blob: dc85ceb908322cad7196339f4df8dd58c37b1cec [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * kernel/sched.c
3 *
4 * Kernel scheduler and related syscalls
5 *
6 * Copyright (C) 1991-2002 Linus Torvalds
7 *
8 * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and
9 * make semaphores SMP safe
10 * 1998-11-19 Implemented schedule_timeout() and related stuff
11 * by Andrea Arcangeli
12 * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar:
13 * hybrid priority-list and round-robin design with
14 * an array-switch method of distributing timeslices
15 * and per-CPU runqueues. Cleanups and useful suggestions
16 * by Davide Libenzi, preemptible kernel bits by Robert Love.
17 * 2003-09-03 Interactivity tuning by Con Kolivas.
18 * 2004-04-02 Scheduler domains code by Nick Piggin
Ingo Molnarc31f2e82007-07-09 18:52:01 +020019 * 2007-04-15 Work begun on replacing all interactivity tuning with a
20 * fair scheduling design by Con Kolivas.
21 * 2007-05-05 Load balancing (smp-nice) and other improvements
22 * by Peter Williams
23 * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith
24 * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri
Ingo Molnarb9131762008-01-25 21:08:19 +010025 * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins,
26 * Thomas Gleixner, Mike Kravetz
Linus Torvalds1da177e2005-04-16 15:20:36 -070027 */
28
29#include <linux/mm.h>
30#include <linux/module.h>
31#include <linux/nmi.h>
32#include <linux/init.h>
Ingo Molnardff06c12007-07-09 18:52:00 +020033#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/highmem.h>
35#include <linux/smp_lock.h>
36#include <asm/mmu_context.h>
37#include <linux/interrupt.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080038#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/completion.h>
40#include <linux/kernel_stat.h>
Ingo Molnar9a11b49a2006-07-03 00:24:33 -070041#include <linux/debug_locks.h>
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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
Gregory Haskins6e0534f2008-05-12 21:21:01 +020079#include "sched_cpupri.h"
Tejun Heo21aa9af2010-06-08 21:40:37 +020080#include "workqueue_sched.h"
Gregory Haskins6e0534f2008-05-12 21:21:01 +020081
Steven Rostedta8d154b2009-04-10 09:36:00 -040082#define CREATE_TRACE_POINTS
Steven Rostedtad8d75f2009-04-14 19:39:12 -040083#include <trace/events/sched.h>
Steven Rostedta8d154b2009-04-10 09:36:00 -040084
Linus Torvalds1da177e2005-04-16 15:20:36 -070085/*
86 * Convert user-nice values [ -20 ... 0 ... 19 ]
87 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
88 * and back.
89 */
90#define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
91#define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
92#define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
93
94/*
95 * 'User priority' is the nice value converted to something we
96 * can work with better when scaling various scheduler parameters,
97 * it's a [ 0 ... 39 ] range.
98 */
99#define USER_PRIO(p) ((p)-MAX_RT_PRIO)
100#define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
101#define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
102
103/*
Ingo Molnard7876a02008-01-25 21:08:19 +0100104 * Helpers for converting nanosecond timing to jiffy resolution
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 */
Eric Dumazetd6322fa2007-11-09 22:39:38 +0100106#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200108#define NICE_0_LOAD SCHED_LOAD_SCALE
109#define NICE_0_SHIFT SCHED_LOAD_SHIFT
110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111/*
112 * These are the 'tuning knobs' of the scheduler:
113 *
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +0200114 * default timeslice is 100 msecs (used only for SCHED_RR tasks).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 * Timeslices get refilled after they expire.
116 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117#define DEF_TIMESLICE (100 * HZ / 1000)
Peter Williams2dd73a42006-06-27 02:54:34 -0700118
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200119/*
120 * single value that denotes runtime == period, ie unlimited time.
121 */
122#define RUNTIME_INF ((u64)~0ULL)
123
Ingo Molnare05606d2007-07-09 18:51:59 +0200124static inline int rt_policy(int policy)
125{
Roel Kluin3f33a7c2008-05-13 23:44:11 +0200126 if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR))
Ingo Molnare05606d2007-07-09 18:51:59 +0200127 return 1;
128 return 0;
129}
130
131static inline int task_has_rt_policy(struct task_struct *p)
132{
133 return rt_policy(p->policy);
134}
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136/*
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200137 * This is the priority-queue data structure of the RT scheduling class:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 */
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200139struct rt_prio_array {
140 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
141 struct list_head queue[MAX_RT_PRIO];
142};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200144struct rt_bandwidth {
Ingo Molnarea736ed2008-03-25 13:51:45 +0100145 /* nests inside the rq lock: */
Thomas Gleixner0986b112009-11-17 15:32:06 +0100146 raw_spinlock_t rt_runtime_lock;
Ingo Molnarea736ed2008-03-25 13:51:45 +0100147 ktime_t rt_period;
148 u64 rt_runtime;
149 struct hrtimer rt_period_timer;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200150};
151
152static struct rt_bandwidth def_rt_bandwidth;
153
154static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
155
156static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
157{
158 struct rt_bandwidth *rt_b =
159 container_of(timer, struct rt_bandwidth, rt_period_timer);
160 ktime_t now;
161 int overrun;
162 int idle = 0;
163
164 for (;;) {
165 now = hrtimer_cb_get_time(timer);
166 overrun = hrtimer_forward(timer, now, rt_b->rt_period);
167
168 if (!overrun)
169 break;
170
171 idle = do_sched_rt_period_timer(rt_b, overrun);
172 }
173
174 return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
175}
176
177static
178void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
179{
180 rt_b->rt_period = ns_to_ktime(period);
181 rt_b->rt_runtime = runtime;
182
Thomas Gleixner0986b112009-11-17 15:32:06 +0100183 raw_spin_lock_init(&rt_b->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200184
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200185 hrtimer_init(&rt_b->rt_period_timer,
186 CLOCK_MONOTONIC, HRTIMER_MODE_REL);
187 rt_b->rt_period_timer.function = sched_rt_period_timer;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200188}
189
Krzysztof Heltc8bfff62008-09-05 23:46:19 +0200190static inline int rt_bandwidth_enabled(void)
191{
192 return sysctl_sched_rt_runtime >= 0;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200193}
194
195static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
196{
197 ktime_t now;
198
Hiroshi Shimamotocac64d02009-02-25 09:59:26 -0800199 if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200200 return;
201
202 if (hrtimer_active(&rt_b->rt_period_timer))
203 return;
204
Thomas Gleixner0986b112009-11-17 15:32:06 +0100205 raw_spin_lock(&rt_b->rt_runtime_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200206 for (;;) {
Peter Zijlstra7f1e2ca2009-03-13 12:21:27 +0100207 unsigned long delta;
208 ktime_t soft, hard;
209
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200210 if (hrtimer_active(&rt_b->rt_period_timer))
211 break;
212
213 now = hrtimer_cb_get_time(&rt_b->rt_period_timer);
214 hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period);
Peter Zijlstra7f1e2ca2009-03-13 12:21:27 +0100215
216 soft = hrtimer_get_softexpires(&rt_b->rt_period_timer);
217 hard = hrtimer_get_expires(&rt_b->rt_period_timer);
218 delta = ktime_to_ns(ktime_sub(hard, soft));
219 __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta,
Arun R Bharadwaj5c333862009-04-16 12:14:37 +0530220 HRTIMER_MODE_ABS_PINNED, 0);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200221 }
Thomas Gleixner0986b112009-11-17 15:32:06 +0100222 raw_spin_unlock(&rt_b->rt_runtime_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200223}
224
225#ifdef CONFIG_RT_GROUP_SCHED
226static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
227{
228 hrtimer_cancel(&rt_b->rt_period_timer);
229}
230#endif
231
Heiko Carstens712555e2008-04-28 11:33:07 +0200232/*
233 * sched_domains_mutex serializes calls to arch_init_sched_domains,
234 * detach_destroy_domains and partition_sched_domains.
235 */
236static DEFINE_MUTEX(sched_domains_mutex);
237
Dhaval Giani7c941432010-01-20 13:26:18 +0100238#ifdef CONFIG_CGROUP_SCHED
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200239
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -0700240#include <linux/cgroup.h>
241
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200242struct cfs_rq;
243
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100244static LIST_HEAD(task_groups);
245
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200246/* task group related information */
Ingo Molnar4cf86d72007-10-15 17:00:14 +0200247struct task_group {
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -0700248 struct cgroup_subsys_state css;
Arun R Bharadwaj6c415b92008-12-01 20:49:05 +0530249
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100250#ifdef CONFIG_FAIR_GROUP_SCHED
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200251 /* schedulable entities of this group on each cpu */
252 struct sched_entity **se;
253 /* runqueue "owned" by this group on each cpu */
254 struct cfs_rq **cfs_rq;
255 unsigned long shares;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100256#endif
257
258#ifdef CONFIG_RT_GROUP_SCHED
259 struct sched_rt_entity **rt_se;
260 struct rt_rq **rt_rq;
261
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200262 struct rt_bandwidth rt_bandwidth;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100263#endif
Srivatsa Vaddagiri6b2d7702008-01-25 21:08:00 +0100264
Srivatsa Vaddagiriae8393e2007-10-29 21:18:11 +0100265 struct rcu_head rcu;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100266 struct list_head list;
Peter Zijlstraf473aa52008-04-19 19:45:00 +0200267
268 struct task_group *parent;
269 struct list_head siblings;
270 struct list_head children;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200271};
272
Peter Zijlstraeff766a2008-04-19 19:45:00 +0200273#define root_task_group init_task_group
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100274
Peter Zijlstra8ed36992008-02-13 15:45:39 +0100275/* task_group_lock serializes add/remove of task groups and also changes to
Srivatsa Vaddagiriec2c5072008-01-25 21:07:59 +0100276 * a task group's cpu shares.
277 */
Peter Zijlstra8ed36992008-02-13 15:45:39 +0100278static DEFINE_SPINLOCK(task_group_lock);
Srivatsa Vaddagiriec2c5072008-01-25 21:07:59 +0100279
Cyrill Gorcunove9036b32009-10-26 22:24:14 +0300280#ifdef CONFIG_FAIR_GROUP_SCHED
281
Peter Zijlstra57310a92009-03-09 13:56:21 +0100282#ifdef CONFIG_SMP
283static int root_task_group_empty(void)
284{
285 return list_empty(&root_task_group.children);
286}
287#endif
288
Srivatsa Vaddagiri93f992c2008-01-25 21:07:59 +0100289# define INIT_TASK_GROUP_LOAD NICE_0_LOAD
Srivatsa Vaddagiri24e377a2007-10-15 17:00:09 +0200290
Miao Xiecb4ad1f2008-04-28 12:54:56 +0800291/*
Lai Jiangshan2e084782008-06-12 16:42:58 +0800292 * A weight of 0 or 1 can cause arithmetics problems.
293 * A weight of a cfs_rq is the sum of weights of which entities
294 * are queued on this cfs_rq, so a weight of a entity should not be
295 * too large, so as the shares value of a task group.
Miao Xiecb4ad1f2008-04-28 12:54:56 +0800296 * (The default weight is 1024 - so there's no practical
297 * limitation from this.)
298 */
Peter Zijlstra18d95a22008-04-19 19:45:00 +0200299#define MIN_SHARES 2
Lai Jiangshan2e084782008-06-12 16:42:58 +0800300#define MAX_SHARES (1UL << 18)
Peter Zijlstra18d95a22008-04-19 19:45:00 +0200301
Srivatsa Vaddagiri93f992c2008-01-25 21:07:59 +0100302static int init_task_group_load = INIT_TASK_GROUP_LOAD;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100303#endif
304
305/* Default task group.
306 * Every task in system belong to this group at bootup.
307 */
Mike Travis434d53b2008-04-04 18:11:04 -0700308struct task_group init_task_group;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200309
Dhaval Giani7c941432010-01-20 13:26:18 +0100310#endif /* CONFIG_CGROUP_SCHED */
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200311
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200312/* CFS-related fields in a runqueue */
313struct cfs_rq {
314 struct load_weight load;
315 unsigned long nr_running;
316
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200317 u64 exec_clock;
Ingo Molnare9acbff2007-10-15 17:00:04 +0200318 u64 min_vruntime;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200319
320 struct rb_root tasks_timeline;
321 struct rb_node *rb_leftmost;
Peter Zijlstra4a55bd52008-04-19 19:45:00 +0200322
323 struct list_head tasks;
324 struct list_head *balance_iterator;
325
326 /*
327 * 'curr' points to currently running entity on this cfs_rq.
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200328 * It is set to NULL otherwise (i.e when none are currently running).
329 */
Peter Zijlstra47932412008-11-04 21:25:09 +0100330 struct sched_entity *curr, *next, *last;
Peter Zijlstraddc97292007-10-15 17:00:10 +0200331
Peter Zijlstra5ac5c4d2008-11-10 10:46:32 +0100332 unsigned int nr_spread_over;
Peter Zijlstraddc97292007-10-15 17:00:10 +0200333
Ingo Molnar62160e32007-10-15 17:00:03 +0200334#ifdef CONFIG_FAIR_GROUP_SCHED
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200335 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
336
Ingo Molnar41a2d6c2007-12-05 15:46:09 +0100337 /*
338 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200339 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
340 * (like users, containers etc.)
341 *
342 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
343 * list is used during load balance.
344 */
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
Peter Zijlstrac8cba852008-06-27 13:41:23 +0200362 /*
363 * this cpu's part of tg->shares
364 */
365 unsigned long shares;
Peter Zijlstraf1d239f2008-06-27 13:41:38 +0200366
367 /*
368 * load.weight at the time we set shares
369 */
370 unsigned long rq_weight;
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200371#endif
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200372#endif
373};
374
375/* Real-Time classes' related field in a runqueue: */
376struct rt_rq {
377 struct rt_prio_array active;
Steven Rostedt63489e42008-01-25 21:08:03 +0100378 unsigned long rt_nr_running;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100379#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
Gregory Haskinse864c492008-12-29 09:39:49 -0500380 struct {
381 int curr; /* highest queued rt task prio */
Gregory Haskins398a1532009-01-14 09:10:04 -0500382#ifdef CONFIG_SMP
Gregory Haskinse864c492008-12-29 09:39:49 -0500383 int next; /* next highest */
Gregory Haskins398a1532009-01-14 09:10:04 -0500384#endif
Gregory Haskinse864c492008-12-29 09:39:49 -0500385 } highest_prio;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100386#endif
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100387#ifdef CONFIG_SMP
Gregory Haskins73fe6aae2008-01-25 21:08:07 +0100388 unsigned long rt_nr_migratory;
Peter Zijlstraa1ba4d82009-04-01 18:40:15 +0200389 unsigned long rt_nr_total;
Gregory Haskinsa22d7fc2008-01-25 21:08:12 +0100390 int overloaded;
Gregory Haskins917b6272008-12-29 09:39:53 -0500391 struct plist_head pushable_tasks;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100392#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100393 int rt_throttled;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100394 u64 rt_time;
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200395 u64 rt_runtime;
Ingo Molnarea736ed2008-03-25 13:51:45 +0100396 /* Nests inside the rq lock: */
Thomas Gleixner0986b112009-11-17 15:32:06 +0100397 raw_spinlock_t rt_runtime_lock;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100398
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100399#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +0100400 unsigned long rt_nr_boosted;
401
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100402 struct rq *rq;
403 struct list_head leaf_rt_rq_list;
404 struct task_group *tg;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100405#endif
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200406};
407
Gregory Haskins57d885f2008-01-25 21:08:18 +0100408#ifdef CONFIG_SMP
409
410/*
411 * We add the notion of a root-domain which will be used to define per-domain
Ingo Molnar0eab9142008-01-25 21:08:19 +0100412 * variables. Each exclusive cpuset essentially defines an island domain by
413 * fully partitioning the member cpus from any other cpuset. Whenever a new
Gregory Haskins57d885f2008-01-25 21:08:18 +0100414 * exclusive cpuset is created, we also create and attach a new root-domain
415 * object.
416 *
Gregory Haskins57d885f2008-01-25 21:08:18 +0100417 */
418struct root_domain {
419 atomic_t refcount;
Rusty Russellc6c49272008-11-25 02:35:05 +1030420 cpumask_var_t span;
421 cpumask_var_t online;
Gregory Haskins637f5082008-01-25 21:08:18 +0100422
Ingo Molnar0eab9142008-01-25 21:08:19 +0100423 /*
Gregory Haskins637f5082008-01-25 21:08:18 +0100424 * The "RT overload" flag: it gets set if a CPU has more than
425 * one runnable RT task.
426 */
Rusty Russellc6c49272008-11-25 02:35:05 +1030427 cpumask_var_t rto_mask;
Ingo Molnar0eab9142008-01-25 21:08:19 +0100428 atomic_t rto_count;
Gregory Haskins6e0534f2008-05-12 21:21:01 +0200429#ifdef CONFIG_SMP
430 struct cpupri cpupri;
431#endif
Gregory Haskins57d885f2008-01-25 21:08:18 +0100432};
433
Gregory Haskinsdc938522008-01-25 21:08:26 +0100434/*
435 * By default the system creates a single root-domain with all cpus as
436 * members (mimicking the global state we have today).
437 */
Gregory Haskins57d885f2008-01-25 21:08:18 +0100438static struct root_domain def_root_domain;
439
440#endif
441
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200442/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 * This is the main, per-CPU runqueue data structure.
444 *
445 * Locking rule: those places that want to lock multiple runqueues
446 * (such as the load balancing or the thread migration code), lock
447 * acquire operations must be ordered by ascending &runqueue.
448 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700449struct rq {
Ingo Molnard8016492007-10-18 21:32:55 +0200450 /* runqueue lock: */
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100451 raw_spinlock_t lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
453 /*
454 * nr_running and cpu_load should be in the same cacheline because
455 * remote CPUs use both these fields when doing load calculation.
456 */
457 unsigned long nr_running;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200458 #define CPU_LOAD_IDX_MAX 5
459 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -0700460 unsigned long last_load_update_tick;
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -0700461#ifdef CONFIG_NO_HZ
Mike Galbraith39c0cbe2010-03-11 17:17:13 +0100462 u64 nohz_stamp;
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -0700463 unsigned char nohz_balance_kick;
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -0700464#endif
Mike Galbraitha64692a2010-03-11 17:16:20 +0100465 unsigned int skip_clock_update;
466
Ingo Molnard8016492007-10-18 21:32:55 +0200467 /* capture load from *all* tasks on this cpu: */
468 struct load_weight load;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200469 unsigned long nr_load_updates;
470 u64 nr_switches;
471
472 struct cfs_rq cfs;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100473 struct rt_rq rt;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100474
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200475#ifdef CONFIG_FAIR_GROUP_SCHED
Ingo Molnard8016492007-10-18 21:32:55 +0200476 /* list of leaf cfs_rq on this cpu: */
477 struct list_head leaf_cfs_rq_list;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100478#endif
479#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100480 struct list_head leaf_rt_rq_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
483 /*
484 * This is part of a global counter where only the total sum
485 * over all CPUs matters. A task can increase this counter on
486 * one CPU and if it got migrated afterwards it may decrease
487 * it on another CPU. Always updated under the runqueue lock:
488 */
489 unsigned long nr_uninterruptible;
490
Ingo Molnar36c8b582006-07-03 00:25:41 -0700491 struct task_struct *curr, *idle;
Christoph Lameterc9819f42006-12-10 02:20:25 -0800492 unsigned long next_balance;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 struct mm_struct *prev_mm;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200494
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200495 u64 clock;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200496
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 atomic_t nr_iowait;
498
499#ifdef CONFIG_SMP
Ingo Molnar0eab9142008-01-25 21:08:19 +0100500 struct root_domain *rd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 struct sched_domain *sd;
502
Peter Zijlstrae51fd5e2010-05-31 12:37:30 +0200503 unsigned long cpu_power;
504
Henrik Austada0a522c2009-02-13 20:35:45 +0100505 unsigned char idle_at_tick;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 /* For active balancing */
Gregory Haskins3f029d32009-07-29 11:08:47 -0400507 int post_schedule;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 int active_balance;
509 int push_cpu;
Tejun Heo969c7922010-05-06 18:49:21 +0200510 struct cpu_stop_work active_balance_work;
Ingo Molnard8016492007-10-18 21:32:55 +0200511 /* cpu of this runqueue: */
512 int cpu;
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -0400513 int online;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
Peter Zijlstraa8a51d52008-06-27 13:41:26 +0200515 unsigned long avg_load_per_task;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
Peter Zijlstrae9e92502009-09-01 10:34:37 +0200517 u64 rt_avg;
518 u64 age_stamp;
Mike Galbraith1b9508f2009-11-04 17:53:50 +0100519 u64 idle_stamp;
520 u64 avg_idle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521#endif
522
Thomas Gleixnerdce48a82009-04-11 10:43:41 +0200523 /* calc_load related fields */
524 unsigned long calc_load_update;
525 long calc_load_active;
526
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100527#ifdef CONFIG_SCHED_HRTICK
Peter Zijlstra31656512008-07-18 18:01:23 +0200528#ifdef CONFIG_SMP
529 int hrtick_csd_pending;
530 struct call_single_data hrtick_csd;
531#endif
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100532 struct hrtimer hrtick_timer;
533#endif
534
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535#ifdef CONFIG_SCHEDSTATS
536 /* latency stats */
537 struct sched_info rq_sched_info;
Ken Chen9c2c4802008-12-16 23:41:22 -0800538 unsigned long long rq_cpu_time;
539 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
541 /* sys_sched_yield() stats */
Ken Chen480b9432007-10-18 21:32:56 +0200542 unsigned int yld_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
544 /* schedule() stats */
Ken Chen480b9432007-10-18 21:32:56 +0200545 unsigned int sched_switch;
546 unsigned int sched_count;
547 unsigned int sched_goidle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
549 /* try_to_wake_up() stats */
Ken Chen480b9432007-10-18 21:32:56 +0200550 unsigned int ttwu_count;
551 unsigned int ttwu_local;
Ingo Molnarb8efb562007-10-15 17:00:10 +0200552
553 /* BKL stats */
Ken Chen480b9432007-10-18 21:32:56 +0200554 unsigned int bkl_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555#endif
556};
557
Fenghua Yuf34e3b62007-07-19 01:48:13 -0700558static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Peter Zijlstra7d478722009-09-14 19:55:44 +0200560static inline
561void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
Ingo Molnardd41f592007-07-09 18:51:59 +0200562{
Peter Zijlstra7d478722009-09-14 19:55:44 +0200563 rq->curr->sched_class->check_preempt_curr(rq, p, flags);
Mike Galbraitha64692a2010-03-11 17:16:20 +0100564
565 /*
566 * A queue event has occurred, and we're going to schedule. In
567 * this case, we can save a useless back to back clock update.
568 */
569 if (test_tsk_need_resched(p))
570 rq->skip_clock_update = 1;
Ingo Molnardd41f592007-07-09 18:51:59 +0200571}
572
Christoph Lameter0a2966b2006-09-25 23:30:51 -0700573static inline int cpu_of(struct rq *rq)
574{
575#ifdef CONFIG_SMP
576 return rq->cpu;
577#else
578 return 0;
579#endif
580}
581
Paul E. McKenney497f0ab2010-02-22 17:04:51 -0800582#define rcu_dereference_check_sched_domain(p) \
Paul E. McKenneyd11c5632010-02-22 17:04:50 -0800583 rcu_dereference_check((p), \
584 rcu_read_lock_sched_held() || \
585 lockdep_is_held(&sched_domains_mutex))
586
Ingo Molnar20d315d2007-07-09 18:51:58 +0200587/*
Nick Piggin674311d2005-06-25 14:57:27 -0700588 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -0700589 * See detach_destroy_domains: synchronize_sched for details.
Nick Piggin674311d2005-06-25 14:57:27 -0700590 *
591 * The domain tree of any CPU may only be accessed from within
592 * preempt-disabled sections.
593 */
Ingo Molnar48f24c42006-07-03 00:25:40 -0700594#define for_each_domain(cpu, __sd) \
Paul E. McKenney497f0ab2010-02-22 17:04:51 -0800595 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
597#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
598#define this_rq() (&__get_cpu_var(runqueues))
599#define task_rq(p) cpu_rq(task_cpu(p))
600#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
Hitoshi Mitake54d35f22009-06-29 14:44:57 +0900601#define raw_rq() (&__raw_get_cpu_var(runqueues))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200603#ifdef CONFIG_CGROUP_SCHED
604
605/*
606 * Return the group to which this tasks belongs.
607 *
608 * We use task_subsys_state_check() and extend the RCU verification
609 * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach()
610 * holds that lock for each task it moves into the cgroup. Therefore
611 * by holding that lock, we pin the task to the current cgroup.
612 */
613static inline struct task_group *task_group(struct task_struct *p)
614{
615 struct cgroup_subsys_state *css;
616
617 css = task_subsys_state_check(p, cpu_cgroup_subsys_id,
618 lockdep_is_held(&task_rq(p)->lock));
619 return container_of(css, struct task_group, css);
620}
621
622/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
623static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
624{
625#ifdef CONFIG_FAIR_GROUP_SCHED
626 p->se.cfs_rq = task_group(p)->cfs_rq[cpu];
627 p->se.parent = task_group(p)->se[cpu];
628#endif
629
630#ifdef CONFIG_RT_GROUP_SCHED
631 p->rt.rt_rq = task_group(p)->rt_rq[cpu];
632 p->rt.parent = task_group(p)->rt_se[cpu];
633#endif
634}
635
636#else /* CONFIG_CGROUP_SCHED */
637
638static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
639static inline struct task_group *task_group(struct task_struct *p)
640{
641 return NULL;
642}
643
644#endif /* CONFIG_CGROUP_SCHED */
645
Ingo Molnaraa9c4c02008-12-17 14:10:57 +0100646inline void update_rq_clock(struct rq *rq)
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200647{
Mike Galbraitha64692a2010-03-11 17:16:20 +0100648 if (!rq->skip_clock_update)
649 rq->clock = sched_clock_cpu(cpu_of(rq));
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200650}
651
Ingo Molnare436d802007-07-19 21:28:35 +0200652/*
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200653 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
654 */
655#ifdef CONFIG_SCHED_DEBUG
656# define const_debug __read_mostly
657#else
658# define const_debug static const
659#endif
660
Ingo Molnar017730c2008-05-12 21:20:52 +0200661/**
662 * runqueue_is_locked
Randy Dunlape17b38b2009-10-11 19:12:00 -0700663 * @cpu: the processor in question.
Ingo Molnar017730c2008-05-12 21:20:52 +0200664 *
665 * Returns true if the current cpu runqueue is locked.
666 * This interface allows printk to be called with the runqueue lock
667 * held and know whether or not it is OK to wake up the klogd.
668 */
Andrew Morton89f19f02009-09-19 11:55:44 -0700669int runqueue_is_locked(int cpu)
Ingo Molnar017730c2008-05-12 21:20:52 +0200670{
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100671 return raw_spin_is_locked(&cpu_rq(cpu)->lock);
Ingo Molnar017730c2008-05-12 21:20:52 +0200672}
673
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200674/*
675 * Debugging: various feature bits
676 */
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200677
678#define SCHED_FEAT(name, enabled) \
679 __SCHED_FEAT_##name ,
680
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200681enum {
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200682#include "sched_features.h"
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200683};
684
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200685#undef SCHED_FEAT
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200686
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200687#define SCHED_FEAT(name, enabled) \
688 (1UL << __SCHED_FEAT_##name) * enabled |
689
690const_debug unsigned int sysctl_sched_features =
691#include "sched_features.h"
692 0;
693
694#undef SCHED_FEAT
695
696#ifdef CONFIG_SCHED_DEBUG
697#define SCHED_FEAT(name, enabled) \
698 #name ,
699
Harvey Harrison983ed7a2008-04-24 18:17:55 -0700700static __read_mostly char *sched_feat_names[] = {
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200701#include "sched_features.h"
702 NULL
703};
704
705#undef SCHED_FEAT
706
Li Zefan34f3a812008-10-30 15:23:32 +0800707static int sched_feat_show(struct seq_file *m, void *v)
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200708{
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200709 int i;
710
711 for (i = 0; sched_feat_names[i]; i++) {
Li Zefan34f3a812008-10-30 15:23:32 +0800712 if (!(sysctl_sched_features & (1UL << i)))
713 seq_puts(m, "NO_");
714 seq_printf(m, "%s ", sched_feat_names[i]);
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200715 }
Li Zefan34f3a812008-10-30 15:23:32 +0800716 seq_puts(m, "\n");
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200717
Li Zefan34f3a812008-10-30 15:23:32 +0800718 return 0;
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200719}
720
721static ssize_t
722sched_feat_write(struct file *filp, const char __user *ubuf,
723 size_t cnt, loff_t *ppos)
724{
725 char buf[64];
726 char *cmp = buf;
727 int neg = 0;
728 int i;
729
730 if (cnt > 63)
731 cnt = 63;
732
733 if (copy_from_user(&buf, ubuf, cnt))
734 return -EFAULT;
735
736 buf[cnt] = 0;
737
Ingo Molnarc24b7c52008-04-18 10:55:34 +0200738 if (strncmp(buf, "NO_", 3) == 0) {
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200739 neg = 1;
740 cmp += 3;
741 }
742
743 for (i = 0; sched_feat_names[i]; i++) {
744 int len = strlen(sched_feat_names[i]);
745
746 if (strncmp(cmp, sched_feat_names[i], len) == 0) {
747 if (neg)
748 sysctl_sched_features &= ~(1UL << i);
749 else
750 sysctl_sched_features |= (1UL << i);
751 break;
752 }
753 }
754
755 if (!sched_feat_names[i])
756 return -EINVAL;
757
Jan Blunck42994722009-11-20 17:40:37 +0100758 *ppos += cnt;
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200759
760 return cnt;
761}
762
Li Zefan34f3a812008-10-30 15:23:32 +0800763static int sched_feat_open(struct inode *inode, struct file *filp)
764{
765 return single_open(filp, sched_feat_show, NULL);
766}
767
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700768static const struct file_operations sched_feat_fops = {
Li Zefan34f3a812008-10-30 15:23:32 +0800769 .open = sched_feat_open,
770 .write = sched_feat_write,
771 .read = seq_read,
772 .llseek = seq_lseek,
773 .release = single_release,
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200774};
775
776static __init int sched_init_debug(void)
777{
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200778 debugfs_create_file("sched_features", 0644, NULL, NULL,
779 &sched_feat_fops);
780
781 return 0;
782}
783late_initcall(sched_init_debug);
784
785#endif
786
787#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200788
789/*
Peter Zijlstrab82d9fd2007-11-09 22:39:39 +0100790 * Number of tasks to iterate in a single balance run.
791 * Limited because this is done with IRQs disabled.
792 */
793const_debug unsigned int sysctl_sched_nr_migrate = 32;
794
795/*
Peter Zijlstra2398f2c2008-06-27 13:41:35 +0200796 * ratelimit for updating the group shares.
Peter Zijlstra55cd5342008-08-04 08:54:26 +0200797 * default: 0.25ms
Peter Zijlstra2398f2c2008-06-27 13:41:35 +0200798 */
Peter Zijlstra55cd5342008-08-04 08:54:26 +0200799unsigned int sysctl_sched_shares_ratelimit = 250000;
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +0100800unsigned int normalized_sysctl_sched_shares_ratelimit = 250000;
Peter Zijlstra2398f2c2008-06-27 13:41:35 +0200801
802/*
Peter Zijlstraffda12a2008-10-17 19:27:02 +0200803 * Inject some fuzzyness into changing the per-cpu group shares
804 * this avoids remote rq-locks at the expense of fairness.
805 * default: 4
806 */
807unsigned int sysctl_sched_shares_thresh = 4;
808
809/*
Peter Zijlstrae9e92502009-09-01 10:34:37 +0200810 * period over which we average the RT time consumption, measured
811 * in ms.
812 *
813 * default: 1s
814 */
815const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC;
816
817/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100818 * period over which we measure -rt task cpu usage in us.
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100819 * default: 1s
820 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100821unsigned int sysctl_sched_rt_period = 1000000;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100822
Ingo Molnar6892b752008-02-13 14:02:36 +0100823static __read_mostly int scheduler_running;
824
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100825/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100826 * part of the period that we allow rt tasks to run in us.
827 * default: 0.95s
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100828 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100829int sysctl_sched_rt_runtime = 950000;
830
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200831static inline u64 global_rt_period(void)
832{
833 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
834}
835
836static inline u64 global_rt_runtime(void)
837{
roel kluine26873b2008-07-22 16:51:15 -0400838 if (sysctl_sched_rt_runtime < 0)
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200839 return RUNTIME_INF;
840
841 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
842}
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100843
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844#ifndef prepare_arch_switch
Nick Piggin4866cde2005-06-25 14:57:23 -0700845# define prepare_arch_switch(next) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846#endif
Nick Piggin4866cde2005-06-25 14:57:23 -0700847#ifndef finish_arch_switch
848# define finish_arch_switch(prev) do { } while (0)
849#endif
850
Dmitry Adamushko051a1d12007-12-18 15:21:13 +0100851static inline int task_current(struct rq *rq, struct task_struct *p)
852{
853 return rq->curr == p;
854}
855
Nick Piggin4866cde2005-06-25 14:57:23 -0700856#ifndef __ARCH_WANT_UNLOCKED_CTXSW
Ingo Molnar70b97a72006-07-03 00:25:42 -0700857static inline int task_running(struct rq *rq, struct task_struct *p)
Nick Piggin4866cde2005-06-25 14:57:23 -0700858{
Dmitry Adamushko051a1d12007-12-18 15:21:13 +0100859 return task_current(rq, p);
Nick Piggin4866cde2005-06-25 14:57:23 -0700860}
861
Ingo Molnar70b97a72006-07-03 00:25:42 -0700862static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
Nick Piggin4866cde2005-06-25 14:57:23 -0700863{
864}
865
Ingo Molnar70b97a72006-07-03 00:25:42 -0700866static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
Nick Piggin4866cde2005-06-25 14:57:23 -0700867{
Ingo Molnarda04c032005-09-13 11:17:59 +0200868#ifdef CONFIG_DEBUG_SPINLOCK
869 /* this is a valid case when another task releases the spinlock */
870 rq->lock.owner = current;
871#endif
Ingo Molnar8a25d5d2006-07-03 00:24:54 -0700872 /*
873 * If we are tracking spinlock dependencies then we have to
874 * fix up the runqueue lock - which gets 'carried over' from
875 * prev into current:
876 */
877 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
878
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100879 raw_spin_unlock_irq(&rq->lock);
Nick Piggin4866cde2005-06-25 14:57:23 -0700880}
881
882#else /* __ARCH_WANT_UNLOCKED_CTXSW */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700883static inline int task_running(struct rq *rq, struct task_struct *p)
Nick Piggin4866cde2005-06-25 14:57:23 -0700884{
885#ifdef CONFIG_SMP
886 return p->oncpu;
887#else
Dmitry Adamushko051a1d12007-12-18 15:21:13 +0100888 return task_current(rq, p);
Nick Piggin4866cde2005-06-25 14:57:23 -0700889#endif
890}
891
Ingo Molnar70b97a72006-07-03 00:25:42 -0700892static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
Nick Piggin4866cde2005-06-25 14:57:23 -0700893{
894#ifdef CONFIG_SMP
895 /*
896 * We can optimise this out completely for !SMP, because the
897 * SMP rebalancing from interrupt is the only thing that cares
898 * here.
899 */
900 next->oncpu = 1;
901#endif
902#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100903 raw_spin_unlock_irq(&rq->lock);
Nick Piggin4866cde2005-06-25 14:57:23 -0700904#else
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100905 raw_spin_unlock(&rq->lock);
Nick Piggin4866cde2005-06-25 14:57:23 -0700906#endif
907}
908
Ingo Molnar70b97a72006-07-03 00:25:42 -0700909static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
Nick Piggin4866cde2005-06-25 14:57:23 -0700910{
911#ifdef CONFIG_SMP
912 /*
913 * After ->oncpu is cleared, the task can be moved to a different CPU.
914 * We must ensure this doesn't happen until the switch is completely
915 * finished.
916 */
917 smp_wmb();
918 prev->oncpu = 0;
919#endif
920#ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
921 local_irq_enable();
922#endif
923}
924#endif /* __ARCH_WANT_UNLOCKED_CTXSW */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
926/*
Peter Zijlstra65cc8e42010-03-25 21:05:16 +0100927 * Check whether the task is waking, we use this to synchronize ->cpus_allowed
928 * against ttwu().
Peter Zijlstra0970d292010-02-15 14:45:54 +0100929 */
930static inline int task_is_waking(struct task_struct *p)
931{
Peter Zijlstra0017d732010-03-24 18:34:10 +0100932 return unlikely(p->state == TASK_WAKING);
Peter Zijlstra0970d292010-02-15 14:45:54 +0100933}
934
935/*
Ingo Molnarb29739f2006-06-27 02:54:51 -0700936 * __task_rq_lock - lock the runqueue a given task resides on.
937 * Must be called interrupts disabled.
938 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700939static inline struct rq *__task_rq_lock(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -0700940 __acquires(rq->lock)
941{
Peter Zijlstra0970d292010-02-15 14:45:54 +0100942 struct rq *rq;
943
Andi Kleen3a5c3592007-10-15 17:00:14 +0200944 for (;;) {
Peter Zijlstra0970d292010-02-15 14:45:54 +0100945 rq = task_rq(p);
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100946 raw_spin_lock(&rq->lock);
Peter Zijlstra65cc8e42010-03-25 21:05:16 +0100947 if (likely(rq == task_rq(p)))
Andi Kleen3a5c3592007-10-15 17:00:14 +0200948 return rq;
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100949 raw_spin_unlock(&rq->lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -0700950 }
Ingo Molnarb29739f2006-06-27 02:54:51 -0700951}
952
953/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 * task_rq_lock - lock the runqueue a given task resides on and disable
Ingo Molnar41a2d6c2007-12-05 15:46:09 +0100955 * interrupts. Note the ordering: we can safely lookup the task_rq without
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 * explicitly disabling preemption.
957 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700958static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 __acquires(rq->lock)
960{
Ingo Molnar70b97a72006-07-03 00:25:42 -0700961 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
Andi Kleen3a5c3592007-10-15 17:00:14 +0200963 for (;;) {
964 local_irq_save(*flags);
965 rq = task_rq(p);
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100966 raw_spin_lock(&rq->lock);
Peter Zijlstra65cc8e42010-03-25 21:05:16 +0100967 if (likely(rq == task_rq(p)))
Andi Kleen3a5c3592007-10-15 17:00:14 +0200968 return rq;
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100969 raw_spin_unlock_irqrestore(&rq->lock, *flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971}
972
Alexey Dobriyana9957442007-10-15 17:00:13 +0200973static void __task_rq_unlock(struct rq *rq)
Ingo Molnarb29739f2006-06-27 02:54:51 -0700974 __releases(rq->lock)
975{
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100976 raw_spin_unlock(&rq->lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -0700977}
978
Ingo Molnar70b97a72006-07-03 00:25:42 -0700979static inline void task_rq_unlock(struct rq *rq, unsigned long *flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 __releases(rq->lock)
981{
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100982 raw_spin_unlock_irqrestore(&rq->lock, *flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983}
984
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985/*
Robert P. J. Daycc2a73b2006-12-10 02:20:00 -0800986 * this_rq_lock - lock this runqueue and disable interrupts.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 */
Alexey Dobriyana9957442007-10-15 17:00:13 +0200988static struct rq *this_rq_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 __acquires(rq->lock)
990{
Ingo Molnar70b97a72006-07-03 00:25:42 -0700991 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
993 local_irq_disable();
994 rq = this_rq();
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100995 raw_spin_lock(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
997 return rq;
998}
999
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001000#ifdef CONFIG_SCHED_HRTICK
1001/*
1002 * Use HR-timers to deliver accurate preemption points.
1003 *
1004 * Its all a bit involved since we cannot program an hrt while holding the
1005 * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a
1006 * reschedule event.
1007 *
1008 * When we get rescheduled we reprogram the hrtick_timer outside of the
1009 * rq->lock.
1010 */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001011
1012/*
1013 * Use hrtick when:
1014 * - enabled by features
1015 * - hrtimer is actually high res
1016 */
1017static inline int hrtick_enabled(struct rq *rq)
1018{
1019 if (!sched_feat(HRTICK))
1020 return 0;
Ingo Molnarba420592008-07-20 11:02:06 +02001021 if (!cpu_active(cpu_of(rq)))
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001022 return 0;
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001023 return hrtimer_is_hres_active(&rq->hrtick_timer);
1024}
1025
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001026static void hrtick_clear(struct rq *rq)
1027{
1028 if (hrtimer_active(&rq->hrtick_timer))
1029 hrtimer_cancel(&rq->hrtick_timer);
1030}
1031
1032/*
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001033 * High-resolution timer tick.
1034 * Runs from hardirq context with interrupts disabled.
1035 */
1036static enum hrtimer_restart hrtick(struct hrtimer *timer)
1037{
1038 struct rq *rq = container_of(timer, struct rq, hrtick_timer);
1039
1040 WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
1041
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001042 raw_spin_lock(&rq->lock);
Peter Zijlstra3e51f332008-05-03 18:29:28 +02001043 update_rq_clock(rq);
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001044 rq->curr->sched_class->task_tick(rq, rq->curr, 1);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001045 raw_spin_unlock(&rq->lock);
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001046
1047 return HRTIMER_NORESTART;
1048}
1049
Rabin Vincent95e904c2008-05-11 05:55:33 +05301050#ifdef CONFIG_SMP
Peter Zijlstra31656512008-07-18 18:01:23 +02001051/*
1052 * called from hardirq (IPI) context
1053 */
1054static void __hrtick_start(void *arg)
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001055{
Peter Zijlstra31656512008-07-18 18:01:23 +02001056 struct rq *rq = arg;
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001057
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001058 raw_spin_lock(&rq->lock);
Peter Zijlstra31656512008-07-18 18:01:23 +02001059 hrtimer_restart(&rq->hrtick_timer);
1060 rq->hrtick_csd_pending = 0;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001061 raw_spin_unlock(&rq->lock);
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001062}
1063
Peter Zijlstra31656512008-07-18 18:01:23 +02001064/*
1065 * Called to set the hrtick timer state.
1066 *
1067 * called with rq->lock held and irqs disabled
1068 */
1069static void hrtick_start(struct rq *rq, u64 delay)
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001070{
Peter Zijlstra31656512008-07-18 18:01:23 +02001071 struct hrtimer *timer = &rq->hrtick_timer;
1072 ktime_t time = ktime_add_ns(timer->base->get_time(), delay);
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001073
Arjan van de Vencc584b22008-09-01 15:02:30 -07001074 hrtimer_set_expires(timer, time);
Peter Zijlstra31656512008-07-18 18:01:23 +02001075
1076 if (rq == this_rq()) {
1077 hrtimer_restart(timer);
1078 } else if (!rq->hrtick_csd_pending) {
Peter Zijlstra6e275632009-02-25 13:59:48 +01001079 __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0);
Peter Zijlstra31656512008-07-18 18:01:23 +02001080 rq->hrtick_csd_pending = 1;
1081 }
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001082}
1083
1084static int
1085hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu)
1086{
1087 int cpu = (int)(long)hcpu;
1088
1089 switch (action) {
1090 case CPU_UP_CANCELED:
1091 case CPU_UP_CANCELED_FROZEN:
1092 case CPU_DOWN_PREPARE:
1093 case CPU_DOWN_PREPARE_FROZEN:
1094 case CPU_DEAD:
1095 case CPU_DEAD_FROZEN:
Peter Zijlstra31656512008-07-18 18:01:23 +02001096 hrtick_clear(cpu_rq(cpu));
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001097 return NOTIFY_OK;
1098 }
1099
1100 return NOTIFY_DONE;
1101}
1102
Rakib Mullickfa748202008-09-22 14:55:45 -07001103static __init void init_hrtick(void)
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001104{
1105 hotcpu_notifier(hotplug_hrtick, 0);
1106}
Peter Zijlstra31656512008-07-18 18:01:23 +02001107#else
1108/*
1109 * Called to set the hrtick timer state.
1110 *
1111 * called with rq->lock held and irqs disabled
1112 */
1113static void hrtick_start(struct rq *rq, u64 delay)
1114{
Peter Zijlstra7f1e2ca2009-03-13 12:21:27 +01001115 __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0,
Arun R Bharadwaj5c333862009-04-16 12:14:37 +05301116 HRTIMER_MODE_REL_PINNED, 0);
Peter Zijlstra31656512008-07-18 18:01:23 +02001117}
1118
Andrew Morton006c75f2008-09-22 14:55:46 -07001119static inline void init_hrtick(void)
Peter Zijlstra31656512008-07-18 18:01:23 +02001120{
1121}
Rabin Vincent95e904c2008-05-11 05:55:33 +05301122#endif /* CONFIG_SMP */
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001123
1124static void init_rq_hrtick(struct rq *rq)
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001125{
Peter Zijlstra31656512008-07-18 18:01:23 +02001126#ifdef CONFIG_SMP
1127 rq->hrtick_csd_pending = 0;
1128
1129 rq->hrtick_csd.flags = 0;
1130 rq->hrtick_csd.func = __hrtick_start;
1131 rq->hrtick_csd.info = rq;
1132#endif
1133
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001134 hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1135 rq->hrtick_timer.function = hrtick;
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001136}
Andrew Morton006c75f2008-09-22 14:55:46 -07001137#else /* CONFIG_SCHED_HRTICK */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001138static inline void hrtick_clear(struct rq *rq)
1139{
1140}
1141
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001142static inline void init_rq_hrtick(struct rq *rq)
1143{
1144}
1145
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001146static inline void init_hrtick(void)
1147{
1148}
Andrew Morton006c75f2008-09-22 14:55:46 -07001149#endif /* CONFIG_SCHED_HRTICK */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001150
Ingo Molnar1b9f19c22007-07-09 18:51:59 +02001151/*
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001152 * resched_task - mark a task 'to be rescheduled now'.
1153 *
1154 * On UP this means the setting of the need_resched flag, on SMP it
1155 * might also involve a cross-CPU call to trigger the scheduler on
1156 * the target CPU.
1157 */
1158#ifdef CONFIG_SMP
1159
1160#ifndef tsk_is_polling
1161#define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG)
1162#endif
1163
Peter Zijlstra31656512008-07-18 18:01:23 +02001164static void resched_task(struct task_struct *p)
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001165{
1166 int cpu;
1167
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001168 assert_raw_spin_locked(&task_rq(p)->lock);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001169
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08001170 if (test_tsk_need_resched(p))
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001171 return;
1172
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08001173 set_tsk_need_resched(p);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001174
1175 cpu = task_cpu(p);
1176 if (cpu == smp_processor_id())
1177 return;
1178
1179 /* NEED_RESCHED must be visible before we test polling */
1180 smp_mb();
1181 if (!tsk_is_polling(p))
1182 smp_send_reschedule(cpu);
1183}
1184
1185static void resched_cpu(int cpu)
1186{
1187 struct rq *rq = cpu_rq(cpu);
1188 unsigned long flags;
1189
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001190 if (!raw_spin_trylock_irqsave(&rq->lock, flags))
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001191 return;
1192 resched_task(cpu_curr(cpu));
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001193 raw_spin_unlock_irqrestore(&rq->lock, flags);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001194}
Thomas Gleixner06d83082008-03-22 09:20:24 +01001195
1196#ifdef CONFIG_NO_HZ
1197/*
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -07001198 * In the semi idle case, use the nearest busy cpu for migrating timers
1199 * from an idle cpu. This is good for power-savings.
1200 *
1201 * We don't do similar optimization for completely idle system, as
1202 * selecting an idle cpu will add more delays to the timers than intended
1203 * (as that cpu's timer base may not be uptodate wrt jiffies etc).
1204 */
1205int get_nohz_timer_target(void)
1206{
1207 int cpu = smp_processor_id();
1208 int i;
1209 struct sched_domain *sd;
1210
1211 for_each_domain(cpu, sd) {
1212 for_each_cpu(i, sched_domain_span(sd))
1213 if (!idle_cpu(i))
1214 return i;
1215 }
1216 return cpu;
1217}
1218/*
Thomas Gleixner06d83082008-03-22 09:20:24 +01001219 * When add_timer_on() enqueues a timer into the timer wheel of an
1220 * idle CPU then this timer might expire before the next timer event
1221 * which is scheduled to wake up that CPU. In case of a completely
1222 * idle system the next event might even be infinite time into the
1223 * future. wake_up_idle_cpu() ensures that the CPU is woken up and
1224 * leaves the inner idle loop so the newly added timer is taken into
1225 * account when the CPU goes back to idle and evaluates the timer
1226 * wheel for the next timer event.
1227 */
1228void wake_up_idle_cpu(int cpu)
1229{
1230 struct rq *rq = cpu_rq(cpu);
1231
1232 if (cpu == smp_processor_id())
1233 return;
1234
1235 /*
1236 * This is safe, as this function is called with the timer
1237 * wheel base lock of (cpu) held. When the CPU is on the way
1238 * to idle and has not yet set rq->curr to idle then it will
1239 * be serialized on the timer wheel base lock and take the new
1240 * timer into account automatically.
1241 */
1242 if (rq->curr != rq->idle)
1243 return;
1244
1245 /*
1246 * We can set TIF_RESCHED on the idle task of the other CPU
1247 * lockless. The worst case is that the other CPU runs the
1248 * idle task through an additional NOOP schedule()
1249 */
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08001250 set_tsk_need_resched(rq->idle);
Thomas Gleixner06d83082008-03-22 09:20:24 +01001251
1252 /* NEED_RESCHED must be visible before we test polling */
1253 smp_mb();
1254 if (!tsk_is_polling(rq->idle))
1255 smp_send_reschedule(cpu);
1256}
Mike Galbraith39c0cbe2010-03-11 17:17:13 +01001257
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02001258#endif /* CONFIG_NO_HZ */
Thomas Gleixner06d83082008-03-22 09:20:24 +01001259
Peter Zijlstrae9e92502009-09-01 10:34:37 +02001260static u64 sched_avg_period(void)
1261{
1262 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1263}
1264
1265static void sched_avg_update(struct rq *rq)
1266{
1267 s64 period = sched_avg_period();
1268
1269 while ((s64)(rq->clock - rq->age_stamp) > period) {
Will Deacon0d98bb22010-05-24 12:11:43 -07001270 /*
1271 * Inline assembly required to prevent the compiler
1272 * optimising this loop into a divmod call.
