blob: f52a8801b7a285fb252ecc6935b55f525813881b [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"
80
Steven Rostedta8d154b2009-04-10 09:36:00 -040081#define CREATE_TRACE_POINTS
Steven Rostedtad8d75f2009-04-14 19:39:12 -040082#include <trace/events/sched.h>
Steven Rostedta8d154b2009-04-10 09:36:00 -040083
Linus Torvalds1da177e2005-04-16 15:20:36 -070084/*
85 * Convert user-nice values [ -20 ... 0 ... 19 ]
86 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
87 * and back.
88 */
89#define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
90#define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
91#define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
92
93/*
94 * 'User priority' is the nice value converted to something we
95 * can work with better when scaling various scheduler parameters,
96 * it's a [ 0 ... 39 ] range.
97 */
98#define USER_PRIO(p) ((p)-MAX_RT_PRIO)
99#define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
100#define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
101
102/*
Ingo Molnard7876a02008-01-25 21:08:19 +0100103 * Helpers for converting nanosecond timing to jiffy resolution
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 */
Eric Dumazetd6322fa2007-11-09 22:39:38 +0100105#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200107#define NICE_0_LOAD SCHED_LOAD_SCALE
108#define NICE_0_SHIFT SCHED_LOAD_SHIFT
109
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110/*
111 * These are the 'tuning knobs' of the scheduler:
112 *
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +0200113 * default timeslice is 100 msecs (used only for SCHED_RR tasks).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 * Timeslices get refilled after they expire.
115 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116#define DEF_TIMESLICE (100 * HZ / 1000)
Peter Williams2dd73a42006-06-27 02:54:34 -0700117
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200118/*
119 * single value that denotes runtime == period, ie unlimited time.
120 */
121#define RUNTIME_INF ((u64)~0ULL)
122
Ingo Molnare05606d2007-07-09 18:51:59 +0200123static inline int rt_policy(int policy)
124{
Roel Kluin3f33a7c2008-05-13 23:44:11 +0200125 if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR))
Ingo Molnare05606d2007-07-09 18:51:59 +0200126 return 1;
127 return 0;
128}
129
130static inline int task_has_rt_policy(struct task_struct *p)
131{
132 return rt_policy(p->policy);
133}
134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135/*
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200136 * This is the priority-queue data structure of the RT scheduling class:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 */
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200138struct rt_prio_array {
139 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
140 struct list_head queue[MAX_RT_PRIO];
141};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200143struct rt_bandwidth {
Ingo Molnarea736ed2008-03-25 13:51:45 +0100144 /* nests inside the rq lock: */
Thomas Gleixner0986b112009-11-17 15:32:06 +0100145 raw_spinlock_t rt_runtime_lock;
Ingo Molnarea736ed2008-03-25 13:51:45 +0100146 ktime_t rt_period;
147 u64 rt_runtime;
148 struct hrtimer rt_period_timer;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200149};
150
151static struct rt_bandwidth def_rt_bandwidth;
152
153static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
154
155static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
156{
157 struct rt_bandwidth *rt_b =
158 container_of(timer, struct rt_bandwidth, rt_period_timer);
159 ktime_t now;
160 int overrun;
161 int idle = 0;
162
163 for (;;) {
164 now = hrtimer_cb_get_time(timer);
165 overrun = hrtimer_forward(timer, now, rt_b->rt_period);
166
167 if (!overrun)
168 break;
169
170 idle = do_sched_rt_period_timer(rt_b, overrun);
171 }
172
173 return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
174}
175
176static
177void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
178{
179 rt_b->rt_period = ns_to_ktime(period);
180 rt_b->rt_runtime = runtime;
181
Thomas Gleixner0986b112009-11-17 15:32:06 +0100182 raw_spin_lock_init(&rt_b->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200183
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200184 hrtimer_init(&rt_b->rt_period_timer,
185 CLOCK_MONOTONIC, HRTIMER_MODE_REL);
186 rt_b->rt_period_timer.function = sched_rt_period_timer;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200187}
188
Krzysztof Heltc8bfff62008-09-05 23:46:19 +0200189static inline int rt_bandwidth_enabled(void)
190{
191 return sysctl_sched_rt_runtime >= 0;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200192}
193
194static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
195{
196 ktime_t now;
197
Hiroshi Shimamotocac64d02009-02-25 09:59:26 -0800198 if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200199 return;
200
201 if (hrtimer_active(&rt_b->rt_period_timer))
202 return;
203
Thomas Gleixner0986b112009-11-17 15:32:06 +0100204 raw_spin_lock(&rt_b->rt_runtime_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200205 for (;;) {
Peter Zijlstra7f1e2ca2009-03-13 12:21:27 +0100206 unsigned long delta;
207 ktime_t soft, hard;
208
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200209 if (hrtimer_active(&rt_b->rt_period_timer))
210 break;
211
212 now = hrtimer_cb_get_time(&rt_b->rt_period_timer);
213 hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period);
Peter Zijlstra7f1e2ca2009-03-13 12:21:27 +0100214
215 soft = hrtimer_get_softexpires(&rt_b->rt_period_timer);
216 hard = hrtimer_get_expires(&rt_b->rt_period_timer);
217 delta = ktime_to_ns(ktime_sub(hard, soft));
218 __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta,
Arun R Bharadwaj5c333862009-04-16 12:14:37 +0530219 HRTIMER_MODE_ABS_PINNED, 0);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200220 }
Thomas Gleixner0986b112009-11-17 15:32:06 +0100221 raw_spin_unlock(&rt_b->rt_runtime_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200222}
223
224#ifdef CONFIG_RT_GROUP_SCHED
225static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
226{
227 hrtimer_cancel(&rt_b->rt_period_timer);
228}
229#endif
230
Heiko Carstens712555e2008-04-28 11:33:07 +0200231/*
232 * sched_domains_mutex serializes calls to arch_init_sched_domains,
233 * detach_destroy_domains and partition_sched_domains.
234 */
235static DEFINE_MUTEX(sched_domains_mutex);
236
Dhaval Giani7c941432010-01-20 13:26:18 +0100237#ifdef CONFIG_CGROUP_SCHED
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200238
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -0700239#include <linux/cgroup.h>
240
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200241struct cfs_rq;
242
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100243static LIST_HEAD(task_groups);
244
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200245/* task group related information */
Ingo Molnar4cf86d72007-10-15 17:00:14 +0200246struct task_group {
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -0700247 struct cgroup_subsys_state css;
Arun R Bharadwaj6c415b92008-12-01 20:49:05 +0530248
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100249#ifdef CONFIG_FAIR_GROUP_SCHED
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200250 /* schedulable entities of this group on each cpu */
251 struct sched_entity **se;
252 /* runqueue "owned" by this group on each cpu */
253 struct cfs_rq **cfs_rq;
254 unsigned long shares;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100255#endif
256
257#ifdef CONFIG_RT_GROUP_SCHED
258 struct sched_rt_entity **rt_se;
259 struct rt_rq **rt_rq;
260
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200261 struct rt_bandwidth rt_bandwidth;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100262#endif
Srivatsa Vaddagiri6b2d7702008-01-25 21:08:00 +0100263
Srivatsa Vaddagiriae8393e2007-10-29 21:18:11 +0100264 struct rcu_head rcu;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100265 struct list_head list;
Peter Zijlstraf473aa52008-04-19 19:45:00 +0200266
267 struct task_group *parent;
268 struct list_head siblings;
269 struct list_head children;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200270};
271
Peter Zijlstraeff766a2008-04-19 19:45:00 +0200272#define root_task_group init_task_group
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100273
Peter Zijlstra8ed36992008-02-13 15:45:39 +0100274/* task_group_lock serializes add/remove of task groups and also changes to
Srivatsa Vaddagiriec2c5072008-01-25 21:07:59 +0100275 * a task group's cpu shares.
276 */
Peter Zijlstra8ed36992008-02-13 15:45:39 +0100277static DEFINE_SPINLOCK(task_group_lock);
Srivatsa Vaddagiriec2c5072008-01-25 21:07:59 +0100278
Cyrill Gorcunove9036b32009-10-26 22:24:14 +0300279#ifdef CONFIG_FAIR_GROUP_SCHED
280
Peter Zijlstra57310a92009-03-09 13:56:21 +0100281#ifdef CONFIG_SMP
282static int root_task_group_empty(void)
283{
284 return list_empty(&root_task_group.children);
285}
286#endif
287
Srivatsa Vaddagiri93f992c2008-01-25 21:07:59 +0100288# define INIT_TASK_GROUP_LOAD NICE_0_LOAD
Srivatsa Vaddagiri24e377a2007-10-15 17:00:09 +0200289
Miao Xiecb4ad1f2008-04-28 12:54:56 +0800290/*
Lai Jiangshan2e084782008-06-12 16:42:58 +0800291 * A weight of 0 or 1 can cause arithmetics problems.
292 * A weight of a cfs_rq is the sum of weights of which entities
293 * are queued on this cfs_rq, so a weight of a entity should not be
294 * too large, so as the shares value of a task group.
Miao Xiecb4ad1f2008-04-28 12:54:56 +0800295 * (The default weight is 1024 - so there's no practical
296 * limitation from this.)
297 */
Peter Zijlstra18d95a22008-04-19 19:45:00 +0200298#define MIN_SHARES 2
Lai Jiangshan2e084782008-06-12 16:42:58 +0800299#define MAX_SHARES (1UL << 18)
Peter Zijlstra18d95a22008-04-19 19:45:00 +0200300
Srivatsa Vaddagiri93f992c2008-01-25 21:07:59 +0100301static int init_task_group_load = INIT_TASK_GROUP_LOAD;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100302#endif
303
304/* Default task group.
305 * Every task in system belong to this group at bootup.
306 */
Mike Travis434d53b2008-04-04 18:11:04 -0700307struct task_group init_task_group;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200308
Dhaval Giani7c941432010-01-20 13:26:18 +0100309#endif /* CONFIG_CGROUP_SCHED */
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +0200310
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200311/* CFS-related fields in a runqueue */
312struct cfs_rq {
313 struct load_weight load;
314 unsigned long nr_running;
315
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200316 u64 exec_clock;
Ingo Molnare9acbff2007-10-15 17:00:04 +0200317 u64 min_vruntime;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200318
319 struct rb_root tasks_timeline;
320 struct rb_node *rb_leftmost;
Peter Zijlstra4a55bd52008-04-19 19:45:00 +0200321
322 struct list_head tasks;
323 struct list_head *balance_iterator;
324
325 /*
326 * 'curr' points to currently running entity on this cfs_rq.
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200327 * It is set to NULL otherwise (i.e when none are currently running).
328 */
Peter Zijlstra47932412008-11-04 21:25:09 +0100329 struct sched_entity *curr, *next, *last;
Peter Zijlstraddc97292007-10-15 17:00:10 +0200330
Peter Zijlstra5ac5c4d2008-11-10 10:46:32 +0100331 unsigned int nr_spread_over;
Peter Zijlstraddc97292007-10-15 17:00:10 +0200332
Ingo Molnar62160e3f2007-10-15 17:00:03 +0200333#ifdef CONFIG_FAIR_GROUP_SCHED
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200334 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
335
Ingo Molnar41a2d6c2007-12-05 15:46:09 +0100336 /*
337 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200338 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
339 * (like users, containers etc.)
340 *
341 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
342 * list is used during load balance.
343 */
Ingo Molnar41a2d6c2007-12-05 15:46:09 +0100344 struct list_head leaf_cfs_rq_list;
345 struct task_group *tg; /* group that "owns" this runqueue */
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200346
347#ifdef CONFIG_SMP
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200348 /*
Peter Zijlstrac8cba852008-06-27 13:41:23 +0200349 * the part of load.weight contributed by tasks
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200350 */
Peter Zijlstrac8cba852008-06-27 13:41:23 +0200351 unsigned long task_weight;
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200352
Peter Zijlstrac8cba852008-06-27 13:41:23 +0200353 /*
354 * h_load = weight * f(tg)
355 *
356 * Where f(tg) is the recursive weight fraction assigned to
357 * this group.
358 */
359 unsigned long h_load;
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200360
Peter Zijlstrac8cba852008-06-27 13:41:23 +0200361 /*
362 * this cpu's part of tg->shares
363 */
364 unsigned long shares;
Peter Zijlstraf1d239f2008-06-27 13:41:38 +0200365
366 /*
367 * load.weight at the time we set shares
368 */
369 unsigned long rq_weight;
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200370#endif
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200371#endif
372};
373
374/* Real-Time classes' related field in a runqueue: */
375struct rt_rq {
376 struct rt_prio_array active;
Steven Rostedt63489e42008-01-25 21:08:03 +0100377 unsigned long rt_nr_running;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100378#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
Gregory Haskinse864c492008-12-29 09:39:49 -0500379 struct {
380 int curr; /* highest queued rt task prio */
Gregory Haskins398a1532009-01-14 09:10:04 -0500381#ifdef CONFIG_SMP
Gregory Haskinse864c492008-12-29 09:39:49 -0500382 int next; /* next highest */
Gregory Haskins398a1532009-01-14 09:10:04 -0500383#endif
Gregory Haskinse864c492008-12-29 09:39:49 -0500384 } highest_prio;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100385#endif
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100386#ifdef CONFIG_SMP
Gregory Haskins73fe6aa2008-01-25 21:08:07 +0100387 unsigned long rt_nr_migratory;
Peter Zijlstraa1ba4d82009-04-01 18:40:15 +0200388 unsigned long rt_nr_total;
Gregory Haskinsa22d7fc2008-01-25 21:08:12 +0100389 int overloaded;
Gregory Haskins917b6272008-12-29 09:39:53 -0500390 struct plist_head pushable_tasks;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100391#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100392 int rt_throttled;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100393 u64 rt_time;
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200394 u64 rt_runtime;
Ingo Molnarea736ed2008-03-25 13:51:45 +0100395 /* Nests inside the rq lock: */
Thomas Gleixner0986b112009-11-17 15:32:06 +0100396 raw_spinlock_t rt_runtime_lock;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100397
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100398#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +0100399 unsigned long rt_nr_boosted;
400
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100401 struct rq *rq;
402 struct list_head leaf_rt_rq_list;
403 struct task_group *tg;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100404#endif
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200405};
406
Gregory Haskins57d885f2008-01-25 21:08:18 +0100407#ifdef CONFIG_SMP
408
409/*
410 * We add the notion of a root-domain which will be used to define per-domain
Ingo Molnar0eab9142008-01-25 21:08:19 +0100411 * variables. Each exclusive cpuset essentially defines an island domain by
412 * fully partitioning the member cpus from any other cpuset. Whenever a new
Gregory Haskins57d885f2008-01-25 21:08:18 +0100413 * exclusive cpuset is created, we also create and attach a new root-domain
414 * object.
415 *
Gregory Haskins57d885f2008-01-25 21:08:18 +0100416 */
417struct root_domain {
418 atomic_t refcount;
Rusty Russellc6c49272008-11-25 02:35:05 +1030419 cpumask_var_t span;
420 cpumask_var_t online;
Gregory Haskins637f5082008-01-25 21:08:18 +0100421
Ingo Molnar0eab9142008-01-25 21:08:19 +0100422 /*
Gregory Haskins637f5082008-01-25 21:08:18 +0100423 * The "RT overload" flag: it gets set if a CPU has more than
424 * one runnable RT task.
425 */
Rusty Russellc6c49272008-11-25 02:35:05 +1030426 cpumask_var_t rto_mask;
Ingo Molnar0eab9142008-01-25 21:08:19 +0100427 atomic_t rto_count;
Gregory Haskins6e0534f2008-05-12 21:21:01 +0200428#ifdef CONFIG_SMP
429 struct cpupri cpupri;
430#endif
Gregory Haskins57d885f2008-01-25 21:08:18 +0100431};
432
Gregory Haskinsdc938522008-01-25 21:08:26 +0100433/*
434 * By default the system creates a single root-domain with all cpus as
435 * members (mimicking the global state we have today).
436 */
Gregory Haskins57d885f2008-01-25 21:08:18 +0100437static struct root_domain def_root_domain;
438
439#endif
440
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200441/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 * This is the main, per-CPU runqueue data structure.
443 *
444 * Locking rule: those places that want to lock multiple runqueues
445 * (such as the load balancing or the thread migration code), lock
446 * acquire operations must be ordered by ascending &runqueue.
447 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700448struct rq {
Ingo Molnard8016492007-10-18 21:32:55 +0200449 /* runqueue lock: */
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100450 raw_spinlock_t lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
452 /*
453 * nr_running and cpu_load should be in the same cacheline because
454 * remote CPUs use both these fields when doing load calculation.
455 */
456 unsigned long nr_running;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200457 #define CPU_LOAD_IDX_MAX 5
458 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -0700459#ifdef CONFIG_NO_HZ
Mike Galbraith39c0cbe2010-03-11 17:17:13 +0100460 u64 nohz_stamp;
Siddha, Suresh B46cb4b72007-05-08 00:32:51 -0700461 unsigned char in_nohz_recently;
462#endif
Mike Galbraitha64692a2010-03-11 17:16:20 +0100463 unsigned int skip_clock_update;
464
Ingo Molnard8016492007-10-18 21:32:55 +0200465 /* capture load from *all* tasks on this cpu: */
466 struct load_weight load;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200467 unsigned long nr_load_updates;
468 u64 nr_switches;
469
470 struct cfs_rq cfs;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100471 struct rt_rq rt;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100472
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200473#ifdef CONFIG_FAIR_GROUP_SCHED
Ingo Molnard8016492007-10-18 21:32:55 +0200474 /* list of leaf cfs_rq on this cpu: */
475 struct list_head leaf_cfs_rq_list;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100476#endif
477#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100478 struct list_head leaf_rt_rq_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
481 /*
482 * This is part of a global counter where only the total sum
483 * over all CPUs matters. A task can increase this counter on
484 * one CPU and if it got migrated afterwards it may decrease
485 * it on another CPU. Always updated under the runqueue lock:
486 */
487 unsigned long nr_uninterruptible;
488
Ingo Molnar36c8b582006-07-03 00:25:41 -0700489 struct task_struct *curr, *idle;
Christoph Lameterc9819f42006-12-10 02:20:25 -0800490 unsigned long next_balance;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 struct mm_struct *prev_mm;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200492
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200493 u64 clock;
Ingo Molnar6aa645e2007-07-09 18:51:58 +0200494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 atomic_t nr_iowait;
496
497#ifdef CONFIG_SMP
Ingo Molnar0eab9142008-01-25 21:08:19 +0100498 struct root_domain *rd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 struct sched_domain *sd;
500
Peter Zijlstrae51fd5e2010-05-31 12:37:30 +0200501 unsigned long cpu_power;
502
Henrik Austada0a522c2009-02-13 20:35:45 +0100503 unsigned char idle_at_tick;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 /* For active balancing */
Gregory Haskins3f029d32009-07-29 11:08:47 -0400505 int post_schedule;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 int active_balance;
507 int push_cpu;
Tejun Heo969c7922010-05-06 18:49:21 +0200508 struct cpu_stop_work active_balance_work;
Ingo Molnard8016492007-10-18 21:32:55 +0200509 /* cpu of this runqueue: */
510 int cpu;
Gregory Haskins1f11eb62008-06-04 15:04:05 -0400511 int online;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Peter Zijlstraa8a51d52008-06-27 13:41:26 +0200513 unsigned long avg_load_per_task;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
Peter Zijlstrae9e92502009-09-01 10:34:37 +0200515 u64 rt_avg;
516 u64 age_stamp;
Mike Galbraith1b9508f2009-11-04 17:53:50 +0100517 u64 idle_stamp;
518 u64 avg_idle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519#endif
520
Thomas Gleixnerdce48a82009-04-11 10:43:41 +0200521 /* calc_load related fields */
522 unsigned long calc_load_update;
523 long calc_load_active;
524
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100525#ifdef CONFIG_SCHED_HRTICK
Peter Zijlstra31656512008-07-18 18:01:23 +0200526#ifdef CONFIG_SMP
527 int hrtick_csd_pending;
528 struct call_single_data hrtick_csd;
529#endif
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100530 struct hrtimer hrtick_timer;
531#endif
532
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533#ifdef CONFIG_SCHEDSTATS
534 /* latency stats */
535 struct sched_info rq_sched_info;
Ken Chen9c2c4802008-12-16 23:41:22 -0800536 unsigned long long rq_cpu_time;
537 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538
539 /* sys_sched_yield() stats */
Ken Chen480b9432007-10-18 21:32:56 +0200540 unsigned int yld_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
542 /* schedule() stats */
Ken Chen480b9432007-10-18 21:32:56 +0200543 unsigned int sched_switch;
544 unsigned int sched_count;
545 unsigned int sched_goidle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
547 /* try_to_wake_up() stats */
Ken Chen480b9432007-10-18 21:32:56 +0200548 unsigned int ttwu_count;
549 unsigned int ttwu_local;
Ingo Molnarb8efb562007-10-15 17:00:10 +0200550
551 /* BKL stats */
Ken Chen480b9432007-10-18 21:32:56 +0200552 unsigned int bkl_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553#endif
554};
555
Fenghua Yuf34e3b62007-07-19 01:48:13 -0700556static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
Peter Zijlstra7d478722009-09-14 19:55:44 +0200558static inline
559void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
Ingo Molnardd41f592007-07-09 18:51:59 +0200560{
Peter Zijlstra7d478722009-09-14 19:55:44 +0200561 rq->curr->sched_class->check_preempt_curr(rq, p, flags);
Mike Galbraitha64692a2010-03-11 17:16:20 +0100562
563 /*
564 * A queue event has occurred, and we're going to schedule. In
565 * this case, we can save a useless back to back clock update.
566 */
567 if (test_tsk_need_resched(p))
568 rq->skip_clock_update = 1;
Ingo Molnardd41f592007-07-09 18:51:59 +0200569}
570
Christoph Lameter0a2966b2006-09-25 23:30:51 -0700571static inline int cpu_of(struct rq *rq)
572{
573#ifdef CONFIG_SMP
574 return rq->cpu;
575#else
576 return 0;
577#endif
578}
579
Paul E. McKenney497f0ab2010-02-22 17:04:51 -0800580#define rcu_dereference_check_sched_domain(p) \
Paul E. McKenneyd11c5632010-02-22 17:04:50 -0800581 rcu_dereference_check((p), \
582 rcu_read_lock_sched_held() || \
583 lockdep_is_held(&sched_domains_mutex))
584
Ingo Molnar20d315d2007-07-09 18:51:58 +0200585/*
Nick Piggin674311d2005-06-25 14:57:27 -0700586 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -0700587 * See detach_destroy_domains: synchronize_sched for details.
Nick Piggin674311d2005-06-25 14:57:27 -0700588 *
589 * The domain tree of any CPU may only be accessed from within
590 * preempt-disabled sections.
591 */
Ingo Molnar48f24c42006-07-03 00:25:40 -0700592#define for_each_domain(cpu, __sd) \
Paul E. McKenney497f0ab2010-02-22 17:04:51 -0800593 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
595#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
596#define this_rq() (&__get_cpu_var(runqueues))
597#define task_rq(p) cpu_rq(task_cpu(p))
598#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
Hitoshi Mitake54d35f22009-06-29 14:44:57 +0900599#define raw_rq() (&__raw_get_cpu_var(runqueues))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200601#ifdef CONFIG_CGROUP_SCHED
602
603/*
604 * Return the group to which this tasks belongs.
605 *
606 * We use task_subsys_state_check() and extend the RCU verification
607 * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach()
608 * holds that lock for each task it moves into the cgroup. Therefore
609 * by holding that lock, we pin the task to the current cgroup.
610 */
611static inline struct task_group *task_group(struct task_struct *p)
612{
613 struct cgroup_subsys_state *css;
614
615 css = task_subsys_state_check(p, cpu_cgroup_subsys_id,
616 lockdep_is_held(&task_rq(p)->lock));
617 return container_of(css, struct task_group, css);
618}
619
620/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
621static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
622{
623#ifdef CONFIG_FAIR_GROUP_SCHED
624 p->se.cfs_rq = task_group(p)->cfs_rq[cpu];
625 p->se.parent = task_group(p)->se[cpu];
626#endif
627
628#ifdef CONFIG_RT_GROUP_SCHED
629 p->rt.rt_rq = task_group(p)->rt_rq[cpu];
630 p->rt.parent = task_group(p)->rt_se[cpu];
631#endif
632}
633
634#else /* CONFIG_CGROUP_SCHED */
635
636static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
637static inline struct task_group *task_group(struct task_struct *p)
638{
639 return NULL;
640}
641
642#endif /* CONFIG_CGROUP_SCHED */
643
Ingo Molnaraa9c4c02008-12-17 14:10:57 +0100644inline void update_rq_clock(struct rq *rq)
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200645{
Mike Galbraitha64692a2010-03-11 17:16:20 +0100646 if (!rq->skip_clock_update)
647 rq->clock = sched_clock_cpu(cpu_of(rq));
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200648}
649
Ingo Molnare436d802007-07-19 21:28:35 +0200650/*
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200651 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
652 */
653#ifdef CONFIG_SCHED_DEBUG
654# define const_debug __read_mostly
655#else
656# define const_debug static const
657#endif
658
Ingo Molnar017730c2008-05-12 21:20:52 +0200659/**
660 * runqueue_is_locked
Randy Dunlape17b38b2009-10-11 19:12:00 -0700661 * @cpu: the processor in question.
Ingo Molnar017730c2008-05-12 21:20:52 +0200662 *
663 * Returns true if the current cpu runqueue is locked.
664 * This interface allows printk to be called with the runqueue lock
665 * held and know whether or not it is OK to wake up the klogd.
666 */
Andrew Morton89f19f02009-09-19 11:55:44 -0700667int runqueue_is_locked(int cpu)
Ingo Molnar017730c2008-05-12 21:20:52 +0200668{
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100669 return raw_spin_is_locked(&cpu_rq(cpu)->lock);
Ingo Molnar017730c2008-05-12 21:20:52 +0200670}
671
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200672/*
673 * Debugging: various feature bits
674 */
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200675
676#define SCHED_FEAT(name, enabled) \
677 __SCHED_FEAT_##name ,
678
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200679enum {
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200680#include "sched_features.h"
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200681};
682
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200683#undef SCHED_FEAT
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200684
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200685#define SCHED_FEAT(name, enabled) \
686 (1UL << __SCHED_FEAT_##name) * enabled |
687
688const_debug unsigned int sysctl_sched_features =
689#include "sched_features.h"
690 0;
691
692#undef SCHED_FEAT
693
694#ifdef CONFIG_SCHED_DEBUG
695#define SCHED_FEAT(name, enabled) \
696 #name ,
697
Harvey Harrison983ed7a2008-04-24 18:17:55 -0700698static __read_mostly char *sched_feat_names[] = {
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200699#include "sched_features.h"
700 NULL
701};
702
703#undef SCHED_FEAT
704
Li Zefan34f3a812008-10-30 15:23:32 +0800705static int sched_feat_show(struct seq_file *m, void *v)
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200706{
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200707 int i;
708
709 for (i = 0; sched_feat_names[i]; i++) {
Li Zefan34f3a812008-10-30 15:23:32 +0800710 if (!(sysctl_sched_features & (1UL << i)))
711 seq_puts(m, "NO_");
712 seq_printf(m, "%s ", sched_feat_names[i]);
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200713 }
Li Zefan34f3a812008-10-30 15:23:32 +0800714 seq_puts(m, "\n");
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200715
Li Zefan34f3a812008-10-30 15:23:32 +0800716 return 0;
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200717}
718
719static ssize_t
720sched_feat_write(struct file *filp, const char __user *ubuf,
721 size_t cnt, loff_t *ppos)
722{
723 char buf[64];
724 char *cmp = buf;
725 int neg = 0;
726 int i;
727
728 if (cnt > 63)
729 cnt = 63;
730
731 if (copy_from_user(&buf, ubuf, cnt))
732 return -EFAULT;
733
734 buf[cnt] = 0;
735
Ingo Molnarc24b7c52008-04-18 10:55:34 +0200736 if (strncmp(buf, "NO_", 3) == 0) {
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200737 neg = 1;
738 cmp += 3;
739 }
740
741 for (i = 0; sched_feat_names[i]; i++) {
742 int len = strlen(sched_feat_names[i]);
743
744 if (strncmp(cmp, sched_feat_names[i], len) == 0) {
745 if (neg)
746 sysctl_sched_features &= ~(1UL << i);
747 else
748 sysctl_sched_features |= (1UL << i);
749 break;
750 }
751 }
752
753 if (!sched_feat_names[i])
754 return -EINVAL;
755
Jan Blunck42994722009-11-20 17:40:37 +0100756 *ppos += cnt;
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200757
758 return cnt;
759}
760
Li Zefan34f3a812008-10-30 15:23:32 +0800761static int sched_feat_open(struct inode *inode, struct file *filp)
762{
763 return single_open(filp, sched_feat_show, NULL);
764}
765
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700766static const struct file_operations sched_feat_fops = {
Li Zefan34f3a812008-10-30 15:23:32 +0800767 .open = sched_feat_open,
768 .write = sched_feat_write,
769 .read = seq_read,
770 .llseek = seq_lseek,
771 .release = single_release,
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200772};
773
774static __init int sched_init_debug(void)
775{
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200776 debugfs_create_file("sched_features", 0644, NULL, NULL,
777 &sched_feat_fops);
778
779 return 0;
780}
781late_initcall(sched_init_debug);
782
783#endif
784
785#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200786
787/*
Peter Zijlstrab82d9fd2007-11-09 22:39:39 +0100788 * Number of tasks to iterate in a single balance run.
