static void calc_global_load_remove(struct rq *rq)
5633{
5634 atomic_long_sub(rq->calc_load_active, &calc_load_tasks);
Thomas Gleixnera468d382009-07-17 14:15:46 +02005635 rq->calc_load_active = 0;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02005636}
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01005637
5638/*
5639 * Migrate all tasks from the rq, sleeping tasks will be migrated by
5640 * try_to_wake_up()->select_task_rq().
5641 *
5642 * Called with rq->lock held even though we'er in stop_machine() and
5643 * there's no concurrency possible, we hold the required locks anyway
5644 * because of lock validation efforts.
5645 */
5646static void migrate_tasks(unsigned int dead_cpu)
5647{
5648 struct rq *rq = cpu_rq(dead_cpu);
5649 struct task_struct *next, *stop = rq->stop;
5650 int dest_cpu;
5651
5652 /*
5653 * Fudge the rq selection such that the below task selection loop
5654 * doesn't get stuck on the currently eligible stop task.
5655 *
5656 * We're currently inside stop_machine() and the rq is either stuck
5657 * in the stop_machine_cpu_stop() loop, or we're executing this code,
5658 * either way we should never end up calling schedule() until we're
5659 * done here.
5660 */
5661 rq->stop = NULL;
5662
5663 for ( ; ; ) {
5664 /*
5665 * There's this thread running, bail when that's the only
5666 * remaining thread.
5667 */
5668 if (rq->nr_running == 1)
5669 break;
5670
5671 next = pick_next_task(rq);
5672 BUG_ON(!next);
5673 next->sched_class->put_prev_task(rq, next);
5674
5675 /* Find suitable destination for @next, with force if needed. */
5676 dest_cpu = select_fallback_rq(dead_cpu, next);
5677 raw_spin_unlock(&rq->lock);
5678
5679 __migrate_task(next, dead_cpu, dest_cpu);
5680
5681 raw_spin_lock(&rq->lock);
5682 }
5683
5684 rq->stop = stop;
5685}
5686
Linus Torvalds1da177e2005-04-16 15:20:36 -07005687#endif /* CONFIG_HOTPLUG_CPU */
5688
Nick Piggine692ab52007-07-26 13:40:43 +02005689#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
5690
5691static struct ctl_table sd_ctl_dir[] = {
Alexey Dobriyane0361852007-08-09 11:16:46 +02005692 {
5693 .procname = "sched_domain",
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005694 .mode = 0555,
Alexey Dobriyane0361852007-08-09 11:16:46 +02005695 },
Eric W. Biederman56992302009-11-05 15:38:40 -08005696 {}
Nick Piggine692ab52007-07-26 13:40:43 +02005697};
5698
5699static struct ctl_table sd_ctl_root[] = {
Alexey Dobriyane0361852007-08-09 11:16:46 +02005700 {
5701 .procname = "kernel",
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005702 .mode = 0555,
Alexey Dobriyane0361852007-08-09 11:16:46 +02005703 .child = sd_ctl_dir,
5704 },
Eric W. Biederman56992302009-11-05 15:38:40 -08005705 {}
Nick Piggine692ab52007-07-26 13:40:43 +02005706};
5707
5708static struct ctl_table *sd_alloc_ctl_entry(int n)
5709{
5710 struct ctl_table *entry =
Milton Miller5cf9f062007-10-15 17:00:19 +02005711 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
Nick Piggine692ab52007-07-26 13:40:43 +02005712
Nick Piggine692ab52007-07-26 13:40:43 +02005713 return entry;
5714}
5715
Milton Miller6382bc92007-10-15 17:00:19 +02005716static void sd_free_ctl_entry(struct ctl_table **tablep)
5717{
Milton Millercd790072007-10-17 16:55:11 +02005718 struct ctl_table *entry;
Milton Miller6382bc92007-10-15 17:00:19 +02005719
Milton Millercd790072007-10-17 16:55:11 +02005720 /*
5721 * In the intermediate directories, both the child directory and
5722 * procname are dynamically allocated and could fail but the mode
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005723 * will always be set. In the lowest directory the names are
Milton Millercd790072007-10-17 16:55:11 +02005724 * static strings and all have proc handlers.
5725 */
5726 for (entry = *tablep; entry->mode; entry++) {
Milton Miller6382bc92007-10-15 17:00:19 +02005727 if (entry->child)
5728 sd_free_ctl_entry(&entry->child);
Milton Millercd790072007-10-17 16:55:11 +02005729 if (entry->proc_handler == NULL)
5730 kfree(entry->procname);
5731 }
Milton Miller6382bc92007-10-15 17:00:19 +02005732
5733 kfree(*tablep);
5734 *tablep = NULL;
5735}
5736
Nick Piggine692ab52007-07-26 13:40:43 +02005737static void
Alexey Dobriyane0361852007-08-09 11:16:46 +02005738set_table_entry(struct ctl_table *entry,
Nick Piggine692ab52007-07-26 13:40:43 +02005739 const char *procname, void *data, int maxlen,
5740 mode_t mode, proc_handler *proc_handler)
5741{
Nick Piggine692ab52007-07-26 13:40:43 +02005742 entry->procname = procname;
5743 entry->data = data;
5744 entry->maxlen = maxlen;
5745 entry->mode = mode;
5746 entry->proc_handler = proc_handler;
5747}
5748
5749static struct ctl_table *
5750sd_alloc_ctl_domain_table(struct sched_domain *sd)
5751{
Ingo Molnara5d8c342008-10-09 11:35:51 +02005752 struct ctl_table *table = sd_alloc_ctl_entry(13);
Nick Piggine692ab52007-07-26 13:40:43 +02005753
Milton Millerad1cdc12007-10-15 17:00:19 +02005754 if (table == NULL)
5755 return NULL;
5756
Alexey Dobriyane0361852007-08-09 11:16:46 +02005757 set_table_entry(&table[0], "min_interval", &sd->min_interval,
Nick Piggine692ab52007-07-26 13:40:43 +02005758 sizeof(long), 0644, proc_doulongvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005759 set_table_entry(&table[1], "max_interval", &sd->max_interval,
Nick Piggine692ab52007-07-26 13:40:43 +02005760 sizeof(long), 0644, proc_doulongvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005761 set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005762 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005763 set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005764 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005765 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005766 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005767 set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005768 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005769 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005770 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005771 set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
Nick Piggine692ab52007-07-26 13:40:43 +02005772 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005773 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
Nick Piggine692ab52007-07-26 13:40:43 +02005774 sizeof(int), 0644, proc_dointvec_minmax);
Zou Nan haiace8b3d2007-10-15 17:00:14 +02005775 set_table_entry(&table[9], "cache_nice_tries",
Nick Piggine692ab52007-07-26 13:40:43 +02005776 &sd->cache_nice_tries,
5777 sizeof(int), 0644, proc_dointvec_minmax);
Zou Nan haiace8b3d2007-10-15 17:00:14 +02005778 set_table_entry(&table[10], "flags", &sd->flags,
Nick Piggine692ab52007-07-26 13:40:43 +02005779 sizeof(int), 0644, proc_dointvec_minmax);
Ingo Molnara5d8c342008-10-09 11:35:51 +02005780 set_table_entry(&table[11], "name", sd->name,
5781 CORENAME_MAX_SIZE, 0444, proc_dostring);
5782 /* &table[12] is terminator */
Nick Piggine692ab52007-07-26 13:40:43 +02005783
5784 return table;
5785}
5786
Ingo Molnar9a4e7152007-11-28 15:52:56 +01005787static ctl_table *sd_alloc_ctl_cpu_table(int cpu)
Nick Piggine692ab52007-07-26 13:40:43 +02005788{
5789 struct ctl_table *entry, *table;
5790 struct sched_domain *sd;
5791 int domain_num = 0, i;
5792 char buf[32];
5793
5794 for_each_domain(cpu, sd)
5795 domain_num++;
5796 entry = table = sd_alloc_ctl_entry(domain_num + 1);
Milton Millerad1cdc12007-10-15 17:00:19 +02005797 if (table == NULL)
5798 return NULL;
Nick Piggine692ab52007-07-26 13:40:43 +02005799
5800 i = 0;
5801 for_each_domain(cpu, sd) {
5802 snprintf(buf, 32, "domain%d", i);
Nick Piggine692ab52007-07-26 13:40:43 +02005803 entry->procname = kstrdup(buf, GFP_KERNEL);
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005804 entry->mode = 0555;
Nick Piggine692ab52007-07-26 13:40:43 +02005805 entry->child = sd_alloc_ctl_domain_table(sd);
5806 entry++;
5807 i++;
5808 }
5809 return table;
5810}
5811
5812static struct ctl_table_header *sd_sysctl_header;
Milton Miller6382bc92007-10-15 17:00:19 +02005813static void register_sched_domain_sysctl(void)
Nick Piggine692ab52007-07-26 13:40:43 +02005814{
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01005815 int i, cpu_num = num_possible_cpus();
Nick Piggine692ab52007-07-26 13:40:43 +02005816 struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
5817 char buf[32];
5818
Milton Miller73785472007-10-24 18:23:48 +02005819 WARN_ON(sd_ctl_dir[0].child);
5820 sd_ctl_dir[0].child = entry;
5821
Milton Millerad1cdc12007-10-15 17:00:19 +02005822 if (entry == NULL)
5823 return;
5824
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01005825 for_each_possible_cpu(i) {
Nick Piggine692ab52007-07-26 13:40:43 +02005826 snprintf(buf, 32, "cpu%d", i);
Nick Piggine692ab52007-07-26 13:40:43 +02005827 entry->procname = kstrdup(buf, GFP_KERNEL);
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005828 entry->mode = 0555;
Nick Piggine692ab52007-07-26 13:40:43 +02005829 entry->child = sd_alloc_ctl_cpu_table(i);
Milton Miller97b6ea72007-10-15 17:00:19 +02005830 entry++;
Nick Piggine692ab52007-07-26 13:40:43 +02005831 }
Milton Miller73785472007-10-24 18:23:48 +02005832
5833 WARN_ON(sd_sysctl_header);
Nick Piggine692ab52007-07-26 13:40:43 +02005834 sd_sysctl_header = register_sysctl_table(sd_ctl_root);
5835}
Milton Miller6382bc92007-10-15 17:00:19 +02005836
Milton Miller73785472007-10-24 18:23:48 +02005837/* may be called multiple times per register */
Milton Miller6382bc92007-10-15 17:00:19 +02005838static void unregister_sched_domain_sysctl(void)
5839{
Milton Miller73785472007-10-24 18:23:48 +02005840 if (sd_sysctl_header)
5841 unregister_sysctl_table(sd_sysctl_header);
Milton Miller6382bc92007-10-15 17:00:19 +02005842 sd_sysctl_header = NULL;
Milton Miller73785472007-10-24 18:23:48 +02005843 if (sd_ctl_dir[0].child)
5844 sd_free_ctl_entry(&sd_ctl_dir[0].child);
Milton Miller6382bc92007-10-15 17:00:19 +02005845}
Nick Piggine692ab52007-07-26 13:40:43 +02005846#else
Milton Miller6382bc92007-10-15 17:00:19 +02005847static void register_sched_domain_sysctl(void)
5848{
5849}
5850static void unregister_sched_domain_sysctl(void)
Nick Piggine692ab52007-07-26 13:40:43 +02005851{
5852}
5853#endif
5854
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005855static void set_rq_online(struct rq *rq)
5856{
5857 if (!rq->online) {
5858 const struct sched_class *class;
5859
Rusty Russellc6c49272008-11-25 02:35:05 +10305860 cpumask_set_cpu(rq->cpu, rq->rd->online);
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005861 rq->online = 1;
5862
5863 for_each_class(class) {
5864 if (class->rq_online)
5865 class->rq_online(rq);
5866 }
5867 }
5868}
5869
5870static void set_rq_offline(struct rq *rq)
5871{
5872 if (rq->online) {
5873 const struct sched_class *class;
5874
5875 for_each_class(class) {
5876 if (class->rq_offline)
5877 class->rq_offline(rq);
5878 }
5879
Rusty Russellc6c49272008-11-25 02:35:05 +10305880 cpumask_clear_cpu(rq->cpu, rq->rd->online);
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005881 rq->online = 0;
5882 }
5883}
5884
Linus Torvalds1da177e2005-04-16 15:20:36 -07005885/*
5886 * migration_call - callback that gets triggered when a CPU is added.
5887 * Here we can start up the necessary migration thread for the new CPU.
5888 */
Ingo Molnar48f24c42006-07-03 00:25:40 -07005889static int __cpuinit
5890migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005891{
Ingo Molnar48f24c42006-07-03 00:25:40 -07005892 int cpu = (long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005893 unsigned long flags;
Tejun Heo969c7922010-05-06 18:49:21 +02005894 struct rq *rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005895
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01005896 switch (action & ~CPU_TASKS_FROZEN) {
Gautham R Shenoy5be93612007-05-09 02:34:04 -07005897
Linus Torvalds1da177e2005-04-16 15:20:36 -07005898 case CPU_UP_PREPARE:
Thomas Gleixnera468d382009-07-17 14:15:46 +02005899 rq->calc_load_update = calc_load_update;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005900 break;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005901
Linus Torvalds1da177e2005-04-16 15:20:36 -07005902 case CPU_ONLINE:
Gregory Haskins1f94ef52008-03-10 16:52:41 -04005903 /* Update our root-domain */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005904 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins1f94ef52008-03-10 16:52:41 -04005905 if (rq->rd) {
Rusty Russellc6c49272008-11-25 02:35:05 +10305906 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005907
5908 set_rq_online(rq);
Gregory Haskins1f94ef52008-03-10 16:52:41 -04005909 }
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005910 raw_spin_unlock_irqrestore(&rq->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005911 break;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005912
Linus Torvalds1da177e2005-04-16 15:20:36 -07005913#ifdef CONFIG_HOTPLUG_CPU
Gregory Haskins08f503b2008-03-10 17:59:11 -04005914 case CPU_DYING:
Gregory Haskins57d885f2008-01-25 21:08:18 +01005915 /* Update our root-domain */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005916 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005917 if (rq->rd) {
Rusty Russellc6c49272008-11-25 02:35:05 +10305918 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005919 set_rq_offline(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005920 }
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01005921 migrate_tasks(cpu);
5922 BUG_ON(rq->nr_running != 1); /* the migration thread */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005923 raw_spin_unlock_irqrestore(&rq->lock, flags);
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01005924
5925 migrate_nr_uninterruptible(rq);
5926 calc_global_load_remove(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005927 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005928#endif
5929 }
5930 return NOTIFY_OK;
5931}
5932
Paul Mackerrasf38b0822009-06-02 21:05:16 +10005933/*
5934 * Register at high priority so that task migration (migrate_all_tasks)
5935 * happens before everything else. This has to be lower priority than
Ingo Molnarcdd6c482009-09-21 12:02:48 +02005936 * the notifier in the perf_event subsystem, though.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005937 */
Chandra Seetharaman26c21432006-06-27 02:54:10 -07005938static struct notifier_block __cpuinitdata migration_notifier = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005939 .notifier_call = migration_call,
Tejun Heo50a323b2010-06-08 21:40:36 +02005940 .priority = CPU_PRI_MIGRATION,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005941};
5942
Tejun Heo3a101d02010-06-08 21:40:36 +02005943static int __cpuinit sched_cpu_active(struct notifier_block *nfb,
5944 unsigned long action, void *hcpu)
5945{
5946 switch (action & ~CPU_TASKS_FROZEN) {
5947 case CPU_ONLINE:
5948 case CPU_DOWN_FAILED:
5949 set_cpu_active((long)hcpu, true);
5950 return NOTIFY_OK;
5951 default:
5952 return NOTIFY_DONE;
5953 }
5954}
5955
5956static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb,
5957 unsigned long action, void *hcpu)
5958{
5959 switch (action & ~CPU_TASKS_FROZEN) {
5960 case CPU_DOWN_PREPARE:
5961 set_cpu_active((long)hcpu, false);
5962 return NOTIFY_OK;
5963 default:
5964 return NOTIFY_DONE;
5965 }
5966}
5967
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -07005968static int __init migration_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005969{
5970 void *cpu = (void *)(long)smp_processor_id();
Akinobu Mita07dccf32006-09-29 02:00:22 -07005971 int err;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005972
Tejun Heo3a101d02010-06-08 21:40:36 +02005973 /* Initialize migration for the boot CPU */
Akinobu Mita07dccf32006-09-29 02:00:22 -07005974 err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
5975 BUG_ON(err == NOTIFY_BAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005976 migration_call(&migration_notifier, CPU_ONLINE, cpu);
5977 register_cpu_notifier(&migration_notifier);
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -07005978
Tejun Heo3a101d02010-06-08 21:40:36 +02005979 /* Register cpu active notifiers */
5980 cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE);
5981 cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE);
5982
Thomas Gleixnera004cd42009-07-21 09:54:05 +02005983 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005984}
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -07005985early_initcall(migration_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005986#endif
5987
5988#ifdef CONFIG_SMP
Christoph Lameter476f3532007-05-06 14:48:58 -07005989
Ingo Molnar3e9830d2007-10-15 17:00:13 +02005990#ifdef CONFIG_SCHED_DEBUG
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005991
Mike Travisf6630112009-11-17 18:22:15 -06005992static __read_mostly int sched_domain_debug_enabled;
5993
5994static int __init sched_domain_debug_setup(char *str)
5995{
5996 sched_domain_debug_enabled = 1;
5997
5998 return 0;
5999}
6000early_param("sched_debug", sched_domain_debug_setup);
6001
Mike Travis7c16ec52008-04-04 18:11:11 -07006002static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
Rusty Russell96f874e2008-11-25 02:35:14 +10306003 struct cpumask *groupmask)
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006004{
6005 struct sched_group *group = sd->groups;
Mike Travis434d53b2008-04-04 18:11:04 -07006006 char str[256];
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006007
Rusty Russell968ea6d2008-12-13 21:55:51 +10306008 cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd));
Rusty Russell96f874e2008-11-25 02:35:14 +10306009 cpumask_clear(groupmask);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006010
6011 printk(KERN_DEBUG "%*s domain %d: ", level, "", level);
6012
6013 if (!(sd->flags & SD_LOAD_BALANCE)) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006014 printk("does not load-balance\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006015 if (sd->parent)
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006016 printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
6017 " has parent");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006018 return -1;
6019 }
6020
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006021 printk(KERN_CONT "span %s level %s\n", str, sd->name);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006022
Rusty Russell758b2cd2008-11-25 02:35:04 +10306023 if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006024 printk(KERN_ERR "ERROR: domain->span does not contain "
6025 "CPU%d\n", cpu);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006026 }
Rusty Russell758b2cd2008-11-25 02:35:04 +10306027 if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006028 printk(KERN_ERR "ERROR: domain->groups does not contain"
6029 " CPU%d\n", cpu);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006030 }
6031
6032 printk(KERN_DEBUG "%*s groups:", level + 1, "");
6033 do {
6034 if (!group) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006035 printk("\n");
6036 printk(KERN_ERR "ERROR: group is NULL\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006037 break;
6038 }
6039
Peter Zijlstra18a38852009-09-01 10:34:39 +02006040 if (!group->cpu_power) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006041 printk(KERN_CONT "\n");
6042 printk(KERN_ERR "ERROR: domain->cpu_power not "
6043 "set\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006044 break;
6045 }
6046
Rusty Russell758b2cd2008-11-25 02:35:04 +10306047 if (!cpumask_weight(sched_group_cpus(group))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006048 printk(KERN_CONT "\n");
6049 printk(KERN_ERR "ERROR: empty group\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006050 break;
6051 }
6052