1273 * See __iter_div_u64_rem() for another example of this.
1274 */
1275 asm("" : "+rm" (rq->age_stamp));
Peter Zijlstrae9e92502009-09-01 10:34:37 +02001276 rq->age_stamp += period;
1277 rq->rt_avg /= 2;
1278 }
1279}
1280
1281static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1282{
1283 rq->rt_avg += rt_delta;
1284 sched_avg_update(rq);
1285}
1286
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02001287#else /* !CONFIG_SMP */
Peter Zijlstra31656512008-07-18 18:01:23 +02001288static void resched_task(struct task_struct *p)
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001289{
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001290 assert_raw_spin_locked(&task_rq(p)->lock);
Peter Zijlstra31656512008-07-18 18:01:23 +02001291 set_tsk_need_resched(p);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001292}
Peter Zijlstrae9e92502009-09-01 10:34:37 +02001293
1294static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1295{
1296}
Suresh Siddhada2b71e2010-08-23 13:42:51 -07001297
1298static void sched_avg_update(struct rq *rq)
1299{
1300}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02001301#endif /* CONFIG_SMP */
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001302
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001303#if BITS_PER_LONG == 32
1304# define WMULT_CONST (~0UL)
1305#else
1306# define WMULT_CONST (1UL << 32)
1307#endif
1308
1309#define WMULT_SHIFT 32
1310
Ingo Molnar194081e2007-08-09 11:16:51 +02001311/*
1312 * Shift right and round:
1313 */
Ingo Molnarcf2ab462007-09-05 14:32:49 +02001314#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
Ingo Molnar194081e2007-08-09 11:16:51 +02001315
Peter Zijlstraa7be37a2008-06-27 13:41:11 +02001316/*
1317 * delta *= weight / lw
1318 */
Ingo Molnarcb1c4fc2007-08-02 17:41:40 +02001319static unsigned long
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001320calc_delta_mine(unsigned long delta_exec, unsigned long weight,
1321 struct load_weight *lw)
1322{
1323 u64 tmp;
1324
Lai Jiangshan7a232e02008-06-12 16:43:07 +08001325 if (!lw->inv_weight) {
1326 if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST))
1327 lw->inv_weight = 1;
1328 else
1329 lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2)
1330 / (lw->weight+1);
1331 }
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001332
1333 tmp = (u64)delta_exec * weight;
1334 /*
1335 * Check whether we'd overflow the 64-bit multiplication:
1336 */
Ingo Molnar194081e2007-08-09 11:16:51 +02001337 if (unlikely(tmp > WMULT_CONST))
Ingo Molnarcf2ab462007-09-05 14:32:49 +02001338 tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
Ingo Molnar194081e2007-08-09 11:16:51 +02001339 WMULT_SHIFT/2);
1340 else
Ingo Molnarcf2ab462007-09-05 14:32:49 +02001341 tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001342
Ingo Molnarecf691d2007-08-02 17:41:40 +02001343 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001344}
1345
Ingo Molnar10919852007-10-15 17:00:04 +02001346static inline void update_load_add(struct load_weight *lw, unsigned long inc)
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001347{
1348 lw->weight += inc;
Ingo Molnare89996a2008-03-14 23:48:28 +01001349 lw->inv_weight = 0;
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001350}
1351
Ingo Molnar10919852007-10-15 17:00:04 +02001352static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001353{
1354 lw->weight -= dec;
Ingo Molnare89996a2008-03-14 23:48:28 +01001355 lw->inv_weight = 0;
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001356}
1357
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358/*
Peter Williams2dd73a42006-06-27 02:54:34 -07001359 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1360 * of tasks with abnormal "nice" values across CPUs the contribution that
1361 * each task makes to its run queue's load is weighted according to its
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01001362 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
Peter Williams2dd73a42006-06-27 02:54:34 -07001363 * scaled version of the new time slice allocation that they receive on time
1364 * slice expiry etc.
1365 */
1366
Peter Zijlstracce7ade2009-01-15 14:53:37 +01001367#define WEIGHT_IDLEPRIO 3
1368#define WMULT_IDLEPRIO 1431655765
Ingo Molnardd41f592007-07-09 18:51:59 +02001369
1370/*
1371 * Nice levels are multiplicative, with a gentle 10% change for every
1372 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1373 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1374 * that remained on nice 0.
1375 *
1376 * The "10% effect" is relative and cumulative: from _any_ nice level,
1377 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
Ingo Molnarf9153ee2007-07-16 09:46:30 +02001378 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1379 * If a task goes up by ~10% and another task goes down by ~10% then
1380 * the relative distance between them is ~25%.)
Ingo Molnardd41f592007-07-09 18:51:59 +02001381 */
1382static const int prio_to_weight[40] = {
Ingo Molnar254753d2007-08-09 11:16:51 +02001383 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1384 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1385 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1386 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1387 /* 0 */ 1024, 820, 655, 526, 423,
1388 /* 5 */ 335, 272, 215, 172, 137,
1389 /* 10 */ 110, 87, 70, 56, 45,
1390 /* 15 */ 36, 29, 23, 18, 15,
Ingo Molnardd41f592007-07-09 18:51:59 +02001391};
1392
Ingo Molnar5714d2d2007-07-16 09:46:31 +02001393/*
1394 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1395 *
1396 * In cases where the weight does not change often, we can use the
1397 * precalculated inverse to speed up arithmetics by turning divisions
1398 * into multiplications:
1399 */
Ingo Molnardd41f592007-07-09 18:51:59 +02001400static const u32 prio_to_wmult[40] = {
Ingo Molnar254753d2007-08-09 11:16:51 +02001401 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1402 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1403 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1404 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1405 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1406 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1407 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1408 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
Ingo Molnardd41f592007-07-09 18:51:59 +02001409};
Peter Williams2dd73a42006-06-27 02:54:34 -07001410
Bharata B Raoef12fef2009-03-31 10:02:22 +05301411/* Time spent by the tasks of the cpu accounting group executing in ... */
1412enum cpuacct_stat_index {
1413 CPUACCT_STAT_USER, /* ... user mode */
1414 CPUACCT_STAT_SYSTEM, /* ... kernel mode */
1415
1416 CPUACCT_STAT_NSTATS,
1417};
1418
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01001419#ifdef CONFIG_CGROUP_CPUACCT
1420static void cpuacct_charge(struct task_struct *tsk, u64 cputime);
Bharata B Raoef12fef2009-03-31 10:02:22 +05301421static void cpuacct_update_stats(struct task_struct *tsk,
1422 enum cpuacct_stat_index idx, cputime_t val);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01001423#else
1424static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
Bharata B Raoef12fef2009-03-31 10:02:22 +05301425static inline void cpuacct_update_stats(struct task_struct *tsk,
1426 enum cpuacct_stat_index idx, cputime_t val) {}
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01001427#endif
1428
Peter Zijlstra18d95a22008-04-19 19:45:00 +02001429static inline void inc_cpu_load(struct rq *rq, unsigned long load)
1430{
1431 update_load_add(&rq->load, load);
1432}
1433
1434static inline void dec_cpu_load(struct rq *rq, unsigned long load)
1435{
1436 update_load_sub(&rq->load, load);
1437}
1438
Ingo Molnar7940ca32008-08-19 13:40:47 +02001439#if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED)
Peter Zijlstraeb755802008-08-19 12:33:05 +02001440typedef int (*tg_visitor)(struct task_group *, void *);
1441
1442/*
1443 * Iterate the full tree, calling @down when first entering a node and @up when
1444 * leaving it for the final time.
1445 */
1446static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
1447{
1448 struct task_group *parent, *child;
1449 int ret;
1450
1451 rcu_read_lock();
1452 parent = &root_task_group;
1453down:
1454 ret = (*down)(parent, data);
1455 if (ret)
1456 goto out_unlock;
1457 list_for_each_entry_rcu(child, &parent->children, siblings) {
1458 parent = child;
1459 goto down;
1460
1461up:
1462 continue;
1463 }
1464 ret = (*up)(parent, data);
1465 if (ret)
1466 goto out_unlock;
1467
1468 child = parent;
1469 parent = parent->parent;
1470 if (parent)
1471 goto up;
1472out_unlock:
1473 rcu_read_unlock();
1474
1475 return ret;
1476}
1477
1478static int tg_nop(struct task_group *tg, void *data)
1479{
1480 return 0;
1481}
1482#endif
1483
Gregory Haskinse7693a32008-01-25 21:08:09 +01001484#ifdef CONFIG_SMP
Peter Zijlstraf5f08f32009-09-10 13:35:28 +02001485/* Used instead of source_load when we know the type == 0 */
1486static unsigned long weighted_cpuload(const int cpu)
1487{
1488 return cpu_rq(cpu)->load.weight;
1489}
1490
1491/*
1492 * Return a low guess at the load of a migration-source cpu weighted
1493 * according to the scheduling class and "nice" value.
1494 *
1495 * We want to under-estimate the load of migration sources, to
1496 * balance conservatively.
1497 */
1498static unsigned long source_load(int cpu, int type)
1499{
1500 struct rq *rq = cpu_rq(cpu);
1501 unsigned long total = weighted_cpuload(cpu);
1502
1503 if (type == 0 || !sched_feat(LB_BIAS))
1504 return total;
1505
1506 return min(rq->cpu_load[type-1], total);
1507}
1508
1509/*
1510 * Return a high guess at the load of a migration-target cpu weighted
1511 * according to the scheduling class and "nice" value.
1512 */
1513static unsigned long target_load(int cpu, int type)
1514{
1515 struct rq *rq = cpu_rq(cpu);
1516 unsigned long total = weighted_cpuload(cpu);
1517
1518 if (type == 0 || !sched_feat(LB_BIAS))
1519 return total;
1520
1521 return max(rq->cpu_load[type-1], total);
1522}
1523
Peter Zijlstraae154be2009-09-10 14:40:57 +02001524static unsigned long power_of(int cpu)
1525{
Peter Zijlstrae51fd5e2010-05-31 12:37:30 +02001526 return cpu_rq(cpu)->cpu_power;
Peter Zijlstraae154be2009-09-10 14:40:57 +02001527}
1528
Gregory Haskinse7693a32008-01-25 21:08:09 +01001529static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001530
Peter Zijlstraa8a51d52008-06-27 13:41:26 +02001531static unsigned long cpu_avg_load_per_task(int cpu)
1532{
1533 struct rq *rq = cpu_rq(cpu);
Ingo Molnaraf6d5962008-11-29 20:45:15 +01001534 unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
Peter Zijlstraa8a51d52008-06-27 13:41:26 +02001535
Steven Rostedt4cd42622008-11-26 21:04:24 -05001536 if (nr_running)
1537 rq->avg_load_per_task = rq->load.weight / nr_running;
Balbir Singha2d47772008-11-12 16:19:00 +05301538 else
1539 rq->avg_load_per_task = 0;
Peter Zijlstraa8a51d52008-06-27 13:41:26 +02001540
1541 return rq->avg_load_per_task;
1542}
1543
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001544#ifdef CONFIG_FAIR_GROUP_SCHED
1545
Tejun Heo43cf38e2010-02-02 14:38:57 +09001546static __read_mostly unsigned long __percpu *update_shares_data;
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001547
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001548static void __set_se_shares(struct sched_entity *se, unsigned long shares);
1549
1550/*
1551 * Calculate and set the cpu's group shares.
1552 */
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001553static void update_group_shares_cpu(struct task_group *tg, int cpu,
1554 unsigned long sd_shares,
1555 unsigned long sd_rq_weight,
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09001556 unsigned long *usd_rq_weight)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001557{
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001558 unsigned long shares, rq_weight;
Peter Zijlstraa50042782009-07-27 14:04:49 +02001559 int boost = 0;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001560
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09001561 rq_weight = usd_rq_weight[cpu];
Peter Zijlstraa50042782009-07-27 14:04:49 +02001562 if (!rq_weight) {
1563 boost = 1;
1564 rq_weight = NICE_0_LOAD;
1565 }
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001566
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001567 /*
Peter Zijlstraa8af7242009-08-21 13:58:54 +02001568 * \Sum_j shares_j * rq_weight_i
1569 * shares_i = -----------------------------
1570 * \Sum_j rq_weight_j
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001571 */
Ken Chenec4e0e22008-11-18 22:41:57 -08001572 shares = (sd_shares * rq_weight) / sd_rq_weight;
Peter Zijlstraffda12a2008-10-17 19:27:02 +02001573 shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001574
Peter Zijlstraffda12a2008-10-17 19:27:02 +02001575 if (abs(shares - tg->se[cpu]->load.weight) >
1576 sysctl_sched_shares_thresh) {
1577 struct rq *rq = cpu_rq(cpu);
1578 unsigned long flags;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001579
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001580 raw_spin_lock_irqsave(&rq->lock, flags);
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001581 tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight;
Peter Zijlstraa50042782009-07-27 14:04:49 +02001582 tg->cfs_rq[cpu]->shares = boost ? 0 : shares;
Peter Zijlstraffda12a2008-10-17 19:27:02 +02001583 __set_se_shares(tg->se[cpu], shares);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001584 raw_spin_unlock_irqrestore(&rq->lock, flags);
Peter Zijlstraffda12a2008-10-17 19:27:02 +02001585 }
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001586}
1587
1588/*
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001589 * Re-compute the task group their per cpu shares over the given domain.
1590 * This needs to be done in a bottom-up fashion because the rq weight of a
1591 * parent group depends on the shares of its child groups.
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001592 */
Peter Zijlstraeb755802008-08-19 12:33:05 +02001593static int tg_shares_up(struct task_group *tg, void *data)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001594{
Peter Zijlstracd8ad402009-12-03 18:00:07 +01001595 unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0;
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09001596 unsigned long *usd_rq_weight;
Peter Zijlstraeb755802008-08-19 12:33:05 +02001597 struct sched_domain *sd = data;
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001598 unsigned long flags;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001599 int i;
1600
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001601 if (!tg->se[0])
1602 return 0;
1603
1604 local_irq_save(flags);
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09001605 usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id());
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001606
Rusty Russell758b2cd2008-11-25 02:35:04 +10301607 for_each_cpu(i, sched_domain_span(sd)) {
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001608 weight = tg->cfs_rq[i]->load.weight;
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09001609 usd_rq_weight[i] = weight;
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001610
Peter Zijlstracd8ad402009-12-03 18:00:07 +01001611 rq_weight += weight;
Ken Chenec4e0e22008-11-18 22:41:57 -08001612 /*
1613 * If there are currently no tasks on the cpu pretend there
1614 * is one of average load so that when a new task gets to
1615 * run here it will not get delayed by group starvation.
1616 */
Ken Chenec4e0e22008-11-18 22:41:57 -08001617 if (!weight)
1618 weight = NICE_0_LOAD;
1619
Peter Zijlstracd8ad402009-12-03 18:00:07 +01001620 sum_weight += weight;
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001621 shares += tg->cfs_rq[i]->shares;
1622 }
1623
Peter Zijlstracd8ad402009-12-03 18:00:07 +01001624 if (!rq_weight)
1625 rq_weight = sum_weight;
1626
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001627 if ((!shares && rq_weight) || shares > tg->shares)
1628 shares = tg->shares;
1629
1630 if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE))
1631 shares = tg->shares;
1632
Rusty Russell758b2cd2008-11-25 02:35:04 +10301633 for_each_cpu(i, sched_domain_span(sd))
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09001634 update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight);
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001635
1636 local_irq_restore(flags);
Peter Zijlstraeb755802008-08-19 12:33:05 +02001637
1638 return 0;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001639}
1640
1641/*
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001642 * Compute the cpu's hierarchical load factor for each task group.
1643 * This needs to be done in a top-down fashion because the load of a child
1644 * group is a fraction of its parents load.
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001645 */
Peter Zijlstraeb755802008-08-19 12:33:05 +02001646static int tg_load_down(struct task_group *tg, void *data)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001647{
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001648 unsigned long load;
Peter Zijlstraeb755802008-08-19 12:33:05 +02001649 long cpu = (long)data;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001650
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001651 if (!tg->parent) {
1652 load = cpu_rq(cpu)->load.weight;
1653 } else {
1654 load = tg->parent->cfs_rq[cpu]->h_load;
1655 load *= tg->cfs_rq[cpu]->shares;
1656 load /= tg->parent->cfs_rq[cpu]->load.weight + 1;
1657 }
1658
1659 tg->cfs_rq[cpu]->h_load = load;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001660
Peter Zijlstraeb755802008-08-19 12:33:05 +02001661 return 0;
Peter Zijlstra4d8d5952008-06-27 13:41:19 +02001662}
1663
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001664static void update_shares(struct sched_domain *sd)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001665{
Peter Zijlstrae7097152009-06-03 15:41:20 +02001666 s64 elapsed;
1667 u64 now;
1668
1669 if (root_task_group_empty())
1670 return;
1671
Peter Zijlstrac6763292010-05-25 10:48:51 +02001672 now = local_clock();
Peter Zijlstrae7097152009-06-03 15:41:20 +02001673 elapsed = now - sd->last_update;
Peter Zijlstra2398f2c2008-06-27 13:41:35 +02001674
1675 if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) {
1676 sd->last_update = now;
Peter Zijlstraeb755802008-08-19 12:33:05 +02001677 walk_tg_tree(tg_nop, tg_shares_up, sd);
Peter Zijlstra2398f2c2008-06-27 13:41:35 +02001678 }
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001679}
1680
Peter Zijlstraeb755802008-08-19 12:33:05 +02001681static void update_h_load(long cpu)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001682{
Peter Zijlstraeb755802008-08-19 12:33:05 +02001683 walk_tg_tree(tg_load_down, tg_nop, (void *)cpu);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001684}
1685
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001686#else
1687
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001688static inline void update_shares(struct sched_domain *sd)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001689{
1690}
1691
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001692#endif
1693
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001694#ifdef CONFIG_PREEMPT
1695
Peter Zijlstrab78bb862009-09-15 14:23:18 +02001696static void double_rq_lock(struct rq *rq1, struct rq *rq2);
1697
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001698/*
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001699 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1700 * way at the expense of forcing extra atomic operations in all
1701 * invocations. This assures that the double_lock is acquired using the
1702 * same underlying policy as the spinlock_t on this architecture, which
1703 * reduces latency compared to the unfair variant below. However, it
1704 * also adds more overhead and therefore may reduce throughput.
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001705 */
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001706static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1707 __releases(this_rq->lock)
1708 __acquires(busiest->lock)
1709 __acquires(this_rq->lock)
1710{
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001711 raw_spin_unlock(&this_rq->lock);
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001712 double_rq_lock(this_rq, busiest);
1713
1714 return 1;
1715}
1716
1717#else
1718/*
1719 * Unfair double_lock_balance: Optimizes throughput at the expense of
1720 * latency by eliminating extra atomic operations when the locks are
1721 * already in proper order on entry. This favors lower cpu-ids and will
1722 * grant the double lock to lower cpus over higher ids under contention,
1723 * regardless of entry order into the function.
1724 */
1725static int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001726 __releases(this_rq->lock)
1727 __acquires(busiest->lock)
1728 __acquires(this_rq->lock)
1729{
1730 int ret = 0;
1731
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001732 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001733 if (busiest < this_rq) {
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001734 raw_spin_unlock(&this_rq->lock);
1735 raw_spin_lock(&busiest->lock);
1736 raw_spin_lock_nested(&this_rq->lock,
1737 SINGLE_DEPTH_NESTING);
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001738 ret = 1;
1739 } else
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001740 raw_spin_lock_nested(&busiest->lock,
1741 SINGLE_DEPTH_NESTING);
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001742 }
1743 return ret;
1744}
1745
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001746#endif /* CONFIG_PREEMPT */
1747
1748/*
1749 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1750 */
1751static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1752{
1753 if (unlikely(!irqs_disabled())) {
1754 /* printk() doesn't work good under rq->lock */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001755 raw_spin_unlock(&this_rq->lock);
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001756 BUG_ON(1);
1757 }
1758
1759 return _double_lock_balance(this_rq, busiest);
1760}
1761
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001762static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1763 __releases(busiest->lock)
1764{
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001765 raw_spin_unlock(&busiest->lock);
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001766 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1767}
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001768
1769/*
1770 * double_rq_lock - safely lock two runqueues
1771 *
1772 * Note this does not disable interrupts like task_rq_lock,
1773 * you need to do so manually before calling.
1774 */
1775static void double_rq_lock(struct rq *rq1, struct rq *rq2)
1776 __acquires(rq1->lock)
1777 __acquires(rq2->lock)
1778{
1779 BUG_ON(!irqs_disabled());
1780 if (rq1 == rq2) {
1781 raw_spin_lock(&rq1->lock);
1782 __acquire(rq2->lock); /* Fake it out ;) */
1783 } else {
1784 if (rq1 < rq2) {
1785 raw_spin_lock(&rq1->lock);
1786 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1787 } else {
1788 raw_spin_lock(&rq2->lock);
1789 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1790 }
1791 }
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001792}
1793
1794/*
1795 * double_rq_unlock - safely unlock two runqueues
1796 *
1797 * Note this does not restore interrupts like task_rq_unlock,
1798 * you need to do so manually after calling.
1799 */
1800static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1801 __releases(rq1->lock)
1802 __releases(rq2->lock)
1803{
1804 raw_spin_unlock(&rq1->lock);
1805 if (rq1 != rq2)
1806 raw_spin_unlock(&rq2->lock);
1807 else
1808 __release(rq2->lock);
1809}
1810
Peter Zijlstra18d95a22008-04-19 19:45:00 +02001811#endif
Peter Zijlstra18d95a22008-04-19 19:45:00 +02001812
1813#ifdef CONFIG_FAIR_GROUP_SCHED
1814static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares)
1815{
Vegard Nossum30432092008-06-27 21:35:50 +02001816#ifdef CONFIG_SMP
Ingo Molnar34e83e82008-06-27 15:42:36 +02001817 cfs_rq->shares = shares;
1818#endif
Peter Zijlstra18d95a22008-04-19 19:45:00 +02001819}
1820#endif
1821
Peter Zijlstra74f51872010-04-22 21:50:19 +02001822static void calc_load_account_idle(struct rq *this_rq);
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01001823static void update_sysctl(void);
Christian Ehrhardtacb4a842009-11-30 12:16:48 +01001824static int get_update_sysctl_factor(void);
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07001825static void update_cpu_load(struct rq *this_rq);
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02001826
Peter Zijlstracd29fe62009-11-27 17:32:46 +01001827static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1828{
1829 set_task_rq(p, cpu);
1830#ifdef CONFIG_SMP
1831 /*
1832 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1833 * successfuly executed on another CPU. We must ensure that updates of
1834 * per-task data have been completed by this moment.
1835 */
1836 smp_wmb();
1837 task_thread_info(p)->cpu = cpu;
1838#endif
1839}
Gregory Haskinse7693a32008-01-25 21:08:09 +01001840
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001841static const struct sched_class rt_sched_class;
Ingo Molnardd41f592007-07-09 18:51:59 +02001842
1843#define sched_class_highest (&rt_sched_class)
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -04001844#define for_each_class(class) \
1845 for (class = sched_class_highest; class; class = class->next)
Ingo Molnardd41f592007-07-09 18:51:59 +02001846
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001847#include "sched_stats.h"
1848
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001849static void inc_nr_running(struct rq *rq)
Ingo Molnar6363ca52008-05-29 11:28:57 +02001850{
1851 rq->nr_running++;
Ingo Molnar6363ca52008-05-29 11:28:57 +02001852}
1853
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001854static void dec_nr_running(struct rq *rq)
Ingo Molnar9c217242007-08-02 17:41:40 +02001855{
1856 rq->nr_running--;
Ingo Molnar9c217242007-08-02 17:41:40 +02001857}
1858
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001859static void set_load_weight(struct task_struct *p)
1860{
1861 if (task_has_rt_policy(p)) {
Peter Zijlstrae51fd5e2010-05-31 12:37:30 +02001862 p->se.load.weight = 0;
1863 p->se.load.inv_weight = WMULT_CONST;
Ingo Molnardd41f592007-07-09 18:51:59 +02001864 return;
1865 }
1866
1867 /*
1868 * SCHED_IDLE tasks get minimal weight:
1869 */
1870 if (p->policy == SCHED_IDLE) {
1871 p->se.load.weight = WEIGHT_IDLEPRIO;
1872 p->se.load.inv_weight = WMULT_IDLEPRIO;
1873 return;
1874 }
1875
1876 p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO];
1877 p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO];
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001878}
1879
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001880static void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
Gregory Haskins2087a1a2008-06-27 14:30:00 -06001881{
Mike Galbraitha64692a2010-03-11 17:16:20 +01001882 update_rq_clock(rq);
Ingo Molnar71f8bd42007-07-09 18:51:59 +02001883 sched_info_queued(p);
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001884 p->sched_class->enqueue_task(rq, p, flags);
Ingo Molnardd41f592007-07-09 18:51:59 +02001885 p->se.on_rq = 1;
1886}
1887
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001888static void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
Ingo Molnardd41f592007-07-09 18:51:59 +02001889{
Mike Galbraitha64692a2010-03-11 17:16:20 +01001890 update_rq_clock(rq);
Ankita Garg46ac22b2008-07-01 14:30:06 +05301891 sched_info_dequeued(p);
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001892 p->sched_class->dequeue_task(rq, p, flags);
Ingo Molnardd41f592007-07-09 18:51:59 +02001893 p->se.on_rq = 0;
Ingo Molnar71f8bd42007-07-09 18:51:59 +02001894}
1895
1896/*
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001897 * activate_task - move a task to the runqueue.
1898 */
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001899static void activate_task(struct rq *rq, struct task_struct *p, int flags)
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001900{
1901 if (task_contributes_to_load(p))
1902 rq->nr_uninterruptible--;
1903
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001904 enqueue_task(rq, p, flags);
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001905 inc_nr_running(rq);
1906}
1907
1908/*
1909 * deactivate_task - remove a task from the runqueue.
1910 */
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001911static void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001912{
1913 if (task_contributes_to_load(p))
1914 rq->nr_uninterruptible++;
1915
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001916 dequeue_task(rq, p, flags);
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001917 dec_nr_running(rq);
1918}
1919
1920#include "sched_idletask.c"
1921#include "sched_fair.c"
1922#include "sched_rt.c"
1923#ifdef CONFIG_SCHED_DEBUG
1924# include "sched_debug.c"
1925#endif
1926
1927/*
Ingo Molnardd41f592007-07-09 18:51:59 +02001928 * __normal_prio - return the priority that is based on the static prio
Ingo Molnar71f8bd42007-07-09 18:51:59 +02001929 */
Ingo Molnar14531182007-07-09 18:51:59 +02001930static inline int __normal_prio(struct task_struct *p)
1931{
Ingo Molnardd41f592007-07-09 18:51:59 +02001932 return p->static_prio;
Ingo Molnar14531182007-07-09 18:51:59 +02001933}
1934
1935/*
Ingo Molnarb29739f2006-06-27 02:54:51 -07001936 * Calculate the expected normal priority: i.e. priority
1937 * without taking RT-inheritance into account. Might be
1938 * boosted by interactivity modifiers. Changes upon fork,
1939 * setprio syscalls, and whenever the interactivity
1940 * estimator recalculates.
1941 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001942static inline int normal_prio(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -07001943{
1944 int prio;
1945
Ingo Molnare05606d2007-07-09 18:51:59 +02001946 if (task_has_rt_policy(p))
Ingo Molnarb29739f2006-06-27 02:54:51 -07001947 prio = MAX_RT_PRIO-1 - p->rt_priority;
1948 else
1949 prio = __normal_prio(p);
1950 return prio;
1951}
1952
1953/*
1954 * Calculate the current priority, i.e. the priority
1955 * taken into account by the scheduler. This value might
1956 * be boosted by RT tasks, or might be boosted by
1957 * interactivity modifiers. Will be RT if the task got
1958 * RT-boosted. If not then it returns p->normal_prio.
1959 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001960static int effective_prio(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -07001961{
1962 p->normal_prio = normal_prio(p);
1963 /*
1964 * If we are RT tasks or we were boosted to RT priority,
1965 * keep the priority unchanged. Otherwise, update priority
1966 * to the normal priority:
1967 */
1968 if (!rt_prio(p->prio))
1969 return p->normal_prio;
1970 return p->prio;
1971}
1972
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973/**
1974 * task_curr - is this task currently executing on a CPU?
1975 * @p: the task in question.
1976 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001977inline int task_curr(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978{
1979 return cpu_curr(task_cpu(p)) == p;
1980}
1981
Steven Rostedtcb469842008-01-25 21:08:22 +01001982static inline void check_class_changed(struct rq *rq, struct task_struct *p,
1983 const struct sched_class *prev_class,
1984 int oldprio, int running)
1985{
1986 if (prev_class != p->sched_class) {
1987 if (prev_class->switched_from)
1988 prev_class->switched_from(rq, p, running);
1989 p->sched_class->switched_to(rq, p, running);
1990 } else
1991 p->sched_class->prio_changed(rq, p, oldprio, running);
1992}
1993
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994#ifdef CONFIG_SMP
Ingo Molnarcc367732007-10-15 17:00:18 +02001995/*
1996 * Is this task likely cache-hot:
1997 */
Gregory Haskinse7693a32008-01-25 21:08:09 +01001998static int
Ingo Molnarcc367732007-10-15 17:00:18 +02001999task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
2000{
2001 s64 delta;
2002
Peter Zijlstrae6c8fba2009-12-16 18:04:33 +01002003 if (p->sched_class != &fair_sched_class)
2004 return 0;
2005
Ingo Molnarf540a602008-03-15 17:10:34 +01002006 /*
2007 * Buddy candidates are cache hot:
2008 */
Mike Galbraithf685cea2009-10-23 23:09:22 +02002009 if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running &&
Peter Zijlstra47932412008-11-04 21:25:09 +01002010 (&p->se == cfs_rq_of(&p->se)->next ||
2011 &p->se == cfs_rq_of(&p->se)->last))
Ingo Molnarf540a602008-03-15 17:10:34 +01002012 return 1;
2013
Ingo Molnar6bc16652007-10-15 17:00:18 +02002014 if (sysctl_sched_migration_cost == -1)
2015 return 1;
2016 if (sysctl_sched_migration_cost == 0)
2017 return 0;
2018
Ingo Molnarcc367732007-10-15 17:00:18 +02002019 delta = now - p->se.exec_start;
2020
2021 return delta < (s64)sysctl_sched_migration_cost;
2022}
2023
Ingo Molnardd41f592007-07-09 18:51:59 +02002024void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
Ingo Molnarc65cc872007-07-09 18:51:58 +02002025{
Peter Zijlstrae2912002009-12-16 18:04:36 +01002026#ifdef CONFIG_SCHED_DEBUG
2027 /*
2028 * We should never call set_task_cpu() on a blocked task,
2029 * ttwu() will sort out the placement.
2030 */
Peter Zijlstra077614e2009-12-17 13:16:31 +01002031 WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
2032 !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE));
Peter Zijlstrae2912002009-12-16 18:04:36 +01002033#endif
2034
Mathieu Desnoyersde1d7282009-05-05 16:49:59 +08002035 trace_sched_migrate_task(p, new_cpu);
Peter Zijlstracbc34ed2008-12-10 08:08:22 +01002036
Peter Zijlstra0c697742009-12-22 15:43:19 +01002037 if (task_cpu(p) != new_cpu) {
2038 p->se.nr_migrations++;
2039 perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0);
2040 }
Ingo Molnardd41f592007-07-09 18:51:59 +02002041
2042 __set_task_cpu(p, new_cpu);
Ingo Molnarc65cc872007-07-09 18:51:58 +02002043}
2044
Tejun Heo969c7922010-05-06 18:49:21 +02002045struct migration_arg {
Ingo Molnar36c8b582006-07-03 00:25:41 -07002046 struct task_struct *task;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 int dest_cpu;
Ingo Molnar70b97a72006-07-03 00:25:42 -07002048};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
Tejun Heo969c7922010-05-06 18:49:21 +02002050static int migration_cpu_stop(void *data);
2051
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052/*
2053 * The task's runqueue lock must be held.
2054 * Returns true if you have to wait for migration thread.
2055 */
Tejun Heo969c7922010-05-06 18:49:21 +02002056static bool migrate_task(struct task_struct *p, int dest_cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057{
Ingo Molnar70b97a72006-07-03 00:25:42 -07002058 struct rq *rq = task_rq(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
2060 /*
2061 * If the task is not on a runqueue (and not running), then
Peter Zijlstrae2912002009-12-16 18:04:36 +01002062 * the next wake-up will properly place the task.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 */
Tejun Heo969c7922010-05-06 18:49:21 +02002064 return p->se.on_rq || task_running(rq, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065}
2066
2067/*
2068 * wait_task_inactive - wait for a thread to unschedule.
2069 *
Roland McGrath85ba2d82008-07-25 19:45:58 -07002070 * If @match_state is nonzero, it's the @p->state value just checked and
2071 * not expected to change. If it changes, i.e. @p might have woken up,
2072 * then return zero. When we succeed in waiting for @p to be off its CPU,
2073 * we return a positive number (its total switch count). If a second call
2074 * a short while later returns the same number, the caller can be sure that
2075 * @p has remained unscheduled the whole time.
2076 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 * The caller must ensure that the task *will* unschedule sometime soon,
2078 * else this function might spin for a *long* time. This function can't
2079 * be called with interrupts off, or it may introduce deadlock with
2080 * smp_call_function() if an IPI is sent by the same process we are
2081 * waiting to become inactive.
2082 */
Roland McGrath85ba2d82008-07-25 19:45:58 -07002083unsigned long wait_task_inactive(struct task_struct *p, long match_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084{
2085 unsigned long flags;
Ingo Molnardd41f592007-07-09 18:51:59 +02002086 int running, on_rq;
Roland McGrath85ba2d82008-07-25 19:45:58 -07002087 unsigned long ncsw;
Ingo Molnar70b97a72006-07-03 00:25:42 -07002088 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089
Andi Kleen3a5c3592007-10-15 17:00:14 +02002090 for (;;) {
2091 /*
2092 * We do the initial early heuristics without holding
2093 * any task-queue locks at all. We'll only try to get
2094 * the runqueue lock when things look like they will
2095 * work out!
2096 */
2097 rq = task_rq(p);
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07002098
Andi Kleen3a5c3592007-10-15 17:00:14 +02002099 /*
2100 * If the task is actively running on another CPU
2101 * still, just relax and busy-wait without holding
2102 * any locks.
2103 *
2104 * NOTE! Since we don't hold any locks, it's not
2105 * even sure that "rq" stays as the right runqueue!
2106 * But we don't care, since "task_running()" will
2107 * return false if the runqueue has changed and p
2108 * is actually now running somewhere else!
2109 */
Roland McGrath85ba2d82008-07-25 19:45:58 -07002110 while (task_running(rq, p)) {
2111 if (match_state && unlikely(p->state != match_state))
2112 return 0;
Andi Kleen3a5c3592007-10-15 17:00:14 +02002113 cpu_relax();
Roland McGrath85ba2d82008-07-25 19:45:58 -07002114 }
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07002115
Andi Kleen3a5c3592007-10-15 17:00:14 +02002116 /*
2117 * Ok, time to look more closely! We need the rq
2118 * lock now, to be *sure*. If we're wrong, we'll
2119 * just go back and repeat.
2120 */
2121 rq = task_rq_lock(p, &flags);
Peter Zijlstra27a9da62010-05-04 20:36:56 +02002122 trace_sched_wait_task(p);
Andi Kleen3a5c3592007-10-15 17:00:14 +02002123 running = task_running(rq, p);
2124 on_rq = p->se.on_rq;
Roland McGrath85ba2d82008-07-25 19:45:58 -07002125 ncsw = 0;
Oleg Nesterovf31e11d2008-08-20 16:54:44 -07002126 if (!match_state || p->state == match_state)
Oleg Nesterov93dcf552008-08-20 16:54:44 -07002127 ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
Andi Kleen3a5c3592007-10-15 17:00:14 +02002128 task_rq_unlock(rq, &flags);
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07002129
Andi Kleen3a5c3592007-10-15 17:00:14 +02002130 /*
Roland McGrath85ba2d82008-07-25 19:45:58 -07002131 * If it changed from the expected state, bail out now.
2132 */
2133 if (unlikely(!ncsw))
2134 break;
2135
2136 /*
Andi Kleen3a5c3592007-10-15 17:00:14 +02002137 * Was it really running after all now that we
2138 * checked with the proper locks actually held?
2139 *
2140 * Oops. Go back and try again..
2141 */
2142 if (unlikely(running)) {
2143 cpu_relax();
2144 continue;
2145 }
2146
2147 /*
2148 * It's not enough that it's not actively running,
2149 * it must be off the runqueue _entirely_, and not
2150 * preempted!
2151 *
Luis Henriques80dd99b2009-03-16 19:58:09 +00002152 * So if it was still runnable (but just not actively
Andi Kleen3a5c3592007-10-15 17:00:14 +02002153 * running right now), it's preempted, and we should
2154 * yield - it could be a while.
2155 */
2156 if (unlikely(on_rq)) {
2157 schedule_timeout_uninterruptible(1);
2158 continue;
2159 }
2160
2161 /*
2162 * Ahh, all good. It wasn't running, and it wasn't
2163 * runnable, which means that it will never become
2164 * running in the future either. We're all done!
2165 */
2166 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 }
Roland McGrath85ba2d82008-07-25 19:45:58 -07002168
2169 return ncsw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170}
2171
2172/***
2173 * kick_process - kick a running thread to enter/exit the kernel
2174 * @p: the to-be-kicked thread
2175 *
2176 * Cause a process which is running on another CPU to enter
2177 * kernel-mode, without any delay. (to get signals handled.)
2178 *
2179 * NOTE: this function doesnt have to take the runqueue lock,
2180 * because all it wants to ensure is that the remote task enters
2181 * the kernel. If the IPI races and the task has been migrated
2182 * to another CPU then no harm is done and the purpose has been
2183 * achieved as well.
2184 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07002185void kick_process(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186{
2187 int cpu;
2188
2189 preempt_disable();
2190 cpu = task_cpu(p);
2191 if ((cpu != smp_processor_id()) && task_curr(p))
2192 smp_send_reschedule(cpu);
2193 preempt_enable();
2194}
Rusty Russellb43e3522009-06-12 22:27:00 -06002195EXPORT_SYMBOL_GPL(kick_process);
Nick Piggin476d1392005-06-25 14:57:29 -07002196#endif /* CONFIG_SMP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197
Thomas Gleixner0793a612008-12-04 20:12:29 +01002198/**
2199 * task_oncpu_function_call - call a function on the cpu on which a task runs
2200 * @p: the task to evaluate
2201 * @func: the function to be called
2202 * @info: the function call argument
2203 *
2204 * Calls the function @func when the task is currently running. This might
2205 * be on the current CPU, which just calls the function directly
2206 */
2207void task_oncpu_function_call(struct task_struct *p,
2208 void (*func) (void *info), void *info)
2209{
2210 int cpu;
2211
2212 preempt_disable();
2213 cpu = task_cpu(p);
2214 if (task_curr(p))
2215 smp_call_function_single(cpu, func, info, 1);
2216 preempt_enable();
2217}
2218
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002219#ifdef CONFIG_SMP
Oleg Nesterov30da6882010-03-15 10:10:19 +01002220/*
2221 * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held.
2222 */
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01002223static int select_fallback_rq(int cpu, struct task_struct *p)
2224{
2225 int dest_cpu;
2226 const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu));
2227
2228 /* Look for allowed, online CPU in same node. */
2229 for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask)
2230 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
2231 return dest_cpu;
2232
2233 /* Any allowed, online CPU? */
2234 dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask);
2235 if (dest_cpu < nr_cpu_ids)
2236 return dest_cpu;
2237
2238 /* No more Mr. Nice Guy. */
Oleg Nesterov897f0b32010-03-15 10:10:03 +01002239 if (unlikely(dest_cpu >= nr_cpu_ids)) {
Oleg Nesterov9084bb82010-03-15 10:10:27 +01002240 dest_cpu = cpuset_cpus_allowed_fallback(p);
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01002241 /*
2242 * Don't tell them about moving exiting tasks or
2243 * kernel threads (both mm NULL), since they never
2244 * leave kernel.
2245 */
2246 if (p->mm && printk_ratelimit()) {
2247 printk(KERN_INFO "process %d (%s) no "
2248 "longer affine to cpu%d\n",
2249 task_pid_nr(p), p->comm, cpu);
2250 }
2251 }
2252
2253 return dest_cpu;
2254}
2255
Peter Zijlstrae2912002009-12-16 18:04:36 +01002256/*
Oleg Nesterov30da6882010-03-15 10:10:19 +01002257 * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable.
Peter Zijlstrae2912002009-12-16 18:04:36 +01002258 */
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002259static inline
Peter Zijlstra0017d732010-03-24 18:34:10 +01002260int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags)
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002261{
Peter Zijlstra0017d732010-03-24 18:34:10 +01002262 int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags);
Peter Zijlstrae2912002009-12-16 18:04:36 +01002263
2264 /*
2265 * In order not to call set_task_cpu() on a blocking task we need
2266 * to rely on ttwu() to place the task on a valid ->cpus_allowed
2267 * cpu.
2268 *
2269 * Since this is common to all placement strategies, this lives here.
2270 *
2271 * [ this allows ->select_task() to simply return task_cpu(p) and
2272 * not worry about this generic constraint ]
2273 */
2274 if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) ||
Peter Zijlstra70f11202009-12-20 17:36:27 +01002275 !cpu_online(cpu)))
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01002276 cpu = select_fallback_rq(task_cpu(p), p);
Peter Zijlstrae2912002009-12-16 18:04:36 +01002277
2278 return cpu;
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002279}
Mike Galbraith09a40af2010-04-15 07:29:59 +02002280
2281static void update_avg(u64 *avg, u64 sample)
2282{
2283 s64 diff = sample - *avg;
2284 *avg += diff >> 3;
2285}
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002286#endif
2287
Tejun Heo9ed38112009-12-03 15:08:03 +09002288static inline void ttwu_activate(struct task_struct *p, struct rq *rq,
2289 bool is_sync, bool is_migrate, bool is_local,
2290 unsigned long en_flags)
2291{
2292 schedstat_inc(p, se.statistics.nr_wakeups);
2293 if (is_sync)
2294 schedstat_inc(p, se.statistics.nr_wakeups_sync);
2295 if (is_migrate)
2296 schedstat_inc(p, se.statistics.nr_wakeups_migrate);
2297 if (is_local)
2298 schedstat_inc(p, se.statistics.nr_wakeups_local);
2299 else
2300 schedstat_inc(p, se.statistics.nr_wakeups_remote);
2301
2302 activate_task(rq, p, en_flags);
2303}
2304
2305static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq,
2306 int wake_flags, bool success)
2307{
2308 trace_sched_wakeup(p, success);
2309 check_preempt_curr(rq, p, wake_flags);
2310
2311 p->state = TASK_RUNNING;
2312#ifdef CONFIG_SMP
2313 if (p->sched_class->task_woken)
2314 p->sched_class->task_woken(rq, p);
2315
2316 if (unlikely(rq->idle_stamp)) {
2317 u64 delta = rq->clock - rq->idle_stamp;
2318 u64 max = 2*sysctl_sched_migration_cost;
2319
2320 if (delta > max)
2321 rq->avg_idle = max;
2322 else
2323 update_avg(&rq->avg_idle, delta);
2324 rq->idle_stamp = 0;
2325 }
2326#endif
Tejun Heo21aa9af2010-06-08 21:40:37 +02002327 /* if a worker is waking up, notify workqueue */
2328 if ((p->flags & PF_WQ_WORKER) && success)
2329 wq_worker_waking_up(p, cpu_of(rq));
Tejun Heo9ed38112009-12-03 15:08:03 +09002330}
2331
2332/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 * try_to_wake_up - wake up a thread
Tejun Heo9ed38112009-12-03 15:08:03 +09002334 * @p: the thread to be awakened
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 * @state: the mask of task states that can be woken
Tejun Heo9ed38112009-12-03 15:08:03 +09002336 * @wake_flags: wake modifier flags (WF_*)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 *
2338 * Put it on the run-queue if it's not already there. The "current"
2339 * thread is always on the run-queue (except when the actual
2340 * re-schedule is in progress), and as such you're allowed to do
2341 * the simpler "current->state = TASK_RUNNING" to mark yourself
2342 * runnable without the overhead of this.
2343 *
Tejun Heo9ed38112009-12-03 15:08:03 +09002344 * Returns %true if @p was woken up, %false if it was already running
2345 * or @state didn't match @p's state.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 */
Peter Zijlstra7d478722009-09-14 19:55:44 +02002347static int try_to_wake_up(struct task_struct *p, unsigned int state,
2348 int wake_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349{
Ingo Molnarcc367732007-10-15 17:00:18 +02002350 int cpu, orig_cpu, this_cpu, success = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 unsigned long flags;
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01002352 unsigned long en_flags = ENQUEUE_WAKEUP;
Dan Carpenterab3b3aa2010-03-06 14:17:52 +03002353 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002355 this_cpu = get_cpu();
Peter Zijlstra2398f2c2008-06-27 13:41:35 +02002356
Linus Torvalds04e2f172008-02-23 18:05:03 -08002357 smp_wmb();
Dan Carpenterab3b3aa2010-03-06 14:17:52 +03002358 rq = task_rq_lock(p, &flags);
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002359 if (!(p->state & state))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 goto out;
2361
Ingo Molnardd41f592007-07-09 18:51:59 +02002362 if (p->se.on_rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 goto out_running;
2364
2365 cpu = task_cpu(p);
Ingo Molnarcc367732007-10-15 17:00:18 +02002366 orig_cpu = cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367
2368#ifdef CONFIG_SMP
2369 if (unlikely(task_running(rq, p)))
2370 goto out_activate;
2371
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002372 /*
2373 * In order to handle concurrent wakeups and release the rq->lock
2374 * we put the task in TASK_WAKING state.
Ingo Molnareb240732009-09-16 21:09:13 +02002375 *
2376 * First fix up the nr_uninterruptible count:
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002377 */
Peter Zijlstracc87f762010-03-26 12:22:14 +01002378 if (task_contributes_to_load(p)) {
2379 if (likely(cpu_online(orig_cpu)))
2380 rq->nr_uninterruptible--;
2381 else
2382 this_rq()->nr_uninterruptible--;
2383 }
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002384 p->state = TASK_WAKING;
Peter Zijlstraefbbd052009-12-16 18:04:40 +01002385
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01002386 if (p->sched_class->task_waking) {
Peter Zijlstraefbbd052009-12-16 18:04:40 +01002387 p->sched_class->task_waking(rq, p);
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01002388 en_flags |= ENQUEUE_WAKING;
Peter Zijlstra0970d292010-02-15 14:45:54 +01002389 }
Peter Zijlstraab19cb22009-11-27 15:44:43 +01002390
Peter Zijlstra0017d732010-03-24 18:34:10 +01002391 cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags);
2392 if (cpu != orig_cpu)
Mike Galbraithf5dc3752009-10-09 08:35:03 +02002393 set_task_cpu(p, cpu);
Peter Zijlstra0017d732010-03-24 18:34:10 +01002394 __task_rq_unlock(rq);
Peter Zijlstraab19cb22009-11-27 15:44:43 +01002395
Peter Zijlstra0970d292010-02-15 14:45:54 +01002396 rq = cpu_rq(cpu);
2397 raw_spin_lock(&rq->lock);
Mike Galbraithf5dc3752009-10-09 08:35:03 +02002398
Peter Zijlstra0970d292010-02-15 14:45:54 +01002399 /*
2400 * We migrated the task without holding either rq->lock, however
2401 * since the task is not on the task list itself, nobody else
2402 * will try and migrate the task, hence the rq should match the
2403 * cpu we just moved it to.
2404 */
2405 WARN_ON(task_cpu(p) != cpu);
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002406 WARN_ON(p->state != TASK_WAKING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407
Gregory Haskinse7693a32008-01-25 21:08:09 +01002408#ifdef CONFIG_SCHEDSTATS
2409 schedstat_inc(rq, ttwu_count);
2410 if (cpu == this_cpu)
2411 schedstat_inc(rq, ttwu_local);
2412 else {
2413 struct sched_domain *sd;
2414 for_each_domain(this_cpu, sd) {
Rusty Russell758b2cd2008-11-25 02:35:04 +10302415 if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
Gregory Haskinse7693a32008-01-25 21:08:09 +01002416 schedstat_inc(sd, ttwu_wake_remote);
2417 break;
2418 }
2419 }
2420 }
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02002421#endif /* CONFIG_SCHEDSTATS */
Gregory Haskinse7693a32008-01-25 21:08:09 +01002422
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423out_activate:
2424#endif /* CONFIG_SMP */
Tejun Heo9ed38112009-12-03 15:08:03 +09002425 ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu,
2426 cpu == this_cpu, en_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 success = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428out_running:
Tejun Heo9ed38112009-12-03 15:08:03 +09002429 ttwu_post_activation(p, rq, wake_flags, success);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430out:
2431 task_rq_unlock(rq, &flags);
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002432 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433
2434 return success;
2435}
2436
David Howells50fa6102009-04-28 15:01:38 +01002437/**
Tejun Heo21aa9af2010-06-08 21:40:37 +02002438 * try_to_wake_up_local - try to wake up a local task with rq lock held
2439 * @p: the thread to be awakened
2440 *
2441 * Put @p on the run-queue if it's not alredy there. The caller must
2442 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2443 * the current task. this_rq() stays locked over invocation.