789 * Limited because this is done with IRQs disabled.
790 */
791const_debug unsigned int sysctl_sched_nr_migrate = 32;
792
793/*
Peter Zijlstra2398f2c2008-06-27 13:41:35 +0200794 * ratelimit for updating the group shares.
Peter Zijlstra55cd5342008-08-04 08:54:26 +0200795 * default: 0.25ms
Peter Zijlstra2398f2c2008-06-27 13:41:35 +0200796 */
Peter Zijlstra55cd5342008-08-04 08:54:26 +0200797unsigned int sysctl_sched_shares_ratelimit = 250000;
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +0100798unsigned int normalized_sysctl_sched_shares_ratelimit = 250000;
Peter Zijlstra2398f2c2008-06-27 13:41:35 +0200799
800/*
Peter Zijlstraffda12a2008-10-17 19:27:02 +0200801 * Inject some fuzzyness into changing the per-cpu group shares
802 * this avoids remote rq-locks at the expense of fairness.
803 * default: 4
804 */
805unsigned int sysctl_sched_shares_thresh = 4;
806
807/*
Peter Zijlstrae9e92502009-09-01 10:34:37 +0200808 * period over which we average the RT time consumption, measured
809 * in ms.
810 *
811 * default: 1s
812 */
813const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC;
814
815/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100816 * period over which we measure -rt task cpu usage in us.
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100817 * default: 1s
818 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100819unsigned int sysctl_sched_rt_period = 1000000;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100820
Ingo Molnar6892b752008-02-13 14:02:36 +0100821static __read_mostly int scheduler_running;
822
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100823/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100824 * part of the period that we allow rt tasks to run in us.
825 * default: 0.95s
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100826 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100827int sysctl_sched_rt_runtime = 950000;
828
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200829static inline u64 global_rt_period(void)
830{
831 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
832}
833
834static inline u64 global_rt_runtime(void)
835{
roel kluine26873b2008-07-22 16:51:15 -0400836 if (sysctl_sched_rt_runtime < 0)
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200837 return RUNTIME_INF;
838
839 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
840}
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100841
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842#ifndef prepare_arch_switch
Nick Piggin4866cde2005-06-25 14:57:23 -0700843# define prepare_arch_switch(next) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844#endif
Nick Piggin4866cde2005-06-25 14:57:23 -0700845#ifndef finish_arch_switch
846# define finish_arch_switch(prev) do { } while (0)
847#endif
848
Dmitry Adamushko051a1d12007-12-18 15:21:13 +0100849static inline int task_current(struct rq *rq, struct task_struct *p)
850{
851 return rq->curr == p;
852}
853
Nick Piggin4866cde2005-06-25 14:57:23 -0700854#ifndef __ARCH_WANT_UNLOCKED_CTXSW
Ingo Molnar70b97a72006-07-03 00:25:42 -0700855static inline int task_running(struct rq *rq, struct task_struct *p)
Nick Piggin4866cde2005-06-25 14:57:23 -0700856{
Dmitry Adamushko051a1d12007-12-18 15:21:13 +0100857 return task_current(rq, p);
Nick Piggin4866cde2005-06-25 14:57:23 -0700858}
859
Ingo Molnar70b97a72006-07-03 00:25:42 -0700860static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
Nick Piggin4866cde2005-06-25 14:57:23 -0700861{
862}
863
Ingo Molnar70b97a72006-07-03 00:25:42 -0700864static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
Nick Piggin4866cde2005-06-25 14:57:23 -0700865{
Ingo Molnarda04c032005-09-13 11:17:59 +0200866#ifdef CONFIG_DEBUG_SPINLOCK
867 /* this is a valid case when another task releases the spinlock */
868 rq->lock.owner = current;
869#endif
Ingo Molnar8a25d5d2006-07-03 00:24:54 -0700870 /*
871 * If we are tracking spinlock dependencies then we have to
872 * fix up the runqueue lock - which gets 'carried over' from
873 * prev into current:
874 */
875 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
876
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100877 raw_spin_unlock_irq(&rq->lock);
Nick Piggin4866cde2005-06-25 14:57:23 -0700878}
879
880#else /* __ARCH_WANT_UNLOCKED_CTXSW */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700881static inline int task_running(struct rq *rq, struct task_struct *p)
Nick Piggin4866cde2005-06-25 14:57:23 -0700882{
883#ifdef CONFIG_SMP
884 return p->oncpu;
885#else
Dmitry Adamushko051a1d12007-12-18 15:21:13 +0100886 return task_current(rq, p);
Nick Piggin4866cde2005-06-25 14:57:23 -0700887#endif
888}
889
Ingo Molnar70b97a72006-07-03 00:25:42 -0700890static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
Nick Piggin4866cde2005-06-25 14:57:23 -0700891{
892#ifdef CONFIG_SMP
893 /*
894 * We can optimise this out completely for !SMP, because the
895 * SMP rebalancing from interrupt is the only thing that cares
896 * here.
897 */
898 next->oncpu = 1;
899#endif
900#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100901 raw_spin_unlock_irq(&rq->lock);
Nick Piggin4866cde2005-06-25 14:57:23 -0700902#else
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100903 raw_spin_unlock(&rq->lock);
Nick Piggin4866cde2005-06-25 14:57:23 -0700904#endif
905}
906
Ingo Molnar70b97a72006-07-03 00:25:42 -0700907static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
Nick Piggin4866cde2005-06-25 14:57:23 -0700908{
909#ifdef CONFIG_SMP
910 /*
911 * After ->oncpu is cleared, the task can be moved to a different CPU.
912 * We must ensure this doesn't happen until the switch is completely
913 * finished.
914 */
915 smp_wmb();
916 prev->oncpu = 0;
917#endif
918#ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
919 local_irq_enable();
920#endif
921}
922#endif /* __ARCH_WANT_UNLOCKED_CTXSW */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
924/*
Peter Zijlstra65cc8e42010-03-25 21:05:16 +0100925 * Check whether the task is waking, we use this to synchronize ->cpus_allowed
926 * against ttwu().
Peter Zijlstra0970d292010-02-15 14:45:54 +0100927 */
928static inline int task_is_waking(struct task_struct *p)
929{
Peter Zijlstra0017d732010-03-24 18:34:10 +0100930 return unlikely(p->state == TASK_WAKING);
Peter Zijlstra0970d292010-02-15 14:45:54 +0100931}
932
933/*
Ingo Molnarb29739f2006-06-27 02:54:51 -0700934 * __task_rq_lock - lock the runqueue a given task resides on.
935 * Must be called interrupts disabled.
936 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700937static inline struct rq *__task_rq_lock(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -0700938 __acquires(rq->lock)
939{
Peter Zijlstra0970d292010-02-15 14:45:54 +0100940 struct rq *rq;
941
Andi Kleen3a5c3592007-10-15 17:00:14 +0200942 for (;;) {
Peter Zijlstra0970d292010-02-15 14:45:54 +0100943 rq = task_rq(p);
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100944 raw_spin_lock(&rq->lock);
Peter Zijlstra65cc8e42010-03-25 21:05:16 +0100945 if (likely(rq == task_rq(p)))
Andi Kleen3a5c3592007-10-15 17:00:14 +0200946 return rq;
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100947 raw_spin_unlock(&rq->lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -0700948 }
Ingo Molnarb29739f2006-06-27 02:54:51 -0700949}
950
951/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 * task_rq_lock - lock the runqueue a given task resides on and disable
Ingo Molnar41a2d6c2007-12-05 15:46:09 +0100953 * interrupts. Note the ordering: we can safely lookup the task_rq without
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 * explicitly disabling preemption.
955 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700956static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 __acquires(rq->lock)
958{
Ingo Molnar70b97a72006-07-03 00:25:42 -0700959 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Andi Kleen3a5c3592007-10-15 17:00:14 +0200961 for (;;) {
962 local_irq_save(*flags);
963 rq = task_rq(p);
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100964 raw_spin_lock(&rq->lock);
Peter Zijlstra65cc8e42010-03-25 21:05:16 +0100965 if (likely(rq == task_rq(p)))
Andi Kleen3a5c3592007-10-15 17:00:14 +0200966 return rq;
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100967 raw_spin_unlock_irqrestore(&rq->lock, *flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969}
970
Alexey Dobriyana9957442007-10-15 17:00:13 +0200971static void __task_rq_unlock(struct rq *rq)
Ingo Molnarb29739f2006-06-27 02:54:51 -0700972 __releases(rq->lock)
973{
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100974 raw_spin_unlock(&rq->lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -0700975}
976
Ingo Molnar70b97a72006-07-03 00:25:42 -0700977static inline void task_rq_unlock(struct rq *rq, unsigned long *flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 __releases(rq->lock)
979{
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100980 raw_spin_unlock_irqrestore(&rq->lock, *flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981}
982
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983/*
Robert P. J. Daycc2a73b2006-12-10 02:20:00 -0800984 * this_rq_lock - lock this runqueue and disable interrupts.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 */
Alexey Dobriyana9957442007-10-15 17:00:13 +0200986static struct rq *this_rq_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 __acquires(rq->lock)
988{
Ingo Molnar70b97a72006-07-03 00:25:42 -0700989 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
991 local_irq_disable();
992 rq = this_rq();
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100993 raw_spin_lock(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
995 return rq;
996}
997
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100998#ifdef CONFIG_SCHED_HRTICK
999/*
1000 * Use HR-timers to deliver accurate preemption points.
1001 *
1002 * Its all a bit involved since we cannot program an hrt while holding the
1003 * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a
1004 * reschedule event.
1005 *
1006 * When we get rescheduled we reprogram the hrtick_timer outside of the
1007 * rq->lock.
1008 */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001009
1010/*
1011 * Use hrtick when:
1012 * - enabled by features
1013 * - hrtimer is actually high res
1014 */
1015static inline int hrtick_enabled(struct rq *rq)
1016{
1017 if (!sched_feat(HRTICK))
1018 return 0;
Ingo Molnarba420592008-07-20 11:02:06 +02001019 if (!cpu_active(cpu_of(rq)))
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001020 return 0;
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001021 return hrtimer_is_hres_active(&rq->hrtick_timer);
1022}
1023
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001024static void hrtick_clear(struct rq *rq)
1025{
1026 if (hrtimer_active(&rq->hrtick_timer))
1027 hrtimer_cancel(&rq->hrtick_timer);
1028}
1029
1030/*
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001031 * High-resolution timer tick.
1032 * Runs from hardirq context with interrupts disabled.
1033 */
1034static enum hrtimer_restart hrtick(struct hrtimer *timer)
1035{
1036 struct rq *rq = container_of(timer, struct rq, hrtick_timer);
1037
1038 WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
1039
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001040 raw_spin_lock(&rq->lock);
Peter Zijlstra3e51f332008-05-03 18:29:28 +02001041 update_rq_clock(rq);
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001042 rq->curr->sched_class->task_tick(rq, rq->curr, 1);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001043 raw_spin_unlock(&rq->lock);
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001044
1045 return HRTIMER_NORESTART;
1046}
1047
Rabin Vincent95e904c2008-05-11 05:55:33 +05301048#ifdef CONFIG_SMP
Peter Zijlstra31656512008-07-18 18:01:23 +02001049/*
1050 * called from hardirq (IPI) context
1051 */
1052static void __hrtick_start(void *arg)
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001053{
Peter Zijlstra31656512008-07-18 18:01:23 +02001054 struct rq *rq = arg;
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001055
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001056 raw_spin_lock(&rq->lock);
Peter Zijlstra31656512008-07-18 18:01:23 +02001057 hrtimer_restart(&rq->hrtick_timer);
1058 rq->hrtick_csd_pending = 0;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001059 raw_spin_unlock(&rq->lock);
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001060}
1061
Peter Zijlstra31656512008-07-18 18:01:23 +02001062/*
1063 * Called to set the hrtick timer state.
1064 *
1065 * called with rq->lock held and irqs disabled
1066 */
1067static void hrtick_start(struct rq *rq, u64 delay)
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001068{
Peter Zijlstra31656512008-07-18 18:01:23 +02001069 struct hrtimer *timer = &rq->hrtick_timer;
1070 ktime_t time = ktime_add_ns(timer->base->get_time(), delay);
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001071
Arjan van de Vencc584b22008-09-01 15:02:30 -07001072 hrtimer_set_expires(timer, time);
Peter Zijlstra31656512008-07-18 18:01:23 +02001073
1074 if (rq == this_rq()) {
1075 hrtimer_restart(timer);
1076 } else if (!rq->hrtick_csd_pending) {
Peter Zijlstra6e275632009-02-25 13:59:48 +01001077 __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0);
Peter Zijlstra31656512008-07-18 18:01:23 +02001078 rq->hrtick_csd_pending = 1;
1079 }
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001080}
1081
1082static int
1083hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu)
1084{
1085 int cpu = (int)(long)hcpu;
1086
1087 switch (action) {
1088 case CPU_UP_CANCELED:
1089 case CPU_UP_CANCELED_FROZEN:
1090 case CPU_DOWN_PREPARE:
1091 case CPU_DOWN_PREPARE_FROZEN:
1092 case CPU_DEAD:
1093 case CPU_DEAD_FROZEN:
Peter Zijlstra31656512008-07-18 18:01:23 +02001094 hrtick_clear(cpu_rq(cpu));
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001095 return NOTIFY_OK;
1096 }
1097
1098 return NOTIFY_DONE;
1099}
1100
Rakib Mullickfa748202008-09-22 14:55:45 -07001101static __init void init_hrtick(void)
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001102{
1103 hotcpu_notifier(hotplug_hrtick, 0);
1104}
Peter Zijlstra31656512008-07-18 18:01:23 +02001105#else
1106/*
1107 * Called to set the hrtick timer state.
1108 *
1109 * called with rq->lock held and irqs disabled
1110 */
1111static void hrtick_start(struct rq *rq, u64 delay)
1112{
Peter Zijlstra7f1e2ca2009-03-13 12:21:27 +01001113 __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0,
Arun R Bharadwaj5c333862009-04-16 12:14:37 +05301114 HRTIMER_MODE_REL_PINNED, 0);
Peter Zijlstra31656512008-07-18 18:01:23 +02001115}
1116
Andrew Morton006c75f2008-09-22 14:55:46 -07001117static inline void init_hrtick(void)
Peter Zijlstra31656512008-07-18 18:01:23 +02001118{
1119}
Rabin Vincent95e904c2008-05-11 05:55:33 +05301120#endif /* CONFIG_SMP */
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001121
1122static void init_rq_hrtick(struct rq *rq)
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001123{
Peter Zijlstra31656512008-07-18 18:01:23 +02001124#ifdef CONFIG_SMP
1125 rq->hrtick_csd_pending = 0;
1126
1127 rq->hrtick_csd.flags = 0;
1128 rq->hrtick_csd.func = __hrtick_start;
1129 rq->hrtick_csd.info = rq;
1130#endif
1131
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001132 hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1133 rq->hrtick_timer.function = hrtick;
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001134}
Andrew Morton006c75f2008-09-22 14:55:46 -07001135#else /* CONFIG_SCHED_HRTICK */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001136static inline void hrtick_clear(struct rq *rq)
1137{
1138}
1139
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001140static inline void init_rq_hrtick(struct rq *rq)
1141{
1142}
1143
Peter Zijlstrab328ca12008-04-29 10:02:46 +02001144static inline void init_hrtick(void)
1145{
1146}
Andrew Morton006c75f2008-09-22 14:55:46 -07001147#endif /* CONFIG_SCHED_HRTICK */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001148
Ingo Molnar1b9f19c2007-07-09 18:51:59 +02001149/*
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001150 * resched_task - mark a task 'to be rescheduled now'.
1151 *
1152 * On UP this means the setting of the need_resched flag, on SMP it
1153 * might also involve a cross-CPU call to trigger the scheduler on
1154 * the target CPU.
1155 */
1156#ifdef CONFIG_SMP
1157
1158#ifndef tsk_is_polling
1159#define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG)
1160#endif
1161
Peter Zijlstra31656512008-07-18 18:01:23 +02001162static void resched_task(struct task_struct *p)
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001163{
1164 int cpu;
1165
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001166 assert_raw_spin_locked(&task_rq(p)->lock);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001167
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08001168 if (test_tsk_need_resched(p))
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001169 return;
1170
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08001171 set_tsk_need_resched(p);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001172
1173 cpu = task_cpu(p);
1174 if (cpu == smp_processor_id())
1175 return;
1176
1177 /* NEED_RESCHED must be visible before we test polling */
1178 smp_mb();
1179 if (!tsk_is_polling(p))
1180 smp_send_reschedule(cpu);
1181}
1182
1183static void resched_cpu(int cpu)
1184{
1185 struct rq *rq = cpu_rq(cpu);
1186 unsigned long flags;
1187
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001188 if (!raw_spin_trylock_irqsave(&rq->lock, flags))
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001189 return;
1190 resched_task(cpu_curr(cpu));
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001191 raw_spin_unlock_irqrestore(&rq->lock, flags);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001192}
Thomas Gleixner06d83082008-03-22 09:20:24 +01001193
1194#ifdef CONFIG_NO_HZ
1195/*
1196 * When add_timer_on() enqueues a timer into the timer wheel of an
1197 * idle CPU then this timer might expire before the next timer event
1198 * which is scheduled to wake up that CPU. In case of a completely
1199 * idle system the next event might even be infinite time into the
1200 * future. wake_up_idle_cpu() ensures that the CPU is woken up and
1201 * leaves the inner idle loop so the newly added timer is taken into
1202 * account when the CPU goes back to idle and evaluates the timer
1203 * wheel for the next timer event.
1204 */
1205void wake_up_idle_cpu(int cpu)
1206{
1207 struct rq *rq = cpu_rq(cpu);
1208
1209 if (cpu == smp_processor_id())
1210 return;
1211
1212 /*
1213 * This is safe, as this function is called with the timer
1214 * wheel base lock of (cpu) held. When the CPU is on the way
1215 * to idle and has not yet set rq->curr to idle then it will
1216 * be serialized on the timer wheel base lock and take the new
1217 * timer into account automatically.
1218 */
1219 if (rq->curr != rq->idle)
1220 return;
1221
1222 /*
1223 * We can set TIF_RESCHED on the idle task of the other CPU
1224 * lockless. The worst case is that the other CPU runs the
1225 * idle task through an additional NOOP schedule()
1226 */
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08001227 set_tsk_need_resched(rq->idle);
Thomas Gleixner06d83082008-03-22 09:20:24 +01001228
1229 /* NEED_RESCHED must be visible before we test polling */
1230 smp_mb();
1231 if (!tsk_is_polling(rq->idle))
1232 smp_send_reschedule(cpu);
1233}
Mike Galbraith39c0cbe2010-03-11 17:17:13 +01001234
1235int nohz_ratelimit(int cpu)
1236{
1237 struct rq *rq = cpu_rq(cpu);
1238 u64 diff = rq->clock - rq->nohz_stamp;
1239
1240 rq->nohz_stamp = rq->clock;
1241
1242 return diff < (NSEC_PER_SEC / HZ) >> 1;
1243}
1244
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02001245#endif /* CONFIG_NO_HZ */
Thomas Gleixner06d83082008-03-22 09:20:24 +01001246
Peter Zijlstrae9e92502009-09-01 10:34:37 +02001247static u64 sched_avg_period(void)
1248{
1249 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1250}
1251
1252static void sched_avg_update(struct rq *rq)
1253{
1254 s64 period = sched_avg_period();
1255
1256 while ((s64)(rq->clock - rq->age_stamp) > period) {
Will Deacon0d98bb22010-05-24 12:11:43 -07001257 /*
1258 * Inline assembly required to prevent the compiler
1259 * optimising this loop into a divmod call.
1260 * See __iter_div_u64_rem() for another example of this.
1261 */
1262 asm("" : "+rm" (rq->age_stamp));
Peter Zijlstrae9e92502009-09-01 10:34:37 +02001263 rq->age_stamp += period;
1264 rq->rt_avg /= 2;
1265 }
1266}
1267
1268static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1269{
1270 rq->rt_avg += rt_delta;
1271 sched_avg_update(rq);
1272}
1273
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02001274#else /* !CONFIG_SMP */
Peter Zijlstra31656512008-07-18 18:01:23 +02001275static void resched_task(struct task_struct *p)
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001276{
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001277 assert_raw_spin_locked(&task_rq(p)->lock);
Peter Zijlstra31656512008-07-18 18:01:23 +02001278 set_tsk_need_resched(p);
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001279}
Peter Zijlstrae9e92502009-09-01 10:34:37 +02001280
1281static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1282{
1283}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02001284#endif /* CONFIG_SMP */
Ingo Molnarc24d20d2007-07-09 18:51:59 +02001285
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001286#if BITS_PER_LONG == 32
1287# define WMULT_CONST (~0UL)
1288#else
1289# define WMULT_CONST (1UL << 32)
1290#endif
1291
1292#define WMULT_SHIFT 32
1293
Ingo Molnar194081e2007-08-09 11:16:51 +02001294/*
1295 * Shift right and round:
1296 */
Ingo Molnarcf2ab462007-09-05 14:32:49 +02001297#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
Ingo Molnar194081e2007-08-09 11:16:51 +02001298
Peter Zijlstraa7be37a2008-06-27 13:41:11 +02001299/*
1300 * delta *= weight / lw
1301 */
Ingo Molnarcb1c4fc2007-08-02 17:41:40 +02001302static unsigned long
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001303calc_delta_mine(unsigned long delta_exec, unsigned long weight,
1304 struct load_weight *lw)
1305{
1306 u64 tmp;
1307
Lai Jiangshan7a232e02008-06-12 16:43:07 +08001308 if (!lw->inv_weight) {
1309 if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST))
1310 lw->inv_weight = 1;
1311 else
1312 lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2)
1313 / (lw->weight+1);
1314 }
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001315
1316 tmp = (u64)delta_exec * weight;
1317 /*
1318 * Check whether we'd overflow the 64-bit multiplication:
1319 */
Ingo Molnar194081e2007-08-09 11:16:51 +02001320 if (unlikely(tmp > WMULT_CONST))
Ingo Molnarcf2ab462007-09-05 14:32:49 +02001321 tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
Ingo Molnar194081e2007-08-09 11:16:51 +02001322 WMULT_SHIFT/2);
1323 else
Ingo Molnarcf2ab462007-09-05 14:32:49 +02001324 tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001325
Ingo Molnarecf691d2007-08-02 17:41:40 +02001326 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001327}
1328
Ingo Molnar10919852007-10-15 17:00:04 +02001329static inline void update_load_add(struct load_weight *lw, unsigned long inc)
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001330{
1331 lw->weight += inc;
Ingo Molnare89996a2008-03-14 23:48:28 +01001332 lw->inv_weight = 0;
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001333}
1334
Ingo Molnar10919852007-10-15 17:00:04 +02001335static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001336{
1337 lw->weight -= dec;
Ingo Molnare89996a2008-03-14 23:48:28 +01001338 lw->inv_weight = 0;
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001339}
1340
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341/*
Peter Williams2dd73a42006-06-27 02:54:34 -07001342 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1343 * of tasks with abnormal "nice" values across CPUs the contribution that
1344 * each task makes to its run queue's load is weighted according to its
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01001345 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
Peter Williams2dd73a42006-06-27 02:54:34 -07001346 * scaled version of the new time slice allocation that they receive on time
1347 * slice expiry etc.
1348 */
1349
Peter Zijlstracce7ade2009-01-15 14:53:37 +01001350#define WEIGHT_IDLEPRIO 3
1351#define WMULT_IDLEPRIO 1431655765
Ingo Molnardd41f592007-07-09 18:51:59 +02001352
1353/*
1354 * Nice levels are multiplicative, with a gentle 10% change for every
1355 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1356 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1357 * that remained on nice 0.
1358 *
1359 * The "10% effect" is relative and cumulative: from _any_ nice level,
1360 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
Ingo Molnarf9153ee2007-07-16 09:46:30 +02001361 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1362 * If a task goes up by ~10% and another task goes down by ~10% then
1363 * the relative distance between them is ~25%.)
Ingo Molnardd41f592007-07-09 18:51:59 +02001364 */
1365static const int prio_to_weight[40] = {
Ingo Molnar254753d2007-08-09 11:16:51 +02001366 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1367 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1368 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1369 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1370 /* 0 */ 1024, 820, 655, 526, 423,
1371 /* 5 */ 335, 272, 215, 172, 137,
1372 /* 10 */ 110, 87, 70, 56, 45,
1373 /* 15 */ 36, 29, 23, 18, 15,
Ingo Molnardd41f592007-07-09 18:51:59 +02001374};
1375
Ingo Molnar5714d2d2007-07-16 09:46:31 +02001376/*
1377 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1378 *
1379 * In cases where the weight does not change often, we can use the
1380 * precalculated inverse to speed up arithmetics by turning divisions
1381 * into multiplications:
1382 */
Ingo Molnardd41f592007-07-09 18:51:59 +02001383static const u32 prio_to_wmult[40] = {
Ingo Molnar254753d2007-08-09 11:16:51 +02001384 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1385 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1386 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1387 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1388 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1389 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1390 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1391 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
Ingo Molnardd41f592007-07-09 18:51:59 +02001392};
Peter Williams2dd73a42006-06-27 02:54:34 -07001393
Bharata B Raoef12fef2009-03-31 10:02:22 +05301394/* Time spent by the tasks of the cpu accounting group executing in ... */
1395enum cpuacct_stat_index {
1396 CPUACCT_STAT_USER, /* ... user mode */
1397 CPUACCT_STAT_SYSTEM, /* ... kernel mode */
1398
1399 CPUACCT_STAT_NSTATS,
1400};
1401
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01001402#ifdef CONFIG_CGROUP_CPUACCT
1403static void cpuacct_charge(struct task_struct *tsk, u64 cputime);
Bharata B Raoef12fef2009-03-31 10:02:22 +05301404static void cpuacct_update_stats(struct task_struct *tsk,
1405 enum cpuacct_stat_index idx, cputime_t val);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01001406#else
1407static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
Bharata B Raoef12fef2009-03-31 10:02:22 +05301408static inline void cpuacct_update_stats(struct task_struct *tsk,
1409 enum cpuacct_stat_index idx, cputime_t val) {}
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01001410#endif
1411
Peter Zijlstra18d95a22008-04-19 19:45:00 +02001412static inline void inc_cpu_load(struct rq *rq, unsigned long load)
1413{
1414 update_load_add(&rq->load, load);
1415}
1416
1417static inline void dec_cpu_load(struct rq *rq, unsigned long load)
1418{
1419 update_load_sub(&rq->load, load);
1420}
1421
Ingo Molnar7940ca32008-08-19 13:40:47 +02001422#if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED)
Peter Zijlstraeb755802008-08-19 12:33:05 +02001423typedef int (*tg_visitor)(struct task_group *, void *);
1424
1425/*
1426 * Iterate the full tree, calling @down when first entering a node and @up when
1427 * leaving it for the final time.
1428 */
1429static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
1430{
1431 struct task_group *parent, *child;
1432 int ret;
1433
1434 rcu_read_lock();
1435 parent = &root_task_group;
1436down:
1437 ret = (*down)(parent, data);
1438 if (ret)
1439 goto out_unlock;
1440 list_for_each_entry_rcu(child, &parent->children, siblings) {
1441 parent = child;
1442 goto down;
1443
1444up:
1445 continue;
1446 }
1447 ret = (*up)(parent, data);
1448 if (ret)
1449 goto out_unlock;
1450
1451 child = parent;
1452 parent = parent->parent;
1453 if (parent)
1454 goto up;
1455out_unlock:
1456 rcu_read_unlock();
1457
1458 return ret;
1459}
1460
1461static int tg_nop(struct task_group *tg, void *data)
1462{
1463 return 0;
1464}
1465#endif
1466
Gregory Haskinse7693a32008-01-25 21:08:09 +01001467#ifdef CONFIG_SMP
Peter Zijlstraf5f08f32009-09-10 13:35:28 +02001468/* Used instead of source_load when we know the type == 0 */
1469static unsigned long weighted_cpuload(const int cpu)
1470{
1471 return cpu_rq(cpu)->load.weight;
1472}
1473
1474/*
1475 * Return a low guess at the load of a migration-source cpu weighted
1476 * according to the scheduling class and "nice" value.
1477 *
1478 * We want to under-estimate the load of migration sources, to
1479 * balance conservatively.
1480 */
1481static unsigned long source_load(int cpu, int type)
1482{
1483 struct rq *rq = cpu_rq(cpu);
1484 unsigned long total = weighted_cpuload(cpu);
1485
1486 if (type == 0 || !sched_feat(LB_BIAS))
1487 return total;
1488
1489 return min(rq->cpu_load[type-1], total);
1490}
1491
1492/*
1493 * Return a high guess at the load of a migration-target cpu weighted
1494 * according to the scheduling class and "nice" value.