Rusty Russell758b2cd2008-11-25 02:35:04 +10306053 if (cpumask_intersects(groupmask, sched_group_cpus(group))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006054 printk(KERN_CONT "\n");
6055 printk(KERN_ERR "ERROR: repeated CPUs\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006056 break;
6057 }
6058
Rusty Russell758b2cd2008-11-25 02:35:04 +10306059 cpumask_or(groupmask, groupmask, sched_group_cpus(group));
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006060
Rusty Russell968ea6d2008-12-13 21:55:51 +10306061 cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
Gautham R Shenoy381512c2009-04-14 09:09:36 +05306062
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006063 printk(KERN_CONT " %s", str);
Peter Zijlstra18a38852009-09-01 10:34:39 +02006064 if (group->cpu_power != SCHED_LOAD_SCALE) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006065 printk(KERN_CONT " (cpu_power = %d)",
6066 group->cpu_power);
Gautham R Shenoy381512c2009-04-14 09:09:36 +05306067 }
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006068
6069 group = group->next;
6070 } while (group != sd->groups);
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006071 printk(KERN_CONT "\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006072
Rusty Russell758b2cd2008-11-25 02:35:04 +10306073 if (!cpumask_equal(sched_domain_span(sd), groupmask))
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006074 printk(KERN_ERR "ERROR: groups don't span domain->span\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006075
Rusty Russell758b2cd2008-11-25 02:35:04 +10306076 if (sd->parent &&
6077 !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006078 printk(KERN_ERR "ERROR: parent span is not a superset "
6079 "of domain->span\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006080 return 0;
6081}
6082
Linus Torvalds1da177e2005-04-16 15:20:36 -07006083static void sched_domain_debug(struct sched_domain *sd, int cpu)
6084{
Rusty Russelld5dd3db2008-11-25 02:35:12 +10306085 cpumask_var_t groupmask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006086 int level = 0;
6087
Mike Travisf6630112009-11-17 18:22:15 -06006088 if (!sched_domain_debug_enabled)
6089 return;
6090
Nick Piggin41c7ce92005-06-25 14:57:24 -07006091 if (!sd) {
6092 printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
6093 return;
6094 }
6095
Linus Torvalds1da177e2005-04-16 15:20:36 -07006096 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
6097
Rusty Russelld5dd3db2008-11-25 02:35:12 +10306098 if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) {
Mike Travis7c16ec52008-04-04 18:11:11 -07006099 printk(KERN_DEBUG "Cannot load-balance (out of memory)\n");
6100 return;
6101 }
6102
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006103 for (;;) {
Mike Travis7c16ec52008-04-04 18:11:11 -07006104 if (sched_domain_debug_one(sd, cpu, level, groupmask))
Linus Torvalds1da177e2005-04-16 15:20:36 -07006105 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006106 level++;
6107 sd = sd->parent;
Miguel Ojeda Sandonis33859f72006-12-10 02:20:38 -08006108 if (!sd)
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006109 break;
6110 }
Rusty Russelld5dd3db2008-11-25 02:35:12 +10306111 free_cpumask_var(groupmask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006112}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006113#else /* !CONFIG_SCHED_DEBUG */
Ingo Molnar48f24c42006-07-03 00:25:40 -07006114# define sched_domain_debug(sd, cpu) do { } while (0)
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006115#endif /* CONFIG_SCHED_DEBUG */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006116
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006117static int sd_degenerate(struct sched_domain *sd)
Suresh Siddha245af2c2005-06-25 14:57:25 -07006118{
Rusty Russell758b2cd2008-11-25 02:35:04 +10306119 if (cpumask_weight(sched_domain_span(sd)) == 1)
Suresh Siddha245af2c2005-06-25 14:57:25 -07006120 return 1;
6121
6122 /* Following flags need at least 2 groups */
6123 if (sd->flags & (SD_LOAD_BALANCE |
6124 SD_BALANCE_NEWIDLE |
6125 SD_BALANCE_FORK |
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006126 SD_BALANCE_EXEC |
6127 SD_SHARE_CPUPOWER |
6128 SD_SHARE_PKG_RESOURCES)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006129 if (sd->groups != sd->groups->next)
6130 return 0;
6131 }
6132
6133 /* Following flags don't use groups */
Peter Zijlstrac88d5912009-09-10 13:50:02 +02006134 if (sd->flags & (SD_WAKE_AFFINE))
Suresh Siddha245af2c2005-06-25 14:57:25 -07006135 return 0;
6136
6137 return 1;
6138}
6139
Ingo Molnar48f24c42006-07-03 00:25:40 -07006140static int
6141sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
Suresh Siddha245af2c2005-06-25 14:57:25 -07006142{
6143 unsigned long cflags = sd->flags, pflags = parent->flags;
6144
6145 if (sd_degenerate(parent))
6146 return 1;
6147
Rusty Russell758b2cd2008-11-25 02:35:04 +10306148 if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
Suresh Siddha245af2c2005-06-25 14:57:25 -07006149 return 0;
6150
Suresh Siddha245af2c2005-06-25 14:57:25 -07006151 /* Flags needing groups don't count if only 1 group in parent */
6152 if (parent->groups == parent->groups->next) {
6153 pflags &= ~(SD_LOAD_BALANCE |
6154 SD_BALANCE_NEWIDLE |
6155 SD_BALANCE_FORK |
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006156 SD_BALANCE_EXEC |
6157 SD_SHARE_CPUPOWER |
6158 SD_SHARE_PKG_RESOURCES);
Ken Chen54364992008-12-07 18:47:37 -08006159 if (nr_node_ids == 1)
6160 pflags &= ~SD_SERIALIZE;
Suresh Siddha245af2c2005-06-25 14:57:25 -07006161 }
6162 if (~cflags & pflags)
6163 return 0;
6164
6165 return 1;
6166}
6167
Rusty Russellc6c49272008-11-25 02:35:05 +10306168static void free_rootdomain(struct root_domain *rd)
6169{
Peter Zijlstra047106a2009-11-16 10:28:09 +01006170 synchronize_sched();
6171
Rusty Russell68e74562008-11-25 02:35:13 +10306172 cpupri_cleanup(&rd->cpupri);
6173
Rusty Russellc6c49272008-11-25 02:35:05 +10306174 free_cpumask_var(rd->rto_mask);
6175 free_cpumask_var(rd->online);
6176 free_cpumask_var(rd->span);
6177 kfree(rd);
6178}
6179
Gregory Haskins57d885f2008-01-25 21:08:18 +01006180static void rq_attach_root(struct rq *rq, struct root_domain *rd)
6181{
Ingo Molnara0490fa2009-02-12 11:35:40 +01006182 struct root_domain *old_rd = NULL;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006183 unsigned long flags;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006184
Thomas Gleixner05fa7852009-11-17 14:28:38 +01006185 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006186
6187 if (rq->rd) {
Ingo Molnara0490fa2009-02-12 11:35:40 +01006188 old_rd = rq->rd;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006189
Rusty Russellc6c49272008-11-25 02:35:05 +10306190 if (cpumask_test_cpu(rq->cpu, old_rd->online))
Gregory Haskins1f11eb62008-06-04 15:04:05 -04006191 set_rq_offline(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006192
Rusty Russellc6c49272008-11-25 02:35:05 +10306193 cpumask_clear_cpu(rq->cpu, old_rd->span);
Gregory Haskinsdc938522008-01-25 21:08:26 +01006194
Ingo Molnara0490fa2009-02-12 11:35:40 +01006195 /*
6196 * If we dont want to free the old_rt yet then
6197 * set old_rd to NULL to skip the freeing later
6198 * in this function:
6199 */
6200 if (!atomic_dec_and_test(&old_rd->refcount))
6201 old_rd = NULL;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006202 }
6203
6204 atomic_inc(&rd->refcount);
6205 rq->rd = rd;
6206
Rusty Russellc6c49272008-11-25 02:35:05 +10306207 cpumask_set_cpu(rq->cpu, rd->span);
Gregory Haskins00aec932009-07-30 10:57:23 -04006208 if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
Gregory Haskins1f11eb62008-06-04 15:04:05 -04006209 set_rq_online(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006210
Thomas Gleixner05fa7852009-11-17 14:28:38 +01006211 raw_spin_unlock_irqrestore(&rq->lock, flags);
Ingo Molnara0490fa2009-02-12 11:35:40 +01006212
6213 if (old_rd)
6214 free_rootdomain(old_rd);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006215}
6216
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006217static int init_rootdomain(struct root_domain *rd)
Gregory Haskins57d885f2008-01-25 21:08:18 +01006218{
6219 memset(rd, 0, sizeof(*rd));
6220
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006221 if (!alloc_cpumask_var(&rd->span, GFP_KERNEL))
Li Zefan0c910d22009-01-06 17:39:06 +08006222 goto out;
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006223 if (!alloc_cpumask_var(&rd->online, GFP_KERNEL))
Rusty Russellc6c49272008-11-25 02:35:05 +10306224 goto free_span;
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006225 if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
Rusty Russellc6c49272008-11-25 02:35:05 +10306226 goto free_online;
Gregory Haskins6e0534f2008-05-12 21:21:01 +02006227
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006228 if (cpupri_init(&rd->cpupri) != 0)
Rusty Russell68e74562008-11-25 02:35:13 +10306229 goto free_rto_mask;
Rusty Russellc6c49272008-11-25 02:35:05 +10306230 return 0;
6231
Rusty Russell68e74562008-11-25 02:35:13 +10306232free_rto_mask:
6233 free_cpumask_var(rd->rto_mask);
Rusty Russellc6c49272008-11-25 02:35:05 +10306234free_online:
6235 free_cpumask_var(rd->online);
6236free_span:
6237 free_cpumask_var(rd->span);
Li Zefan0c910d22009-01-06 17:39:06 +08006238out:
Rusty Russellc6c49272008-11-25 02:35:05 +10306239 return -ENOMEM;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006240}
6241
6242static void init_defrootdomain(void)
6243{
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006244 init_rootdomain(&def_root_domain);
Rusty Russellc6c49272008-11-25 02:35:05 +10306245
Gregory Haskins57d885f2008-01-25 21:08:18 +01006246 atomic_set(&def_root_domain.refcount, 1);
6247}
6248
Gregory Haskinsdc938522008-01-25 21:08:26 +01006249static struct root_domain *alloc_rootdomain(void)
Gregory Haskins57d885f2008-01-25 21:08:18 +01006250{
6251 struct root_domain *rd;
6252
6253 rd = kmalloc(sizeof(*rd), GFP_KERNEL);
6254 if (!rd)
6255 return NULL;
6256
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006257 if (init_rootdomain(rd) != 0) {
Rusty Russellc6c49272008-11-25 02:35:05 +10306258 kfree(rd);
6259 return NULL;
6260 }
Gregory Haskins57d885f2008-01-25 21:08:18 +01006261
6262 return rd;
6263}
6264
Linus Torvalds1da177e2005-04-16 15:20:36 -07006265/*
Ingo Molnar0eab9142008-01-25 21:08:19 +01006266 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
Linus Torvalds1da177e2005-04-16 15:20:36 -07006267 * hold the hotplug lock.
6268 */
Ingo Molnar0eab9142008-01-25 21:08:19 +01006269static void
6270cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006271{
Ingo Molnar70b97a72006-07-03 00:25:42 -07006272 struct rq *rq = cpu_rq(cpu);
Suresh Siddha245af2c2005-06-25 14:57:25 -07006273 struct sched_domain *tmp;
6274
Peter Zijlstra669c55e2010-04-16 14:59:29 +02006275 for (tmp = sd; tmp; tmp = tmp->parent)
6276 tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
6277
Suresh Siddha245af2c2005-06-25 14:57:25 -07006278 /* Remove the sched domains which do not contribute to scheduling. */
Li Zefanf29c9b12008-11-06 09:45:16 +08006279 for (tmp = sd; tmp; ) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006280 struct sched_domain *parent = tmp->parent;
6281 if (!parent)
6282 break;
Li Zefanf29c9b12008-11-06 09:45:16 +08006283
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006284 if (sd_parent_degenerate(tmp, parent)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006285 tmp->parent = parent->parent;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006286 if (parent->parent)
6287 parent->parent->child = tmp;
Li Zefanf29c9b12008-11-06 09:45:16 +08006288 } else
6289 tmp = tmp->parent;
Suresh Siddha245af2c2005-06-25 14:57:25 -07006290 }
6291
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006292 if (sd && sd_degenerate(sd)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006293 sd = sd->parent;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006294 if (sd)
6295 sd->child = NULL;
6296 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006297
6298 sched_domain_debug(sd, cpu);
6299
Gregory Haskins57d885f2008-01-25 21:08:18 +01006300 rq_attach_root(rq, rd);
Nick Piggin674311d2005-06-25 14:57:27 -07006301 rcu_assign_pointer(rq->sd, sd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006302}
6303
6304/* cpus with isolated domains */
Rusty Russelldcc30a32008-11-25 02:35:12 +10306305static cpumask_var_t cpu_isolated_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006306
6307/* Setup the mask of cpus configured for isolated domains */
6308static int __init isolated_cpu_setup(char *str)
6309{
Rusty Russellbdddd292009-12-02 14:09:16 +10306310 alloc_bootmem_cpumask_var(&cpu_isolated_map);
Rusty Russell968ea6d2008-12-13 21:55:51 +10306311 cpulist_parse(str, cpu_isolated_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006312 return 1;
6313}
6314
Ingo Molnar8927f492007-10-15 17:00:13 +02006315__setup("isolcpus=", isolated_cpu_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006316
6317/*
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006318 * init_sched_build_groups takes the cpumask we wish to span, and a pointer
6319 * to a function which identifies what group(along with sched group) a CPU
Rusty Russell96f874e2008-11-25 02:35:14 +10306320 * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids
6321 * (due to the fact that we keep track of groups covered with a struct cpumask).
Linus Torvalds1da177e2005-04-16 15:20:36 -07006322 *
6323 * init_sched_build_groups will build a circular linked list of the groups
6324 * covered by the given span, and will set each group's ->cpumask correctly,
6325 * and ->cpu_power to 0.
6326 */
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006327static void
Rusty Russell96f874e2008-11-25 02:35:14 +10306328init_sched_build_groups(const struct cpumask *span,
6329 const struct cpumask *cpu_map,
6330 int (*group_fn)(int cpu, const struct cpumask *cpu_map,
Mike Travis7c16ec52008-04-04 18:11:11 -07006331 struct sched_group **sg,
Rusty Russell96f874e2008-11-25 02:35:14 +10306332 struct cpumask *tmpmask),
6333 struct cpumask *covered, struct cpumask *tmpmask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006334{
6335 struct sched_group *first = NULL, *last = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006336 int i;
6337
Rusty Russell96f874e2008-11-25 02:35:14 +10306338 cpumask_clear(covered);
Mike Travis7c16ec52008-04-04 18:11:11 -07006339
Rusty Russellabcd0832008-11-25 02:35:02 +10306340 for_each_cpu(i, span) {
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006341 struct sched_group *sg;
Mike Travis7c16ec52008-04-04 18:11:11 -07006342 int group = group_fn(i, cpu_map, &sg, tmpmask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006343 int j;
6344
Rusty Russell758b2cd2008-11-25 02:35:04 +10306345 if (cpumask_test_cpu(i, covered))
Linus Torvalds1da177e2005-04-16 15:20:36 -07006346 continue;
6347
Rusty Russell758b2cd2008-11-25 02:35:04 +10306348 cpumask_clear(sched_group_cpus(sg));
Peter Zijlstra18a38852009-09-01 10:34:39 +02006349 sg->cpu_power = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006350
Rusty Russellabcd0832008-11-25 02:35:02 +10306351 for_each_cpu(j, span) {
Mike Travis7c16ec52008-04-04 18:11:11 -07006352 if (group_fn(j, cpu_map, NULL, tmpmask) != group)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006353 continue;
6354
Rusty Russell96f874e2008-11-25 02:35:14 +10306355 cpumask_set_cpu(j, covered);
Rusty Russell758b2cd2008-11-25 02:35:04 +10306356 cpumask_set_cpu(j, sched_group_cpus(sg));
Linus Torvalds1da177e2005-04-16 15:20:36 -07006357 }
6358 if (!first)
6359 first = sg;
6360 if (last)
6361 last->next = sg;
6362 last = sg;
6363 }
6364 last->next = first;
6365}
6366
John Hawkes9c1cfda2005-09-06 15:18:14 -07006367#define SD_NODES_PER_DOMAIN 16
Linus Torvalds1da177e2005-04-16 15:20:36 -07006368
John Hawkes9c1cfda2005-09-06 15:18:14 -07006369#ifdef CONFIG_NUMA
akpm@osdl.org198e2f12006-01-12 01:05:30 -08006370
John Hawkes9c1cfda2005-09-06 15:18:14 -07006371/**
6372 * find_next_best_node - find the next node to include in a sched_domain
6373 * @node: node whose sched_domain we're building
6374 * @used_nodes: nodes already in the sched_domain
6375 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006376 * Find the next node to include in a given scheduling domain. Simply
John Hawkes9c1cfda2005-09-06 15:18:14 -07006377 * finds the closest node not already in the @used_nodes map.
6378 *
6379 * Should use nodemask_t.
6380 */
Mike Travisc5f59f02008-04-04 18:11:10 -07006381static int find_next_best_node(int node, nodemask_t *used_nodes)
John Hawkes9c1cfda2005-09-06 15:18:14 -07006382{
6383 int i, n, val, min_val, best_node = 0;
6384
6385 min_val = INT_MAX;
6386
Mike Travis076ac2a2008-05-12 21:21:12 +02006387 for (i = 0; i < nr_node_ids; i++) {
John Hawkes9c1cfda2005-09-06 15:18:14 -07006388 /* Start at @node */
Mike Travis076ac2a2008-05-12 21:21:12 +02006389 n = (node + i) % nr_node_ids;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006390
6391 if (!nr_cpus_node(n))
6392 continue;
6393
6394 /* Skip already used nodes */
Mike Travisc5f59f02008-04-04 18:11:10 -07006395 if (node_isset(n, *used_nodes))
John Hawkes9c1cfda2005-09-06 15:18:14 -07006396 continue;
6397
6398 /* Simple min distance search */
6399 val = node_distance(node, n);
6400
6401 if (val < min_val) {
6402 min_val = val;
6403 best_node = n;
6404 }
6405 }
6406
Mike Travisc5f59f02008-04-04 18:11:10 -07006407 node_set(best_node, *used_nodes);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006408 return best_node;
6409}
6410
6411/**
6412 * sched_domain_node_span - get a cpumask for a node's sched_domain
6413 * @node: node whose cpumask we're constructing
Randy Dunlap73486722008-04-22 10:07:22 -07006414 * @span: resulting cpumask
John Hawkes9c1cfda2005-09-06 15:18:14 -07006415 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006416 * Given a node, construct a good cpumask for its sched_domain to span. It
John Hawkes9c1cfda2005-09-06 15:18:14 -07006417 * should be one that prevents unnecessary balancing, but also spreads tasks
6418 * out optimally.
6419 */
Rusty Russell96f874e2008-11-25 02:35:14 +10306420static void sched_domain_node_span(int node, struct cpumask *span)
John Hawkes9c1cfda2005-09-06 15:18:14 -07006421{
Mike Travisc5f59f02008-04-04 18:11:10 -07006422 nodemask_t used_nodes;
Ingo Molnar48f24c42006-07-03 00:25:40 -07006423 int i;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006424
Mike Travis6ca09df2008-12-31 18:08:45 -08006425 cpumask_clear(span);
Mike Travisc5f59f02008-04-04 18:11:10 -07006426 nodes_clear(used_nodes);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006427
Mike Travis6ca09df2008-12-31 18:08:45 -08006428 cpumask_or(span, span, cpumask_of_node(node));
Mike Travisc5f59f02008-04-04 18:11:10 -07006429 node_set(node, used_nodes);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006430
6431 for (i = 1; i < SD_NODES_PER_DOMAIN; i++) {
Mike Travisc5f59f02008-04-04 18:11:10 -07006432 int next_node = find_next_best_node(node, &used_nodes);
Ingo Molnar48f24c42006-07-03 00:25:40 -07006433
Mike Travis6ca09df2008-12-31 18:08:45 -08006434 cpumask_or(span, span, cpumask_of_node(next_node));
John Hawkes9c1cfda2005-09-06 15:18:14 -07006435 }
John Hawkes9c1cfda2005-09-06 15:18:14 -07006436}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006437#endif /* CONFIG_NUMA */
John Hawkes9c1cfda2005-09-06 15:18:14 -07006438
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006439int sched_smt_power_savings = 0, sched_mc_power_savings = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07006440
John Hawkes9c1cfda2005-09-06 15:18:14 -07006441/*
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306442 * The cpus mask in sched_group and sched_domain hangs off the end.