2444 */
2445static void try_to_wake_up_local(struct task_struct *p)
2446{
2447 struct rq *rq = task_rq(p);
2448 bool success = false;
2449
2450 BUG_ON(rq != this_rq());
2451 BUG_ON(p == current);
2452 lockdep_assert_held(&rq->lock);
2453
2454 if (!(p->state & TASK_NORMAL))
2455 return;
2456
2457 if (!p->se.on_rq) {
2458 if (likely(!task_running(rq, p))) {
2459 schedstat_inc(rq, ttwu_count);
2460 schedstat_inc(rq, ttwu_local);
2461 }
2462 ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP);
2463 success = true;
2464 }
2465 ttwu_post_activation(p, rq, 0, success);
2466}
2467
2468/**
David Howells50fa6102009-04-28 15:01:38 +01002469 * wake_up_process - Wake up a specific process
2470 * @p: The process to be woken up.
2471 *
2472 * Attempt to wake up the nominated process and move it to the set of runnable
2473 * processes. Returns 1 if the process was woken up, 0 if it was already
2474 * running.
2475 *
2476 * It may be assumed that this function implies a write memory barrier before
2477 * changing the task state if and only if any tasks are woken up.
2478 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08002479int wake_up_process(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480{
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05002481 return try_to_wake_up(p, TASK_ALL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483EXPORT_SYMBOL(wake_up_process);
2484
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08002485int wake_up_state(struct task_struct *p, unsigned int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486{
2487 return try_to_wake_up(p, state, 0);
2488}
2489
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490/*
2491 * Perform scheduler related setup for a newly forked process p.
2492 * p is forked by current.
Ingo Molnardd41f592007-07-09 18:51:59 +02002493 *
2494 * __sched_fork() is basic setup used by init_idle() too:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 */
Ingo Molnardd41f592007-07-09 18:51:59 +02002496static void __sched_fork(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497{
Ingo Molnardd41f592007-07-09 18:51:59 +02002498 p->se.exec_start = 0;
2499 p->se.sum_exec_runtime = 0;
Ingo Molnarf6cf8912007-08-28 12:53:24 +02002500 p->se.prev_sum_exec_runtime = 0;
Ingo Molnar6c594c22008-12-14 12:34:15 +01002501 p->se.nr_migrations = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02002502
2503#ifdef CONFIG_SCHEDSTATS
Lucas De Marchi41acab82010-03-10 23:37:45 -03002504 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02002505#endif
Nick Piggin476d1392005-06-25 14:57:29 -07002506
Peter Zijlstrafa717062008-01-25 21:08:27 +01002507 INIT_LIST_HEAD(&p->rt.run_list);
Ingo Molnardd41f592007-07-09 18:51:59 +02002508 p->se.on_rq = 0;
Peter Zijlstra4a55bd52008-04-19 19:45:00 +02002509 INIT_LIST_HEAD(&p->se.group_node);
Nick Piggin476d1392005-06-25 14:57:29 -07002510
Avi Kivitye107be32007-07-26 13:40:43 +02002511#ifdef CONFIG_PREEMPT_NOTIFIERS
2512 INIT_HLIST_HEAD(&p->preempt_notifiers);
2513#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02002514}
2515
2516/*
2517 * fork()/clone()-time setup:
2518 */
2519void sched_fork(struct task_struct *p, int clone_flags)
2520{
2521 int cpu = get_cpu();
2522
2523 __sched_fork(p);
Peter Zijlstra06b83b52009-12-16 18:04:35 +01002524 /*
Peter Zijlstra0017d732010-03-24 18:34:10 +01002525 * We mark the process as running here. This guarantees that
Peter Zijlstra06b83b52009-12-16 18:04:35 +01002526 * nobody will actually run it, and a signal or other external
2527 * event cannot wake it up and insert it on the runqueue either.
2528 */
Peter Zijlstra0017d732010-03-24 18:34:10 +01002529 p->state = TASK_RUNNING;
Ingo Molnardd41f592007-07-09 18:51:59 +02002530
Ingo Molnarb29739f2006-06-27 02:54:51 -07002531 /*
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02002532 * Revert to default priority/policy on fork if requested.
2533 */
2534 if (unlikely(p->sched_reset_on_fork)) {
Peter Williamsf83f9ac2009-09-24 06:47:10 +00002535 if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) {
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02002536 p->policy = SCHED_NORMAL;
Peter Williamsf83f9ac2009-09-24 06:47:10 +00002537 p->normal_prio = p->static_prio;
2538 }
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02002539
Mike Galbraith6c697bd2009-06-17 10:48:02 +02002540 if (PRIO_TO_NICE(p->static_prio) < 0) {
2541 p->static_prio = NICE_TO_PRIO(0);
Peter Williamsf83f9ac2009-09-24 06:47:10 +00002542 p->normal_prio = p->static_prio;
Mike Galbraith6c697bd2009-06-17 10:48:02 +02002543 set_load_weight(p);
2544 }
2545
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02002546 /*
2547 * We don't need the reset flag anymore after the fork. It has
2548 * fulfilled its duty:
2549 */
2550 p->sched_reset_on_fork = 0;
2551 }
Lennart Poetteringca94c442009-06-15 17:17:47 +02002552
Peter Williamsf83f9ac2009-09-24 06:47:10 +00002553 /*
2554 * Make sure we do not leak PI boosting priority to the child.
2555 */
2556 p->prio = current->normal_prio;
2557
Hiroshi Shimamoto2ddbf952007-10-15 17:00:11 +02002558 if (!rt_prio(p->prio))
2559 p->sched_class = &fair_sched_class;
Ingo Molnarb29739f2006-06-27 02:54:51 -07002560
Peter Zijlstracd29fe62009-11-27 17:32:46 +01002561 if (p->sched_class->task_fork)
2562 p->sched_class->task_fork(p);
2563
Peter Zijlstra86951592010-06-22 11:44:53 +02002564 /*
2565 * The child is not yet in the pid-hash so no cgroup attach races,
2566 * and the cgroup is pinned to this child due to cgroup_fork()
2567 * is ran before sched_fork().
2568 *
2569 * Silence PROVE_RCU.
2570 */
2571 rcu_read_lock();
Peter Zijlstra5f3edc12009-09-10 13:42:00 +02002572 set_task_cpu(p, cpu);
Peter Zijlstra86951592010-06-22 11:44:53 +02002573 rcu_read_unlock();
Peter Zijlstra5f3edc12009-09-10 13:42:00 +02002574
Chandra Seetharaman52f17b62006-07-14 00:24:38 -07002575#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
Ingo Molnardd41f592007-07-09 18:51:59 +02002576 if (likely(sched_info_on()))
Chandra Seetharaman52f17b62006-07-14 00:24:38 -07002577 memset(&p->sched_info, 0, sizeof(p->sched_info));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578#endif
Chen, Kenneth Wd6077cb2006-02-14 13:53:10 -08002579#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
Nick Piggin4866cde2005-06-25 14:57:23 -07002580 p->oncpu = 0;
2581#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582#ifdef CONFIG_PREEMPT
Nick Piggin4866cde2005-06-25 14:57:23 -07002583 /* Want to start with kernel preemption disabled. */
Al Viroa1261f542005-11-13 16:06:55 -08002584 task_thread_info(p)->preempt_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585#endif
Gregory Haskins917b6272008-12-29 09:39:53 -05002586 plist_node_init(&p->pushable_tasks, MAX_PRIO);
2587
Nick Piggin476d1392005-06-25 14:57:29 -07002588 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589}
2590
2591/*
2592 * wake_up_new_task - wake up a newly created task for the first time.
2593 *
2594 * This function will do some initial scheduler statistics housekeeping
2595 * that must be done for every newly created context, then puts the task
2596 * on the runqueue and wakes it.
2597 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08002598void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599{
2600 unsigned long flags;
Ingo Molnardd41f592007-07-09 18:51:59 +02002601 struct rq *rq;
Andrew Mortonc8906922010-03-11 14:08:43 -08002602 int cpu __maybe_unused = get_cpu();
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002603
2604#ifdef CONFIG_SMP
Peter Zijlstra0017d732010-03-24 18:34:10 +01002605 rq = task_rq_lock(p, &flags);
2606 p->state = TASK_WAKING;
2607
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002608 /*
2609 * Fork balancing, do it here and not earlier because:
2610 * - cpus_allowed can change in the fork path
2611 * - any previously selected cpu might disappear through hotplug
2612 *
Peter Zijlstra0017d732010-03-24 18:34:10 +01002613 * We set TASK_WAKING so that select_task_rq() can drop rq->lock
2614 * without people poking at ->cpus_allowed.
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002615 */
Peter Zijlstra0017d732010-03-24 18:34:10 +01002616 cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0);
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002617 set_task_cpu(p, cpu);
Peter Zijlstra0017d732010-03-24 18:34:10 +01002618
2619 p->state = TASK_RUNNING;
2620 task_rq_unlock(rq, &flags);
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002621#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622
Peter Zijlstra0017d732010-03-24 18:34:10 +01002623 rq = task_rq_lock(p, &flags);
Peter Zijlstracd29fe62009-11-27 17:32:46 +01002624 activate_task(rq, p, 0);
Peter Zijlstra27a9da62010-05-04 20:36:56 +02002625 trace_sched_wakeup_new(p, 1);
Peter Zijlstraa7558e02009-09-14 20:02:34 +02002626 check_preempt_curr(rq, p, WF_FORK);
Steven Rostedt9a897c52008-01-25 21:08:22 +01002627#ifdef CONFIG_SMP
Peter Zijlstraefbbd052009-12-16 18:04:40 +01002628 if (p->sched_class->task_woken)
2629 p->sched_class->task_woken(rq, p);
Steven Rostedt9a897c52008-01-25 21:08:22 +01002630#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02002631 task_rq_unlock(rq, &flags);
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002632 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633}
2634
Avi Kivitye107be32007-07-26 13:40:43 +02002635#ifdef CONFIG_PREEMPT_NOTIFIERS
2636
2637/**
Luis Henriques80dd99b2009-03-16 19:58:09 +00002638 * preempt_notifier_register - tell me when current is being preempted & rescheduled
Randy Dunlap421cee22007-07-31 00:37:50 -07002639 * @notifier: notifier struct to register
Avi Kivitye107be32007-07-26 13:40:43 +02002640 */
2641void preempt_notifier_register(struct preempt_notifier *notifier)
2642{
2643 hlist_add_head(&notifier->link, &current->preempt_notifiers);
2644}
2645EXPORT_SYMBOL_GPL(preempt_notifier_register);
2646
2647/**
2648 * preempt_notifier_unregister - no longer interested in preemption notifications
Randy Dunlap421cee22007-07-31 00:37:50 -07002649 * @notifier: notifier struct to unregister
Avi Kivitye107be32007-07-26 13:40:43 +02002650 *
2651 * This is safe to call from within a preemption notifier.
2652 */
2653void preempt_notifier_unregister(struct preempt_notifier *notifier)
2654{
2655 hlist_del(&notifier->link);
2656}
2657EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
2658
2659static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2660{
2661 struct preempt_notifier *notifier;
2662 struct hlist_node *node;
2663
2664 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2665 notifier->ops->sched_in(notifier, raw_smp_processor_id());
2666}
2667
2668static void
2669fire_sched_out_preempt_notifiers(struct task_struct *curr,
2670 struct task_struct *next)
2671{
2672 struct preempt_notifier *notifier;
2673 struct hlist_node *node;
2674
2675 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2676 notifier->ops->sched_out(notifier, next);
2677}
2678
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02002679#else /* !CONFIG_PREEMPT_NOTIFIERS */
Avi Kivitye107be32007-07-26 13:40:43 +02002680
2681static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2682{
2683}
2684
2685static void
2686fire_sched_out_preempt_notifiers(struct task_struct *curr,
2687 struct task_struct *next)
2688{
2689}
2690
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02002691#endif /* CONFIG_PREEMPT_NOTIFIERS */
Avi Kivitye107be32007-07-26 13:40:43 +02002692
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693/**
Nick Piggin4866cde2005-06-25 14:57:23 -07002694 * prepare_task_switch - prepare to switch tasks
2695 * @rq: the runqueue preparing to switch
Randy Dunlap421cee22007-07-31 00:37:50 -07002696 * @prev: the current task that is being switched out
Nick Piggin4866cde2005-06-25 14:57:23 -07002697 * @next: the task we are going to switch to.
2698 *
2699 * This is called with the rq lock held and interrupts off. It must
2700 * be paired with a subsequent finish_task_switch after the context
2701 * switch.
2702 *
2703 * prepare_task_switch sets up locking and calls architecture specific
2704 * hooks.
2705 */
Avi Kivitye107be32007-07-26 13:40:43 +02002706static inline void
2707prepare_task_switch(struct rq *rq, struct task_struct *prev,
2708 struct task_struct *next)
Nick Piggin4866cde2005-06-25 14:57:23 -07002709{
Avi Kivitye107be32007-07-26 13:40:43 +02002710 fire_sched_out_preempt_notifiers(prev, next);
Nick Piggin4866cde2005-06-25 14:57:23 -07002711 prepare_lock_switch(rq, next);
2712 prepare_arch_switch(next);
2713}
2714
2715/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 * finish_task_switch - clean up after a task-switch
Jeff Garzik344baba2005-09-07 01:15:17 -04002717 * @rq: runqueue associated with task-switch
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 * @prev: the thread we just switched away from.
2719 *
Nick Piggin4866cde2005-06-25 14:57:23 -07002720 * finish_task_switch must be called after the context switch, paired
2721 * with a prepare_task_switch call before the context switch.
2722 * finish_task_switch will reconcile locking set up by prepare_task_switch,
2723 * and do any other architecture-specific cleanup actions.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 *
2725 * Note that we may have delayed dropping an mm in context_switch(). If
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01002726 * so, we finish that here outside of the runqueue lock. (Doing it
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 * with the lock held can cause deadlocks; see schedule() for
2728 * details.)
2729 */
Alexey Dobriyana9957442007-10-15 17:00:13 +02002730static void finish_task_switch(struct rq *rq, struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 __releases(rq->lock)
2732{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 struct mm_struct *mm = rq->prev_mm;
Oleg Nesterov55a101f2006-09-29 02:01:10 -07002734 long prev_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735
2736 rq->prev_mm = NULL;
2737
2738 /*
2739 * A task struct has one reference for the use as "current".
Oleg Nesterovc394cc92006-09-29 02:01:11 -07002740 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
Oleg Nesterov55a101f2006-09-29 02:01:10 -07002741 * schedule one last time. The schedule call will never return, and
2742 * the scheduled task must drop that reference.
Oleg Nesterovc394cc92006-09-29 02:01:11 -07002743 * The test for TASK_DEAD must occur while the runqueue locks are
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 * still held, otherwise prev could be scheduled on another cpu, die
2745 * there before we look at prev->state, and then the reference would
2746 * be dropped twice.
2747 * Manfred Spraul <manfred@colorfullife.com>
2748 */
Oleg Nesterov55a101f2006-09-29 02:01:10 -07002749 prev_state = prev->state;
Nick Piggin4866cde2005-06-25 14:57:23 -07002750 finish_arch_switch(prev);
Jamie Iles8381f652010-01-08 15:27:33 +00002751#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2752 local_irq_disable();
2753#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
Peter Zijlstra49f47432009-12-27 11:51:52 +01002754 perf_event_task_sched_in(current);
Jamie Iles8381f652010-01-08 15:27:33 +00002755#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2756 local_irq_enable();
2757#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
Nick Piggin4866cde2005-06-25 14:57:23 -07002758 finish_lock_switch(rq, prev);
Steven Rostedte8fa1362008-01-25 21:08:05 +01002759
Avi Kivitye107be32007-07-26 13:40:43 +02002760 fire_sched_in_preempt_notifiers(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 if (mm)
2762 mmdrop(mm);
Oleg Nesterovc394cc92006-09-29 02:01:11 -07002763 if (unlikely(prev_state == TASK_DEAD)) {
bibo maoc6fd91f2006-03-26 01:38:20 -08002764 /*
2765 * Remove function-return probe instances associated with this
2766 * task and put them back on the free list.
Ingo Molnar9761eea2007-07-09 18:52:00 +02002767 */
bibo maoc6fd91f2006-03-26 01:38:20 -08002768 kprobe_flush_task(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 put_task_struct(prev);
bibo maoc6fd91f2006-03-26 01:38:20 -08002770 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771}
2772
Gregory Haskins3f029d32009-07-29 11:08:47 -04002773#ifdef CONFIG_SMP
2774
2775/* assumes rq->lock is held */
2776static inline void pre_schedule(struct rq *rq, struct task_struct *prev)
2777{
2778 if (prev->sched_class->pre_schedule)
2779 prev->sched_class->pre_schedule(rq, prev);
2780}
2781
2782/* rq->lock is NOT held, but preemption is disabled */
2783static inline void post_schedule(struct rq *rq)
2784{
2785 if (rq->post_schedule) {
2786 unsigned long flags;
2787
Thomas Gleixner05fa7852009-11-17 14:28:38 +01002788 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins3f029d32009-07-29 11:08:47 -04002789 if (rq->curr->sched_class->post_schedule)
2790 rq->curr->sched_class->post_schedule(rq);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01002791 raw_spin_unlock_irqrestore(&rq->lock, flags);
Gregory Haskins3f029d32009-07-29 11:08:47 -04002792
2793 rq->post_schedule = 0;
2794 }
2795}
2796
2797#else
2798
2799static inline void pre_schedule(struct rq *rq, struct task_struct *p)
2800{
2801}
2802
2803static inline void post_schedule(struct rq *rq)
2804{
2805}
2806
2807#endif
2808
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809/**
2810 * schedule_tail - first thing a freshly forked thread must call.
2811 * @prev: the thread we just switched away from.
2812 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07002813asmlinkage void schedule_tail(struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 __releases(rq->lock)
2815{
Ingo Molnar70b97a72006-07-03 00:25:42 -07002816 struct rq *rq = this_rq();
2817
Nick Piggin4866cde2005-06-25 14:57:23 -07002818 finish_task_switch(rq, prev);
Steven Rostedtda19ab52009-07-29 00:21:22 -04002819
Gregory Haskins3f029d32009-07-29 11:08:47 -04002820 /*
2821 * FIXME: do we need to worry about rq being invalidated by the
2822 * task_switch?
2823 */
2824 post_schedule(rq);
Steven Rostedtda19ab52009-07-29 00:21:22 -04002825
Nick Piggin4866cde2005-06-25 14:57:23 -07002826#ifdef __ARCH_WANT_UNLOCKED_CTXSW
2827 /* In this case, finish_task_switch does not reenable preemption */
2828 preempt_enable();
2829#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 if (current->set_child_tid)
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002831 put_user(task_pid_vnr(current), current->set_child_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832}
2833
2834/*
2835 * context_switch - switch to the new MM and the new
2836 * thread's register state.
2837 */
Ingo Molnardd41f592007-07-09 18:51:59 +02002838static inline void
Ingo Molnar70b97a72006-07-03 00:25:42 -07002839context_switch(struct rq *rq, struct task_struct *prev,
Ingo Molnar36c8b582006-07-03 00:25:41 -07002840 struct task_struct *next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841{
Ingo Molnardd41f592007-07-09 18:51:59 +02002842 struct mm_struct *mm, *oldmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843
Avi Kivitye107be32007-07-26 13:40:43 +02002844 prepare_task_switch(rq, prev, next);
Peter Zijlstra27a9da62010-05-04 20:36:56 +02002845 trace_sched_switch(prev, next);
Ingo Molnardd41f592007-07-09 18:51:59 +02002846 mm = next->mm;
2847 oldmm = prev->active_mm;
Zachary Amsden9226d122007-02-13 13:26:21 +01002848 /*
2849 * For paravirt, this is coupled with an exit in switch_to to
2850 * combine the page table reload and the switch backend into
2851 * one hypercall.
2852 */
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -08002853 arch_start_context_switch(prev);
Zachary Amsden9226d122007-02-13 13:26:21 +01002854
Tim Blechmann710390d2009-11-24 11:55:27 +01002855 if (likely(!mm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 next->active_mm = oldmm;
2857 atomic_inc(&oldmm->mm_count);
2858 enter_lazy_tlb(oldmm, next);
2859 } else
2860 switch_mm(oldmm, mm, next);
2861
Tim Blechmann710390d2009-11-24 11:55:27 +01002862 if (likely(!prev->mm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 prev->active_mm = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 rq->prev_mm = oldmm;
2865 }
Ingo Molnar3a5f5e42006-07-14 00:24:27 -07002866 /*
2867 * Since the runqueue lock will be released by the next
2868 * task (which is an invalid locking op but in the case
2869 * of the scheduler it's an obvious special-case), so we
2870 * do an early lockdep release here:
2871 */
2872#ifndef __ARCH_WANT_UNLOCKED_CTXSW
Ingo Molnar8a25d5d2006-07-03 00:24:54 -07002873 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
Ingo Molnar3a5f5e42006-07-14 00:24:27 -07002874#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875
2876 /* Here we just switch the register state and the stack. */
2877 switch_to(prev, next, prev);
2878
Ingo Molnardd41f592007-07-09 18:51:59 +02002879 barrier();
2880 /*
2881 * this_rq must be evaluated again because prev may have moved
2882 * CPUs since it called schedule(), thus the 'rq' on its stack
2883 * frame will be invalid.
2884 */
2885 finish_task_switch(this_rq(), prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886}
2887
2888/*
2889 * nr_running, nr_uninterruptible and nr_context_switches:
2890 *
2891 * externally visible scheduler statistics: current number of runnable
2892 * threads, current number of uninterruptible-sleeping threads, total
2893 * number of context switches performed since bootup.
2894 */
2895unsigned long nr_running(void)
2896{
2897 unsigned long i, sum = 0;
2898
2899 for_each_online_cpu(i)
2900 sum += cpu_rq(i)->nr_running;
2901
2902 return sum;
2903}
2904
2905unsigned long nr_uninterruptible(void)
2906{
2907 unsigned long i, sum = 0;
2908
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002909 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910 sum += cpu_rq(i)->nr_uninterruptible;
2911
2912 /*
2913 * Since we read the counters lockless, it might be slightly
2914 * inaccurate. Do not allow it to go below zero though:
2915 */
2916 if (unlikely((long)sum < 0))
2917 sum = 0;
2918
2919 return sum;
2920}
2921
2922unsigned long long nr_context_switches(void)
2923{
Steven Rostedtcc94abf2006-06-27 02:54:31 -07002924 int i;
2925 unsigned long long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002927 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 sum += cpu_rq(i)->nr_switches;
2929
2930 return sum;
2931}
2932
2933unsigned long nr_iowait(void)
2934{
2935 unsigned long i, sum = 0;
2936
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002937 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938 sum += atomic_read(&cpu_rq(i)->nr_iowait);
2939
2940 return sum;
2941}
2942
Peter Zijlstra8c215bd2010-07-01 09:07:17 +02002943unsigned long nr_iowait_cpu(int cpu)
Arjan van de Ven69d25872009-09-21 17:04:08 -07002944{
Peter Zijlstra8c215bd2010-07-01 09:07:17 +02002945 struct rq *this = cpu_rq(cpu);
Arjan van de Ven69d25872009-09-21 17:04:08 -07002946 return atomic_read(&this->nr_iowait);
2947}
2948
2949unsigned long this_cpu_load(void)
2950{
2951 struct rq *this = this_rq();
2952 return this->cpu_load[0];
2953}
2954
2955
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002956/* Variables and functions for calc_load */
2957static atomic_long_t calc_load_tasks;
2958static unsigned long calc_load_update;
2959unsigned long avenrun[3];
2960EXPORT_SYMBOL(avenrun);
2961
Peter Zijlstra74f51872010-04-22 21:50:19 +02002962static long calc_load_fold_active(struct rq *this_rq)
2963{
2964 long nr_active, delta = 0;
2965
2966 nr_active = this_rq->nr_running;
2967 nr_active += (long) this_rq->nr_uninterruptible;
2968
2969 if (nr_active != this_rq->calc_load_active) {
2970 delta = nr_active - this_rq->calc_load_active;
2971 this_rq->calc_load_active = nr_active;
2972 }
2973
2974 return delta;
2975}
2976
2977#ifdef CONFIG_NO_HZ
2978/*
2979 * For NO_HZ we delay the active fold to the next LOAD_FREQ update.
2980 *
2981 * When making the ILB scale, we should try to pull this in as well.
2982 */
2983static atomic_long_t calc_load_tasks_idle;
2984
2985static void calc_load_account_idle(struct rq *this_rq)
2986{
2987 long delta;
2988
2989 delta = calc_load_fold_active(this_rq);
2990 if (delta)
2991 atomic_long_add(delta, &calc_load_tasks_idle);
2992}
2993
2994static long calc_load_fold_idle(void)
2995{
2996 long delta = 0;
2997
2998 /*
2999 * Its got a race, we don't care...
3000 */
3001 if (atomic_long_read(&calc_load_tasks_idle))
3002 delta = atomic_long_xchg(&calc_load_tasks_idle, 0);
3003
3004 return delta;
3005}
3006#else
3007static void calc_load_account_idle(struct rq *this_rq)
3008{
3009}
3010
3011static inline long calc_load_fold_idle(void)
3012{
3013 return 0;
3014}
3015#endif
3016
Thomas Gleixner2d024942009-05-02 20:08:52 +02003017/**
3018 * get_avenrun - get the load average array
3019 * @loads: pointer to dest load array
3020 * @offset: offset to add
3021 * @shift: shift count to shift the result left
3022 *
3023 * These values are estimates at best, so no need for locking.
3024 */
3025void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
3026{
3027 loads[0] = (avenrun[0] + offset) << shift;
3028 loads[1] = (avenrun[1] + offset) << shift;
3029 loads[2] = (avenrun[2] + offset) << shift;
3030}
3031
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003032static unsigned long
3033calc_load(unsigned long load, unsigned long exp, unsigned long active)
Jack Steinerdb1b1fe2006-03-31 02:31:21 -08003034{
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003035 load *= exp;
3036 load += active * (FIXED_1 - exp);
3037 return load >> FSHIFT;
3038}
Jack Steinerdb1b1fe2006-03-31 02:31:21 -08003039
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003040/*
3041 * calc_load - update the avenrun load estimates 10 ticks after the
3042 * CPUs have updated calc_load_tasks.
3043 */
3044void calc_global_load(void)
3045{
3046 unsigned long upd = calc_load_update + 10;
3047 long active;
3048
3049 if (time_before(jiffies, upd))
3050 return;
3051
3052 active = atomic_long_read(&calc_load_tasks);
3053 active = active > 0 ? active * FIXED_1 : 0;
3054
3055 avenrun[0] = calc_load(avenrun[0], EXP_1, active);
3056 avenrun[1] = calc_load(avenrun[1], EXP_5, active);
3057 avenrun[2] = calc_load(avenrun[2], EXP_15, active);
3058
3059 calc_load_update += LOAD_FREQ;
3060}
3061
3062/*
Peter Zijlstra74f51872010-04-22 21:50:19 +02003063 * Called from update_cpu_load() to periodically update this CPU's
3064 * active count.
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003065 */
3066static void calc_load_account_active(struct rq *this_rq)
3067{
Peter Zijlstra74f51872010-04-22 21:50:19 +02003068 long delta;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003069
Peter Zijlstra74f51872010-04-22 21:50:19 +02003070 if (time_before(jiffies, this_rq->calc_load_update))
3071 return;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003072
Peter Zijlstra74f51872010-04-22 21:50:19 +02003073 delta = calc_load_fold_active(this_rq);
3074 delta += calc_load_fold_idle();
3075 if (delta)
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003076 atomic_long_add(delta, &calc_load_tasks);
Peter Zijlstra74f51872010-04-22 21:50:19 +02003077
3078 this_rq->calc_load_update += LOAD_FREQ;
Jack Steinerdb1b1fe2006-03-31 02:31:21 -08003079}
3080
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081/*
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003082 * The exact cpuload at various idx values, calculated at every tick would be
3083 * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load
3084 *
3085 * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called
3086 * on nth tick when cpu may be busy, then we have:
3087 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3088 * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load
3089 *
3090 * decay_load_missed() below does efficient calculation of
3091 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3092 * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load
3093 *
3094 * The calculation is approximated on a 128 point scale.
3095 * degrade_zero_ticks is the number of ticks after which load at any
3096 * particular idx is approximated to be zero.
3097 * degrade_factor is a precomputed table, a row for each load idx.
3098 * Each column corresponds to degradation factor for a power of two ticks,
3099 * based on 128 point scale.
3100 * Example:
3101 * row 2, col 3 (=12) says that the degradation at load idx 2 after
3102 * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8).
3103 *
3104 * With this power of 2 load factors, we can degrade the load n times
3105 * by looking at 1 bits in n and doing as many mult/shift instead of
3106 * n mult/shifts needed by the exact degradation.
3107 */
3108#define DEGRADE_SHIFT 7
3109static const unsigned char
3110 degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128};
3111static const unsigned char
3112 degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = {
3113 {0, 0, 0, 0, 0, 0, 0, 0},
3114 {64, 32, 8, 0, 0, 0, 0, 0},
3115 {96, 72, 40, 12, 1, 0, 0},
3116 {112, 98, 75, 43, 15, 1, 0},
3117 {120, 112, 98, 76, 45, 16, 2} };
3118
3119/*
3120 * Update cpu_load for any missed ticks, due to tickless idle. The backlog
3121 * would be when CPU is idle and so we just decay the old load without
3122 * adding any new load.
3123 */
3124static unsigned long
3125decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
3126{
3127 int j = 0;
3128
3129 if (!missed_updates)
3130 return load;
3131
3132 if (missed_updates >= degrade_zero_ticks[idx])
3133 return 0;
3134
3135 if (idx == 1)
3136 return load >> missed_updates;
3137
3138 while (missed_updates) {
3139 if (missed_updates % 2)
3140 load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT;
3141
3142 missed_updates >>= 1;
3143 j++;
3144 }
3145 return load;
3146}
3147
3148/*
Ingo Molnardd41f592007-07-09 18:51:59 +02003149 * Update rq->cpu_load[] statistics. This function is usually called every
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003150 * scheduler tick (TICK_NSEC). With tickless idle this will not be called
3151 * every tick. We fix it up based on jiffies.
Ingo Molnar48f24c42006-07-03 00:25:40 -07003152 */
Ingo Molnardd41f592007-07-09 18:51:59 +02003153static void update_cpu_load(struct rq *this_rq)
Ingo Molnar48f24c42006-07-03 00:25:40 -07003154{
Dmitry Adamushko495eca42007-10-15 17:00:06 +02003155 unsigned long this_load = this_rq->load.weight;
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003156 unsigned long curr_jiffies = jiffies;
3157 unsigned long pending_updates;
Ingo Molnardd41f592007-07-09 18:51:59 +02003158 int i, scale;
3159
3160 this_rq->nr_load_updates++;
Ingo Molnardd41f592007-07-09 18:51:59 +02003161
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003162 /* Avoid repeated calls on same jiffy, when moving in and out of idle */
3163 if (curr_jiffies == this_rq->last_load_update_tick)
3164 return;
3165
3166 pending_updates = curr_jiffies - this_rq->last_load_update_tick;
3167 this_rq->last_load_update_tick = curr_jiffies;
3168
Ingo Molnardd41f592007-07-09 18:51:59 +02003169 /* Update our load: */
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003170 this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */
3171 for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
Ingo Molnardd41f592007-07-09 18:51:59 +02003172 unsigned long old_load, new_load;
3173
3174 /* scale is effectively 1 << i now, and >> i divides by scale */
3175
3176 old_load = this_rq->cpu_load[i];
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003177 old_load = decay_load_missed(old_load, pending_updates - 1, i);
Ingo Molnardd41f592007-07-09 18:51:59 +02003178 new_load = this_load;
Ingo Molnara25707f2007-10-15 17:00:03 +02003179 /*
3180 * Round up the averaging division if load is increasing. This
3181 * prevents us from getting stuck on 9 if the load is 10, for
3182 * example.
3183 */
3184 if (new_load > old_load)
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003185 new_load += scale - 1;
3186
3187 this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i;
Ingo Molnardd41f592007-07-09 18:51:59 +02003188 }
Suresh Siddhada2b71e2010-08-23 13:42:51 -07003189
3190 sched_avg_update(this_rq);
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003191}
3192
3193static void update_cpu_load_active(struct rq *this_rq)
3194{
3195 update_cpu_load(this_rq);
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003196
Peter Zijlstra74f51872010-04-22 21:50:19 +02003197 calc_load_account_active(this_rq);
Ingo Molnar48f24c42006-07-03 00:25:40 -07003198}
3199
Ingo Molnardd41f592007-07-09 18:51:59 +02003200#ifdef CONFIG_SMP
3201
Ingo Molnar48f24c42006-07-03 00:25:40 -07003202/*
Peter Zijlstra38022902009-12-16 18:04:37 +01003203 * sched_exec - execve() is a valuable balancing opportunity, because at
3204 * this point the task has the smallest effective memory and cache footprint.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003205 */
Peter Zijlstra38022902009-12-16 18:04:37 +01003206void sched_exec(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207{
Peter Zijlstra38022902009-12-16 18:04:37 +01003208 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003209 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07003210 struct rq *rq;
Peter Zijlstra0017d732010-03-24 18:34:10 +01003211 int dest_cpu;
Peter Zijlstra38022902009-12-16 18:04:37 +01003212
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213 rq = task_rq_lock(p, &flags);
Peter Zijlstra0017d732010-03-24 18:34:10 +01003214 dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0);
3215 if (dest_cpu == smp_processor_id())
3216 goto unlock;
Peter Zijlstra38022902009-12-16 18:04:37 +01003217
3218 /*
3219 * select_task_rq() can race against ->cpus_allowed
3220 */
Oleg Nesterov30da6882010-03-15 10:10:19 +01003221 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) &&
Tejun Heo969c7922010-05-06 18:49:21 +02003222 likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) {
3223 struct migration_arg arg = { p, dest_cpu };
Ingo Molnar36c8b582006-07-03 00:25:41 -07003224
Linus Torvalds1da177e2005-04-16 15:20:36 -07003225 task_rq_unlock(rq, &flags);
Tejun Heo969c7922010-05-06 18:49:21 +02003226 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227 return;
3228 }
Peter Zijlstra0017d732010-03-24 18:34:10 +01003229unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230 task_rq_unlock(rq, &flags);
3231}
3232
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233#endif
3234
Linus Torvalds1da177e2005-04-16 15:20:36 -07003235DEFINE_PER_CPU(struct kernel_stat, kstat);
3236
3237EXPORT_PER_CPU_SYMBOL(kstat);
3238
3239/*
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003240 * Return any ns on the sched_clock that have not yet been accounted in
Frank Mayharf06febc2008-09-12 09:54:39 -07003241 * @p in case that task is currently running.
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003242 *
3243 * Called with task_rq_lock() held on @rq.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244 */
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003245static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq)
3246{
3247 u64 ns = 0;
3248
3249 if (task_current(rq, p)) {
3250 update_rq_clock(rq);
3251 ns = rq->clock - p->se.exec_start;
3252 if ((s64)ns < 0)
3253 ns = 0;
3254 }
3255
3256 return ns;
3257}
3258
Frank Mayharbb34d922008-09-12 09:54:39 -07003259unsigned long long task_delta_exec(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261 unsigned long flags;
Ingo Molnar41b86e92007-07-09 18:51:58 +02003262 struct rq *rq;
Frank Mayharbb34d922008-09-12 09:54:39 -07003263 u64 ns = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07003264
Ingo Molnar41b86e92007-07-09 18:51:58 +02003265 rq = task_rq_lock(p, &flags);
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003266 ns = do_task_delta_exec(p, rq);
3267 task_rq_unlock(rq, &flags);
Ingo Molnar15084872008-09-30 08:28:17 +02003268
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003269 return ns;
3270}
Frank Mayharf06febc2008-09-12 09:54:39 -07003271
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003272/*
3273 * Return accounted runtime for the task.
3274 * In case the task is currently running, return the runtime plus current's
3275 * pending runtime that have not been accounted yet.
3276 */
3277unsigned long long task_sched_runtime(struct task_struct *p)
3278{
3279 unsigned long flags;
3280 struct rq *rq;
3281 u64 ns = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07003282
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003283 rq = task_rq_lock(p, &flags);
3284 ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq);
3285 task_rq_unlock(rq, &flags);
3286
3287 return ns;
3288}
3289
3290/*
3291 * Return sum_exec_runtime for the thread group.
3292 * In case the task is currently running, return the sum plus current's
3293 * pending runtime that have not been accounted yet.
3294 *
3295 * Note that the thread group might have other running tasks as well,
3296 * so the return value not includes other pending runtime that other
3297 * running tasks might have.
3298 */
3299unsigned long long thread_group_sched_runtime(struct task_struct *p)
3300{
3301 struct task_cputime totals;
3302 unsigned long flags;
3303 struct rq *rq;
3304 u64 ns;
3305
3306 rq = task_rq_lock(p, &flags);
3307 thread_group_cputime(p, &totals);
3308 ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309 task_rq_unlock(rq, &flags);
3310
3311 return ns;
3312}
3313
3314/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 * Account user cpu time to a process.
3316 * @p: the process that the cpu time gets accounted to
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317 * @cputime: the cpu time spent in user space since the last update
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003318 * @cputime_scaled: cputime scaled by cpu frequency
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 */
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003320void account_user_time(struct task_struct *p, cputime_t cputime,
3321 cputime_t cputime_scaled)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322{
3323 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3324 cputime64_t tmp;
3325
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003326 /* Add user time to process. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327 p->utime = cputime_add(p->utime, cputime);
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003328 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
Frank Mayharf06febc2008-09-12 09:54:39 -07003329 account_group_user_time(p, cputime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330
3331 /* Add user time to cpustat. */
3332 tmp = cputime_to_cputime64(cputime);
3333 if (TASK_NICE(p) > 0)
3334 cpustat->nice = cputime64_add(cpustat->nice, tmp);
3335 else
3336 cpustat->user = cputime64_add(cpustat->user, tmp);
Bharata B Raoef12fef2009-03-31 10:02:22 +05303337
3338 cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime);
Jonathan Lim49b5cf32008-07-25 01:48:40 -07003339 /* Account for user time used */
3340 acct_update_integrals(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341}
3342
3343/*
Laurent Vivier94886b82007-10-15 17:00:19 +02003344 * Account guest cpu time to a process.
3345 * @p: the process that the cpu time gets accounted to
3346 * @cputime: the cpu time spent in virtual machine since the last update
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003347 * @cputime_scaled: cputime scaled by cpu frequency
Laurent Vivier94886b82007-10-15 17:00:19 +02003348 */
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003349static void account_guest_time(struct task_struct *p, cputime_t cputime,
3350 cputime_t cputime_scaled)
Laurent Vivier94886b82007-10-15 17:00:19 +02003351{
3352 cputime64_t tmp;
3353 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3354
3355 tmp = cputime_to_cputime64(cputime);
3356
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003357 /* Add guest time to process. */
Laurent Vivier94886b82007-10-15 17:00:19 +02003358 p->utime = cputime_add(p->utime, cputime);
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003359 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
Frank Mayharf06febc2008-09-12 09:54:39 -07003360 account_group_user_time(p, cputime);
Laurent Vivier94886b82007-10-15 17:00:19 +02003361 p->gtime = cputime_add(p->gtime, cputime);
3362
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003363 /* Add guest time to cpustat. */
Ryota Ozakice0e7b22009-10-24 01:20:10 +09003364 if (TASK_NICE(p) > 0) {
3365 cpustat->nice = cputime64_add(cpustat->nice, tmp);
3366 cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp);
3367 } else {
3368 cpustat->user = cputime64_add(cpustat->user, tmp);
3369 cpustat->guest = cputime64_add(cpustat->guest, tmp);
3370 }
Laurent Vivier94886b82007-10-15 17:00:19 +02003371}
3372
3373/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374 * Account system cpu time to a process.
3375 * @p: the process that the cpu time gets accounted to
3376 * @hardirq_offset: the offset to subtract from hardirq_count()
3377 * @cputime: the cpu time spent in kernel space since the last update
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003378 * @cputime_scaled: cputime scaled by cpu frequency
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379 */
3380void account_system_time(struct task_struct *p, int hardirq_offset,
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003381 cputime_t cputime, cputime_t cputime_scaled)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382{
3383 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003384 cputime64_t tmp;
3385
Harvey Harrison983ed7a2008-04-24 18:17:55 -07003386 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003387 account_guest_time(p, cputime, cputime_scaled);
Harvey Harrison983ed7a2008-04-24 18:17:55 -07003388 return;
3389 }
Laurent Vivier94886b82007-10-15 17:00:19 +02003390
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003391 /* Add system time to process. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392 p->stime = cputime_add(p->stime, cputime);
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003393 p->stimescaled = cputime_add(p->stimescaled, cputime_scaled);
Frank Mayharf06febc2008-09-12 09:54:39 -07003394 account_group_system_time(p, cputime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395
3396 /* Add system time to cpustat. */
3397 tmp = cputime_to_cputime64(cputime);
3398 if (hardirq_count() - hardirq_offset)
3399 cpustat->irq = cputime64_add(cpustat->irq, tmp);
3400 else if (softirq_count())
3401 cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402 else
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003403 cpustat->system = cputime64_add(cpustat->system, tmp);
3404
Bharata B Raoef12fef2009-03-31 10:02:22 +05303405 cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime);
3406
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407 /* Account for system time used */
3408 acct_update_integrals(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409}
3410
3411/*
3412 * Account for involuntary wait time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413 * @steal: the cpu time spent in involuntary wait
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414 */
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003415void account_steal_time(cputime_t cputime)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003416{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003418 cputime64_t cputime64 = cputime_to_cputime64(cputime);
3419
3420 cpustat->steal = cputime64_add(cpustat->steal, cputime64);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421}
3422
Christoph Lameter7835b982006-12-10 02:20:22 -08003423/*
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003424 * Account for idle time.
3425 * @cputime: the cpu time spent in idle wait
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 */
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003427void account_idle_time(cputime_t cputime)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428{
3429 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003430 cputime64_t cputime64 = cputime_to_cputime64(cputime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431 struct rq *rq = this_rq();
3432
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003433 if (atomic_read(&rq->nr_iowait) > 0)
3434 cpustat->iowait = cputime64_add(cpustat->iowait, cputime64);
3435 else
3436 cpustat->idle = cputime64_add(cpustat->idle, cputime64);
Christoph Lameter7835b982006-12-10 02:20:22 -08003437}
3438
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003439#ifndef CONFIG_VIRT_CPU_ACCOUNTING
3440
3441/*
3442 * Account a single tick of cpu time.
3443 * @p: the process that the cpu time gets accounted to
3444 * @user_tick: indicates if the tick is a user or a system tick
3445 */
3446void account_process_tick(struct task_struct *p, int user_tick)
3447{
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02003448 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003449 struct rq *rq = this_rq();
3450
3451 if (user_tick)
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02003452 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
Eric Dumazetf5f293a2009-04-29 14:44:49 +02003453 else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02003454 account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003455 one_jiffy_scaled);
3456 else
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02003457 account_idle_time(cputime_one_jiffy);
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003458}
3459
3460/*
3461 * Account multiple ticks of steal time.
3462 * @p: the process from which the cpu time has been stolen
3463 * @ticks: number of stolen ticks
3464 */
3465void account_steal_ticks(unsigned long ticks)
3466{
3467 account_steal_time(jiffies_to_cputime(ticks));
3468}
3469
3470/*
3471 * Account multiple ticks of idle time.
3472 * @ticks: number of stolen ticks
3473 */
3474void account_idle_ticks(unsigned long ticks)
3475{
3476 account_idle_time(jiffies_to_cputime(ticks));
3477}
3478
3479#endif
3480
Christoph Lameter7835b982006-12-10 02:20:22 -08003481/*
Balbir Singh49048622008-09-05 18:12:23 +02003482 * Use precise platform statistics if available:
3483 */
3484#ifdef CONFIG_VIRT_CPU_ACCOUNTING
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003485void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
Balbir Singh49048622008-09-05 18:12:23 +02003486{
Hidetoshi Setod99ca3b2009-12-02 17:26:47 +09003487 *ut = p->utime;
3488 *st = p->stime;
Balbir Singh49048622008-09-05 18:12:23 +02003489}
3490
Hidetoshi Seto0cf55e12009-12-02 17:28:07 +09003491void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
Balbir Singh49048622008-09-05 18:12:23 +02003492{
Hidetoshi Seto0cf55e12009-12-02 17:28:07 +09003493 struct task_cputime cputime;
3494
3495 thread_group_cputime(p, &cputime);
3496
3497 *ut = cputime.utime;
3498 *st = cputime.stime;
Balbir Singh49048622008-09-05 18:12:23 +02003499}
3500#else
Hidetoshi Seto761b1d22009-11-12 13:33:45 +09003501
3502#ifndef nsecs_to_cputime
Hidetoshi Setob7b20df92009-11-26 14:49:27 +09003503# define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs)
Hidetoshi Seto761b1d22009-11-12 13:33:45 +09003504#endif
3505
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003506void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
Balbir Singh49048622008-09-05 18:12:23 +02003507{
Hidetoshi Setod99ca3b2009-12-02 17:26:47 +09003508 cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime);
Balbir Singh49048622008-09-05 18:12:23 +02003509
3510 /*
3511 * Use CFS's precise accounting:
3512 */
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003513 rtime = nsecs_to_cputime(p->se.sum_exec_runtime);
Balbir Singh49048622008-09-05 18:12:23 +02003514
3515 if (total) {
Stanislaw Gruszkae75e8632010-09-14 16:35:14 +02003516 u64 temp = rtime;
Balbir Singh49048622008-09-05 18:12:23 +02003517
Stanislaw Gruszkae75e8632010-09-14 16:35:14 +02003518 temp *= utime;
Balbir Singh49048622008-09-05 18:12:23 +02003519 do_div(temp, total);
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003520 utime = (cputime_t)temp;
3521 } else
3522 utime = rtime;
Balbir Singh49048622008-09-05 18:12:23 +02003523
3524 /*
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003525 * Compare with previous values, to keep monotonicity:
Balbir Singh49048622008-09-05 18:12:23 +02003526 */
Hidetoshi Seto761b1d22009-11-12 13:33:45 +09003527 p->prev_utime = max(p->prev_utime, utime);
Hidetoshi Setod99ca3b2009-12-02 17:26:47 +09003528 p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime));
Balbir Singh49048622008-09-05 18:12:23 +02003529
Hidetoshi Setod99ca3b2009-12-02 17:26:47 +09003530 *ut = p->prev_utime;
3531 *st = p->prev_stime;
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003532}
Balbir Singh49048622008-09-05 18:12:23 +02003533
Hidetoshi Seto0cf55e12009-12-02 17:28:07 +09003534/*
3535 * Must be called with siglock held.
3536 */
3537void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3538{
3539 struct signal_struct *sig = p->signal;
3540 struct task_cputime cputime;
3541 cputime_t rtime, utime, total;
3542
3543 thread_group_cputime(p, &cputime);
3544
3545 total = cputime_add(cputime.utime, cputime.stime);
3546 rtime = nsecs_to_cputime(cputime.sum_exec_runtime);
3547
3548 if (total) {
Stanislaw Gruszkae75e8632010-09-14 16:35:14 +02003549 u64 temp = rtime;
Hidetoshi Seto0cf55e12009-12-02 17:28:07 +09003550
Stanislaw Gruszkae75e8632010-09-14 16:35:14 +02003551 temp *= cputime.utime;
Hidetoshi Seto0cf55e12009-12-02 17:28:07 +09003552 do_div(temp, total);
3553 utime = (cputime_t)temp;
3554 } else
3555 utime = rtime;
3556
3557 sig->prev_utime = max(sig->prev_utime, utime);
3558 sig->prev_stime = max(sig->prev_stime,
3559 cputime_sub(rtime, sig->prev_utime));
3560
3561 *ut = sig->prev_utime;
3562 *st = sig->prev_stime;
Balbir Singh49048622008-09-05 18:12:23 +02003563}
3564#endif
3565
Balbir Singh49048622008-09-05 18:12:23 +02003566/*
Christoph Lameter7835b982006-12-10 02:20:22 -08003567 * This function gets called by the timer code, with HZ frequency.
3568 * We call it with interrupts disabled.
3569 *
3570 * It also gets called by the fork code, when changing the parent's
3571 * timeslices.
3572 */
3573void scheduler_tick(void)
3574{
Christoph Lameter7835b982006-12-10 02:20:22 -08003575 int cpu = smp_processor_id();
3576 struct rq *rq = cpu_rq(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02003577 struct task_struct *curr = rq->curr;
Peter Zijlstra3e51f332008-05-03 18:29:28 +02003578
3579 sched_clock_tick();
Christoph Lameter7835b982006-12-10 02:20:22 -08003580
Thomas Gleixner05fa7852009-11-17 14:28:38 +01003581 raw_spin_lock(&rq->lock);
Peter Zijlstra3e51f332008-05-03 18:29:28 +02003582 update_rq_clock(rq);
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07003583 update_cpu_load_active(rq);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01003584 curr->sched_class->task_tick(rq, curr, 0);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01003585 raw_spin_unlock(&rq->lock);
Ingo Molnardd41f592007-07-09 18:51:59 +02003586
Peter Zijlstra49f47432009-12-27 11:51:52 +01003587 perf_event_task_tick(curr);
Peter Zijlstrae220d2d2009-05-23 18:28:55 +02003588
Christoph Lametere418e1c2006-12-10 02:20:23 -08003589#ifdef CONFIG_SMP
Ingo Molnardd41f592007-07-09 18:51:59 +02003590 rq->idle_at_tick = idle_cpu(cpu);
3591 trigger_load_balance(rq, cpu);
Christoph Lametere418e1c2006-12-10 02:20:23 -08003592#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593}
3594
Lai Jiangshan132380a2009-04-02 14:18:25 +08003595notrace unsigned long get_parent_ip(unsigned long addr)
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003596{
3597 if (in_lock_functions(addr)) {
3598 addr = CALLER_ADDR2;
3599 if (in_lock_functions(addr))
3600 addr = CALLER_ADDR3;
3601 }
3602 return addr;
3603}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003604
Steven Rostedt7e49fcc2009-01-22 19:01:40 -05003605#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
3606 defined(CONFIG_PREEMPT_TRACER))
3607
Srinivasa Ds43627582008-02-23 15:24:04 -08003608void __kprobes add_preempt_count(int val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003609{
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003610#ifdef CONFIG_DEBUG_PREEMPT
Linus Torvalds1da177e2005-04-16 15:20:36 -07003611 /*
3612 * Underflow?