1495 */
1496static unsigned long target_load(int cpu, int type)
1497{
1498 struct rq *rq = cpu_rq(cpu);
1499 unsigned long total = weighted_cpuload(cpu);
1500
1501 if (type == 0 || !sched_feat(LB_BIAS))
1502 return total;
1503
1504 return max(rq->cpu_load[type-1], total);
1505}
1506
Peter Zijlstraae154be2009-09-10 14:40:57 +02001507static unsigned long power_of(int cpu)
1508{
Peter Zijlstrae51fd5e2010-05-31 12:37:30 +02001509 return cpu_rq(cpu)->cpu_power;
Peter Zijlstraae154be2009-09-10 14:40:57 +02001510}
1511
Gregory Haskinse7693a32008-01-25 21:08:09 +01001512static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001513
Peter Zijlstraa8a51d52008-06-27 13:41:26 +02001514static unsigned long cpu_avg_load_per_task(int cpu)
1515{
1516 struct rq *rq = cpu_rq(cpu);
Ingo Molnaraf6d5962008-11-29 20:45:15 +01001517 unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
Peter Zijlstraa8a51d52008-06-27 13:41:26 +02001518
Steven Rostedt4cd42622008-11-26 21:04:24 -05001519 if (nr_running)
1520 rq->avg_load_per_task = rq->load.weight / nr_running;
Balbir Singha2d47772008-11-12 16:19:00 +05301521 else
1522 rq->avg_load_per_task = 0;
Peter Zijlstraa8a51d52008-06-27 13:41:26 +02001523
1524 return rq->avg_load_per_task;
1525}
1526
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001527#ifdef CONFIG_FAIR_GROUP_SCHED
1528
Tejun Heo43cf38e2010-02-02 14:38:57 +09001529static __read_mostly unsigned long __percpu *update_shares_data;
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001530
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001531static void __set_se_shares(struct sched_entity *se, unsigned long shares);
1532
1533/*
1534 * Calculate and set the cpu's group shares.
1535 */
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001536static void update_group_shares_cpu(struct task_group *tg, int cpu,
1537 unsigned long sd_shares,
1538 unsigned long sd_rq_weight,
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09001539 unsigned long *usd_rq_weight)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001540{
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001541 unsigned long shares, rq_weight;
Peter Zijlstraa5004272009-07-27 14:04:49 +02001542 int boost = 0;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001543
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09001544 rq_weight = usd_rq_weight[cpu];
Peter Zijlstraa5004272009-07-27 14:04:49 +02001545 if (!rq_weight) {
1546 boost = 1;
1547 rq_weight = NICE_0_LOAD;
1548 }
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001549
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001550 /*
Peter Zijlstraa8af7242009-08-21 13:58:54 +02001551 * \Sum_j shares_j * rq_weight_i
1552 * shares_i = -----------------------------
1553 * \Sum_j rq_weight_j
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001554 */
Ken Chenec4e0e22008-11-18 22:41:57 -08001555 shares = (sd_shares * rq_weight) / sd_rq_weight;
Peter Zijlstraffda12a2008-10-17 19:27:02 +02001556 shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001557
Peter Zijlstraffda12a2008-10-17 19:27:02 +02001558 if (abs(shares - tg->se[cpu]->load.weight) >
1559 sysctl_sched_shares_thresh) {
1560 struct rq *rq = cpu_rq(cpu);
1561 unsigned long flags;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001562
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001563 raw_spin_lock_irqsave(&rq->lock, flags);
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001564 tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight;
Peter Zijlstraa5004272009-07-27 14:04:49 +02001565 tg->cfs_rq[cpu]->shares = boost ? 0 : shares;
Peter Zijlstraffda12a2008-10-17 19:27:02 +02001566 __set_se_shares(tg->se[cpu], shares);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001567 raw_spin_unlock_irqrestore(&rq->lock, flags);
Peter Zijlstraffda12a2008-10-17 19:27:02 +02001568 }
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001569}
1570
1571/*
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001572 * Re-compute the task group their per cpu shares over the given domain.
1573 * This needs to be done in a bottom-up fashion because the rq weight of a
1574 * parent group depends on the shares of its child groups.
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001575 */
Peter Zijlstraeb755802008-08-19 12:33:05 +02001576static int tg_shares_up(struct task_group *tg, void *data)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001577{
Peter Zijlstracd8ad402009-12-03 18:00:07 +01001578 unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0;
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09001579 unsigned long *usd_rq_weight;
Peter Zijlstraeb755802008-08-19 12:33:05 +02001580 struct sched_domain *sd = data;
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001581 unsigned long flags;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001582 int i;
1583
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001584 if (!tg->se[0])
1585 return 0;
1586
1587 local_irq_save(flags);
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09001588 usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id());
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001589
Rusty Russell758b2cd2008-11-25 02:35:04 +10301590 for_each_cpu(i, sched_domain_span(sd)) {
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001591 weight = tg->cfs_rq[i]->load.weight;
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09001592 usd_rq_weight[i] = weight;
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001593
Peter Zijlstracd8ad402009-12-03 18:00:07 +01001594 rq_weight += weight;
Ken Chenec4e0e22008-11-18 22:41:57 -08001595 /*
1596 * If there are currently no tasks on the cpu pretend there
1597 * is one of average load so that when a new task gets to
1598 * run here it will not get delayed by group starvation.
1599 */
Ken Chenec4e0e22008-11-18 22:41:57 -08001600 if (!weight)
1601 weight = NICE_0_LOAD;
1602
Peter Zijlstracd8ad402009-12-03 18:00:07 +01001603 sum_weight += weight;
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001604 shares += tg->cfs_rq[i]->shares;
1605 }
1606
Peter Zijlstracd8ad402009-12-03 18:00:07 +01001607 if (!rq_weight)
1608 rq_weight = sum_weight;
1609
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001610 if ((!shares && rq_weight) || shares > tg->shares)
1611 shares = tg->shares;
1612
1613 if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE))
1614 shares = tg->shares;
1615
Rusty Russell758b2cd2008-11-25 02:35:04 +10301616 for_each_cpu(i, sched_domain_span(sd))
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09001617 update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight);
Peter Zijlstra34d76c42009-08-27 13:08:56 +02001618
1619 local_irq_restore(flags);
Peter Zijlstraeb755802008-08-19 12:33:05 +02001620
1621 return 0;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001622}
1623
1624/*
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001625 * Compute the cpu's hierarchical load factor for each task group.
1626 * This needs to be done in a top-down fashion because the load of a child
1627 * group is a fraction of its parents load.
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001628 */
Peter Zijlstraeb755802008-08-19 12:33:05 +02001629static int tg_load_down(struct task_group *tg, void *data)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001630{
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001631 unsigned long load;
Peter Zijlstraeb755802008-08-19 12:33:05 +02001632 long cpu = (long)data;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001633
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001634 if (!tg->parent) {
1635 load = cpu_rq(cpu)->load.weight;
1636 } else {
1637 load = tg->parent->cfs_rq[cpu]->h_load;
1638 load *= tg->cfs_rq[cpu]->shares;
1639 load /= tg->parent->cfs_rq[cpu]->load.weight + 1;
1640 }
1641
1642 tg->cfs_rq[cpu]->h_load = load;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001643
Peter Zijlstraeb755802008-08-19 12:33:05 +02001644 return 0;
Peter Zijlstra4d8d5952008-06-27 13:41:19 +02001645}
1646
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001647static void update_shares(struct sched_domain *sd)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001648{
Peter Zijlstrae7097152009-06-03 15:41:20 +02001649 s64 elapsed;
1650 u64 now;
1651
1652 if (root_task_group_empty())
1653 return;
1654
1655 now = cpu_clock(raw_smp_processor_id());
1656 elapsed = now - sd->last_update;
Peter Zijlstra2398f2c2008-06-27 13:41:35 +02001657
1658 if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) {
1659 sd->last_update = now;
Peter Zijlstraeb755802008-08-19 12:33:05 +02001660 walk_tg_tree(tg_nop, tg_shares_up, sd);
Peter Zijlstra2398f2c2008-06-27 13:41:35 +02001661 }
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001662}
1663
Peter Zijlstraeb755802008-08-19 12:33:05 +02001664static void update_h_load(long cpu)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001665{
Peter Zijlstraeb755802008-08-19 12:33:05 +02001666 walk_tg_tree(tg_load_down, tg_nop, (void *)cpu);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001667}
1668
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001669#else
1670
Peter Zijlstrac8cba852008-06-27 13:41:23 +02001671static inline void update_shares(struct sched_domain *sd)
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001672{
1673}
1674
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001675#endif
1676
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001677#ifdef CONFIG_PREEMPT
1678
Peter Zijlstrab78bb862009-09-15 14:23:18 +02001679static void double_rq_lock(struct rq *rq1, struct rq *rq2);
1680
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001681/*
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001682 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1683 * way at the expense of forcing extra atomic operations in all
1684 * invocations. This assures that the double_lock is acquired using the
1685 * same underlying policy as the spinlock_t on this architecture, which
1686 * reduces latency compared to the unfair variant below. However, it
1687 * also adds more overhead and therefore may reduce throughput.
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001688 */
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001689static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1690 __releases(this_rq->lock)
1691 __acquires(busiest->lock)
1692 __acquires(this_rq->lock)
1693{
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001694 raw_spin_unlock(&this_rq->lock);
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001695 double_rq_lock(this_rq, busiest);
1696
1697 return 1;
1698}
1699
1700#else
1701/*
1702 * Unfair double_lock_balance: Optimizes throughput at the expense of
1703 * latency by eliminating extra atomic operations when the locks are
1704 * already in proper order on entry. This favors lower cpu-ids and will
1705 * grant the double lock to lower cpus over higher ids under contention,
1706 * regardless of entry order into the function.
1707 */
1708static int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001709 __releases(this_rq->lock)
1710 __acquires(busiest->lock)
1711 __acquires(this_rq->lock)
1712{
1713 int ret = 0;
1714
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001715 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001716 if (busiest < this_rq) {
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001717 raw_spin_unlock(&this_rq->lock);
1718 raw_spin_lock(&busiest->lock);
1719 raw_spin_lock_nested(&this_rq->lock,
1720 SINGLE_DEPTH_NESTING);
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001721 ret = 1;
1722 } else
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001723 raw_spin_lock_nested(&busiest->lock,
1724 SINGLE_DEPTH_NESTING);
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001725 }
1726 return ret;
1727}
1728
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001729#endif /* CONFIG_PREEMPT */
1730
1731/*
1732 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1733 */
1734static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1735{
1736 if (unlikely(!irqs_disabled())) {
1737 /* printk() doesn't work good under rq->lock */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001738 raw_spin_unlock(&this_rq->lock);
Gregory Haskins8f45e2b2008-12-29 09:39:51 -05001739 BUG_ON(1);
1740 }
1741
1742 return _double_lock_balance(this_rq, busiest);
1743}
1744
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001745static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1746 __releases(busiest->lock)
1747{
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001748 raw_spin_unlock(&busiest->lock);
Alexey Dobriyan70574a92008-11-28 22:08:00 +03001749 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1750}
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001751
1752/*
1753 * double_rq_lock - safely lock two runqueues
1754 *
1755 * Note this does not disable interrupts like task_rq_lock,
1756 * you need to do so manually before calling.
1757 */
1758static void double_rq_lock(struct rq *rq1, struct rq *rq2)
1759 __acquires(rq1->lock)
1760 __acquires(rq2->lock)
1761{
1762 BUG_ON(!irqs_disabled());
1763 if (rq1 == rq2) {
1764 raw_spin_lock(&rq1->lock);
1765 __acquire(rq2->lock); /* Fake it out ;) */
1766 } else {
1767 if (rq1 < rq2) {
1768 raw_spin_lock(&rq1->lock);
1769 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1770 } else {
1771 raw_spin_lock(&rq2->lock);
1772 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1773 }
1774 }
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001775}
1776
1777/*
1778 * double_rq_unlock - safely unlock two runqueues
1779 *
1780 * Note this does not restore interrupts like task_rq_unlock,
1781 * you need to do so manually after calling.
1782 */
1783static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1784 __releases(rq1->lock)
1785 __releases(rq2->lock)
1786{
1787 raw_spin_unlock(&rq1->lock);
1788 if (rq1 != rq2)
1789 raw_spin_unlock(&rq2->lock);
1790 else
1791 __release(rq2->lock);
1792}
1793
Peter Zijlstra18d95a22008-04-19 19:45:00 +02001794#endif
Peter Zijlstra18d95a22008-04-19 19:45:00 +02001795
1796#ifdef CONFIG_FAIR_GROUP_SCHED
1797static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares)
1798{
Vegard Nossum30432092008-06-27 21:35:50 +02001799#ifdef CONFIG_SMP
Ingo Molnar34e83e82008-06-27 15:42:36 +02001800 cfs_rq->shares = shares;
1801#endif
Peter Zijlstra18d95a22008-04-19 19:45:00 +02001802}
1803#endif
1804
Peter Zijlstra74f51872010-04-22 21:50:19 +02001805static void calc_load_account_idle(struct rq *this_rq);
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01001806static void update_sysctl(void);
Christian Ehrhardtacb4a842009-11-30 12:16:48 +01001807static int get_update_sysctl_factor(void);
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02001808
Peter Zijlstracd29fe62009-11-27 17:32:46 +01001809static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1810{
1811 set_task_rq(p, cpu);
1812#ifdef CONFIG_SMP
1813 /*
1814 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1815 * successfuly executed on another CPU. We must ensure that updates of
1816 * per-task data have been completed by this moment.
1817 */
1818 smp_wmb();
1819 task_thread_info(p)->cpu = cpu;
1820#endif
1821}
Gregory Haskinse7693a32008-01-25 21:08:09 +01001822
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001823static const struct sched_class rt_sched_class;
Ingo Molnardd41f592007-07-09 18:51:59 +02001824
1825#define sched_class_highest (&rt_sched_class)
Gregory Haskins1f11eb62008-06-04 15:04:05 -04001826#define for_each_class(class) \
1827 for (class = sched_class_highest; class; class = class->next)
Ingo Molnardd41f592007-07-09 18:51:59 +02001828
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001829#include "sched_stats.h"
1830
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001831static void inc_nr_running(struct rq *rq)
Ingo Molnar6363ca52008-05-29 11:28:57 +02001832{
1833 rq->nr_running++;
Ingo Molnar6363ca52008-05-29 11:28:57 +02001834}
1835
Peter Zijlstrac09595f2008-06-27 13:41:14 +02001836static void dec_nr_running(struct rq *rq)
Ingo Molnar9c217242007-08-02 17:41:40 +02001837{
1838 rq->nr_running--;
Ingo Molnar9c217242007-08-02 17:41:40 +02001839}
1840
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001841static void set_load_weight(struct task_struct *p)
1842{
1843 if (task_has_rt_policy(p)) {
Peter Zijlstrae51fd5e2010-05-31 12:37:30 +02001844 p->se.load.weight = 0;
1845 p->se.load.inv_weight = WMULT_CONST;
Ingo Molnardd41f592007-07-09 18:51:59 +02001846 return;
1847 }
1848
1849 /*
1850 * SCHED_IDLE tasks get minimal weight:
1851 */
1852 if (p->policy == SCHED_IDLE) {
1853 p->se.load.weight = WEIGHT_IDLEPRIO;
1854 p->se.load.inv_weight = WMULT_IDLEPRIO;
1855 return;
1856 }
1857
1858 p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO];
1859 p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO];
Ingo Molnar45bf76d2007-07-09 18:51:59 +02001860}
1861
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001862static void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
Gregory Haskins2087a1a2008-06-27 14:30:00 -06001863{
Mike Galbraitha64692a2010-03-11 17:16:20 +01001864 update_rq_clock(rq);
Ingo Molnar71f8bd42007-07-09 18:51:59 +02001865 sched_info_queued(p);
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001866 p->sched_class->enqueue_task(rq, p, flags);
Ingo Molnardd41f592007-07-09 18:51:59 +02001867 p->se.on_rq = 1;
1868}
1869
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001870static void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
Ingo Molnardd41f592007-07-09 18:51:59 +02001871{
Mike Galbraitha64692a2010-03-11 17:16:20 +01001872 update_rq_clock(rq);
Ankita Garg46ac22b2008-07-01 14:30:06 +05301873 sched_info_dequeued(p);
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001874 p->sched_class->dequeue_task(rq, p, flags);
Ingo Molnardd41f592007-07-09 18:51:59 +02001875 p->se.on_rq = 0;
Ingo Molnar71f8bd42007-07-09 18:51:59 +02001876}
1877
1878/*
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001879 * activate_task - move a task to the runqueue.
1880 */
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001881static void activate_task(struct rq *rq, struct task_struct *p, int flags)
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001882{
1883 if (task_contributes_to_load(p))
1884 rq->nr_uninterruptible--;
1885
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001886 enqueue_task(rq, p, flags);
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001887 inc_nr_running(rq);
1888}
1889
1890/*
1891 * deactivate_task - remove a task from the runqueue.
1892 */
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001893static void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001894{
1895 if (task_contributes_to_load(p))
1896 rq->nr_uninterruptible++;
1897
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01001898 dequeue_task(rq, p, flags);
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +01001899 dec_nr_running(rq);
1900}
1901
1902#include "sched_idletask.c"
1903#include "sched_fair.c"
1904#include "sched_rt.c"
1905#ifdef CONFIG_SCHED_DEBUG
1906# include "sched_debug.c"
1907#endif
1908
1909/*
Ingo Molnardd41f592007-07-09 18:51:59 +02001910 * __normal_prio - return the priority that is based on the static prio
Ingo Molnar71f8bd42007-07-09 18:51:59 +02001911 */
Ingo Molnar14531182007-07-09 18:51:59 +02001912static inline int __normal_prio(struct task_struct *p)
1913{
Ingo Molnardd41f592007-07-09 18:51:59 +02001914 return p->static_prio;
Ingo Molnar14531182007-07-09 18:51:59 +02001915}
1916
1917/*
Ingo Molnarb29739f2006-06-27 02:54:51 -07001918 * Calculate the expected normal priority: i.e. priority
1919 * without taking RT-inheritance into account. Might be
1920 * boosted by interactivity modifiers. Changes upon fork,
1921 * setprio syscalls, and whenever the interactivity
1922 * estimator recalculates.
1923 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001924static inline int normal_prio(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -07001925{
1926 int prio;
1927
Ingo Molnare05606d2007-07-09 18:51:59 +02001928 if (task_has_rt_policy(p))
Ingo Molnarb29739f2006-06-27 02:54:51 -07001929 prio = MAX_RT_PRIO-1 - p->rt_priority;
1930 else
1931 prio = __normal_prio(p);
1932 return prio;
1933}
1934
1935/*
1936 * Calculate the current priority, i.e. the priority
1937 * taken into account by the scheduler. This value might
1938 * be boosted by RT tasks, or might be boosted by
1939 * interactivity modifiers. Will be RT if the task got
1940 * RT-boosted. If not then it returns p->normal_prio.
1941 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001942static int effective_prio(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -07001943{
1944 p->normal_prio = normal_prio(p);
1945 /*
1946 * If we are RT tasks or we were boosted to RT priority,
1947 * keep the priority unchanged. Otherwise, update priority
1948 * to the normal priority:
1949 */
1950 if (!rt_prio(p->prio))
1951 return p->normal_prio;
1952 return p->prio;
1953}
1954
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955/**
1956 * task_curr - is this task currently executing on a CPU?
1957 * @p: the task in question.
1958 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001959inline int task_curr(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960{
1961 return cpu_curr(task_cpu(p)) == p;
1962}
1963
Steven Rostedtcb469842008-01-25 21:08:22 +01001964static inline void check_class_changed(struct rq *rq, struct task_struct *p,
1965 const struct sched_class *prev_class,
1966 int oldprio, int running)
1967{
1968 if (prev_class != p->sched_class) {
1969 if (prev_class->switched_from)
1970 prev_class->switched_from(rq, p, running);
1971 p->sched_class->switched_to(rq, p, running);
1972 } else
1973 p->sched_class->prio_changed(rq, p, oldprio, running);
1974}
1975
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976#ifdef CONFIG_SMP
Ingo Molnarcc367732007-10-15 17:00:18 +02001977/*
1978 * Is this task likely cache-hot:
1979 */
Gregory Haskinse7693a32008-01-25 21:08:09 +01001980static int
Ingo Molnarcc367732007-10-15 17:00:18 +02001981task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
1982{
1983 s64 delta;
1984
Peter Zijlstrae6c8fba2009-12-16 18:04:33 +01001985 if (p->sched_class != &fair_sched_class)
1986 return 0;
1987
Ingo Molnarf540a602008-03-15 17:10:34 +01001988 /*
1989 * Buddy candidates are cache hot:
1990 */
Mike Galbraithf685cea2009-10-23 23:09:22 +02001991 if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running &&
Peter Zijlstra47932412008-11-04 21:25:09 +01001992 (&p->se == cfs_rq_of(&p->se)->next ||
1993 &p->se == cfs_rq_of(&p->se)->last))
Ingo Molnarf540a602008-03-15 17:10:34 +01001994 return 1;
1995
Ingo Molnar6bc16652007-10-15 17:00:18 +02001996 if (sysctl_sched_migration_cost == -1)
1997 return 1;
1998 if (sysctl_sched_migration_cost == 0)
1999 return 0;
2000
Ingo Molnarcc367732007-10-15 17:00:18 +02002001 delta = now - p->se.exec_start;
2002
2003 return delta < (s64)sysctl_sched_migration_cost;
2004}
2005
Ingo Molnardd41f592007-07-09 18:51:59 +02002006void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
Ingo Molnarc65cc872007-07-09 18:51:58 +02002007{
Peter Zijlstrae2912002009-12-16 18:04:36 +01002008#ifdef CONFIG_SCHED_DEBUG
2009 /*
2010 * We should never call set_task_cpu() on a blocked task,
2011 * ttwu() will sort out the placement.
2012 */
Peter Zijlstra077614e2009-12-17 13:16:31 +01002013 WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
2014 !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE));
Peter Zijlstrae2912002009-12-16 18:04:36 +01002015#endif
2016
Mathieu Desnoyersde1d7282009-05-05 16:49:59 +08002017 trace_sched_migrate_task(p, new_cpu);
Peter Zijlstracbc34ed2008-12-10 08:08:22 +01002018
Peter Zijlstra0c697742009-12-22 15:43:19 +01002019 if (task_cpu(p) != new_cpu) {
2020 p->se.nr_migrations++;
2021 perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0);
2022 }
Ingo Molnardd41f592007-07-09 18:51:59 +02002023
2024 __set_task_cpu(p, new_cpu);
Ingo Molnarc65cc872007-07-09 18:51:58 +02002025}
2026
Tejun Heo969c7922010-05-06 18:49:21 +02002027struct migration_arg {
Ingo Molnar36c8b582006-07-03 00:25:41 -07002028 struct task_struct *task;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 int dest_cpu;
Ingo Molnar70b97a72006-07-03 00:25:42 -07002030};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031
Tejun Heo969c7922010-05-06 18:49:21 +02002032static int migration_cpu_stop(void *data);
2033
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034/*
2035 * The task's runqueue lock must be held.
2036 * Returns true if you have to wait for migration thread.
2037 */
Tejun Heo969c7922010-05-06 18:49:21 +02002038static bool migrate_task(struct task_struct *p, int dest_cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039{
Ingo Molnar70b97a72006-07-03 00:25:42 -07002040 struct rq *rq = task_rq(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
2042 /*
2043 * If the task is not on a runqueue (and not running), then
Peter Zijlstrae2912002009-12-16 18:04:36 +01002044 * the next wake-up will properly place the task.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 */
Tejun Heo969c7922010-05-06 18:49:21 +02002046 return p->se.on_rq || task_running(rq, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047}
2048
2049/*
2050 * wait_task_inactive - wait for a thread to unschedule.
2051 *
Roland McGrath85ba2d82008-07-25 19:45:58 -07002052 * If @match_state is nonzero, it's the @p->state value just checked and
2053 * not expected to change. If it changes, i.e. @p might have woken up,
2054 * then return zero. When we succeed in waiting for @p to be off its CPU,
2055 * we return a positive number (its total switch count). If a second call
2056 * a short while later returns the same number, the caller can be sure that
2057 * @p has remained unscheduled the whole time.
2058 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 * The caller must ensure that the task *will* unschedule sometime soon,
2060 * else this function might spin for a *long* time. This function can't
2061 * be called with interrupts off, or it may introduce deadlock with
2062 * smp_call_function() if an IPI is sent by the same process we are
2063 * waiting to become inactive.
2064 */
Roland McGrath85ba2d82008-07-25 19:45:58 -07002065unsigned long wait_task_inactive(struct task_struct *p, long match_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066{
2067 unsigned long flags;
Ingo Molnardd41f592007-07-09 18:51:59 +02002068 int running, on_rq;
Roland McGrath85ba2d82008-07-25 19:45:58 -07002069 unsigned long ncsw;
Ingo Molnar70b97a72006-07-03 00:25:42 -07002070 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071
Andi Kleen3a5c3592007-10-15 17:00:14 +02002072 for (;;) {
2073 /*
2074 * We do the initial early heuristics without holding
2075 * any task-queue locks at all. We'll only try to get
2076 * the runqueue lock when things look like they will
2077 * work out!
2078 */
2079 rq = task_rq(p);
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07002080
Andi Kleen3a5c3592007-10-15 17:00:14 +02002081 /*
2082 * If the task is actively running on another CPU
2083 * still, just relax and busy-wait without holding
2084 * any locks.
2085 *
2086 * NOTE! Since we don't hold any locks, it's not
2087 * even sure that "rq" stays as the right runqueue!
2088 * But we don't care, since "task_running()" will
2089 * return false if the runqueue has changed and p
2090 * is actually now running somewhere else!
2091 */
Roland McGrath85ba2d82008-07-25 19:45:58 -07002092 while (task_running(rq, p)) {
2093 if (match_state && unlikely(p->state != match_state))
2094 return 0;
Andi Kleen3a5c3592007-10-15 17:00:14 +02002095 cpu_relax();
Roland McGrath85ba2d82008-07-25 19:45:58 -07002096 }
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07002097
Andi Kleen3a5c3592007-10-15 17:00:14 +02002098 /*
2099 * Ok, time to look more closely! We need the rq
2100 * lock now, to be *sure*. If we're wrong, we'll
2101 * just go back and repeat.
2102 */
2103 rq = task_rq_lock(p, &flags);
Peter Zijlstra27a9da62010-05-04 20:36:56 +02002104 trace_sched_wait_task(p);
Andi Kleen3a5c3592007-10-15 17:00:14 +02002105 running = task_running(rq, p);
2106 on_rq = p->se.on_rq;
Roland McGrath85ba2d82008-07-25 19:45:58 -07002107 ncsw = 0;
Oleg Nesterovf31e11d2008-08-20 16:54:44 -07002108 if (!match_state || p->state == match_state)
Oleg Nesterov93dcf552008-08-20 16:54:44 -07002109 ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
Andi Kleen3a5c3592007-10-15 17:00:14 +02002110 task_rq_unlock(rq, &flags);
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07002111
Andi Kleen3a5c3592007-10-15 17:00:14 +02002112 /*
Roland McGrath85ba2d82008-07-25 19:45:58 -07002113 * If it changed from the expected state, bail out now.
2114 */
2115 if (unlikely(!ncsw))
2116 break;
2117
2118 /*
Andi Kleen3a5c3592007-10-15 17:00:14 +02002119 * Was it really running after all now that we
2120 * checked with the proper locks actually held?
2121 *
2122 * Oops. Go back and try again..
2123 */
2124 if (unlikely(running)) {
2125 cpu_relax();
2126 continue;
2127 }
2128
2129 /*
2130 * It's not enough that it's not actively running,
2131 * it must be off the runqueue _entirely_, and not
2132 * preempted!
2133 *
Luis Henriques80dd99b2009-03-16 19:58:09 +00002134 * So if it was still runnable (but just not actively
Andi Kleen3a5c3592007-10-15 17:00:14 +02002135 * running right now), it's preempted, and we should
2136 * yield - it could be a while.
2137 */
2138 if (unlikely(on_rq)) {
2139 schedule_timeout_uninterruptible(1);
2140 continue;
2141 }
2142
2143 /*
2144 * Ahh, all good. It wasn't running, and it wasn't
2145 * runnable, which means that it will never become
2146 * running in the future either. We're all done!
2147 */
2148 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 }
Roland McGrath85ba2d82008-07-25 19:45:58 -07002150
2151 return ncsw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152}
2153
2154/***
2155 * kick_process - kick a running thread to enter/exit the kernel
2156 * @p: the to-be-kicked thread
2157 *
2158 * Cause a process which is running on another CPU to enter
2159 * kernel-mode, without any delay. (to get signals handled.)
2160 *
2161 * NOTE: this function doesnt have to take the runqueue lock,
2162 * because all it wants to ensure is that the remote task enters
2163 * the kernel. If the IPI races and the task has been migrated
2164 * to another CPU then no harm is done and the purpose has been
2165 * achieved as well.