Ingo Molnar4200efd2009-05-19 09:22:19 +02006443 *
6444 * ( See the the comments in include/linux/sched.h:struct sched_group
6445 * and struct sched_domain. )
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306446 */
6447struct static_sched_group {
6448 struct sched_group sg;
6449 DECLARE_BITMAP(cpus, CONFIG_NR_CPUS);
6450};
6451
6452struct static_sched_domain {
6453 struct sched_domain sd;
6454 DECLARE_BITMAP(span, CONFIG_NR_CPUS);
6455};
6456
Andreas Herrmann49a02c52009-08-18 12:51:52 +02006457struct s_data {
6458#ifdef CONFIG_NUMA
6459 int sd_allnodes;
6460 cpumask_var_t domainspan;
6461 cpumask_var_t covered;
6462 cpumask_var_t notcovered;
6463#endif
6464 cpumask_var_t nodemask;
6465 cpumask_var_t this_sibling_map;
6466 cpumask_var_t this_core_map;
Heiko Carstens01a08542010-08-31 10:28:16 +02006467 cpumask_var_t this_book_map;
Andreas Herrmann49a02c52009-08-18 12:51:52 +02006468 cpumask_var_t send_covered;
6469 cpumask_var_t tmpmask;
6470 struct sched_group **sched_group_nodes;
6471 struct root_domain *rd;
6472};
6473
Andreas Herrmann2109b992009-08-18 12:53:00 +02006474enum s_alloc {
6475 sa_sched_groups = 0,
6476 sa_rootdomain,
6477 sa_tmpmask,
6478 sa_send_covered,
Heiko Carstens01a08542010-08-31 10:28:16 +02006479 sa_this_book_map,
Andreas Herrmann2109b992009-08-18 12:53:00 +02006480 sa_this_core_map,
6481 sa_this_sibling_map,
6482 sa_nodemask,
6483 sa_sched_group_nodes,
6484#ifdef CONFIG_NUMA
6485 sa_notcovered,
6486 sa_covered,
6487 sa_domainspan,
6488#endif
6489 sa_none,
6490};
6491
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306492/*
Ingo Molnar48f24c42006-07-03 00:25:40 -07006493 * SMT sched-domains:
John Hawkes9c1cfda2005-09-06 15:18:14 -07006494 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006495#ifdef CONFIG_SCHED_SMT
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306496static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains);
Tejun Heo1871e522009-10-29 22:34:13 +09006497static DEFINE_PER_CPU(struct static_sched_group, sched_groups);
Ingo Molnar48f24c42006-07-03 00:25:40 -07006498
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006499static int
Rusty Russell96f874e2008-11-25 02:35:14 +10306500cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map,
6501 struct sched_group **sg, struct cpumask *unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006502{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006503 if (sg)
Tejun Heo1871e522009-10-29 22:34:13 +09006504 *sg = &per_cpu(sched_groups, cpu).sg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006505 return cpu;
6506}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006507#endif /* CONFIG_SCHED_SMT */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006508
Ingo Molnar48f24c42006-07-03 00:25:40 -07006509/*
6510 * multi-core sched-domains:
6511 */
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006512#ifdef CONFIG_SCHED_MC
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306513static DEFINE_PER_CPU(struct static_sched_domain, core_domains);
6514static DEFINE_PER_CPU(struct static_sched_group, sched_group_core);
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006515
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006516static int
Rusty Russell96f874e2008-11-25 02:35:14 +10306517cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
6518 struct sched_group **sg, struct cpumask *mask)
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006519{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006520 int group;
Heiko Carstensf2698932010-08-31 10:28:15 +02006521#ifdef CONFIG_SCHED_SMT
Rusty Russellc69fc562009-03-13 14:49:46 +10306522 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
Rusty Russell96f874e2008-11-25 02:35:14 +10306523 group = cpumask_first(mask);
Heiko Carstensf2698932010-08-31 10:28:15 +02006524#else
6525 group = cpu;
6526#endif
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006527 if (sg)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306528 *sg = &per_cpu(sched_group_core, group).sg;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006529 return group;
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006530}
Heiko Carstensf2698932010-08-31 10:28:15 +02006531#endif /* CONFIG_SCHED_MC */
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006532
Heiko Carstens01a08542010-08-31 10:28:16 +02006533/*
6534 * book sched-domains:
6535 */
6536#ifdef CONFIG_SCHED_BOOK
6537static DEFINE_PER_CPU(struct static_sched_domain, book_domains);
6538static DEFINE_PER_CPU(struct static_sched_group, sched_group_book);
6539
Linus Torvalds1da177e2005-04-16 15:20:36 -07006540static int
Heiko Carstens01a08542010-08-31 10:28:16 +02006541cpu_to_book_group(int cpu, const struct cpumask *cpu_map,
6542 struct sched_group **sg, struct cpumask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006543{
Heiko Carstens01a08542010-08-31 10:28:16 +02006544 int group = cpu;
6545#ifdef CONFIG_SCHED_MC
6546 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
6547 group = cpumask_first(mask);
6548#elif defined(CONFIG_SCHED_SMT)
6549 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
6550 group = cpumask_first(mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006551#endif
Heiko Carstens01a08542010-08-31 10:28:16 +02006552 if (sg)
6553 *sg = &per_cpu(sched_group_book, group).sg;
6554 return group;
6555}
6556#endif /* CONFIG_SCHED_BOOK */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006557
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306558static DEFINE_PER_CPU(struct static_sched_domain, phys_domains);
6559static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys);
Ingo Molnar48f24c42006-07-03 00:25:40 -07006560
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006561static int
Rusty Russell96f874e2008-11-25 02:35:14 +10306562cpu_to_phys_group(int cpu, const struct cpumask *cpu_map,
6563 struct sched_group **sg, struct cpumask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006564{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006565 int group;
Heiko Carstens01a08542010-08-31 10:28:16 +02006566#ifdef CONFIG_SCHED_BOOK
6567 cpumask_and(mask, cpu_book_mask(cpu), cpu_map);
6568 group = cpumask_first(mask);
6569#elif defined(CONFIG_SCHED_MC)
Mike Travis6ca09df2008-12-31 18:08:45 -08006570 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
Rusty Russell96f874e2008-11-25 02:35:14 +10306571 group = cpumask_first(mask);
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006572#elif defined(CONFIG_SCHED_SMT)
Rusty Russellc69fc562009-03-13 14:49:46 +10306573 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
Rusty Russell96f874e2008-11-25 02:35:14 +10306574 group = cpumask_first(mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006575#else
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006576 group = cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006577#endif
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006578 if (sg)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306579 *sg = &per_cpu(sched_group_phys, group).sg;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006580 return group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006581}
6582
6583#ifdef CONFIG_NUMA
John Hawkes9c1cfda2005-09-06 15:18:14 -07006584/*
6585 * The init_sched_build_groups can't handle what we want to do with node
6586 * groups, so roll our own. Now each node has its own list of groups which
6587 * gets dynamically allocated.
6588 */
Rusty Russell62ea9ce2009-01-11 01:04:16 +01006589static DEFINE_PER_CPU(struct static_sched_domain, node_domains);
Mike Travis434d53b2008-04-04 18:11:04 -07006590static struct sched_group ***sched_group_nodes_bycpu;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006591
Rusty Russell62ea9ce2009-01-11 01:04:16 +01006592static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains);
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306593static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006594
Rusty Russell96f874e2008-11-25 02:35:14 +10306595static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map,
6596 struct sched_group **sg,
6597 struct cpumask *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006598{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006599 int group;
6600
Mike Travis6ca09df2008-12-31 18:08:45 -08006601 cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map);
Rusty Russell96f874e2008-11-25 02:35:14 +10306602 group = cpumask_first(nodemask);
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006603
6604 if (sg)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306605 *sg = &per_cpu(sched_group_allnodes, group).sg;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006606 return group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006607}
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006608
Siddha, Suresh B08069032006-03-27 01:15:23 -08006609static void init_numa_sched_groups_power(struct sched_group *group_head)
6610{
6611 struct sched_group *sg = group_head;
6612 int j;
6613
6614 if (!sg)
6615 return;
Andi Kleen3a5c3592007-10-15 17:00:14 +02006616 do {
Rusty Russell758b2cd2008-11-25 02:35:04 +10306617 for_each_cpu(j, sched_group_cpus(sg)) {
Andi Kleen3a5c3592007-10-15 17:00:14 +02006618 struct sched_domain *sd;
Siddha, Suresh B08069032006-03-27 01:15:23 -08006619
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306620 sd = &per_cpu(phys_domains, j).sd;
Miao Xie13318a72009-04-15 09:59:10 +08006621 if (j != group_first_cpu(sd->groups)) {
Andi Kleen3a5c3592007-10-15 17:00:14 +02006622 /*
6623 * Only add "power" once for each
6624 * physical package.
6625 */
6626 continue;
6627 }
6628
Peter Zijlstra18a38852009-09-01 10:34:39 +02006629 sg->cpu_power += sd->groups->cpu_power;
Siddha, Suresh B08069032006-03-27 01:15:23 -08006630 }
Andi Kleen3a5c3592007-10-15 17:00:14 +02006631 sg = sg->next;
6632 } while (sg != group_head);
Siddha, Suresh B08069032006-03-27 01:15:23 -08006633}
Andreas Herrmann0601a882009-08-18 13:01:11 +02006634
6635static int build_numa_sched_groups(struct s_data *d,
6636 const struct cpumask *cpu_map, int num)
6637{
6638 struct sched_domain *sd;
6639 struct sched_group *sg, *prev;
6640 int n, j;
6641
6642 cpumask_clear(d->covered);
6643 cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map);
6644 if (cpumask_empty(d->nodemask)) {
6645 d->sched_group_nodes[num] = NULL;
6646 goto out;
6647 }
6648
6649 sched_domain_node_span(num, d->domainspan);
6650 cpumask_and(d->domainspan, d->domainspan, cpu_map);
6651
6652 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
6653 GFP_KERNEL, num);
6654 if (!sg) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006655 printk(KERN_WARNING "Can not alloc domain group for node %d\n",
6656 num);
Andreas Herrmann0601a882009-08-18 13:01:11 +02006657 return -ENOMEM;
6658 }
6659 d->sched_group_nodes[num] = sg;
6660
6661 for_each_cpu(j, d->nodemask) {
6662 sd = &per_cpu(node_domains, j).sd;
6663 sd->groups = sg;
6664 }
6665
Peter Zijlstra18a38852009-09-01 10:34:39 +02006666 sg->cpu_power = 0;
Andreas Herrmann0601a882009-08-18 13:01:11 +02006667 cpumask_copy(sched_group_cpus(sg), d->nodemask);
6668 sg->next = sg;
6669 cpumask_or(d->covered, d->covered, d->nodemask);
6670
6671 prev = sg;
6672 for (j = 0; j < nr_node_ids; j++) {
6673 n = (num + j) % nr_node_ids;
6674 cpumask_complement(d->notcovered, d->covered);
6675 cpumask_and(d->tmpmask, d->notcovered, cpu_map);
6676 cpumask_and(d->tmpmask, d->tmpmask, d->domainspan);
6677 if (cpumask_empty(d->tmpmask))
6678 break;
6679 cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n));
6680 if (cpumask_empty(d->tmpmask))
6681 continue;
6682 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
6683 GFP_KERNEL, num);
6684 if (!sg) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006685 printk(KERN_WARNING
6686 "Can not alloc domain group for node %d\n", j);
Andreas Herrmann0601a882009-08-18 13:01:11 +02006687 return -ENOMEM;
6688 }
Peter Zijlstra18a38852009-09-01 10:34:39 +02006689 sg->cpu_power = 0;
Andreas Herrmann0601a882009-08-18 13:01:11 +02006690 cpumask_copy(sched_group_cpus(sg), d->tmpmask);
6691 sg->next = prev->next;
6692 cpumask_or(d->covered, d->covered, d->tmpmask);
6693 prev->next = sg;
6694 prev = sg;
6695 }
6696out:
6697 return 0;
6698}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006699#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006700
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006701#ifdef CONFIG_NUMA
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006702/* Free memory allocated for various sched_group structures */
Rusty Russell96f874e2008-11-25 02:35:14 +10306703static void free_sched_groups(const struct cpumask *cpu_map,
6704 struct cpumask *nodemask)
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006705{
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006706 int cpu, i;
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006707
Rusty Russellabcd0832008-11-25 02:35:02 +10306708 for_each_cpu(cpu, cpu_map) {
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006709 struct sched_group **sched_group_nodes
6710 = sched_group_nodes_bycpu[cpu];
6711
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006712 if (!sched_group_nodes)
6713 continue;
6714
Mike Travis076ac2a2008-05-12 21:21:12 +02006715 for (i = 0; i < nr_node_ids; i++) {
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006716 struct sched_group *oldsg, *sg = sched_group_nodes[i];
6717
Mike Travis6ca09df2008-12-31 18:08:45 -08006718 cpumask_and(nodemask, cpumask_of_node(i), cpu_map);
Rusty Russell96f874e2008-11-25 02:35:14 +10306719 if (cpumask_empty(nodemask))
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006720 continue;
6721
6722 if (sg == NULL)
6723 continue;
6724 sg = sg->next;
6725next_sg:
6726 oldsg = sg;
6727 sg = sg->next;
6728 kfree(oldsg);
6729 if (oldsg != sched_group_nodes[i])
6730 goto next_sg;
6731 }
6732 kfree(sched_group_nodes);
6733 sched_group_nodes_bycpu[cpu] = NULL;
6734 }
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006735}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006736#else /* !CONFIG_NUMA */
Rusty Russell96f874e2008-11-25 02:35:14 +10306737static void free_sched_groups(const struct cpumask *cpu_map,
6738 struct cpumask *nodemask)
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006739{
6740}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006741#endif /* CONFIG_NUMA */
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006742
Linus Torvalds1da177e2005-04-16 15:20:36 -07006743/*
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006744 * Initialize sched groups cpu_power.
6745 *
6746 * cpu_power indicates the capacity of sched group, which is used while
6747 * distributing the load between different sched groups in a sched domain.
6748 * Typically cpu_power for all the groups in a sched domain will be same unless
6749 * there are asymmetries in the topology. If there are asymmetries, group
6750 * having more cpu_power will pickup more load compared to the group having
6751 * less cpu_power.
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006752 */
6753static void init_sched_groups_power(int cpu, struct sched_domain *sd)
6754{
6755 struct sched_domain *child;
6756 struct sched_group *group;
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006757 long power;
6758 int weight;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006759
6760 WARN_ON(!sd || !sd->groups);
6761
Miao Xie13318a72009-04-15 09:59:10 +08006762 if (cpu != group_first_cpu(sd->groups))
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006763 return;
6764
Suresh Siddhaaae6d3d2010-09-17 15:02:32 -07006765 sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups));
6766
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006767 child = sd->child;
6768
Peter Zijlstra18a38852009-09-01 10:34:39 +02006769 sd->groups->cpu_power = 0;
Eric Dumazet5517d862007-05-08 00:32:57 -07006770
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006771 if (!child) {
6772 power = SCHED_LOAD_SCALE;
6773 weight = cpumask_weight(sched_domain_span(sd));
6774 /*
6775 * SMT siblings share the power of a single core.
Peter Zijlstraa52bfd72009-09-01 10:34:35 +02006776 * Usually multiple threads get a better yield out of
6777 * that one core than a single thread would have,
6778 * reflect that in sd->smt_gain.
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006779 */
Peter Zijlstraa52bfd72009-09-01 10:34:35 +02006780 if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) {
6781 power *= sd->smt_gain;
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006782 power /= weight;
Peter Zijlstraa52bfd72009-09-01 10:34:35 +02006783 power >>= SCHED_LOAD_SHIFT;
6784 }
Peter Zijlstra18a38852009-09-01 10:34:39 +02006785 sd->groups->cpu_power += power;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006786 return;
6787 }
6788
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006789 /*
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006790 * Add cpu_power of each child group to this groups cpu_power.