3613 */
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07003614 if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
3615 return;
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003616#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003617 preempt_count() += val;
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003618#ifdef CONFIG_DEBUG_PREEMPT
Linus Torvalds1da177e2005-04-16 15:20:36 -07003619 /*
3620 * Spinlock count overflowing soon?
3621 */
Miguel Ojeda Sandonis33859f72006-12-10 02:20:38 -08003622 DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
3623 PREEMPT_MASK - 10);
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003624#endif
3625 if (preempt_count() == val)
3626 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627}
3628EXPORT_SYMBOL(add_preempt_count);
3629
Srinivasa Ds43627582008-02-23 15:24:04 -08003630void __kprobes sub_preempt_count(int val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631{
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003632#ifdef CONFIG_DEBUG_PREEMPT
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633 /*
3634 * Underflow?
3635 */
Ingo Molnar01e3eb82009-01-12 13:00:50 +01003636 if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07003637 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638 /*
3639 * Is the spinlock portion underflowing?
3640 */
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07003641 if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
3642 !(preempt_count() & PREEMPT_MASK)))
3643 return;
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003644#endif
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07003645
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003646 if (preempt_count() == val)
3647 trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003648 preempt_count() -= val;
3649}
3650EXPORT_SYMBOL(sub_preempt_count);
3651
3652#endif
3653
3654/*
Ingo Molnardd41f592007-07-09 18:51:59 +02003655 * Print scheduling while atomic bug:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656 */
Ingo Molnardd41f592007-07-09 18:51:59 +02003657static noinline void __schedule_bug(struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658{
Satyam Sharma838225b2007-10-24 18:23:50 +02003659 struct pt_regs *regs = get_irq_regs();
3660
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01003661 printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
3662 prev->comm, prev->pid, preempt_count());
Satyam Sharma838225b2007-10-24 18:23:50 +02003663
Ingo Molnardd41f592007-07-09 18:51:59 +02003664 debug_show_held_locks(prev);
Arjan van de Vene21f5b12008-05-23 09:05:58 -07003665 print_modules();
Ingo Molnardd41f592007-07-09 18:51:59 +02003666 if (irqs_disabled())
3667 print_irqtrace_events(prev);
Satyam Sharma838225b2007-10-24 18:23:50 +02003668
3669 if (regs)
3670 show_regs(regs);
3671 else
3672 dump_stack();
Ingo Molnardd41f592007-07-09 18:51:59 +02003673}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674
Ingo Molnardd41f592007-07-09 18:51:59 +02003675/*
3676 * Various schedule()-time debugging checks and statistics:
3677 */
3678static inline void schedule_debug(struct task_struct *prev)
3679{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680 /*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003681 * Test if we are atomic. Since do_exit() needs to call into
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682 * schedule() atomically, we ignore that path for now.
3683 * Otherwise, whine if we are scheduling when we should not be.
3684 */
Roel Kluin3f33a7c2008-05-13 23:44:11 +02003685 if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
Ingo Molnardd41f592007-07-09 18:51:59 +02003686 __schedule_bug(prev);
3687
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688 profile_hit(SCHED_PROFILING, __builtin_return_address(0));
3689
Ingo Molnar2d723762007-10-15 17:00:12 +02003690 schedstat_inc(this_rq(), sched_count);
Ingo Molnarb8efb562007-10-15 17:00:10 +02003691#ifdef CONFIG_SCHEDSTATS
3692 if (unlikely(prev->lock_depth >= 0)) {
Ingo Molnar2d723762007-10-15 17:00:12 +02003693 schedstat_inc(this_rq(), bkl_count);
3694 schedstat_inc(prev, sched_info.bkl_count);
Ingo Molnarb8efb562007-10-15 17:00:10 +02003695 }
3696#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02003697}
3698
Peter Zijlstra6cecd082009-11-30 13:00:37 +01003699static void put_prev_task(struct rq *rq, struct task_struct *prev)
Mike Galbraithdf1c99d2009-03-10 19:08:11 +01003700{
Mike Galbraitha64692a2010-03-11 17:16:20 +01003701 if (prev->se.on_rq)
3702 update_rq_clock(rq);
3703 rq->skip_clock_update = 0;
Peter Zijlstra6cecd082009-11-30 13:00:37 +01003704 prev->sched_class->put_prev_task(rq, prev);
Mike Galbraithdf1c99d2009-03-10 19:08:11 +01003705}
3706
Ingo Molnardd41f592007-07-09 18:51:59 +02003707/*
3708 * Pick up the highest-prio task:
3709 */
3710static inline struct task_struct *
Wang Chenb67802e2009-03-02 13:55:26 +08003711pick_next_task(struct rq *rq)
Ingo Molnardd41f592007-07-09 18:51:59 +02003712{
Ingo Molnar5522d5d2007-10-15 17:00:12 +02003713 const struct sched_class *class;
Ingo Molnardd41f592007-07-09 18:51:59 +02003714 struct task_struct *p;
3715
3716 /*
3717 * Optimization: we know that if all tasks are in
3718 * the fair class we can call that function directly:
3719 */
3720 if (likely(rq->nr_running == rq->cfs.nr_running)) {
Ingo Molnarfb8d4722007-08-09 11:16:48 +02003721 p = fair_sched_class.pick_next_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02003722 if (likely(p))
3723 return p;
3724 }
3725
3726 class = sched_class_highest;
3727 for ( ; ; ) {
Ingo Molnarfb8d4722007-08-09 11:16:48 +02003728 p = class->pick_next_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02003729 if (p)
3730 return p;
3731 /*
3732 * Will never be NULL as the idle class always
3733 * returns a non-NULL p:
3734 */
3735 class = class->next;
3736 }
3737}
3738
3739/*
3740 * schedule() is the main scheduler function.
3741 */
Peter Zijlstraff743342009-03-13 12:21:26 +01003742asmlinkage void __sched schedule(void)
Ingo Molnardd41f592007-07-09 18:51:59 +02003743{
3744 struct task_struct *prev, *next;
Harvey Harrison67ca7bd2008-02-15 09:56:36 -08003745 unsigned long *switch_count;
Ingo Molnardd41f592007-07-09 18:51:59 +02003746 struct rq *rq;
Peter Zijlstra31656512008-07-18 18:01:23 +02003747 int cpu;
Ingo Molnardd41f592007-07-09 18:51:59 +02003748
Peter Zijlstraff743342009-03-13 12:21:26 +01003749need_resched:
3750 preempt_disable();
Ingo Molnardd41f592007-07-09 18:51:59 +02003751 cpu = smp_processor_id();
3752 rq = cpu_rq(cpu);
Paul E. McKenney25502a62010-04-01 17:37:01 -07003753 rcu_note_context_switch(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02003754 prev = rq->curr;
Ingo Molnardd41f592007-07-09 18:51:59 +02003755
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756 release_kernel_lock(prev);
3757need_resched_nonpreemptible:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003758
Ingo Molnardd41f592007-07-09 18:51:59 +02003759 schedule_debug(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003760
Peter Zijlstra31656512008-07-18 18:01:23 +02003761 if (sched_feat(HRTICK))
Mike Galbraithf333fdc2008-05-12 21:20:55 +02003762 hrtick_clear(rq);
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003763
Thomas Gleixner05fa7852009-11-17 14:28:38 +01003764 raw_spin_lock_irq(&rq->lock);
Ingo Molnar1e819952007-10-15 17:00:13 +02003765 clear_tsk_need_resched(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766
Oleg Nesterov246d86b2010-05-19 14:57:11 +02003767 switch_count = &prev->nivcsw;
Ingo Molnardd41f592007-07-09 18:51:59 +02003768 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
Tejun Heo21aa9af2010-06-08 21:40:37 +02003769 if (unlikely(signal_pending_state(prev->state, prev))) {
Ingo Molnardd41f592007-07-09 18:51:59 +02003770 prev->state = TASK_RUNNING;
Tejun Heo21aa9af2010-06-08 21:40:37 +02003771 } else {
3772 /*
3773 * If a worker is going to sleep, notify and
3774 * ask workqueue whether it wants to wake up a
3775 * task to maintain concurrency. If so, wake
3776 * up the task.
3777 */
3778 if (prev->flags & PF_WQ_WORKER) {
3779 struct task_struct *to_wakeup;
3780
3781 to_wakeup = wq_worker_sleeping(prev, cpu);
3782 if (to_wakeup)
3783 try_to_wake_up_local(to_wakeup);
3784 }
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01003785 deactivate_task(rq, prev, DEQUEUE_SLEEP);
Tejun Heo21aa9af2010-06-08 21:40:37 +02003786 }
Ingo Molnardd41f592007-07-09 18:51:59 +02003787 switch_count = &prev->nvcsw;
3788 }
3789
Gregory Haskins3f029d32009-07-29 11:08:47 -04003790 pre_schedule(rq, prev);
Steven Rostedtf65eda42008-01-25 21:08:07 +01003791
Ingo Molnardd41f592007-07-09 18:51:59 +02003792 if (unlikely(!rq->nr_running))
3793 idle_balance(cpu, rq);
3794
Mike Galbraithdf1c99d2009-03-10 19:08:11 +01003795 put_prev_task(rq, prev);
Wang Chenb67802e2009-03-02 13:55:26 +08003796 next = pick_next_task(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003797
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 if (likely(prev != next)) {
David Simner673a90a2008-04-29 10:08:59 +01003799 sched_info_switch(prev, next);
Peter Zijlstra49f47432009-12-27 11:51:52 +01003800 perf_event_task_sched_out(prev, next);
David Simner673a90a2008-04-29 10:08:59 +01003801
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802 rq->nr_switches++;
3803 rq->curr = next;
3804 ++*switch_count;
3805
Ingo Molnardd41f592007-07-09 18:51:59 +02003806 context_switch(rq, prev, next); /* unlocks the rq */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003807 /*
Oleg Nesterov246d86b2010-05-19 14:57:11 +02003808 * The context switch have flipped the stack from under us
3809 * and restored the local variables which were saved when
3810 * this task called schedule() in the past. prev == current
3811 * is still correct, but it can be moved to another cpu/rq.
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003812 */
3813 cpu = smp_processor_id();
3814 rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815 } else
Thomas Gleixner05fa7852009-11-17 14:28:38 +01003816 raw_spin_unlock_irq(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003817
Gregory Haskins3f029d32009-07-29 11:08:47 -04003818 post_schedule(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819
Oleg Nesterov246d86b2010-05-19 14:57:11 +02003820 if (unlikely(reacquire_kernel_lock(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821 goto need_resched_nonpreemptible;
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003822
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823 preempt_enable_no_resched();
Peter Zijlstraff743342009-03-13 12:21:26 +01003824 if (need_resched())
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825 goto need_resched;
3826}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827EXPORT_SYMBOL(schedule);
3828
Frederic Weisbeckerc08f7822009-12-02 20:49:17 +01003829#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003830/*
3831 * Look out! "owner" is an entirely speculative pointer
3832 * access and not reliable.
3833 */
3834int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner)
3835{
3836 unsigned int cpu;
3837 struct rq *rq;
3838
3839 if (!sched_feat(OWNER_SPIN))
3840 return 0;
3841
3842#ifdef CONFIG_DEBUG_PAGEALLOC
3843 /*
3844 * Need to access the cpu field knowing that
3845 * DEBUG_PAGEALLOC could have unmapped it if
3846 * the mutex owner just released it and exited.
3847 */
3848 if (probe_kernel_address(&owner->cpu, cpu))
Benjamin Herrenschmidt4b402212010-04-16 23:20:00 +02003849 return 0;
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003850#else
3851 cpu = owner->cpu;
3852#endif
3853
3854 /*
3855 * Even if the access succeeded (likely case),
3856 * the cpu field may no longer be valid.
3857 */
3858 if (cpu >= nr_cpumask_bits)
Benjamin Herrenschmidt4b402212010-04-16 23:20:00 +02003859 return 0;
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003860
3861 /*
3862 * We need to validate that we can do a
3863 * get_cpu() and that we have the percpu area.
3864 */
3865 if (!cpu_online(cpu))
Benjamin Herrenschmidt4b402212010-04-16 23:20:00 +02003866 return 0;
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003867
3868 rq = cpu_rq(cpu);
3869
3870 for (;;) {
3871 /*
3872 * Owner changed, break to re-assess state.
3873 */
Tim Chen9d0f4dc2010-08-18 15:00:27 -07003874 if (lock->owner != owner) {
3875 /*
3876 * If the lock has switched to a different owner,
3877 * we likely have heavy contention. Return 0 to quit
3878 * optimistic spinning and not contend further:
3879 */
3880 if (lock->owner)
3881 return 0;
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003882 break;
Tim Chen9d0f4dc2010-08-18 15:00:27 -07003883 }
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003884
3885 /*
3886 * Is that owner really running on that cpu?
3887 */
3888 if (task_thread_info(rq->curr) != owner || need_resched())
3889 return 0;
3890
3891 cpu_relax();
3892 }
Benjamin Herrenschmidt4b402212010-04-16 23:20:00 +02003893
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003894 return 1;
3895}
3896#endif
3897
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898#ifdef CONFIG_PREEMPT
3899/*
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003900 * this is the entry point to schedule() from in-kernel preemption
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003901 * off of preempt_enable. Kernel preemptions off return from interrupt
Linus Torvalds1da177e2005-04-16 15:20:36 -07003902 * occur there and call schedule directly.
3903 */
Steven Rostedtd1f74e22010-06-02 21:52:29 -04003904asmlinkage void __sched notrace preempt_schedule(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905{
3906 struct thread_info *ti = current_thread_info();
Ingo Molnar6478d882008-01-25 21:08:33 +01003907
Linus Torvalds1da177e2005-04-16 15:20:36 -07003908 /*
3909 * If there is a non-zero preempt_count or interrupts are disabled,
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003910 * we do not want to preempt the current task. Just return..
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911 */
Nick Pigginbeed33a2006-10-11 01:21:52 -07003912 if (likely(ti->preempt_count || irqs_disabled()))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003913 return;
3914
Andi Kleen3a5c3592007-10-15 17:00:14 +02003915 do {
Steven Rostedtd1f74e22010-06-02 21:52:29 -04003916 add_preempt_count_notrace(PREEMPT_ACTIVE);
Andi Kleen3a5c3592007-10-15 17:00:14 +02003917 schedule();
Steven Rostedtd1f74e22010-06-02 21:52:29 -04003918 sub_preempt_count_notrace(PREEMPT_ACTIVE);
Andi Kleen3a5c3592007-10-15 17:00:14 +02003919
3920 /*
3921 * Check again in case we missed a preemption opportunity
3922 * between schedule and now.
3923 */
3924 barrier();
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08003925 } while (need_resched());
Linus Torvalds1da177e2005-04-16 15:20:36 -07003926}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003927EXPORT_SYMBOL(preempt_schedule);
3928
3929/*
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003930 * this is the entry point to schedule() from kernel preemption
Linus Torvalds1da177e2005-04-16 15:20:36 -07003931 * off of irq context.
3932 * Note, that this is called and return with irqs disabled. This will
3933 * protect us against recursive calling from irq.
3934 */
3935asmlinkage void __sched preempt_schedule_irq(void)
3936{
3937 struct thread_info *ti = current_thread_info();
Ingo Molnar6478d882008-01-25 21:08:33 +01003938
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003939 /* Catch callers which need to be fixed */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003940 BUG_ON(ti->preempt_count || !irqs_disabled());
3941
Andi Kleen3a5c3592007-10-15 17:00:14 +02003942 do {
3943 add_preempt_count(PREEMPT_ACTIVE);
Andi Kleen3a5c3592007-10-15 17:00:14 +02003944 local_irq_enable();
3945 schedule();
3946 local_irq_disable();
Andi Kleen3a5c3592007-10-15 17:00:14 +02003947 sub_preempt_count(PREEMPT_ACTIVE);
3948
3949 /*
3950 * Check again in case we missed a preemption opportunity
3951 * between schedule and now.
3952 */
3953 barrier();
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08003954 } while (need_resched());
Linus Torvalds1da177e2005-04-16 15:20:36 -07003955}
3956
3957#endif /* CONFIG_PREEMPT */
3958
Peter Zijlstra63859d42009-09-15 19:14:42 +02003959int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07003960 void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003961{
Peter Zijlstra63859d42009-09-15 19:14:42 +02003962 return try_to_wake_up(curr->private, mode, wake_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003963}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003964EXPORT_SYMBOL(default_wake_function);
3965
3966/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003967 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
3968 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
Linus Torvalds1da177e2005-04-16 15:20:36 -07003969 * number) then we wake all the non-exclusive tasks and one exclusive task.
3970 *
3971 * There are circumstances in which we can try to wake a task which has already
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003972 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
Linus Torvalds1da177e2005-04-16 15:20:36 -07003973 * zero in this (rare) case, and we handle it by continuing to scan the queue.
3974 */
Johannes Weiner78ddb082009-04-14 16:53:05 +02003975static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
Peter Zijlstra63859d42009-09-15 19:14:42 +02003976 int nr_exclusive, int wake_flags, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977{
Matthias Kaehlcke2e458742007-10-15 17:00:02 +02003978 wait_queue_t *curr, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979
Matthias Kaehlcke2e458742007-10-15 17:00:02 +02003980 list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
Ingo Molnar48f24c42006-07-03 00:25:40 -07003981 unsigned flags = curr->flags;
3982
Peter Zijlstra63859d42009-09-15 19:14:42 +02003983 if (curr->func(curr, mode, wake_flags, key) &&
Ingo Molnar48f24c42006-07-03 00:25:40 -07003984 (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003985 break;
3986 }
3987}
3988
3989/**
3990 * __wake_up - wake up threads blocked on a waitqueue.
3991 * @q: the waitqueue
3992 * @mode: which threads
3993 * @nr_exclusive: how many wake-one or wake-many threads to wake up
Martin Waitz67be2dd2005-05-01 08:59:26 -07003994 * @key: is directly passed to the wakeup function
David Howells50fa6102009-04-28 15:01:38 +01003995 *
3996 * It may be assumed that this function implies a write memory barrier before
3997 * changing the task state if and only if any tasks are woken up.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08003999void __wake_up(wait_queue_head_t *q, unsigned int mode,
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004000 int nr_exclusive, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001{
4002 unsigned long flags;
4003
4004 spin_lock_irqsave(&q->lock, flags);
4005 __wake_up_common(q, mode, nr_exclusive, 0, key);
4006 spin_unlock_irqrestore(&q->lock, flags);
4007}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004008EXPORT_SYMBOL(__wake_up);
4009
4010/*
4011 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
4012 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08004013void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004014{
4015 __wake_up_common(q, mode, 1, 0, NULL);
4016}
Michal Nazarewicz22c43c82010-05-05 12:53:11 +02004017EXPORT_SYMBOL_GPL(__wake_up_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018
Davide Libenzi4ede8162009-03-31 15:24:20 -07004019void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
4020{
4021 __wake_up_common(q, mode, 1, 0, key);
4022}
4023
Linus Torvalds1da177e2005-04-16 15:20:36 -07004024/**
Davide Libenzi4ede8162009-03-31 15:24:20 -07004025 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004026 * @q: the waitqueue
4027 * @mode: which threads
4028 * @nr_exclusive: how many wake-one or wake-many threads to wake up
Davide Libenzi4ede8162009-03-31 15:24:20 -07004029 * @key: opaque value to be passed to wakeup targets
Linus Torvalds1da177e2005-04-16 15:20:36 -07004030 *
4031 * The sync wakeup differs that the waker knows that it will schedule
4032 * away soon, so while the target thread will be woken up, it will not
4033 * be migrated to another CPU - ie. the two threads are 'synchronized'
4034 * with each other. This can prevent needless bouncing between CPUs.
4035 *
4036 * On UP it can prevent extra preemption.
David Howells50fa6102009-04-28 15:01:38 +01004037 *
4038 * It may be assumed that this function implies a write memory barrier before
4039 * changing the task state if and only if any tasks are woken up.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040 */
Davide Libenzi4ede8162009-03-31 15:24:20 -07004041void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
4042 int nr_exclusive, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004043{
4044 unsigned long flags;
Peter Zijlstra7d478722009-09-14 19:55:44 +02004045 int wake_flags = WF_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004046
4047 if (unlikely(!q))
4048 return;
4049
4050 if (unlikely(!nr_exclusive))
Peter Zijlstra7d478722009-09-14 19:55:44 +02004051 wake_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004052
4053 spin_lock_irqsave(&q->lock, flags);
Peter Zijlstra7d478722009-09-14 19:55:44 +02004054 __wake_up_common(q, mode, nr_exclusive, wake_flags, key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004055 spin_unlock_irqrestore(&q->lock, flags);
4056}
Davide Libenzi4ede8162009-03-31 15:24:20 -07004057EXPORT_SYMBOL_GPL(__wake_up_sync_key);
4058
4059/*
4060 * __wake_up_sync - see __wake_up_sync_key()
4061 */
4062void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
4063{
4064 __wake_up_sync_key(q, mode, nr_exclusive, NULL);
4065}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004066EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
4067
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004068/**
4069 * complete: - signals a single thread waiting on this completion
4070 * @x: holds the state of this particular completion
4071 *
4072 * This will wake up a single thread waiting on this completion. Threads will be
4073 * awakened in the same order in which they were queued.
4074 *
4075 * See also complete_all(), wait_for_completion() and related routines.
David Howells50fa6102009-04-28 15:01:38 +01004076 *
4077 * It may be assumed that this function implies a write memory barrier before
4078 * changing the task state if and only if any tasks are woken up.
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004079 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02004080void complete(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004081{
4082 unsigned long flags;
4083
4084 spin_lock_irqsave(&x->wait.lock, flags);
4085 x->done++;
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05004086 __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004087 spin_unlock_irqrestore(&x->wait.lock, flags);
4088}
4089EXPORT_SYMBOL(complete);
4090
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004091/**
4092 * complete_all: - signals all threads waiting on this completion
4093 * @x: holds the state of this particular completion
4094 *
4095 * This will wake up all threads waiting on this particular completion event.
David Howells50fa6102009-04-28 15:01:38 +01004096 *
4097 * It may be assumed that this function implies a write memory barrier before
4098 * changing the task state if and only if any tasks are woken up.
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004099 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02004100void complete_all(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004101{
4102 unsigned long flags;
4103
4104 spin_lock_irqsave(&x->wait.lock, flags);
4105 x->done += UINT_MAX/2;
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05004106 __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004107 spin_unlock_irqrestore(&x->wait.lock, flags);
4108}
4109EXPORT_SYMBOL(complete_all);
4110
Andi Kleen8cbbe862007-10-15 17:00:14 +02004111static inline long __sched
4112do_wait_for_common(struct completion *x, long timeout, int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004113{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004114 if (!x->done) {
4115 DECLARE_WAITQUEUE(wait, current);
4116
Changli Gaoa93d2f172010-05-07 14:33:26 +08004117 __add_wait_queue_tail_exclusive(&x->wait, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004118 do {
Oleg Nesterov94d3d822008-08-20 16:54:41 -07004119 if (signal_pending_state(state, current)) {
Oleg Nesterovea71a542008-06-20 18:32:20 +04004120 timeout = -ERESTARTSYS;
4121 break;
Andi Kleen8cbbe862007-10-15 17:00:14 +02004122 }
4123 __set_current_state(state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004124 spin_unlock_irq(&x->wait.lock);
Andi Kleen8cbbe862007-10-15 17:00:14 +02004125 timeout = schedule_timeout(timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004126 spin_lock_irq(&x->wait.lock);
Oleg Nesterovea71a542008-06-20 18:32:20 +04004127 } while (!x->done && timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004128 __remove_wait_queue(&x->wait, &wait);
Oleg Nesterovea71a542008-06-20 18:32:20 +04004129 if (!x->done)
4130 return timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004131 }
4132 x->done--;
Oleg Nesterovea71a542008-06-20 18:32:20 +04004133 return timeout ?: 1;
Andi Kleen8cbbe862007-10-15 17:00:14 +02004134}
4135
4136static long __sched
4137wait_for_common(struct completion *x, long timeout, int state)
4138{
4139 might_sleep();
4140
4141 spin_lock_irq(&x->wait.lock);
4142 timeout = do_wait_for_common(x, timeout, state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004143 spin_unlock_irq(&x->wait.lock);
Andi Kleen8cbbe862007-10-15 17:00:14 +02004144 return timeout;
4145}
4146
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004147/**
4148 * wait_for_completion: - waits for completion of a task
4149 * @x: holds the state of this particular completion
4150 *
4151 * This waits to be signaled for completion of a specific task. It is NOT
4152 * interruptible and there is no timeout.
4153 *
4154 * See also similar routines (i.e. wait_for_completion_timeout()) with timeout
4155 * and interrupt capability. Also see complete().
4156 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02004157void __sched wait_for_completion(struct completion *x)
Andi Kleen8cbbe862007-10-15 17:00:14 +02004158{
4159 wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004160}
4161EXPORT_SYMBOL(wait_for_completion);
4162
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004163/**
4164 * wait_for_completion_timeout: - waits for completion of a task (w/timeout)
4165 * @x: holds the state of this particular completion
4166 * @timeout: timeout value in jiffies
4167 *
4168 * This waits for either a completion of a specific task to be signaled or for a
4169 * specified timeout to expire. The timeout is in jiffies. It is not
4170 * interruptible.
4171 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02004172unsigned long __sched
Linus Torvalds1da177e2005-04-16 15:20:36 -07004173wait_for_completion_timeout(struct completion *x, unsigned long timeout)
4174{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004175 return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004176}
4177EXPORT_SYMBOL(wait_for_completion_timeout);
4178
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004179/**
4180 * wait_for_completion_interruptible: - waits for completion of a task (w/intr)
4181 * @x: holds the state of this particular completion
4182 *
4183 * This waits for completion of a specific task to be signaled. It is
4184 * interruptible.
4185 */
Andi Kleen8cbbe862007-10-15 17:00:14 +02004186int __sched wait_for_completion_interruptible(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004187{
Andi Kleen51e97992007-10-18 21:32:55 +02004188 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
4189 if (t == -ERESTARTSYS)
4190 return t;
4191 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004192}
4193EXPORT_SYMBOL(wait_for_completion_interruptible);
4194
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004195/**
4196 * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr))
4197 * @x: holds the state of this particular completion
4198 * @timeout: timeout value in jiffies
4199 *
4200 * This waits for either a completion of a specific task to be signaled or for a
4201 * specified timeout to expire. It is interruptible. The timeout is in jiffies.
4202 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02004203unsigned long __sched
Linus Torvalds1da177e2005-04-16 15:20:36 -07004204wait_for_completion_interruptible_timeout(struct completion *x,
4205 unsigned long timeout)
4206{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004207 return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004208}
4209EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
4210
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004211/**
4212 * wait_for_completion_killable: - waits for completion of a task (killable)
4213 * @x: holds the state of this particular completion
4214 *
4215 * This waits to be signaled for completion of a specific task. It can be
4216 * interrupted by a kill signal.
4217 */
Matthew Wilcox009e5772007-12-06 12:29:54 -05004218int __sched wait_for_completion_killable(struct completion *x)
4219{
4220 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
4221 if (t == -ERESTARTSYS)
4222 return t;
4223 return 0;
4224}
4225EXPORT_SYMBOL(wait_for_completion_killable);
4226
Dave Chinnerbe4de352008-08-15 00:40:44 -07004227/**
Sage Weil0aa12fb2010-05-29 09:12:30 -07004228 * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable))
4229 * @x: holds the state of this particular completion
4230 * @timeout: timeout value in jiffies
4231 *
4232 * This waits for either a completion of a specific task to be
4233 * signaled or for a specified timeout to expire. It can be
4234 * interrupted by a kill signal. The timeout is in jiffies.
4235 */
4236unsigned long __sched
4237wait_for_completion_killable_timeout(struct completion *x,
4238 unsigned long timeout)
4239{
4240 return wait_for_common(x, timeout, TASK_KILLABLE);
4241}
4242EXPORT_SYMBOL(wait_for_completion_killable_timeout);
4243
4244/**
Dave Chinnerbe4de352008-08-15 00:40:44 -07004245 * try_wait_for_completion - try to decrement a completion without blocking
4246 * @x: completion structure
4247 *
4248 * Returns: 0 if a decrement cannot be done without blocking
4249 * 1 if a decrement succeeded.
4250 *
4251 * If a completion is being used as a counting completion,
4252 * attempt to decrement the counter without blocking. This
4253 * enables us to avoid waiting if the resource the completion
4254 * is protecting is not available.
4255 */
4256bool try_wait_for_completion(struct completion *x)
4257{
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004258 unsigned long flags;
Dave Chinnerbe4de352008-08-15 00:40:44 -07004259 int ret = 1;
4260
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004261 spin_lock_irqsave(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07004262 if (!x->done)
4263 ret = 0;
4264 else
4265 x->done--;
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004266 spin_unlock_irqrestore(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07004267 return ret;
4268}
4269EXPORT_SYMBOL(try_wait_for_completion);
4270
4271/**
4272 * completion_done - Test to see if a completion has any waiters
4273 * @x: completion structure
4274 *
4275 * Returns: 0 if there are waiters (wait_for_completion() in progress)
4276 * 1 if there are no waiters.
4277 *
4278 */
4279bool completion_done(struct completion *x)
4280{
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004281 unsigned long flags;
Dave Chinnerbe4de352008-08-15 00:40:44 -07004282 int ret = 1;
4283
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004284 spin_lock_irqsave(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07004285 if (!x->done)
4286 ret = 0;
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004287 spin_unlock_irqrestore(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07004288 return ret;
4289}
4290EXPORT_SYMBOL(completion_done);
4291
Andi Kleen8cbbe862007-10-15 17:00:14 +02004292static long __sched
4293sleep_on_common(wait_queue_head_t *q, int state, long timeout)
Ingo Molnar0fec1712007-07-09 18:52:01 +02004294{
4295 unsigned long flags;
4296 wait_queue_t wait;
4297
4298 init_waitqueue_entry(&wait, current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004299
Andi Kleen8cbbe862007-10-15 17:00:14 +02004300 __set_current_state(state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004301
Andi Kleen8cbbe862007-10-15 17:00:14 +02004302 spin_lock_irqsave(&q->lock, flags);
4303 __add_wait_queue(q, &wait);
4304 spin_unlock(&q->lock);
4305 timeout = schedule_timeout(timeout);
4306 spin_lock_irq(&q->lock);
4307 __remove_wait_queue(q, &wait);
4308 spin_unlock_irqrestore(&q->lock, flags);
4309
4310 return timeout;
4311}
4312
4313void __sched interruptible_sleep_on(wait_queue_head_t *q)
4314{
4315 sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004316}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004317EXPORT_SYMBOL(interruptible_sleep_on);
4318
Ingo Molnar0fec1712007-07-09 18:52:01 +02004319long __sched
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004320interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004321{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004322 return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004323}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004324EXPORT_SYMBOL(interruptible_sleep_on_timeout);
4325
Ingo Molnar0fec1712007-07-09 18:52:01 +02004326void __sched sleep_on(wait_queue_head_t *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004327{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004328 sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004329}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004330EXPORT_SYMBOL(sleep_on);
4331
Ingo Molnar0fec1712007-07-09 18:52:01 +02004332long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004333{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004334 return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004335}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004336EXPORT_SYMBOL(sleep_on_timeout);
4337
Ingo Molnarb29739f2006-06-27 02:54:51 -07004338#ifdef CONFIG_RT_MUTEXES
4339
4340/*
4341 * rt_mutex_setprio - set the current priority of a task
4342 * @p: task
4343 * @prio: prio value (kernel-internal form)
4344 *
4345 * This function changes the 'effective' priority of a task. It does
4346 * not touch ->normal_prio like __setscheduler().
4347 *
4348 * Used by the rt_mutex code to implement priority inheritance logic.
4349 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004350void rt_mutex_setprio(struct task_struct *p, int prio)
Ingo Molnarb29739f2006-06-27 02:54:51 -07004351{
4352 unsigned long flags;
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004353 int oldprio, on_rq, running;
Ingo Molnar70b97a72006-07-03 00:25:42 -07004354 struct rq *rq;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01004355 const struct sched_class *prev_class;
Ingo Molnarb29739f2006-06-27 02:54:51 -07004356
4357 BUG_ON(prio < 0 || prio > MAX_PRIO);
4358
4359 rq = task_rq_lock(p, &flags);
4360
Andrew Mortond5f9f942007-05-08 20:27:06 -07004361 oldprio = p->prio;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01004362 prev_class = p->sched_class;
Ingo Molnardd41f592007-07-09 18:51:59 +02004363 on_rq = p->se.on_rq;
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01004364 running = task_current(rq, p);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004365 if (on_rq)
Ingo Molnar69be72c2007-08-09 11:16:49 +02004366 dequeue_task(rq, p, 0);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004367 if (running)
4368 p->sched_class->put_prev_task(rq, p);
Ingo Molnardd41f592007-07-09 18:51:59 +02004369
4370 if (rt_prio(prio))
4371 p->sched_class = &rt_sched_class;
4372 else
4373 p->sched_class = &fair_sched_class;
4374
Ingo Molnarb29739f2006-06-27 02:54:51 -07004375 p->prio = prio;
4376
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004377 if (running)
4378 p->sched_class->set_curr_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02004379 if (on_rq) {
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01004380 enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0);
Steven Rostedtcb469842008-01-25 21:08:22 +01004381
4382 check_class_changed(rq, p, prev_class, oldprio, running);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004383 }
4384 task_rq_unlock(rq, &flags);
4385}
4386
4387#endif
4388
Ingo Molnar36c8b582006-07-03 00:25:41 -07004389void set_user_nice(struct task_struct *p, long nice)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004390{
Ingo Molnardd41f592007-07-09 18:51:59 +02004391 int old_prio, delta, on_rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004392 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07004393 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004394
4395 if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
4396 return;
4397 /*
4398 * We have to be careful, if called from sys_setpriority(),
4399 * the task might be in the middle of scheduling on another CPU.
4400 */
4401 rq = task_rq_lock(p, &flags);
4402 /*
4403 * The RT priorities are set via sched_setscheduler(), but we still
4404 * allow the 'normal' nice value to be set - but as expected
4405 * it wont have any effect on scheduling until the task is
Ingo Molnardd41f592007-07-09 18:51:59 +02004406 * SCHED_FIFO/SCHED_RR:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004407 */
Ingo Molnare05606d2007-07-09 18:51:59 +02004408 if (task_has_rt_policy(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004409 p->static_prio = NICE_TO_PRIO(nice);
4410 goto out_unlock;
4411 }
Ingo Molnardd41f592007-07-09 18:51:59 +02004412 on_rq = p->se.on_rq;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02004413 if (on_rq)
Ingo Molnar69be72c2007-08-09 11:16:49 +02004414 dequeue_task(rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004415
Linus Torvalds1da177e2005-04-16 15:20:36 -07004416 p->static_prio = NICE_TO_PRIO(nice);
Peter Williams2dd73a42006-06-27 02:54:34 -07004417 set_load_weight(p);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004418 old_prio = p->prio;
4419 p->prio = effective_prio(p);
4420 delta = p->prio - old_prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004421
Ingo Molnardd41f592007-07-09 18:51:59 +02004422 if (on_rq) {
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01004423 enqueue_task(rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004424 /*
Andrew Mortond5f9f942007-05-08 20:27:06 -07004425 * If the task increased its priority or is running and
4426 * lowered its priority, then reschedule its CPU:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004427 */
Andrew Mortond5f9f942007-05-08 20:27:06 -07004428 if (delta < 0 || (delta > 0 && task_running(rq, p)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004429 resched_task(rq->curr);
4430 }
4431out_unlock:
4432 task_rq_unlock(rq, &flags);
4433}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004434EXPORT_SYMBOL(set_user_nice);
4435
Matt Mackalle43379f2005-05-01 08:59:00 -07004436/*
4437 * can_nice - check if a task can reduce its nice value
4438 * @p: task
4439 * @nice: nice value
4440 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004441int can_nice(const struct task_struct *p, const int nice)
Matt Mackalle43379f2005-05-01 08:59:00 -07004442{
Matt Mackall024f4742005-08-18 11:24:19 -07004443 /* convert nice value [19,-20] to rlimit style value [1,40] */
4444 int nice_rlim = 20 - nice;
Ingo Molnar48f24c42006-07-03 00:25:40 -07004445
Jiri Slaby78d7d402010-03-05 13:42:54 -08004446 return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
Matt Mackalle43379f2005-05-01 08:59:00 -07004447 capable(CAP_SYS_NICE));
4448}
4449
Linus Torvalds1da177e2005-04-16 15:20:36 -07004450#ifdef __ARCH_WANT_SYS_NICE
4451
4452/*
4453 * sys_nice - change the priority of the current process.
4454 * @increment: priority increment
4455 *
4456 * sys_setpriority is a more generic, but much slower function that
4457 * does similar things.
4458 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004459SYSCALL_DEFINE1(nice, int, increment)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004460{
Ingo Molnar48f24c42006-07-03 00:25:40 -07004461 long nice, retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004462
4463 /*
4464 * Setpriority might change our priority at the same moment.
4465 * We don't have to worry. Conceptually one call occurs first
4466 * and we have a single winner.
4467 */
Matt Mackalle43379f2005-05-01 08:59:00 -07004468 if (increment < -40)
4469 increment = -40;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004470 if (increment > 40)
4471 increment = 40;
4472
Américo Wang2b8f8362009-02-16 18:54:21 +08004473 nice = TASK_NICE(current) + increment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004474 if (nice < -20)
4475 nice = -20;
4476 if (nice > 19)
4477 nice = 19;
4478
Matt Mackalle43379f2005-05-01 08:59:00 -07004479 if (increment < 0 && !can_nice(current, nice))
4480 return -EPERM;
4481
Linus Torvalds1da177e2005-04-16 15:20:36 -07004482 retval = security_task_setnice(current, nice);
4483 if (retval)
4484 return retval;
4485
4486 set_user_nice(current, nice);
4487 return 0;
4488}
4489
4490#endif
4491
4492/**
4493 * task_prio - return the priority value of a given task.
4494 * @p: the task in question.
4495 *
4496 * This is the priority value as seen by users in /proc.
4497 * RT tasks are offset by -200. Normal tasks are centered
4498 * around 0, value goes from -16 to +15.
4499 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004500int task_prio(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004501{
4502 return p->prio - MAX_RT_PRIO;
4503}
4504
4505/**
4506 * task_nice - return the nice value of a given task.
4507 * @p: the task in question.
4508 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004509int task_nice(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004510{
4511 return TASK_NICE(p);
4512}
Pavel Roskin150d8be2008-03-05 16:56:37 -05004513EXPORT_SYMBOL(task_nice);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004514
4515/**
4516 * idle_cpu - is a given cpu idle currently?
4517 * @cpu: the processor in question.
4518 */
4519int idle_cpu(int cpu)
4520{
4521 return cpu_curr(cpu) == cpu_rq(cpu)->idle;
4522}
4523
Linus Torvalds1da177e2005-04-16 15:20:36 -07004524/**
4525 * idle_task - return the idle task for a given cpu.
4526 * @cpu: the processor in question.
4527 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004528struct task_struct *idle_task(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004529{
4530 return cpu_rq(cpu)->idle;
4531}
4532
4533/**
4534 * find_process_by_pid - find a process with a matching PID value.
4535 * @pid: the pid in question.
4536 */
Alexey Dobriyana9957442007-10-15 17:00:13 +02004537static struct task_struct *find_process_by_pid(pid_t pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004538{
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07004539 return pid ? find_task_by_vpid(pid) : current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004540}
4541
4542/* Actually do priority change: must hold rq lock. */
Ingo Molnardd41f592007-07-09 18:51:59 +02004543static void
4544__setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004545{
Ingo Molnardd41f592007-07-09 18:51:59 +02004546 BUG_ON(p->se.on_rq);
Ingo Molnar48f24c42006-07-03 00:25:40 -07004547
Linus Torvalds1da177e2005-04-16 15:20:36 -07004548 p->policy = policy;
4549 p->rt_priority = prio;
Ingo Molnarb29739f2006-06-27 02:54:51 -07004550 p->normal_prio = normal_prio(p);
4551 /* we are holding p->pi_lock already */
4552 p->prio = rt_mutex_getprio(p);
Peter Zijlstraffd44db2009-11-10 20:12:01 +01004553 if (rt_prio(p->prio))
4554 p->sched_class = &rt_sched_class;
4555 else
4556 p->sched_class = &fair_sched_class;
Peter Williams2dd73a42006-06-27 02:54:34 -07004557 set_load_weight(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004558}
4559
David Howellsc69e8d92008-11-14 10:39:19 +11004560/*
4561 * check the target process has a UID that matches the current process's
4562 */
4563static bool check_same_owner(struct task_struct *p)
4564{
4565 const struct cred *cred = current_cred(), *pcred;
4566 bool match;
4567
4568 rcu_read_lock();
4569 pcred = __task_cred(p);
4570 match = (cred->euid == pcred->euid ||
4571 cred->euid == pcred->uid);
4572 rcu_read_unlock();
4573 return match;
4574}
4575
Rusty Russell961ccdd2008-06-23 13:55:38 +10004576static int __sched_setscheduler(struct task_struct *p, int policy,
4577 struct sched_param *param, bool user)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578{
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004579 int retval, oldprio, oldpolicy = -1, on_rq, running;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004580 unsigned long flags;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01004581 const struct sched_class *prev_class;
Ingo Molnar70b97a72006-07-03 00:25:42 -07004582 struct rq *rq;
Lennart Poetteringca94c442009-06-15 17:17:47 +02004583 int reset_on_fork;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004584
Steven Rostedt66e53932006-06-27 02:54:44 -07004585 /* may grab non-irq protected spin_locks */
4586 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07004587recheck:
4588 /* double check policy once rq lock held */
Lennart Poetteringca94c442009-06-15 17:17:47 +02004589 if (policy < 0) {
4590 reset_on_fork = p->sched_reset_on_fork;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004591 policy = oldpolicy = p->policy;
Lennart Poetteringca94c442009-06-15 17:17:47 +02004592 } else {
4593 reset_on_fork = !!(policy & SCHED_RESET_ON_FORK);
4594 policy &= ~SCHED_RESET_ON_FORK;
4595
4596 if (policy != SCHED_FIFO && policy != SCHED_RR &&
4597 policy != SCHED_NORMAL && policy != SCHED_BATCH &&
4598 policy != SCHED_IDLE)
4599 return -EINVAL;
4600 }
4601
Linus Torvalds1da177e2005-04-16 15:20:36 -07004602 /*
4603 * Valid priorities for SCHED_FIFO and SCHED_RR are
Ingo Molnardd41f592007-07-09 18:51:59 +02004604 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
4605 * SCHED_BATCH and SCHED_IDLE is 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004606 */
4607 if (param->sched_priority < 0 ||
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004608 (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
Steven Rostedtd46523e2005-07-25 16:28:39 -04004609 (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004610 return -EINVAL;
Ingo Molnare05606d2007-07-09 18:51:59 +02004611 if (rt_policy(policy) != (param->sched_priority != 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004612 return -EINVAL;
4613
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004614 /*
4615 * Allow unprivileged RT tasks to decrease priority:
4616 */
Rusty Russell961ccdd2008-06-23 13:55:38 +10004617 if (user && !capable(CAP_SYS_NICE)) {
Ingo Molnare05606d2007-07-09 18:51:59 +02004618 if (rt_policy(policy)) {
Oleg Nesterova44702e82010-06-11 01:09:44 +02004619 unsigned long rlim_rtprio =
4620 task_rlimit(p, RLIMIT_RTPRIO);
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004621
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07004622 /* can't set/change the rt policy */
4623 if (policy != p->policy && !rlim_rtprio)
4624 return -EPERM;
4625
4626 /* can't increase priority */
4627 if (param->sched_priority > p->rt_priority &&
4628 param->sched_priority > rlim_rtprio)
4629 return -EPERM;
4630 }
Ingo Molnardd41f592007-07-09 18:51:59 +02004631 /*
4632 * Like positive nice levels, dont allow tasks to
4633 * move out of SCHED_IDLE either:
4634 */
4635 if (p->policy == SCHED_IDLE && policy != SCHED_IDLE)
4636 return -EPERM;
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07004637
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004638 /* can't change other user's priorities */
David Howellsc69e8d92008-11-14 10:39:19 +11004639 if (!check_same_owner(p))
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004640 return -EPERM;
Lennart Poetteringca94c442009-06-15 17:17:47 +02004641
4642 /* Normal users shall not reset the sched_reset_on_fork flag */
4643 if (p->sched_reset_on_fork && !reset_on_fork)
4644 return -EPERM;
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004645 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004646
Jeremy Fitzhardinge725aad22008-08-03 09:33:03 -07004647 if (user) {
Jeremy Fitzhardinge725aad22008-08-03 09:33:03 -07004648 retval = security_task_setscheduler(p, policy, param);
4649 if (retval)
4650 return retval;
4651 }
4652
Linus Torvalds1da177e2005-04-16 15:20:36 -07004653 /*
Ingo Molnarb29739f2006-06-27 02:54:51 -07004654 * make sure no PI-waiters arrive (or leave) while we are
4655 * changing the priority of the task:
4656 */
Thomas Gleixner1d615482009-11-17 14:54:03 +01004657 raw_spin_lock_irqsave(&p->pi_lock, flags);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004658 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004659 * To be able to change p->policy safely, the apropriate
4660 * runqueue lock must be held.
4661 */
Ingo Molnarb29739f2006-06-27 02:54:51 -07004662 rq = __task_rq_lock(p);
Peter Zijlstradc61b1d2010-06-08 11:40:42 +02004663
4664#ifdef CONFIG_RT_GROUP_SCHED
4665 if (user) {
4666 /*
4667 * Do not allow realtime tasks into groups that have no runtime
4668 * assigned.
4669 */
4670 if (rt_bandwidth_enabled() && rt_policy(policy) &&
4671 task_group(p)->rt_bandwidth.rt_runtime == 0) {
4672 __task_rq_unlock(rq);
4673 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4674 return -EPERM;
4675 }
4676 }
4677#endif
4678
Linus Torvalds1da177e2005-04-16 15:20:36 -07004679 /* recheck policy now with rq lock held */
4680 if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
4681 policy = oldpolicy = -1;
Ingo Molnarb29739f2006-06-27 02:54:51 -07004682 __task_rq_unlock(rq);
Thomas Gleixner1d615482009-11-17 14:54:03 +01004683 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004684 goto recheck;
4685 }
Ingo Molnardd41f592007-07-09 18:51:59 +02004686 on_rq = p->se.on_rq;
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01004687 running = task_current(rq, p);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004688 if (on_rq)
Ingo Molnar2e1cb742007-08-09 11:16:49 +02004689 deactivate_task(rq, p, 0);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004690 if (running)
4691 p->sched_class->put_prev_task(rq, p);
Dmitry Adamushkof6b532052007-10-15 17:00:08 +02004692
Lennart Poetteringca94c442009-06-15 17:17:47 +02004693 p->sched_reset_on_fork = reset_on_fork;
4694
Linus Torvalds1da177e2005-04-16 15:20:36 -07004695 oldprio = p->prio;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01004696 prev_class = p->sched_class;
Ingo Molnardd41f592007-07-09 18:51:59 +02004697 __setscheduler(rq, p, policy, param->sched_priority);
Dmitry Adamushkof6b532052007-10-15 17:00:08 +02004698
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004699 if (running)
4700 p->sched_class->set_curr_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02004701 if (on_rq) {
4702 activate_task(rq, p, 0);
Steven Rostedtcb469842008-01-25 21:08:22 +01004703
4704 check_class_changed(rq, p, prev_class, oldprio, running);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004705 }
Ingo Molnarb29739f2006-06-27 02:54:51 -07004706 __task_rq_unlock(rq);
Thomas Gleixner1d615482009-11-17 14:54:03 +01004707 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004708
Thomas Gleixner95e02ca2006-06-27 02:55:02 -07004709 rt_mutex_adjust_pi(p);
4710
Linus Torvalds1da177e2005-04-16 15:20:36 -07004711 return 0;
4712}
Rusty Russell961ccdd2008-06-23 13:55:38 +10004713
4714/**
4715 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
4716 * @p: the task in question.
4717 * @policy: new policy.
4718 * @param: structure containing the new RT priority.
4719 *
4720 * NOTE that the task may be already dead.
4721 */
4722int sched_setscheduler(struct task_struct *p, int policy,
4723 struct sched_param *param)
4724{
4725 return __sched_setscheduler(p, policy, param, true);
4726}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004727EXPORT_SYMBOL_GPL(sched_setscheduler);
4728
Rusty Russell961ccdd2008-06-23 13:55:38 +10004729/**
4730 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
4731 * @p: the task in question.
4732 * @policy: new policy.
4733 * @param: structure containing the new RT priority.
4734 *
4735 * Just like sched_setscheduler, only don't bother checking if the
4736 * current context has permission. For example, this is needed in
4737 * stop_machine(): we create temporary high priority worker threads,
4738 * but our caller might not have that capability.