2166 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07002167void kick_process(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168{
2169 int cpu;
2170
2171 preempt_disable();
2172 cpu = task_cpu(p);
2173 if ((cpu != smp_processor_id()) && task_curr(p))
2174 smp_send_reschedule(cpu);
2175 preempt_enable();
2176}
Rusty Russellb43e3522009-06-12 22:27:00 -06002177EXPORT_SYMBOL_GPL(kick_process);
Nick Piggin476d1392005-06-25 14:57:29 -07002178#endif /* CONFIG_SMP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179
Thomas Gleixner0793a612008-12-04 20:12:29 +01002180/**
2181 * task_oncpu_function_call - call a function on the cpu on which a task runs
2182 * @p: the task to evaluate
2183 * @func: the function to be called
2184 * @info: the function call argument
2185 *
2186 * Calls the function @func when the task is currently running. This might
2187 * be on the current CPU, which just calls the function directly
2188 */
2189void task_oncpu_function_call(struct task_struct *p,
2190 void (*func) (void *info), void *info)
2191{
2192 int cpu;
2193
2194 preempt_disable();
2195 cpu = task_cpu(p);
2196 if (task_curr(p))
2197 smp_call_function_single(cpu, func, info, 1);
2198 preempt_enable();
2199}
2200
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002201#ifdef CONFIG_SMP
Oleg Nesterov30da6882010-03-15 10:10:19 +01002202/*
2203 * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held.
2204 */
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01002205static int select_fallback_rq(int cpu, struct task_struct *p)
2206{
2207 int dest_cpu;
2208 const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu));
2209
2210 /* Look for allowed, online CPU in same node. */
2211 for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask)
2212 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
2213 return dest_cpu;
2214
2215 /* Any allowed, online CPU? */
2216 dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask);
2217 if (dest_cpu < nr_cpu_ids)
2218 return dest_cpu;
2219
2220 /* No more Mr. Nice Guy. */
Oleg Nesterov897f0b32010-03-15 10:10:03 +01002221 if (unlikely(dest_cpu >= nr_cpu_ids)) {
Oleg Nesterov9084bb82010-03-15 10:10:27 +01002222 dest_cpu = cpuset_cpus_allowed_fallback(p);
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01002223 /*
2224 * Don't tell them about moving exiting tasks or
2225 * kernel threads (both mm NULL), since they never
2226 * leave kernel.
2227 */
2228 if (p->mm && printk_ratelimit()) {
2229 printk(KERN_INFO "process %d (%s) no "
2230 "longer affine to cpu%d\n",
2231 task_pid_nr(p), p->comm, cpu);
2232 }
2233 }
2234
2235 return dest_cpu;
2236}
2237
Peter Zijlstrae2912002009-12-16 18:04:36 +01002238/*
Oleg Nesterov30da6882010-03-15 10:10:19 +01002239 * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable.
Peter Zijlstrae2912002009-12-16 18:04:36 +01002240 */
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002241static inline
Peter Zijlstra0017d732010-03-24 18:34:10 +01002242int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags)
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002243{
Peter Zijlstra0017d732010-03-24 18:34:10 +01002244 int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags);
Peter Zijlstrae2912002009-12-16 18:04:36 +01002245
2246 /*
2247 * In order not to call set_task_cpu() on a blocking task we need
2248 * to rely on ttwu() to place the task on a valid ->cpus_allowed
2249 * cpu.
2250 *
2251 * Since this is common to all placement strategies, this lives here.
2252 *
2253 * [ this allows ->select_task() to simply return task_cpu(p) and
2254 * not worry about this generic constraint ]
2255 */
2256 if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) ||
Peter Zijlstra70f11202009-12-20 17:36:27 +01002257 !cpu_online(cpu)))
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01002258 cpu = select_fallback_rq(task_cpu(p), p);
Peter Zijlstrae2912002009-12-16 18:04:36 +01002259
2260 return cpu;
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002261}
Mike Galbraith09a40af2010-04-15 07:29:59 +02002262
2263static void update_avg(u64 *avg, u64 sample)
2264{
2265 s64 diff = sample - *avg;
2266 *avg += diff >> 3;
2267}
Peter Zijlstra970b13b2009-11-25 13:31:39 +01002268#endif
2269
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270/***
2271 * try_to_wake_up - wake up a thread
2272 * @p: the to-be-woken-up thread
2273 * @state: the mask of task states that can be woken
2274 * @sync: do a synchronous wakeup?
2275 *
2276 * Put it on the run-queue if it's not already there. The "current"
2277 * thread is always on the run-queue (except when the actual
2278 * re-schedule is in progress), and as such you're allowed to do
2279 * the simpler "current->state = TASK_RUNNING" to mark yourself
2280 * runnable without the overhead of this.
2281 *
2282 * returns failure only if the task is already active.
2283 */
Peter Zijlstra7d478722009-09-14 19:55:44 +02002284static int try_to_wake_up(struct task_struct *p, unsigned int state,
2285 int wake_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286{
Ingo Molnarcc367732007-10-15 17:00:18 +02002287 int cpu, orig_cpu, this_cpu, success = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 unsigned long flags;
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01002289 unsigned long en_flags = ENQUEUE_WAKEUP;
Dan Carpenterab3b3aa2010-03-06 14:17:52 +03002290 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002292 this_cpu = get_cpu();
Peter Zijlstra2398f2c2008-06-27 13:41:35 +02002293
Linus Torvalds04e2f172008-02-23 18:05:03 -08002294 smp_wmb();
Dan Carpenterab3b3aa2010-03-06 14:17:52 +03002295 rq = task_rq_lock(p, &flags);
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002296 if (!(p->state & state))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 goto out;
2298
Ingo Molnardd41f592007-07-09 18:51:59 +02002299 if (p->se.on_rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 goto out_running;
2301
2302 cpu = task_cpu(p);
Ingo Molnarcc367732007-10-15 17:00:18 +02002303 orig_cpu = cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
2305#ifdef CONFIG_SMP
2306 if (unlikely(task_running(rq, p)))
2307 goto out_activate;
2308
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002309 /*
2310 * In order to handle concurrent wakeups and release the rq->lock
2311 * we put the task in TASK_WAKING state.
Ingo Molnareb240732009-09-16 21:09:13 +02002312 *
2313 * First fix up the nr_uninterruptible count:
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002314 */
Peter Zijlstracc87f762010-03-26 12:22:14 +01002315 if (task_contributes_to_load(p)) {
2316 if (likely(cpu_online(orig_cpu)))
2317 rq->nr_uninterruptible--;
2318 else
2319 this_rq()->nr_uninterruptible--;
2320 }
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002321 p->state = TASK_WAKING;
Peter Zijlstraefbbd052009-12-16 18:04:40 +01002322
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01002323 if (p->sched_class->task_waking) {
Peter Zijlstraefbbd052009-12-16 18:04:40 +01002324 p->sched_class->task_waking(rq, p);
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01002325 en_flags |= ENQUEUE_WAKING;
Peter Zijlstra0970d292010-02-15 14:45:54 +01002326 }
Peter Zijlstraab19cb22009-11-27 15:44:43 +01002327
Peter Zijlstra0017d732010-03-24 18:34:10 +01002328 cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags);
2329 if (cpu != orig_cpu)
Mike Galbraithf5dc3752009-10-09 08:35:03 +02002330 set_task_cpu(p, cpu);
Peter Zijlstra0017d732010-03-24 18:34:10 +01002331 __task_rq_unlock(rq);
Peter Zijlstraab19cb22009-11-27 15:44:43 +01002332
Peter Zijlstra0970d292010-02-15 14:45:54 +01002333 rq = cpu_rq(cpu);
2334 raw_spin_lock(&rq->lock);
Mike Galbraithf5dc3752009-10-09 08:35:03 +02002335
Peter Zijlstra0970d292010-02-15 14:45:54 +01002336 /*
2337 * We migrated the task without holding either rq->lock, however
2338 * since the task is not on the task list itself, nobody else
2339 * will try and migrate the task, hence the rq should match the
2340 * cpu we just moved it to.
2341 */
2342 WARN_ON(task_cpu(p) != cpu);
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002343 WARN_ON(p->state != TASK_WAKING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344
Gregory Haskinse7693a32008-01-25 21:08:09 +01002345#ifdef CONFIG_SCHEDSTATS
2346 schedstat_inc(rq, ttwu_count);
2347 if (cpu == this_cpu)
2348 schedstat_inc(rq, ttwu_local);
2349 else {
2350 struct sched_domain *sd;
2351 for_each_domain(this_cpu, sd) {
Rusty Russell758b2cd2008-11-25 02:35:04 +10302352 if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
Gregory Haskinse7693a32008-01-25 21:08:09 +01002353 schedstat_inc(sd, ttwu_wake_remote);
2354 break;
2355 }
2356 }
2357 }
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02002358#endif /* CONFIG_SCHEDSTATS */
Gregory Haskinse7693a32008-01-25 21:08:09 +01002359
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360out_activate:
2361#endif /* CONFIG_SMP */
Lucas De Marchi41acab82010-03-10 23:37:45 -03002362 schedstat_inc(p, se.statistics.nr_wakeups);
Peter Zijlstra7d478722009-09-14 19:55:44 +02002363 if (wake_flags & WF_SYNC)
Lucas De Marchi41acab82010-03-10 23:37:45 -03002364 schedstat_inc(p, se.statistics.nr_wakeups_sync);
Ingo Molnarcc367732007-10-15 17:00:18 +02002365 if (orig_cpu != cpu)
Lucas De Marchi41acab82010-03-10 23:37:45 -03002366 schedstat_inc(p, se.statistics.nr_wakeups_migrate);
Ingo Molnarcc367732007-10-15 17:00:18 +02002367 if (cpu == this_cpu)
Lucas De Marchi41acab82010-03-10 23:37:45 -03002368 schedstat_inc(p, se.statistics.nr_wakeups_local);
Ingo Molnarcc367732007-10-15 17:00:18 +02002369 else
Lucas De Marchi41acab82010-03-10 23:37:45 -03002370 schedstat_inc(p, se.statistics.nr_wakeups_remote);
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01002371 activate_task(rq, p, en_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 success = 1;
2373
2374out_running:
Peter Zijlstra27a9da62010-05-04 20:36:56 +02002375 trace_sched_wakeup(p, success);
Peter Zijlstra7d478722009-09-14 19:55:44 +02002376 check_preempt_curr(rq, p, wake_flags);
Ingo Molnar4ae7d5c2008-03-19 01:42:00 +01002377
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 p->state = TASK_RUNNING;
Steven Rostedt9a897c52008-01-25 21:08:22 +01002379#ifdef CONFIG_SMP
Peter Zijlstraefbbd052009-12-16 18:04:40 +01002380 if (p->sched_class->task_woken)
2381 p->sched_class->task_woken(rq, p);
Mike Galbraitheae0c9d2009-11-10 03:50:02 +01002382
2383 if (unlikely(rq->idle_stamp)) {
2384 u64 delta = rq->clock - rq->idle_stamp;
2385 u64 max = 2*sysctl_sched_migration_cost;
2386
2387 if (delta > max)
2388 rq->avg_idle = max;
2389 else
2390 update_avg(&rq->avg_idle, delta);
2391 rq->idle_stamp = 0;
2392 }
Steven Rostedt9a897c52008-01-25 21:08:22 +01002393#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394out:
2395 task_rq_unlock(rq, &flags);
Peter Zijlstrae9c84312009-09-15 14:43:03 +02002396 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397
2398 return success;
2399}
2400
David Howells50fa6102009-04-28 15:01:38 +01002401/**
2402 * wake_up_process - Wake up a specific process
2403 * @p: The process to be woken up.
2404 *
2405 * Attempt to wake up the nominated process and move it to the set of runnable
2406 * processes. Returns 1 if the process was woken up, 0 if it was already
2407 * running.
2408 *
2409 * It may be assumed that this function implies a write memory barrier before
2410 * changing the task state if and only if any tasks are woken up.
2411 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08002412int wake_up_process(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413{
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05002414 return try_to_wake_up(p, TASK_ALL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416EXPORT_SYMBOL(wake_up_process);
2417
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08002418int wake_up_state(struct task_struct *p, unsigned int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419{
2420 return try_to_wake_up(p, state, 0);
2421}
2422
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423/*
2424 * Perform scheduler related setup for a newly forked process p.
2425 * p is forked by current.
Ingo Molnardd41f592007-07-09 18:51:59 +02002426 *
2427 * __sched_fork() is basic setup used by init_idle() too:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 */
Ingo Molnardd41f592007-07-09 18:51:59 +02002429static void __sched_fork(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430{
Ingo Molnardd41f592007-07-09 18:51:59 +02002431 p->se.exec_start = 0;
2432 p->se.sum_exec_runtime = 0;
Ingo Molnarf6cf8912007-08-28 12:53:24 +02002433 p->se.prev_sum_exec_runtime = 0;
Ingo Molnar6c594c22008-12-14 12:34:15 +01002434 p->se.nr_migrations = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02002435
2436#ifdef CONFIG_SCHEDSTATS
Lucas De Marchi41acab82010-03-10 23:37:45 -03002437 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02002438#endif
Nick Piggin476d1392005-06-25 14:57:29 -07002439
Peter Zijlstrafa717062008-01-25 21:08:27 +01002440 INIT_LIST_HEAD(&p->rt.run_list);
Ingo Molnardd41f592007-07-09 18:51:59 +02002441 p->se.on_rq = 0;
Peter Zijlstra4a55bd52008-04-19 19:45:00 +02002442 INIT_LIST_HEAD(&p->se.group_node);
Nick Piggin476d1392005-06-25 14:57:29 -07002443
Avi Kivitye107be32007-07-26 13:40:43 +02002444#ifdef CONFIG_PREEMPT_NOTIFIERS
2445 INIT_HLIST_HEAD(&p->preempt_notifiers);
2446#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02002447}
2448
2449/*
2450 * fork()/clone()-time setup:
2451 */
2452void sched_fork(struct task_struct *p, int clone_flags)
2453{
2454 int cpu = get_cpu();
2455
2456 __sched_fork(p);
Peter Zijlstra06b83b52009-12-16 18:04:35 +01002457 /*
Peter Zijlstra0017d732010-03-24 18:34:10 +01002458 * We mark the process as running here. This guarantees that
Peter Zijlstra06b83b52009-12-16 18:04:35 +01002459 * nobody will actually run it, and a signal or other external
2460 * event cannot wake it up and insert it on the runqueue either.
2461 */
Peter Zijlstra0017d732010-03-24 18:34:10 +01002462 p->state = TASK_RUNNING;
Ingo Molnardd41f592007-07-09 18:51:59 +02002463
Ingo Molnarb29739f2006-06-27 02:54:51 -07002464 /*
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02002465 * Revert to default priority/policy on fork if requested.
2466 */
2467 if (unlikely(p->sched_reset_on_fork)) {
Peter Williamsf83f9ac2009-09-24 06:47:10 +00002468 if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) {
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02002469 p->policy = SCHED_NORMAL;
Peter Williamsf83f9ac2009-09-24 06:47:10 +00002470 p->normal_prio = p->static_prio;
2471 }
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02002472
Mike Galbraith6c697bdf2009-06-17 10:48:02 +02002473 if (PRIO_TO_NICE(p->static_prio) < 0) {
2474 p->static_prio = NICE_TO_PRIO(0);
Peter Williamsf83f9ac2009-09-24 06:47:10 +00002475 p->normal_prio = p->static_prio;
Mike Galbraith6c697bdf2009-06-17 10:48:02 +02002476 set_load_weight(p);
2477 }
2478
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02002479 /*
2480 * We don't need the reset flag anymore after the fork. It has
2481 * fulfilled its duty:
2482 */
2483 p->sched_reset_on_fork = 0;
2484 }
Lennart Poetteringca94c442009-06-15 17:17:47 +02002485
Peter Williamsf83f9ac2009-09-24 06:47:10 +00002486 /*
2487 * Make sure we do not leak PI boosting priority to the child.
2488 */
2489 p->prio = current->normal_prio;
2490
Hiroshi Shimamoto2ddbf952007-10-15 17:00:11 +02002491 if (!rt_prio(p->prio))
2492 p->sched_class = &fair_sched_class;
Ingo Molnarb29739f2006-06-27 02:54:51 -07002493
Peter Zijlstracd29fe62009-11-27 17:32:46 +01002494 if (p->sched_class->task_fork)
2495 p->sched_class->task_fork(p);
2496
Peter Zijlstra86951592010-06-22 11:44:53 +02002497 /*
2498 * The child is not yet in the pid-hash so no cgroup attach races,
2499 * and the cgroup is pinned to this child due to cgroup_fork()
2500 * is ran before sched_fork().
2501 *
2502 * Silence PROVE_RCU.
2503 */
2504 rcu_read_lock();
Peter Zijlstra5f3edc12009-09-10 13:42:00 +02002505 set_task_cpu(p, cpu);
Peter Zijlstra86951592010-06-22 11:44:53 +02002506 rcu_read_unlock();
Peter Zijlstra5f3edc12009-09-10 13:42:00 +02002507
Chandra Seetharaman52f17b62006-07-14 00:24:38 -07002508#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
Ingo Molnardd41f592007-07-09 18:51:59 +02002509 if (likely(sched_info_on()))
Chandra Seetharaman52f17b62006-07-14 00:24:38 -07002510 memset(&p->sched_info, 0, sizeof(p->sched_info));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511#endif
Chen, Kenneth Wd6077cb2006-02-14 13:53:10 -08002512#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
Nick Piggin4866cde2005-06-25 14:57:23 -07002513 p->oncpu = 0;
2514#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515#ifdef CONFIG_PREEMPT
Nick Piggin4866cde2005-06-25 14:57:23 -07002516 /* Want to start with kernel preemption disabled. */
Al Viroa1261f52005-11-13 16:06:55 -08002517 task_thread_info(p)->preempt_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518#endif
Gregory Haskins917b6272008-12-29 09:39:53 -05002519 plist_node_init(&p->pushable_tasks, MAX_PRIO);
2520
Nick Piggin476d1392005-06-25 14:57:29 -07002521 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522}
2523
2524/*
2525 * wake_up_new_task - wake up a newly created task for the first time.
2526 *
2527 * This function will do some initial scheduler statistics housekeeping
2528 * that must be done for every newly created context, then puts the task
2529 * on the runqueue and wakes it.
2530 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08002531void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532{
2533 unsigned long flags;
Ingo Molnardd41f592007-07-09 18:51:59 +02002534 struct rq *rq;
Andrew Mortonc8906922010-03-11 14:08:43 -08002535 int cpu __maybe_unused = get_cpu();
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002536
2537#ifdef CONFIG_SMP
Peter Zijlstra0017d732010-03-24 18:34:10 +01002538 rq = task_rq_lock(p, &flags);
2539 p->state = TASK_WAKING;
2540
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002541 /*
2542 * Fork balancing, do it here and not earlier because:
2543 * - cpus_allowed can change in the fork path
2544 * - any previously selected cpu might disappear through hotplug
2545 *
Peter Zijlstra0017d732010-03-24 18:34:10 +01002546 * We set TASK_WAKING so that select_task_rq() can drop rq->lock
2547 * without people poking at ->cpus_allowed.
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002548 */
Peter Zijlstra0017d732010-03-24 18:34:10 +01002549 cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0);
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002550 set_task_cpu(p, cpu);
Peter Zijlstra0017d732010-03-24 18:34:10 +01002551
2552 p->state = TASK_RUNNING;
2553 task_rq_unlock(rq, &flags);
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002554#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555
Peter Zijlstra0017d732010-03-24 18:34:10 +01002556 rq = task_rq_lock(p, &flags);
Peter Zijlstracd29fe62009-11-27 17:32:46 +01002557 activate_task(rq, p, 0);
Peter Zijlstra27a9da62010-05-04 20:36:56 +02002558 trace_sched_wakeup_new(p, 1);
Peter Zijlstraa7558e02009-09-14 20:02:34 +02002559 check_preempt_curr(rq, p, WF_FORK);
Steven Rostedt9a897c52008-01-25 21:08:22 +01002560#ifdef CONFIG_SMP
Peter Zijlstraefbbd052009-12-16 18:04:40 +01002561 if (p->sched_class->task_woken)
2562 p->sched_class->task_woken(rq, p);
Steven Rostedt9a897c52008-01-25 21:08:22 +01002563#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02002564 task_rq_unlock(rq, &flags);
Peter Zijlstrafabf3182010-01-21 21:04:57 +01002565 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566}
2567
Avi Kivitye107be32007-07-26 13:40:43 +02002568#ifdef CONFIG_PREEMPT_NOTIFIERS
2569
2570/**
Luis Henriques80dd99b2009-03-16 19:58:09 +00002571 * preempt_notifier_register - tell me when current is being preempted & rescheduled
Randy Dunlap421cee22007-07-31 00:37:50 -07002572 * @notifier: notifier struct to register
Avi Kivitye107be32007-07-26 13:40:43 +02002573 */
2574void preempt_notifier_register(struct preempt_notifier *notifier)
2575{
2576 hlist_add_head(&notifier->link, &current->preempt_notifiers);
2577}
2578EXPORT_SYMBOL_GPL(preempt_notifier_register);
2579
2580/**
2581 * preempt_notifier_unregister - no longer interested in preemption notifications
Randy Dunlap421cee22007-07-31 00:37:50 -07002582 * @notifier: notifier struct to unregister
Avi Kivitye107be32007-07-26 13:40:43 +02002583 *
2584 * This is safe to call from within a preemption notifier.
2585 */
2586void preempt_notifier_unregister(struct preempt_notifier *notifier)
2587{
2588 hlist_del(&notifier->link);
2589}
2590EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
2591
2592static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2593{
2594 struct preempt_notifier *notifier;
2595 struct hlist_node *node;
2596
2597 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2598 notifier->ops->sched_in(notifier, raw_smp_processor_id());
2599}
2600
2601static void
2602fire_sched_out_preempt_notifiers(struct task_struct *curr,
2603 struct task_struct *next)
2604{
2605 struct preempt_notifier *notifier;
2606 struct hlist_node *node;
2607
2608 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2609 notifier->ops->sched_out(notifier, next);
2610}
2611
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02002612#else /* !CONFIG_PREEMPT_NOTIFIERS */
Avi Kivitye107be32007-07-26 13:40:43 +02002613
2614static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2615{
2616}
2617
2618static void
2619fire_sched_out_preempt_notifiers(struct task_struct *curr,
2620 struct task_struct *next)
2621{
2622}
2623
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02002624#endif /* CONFIG_PREEMPT_NOTIFIERS */
Avi Kivitye107be32007-07-26 13:40:43 +02002625
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626/**
Nick Piggin4866cde2005-06-25 14:57:23 -07002627 * prepare_task_switch - prepare to switch tasks
2628 * @rq: the runqueue preparing to switch
Randy Dunlap421cee22007-07-31 00:37:50 -07002629 * @prev: the current task that is being switched out
Nick Piggin4866cde2005-06-25 14:57:23 -07002630 * @next: the task we are going to switch to.
2631 *
2632 * This is called with the rq lock held and interrupts off. It must
2633 * be paired with a subsequent finish_task_switch after the context
2634 * switch.
2635 *
2636 * prepare_task_switch sets up locking and calls architecture specific
2637 * hooks.
2638 */
Avi Kivitye107be32007-07-26 13:40:43 +02002639static inline void
2640prepare_task_switch(struct rq *rq, struct task_struct *prev,
2641 struct task_struct *next)
Nick Piggin4866cde2005-06-25 14:57:23 -07002642{
Avi Kivitye107be32007-07-26 13:40:43 +02002643 fire_sched_out_preempt_notifiers(prev, next);
Nick Piggin4866cde2005-06-25 14:57:23 -07002644 prepare_lock_switch(rq, next);
2645 prepare_arch_switch(next);
2646}
2647
2648/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 * finish_task_switch - clean up after a task-switch
Jeff Garzik344baba2005-09-07 01:15:17 -04002650 * @rq: runqueue associated with task-switch
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 * @prev: the thread we just switched away from.
2652 *
Nick Piggin4866cde2005-06-25 14:57:23 -07002653 * finish_task_switch must be called after the context switch, paired
2654 * with a prepare_task_switch call before the context switch.
2655 * finish_task_switch will reconcile locking set up by prepare_task_switch,
2656 * and do any other architecture-specific cleanup actions.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 *
2658 * Note that we may have delayed dropping an mm in context_switch(). If
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01002659 * so, we finish that here outside of the runqueue lock. (Doing it
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 * with the lock held can cause deadlocks; see schedule() for
2661 * details.)
2662 */
Alexey Dobriyana9957442007-10-15 17:00:13 +02002663static void finish_task_switch(struct rq *rq, struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 __releases(rq->lock)
2665{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 struct mm_struct *mm = rq->prev_mm;
Oleg Nesterov55a101f2006-09-29 02:01:10 -07002667 long prev_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668
2669 rq->prev_mm = NULL;
2670
2671 /*
2672 * A task struct has one reference for the use as "current".
Oleg Nesterovc394cc92006-09-29 02:01:11 -07002673 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
Oleg Nesterov55a101f2006-09-29 02:01:10 -07002674 * schedule one last time. The schedule call will never return, and
2675 * the scheduled task must drop that reference.
Oleg Nesterovc394cc92006-09-29 02:01:11 -07002676 * The test for TASK_DEAD must occur while the runqueue locks are
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 * still held, otherwise prev could be scheduled on another cpu, die
2678 * there before we look at prev->state, and then the reference would
2679 * be dropped twice.
2680 * Manfred Spraul <manfred@colorfullife.com>
2681 */
Oleg Nesterov55a101f2006-09-29 02:01:10 -07002682 prev_state = prev->state;
Nick Piggin4866cde2005-06-25 14:57:23 -07002683 finish_arch_switch(prev);
Jamie Iles8381f652010-01-08 15:27:33 +00002684#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2685 local_irq_disable();
2686#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
Peter Zijlstra49f47432009-12-27 11:51:52 +01002687 perf_event_task_sched_in(current);
Jamie Iles8381f652010-01-08 15:27:33 +00002688#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2689 local_irq_enable();
2690#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
Nick Piggin4866cde2005-06-25 14:57:23 -07002691 finish_lock_switch(rq, prev);
Steven Rostedte8fa1362008-01-25 21:08:05 +01002692
Avi Kivitye107be32007-07-26 13:40:43 +02002693 fire_sched_in_preempt_notifiers(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 if (mm)
2695 mmdrop(mm);
Oleg Nesterovc394cc92006-09-29 02:01:11 -07002696 if (unlikely(prev_state == TASK_DEAD)) {
bibo maoc6fd91f2006-03-26 01:38:20 -08002697 /*
2698 * Remove function-return probe instances associated with this
2699 * task and put them back on the free list.
Ingo Molnar9761eea2007-07-09 18:52:00 +02002700 */
bibo maoc6fd91f2006-03-26 01:38:20 -08002701 kprobe_flush_task(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 put_task_struct(prev);
bibo maoc6fd91f2006-03-26 01:38:20 -08002703 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704}
2705
Gregory Haskins3f029d32009-07-29 11:08:47 -04002706#ifdef CONFIG_SMP
2707
2708/* assumes rq->lock is held */
2709static inline void pre_schedule(struct rq *rq, struct task_struct *prev)
2710{
2711 if (prev->sched_class->pre_schedule)
2712 prev->sched_class->pre_schedule(rq, prev);
2713}
2714
2715/* rq->lock is NOT held, but preemption is disabled */
2716static inline void post_schedule(struct rq *rq)
2717{
2718 if (rq->post_schedule) {
2719 unsigned long flags;
2720
Thomas Gleixner05fa7852009-11-17 14:28:38 +01002721 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins3f029d32009-07-29 11:08:47 -04002722 if (rq->curr->sched_class->post_schedule)
2723 rq->curr->sched_class->post_schedule(rq);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01002724 raw_spin_unlock_irqrestore(&rq->lock, flags);
Gregory Haskins3f029d32009-07-29 11:08:47 -04002725
2726 rq->post_schedule = 0;
2727 }
2728}
2729
2730#else
2731
2732static inline void pre_schedule(struct rq *rq, struct task_struct *p)
2733{
2734}
2735
2736static inline void post_schedule(struct rq *rq)
2737{
2738}
2739
2740#endif
2741
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742/**
2743 * schedule_tail - first thing a freshly forked thread must call.
2744 * @prev: the thread we just switched away from.
2745 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07002746asmlinkage void schedule_tail(struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 __releases(rq->lock)
2748{
Ingo Molnar70b97a72006-07-03 00:25:42 -07002749 struct rq *rq = this_rq();
2750
Nick Piggin4866cde2005-06-25 14:57:23 -07002751 finish_task_switch(rq, prev);
Steven Rostedtda19ab52009-07-29 00:21:22 -04002752
Gregory Haskins3f029d32009-07-29 11:08:47 -04002753 /*
2754 * FIXME: do we need to worry about rq being invalidated by the
2755 * task_switch?