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006791 */
6792 group = child->groups;
6793 do {
Peter Zijlstra18a38852009-09-01 10:34:39 +02006794 sd->groups->cpu_power += group->cpu_power;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006795 group = group->next;
6796 } while (group != child->groups);
6797}
6798
6799/*
Mike Travis7c16ec52008-04-04 18:11:11 -07006800 * Initializers for schedule domains
6801 * Non-inlined to reduce accumulated stack pressure in build_sched_domains()
6802 */
6803
Ingo Molnara5d8c342008-10-09 11:35:51 +02006804#ifdef CONFIG_SCHED_DEBUG
6805# define SD_INIT_NAME(sd, type) sd->name = #type
6806#else
6807# define SD_INIT_NAME(sd, type) do { } while (0)
6808#endif
6809
Mike Travis7c16ec52008-04-04 18:11:11 -07006810#define SD_INIT(sd, type) sd_init_##type(sd)
Ingo Molnara5d8c342008-10-09 11:35:51 +02006811
Mike Travis7c16ec52008-04-04 18:11:11 -07006812#define SD_INIT_FUNC(type) \
6813static noinline void sd_init_##type(struct sched_domain *sd) \
6814{ \
6815 memset(sd, 0, sizeof(*sd)); \
6816 *sd = SD_##type##_INIT; \
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006817 sd->level = SD_LV_##type; \
Ingo Molnara5d8c342008-10-09 11:35:51 +02006818 SD_INIT_NAME(sd, type); \
Mike Travis7c16ec52008-04-04 18:11:11 -07006819}
6820
6821SD_INIT_FUNC(CPU)
6822#ifdef CONFIG_NUMA
6823 SD_INIT_FUNC(ALLNODES)
6824 SD_INIT_FUNC(NODE)
6825#endif
6826#ifdef CONFIG_SCHED_SMT
6827 SD_INIT_FUNC(SIBLING)
6828#endif
6829#ifdef CONFIG_SCHED_MC
6830 SD_INIT_FUNC(MC)
6831#endif
Heiko Carstens01a08542010-08-31 10:28:16 +02006832#ifdef CONFIG_SCHED_BOOK
6833 SD_INIT_FUNC(BOOK)
6834#endif
Mike Travis7c16ec52008-04-04 18:11:11 -07006835
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006836static int default_relax_domain_level = -1;
6837
6838static int __init setup_relax_domain_level(char *str)
6839{
Li Zefan30e0e172008-05-13 10:27:17 +08006840 unsigned long val;
6841
6842 val = simple_strtoul(str, NULL, 0);
6843 if (val < SD_LV_MAX)
6844 default_relax_domain_level = val;
6845
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006846 return 1;
6847}
6848__setup("relax_domain_level=", setup_relax_domain_level);
6849
6850static void set_domain_attribute(struct sched_domain *sd,
6851 struct sched_domain_attr *attr)
6852{
6853 int request;
6854
6855 if (!attr || attr->relax_domain_level < 0) {
6856 if (default_relax_domain_level < 0)
6857 return;
6858 else
6859 request = default_relax_domain_level;
6860 } else
6861 request = attr->relax_domain_level;
6862 if (request < sd->level) {
6863 /* turn off idle balance on this domain */
Peter Zijlstrac88d5912009-09-10 13:50:02 +02006864 sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006865 } else {
6866 /* turn on idle balance on this domain */
Peter Zijlstrac88d5912009-09-10 13:50:02 +02006867 sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006868 }
6869}
6870
Andreas Herrmann2109b992009-08-18 12:53:00 +02006871static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
6872 const struct cpumask *cpu_map)
6873{
6874 switch (what) {
6875 case sa_sched_groups:
6876 free_sched_groups(cpu_map, d->tmpmask); /* fall through */
6877 d->sched_group_nodes = NULL;
6878 case sa_rootdomain:
6879 free_rootdomain(d->rd); /* fall through */
6880 case sa_tmpmask:
6881 free_cpumask_var(d->tmpmask); /* fall through */
6882 case sa_send_covered:
6883 free_cpumask_var(d->send_covered); /* fall through */
Heiko Carstens01a08542010-08-31 10:28:16 +02006884 case sa_this_book_map:
6885 free_cpumask_var(d->this_book_map); /* fall through */
Andreas Herrmann2109b992009-08-18 12:53:00 +02006886 case sa_this_core_map:
6887 free_cpumask_var(d->this_core_map); /* fall through */
6888 case sa_this_sibling_map:
6889 free_cpumask_var(d->this_sibling_map); /* fall through */
6890 case sa_nodemask:
6891 free_cpumask_var(d->nodemask); /* fall through */
6892 case sa_sched_group_nodes:
6893#ifdef CONFIG_NUMA
6894 kfree(d->sched_group_nodes); /* fall through */
6895 case sa_notcovered:
6896 free_cpumask_var(d->notcovered); /* fall through */
6897 case sa_covered:
6898 free_cpumask_var(d->covered); /* fall through */
6899 case sa_domainspan:
6900 free_cpumask_var(d->domainspan); /* fall through */
6901#endif
6902 case sa_none:
6903 break;
6904 }
6905}
6906
6907static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
6908 const struct cpumask *cpu_map)
6909{
6910#ifdef CONFIG_NUMA
6911 if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL))
6912 return sa_none;
6913 if (!alloc_cpumask_var(&d->covered, GFP_KERNEL))
6914 return sa_domainspan;
6915 if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL))
6916 return sa_covered;
6917 /* Allocate the per-node list of sched groups */
6918 d->sched_group_nodes = kcalloc(nr_node_ids,
6919 sizeof(struct sched_group *), GFP_KERNEL);
6920 if (!d->sched_group_nodes) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006921 printk(KERN_WARNING "Can not alloc sched group node list\n");
Andreas Herrmann2109b992009-08-18 12:53:00 +02006922 return sa_notcovered;
6923 }
6924 sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes;
6925#endif
6926 if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL))
6927 return sa_sched_group_nodes;
6928 if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL))
6929 return sa_nodemask;
6930 if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL))
6931 return sa_this_sibling_map;
Heiko Carstens01a08542010-08-31 10:28:16 +02006932 if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL))
Andreas Herrmann2109b992009-08-18 12:53:00 +02006933 return sa_this_core_map;
Heiko Carstens01a08542010-08-31 10:28:16 +02006934 if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL))
6935 return sa_this_book_map;
Andreas Herrmann2109b992009-08-18 12:53:00 +02006936 if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL))
6937 return sa_send_covered;
6938 d->rd = alloc_rootdomain();
6939 if (!d->rd) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006940 printk(KERN_WARNING "Cannot alloc root domain\n");
Andreas Herrmann2109b992009-08-18 12:53:00 +02006941 return sa_tmpmask;
6942 }
6943 return sa_rootdomain;
6944}
6945
Andreas Herrmann7f4588f2009-08-18 12:54:06 +02006946static struct sched_domain *__build_numa_sched_domains(struct s_data *d,
6947 const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i)
6948{
6949 struct sched_domain *sd = NULL;
6950#ifdef CONFIG_NUMA
6951 struct sched_domain *parent;
6952
6953 d->sd_allnodes = 0;
6954 if (cpumask_weight(cpu_map) >
6955 SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) {
6956 sd = &per_cpu(allnodes_domains, i).sd;
6957 SD_INIT(sd, ALLNODES);
6958 set_domain_attribute(sd, attr);
6959 cpumask_copy(sched_domain_span(sd), cpu_map);
6960 cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask);
6961 d->sd_allnodes = 1;
6962 }
6963 parent = sd;
6964
6965 sd = &per_cpu(node_domains, i).sd;
6966 SD_INIT(sd, NODE);
6967 set_domain_attribute(sd, attr);
6968 sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd));
6969 sd->parent = parent;
6970 if (parent)
6971 parent->child = sd;
6972 cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map);
6973#endif
6974 return sd;
6975}
6976
Andreas Herrmann87cce662009-08-18 12:54:55 +02006977static struct sched_domain *__build_cpu_sched_domain(struct s_data *d,
6978 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
6979 struct sched_domain *parent, int i)
6980{
6981 struct sched_domain *sd;
6982 sd = &per_cpu(phys_domains, i).sd;
6983 SD_INIT(sd, CPU);
6984 set_domain_attribute(sd, attr);
6985 cpumask_copy(sched_domain_span(sd), d->nodemask);
6986 sd->parent = parent;
6987 if (parent)
6988 parent->child = sd;
6989 cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask);
6990 return sd;
6991}
6992
Heiko Carstens01a08542010-08-31 10:28:16 +02006993static struct sched_domain *__build_book_sched_domain(struct s_data *d,
6994 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
6995 struct sched_domain *parent, int i)
6996{
6997 struct sched_domain *sd = parent;
6998#ifdef CONFIG_SCHED_BOOK
6999 sd = &per_cpu(book_domains, i).sd;
7000 SD_INIT(sd, BOOK);
7001 set_domain_attribute(sd, attr);
7002 cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i));
7003 sd->parent = parent;
7004 parent->child = sd;
7005 cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask);
7006#endif
7007 return sd;
7008}
7009
Andreas Herrmann410c4082009-08-18 12:56:14 +02007010static struct sched_domain *__build_mc_sched_domain(struct s_data *d,
7011 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7012 struct sched_domain *parent, int i)
7013{
7014 struct sched_domain *sd = parent;
7015#ifdef CONFIG_SCHED_MC
7016 sd = &per_cpu(core_domains, i).sd;
7017 SD_INIT(sd, MC);
7018 set_domain_attribute(sd, attr);
7019 cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i));
7020 sd->parent = parent;
7021 parent->child = sd;
7022 cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask);
7023#endif
7024 return sd;
7025}
7026
Andreas Herrmannd8173532009-08-18 12:57:03 +02007027static struct sched_domain *__build_smt_sched_domain(struct s_data *d,
7028 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7029 struct sched_domain *parent, int i)
7030{
7031 struct sched_domain *sd = parent;
7032#ifdef CONFIG_SCHED_SMT
7033 sd = &per_cpu(cpu_domains, i).sd;
7034 SD_INIT(sd, SIBLING);
7035 set_domain_attribute(sd, attr);
7036 cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i));
7037 sd->parent = parent;
7038 parent->child = sd;
7039 cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask);
7040#endif
7041 return sd;
7042}
7043
Andreas Herrmann0e8e85c2009-08-18 12:57:51 +02007044static void build_sched_groups(struct s_data *d, enum sched_domain_level l,
7045 const struct cpumask *cpu_map, int cpu)
7046{
7047 switch (l) {
7048#ifdef CONFIG_SCHED_SMT
7049 case SD_LV_SIBLING: /* set up CPU (sibling) groups */
7050 cpumask_and(d->this_sibling_map, cpu_map,
7051 topology_thread_cpumask(cpu));
7052 if (cpu == cpumask_first(d->this_sibling_map))
7053 init_sched_build_groups(d->this_sibling_map, cpu_map,
7054 &cpu_to_cpu_group,
7055 d->send_covered, d->tmpmask);
7056 break;
7057#endif
Andreas Herrmanna2af04c2009-08-18 12:58:38 +02007058#ifdef CONFIG_SCHED_MC
7059 case SD_LV_MC: /* set up multi-core groups */
7060 cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu));
7061 if (cpu == cpumask_first(d->this_core_map))
7062 init_sched_build_groups(d->this_core_map, cpu_map,
7063 &cpu_to_core_group,
7064 d->send_covered, d->tmpmask);
7065 break;
7066#endif
Heiko Carstens01a08542010-08-31 10:28:16 +02007067#ifdef CONFIG_SCHED_BOOK
7068 case SD_LV_BOOK: /* set up book groups */
7069 cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu));
7070 if (cpu == cpumask_first(d->this_book_map))
7071 init_sched_build_groups(d->this_book_map, cpu_map,
7072 &cpu_to_book_group,
7073 d->send_covered, d->tmpmask);
7074 break;
7075#endif
Andreas Herrmann86548092009-08-18 12:59:28 +02007076 case SD_LV_CPU: /* set up physical groups */
7077 cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map);
7078 if (!cpumask_empty(d->nodemask))
7079 init_sched_build_groups(d->nodemask, cpu_map,
7080 &cpu_to_phys_group,
7081 d->send_covered, d->tmpmask);
7082 break;
Andreas Herrmannde616e32009-08-18 13:00:13 +02007083#ifdef CONFIG_NUMA
7084 case SD_LV_ALLNODES:
7085 init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group,
7086 d->send_covered, d->tmpmask);
7087 break;
7088#endif
Andreas Herrmann0e8e85c2009-08-18 12:57:51 +02007089 default:
7090 break;
7091 }
7092}
7093
Mike Travis7c16ec52008-04-04 18:11:11 -07007094/*
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007095 * Build sched domains for a given set of cpus and attach the sched domains
7096 * to the individual cpus
Linus Torvalds1da177e2005-04-16 15:20:36 -07007097 */
Rusty Russell96f874e2008-11-25 02:35:14 +10307098static int __build_sched_domains(const struct cpumask *cpu_map,
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007099 struct sched_domain_attr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007100{
Andreas Herrmann2109b992009-08-18 12:53:00 +02007101 enum s_alloc alloc_state = sa_none;
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007102 struct s_data d;
Andreas Herrmann294b0c92009-08-18 13:02:29 +02007103 struct sched_domain *sd;
Andreas Herrmann2109b992009-08-18 12:53:00 +02007104 int i;
John Hawkesd1b55132005-09-06 15:18:14 -07007105#ifdef CONFIG_NUMA
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007106 d.sd_allnodes = 0;
Rusty Russell3404c8d2008-11-25 02:35:03 +10307107#endif
7108
Andreas Herrmann2109b992009-08-18 12:53:00 +02007109 alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
7110 if (alloc_state != sa_rootdomain)
7111 goto error;
7112 alloc_state = sa_sched_groups;
Mike Travis7c16ec52008-04-04 18:11:11 -07007113
Linus Torvalds1da177e2005-04-16 15:20:36 -07007114 /*
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007115 * Set up domains for cpus specified by the cpu_map.
Linus Torvalds1da177e2005-04-16 15:20:36 -07007116 */
Rusty Russellabcd0832008-11-25 02:35:02 +10307117 for_each_cpu(i, cpu_map) {
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007118 cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)),
7119 cpu_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007120
Andreas Herrmann7f4588f2009-08-18 12:54:06 +02007121 sd = __build_numa_sched_domains(&d, cpu_map, attr, i);
Andreas Herrmann87cce662009-08-18 12:54:55 +02007122 sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i);
Heiko Carstens01a08542010-08-31 10:28:16 +02007123 sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i);
Andreas Herrmann410c4082009-08-18 12:56:14 +02007124 sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i);
Andreas Herrmannd8173532009-08-18 12:57:03 +02007125 sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007126 }
7127
Rusty Russellabcd0832008-11-25 02:35:02 +10307128 for_each_cpu(i, cpu_map) {
Andreas Herrmann0e8e85c2009-08-18 12:57:51 +02007129 build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i);
Heiko Carstens01a08542010-08-31 10:28:16 +02007130 build_sched_groups(&d, SD_LV_BOOK, cpu_map, i);
Andreas Herrmanna2af04c2009-08-18 12:58:38 +02007131 build_sched_groups(&d, SD_LV_MC, cpu_map, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007132 }
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08007133
Linus Torvalds1da177e2005-04-16 15:20:36 -07007134 /* Set up physical groups */
Andreas Herrmann86548092009-08-18 12:59:28 +02007135 for (i = 0; i < nr_node_ids; i++)
7136 build_sched_groups(&d, SD_LV_CPU, cpu_map, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007137
7138#ifdef CONFIG_NUMA
7139 /* Set up node groups */
Andreas Herrmannde616e32009-08-18 13:00:13 +02007140 if (d.sd_allnodes)
7141 build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0);
John Hawkes9c1cfda2005-09-06 15:18:14 -07007142
Andreas Herrmann0601a882009-08-18 13:01:11 +02007143 for (i = 0; i < nr_node_ids; i++)
7144 if (build_numa_sched_groups(&d, cpu_map, i))
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07007145 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007146#endif
7147
7148 /* Calculate CPU power for physical packages and nodes */
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007149#ifdef CONFIG_SCHED_SMT
Rusty Russellabcd0832008-11-25 02:35:02 +10307150 for_each_cpu(i, cpu_map) {
Andreas Herrmann294b0c92009-08-18 13:02:29 +02007151 sd = &per_cpu(cpu_domains, i).sd;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07007152 init_sched_groups_power(i, sd);
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007153 }
7154#endif
7155#ifdef CONFIG_SCHED_MC
Rusty Russellabcd0832008-11-25 02:35:02 +10307156 for_each_cpu(i, cpu_map) {
Andreas Herrmann294b0c92009-08-18 13:02:29 +02007157 sd = &per_cpu(core_domains, i).sd;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07007158 init_sched_groups_power(i, sd);
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007159 }
7160#endif
Heiko Carstens01a08542010-08-31 10:28:16 +02007161#ifdef CONFIG_SCHED_BOOK
7162 for_each_cpu(i, cpu_map) {
7163 sd = &per_cpu(book_domains, i).sd;
7164 init_sched_groups_power(i, sd);
7165 }
7166#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07007167
Rusty Russellabcd0832008-11-25 02:35:02 +10307168 for_each_cpu(i, cpu_map) {
Andreas Herrmann294b0c92009-08-18 13:02:29 +02007169 sd = &per_cpu(phys_domains, i).sd;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07007170 init_sched_groups_power(i, sd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007171 }
7172
John Hawkes9c1cfda2005-09-06 15:18:14 -07007173#ifdef CONFIG_NUMA
Mike Travis076ac2a2008-05-12 21:21:12 +02007174 for (i = 0; i < nr_node_ids; i++)
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007175 init_numa_sched_groups_power(d.sched_group_nodes[i]);
John Hawkes9c1cfda2005-09-06 15:18:14 -07007176
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007177 if (d.sd_allnodes) {
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08007178 struct sched_group *sg;
Siddha, Suresh Bf712c0c2006-07-30 03:02:59 -07007179
Rusty Russell96f874e2008-11-25 02:35:14 +10307180 cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg,
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007181 d.tmpmask);
Siddha, Suresh Bf712c0c2006-07-30 03:02:59 -07007182 init_numa_sched_groups_power(sg);
7183 }
John Hawkes9c1cfda2005-09-06 15:18:14 -07007184#endif
7185
Linus Torvalds1da177e2005-04-16 15:20:36 -07007186 /* Attach the domains */
Rusty Russellabcd0832008-11-25 02:35:02 +10307187 for_each_cpu(i, cpu_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007188#ifdef CONFIG_SCHED_SMT
Rusty Russell6c99e9a2008-11-25 02:35:04 +10307189 sd = &per_cpu(cpu_domains, i).sd;
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08007190#elif defined(CONFIG_SCHED_MC)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10307191 sd = &per_cpu(core_domains, i).sd;
Heiko Carstens01a08542010-08-31 10:28:16 +02007192#elif defined(CONFIG_SCHED_BOOK)
7193 sd = &per_cpu(book_domains, i).sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007194#else
Rusty Russell6c99e9a2008-11-25 02:35:04 +10307195 sd = &per_cpu(phys_domains, i).sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007196#endif
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007197 cpu_attach_domain(sd, d.rd, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007198 }
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07007199
Andreas Herrmann2109b992009-08-18 12:53:00 +02007200 d.sched_group_nodes = NULL; /* don't free this we still need it */
7201 __free_domain_allocs(&d, sa_tmpmask, cpu_map);
7202 return 0;
Rusty Russell3404c8d2008-11-25 02:35:03 +10307203
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07007204error:
Andreas Herrmann2109b992009-08-18 12:53:00 +02007205 __free_domain_allocs(&d, alloc_state, cpu_map);
7206 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007207}
Paul Jackson029190c2007-10-18 23:40:20 -07007208
Rusty Russell96f874e2008-11-25 02:35:14 +10307209static int build_sched_domains(const struct cpumask *cpu_map)
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007210{
7211 return __build_sched_domains(cpu_map, NULL);
7212}
7213
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307214static cpumask_var_t *doms_cur; /* current sched domains */
Paul Jackson029190c2007-10-18 23:40:20 -07007215static int ndoms_cur; /* number of sched domains in 'doms_cur' */
Ingo Molnar4285f5942008-05-16 17:47:14 +02007216static struct sched_domain_attr *dattr_cur;
7217 /* attribues of custom domains in 'doms_cur' */
Paul Jackson029190c2007-10-18 23:40:20 -07007218
7219/*
7220 * Special case: If a kmalloc of a doms_cur partition (array of
Rusty Russell42128232008-11-25 02:35:12 +10307221 * cpumask) fails, then fallback to a single sched domain,
7222 * as determined by the single cpumask fallback_doms.
Paul Jackson029190c2007-10-18 23:40:20 -07007223 */
Rusty Russell42128232008-11-25 02:35:12 +10307224static cpumask_var_t fallback_doms;
Paul Jackson029190c2007-10-18 23:40:20 -07007225
Heiko Carstensee79d1b2008-12-09 18:49:50 +01007226/*
7227 * arch_update_cpu_topology lets virtualized architectures update the
7228 * cpu core maps. It is supposed to return 1 if the topology changed
7229 * or 0 if it stayed the same.
7230 */
7231int __attribute__((weak)) arch_update_cpu_topology(void)
Heiko Carstens22e52b02008-03-12 18:31:59 +01007232{
Heiko Carstensee79d1b2008-12-09 18:49:50 +01007233 return 0;
Heiko Carstens22e52b02008-03-12 18:31:59 +01007234}
7235
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307236cpumask_var_t *alloc_sched_domains(unsigned int ndoms)
7237{
7238 int i;
7239 cpumask_var_t *doms;
7240
7241 doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL);
7242 if (!doms)
7243 return NULL;
7244 for (i = 0; i < ndoms; i++) {
7245 if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) {
7246 free_sched_domains(doms, i);
7247 return NULL;
7248 }
7249 }
7250 return doms;
7251}
7252
7253void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms)
7254{
7255 unsigned int i;
7256 for (i = 0; i < ndoms; i++)
7257 free_cpumask_var(doms[i]);
7258 kfree(doms);
7259}
7260
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007261/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007262 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
Paul Jackson029190c2007-10-18 23:40:20 -07007263 * For now this just excludes isolated cpus, but could be used to
7264 * exclude other special cases in the future.
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007265 */
Rusty Russell96f874e2008-11-25 02:35:14 +10307266static int arch_init_sched_domains(const struct cpumask *cpu_map)
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007267{
Milton Miller73785472007-10-24 18:23:48 +02007268 int err;
7269
Heiko Carstens22e52b02008-03-12 18:31:59 +01007270 arch_update_cpu_topology();
Paul Jackson029190c2007-10-18 23:40:20 -07007271 ndoms_cur = 1;
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307272 doms_cur = alloc_sched_domains(ndoms_cur);
Paul Jackson029190c2007-10-18 23:40:20 -07007273 if (!doms_cur)
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307274 doms_cur = &fallback_doms;
7275 cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007276 dattr_cur = NULL;
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307277 err = build_sched_domains(doms_cur[0]);
Milton Miller6382bc92007-10-15 17:00:19 +02007278 register_sched_domain_sysctl();
Milton Miller73785472007-10-24 18:23:48 +02007279
7280 return err;
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007281}
7282
Rusty Russell96f874e2008-11-25 02:35:14 +10307283static void arch_destroy_sched_domains(const struct cpumask *cpu_map,
7284 struct cpumask *tmpmask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007285{
Mike Travis7c16ec52008-04-04 18:11:11 -07007286 free_sched_groups(cpu_map, tmpmask);
John Hawkes9c1cfda2005-09-06 15:18:14 -07007287}
Linus Torvalds1da177e2005-04-16 15:20:36 -07007288
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007289/*
7290 * Detach sched domains from a group of cpus specified in cpu_map
7291 * These cpus will now be attached to the NULL domain
7292 */
Rusty Russell96f874e2008-11-25 02:35:14 +10307293static void detach_destroy_domains(const struct cpumask *cpu_map)
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007294{
Rusty Russell96f874e2008-11-25 02:35:14 +10307295 /* Save because hotplug lock held. */
7296 static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS);
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007297 int i;
7298
Rusty Russellabcd0832008-11-25 02:35:02 +10307299 for_each_cpu(i, cpu_map)
Gregory Haskins57d885f2008-01-25 21:08:18 +01007300 cpu_attach_domain(NULL, &def_root_domain, i);
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007301 synchronize_sched();
Rusty Russell96f874e2008-11-25 02:35:14 +10307302 arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask));
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007303}
7304
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007305/* handle null as "default" */
7306static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
7307 struct sched_domain_attr *new, int idx_new)
7308{
7309 struct sched_domain_attr tmp;
7310
7311 /* fast path */
7312 if (!new && !cur)
7313 return 1;
7314
7315 tmp = SD_ATTR_INIT;
7316 return !memcmp(cur ? (cur + idx_cur) : &tmp,
7317 new ? (new + idx_new) : &tmp,
7318 sizeof(struct sched_domain_attr));
7319}
7320
Paul Jackson029190c2007-10-18 23:40:20 -07007321/*
7322 * Partition sched domains as specified by the 'ndoms_new'
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007323 * cpumasks in the array doms_new[] of cpumasks. This compares
Paul Jackson029190c2007-10-18 23:40:20 -07007324 * doms_new[] to the current sched domain partitioning, doms_cur[].