4739 */
4740int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4741 struct sched_param *param)
4742{
4743 return __sched_setscheduler(p, policy, param, false);
4744}
4745
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004746static int
4747do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004748{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004749 struct sched_param lparam;
4750 struct task_struct *p;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004751 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004752
4753 if (!param || pid < 0)
4754 return -EINVAL;
4755 if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
4756 return -EFAULT;
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004757
4758 rcu_read_lock();
4759 retval = -ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004760 p = find_process_by_pid(pid);
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004761 if (p != NULL)
4762 retval = sched_setscheduler(p, policy, &lparam);
4763 rcu_read_unlock();
Ingo Molnar36c8b582006-07-03 00:25:41 -07004764
Linus Torvalds1da177e2005-04-16 15:20:36 -07004765 return retval;
4766}
4767
4768/**
4769 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
4770 * @pid: the pid in question.
4771 * @policy: new policy.
4772 * @param: structure containing the new RT priority.
4773 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004774SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
4775 struct sched_param __user *, param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004776{
Jason Baronc21761f2006-01-18 17:43:03 -08004777 /* negative values for policy are not valid */
4778 if (policy < 0)
4779 return -EINVAL;
4780
Linus Torvalds1da177e2005-04-16 15:20:36 -07004781 return do_sched_setscheduler(pid, policy, param);
4782}
4783
4784/**
4785 * sys_sched_setparam - set/change the RT priority of a thread
4786 * @pid: the pid in question.
4787 * @param: structure containing the new RT priority.
4788 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004789SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004790{
4791 return do_sched_setscheduler(pid, -1, param);
4792}
4793
4794/**
4795 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
4796 * @pid: the pid in question.
4797 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004798SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004799{
Ingo Molnar36c8b582006-07-03 00:25:41 -07004800 struct task_struct *p;
Andi Kleen3a5c3592007-10-15 17:00:14 +02004801 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004802
4803 if (pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02004804 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004805
4806 retval = -ESRCH;
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004807 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004808 p = find_process_by_pid(pid);
4809 if (p) {
4810 retval = security_task_getscheduler(p);
4811 if (!retval)
Lennart Poetteringca94c442009-06-15 17:17:47 +02004812 retval = p->policy
4813 | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004814 }
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004815 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004816 return retval;
4817}
4818
4819/**
Lennart Poetteringca94c442009-06-15 17:17:47 +02004820 * sys_sched_getparam - get the RT priority of a thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07004821 * @pid: the pid in question.
4822 * @param: structure containing the RT priority.
4823 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004824SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004825{
4826 struct sched_param lp;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004827 struct task_struct *p;
Andi Kleen3a5c3592007-10-15 17:00:14 +02004828 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004829
4830 if (!param || pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02004831 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004832
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004833 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004834 p = find_process_by_pid(pid);
4835 retval = -ESRCH;
4836 if (!p)
4837 goto out_unlock;
4838
4839 retval = security_task_getscheduler(p);
4840 if (retval)
4841 goto out_unlock;
4842
4843 lp.sched_priority = p->rt_priority;
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004844 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004845
4846 /*
4847 * This one might sleep, we cannot do it with a spinlock held ...
4848 */
4849 retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
4850
Linus Torvalds1da177e2005-04-16 15:20:36 -07004851 return retval;
4852
4853out_unlock:
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004854 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004855 return retval;
4856}
4857
Rusty Russell96f874e22008-11-25 02:35:14 +10304858long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004859{
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304860 cpumask_var_t cpus_allowed, new_mask;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004861 struct task_struct *p;
4862 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004863
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004864 get_online_cpus();
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004865 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004866
4867 p = find_process_by_pid(pid);
4868 if (!p) {
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004869 rcu_read_unlock();
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004870 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004871 return -ESRCH;
4872 }
4873
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004874 /* Prevent p going away */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004875 get_task_struct(p);
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004876 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004877
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304878 if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
4879 retval = -ENOMEM;
4880 goto out_put_task;
4881 }
4882 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
4883 retval = -ENOMEM;
4884 goto out_free_cpus_allowed;
4885 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004886 retval = -EPERM;
David Howellsc69e8d92008-11-14 10:39:19 +11004887 if (!check_same_owner(p) && !capable(CAP_SYS_NICE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004888 goto out_unlock;
4889
David Quigleye7834f82006-06-23 02:03:59 -07004890 retval = security_task_setscheduler(p, 0, NULL);
4891 if (retval)
4892 goto out_unlock;
4893
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304894 cpuset_cpus_allowed(p, cpus_allowed);
4895 cpumask_and(new_mask, in_mask, cpus_allowed);
Paul Menage8707d8b2007-10-18 23:40:22 -07004896 again:
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304897 retval = set_cpus_allowed_ptr(p, new_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004898
Paul Menage8707d8b2007-10-18 23:40:22 -07004899 if (!retval) {
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304900 cpuset_cpus_allowed(p, cpus_allowed);
4901 if (!cpumask_subset(new_mask, cpus_allowed)) {
Paul Menage8707d8b2007-10-18 23:40:22 -07004902 /*
4903 * We must have raced with a concurrent cpuset
4904 * update. Just reset the cpus_allowed to the
4905 * cpuset's cpus_allowed
4906 */
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304907 cpumask_copy(new_mask, cpus_allowed);
Paul Menage8707d8b2007-10-18 23:40:22 -07004908 goto again;
4909 }
4910 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004911out_unlock:
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304912 free_cpumask_var(new_mask);
4913out_free_cpus_allowed:
4914 free_cpumask_var(cpus_allowed);
4915out_put_task:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004916 put_task_struct(p);
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004917 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004918 return retval;
4919}
4920
4921static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
Rusty Russell96f874e22008-11-25 02:35:14 +10304922 struct cpumask *new_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004923{
Rusty Russell96f874e22008-11-25 02:35:14 +10304924 if (len < cpumask_size())
4925 cpumask_clear(new_mask);
4926 else if (len > cpumask_size())
4927 len = cpumask_size();
4928
Linus Torvalds1da177e2005-04-16 15:20:36 -07004929 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
4930}
4931
4932/**
4933 * sys_sched_setaffinity - set the cpu affinity of a process
4934 * @pid: pid of the process
4935 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4936 * @user_mask_ptr: user-space pointer to the new cpu mask
4937 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004938SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
4939 unsigned long __user *, user_mask_ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004940{
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304941 cpumask_var_t new_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004942 int retval;
4943
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304944 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
4945 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004946
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304947 retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
4948 if (retval == 0)
4949 retval = sched_setaffinity(pid, new_mask);
4950 free_cpumask_var(new_mask);
4951 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004952}
4953
Rusty Russell96f874e22008-11-25 02:35:14 +10304954long sched_getaffinity(pid_t pid, struct cpumask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004955{
Ingo Molnar36c8b582006-07-03 00:25:41 -07004956 struct task_struct *p;
Thomas Gleixner31605682009-12-08 20:24:16 +00004957 unsigned long flags;
4958 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004959 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004960
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004961 get_online_cpus();
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004962 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004963
4964 retval = -ESRCH;
4965 p = find_process_by_pid(pid);
4966 if (!p)
4967 goto out_unlock;
4968
David Quigleye7834f82006-06-23 02:03:59 -07004969 retval = security_task_getscheduler(p);
4970 if (retval)
4971 goto out_unlock;
4972
Thomas Gleixner31605682009-12-08 20:24:16 +00004973 rq = task_rq_lock(p, &flags);
Rusty Russell96f874e22008-11-25 02:35:14 +10304974 cpumask_and(mask, &p->cpus_allowed, cpu_online_mask);
Thomas Gleixner31605682009-12-08 20:24:16 +00004975 task_rq_unlock(rq, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004976
4977out_unlock:
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004978 rcu_read_unlock();
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004979 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004980
Ulrich Drepper9531b622007-08-09 11:16:46 +02004981 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004982}
4983
4984/**
4985 * sys_sched_getaffinity - get the cpu affinity of a process
4986 * @pid: pid of the process
4987 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4988 * @user_mask_ptr: user-space pointer to hold the current cpu mask
4989 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004990SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
4991 unsigned long __user *, user_mask_ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004992{
4993 int ret;
Rusty Russellf17c8602008-11-25 02:35:11 +10304994 cpumask_var_t mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004995
Anton Blanchard84fba5e2010-04-06 17:02:19 +10004996 if ((len * BITS_PER_BYTE) < nr_cpu_ids)
KOSAKI Motohirocd3d8032010-03-12 16:15:36 +09004997 return -EINVAL;
4998 if (len & (sizeof(unsigned long)-1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004999 return -EINVAL;
5000
Rusty Russellf17c8602008-11-25 02:35:11 +10305001 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
5002 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005003
Rusty Russellf17c8602008-11-25 02:35:11 +10305004 ret = sched_getaffinity(pid, mask);
5005 if (ret == 0) {
KOSAKI Motohiro8bc037f2010-03-17 09:36:58 +09005006 size_t retlen = min_t(size_t, len, cpumask_size());
KOSAKI Motohirocd3d8032010-03-12 16:15:36 +09005007
5008 if (copy_to_user(user_mask_ptr, mask, retlen))
Rusty Russellf17c8602008-11-25 02:35:11 +10305009 ret = -EFAULT;
5010 else
KOSAKI Motohirocd3d8032010-03-12 16:15:36 +09005011 ret = retlen;
Rusty Russellf17c8602008-11-25 02:35:11 +10305012 }
5013 free_cpumask_var(mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005014
Rusty Russellf17c8602008-11-25 02:35:11 +10305015 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005016}
5017
5018/**
5019 * sys_sched_yield - yield the current processor to other threads.
5020 *
Ingo Molnardd41f592007-07-09 18:51:59 +02005021 * This function yields the current CPU to other tasks. If there are no
5022 * other threads running on this CPU then this function will return.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005023 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01005024SYSCALL_DEFINE0(sched_yield)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005025{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005026 struct rq *rq = this_rq_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005027
Ingo Molnar2d723762007-10-15 17:00:12 +02005028 schedstat_inc(rq, yld_count);
Dmitry Adamushko4530d7a2007-10-15 17:00:08 +02005029 current->sched_class->yield_task(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005030
5031 /*
5032 * Since we are going to call schedule() anyway, there's
5033 * no need to preempt or enable interrupts:
5034 */
5035 __release(rq->lock);
Ingo Molnar8a25d5d2006-07-03 00:24:54 -07005036 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
Thomas Gleixner9828ea92009-12-03 20:55:53 +01005037 do_raw_spin_unlock(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005038 preempt_enable_no_resched();
5039
5040 schedule();
5041
5042 return 0;
5043}
5044
Peter Zijlstrad86ee482009-07-10 14:57:57 +02005045static inline int should_resched(void)
5046{
5047 return need_resched() && !(preempt_count() & PREEMPT_ACTIVE);
5048}
5049
Andrew Mortone7b38402006-06-30 01:56:00 -07005050static void __cond_resched(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005051{
Frederic Weisbeckere7aaaa62009-07-16 15:44:29 +02005052 add_preempt_count(PREEMPT_ACTIVE);
5053 schedule();
5054 sub_preempt_count(PREEMPT_ACTIVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005055}
5056
Herbert Xu02b67cc32008-01-25 21:08:28 +01005057int __sched _cond_resched(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005058{
Peter Zijlstrad86ee482009-07-10 14:57:57 +02005059 if (should_resched()) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005060 __cond_resched();
5061 return 1;
5062 }
5063 return 0;
5064}
Herbert Xu02b67cc32008-01-25 21:08:28 +01005065EXPORT_SYMBOL(_cond_resched);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005066
5067/*
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02005068 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005069 * call schedule, and on return reacquire the lock.
5070 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005071 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
Linus Torvalds1da177e2005-04-16 15:20:36 -07005072 * operations here to prevent schedule() from being called twice (once via
5073 * spin_unlock(), once by hand).
5074 */
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02005075int __cond_resched_lock(spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005076{
Peter Zijlstrad86ee482009-07-10 14:57:57 +02005077 int resched = should_resched();
Jan Kara6df3cec2005-06-13 15:52:32 -07005078 int ret = 0;
5079
Peter Zijlstraf607c662009-07-20 19:16:29 +02005080 lockdep_assert_held(lock);
5081
Nick Piggin95c354f2008-01-30 13:31:20 +01005082 if (spin_needbreak(lock) || resched) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005083 spin_unlock(lock);
Peter Zijlstrad86ee482009-07-10 14:57:57 +02005084 if (resched)
Nick Piggin95c354f2008-01-30 13:31:20 +01005085 __cond_resched();
5086 else
5087 cpu_relax();
Jan Kara6df3cec2005-06-13 15:52:32 -07005088 ret = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005089 spin_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005090 }
Jan Kara6df3cec2005-06-13 15:52:32 -07005091 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005092}
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02005093EXPORT_SYMBOL(__cond_resched_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005094
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02005095int __sched __cond_resched_softirq(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005096{
5097 BUG_ON(!in_softirq());
5098
Peter Zijlstrad86ee482009-07-10 14:57:57 +02005099 if (should_resched()) {
Thomas Gleixner98d825672007-05-23 13:58:18 -07005100 local_bh_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005101 __cond_resched();
5102 local_bh_disable();
5103 return 1;
5104 }
5105 return 0;
5106}
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02005107EXPORT_SYMBOL(__cond_resched_softirq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005108
Linus Torvalds1da177e2005-04-16 15:20:36 -07005109/**
5110 * yield - yield the current processor to other threads.
5111 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08005112 * This is a shortcut for kernel-space yielding - it marks the
Linus Torvalds1da177e2005-04-16 15:20:36 -07005113 * thread runnable and calls sys_sched_yield().
5114 */
5115void __sched yield(void)
5116{
5117 set_current_state(TASK_RUNNING);
5118 sys_sched_yield();
5119}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005120EXPORT_SYMBOL(yield);
5121
5122/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005123 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
Linus Torvalds1da177e2005-04-16 15:20:36 -07005124 * that process accounting knows that this is a task in IO wait state.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005125 */
5126void __sched io_schedule(void)
5127{
Hitoshi Mitake54d35f22009-06-29 14:44:57 +09005128 struct rq *rq = raw_rq();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005129
Shailabh Nagar0ff92242006-07-14 00:24:37 -07005130 delayacct_blkio_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005131 atomic_inc(&rq->nr_iowait);
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07005132 current->in_iowait = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005133 schedule();
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07005134 current->in_iowait = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005135 atomic_dec(&rq->nr_iowait);
Shailabh Nagar0ff92242006-07-14 00:24:37 -07005136 delayacct_blkio_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005137}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005138EXPORT_SYMBOL(io_schedule);
5139
5140long __sched io_schedule_timeout(long timeout)
5141{
Hitoshi Mitake54d35f22009-06-29 14:44:57 +09005142 struct rq *rq = raw_rq();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005143 long ret;
5144
Shailabh Nagar0ff92242006-07-14 00:24:37 -07005145 delayacct_blkio_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005146 atomic_inc(&rq->nr_iowait);
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07005147 current->in_iowait = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005148 ret = schedule_timeout(timeout);
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07005149 current->in_iowait = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005150 atomic_dec(&rq->nr_iowait);
Shailabh Nagar0ff92242006-07-14 00:24:37 -07005151 delayacct_blkio_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005152 return ret;
5153}
5154
5155/**
5156 * sys_sched_get_priority_max - return maximum RT priority.
5157 * @policy: scheduling class.
5158 *
5159 * this syscall returns the maximum rt_priority that can be used
5160 * by a given scheduling class.
5161 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01005162SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005163{
5164 int ret = -EINVAL;
5165
5166 switch (policy) {
5167 case SCHED_FIFO:
5168 case SCHED_RR:
5169 ret = MAX_USER_RT_PRIO-1;
5170 break;
5171 case SCHED_NORMAL:
Ingo Molnarb0a94992006-01-14 13:20:41 -08005172 case SCHED_BATCH:
Ingo Molnardd41f592007-07-09 18:51:59 +02005173 case SCHED_IDLE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005174 ret = 0;
5175 break;
5176 }
5177 return ret;
5178}
5179
5180/**
5181 * sys_sched_get_priority_min - return minimum RT priority.
5182 * @policy: scheduling class.
5183 *
5184 * this syscall returns the minimum rt_priority that can be used
5185 * by a given scheduling class.
5186 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01005187SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005188{
5189 int ret = -EINVAL;
5190
5191 switch (policy) {
5192 case SCHED_FIFO:
5193 case SCHED_RR:
5194 ret = 1;
5195 break;
5196 case SCHED_NORMAL:
Ingo Molnarb0a94992006-01-14 13:20:41 -08005197 case SCHED_BATCH:
Ingo Molnardd41f592007-07-09 18:51:59 +02005198 case SCHED_IDLE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005199 ret = 0;
5200 }
5201 return ret;
5202}
5203
5204/**
5205 * sys_sched_rr_get_interval - return the default timeslice of a process.
5206 * @pid: pid of the process.
5207 * @interval: userspace pointer to the timeslice value.
5208 *
5209 * this syscall writes the default timeslice value of a given process
5210 * into the user-space timespec buffer. A value of '0' means infinity.
5211 */
Heiko Carstens17da2bd2009-01-14 14:14:10 +01005212SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
Heiko Carstens754fe8d2009-01-14 14:14:09 +01005213 struct timespec __user *, interval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005214{
Ingo Molnar36c8b582006-07-03 00:25:41 -07005215 struct task_struct *p;
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005216 unsigned int time_slice;
Thomas Gleixnerdba091b2009-12-09 09:32:03 +01005217 unsigned long flags;
5218 struct rq *rq;
Andi Kleen3a5c3592007-10-15 17:00:14 +02005219 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005220 struct timespec t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005221
5222 if (pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02005223 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005224
5225 retval = -ESRCH;
Thomas Gleixner1a551ae2009-12-09 10:15:11 +00005226 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005227 p = find_process_by_pid(pid);
5228 if (!p)
5229 goto out_unlock;
5230
5231 retval = security_task_getscheduler(p);
5232 if (retval)
5233 goto out_unlock;
5234
Thomas Gleixnerdba091b2009-12-09 09:32:03 +01005235 rq = task_rq_lock(p, &flags);
5236 time_slice = p->sched_class->get_rr_interval(rq, p);
5237 task_rq_unlock(rq, &flags);
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005238
Thomas Gleixner1a551ae2009-12-09 10:15:11 +00005239 rcu_read_unlock();
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005240 jiffies_to_timespec(time_slice, &t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005241 retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005242 return retval;
Andi Kleen3a5c3592007-10-15 17:00:14 +02005243
Linus Torvalds1da177e2005-04-16 15:20:36 -07005244out_unlock:
Thomas Gleixner1a551ae2009-12-09 10:15:11 +00005245 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005246 return retval;
5247}
5248
Steven Rostedt7c731e02008-05-12 21:20:41 +02005249static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
Ingo Molnar36c8b582006-07-03 00:25:41 -07005250
Ingo Molnar82a1fcb2008-01-25 21:08:02 +01005251void sched_show_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005252{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005253 unsigned long free = 0;
Ingo Molnar36c8b582006-07-03 00:25:41 -07005254 unsigned state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005255
Linus Torvalds1da177e2005-04-16 15:20:36 -07005256 state = p->state ? __ffs(p->state) + 1 : 0;
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005257 printk(KERN_INFO "%-13.13s %c", p->comm,
Andreas Mohr2ed6e342006-07-10 04:43:52 -07005258 state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
Ingo Molnar4bd77322007-07-11 21:21:47 +02005259#if BITS_PER_LONG == 32
Linus Torvalds1da177e2005-04-16 15:20:36 -07005260 if (state == TASK_RUNNING)
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005261 printk(KERN_CONT " running ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005262 else
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005263 printk(KERN_CONT " %08lx ", thread_saved_pc(p));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005264#else
5265 if (state == TASK_RUNNING)
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005266 printk(KERN_CONT " running task ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005267 else
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005268 printk(KERN_CONT " %016lx ", thread_saved_pc(p));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005269#endif
5270#ifdef CONFIG_DEBUG_STACK_USAGE
Eric Sandeen7c9f8862008-04-22 16:38:23 -05005271 free = stack_not_used(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005272#endif
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005273 printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
David Rientjesaa47b7e2009-05-04 01:38:05 -07005274 task_pid_nr(p), task_pid_nr(p->real_parent),
5275 (unsigned long)task_thread_info(p)->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005276
Nick Piggin5fb5e6d2008-01-25 21:08:34 +01005277 show_stack(p, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005278}
5279
Ingo Molnare59e2ae2006-12-06 20:35:59 -08005280void show_state_filter(unsigned long state_filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005281{
Ingo Molnar36c8b582006-07-03 00:25:41 -07005282 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005283
Ingo Molnar4bd77322007-07-11 21:21:47 +02005284#if BITS_PER_LONG == 32
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005285 printk(KERN_INFO
5286 " task PC stack pid father\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005287#else
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005288 printk(KERN_INFO
5289 " task PC stack pid father\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005290#endif
5291 read_lock(&tasklist_lock);
5292 do_each_thread(g, p) {
5293 /*
5294 * reset the NMI-timeout, listing all files on a slow
5295 * console might take alot of time:
5296 */
5297 touch_nmi_watchdog();
Ingo Molnar39bc89f2007-04-25 20:50:03 -07005298 if (!state_filter || (p->state & state_filter))
Ingo Molnar82a1fcb2008-01-25 21:08:02 +01005299 sched_show_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005300 } while_each_thread(g, p);
5301
Jeremy Fitzhardinge04c91672007-05-08 00:28:05 -07005302 touch_all_softlockup_watchdogs();
5303
Ingo Molnardd41f592007-07-09 18:51:59 +02005304#ifdef CONFIG_SCHED_DEBUG
5305 sysrq_sched_debug_show();
5306#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005307 read_unlock(&tasklist_lock);
Ingo Molnare59e2ae2006-12-06 20:35:59 -08005308 /*
5309 * Only show locks if all tasks are dumped:
5310 */
Shmulik Ladkani93335a22009-11-25 15:23:41 +02005311 if (!state_filter)
Ingo Molnare59e2ae2006-12-06 20:35:59 -08005312 debug_show_all_locks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005313}
5314
Ingo Molnar1df21052007-07-09 18:51:58 +02005315void __cpuinit init_idle_bootup_task(struct task_struct *idle)
5316{
Ingo Molnardd41f592007-07-09 18:51:59 +02005317 idle->sched_class = &idle_sched_class;
Ingo Molnar1df21052007-07-09 18:51:58 +02005318}
5319
Ingo Molnarf340c0d2005-06-28 16:40:42 +02005320/**
5321 * init_idle - set up an idle thread for a given CPU
5322 * @idle: task in question
5323 * @cpu: cpu the idle task belongs to
5324 *
5325 * NOTE: this function does not set the idle thread's NEED_RESCHED
5326 * flag, to make booting more robust.
5327 */
Nick Piggin5c1e1762006-10-03 01:14:04 -07005328void __cpuinit init_idle(struct task_struct *idle, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005329{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005330 struct rq *rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005331 unsigned long flags;
5332
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005333 raw_spin_lock_irqsave(&rq->lock, flags);
Ingo Molnar5cbd54e2008-11-12 20:05:50 +01005334
Ingo Molnardd41f592007-07-09 18:51:59 +02005335 __sched_fork(idle);
Peter Zijlstra06b83b52009-12-16 18:04:35 +01005336 idle->state = TASK_RUNNING;
Ingo Molnardd41f592007-07-09 18:51:59 +02005337 idle->se.exec_start = sched_clock();
5338
Rusty Russell96f874e22008-11-25 02:35:14 +10305339 cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu));
Ingo Molnardd41f592007-07-09 18:51:59 +02005340 __set_task_cpu(idle, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005341
Linus Torvalds1da177e2005-04-16 15:20:36 -07005342 rq->curr = rq->idle = idle;
Nick Piggin4866cde2005-06-25 14:57:23 -07005343#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
5344 idle->oncpu = 1;
5345#endif
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005346 raw_spin_unlock_irqrestore(&rq->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005347
5348 /* Set the preempt count _outside_ the spinlocks! */
Linus Torvalds8e3e0762008-05-10 20:58:02 -07005349#if defined(CONFIG_PREEMPT)
5350 task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0);
5351#else
Al Viroa1261f542005-11-13 16:06:55 -08005352 task_thread_info(idle)->preempt_count = 0;
Linus Torvalds8e3e0762008-05-10 20:58:02 -07005353#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02005354 /*
5355 * The idle tasks have their own, simple scheduling class:
5356 */
5357 idle->sched_class = &idle_sched_class;
Frederic Weisbeckerfb526072008-11-25 21:07:04 +01005358 ftrace_graph_init_task(idle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005359}
5360
5361/*
5362 * In a system that switches off the HZ timer nohz_cpu_mask
5363 * indicates which cpus entered this state. This is used
5364 * in the rcu update to wait only for active cpus. For system
5365 * which do not switch off the HZ timer nohz_cpu_mask should
Rusty Russell6a7b3dc2008-11-25 02:35:04 +10305366 * always be CPU_BITS_NONE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005367 */
Rusty Russell6a7b3dc2008-11-25 02:35:04 +10305368cpumask_var_t nohz_cpu_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005369
Ingo Molnar19978ca2007-11-09 22:39:38 +01005370/*
5371 * Increase the granularity value when there are more CPUs,
5372 * because with more CPUs the 'effective latency' as visible
5373 * to users decreases. But the relationship is not linear,
5374 * so pick a second-best guess by going with the log2 of the
5375 * number of CPUs.
5376 *
5377 * This idea comes from the SD scheduler of Con Kolivas:
5378 */
Christian Ehrhardtacb4a842009-11-30 12:16:48 +01005379static int get_update_sysctl_factor(void)
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01005380{
Mike Galbraith4ca3ef72009-12-10 09:25:53 +01005381 unsigned int cpus = min_t(int, num_online_cpus(), 8);
Christian Ehrhardt1983a922009-11-30 12:16:47 +01005382 unsigned int factor;
5383
5384 switch (sysctl_sched_tunable_scaling) {
5385 case SCHED_TUNABLESCALING_NONE:
5386 factor = 1;
5387 break;
5388 case SCHED_TUNABLESCALING_LINEAR:
5389 factor = cpus;
5390 break;
5391 case SCHED_TUNABLESCALING_LOG:
5392 default:
5393 factor = 1 + ilog2(cpus);
5394 break;
5395 }
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01005396
Christian Ehrhardtacb4a842009-11-30 12:16:48 +01005397 return factor;
5398}
5399
5400static void update_sysctl(void)
5401{
5402 unsigned int factor = get_update_sysctl_factor();
5403
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01005404#define SET_SYSCTL(name) \
5405 (sysctl_##name = (factor) * normalized_sysctl_##name)
5406 SET_SYSCTL(sched_min_granularity);
5407 SET_SYSCTL(sched_latency);
5408 SET_SYSCTL(sched_wakeup_granularity);
5409 SET_SYSCTL(sched_shares_ratelimit);
5410#undef SET_SYSCTL
5411}
5412
Ingo Molnar19978ca2007-11-09 22:39:38 +01005413static inline void sched_init_granularity(void)
5414{
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01005415 update_sysctl();
Ingo Molnar19978ca2007-11-09 22:39:38 +01005416}
5417
Linus Torvalds1da177e2005-04-16 15:20:36 -07005418#ifdef CONFIG_SMP
5419/*
5420 * This is how migration works:
5421 *
Tejun Heo969c7922010-05-06 18:49:21 +02005422 * 1) we invoke migration_cpu_stop() on the target CPU using
5423 * stop_one_cpu().
5424 * 2) stopper starts to run (implicitly forcing the migrated thread
5425 * off the CPU)
5426 * 3) it checks whether the migrated task is still in the wrong runqueue.
5427 * 4) if it's in the wrong runqueue then the migration thread removes
Linus Torvalds1da177e2005-04-16 15:20:36 -07005428 * it and puts it into the right queue.
Tejun Heo969c7922010-05-06 18:49:21 +02005429 * 5) stopper completes and stop_one_cpu() returns and the migration
5430 * is done.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005431 */
5432
5433/*
5434 * Change a given task's CPU affinity. Migrate the thread to a
5435 * proper CPU and schedule it away if the CPU it's executing on
5436 * is removed from the allowed bitmask.
5437 *
5438 * NOTE: the caller must have a valid reference to the task, the
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005439 * task must not exit() & deallocate itself prematurely. The
Linus Torvalds1da177e2005-04-16 15:20:36 -07005440 * call is not atomic; no spinlocks may be held.
5441 */
Rusty Russell96f874e22008-11-25 02:35:14 +10305442int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005443{
5444 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07005445 struct rq *rq;
Tejun Heo969c7922010-05-06 18:49:21 +02005446 unsigned int dest_cpu;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005447 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005448
Peter Zijlstra65cc8e42010-03-25 21:05:16 +01005449 /*
5450 * Serialize against TASK_WAKING so that ttwu() and wunt() can
5451 * drop the rq->lock and still rely on ->cpus_allowed.
5452 */
5453again:
5454 while (task_is_waking(p))
5455 cpu_relax();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005456 rq = task_rq_lock(p, &flags);
Peter Zijlstra65cc8e42010-03-25 21:05:16 +01005457 if (task_is_waking(p)) {
5458 task_rq_unlock(rq, &flags);
5459 goto again;
5460 }
Peter Zijlstrae2912002009-12-16 18:04:36 +01005461
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01005462 if (!cpumask_intersects(new_mask, cpu_active_mask)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005463 ret = -EINVAL;
5464 goto out;
5465 }
5466
David Rientjes9985b0b2008-06-05 12:57:11 -07005467 if (unlikely((p->flags & PF_THREAD_BOUND) && p != current &&
Rusty Russell96f874e22008-11-25 02:35:14 +10305468 !cpumask_equal(&p->cpus_allowed, new_mask))) {
David Rientjes9985b0b2008-06-05 12:57:11 -07005469 ret = -EINVAL;
5470 goto out;
5471 }
5472
Gregory Haskins73fe6aae2008-01-25 21:08:07 +01005473 if (p->sched_class->set_cpus_allowed)
Mike Traviscd8ba7c2008-03-26 14:23:49 -07005474 p->sched_class->set_cpus_allowed(p, new_mask);
Gregory Haskins73fe6aae2008-01-25 21:08:07 +01005475 else {
Rusty Russell96f874e22008-11-25 02:35:14 +10305476 cpumask_copy(&p->cpus_allowed, new_mask);
5477 p->rt.nr_cpus_allowed = cpumask_weight(new_mask);
Gregory Haskins73fe6aae2008-01-25 21:08:07 +01005478 }
5479
Linus Torvalds1da177e2005-04-16 15:20:36 -07005480 /* Can the task run on the task's current CPU? If so, we're done */
Rusty Russell96f874e22008-11-25 02:35:14 +10305481 if (cpumask_test_cpu(task_cpu(p), new_mask))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005482 goto out;
5483
Tejun Heo969c7922010-05-06 18:49:21 +02005484 dest_cpu = cpumask_any_and(cpu_active_mask, new_mask);
5485 if (migrate_task(p, dest_cpu)) {
5486 struct migration_arg arg = { p, dest_cpu };
Linus Torvalds1da177e2005-04-16 15:20:36 -07005487 /* Need help from migration thread: drop lock and wait. */
5488 task_rq_unlock(rq, &flags);
Tejun Heo969c7922010-05-06 18:49:21 +02005489 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005490 tlb_migrate_finish(p->mm);
5491 return 0;
5492 }
5493out:
5494 task_rq_unlock(rq, &flags);
Ingo Molnar48f24c42006-07-03 00:25:40 -07005495
Linus Torvalds1da177e2005-04-16 15:20:36 -07005496 return ret;
5497}
Mike Traviscd8ba7c2008-03-26 14:23:49 -07005498EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005499
5500/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005501 * Move (not current) task off this cpu, onto dest cpu. We're doing
Linus Torvalds1da177e2005-04-16 15:20:36 -07005502 * this because either it can't run here any more (set_cpus_allowed()
5503 * away from this CPU, or CPU going down), or because we're
5504 * attempting to rebalance this task on exec (sched_exec).
5505 *
5506 * So we race with normal scheduler movements, but that's OK, as long
5507 * as the task is no longer on this CPU.
Kirill Korotaevefc30812006-06-27 02:54:32 -07005508 *
5509 * Returns non-zero if task was successfully migrated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005510 */
Kirill Korotaevefc30812006-06-27 02:54:32 -07005511static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005512{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005513 struct rq *rq_dest, *rq_src;
Peter Zijlstrae2912002009-12-16 18:04:36 +01005514 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005515
Max Krasnyanskye761b772008-07-15 04:43:49 -07005516 if (unlikely(!cpu_active(dest_cpu)))
Kirill Korotaevefc30812006-06-27 02:54:32 -07005517 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005518
5519 rq_src = cpu_rq(src_cpu);
5520 rq_dest = cpu_rq(dest_cpu);
5521
5522 double_rq_lock(rq_src, rq_dest);
5523 /* Already moved. */
5524 if (task_cpu(p) != src_cpu)
Linus Torvaldsb1e38732008-07-10 11:25:03 -07005525 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005526 /* Affinity changed (again). */
Rusty Russell96f874e22008-11-25 02:35:14 +10305527 if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
Linus Torvaldsb1e38732008-07-10 11:25:03 -07005528 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005529
Peter Zijlstrae2912002009-12-16 18:04:36 +01005530 /*
5531 * If we're not on a rq, the next wake-up will ensure we're
5532 * placed properly.
5533 */
5534 if (p->se.on_rq) {
Ingo Molnar2e1cb742007-08-09 11:16:49 +02005535 deactivate_task(rq_src, p, 0);
Peter Zijlstrae2912002009-12-16 18:04:36 +01005536 set_task_cpu(p, dest_cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02005537 activate_task(rq_dest, p, 0);
Peter Zijlstra15afe092008-09-20 23:38:02 +02005538 check_preempt_curr(rq_dest, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005539 }
Linus Torvaldsb1e38732008-07-10 11:25:03 -07005540done:
Kirill Korotaevefc30812006-06-27 02:54:32 -07005541 ret = 1;
Linus Torvaldsb1e38732008-07-10 11:25:03 -07005542fail:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005543 double_rq_unlock(rq_src, rq_dest);
Kirill Korotaevefc30812006-06-27 02:54:32 -07005544 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005545}
5546
5547/*
Tejun Heo969c7922010-05-06 18:49:21 +02005548 * migration_cpu_stop - this will be executed by a highprio stopper thread
5549 * and performs thread migration by bumping thread off CPU then
5550 * 'pushing' onto another runqueue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005551 */
Tejun Heo969c7922010-05-06 18:49:21 +02005552static int migration_cpu_stop(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005553{
Tejun Heo969c7922010-05-06 18:49:21 +02005554 struct migration_arg *arg = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005555
Tejun Heo969c7922010-05-06 18:49:21 +02005556 /*
5557 * The original target cpu might have gone down and we might
5558 * be on another cpu but it doesn't matter.
5559 */
5560 local_irq_disable();
5561 __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu);
5562 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005563 return 0;
5564}
5565
5566#ifdef CONFIG_HOTPLUG_CPU
Kirill Korotaev054b9102006-12-10 02:20:11 -08005567/*
Robert P. J. Day3a4fa0a2007-10-19 23:10:43 +02005568 * Figure out where task on dead CPU should go, use force if necessary.
Kirill Korotaev054b9102006-12-10 02:20:11 -08005569 */
Oleg Nesterov6a1bdc12010-03-15 10:10:23 +01005570void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005571{
Oleg Nesterov1445c082010-03-15 10:10:10 +01005572 struct rq *rq = cpu_rq(dead_cpu);
5573 int needs_cpu, uninitialized_var(dest_cpu);
5574 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005575
Oleg Nesterov1445c082010-03-15 10:10:10 +01005576 local_irq_save(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005577
Oleg Nesterov1445c082010-03-15 10:10:10 +01005578 raw_spin_lock(&rq->lock);
5579 needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING);
5580 if (needs_cpu)
5581 dest_cpu = select_fallback_rq(dead_cpu, p);
5582 raw_spin_unlock(&rq->lock);
Oleg Nesterovc1804d52010-03-15 10:10:14 +01005583 /*
5584 * It can only fail if we race with set_cpus_allowed(),
5585 * in the racer should migrate the task anyway.
5586 */
Oleg Nesterov1445c082010-03-15 10:10:10 +01005587 if (needs_cpu)
Oleg Nesterovc1804d52010-03-15 10:10:14 +01005588 __migrate_task(p, dead_cpu, dest_cpu);
Oleg Nesterov1445c082010-03-15 10:10:10 +01005589 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005590}
5591
5592/*
5593 * While a dead CPU has no uninterruptible tasks queued at this point,
5594 * it might still have a nonzero ->nr_uninterruptible counter, because
5595 * for performance reasons the counter is not stricly tracking tasks to
5596 * their home CPUs. So we just add the counter to another CPU's counter,
5597 * to keep the global sum constant after CPU-down:
5598 */
Ingo Molnar70b97a72006-07-03 00:25:42 -07005599static void migrate_nr_uninterruptible(struct rq *rq_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005600{
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01005601 struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005602 unsigned long flags;
5603
5604 local_irq_save(flags);
5605 double_rq_lock(rq_src, rq_dest);
5606 rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible;
5607 rq_src->nr_uninterruptible = 0;
5608 double_rq_unlock(rq_src, rq_dest);
5609 local_irq_restore(flags);
5610}
5611
5612/* Run through task list and migrate tasks from the dead cpu. */
5613static void migrate_live_tasks(int src_cpu)
5614{
Ingo Molnar48f24c42006-07-03 00:25:40 -07005615 struct task_struct *p, *t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005616
Oleg Nesterovf7b4cdd2007-10-16 23:30:56 -07005617 read_lock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005618
Ingo Molnar48f24c42006-07-03 00:25:40 -07005619 do_each_thread(t, p) {
5620 if (p == current)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005621 continue;
5622
Ingo Molnar48f24c42006-07-03 00:25:40 -07005623 if (task_cpu(p) == src_cpu)
5624 move_task_off_dead_cpu(src_cpu, p);
5625 } while_each_thread(t, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005626
Oleg Nesterovf7b4cdd2007-10-16 23:30:56 -07005627 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005628}
5629
Ingo Molnardd41f592007-07-09 18:51:59 +02005630/*
5631 * Schedules idle task to be the next runnable task on current CPU.
Dmitry Adamushko94bc9a72007-11-15 20:57:40 +01005632 * It does so by boosting its priority to highest possible.
5633 * Used by CPU offline code.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005634 */
5635void sched_idle_next(void)
5636{
Ingo Molnar48f24c42006-07-03 00:25:40 -07005637 int this_cpu = smp_processor_id();
Ingo Molnar70b97a72006-07-03 00:25:42 -07005638 struct rq *rq = cpu_rq(this_cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005639 struct task_struct *p = rq->idle;
5640 unsigned long flags;
5641
5642 /* cpu has to be offline */
Ingo Molnar48f24c42006-07-03 00:25:40 -07005643 BUG_ON(cpu_online(this_cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005644
Ingo Molnar48f24c42006-07-03 00:25:40 -07005645 /*
5646 * Strictly not necessary since rest of the CPUs are stopped by now
5647 * and interrupts disabled on the current cpu.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005648 */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005649 raw_spin_lock_irqsave(&rq->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005650
Ingo Molnardd41f592007-07-09 18:51:59 +02005651 __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1);
Ingo Molnar48f24c42006-07-03 00:25:40 -07005652
Dmitry Adamushko94bc9a72007-11-15 20:57:40 +01005653 activate_task(rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005654
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005655 raw_spin_unlock_irqrestore(&rq->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005656}
5657
Ingo Molnar48f24c42006-07-03 00:25:40 -07005658/*
5659 * Ensures that the idle task is using init_mm right before its cpu goes
Linus Torvalds1da177e2005-04-16 15:20:36 -07005660 * offline.
5661 */
5662void idle_task_exit(void)
5663{
5664 struct mm_struct *mm = current->active_mm;
5665
5666 BUG_ON(cpu_online(smp_processor_id()));
5667
5668 if (mm != &init_mm)
5669 switch_mm(mm, &init_mm, current);
5670 mmdrop(mm);
5671}
5672
Kirill Korotaev054b9102006-12-10 02:20:11 -08005673/* called under rq->lock with disabled interrupts */
Ingo Molnar36c8b582006-07-03 00:25:41 -07005674static void migrate_dead(unsigned int dead_cpu, struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005675{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005676 struct rq *rq = cpu_rq(dead_cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005677
5678 /* Must be exiting, otherwise would be on tasklist. */
Eugene Teo270f7222007-10-18 23:40:38 -07005679 BUG_ON(!p->exit_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005680
5681 /* Cannot have done final schedule yet: would have vanished. */
Oleg Nesterovc394cc92006-09-29 02:01:11 -07005682 BUG_ON(p->state == TASK_DEAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005683
Ingo Molnar48f24c42006-07-03 00:25:40 -07005684 get_task_struct(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005685
5686 /*
5687 * Drop lock around migration; if someone else moves it,
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005688 * that's OK. No task can be added to this CPU, so iteration is
Linus Torvalds1da177e2005-04-16 15:20:36 -07005689 * fine.
5690 */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005691 raw_spin_unlock_irq(&rq->lock);
Ingo Molnar48f24c42006-07-03 00:25:40 -07005692 move_task_off_dead_cpu(dead_cpu, p);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005693 raw_spin_lock_irq(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005694
Ingo Molnar48f24c42006-07-03 00:25:40 -07005695 put_task_struct(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005696}
5697
5698/* release_task() removes task from tasklist, so we won't find dead tasks. */
5699static void migrate_dead_tasks(unsigned int dead_cpu)
5700{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005701 struct rq *rq = cpu_rq(dead_cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02005702 struct task_struct *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005703
Ingo Molnardd41f592007-07-09 18:51:59 +02005704 for ( ; ; ) {
5705 if (!rq->nr_running)
5706 break;
Wang Chenb67802e2009-03-02 13:55:26 +08005707 next = pick_next_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02005708 if (!next)
5709 break;
Dmitry Adamushko79c53792008-06-29 00:16:56 +02005710 next->sched_class->put_prev_task(rq, next);
Ingo Molnardd41f592007-07-09 18:51:59 +02005711 migrate_dead(dead_cpu, next);
Nick Piggine692ab52007-07-26 13:40:43 +02005712
Linus Torvalds1da177e2005-04-16 15:20:36 -07005713 }
5714}
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02005715
5716/*
5717 * remove the tasks which were accounted by rq from calc_load_tasks.
5718 */
5719static void calc_global_load_remove(struct rq *rq)
5720{
5721 atomic_long_sub(rq->calc_load_active, &calc_load_tasks);
Thomas Gleixnera468d382009-07-17 14:15:46 +02005722 rq->calc_load_active = 0;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02005723}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005724#endif /* CONFIG_HOTPLUG_CPU */
5725
Nick Piggine692ab52007-07-26 13:40:43 +02005726#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
5727
5728static struct ctl_table sd_ctl_dir[] = {
Alexey Dobriyane0361852007-08-09 11:16:46 +02005729 {
5730 .procname = "sched_domain",
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005731 .mode = 0555,
Alexey Dobriyane0361852007-08-09 11:16:46 +02005732 },
Eric W. Biederman56992302009-11-05 15:38:40 -08005733 {}
Nick Piggine692ab52007-07-26 13:40:43 +02005734};
5735
5736static struct ctl_table sd_ctl_root[] = {
Alexey Dobriyane0361852007-08-09 11:16:46 +02005737 {
5738 .procname = "kernel",
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005739 .mode = 0555,
Alexey Dobriyane0361852007-08-09 11:16:46 +02005740 .child = sd_ctl_dir,
5741 },
Eric W. Biederman56992302009-11-05 15:38:40 -08005742 {}
Nick Piggine692ab52007-07-26 13:40:43 +02005743};
5744
5745static struct ctl_table *sd_alloc_ctl_entry(int n)
5746{
5747 struct ctl_table *entry =
Milton Miller5cf9f062007-10-15 17:00:19 +02005748 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
Nick Piggine692ab52007-07-26 13:40:43 +02005749
Nick Piggine692ab52007-07-26 13:40:43 +02005750 return entry;
5751}
5752
Milton Miller6382bc92007-10-15 17:00:19 +02005753static void sd_free_ctl_entry(struct ctl_table **tablep)
5754{
Milton Millercd7900762007-10-17 16:55:11 +02005755 struct ctl_table *entry;
Milton Miller6382bc92007-10-15 17:00:19 +02005756
Milton Millercd7900762007-10-17 16:55:11 +02005757 /*
5758 * In the intermediate directories, both the child directory and
5759 * procname are dynamically allocated and could fail but the mode
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005760 * will always be set. In the lowest directory the names are
Milton Millercd7900762007-10-17 16:55:11 +02005761 * static strings and all have proc handlers.
5762 */
5763 for (entry = *tablep; entry->mode; entry++) {
Milton Miller6382bc92007-10-15 17:00:19 +02005764 if (entry->child)
5765 sd_free_ctl_entry(&entry->child);
Milton Millercd7900762007-10-17 16:55:11 +02005766 if (entry->proc_handler == NULL)
5767 kfree(entry->procname);
5768 }
Milton Miller6382bc92007-10-15 17:00:19 +02005769
5770 kfree(*tablep);
5771 *tablep = NULL;
5772}
5773
Nick Piggine692ab52007-07-26 13:40:43 +02005774static void
Alexey Dobriyane0361852007-08-09 11:16:46 +02005775set_table_entry(struct ctl_table *entry,
Nick Piggine692ab52007-07-26 13:40:43 +02005776 const char *procname, void *data, int maxlen,
5777 mode_t mode, proc_handler *proc_handler)
5778{
Nick Piggine692ab52007-07-26 13:40:43 +02005779 entry->procname = procname;
5780 entry->data = data;
5781 entry->maxlen = maxlen;
5782 entry->mode = mode;
5783 entry->proc_handler = proc_handler;
5784}
5785
5786static struct ctl_table *
5787sd_alloc_ctl_domain_table(struct sched_domain *sd)
5788{
Ingo Molnara5d8c342008-10-09 11:35:51 +02005789 struct ctl_table *table = sd_alloc_ctl_entry(13);
Nick Piggine692ab52007-07-26 13:40:43 +02005790
Milton Millerad1cdc12007-10-15 17:00:19 +02005791 if (table == NULL)
5792 return NULL;
5793
Alexey Dobriyane0361852007-08-09 11:16:46 +02005794 set_table_entry(&table[0], "min_interval", &sd->min_interval,
Nick Piggine692ab52007-07-26 13:40:43 +02005795 sizeof(long), 0644, proc_doulongvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005796 set_table_entry(&table[1], "max_interval", &sd->max_interval,
Nick Piggine692ab52007-07-26 13:40:43 +02005797 sizeof(long), 0644, proc_doulongvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005798 set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005799 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005800 set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005801 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005802 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005803 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005804 set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005805 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005806 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005807 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005808 set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
Nick Piggine692ab52007-07-26 13:40:43 +02005809 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005810 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
Nick Piggine692ab52007-07-26 13:40:43 +02005811 sizeof(int), 0644, proc_dointvec_minmax);
Zou Nan haiace8b3d2007-10-15 17:00:14 +02005812 set_table_entry(&table[9], "cache_nice_tries",
Nick Piggine692ab52007-07-26 13:40:43 +02005813 &sd->cache_nice_tries,
5814 sizeof(int), 0644, proc_dointvec_minmax);
Zou Nan haiace8b3d2007-10-15 17:00:14 +02005815 set_table_entry(&table[10], "flags", &sd->flags,
Nick Piggine692ab52007-07-26 13:40:43 +02005816 sizeof(int), 0644, proc_dointvec_minmax);
Ingo Molnara5d8c342008-10-09 11:35:51 +02005817 set_table_entry(&table[11], "name", sd->name,
5818 CORENAME_MAX_SIZE, 0444, proc_dostring);
5819 /* &table[12] is terminator */
Nick Piggine692ab52007-07-26 13:40:43 +02005820
5821 return table;
5822}
5823
Ingo Molnar9a4e7152007-11-28 15:52:56 +01005824static ctl_table *sd_alloc_ctl_cpu_table(int cpu)
Nick Piggine692ab52007-07-26 13:40:43 +02005825{
5826 struct ctl_table *entry, *table;
5827 struct sched_domain *sd;
5828 int domain_num = 0, i;
5829 char buf[32];
5830
5831 for_each_domain(cpu, sd)
5832 domain_num++;
5833 entry = table = sd_alloc_ctl_entry(domain_num + 1);
Milton Millerad1cdc12007-10-15 17:00:19 +02005834 if (table == NULL)
5835 return NULL;
Nick Piggine692ab52007-07-26 13:40:43 +02005836
5837 i = 0;
5838 for_each_domain(cpu, sd) {
5839 snprintf(buf, 32, "domain%d", i);
Nick Piggine692ab52007-07-26 13:40:43 +02005840 entry->procname = kstrdup(buf, GFP_KERNEL);
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005841 entry->mode = 0555;
Nick Piggine692ab52007-07-26 13:40:43 +02005842 entry->child = sd_alloc_ctl_domain_table(sd);
5843 entry++;
5844 i++;
5845 }
5846 return table;
5847}
5848
5849static struct ctl_table_header *sd_sysctl_header;
Milton Miller6382bc92007-10-15 17:00:19 +02005850static void register_sched_domain_sysctl(void)
Nick Piggine692ab52007-07-26 13:40:43 +02005851{
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01005852 int i, cpu_num = num_possible_cpus();
Nick Piggine692ab52007-07-26 13:40:43 +02005853 struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
5854 char buf[32];
5855
Milton Miller73785472007-10-24 18:23:48 +02005856 WARN_ON(sd_ctl_dir[0].child);
5857 sd_ctl_dir[0].child = entry;
5858
Milton Millerad1cdc12007-10-15 17:00:19 +02005859 if (entry == NULL)
5860 return;
5861
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01005862 for_each_possible_cpu(i) {
Nick Piggine692ab52007-07-26 13:40:43 +02005863 snprintf(buf, 32, "cpu%d", i);
Nick Piggine692ab52007-07-26 13:40:43 +02005864 entry->procname = kstrdup(buf, GFP_KERNEL);
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005865 entry->mode = 0555;
Nick Piggine692ab52007-07-26 13:40:43 +02005866 entry->child = sd_alloc_ctl_cpu_table(i);
Milton Miller97b6ea72007-10-15 17:00:19 +02005867 entry++;
Nick Piggine692ab52007-07-26 13:40:43 +02005868 }
Milton Miller73785472007-10-24 18:23:48 +02005869
5870 WARN_ON(sd_sysctl_header);
Nick Piggine692ab52007-07-26 13:40:43 +02005871 sd_sysctl_header = register_sysctl_table(sd_ctl_root);
5872}
Milton Miller6382bc92007-10-15 17:00:19 +02005873
Milton Miller73785472007-10-24 18:23:48 +02005874/* may be called multiple times per register */
Milton Miller6382bc92007-10-15 17:00:19 +02005875static void unregister_sched_domain_sysctl(void)
5876{
Milton Miller73785472007-10-24 18:23:48 +02005877 if (sd_sysctl_header)
5878 unregister_sysctl_table(sd_sysctl_header);
Milton Miller6382bc92007-10-15 17:00:19 +02005879 sd_sysctl_header = NULL;
Milton Miller73785472007-10-24 18:23:48 +02005880 if (sd_ctl_dir[0].child)
5881 sd_free_ctl_entry(&sd_ctl_dir[0].child);
Milton Miller6382bc92007-10-15 17:00:19 +02005882}
Nick Piggine692ab52007-07-26 13:40:43 +02005883#else
Milton Miller6382bc92007-10-15 17:00:19 +02005884static void register_sched_domain_sysctl(void)
5885{
5886}
5887static void unregister_sched_domain_sysctl(void)
Nick Piggine692ab52007-07-26 13:40:43 +02005888{
5889}
5890#endif
5891
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -04005892static void set_rq_online(struct rq *rq)
5893{
5894 if (!rq->online) {
5895 const struct sched_class *class;
5896
Rusty Russellc6c49272008-11-25 02:35:05 +10305897 cpumask_set_cpu(rq->cpu, rq->rd->online);
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -04005898 rq->online = 1;
5899
5900 for_each_class(class) {
5901 if (class->rq_online)
5902 class->rq_online(rq);
5903 }
5904 }
5905}
5906
5907static void set_rq_offline(struct rq *rq)
5908{
5909 if (rq->online) {
5910 const struct sched_class *class;
5911
5912 for_each_class(class) {
5913 if (class->rq_offline)
5914 class->rq_offline(rq);
5915 }
5916
Rusty Russellc6c49272008-11-25 02:35:05 +10305917 cpumask_clear_cpu(rq->cpu, rq->rd->online);
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -04005918 rq->online = 0;
5919 }
5920}
5921
Linus Torvalds1da177e2005-04-16 15:20:36 -07005922/*
5923 * migration_call - callback that gets triggered when a CPU is added.