2756 */
2757 post_schedule(rq);
Steven Rostedtda19ab52009-07-29 00:21:22 -04002758
Nick Piggin4866cde2005-06-25 14:57:23 -07002759#ifdef __ARCH_WANT_UNLOCKED_CTXSW
2760 /* In this case, finish_task_switch does not reenable preemption */
2761 preempt_enable();
2762#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 if (current->set_child_tid)
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002764 put_user(task_pid_vnr(current), current->set_child_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765}
2766
2767/*
2768 * context_switch - switch to the new MM and the new
2769 * thread's register state.
2770 */
Ingo Molnardd41f592007-07-09 18:51:59 +02002771static inline void
Ingo Molnar70b97a72006-07-03 00:25:42 -07002772context_switch(struct rq *rq, struct task_struct *prev,
Ingo Molnar36c8b582006-07-03 00:25:41 -07002773 struct task_struct *next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774{
Ingo Molnardd41f592007-07-09 18:51:59 +02002775 struct mm_struct *mm, *oldmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776
Avi Kivitye107be32007-07-26 13:40:43 +02002777 prepare_task_switch(rq, prev, next);
Peter Zijlstra27a9da62010-05-04 20:36:56 +02002778 trace_sched_switch(prev, next);
Ingo Molnardd41f592007-07-09 18:51:59 +02002779 mm = next->mm;
2780 oldmm = prev->active_mm;
Zachary Amsden9226d122007-02-13 13:26:21 +01002781 /*
2782 * For paravirt, this is coupled with an exit in switch_to to
2783 * combine the page table reload and the switch backend into
2784 * one hypercall.
2785 */
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -08002786 arch_start_context_switch(prev);
Zachary Amsden9226d122007-02-13 13:26:21 +01002787
Tim Blechmann710390d2009-11-24 11:55:27 +01002788 if (likely(!mm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 next->active_mm = oldmm;
2790 atomic_inc(&oldmm->mm_count);
2791 enter_lazy_tlb(oldmm, next);
2792 } else
2793 switch_mm(oldmm, mm, next);
2794
Tim Blechmann710390d2009-11-24 11:55:27 +01002795 if (likely(!prev->mm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 prev->active_mm = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 rq->prev_mm = oldmm;
2798 }
Ingo Molnar3a5f5e42006-07-14 00:24:27 -07002799 /*
2800 * Since the runqueue lock will be released by the next
2801 * task (which is an invalid locking op but in the case
2802 * of the scheduler it's an obvious special-case), so we
2803 * do an early lockdep release here:
2804 */
2805#ifndef __ARCH_WANT_UNLOCKED_CTXSW
Ingo Molnar8a25d5d2006-07-03 00:24:54 -07002806 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
Ingo Molnar3a5f5e42006-07-14 00:24:27 -07002807#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808
2809 /* Here we just switch the register state and the stack. */
2810 switch_to(prev, next, prev);
2811
Ingo Molnardd41f592007-07-09 18:51:59 +02002812 barrier();
2813 /*
2814 * this_rq must be evaluated again because prev may have moved
2815 * CPUs since it called schedule(), thus the 'rq' on its stack
2816 * frame will be invalid.
2817 */
2818 finish_task_switch(this_rq(), prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819}
2820
2821/*
2822 * nr_running, nr_uninterruptible and nr_context_switches:
2823 *
2824 * externally visible scheduler statistics: current number of runnable
2825 * threads, current number of uninterruptible-sleeping threads, total
2826 * number of context switches performed since bootup.
2827 */
2828unsigned long nr_running(void)
2829{
2830 unsigned long i, sum = 0;
2831
2832 for_each_online_cpu(i)
2833 sum += cpu_rq(i)->nr_running;
2834
2835 return sum;
2836}
2837
2838unsigned long nr_uninterruptible(void)
2839{
2840 unsigned long i, sum = 0;
2841
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002842 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 sum += cpu_rq(i)->nr_uninterruptible;
2844
2845 /*
2846 * Since we read the counters lockless, it might be slightly
2847 * inaccurate. Do not allow it to go below zero though:
2848 */
2849 if (unlikely((long)sum < 0))
2850 sum = 0;
2851
2852 return sum;
2853}
2854
2855unsigned long long nr_context_switches(void)
2856{
Steven Rostedtcc94abf2006-06-27 02:54:31 -07002857 int i;
2858 unsigned long long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002860 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861 sum += cpu_rq(i)->nr_switches;
2862
2863 return sum;
2864}
2865
2866unsigned long nr_iowait(void)
2867{
2868 unsigned long i, sum = 0;
2869
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08002870 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 sum += atomic_read(&cpu_rq(i)->nr_iowait);
2872
2873 return sum;
2874}
2875
Peter Zijlstra8c215bd2010-07-01 09:07:17 +02002876unsigned long nr_iowait_cpu(int cpu)
Arjan van de Ven69d25872009-09-21 17:04:08 -07002877{
Peter Zijlstra8c215bd2010-07-01 09:07:17 +02002878 struct rq *this = cpu_rq(cpu);
Arjan van de Ven69d25872009-09-21 17:04:08 -07002879 return atomic_read(&this->nr_iowait);
2880}
2881
2882unsigned long this_cpu_load(void)
2883{
2884 struct rq *this = this_rq();
2885 return this->cpu_load[0];
2886}
2887
2888
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002889/* Variables and functions for calc_load */
2890static atomic_long_t calc_load_tasks;
2891static unsigned long calc_load_update;
2892unsigned long avenrun[3];
2893EXPORT_SYMBOL(avenrun);
2894
Peter Zijlstra74f51872010-04-22 21:50:19 +02002895static long calc_load_fold_active(struct rq *this_rq)
2896{
2897 long nr_active, delta = 0;
2898
2899 nr_active = this_rq->nr_running;
2900 nr_active += (long) this_rq->nr_uninterruptible;
2901
2902 if (nr_active != this_rq->calc_load_active) {
2903 delta = nr_active - this_rq->calc_load_active;
2904 this_rq->calc_load_active = nr_active;
2905 }
2906
2907 return delta;
2908}
2909
2910#ifdef CONFIG_NO_HZ
2911/*
2912 * For NO_HZ we delay the active fold to the next LOAD_FREQ update.
2913 *
2914 * When making the ILB scale, we should try to pull this in as well.
2915 */
2916static atomic_long_t calc_load_tasks_idle;
2917
2918static void calc_load_account_idle(struct rq *this_rq)
2919{
2920 long delta;
2921
2922 delta = calc_load_fold_active(this_rq);
2923 if (delta)
2924 atomic_long_add(delta, &calc_load_tasks_idle);
2925}
2926
2927static long calc_load_fold_idle(void)
2928{
2929 long delta = 0;
2930
2931 /*
2932 * Its got a race, we don't care...
2933 */
2934 if (atomic_long_read(&calc_load_tasks_idle))
2935 delta = atomic_long_xchg(&calc_load_tasks_idle, 0);
2936
2937 return delta;
2938}
2939#else
2940static void calc_load_account_idle(struct rq *this_rq)
2941{
2942}
2943
2944static inline long calc_load_fold_idle(void)
2945{
2946 return 0;
2947}
2948#endif
2949
Thomas Gleixner2d024942009-05-02 20:08:52 +02002950/**
2951 * get_avenrun - get the load average array
2952 * @loads: pointer to dest load array
2953 * @offset: offset to add
2954 * @shift: shift count to shift the result left
2955 *
2956 * These values are estimates at best, so no need for locking.
2957 */
2958void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
2959{
2960 loads[0] = (avenrun[0] + offset) << shift;
2961 loads[1] = (avenrun[1] + offset) << shift;
2962 loads[2] = (avenrun[2] + offset) << shift;
2963}
2964
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002965static unsigned long
2966calc_load(unsigned long load, unsigned long exp, unsigned long active)
Jack Steinerdb1b1fe2006-03-31 02:31:21 -08002967{
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002968 load *= exp;
2969 load += active * (FIXED_1 - exp);
2970 return load >> FSHIFT;
2971}
Jack Steinerdb1b1fe2006-03-31 02:31:21 -08002972
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002973/*
2974 * calc_load - update the avenrun load estimates 10 ticks after the
2975 * CPUs have updated calc_load_tasks.
2976 */
2977void calc_global_load(void)
2978{
2979 unsigned long upd = calc_load_update + 10;
2980 long active;
2981
2982 if (time_before(jiffies, upd))
2983 return;
2984
2985 active = atomic_long_read(&calc_load_tasks);
2986 active = active > 0 ? active * FIXED_1 : 0;
2987
2988 avenrun[0] = calc_load(avenrun[0], EXP_1, active);
2989 avenrun[1] = calc_load(avenrun[1], EXP_5, active);
2990 avenrun[2] = calc_load(avenrun[2], EXP_15, active);
2991
2992 calc_load_update += LOAD_FREQ;
2993}
2994
2995/*
Peter Zijlstra74f51872010-04-22 21:50:19 +02002996 * Called from update_cpu_load() to periodically update this CPU's
2997 * active count.
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002998 */
2999static void calc_load_account_active(struct rq *this_rq)
3000{
Peter Zijlstra74f51872010-04-22 21:50:19 +02003001 long delta;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003002
Peter Zijlstra74f51872010-04-22 21:50:19 +02003003 if (time_before(jiffies, this_rq->calc_load_update))
3004 return;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003005
Peter Zijlstra74f51872010-04-22 21:50:19 +02003006 delta = calc_load_fold_active(this_rq);
3007 delta += calc_load_fold_idle();
3008 if (delta)
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003009 atomic_long_add(delta, &calc_load_tasks);
Peter Zijlstra74f51872010-04-22 21:50:19 +02003010
3011 this_rq->calc_load_update += LOAD_FREQ;
Jack Steinerdb1b1fe2006-03-31 02:31:21 -08003012}
3013
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014/*
Ingo Molnardd41f592007-07-09 18:51:59 +02003015 * Update rq->cpu_load[] statistics. This function is usually called every
3016 * scheduler tick (TICK_NSEC).
Ingo Molnar48f24c42006-07-03 00:25:40 -07003017 */
Ingo Molnardd41f592007-07-09 18:51:59 +02003018static void update_cpu_load(struct rq *this_rq)
Ingo Molnar48f24c42006-07-03 00:25:40 -07003019{
Dmitry Adamushko495eca42007-10-15 17:00:06 +02003020 unsigned long this_load = this_rq->load.weight;
Ingo Molnardd41f592007-07-09 18:51:59 +02003021 int i, scale;
3022
3023 this_rq->nr_load_updates++;
Ingo Molnardd41f592007-07-09 18:51:59 +02003024
3025 /* Update our load: */
3026 for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
3027 unsigned long old_load, new_load;
3028
3029 /* scale is effectively 1 << i now, and >> i divides by scale */
3030
3031 old_load = this_rq->cpu_load[i];
3032 new_load = this_load;
Ingo Molnara25707f2007-10-15 17:00:03 +02003033 /*
3034 * Round up the averaging division if load is increasing. This
3035 * prevents us from getting stuck on 9 if the load is 10, for
3036 * example.
3037 */
3038 if (new_load > old_load)
3039 new_load += scale-1;
Ingo Molnardd41f592007-07-09 18:51:59 +02003040 this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i;
3041 }
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02003042
Peter Zijlstra74f51872010-04-22 21:50:19 +02003043 calc_load_account_active(this_rq);
Ingo Molnar48f24c42006-07-03 00:25:40 -07003044}
3045
Ingo Molnardd41f592007-07-09 18:51:59 +02003046#ifdef CONFIG_SMP
3047
Ingo Molnar48f24c42006-07-03 00:25:40 -07003048/*
Peter Zijlstra38022902009-12-16 18:04:37 +01003049 * sched_exec - execve() is a valuable balancing opportunity, because at
3050 * this point the task has the smallest effective memory and cache footprint.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 */
Peter Zijlstra38022902009-12-16 18:04:37 +01003052void sched_exec(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053{
Peter Zijlstra38022902009-12-16 18:04:37 +01003054 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07003056 struct rq *rq;
Peter Zijlstra0017d732010-03-24 18:34:10 +01003057 int dest_cpu;
Peter Zijlstra38022902009-12-16 18:04:37 +01003058
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059 rq = task_rq_lock(p, &flags);
Peter Zijlstra0017d732010-03-24 18:34:10 +01003060 dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0);
3061 if (dest_cpu == smp_processor_id())
3062 goto unlock;
Peter Zijlstra38022902009-12-16 18:04:37 +01003063
3064 /*
3065 * select_task_rq() can race against ->cpus_allowed
3066 */
Oleg Nesterov30da6882010-03-15 10:10:19 +01003067 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) &&
Tejun Heo969c7922010-05-06 18:49:21 +02003068 likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) {
3069 struct migration_arg arg = { p, dest_cpu };
Ingo Molnar36c8b582006-07-03 00:25:41 -07003070
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071 task_rq_unlock(rq, &flags);
Tejun Heo969c7922010-05-06 18:49:21 +02003072 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 return;
3074 }
Peter Zijlstra0017d732010-03-24 18:34:10 +01003075unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076 task_rq_unlock(rq, &flags);
3077}
3078
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079#endif
3080
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081DEFINE_PER_CPU(struct kernel_stat, kstat);
3082
3083EXPORT_PER_CPU_SYMBOL(kstat);
3084
3085/*
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003086 * Return any ns on the sched_clock that have not yet been accounted in
Frank Mayharf06febc2008-09-12 09:54:39 -07003087 * @p in case that task is currently running.
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003088 *
3089 * Called with task_rq_lock() held on @rq.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 */
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003091static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq)
3092{
3093 u64 ns = 0;
3094
3095 if (task_current(rq, p)) {
3096 update_rq_clock(rq);
3097 ns = rq->clock - p->se.exec_start;
3098 if ((s64)ns < 0)
3099 ns = 0;
3100 }
3101
3102 return ns;
3103}
3104
Frank Mayharbb34d922008-09-12 09:54:39 -07003105unsigned long long task_delta_exec(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107 unsigned long flags;
Ingo Molnar41b86e92007-07-09 18:51:58 +02003108 struct rq *rq;
Frank Mayharbb34d922008-09-12 09:54:39 -07003109 u64 ns = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07003110
Ingo Molnar41b86e92007-07-09 18:51:58 +02003111 rq = task_rq_lock(p, &flags);
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003112 ns = do_task_delta_exec(p, rq);
3113 task_rq_unlock(rq, &flags);
Ingo Molnar15084872008-09-30 08:28:17 +02003114
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003115 return ns;
3116}
Frank Mayharf06febc2008-09-12 09:54:39 -07003117
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003118/*
3119 * Return accounted runtime for the task.
3120 * In case the task is currently running, return the runtime plus current's
3121 * pending runtime that have not been accounted yet.
3122 */
3123unsigned long long task_sched_runtime(struct task_struct *p)
3124{
3125 unsigned long flags;
3126 struct rq *rq;
3127 u64 ns = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07003128
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09003129 rq = task_rq_lock(p, &flags);
3130 ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq);
3131 task_rq_unlock(rq, &flags);
3132
3133 return ns;
3134}
3135
3136/*
3137 * Return sum_exec_runtime for the thread group.
3138 * In case the task is currently running, return the sum plus current's
3139 * pending runtime that have not been accounted yet.
3140 *
3141 * Note that the thread group might have other running tasks as well,
3142 * so the return value not includes other pending runtime that other
3143 * running tasks might have.
3144 */
3145unsigned long long thread_group_sched_runtime(struct task_struct *p)
3146{
3147 struct task_cputime totals;
3148 unsigned long flags;
3149 struct rq *rq;
3150 u64 ns;
3151
3152 rq = task_rq_lock(p, &flags);
3153 thread_group_cputime(p, &totals);
3154 ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155 task_rq_unlock(rq, &flags);
3156
3157 return ns;
3158}
3159
3160/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 * Account user cpu time to a process.
3162 * @p: the process that the cpu time gets accounted to
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 * @cputime: the cpu time spent in user space since the last update
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003164 * @cputime_scaled: cputime scaled by cpu frequency
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165 */
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003166void account_user_time(struct task_struct *p, cputime_t cputime,
3167 cputime_t cputime_scaled)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168{
3169 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3170 cputime64_t tmp;
3171
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003172 /* Add user time to process. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173 p->utime = cputime_add(p->utime, cputime);
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003174 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
Frank Mayharf06febc2008-09-12 09:54:39 -07003175 account_group_user_time(p, cputime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176
3177 /* Add user time to cpustat. */
3178 tmp = cputime_to_cputime64(cputime);
3179 if (TASK_NICE(p) > 0)
3180 cpustat->nice = cputime64_add(cpustat->nice, tmp);
3181 else
3182 cpustat->user = cputime64_add(cpustat->user, tmp);
Bharata B Raoef12fef2009-03-31 10:02:22 +05303183
3184 cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime);
Jonathan Lim49b5cf32008-07-25 01:48:40 -07003185 /* Account for user time used */
3186 acct_update_integrals(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187}
3188
3189/*
Laurent Vivier94886b82007-10-15 17:00:19 +02003190 * Account guest cpu time to a process.
3191 * @p: the process that the cpu time gets accounted to
3192 * @cputime: the cpu time spent in virtual machine since the last update
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003193 * @cputime_scaled: cputime scaled by cpu frequency
Laurent Vivier94886b82007-10-15 17:00:19 +02003194 */
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003195static void account_guest_time(struct task_struct *p, cputime_t cputime,
3196 cputime_t cputime_scaled)
Laurent Vivier94886b82007-10-15 17:00:19 +02003197{
3198 cputime64_t tmp;
3199 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3200
3201 tmp = cputime_to_cputime64(cputime);
3202
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003203 /* Add guest time to process. */
Laurent Vivier94886b82007-10-15 17:00:19 +02003204 p->utime = cputime_add(p->utime, cputime);
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003205 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
Frank Mayharf06febc2008-09-12 09:54:39 -07003206 account_group_user_time(p, cputime);
Laurent Vivier94886b82007-10-15 17:00:19 +02003207 p->gtime = cputime_add(p->gtime, cputime);
3208
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003209 /* Add guest time to cpustat. */
Ryota Ozakice0e7b22009-10-24 01:20:10 +09003210 if (TASK_NICE(p) > 0) {
3211 cpustat->nice = cputime64_add(cpustat->nice, tmp);
3212 cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp);
3213 } else {
3214 cpustat->user = cputime64_add(cpustat->user, tmp);
3215 cpustat->guest = cputime64_add(cpustat->guest, tmp);
3216 }
Laurent Vivier94886b82007-10-15 17:00:19 +02003217}
3218
3219/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220 * Account system cpu time to a process.
3221 * @p: the process that the cpu time gets accounted to
3222 * @hardirq_offset: the offset to subtract from hardirq_count()
3223 * @cputime: the cpu time spent in kernel space since the last update
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003224 * @cputime_scaled: cputime scaled by cpu frequency
Linus Torvalds1da177e2005-04-16 15:20:36 -07003225 */
3226void account_system_time(struct task_struct *p, int hardirq_offset,
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003227 cputime_t cputime, cputime_t cputime_scaled)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003228{
3229 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230 cputime64_t tmp;
3231
Harvey Harrison983ed7a2008-04-24 18:17:55 -07003232 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003233 account_guest_time(p, cputime, cputime_scaled);
Harvey Harrison983ed7a2008-04-24 18:17:55 -07003234 return;
3235 }
Laurent Vivier94886b82007-10-15 17:00:19 +02003236
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003237 /* Add system time to process. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238 p->stime = cputime_add(p->stime, cputime);
Martin Schwidefsky457533a2008-12-31 15:11:37 +01003239 p->stimescaled = cputime_add(p->stimescaled, cputime_scaled);
Frank Mayharf06febc2008-09-12 09:54:39 -07003240 account_group_system_time(p, cputime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241
3242 /* Add system time to cpustat. */
3243 tmp = cputime_to_cputime64(cputime);
3244 if (hardirq_count() - hardirq_offset)
3245 cpustat->irq = cputime64_add(cpustat->irq, tmp);
3246 else if (softirq_count())
3247 cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248 else
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003249 cpustat->system = cputime64_add(cpustat->system, tmp);
3250
Bharata B Raoef12fef2009-03-31 10:02:22 +05303251 cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime);
3252
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253 /* Account for system time used */
3254 acct_update_integrals(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255}
3256
3257/*
3258 * Account for involuntary wait time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259 * @steal: the cpu time spent in involuntary wait
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260 */
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003261void account_steal_time(cputime_t cputime)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003264 cputime64_t cputime64 = cputime_to_cputime64(cputime);
3265
3266 cpustat->steal = cputime64_add(cpustat->steal, cputime64);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267}
3268
Christoph Lameter7835b982006-12-10 02:20:22 -08003269/*
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003270 * Account for idle time.
3271 * @cputime: the cpu time spent in idle wait
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272 */
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003273void account_idle_time(cputime_t cputime)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274{
3275 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003276 cputime64_t cputime64 = cputime_to_cputime64(cputime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277 struct rq *rq = this_rq();
3278
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003279 if (atomic_read(&rq->nr_iowait) > 0)
3280 cpustat->iowait = cputime64_add(cpustat->iowait, cputime64);
3281 else
3282 cpustat->idle = cputime64_add(cpustat->idle, cputime64);
Christoph Lameter7835b982006-12-10 02:20:22 -08003283}
3284
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003285#ifndef CONFIG_VIRT_CPU_ACCOUNTING
3286
3287/*
3288 * Account a single tick of cpu time.
3289 * @p: the process that the cpu time gets accounted to
3290 * @user_tick: indicates if the tick is a user or a system tick
3291 */
3292void account_process_tick(struct task_struct *p, int user_tick)
3293{
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02003294 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003295 struct rq *rq = this_rq();
3296
3297 if (user_tick)
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02003298 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
Eric Dumazetf5f293a2009-04-29 14:44:49 +02003299 else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02003300 account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003301 one_jiffy_scaled);
3302 else
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02003303 account_idle_time(cputime_one_jiffy);
Martin Schwidefsky79741dd2008-12-31 15:11:38 +01003304}
3305
3306/*
3307 * Account multiple ticks of steal time.
3308 * @p: the process from which the cpu time has been stolen
3309 * @ticks: number of stolen ticks
3310 */
3311void account_steal_ticks(unsigned long ticks)
3312{
3313 account_steal_time(jiffies_to_cputime(ticks));
3314}
3315
3316/*
3317 * Account multiple ticks of idle time.
3318 * @ticks: number of stolen ticks
3319 */
3320void account_idle_ticks(unsigned long ticks)
3321{
3322 account_idle_time(jiffies_to_cputime(ticks));
3323}
3324
3325#endif
3326
Christoph Lameter7835b982006-12-10 02:20:22 -08003327/*
Balbir Singh49048622008-09-05 18:12:23 +02003328 * Use precise platform statistics if available:
3329 */
3330#ifdef CONFIG_VIRT_CPU_ACCOUNTING
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003331void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
Balbir Singh49048622008-09-05 18:12:23 +02003332{
Hidetoshi Setod99ca3b2009-12-02 17:26:47 +09003333 *ut = p->utime;
3334 *st = p->stime;
Balbir Singh49048622008-09-05 18:12:23 +02003335}
3336
Hidetoshi Seto0cf55e12009-12-02 17:28:07 +09003337void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
Balbir Singh49048622008-09-05 18:12:23 +02003338{
Hidetoshi Seto0cf55e12009-12-02 17:28:07 +09003339 struct task_cputime cputime;
3340
3341 thread_group_cputime(p, &cputime);
3342
3343 *ut = cputime.utime;
3344 *st = cputime.stime;
Balbir Singh49048622008-09-05 18:12:23 +02003345}
3346#else
Hidetoshi Seto761b1d22009-11-12 13:33:45 +09003347
3348#ifndef nsecs_to_cputime
Hidetoshi Setob7b20df92009-11-26 14:49:27 +09003349# define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs)
Hidetoshi Seto761b1d22009-11-12 13:33:45 +09003350#endif
3351
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003352void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
Balbir Singh49048622008-09-05 18:12:23 +02003353{
Hidetoshi Setod99ca3b2009-12-02 17:26:47 +09003354 cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime);
Balbir Singh49048622008-09-05 18:12:23 +02003355
3356 /*
3357 * Use CFS's precise accounting:
3358 */
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003359 rtime = nsecs_to_cputime(p->se.sum_exec_runtime);
Balbir Singh49048622008-09-05 18:12:23 +02003360
3361 if (total) {
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003362 u64 temp;
Balbir Singh49048622008-09-05 18:12:23 +02003363
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003364 temp = (u64)(rtime * utime);
Balbir Singh49048622008-09-05 18:12:23 +02003365 do_div(temp, total);
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003366 utime = (cputime_t)temp;
3367 } else
3368 utime = rtime;
Balbir Singh49048622008-09-05 18:12:23 +02003369
3370 /*
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003371 * Compare with previous values, to keep monotonicity:
Balbir Singh49048622008-09-05 18:12:23 +02003372 */
Hidetoshi Seto761b1d22009-11-12 13:33:45 +09003373 p->prev_utime = max(p->prev_utime, utime);
Hidetoshi Setod99ca3b2009-12-02 17:26:47 +09003374 p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime));
Balbir Singh49048622008-09-05 18:12:23 +02003375
Hidetoshi Setod99ca3b2009-12-02 17:26:47 +09003376 *ut = p->prev_utime;
3377 *st = p->prev_stime;
Hidetoshi Setod180c5b2009-11-26 14:48:30 +09003378}
Balbir Singh49048622008-09-05 18:12:23 +02003379
Hidetoshi Seto0cf55e12009-12-02 17:28:07 +09003380/*
3381 * Must be called with siglock held.
3382 */
3383void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3384{
3385 struct signal_struct *sig = p->signal;
3386 struct task_cputime cputime;
3387 cputime_t rtime, utime, total;
3388
3389 thread_group_cputime(p, &cputime);
3390
3391 total = cputime_add(cputime.utime, cputime.stime);
3392 rtime = nsecs_to_cputime(cputime.sum_exec_runtime);
3393
3394 if (total) {
3395 u64 temp;
3396
3397 temp = (u64)(rtime * cputime.utime);
3398 do_div(temp, total);
3399 utime = (cputime_t)temp;
3400 } else
3401 utime = rtime;
3402
3403 sig->prev_utime = max(sig->prev_utime, utime);
3404 sig->prev_stime = max(sig->prev_stime,
3405 cputime_sub(rtime, sig->prev_utime));
3406
3407 *ut = sig->prev_utime;
3408 *st = sig->prev_stime;
Balbir Singh49048622008-09-05 18:12:23 +02003409}
3410#endif
3411
Balbir Singh49048622008-09-05 18:12:23 +02003412/*
Christoph Lameter7835b982006-12-10 02:20:22 -08003413 * This function gets called by the timer code, with HZ frequency.
3414 * We call it with interrupts disabled.
3415 *
3416 * It also gets called by the fork code, when changing the parent's
3417 * timeslices.
3418 */
3419void scheduler_tick(void)
3420{
Christoph Lameter7835b982006-12-10 02:20:22 -08003421 int cpu = smp_processor_id();
3422 struct rq *rq = cpu_rq(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02003423 struct task_struct *curr = rq->curr;
Peter Zijlstra3e51f332008-05-03 18:29:28 +02003424
3425 sched_clock_tick();
Christoph Lameter7835b982006-12-10 02:20:22 -08003426
Thomas Gleixner05fa7852009-11-17 14:28:38 +01003427 raw_spin_lock(&rq->lock);
Peter Zijlstra3e51f332008-05-03 18:29:28 +02003428 update_rq_clock(rq);
Ingo Molnarf1a438d2007-08-09 11:16:45 +02003429 update_cpu_load(rq);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01003430 curr->sched_class->task_tick(rq, curr, 0);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01003431 raw_spin_unlock(&rq->lock);
Ingo Molnardd41f592007-07-09 18:51:59 +02003432
Peter Zijlstra49f47432009-12-27 11:51:52 +01003433 perf_event_task_tick(curr);
Peter Zijlstrae220d2d2009-05-23 18:28:55 +02003434
Christoph Lametere418e1c2006-12-10 02:20:23 -08003435#ifdef CONFIG_SMP
Ingo Molnardd41f592007-07-09 18:51:59 +02003436 rq->idle_at_tick = idle_cpu(cpu);
3437 trigger_load_balance(rq, cpu);
Christoph Lametere418e1c2006-12-10 02:20:23 -08003438#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003439}
3440
Lai Jiangshan132380a2009-04-02 14:18:25 +08003441notrace unsigned long get_parent_ip(unsigned long addr)
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003442{
3443 if (in_lock_functions(addr)) {
3444 addr = CALLER_ADDR2;
3445 if (in_lock_functions(addr))
3446 addr = CALLER_ADDR3;
3447 }
3448 return addr;
3449}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450
Steven Rostedt7e49fcc2009-01-22 19:01:40 -05003451#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
3452 defined(CONFIG_PREEMPT_TRACER))
3453
Srinivasa Ds43627582008-02-23 15:24:04 -08003454void __kprobes add_preempt_count(int val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003455{
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003456#ifdef CONFIG_DEBUG_PREEMPT
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457 /*
3458 * Underflow?