7325 * It destroys each deleted domain and builds each new domain.
7326 *
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307327 * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'.
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007328 * The masks don't intersect (don't overlap.) We should setup one
7329 * sched domain for each mask. CPUs not in any of the cpumasks will
7330 * not be load balanced. If the same cpumask appears both in the
Paul Jackson029190c2007-10-18 23:40:20 -07007331 * current 'doms_cur' domains and in the new 'doms_new', we can leave
7332 * it as it is.
7333 *
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307334 * The passed in 'doms_new' should be allocated using
7335 * alloc_sched_domains. This routine takes ownership of it and will
7336 * free_sched_domains it when done with it. If the caller failed the
7337 * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1,
7338 * and partition_sched_domains() will fallback to the single partition
7339 * 'fallback_doms', it also forces the domains to be rebuilt.
Paul Jackson029190c2007-10-18 23:40:20 -07007340 *
Rusty Russell96f874e2008-11-25 02:35:14 +10307341 * If doms_new == NULL it will be replaced with cpu_online_mask.
Li Zefan700018e2008-11-18 14:02:03 +08007342 * ndoms_new == 0 is a special case for destroying existing domains,
7343 * and it will not create the default domain.
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07007344 *
Paul Jackson029190c2007-10-18 23:40:20 -07007345 * Call with hotplug lock held
7346 */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307347void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007348 struct sched_domain_attr *dattr_new)
Paul Jackson029190c2007-10-18 23:40:20 -07007349{
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07007350 int i, j, n;
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01007351 int new_topology;
Paul Jackson029190c2007-10-18 23:40:20 -07007352
Heiko Carstens712555e2008-04-28 11:33:07 +02007353 mutex_lock(&sched_domains_mutex);
Srivatsa Vaddagiria1835612008-01-25 21:08:00 +01007354
Milton Miller73785472007-10-24 18:23:48 +02007355 /* always unregister in case we don't destroy any domains */
7356 unregister_sched_domain_sysctl();
7357
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01007358 /* Let architecture update cpu core mappings. */
7359 new_topology = arch_update_cpu_topology();
7360
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07007361 n = doms_new ? ndoms_new : 0;
Paul Jackson029190c2007-10-18 23:40:20 -07007362
7363 /* Destroy deleted domains */
7364 for (i = 0; i < ndoms_cur; i++) {
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01007365 for (j = 0; j < n && !new_topology; j++) {
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307366 if (cpumask_equal(doms_cur[i], doms_new[j])
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007367 && dattrs_equal(dattr_cur, i, dattr_new, j))
Paul Jackson029190c2007-10-18 23:40:20 -07007368 goto match1;
7369 }
7370 /* no match - a current sched domain not in new doms_new[] */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307371 detach_destroy_domains(doms_cur[i]);
Paul Jackson029190c2007-10-18 23:40:20 -07007372match1:
7373 ;
7374 }
7375
Max Krasnyanskye761b772008-07-15 04:43:49 -07007376 if (doms_new == NULL) {
7377 ndoms_cur = 0;
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307378 doms_new = &fallback_doms;
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01007379 cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
Li Zefanfaa2f982008-11-04 16:20:23 +08007380 WARN_ON_ONCE(dattr_new);
Max Krasnyanskye761b772008-07-15 04:43:49 -07007381 }
7382
Paul Jackson029190c2007-10-18 23:40:20 -07007383 /* Build new domains */
7384 for (i = 0; i < ndoms_new; i++) {
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01007385 for (j = 0; j < ndoms_cur && !new_topology; j++) {
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307386 if (cpumask_equal(doms_new[i], doms_cur[j])
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007387 && dattrs_equal(dattr_new, i, dattr_cur, j))
Paul Jackson029190c2007-10-18 23:40:20 -07007388 goto match2;
7389 }
7390 /* no match - add a new doms_new */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307391 __build_sched_domains(doms_new[i],
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007392 dattr_new ? dattr_new + i : NULL);
Paul Jackson029190c2007-10-18 23:40:20 -07007393match2:
7394 ;
7395 }
7396
7397 /* Remember the new sched domains */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307398 if (doms_cur != &fallback_doms)
7399 free_sched_domains(doms_cur, ndoms_cur);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007400 kfree(dattr_cur); /* kfree(NULL) is safe */
Paul Jackson029190c2007-10-18 23:40:20 -07007401 doms_cur = doms_new;
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007402 dattr_cur = dattr_new;
Paul Jackson029190c2007-10-18 23:40:20 -07007403 ndoms_cur = ndoms_new;
Milton Miller73785472007-10-24 18:23:48 +02007404
7405 register_sched_domain_sysctl();
Srivatsa Vaddagiria1835612008-01-25 21:08:00 +01007406
Heiko Carstens712555e2008-04-28 11:33:07 +02007407 mutex_unlock(&sched_domains_mutex);
Paul Jackson029190c2007-10-18 23:40:20 -07007408}
7409
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007410#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
Li Zefanc70f22d2009-01-05 19:07:50 +08007411static void arch_reinit_sched_domains(void)
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007412{
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007413 get_online_cpus();
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07007414
7415 /* Destroy domains first to force the rebuild */
7416 partition_sched_domains(0, NULL, NULL);
7417
Max Krasnyanskye761b772008-07-15 04:43:49 -07007418 rebuild_sched_domains();
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007419 put_online_cpus();
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007420}
7421
7422static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt)
7423{
Gautham R Shenoyafb8a9b2008-12-18 23:26:09 +05307424 unsigned int level = 0;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007425
Gautham R Shenoyafb8a9b2008-12-18 23:26:09 +05307426 if (sscanf(buf, "%u", &level) != 1)
7427 return -EINVAL;
7428
7429 /*
7430 * level is always be positive so don't check for
7431 * level < POWERSAVINGS_BALANCE_NONE which is 0
7432 * What happens on 0 or 1 byte write,
7433 * need to check for count as well?
7434 */
7435
7436 if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS)
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007437 return -EINVAL;
7438
7439 if (smt)
Gautham R Shenoyafb8a9b2008-12-18 23:26:09 +05307440 sched_smt_power_savings = level;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007441 else
Gautham R Shenoyafb8a9b2008-12-18 23:26:09 +05307442 sched_mc_power_savings = level;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007443
Li Zefanc70f22d2009-01-05 19:07:50 +08007444 arch_reinit_sched_domains();
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007445
Li Zefanc70f22d2009-01-05 19:07:50 +08007446 return count;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007447}
7448
Adrian Bunk6707de002007-08-12 18:08:19 +02007449#ifdef CONFIG_SCHED_MC
Andi Kleenf718cd42008-07-29 22:33:52 -07007450static ssize_t sched_mc_power_savings_show(struct sysdev_class *class,
Andi Kleenc9be0a32010-01-05 12:47:58 +01007451 struct sysdev_class_attribute *attr,
Andi Kleenf718cd42008-07-29 22:33:52 -07007452 char *page)
Adrian Bunk6707de002007-08-12 18:08:19 +02007453{
7454 return sprintf(page, "%u\n", sched_mc_power_savings);
7455}
Andi Kleenf718cd42008-07-29 22:33:52 -07007456static ssize_t sched_mc_power_savings_store(struct sysdev_class *class,
Andi Kleenc9be0a32010-01-05 12:47:58 +01007457 struct sysdev_class_attribute *attr,
Adrian Bunk6707de002007-08-12 18:08:19 +02007458 const char *buf, size_t count)
7459{
7460 return sched_power_savings_store(buf, count, 0);
7461}
Andi Kleenf718cd42008-07-29 22:33:52 -07007462static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644,
7463 sched_mc_power_savings_show,
7464 sched_mc_power_savings_store);
Adrian Bunk6707de002007-08-12 18:08:19 +02007465#endif
7466
7467#ifdef CONFIG_SCHED_SMT
Andi Kleenf718cd42008-07-29 22:33:52 -07007468static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev,
Andi Kleenc9be0a32010-01-05 12:47:58 +01007469 struct sysdev_class_attribute *attr,
Andi Kleenf718cd42008-07-29 22:33:52 -07007470 char *page)
Adrian Bunk6707de002007-08-12 18:08:19 +02007471{
7472 return sprintf(page, "%u\n", sched_smt_power_savings);
7473}
Andi Kleenf718cd42008-07-29 22:33:52 -07007474static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev,
Andi Kleenc9be0a32010-01-05 12:47:58 +01007475 struct sysdev_class_attribute *attr,
Adrian Bunk6707de002007-08-12 18:08:19 +02007476 const char *buf, size_t count)
7477{
7478 return sched_power_savings_store(buf, count, 1);
7479}
Andi Kleenf718cd42008-07-29 22:33:52 -07007480static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644,
7481 sched_smt_power_savings_show,
Adrian Bunk6707de002007-08-12 18:08:19 +02007482 sched_smt_power_savings_store);
7483#endif
7484
Li Zefan39aac642009-01-05 19:18:02 +08007485int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls)
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007486{
7487 int err = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07007488
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007489#ifdef CONFIG_SCHED_SMT
7490 if (smt_capable())
7491 err = sysfs_create_file(&cls->kset.kobj,
7492 &attr_sched_smt_power_savings.attr);
7493#endif
7494#ifdef CONFIG_SCHED_MC
7495 if (!err && mc_capable())
7496 err = sysfs_create_file(&cls->kset.kobj,
7497 &attr_sched_mc_power_savings.attr);
7498#endif
7499 return err;
7500}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007501#endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007502
Linus Torvalds1da177e2005-04-16 15:20:36 -07007503/*
Tejun Heo3a101d02010-06-08 21:40:36 +02007504 * Update cpusets according to cpu_active mask. If cpusets are
7505 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
7506 * around partition_sched_domains().
Linus Torvalds1da177e2005-04-16 15:20:36 -07007507 */
Tejun Heo0b2e9182010-06-21 23:53:31 +02007508static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action,
7509 void *hcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007510{
Tejun Heo3a101d02010-06-08 21:40:36 +02007511 switch (action & ~CPU_TASKS_FROZEN) {
Max Krasnyanskye761b772008-07-15 04:43:49 -07007512 case CPU_ONLINE:
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01007513 case CPU_DOWN_FAILED:
Tejun Heo3a101d02010-06-08 21:40:36 +02007514 cpuset_update_active_cpus();
Max Krasnyanskye761b772008-07-15 04:43:49 -07007515 return NOTIFY_OK;
Max Krasnyanskye761b772008-07-15 04:43:49 -07007516 default:
7517 return NOTIFY_DONE;
7518 }
7519}
Tejun Heo3a101d02010-06-08 21:40:36 +02007520
Tejun Heo0b2e9182010-06-21 23:53:31 +02007521static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action,
7522 void *hcpu)
Tejun Heo3a101d02010-06-08 21:40:36 +02007523{
7524 switch (action & ~CPU_TASKS_FROZEN) {
7525 case CPU_DOWN_PREPARE:
7526 cpuset_update_active_cpus();
7527 return NOTIFY_OK;
7528 default:
7529 return NOTIFY_DONE;
7530 }
7531}
Max Krasnyanskye761b772008-07-15 04:43:49 -07007532
7533static int update_runtime(struct notifier_block *nfb,
7534 unsigned long action, void *hcpu)
7535{
Peter Zijlstra7def2be2008-06-05 14:49:58 +02007536 int cpu = (int)(long)hcpu;
7537
Linus Torvalds1da177e2005-04-16 15:20:36 -07007538 switch (action) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007539 case CPU_DOWN_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007540 case CPU_DOWN_PREPARE_FROZEN:
Peter Zijlstra7def2be2008-06-05 14:49:58 +02007541 disable_runtime(cpu_rq(cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -07007542 return NOTIFY_OK;
7543
Linus Torvalds1da177e2005-04-16 15:20:36 -07007544 case CPU_DOWN_FAILED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007545 case CPU_DOWN_FAILED_FROZEN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07007546 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007547 case CPU_ONLINE_FROZEN:
Peter Zijlstra7def2be2008-06-05 14:49:58 +02007548 enable_runtime(cpu_rq(cpu));
Max Krasnyanskye761b772008-07-15 04:43:49 -07007549 return NOTIFY_OK;
7550
Linus Torvalds1da177e2005-04-16 15:20:36 -07007551 default:
7552 return NOTIFY_DONE;
7553 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007554}
Linus Torvalds1da177e2005-04-16 15:20:36 -07007555
7556void __init sched_init_smp(void)
7557{
Rusty Russelldcc30a32008-11-25 02:35:12 +10307558 cpumask_var_t non_isolated_cpus;
7559
7560 alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
Yong Zhangcb5fd132009-09-14 20:20:16 +08007561 alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
Nick Piggin5c1e1762006-10-03 01:14:04 -07007562
Mike Travis434d53b2008-04-04 18:11:04 -07007563#if defined(CONFIG_NUMA)
7564 sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **),
7565 GFP_KERNEL);
7566 BUG_ON(sched_group_nodes_bycpu == NULL);
7567#endif
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007568 get_online_cpus();
Heiko Carstens712555e2008-04-28 11:33:07 +02007569 mutex_lock(&sched_domains_mutex);
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01007570 arch_init_sched_domains(cpu_active_mask);
Rusty Russelldcc30a32008-11-25 02:35:12 +10307571 cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map);
7572 if (cpumask_empty(non_isolated_cpus))
7573 cpumask_set_cpu(smp_processor_id(), non_isolated_cpus);
Heiko Carstens712555e2008-04-28 11:33:07 +02007574 mutex_unlock(&sched_domains_mutex);
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007575 put_online_cpus();
Max Krasnyanskye761b772008-07-15 04:43:49 -07007576
Tejun Heo3a101d02010-06-08 21:40:36 +02007577 hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE);
7578 hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE);
Max Krasnyanskye761b772008-07-15 04:43:49 -07007579
7580 /* RT runtime code needs to handle some hotplug events */
7581 hotcpu_notifier(update_runtime, 0);
7582
Peter Zijlstrab328ca12008-04-29 10:02:46 +02007583 init_hrtick();
Nick Piggin5c1e1762006-10-03 01:14:04 -07007584
7585 /* Move init over to a non-isolated CPU */
Rusty Russelldcc30a32008-11-25 02:35:12 +10307586 if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
Nick Piggin5c1e1762006-10-03 01:14:04 -07007587 BUG();
Ingo Molnar19978ca2007-11-09 22:39:38 +01007588 sched_init_granularity();
Rusty Russelldcc30a32008-11-25 02:35:12 +10307589 free_cpumask_var(non_isolated_cpus);
Rusty Russell42128232008-11-25 02:35:12 +10307590
Rusty Russell0e3900e2008-11-25 02:35:13 +10307591 init_sched_rt_class();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007592}
7593#else
7594void __init sched_init_smp(void)
7595{
Ingo Molnar19978ca2007-11-09 22:39:38 +01007596 sched_init_granularity();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007597}
7598#endif /* CONFIG_SMP */
7599
Arun R Bharadwajcd1bb942009-04-16 12:15:34 +05307600const_debug unsigned int sysctl_timer_migration = 1;
7601
Linus Torvalds1da177e2005-04-16 15:20:36 -07007602int in_sched_functions(unsigned long addr)
7603{
Linus Torvalds1da177e2005-04-16 15:20:36 -07007604 return in_lock_functions(addr) ||
7605 (addr >= (unsigned long)__sched_text_start
7606 && addr < (unsigned long)__sched_text_end);
7607}
7608
Alexey Dobriyana9957442007-10-15 17:00:13 +02007609static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq)
Ingo Molnardd41f592007-07-09 18:51:59 +02007610{
7611 cfs_rq->tasks_timeline = RB_ROOT;
Peter Zijlstra4a55bd52008-04-19 19:45:00 +02007612 INIT_LIST_HEAD(&cfs_rq->tasks);
Ingo Molnardd41f592007-07-09 18:51:59 +02007613#ifdef CONFIG_FAIR_GROUP_SCHED
7614 cfs_rq->rq = rq;
7615#endif
Peter Zijlstra67e9fb22007-10-15 17:00:10 +02007616 cfs_rq->min_vruntime = (u64)(-(1LL << 20));
Ingo Molnardd41f592007-07-09 18:51:59 +02007617}
7618
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007619static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq)
7620{
7621 struct rt_prio_array *array;
7622 int i;
7623
7624 array = &rt_rq->active;
7625 for (i = 0; i < MAX_RT_PRIO; i++) {
7626 INIT_LIST_HEAD(array->queue + i);
7627 __clear_bit(i, array->bitmap);
7628 }
7629 /* delimiter for bitsearch: */
7630 __set_bit(MAX_RT_PRIO, array->bitmap);
7631
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007632#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
Gregory Haskinse864c492008-12-29 09:39:49 -05007633 rt_rq->highest_prio.curr = MAX_RT_PRIO;
Gregory Haskins398a1532009-01-14 09:10:04 -05007634#ifdef CONFIG_SMP
Gregory Haskinse864c492008-12-29 09:39:49 -05007635 rt_rq->highest_prio.next = MAX_RT_PRIO;
Peter Zijlstra48d5e252008-01-25 21:08:31 +01007636#endif
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007637#endif
7638#ifdef CONFIG_SMP
7639 rt_rq->rt_nr_migratory = 0;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007640 rt_rq->overloaded = 0;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01007641 plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007642#endif
7643
7644 rt_rq->rt_time = 0;
7645 rt_rq->rt_throttled = 0;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007646 rt_rq->rt_runtime = 0;
Thomas Gleixner0986b112009-11-17 15:32:06 +01007647 raw_spin_lock_init(&rt_rq->rt_runtime_lock);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007648
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007649#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +01007650 rt_rq->rt_nr_boosted = 0;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007651 rt_rq->rq = rq;
7652#endif
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007653}
7654
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007655#ifdef CONFIG_FAIR_GROUP_SCHED
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007656static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -08007657 struct sched_entity *se, int cpu,
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007658 struct sched_entity *parent)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007659{
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007660 struct rq *rq = cpu_rq(cpu);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007661 tg->cfs_rq[cpu] = cfs_rq;
7662 init_cfs_rq(cfs_rq, rq);
7663 cfs_rq->tg = tg;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007664
7665 tg->se[cpu] = se;
Dhaval Giani354d60c2008-04-19 19:44:59 +02007666 /* se could be NULL for init_task_group */
7667 if (!se)
7668 return;
7669
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007670 if (!parent)
7671 se->cfs_rq = &rq->cfs;
7672 else
7673 se->cfs_rq = parent->my_q;
7674
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007675 se->my_q = cfs_rq;
Paul Turner94371782010-11-15 15:47:10 -08007676 update_load_set(&se->load, 0);
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007677 se->parent = parent;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007678}
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007679#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007680
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007681#ifdef CONFIG_RT_GROUP_SCHED
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007682static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -08007683 struct sched_rt_entity *rt_se, int cpu,
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007684 struct sched_rt_entity *parent)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007685{
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007686 struct rq *rq = cpu_rq(cpu);
7687
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007688 tg->rt_rq[cpu] = rt_rq;
7689 init_rt_rq(rt_rq, rq);
7690 rt_rq->tg = tg;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007691 rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007692
7693 tg->rt_se[cpu] = rt_se;
Dhaval Giani354d60c2008-04-19 19:44:59 +02007694 if (!rt_se)
7695 return;
7696
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007697 if (!parent)
7698 rt_se->rt_rq = &rq->rt;
7699 else
7700 rt_se->rt_rq = parent->my_q;
7701
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007702 rt_se->my_q = rt_rq;
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007703 rt_se->parent = parent;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007704 INIT_LIST_HEAD(&rt_se->run_list);
7705}
7706#endif
7707
Linus Torvalds1da177e2005-04-16 15:20:36 -07007708void __init sched_init(void)
7709{
Ingo Molnardd41f592007-07-09 18:51:59 +02007710 int i, j;
Mike Travis434d53b2008-04-04 18:11:04 -07007711 unsigned long alloc_size = 0, ptr;
7712
7713#ifdef CONFIG_FAIR_GROUP_SCHED
7714 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
7715#endif
7716#ifdef CONFIG_RT_GROUP_SCHED
7717 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
7718#endif
Rusty Russelldf7c8e82009-03-19 15:22:20 +10307719#ifdef CONFIG_CPUMASK_OFFSTACK
Rusty Russell8c083f02009-03-19 15:22:20 +10307720 alloc_size += num_possible_cpus() * cpumask_size();
Rusty Russelldf7c8e82009-03-19 15:22:20 +10307721#endif
Mike Travis434d53b2008-04-04 18:11:04 -07007722 if (alloc_size) {
Pekka Enberg36b7b6d2009-06-10 23:42:36 +03007723 ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
Mike Travis434d53b2008-04-04 18:11:04 -07007724
7725#ifdef CONFIG_FAIR_GROUP_SCHED
7726 init_task_group.se = (struct sched_entity **)ptr;
7727 ptr += nr_cpu_ids * sizeof(void **);
7728
7729 init_task_group.cfs_rq = (struct cfs_rq **)ptr;
7730 ptr += nr_cpu_ids * sizeof(void **);
Peter Zijlstraeff766a2008-04-19 19:45:00 +02007731
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007732#endif /* CONFIG_FAIR_GROUP_SCHED */
Mike Travis434d53b2008-04-04 18:11:04 -07007733#ifdef CONFIG_RT_GROUP_SCHED
7734 init_task_group.rt_se = (struct sched_rt_entity **)ptr;
7735 ptr += nr_cpu_ids * sizeof(void **);
7736
7737 init_task_group.rt_rq = (struct rt_rq **)ptr;
Peter Zijlstraeff766a2008-04-19 19:45:00 +02007738 ptr += nr_cpu_ids * sizeof(void **);
7739
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007740#endif /* CONFIG_RT_GROUP_SCHED */
Rusty Russelldf7c8e82009-03-19 15:22:20 +10307741#ifdef CONFIG_CPUMASK_OFFSTACK
7742 for_each_possible_cpu(i) {
7743 per_cpu(load_balance_tmpmask, i) = (void *)ptr;
7744 ptr += cpumask_size();
7745 }
7746#endif /* CONFIG_CPUMASK_OFFSTACK */
Mike Travis434d53b2008-04-04 18:11:04 -07007747 }
Ingo Molnardd41f592007-07-09 18:51:59 +02007748
Gregory Haskins57d885f2008-01-25 21:08:18 +01007749#ifdef CONFIG_SMP
7750 init_defrootdomain();
7751#endif
7752
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007753 init_rt_bandwidth(&def_rt_bandwidth,
7754 global_rt_period(), global_rt_runtime());
7755
7756#ifdef CONFIG_RT_GROUP_SCHED
7757 init_rt_bandwidth(&init_task_group.rt_bandwidth,
7758 global_rt_period(), global_rt_runtime());
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007759#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007760
Dhaval Giani7c941432010-01-20 13:26:18 +01007761#ifdef CONFIG_CGROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007762 list_add(&init_task_group.list, &task_groups);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02007763 INIT_LIST_HEAD(&init_task_group.children);
Mike Galbraith5091faa2010-11-30 14:18:03 +01007764 autogroup_init(&init_task);
Dhaval Giani7c941432010-01-20 13:26:18 +01007765#endif /* CONFIG_CGROUP_SCHED */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007766
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08007767 for_each_possible_cpu(i) {
Ingo Molnar70b97a72006-07-03 00:25:42 -07007768 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007769
7770 rq = cpu_rq(i);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01007771 raw_spin_lock_init(&rq->lock);
Nick Piggin78979862005-06-25 14:57:13 -07007772 rq->nr_running = 0;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02007773 rq->calc_load_active = 0;
7774 rq->calc_load_update = jiffies + LOAD_FREQ;
Ingo Molnardd41f592007-07-09 18:51:59 +02007775 init_cfs_rq(&rq->cfs, rq);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007776 init_rt_rq(&rq->rt, rq);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007777#ifdef CONFIG_FAIR_GROUP_SCHED
7778 init_task_group.shares = init_task_group_load;
7779 INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
Dhaval Giani354d60c2008-04-19 19:44:59 +02007780#ifdef CONFIG_CGROUP_SCHED
7781 /*
7782 * How much cpu bandwidth does init_task_group get?