5924 * Here we can start up the necessary migration thread for the new CPU.
5925 */
Ingo Molnar48f24c42006-07-03 00:25:40 -07005926static int __cpuinit
5927migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005928{
Ingo Molnar48f24c42006-07-03 00:25:40 -07005929 int cpu = (long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005930 unsigned long flags;
Tejun Heo969c7922010-05-06 18:49:21 +02005931 struct rq *rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005932
5933 switch (action) {
Gautham R Shenoy5be93612007-05-09 02:34:04 -07005934
Linus Torvalds1da177e2005-04-16 15:20:36 -07005935 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005936 case CPU_UP_PREPARE_FROZEN:
Thomas Gleixnera468d382009-07-17 14:15:46 +02005937 rq->calc_load_update = calc_load_update;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005938 break;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005939
Linus Torvalds1da177e2005-04-16 15:20:36 -07005940 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005941 case CPU_ONLINE_FROZEN:
Gregory Haskins1f94ef52008-03-10 16:52:41 -04005942 /* Update our root-domain */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005943 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins1f94ef52008-03-10 16:52:41 -04005944 if (rq->rd) {
Rusty Russellc6c49272008-11-25 02:35:05 +10305945 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -04005946
5947 set_rq_online(rq);
Gregory Haskins1f94ef52008-03-10 16:52:41 -04005948 }
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005949 raw_spin_unlock_irqrestore(&rq->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005950 break;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005951
Linus Torvalds1da177e2005-04-16 15:20:36 -07005952#ifdef CONFIG_HOTPLUG_CPU
Linus Torvalds1da177e2005-04-16 15:20:36 -07005953 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005954 case CPU_DEAD_FROZEN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005955 migrate_live_tasks(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005956 /* Idle task back to normal (off runqueue, low prio) */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005957 raw_spin_lock_irq(&rq->lock);
Ingo Molnar2e1cb742007-08-09 11:16:49 +02005958 deactivate_task(rq, rq->idle, 0);
Ingo Molnardd41f592007-07-09 18:51:59 +02005959 __setscheduler(rq, rq->idle, SCHED_NORMAL, 0);
5960 rq->idle->sched_class = &idle_sched_class;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005961 migrate_dead_tasks(cpu);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005962 raw_spin_unlock_irq(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005963 migrate_nr_uninterruptible(rq);
5964 BUG_ON(rq->nr_running != 0);
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02005965 calc_global_load_remove(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005966 break;
Gregory Haskins57d885f2008-01-25 21:08:18 +01005967
Gregory Haskins08f503b2008-03-10 17:59:11 -04005968 case CPU_DYING:
5969 case CPU_DYING_FROZEN:
Gregory Haskins57d885f2008-01-25 21:08:18 +01005970 /* Update our root-domain */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005971 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005972 if (rq->rd) {
Rusty Russellc6c49272008-11-25 02:35:05 +10305973 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -04005974 set_rq_offline(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005975 }
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005976 raw_spin_unlock_irqrestore(&rq->lock, flags);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005977 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005978#endif
5979 }
5980 return NOTIFY_OK;
5981}
5982
Paul Mackerrasf38b0822009-06-02 21:05:16 +10005983/*
5984 * Register at high priority so that task migration (migrate_all_tasks)
5985 * happens before everything else. This has to be lower priority than
Ingo Molnarcdd6c482009-09-21 12:02:48 +02005986 * the notifier in the perf_event subsystem, though.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005987 */
Chandra Seetharaman26c21432006-06-27 02:54:10 -07005988static struct notifier_block __cpuinitdata migration_notifier = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005989 .notifier_call = migration_call,
Tejun Heo50a323b2010-06-08 21:40:36 +02005990 .priority = CPU_PRI_MIGRATION,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005991};
5992
Tejun Heo3a101d02010-06-08 21:40:36 +02005993static int __cpuinit sched_cpu_active(struct notifier_block *nfb,
5994 unsigned long action, void *hcpu)
5995{
5996 switch (action & ~CPU_TASKS_FROZEN) {
5997 case CPU_ONLINE:
5998 case CPU_DOWN_FAILED:
5999 set_cpu_active((long)hcpu, true);
6000 return NOTIFY_OK;
6001 default:
6002 return NOTIFY_DONE;
6003 }
6004}
6005
6006static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb,
6007 unsigned long action, void *hcpu)
6008{
6009 switch (action & ~CPU_TASKS_FROZEN) {
6010 case CPU_DOWN_PREPARE:
6011 set_cpu_active((long)hcpu, false);
6012 return NOTIFY_OK;
6013 default:
6014 return NOTIFY_DONE;
6015 }
6016}
6017
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -07006018static int __init migration_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006019{
6020 void *cpu = (void *)(long)smp_processor_id();
Akinobu Mita07dccf32006-09-29 02:00:22 -07006021 int err;
Ingo Molnar48f24c42006-07-03 00:25:40 -07006022
Tejun Heo3a101d02010-06-08 21:40:36 +02006023 /* Initialize migration for the boot CPU */
Akinobu Mita07dccf32006-09-29 02:00:22 -07006024 err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
6025 BUG_ON(err == NOTIFY_BAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006026 migration_call(&migration_notifier, CPU_ONLINE, cpu);
6027 register_cpu_notifier(&migration_notifier);
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -07006028
Tejun Heo3a101d02010-06-08 21:40:36 +02006029 /* Register cpu active notifiers */
6030 cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE);
6031 cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE);
6032
Thomas Gleixnera004cd42009-07-21 09:54:05 +02006033 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006034}
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -07006035early_initcall(migration_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006036#endif
6037
6038#ifdef CONFIG_SMP
Christoph Lameter476f3532007-05-06 14:48:58 -07006039
Ingo Molnar3e9830d2007-10-15 17:00:13 +02006040#ifdef CONFIG_SCHED_DEBUG
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006041
Mike Travisf6630112009-11-17 18:22:15 -06006042static __read_mostly int sched_domain_debug_enabled;
6043
6044static int __init sched_domain_debug_setup(char *str)
6045{
6046 sched_domain_debug_enabled = 1;
6047
6048 return 0;
6049}
6050early_param("sched_debug", sched_domain_debug_setup);
6051
Mike Travis7c16ec52008-04-04 18:11:11 -07006052static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
Rusty Russell96f874e22008-11-25 02:35:14 +10306053 struct cpumask *groupmask)
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006054{
6055 struct sched_group *group = sd->groups;
Mike Travis434d53b2008-04-04 18:11:04 -07006056 char str[256];
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006057
Rusty Russell968ea6d2008-12-13 21:55:51 +10306058 cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd));
Rusty Russell96f874e22008-11-25 02:35:14 +10306059 cpumask_clear(groupmask);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006060
6061 printk(KERN_DEBUG "%*s domain %d: ", level, "", level);
6062
6063 if (!(sd->flags & SD_LOAD_BALANCE)) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006064 printk("does not load-balance\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006065 if (sd->parent)
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006066 printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
6067 " has parent");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006068 return -1;
6069 }
6070
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006071 printk(KERN_CONT "span %s level %s\n", str, sd->name);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006072
Rusty Russell758b2cd2008-11-25 02:35:04 +10306073 if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006074 printk(KERN_ERR "ERROR: domain->span does not contain "
6075 "CPU%d\n", cpu);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006076 }
Rusty Russell758b2cd2008-11-25 02:35:04 +10306077 if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006078 printk(KERN_ERR "ERROR: domain->groups does not contain"
6079 " CPU%d\n", cpu);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006080 }
6081
6082 printk(KERN_DEBUG "%*s groups:", level + 1, "");
6083 do {
6084 if (!group) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006085 printk("\n");
6086 printk(KERN_ERR "ERROR: group is NULL\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006087 break;
6088 }
6089
Peter Zijlstra18a38852009-09-01 10:34:39 +02006090 if (!group->cpu_power) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006091 printk(KERN_CONT "\n");
6092 printk(KERN_ERR "ERROR: domain->cpu_power not "
6093 "set\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006094 break;
6095 }
6096
Rusty Russell758b2cd2008-11-25 02:35:04 +10306097 if (!cpumask_weight(sched_group_cpus(group))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006098 printk(KERN_CONT "\n");
6099 printk(KERN_ERR "ERROR: empty group\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006100 break;
6101 }
6102
Rusty Russell758b2cd2008-11-25 02:35:04 +10306103 if (cpumask_intersects(groupmask, sched_group_cpus(group))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006104 printk(KERN_CONT "\n");
6105 printk(KERN_ERR "ERROR: repeated CPUs\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006106 break;
6107 }
6108
Rusty Russell758b2cd2008-11-25 02:35:04 +10306109 cpumask_or(groupmask, groupmask, sched_group_cpus(group));
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006110
Rusty Russell968ea6d2008-12-13 21:55:51 +10306111 cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
Gautham R Shenoy381512c2009-04-14 09:09:36 +05306112
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006113 printk(KERN_CONT " %s", str);
Peter Zijlstra18a38852009-09-01 10:34:39 +02006114 if (group->cpu_power != SCHED_LOAD_SCALE) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006115 printk(KERN_CONT " (cpu_power = %d)",
6116 group->cpu_power);
Gautham R Shenoy381512c2009-04-14 09:09:36 +05306117 }
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006118
6119 group = group->next;
6120 } while (group != sd->groups);
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006121 printk(KERN_CONT "\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006122
Rusty Russell758b2cd2008-11-25 02:35:04 +10306123 if (!cpumask_equal(sched_domain_span(sd), groupmask))
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006124 printk(KERN_ERR "ERROR: groups don't span domain->span\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006125
Rusty Russell758b2cd2008-11-25 02:35:04 +10306126 if (sd->parent &&
6127 !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006128 printk(KERN_ERR "ERROR: parent span is not a superset "
6129 "of domain->span\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006130 return 0;
6131}
6132
Linus Torvalds1da177e2005-04-16 15:20:36 -07006133static void sched_domain_debug(struct sched_domain *sd, int cpu)
6134{
Rusty Russelld5dd3db2008-11-25 02:35:12 +10306135 cpumask_var_t groupmask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006136 int level = 0;
6137
Mike Travisf6630112009-11-17 18:22:15 -06006138 if (!sched_domain_debug_enabled)
6139 return;
6140
Nick Piggin41c7ce92005-06-25 14:57:24 -07006141 if (!sd) {
6142 printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
6143 return;
6144 }
6145
Linus Torvalds1da177e2005-04-16 15:20:36 -07006146 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
6147
Rusty Russelld5dd3db2008-11-25 02:35:12 +10306148 if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) {
Mike Travis7c16ec52008-04-04 18:11:11 -07006149 printk(KERN_DEBUG "Cannot load-balance (out of memory)\n");
6150 return;
6151 }
6152
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006153 for (;;) {
Mike Travis7c16ec52008-04-04 18:11:11 -07006154 if (sched_domain_debug_one(sd, cpu, level, groupmask))
Linus Torvalds1da177e2005-04-16 15:20:36 -07006155 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006156 level++;
6157 sd = sd->parent;
Miguel Ojeda Sandonis33859f72006-12-10 02:20:38 -08006158 if (!sd)
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02006159 break;
6160 }
Rusty Russelld5dd3db2008-11-25 02:35:12 +10306161 free_cpumask_var(groupmask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006162}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006163#else /* !CONFIG_SCHED_DEBUG */
Ingo Molnar48f24c42006-07-03 00:25:40 -07006164# define sched_domain_debug(sd, cpu) do { } while (0)
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006165#endif /* CONFIG_SCHED_DEBUG */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006166
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006167static int sd_degenerate(struct sched_domain *sd)
Suresh Siddha245af2c2005-06-25 14:57:25 -07006168{
Rusty Russell758b2cd2008-11-25 02:35:04 +10306169 if (cpumask_weight(sched_domain_span(sd)) == 1)
Suresh Siddha245af2c2005-06-25 14:57:25 -07006170 return 1;
6171
6172 /* Following flags need at least 2 groups */
6173 if (sd->flags & (SD_LOAD_BALANCE |
6174 SD_BALANCE_NEWIDLE |
6175 SD_BALANCE_FORK |
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006176 SD_BALANCE_EXEC |
6177 SD_SHARE_CPUPOWER |
6178 SD_SHARE_PKG_RESOURCES)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006179 if (sd->groups != sd->groups->next)
6180 return 0;
6181 }
6182
6183 /* Following flags don't use groups */
Peter Zijlstrac88d5912009-09-10 13:50:02 +02006184 if (sd->flags & (SD_WAKE_AFFINE))
Suresh Siddha245af2c2005-06-25 14:57:25 -07006185 return 0;
6186
6187 return 1;
6188}
6189
Ingo Molnar48f24c42006-07-03 00:25:40 -07006190static int
6191sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
Suresh Siddha245af2c2005-06-25 14:57:25 -07006192{
6193 unsigned long cflags = sd->flags, pflags = parent->flags;
6194
6195 if (sd_degenerate(parent))
6196 return 1;
6197
Rusty Russell758b2cd2008-11-25 02:35:04 +10306198 if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
Suresh Siddha245af2c2005-06-25 14:57:25 -07006199 return 0;
6200
Suresh Siddha245af2c2005-06-25 14:57:25 -07006201 /* Flags needing groups don't count if only 1 group in parent */
6202 if (parent->groups == parent->groups->next) {
6203 pflags &= ~(SD_LOAD_BALANCE |
6204 SD_BALANCE_NEWIDLE |
6205 SD_BALANCE_FORK |
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006206 SD_BALANCE_EXEC |
6207 SD_SHARE_CPUPOWER |
6208 SD_SHARE_PKG_RESOURCES);
Ken Chen54364992008-12-07 18:47:37 -08006209 if (nr_node_ids == 1)
6210 pflags &= ~SD_SERIALIZE;
Suresh Siddha245af2c2005-06-25 14:57:25 -07006211 }
6212 if (~cflags & pflags)
6213 return 0;
6214
6215 return 1;
6216}
6217
Rusty Russellc6c49272008-11-25 02:35:05 +10306218static void free_rootdomain(struct root_domain *rd)
6219{
Peter Zijlstra047106a2009-11-16 10:28:09 +01006220 synchronize_sched();
6221
Rusty Russell68e74562008-11-25 02:35:13 +10306222 cpupri_cleanup(&rd->cpupri);
6223
Rusty Russellc6c49272008-11-25 02:35:05 +10306224 free_cpumask_var(rd->rto_mask);
6225 free_cpumask_var(rd->online);
6226 free_cpumask_var(rd->span);
6227 kfree(rd);
6228}
6229
Gregory Haskins57d885f2008-01-25 21:08:18 +01006230static void rq_attach_root(struct rq *rq, struct root_domain *rd)
6231{
Ingo Molnara0490fa2009-02-12 11:35:40 +01006232 struct root_domain *old_rd = NULL;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006233 unsigned long flags;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006234
Thomas Gleixner05fa7852009-11-17 14:28:38 +01006235 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006236
6237 if (rq->rd) {
Ingo Molnara0490fa2009-02-12 11:35:40 +01006238 old_rd = rq->rd;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006239
Rusty Russellc6c49272008-11-25 02:35:05 +10306240 if (cpumask_test_cpu(rq->cpu, old_rd->online))
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -04006241 set_rq_offline(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006242
Rusty Russellc6c49272008-11-25 02:35:05 +10306243 cpumask_clear_cpu(rq->cpu, old_rd->span);
Gregory Haskinsdc938522008-01-25 21:08:26 +01006244
Ingo Molnara0490fa2009-02-12 11:35:40 +01006245 /*
6246 * If we dont want to free the old_rt yet then
6247 * set old_rd to NULL to skip the freeing later
6248 * in this function:
6249 */
6250 if (!atomic_dec_and_test(&old_rd->refcount))
6251 old_rd = NULL;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006252 }
6253
6254 atomic_inc(&rd->refcount);
6255 rq->rd = rd;
6256
Rusty Russellc6c49272008-11-25 02:35:05 +10306257 cpumask_set_cpu(rq->cpu, rd->span);
Gregory Haskins00aec932009-07-30 10:57:23 -04006258 if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -04006259 set_rq_online(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006260
Thomas Gleixner05fa7852009-11-17 14:28:38 +01006261 raw_spin_unlock_irqrestore(&rq->lock, flags);
Ingo Molnara0490fa2009-02-12 11:35:40 +01006262
6263 if (old_rd)
6264 free_rootdomain(old_rd);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006265}
6266
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006267static int init_rootdomain(struct root_domain *rd)
Gregory Haskins57d885f2008-01-25 21:08:18 +01006268{
6269 memset(rd, 0, sizeof(*rd));
6270
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006271 if (!alloc_cpumask_var(&rd->span, GFP_KERNEL))
Li Zefan0c910d22009-01-06 17:39:06 +08006272 goto out;
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006273 if (!alloc_cpumask_var(&rd->online, GFP_KERNEL))
Rusty Russellc6c49272008-11-25 02:35:05 +10306274 goto free_span;
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006275 if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
Rusty Russellc6c49272008-11-25 02:35:05 +10306276 goto free_online;
Gregory Haskins6e0534f2008-05-12 21:21:01 +02006277
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006278 if (cpupri_init(&rd->cpupri) != 0)
Rusty Russell68e74562008-11-25 02:35:13 +10306279 goto free_rto_mask;
Rusty Russellc6c49272008-11-25 02:35:05 +10306280 return 0;
6281
Rusty Russell68e74562008-11-25 02:35:13 +10306282free_rto_mask:
6283 free_cpumask_var(rd->rto_mask);
Rusty Russellc6c49272008-11-25 02:35:05 +10306284free_online:
6285 free_cpumask_var(rd->online);
6286free_span:
6287 free_cpumask_var(rd->span);
Li Zefan0c910d22009-01-06 17:39:06 +08006288out:
Rusty Russellc6c49272008-11-25 02:35:05 +10306289 return -ENOMEM;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006290}
6291
6292static void init_defrootdomain(void)
6293{
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006294 init_rootdomain(&def_root_domain);
Rusty Russellc6c49272008-11-25 02:35:05 +10306295
Gregory Haskins57d885f2008-01-25 21:08:18 +01006296 atomic_set(&def_root_domain.refcount, 1);
6297}
6298
Gregory Haskinsdc938522008-01-25 21:08:26 +01006299static struct root_domain *alloc_rootdomain(void)
Gregory Haskins57d885f2008-01-25 21:08:18 +01006300{
6301 struct root_domain *rd;
6302
6303 rd = kmalloc(sizeof(*rd), GFP_KERNEL);
6304 if (!rd)
6305 return NULL;
6306
Pekka Enberg68c38fc2010-07-15 23:18:22 +03006307 if (init_rootdomain(rd) != 0) {
Rusty Russellc6c49272008-11-25 02:35:05 +10306308 kfree(rd);
6309 return NULL;
6310 }
Gregory Haskins57d885f2008-01-25 21:08:18 +01006311
6312 return rd;
6313}
6314
Linus Torvalds1da177e2005-04-16 15:20:36 -07006315/*
Ingo Molnar0eab9142008-01-25 21:08:19 +01006316 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
Linus Torvalds1da177e2005-04-16 15:20:36 -07006317 * hold the hotplug lock.
6318 */
Ingo Molnar0eab9142008-01-25 21:08:19 +01006319static void
6320cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006321{
Ingo Molnar70b97a72006-07-03 00:25:42 -07006322 struct rq *rq = cpu_rq(cpu);
Suresh Siddha245af2c2005-06-25 14:57:25 -07006323 struct sched_domain *tmp;
6324
Peter Zijlstra669c55e2010-04-16 14:59:29 +02006325 for (tmp = sd; tmp; tmp = tmp->parent)
6326 tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
6327
Suresh Siddha245af2c2005-06-25 14:57:25 -07006328 /* Remove the sched domains which do not contribute to scheduling. */
Li Zefanf29c9b12008-11-06 09:45:16 +08006329 for (tmp = sd; tmp; ) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006330 struct sched_domain *parent = tmp->parent;
6331 if (!parent)
6332 break;
Li Zefanf29c9b12008-11-06 09:45:16 +08006333
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006334 if (sd_parent_degenerate(tmp, parent)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006335 tmp->parent = parent->parent;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006336 if (parent->parent)
6337 parent->parent->child = tmp;
Li Zefanf29c9b12008-11-06 09:45:16 +08006338 } else
6339 tmp = tmp->parent;
Suresh Siddha245af2c2005-06-25 14:57:25 -07006340 }
6341
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006342 if (sd && sd_degenerate(sd)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006343 sd = sd->parent;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006344 if (sd)
6345 sd->child = NULL;
6346 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006347
6348 sched_domain_debug(sd, cpu);
6349
Gregory Haskins57d885f2008-01-25 21:08:18 +01006350 rq_attach_root(rq, rd);
Nick Piggin674311d2005-06-25 14:57:27 -07006351 rcu_assign_pointer(rq->sd, sd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006352}
6353
6354/* cpus with isolated domains */
Rusty Russelldcc30a32008-11-25 02:35:12 +10306355static cpumask_var_t cpu_isolated_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006356
6357/* Setup the mask of cpus configured for isolated domains */
6358static int __init isolated_cpu_setup(char *str)
6359{
Rusty Russellbdddd292009-12-02 14:09:16 +10306360 alloc_bootmem_cpumask_var(&cpu_isolated_map);
Rusty Russell968ea6d2008-12-13 21:55:51 +10306361 cpulist_parse(str, cpu_isolated_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006362 return 1;
6363}
6364
Ingo Molnar8927f492007-10-15 17:00:13 +02006365__setup("isolcpus=", isolated_cpu_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006366
6367/*
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006368 * init_sched_build_groups takes the cpumask we wish to span, and a pointer
6369 * to a function which identifies what group(along with sched group) a CPU
Rusty Russell96f874e22008-11-25 02:35:14 +10306370 * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids
6371 * (due to the fact that we keep track of groups covered with a struct cpumask).
Linus Torvalds1da177e2005-04-16 15:20:36 -07006372 *
6373 * init_sched_build_groups will build a circular linked list of the groups
6374 * covered by the given span, and will set each group's ->cpumask correctly,
6375 * and ->cpu_power to 0.
6376 */
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006377static void
Rusty Russell96f874e22008-11-25 02:35:14 +10306378init_sched_build_groups(const struct cpumask *span,
6379 const struct cpumask *cpu_map,
6380 int (*group_fn)(int cpu, const struct cpumask *cpu_map,
Mike Travis7c16ec52008-04-04 18:11:11 -07006381 struct sched_group **sg,
Rusty Russell96f874e22008-11-25 02:35:14 +10306382 struct cpumask *tmpmask),
6383 struct cpumask *covered, struct cpumask *tmpmask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006384{
6385 struct sched_group *first = NULL, *last = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006386 int i;
6387
Rusty Russell96f874e22008-11-25 02:35:14 +10306388 cpumask_clear(covered);
Mike Travis7c16ec52008-04-04 18:11:11 -07006389
Rusty Russellabcd0832008-11-25 02:35:02 +10306390 for_each_cpu(i, span) {
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006391 struct sched_group *sg;
Mike Travis7c16ec52008-04-04 18:11:11 -07006392 int group = group_fn(i, cpu_map, &sg, tmpmask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006393 int j;
6394
Rusty Russell758b2cd2008-11-25 02:35:04 +10306395 if (cpumask_test_cpu(i, covered))
Linus Torvalds1da177e2005-04-16 15:20:36 -07006396 continue;
6397
Rusty Russell758b2cd2008-11-25 02:35:04 +10306398 cpumask_clear(sched_group_cpus(sg));
Peter Zijlstra18a38852009-09-01 10:34:39 +02006399 sg->cpu_power = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006400
Rusty Russellabcd0832008-11-25 02:35:02 +10306401 for_each_cpu(j, span) {
Mike Travis7c16ec52008-04-04 18:11:11 -07006402 if (group_fn(j, cpu_map, NULL, tmpmask) != group)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006403 continue;
6404
Rusty Russell96f874e22008-11-25 02:35:14 +10306405 cpumask_set_cpu(j, covered);
Rusty Russell758b2cd2008-11-25 02:35:04 +10306406 cpumask_set_cpu(j, sched_group_cpus(sg));
Linus Torvalds1da177e2005-04-16 15:20:36 -07006407 }
6408 if (!first)
6409 first = sg;
6410 if (last)
6411 last->next = sg;
6412 last = sg;
6413 }
6414 last->next = first;
6415}
6416
John Hawkes9c1cfda2005-09-06 15:18:14 -07006417#define SD_NODES_PER_DOMAIN 16
Linus Torvalds1da177e2005-04-16 15:20:36 -07006418
John Hawkes9c1cfda2005-09-06 15:18:14 -07006419#ifdef CONFIG_NUMA
akpm@osdl.org198e2f12006-01-12 01:05:30 -08006420
John Hawkes9c1cfda2005-09-06 15:18:14 -07006421/**
6422 * find_next_best_node - find the next node to include in a sched_domain
6423 * @node: node whose sched_domain we're building
6424 * @used_nodes: nodes already in the sched_domain
6425 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006426 * Find the next node to include in a given scheduling domain. Simply
John Hawkes9c1cfda2005-09-06 15:18:14 -07006427 * finds the closest node not already in the @used_nodes map.
6428 *
6429 * Should use nodemask_t.
6430 */
Mike Travisc5f59f02008-04-04 18:11:10 -07006431static int find_next_best_node(int node, nodemask_t *used_nodes)
John Hawkes9c1cfda2005-09-06 15:18:14 -07006432{
6433 int i, n, val, min_val, best_node = 0;
6434
6435 min_val = INT_MAX;
6436
Mike Travis076ac2a2008-05-12 21:21:12 +02006437 for (i = 0; i < nr_node_ids; i++) {
John Hawkes9c1cfda2005-09-06 15:18:14 -07006438 /* Start at @node */
Mike Travis076ac2a2008-05-12 21:21:12 +02006439 n = (node + i) % nr_node_ids;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006440
6441 if (!nr_cpus_node(n))
6442 continue;
6443
6444 /* Skip already used nodes */
Mike Travisc5f59f02008-04-04 18:11:10 -07006445 if (node_isset(n, *used_nodes))
John Hawkes9c1cfda2005-09-06 15:18:14 -07006446 continue;
6447
6448 /* Simple min distance search */
6449 val = node_distance(node, n);
6450
6451 if (val < min_val) {
6452 min_val = val;
6453 best_node = n;
6454 }
6455 }
6456
Mike Travisc5f59f02008-04-04 18:11:10 -07006457 node_set(best_node, *used_nodes);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006458 return best_node;
6459}
6460
6461/**
6462 * sched_domain_node_span - get a cpumask for a node's sched_domain
6463 * @node: node whose cpumask we're constructing
Randy Dunlap73486722008-04-22 10:07:22 -07006464 * @span: resulting cpumask
John Hawkes9c1cfda2005-09-06 15:18:14 -07006465 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006466 * Given a node, construct a good cpumask for its sched_domain to span. It
John Hawkes9c1cfda2005-09-06 15:18:14 -07006467 * should be one that prevents unnecessary balancing, but also spreads tasks
6468 * out optimally.
6469 */
Rusty Russell96f874e22008-11-25 02:35:14 +10306470static void sched_domain_node_span(int node, struct cpumask *span)
John Hawkes9c1cfda2005-09-06 15:18:14 -07006471{
Mike Travisc5f59f02008-04-04 18:11:10 -07006472 nodemask_t used_nodes;
Ingo Molnar48f24c42006-07-03 00:25:40 -07006473 int i;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006474
Mike Travis6ca09df2008-12-31 18:08:45 -08006475 cpumask_clear(span);
Mike Travisc5f59f02008-04-04 18:11:10 -07006476 nodes_clear(used_nodes);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006477
Mike Travis6ca09df2008-12-31 18:08:45 -08006478 cpumask_or(span, span, cpumask_of_node(node));
Mike Travisc5f59f02008-04-04 18:11:10 -07006479 node_set(node, used_nodes);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006480
6481 for (i = 1; i < SD_NODES_PER_DOMAIN; i++) {
Mike Travisc5f59f02008-04-04 18:11:10 -07006482 int next_node = find_next_best_node(node, &used_nodes);
Ingo Molnar48f24c42006-07-03 00:25:40 -07006483
Mike Travis6ca09df2008-12-31 18:08:45 -08006484 cpumask_or(span, span, cpumask_of_node(next_node));
John Hawkes9c1cfda2005-09-06 15:18:14 -07006485 }
John Hawkes9c1cfda2005-09-06 15:18:14 -07006486}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006487#endif /* CONFIG_NUMA */
John Hawkes9c1cfda2005-09-06 15:18:14 -07006488
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006489int sched_smt_power_savings = 0, sched_mc_power_savings = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07006490
John Hawkes9c1cfda2005-09-06 15:18:14 -07006491/*
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306492 * The cpus mask in sched_group and sched_domain hangs off the end.
Ingo Molnar4200efd2009-05-19 09:22:19 +02006493 *
6494 * ( See the the comments in include/linux/sched.h:struct sched_group
6495 * and struct sched_domain. )
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306496 */
6497struct static_sched_group {
6498 struct sched_group sg;
6499 DECLARE_BITMAP(cpus, CONFIG_NR_CPUS);
6500};
6501
6502struct static_sched_domain {
6503 struct sched_domain sd;
6504 DECLARE_BITMAP(span, CONFIG_NR_CPUS);
6505};
6506
Andreas Herrmann49a02c52009-08-18 12:51:52 +02006507struct s_data {
6508#ifdef CONFIG_NUMA
6509 int sd_allnodes;
6510 cpumask_var_t domainspan;
6511 cpumask_var_t covered;
6512 cpumask_var_t notcovered;
6513#endif
6514 cpumask_var_t nodemask;
6515 cpumask_var_t this_sibling_map;
6516 cpumask_var_t this_core_map;
6517 cpumask_var_t send_covered;
6518 cpumask_var_t tmpmask;
6519 struct sched_group **sched_group_nodes;
6520 struct root_domain *rd;
6521};
6522
Andreas Herrmann2109b992009-08-18 12:53:00 +02006523enum s_alloc {
6524 sa_sched_groups = 0,
6525 sa_rootdomain,
6526 sa_tmpmask,
6527 sa_send_covered,
6528 sa_this_core_map,
6529 sa_this_sibling_map,
6530 sa_nodemask,
6531 sa_sched_group_nodes,
6532#ifdef CONFIG_NUMA
6533 sa_notcovered,
6534 sa_covered,
6535 sa_domainspan,
6536#endif
6537 sa_none,
6538};
6539
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306540/*
Ingo Molnar48f24c42006-07-03 00:25:40 -07006541 * SMT sched-domains:
John Hawkes9c1cfda2005-09-06 15:18:14 -07006542 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006543#ifdef CONFIG_SCHED_SMT
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306544static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains);
Tejun Heo1871e522009-10-29 22:34:13 +09006545static DEFINE_PER_CPU(struct static_sched_group, sched_groups);
Ingo Molnar48f24c42006-07-03 00:25:40 -07006546
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006547static int
Rusty Russell96f874e22008-11-25 02:35:14 +10306548cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map,
6549 struct sched_group **sg, struct cpumask *unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006550{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006551 if (sg)
Tejun Heo1871e522009-10-29 22:34:13 +09006552 *sg = &per_cpu(sched_groups, cpu).sg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006553 return cpu;
6554}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006555#endif /* CONFIG_SCHED_SMT */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006556
Ingo Molnar48f24c42006-07-03 00:25:40 -07006557/*
6558 * multi-core sched-domains:
6559 */
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006560#ifdef CONFIG_SCHED_MC
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306561static DEFINE_PER_CPU(struct static_sched_domain, core_domains);
6562static DEFINE_PER_CPU(struct static_sched_group, sched_group_core);
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006563#endif /* CONFIG_SCHED_MC */
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006564
6565#if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT)
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006566static int
Rusty Russell96f874e22008-11-25 02:35:14 +10306567cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
6568 struct sched_group **sg, struct cpumask *mask)
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006569{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006570 int group;
Mike Travis7c16ec52008-04-04 18:11:11 -07006571
Rusty Russellc69fc562009-03-13 14:49:46 +10306572 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
Rusty Russell96f874e22008-11-25 02:35:14 +10306573 group = cpumask_first(mask);
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006574 if (sg)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306575 *sg = &per_cpu(sched_group_core, group).sg;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006576 return group;
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006577}
6578#elif defined(CONFIG_SCHED_MC)
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006579static int
Rusty Russell96f874e22008-11-25 02:35:14 +10306580cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
6581 struct sched_group **sg, struct cpumask *unused)
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006582{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006583 if (sg)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306584 *sg = &per_cpu(sched_group_core, cpu).sg;
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006585 return cpu;
6586}
6587#endif
6588
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306589static DEFINE_PER_CPU(struct static_sched_domain, phys_domains);
6590static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys);
Ingo Molnar48f24c42006-07-03 00:25:40 -07006591
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006592static int
Rusty Russell96f874e22008-11-25 02:35:14 +10306593cpu_to_phys_group(int cpu, const struct cpumask *cpu_map,
6594 struct sched_group **sg, struct cpumask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006595{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006596 int group;
Ingo Molnar48f24c42006-07-03 00:25:40 -07006597#ifdef CONFIG_SCHED_MC
Mike Travis6ca09df2008-12-31 18:08:45 -08006598 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
Rusty Russell96f874e22008-11-25 02:35:14 +10306599 group = cpumask_first(mask);
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006600#elif defined(CONFIG_SCHED_SMT)
Rusty Russellc69fc562009-03-13 14:49:46 +10306601 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
Rusty Russell96f874e22008-11-25 02:35:14 +10306602 group = cpumask_first(mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006603#else
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006604 group = cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006605#endif
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006606 if (sg)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306607 *sg = &per_cpu(sched_group_phys, group).sg;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006608 return group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006609}
6610
6611#ifdef CONFIG_NUMA
John Hawkes9c1cfda2005-09-06 15:18:14 -07006612/*
6613 * The init_sched_build_groups can't handle what we want to do with node
6614 * groups, so roll our own. Now each node has its own list of groups which
6615 * gets dynamically allocated.
6616 */
Rusty Russell62ea9ce2009-01-11 01:04:16 +01006617static DEFINE_PER_CPU(struct static_sched_domain, node_domains);
Mike Travis434d53b2008-04-04 18:11:04 -07006618static struct sched_group ***sched_group_nodes_bycpu;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006619
Rusty Russell62ea9ce2009-01-11 01:04:16 +01006620static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains);
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306621static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006622
Rusty Russell96f874e22008-11-25 02:35:14 +10306623static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map,
6624 struct sched_group **sg,
6625 struct cpumask *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006626{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006627 int group;
6628
Mike Travis6ca09df2008-12-31 18:08:45 -08006629 cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map);
Rusty Russell96f874e22008-11-25 02:35:14 +10306630 group = cpumask_first(nodemask);
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006631
6632 if (sg)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306633 *sg = &per_cpu(sched_group_allnodes, group).sg;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006634 return group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006635}
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006636
Siddha, Suresh B08069032006-03-27 01:15:23 -08006637static void init_numa_sched_groups_power(struct sched_group *group_head)
6638{
6639 struct sched_group *sg = group_head;
6640 int j;
6641
6642 if (!sg)
6643 return;
Andi Kleen3a5c3592007-10-15 17:00:14 +02006644 do {
Rusty Russell758b2cd2008-11-25 02:35:04 +10306645 for_each_cpu(j, sched_group_cpus(sg)) {
Andi Kleen3a5c3592007-10-15 17:00:14 +02006646 struct sched_domain *sd;
Siddha, Suresh B08069032006-03-27 01:15:23 -08006647
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306648 sd = &per_cpu(phys_domains, j).sd;
Miao Xie13318a72009-04-15 09:59:10 +08006649 if (j != group_first_cpu(sd->groups)) {
Andi Kleen3a5c3592007-10-15 17:00:14 +02006650 /*
6651 * Only add "power" once for each
6652 * physical package.
6653 */
6654 continue;
6655 }
6656
Peter Zijlstra18a38852009-09-01 10:34:39 +02006657 sg->cpu_power += sd->groups->cpu_power;
Siddha, Suresh B08069032006-03-27 01:15:23 -08006658 }
Andi Kleen3a5c3592007-10-15 17:00:14 +02006659 sg = sg->next;
6660 } while (sg != group_head);
Siddha, Suresh B08069032006-03-27 01:15:23 -08006661}
Andreas Herrmann0601a882009-08-18 13:01:11 +02006662
6663static int build_numa_sched_groups(struct s_data *d,
6664 const struct cpumask *cpu_map, int num)
6665{
6666 struct sched_domain *sd;
6667 struct sched_group *sg, *prev;
6668 int n, j;
6669
6670 cpumask_clear(d->covered);
6671 cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map);
6672 if (cpumask_empty(d->nodemask)) {
6673 d->sched_group_nodes[num] = NULL;
6674 goto out;
6675 }
6676
6677 sched_domain_node_span(num, d->domainspan);
6678 cpumask_and(d->domainspan, d->domainspan, cpu_map);
6679
6680 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
6681 GFP_KERNEL, num);
6682 if (!sg) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006683 printk(KERN_WARNING "Can not alloc domain group for node %d\n",
6684 num);
Andreas Herrmann0601a882009-08-18 13:01:11 +02006685 return -ENOMEM;
6686 }
6687 d->sched_group_nodes[num] = sg;
6688
6689 for_each_cpu(j, d->nodemask) {
6690 sd = &per_cpu(node_domains, j).sd;
6691 sd->groups = sg;
6692 }
6693
Peter Zijlstra18a38852009-09-01 10:34:39 +02006694 sg->cpu_power = 0;
Andreas Herrmann0601a882009-08-18 13:01:11 +02006695 cpumask_copy(sched_group_cpus(sg), d->nodemask);
6696 sg->next = sg;
6697 cpumask_or(d->covered, d->covered, d->nodemask);
6698
6699 prev = sg;
6700 for (j = 0; j < nr_node_ids; j++) {
6701 n = (num + j) % nr_node_ids;
6702 cpumask_complement(d->notcovered, d->covered);
6703 cpumask_and(d->tmpmask, d->notcovered, cpu_map);
6704 cpumask_and(d->tmpmask, d->tmpmask, d->domainspan);
6705 if (cpumask_empty(d->tmpmask))
6706 break;
6707 cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n));
6708 if (cpumask_empty(d->tmpmask))
6709 continue;
6710 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
6711 GFP_KERNEL, num);
6712 if (!sg) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006713 printk(KERN_WARNING
6714 "Can not alloc domain group for node %d\n", j);
Andreas Herrmann0601a882009-08-18 13:01:11 +02006715 return -ENOMEM;
6716 }
Peter Zijlstra18a38852009-09-01 10:34:39 +02006717 sg->cpu_power = 0;
Andreas Herrmann0601a882009-08-18 13:01:11 +02006718 cpumask_copy(sched_group_cpus(sg), d->tmpmask);
6719 sg->next = prev->next;
6720 cpumask_or(d->covered, d->covered, d->tmpmask);
6721 prev->next = sg;
6722 prev = sg;
6723 }
6724out:
6725 return 0;
6726}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006727#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006728
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006729#ifdef CONFIG_NUMA
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006730/* Free memory allocated for various sched_group structures */
Rusty Russell96f874e22008-11-25 02:35:14 +10306731static void free_sched_groups(const struct cpumask *cpu_map,
6732 struct cpumask *nodemask)
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006733{
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006734 int cpu, i;
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006735
Rusty Russellabcd0832008-11-25 02:35:02 +10306736 for_each_cpu(cpu, cpu_map) {
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006737 struct sched_group **sched_group_nodes
6738 = sched_group_nodes_bycpu[cpu];
6739
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006740 if (!sched_group_nodes)
6741 continue;
6742
Mike Travis076ac2a2008-05-12 21:21:12 +02006743 for (i = 0; i < nr_node_ids; i++) {
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006744 struct sched_group *oldsg, *sg = sched_group_nodes[i];
6745
Mike Travis6ca09df2008-12-31 18:08:45 -08006746 cpumask_and(nodemask, cpumask_of_node(i), cpu_map);
Rusty Russell96f874e22008-11-25 02:35:14 +10306747 if (cpumask_empty(nodemask))
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006748 continue;
6749
6750 if (sg == NULL)
6751 continue;
6752 sg = sg->next;
6753next_sg:
6754 oldsg = sg;
6755 sg = sg->next;
6756 kfree(oldsg);
6757 if (oldsg != sched_group_nodes[i])
6758 goto next_sg;
6759 }
6760 kfree(sched_group_nodes);
6761 sched_group_nodes_bycpu[cpu] = NULL;
6762 }
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006763}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006764#else /* !CONFIG_NUMA */
Rusty Russell96f874e22008-11-25 02:35:14 +10306765static void free_sched_groups(const struct cpumask *cpu_map,
6766 struct cpumask *nodemask)
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006767{
6768}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006769#endif /* CONFIG_NUMA */
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006770
Linus Torvalds1da177e2005-04-16 15:20:36 -07006771/*
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006772 * Initialize sched groups cpu_power.
6773 *
6774 * cpu_power indicates the capacity of sched group, which is used while
6775 * distributing the load between different sched groups in a sched domain.
6776 * Typically cpu_power for all the groups in a sched domain will be same unless
6777 * there are asymmetries in the topology. If there are asymmetries, group
6778 * having more cpu_power will pickup more load compared to the group having
6779 * less cpu_power.
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006780 */
6781static void init_sched_groups_power(int cpu, struct sched_domain *sd)
6782{
6783 struct sched_domain *child;
6784 struct sched_group *group;
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006785 long power;
6786 int weight;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006787
6788 WARN_ON(!sd || !sd->groups);
6789
Miao Xie13318a72009-04-15 09:59:10 +08006790 if (cpu != group_first_cpu(sd->groups))
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006791 return;
6792
6793 child = sd->child;
6794
Peter Zijlstra18a38852009-09-01 10:34:39 +02006795 sd->groups->cpu_power = 0;
Eric Dumazet5517d862007-05-08 00:32:57 -07006796
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006797 if (!child) {
6798 power = SCHED_LOAD_SCALE;
6799 weight = cpumask_weight(sched_domain_span(sd));
6800 /*
6801 * SMT siblings share the power of a single core.
Peter Zijlstraa52bfd732009-09-01 10:34:35 +02006802 * Usually multiple threads get a better yield out of
6803 * that one core than a single thread would have,
6804 * reflect that in sd->smt_gain.
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006805 */
Peter Zijlstraa52bfd732009-09-01 10:34:35 +02006806 if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) {
6807 power *= sd->smt_gain;
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006808 power /= weight;
Peter Zijlstraa52bfd732009-09-01 10:34:35 +02006809 power >>= SCHED_LOAD_SHIFT;
6810 }
Peter Zijlstra18a38852009-09-01 10:34:39 +02006811 sd->groups->cpu_power += power;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006812 return;
6813 }
6814
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006815 /*
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006816 * Add cpu_power of each child group to this groups cpu_power.