3459 */
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07003460 if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
3461 return;
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003462#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003463 preempt_count() += val;
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003464#ifdef CONFIG_DEBUG_PREEMPT
Linus Torvalds1da177e2005-04-16 15:20:36 -07003465 /*
3466 * Spinlock count overflowing soon?
3467 */
Miguel Ojeda Sandonis33859f72006-12-10 02:20:38 -08003468 DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
3469 PREEMPT_MASK - 10);
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003470#endif
3471 if (preempt_count() == val)
3472 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003473}
3474EXPORT_SYMBOL(add_preempt_count);
3475
Srinivasa Ds43627582008-02-23 15:24:04 -08003476void __kprobes sub_preempt_count(int val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477{
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003478#ifdef CONFIG_DEBUG_PREEMPT
Linus Torvalds1da177e2005-04-16 15:20:36 -07003479 /*
3480 * Underflow?
3481 */
Ingo Molnar01e3eb82009-01-12 13:00:50 +01003482 if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07003483 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484 /*
3485 * Is the spinlock portion underflowing?
3486 */
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07003487 if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
3488 !(preempt_count() & PREEMPT_MASK)))
3489 return;
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003490#endif
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07003491
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02003492 if (preempt_count() == val)
3493 trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003494 preempt_count() -= val;
3495}
3496EXPORT_SYMBOL(sub_preempt_count);
3497
3498#endif
3499
3500/*
Ingo Molnardd41f592007-07-09 18:51:59 +02003501 * Print scheduling while atomic bug:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502 */
Ingo Molnardd41f592007-07-09 18:51:59 +02003503static noinline void __schedule_bug(struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003504{
Satyam Sharma838225b2007-10-24 18:23:50 +02003505 struct pt_regs *regs = get_irq_regs();
3506
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01003507 printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
3508 prev->comm, prev->pid, preempt_count());
Satyam Sharma838225b2007-10-24 18:23:50 +02003509
Ingo Molnardd41f592007-07-09 18:51:59 +02003510 debug_show_held_locks(prev);
Arjan van de Vene21f5b12008-05-23 09:05:58 -07003511 print_modules();
Ingo Molnardd41f592007-07-09 18:51:59 +02003512 if (irqs_disabled())
3513 print_irqtrace_events(prev);
Satyam Sharma838225b2007-10-24 18:23:50 +02003514
3515 if (regs)
3516 show_regs(regs);
3517 else
3518 dump_stack();
Ingo Molnardd41f592007-07-09 18:51:59 +02003519}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003520
Ingo Molnardd41f592007-07-09 18:51:59 +02003521/*
3522 * Various schedule()-time debugging checks and statistics:
3523 */
3524static inline void schedule_debug(struct task_struct *prev)
3525{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003526 /*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003527 * Test if we are atomic. Since do_exit() needs to call into
Linus Torvalds1da177e2005-04-16 15:20:36 -07003528 * schedule() atomically, we ignore that path for now.
3529 * Otherwise, whine if we are scheduling when we should not be.
3530 */
Roel Kluin3f33a7c2008-05-13 23:44:11 +02003531 if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
Ingo Molnardd41f592007-07-09 18:51:59 +02003532 __schedule_bug(prev);
3533
Linus Torvalds1da177e2005-04-16 15:20:36 -07003534 profile_hit(SCHED_PROFILING, __builtin_return_address(0));
3535
Ingo Molnar2d723762007-10-15 17:00:12 +02003536 schedstat_inc(this_rq(), sched_count);
Ingo Molnarb8efb562007-10-15 17:00:10 +02003537#ifdef CONFIG_SCHEDSTATS
3538 if (unlikely(prev->lock_depth >= 0)) {
Ingo Molnar2d723762007-10-15 17:00:12 +02003539 schedstat_inc(this_rq(), bkl_count);
3540 schedstat_inc(prev, sched_info.bkl_count);
Ingo Molnarb8efb562007-10-15 17:00:10 +02003541 }
3542#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02003543}
3544
Peter Zijlstra6cecd082009-11-30 13:00:37 +01003545static void put_prev_task(struct rq *rq, struct task_struct *prev)
Mike Galbraithdf1c99d2009-03-10 19:08:11 +01003546{
Mike Galbraitha64692a2010-03-11 17:16:20 +01003547 if (prev->se.on_rq)
3548 update_rq_clock(rq);
3549 rq->skip_clock_update = 0;
Peter Zijlstra6cecd082009-11-30 13:00:37 +01003550 prev->sched_class->put_prev_task(rq, prev);
Mike Galbraithdf1c99d2009-03-10 19:08:11 +01003551}
3552
Ingo Molnardd41f592007-07-09 18:51:59 +02003553/*
3554 * Pick up the highest-prio task:
3555 */
3556static inline struct task_struct *
Wang Chenb67802e2009-03-02 13:55:26 +08003557pick_next_task(struct rq *rq)
Ingo Molnardd41f592007-07-09 18:51:59 +02003558{
Ingo Molnar5522d5d2007-10-15 17:00:12 +02003559 const struct sched_class *class;
Ingo Molnardd41f592007-07-09 18:51:59 +02003560 struct task_struct *p;
3561
3562 /*
3563 * Optimization: we know that if all tasks are in
3564 * the fair class we can call that function directly:
3565 */
3566 if (likely(rq->nr_running == rq->cfs.nr_running)) {
Ingo Molnarfb8d4722007-08-09 11:16:48 +02003567 p = fair_sched_class.pick_next_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02003568 if (likely(p))
3569 return p;
3570 }
3571
3572 class = sched_class_highest;
3573 for ( ; ; ) {
Ingo Molnarfb8d4722007-08-09 11:16:48 +02003574 p = class->pick_next_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02003575 if (p)
3576 return p;
3577 /*
3578 * Will never be NULL as the idle class always
3579 * returns a non-NULL p:
3580 */
3581 class = class->next;
3582 }
3583}
3584
3585/*
3586 * schedule() is the main scheduler function.
3587 */
Peter Zijlstraff743342009-03-13 12:21:26 +01003588asmlinkage void __sched schedule(void)
Ingo Molnardd41f592007-07-09 18:51:59 +02003589{
3590 struct task_struct *prev, *next;
Harvey Harrison67ca7bd2008-02-15 09:56:36 -08003591 unsigned long *switch_count;
Ingo Molnardd41f592007-07-09 18:51:59 +02003592 struct rq *rq;
Peter Zijlstra31656512008-07-18 18:01:23 +02003593 int cpu;
Ingo Molnardd41f592007-07-09 18:51:59 +02003594
Peter Zijlstraff743342009-03-13 12:21:26 +01003595need_resched:
3596 preempt_disable();
Ingo Molnardd41f592007-07-09 18:51:59 +02003597 cpu = smp_processor_id();
3598 rq = cpu_rq(cpu);
Paul E. McKenney25502a62010-04-01 17:37:01 -07003599 rcu_note_context_switch(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02003600 prev = rq->curr;
3601 switch_count = &prev->nivcsw;
3602
Linus Torvalds1da177e2005-04-16 15:20:36 -07003603 release_kernel_lock(prev);
3604need_resched_nonpreemptible:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003605
Ingo Molnardd41f592007-07-09 18:51:59 +02003606 schedule_debug(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003607
Peter Zijlstra31656512008-07-18 18:01:23 +02003608 if (sched_feat(HRTICK))
Mike Galbraithf333fdc2008-05-12 21:20:55 +02003609 hrtick_clear(rq);
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003610
Thomas Gleixner05fa7852009-11-17 14:28:38 +01003611 raw_spin_lock_irq(&rq->lock);
Ingo Molnar1e819952007-10-15 17:00:13 +02003612 clear_tsk_need_resched(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003613
Ingo Molnardd41f592007-07-09 18:51:59 +02003614 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
Oleg Nesterov16882c12008-06-08 21:20:41 +04003615 if (unlikely(signal_pending_state(prev->state, prev)))
Ingo Molnardd41f592007-07-09 18:51:59 +02003616 prev->state = TASK_RUNNING;
Oleg Nesterov16882c12008-06-08 21:20:41 +04003617 else
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01003618 deactivate_task(rq, prev, DEQUEUE_SLEEP);
Ingo Molnardd41f592007-07-09 18:51:59 +02003619 switch_count = &prev->nvcsw;
3620 }
3621
Gregory Haskins3f029d32009-07-29 11:08:47 -04003622 pre_schedule(rq, prev);
Steven Rostedtf65eda42008-01-25 21:08:07 +01003623
Ingo Molnardd41f592007-07-09 18:51:59 +02003624 if (unlikely(!rq->nr_running))
3625 idle_balance(cpu, rq);
3626
Mike Galbraithdf1c99d2009-03-10 19:08:11 +01003627 put_prev_task(rq, prev);
Wang Chenb67802e2009-03-02 13:55:26 +08003628 next = pick_next_task(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629
Linus Torvalds1da177e2005-04-16 15:20:36 -07003630 if (likely(prev != next)) {
David Simner673a90a2008-04-29 10:08:59 +01003631 sched_info_switch(prev, next);
Peter Zijlstra49f47432009-12-27 11:51:52 +01003632 perf_event_task_sched_out(prev, next);
David Simner673a90a2008-04-29 10:08:59 +01003633
Linus Torvalds1da177e2005-04-16 15:20:36 -07003634 rq->nr_switches++;
3635 rq->curr = next;
3636 ++*switch_count;
3637
Ingo Molnardd41f592007-07-09 18:51:59 +02003638 context_switch(rq, prev, next); /* unlocks the rq */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003639 /*
3640 * the context switch might have flipped the stack from under
3641 * us, hence refresh the local variables.
3642 */
3643 cpu = smp_processor_id();
3644 rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645 } else
Thomas Gleixner05fa7852009-11-17 14:28:38 +01003646 raw_spin_unlock_irq(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003647
Gregory Haskins3f029d32009-07-29 11:08:47 -04003648 post_schedule(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649
Yong Zhang6d558c32010-01-11 14:21:25 +08003650 if (unlikely(reacquire_kernel_lock(current) < 0)) {
3651 prev = rq->curr;
3652 switch_count = &prev->nivcsw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003653 goto need_resched_nonpreemptible;
Yong Zhang6d558c32010-01-11 14:21:25 +08003654 }
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01003655
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656 preempt_enable_no_resched();
Peter Zijlstraff743342009-03-13 12:21:26 +01003657 if (need_resched())
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658 goto need_resched;
3659}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660EXPORT_SYMBOL(schedule);
3661
Frederic Weisbeckerc08f7822009-12-02 20:49:17 +01003662#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003663/*
3664 * Look out! "owner" is an entirely speculative pointer
3665 * access and not reliable.
3666 */
3667int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner)
3668{
3669 unsigned int cpu;
3670 struct rq *rq;
3671
3672 if (!sched_feat(OWNER_SPIN))
3673 return 0;
3674
3675#ifdef CONFIG_DEBUG_PAGEALLOC
3676 /*
3677 * Need to access the cpu field knowing that
3678 * DEBUG_PAGEALLOC could have unmapped it if
3679 * the mutex owner just released it and exited.
3680 */
3681 if (probe_kernel_address(&owner->cpu, cpu))
Benjamin Herrenschmidt4b402212010-04-16 23:20:00 +02003682 return 0;
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003683#else
3684 cpu = owner->cpu;
3685#endif
3686
3687 /*
3688 * Even if the access succeeded (likely case),
3689 * the cpu field may no longer be valid.
3690 */
3691 if (cpu >= nr_cpumask_bits)
Benjamin Herrenschmidt4b402212010-04-16 23:20:00 +02003692 return 0;
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003693
3694 /*
3695 * We need to validate that we can do a
3696 * get_cpu() and that we have the percpu area.
3697 */
3698 if (!cpu_online(cpu))
Benjamin Herrenschmidt4b402212010-04-16 23:20:00 +02003699 return 0;
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003700
3701 rq = cpu_rq(cpu);
3702
3703 for (;;) {
3704 /*
3705 * Owner changed, break to re-assess state.
3706 */
3707 if (lock->owner != owner)
3708 break;
3709
3710 /*
3711 * Is that owner really running on that cpu?
3712 */
3713 if (task_thread_info(rq->curr) != owner || need_resched())
3714 return 0;
3715
3716 cpu_relax();
3717 }
Benjamin Herrenschmidt4b402212010-04-16 23:20:00 +02003718
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01003719 return 1;
3720}
3721#endif
3722
Linus Torvalds1da177e2005-04-16 15:20:36 -07003723#ifdef CONFIG_PREEMPT
3724/*
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003725 * this is the entry point to schedule() from in-kernel preemption
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003726 * off of preempt_enable. Kernel preemptions off return from interrupt
Linus Torvalds1da177e2005-04-16 15:20:36 -07003727 * occur there and call schedule directly.
3728 */
3729asmlinkage void __sched preempt_schedule(void)
3730{
3731 struct thread_info *ti = current_thread_info();
Ingo Molnar6478d882008-01-25 21:08:33 +01003732
Linus Torvalds1da177e2005-04-16 15:20:36 -07003733 /*
3734 * If there is a non-zero preempt_count or interrupts are disabled,
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003735 * we do not want to preempt the current task. Just return..
Linus Torvalds1da177e2005-04-16 15:20:36 -07003736 */
Nick Pigginbeed33a2006-10-11 01:21:52 -07003737 if (likely(ti->preempt_count || irqs_disabled()))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738 return;
3739
Andi Kleen3a5c3592007-10-15 17:00:14 +02003740 do {
3741 add_preempt_count(PREEMPT_ACTIVE);
Andi Kleen3a5c3592007-10-15 17:00:14 +02003742 schedule();
Andi Kleen3a5c3592007-10-15 17:00:14 +02003743 sub_preempt_count(PREEMPT_ACTIVE);
3744
3745 /*
3746 * Check again in case we missed a preemption opportunity
3747 * between schedule and now.
3748 */
3749 barrier();
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08003750 } while (need_resched());
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003752EXPORT_SYMBOL(preempt_schedule);
3753
3754/*
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003755 * this is the entry point to schedule() from kernel preemption
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756 * off of irq context.
3757 * Note, that this is called and return with irqs disabled. This will
3758 * protect us against recursive calling from irq.
3759 */
3760asmlinkage void __sched preempt_schedule_irq(void)
3761{
3762 struct thread_info *ti = current_thread_info();
Ingo Molnar6478d882008-01-25 21:08:33 +01003763
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003764 /* Catch callers which need to be fixed */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765 BUG_ON(ti->preempt_count || !irqs_disabled());
3766
Andi Kleen3a5c3592007-10-15 17:00:14 +02003767 do {
3768 add_preempt_count(PREEMPT_ACTIVE);
Andi Kleen3a5c3592007-10-15 17:00:14 +02003769 local_irq_enable();
3770 schedule();
3771 local_irq_disable();
Andi Kleen3a5c3592007-10-15 17:00:14 +02003772 sub_preempt_count(PREEMPT_ACTIVE);
3773
3774 /*
3775 * Check again in case we missed a preemption opportunity
3776 * between schedule and now.
3777 */
3778 barrier();
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08003779 } while (need_resched());
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780}
3781
3782#endif /* CONFIG_PREEMPT */
3783
Peter Zijlstra63859d42009-09-15 19:14:42 +02003784int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07003785 void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786{
Peter Zijlstra63859d42009-09-15 19:14:42 +02003787 return try_to_wake_up(curr->private, mode, wake_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789EXPORT_SYMBOL(default_wake_function);
3790
3791/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003792 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
3793 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794 * number) then we wake all the non-exclusive tasks and one exclusive task.
3795 *
3796 * There are circumstances in which we can try to wake a task which has already
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003797 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 * zero in this (rare) case, and we handle it by continuing to scan the queue.
3799 */
Johannes Weiner78ddb082009-04-14 16:53:05 +02003800static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
Peter Zijlstra63859d42009-09-15 19:14:42 +02003801 int nr_exclusive, int wake_flags, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802{
Matthias Kaehlcke2e458742007-10-15 17:00:02 +02003803 wait_queue_t *curr, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804
Matthias Kaehlcke2e458742007-10-15 17:00:02 +02003805 list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
Ingo Molnar48f24c42006-07-03 00:25:40 -07003806 unsigned flags = curr->flags;
3807
Peter Zijlstra63859d42009-09-15 19:14:42 +02003808 if (curr->func(curr, mode, wake_flags, key) &&
Ingo Molnar48f24c42006-07-03 00:25:40 -07003809 (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003810 break;
3811 }
3812}
3813
3814/**
3815 * __wake_up - wake up threads blocked on a waitqueue.
3816 * @q: the waitqueue
3817 * @mode: which threads
3818 * @nr_exclusive: how many wake-one or wake-many threads to wake up
Martin Waitz67be2dd2005-05-01 08:59:26 -07003819 * @key: is directly passed to the wakeup function
David Howells50fa6102009-04-28 15:01:38 +01003820 *
3821 * It may be assumed that this function implies a write memory barrier before
3822 * changing the task state if and only if any tasks are woken up.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08003824void __wake_up(wait_queue_head_t *q, unsigned int mode,
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07003825 int nr_exclusive, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003826{
3827 unsigned long flags;
3828
3829 spin_lock_irqsave(&q->lock, flags);
3830 __wake_up_common(q, mode, nr_exclusive, 0, key);
3831 spin_unlock_irqrestore(&q->lock, flags);
3832}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003833EXPORT_SYMBOL(__wake_up);
3834
3835/*
3836 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
3837 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08003838void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839{
3840 __wake_up_common(q, mode, 1, 0, NULL);
3841}
Michal Nazarewicz22c43c82010-05-05 12:53:11 +02003842EXPORT_SYMBOL_GPL(__wake_up_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003843
Davide Libenzi4ede8162009-03-31 15:24:20 -07003844void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
3845{
3846 __wake_up_common(q, mode, 1, 0, key);
3847}
3848
Linus Torvalds1da177e2005-04-16 15:20:36 -07003849/**
Davide Libenzi4ede8162009-03-31 15:24:20 -07003850 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003851 * @q: the waitqueue
3852 * @mode: which threads
3853 * @nr_exclusive: how many wake-one or wake-many threads to wake up
Davide Libenzi4ede8162009-03-31 15:24:20 -07003854 * @key: opaque value to be passed to wakeup targets
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855 *
3856 * The sync wakeup differs that the waker knows that it will schedule
3857 * away soon, so while the target thread will be woken up, it will not
3858 * be migrated to another CPU - ie. the two threads are 'synchronized'
3859 * with each other. This can prevent needless bouncing between CPUs.
3860 *
3861 * On UP it can prevent extra preemption.
David Howells50fa6102009-04-28 15:01:38 +01003862 *
3863 * It may be assumed that this function implies a write memory barrier before
3864 * changing the task state if and only if any tasks are woken up.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003865 */
Davide Libenzi4ede8162009-03-31 15:24:20 -07003866void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
3867 int nr_exclusive, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868{
3869 unsigned long flags;
Peter Zijlstra7d478722009-09-14 19:55:44 +02003870 int wake_flags = WF_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003871
3872 if (unlikely(!q))
3873 return;
3874
3875 if (unlikely(!nr_exclusive))
Peter Zijlstra7d478722009-09-14 19:55:44 +02003876 wake_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003877
3878 spin_lock_irqsave(&q->lock, flags);
Peter Zijlstra7d478722009-09-14 19:55:44 +02003879 __wake_up_common(q, mode, nr_exclusive, wake_flags, key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880 spin_unlock_irqrestore(&q->lock, flags);
3881}
Davide Libenzi4ede8162009-03-31 15:24:20 -07003882EXPORT_SYMBOL_GPL(__wake_up_sync_key);
3883
3884/*
3885 * __wake_up_sync - see __wake_up_sync_key()
3886 */
3887void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
3888{
3889 __wake_up_sync_key(q, mode, nr_exclusive, NULL);
3890}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
3892
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003893/**
3894 * complete: - signals a single thread waiting on this completion
3895 * @x: holds the state of this particular completion
3896 *
3897 * This will wake up a single thread waiting on this completion. Threads will be
3898 * awakened in the same order in which they were queued.
3899 *
3900 * See also complete_all(), wait_for_completion() and related routines.
David Howells50fa6102009-04-28 15:01:38 +01003901 *
3902 * It may be assumed that this function implies a write memory barrier before
3903 * changing the task state if and only if any tasks are woken up.
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003904 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02003905void complete(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003906{
3907 unsigned long flags;
3908
3909 spin_lock_irqsave(&x->wait.lock, flags);
3910 x->done++;
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05003911 __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912 spin_unlock_irqrestore(&x->wait.lock, flags);
3913}
3914EXPORT_SYMBOL(complete);
3915
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003916/**
3917 * complete_all: - signals all threads waiting on this completion
3918 * @x: holds the state of this particular completion
3919 *
3920 * This will wake up all threads waiting on this particular completion event.
David Howells50fa6102009-04-28 15:01:38 +01003921 *
3922 * It may be assumed that this function implies a write memory barrier before
3923 * changing the task state if and only if any tasks are woken up.
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003924 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02003925void complete_all(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003926{
3927 unsigned long flags;
3928
3929 spin_lock_irqsave(&x->wait.lock, flags);
3930 x->done += UINT_MAX/2;
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05003931 __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932 spin_unlock_irqrestore(&x->wait.lock, flags);
3933}
3934EXPORT_SYMBOL(complete_all);
3935
Andi Kleen8cbbe862007-10-15 17:00:14 +02003936static inline long __sched
3937do_wait_for_common(struct completion *x, long timeout, int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003938{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003939 if (!x->done) {
3940 DECLARE_WAITQUEUE(wait, current);
3941
Changli Gaoa93d2f12010-05-07 14:33:26 +08003942 __add_wait_queue_tail_exclusive(&x->wait, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003943 do {
Oleg Nesterov94d3d822008-08-20 16:54:41 -07003944 if (signal_pending_state(state, current)) {
Oleg Nesterovea71a542008-06-20 18:32:20 +04003945 timeout = -ERESTARTSYS;
3946 break;
Andi Kleen8cbbe862007-10-15 17:00:14 +02003947 }
3948 __set_current_state(state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949 spin_unlock_irq(&x->wait.lock);
Andi Kleen8cbbe862007-10-15 17:00:14 +02003950 timeout = schedule_timeout(timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003951 spin_lock_irq(&x->wait.lock);
Oleg Nesterovea71a542008-06-20 18:32:20 +04003952 } while (!x->done && timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003953 __remove_wait_queue(&x->wait, &wait);
Oleg Nesterovea71a542008-06-20 18:32:20 +04003954 if (!x->done)
3955 return timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003956 }
3957 x->done--;
Oleg Nesterovea71a542008-06-20 18:32:20 +04003958 return timeout ?: 1;
Andi Kleen8cbbe862007-10-15 17:00:14 +02003959}
3960
3961static long __sched
3962wait_for_common(struct completion *x, long timeout, int state)
3963{
3964 might_sleep();
3965
3966 spin_lock_irq(&x->wait.lock);
3967 timeout = do_wait_for_common(x, timeout, state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968 spin_unlock_irq(&x->wait.lock);
Andi Kleen8cbbe862007-10-15 17:00:14 +02003969 return timeout;
3970}
3971
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003972/**
3973 * wait_for_completion: - waits for completion of a task
3974 * @x: holds the state of this particular completion
3975 *
3976 * This waits to be signaled for completion of a specific task. It is NOT
3977 * interruptible and there is no timeout.
3978 *
3979 * See also similar routines (i.e. wait_for_completion_timeout()) with timeout
3980 * and interrupt capability. Also see complete().
3981 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02003982void __sched wait_for_completion(struct completion *x)
Andi Kleen8cbbe862007-10-15 17:00:14 +02003983{
3984 wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003985}
3986EXPORT_SYMBOL(wait_for_completion);
3987
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003988/**
3989 * wait_for_completion_timeout: - waits for completion of a task (w/timeout)
3990 * @x: holds the state of this particular completion
3991 * @timeout: timeout value in jiffies
3992 *
3993 * This waits for either a completion of a specific task to be signaled or for a
3994 * specified timeout to expire. The timeout is in jiffies. It is not
3995 * interruptible.
3996 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02003997unsigned long __sched
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998wait_for_completion_timeout(struct completion *x, unsigned long timeout)
3999{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004000 return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001}
4002EXPORT_SYMBOL(wait_for_completion_timeout);
4003
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004004/**
4005 * wait_for_completion_interruptible: - waits for completion of a task (w/intr)
4006 * @x: holds the state of this particular completion
4007 *
4008 * This waits for completion of a specific task to be signaled. It is
4009 * interruptible.
4010 */
Andi Kleen8cbbe862007-10-15 17:00:14 +02004011int __sched wait_for_completion_interruptible(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004012{
Andi Kleen51e97992007-10-18 21:32:55 +02004013 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
4014 if (t == -ERESTARTSYS)
4015 return t;
4016 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004017}
4018EXPORT_SYMBOL(wait_for_completion_interruptible);
4019
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004020/**
4021 * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr))
4022 * @x: holds the state of this particular completion
4023 * @timeout: timeout value in jiffies
4024 *
4025 * This waits for either a completion of a specific task to be signaled or for a
4026 * specified timeout to expire. It is interruptible. The timeout is in jiffies.
4027 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02004028unsigned long __sched
Linus Torvalds1da177e2005-04-16 15:20:36 -07004029wait_for_completion_interruptible_timeout(struct completion *x,
4030 unsigned long timeout)
4031{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004032 return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004033}
4034EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
4035
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02004036/**
4037 * wait_for_completion_killable: - waits for completion of a task (killable)
4038 * @x: holds the state of this particular completion
4039 *
4040 * This waits to be signaled for completion of a specific task. It can be
4041 * interrupted by a kill signal.
4042 */
Matthew Wilcox009e5772007-12-06 12:29:54 -05004043int __sched wait_for_completion_killable(struct completion *x)
4044{
4045 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
4046 if (t == -ERESTARTSYS)
4047 return t;
4048 return 0;
4049}
4050EXPORT_SYMBOL(wait_for_completion_killable);
4051
Dave Chinnerbe4de352008-08-15 00:40:44 -07004052/**
Sage Weil0aa12fb2010-05-29 09:12:30 -07004053 * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable))
4054 * @x: holds the state of this particular completion
4055 * @timeout: timeout value in jiffies
4056 *
4057 * This waits for either a completion of a specific task to be
4058 * signaled or for a specified timeout to expire. It can be
4059 * interrupted by a kill signal. The timeout is in jiffies.
4060 */
4061unsigned long __sched
4062wait_for_completion_killable_timeout(struct completion *x,
4063 unsigned long timeout)
4064{
4065 return wait_for_common(x, timeout, TASK_KILLABLE);
4066}
4067EXPORT_SYMBOL(wait_for_completion_killable_timeout);
4068
4069/**
Dave Chinnerbe4de352008-08-15 00:40:44 -07004070 * try_wait_for_completion - try to decrement a completion without blocking
4071 * @x: completion structure
4072 *
4073 * Returns: 0 if a decrement cannot be done without blocking
4074 * 1 if a decrement succeeded.
4075 *
4076 * If a completion is being used as a counting completion,
4077 * attempt to decrement the counter without blocking. This
4078 * enables us to avoid waiting if the resource the completion
4079 * is protecting is not available.
4080 */
4081bool try_wait_for_completion(struct completion *x)
4082{
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004083 unsigned long flags;
Dave Chinnerbe4de352008-08-15 00:40:44 -07004084 int ret = 1;
4085
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004086 spin_lock_irqsave(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07004087 if (!x->done)
4088 ret = 0;
4089 else
4090 x->done--;
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004091 spin_unlock_irqrestore(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07004092 return ret;
4093}
4094EXPORT_SYMBOL(try_wait_for_completion);
4095
4096/**
4097 * completion_done - Test to see if a completion has any waiters
4098 * @x: completion structure
4099 *
4100 * Returns: 0 if there are waiters (wait_for_completion() in progress)
4101 * 1 if there are no waiters.