7783 *
7784 * In case of task-groups formed thr' the cgroup filesystem, it
7785 * gets 100% of the cpu resources in the system. This overall
7786 * system cpu resource is divided among the tasks of
7787 * init_task_group and its child task-groups in a fair manner,
7788 * based on each entity's (task or task-group's) weight
7789 * (se->load.weight).
7790 *
7791 * In other words, if init_task_group has 10 tasks of weight
7792 * 1024) and two child groups A0 and A1 (of weight 1024 each),
7793 * then A0's share of the cpu resource is:
7794 *
Ingo Molnar0d905bc2009-05-04 19:13:30 +02007795 * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
Dhaval Giani354d60c2008-04-19 19:44:59 +02007796 *
7797 * We achieve this by letting init_task_group's tasks sit
7798 * directly in rq->cfs (i.e init_task_group->se[] = NULL).
7799 */
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -08007800 init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, NULL);
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007801#endif
Dhaval Giani354d60c2008-04-19 19:44:59 +02007802#endif /* CONFIG_FAIR_GROUP_SCHED */
7803
7804 rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007805#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007806 INIT_LIST_HEAD(&rq->leaf_rt_rq_list);
Dhaval Giani354d60c2008-04-19 19:44:59 +02007807#ifdef CONFIG_CGROUP_SCHED
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -08007808 init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, NULL);
Dhaval Giani354d60c2008-04-19 19:44:59 +02007809#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007810#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07007811
Ingo Molnardd41f592007-07-09 18:51:59 +02007812 for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
7813 rq->cpu_load[j] = 0;
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07007814
7815 rq->last_load_update_tick = jiffies;
7816
Linus Torvalds1da177e2005-04-16 15:20:36 -07007817#ifdef CONFIG_SMP
Nick Piggin41c7ce92005-06-25 14:57:24 -07007818 rq->sd = NULL;
Gregory Haskins57d885f2008-01-25 21:08:18 +01007819 rq->rd = NULL;
Peter Zijlstrae51fd5e2010-05-31 12:37:30 +02007820 rq->cpu_power = SCHED_LOAD_SCALE;
Gregory Haskins3f029d32009-07-29 11:08:47 -04007821 rq->post_schedule = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007822 rq->active_balance = 0;
Ingo Molnardd41f592007-07-09 18:51:59 +02007823 rq->next_balance = jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007824 rq->push_cpu = 0;
Christoph Lameter0a2966b2006-09-25 23:30:51 -07007825 rq->cpu = i;
Gregory Haskins1f11eb62008-06-04 15:04:05 -04007826 rq->online = 0;
Mike Galbraitheae0c9d2009-11-10 03:50:02 +01007827 rq->idle_stamp = 0;
7828 rq->avg_idle = 2*sysctl_sched_migration_cost;
Gregory Haskinsdc938522008-01-25 21:08:26 +01007829 rq_attach_root(rq, &def_root_domain);
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -07007830#ifdef CONFIG_NO_HZ
7831 rq->nohz_balance_kick = 0;
7832 init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i));
7833#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07007834#endif
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01007835 init_rq_hrtick(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007836 atomic_set(&rq->nr_iowait, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007837 }
7838
Peter Williams2dd73a42006-06-27 02:54:34 -07007839 set_load_weight(&init_task);
Heiko Carstensb50f60c2006-07-30 03:03:52 -07007840
Avi Kivitye107be32007-07-26 13:40:43 +02007841#ifdef CONFIG_PREEMPT_NOTIFIERS
7842 INIT_HLIST_HEAD(&init_task.preempt_notifiers);
7843#endif
7844
Christoph Lameterc9819f42006-12-10 02:20:25 -08007845#ifdef CONFIG_SMP
Carlos R. Mafra962cf362008-05-15 11:15:37 -03007846 open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);
Christoph Lameterc9819f42006-12-10 02:20:25 -08007847#endif
7848
Heiko Carstensb50f60c2006-07-30 03:03:52 -07007849#ifdef CONFIG_RT_MUTEXES
Thomas Gleixner1d615482009-11-17 14:54:03 +01007850 plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock);
Heiko Carstensb50f60c2006-07-30 03:03:52 -07007851#endif
7852
Linus Torvalds1da177e2005-04-16 15:20:36 -07007853 /*
7854 * The boot idle thread does lazy MMU switching as well:
7855 */
7856 atomic_inc(&init_mm.mm_count);
7857 enter_lazy_tlb(&init_mm, current);
7858
7859 /*
7860 * Make us the idle thread. Technically, schedule() should not be
7861 * called from this thread, however somewhere below it might be,
7862 * but because we are the idle thread, we just pick up running again
7863 * when this runqueue becomes "idle".
7864 */
7865 init_idle(current, smp_processor_id());
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02007866
7867 calc_load_update = jiffies + LOAD_FREQ;
7868
Ingo Molnardd41f592007-07-09 18:51:59 +02007869 /*
7870 * During early bootup we pretend to be a normal task:
7871 */
7872 current->sched_class = &fair_sched_class;
Ingo Molnar6892b752008-02-13 14:02:36 +01007873
Rusty Russell6a7b3dc2008-11-25 02:35:04 +10307874 /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */
Rusty Russell49557e62009-11-02 20:37:20 +10307875 zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT);
Rusty Russellbf4d83f2008-11-25 09:57:51 +10307876#ifdef CONFIG_SMP
Rusty Russell7d1e6a92008-11-25 02:35:09 +10307877#ifdef CONFIG_NO_HZ
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -07007878 zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
7879 alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT);
7880 atomic_set(&nohz.load_balancer, nr_cpu_ids);
7881 atomic_set(&nohz.first_pick_cpu, nr_cpu_ids);
7882 atomic_set(&nohz.second_pick_cpu, nr_cpu_ids);
Rusty Russell7d1e6a92008-11-25 02:35:09 +10307883#endif
Rusty Russellbdddd292009-12-02 14:09:16 +10307884 /* May be allocated at isolcpus cmdline parse time */
7885 if (cpu_isolated_map == NULL)
7886 zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT);
Rusty Russellbf4d83f2008-11-25 09:57:51 +10307887#endif /* SMP */
Rusty Russell6a7b3dc2008-11-25 02:35:04 +10307888
Ingo Molnarcdd6c482009-09-21 12:02:48 +02007889 perf_event_init();
Ingo Molnar0d905bc2009-05-04 19:13:30 +02007890
Ingo Molnar6892b752008-02-13 14:02:36 +01007891 scheduler_running = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007892}
7893
7894#ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
Frederic Weisbeckere4aafea2009-07-16 15:44:29 +02007895static inline int preempt_count_equals(int preempt_offset)
7896{
Frederic Weisbecker234da7b2009-12-16 20:21:05 +01007897 int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth();
Frederic Weisbeckere4aafea2009-07-16 15:44:29 +02007898
7899 return (nested == PREEMPT_INATOMIC_BASE + preempt_offset);
7900}
7901
Simon Kagstromd8948372009-12-23 11:08:18 +01007902void __might_sleep(const char *file, int line, int preempt_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007903{
Ingo Molnar48f24c42006-07-03 00:25:40 -07007904#ifdef in_atomic
Linus Torvalds1da177e2005-04-16 15:20:36 -07007905 static unsigned long prev_jiffy; /* ratelimiting */
7906
Frederic Weisbeckere4aafea2009-07-16 15:44:29 +02007907 if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) ||
7908 system_state != SYSTEM_RUNNING || oops_in_progress)
Ingo Molnaraef745f2008-08-28 11:34:43 +02007909 return;
7910 if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
7911 return;
7912 prev_jiffy = jiffies;
7913
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01007914 printk(KERN_ERR
7915 "BUG: sleeping function called from invalid context at %s:%d\n",
7916 file, line);
7917 printk(KERN_ERR
7918 "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n",
7919 in_atomic(), irqs_disabled(),
7920 current->pid, current->comm);
Ingo Molnaraef745f2008-08-28 11:34:43 +02007921
7922 debug_show_held_locks(current);
7923 if (irqs_disabled())
7924 print_irqtrace_events(current);
7925 dump_stack();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007926#endif
7927}
7928EXPORT_SYMBOL(__might_sleep);
7929#endif
7930
7931#ifdef CONFIG_MAGIC_SYSRQ
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02007932static void normalize_task(struct rq *rq, struct task_struct *p)
7933{
7934 int on_rq;
Peter Zijlstra3e51f332008-05-03 18:29:28 +02007935
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02007936 on_rq = p->se.on_rq;
7937 if (on_rq)
7938 deactivate_task(rq, p, 0);
7939 __setscheduler(rq, p, SCHED_NORMAL, 0);
7940 if (on_rq) {
7941 activate_task(rq, p, 0);
7942 resched_task(rq->curr);
7943 }
7944}
7945
Linus Torvalds1da177e2005-04-16 15:20:36 -07007946void normalize_rt_tasks(void)
7947{
Ingo Molnara0f98a12007-06-17 18:37:45 +02007948 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007949 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07007950 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007951
Peter Zijlstra4cf5d772008-02-13 15:45:39 +01007952 read_lock_irqsave(&tasklist_lock, flags);
Ingo Molnara0f98a12007-06-17 18:37:45 +02007953 do_each_thread(g, p) {
Ingo Molnar178be792007-10-15 17:00:18 +02007954 /*
7955 * Only normalize user tasks:
7956 */
7957 if (!p->mm)
7958 continue;
7959
Ingo Molnardd41f592007-07-09 18:51:59 +02007960 p->se.exec_start = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02007961#ifdef CONFIG_SCHEDSTATS
Lucas De Marchi41acab82010-03-10 23:37:45 -03007962 p->se.statistics.wait_start = 0;
7963 p->se.statistics.sleep_start = 0;
7964 p->se.statistics.block_start = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02007965#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02007966
7967 if (!rt_task(p)) {
7968 /*
7969 * Renice negative nice level userspace
7970 * tasks back to 0:
7971 */
7972 if (TASK_NICE(p) < 0 && p->mm)
7973 set_user_nice(p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007974 continue;
Ingo Molnardd41f592007-07-09 18:51:59 +02007975 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007976
Thomas Gleixner1d615482009-11-17 14:54:03 +01007977 raw_spin_lock(&p->pi_lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -07007978 rq = __task_rq_lock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007979
Ingo Molnar178be792007-10-15 17:00:18 +02007980 normalize_task(rq, p);
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02007981
Ingo Molnarb29739f2006-06-27 02:54:51 -07007982 __task_rq_unlock(rq);
Thomas Gleixner1d615482009-11-17 14:54:03 +01007983 raw_spin_unlock(&p->pi_lock);
Ingo Molnara0f98a12007-06-17 18:37:45 +02007984 } while_each_thread(g, p);
7985
Peter Zijlstra4cf5d772008-02-13 15:45:39 +01007986 read_unlock_irqrestore(&tasklist_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007987}
7988
7989#endif /* CONFIG_MAGIC_SYSRQ */
Linus Torvalds1df5c102005-09-12 07:59:21 -07007990
Jason Wessel67fc4e02010-05-20 21:04:21 -05007991#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
Linus Torvalds1df5c102005-09-12 07:59:21 -07007992/*
Jason Wessel67fc4e02010-05-20 21:04:21 -05007993 * These functions are only useful for the IA64 MCA handling, or kdb.
Linus Torvalds1df5c102005-09-12 07:59:21 -07007994 *
7995 * They can only be called when the whole system has been
7996 * stopped - every CPU needs to be quiescent, and no scheduling
7997 * activity can take place. Using them for anything else would
7998 * be a serious bug, and as a result, they aren't even visible
7999 * under any other configuration.
8000 */
8001
8002/**
8003 * curr_task - return the current task for a given cpu.
8004 * @cpu: the processor in question.
8005 *
8006 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
8007 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07008008struct task_struct *curr_task(int cpu)
Linus Torvalds1df5c102005-09-12 07:59:21 -07008009{
8010 return cpu_curr(cpu);
8011}
8012
Jason Wessel67fc4e02010-05-20 21:04:21 -05008013#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */
8014
8015#ifdef CONFIG_IA64
Linus Torvalds1df5c102005-09-12 07:59:21 -07008016/**
8017 * set_curr_task - set the current task for a given cpu.
8018 * @cpu: the processor in question.
8019 * @p: the task pointer to set.
8020 *
8021 * Description: This function must only be used when non-maskable interrupts
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01008022 * are serviced on a separate stack. It allows the architecture to switch the
8023 * notion of the current task on a cpu in a non-blocking manner. This function
Linus Torvalds1df5c102005-09-12 07:59:21 -07008024 * must be called with all CPU's synchronized, and interrupts disabled, the
8025 * and caller must save the original value of the current task (see
8026 * curr_task() above) and restore that value before reenabling interrupts and
8027 * re-starting the system.
8028 *
8029 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
8030 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07008031void set_curr_task(int cpu, struct task_struct *p)
Linus Torvalds1df5c102005-09-12 07:59:21 -07008032{
8033 cpu_curr(cpu) = p;
8034}
8035
8036#endif
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008037
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008038#ifdef CONFIG_FAIR_GROUP_SCHED
8039static void free_fair_sched_group(struct task_group *tg)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008040{
8041 int i;
8042
8043 for_each_possible_cpu(i) {
8044 if (tg->cfs_rq)
8045 kfree(tg->cfs_rq[i]);
8046 if (tg->se)
8047 kfree(tg->se[i]);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008048 }
8049
8050 kfree(tg->cfs_rq);
8051 kfree(tg->se);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008052}
8053
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008054static
8055int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008056{
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008057 struct cfs_rq *cfs_rq;
Li Zefaneab17222008-10-29 17:03:22 +08008058 struct sched_entity *se;
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008059 struct rq *rq;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008060 int i;
8061
Mike Travis434d53b2008-04-04 18:11:04 -07008062 tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008063 if (!tg->cfs_rq)
8064 goto err;
Mike Travis434d53b2008-04-04 18:11:04 -07008065 tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008066 if (!tg->se)
8067 goto err;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008068
8069 tg->shares = NICE_0_LOAD;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008070
8071 for_each_possible_cpu(i) {
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008072 rq = cpu_rq(i);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008073
Li Zefaneab17222008-10-29 17:03:22 +08008074 cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
8075 GFP_KERNEL, cpu_to_node(i));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008076 if (!cfs_rq)
8077 goto err;
8078
Li Zefaneab17222008-10-29 17:03:22 +08008079 se = kzalloc_node(sizeof(struct sched_entity),
8080 GFP_KERNEL, cpu_to_node(i));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008081 if (!se)
Phil Carmodydfc12eb2009-12-10 14:29:37 +02008082 goto err_free_rq;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008083
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -08008084 init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008085 }
8086
8087 return 1;
8088
Peter Zijlstra49246272010-10-17 21:46:10 +02008089err_free_rq:
Phil Carmodydfc12eb2009-12-10 14:29:37 +02008090 kfree(cfs_rq);
Peter Zijlstra49246272010-10-17 21:46:10 +02008091err:
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008092 return 0;
8093}
8094
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008095static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
8096{
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -08008097 struct rq *rq = cpu_rq(cpu);
8098 unsigned long flags;
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -08008099
8100 /*
8101 * Only empty task groups can be destroyed; so we can speculatively
8102 * check on_list without danger of it being re-added.