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006817 */
6818 group = child->groups;
6819 do {
Peter Zijlstra18a38852009-09-01 10:34:39 +02006820 sd->groups->cpu_power += group->cpu_power;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006821 group = group->next;
6822 } while (group != child->groups);
6823}
6824
6825/*
Mike Travis7c16ec52008-04-04 18:11:11 -07006826 * Initializers for schedule domains
6827 * Non-inlined to reduce accumulated stack pressure in build_sched_domains()
6828 */
6829
Ingo Molnara5d8c342008-10-09 11:35:51 +02006830#ifdef CONFIG_SCHED_DEBUG
6831# define SD_INIT_NAME(sd, type) sd->name = #type
6832#else
6833# define SD_INIT_NAME(sd, type) do { } while (0)
6834#endif
6835
Mike Travis7c16ec52008-04-04 18:11:11 -07006836#define SD_INIT(sd, type) sd_init_##type(sd)
Ingo Molnara5d8c342008-10-09 11:35:51 +02006837
Mike Travis7c16ec52008-04-04 18:11:11 -07006838#define SD_INIT_FUNC(type) \
6839static noinline void sd_init_##type(struct sched_domain *sd) \
6840{ \
6841 memset(sd, 0, sizeof(*sd)); \
6842 *sd = SD_##type##_INIT; \
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006843 sd->level = SD_LV_##type; \
Ingo Molnara5d8c342008-10-09 11:35:51 +02006844 SD_INIT_NAME(sd, type); \
Mike Travis7c16ec52008-04-04 18:11:11 -07006845}
6846
6847SD_INIT_FUNC(CPU)
6848#ifdef CONFIG_NUMA
6849 SD_INIT_FUNC(ALLNODES)
6850 SD_INIT_FUNC(NODE)
6851#endif
6852#ifdef CONFIG_SCHED_SMT
6853 SD_INIT_FUNC(SIBLING)
6854#endif
6855#ifdef CONFIG_SCHED_MC
6856 SD_INIT_FUNC(MC)
6857#endif
6858
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006859static int default_relax_domain_level = -1;
6860
6861static int __init setup_relax_domain_level(char *str)
6862{
Li Zefan30e0e172008-05-13 10:27:17 +08006863 unsigned long val;
6864
6865 val = simple_strtoul(str, NULL, 0);
6866 if (val < SD_LV_MAX)
6867 default_relax_domain_level = val;
6868
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006869 return 1;
6870}
6871__setup("relax_domain_level=", setup_relax_domain_level);
6872
6873static void set_domain_attribute(struct sched_domain *sd,
6874 struct sched_domain_attr *attr)
6875{
6876 int request;
6877
6878 if (!attr || attr->relax_domain_level < 0) {
6879 if (default_relax_domain_level < 0)
6880 return;
6881 else
6882 request = default_relax_domain_level;
6883 } else
6884 request = attr->relax_domain_level;
6885 if (request < sd->level) {
6886 /* turn off idle balance on this domain */
Peter Zijlstrac88d5912009-09-10 13:50:02 +02006887 sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006888 } else {
6889 /* turn on idle balance on this domain */
Peter Zijlstrac88d5912009-09-10 13:50:02 +02006890 sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006891 }
6892}
6893
Andreas Herrmann2109b992009-08-18 12:53:00 +02006894static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
6895 const struct cpumask *cpu_map)
6896{
6897 switch (what) {
6898 case sa_sched_groups:
6899 free_sched_groups(cpu_map, d->tmpmask); /* fall through */
6900 d->sched_group_nodes = NULL;
6901 case sa_rootdomain:
6902 free_rootdomain(d->rd); /* fall through */
6903 case sa_tmpmask:
6904 free_cpumask_var(d->tmpmask); /* fall through */
6905 case sa_send_covered:
6906 free_cpumask_var(d->send_covered); /* fall through */
6907 case sa_this_core_map:
6908 free_cpumask_var(d->this_core_map); /* fall through */
6909 case sa_this_sibling_map:
6910 free_cpumask_var(d->this_sibling_map); /* fall through */
6911 case sa_nodemask:
6912 free_cpumask_var(d->nodemask); /* fall through */
6913 case sa_sched_group_nodes:
6914#ifdef CONFIG_NUMA
6915 kfree(d->sched_group_nodes); /* fall through */
6916 case sa_notcovered:
6917 free_cpumask_var(d->notcovered); /* fall through */
6918 case sa_covered:
6919 free_cpumask_var(d->covered); /* fall through */
6920 case sa_domainspan:
6921 free_cpumask_var(d->domainspan); /* fall through */
6922#endif
6923 case sa_none:
6924 break;
6925 }
6926}
6927
6928static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
6929 const struct cpumask *cpu_map)
6930{
6931#ifdef CONFIG_NUMA
6932 if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL))
6933 return sa_none;
6934 if (!alloc_cpumask_var(&d->covered, GFP_KERNEL))
6935 return sa_domainspan;
6936 if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL))
6937 return sa_covered;
6938 /* Allocate the per-node list of sched groups */
6939 d->sched_group_nodes = kcalloc(nr_node_ids,
6940 sizeof(struct sched_group *), GFP_KERNEL);
6941 if (!d->sched_group_nodes) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006942 printk(KERN_WARNING "Can not alloc sched group node list\n");
Andreas Herrmann2109b992009-08-18 12:53:00 +02006943 return sa_notcovered;
6944 }
6945 sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes;
6946#endif
6947 if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL))
6948 return sa_sched_group_nodes;
6949 if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL))
6950 return sa_nodemask;
6951 if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL))
6952 return sa_this_sibling_map;
6953 if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL))
6954 return sa_this_core_map;
6955 if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL))
6956 return sa_send_covered;
6957 d->rd = alloc_rootdomain();
6958 if (!d->rd) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006959 printk(KERN_WARNING "Cannot alloc root domain\n");
Andreas Herrmann2109b992009-08-18 12:53:00 +02006960 return sa_tmpmask;
6961 }
6962 return sa_rootdomain;
6963}
6964
Andreas Herrmann7f4588f2009-08-18 12:54:06 +02006965static struct sched_domain *__build_numa_sched_domains(struct s_data *d,
6966 const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i)
6967{
6968 struct sched_domain *sd = NULL;
6969#ifdef CONFIG_NUMA
6970 struct sched_domain *parent;
6971
6972 d->sd_allnodes = 0;
6973 if (cpumask_weight(cpu_map) >
6974 SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) {
6975 sd = &per_cpu(allnodes_domains, i).sd;
6976 SD_INIT(sd, ALLNODES);
6977 set_domain_attribute(sd, attr);
6978 cpumask_copy(sched_domain_span(sd), cpu_map);
6979 cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask);
6980 d->sd_allnodes = 1;
6981 }
6982 parent = sd;
6983
6984 sd = &per_cpu(node_domains, i).sd;
6985 SD_INIT(sd, NODE);
6986 set_domain_attribute(sd, attr);
6987 sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd));
6988 sd->parent = parent;
6989 if (parent)
6990 parent->child = sd;
6991 cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map);
6992#endif
6993 return sd;
6994}
6995
Andreas Herrmann87cce662009-08-18 12:54:55 +02006996static struct sched_domain *__build_cpu_sched_domain(struct s_data *d,
6997 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
6998 struct sched_domain *parent, int i)
6999{
7000 struct sched_domain *sd;
7001 sd = &per_cpu(phys_domains, i).sd;
7002 SD_INIT(sd, CPU);
7003 set_domain_attribute(sd, attr);
7004 cpumask_copy(sched_domain_span(sd), d->nodemask);
7005 sd->parent = parent;
7006 if (parent)
7007 parent->child = sd;
7008 cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask);
7009 return sd;
7010}
7011
Andreas Herrmann410c4082009-08-18 12:56:14 +02007012static struct sched_domain *__build_mc_sched_domain(struct s_data *d,
7013 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7014 struct sched_domain *parent, int i)
7015{
7016 struct sched_domain *sd = parent;
7017#ifdef CONFIG_SCHED_MC
7018 sd = &per_cpu(core_domains, i).sd;
7019 SD_INIT(sd, MC);
7020 set_domain_attribute(sd, attr);
7021 cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i));
7022 sd->parent = parent;
7023 parent->child = sd;
7024 cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask);
7025#endif
7026 return sd;
7027}
7028
Andreas Herrmannd8173532009-08-18 12:57:03 +02007029static struct sched_domain *__build_smt_sched_domain(struct s_data *d,
7030 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7031 struct sched_domain *parent, int i)
7032{
7033 struct sched_domain *sd = parent;
7034#ifdef CONFIG_SCHED_SMT
7035 sd = &per_cpu(cpu_domains, i).sd;
7036 SD_INIT(sd, SIBLING);
7037 set_domain_attribute(sd, attr);
7038 cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i));
7039 sd->parent = parent;
7040 parent->child = sd;
7041 cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask);
7042#endif
7043 return sd;
7044}
7045
Andreas Herrmann0e8e85c2009-08-18 12:57:51 +02007046static void build_sched_groups(struct s_data *d, enum sched_domain_level l,
7047 const struct cpumask *cpu_map, int cpu)
7048{
7049 switch (l) {
7050#ifdef CONFIG_SCHED_SMT
7051 case SD_LV_SIBLING: /* set up CPU (sibling) groups */
7052 cpumask_and(d->this_sibling_map, cpu_map,
7053 topology_thread_cpumask(cpu));
7054 if (cpu == cpumask_first(d->this_sibling_map))
7055 init_sched_build_groups(d->this_sibling_map, cpu_map,
7056 &cpu_to_cpu_group,
7057 d->send_covered, d->tmpmask);
7058 break;
7059#endif
Andreas Herrmanna2af04c2009-08-18 12:58:38 +02007060#ifdef CONFIG_SCHED_MC
7061 case SD_LV_MC: /* set up multi-core groups */
7062 cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu));
7063 if (cpu == cpumask_first(d->this_core_map))
7064 init_sched_build_groups(d->this_core_map, cpu_map,
7065 &cpu_to_core_group,
7066 d->send_covered, d->tmpmask);
7067 break;
7068#endif
Andreas Herrmann86548092009-08-18 12:59:28 +02007069 case SD_LV_CPU: /* set up physical groups */
7070 cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map);
7071 if (!cpumask_empty(d->nodemask))
7072 init_sched_build_groups(d->nodemask, cpu_map,
7073 &cpu_to_phys_group,
7074 d->send_covered, d->tmpmask);
7075 break;
Andreas Herrmannde616e32009-08-18 13:00:13 +02007076#ifdef CONFIG_NUMA
7077 case SD_LV_ALLNODES:
7078 init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group,
7079 d->send_covered, d->tmpmask);
7080 break;
7081#endif
Andreas Herrmann0e8e85c2009-08-18 12:57:51 +02007082 default:
7083 break;
7084 }
7085}
7086
Mike Travis7c16ec52008-04-04 18:11:11 -07007087/*
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007088 * Build sched domains for a given set of cpus and attach the sched domains
7089 * to the individual cpus
Linus Torvalds1da177e2005-04-16 15:20:36 -07007090 */
Rusty Russell96f874e22008-11-25 02:35:14 +10307091static int __build_sched_domains(const struct cpumask *cpu_map,
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007092 struct sched_domain_attr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007093{
Andreas Herrmann2109b992009-08-18 12:53:00 +02007094 enum s_alloc alloc_state = sa_none;
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007095 struct s_data d;
Andreas Herrmann294b0c92009-08-18 13:02:29 +02007096 struct sched_domain *sd;
Andreas Herrmann2109b992009-08-18 12:53:00 +02007097 int i;
John Hawkesd1b55132005-09-06 15:18:14 -07007098#ifdef CONFIG_NUMA
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007099 d.sd_allnodes = 0;
Rusty Russell3404c8d2008-11-25 02:35:03 +10307100#endif
7101
Andreas Herrmann2109b992009-08-18 12:53:00 +02007102 alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
7103 if (alloc_state != sa_rootdomain)
7104 goto error;
7105 alloc_state = sa_sched_groups;
Mike Travis7c16ec52008-04-04 18:11:11 -07007106
Linus Torvalds1da177e2005-04-16 15:20:36 -07007107 /*
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007108 * Set up domains for cpus specified by the cpu_map.
Linus Torvalds1da177e2005-04-16 15:20:36 -07007109 */
Rusty Russellabcd0832008-11-25 02:35:02 +10307110 for_each_cpu(i, cpu_map) {
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007111 cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)),
7112 cpu_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007113
Andreas Herrmann7f4588f2009-08-18 12:54:06 +02007114 sd = __build_numa_sched_domains(&d, cpu_map, attr, i);
Andreas Herrmann87cce662009-08-18 12:54:55 +02007115 sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i);
Andreas Herrmann410c4082009-08-18 12:56:14 +02007116 sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i);
Andreas Herrmannd8173532009-08-18 12:57:03 +02007117 sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007118 }
7119
Rusty Russellabcd0832008-11-25 02:35:02 +10307120 for_each_cpu(i, cpu_map) {
Andreas Herrmann0e8e85c2009-08-18 12:57:51 +02007121 build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i);
Andreas Herrmanna2af04c2009-08-18 12:58:38 +02007122 build_sched_groups(&d, SD_LV_MC, cpu_map, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007123 }
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08007124
Linus Torvalds1da177e2005-04-16 15:20:36 -07007125 /* Set up physical groups */
Andreas Herrmann86548092009-08-18 12:59:28 +02007126 for (i = 0; i < nr_node_ids; i++)
7127 build_sched_groups(&d, SD_LV_CPU, cpu_map, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007128
7129#ifdef CONFIG_NUMA
7130 /* Set up node groups */
Andreas Herrmannde616e32009-08-18 13:00:13 +02007131 if (d.sd_allnodes)
7132 build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0);
John Hawkes9c1cfda2005-09-06 15:18:14 -07007133
Andreas Herrmann0601a882009-08-18 13:01:11 +02007134 for (i = 0; i < nr_node_ids; i++)
7135 if (build_numa_sched_groups(&d, cpu_map, i))
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07007136 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007137#endif
7138
7139 /* Calculate CPU power for physical packages and nodes */
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007140#ifdef CONFIG_SCHED_SMT
Rusty Russellabcd0832008-11-25 02:35:02 +10307141 for_each_cpu(i, cpu_map) {
Andreas Herrmann294b0c92009-08-18 13:02:29 +02007142 sd = &per_cpu(cpu_domains, i).sd;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07007143 init_sched_groups_power(i, sd);
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007144 }
7145#endif
7146#ifdef CONFIG_SCHED_MC
Rusty Russellabcd0832008-11-25 02:35:02 +10307147 for_each_cpu(i, cpu_map) {
Andreas Herrmann294b0c92009-08-18 13:02:29 +02007148 sd = &per_cpu(core_domains, i).sd;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07007149 init_sched_groups_power(i, sd);
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007150 }
7151#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07007152
Rusty Russellabcd0832008-11-25 02:35:02 +10307153 for_each_cpu(i, cpu_map) {
Andreas Herrmann294b0c92009-08-18 13:02:29 +02007154 sd = &per_cpu(phys_domains, i).sd;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07007155 init_sched_groups_power(i, sd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007156 }
7157
John Hawkes9c1cfda2005-09-06 15:18:14 -07007158#ifdef CONFIG_NUMA
Mike Travis076ac2a2008-05-12 21:21:12 +02007159 for (i = 0; i < nr_node_ids; i++)
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007160 init_numa_sched_groups_power(d.sched_group_nodes[i]);
John Hawkes9c1cfda2005-09-06 15:18:14 -07007161
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007162 if (d.sd_allnodes) {
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08007163 struct sched_group *sg;
Siddha, Suresh Bf712c0c72006-07-30 03:02:59 -07007164
Rusty Russell96f874e22008-11-25 02:35:14 +10307165 cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg,
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007166 d.tmpmask);
Siddha, Suresh Bf712c0c72006-07-30 03:02:59 -07007167 init_numa_sched_groups_power(sg);
7168 }
John Hawkes9c1cfda2005-09-06 15:18:14 -07007169#endif
7170
Linus Torvalds1da177e2005-04-16 15:20:36 -07007171 /* Attach the domains */
Rusty Russellabcd0832008-11-25 02:35:02 +10307172 for_each_cpu(i, cpu_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007173#ifdef CONFIG_SCHED_SMT
Rusty Russell6c99e9a2008-11-25 02:35:04 +10307174 sd = &per_cpu(cpu_domains, i).sd;
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08007175#elif defined(CONFIG_SCHED_MC)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10307176 sd = &per_cpu(core_domains, i).sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007177#else
Rusty Russell6c99e9a2008-11-25 02:35:04 +10307178 sd = &per_cpu(phys_domains, i).sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007179#endif
Andreas Herrmann49a02c52009-08-18 12:51:52 +02007180 cpu_attach_domain(sd, d.rd, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007181 }
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07007182
Andreas Herrmann2109b992009-08-18 12:53:00 +02007183 d.sched_group_nodes = NULL; /* don't free this we still need it */
7184 __free_domain_allocs(&d, sa_tmpmask, cpu_map);
7185 return 0;
Rusty Russell3404c8d2008-11-25 02:35:03 +10307186
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07007187error:
Andreas Herrmann2109b992009-08-18 12:53:00 +02007188 __free_domain_allocs(&d, alloc_state, cpu_map);
7189 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007190}
Paul Jackson029190c2007-10-18 23:40:20 -07007191
Rusty Russell96f874e22008-11-25 02:35:14 +10307192static int build_sched_domains(const struct cpumask *cpu_map)
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007193{
7194 return __build_sched_domains(cpu_map, NULL);
7195}
7196
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307197static cpumask_var_t *doms_cur; /* current sched domains */
Paul Jackson029190c2007-10-18 23:40:20 -07007198static int ndoms_cur; /* number of sched domains in 'doms_cur' */
Ingo Molnar4285f5942008-05-16 17:47:14 +02007199static struct sched_domain_attr *dattr_cur;
7200 /* attribues of custom domains in 'doms_cur' */
Paul Jackson029190c2007-10-18 23:40:20 -07007201
7202/*
7203 * Special case: If a kmalloc of a doms_cur partition (array of
Rusty Russell42128232008-11-25 02:35:12 +10307204 * cpumask) fails, then fallback to a single sched domain,
7205 * as determined by the single cpumask fallback_doms.
Paul Jackson029190c2007-10-18 23:40:20 -07007206 */
Rusty Russell42128232008-11-25 02:35:12 +10307207static cpumask_var_t fallback_doms;
Paul Jackson029190c2007-10-18 23:40:20 -07007208
Heiko Carstensee79d1b2008-12-09 18:49:50 +01007209/*
7210 * arch_update_cpu_topology lets virtualized architectures update the
7211 * cpu core maps. It is supposed to return 1 if the topology changed
7212 * or 0 if it stayed the same.
7213 */
7214int __attribute__((weak)) arch_update_cpu_topology(void)
Heiko Carstens22e52b02008-03-12 18:31:59 +01007215{
Heiko Carstensee79d1b2008-12-09 18:49:50 +01007216 return 0;
Heiko Carstens22e52b02008-03-12 18:31:59 +01007217}
7218
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307219cpumask_var_t *alloc_sched_domains(unsigned int ndoms)
7220{
7221 int i;
7222 cpumask_var_t *doms;
7223
7224 doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL);
7225 if (!doms)
7226 return NULL;
7227 for (i = 0; i < ndoms; i++) {
7228 if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) {
7229 free_sched_domains(doms, i);
7230 return NULL;
7231 }
7232 }
7233 return doms;
7234}
7235
7236void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms)
7237{
7238 unsigned int i;
7239 for (i = 0; i < ndoms; i++)
7240 free_cpumask_var(doms[i]);
7241 kfree(doms);
7242}
7243
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007244/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007245 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
Paul Jackson029190c2007-10-18 23:40:20 -07007246 * For now this just excludes isolated cpus, but could be used to
7247 * exclude other special cases in the future.
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007248 */
Rusty Russell96f874e22008-11-25 02:35:14 +10307249static int arch_init_sched_domains(const struct cpumask *cpu_map)
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007250{
Milton Miller73785472007-10-24 18:23:48 +02007251 int err;
7252
Heiko Carstens22e52b02008-03-12 18:31:59 +01007253 arch_update_cpu_topology();
Paul Jackson029190c2007-10-18 23:40:20 -07007254 ndoms_cur = 1;
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307255 doms_cur = alloc_sched_domains(ndoms_cur);
Paul Jackson029190c2007-10-18 23:40:20 -07007256 if (!doms_cur)
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307257 doms_cur = &fallback_doms;
7258 cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007259 dattr_cur = NULL;
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307260 err = build_sched_domains(doms_cur[0]);
Milton Miller6382bc92007-10-15 17:00:19 +02007261 register_sched_domain_sysctl();
Milton Miller73785472007-10-24 18:23:48 +02007262
7263 return err;
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007264}
7265
Rusty Russell96f874e22008-11-25 02:35:14 +10307266static void arch_destroy_sched_domains(const struct cpumask *cpu_map,
7267 struct cpumask *tmpmask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007268{
Mike Travis7c16ec52008-04-04 18:11:11 -07007269 free_sched_groups(cpu_map, tmpmask);
John Hawkes9c1cfda2005-09-06 15:18:14 -07007270}
Linus Torvalds1da177e2005-04-16 15:20:36 -07007271
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007272/*
7273 * Detach sched domains from a group of cpus specified in cpu_map
7274 * These cpus will now be attached to the NULL domain
7275 */
Rusty Russell96f874e22008-11-25 02:35:14 +10307276static void detach_destroy_domains(const struct cpumask *cpu_map)
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007277{
Rusty Russell96f874e22008-11-25 02:35:14 +10307278 /* Save because hotplug lock held. */
7279 static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS);
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007280 int i;
7281
Rusty Russellabcd0832008-11-25 02:35:02 +10307282 for_each_cpu(i, cpu_map)
Gregory Haskins57d885f2008-01-25 21:08:18 +01007283 cpu_attach_domain(NULL, &def_root_domain, i);
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007284 synchronize_sched();
Rusty Russell96f874e22008-11-25 02:35:14 +10307285 arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask));
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007286}
7287
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007288/* handle null as "default" */
7289static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
7290 struct sched_domain_attr *new, int idx_new)
7291{
7292 struct sched_domain_attr tmp;
7293
7294 /* fast path */
7295 if (!new && !cur)
7296 return 1;
7297
7298 tmp = SD_ATTR_INIT;
7299 return !memcmp(cur ? (cur + idx_cur) : &tmp,
7300 new ? (new + idx_new) : &tmp,
7301 sizeof(struct sched_domain_attr));
7302}
7303
Paul Jackson029190c2007-10-18 23:40:20 -07007304/*
7305 * Partition sched domains as specified by the 'ndoms_new'
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007306 * cpumasks in the array doms_new[] of cpumasks. This compares
Paul Jackson029190c2007-10-18 23:40:20 -07007307 * doms_new[] to the current sched domain partitioning, doms_cur[].
7308 * It destroys each deleted domain and builds each new domain.
7309 *
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307310 * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'.
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007311 * The masks don't intersect (don't overlap.) We should setup one
7312 * sched domain for each mask. CPUs not in any of the cpumasks will
7313 * not be load balanced. If the same cpumask appears both in the
Paul Jackson029190c2007-10-18 23:40:20 -07007314 * current 'doms_cur' domains and in the new 'doms_new', we can leave
7315 * it as it is.
7316 *
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307317 * The passed in 'doms_new' should be allocated using
7318 * alloc_sched_domains. This routine takes ownership of it and will
7319 * free_sched_domains it when done with it. If the caller failed the
7320 * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1,
7321 * and partition_sched_domains() will fallback to the single partition
7322 * 'fallback_doms', it also forces the domains to be rebuilt.
Paul Jackson029190c2007-10-18 23:40:20 -07007323 *
Rusty Russell96f874e22008-11-25 02:35:14 +10307324 * If doms_new == NULL it will be replaced with cpu_online_mask.
Li Zefan700018e2008-11-18 14:02:03 +08007325 * ndoms_new == 0 is a special case for destroying existing domains,
7326 * and it will not create the default domain.
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07007327 *
Paul Jackson029190c2007-10-18 23:40:20 -07007328 * Call with hotplug lock held
7329 */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307330void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007331 struct sched_domain_attr *dattr_new)
Paul Jackson029190c2007-10-18 23:40:20 -07007332{
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07007333 int i, j, n;
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01007334 int new_topology;
Paul Jackson029190c2007-10-18 23:40:20 -07007335
Heiko Carstens712555e2008-04-28 11:33:07 +02007336 mutex_lock(&sched_domains_mutex);
Srivatsa Vaddagiria1835612008-01-25 21:08:00 +01007337
Milton Miller73785472007-10-24 18:23:48 +02007338 /* always unregister in case we don't destroy any domains */
7339 unregister_sched_domain_sysctl();
7340
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01007341 /* Let architecture update cpu core mappings. */
7342 new_topology = arch_update_cpu_topology();
7343
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07007344 n = doms_new ? ndoms_new : 0;
Paul Jackson029190c2007-10-18 23:40:20 -07007345
7346 /* Destroy deleted domains */
7347 for (i = 0; i < ndoms_cur; i++) {
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01007348 for (j = 0; j < n && !new_topology; j++) {
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307349 if (cpumask_equal(doms_cur[i], doms_new[j])
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007350 && dattrs_equal(dattr_cur, i, dattr_new, j))
Paul Jackson029190c2007-10-18 23:40:20 -07007351 goto match1;
7352 }
7353 /* no match - a current sched domain not in new doms_new[] */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307354 detach_destroy_domains(doms_cur[i]);
Paul Jackson029190c2007-10-18 23:40:20 -07007355match1:
7356 ;
7357 }
7358
Max Krasnyanskye761b772008-07-15 04:43:49 -07007359 if (doms_new == NULL) {
7360 ndoms_cur = 0;
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307361 doms_new = &fallback_doms;
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01007362 cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
Li Zefanfaa2f982008-11-04 16:20:23 +08007363 WARN_ON_ONCE(dattr_new);
Max Krasnyanskye761b772008-07-15 04:43:49 -07007364 }
7365
Paul Jackson029190c2007-10-18 23:40:20 -07007366 /* Build new domains */
7367 for (i = 0; i < ndoms_new; i++) {
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01007368 for (j = 0; j < ndoms_cur && !new_topology; j++) {
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307369 if (cpumask_equal(doms_new[i], doms_cur[j])
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007370 && dattrs_equal(dattr_new, i, dattr_cur, j))
Paul Jackson029190c2007-10-18 23:40:20 -07007371 goto match2;
7372 }
7373 /* no match - add a new doms_new */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307374 __build_sched_domains(doms_new[i],
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007375 dattr_new ? dattr_new + i : NULL);
Paul Jackson029190c2007-10-18 23:40:20 -07007376match2:
7377 ;
7378 }
7379
7380 /* Remember the new sched domains */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307381 if (doms_cur != &fallback_doms)
7382 free_sched_domains(doms_cur, ndoms_cur);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007383 kfree(dattr_cur); /* kfree(NULL) is safe */
Paul Jackson029190c2007-10-18 23:40:20 -07007384 doms_cur = doms_new;
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007385 dattr_cur = dattr_new;
Paul Jackson029190c2007-10-18 23:40:20 -07007386 ndoms_cur = ndoms_new;
Milton Miller73785472007-10-24 18:23:48 +02007387
7388 register_sched_domain_sysctl();
Srivatsa Vaddagiria1835612008-01-25 21:08:00 +01007389
Heiko Carstens712555e2008-04-28 11:33:07 +02007390 mutex_unlock(&sched_domains_mutex);
Paul Jackson029190c2007-10-18 23:40:20 -07007391}
7392
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007393#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
Li Zefanc70f22d2009-01-05 19:07:50 +08007394static void arch_reinit_sched_domains(void)
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007395{
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007396 get_online_cpus();
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07007397
7398 /* Destroy domains first to force the rebuild */
7399 partition_sched_domains(0, NULL, NULL);
7400
Max Krasnyanskye761b772008-07-15 04:43:49 -07007401 rebuild_sched_domains();
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007402 put_online_cpus();
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007403}
7404
7405static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt)
7406{
Gautham R Shenoyafb8a9b2008-12-18 23:26:09 +05307407 unsigned int level = 0;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007408
Gautham R Shenoyafb8a9b2008-12-18 23:26:09 +05307409 if (sscanf(buf, "%u", &level) != 1)
7410 return -EINVAL;
7411
7412 /*
7413 * level is always be positive so don't check for
7414 * level < POWERSAVINGS_BALANCE_NONE which is 0
7415 * What happens on 0 or 1 byte write,
7416 * need to check for count as well?
7417 */
7418
7419 if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS)
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007420 return -EINVAL;
7421
7422 if (smt)
Gautham R Shenoyafb8a9b2008-12-18 23:26:09 +05307423 sched_smt_power_savings = level;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007424 else
Gautham R Shenoyafb8a9b2008-12-18 23:26:09 +05307425 sched_mc_power_savings = level;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007426
Li Zefanc70f22d2009-01-05 19:07:50 +08007427 arch_reinit_sched_domains();
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007428
Li Zefanc70f22d2009-01-05 19:07:50 +08007429 return count;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007430}
7431
Adrian Bunk6707de002007-08-12 18:08:19 +02007432#ifdef CONFIG_SCHED_MC
Andi Kleenf718cd42008-07-29 22:33:52 -07007433static ssize_t sched_mc_power_savings_show(struct sysdev_class *class,
Andi Kleenc9be0a32010-01-05 12:47:58 +01007434 struct sysdev_class_attribute *attr,
Andi Kleenf718cd42008-07-29 22:33:52 -07007435 char *page)
Adrian Bunk6707de002007-08-12 18:08:19 +02007436{
7437 return sprintf(page, "%u\n", sched_mc_power_savings);
7438}
Andi Kleenf718cd42008-07-29 22:33:52 -07007439static ssize_t sched_mc_power_savings_store(struct sysdev_class *class,
Andi Kleenc9be0a32010-01-05 12:47:58 +01007440 struct sysdev_class_attribute *attr,
Adrian Bunk6707de002007-08-12 18:08:19 +02007441 const char *buf, size_t count)
7442{
7443 return sched_power_savings_store(buf, count, 0);
7444}
Andi Kleenf718cd42008-07-29 22:33:52 -07007445static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644,
7446 sched_mc_power_savings_show,
7447 sched_mc_power_savings_store);
Adrian Bunk6707de002007-08-12 18:08:19 +02007448#endif
7449
7450#ifdef CONFIG_SCHED_SMT
Andi Kleenf718cd42008-07-29 22:33:52 -07007451static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev,
Andi Kleenc9be0a32010-01-05 12:47:58 +01007452 struct sysdev_class_attribute *attr,
Andi Kleenf718cd42008-07-29 22:33:52 -07007453 char *page)
Adrian Bunk6707de002007-08-12 18:08:19 +02007454{
7455 return sprintf(page, "%u\n", sched_smt_power_savings);
7456}
Andi Kleenf718cd42008-07-29 22:33:52 -07007457static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev,
Andi Kleenc9be0a32010-01-05 12:47:58 +01007458 struct sysdev_class_attribute *attr,
Adrian Bunk6707de002007-08-12 18:08:19 +02007459 const char *buf, size_t count)
7460{
7461 return sched_power_savings_store(buf, count, 1);
7462}
Andi Kleenf718cd42008-07-29 22:33:52 -07007463static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644,
7464 sched_smt_power_savings_show,
Adrian Bunk6707de002007-08-12 18:08:19 +02007465 sched_smt_power_savings_store);
7466#endif
7467
Li Zefan39aac642009-01-05 19:18:02 +08007468int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls)
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007469{
7470 int err = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07007471
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007472#ifdef CONFIG_SCHED_SMT
7473 if (smt_capable())
7474 err = sysfs_create_file(&cls->kset.kobj,
7475 &attr_sched_smt_power_savings.attr);
7476#endif
7477#ifdef CONFIG_SCHED_MC
7478 if (!err && mc_capable())
7479 err = sysfs_create_file(&cls->kset.kobj,
7480 &attr_sched_mc_power_savings.attr);
7481#endif
7482 return err;
7483}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007484#endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007485
Linus Torvalds1da177e2005-04-16 15:20:36 -07007486/*
Tejun Heo3a101d02010-06-08 21:40:36 +02007487 * Update cpusets according to cpu_active mask. If cpusets are
7488 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
7489 * around partition_sched_domains().
Linus Torvalds1da177e2005-04-16 15:20:36 -07007490 */
Tejun Heo0b2e9182010-06-21 23:53:31 +02007491static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action,
7492 void *hcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007493{
Tejun Heo3a101d02010-06-08 21:40:36 +02007494 switch (action & ~CPU_TASKS_FROZEN) {
Max Krasnyanskye761b772008-07-15 04:43:49 -07007495 case CPU_ONLINE:
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01007496 case CPU_DOWN_FAILED:
Tejun Heo3a101d02010-06-08 21:40:36 +02007497 cpuset_update_active_cpus();
Max Krasnyanskye761b772008-07-15 04:43:49 -07007498 return NOTIFY_OK;
Max Krasnyanskye761b772008-07-15 04:43:49 -07007499 default:
7500 return NOTIFY_DONE;
7501 }
7502}
Tejun Heo3a101d02010-06-08 21:40:36 +02007503
Tejun Heo0b2e9182010-06-21 23:53:31 +02007504static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action,
7505 void *hcpu)
Tejun Heo3a101d02010-06-08 21:40:36 +02007506{
7507 switch (action & ~CPU_TASKS_FROZEN) {
7508 case CPU_DOWN_PREPARE:
7509 cpuset_update_active_cpus();
7510 return NOTIFY_OK;
7511 default:
7512 return NOTIFY_DONE;
7513 }
7514}
Max Krasnyanskye761b772008-07-15 04:43:49 -07007515
7516static int update_runtime(struct notifier_block *nfb,
7517 unsigned long action, void *hcpu)
7518{
Peter Zijlstra7def2be2008-06-05 14:49:58 +02007519 int cpu = (int)(long)hcpu;
7520
Linus Torvalds1da177e2005-04-16 15:20:36 -07007521 switch (action) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007522 case CPU_DOWN_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007523 case CPU_DOWN_PREPARE_FROZEN:
Peter Zijlstra7def2be2008-06-05 14:49:58 +02007524 disable_runtime(cpu_rq(cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -07007525 return NOTIFY_OK;
7526
Linus Torvalds1da177e2005-04-16 15:20:36 -07007527 case CPU_DOWN_FAILED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007528 case CPU_DOWN_FAILED_FROZEN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07007529 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007530 case CPU_ONLINE_FROZEN:
Peter Zijlstra7def2be2008-06-05 14:49:58 +02007531 enable_runtime(cpu_rq(cpu));
Max Krasnyanskye761b772008-07-15 04:43:49 -07007532 return NOTIFY_OK;
7533
Linus Torvalds1da177e2005-04-16 15:20:36 -07007534 default:
7535 return NOTIFY_DONE;
7536 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007537}
Linus Torvalds1da177e2005-04-16 15:20:36 -07007538
7539void __init sched_init_smp(void)
7540{
Rusty Russelldcc30a32008-11-25 02:35:12 +10307541 cpumask_var_t non_isolated_cpus;
7542
7543 alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
Yong Zhangcb5fd132009-09-14 20:20:16 +08007544 alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
Nick Piggin5c1e1762006-10-03 01:14:04 -07007545
Mike Travis434d53b2008-04-04 18:11:04 -07007546#if defined(CONFIG_NUMA)
7547 sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **),
7548 GFP_KERNEL);
7549 BUG_ON(sched_group_nodes_bycpu == NULL);
7550#endif
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007551 get_online_cpus();
Heiko Carstens712555e2008-04-28 11:33:07 +02007552 mutex_lock(&sched_domains_mutex);
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01007553 arch_init_sched_domains(cpu_active_mask);
Rusty Russelldcc30a32008-11-25 02:35:12 +10307554 cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map);
7555 if (cpumask_empty(non_isolated_cpus))
7556 cpumask_set_cpu(smp_processor_id(), non_isolated_cpus);
Heiko Carstens712555e2008-04-28 11:33:07 +02007557 mutex_unlock(&sched_domains_mutex);
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007558 put_online_cpus();
Max Krasnyanskye761b772008-07-15 04:43:49 -07007559
Tejun Heo3a101d02010-06-08 21:40:36 +02007560 hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE);
7561 hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE);
Max Krasnyanskye761b772008-07-15 04:43:49 -07007562
7563 /* RT runtime code needs to handle some hotplug events */
7564 hotcpu_notifier(update_runtime, 0);
7565
Peter Zijlstrab328ca12008-04-29 10:02:46 +02007566 init_hrtick();
Nick Piggin5c1e1762006-10-03 01:14:04 -07007567
7568 /* Move init over to a non-isolated CPU */
Rusty Russelldcc30a32008-11-25 02:35:12 +10307569 if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
Nick Piggin5c1e1762006-10-03 01:14:04 -07007570 BUG();
Ingo Molnar19978ca2007-11-09 22:39:38 +01007571 sched_init_granularity();
Rusty Russelldcc30a32008-11-25 02:35:12 +10307572 free_cpumask_var(non_isolated_cpus);
Rusty Russell42128232008-11-25 02:35:12 +10307573
Rusty Russell0e3900e2008-11-25 02:35:13 +10307574 init_sched_rt_class();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007575}
7576#else
7577void __init sched_init_smp(void)
7578{
Ingo Molnar19978ca2007-11-09 22:39:38 +01007579 sched_init_granularity();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007580}
7581#endif /* CONFIG_SMP */
7582
Arun R Bharadwajcd1bb942009-04-16 12:15:34 +05307583const_debug unsigned int sysctl_timer_migration = 1;
7584
Linus Torvalds1da177e2005-04-16 15:20:36 -07007585int in_sched_functions(unsigned long addr)
7586{
Linus Torvalds1da177e2005-04-16 15:20:36 -07007587 return in_lock_functions(addr) ||
7588 (addr >= (unsigned long)__sched_text_start
7589 && addr < (unsigned long)__sched_text_end);
7590}
7591
Alexey Dobriyana9957442007-10-15 17:00:13 +02007592static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq)
Ingo Molnardd41f592007-07-09 18:51:59 +02007593{
7594 cfs_rq->tasks_timeline = RB_ROOT;
Peter Zijlstra4a55bd52008-04-19 19:45:00 +02007595 INIT_LIST_HEAD(&cfs_rq->tasks);
Ingo Molnardd41f592007-07-09 18:51:59 +02007596#ifdef CONFIG_FAIR_GROUP_SCHED
7597 cfs_rq->rq = rq;
7598#endif
Peter Zijlstra67e9fb22007-10-15 17:00:10 +02007599 cfs_rq->min_vruntime = (u64)(-(1LL << 20));
Ingo Molnardd41f592007-07-09 18:51:59 +02007600}
7601
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007602static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq)
7603{
7604 struct rt_prio_array *array;
7605 int i;
7606
7607 array = &rt_rq->active;
7608 for (i = 0; i < MAX_RT_PRIO; i++) {
7609 INIT_LIST_HEAD(array->queue + i);
7610 __clear_bit(i, array->bitmap);
7611 }
7612 /* delimiter for bitsearch: */
7613 __set_bit(MAX_RT_PRIO, array->bitmap);
7614
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007615#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
Gregory Haskinse864c492008-12-29 09:39:49 -05007616 rt_rq->highest_prio.curr = MAX_RT_PRIO;
Gregory Haskins398a1532009-01-14 09:10:04 -05007617#ifdef CONFIG_SMP
Gregory Haskinse864c492008-12-29 09:39:49 -05007618 rt_rq->highest_prio.next = MAX_RT_PRIO;
Peter Zijlstra48d5e252008-01-25 21:08:31 +01007619#endif
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007620#endif
7621#ifdef CONFIG_SMP
7622 rt_rq->rt_nr_migratory = 0;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007623 rt_rq->overloaded = 0;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01007624 plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007625#endif
7626
7627 rt_rq->rt_time = 0;
7628 rt_rq->rt_throttled = 0;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007629 rt_rq->rt_runtime = 0;
Thomas Gleixner0986b112009-11-17 15:32:06 +01007630 raw_spin_lock_init(&rt_rq->rt_runtime_lock);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007631
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007632#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +01007633 rt_rq->rt_nr_boosted = 0;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007634 rt_rq->rq = rq;
7635#endif
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007636}
7637
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007638#ifdef CONFIG_FAIR_GROUP_SCHED
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007639static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
7640 struct sched_entity *se, int cpu, int add,
7641 struct sched_entity *parent)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007642{
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007643 struct rq *rq = cpu_rq(cpu);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007644 tg->cfs_rq[cpu] = cfs_rq;
7645 init_cfs_rq(cfs_rq, rq);
7646 cfs_rq->tg = tg;
7647 if (add)
7648 list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list);
7649
7650 tg->se[cpu] = se;
Dhaval Giani354d60c2008-04-19 19:44:59 +02007651 /* se could be NULL for init_task_group */
7652 if (!se)
7653 return;
7654
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007655 if (!parent)
7656 se->cfs_rq = &rq->cfs;
7657 else
7658 se->cfs_rq = parent->my_q;
7659
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007660 se->my_q = cfs_rq;
7661 se->load.weight = tg->shares;
Peter Zijlstrae05510d2008-05-05 23:56:17 +02007662 se->load.inv_weight = 0;
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007663 se->parent = parent;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007664}
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007665#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007666
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007667#ifdef CONFIG_RT_GROUP_SCHED
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007668static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
7669 struct sched_rt_entity *rt_se, int cpu, int add,
7670 struct sched_rt_entity *parent)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007671{
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007672 struct rq *rq = cpu_rq(cpu);
7673
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007674 tg->rt_rq[cpu] = rt_rq;
7675 init_rt_rq(rt_rq, rq);
7676 rt_rq->tg = tg;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007677 rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007678 if (add)
7679 list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list);
7680
7681 tg->rt_se[cpu] = rt_se;
Dhaval Giani354d60c2008-04-19 19:44:59 +02007682 if (!rt_se)
7683 return;
7684
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007685 if (!parent)
7686 rt_se->rt_rq = &rq->rt;
7687 else
7688 rt_se->rt_rq = parent->my_q;
7689
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007690 rt_se->my_q = rt_rq;
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007691 rt_se->parent = parent;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007692 INIT_LIST_HEAD(&rt_se->run_list);
7693}
7694#endif
7695
Linus Torvalds1da177e2005-04-16 15:20:36 -07007696void __init sched_init(void)
7697{
Ingo Molnardd41f592007-07-09 18:51:59 +02007698 int i, j;
Mike Travis434d53b2008-04-04 18:11:04 -07007699 unsigned long alloc_size = 0, ptr;
7700
7701#ifdef CONFIG_FAIR_GROUP_SCHED
7702 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
7703#endif
7704#ifdef CONFIG_RT_GROUP_SCHED
7705 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
7706#endif
Rusty Russelldf7c8e82009-03-19 15:22:20 +10307707#ifdef CONFIG_CPUMASK_OFFSTACK
Rusty Russell8c083f02009-03-19 15:22:20 +10307708 alloc_size += num_possible_cpus() * cpumask_size();
Rusty Russelldf7c8e82009-03-19 15:22:20 +10307709#endif
Mike Travis434d53b2008-04-04 18:11:04 -07007710 if (alloc_size) {
Pekka Enberg36b7b6d2009-06-10 23:42:36 +03007711 ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
Mike Travis434d53b2008-04-04 18:11:04 -07007712
7713#ifdef CONFIG_FAIR_GROUP_SCHED
7714 init_task_group.se = (struct sched_entity **)ptr;
7715 ptr += nr_cpu_ids * sizeof(void **);
7716
7717 init_task_group.cfs_rq = (struct cfs_rq **)ptr;
7718 ptr += nr_cpu_ids * sizeof(void **);
Peter Zijlstraeff766a2008-04-19 19:45:00 +02007719
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007720#endif /* CONFIG_FAIR_GROUP_SCHED */
Mike Travis434d53b2008-04-04 18:11:04 -07007721#ifdef CONFIG_RT_GROUP_SCHED
7722 init_task_group.rt_se = (struct sched_rt_entity **)ptr;
7723 ptr += nr_cpu_ids * sizeof(void **);
7724
7725 init_task_group.rt_rq = (struct rt_rq **)ptr;
Peter Zijlstraeff766a2008-04-19 19:45:00 +02007726 ptr += nr_cpu_ids * sizeof(void **);
7727
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007728#endif /* CONFIG_RT_GROUP_SCHED */
Rusty Russelldf7c8e82009-03-19 15:22:20 +10307729#ifdef CONFIG_CPUMASK_OFFSTACK
7730 for_each_possible_cpu(i) {
7731 per_cpu(load_balance_tmpmask, i) = (void *)ptr;
7732 ptr += cpumask_size();
7733 }
7734#endif /* CONFIG_CPUMASK_OFFSTACK */
Mike Travis434d53b2008-04-04 18:11:04 -07007735 }
Ingo Molnardd41f592007-07-09 18:51:59 +02007736
Gregory Haskins57d885f2008-01-25 21:08:18 +01007737#ifdef CONFIG_SMP
7738 init_defrootdomain();
7739#endif
7740
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007741 init_rt_bandwidth(&def_rt_bandwidth,
7742 global_rt_period(), global_rt_runtime());
7743
7744#ifdef CONFIG_RT_GROUP_SCHED
7745 init_rt_bandwidth(&init_task_group.rt_bandwidth,
7746 global_rt_period(), global_rt_runtime());
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007747#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007748
Dhaval Giani7c941432010-01-20 13:26:18 +01007749#ifdef CONFIG_CGROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007750 list_add(&init_task_group.list, &task_groups);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02007751 INIT_LIST_HEAD(&init_task_group.children);
7752
Dhaval Giani7c941432010-01-20 13:26:18 +01007753#endif /* CONFIG_CGROUP_SCHED */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007754
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09007755#if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP
7756 update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long),
7757 __alignof__(unsigned long));
7758#endif
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08007759 for_each_possible_cpu(i) {
Ingo Molnar70b97a72006-07-03 00:25:42 -07007760 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007761
7762 rq = cpu_rq(i);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01007763 raw_spin_lock_init(&rq->lock);
Nick Piggin78979862005-06-25 14:57:13 -07007764 rq->nr_running = 0;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02007765 rq->calc_load_active = 0;
7766 rq->calc_load_update = jiffies + LOAD_FREQ;
Ingo Molnardd41f592007-07-09 18:51:59 +02007767 init_cfs_rq(&rq->cfs, rq);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007768 init_rt_rq(&rq->rt, rq);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007769#ifdef CONFIG_FAIR_GROUP_SCHED
7770 init_task_group.shares = init_task_group_load;
7771 INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
Dhaval Giani354d60c2008-04-19 19:44:59 +02007772#ifdef CONFIG_CGROUP_SCHED
7773 /*
7774 * How much cpu bandwidth does init_task_group get?
7775 *
7776 * In case of task-groups formed thr' the cgroup filesystem, it
7777 * gets 100% of the cpu resources in the system. This overall
7778 * system cpu resource is divided among the tasks of
7779 * init_task_group and its child task-groups in a fair manner,
7780 * based on each entity's (task or task-group's) weight
7781 * (se->load.weight).
7782 *
7783 * In other words, if init_task_group has 10 tasks of weight
7784 * 1024) and two child groups A0 and A1 (of weight 1024 each),
7785 * then A0's share of the cpu resource is:
7786 *
Ingo Molnar0d905bc2009-05-04 19:13:30 +02007787 * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
Dhaval Giani354d60c2008-04-19 19:44:59 +02007788 *
7789 * We achieve this by letting init_task_group's tasks sit
7790 * directly in rq->cfs (i.e init_task_group->se[] = NULL).
7791 */
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007792 init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL);
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007793#endif
Dhaval Giani354d60c2008-04-19 19:44:59 +02007794#endif /* CONFIG_FAIR_GROUP_SCHED */
7795
7796 rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007797#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007798 INIT_LIST_HEAD(&rq->leaf_rt_rq_list);
Dhaval Giani354d60c2008-04-19 19:44:59 +02007799#ifdef CONFIG_CGROUP_SCHED
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007800 init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL);
Dhaval Giani354d60c2008-04-19 19:44:59 +02007801#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007802#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07007803
Ingo Molnardd41f592007-07-09 18:51:59 +02007804 for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
7805 rq->cpu_load[j] = 0;
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07007806
7807 rq->last_load_update_tick = jiffies;
7808
Linus Torvalds1da177e2005-04-16 15:20:36 -07007809#ifdef CONFIG_SMP
Nick Piggin41c7ce92005-06-25 14:57:24 -07007810 rq->sd = NULL;
Gregory Haskins57d885f2008-01-25 21:08:18 +01007811 rq->rd = NULL;
Peter Zijlstrae51fd5e2010-05-31 12:37:30 +02007812 rq->cpu_power = SCHED_LOAD_SCALE;
Gregory Haskins3f029d32009-07-29 11:08:47 -04007813 rq->post_schedule = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007814 rq->active_balance = 0;
Ingo Molnardd41f592007-07-09 18:51:59 +02007815 rq->next_balance = jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007816 rq->push_cpu = 0;
Christoph Lameter0a2966b2006-09-25 23:30:51 -07007817 rq->cpu = i;
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -04007818 rq->online = 0;
Mike Galbraitheae0c9d2009-11-10 03:50:02 +01007819 rq->idle_stamp = 0;
7820 rq->avg_idle = 2*sysctl_sched_migration_cost;
Gregory Haskinsdc938522008-01-25 21:08:26 +01007821 rq_attach_root(rq, &def_root_domain);
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -07007822#ifdef CONFIG_NO_HZ
7823 rq->nohz_balance_kick = 0;
7824 init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i));
7825#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07007826#endif
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01007827 init_rq_hrtick(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007828 atomic_set(&rq->nr_iowait, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007829 }
7830
Peter Williams2dd73a42006-06-27 02:54:34 -07007831 set_load_weight(&init_task);
Heiko Carstensb50f60c2006-07-30 03:03:52 -07007832
Avi Kivitye107be32007-07-26 13:40:43 +02007833#ifdef CONFIG_PREEMPT_NOTIFIERS
7834 INIT_HLIST_HEAD(&init_task.preempt_notifiers);
7835#endif
7836
Christoph Lameterc9819f42006-12-10 02:20:25 -08007837#ifdef CONFIG_SMP
Carlos R. Mafra962cf362008-05-15 11:15:37 -03007838 open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);
Christoph Lameterc9819f42006-12-10 02:20:25 -08007839#endif
7840
Heiko Carstensb50f60c2006-07-30 03:03:52 -07007841#ifdef CONFIG_RT_MUTEXES
Thomas Gleixner1d615482009-11-17 14:54:03 +01007842 plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock);
Heiko Carstensb50f60c2006-07-30 03:03:52 -07007843#endif
7844
Linus Torvalds1da177e2005-04-16 15:20:36 -07007845 /*
7846 * The boot idle thread does lazy MMU switching as well:
7847 */
7848 atomic_inc(&init_mm.mm_count);
7849 enter_lazy_tlb(&init_mm, current);
7850
7851 /*
7852 * Make us the idle thread. Technically, schedule() should not be
7853 * called from this thread, however somewhere below it might be,
7854 * but because we are the idle thread, we just pick up running again
7855 * when this runqueue becomes "idle".