4102 *
4103 */
4104bool completion_done(struct completion *x)
4105{
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004106 unsigned long flags;
Dave Chinnerbe4de352008-08-15 00:40:44 -07004107 int ret = 1;
4108
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004109 spin_lock_irqsave(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07004110 if (!x->done)
4111 ret = 0;
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01004112 spin_unlock_irqrestore(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07004113 return ret;
4114}
4115EXPORT_SYMBOL(completion_done);
4116
Andi Kleen8cbbe862007-10-15 17:00:14 +02004117static long __sched
4118sleep_on_common(wait_queue_head_t *q, int state, long timeout)
Ingo Molnar0fec1712007-07-09 18:52:01 +02004119{
4120 unsigned long flags;
4121 wait_queue_t wait;
4122
4123 init_waitqueue_entry(&wait, current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004124
Andi Kleen8cbbe862007-10-15 17:00:14 +02004125 __set_current_state(state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004126
Andi Kleen8cbbe862007-10-15 17:00:14 +02004127 spin_lock_irqsave(&q->lock, flags);
4128 __add_wait_queue(q, &wait);
4129 spin_unlock(&q->lock);
4130 timeout = schedule_timeout(timeout);
4131 spin_lock_irq(&q->lock);
4132 __remove_wait_queue(q, &wait);
4133 spin_unlock_irqrestore(&q->lock, flags);
4134
4135 return timeout;
4136}
4137
4138void __sched interruptible_sleep_on(wait_queue_head_t *q)
4139{
4140 sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004141}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004142EXPORT_SYMBOL(interruptible_sleep_on);
4143
Ingo Molnar0fec1712007-07-09 18:52:01 +02004144long __sched
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004145interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004146{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004147 return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004148}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004149EXPORT_SYMBOL(interruptible_sleep_on_timeout);
4150
Ingo Molnar0fec1712007-07-09 18:52:01 +02004151void __sched sleep_on(wait_queue_head_t *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004152{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004153 sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004154}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004155EXPORT_SYMBOL(sleep_on);
4156
Ingo Molnar0fec1712007-07-09 18:52:01 +02004157long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158{
Andi Kleen8cbbe862007-10-15 17:00:14 +02004159 return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004160}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004161EXPORT_SYMBOL(sleep_on_timeout);
4162
Ingo Molnarb29739f2006-06-27 02:54:51 -07004163#ifdef CONFIG_RT_MUTEXES
4164
4165/*
4166 * rt_mutex_setprio - set the current priority of a task
4167 * @p: task
4168 * @prio: prio value (kernel-internal form)
4169 *
4170 * This function changes the 'effective' priority of a task. It does
4171 * not touch ->normal_prio like __setscheduler().
4172 *
4173 * Used by the rt_mutex code to implement priority inheritance logic.
4174 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004175void rt_mutex_setprio(struct task_struct *p, int prio)
Ingo Molnarb29739f2006-06-27 02:54:51 -07004176{
4177 unsigned long flags;
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004178 int oldprio, on_rq, running;
Ingo Molnar70b97a72006-07-03 00:25:42 -07004179 struct rq *rq;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01004180 const struct sched_class *prev_class;
Ingo Molnarb29739f2006-06-27 02:54:51 -07004181
4182 BUG_ON(prio < 0 || prio > MAX_PRIO);
4183
4184 rq = task_rq_lock(p, &flags);
4185
Andrew Mortond5f9f942007-05-08 20:27:06 -07004186 oldprio = p->prio;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01004187 prev_class = p->sched_class;
Ingo Molnardd41f592007-07-09 18:51:59 +02004188 on_rq = p->se.on_rq;
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01004189 running = task_current(rq, p);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004190 if (on_rq)
Ingo Molnar69be72c2007-08-09 11:16:49 +02004191 dequeue_task(rq, p, 0);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004192 if (running)
4193 p->sched_class->put_prev_task(rq, p);
Ingo Molnardd41f592007-07-09 18:51:59 +02004194
4195 if (rt_prio(prio))
4196 p->sched_class = &rt_sched_class;
4197 else
4198 p->sched_class = &fair_sched_class;
4199
Ingo Molnarb29739f2006-06-27 02:54:51 -07004200 p->prio = prio;
4201
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004202 if (running)
4203 p->sched_class->set_curr_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02004204 if (on_rq) {
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01004205 enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0);
Steven Rostedtcb469842008-01-25 21:08:22 +01004206
4207 check_class_changed(rq, p, prev_class, oldprio, running);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004208 }
4209 task_rq_unlock(rq, &flags);
4210}
4211
4212#endif
4213
Ingo Molnar36c8b582006-07-03 00:25:41 -07004214void set_user_nice(struct task_struct *p, long nice)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004215{
Ingo Molnardd41f592007-07-09 18:51:59 +02004216 int old_prio, delta, on_rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004217 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07004218 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004219
4220 if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
4221 return;
4222 /*
4223 * We have to be careful, if called from sys_setpriority(),
4224 * the task might be in the middle of scheduling on another CPU.
4225 */
4226 rq = task_rq_lock(p, &flags);
4227 /*
4228 * The RT priorities are set via sched_setscheduler(), but we still
4229 * allow the 'normal' nice value to be set - but as expected
4230 * it wont have any effect on scheduling until the task is
Ingo Molnardd41f592007-07-09 18:51:59 +02004231 * SCHED_FIFO/SCHED_RR:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004232 */
Ingo Molnare05606d2007-07-09 18:51:59 +02004233 if (task_has_rt_policy(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004234 p->static_prio = NICE_TO_PRIO(nice);
4235 goto out_unlock;
4236 }
Ingo Molnardd41f592007-07-09 18:51:59 +02004237 on_rq = p->se.on_rq;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02004238 if (on_rq)
Ingo Molnar69be72c2007-08-09 11:16:49 +02004239 dequeue_task(rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004240
Linus Torvalds1da177e2005-04-16 15:20:36 -07004241 p->static_prio = NICE_TO_PRIO(nice);
Peter Williams2dd73a42006-06-27 02:54:34 -07004242 set_load_weight(p);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004243 old_prio = p->prio;
4244 p->prio = effective_prio(p);
4245 delta = p->prio - old_prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004246
Ingo Molnardd41f592007-07-09 18:51:59 +02004247 if (on_rq) {
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01004248 enqueue_task(rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004249 /*
Andrew Mortond5f9f942007-05-08 20:27:06 -07004250 * If the task increased its priority or is running and
4251 * lowered its priority, then reschedule its CPU:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004252 */
Andrew Mortond5f9f942007-05-08 20:27:06 -07004253 if (delta < 0 || (delta > 0 && task_running(rq, p)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004254 resched_task(rq->curr);
4255 }
4256out_unlock:
4257 task_rq_unlock(rq, &flags);
4258}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004259EXPORT_SYMBOL(set_user_nice);
4260
Matt Mackalle43379f2005-05-01 08:59:00 -07004261/*
4262 * can_nice - check if a task can reduce its nice value
4263 * @p: task
4264 * @nice: nice value
4265 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004266int can_nice(const struct task_struct *p, const int nice)
Matt Mackalle43379f2005-05-01 08:59:00 -07004267{
Matt Mackall024f4742005-08-18 11:24:19 -07004268 /* convert nice value [19,-20] to rlimit style value [1,40] */
4269 int nice_rlim = 20 - nice;
Ingo Molnar48f24c42006-07-03 00:25:40 -07004270
Jiri Slaby78d7d402010-03-05 13:42:54 -08004271 return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
Matt Mackalle43379f2005-05-01 08:59:00 -07004272 capable(CAP_SYS_NICE));
4273}
4274
Linus Torvalds1da177e2005-04-16 15:20:36 -07004275#ifdef __ARCH_WANT_SYS_NICE
4276
4277/*
4278 * sys_nice - change the priority of the current process.
4279 * @increment: priority increment
4280 *
4281 * sys_setpriority is a more generic, but much slower function that
4282 * does similar things.
4283 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004284SYSCALL_DEFINE1(nice, int, increment)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004285{
Ingo Molnar48f24c42006-07-03 00:25:40 -07004286 long nice, retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004287
4288 /*
4289 * Setpriority might change our priority at the same moment.
4290 * We don't have to worry. Conceptually one call occurs first
4291 * and we have a single winner.
4292 */
Matt Mackalle43379f2005-05-01 08:59:00 -07004293 if (increment < -40)
4294 increment = -40;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004295 if (increment > 40)
4296 increment = 40;
4297
Américo Wang2b8f8362009-02-16 18:54:21 +08004298 nice = TASK_NICE(current) + increment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004299 if (nice < -20)
4300 nice = -20;
4301 if (nice > 19)
4302 nice = 19;
4303
Matt Mackalle43379f2005-05-01 08:59:00 -07004304 if (increment < 0 && !can_nice(current, nice))
4305 return -EPERM;
4306
Linus Torvalds1da177e2005-04-16 15:20:36 -07004307 retval = security_task_setnice(current, nice);
4308 if (retval)
4309 return retval;
4310
4311 set_user_nice(current, nice);
4312 return 0;
4313}
4314
4315#endif
4316
4317/**
4318 * task_prio - return the priority value of a given task.
4319 * @p: the task in question.
4320 *
4321 * This is the priority value as seen by users in /proc.
4322 * RT tasks are offset by -200. Normal tasks are centered
4323 * around 0, value goes from -16 to +15.
4324 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004325int task_prio(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004326{
4327 return p->prio - MAX_RT_PRIO;
4328}
4329
4330/**
4331 * task_nice - return the nice value of a given task.
4332 * @p: the task in question.
4333 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004334int task_nice(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004335{
4336 return TASK_NICE(p);
4337}
Pavel Roskin150d8be2008-03-05 16:56:37 -05004338EXPORT_SYMBOL(task_nice);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004339
4340/**
4341 * idle_cpu - is a given cpu idle currently?
4342 * @cpu: the processor in question.
4343 */
4344int idle_cpu(int cpu)
4345{
4346 return cpu_curr(cpu) == cpu_rq(cpu)->idle;
4347}
4348
Linus Torvalds1da177e2005-04-16 15:20:36 -07004349/**
4350 * idle_task - return the idle task for a given cpu.
4351 * @cpu: the processor in question.
4352 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07004353struct task_struct *idle_task(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004354{
4355 return cpu_rq(cpu)->idle;
4356}
4357
4358/**
4359 * find_process_by_pid - find a process with a matching PID value.
4360 * @pid: the pid in question.
4361 */
Alexey Dobriyana9957442007-10-15 17:00:13 +02004362static struct task_struct *find_process_by_pid(pid_t pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004363{
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07004364 return pid ? find_task_by_vpid(pid) : current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004365}
4366
4367/* Actually do priority change: must hold rq lock. */
Ingo Molnardd41f592007-07-09 18:51:59 +02004368static void
4369__setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004370{
Ingo Molnardd41f592007-07-09 18:51:59 +02004371 BUG_ON(p->se.on_rq);
Ingo Molnar48f24c42006-07-03 00:25:40 -07004372
Linus Torvalds1da177e2005-04-16 15:20:36 -07004373 p->policy = policy;
4374 p->rt_priority = prio;
Ingo Molnarb29739f2006-06-27 02:54:51 -07004375 p->normal_prio = normal_prio(p);
4376 /* we are holding p->pi_lock already */
4377 p->prio = rt_mutex_getprio(p);
Peter Zijlstraffd44db2009-11-10 20:12:01 +01004378 if (rt_prio(p->prio))
4379 p->sched_class = &rt_sched_class;
4380 else
4381 p->sched_class = &fair_sched_class;
Peter Williams2dd73a42006-06-27 02:54:34 -07004382 set_load_weight(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004383}
4384
David Howellsc69e8d92008-11-14 10:39:19 +11004385/*
4386 * check the target process has a UID that matches the current process's
4387 */
4388static bool check_same_owner(struct task_struct *p)
4389{
4390 const struct cred *cred = current_cred(), *pcred;
4391 bool match;
4392
4393 rcu_read_lock();
4394 pcred = __task_cred(p);
4395 match = (cred->euid == pcred->euid ||
4396 cred->euid == pcred->uid);
4397 rcu_read_unlock();
4398 return match;
4399}
4400
Rusty Russell961ccdd2008-06-23 13:55:38 +10004401static int __sched_setscheduler(struct task_struct *p, int policy,
4402 struct sched_param *param, bool user)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004403{
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02004404 int retval, oldprio, oldpolicy = -1, on_rq, running;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004405 unsigned long flags;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01004406 const struct sched_class *prev_class;
Ingo Molnar70b97a72006-07-03 00:25:42 -07004407 struct rq *rq;
Lennart Poetteringca94c442009-06-15 17:17:47 +02004408 int reset_on_fork;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004409
Steven Rostedt66e53932006-06-27 02:54:44 -07004410 /* may grab non-irq protected spin_locks */
4411 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07004412recheck:
4413 /* double check policy once rq lock held */
Lennart Poetteringca94c442009-06-15 17:17:47 +02004414 if (policy < 0) {
4415 reset_on_fork = p->sched_reset_on_fork;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004416 policy = oldpolicy = p->policy;
Lennart Poetteringca94c442009-06-15 17:17:47 +02004417 } else {
4418 reset_on_fork = !!(policy & SCHED_RESET_ON_FORK);
4419 policy &= ~SCHED_RESET_ON_FORK;
4420
4421 if (policy != SCHED_FIFO && policy != SCHED_RR &&
4422 policy != SCHED_NORMAL && policy != SCHED_BATCH &&
4423 policy != SCHED_IDLE)
4424 return -EINVAL;
4425 }
4426
Linus Torvalds1da177e2005-04-16 15:20:36 -07004427 /*
4428 * Valid priorities for SCHED_FIFO and SCHED_RR are
Ingo Molnardd41f592007-07-09 18:51:59 +02004429 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
4430 * SCHED_BATCH and SCHED_IDLE is 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004431 */
4432 if (param->sched_priority < 0 ||
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004433 (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
Steven Rostedtd46523e2005-07-25 16:28:39 -04004434 (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004435 return -EINVAL;
Ingo Molnare05606d2007-07-09 18:51:59 +02004436 if (rt_policy(policy) != (param->sched_priority != 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004437 return -EINVAL;
4438
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004439 /*
4440 * Allow unprivileged RT tasks to decrease priority:
4441 */
Rusty Russell961ccdd2008-06-23 13:55:38 +10004442 if (user && !capable(CAP_SYS_NICE)) {
Ingo Molnare05606d2007-07-09 18:51:59 +02004443 if (rt_policy(policy)) {
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07004444 unsigned long rlim_rtprio;
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004445
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07004446 if (!lock_task_sighand(p, &flags))
4447 return -ESRCH;
Jiri Slaby78d7d402010-03-05 13:42:54 -08004448 rlim_rtprio = task_rlimit(p, RLIMIT_RTPRIO);
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07004449 unlock_task_sighand(p, &flags);
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004450
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07004451 /* can't set/change the rt policy */
4452 if (policy != p->policy && !rlim_rtprio)
4453 return -EPERM;
4454
4455 /* can't increase priority */
4456 if (param->sched_priority > p->rt_priority &&
4457 param->sched_priority > rlim_rtprio)
4458 return -EPERM;
4459 }
Ingo Molnardd41f592007-07-09 18:51:59 +02004460 /*
4461 * Like positive nice levels, dont allow tasks to
4462 * move out of SCHED_IDLE either:
4463 */
4464 if (p->policy == SCHED_IDLE && policy != SCHED_IDLE)
4465 return -EPERM;
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07004466
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004467 /* can't change other user's priorities */
David Howellsc69e8d92008-11-14 10:39:19 +11004468 if (!check_same_owner(p))
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004469 return -EPERM;
Lennart Poetteringca94c442009-06-15 17:17:47 +02004470
4471 /* Normal users shall not reset the sched_reset_on_fork flag */
4472 if (p->sched_reset_on_fork && !reset_on_fork)
4473 return -EPERM;
Olivier Croquette37e4ab32005-06-25 14:57:32 -07004474 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004475
Jeremy Fitzhardinge725aad22008-08-03 09:33:03 -07004476 if (user) {
Jeremy Fitzhardinge725aad22008-08-03 09:33:03 -07004477 retval = security_task_setscheduler(p, policy, param);
4478 if (retval)
4479 return retval;
4480 }
4481
Linus Torvalds1da177e2005-04-16 15:20:36 -07004482 /*
Ingo Molnarb29739f2006-06-27 02:54:51 -07004483 * make sure no PI-waiters arrive (or leave) while we are
4484 * changing the priority of the task:
4485 */
Thomas Gleixner1d615482009-11-17 14:54:03 +01004486 raw_spin_lock_irqsave(&p->pi_lock, flags);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004487 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004488 * To be able to change p->policy safely, the apropriate
4489 * runqueue lock must be held.
4490 */
Ingo Molnarb29739f2006-06-27 02:54:51 -07004491 rq = __task_rq_lock(p);
Peter Zijlstradc61b1d2010-06-08 11:40:42 +02004492
4493#ifdef CONFIG_RT_GROUP_SCHED
4494 if (user) {
4495 /*
4496 * Do not allow realtime tasks into groups that have no runtime
4497 * assigned.
4498 */
4499 if (rt_bandwidth_enabled() && rt_policy(policy) &&
4500 task_group(p)->rt_bandwidth.rt_runtime == 0) {
4501 __task_rq_unlock(rq);
4502 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4503 return -EPERM;
4504 }
4505 }
4506#endif
4507
Linus Torvalds1da177e2005-04-16 15:20:36 -07004508 /* recheck policy now with rq lock held */
4509 if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
4510 policy = oldpolicy = -1;
Ingo Molnarb29739f2006-06-27 02:54:51 -07004511 __task_rq_unlock(rq);
Thomas Gleixner1d615482009-11-17 14:54:03 +01004512 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004513 goto recheck;
4514 }
Ingo Molnardd41f592007-07-09 18:51:59 +02004515 on_rq = p->se.on_rq;
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01004516 running = task_current(rq, p);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004517 if (on_rq)
Ingo Molnar2e1cb742007-08-09 11:16:49 +02004518 deactivate_task(rq, p, 0);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004519 if (running)
4520 p->sched_class->put_prev_task(rq, p);
Dmitry Adamushkof6b53202007-10-15 17:00:08 +02004521
Lennart Poetteringca94c442009-06-15 17:17:47 +02004522 p->sched_reset_on_fork = reset_on_fork;
4523
Linus Torvalds1da177e2005-04-16 15:20:36 -07004524 oldprio = p->prio;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01004525 prev_class = p->sched_class;
Ingo Molnardd41f592007-07-09 18:51:59 +02004526 __setscheduler(rq, p, policy, param->sched_priority);
Dmitry Adamushkof6b53202007-10-15 17:00:08 +02004527
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07004528 if (running)
4529 p->sched_class->set_curr_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02004530 if (on_rq) {
4531 activate_task(rq, p, 0);
Steven Rostedtcb469842008-01-25 21:08:22 +01004532
4533 check_class_changed(rq, p, prev_class, oldprio, running);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004534 }
Ingo Molnarb29739f2006-06-27 02:54:51 -07004535 __task_rq_unlock(rq);
Thomas Gleixner1d615482009-11-17 14:54:03 +01004536 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
Ingo Molnarb29739f2006-06-27 02:54:51 -07004537
Thomas Gleixner95e02ca2006-06-27 02:55:02 -07004538 rt_mutex_adjust_pi(p);
4539
Linus Torvalds1da177e2005-04-16 15:20:36 -07004540 return 0;
4541}
Rusty Russell961ccdd2008-06-23 13:55:38 +10004542
4543/**
4544 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
4545 * @p: the task in question.
4546 * @policy: new policy.
4547 * @param: structure containing the new RT priority.
4548 *
4549 * NOTE that the task may be already dead.
4550 */
4551int sched_setscheduler(struct task_struct *p, int policy,
4552 struct sched_param *param)
4553{
4554 return __sched_setscheduler(p, policy, param, true);
4555}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004556EXPORT_SYMBOL_GPL(sched_setscheduler);
4557
Rusty Russell961ccdd2008-06-23 13:55:38 +10004558/**
4559 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
4560 * @p: the task in question.
4561 * @policy: new policy.
4562 * @param: structure containing the new RT priority.
4563 *
4564 * Just like sched_setscheduler, only don't bother checking if the
4565 * current context has permission. For example, this is needed in
4566 * stop_machine(): we create temporary high priority worker threads,
4567 * but our caller might not have that capability.
4568 */
4569int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4570 struct sched_param *param)
4571{
4572 return __sched_setscheduler(p, policy, param, false);
4573}
4574
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07004575static int
4576do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004577{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578 struct sched_param lparam;
4579 struct task_struct *p;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004580 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004581
4582 if (!param || pid < 0)
4583 return -EINVAL;
4584 if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
4585 return -EFAULT;
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004586
4587 rcu_read_lock();
4588 retval = -ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004589 p = find_process_by_pid(pid);
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07004590 if (p != NULL)
4591 retval = sched_setscheduler(p, policy, &lparam);
4592 rcu_read_unlock();
Ingo Molnar36c8b582006-07-03 00:25:41 -07004593
Linus Torvalds1da177e2005-04-16 15:20:36 -07004594 return retval;
4595}
4596
4597/**
4598 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
4599 * @pid: the pid in question.
4600 * @policy: new policy.
4601 * @param: structure containing the new RT priority.
4602 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004603SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
4604 struct sched_param __user *, param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004605{
Jason Baronc21761f2006-01-18 17:43:03 -08004606 /* negative values for policy are not valid */
4607 if (policy < 0)
4608 return -EINVAL;
4609
Linus Torvalds1da177e2005-04-16 15:20:36 -07004610 return do_sched_setscheduler(pid, policy, param);
4611}
4612
4613/**
4614 * sys_sched_setparam - set/change the RT priority of a thread
4615 * @pid: the pid in question.
4616 * @param: structure containing the new RT priority.
4617 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004618SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004619{
4620 return do_sched_setscheduler(pid, -1, param);
4621}
4622
4623/**
4624 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
4625 * @pid: the pid in question.
4626 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004627SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004628{
Ingo Molnar36c8b582006-07-03 00:25:41 -07004629 struct task_struct *p;
Andi Kleen3a5c3592007-10-15 17:00:14 +02004630 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004631
4632 if (pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02004633 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004634
4635 retval = -ESRCH;
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004636 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004637 p = find_process_by_pid(pid);
4638 if (p) {
4639 retval = security_task_getscheduler(p);
4640 if (!retval)
Lennart Poetteringca94c442009-06-15 17:17:47 +02004641 retval = p->policy
4642 | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004643 }
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004644 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004645 return retval;
4646}
4647
4648/**
Lennart Poetteringca94c442009-06-15 17:17:47 +02004649 * sys_sched_getparam - get the RT priority of a thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07004650 * @pid: the pid in question.
4651 * @param: structure containing the RT priority.
4652 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004653SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004654{
4655 struct sched_param lp;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004656 struct task_struct *p;
Andi Kleen3a5c3592007-10-15 17:00:14 +02004657 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004658
4659 if (!param || pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02004660 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004661
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004662 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004663 p = find_process_by_pid(pid);
4664 retval = -ESRCH;
4665 if (!p)
4666 goto out_unlock;
4667
4668 retval = security_task_getscheduler(p);
4669 if (retval)
4670 goto out_unlock;
4671
4672 lp.sched_priority = p->rt_priority;
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004673 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004674
4675 /*
4676 * This one might sleep, we cannot do it with a spinlock held ...
4677 */
4678 retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
4679
Linus Torvalds1da177e2005-04-16 15:20:36 -07004680 return retval;
4681
4682out_unlock:
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004683 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004684 return retval;
4685}
4686
Rusty Russell96f874e2008-11-25 02:35:14 +10304687long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004688{
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304689 cpumask_var_t cpus_allowed, new_mask;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004690 struct task_struct *p;
4691 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004692
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004693 get_online_cpus();
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004694 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004695
4696 p = find_process_by_pid(pid);
4697 if (!p) {
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004698 rcu_read_unlock();
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004699 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004700 return -ESRCH;
4701 }
4702
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004703 /* Prevent p going away */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004704 get_task_struct(p);
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004705 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004706
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304707 if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
4708 retval = -ENOMEM;
4709 goto out_put_task;
4710 }
4711 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
4712 retval = -ENOMEM;
4713 goto out_free_cpus_allowed;
4714 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004715 retval = -EPERM;
David Howellsc69e8d92008-11-14 10:39:19 +11004716 if (!check_same_owner(p) && !capable(CAP_SYS_NICE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004717 goto out_unlock;
4718
David Quigleye7834f82006-06-23 02:03:59 -07004719 retval = security_task_setscheduler(p, 0, NULL);
4720 if (retval)
4721 goto out_unlock;
4722
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304723 cpuset_cpus_allowed(p, cpus_allowed);
4724 cpumask_and(new_mask, in_mask, cpus_allowed);
Paul Menage8707d8b2007-10-18 23:40:22 -07004725 again:
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304726 retval = set_cpus_allowed_ptr(p, new_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004727
Paul Menage8707d8b2007-10-18 23:40:22 -07004728 if (!retval) {
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304729 cpuset_cpus_allowed(p, cpus_allowed);
4730 if (!cpumask_subset(new_mask, cpus_allowed)) {
Paul Menage8707d8b2007-10-18 23:40:22 -07004731 /*
4732 * We must have raced with a concurrent cpuset
4733 * update. Just reset the cpus_allowed to the
4734 * cpuset's cpus_allowed
4735 */
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304736 cpumask_copy(new_mask, cpus_allowed);
Paul Menage8707d8b2007-10-18 23:40:22 -07004737 goto again;
4738 }
4739 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004740out_unlock:
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304741 free_cpumask_var(new_mask);
4742out_free_cpus_allowed:
4743 free_cpumask_var(cpus_allowed);
4744out_put_task:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004745 put_task_struct(p);
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004746 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004747 return retval;
4748}
4749
4750static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
Rusty Russell96f874e2008-11-25 02:35:14 +10304751 struct cpumask *new_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004752{
Rusty Russell96f874e2008-11-25 02:35:14 +10304753 if (len < cpumask_size())
4754 cpumask_clear(new_mask);
4755 else if (len > cpumask_size())
4756 len = cpumask_size();
4757
Linus Torvalds1da177e2005-04-16 15:20:36 -07004758 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
4759}
4760
4761/**
4762 * sys_sched_setaffinity - set the cpu affinity of a process
4763 * @pid: pid of the process
4764 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4765 * @user_mask_ptr: user-space pointer to the new cpu mask
4766 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004767SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
4768 unsigned long __user *, user_mask_ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004769{
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304770 cpumask_var_t new_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004771 int retval;
4772
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304773 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
4774 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004775
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304776 retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
4777 if (retval == 0)
4778 retval = sched_setaffinity(pid, new_mask);
4779 free_cpumask_var(new_mask);
4780 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004781}
4782
Rusty Russell96f874e2008-11-25 02:35:14 +10304783long sched_getaffinity(pid_t pid, struct cpumask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004784{
Ingo Molnar36c8b582006-07-03 00:25:41 -07004785 struct task_struct *p;
Thomas Gleixner31605682009-12-08 20:24:16 +00004786 unsigned long flags;
4787 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004788 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004789
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004790 get_online_cpus();
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004791 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004792
4793 retval = -ESRCH;
4794 p = find_process_by_pid(pid);
4795 if (!p)
4796 goto out_unlock;
4797
David Quigleye7834f82006-06-23 02:03:59 -07004798 retval = security_task_getscheduler(p);
4799 if (retval)
4800 goto out_unlock;
4801
Thomas Gleixner31605682009-12-08 20:24:16 +00004802 rq = task_rq_lock(p, &flags);
Rusty Russell96f874e2008-11-25 02:35:14 +10304803 cpumask_and(mask, &p->cpus_allowed, cpu_online_mask);
Thomas Gleixner31605682009-12-08 20:24:16 +00004804 task_rq_unlock(rq, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004805
4806out_unlock:
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004807 rcu_read_unlock();
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004808 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004809
Ulrich Drepper9531b622007-08-09 11:16:46 +02004810 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004811}
4812
4813/**
4814 * sys_sched_getaffinity - get the cpu affinity of a process
4815 * @pid: pid of the process
4816 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4817 * @user_mask_ptr: user-space pointer to hold the current cpu mask
4818 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004819SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
4820 unsigned long __user *, user_mask_ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004821{
4822 int ret;
Rusty Russellf17c8602008-11-25 02:35:11 +10304823 cpumask_var_t mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004824
Anton Blanchard84fba5e2010-04-06 17:02:19 +10004825 if ((len * BITS_PER_BYTE) < nr_cpu_ids)
KOSAKI Motohirocd3d8032010-03-12 16:15:36 +09004826 return -EINVAL;
4827 if (len & (sizeof(unsigned long)-1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004828 return -EINVAL;
4829
Rusty Russellf17c8602008-11-25 02:35:11 +10304830 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
4831 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004832
Rusty Russellf17c8602008-11-25 02:35:11 +10304833 ret = sched_getaffinity(pid, mask);
4834 if (ret == 0) {
KOSAKI Motohiro8bc037f2010-03-17 09:36:58 +09004835 size_t retlen = min_t(size_t, len, cpumask_size());
KOSAKI Motohirocd3d8032010-03-12 16:15:36 +09004836
4837 if (copy_to_user(user_mask_ptr, mask, retlen))
Rusty Russellf17c8602008-11-25 02:35:11 +10304838 ret = -EFAULT;
4839 else
KOSAKI Motohirocd3d8032010-03-12 16:15:36 +09004840 ret = retlen;
Rusty Russellf17c8602008-11-25 02:35:11 +10304841 }
4842 free_cpumask_var(mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004843
Rusty Russellf17c8602008-11-25 02:35:11 +10304844 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004845}
4846
4847/**
4848 * sys_sched_yield - yield the current processor to other threads.