8103 */
8104 if (!tg->cfs_rq[cpu]->on_list)
8105 return;
8106
8107 raw_spin_lock_irqsave(&rq->lock, flags);
Paul Turner822bc182010-11-29 16:55:40 -08008108 list_del_leaf_cfs_rq(tg->cfs_rq[cpu]);
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -08008109 raw_spin_unlock_irqrestore(&rq->lock, flags);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008110}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008111#else /* !CONFG_FAIR_GROUP_SCHED */
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008112static inline void free_fair_sched_group(struct task_group *tg)
8113{
8114}
8115
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008116static inline
8117int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008118{
8119 return 1;
8120}
8121
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008122static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
8123{
8124}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008125#endif /* CONFIG_FAIR_GROUP_SCHED */
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008126
8127#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008128static void free_rt_sched_group(struct task_group *tg)
8129{
8130 int i;
8131
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008132 destroy_rt_bandwidth(&tg->rt_bandwidth);
8133
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008134 for_each_possible_cpu(i) {
8135 if (tg->rt_rq)
8136 kfree(tg->rt_rq[i]);
8137 if (tg->rt_se)
8138 kfree(tg->rt_se[i]);
8139 }
8140
8141 kfree(tg->rt_rq);
8142 kfree(tg->rt_se);
8143}
8144
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008145static
8146int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008147{
8148 struct rt_rq *rt_rq;
Li Zefaneab17222008-10-29 17:03:22 +08008149 struct sched_rt_entity *rt_se;
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008150 struct rq *rq;
8151 int i;
8152
Mike Travis434d53b2008-04-04 18:11:04 -07008153 tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008154 if (!tg->rt_rq)
8155 goto err;
Mike Travis434d53b2008-04-04 18:11:04 -07008156 tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008157 if (!tg->rt_se)
8158 goto err;
8159
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008160 init_rt_bandwidth(&tg->rt_bandwidth,
8161 ktime_to_ns(def_rt_bandwidth.rt_period), 0);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008162
8163 for_each_possible_cpu(i) {
8164 rq = cpu_rq(i);
8165
Li Zefaneab17222008-10-29 17:03:22 +08008166 rt_rq = kzalloc_node(sizeof(struct rt_rq),
8167 GFP_KERNEL, cpu_to_node(i));
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008168 if (!rt_rq)
8169 goto err;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008170
Li Zefaneab17222008-10-29 17:03:22 +08008171 rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
8172 GFP_KERNEL, cpu_to_node(i));
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008173 if (!rt_se)
Phil Carmodydfc12eb2009-12-10 14:29:37 +02008174 goto err_free_rq;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008175
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -08008176 init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008177 }
8178
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008179 return 1;
8180
Peter Zijlstra49246272010-10-17 21:46:10 +02008181err_free_rq:
Phil Carmodydfc12eb2009-12-10 14:29:37 +02008182 kfree(rt_rq);
Peter Zijlstra49246272010-10-17 21:46:10 +02008183err:
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008184 return 0;
8185}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008186#else /* !CONFIG_RT_GROUP_SCHED */
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008187static inline void free_rt_sched_group(struct task_group *tg)
8188{
8189}
8190
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008191static inline
8192int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008193{
8194 return 1;
8195}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008196#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008197
Dhaval Giani7c941432010-01-20 13:26:18 +01008198#ifdef CONFIG_CGROUP_SCHED
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008199static void free_sched_group(struct task_group *tg)
8200{
8201 free_fair_sched_group(tg);
8202 free_rt_sched_group(tg);
8203 kfree(tg);
8204}
8205
8206/* allocate runqueue etc for a new task group */
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008207struct task_group *sched_create_group(struct task_group *parent)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008208{
8209 struct task_group *tg;
8210 unsigned long flags;
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008211
8212 tg = kzalloc(sizeof(*tg), GFP_KERNEL);
8213 if (!tg)
8214 return ERR_PTR(-ENOMEM);
8215
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008216 if (!alloc_fair_sched_group(tg, parent))
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008217 goto err;
8218
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008219 if (!alloc_rt_sched_group(tg, parent))
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008220 goto err;
8221
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008222 spin_lock_irqsave(&task_group_lock, flags);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008223 list_add_rcu(&tg->list, &task_groups);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02008224
8225 WARN_ON(!parent); /* root should already exist */
8226
8227 tg->parent = parent;
Peter Zijlstraf473aa52008-04-19 19:45:00 +02008228 INIT_LIST_HEAD(&tg->children);
Zhang, Yanmin09f27242030-08-14 15:56:40 +08008229 list_add_rcu(&tg->siblings, &parent->children);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008230 spin_unlock_irqrestore(&task_group_lock, flags);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008231
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008232 return tg;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008233
8234err:
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008235 free_sched_group(tg);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008236 return ERR_PTR(-ENOMEM);
8237}
8238
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008239/* rcu callback to free various structures associated with a task group */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008240static void free_sched_group_rcu(struct rcu_head *rhp)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008241{
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008242 /* now it should be safe to free those cfs_rqs */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008243 free_sched_group(container_of(rhp, struct task_group, rcu));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008244}
8245
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008246/* Destroy runqueue etc associated with a task group */
Ingo Molnar4cf86d72007-10-15 17:00:14 +02008247void sched_destroy_group(struct task_group *tg)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008248{
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008249 unsigned long flags;
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008250 int i;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008251
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -08008252 /* end participation in shares distribution */
8253 for_each_possible_cpu(i)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008254 unregister_fair_sched_group(tg, i);
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -08008255
8256 spin_lock_irqsave(&task_group_lock, flags);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008257 list_del_rcu(&tg->list);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02008258 list_del_rcu(&tg->siblings);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008259 spin_unlock_irqrestore(&task_group_lock, flags);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008260
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008261 /* wait for possible concurrent references to cfs_rqs complete */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008262 call_rcu(&tg->rcu, free_sched_group_rcu);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008263}
8264
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008265/* change task's runqueue when it moves between groups.
Ingo Molnar3a252012007-10-15 17:00:12 +02008266 * The caller of this function should have put the task in its new group
8267 * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to
8268 * reflect its new group.
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008269 */
8270void sched_move_task(struct task_struct *tsk)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008271{
8272 int on_rq, running;
8273 unsigned long flags;
8274 struct rq *rq;
8275
8276 rq = task_rq_lock(tsk, &flags);
8277
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01008278 running = task_current(rq, tsk);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008279 on_rq = tsk->se.on_rq;
8280
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07008281 if (on_rq)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008282 dequeue_task(rq, tsk, 0);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07008283 if (unlikely(running))
8284 tsk->sched_class->put_prev_task(rq, tsk);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008285
Peter Zijlstra810b3812008-02-29 15:21:01 -05008286#ifdef CONFIG_FAIR_GROUP_SCHED
Peter Zijlstrab2b5ce02010-10-15 15:24:15 +02008287 if (tsk->sched_class->task_move_group)
8288 tsk->sched_class->task_move_group(tsk, on_rq);
8289 else
Peter Zijlstra810b3812008-02-29 15:21:01 -05008290#endif
Peter Zijlstrab2b5ce02010-10-15 15:24:15 +02008291 set_task_rq(tsk, task_cpu(tsk));
Peter Zijlstra810b3812008-02-29 15:21:01 -05008292
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07008293 if (unlikely(running))
8294 tsk->sched_class->set_curr_task(rq);
8295 if (on_rq)
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01008296 enqueue_task(rq, tsk, 0);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008297
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008298 task_rq_unlock(rq, &flags);
8299}
Dhaval Giani7c941432010-01-20 13:26:18 +01008300#endif /* CONFIG_CGROUP_SCHED */
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008301
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008302#ifdef CONFIG_FAIR_GROUP_SCHED
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008303static DEFINE_MUTEX(shares_mutex);
8304
Ingo Molnar4cf86d72007-10-15 17:00:14 +02008305int sched_group_set_shares(struct task_group *tg, unsigned long shares)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008306{
8307 int i;
Paul Turner94371782010-11-15 15:47:10 -08008308 unsigned long flags;
Ingo Molnarc61935f2008-01-22 11:24:58 +01008309
Peter Zijlstra62fb1852008-02-25 17:34:02 +01008310 /*
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008311 * We can't change the weight of the root cgroup.
8312 */
8313 if (!tg->se[0])
8314 return -EINVAL;
8315
Peter Zijlstra18d95a22008-04-19 19:45:00 +02008316 if (shares < MIN_SHARES)
8317 shares = MIN_SHARES;
Miao Xiecb4ad1f2008-04-28 12:54:56 +08008318 else if (shares > MAX_SHARES)
8319 shares = MAX_SHARES;
Peter Zijlstra62fb1852008-02-25 17:34:02 +01008320
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008321 mutex_lock(&shares_mutex);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008322 if (tg->shares == shares)
Dhaval Giani5cb350b2007-10-15 17:00:14 +02008323 goto done;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008324
Srivatsa Vaddagiri6b2d7702008-01-25 21:08:00 +01008325 tg->shares = shares;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008326 for_each_possible_cpu(i) {
Paul Turner94371782010-11-15 15:47:10 -08008327 struct rq *rq = cpu_rq(i);
8328 struct sched_entity *se;
8329
8330 se = tg->se[i];
8331 /* Propagate contribution to hierarchy */
8332 raw_spin_lock_irqsave(&rq->lock, flags);
8333 for_each_sched_entity(se)
8334 update_cfs_shares(group_cfs_rq(se), 0);
8335 raw_spin_unlock_irqrestore(&rq->lock, flags);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008336 }
Srivatsa Vaddagiri6b2d7702008-01-25 21:08:00 +01008337
Dhaval Giani5cb350b2007-10-15 17:00:14 +02008338done:
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008339 mutex_unlock(&shares_mutex);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008340 return 0;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008341}
8342
Dhaval Giani5cb350b2007-10-15 17:00:14 +02008343unsigned long sched_group_shares(struct task_group *tg)
8344{
8345 return tg->shares;
8346}
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008347#endif
Dhaval Giani5cb350b2007-10-15 17:00:14 +02008348
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008349#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008350/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008351 * Ensure that the real time constraints are schedulable.
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008352 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008353static DEFINE_MUTEX(rt_constraints_mutex);
8354
8355static unsigned long to_ratio(u64 period, u64 runtime)
8356{
8357 if (runtime == RUNTIME_INF)
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008358 return 1ULL << 20;
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008359
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008360 return div64_u64(runtime << 20, period);
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008361}
8362
Dhaval Giani521f1a242008-02-28 15:21:56 +05308363/* Must be called with tasklist_lock held */
8364static inline int tg_has_rt_tasks(struct task_group *tg)
8365{
8366 struct task_struct *g, *p;
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008367
Dhaval Giani521f1a242008-02-28 15:21:56 +05308368 do_each_thread(g, p) {
8369 if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg)
8370 return 1;
8371 } while_each_thread(g, p);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008372
Dhaval Giani521f1a242008-02-28 15:21:56 +05308373 return 0;
8374}
8375
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008376struct rt_schedulable_data {
8377 struct task_group *tg;
8378 u64 rt_period;
8379 u64 rt_runtime;
8380};
8381
8382static int tg_schedulable(struct task_group *tg, void *data)
8383{
8384 struct rt_schedulable_data *d = data;
8385 struct task_group *child;
8386 unsigned long total, sum = 0;
8387 u64 period, runtime;
8388
8389 period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8390 runtime = tg->rt_bandwidth.rt_runtime;
8391
8392 if (tg == d->tg) {
8393 period = d->rt_period;
8394 runtime = d->rt_runtime;
8395 }
8396
Peter Zijlstra4653f802008-09-23 15:33:44 +02008397 /*
8398 * Cannot have more runtime than the period.
8399 */
8400 if (runtime > period && runtime != RUNTIME_INF)
8401 return -EINVAL;
8402
8403 /*
8404 * Ensure we don't starve existing RT tasks.
8405 */
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008406 if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
8407 return -EBUSY;
8408
8409 total = to_ratio(period, runtime);
8410
Peter Zijlstra4653f802008-09-23 15:33:44 +02008411 /*
8412 * Nobody can have more than the global setting allows.
8413 */
8414 if (total > to_ratio(global_rt_period(), global_rt_runtime()))
8415 return -EINVAL;
8416
8417 /*
8418 * The sum of our children's runtime should not exceed our own.
8419 */
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008420 list_for_each_entry_rcu(child, &tg->children, siblings) {
8421 period = ktime_to_ns(child->rt_bandwidth.rt_period);
8422 runtime = child->rt_bandwidth.rt_runtime;
8423
8424 if (child == d->tg) {
8425 period = d->rt_period;
8426 runtime = d->rt_runtime;
8427 }
8428
8429 sum += to_ratio(period, runtime);
8430 }
8431
8432 if (sum > total)
8433 return -EINVAL;
8434
8435 return 0;
8436}
8437
8438static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
8439{
8440 struct rt_schedulable_data data = {
8441 .tg = tg,
8442 .rt_period = period,
8443 .rt_runtime = runtime,
8444 };
8445
8446 return walk_tg_tree(tg_schedulable, tg_nop, &data);
8447}
8448
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008449static int tg_set_bandwidth(struct task_group *tg,
8450 u64 rt_period, u64 rt_runtime)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008451{
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008452 int i, err = 0;
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008453
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008454 mutex_lock(&rt_constraints_mutex);
Dhaval Giani521f1a242008-02-28 15:21:56 +05308455 read_lock(&tasklist_lock);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008456 err = __rt_schedulable(tg, rt_period, rt_runtime);
8457 if (err)
Dhaval Giani521f1a242008-02-28 15:21:56 +05308458 goto unlock;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008459
Thomas Gleixner0986b112009-11-17 15:32:06 +01008460 raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008461 tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
8462 tg->rt_bandwidth.rt_runtime = rt_runtime;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008463
8464 for_each_possible_cpu(i) {
8465 struct rt_rq *rt_rq = tg->rt_rq[i];
8466
Thomas Gleixner0986b112009-11-17 15:32:06 +01008467 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008468 rt_rq->rt_runtime = rt_runtime;
Thomas Gleixner0986b112009-11-17 15:32:06 +01008469 raw_spin_unlock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008470 }
Thomas Gleixner0986b112009-11-17 15:32:06 +01008471 raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
Peter Zijlstra49246272010-10-17 21:46:10 +02008472unlock:
Dhaval Giani521f1a242008-02-28 15:21:56 +05308473 read_unlock(&tasklist_lock);
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008474 mutex_unlock(&rt_constraints_mutex);
8475
8476 return err;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008477}
8478
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008479int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
8480{
8481 u64 rt_runtime, rt_period;
8482
8483 rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8484 rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
8485 if (rt_runtime_us < 0)
8486 rt_runtime = RUNTIME_INF;
8487
8488 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8489}
8490
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008491long sched_group_rt_runtime(struct task_group *tg)
8492{
8493 u64 rt_runtime_us;
8494
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008495 if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008496 return -1;
8497
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008498 rt_runtime_us = tg->rt_bandwidth.rt_runtime;
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008499 do_div(rt_runtime_us, NSEC_PER_USEC);
8500 return rt_runtime_us;
8501}
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008502
8503int sched_group_set_rt_period(struct task_group *tg, long rt_period_us)
8504{
8505 u64 rt_runtime, rt_period;
8506
8507 rt_period = (u64)rt_period_us * NSEC_PER_USEC;
8508 rt_runtime = tg->rt_bandwidth.rt_runtime;
8509
Raistlin619b0482008-06-26 18:54:09 +02008510 if (rt_period == 0)
8511 return -EINVAL;
8512
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008513 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8514}
8515
8516long sched_group_rt_period(struct task_group *tg)
8517{
8518 u64 rt_period_us;
8519
8520 rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
8521 do_div(rt_period_us, NSEC_PER_USEC);
8522 return rt_period_us;
8523}
8524
8525static int sched_rt_global_constraints(void)
8526{
Peter Zijlstra4653f802008-09-23 15:33:44 +02008527 u64 runtime, period;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008528 int ret = 0;
8529
Hiroshi Shimamotoec5d4982008-09-10 17:00:19 -07008530 if (sysctl_sched_rt_period <= 0)
8531 return -EINVAL;
8532
Peter Zijlstra4653f802008-09-23 15:33:44 +02008533 runtime = global_rt_runtime();
8534 period = global_rt_period();
8535
8536 /*
8537 * Sanity check on the sysctl variables.
8538 */
8539 if (runtime > period && runtime != RUNTIME_INF)
8540 return -EINVAL;
Peter Zijlstra10b612f2008-06-19 14:22:27 +02008541
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008542 mutex_lock(&rt_constraints_mutex);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008543 read_lock(&tasklist_lock);
Peter Zijlstra4653f802008-09-23 15:33:44 +02008544 ret = __rt_schedulable(NULL, 0, 0);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008545 read_unlock(&tasklist_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008546 mutex_unlock(&rt_constraints_mutex);
8547
8548 return ret;
8549}
Dhaval Giani54e99122009-02-27 15:13:54 +05308550
8551int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
8552{
8553 /* Don't accept realtime tasks when there is no way for them to run */
8554 if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0)
8555 return 0;
8556
8557 return 1;
8558}
8559
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008560#else /* !CONFIG_RT_GROUP_SCHED */
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008561static int sched_rt_global_constraints(void)
8562{
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008563 unsigned long flags;
8564 int i;
8565
Hiroshi Shimamotoec5d4982008-09-10 17:00:19 -07008566 if (sysctl_sched_rt_period <= 0)
8567 return -EINVAL;
8568
Peter Zijlstra60aa6052009-05-05 17:50:21 +02008569 /*
8570 * There's always some RT tasks in the root group
8571 * -- migration, kstopmachine etc..