7856 */
7857 init_idle(current, smp_processor_id());
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02007858
7859 calc_load_update = jiffies + LOAD_FREQ;
7860
Ingo Molnardd41f592007-07-09 18:51:59 +02007861 /*
7862 * During early bootup we pretend to be a normal task:
7863 */
7864 current->sched_class = &fair_sched_class;
Ingo Molnar6892b752008-02-13 14:02:36 +01007865
Rusty Russell6a7b3dc2008-11-25 02:35:04 +10307866 /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */
Rusty Russell49557e62009-11-02 20:37:20 +10307867 zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT);
Rusty Russellbf4d83f2008-11-25 09:57:51 +10307868#ifdef CONFIG_SMP
Rusty Russell7d1e6a92008-11-25 02:35:09 +10307869#ifdef CONFIG_NO_HZ
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -07007870 zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
7871 alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT);
7872 atomic_set(&nohz.load_balancer, nr_cpu_ids);
7873 atomic_set(&nohz.first_pick_cpu, nr_cpu_ids);
7874 atomic_set(&nohz.second_pick_cpu, nr_cpu_ids);
Rusty Russell7d1e6a92008-11-25 02:35:09 +10307875#endif
Rusty Russellbdddd292009-12-02 14:09:16 +10307876 /* May be allocated at isolcpus cmdline parse time */
7877 if (cpu_isolated_map == NULL)
7878 zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT);
Rusty Russellbf4d83f2008-11-25 09:57:51 +10307879#endif /* SMP */
Rusty Russell6a7b3dc2008-11-25 02:35:04 +10307880
Ingo Molnarcdd6c482009-09-21 12:02:48 +02007881 perf_event_init();
Ingo Molnar0d905bc2009-05-04 19:13:30 +02007882
Ingo Molnar6892b752008-02-13 14:02:36 +01007883 scheduler_running = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007884}
7885
7886#ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
Frederic Weisbeckere4aafea2009-07-16 15:44:29 +02007887static inline int preempt_count_equals(int preempt_offset)
7888{
Frederic Weisbecker234da7b2009-12-16 20:21:05 +01007889 int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth();
Frederic Weisbeckere4aafea2009-07-16 15:44:29 +02007890
7891 return (nested == PREEMPT_INATOMIC_BASE + preempt_offset);
7892}
7893
Simon Kagstromd8948372009-12-23 11:08:18 +01007894void __might_sleep(const char *file, int line, int preempt_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007895{
Ingo Molnar48f24c42006-07-03 00:25:40 -07007896#ifdef in_atomic
Linus Torvalds1da177e2005-04-16 15:20:36 -07007897 static unsigned long prev_jiffy; /* ratelimiting */
7898
Frederic Weisbeckere4aafea2009-07-16 15:44:29 +02007899 if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) ||
7900 system_state != SYSTEM_RUNNING || oops_in_progress)
Ingo Molnaraef745f2008-08-28 11:34:43 +02007901 return;
7902 if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
7903 return;
7904 prev_jiffy = jiffies;
7905
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01007906 printk(KERN_ERR
7907 "BUG: sleeping function called from invalid context at %s:%d\n",
7908 file, line);
7909 printk(KERN_ERR
7910 "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n",
7911 in_atomic(), irqs_disabled(),
7912 current->pid, current->comm);
Ingo Molnaraef745f2008-08-28 11:34:43 +02007913
7914 debug_show_held_locks(current);
7915 if (irqs_disabled())
7916 print_irqtrace_events(current);
7917 dump_stack();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007918#endif
7919}
7920EXPORT_SYMBOL(__might_sleep);
7921#endif
7922
7923#ifdef CONFIG_MAGIC_SYSRQ
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02007924static void normalize_task(struct rq *rq, struct task_struct *p)
7925{
7926 int on_rq;
Peter Zijlstra3e51f332008-05-03 18:29:28 +02007927
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02007928 on_rq = p->se.on_rq;
7929 if (on_rq)
7930 deactivate_task(rq, p, 0);
7931 __setscheduler(rq, p, SCHED_NORMAL, 0);
7932 if (on_rq) {
7933 activate_task(rq, p, 0);
7934 resched_task(rq->curr);
7935 }
7936}
7937
Linus Torvalds1da177e2005-04-16 15:20:36 -07007938void normalize_rt_tasks(void)
7939{
Ingo Molnara0f98a12007-06-17 18:37:45 +02007940 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007941 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07007942 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007943
Peter Zijlstra4cf5d772008-02-13 15:45:39 +01007944 read_lock_irqsave(&tasklist_lock, flags);
Ingo Molnara0f98a12007-06-17 18:37:45 +02007945 do_each_thread(g, p) {
Ingo Molnar178be792007-10-15 17:00:18 +02007946 /*
7947 * Only normalize user tasks:
7948 */
7949 if (!p->mm)
7950 continue;
7951
Ingo Molnardd41f592007-07-09 18:51:59 +02007952 p->se.exec_start = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02007953#ifdef CONFIG_SCHEDSTATS
Lucas De Marchi41acab82010-03-10 23:37:45 -03007954 p->se.statistics.wait_start = 0;
7955 p->se.statistics.sleep_start = 0;
7956 p->se.statistics.block_start = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02007957#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02007958
7959 if (!rt_task(p)) {
7960 /*
7961 * Renice negative nice level userspace
7962 * tasks back to 0:
7963 */
7964 if (TASK_NICE(p) < 0 && p->mm)
7965 set_user_nice(p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007966 continue;
Ingo Molnardd41f592007-07-09 18:51:59 +02007967 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007968
Thomas Gleixner1d615482009-11-17 14:54:03 +01007969 raw_spin_lock(&p->pi_lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -07007970 rq = __task_rq_lock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007971
Ingo Molnar178be792007-10-15 17:00:18 +02007972 normalize_task(rq, p);
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02007973
Ingo Molnarb29739f2006-06-27 02:54:51 -07007974 __task_rq_unlock(rq);
Thomas Gleixner1d615482009-11-17 14:54:03 +01007975 raw_spin_unlock(&p->pi_lock);
Ingo Molnara0f98a12007-06-17 18:37:45 +02007976 } while_each_thread(g, p);
7977
Peter Zijlstra4cf5d772008-02-13 15:45:39 +01007978 read_unlock_irqrestore(&tasklist_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007979}
7980
7981#endif /* CONFIG_MAGIC_SYSRQ */
Linus Torvalds1df5c102005-09-12 07:59:21 -07007982
Jason Wessel67fc4e02010-05-20 21:04:21 -05007983#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
Linus Torvalds1df5c102005-09-12 07:59:21 -07007984/*
Jason Wessel67fc4e02010-05-20 21:04:21 -05007985 * These functions are only useful for the IA64 MCA handling, or kdb.
Linus Torvalds1df5c102005-09-12 07:59:21 -07007986 *
7987 * They can only be called when the whole system has been
7988 * stopped - every CPU needs to be quiescent, and no scheduling
7989 * activity can take place. Using them for anything else would
7990 * be a serious bug, and as a result, they aren't even visible
7991 * under any other configuration.
7992 */
7993
7994/**
7995 * curr_task - return the current task for a given cpu.
7996 * @cpu: the processor in question.
7997 *
7998 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
7999 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07008000struct task_struct *curr_task(int cpu)
Linus Torvalds1df5c102005-09-12 07:59:21 -07008001{
8002 return cpu_curr(cpu);
8003}
8004
Jason Wessel67fc4e02010-05-20 21:04:21 -05008005#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */
8006
8007#ifdef CONFIG_IA64
Linus Torvalds1df5c102005-09-12 07:59:21 -07008008/**
8009 * set_curr_task - set the current task for a given cpu.
8010 * @cpu: the processor in question.
8011 * @p: the task pointer to set.
8012 *
8013 * Description: This function must only be used when non-maskable interrupts
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01008014 * are serviced on a separate stack. It allows the architecture to switch the
8015 * notion of the current task on a cpu in a non-blocking manner. This function
Linus Torvalds1df5c102005-09-12 07:59:21 -07008016 * must be called with all CPU's synchronized, and interrupts disabled, the
8017 * and caller must save the original value of the current task (see
8018 * curr_task() above) and restore that value before reenabling interrupts and
8019 * re-starting the system.
8020 *
8021 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
8022 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07008023void set_curr_task(int cpu, struct task_struct *p)
Linus Torvalds1df5c102005-09-12 07:59:21 -07008024{
8025 cpu_curr(cpu) = p;
8026}
8027
8028#endif
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008029
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008030#ifdef CONFIG_FAIR_GROUP_SCHED
8031static void free_fair_sched_group(struct task_group *tg)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008032{
8033 int i;
8034
8035 for_each_possible_cpu(i) {
8036 if (tg->cfs_rq)
8037 kfree(tg->cfs_rq[i]);
8038 if (tg->se)
8039 kfree(tg->se[i]);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008040 }
8041
8042 kfree(tg->cfs_rq);
8043 kfree(tg->se);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008044}
8045
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008046static
8047int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008048{
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008049 struct cfs_rq *cfs_rq;
Li Zefaneab17222008-10-29 17:03:22 +08008050 struct sched_entity *se;
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008051 struct rq *rq;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008052 int i;
8053
Mike Travis434d53b2008-04-04 18:11:04 -07008054 tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008055 if (!tg->cfs_rq)
8056 goto err;
Mike Travis434d53b2008-04-04 18:11:04 -07008057 tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008058 if (!tg->se)
8059 goto err;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008060
8061 tg->shares = NICE_0_LOAD;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008062
8063 for_each_possible_cpu(i) {
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008064 rq = cpu_rq(i);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008065
Li Zefaneab17222008-10-29 17:03:22 +08008066 cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
8067 GFP_KERNEL, cpu_to_node(i));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008068 if (!cfs_rq)
8069 goto err;
8070
Li Zefaneab17222008-10-29 17:03:22 +08008071 se = kzalloc_node(sizeof(struct sched_entity),
8072 GFP_KERNEL, cpu_to_node(i));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008073 if (!se)
Phil Carmodydfc12eb2009-12-10 14:29:37 +02008074 goto err_free_rq;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008075
Li Zefaneab17222008-10-29 17:03:22 +08008076 init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008077 }
8078
8079 return 1;
8080
Phil Carmodydfc12eb2009-12-10 14:29:37 +02008081 err_free_rq:
8082 kfree(cfs_rq);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008083 err:
8084 return 0;
8085}
8086
8087static inline void register_fair_sched_group(struct task_group *tg, int cpu)
8088{
8089 list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list,
8090 &cpu_rq(cpu)->leaf_cfs_rq_list);
8091}
8092
8093static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
8094{
8095 list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list);
8096}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008097#else /* !CONFG_FAIR_GROUP_SCHED */
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008098static inline void free_fair_sched_group(struct task_group *tg)
8099{
8100}
8101
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008102static inline
8103int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008104{
8105 return 1;
8106}
8107
8108static inline void register_fair_sched_group(struct task_group *tg, int cpu)
8109{
8110}
8111
8112static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
8113{
8114}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008115#endif /* CONFIG_FAIR_GROUP_SCHED */
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008116
8117#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008118static void free_rt_sched_group(struct task_group *tg)
8119{
8120 int i;
8121
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008122 destroy_rt_bandwidth(&tg->rt_bandwidth);
8123
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008124 for_each_possible_cpu(i) {
8125 if (tg->rt_rq)
8126 kfree(tg->rt_rq[i]);
8127 if (tg->rt_se)
8128 kfree(tg->rt_se[i]);
8129 }
8130
8131 kfree(tg->rt_rq);
8132 kfree(tg->rt_se);
8133}
8134
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008135static
8136int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008137{
8138 struct rt_rq *rt_rq;
Li Zefaneab17222008-10-29 17:03:22 +08008139 struct sched_rt_entity *rt_se;
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008140 struct rq *rq;
8141 int i;
8142
Mike Travis434d53b2008-04-04 18:11:04 -07008143 tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008144 if (!tg->rt_rq)
8145 goto err;
Mike Travis434d53b2008-04-04 18:11:04 -07008146 tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008147 if (!tg->rt_se)
8148 goto err;
8149
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008150 init_rt_bandwidth(&tg->rt_bandwidth,
8151 ktime_to_ns(def_rt_bandwidth.rt_period), 0);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008152
8153 for_each_possible_cpu(i) {
8154 rq = cpu_rq(i);
8155
Li Zefaneab17222008-10-29 17:03:22 +08008156 rt_rq = kzalloc_node(sizeof(struct rt_rq),
8157 GFP_KERNEL, cpu_to_node(i));
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008158 if (!rt_rq)
8159 goto err;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008160
Li Zefaneab17222008-10-29 17:03:22 +08008161 rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
8162 GFP_KERNEL, cpu_to_node(i));
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008163 if (!rt_se)
Phil Carmodydfc12eb2009-12-10 14:29:37 +02008164 goto err_free_rq;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008165
Li Zefaneab17222008-10-29 17:03:22 +08008166 init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008167 }
8168
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008169 return 1;
8170
Phil Carmodydfc12eb2009-12-10 14:29:37 +02008171 err_free_rq:
8172 kfree(rt_rq);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008173 err:
8174 return 0;
8175}
8176
8177static inline void register_rt_sched_group(struct task_group *tg, int cpu)
8178{
8179 list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list,
8180 &cpu_rq(cpu)->leaf_rt_rq_list);
8181}
8182
8183static inline void unregister_rt_sched_group(struct task_group *tg, int cpu)
8184{
8185 list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list);
8186}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008187#else /* !CONFIG_RT_GROUP_SCHED */
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008188static inline void free_rt_sched_group(struct task_group *tg)
8189{
8190}
8191
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008192static inline
8193int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008194{
8195 return 1;
8196}
8197
8198static inline void register_rt_sched_group(struct task_group *tg, int cpu)
8199{
8200}
8201
8202static inline void unregister_rt_sched_group(struct task_group *tg, int cpu)
8203{
8204}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008205#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008206
Dhaval Giani7c941432010-01-20 13:26:18 +01008207#ifdef CONFIG_CGROUP_SCHED
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008208static void free_sched_group(struct task_group *tg)
8209{
8210 free_fair_sched_group(tg);
8211 free_rt_sched_group(tg);
8212 kfree(tg);
8213}
8214
8215/* allocate runqueue etc for a new task group */
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008216struct task_group *sched_create_group(struct task_group *parent)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008217{
8218 struct task_group *tg;
8219 unsigned long flags;
8220 int i;
8221
8222 tg = kzalloc(sizeof(*tg), GFP_KERNEL);
8223 if (!tg)
8224 return ERR_PTR(-ENOMEM);
8225
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008226 if (!alloc_fair_sched_group(tg, parent))
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008227 goto err;
8228
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008229 if (!alloc_rt_sched_group(tg, parent))
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008230 goto err;
8231
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008232 spin_lock_irqsave(&task_group_lock, flags);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008233 for_each_possible_cpu(i) {
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008234 register_fair_sched_group(tg, i);
8235 register_rt_sched_group(tg, i);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008236 }
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008237 list_add_rcu(&tg->list, &task_groups);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02008238
8239 WARN_ON(!parent); /* root should already exist */
8240
8241 tg->parent = parent;
Peter Zijlstraf473aa52008-04-19 19:45:00 +02008242 INIT_LIST_HEAD(&tg->children);
Zhang, Yanmin09f27242030-08-14 15:56:40 +08008243 list_add_rcu(&tg->siblings, &parent->children);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008244 spin_unlock_irqrestore(&task_group_lock, flags);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008245
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008246 return tg;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008247
8248err:
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008249 free_sched_group(tg);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008250 return ERR_PTR(-ENOMEM);
8251}
8252
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008253/* rcu callback to free various structures associated with a task group */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008254static void free_sched_group_rcu(struct rcu_head *rhp)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008255{
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008256 /* now it should be safe to free those cfs_rqs */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008257 free_sched_group(container_of(rhp, struct task_group, rcu));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008258}
8259
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008260/* Destroy runqueue etc associated with a task group */
Ingo Molnar4cf86d72007-10-15 17:00:14 +02008261void sched_destroy_group(struct task_group *tg)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008262{
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008263 unsigned long flags;
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008264 int i;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008265
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008266 spin_lock_irqsave(&task_group_lock, flags);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008267 for_each_possible_cpu(i) {
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008268 unregister_fair_sched_group(tg, i);
8269 unregister_rt_sched_group(tg, i);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008270 }
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008271 list_del_rcu(&tg->list);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02008272 list_del_rcu(&tg->siblings);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008273 spin_unlock_irqrestore(&task_group_lock, flags);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008274
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008275 /* wait for possible concurrent references to cfs_rqs complete */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008276 call_rcu(&tg->rcu, free_sched_group_rcu);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008277}
8278
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008279/* change task's runqueue when it moves between groups.
Ingo Molnar3a252012007-10-15 17:00:12 +02008280 * The caller of this function should have put the task in its new group
8281 * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to
8282 * reflect its new group.
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008283 */
8284void sched_move_task(struct task_struct *tsk)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008285{
8286 int on_rq, running;
8287 unsigned long flags;
8288 struct rq *rq;
8289
8290 rq = task_rq_lock(tsk, &flags);
8291
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01008292 running = task_current(rq, tsk);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008293 on_rq = tsk->se.on_rq;
8294
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07008295 if (on_rq)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008296 dequeue_task(rq, tsk, 0);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07008297 if (unlikely(running))
8298 tsk->sched_class->put_prev_task(rq, tsk);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008299
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008300 set_task_rq(tsk, task_cpu(tsk));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008301
Peter Zijlstra810b3812008-02-29 15:21:01 -05008302#ifdef CONFIG_FAIR_GROUP_SCHED
8303 if (tsk->sched_class->moved_group)
Peter Zijlstra88ec22d2009-12-16 18:04:41 +01008304 tsk->sched_class->moved_group(tsk, on_rq);
Peter Zijlstra810b3812008-02-29 15:21:01 -05008305#endif
8306
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07008307 if (unlikely(running))
8308 tsk->sched_class->set_curr_task(rq);
8309 if (on_rq)
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01008310 enqueue_task(rq, tsk, 0);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008311
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008312 task_rq_unlock(rq, &flags);
8313}
Dhaval Giani7c941432010-01-20 13:26:18 +01008314#endif /* CONFIG_CGROUP_SCHED */
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008315
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008316#ifdef CONFIG_FAIR_GROUP_SCHED
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008317static void __set_se_shares(struct sched_entity *se, unsigned long shares)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008318{
8319 struct cfs_rq *cfs_rq = se->cfs_rq;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008320 int on_rq;
8321
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008322 on_rq = se->on_rq;
Peter Zijlstra62fb1852008-02-25 17:34:02 +01008323 if (on_rq)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008324 dequeue_entity(cfs_rq, se, 0);
8325
8326 se->load.weight = shares;
Peter Zijlstrae05510d2008-05-05 23:56:17 +02008327 se->load.inv_weight = 0;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008328
Peter Zijlstra62fb1852008-02-25 17:34:02 +01008329 if (on_rq)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008330 enqueue_entity(cfs_rq, se, 0);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008331}
Peter Zijlstra62fb1852008-02-25 17:34:02 +01008332
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008333static void set_se_shares(struct sched_entity *se, unsigned long shares)
8334{
8335 struct cfs_rq *cfs_rq = se->cfs_rq;
8336 struct rq *rq = cfs_rq->rq;
8337 unsigned long flags;
8338
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008339 raw_spin_lock_irqsave(&rq->lock, flags);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008340 __set_se_shares(se, shares);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008341 raw_spin_unlock_irqrestore(&rq->lock, flags);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008342}
8343
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008344static DEFINE_MUTEX(shares_mutex);
8345
Ingo Molnar4cf86d72007-10-15 17:00:14 +02008346int sched_group_set_shares(struct task_group *tg, unsigned long shares)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008347{
8348 int i;
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008349 unsigned long flags;
Ingo Molnarc61935f2008-01-22 11:24:58 +01008350
Peter Zijlstra62fb1852008-02-25 17:34:02 +01008351 /*
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008352 * We can't change the weight of the root cgroup.
8353 */
8354 if (!tg->se[0])
8355 return -EINVAL;
8356
Peter Zijlstra18d95a22008-04-19 19:45:00 +02008357 if (shares < MIN_SHARES)
8358 shares = MIN_SHARES;
Miao Xiecb4ad1f2008-04-28 12:54:56 +08008359 else if (shares > MAX_SHARES)
8360 shares = MAX_SHARES;
Peter Zijlstra62fb1852008-02-25 17:34:02 +01008361
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008362 mutex_lock(&shares_mutex);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008363 if (tg->shares == shares)
Dhaval Giani5cb350b2007-10-15 17:00:14 +02008364 goto done;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008365
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008366 spin_lock_irqsave(&task_group_lock, flags);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008367 for_each_possible_cpu(i)
8368 unregister_fair_sched_group(tg, i);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02008369 list_del_rcu(&tg->siblings);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008370 spin_unlock_irqrestore(&task_group_lock, flags);
Srivatsa Vaddagiri6b2d7702008-01-25 21:08:00 +01008371
8372 /* wait for any ongoing reference to this group to finish */
8373 synchronize_sched();
8374
8375 /*
8376 * Now we are free to modify the group's share on each cpu
8377 * w/o tripping rebalance_share or load_balance_fair.
8378 */
8379 tg->shares = shares;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008380 for_each_possible_cpu(i) {
8381 /*
8382 * force a rebalance
8383 */
8384 cfs_rq_set_shares(tg->cfs_rq[i], 0);
Miao Xiecb4ad1f2008-04-28 12:54:56 +08008385 set_se_shares(tg->se[i], shares);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008386 }
Srivatsa Vaddagiri6b2d7702008-01-25 21:08:00 +01008387
8388 /*
8389 * Enable load balance activity on this group, by inserting it back on
8390 * each cpu's rq->leaf_cfs_rq_list.
8391 */
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008392 spin_lock_irqsave(&task_group_lock, flags);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008393 for_each_possible_cpu(i)
8394 register_fair_sched_group(tg, i);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02008395 list_add_rcu(&tg->siblings, &tg->parent->children);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008396 spin_unlock_irqrestore(&task_group_lock, flags);
Dhaval Giani5cb350b2007-10-15 17:00:14 +02008397done:
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008398 mutex_unlock(&shares_mutex);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008399 return 0;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008400}
8401
Dhaval Giani5cb350b2007-10-15 17:00:14 +02008402unsigned long sched_group_shares(struct task_group *tg)
8403{
8404 return tg->shares;
8405}
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008406#endif
Dhaval Giani5cb350b2007-10-15 17:00:14 +02008407
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008408#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008409/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008410 * Ensure that the real time constraints are schedulable.
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008411 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008412static DEFINE_MUTEX(rt_constraints_mutex);
8413
8414static unsigned long to_ratio(u64 period, u64 runtime)
8415{
8416 if (runtime == RUNTIME_INF)
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008417 return 1ULL << 20;
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008418
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008419 return div64_u64(runtime << 20, period);
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008420}
8421
Dhaval Giani521f1a242008-02-28 15:21:56 +05308422/* Must be called with tasklist_lock held */
8423static inline int tg_has_rt_tasks(struct task_group *tg)
8424{
8425 struct task_struct *g, *p;
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008426
Dhaval Giani521f1a242008-02-28 15:21:56 +05308427 do_each_thread(g, p) {
8428 if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg)
8429 return 1;
8430 } while_each_thread(g, p);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008431
Dhaval Giani521f1a242008-02-28 15:21:56 +05308432 return 0;
8433}
8434
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008435struct rt_schedulable_data {
8436 struct task_group *tg;
8437 u64 rt_period;
8438 u64 rt_runtime;
8439};
8440
8441static int tg_schedulable(struct task_group *tg, void *data)
8442{
8443 struct rt_schedulable_data *d = data;
8444 struct task_group *child;
8445 unsigned long total, sum = 0;
8446 u64 period, runtime;
8447
8448 period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8449 runtime = tg->rt_bandwidth.rt_runtime;
8450
8451 if (tg == d->tg) {
8452 period = d->rt_period;
8453 runtime = d->rt_runtime;
8454 }
8455
Peter Zijlstra4653f802008-09-23 15:33:44 +02008456 /*
8457 * Cannot have more runtime than the period.
8458 */
8459 if (runtime > period && runtime != RUNTIME_INF)
8460 return -EINVAL;
8461
8462 /*
8463 * Ensure we don't starve existing RT tasks.
8464 */
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008465 if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
8466 return -EBUSY;
8467
8468 total = to_ratio(period, runtime);
8469
Peter Zijlstra4653f802008-09-23 15:33:44 +02008470 /*
8471 * Nobody can have more than the global setting allows.
8472 */
8473 if (total > to_ratio(global_rt_period(), global_rt_runtime()))
8474 return -EINVAL;
8475
8476 /*
8477 * The sum of our children's runtime should not exceed our own.
8478 */
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008479 list_for_each_entry_rcu(child, &tg->children, siblings) {
8480 period = ktime_to_ns(child->rt_bandwidth.rt_period);
8481 runtime = child->rt_bandwidth.rt_runtime;
8482
8483 if (child == d->tg) {
8484 period = d->rt_period;
8485 runtime = d->rt_runtime;
8486 }
8487
8488 sum += to_ratio(period, runtime);
8489 }
8490
8491 if (sum > total)
8492 return -EINVAL;
8493
8494 return 0;
8495}
8496
8497static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
8498{
8499 struct rt_schedulable_data data = {
8500 .tg = tg,
8501 .rt_period = period,
8502 .rt_runtime = runtime,
8503 };
8504
8505 return walk_tg_tree(tg_schedulable, tg_nop, &data);
8506}
8507
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008508static int tg_set_bandwidth(struct task_group *tg,
8509 u64 rt_period, u64 rt_runtime)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008510{
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008511 int i, err = 0;
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008512
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008513 mutex_lock(&rt_constraints_mutex);
Dhaval Giani521f1a242008-02-28 15:21:56 +05308514 read_lock(&tasklist_lock);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008515 err = __rt_schedulable(tg, rt_period, rt_runtime);
8516 if (err)
Dhaval Giani521f1a242008-02-28 15:21:56 +05308517 goto unlock;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008518
Thomas Gleixner0986b112009-11-17 15:32:06 +01008519 raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008520 tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
8521 tg->rt_bandwidth.rt_runtime = rt_runtime;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008522
8523 for_each_possible_cpu(i) {
8524 struct rt_rq *rt_rq = tg->rt_rq[i];
8525
Thomas Gleixner0986b112009-11-17 15:32:06 +01008526 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008527 rt_rq->rt_runtime = rt_runtime;
Thomas Gleixner0986b112009-11-17 15:32:06 +01008528 raw_spin_unlock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008529 }
Thomas Gleixner0986b112009-11-17 15:32:06 +01008530 raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008531 unlock:
Dhaval Giani521f1a242008-02-28 15:21:56 +05308532 read_unlock(&tasklist_lock);
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008533 mutex_unlock(&rt_constraints_mutex);
8534
8535 return err;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008536}
8537
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008538int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
8539{
8540 u64 rt_runtime, rt_period;
8541
8542 rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8543 rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
8544 if (rt_runtime_us < 0)
8545 rt_runtime = RUNTIME_INF;
8546
8547 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8548}
8549
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008550long sched_group_rt_runtime(struct task_group *tg)
8551{
8552 u64 rt_runtime_us;
8553
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008554 if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008555 return -1;
8556
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008557 rt_runtime_us = tg->rt_bandwidth.rt_runtime;
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008558 do_div(rt_runtime_us, NSEC_PER_USEC);
8559 return rt_runtime_us;
8560}
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008561
8562int sched_group_set_rt_period(struct task_group *tg, long rt_period_us)
8563{
8564 u64 rt_runtime, rt_period;
8565
8566 rt_period = (u64)rt_period_us * NSEC_PER_USEC;
8567 rt_runtime = tg->rt_bandwidth.rt_runtime;
8568
Raistlin619b0482008-06-26 18:54:09 +02008569 if (rt_period == 0)
8570 return -EINVAL;
8571
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008572 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8573}
8574
8575long sched_group_rt_period(struct task_group *tg)
8576{
8577 u64 rt_period_us;
8578
8579 rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
8580 do_div(rt_period_us, NSEC_PER_USEC);
8581 return rt_period_us;
8582}
8583
8584static int sched_rt_global_constraints(void)
8585{
Peter Zijlstra4653f802008-09-23 15:33:44 +02008586 u64 runtime, period;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008587 int ret = 0;
8588
Hiroshi Shimamotoec5d4982008-09-10 17:00:19 -07008589 if (sysctl_sched_rt_period <= 0)
8590 return -EINVAL;
8591
Peter Zijlstra4653f802008-09-23 15:33:44 +02008592 runtime = global_rt_runtime();
8593 period = global_rt_period();
8594
8595 /*
8596 * Sanity check on the sysctl variables.
8597 */
8598 if (runtime > period && runtime != RUNTIME_INF)
8599 return -EINVAL;
Peter Zijlstra10b612f2008-06-19 14:22:27 +02008600
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008601 mutex_lock(&rt_constraints_mutex);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008602 read_lock(&tasklist_lock);
Peter Zijlstra4653f802008-09-23 15:33:44 +02008603 ret = __rt_schedulable(NULL, 0, 0);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008604 read_unlock(&tasklist_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008605 mutex_unlock(&rt_constraints_mutex);
8606
8607 return ret;
8608}
Dhaval Giani54e99122009-02-27 15:13:54 +05308609
8610int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
8611{
8612 /* Don't accept realtime tasks when there is no way for them to run */
8613 if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0)
8614 return 0;
8615
8616 return 1;
8617}
8618
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008619#else /* !CONFIG_RT_GROUP_SCHED */
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008620static int sched_rt_global_constraints(void)
8621{
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008622 unsigned long flags;
8623 int i;
8624
Hiroshi Shimamotoec5d4982008-09-10 17:00:19 -07008625 if (sysctl_sched_rt_period <= 0)
8626 return -EINVAL;
8627
Peter Zijlstra60aa6052009-05-05 17:50:21 +02008628 /*
8629 * There's always some RT tasks in the root group
8630 * -- migration, kstopmachine etc..
8631 */
8632 if (sysctl_sched_rt_runtime == 0)
8633 return -EBUSY;
8634
Thomas Gleixner0986b112009-11-17 15:32:06 +01008635 raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008636 for_each_possible_cpu(i) {
8637 struct rt_rq *rt_rq = &cpu_rq(i)->rt;
8638
Thomas Gleixner0986b112009-11-17 15:32:06 +01008639 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008640 rt_rq->rt_runtime = global_rt_runtime();
Thomas Gleixner0986b112009-11-17 15:32:06 +01008641 raw_spin_unlock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008642 }
Thomas Gleixner0986b112009-11-17 15:32:06 +01008643 raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008644
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008645 return 0;
8646}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008647#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008648
8649int sched_rt_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07008650 void __user *buffer, size_t *lenp,
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008651 loff_t *ppos)
8652{
8653 int ret;
8654 int old_period, old_runtime;
8655 static DEFINE_MUTEX(mutex);
8656
8657 mutex_lock(&mutex);
8658 old_period = sysctl_sched_rt_period;
8659 old_runtime = sysctl_sched_rt_runtime;
8660
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07008661 ret = proc_dointvec(table, write, buffer, lenp, ppos);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008662
8663 if (!ret && write) {
8664 ret = sched_rt_global_constraints();
8665 if (ret) {
8666 sysctl_sched_rt_period = old_period;
8667 sysctl_sched_rt_runtime = old_runtime;
8668 } else {
8669 def_rt_bandwidth.rt_runtime = global_rt_runtime();
8670 def_rt_bandwidth.rt_period =
8671 ns_to_ktime(global_rt_period());
8672 }
8673 }
8674 mutex_unlock(&mutex);
8675
8676 return ret;
8677}
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008678
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008679#ifdef CONFIG_CGROUP_SCHED
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008680
8681/* return corresponding task_group object of a cgroup */
Paul Menage2b01dfe2007-10-24 18:23:50 +02008682static inline struct task_group *cgroup_tg(struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008683{
Paul Menage2b01dfe2007-10-24 18:23:50 +02008684 return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id),
8685 struct task_group, css);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008686}
8687
8688static struct cgroup_subsys_state *
Paul Menage2b01dfe2007-10-24 18:23:50 +02008689cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008690{
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008691 struct task_group *tg, *parent;
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008692
Paul Menage2b01dfe2007-10-24 18:23:50 +02008693 if (!cgrp->parent) {
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008694 /* This is early initialization for the top cgroup */
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008695 return &init_task_group.css;
8696 }
8697
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008698 parent = cgroup_tg(cgrp->parent);
8699 tg = sched_create_group(parent);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008700 if (IS_ERR(tg))
8701 return ERR_PTR(-ENOMEM);
8702
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008703 return &tg->css;
8704}
8705
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01008706static void
8707cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008708{
Paul Menage2b01dfe2007-10-24 18:23:50 +02008709 struct task_group *tg = cgroup_tg(cgrp);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008710
8711 sched_destroy_group(tg);
8712}
8713
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01008714static int
Ben Blumbe367d02009-09-23 15:56:31 -07008715cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008716{
Peter Zijlstrab68aa232008-02-13 15:45:40 +01008717#ifdef CONFIG_RT_GROUP_SCHED
Dhaval Giani54e99122009-02-27 15:13:54 +05308718 if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk))
Peter Zijlstrab68aa232008-02-13 15:45:40 +01008719 return -EINVAL;
8720#else
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008721 /* We don't support RT-tasks being in separate groups */
8722 if (tsk->sched_class != &fair_sched_class)
8723 return -EINVAL;
Peter Zijlstrab68aa232008-02-13 15:45:40 +01008724#endif
Ben Blumbe367d02009-09-23 15:56:31 -07008725 return 0;
8726}
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008727
Ben Blumbe367d02009-09-23 15:56:31 -07008728static int
8729cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
8730 struct task_struct *tsk, bool threadgroup)
8731{
8732 int retval = cpu_cgroup_can_attach_task(cgrp, tsk);
8733 if (retval)
8734 return retval;
8735 if (threadgroup) {
8736 struct task_struct *c;
8737 rcu_read_lock();
8738 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) {
8739 retval = cpu_cgroup_can_attach_task(cgrp, c);
8740 if (retval) {
8741 rcu_read_unlock();
8742 return retval;
8743 }
8744 }
8745 rcu_read_unlock();
8746 }
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008747 return 0;
8748}
8749
8750static void
Paul Menage2b01dfe2007-10-24 18:23:50 +02008751cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
Ben Blumbe367d02009-09-23 15:56:31 -07008752 struct cgroup *old_cont, struct task_struct *tsk,
8753 bool threadgroup)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008754{
8755 sched_move_task(tsk);
Ben Blumbe367d02009-09-23 15:56:31 -07008756 if (threadgroup) {
8757 struct task_struct *c;
8758 rcu_read_lock();
8759 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) {
8760 sched_move_task(c);
8761 }
8762 rcu_read_unlock();
8763 }
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008764}
8765
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008766#ifdef CONFIG_FAIR_GROUP_SCHED
Paul Menagef4c753b2008-04-29 00:59:56 -07008767static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype,
Paul Menage2b01dfe2007-10-24 18:23:50 +02008768 u64 shareval)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008769{
Paul Menage2b01dfe2007-10-24 18:23:50 +02008770 return sched_group_set_shares(cgroup_tg(cgrp), shareval);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008771}
8772
Paul Menagef4c753b2008-04-29 00:59:56 -07008773static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008774{
Paul Menage2b01dfe2007-10-24 18:23:50 +02008775 struct task_group *tg = cgroup_tg(cgrp);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008776
8777 return (u64) tg->shares;
8778}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008779#endif /* CONFIG_FAIR_GROUP_SCHED */
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008780
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008781#ifdef CONFIG_RT_GROUP_SCHED
Mirco Tischler0c708142008-05-14 16:05:46 -07008782static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft,
Paul Menage06ecb272008-04-29 01:00:06 -07008783 s64 val)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008784{
Paul Menage06ecb272008-04-29 01:00:06 -07008785 return sched_group_set_rt_runtime(cgroup_tg(cgrp), val);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008786}
8787
Paul Menage06ecb272008-04-29 01:00:06 -07008788static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008789{
Paul Menage06ecb272008-04-29 01:00:06 -07008790 return sched_group_rt_runtime(cgroup_tg(cgrp));
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008791}
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008792
8793static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype,
8794 u64 rt_period_us)
8795{
8796 return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us);
8797}
8798
8799static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft)
8800{
8801 return sched_group_rt_period(cgroup_tg(cgrp));
8802}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008803#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008804
Paul Menagefe5c7cc2007-10-29 21:18:11 +01008805static struct cftype cpu_files[] = {
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008806#ifdef CONFIG_FAIR_GROUP_SCHED
Paul Menagefe5c7cc2007-10-29 21:18:11 +01008807 {
8808 .name = "shares",
Paul Menagef4c753b2008-04-29 00:59:56 -07008809 .read_u64 = cpu_shares_read_u64,
8810 .write_u64 = cpu_shares_write_u64,
Paul Menagefe5c7cc2007-10-29 21:18:11 +01008811 },
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008812#endif
8813#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008814 {
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008815 .name = "rt_runtime_us",
Paul Menage06ecb272008-04-29 01:00:06 -07008816 .read_s64 = cpu_rt_runtime_read,
8817 .write_s64 = cpu_rt_runtime_write,
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008818 },
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008819 {
8820 .name = "rt_period_us",
Paul Menagef4c753b2008-04-29 00:59:56 -07008821 .read_u64 = cpu_rt_period_read_uint,
8822 .write_u64 = cpu_rt_period_write_uint,
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008823 },
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008824#endif
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008825};
8826
8827static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont)
8828{
Paul Menagefe5c7cc2007-10-29 21:18:11 +01008829 return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files));
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008830}
8831
8832struct cgroup_subsys cpu_cgroup_subsys = {
Ingo Molnar38605ca2007-10-29 21:18:11 +01008833 .name = "cpu",
8834 .create = cpu_cgroup_create,
8835 .destroy = cpu_cgroup_destroy,
8836 .can_attach = cpu_cgroup_can_attach,
8837 .attach = cpu_cgroup_attach,
8838 .populate = cpu_cgroup_populate,
8839 .subsys_id = cpu_cgroup_subsys_id,
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008840 .early_init = 1,
8841};
8842
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008843#endif /* CONFIG_CGROUP_SCHED */
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008844
8845#ifdef CONFIG_CGROUP_CPUACCT
8846
8847/*
8848 * CPU accounting code for task groups.
8849 *
8850 * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
8851 * (balbir@in.ibm.com).
8852 */
8853
Bharata B Rao934352f2008-11-10 20:41:13 +05308854/* track cpu usage of a group of tasks and its child groups */
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008855struct cpuacct {
8856 struct cgroup_subsys_state css;
8857 /* cpuusage holds pointer to a u64-type object on every cpu */
Tejun Heo43cf38e2010-02-02 14:38:57 +09008858 u64 __percpu *cpuusage;
Bharata B Raoef12fef2009-03-31 10:02:22 +05308859 struct percpu_counter cpustat[CPUACCT_STAT_NSTATS];
Bharata B Rao934352f2008-11-10 20:41:13 +05308860 struct cpuacct *parent;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008861};
8862
8863struct cgroup_subsys cpuacct_subsys;
8864
8865/* return cpu accounting group corresponding to this container */
Dhaval Giani32cd7562008-02-29 10:02:43 +05308866static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008867{
Dhaval Giani32cd7562008-02-29 10:02:43 +05308868 return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id),
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008869 struct cpuacct, css);
8870}
8871
8872/* return cpu accounting group to which this task belongs */
8873static inline struct cpuacct *task_ca(struct task_struct *tsk)
8874{
8875 return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
8876 struct cpuacct, css);
8877}
8878
8879/* create a new cpu accounting group */
8880static struct cgroup_subsys_state *cpuacct_create(
Dhaval Giani32cd7562008-02-29 10:02:43 +05308881 struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008882{
8883 struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
Bharata B Raoef12fef2009-03-31 10:02:22 +05308884 int i;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008885
8886 if (!ca)
Bharata B Raoef12fef2009-03-31 10:02:22 +05308887 goto out;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008888
8889 ca->cpuusage = alloc_percpu(u64);
Bharata B Raoef12fef2009-03-31 10:02:22 +05308890 if (!ca->cpuusage)
8891 goto out_free_ca;
8892
8893 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
8894 if (percpu_counter_init(&ca->cpustat[i], 0))
8895 goto out_free_counters;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008896
Bharata B Rao934352f2008-11-10 20:41:13 +05308897 if (cgrp->parent)
8898 ca->parent = cgroup_ca(cgrp->parent);
8899
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008900 return &ca->css;
Bharata B Raoef12fef2009-03-31 10:02:22 +05308901
8902out_free_counters:
8903 while (--i >= 0)
8904 percpu_counter_destroy(&ca->cpustat[i]);
8905 free_percpu(ca->cpuusage);
8906out_free_ca:
8907 kfree(ca);
8908out:
8909 return ERR_PTR(-ENOMEM);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008910}
8911
8912/* destroy an existing cpu accounting group */
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01008913static void
Dhaval Giani32cd7562008-02-29 10:02:43 +05308914cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008915{
Dhaval Giani32cd7562008-02-29 10:02:43 +05308916 struct cpuacct *ca = cgroup_ca(cgrp);
Bharata B Raoef12fef2009-03-31 10:02:22 +05308917 int i;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008918
Bharata B Raoef12fef2009-03-31 10:02:22 +05308919 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
8920 percpu_counter_destroy(&ca->cpustat[i]);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008921 free_percpu(ca->cpuusage);
8922 kfree(ca);
8923}
8924
Ken Chen720f5492008-12-15 22:02:01 -08008925static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu)
8926{
Rusty Russellb36128c2009-02-20 16:29:08 +09008927 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
Ken Chen720f5492008-12-15 22:02:01 -08008928 u64 data;
8929
8930#ifndef CONFIG_64BIT
8931 /*
8932 * Take rq->lock to make 64-bit read safe on 32-bit platforms.
8933 */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008934 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08008935 data = *cpuusage;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008936 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08008937#else
8938 data = *cpuusage;
8939#endif
8940
8941 return data;
8942}
8943
8944static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val)
8945{
Rusty Russellb36128c2009-02-20 16:29:08 +09008946 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
Ken Chen720f5492008-12-15 22:02:01 -08008947
8948#ifndef CONFIG_64BIT
8949 /*
8950 * Take rq->lock to make 64-bit write safe on 32-bit platforms.
8951 */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008952 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08008953 *cpuusage = val;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008954 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08008955#else
8956 *cpuusage = val;
8957#endif
8958}
8959
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008960/* return total cpu usage (in nanoseconds) of a group */
Dhaval Giani32cd7562008-02-29 10:02:43 +05308961static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008962{
Dhaval Giani32cd7562008-02-29 10:02:43 +05308963 struct cpuacct *ca = cgroup_ca(cgrp);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008964 u64 totalcpuusage = 0;
8965 int i;
8966
Ken Chen720f5492008-12-15 22:02:01 -08008967 for_each_present_cpu(i)
8968 totalcpuusage += cpuacct_cpuusage_read(ca, i);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008969
8970 return totalcpuusage;
8971}
8972
Dhaval Giani0297b802008-02-29 10:02:44 +05308973static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype,
8974 u64 reset)
8975{
8976 struct cpuacct *ca = cgroup_ca(cgrp);
8977 int err = 0;
8978 int i;
8979
8980 if (reset) {
8981 err = -EINVAL;
8982 goto out;
8983 }
8984
Ken Chen720f5492008-12-15 22:02:01 -08008985 for_each_present_cpu(i)
8986 cpuacct_cpuusage_write(ca, i, 0);
Dhaval Giani0297b802008-02-29 10:02:44 +05308987
Dhaval Giani0297b802008-02-29 10:02:44 +05308988out:
8989 return err;
8990}
8991
Ken Chene9515c32008-12-15 22:04:15 -08008992static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft,
8993 struct seq_file *m)
8994{
8995 struct cpuacct *ca = cgroup_ca(cgroup);
8996 u64 percpu;
8997 int i;
8998
8999 for_each_present_cpu(i) {
9000 percpu = cpuacct_cpuusage_read(ca, i);
9001 seq_printf(m, "%llu ", (unsigned long long) percpu);
9002 }
9003 seq_printf(m, "\n");
9004 return 0;
9005}
9006
Bharata B Raoef12fef2009-03-31 10:02:22 +05309007static const char *cpuacct_stat_desc[] = {
9008 [CPUACCT_STAT_USER] = "user",
9009 [CPUACCT_STAT_SYSTEM] = "system",
9010};
9011
9012static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft,
9013 struct cgroup_map_cb *cb)
9014{
9015 struct cpuacct *ca = cgroup_ca(cgrp);
9016 int i;
9017
9018 for (i = 0; i < CPUACCT_STAT_NSTATS; i++) {
9019 s64 val = percpu_counter_read(&ca->cpustat[i]);
9020 val = cputime64_to_clock_t(val);
9021 cb->fill(cb, cpuacct_stat_desc[i], val);
9022 }
9023 return 0;
9024}
9025
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009026static struct cftype files[] = {
9027 {
9028 .name = "usage",
Paul Menagef4c753b2008-04-29 00:59:56 -07009029 .read_u64 = cpuusage_read,
9030 .write_u64 = cpuusage_write,
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009031 },
Ken Chene9515c32008-12-15 22:04:15 -08009032 {
9033 .name = "usage_percpu",
9034 .read_seq_string = cpuacct_percpu_seq_read,
9035 },
Bharata B Raoef12fef2009-03-31 10:02:22 +05309036 {
9037 .name = "stat",
9038 .read_map = cpuacct_stats_show,
9039 },
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009040};
9041
Dhaval Giani32cd7562008-02-29 10:02:43 +05309042static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009043{
Dhaval Giani32cd7562008-02-29 10:02:43 +05309044 return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files));
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009045}
9046
9047/*
9048 * charge this task's execution time to its accounting group.
9049 *
9050 * called with rq->lock held.
9051 */
9052static void cpuacct_charge(struct task_struct *tsk, u64 cputime)
9053{
9054 struct cpuacct *ca;
Bharata B Rao934352f2008-11-10 20:41:13 +05309055 int cpu;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009056
Li Zefanc40c6f82009-02-26 15:40:15 +08009057 if (unlikely(!cpuacct_subsys.active))
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009058 return;
9059
Bharata B Rao934352f2008-11-10 20:41:13 +05309060 cpu = task_cpu(tsk);
Bharata B Raoa18b83b2009-03-23 10:02:53 +05309061
9062 rcu_read_lock();
9063
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009064 ca = task_ca(tsk);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009065
Bharata B Rao934352f2008-11-10 20:41:13 +05309066 for (; ca; ca = ca->parent) {
Rusty Russellb36128c2009-02-20 16:29:08 +09009067 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009068 *cpuusage += cputime;
9069 }
Bharata B Raoa18b83b2009-03-23 10:02:53 +05309070
9071 rcu_read_unlock();
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009072}
9073
Bharata B Raoef12fef2009-03-31 10:02:22 +05309074/*
Anton Blanchardfa535a72010-02-02 14:46:13 -08009075 * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large
9076 * in cputime_t units. As a result, cpuacct_update_stats calls
9077 * percpu_counter_add with values large enough to always overflow the
9078 * per cpu batch limit causing bad SMP scalability.
9079 *
9080 * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we
9081 * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled
9082 * and enabled. We cap it at INT_MAX which is the largest allowed batch value.
9083 */
9084#ifdef CONFIG_SMP
9085#define CPUACCT_BATCH \
9086 min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX)
9087#else
9088#define CPUACCT_BATCH 0
9089#endif
9090
9091/*
Bharata B Raoef12fef2009-03-31 10:02:22 +05309092 * Charge the system/user time to the task's accounting group.
9093 */
9094static void cpuacct_update_stats(struct task_struct *tsk,
9095 enum cpuacct_stat_index idx, cputime_t val)
9096{
9097 struct cpuacct *ca;
Anton Blanchardfa535a72010-02-02 14:46:13 -08009098 int batch = CPUACCT_BATCH;
Bharata B Raoef12fef2009-03-31 10:02:22 +05309099
9100 if (unlikely(!cpuacct_subsys.active))
9101 return;
9102
9103 rcu_read_lock();
9104 ca = task_ca(tsk);
9105
9106 do {
Anton Blanchardfa535a72010-02-02 14:46:13 -08009107 __percpu_counter_add(&ca->cpustat[idx], val, batch);
Bharata B Raoef12fef2009-03-31 10:02:22 +05309108 ca = ca->parent;
9109 } while (ca);
9110 rcu_read_unlock();
9111}
9112
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01009113struct cgroup_subsys cpuacct_subsys = {
9114 .name = "cpuacct",
9115 .create = cpuacct_create,
9116 .destroy = cpuacct_destroy,
9117 .populate = cpuacct_populate,
9118 .subsys_id = cpuacct_subsys_id,
9119};
9120#endif /* CONFIG_CGROUP_CPUACCT */
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009121
9122#ifndef CONFIG_SMP
9123
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009124void synchronize_sched_expedited(void)
9125{
Paul E. McKenneyfc390cd2010-05-06 11:42:52 -07009126 barrier();
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009127}
9128EXPORT_SYMBOL_GPL(synchronize_sched_expedited);
9129
9130#else /* #ifndef CONFIG_SMP */
9131
Paul E. McKenneycc631fb2010-05-06 18:49:21 +02009132static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0);
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009133
Paul E. McKenneycc631fb2010-05-06 18:49:21 +02009134static int synchronize_sched_expedited_cpu_stop(void *data)
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009135{
Tejun Heo969c7922010-05-06 18:49:21 +02009136 /*
9137 * There must be a full memory barrier on each affected CPU
9138 * between the time that try_stop_cpus() is called and the
9139 * time that it returns.
9140 *
9141 * In the current initial implementation of cpu_stop, the
9142 * above condition is already met when the control reaches
9143 * this point and the following smp_mb() is not strictly
9144 * necessary. Do smp_mb() anyway for documentation and
9145 * robustness against future implementation changes.
9146 */
Paul E. McKenneycc631fb2010-05-06 18:49:21 +02009147 smp_mb(); /* See above comment block. */
Tejun Heo969c7922010-05-06 18:49:21 +02009148 return 0;
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009149}
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009150
9151/*
9152 * Wait for an rcu-sched grace period to elapse, but use "big hammer"
9153 * approach to force grace period to end quickly. This consumes
9154 * significant time on all CPUs, and is thus not recommended for
9155 * any sort of common-case code.
9156 *
9157 * Note that it is illegal to call this function while holding any
9158 * lock that is acquired by a CPU-hotplug notifier. Failing to
9159 * observe this restriction will result in deadlock.
9160 */
9161void synchronize_sched_expedited(void)
9162{
Tejun Heo969c7922010-05-06 18:49:21 +02009163 int snap, trycount = 0;
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009164
9165 smp_mb(); /* ensure prior mod happens before capturing snap. */
Tejun Heo969c7922010-05-06 18:49:21 +02009166 snap = atomic_read(&synchronize_sched_expedited_count) + 1;
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009167 get_online_cpus();
Tejun Heo969c7922010-05-06 18:49:21 +02009168 while (try_stop_cpus(cpu_online_mask,
9169 synchronize_sched_expedited_cpu_stop,
Tejun Heo94458d52010-05-06 18:49:21 +02009170 NULL) == -EAGAIN) {
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009171 put_online_cpus();
9172 if (trycount++ < 10)
9173 udelay(trycount * num_online_cpus());
9174 else {
9175 synchronize_sched();
9176 return;
9177 }
Tejun Heo969c7922010-05-06 18:49:21 +02009178 if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) {
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009179 smp_mb(); /* ensure test happens before caller kfree */
9180 return;
9181 }
9182 get_online_cpus();
9183 }
Tejun Heo969c7922010-05-06 18:49:21 +02009184 atomic_inc(&synchronize_sched_expedited_count);
Paul E. McKenneycc631fb2010-05-06 18:49:21 +02009185 smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009186 put_online_cpus();
Paul E. McKenney03b042b2009-06-25 09:08:16 -07009187}
9188EXPORT_SYMBOL_GPL(synchronize_sched_expedited);
9189
9190#endif /* #else #ifndef CONFIG_SMP */