8572 */
8573 if (sysctl_sched_rt_runtime == 0)
8574 return -EBUSY;
8575
Thomas Gleixner0986b112009-11-17 15:32:06 +01008576 raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008577 for_each_possible_cpu(i) {
8578 struct rt_rq *rt_rq = &cpu_rq(i)->rt;
8579
Thomas Gleixner0986b112009-11-17 15:32:06 +01008580 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008581 rt_rq->rt_runtime = global_rt_runtime();
Thomas Gleixner0986b112009-11-17 15:32:06 +01008582 raw_spin_unlock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008583 }
Thomas Gleixner0986b112009-11-17 15:32:06 +01008584 raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008585
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008586 return 0;
8587}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008588#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008589
8590int sched_rt_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07008591 void __user *buffer, size_t *lenp,
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008592 loff_t *ppos)
8593{
8594 int ret;
8595 int old_period, old_runtime;
8596 static DEFINE_MUTEX(mutex);
8597
8598 mutex_lock(&mutex);
8599 old_period = sysctl_sched_rt_period;
8600 old_runtime = sysctl_sched_rt_runtime;
8601
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07008602 ret = proc_dointvec(table, write, buffer, lenp, ppos);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008603
8604 if (!ret && write) {
8605 ret = sched_rt_global_constraints();
8606 if (ret) {
8607 sysctl_sched_rt_period = old_period;
8608 sysctl_sched_rt_runtime = old_runtime;
8609 } else {
8610 def_rt_bandwidth.rt_runtime = global_rt_runtime();
8611 def_rt_bandwidth.rt_period =
8612 ns_to_ktime(global_rt_period());
8613 }
8614 }
8615 mutex_unlock(&mutex);
8616
8617 return ret;
8618}
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008619
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008620#ifdef CONFIG_CGROUP_SCHED
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008621
8622/* return corresponding task_group object of a cgroup */
Paul Menage2b01dfe2007-10-24 18:23:50 +02008623static inline struct task_group *cgroup_tg(struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008624{
Paul Menage2b01dfe2007-10-24 18:23:50 +02008625 return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id),
8626 struct task_group, css);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008627}
8628
8629static struct cgroup_subsys_state *
Paul Menage2b01dfe2007-10-24 18:23:50 +02008630cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008631{
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008632 struct task_group *tg, *parent;
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008633
Paul Menage2b01dfe2007-10-24 18:23:50 +02008634 if (!cgrp->parent) {
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008635 /* This is early initialization for the top cgroup */
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008636 return &init_task_group.css;
8637 }
8638
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008639 parent = cgroup_tg(cgrp->parent);
8640 tg = sched_create_group(parent);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008641 if (IS_ERR(tg))
8642 return ERR_PTR(-ENOMEM);
8643
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008644 return &tg->css;
8645}
8646
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01008647static void
8648cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008649{
Paul Menage2b01dfe2007-10-24 18:23:50 +02008650 struct task_group *tg = cgroup_tg(cgrp);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008651
8652 sched_destroy_group(tg);
8653}
8654
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01008655static int
Ben Blumbe367d02009-09-23 15:56:31 -07008656cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008657{
Peter Zijlstrab68aa232008-02-13 15:45:40 +01008658#ifdef CONFIG_RT_GROUP_SCHED
Dhaval Giani54e99122009-02-27 15:13:54 +05308659 if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk))
Peter Zijlstrab68aa232008-02-13 15:45:40 +01008660 return -EINVAL;
8661#else
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008662 /* We don't support RT-tasks being in separate groups */
8663 if (tsk->sched_class != &fair_sched_class)
8664 return -EINVAL;
Peter Zijlstrab68aa232008-02-13 15:45:40 +01008665#endif
Ben Blumbe367d02009-09-23 15:56:31 -07008666 return 0;
8667}
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008668
Ben Blumbe367d02009-09-23 15:56:31 -07008669static int
8670cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
8671 struct task_struct *tsk, bool threadgroup)
8672{
8673 int retval = cpu_cgroup_can_attach_task(cgrp, tsk);
8674 if (retval)
8675 return retval;
8676 if (threadgroup) {
8677 struct task_struct *c;
8678 rcu_read_lock();
8679 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) {
8680 retval = cpu_cgroup_can_attach_task(cgrp, c);
8681 if (retval) {
8682 rcu_read_unlock();
8683 return retval;
8684 }
8685 }
8686 rcu_read_unlock();
8687 }
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008688 return 0;
8689}
8690
8691static void
Paul Menage2b01dfe2007-10-24 18:23:50 +02008692cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
Ben Blumbe367d02009-09-23 15:56:31 -07008693 struct cgroup *old_cont, struct task_struct *tsk,
8694 bool threadgroup)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008695{
8696 sched_move_task(tsk);
Ben Blumbe367d02009-09-23 15:56:31 -07008697 if (threadgroup) {
8698 struct task_struct *c;
8699 rcu_read_lock();
8700 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) {
8701 sched_move_task(c);
8702 }
8703 rcu_read_unlock();
8704 }
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008705}
8706
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008707#ifdef CONFIG_FAIR_GROUP_SCHED
Paul Menagef4c753b2008-04-29 00:59:56 -07008708static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype,
Paul Menage2b01dfe2007-10-24 18:23:50 +02008709 u64 shareval)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008710{
Paul Menage2b01dfe2007-10-24 18:23:50 +02008711 return sched_group_set_shares(cgroup_tg(cgrp), shareval);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008712}
8713
Paul Menagef4c753b2008-04-29 00:59:56 -07008714static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008715{
Paul Menage2b01dfe2007-10-24 18:23:50 +02008716 struct task_group *tg = cgroup_tg(cgrp);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008717
8718 return (u64) tg->shares;
8719}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008720#endif /* CONFIG_FAIR_GROUP_SCHED */
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008721
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008722#ifdef CONFIG_RT_GROUP_SCHED
Mirco Tischler0c708142008-05-14 16:05:46 -07008723static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft,
Paul Menage06ecb272008-04-29 01:00:06 -07008724 s64 val)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008725{
Paul Menage06ecb272008-04-29 01:00:06 -07008726 return sched_group_set_rt_runtime(cgroup_tg(cgrp), val);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008727}
8728
Paul Menage06ecb272008-04-29 01:00:06 -07008729static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008730{
Paul Menage06ecb272008-04-29 01:00:06 -07008731 return sched_group_rt_runtime(cgroup_tg(cgrp));
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008732}
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008733
8734static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype,
8735 u64 rt_period_us)
8736{
8737 return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us);
8738}
8739
8740static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft)
8741{
8742 return sched_group_rt_period(cgroup_tg(cgrp));
8743}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008744#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008745
Paul Menagefe5c7cc2007-10-29 21:18:11 +01008746static struct cftype cpu_files[] = {
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008747#ifdef CONFIG_FAIR_GROUP_SCHED
Paul Menagefe5c7cc2007-10-29 21:18:11 +01008748 {
8749 .name = "shares",
Paul Menagef4c753b2008-04-29 00:59:56 -07008750 .read_u64 = cpu_shares_read_u64,
8751 .write_u64 = cpu_shares_write_u64,
Paul Menagefe5c7cc2007-10-29 21:18:11 +01008752 },
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008753#endif
8754#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008755 {
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008756 .name = "rt_runtime_us",
Paul Menage06ecb272008-04-29 01:00:06 -07008757 .read_s64 = cpu_rt_runtime_read,
8758 .write_s64 = cpu_rt_runtime_write,
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008759 },
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008760 {
8761 .name = "rt_period_us",
Paul Menagef4c753b2008-04-29 00:59:56 -07008762 .read_u64 = cpu_rt_period_read_uint,
8763 .write_u64 = cpu_rt_period_write_uint,
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008764 },
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008765#endif
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008766};
8767
8768static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont)
8769{
Paul Menagefe5c7cc2007-10-29 21:18:11 +01008770 return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files));
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008771}
8772
8773struct cgroup_subsys cpu_cgroup_subsys = {
Ingo Molnar38605ca2007-10-29 21:18:11 +01008774 .name = "cpu",
8775 .create = cpu_cgroup_create,
8776 .destroy = cpu_cgroup_destroy,
8777 .can_attach = cpu_cgroup_can_attach,
8778 .attach = cpu_cgroup_attach,
8779 .populate = cpu_cgroup_populate,
8780 .subsys_id = cpu_cgroup_subsys_id,
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008781 .early_init = 1,
8782};
8783
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008784#endif /* CONFIG_CGROUP_SCHED */
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008785
8786#ifdef CONFIG_CGROUP_CPUACCT
8787
8788/*
8789 * CPU accounting code for task groups.
8790 *
8791 * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
8792 * (balbir@in.ibm.com).
8793 */
8794
Bharata B Rao934352f2008-11-10 20:41:13 +05308795/* track cpu usage of a group of tasks and its child groups */
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008796struct cpuacct {
8797 struct cgroup_subsys_state css;
8798 /* cpuusage holds pointer to a u64-type object on every cpu */
Tejun Heo43cf38e2010-02-02 14:38:57 +09008799 u64 __percpu *cpuusage;
Bharata B Raoef12fef2009-03-31 10:02:22 +05308800 struct percpu_counter cpustat[CPUACCT_STAT_NSTATS];
Bharata B Rao934352f2008-11-10 20:41:13 +05308801 struct cpuacct *parent;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008802};
8803
8804struct cgroup_subsys cpuacct_subsys;
8805
8806/* return cpu accounting group corresponding to this container */
Dhaval Giani32cd7562008-02-29 10:02:43 +05308807static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008808{
Dhaval Giani32cd7562008-02-29 10:02:43 +05308809 return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id),
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008810 struct cpuacct, css);
8811}
8812
8813/* return cpu accounting group to which this task belongs */
8814static inline struct cpuacct *task_ca(struct task_struct *tsk)
8815{
8816 return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
8817 struct cpuacct, css);
8818}
8819
8820/* create a new cpu accounting group */
8821static struct cgroup_subsys_state *cpuacct_create(
Dhaval Giani32cd7562008-02-29 10:02:43 +05308822 struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008823{
8824 struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
Bharata B Raoef12fef2009-03-31 10:02:22 +05308825 int i;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008826
8827 if (!ca)
Bharata B Raoef12fef2009-03-31 10:02:22 +05308828 goto out;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008829
8830 ca->cpuusage = alloc_percpu(u64);
Bharata B Raoef12fef2009-03-31 10:02:22 +05308831 if (!ca->cpuusage)
8832 goto out_free_ca;
8833
8834 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
8835 if (percpu_counter_init(&ca->cpustat[i], 0))
8836 goto out_free_counters;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008837
Bharata B Rao934352f2008-11-10 20:41:13 +05308838 if (cgrp->parent)
8839 ca->parent = cgroup_ca(cgrp->parent);
8840
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008841 return &ca->css;
Bharata B Raoef12fef2009-03-31 10:02:22 +05308842
8843out_free_counters:
8844 while (--i >= 0)
8845 percpu_counter_destroy(&ca->cpustat[i]);
8846 free_percpu(ca->cpuusage);
8847out_free_ca:
8848 kfree(ca);
8849out:
8850 return ERR_PTR(-ENOMEM);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008851}
8852
8853/* destroy an existing cpu accounting group */
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01008854static void
Dhaval Giani32cd7562008-02-29 10:02:43 +05308855cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008856{
Dhaval Giani32cd7562008-02-29 10:02:43 +05308857 struct cpuacct *ca = cgroup_ca(cgrp);
Bharata B Raoef12fef2009-03-31 10:02:22 +05308858 int i;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008859
Bharata B Raoef12fef2009-03-31 10:02:22 +05308860 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
8861 percpu_counter_destroy(&ca->cpustat[i]);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008862 free_percpu(ca->cpuusage);
8863 kfree(ca);
8864}
8865
Ken Chen720f5492008-12-15 22:02:01 -08008866static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu)
8867{
Rusty Russellb36128c2009-02-20 16:29:08 +09008868 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
Ken Chen720f5492008-12-15 22:02:01 -08008869 u64 data;
8870
8871#ifndef CONFIG_64BIT
8872 /*
8873 * Take rq->lock to make 64-bit read safe on 32-bit platforms.
8874 */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008875 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08008876 data = *cpuusage;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008877 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08008878#else
8879 data = *cpuusage;
8880#endif
8881
8882 return data;
8883}
8884
8885static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val)
8886{
Rusty Russellb36128c2009-02-20 16:29:08 +09008887 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
Ken Chen720f5492008-12-15 22:02:01 -08008888
8889#ifndef CONFIG_64BIT
8890 /*
8891 * Take rq->lock to make 64-bit write safe on 32-bit platforms.
8892 */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008893 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08008894 *cpuusage = val;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008895 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08008896#else
8897 *cpuusage = val;
8898#endif
8899}
8900
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008901/* return total cpu usage (in nanoseconds) of a group */
Dhaval Giani32cd7562008-02-29 10:02:43 +05308902static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008903{
Dhaval Giani32cd7562008-02-29 10:02:43 +05308904 struct cpuacct *ca = cgroup_ca(cgrp);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008905 u64 totalcpuusage = 0;
8906 int i;
8907
Ken Chen720f5492008-12-15 22:02:01 -08008908 for_each_present_cpu(i)
8909 totalcpuusage += cpuacct_cpuusage_read(ca, i);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008910
8911 return totalcpuusage;
8912}
8913
Dhaval Giani0297b802008-02-29 10:02:44 +05308914static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype,
8915 u64 reset)
8916{
8917 struct cpuacct *ca = cgroup_ca(cgrp);
8918 int err = 0;
8919 int i;
8920
8921 if (reset) {
8922 err = -EINVAL;
8923 goto out;
8924 }
8925
Ken Chen720f5492008-12-15 22:02:01 -08008926 for_each_present_cpu(i)
8927 cpuacct_cpuusage_write(ca, i, 0);
Dhaval Giani0297b802008-02-29 10:02:44 +05308928
Dhaval Giani0297b802008-02-29 10:02:44 +05308929out:
8930 return err;
8931}
8932
Ken Chene9515c32008-12-15 22:04:15 -08008933static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft,
8934 struct seq_file *m)
8935{
8936 struct cpuacct *ca = cgroup_ca(cgroup);
8937 u64 percpu;
8938 int i;
8939
8940 for_each_present_cpu(i) {
8941 percpu = cpuacct_cpuusage_read(ca, i);
8942 seq_printf(m, "%llu ", (unsigned long long) percpu);
8943 }
8944 seq_printf(m, "\n");
8945 return 0;
8946}
8947
Bharata B Raoef12fef2009-03-31 10:02:22 +05308948static const char *cpuacct_stat_desc[] = {
8949 [CPUACCT_STAT_USER] = "user",
8950 [CPUACCT_STAT_SYSTEM] = "system",
8951};
8952
8953static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft,
8954 struct cgroup_map_cb *cb)
8955{
8956 struct cpuacct *ca = cgroup_ca(cgrp);
8957 int i;
8958
8959 for (i = 0; i < CPUACCT_STAT_NSTATS; i++) {
8960 s64 val = percpu_counter_read(&ca->cpustat[i]);
8961 val = cputime64_to_clock_t(val);
8962 cb->fill(cb, cpuacct_stat_desc[i], val);
8963 }
8964 return 0;
8965}
8966
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008967static struct cftype files[] = {
8968 {
8969 .name = "usage",
Paul Menagef4c753b2008-04-29 00:59:56 -07008970 .read_u64 = cpuusage_read,
8971 .write_u64 = cpuusage_write,
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008972 },
Ken Chene9515c32008-12-15 22:04:15 -08008973 {
8974 .name = "usage_percpu",
8975 .read_seq_string = cpuacct_percpu_seq_read,
8976 },
Bharata B Raoef12fef2009-03-31 10:02:22 +05308977 {
8978 .name = "stat",
8979 .read_map = cpuacct_stats_show,
8980 },
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008981};
8982
Dhaval Giani32cd7562008-02-29 10:02:43 +05308983static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008984{
Dhaval Giani32cd7562008-02-29 10:02:43 +05308985 return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files));
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008986}
8987
8988/*
8989 * charge this task's execution time to its accounting group.
8990 *
8991 * called with rq->lock held.
8992 */
8993static void cpuacct_charge(struct task_struct *tsk, u64 cputime)
8994{
8995 struct cpuacct *ca;
Bharata B Rao934352f2008-11-10 20:41:13 +05308996 int cpu;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008997
Li Zefanc40c6f82009-02-26 15:40:15 +08008998 if (unlikely(!cpuacct_subsys.active))
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008999 return;
9000
Bharata B Rao934352f2008-11-10 20:41:13 +05309001 cpu = task_cpu(tsk);
Bharata B Raoa18b83b2009-03-23 10:02:53 +05309002
9003 rcu_read_lock();
9004
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009005 ca = task_ca(tsk);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009006
Bharata B Rao934352f2008-11-10 20:41:13 +05309007 for (; ca; ca = ca->parent) {
Rusty Russellb36128c2009-02-20 16:29:08 +09009008 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009009 *cpuusage += cputime;
9010 }
Bharata B Raoa18b83b2009-03-23 10:02:53 +05309011
9012 rcu_read_unlock();
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009013}
9014
Bharata B Raoef12fef2009-03-31 10:02:22 +05309015/*
Anton Blanchardfa535a72010-02-02 14:46:13 -08009016 * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large
9017 * in cputime_t units. As a result, cpuacct_update_stats calls
9018 * percpu_counter_add with values large enough to always overflow the
9019 * per cpu batch limit causing bad SMP scalability.
9020 *
9021 * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we
9022 * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled
9023 * and enabled. We cap it at INT_MAX which is the largest allowed batch value.
9024 */
9025#ifdef CONFIG_SMP
9026#define CPUACCT_BATCH \
9027 min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX)
9028#else
9029#define CPUACCT_BATCH 0
9030#endif
9031
9032/*
Bharata B Raoef12fef2009-03-31 10:02:22 +05309033 * Charge the system/user time to the task's accounting group.
9034 */
9035static void cpuacct_update_stats(struct task_struct *tsk,
9036 enum cpuacct_stat_index idx, cputime_t val)
9037{
9038 struct cpuacct *ca;
Anton Blanchardfa535a72010-02-02 14:46:13 -08009039 int batch = CPUACCT_BATCH;
Bharata B Raoef12fef2009-03-31 10:02:22 +05309040
9041 if (unlikely(!cpuacct_subsys.active))
9042 return;
9043
9044 rcu_read_lock();
9045 ca = task_ca(tsk);
9046
9047 do {
Anton Blanchardfa535a72010-02-02 14:46:13 -08009048 __percpu_counter_add(&ca->cpustat[idx], val, batch);
Bharata B Raoef12fef2009-03-31 10:02:22 +05309049 ca = ca->parent;
9050 } while (ca);
9051 rcu_read_unlock();
9052}
9053
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009054struct cgroup_subsys cpuacct_subsys = {
9055 .name = "cpuacct",
9056 .create = cpuacct_create,
9057 .destroy = cpuacct_destroy,
9058 .populate = cpuacct_populate,
9059 .subsys_id = cpuacct_subsys_id,
9060};
9061#endif /* CONFIG_CGROUP_CPUACCT */
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009062
9063#ifndef CONFIG_SMP
9064
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009065void synchronize_sched_expedited(void)
9066{
Paul E. McKenneyfc390cd2010-05-06 11:42:52 -07009067 barrier();
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009068}
9069EXPORT_SYMBOL_GPL(synchronize_sched_expedited);
9070
9071#else /* #ifndef CONFIG_SMP */
9072
Paul E. McKenneycc631fb2010-05-06 18:49:21 +02009073static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0);
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009074
Paul E. McKenneycc631fb2010-05-06 18:49:21 +02009075static int synchronize_sched_expedited_cpu_stop(void *data)
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009076{
Tejun Heo969c7922010-05-06 18:49:21 +02009077 /*
9078 * There must be a full memory barrier on each affected CPU
9079 * between the time that try_stop_cpus() is called and the
9080 * time that it returns.
9081 *
9082 * In the current initial implementation of cpu_stop, the
9083 * above condition is already met when the control reaches
9084 * this point and the following smp_mb() is not strictly
9085 * necessary. Do smp_mb() anyway for documentation and
9086 * robustness against future implementation changes.
9087 */
Paul E. McKenneycc631fb2010-05-06 18:49:21 +02009088 smp_mb(); /* See above comment block. */
Tejun Heo969c7922010-05-06 18:49:21 +02009089 return 0;
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009090}
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009091
9092/*
9093 * Wait for an rcu-sched grace period to elapse, but use "big hammer"
9094 * approach to force grace period to end quickly. This consumes
9095 * significant time on all CPUs, and is thus not recommended for
9096 * any sort of common-case code.
9097 *
9098 * Note that it is illegal to call this function while holding any
9099 * lock that is acquired by a CPU-hotplug notifier. Failing to
9100 * observe this restriction will result in deadlock.
9101 */
9102void synchronize_sched_expedited(void)
9103{
Tejun Heo969c7922010-05-06 18:49:21 +02009104 int snap, trycount = 0;
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009105
9106 smp_mb(); /* ensure prior mod happens before capturing snap. */
Tejun Heo969c7922010-05-06 18:49:21 +02009107 snap = atomic_read(&synchronize_sched_expedited_count) + 1;
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009108 get_online_cpus();
Tejun Heo969c7922010-05-06 18:49:21 +02009109 while (try_stop_cpus(cpu_online_mask,
9110 synchronize_sched_expedited_cpu_stop,
Tejun Heo94458d52010-05-06 18:49:21 +02009111 NULL) == -EAGAIN) {
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009112 put_online_cpus();
9113 if (trycount++ < 10)
9114 udelay(trycount * num_online_cpus());
9115 else {
9116 synchronize_sched();
9117 return;
9118 }
Tejun Heo969c7922010-05-06 18:49:21 +02009119 if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) {
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009120 smp_mb(); /* ensure test happens before caller kfree */
9121 return;
9122 }
9123 get_online_cpus();
9124 }
Tejun Heo969c7922010-05-06 18:49:21 +02009125 atomic_inc(&synchronize_sched_expedited_count);
Paul E. McKenneycc631fb2010-05-06 18:49:21 +02009126 smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009127 put_online_cpus();
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009128}
9129EXPORT_SYMBOL_GPL(synchronize_sched_expedited);
9130
9131#endif /* #else #ifndef CONFIG_SMP */