4849 *
Ingo Molnardd41f592007-07-09 18:51:59 +02004850 * This function yields the current CPU to other tasks. If there are no
4851 * other threads running on this CPU then this function will return.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004852 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004853SYSCALL_DEFINE0(sched_yield)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004854{
Ingo Molnar70b97a72006-07-03 00:25:42 -07004855 struct rq *rq = this_rq_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004856
Ingo Molnar2d723762007-10-15 17:00:12 +02004857 schedstat_inc(rq, yld_count);
Dmitry Adamushko4530d7a2007-10-15 17:00:08 +02004858 current->sched_class->yield_task(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004859
4860 /*
4861 * Since we are going to call schedule() anyway, there's
4862 * no need to preempt or enable interrupts:
4863 */
4864 __release(rq->lock);
Ingo Molnar8a25d5d2006-07-03 00:24:54 -07004865 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
Thomas Gleixner9828ea92009-12-03 20:55:53 +01004866 do_raw_spin_unlock(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004867 preempt_enable_no_resched();
4868
4869 schedule();
4870
4871 return 0;
4872}
4873
Peter Zijlstrad86ee482009-07-10 14:57:57 +02004874static inline int should_resched(void)
4875{
4876 return need_resched() && !(preempt_count() & PREEMPT_ACTIVE);
4877}
4878
Andrew Mortone7b38402006-06-30 01:56:00 -07004879static void __cond_resched(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004880{
Frederic Weisbeckere7aaaa62009-07-16 15:44:29 +02004881 add_preempt_count(PREEMPT_ACTIVE);
4882 schedule();
4883 sub_preempt_count(PREEMPT_ACTIVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004884}
4885
Herbert Xu02b67cc32008-01-25 21:08:28 +01004886int __sched _cond_resched(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004887{
Peter Zijlstrad86ee482009-07-10 14:57:57 +02004888 if (should_resched()) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004889 __cond_resched();
4890 return 1;
4891 }
4892 return 0;
4893}
Herbert Xu02b67cc32008-01-25 21:08:28 +01004894EXPORT_SYMBOL(_cond_resched);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004895
4896/*
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02004897 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004898 * call schedule, and on return reacquire the lock.
4899 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01004900 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
Linus Torvalds1da177e2005-04-16 15:20:36 -07004901 * operations here to prevent schedule() from being called twice (once via
4902 * spin_unlock(), once by hand).
4903 */
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02004904int __cond_resched_lock(spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004905{
Peter Zijlstrad86ee482009-07-10 14:57:57 +02004906 int resched = should_resched();
Jan Kara6df3cec2005-06-13 15:52:32 -07004907 int ret = 0;
4908
Peter Zijlstraf607c662009-07-20 19:16:29 +02004909 lockdep_assert_held(lock);
4910
Nick Piggin95c354f2008-01-30 13:31:20 +01004911 if (spin_needbreak(lock) || resched) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004912 spin_unlock(lock);
Peter Zijlstrad86ee482009-07-10 14:57:57 +02004913 if (resched)
Nick Piggin95c354f2008-01-30 13:31:20 +01004914 __cond_resched();
4915 else
4916 cpu_relax();
Jan Kara6df3cec2005-06-13 15:52:32 -07004917 ret = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004918 spin_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004919 }
Jan Kara6df3cec2005-06-13 15:52:32 -07004920 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004921}
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02004922EXPORT_SYMBOL(__cond_resched_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004923
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02004924int __sched __cond_resched_softirq(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004925{
4926 BUG_ON(!in_softirq());
4927
Peter Zijlstrad86ee482009-07-10 14:57:57 +02004928 if (should_resched()) {
Thomas Gleixner98d825672007-05-23 13:58:18 -07004929 local_bh_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004930 __cond_resched();
4931 local_bh_disable();
4932 return 1;
4933 }
4934 return 0;
4935}
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02004936EXPORT_SYMBOL(__cond_resched_softirq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004937
Linus Torvalds1da177e2005-04-16 15:20:36 -07004938/**
4939 * yield - yield the current processor to other threads.
4940 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08004941 * This is a shortcut for kernel-space yielding - it marks the
Linus Torvalds1da177e2005-04-16 15:20:36 -07004942 * thread runnable and calls sys_sched_yield().
4943 */
4944void __sched yield(void)
4945{
4946 set_current_state(TASK_RUNNING);
4947 sys_sched_yield();
4948}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004949EXPORT_SYMBOL(yield);
4950
4951/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01004952 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
Linus Torvalds1da177e2005-04-16 15:20:36 -07004953 * that process accounting knows that this is a task in IO wait state.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004954 */
4955void __sched io_schedule(void)
4956{
Hitoshi Mitake54d35f22009-06-29 14:44:57 +09004957 struct rq *rq = raw_rq();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004958
Shailabh Nagar0ff92242006-07-14 00:24:37 -07004959 delayacct_blkio_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004960 atomic_inc(&rq->nr_iowait);
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07004961 current->in_iowait = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004962 schedule();
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07004963 current->in_iowait = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004964 atomic_dec(&rq->nr_iowait);
Shailabh Nagar0ff92242006-07-14 00:24:37 -07004965 delayacct_blkio_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004966}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004967EXPORT_SYMBOL(io_schedule);
4968
4969long __sched io_schedule_timeout(long timeout)
4970{
Hitoshi Mitake54d35f22009-06-29 14:44:57 +09004971 struct rq *rq = raw_rq();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004972 long ret;
4973
Shailabh Nagar0ff92242006-07-14 00:24:37 -07004974 delayacct_blkio_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004975 atomic_inc(&rq->nr_iowait);
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07004976 current->in_iowait = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004977 ret = schedule_timeout(timeout);
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07004978 current->in_iowait = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004979 atomic_dec(&rq->nr_iowait);
Shailabh Nagar0ff92242006-07-14 00:24:37 -07004980 delayacct_blkio_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004981 return ret;
4982}
4983
4984/**
4985 * sys_sched_get_priority_max - return maximum RT priority.
4986 * @policy: scheduling class.
4987 *
4988 * this syscall returns the maximum rt_priority that can be used
4989 * by a given scheduling class.
4990 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004991SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004992{
4993 int ret = -EINVAL;
4994
4995 switch (policy) {
4996 case SCHED_FIFO:
4997 case SCHED_RR:
4998 ret = MAX_USER_RT_PRIO-1;
4999 break;
5000 case SCHED_NORMAL:
Ingo Molnarb0a94992006-01-14 13:20:41 -08005001 case SCHED_BATCH:
Ingo Molnardd41f592007-07-09 18:51:59 +02005002 case SCHED_IDLE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005003 ret = 0;
5004 break;
5005 }
5006 return ret;
5007}
5008
5009/**
5010 * sys_sched_get_priority_min - return minimum RT priority.
5011 * @policy: scheduling class.
5012 *
5013 * this syscall returns the minimum rt_priority that can be used
5014 * by a given scheduling class.
5015 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01005016SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005017{
5018 int ret = -EINVAL;
5019
5020 switch (policy) {
5021 case SCHED_FIFO:
5022 case SCHED_RR:
5023 ret = 1;
5024 break;
5025 case SCHED_NORMAL:
Ingo Molnarb0a94992006-01-14 13:20:41 -08005026 case SCHED_BATCH:
Ingo Molnardd41f592007-07-09 18:51:59 +02005027 case SCHED_IDLE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005028 ret = 0;
5029 }
5030 return ret;
5031}
5032
5033/**
5034 * sys_sched_rr_get_interval - return the default timeslice of a process.
5035 * @pid: pid of the process.
5036 * @interval: userspace pointer to the timeslice value.
5037 *
5038 * this syscall writes the default timeslice value of a given process
5039 * into the user-space timespec buffer. A value of '0' means infinity.
5040 */
Heiko Carstens17da2bd2009-01-14 14:14:10 +01005041SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
Heiko Carstens754fe8d2009-01-14 14:14:09 +01005042 struct timespec __user *, interval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005043{
Ingo Molnar36c8b582006-07-03 00:25:41 -07005044 struct task_struct *p;
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005045 unsigned int time_slice;
Thomas Gleixnerdba091b2009-12-09 09:32:03 +01005046 unsigned long flags;
5047 struct rq *rq;
Andi Kleen3a5c3592007-10-15 17:00:14 +02005048 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005049 struct timespec t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005050
5051 if (pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02005052 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005053
5054 retval = -ESRCH;
Thomas Gleixner1a551ae2009-12-09 10:15:11 +00005055 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005056 p = find_process_by_pid(pid);
5057 if (!p)
5058 goto out_unlock;
5059
5060 retval = security_task_getscheduler(p);
5061 if (retval)
5062 goto out_unlock;
5063
Thomas Gleixnerdba091b2009-12-09 09:32:03 +01005064 rq = task_rq_lock(p, &flags);
5065 time_slice = p->sched_class->get_rr_interval(rq, p);
5066 task_rq_unlock(rq, &flags);
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005067
Thomas Gleixner1a551ae2009-12-09 10:15:11 +00005068 rcu_read_unlock();
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02005069 jiffies_to_timespec(time_slice, &t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005070 retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005071 return retval;
Andi Kleen3a5c3592007-10-15 17:00:14 +02005072
Linus Torvalds1da177e2005-04-16 15:20:36 -07005073out_unlock:
Thomas Gleixner1a551ae2009-12-09 10:15:11 +00005074 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005075 return retval;
5076}
5077
Steven Rostedt7c731e02008-05-12 21:20:41 +02005078static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
Ingo Molnar36c8b582006-07-03 00:25:41 -07005079
Ingo Molnar82a1fcb2008-01-25 21:08:02 +01005080void sched_show_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005081{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005082 unsigned long free = 0;
Ingo Molnar36c8b582006-07-03 00:25:41 -07005083 unsigned state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005084
Linus Torvalds1da177e2005-04-16 15:20:36 -07005085 state = p->state ? __ffs(p->state) + 1 : 0;
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005086 printk(KERN_INFO "%-13.13s %c", p->comm,
Andreas Mohr2ed6e342006-07-10 04:43:52 -07005087 state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
Ingo Molnar4bd77322007-07-11 21:21:47 +02005088#if BITS_PER_LONG == 32
Linus Torvalds1da177e2005-04-16 15:20:36 -07005089 if (state == TASK_RUNNING)
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005090 printk(KERN_CONT " running ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005091 else
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005092 printk(KERN_CONT " %08lx ", thread_saved_pc(p));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005093#else
5094 if (state == TASK_RUNNING)
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005095 printk(KERN_CONT " running task ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005096 else
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005097 printk(KERN_CONT " %016lx ", thread_saved_pc(p));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005098#endif
5099#ifdef CONFIG_DEBUG_STACK_USAGE
Eric Sandeen7c9f8862008-04-22 16:38:23 -05005100 free = stack_not_used(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005101#endif
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005102 printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
David Rientjesaa47b7e2009-05-04 01:38:05 -07005103 task_pid_nr(p), task_pid_nr(p->real_parent),
5104 (unsigned long)task_thread_info(p)->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005105
Nick Piggin5fb5e6d2008-01-25 21:08:34 +01005106 show_stack(p, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005107}
5108
Ingo Molnare59e2ae2006-12-06 20:35:59 -08005109void show_state_filter(unsigned long state_filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005110{
Ingo Molnar36c8b582006-07-03 00:25:41 -07005111 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005112
Ingo Molnar4bd77322007-07-11 21:21:47 +02005113#if BITS_PER_LONG == 32
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005114 printk(KERN_INFO
5115 " task PC stack pid father\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005116#else
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005117 printk(KERN_INFO
5118 " task PC stack pid father\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005119#endif
5120 read_lock(&tasklist_lock);
5121 do_each_thread(g, p) {
5122 /*
5123 * reset the NMI-timeout, listing all files on a slow
5124 * console might take alot of time:
5125 */
5126 touch_nmi_watchdog();
Ingo Molnar39bc89f2007-04-25 20:50:03 -07005127 if (!state_filter || (p->state & state_filter))
Ingo Molnar82a1fcb2008-01-25 21:08:02 +01005128 sched_show_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005129 } while_each_thread(g, p);
5130
Jeremy Fitzhardinge04c91672007-05-08 00:28:05 -07005131 touch_all_softlockup_watchdogs();
5132
Ingo Molnardd41f592007-07-09 18:51:59 +02005133#ifdef CONFIG_SCHED_DEBUG
5134 sysrq_sched_debug_show();
5135#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005136 read_unlock(&tasklist_lock);
Ingo Molnare59e2ae2006-12-06 20:35:59 -08005137 /*
5138 * Only show locks if all tasks are dumped:
5139 */
Shmulik Ladkani93335a22009-11-25 15:23:41 +02005140 if (!state_filter)
Ingo Molnare59e2ae2006-12-06 20:35:59 -08005141 debug_show_all_locks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005142}
5143
Ingo Molnar1df21052007-07-09 18:51:58 +02005144void __cpuinit init_idle_bootup_task(struct task_struct *idle)
5145{
Ingo Molnardd41f592007-07-09 18:51:59 +02005146 idle->sched_class = &idle_sched_class;
Ingo Molnar1df21052007-07-09 18:51:58 +02005147}
5148
Ingo Molnarf340c0d2005-06-28 16:40:42 +02005149/**
5150 * init_idle - set up an idle thread for a given CPU
5151 * @idle: task in question
5152 * @cpu: cpu the idle task belongs to
5153 *
5154 * NOTE: this function does not set the idle thread's NEED_RESCHED
5155 * flag, to make booting more robust.
5156 */
Nick Piggin5c1e1762006-10-03 01:14:04 -07005157void __cpuinit init_idle(struct task_struct *idle, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005158{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005159 struct rq *rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005160 unsigned long flags;
5161
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005162 raw_spin_lock_irqsave(&rq->lock, flags);
Ingo Molnar5cbd54e2008-11-12 20:05:50 +01005163
Ingo Molnardd41f592007-07-09 18:51:59 +02005164 __sched_fork(idle);
Peter Zijlstra06b83b52009-12-16 18:04:35 +01005165 idle->state = TASK_RUNNING;
Ingo Molnardd41f592007-07-09 18:51:59 +02005166 idle->se.exec_start = sched_clock();
5167
Rusty Russell96f874e2008-11-25 02:35:14 +10305168 cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu));
Ingo Molnardd41f592007-07-09 18:51:59 +02005169 __set_task_cpu(idle, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005170
Linus Torvalds1da177e2005-04-16 15:20:36 -07005171 rq->curr = rq->idle = idle;
Nick Piggin4866cde2005-06-25 14:57:23 -07005172#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
5173 idle->oncpu = 1;
5174#endif
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005175 raw_spin_unlock_irqrestore(&rq->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005176
5177 /* Set the preempt count _outside_ the spinlocks! */
Linus Torvalds8e3e0762008-05-10 20:58:02 -07005178#if defined(CONFIG_PREEMPT)
5179 task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0);
5180#else
Al Viroa1261f52005-11-13 16:06:55 -08005181 task_thread_info(idle)->preempt_count = 0;
Linus Torvalds8e3e0762008-05-10 20:58:02 -07005182#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02005183 /*
5184 * The idle tasks have their own, simple scheduling class:
5185 */
5186 idle->sched_class = &idle_sched_class;
Frederic Weisbeckerfb526072008-11-25 21:07:04 +01005187 ftrace_graph_init_task(idle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005188}
5189
5190/*
5191 * In a system that switches off the HZ timer nohz_cpu_mask
5192 * indicates which cpus entered this state. This is used
5193 * in the rcu update to wait only for active cpus. For system
5194 * which do not switch off the HZ timer nohz_cpu_mask should
Rusty Russell6a7b3dc2008-11-25 02:35:04 +10305195 * always be CPU_BITS_NONE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005196 */
Rusty Russell6a7b3dc2008-11-25 02:35:04 +10305197cpumask_var_t nohz_cpu_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005198
Ingo Molnar19978ca2007-11-09 22:39:38 +01005199/*
5200 * Increase the granularity value when there are more CPUs,
5201 * because with more CPUs the 'effective latency' as visible
5202 * to users decreases. But the relationship is not linear,
5203 * so pick a second-best guess by going with the log2 of the
5204 * number of CPUs.
5205 *
5206 * This idea comes from the SD scheduler of Con Kolivas:
5207 */
Christian Ehrhardtacb4a842009-11-30 12:16:48 +01005208static int get_update_sysctl_factor(void)
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01005209{
Mike Galbraith4ca3ef72009-12-10 09:25:53 +01005210 unsigned int cpus = min_t(int, num_online_cpus(), 8);
Christian Ehrhardt1983a922009-11-30 12:16:47 +01005211 unsigned int factor;
5212
5213 switch (sysctl_sched_tunable_scaling) {
5214 case SCHED_TUNABLESCALING_NONE:
5215 factor = 1;
5216 break;
5217 case SCHED_TUNABLESCALING_LINEAR:
5218 factor = cpus;
5219 break;
5220 case SCHED_TUNABLESCALING_LOG:
5221 default:
5222 factor = 1 + ilog2(cpus);
5223 break;
5224 }
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01005225
Christian Ehrhardtacb4a842009-11-30 12:16:48 +01005226 return factor;
5227}
5228
5229static void update_sysctl(void)
5230{
5231 unsigned int factor = get_update_sysctl_factor();
5232
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01005233#define SET_SYSCTL(name) \
5234 (sysctl_##name = (factor) * normalized_sysctl_##name)
5235 SET_SYSCTL(sched_min_granularity);
5236 SET_SYSCTL(sched_latency);
5237 SET_SYSCTL(sched_wakeup_granularity);
5238 SET_SYSCTL(sched_shares_ratelimit);
5239#undef SET_SYSCTL
5240}
5241
Ingo Molnar19978ca2007-11-09 22:39:38 +01005242static inline void sched_init_granularity(void)
5243{
Christian Ehrhardt0bcdcf22009-11-30 12:16:46 +01005244 update_sysctl();
Ingo Molnar19978ca2007-11-09 22:39:38 +01005245}
5246
Linus Torvalds1da177e2005-04-16 15:20:36 -07005247#ifdef CONFIG_SMP
5248/*
5249 * This is how migration works:
5250 *
Tejun Heo969c7922010-05-06 18:49:21 +02005251 * 1) we invoke migration_cpu_stop() on the target CPU using
5252 * stop_one_cpu().
5253 * 2) stopper starts to run (implicitly forcing the migrated thread
5254 * off the CPU)
5255 * 3) it checks whether the migrated task is still in the wrong runqueue.
5256 * 4) if it's in the wrong runqueue then the migration thread removes
Linus Torvalds1da177e2005-04-16 15:20:36 -07005257 * it and puts it into the right queue.
Tejun Heo969c7922010-05-06 18:49:21 +02005258 * 5) stopper completes and stop_one_cpu() returns and the migration
5259 * is done.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005260 */
5261
5262/*
5263 * Change a given task's CPU affinity. Migrate the thread to a
5264 * proper CPU and schedule it away if the CPU it's executing on
5265 * is removed from the allowed bitmask.
5266 *
5267 * NOTE: the caller must have a valid reference to the task, the
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005268 * task must not exit() & deallocate itself prematurely. The
Linus Torvalds1da177e2005-04-16 15:20:36 -07005269 * call is not atomic; no spinlocks may be held.
5270 */
Rusty Russell96f874e2008-11-25 02:35:14 +10305271int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005272{
5273 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07005274 struct rq *rq;
Tejun Heo969c7922010-05-06 18:49:21 +02005275 unsigned int dest_cpu;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005276 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005277
Peter Zijlstra65cc8e42010-03-25 21:05:16 +01005278 /*
5279 * Serialize against TASK_WAKING so that ttwu() and wunt() can
5280 * drop the rq->lock and still rely on ->cpus_allowed.
5281 */
5282again:
5283 while (task_is_waking(p))
5284 cpu_relax();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005285 rq = task_rq_lock(p, &flags);
Peter Zijlstra65cc8e42010-03-25 21:05:16 +01005286 if (task_is_waking(p)) {
5287 task_rq_unlock(rq, &flags);
5288 goto again;
5289 }
Peter Zijlstrae2912002009-12-16 18:04:36 +01005290
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01005291 if (!cpumask_intersects(new_mask, cpu_active_mask)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005292 ret = -EINVAL;
5293 goto out;
5294 }
5295
David Rientjes9985b0b2008-06-05 12:57:11 -07005296 if (unlikely((p->flags & PF_THREAD_BOUND) && p != current &&
Rusty Russell96f874e2008-11-25 02:35:14 +10305297 !cpumask_equal(&p->cpus_allowed, new_mask))) {
David Rientjes9985b0b2008-06-05 12:57:11 -07005298 ret = -EINVAL;
5299 goto out;
5300 }
5301
Gregory Haskins73fe6aa2008-01-25 21:08:07 +01005302 if (p->sched_class->set_cpus_allowed)
Mike Traviscd8ba7c2008-03-26 14:23:49 -07005303 p->sched_class->set_cpus_allowed(p, new_mask);
Gregory Haskins73fe6aa2008-01-25 21:08:07 +01005304 else {
Rusty Russell96f874e2008-11-25 02:35:14 +10305305 cpumask_copy(&p->cpus_allowed, new_mask);
5306 p->rt.nr_cpus_allowed = cpumask_weight(new_mask);
Gregory Haskins73fe6aa2008-01-25 21:08:07 +01005307 }
5308
Linus Torvalds1da177e2005-04-16 15:20:36 -07005309 /* Can the task run on the task's current CPU? If so, we're done */
Rusty Russell96f874e2008-11-25 02:35:14 +10305310 if (cpumask_test_cpu(task_cpu(p), new_mask))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005311 goto out;
5312
Tejun Heo969c7922010-05-06 18:49:21 +02005313 dest_cpu = cpumask_any_and(cpu_active_mask, new_mask);
5314 if (migrate_task(p, dest_cpu)) {
5315 struct migration_arg arg = { p, dest_cpu };
Linus Torvalds1da177e2005-04-16 15:20:36 -07005316 /* Need help from migration thread: drop lock and wait. */
5317 task_rq_unlock(rq, &flags);
Tejun Heo969c7922010-05-06 18:49:21 +02005318 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005319 tlb_migrate_finish(p->mm);
5320 return 0;
5321 }
5322out:
5323 task_rq_unlock(rq, &flags);
Ingo Molnar48f24c42006-07-03 00:25:40 -07005324
Linus Torvalds1da177e2005-04-16 15:20:36 -07005325 return ret;
5326}
Mike Traviscd8ba7c2008-03-26 14:23:49 -07005327EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005328
5329/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005330 * Move (not current) task off this cpu, onto dest cpu. We're doing
Linus Torvalds1da177e2005-04-16 15:20:36 -07005331 * this because either it can't run here any more (set_cpus_allowed()
5332 * away from this CPU, or CPU going down), or because we're
5333 * attempting to rebalance this task on exec (sched_exec).
5334 *
5335 * So we race with normal scheduler movements, but that's OK, as long
5336 * as the task is no longer on this CPU.
Kirill Korotaevefc30812006-06-27 02:54:32 -07005337 *
5338 * Returns non-zero if task was successfully migrated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005339 */
Kirill Korotaevefc30812006-06-27 02:54:32 -07005340static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005341{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005342 struct rq *rq_dest, *rq_src;
Peter Zijlstrae2912002009-12-16 18:04:36 +01005343 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005344
Max Krasnyanskye761b772008-07-15 04:43:49 -07005345 if (unlikely(!cpu_active(dest_cpu)))
Kirill Korotaevefc30812006-06-27 02:54:32 -07005346 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005347
5348 rq_src = cpu_rq(src_cpu);
5349 rq_dest = cpu_rq(dest_cpu);
5350
5351 double_rq_lock(rq_src, rq_dest);
5352 /* Already moved. */
5353 if (task_cpu(p) != src_cpu)
Linus Torvaldsb1e38732008-07-10 11:25:03 -07005354 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005355 /* Affinity changed (again). */
Rusty Russell96f874e2008-11-25 02:35:14 +10305356 if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
Linus Torvaldsb1e38732008-07-10 11:25:03 -07005357 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005358
Peter Zijlstrae2912002009-12-16 18:04:36 +01005359 /*
5360 * If we're not on a rq, the next wake-up will ensure we're
5361 * placed properly.
5362 */
5363 if (p->se.on_rq) {
Ingo Molnar2e1cb742007-08-09 11:16:49 +02005364 deactivate_task(rq_src, p, 0);
Peter Zijlstrae2912002009-12-16 18:04:36 +01005365 set_task_cpu(p, dest_cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02005366 activate_task(rq_dest, p, 0);
Peter Zijlstra15afe092008-09-20 23:38:02 +02005367 check_preempt_curr(rq_dest, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005368 }
Linus Torvaldsb1e38732008-07-10 11:25:03 -07005369done:
Kirill Korotaevefc30812006-06-27 02:54:32 -07005370 ret = 1;
Linus Torvaldsb1e38732008-07-10 11:25:03 -07005371fail:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005372 double_rq_unlock(rq_src, rq_dest);
Kirill Korotaevefc30812006-06-27 02:54:32 -07005373 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005374}
5375
5376/*
Tejun Heo969c7922010-05-06 18:49:21 +02005377 * migration_cpu_stop - this will be executed by a highprio stopper thread
5378 * and performs thread migration by bumping thread off CPU then
5379 * 'pushing' onto another runqueue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005380 */
Tejun Heo969c7922010-05-06 18:49:21 +02005381static int migration_cpu_stop(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005382{
Tejun Heo969c7922010-05-06 18:49:21 +02005383 struct migration_arg *arg = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005384
Tejun Heo969c7922010-05-06 18:49:21 +02005385 /*
5386 * The original target cpu might have gone down and we might
5387 * be on another cpu but it doesn't matter.
5388 */
5389 local_irq_disable();
5390 __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu);
5391 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005392 return 0;
5393}
5394
5395#ifdef CONFIG_HOTPLUG_CPU
Kirill Korotaev054b9102006-12-10 02:20:11 -08005396/*
Robert P. J. Day3a4fa0a2007-10-19 23:10:43 +02005397 * Figure out where task on dead CPU should go, use force if necessary.
Kirill Korotaev054b9102006-12-10 02:20:11 -08005398 */
Oleg Nesterov6a1bdc12010-03-15 10:10:23 +01005399void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005400{
Oleg Nesterov1445c082010-03-15 10:10:10 +01005401 struct rq *rq = cpu_rq(dead_cpu);
5402 int needs_cpu, uninitialized_var(dest_cpu);
5403 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005404
Oleg Nesterov1445c082010-03-15 10:10:10 +01005405 local_irq_save(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005406
Oleg Nesterov1445c082010-03-15 10:10:10 +01005407 raw_spin_lock(&rq->lock);
5408 needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING);
5409 if (needs_cpu)
5410 dest_cpu = select_fallback_rq(dead_cpu, p);
5411 raw_spin_unlock(&rq->lock);
Oleg Nesterovc1804d52010-03-15 10:10:14 +01005412 /*
5413 * It can only fail if we race with set_cpus_allowed(),
5414 * in the racer should migrate the task anyway.
5415 */
Oleg Nesterov1445c082010-03-15 10:10:10 +01005416 if (needs_cpu)
Oleg Nesterovc1804d52010-03-15 10:10:14 +01005417 __migrate_task(p, dead_cpu, dest_cpu);
Oleg Nesterov1445c082010-03-15 10:10:10 +01005418 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005419}
5420
5421/*
5422 * While a dead CPU has no uninterruptible tasks queued at this point,
5423 * it might still have a nonzero ->nr_uninterruptible counter, because
5424 * for performance reasons the counter is not stricly tracking tasks to
5425 * their home CPUs. So we just add the counter to another CPU's counter,
5426 * to keep the global sum constant after CPU-down:
5427 */
Ingo Molnar70b97a72006-07-03 00:25:42 -07005428static void migrate_nr_uninterruptible(struct rq *rq_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005429{
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01005430 struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005431 unsigned long flags;
5432
5433 local_irq_save(flags);
5434 double_rq_lock(rq_src, rq_dest);
5435 rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible;
5436 rq_src->nr_uninterruptible = 0;
5437 double_rq_unlock(rq_src, rq_dest);
5438 local_irq_restore(flags);
5439}
5440
5441/* Run through task list and migrate tasks from the dead cpu. */
5442static void migrate_live_tasks(int src_cpu)
5443{
Ingo Molnar48f24c42006-07-03 00:25:40 -07005444 struct task_struct *p, *t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005445
Oleg Nesterovf7b4cdd2007-10-16 23:30:56 -07005446 read_lock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005447
Ingo Molnar48f24c42006-07-03 00:25:40 -07005448 do_each_thread(t, p) {
5449 if (p == current)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005450 continue;
5451
Ingo Molnar48f24c42006-07-03 00:25:40 -07005452 if (task_cpu(p) == src_cpu)
5453 move_task_off_dead_cpu(src_cpu, p);
5454 } while_each_thread(t, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005455
Oleg Nesterovf7b4cdd2007-10-16 23:30:56 -07005456 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005457}
5458
Ingo Molnardd41f592007-07-09 18:51:59 +02005459/*
5460 * Schedules idle task to be the next runnable task on current CPU.
Dmitry Adamushko94bc9a72007-11-15 20:57:40 +01005461 * It does so by boosting its priority to highest possible.
5462 * Used by CPU offline code.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005463 */
5464void sched_idle_next(void)
5465{
Ingo Molnar48f24c42006-07-03 00:25:40 -07005466 int this_cpu = smp_processor_id();
Ingo Molnar70b97a72006-07-03 00:25:42 -07005467 struct rq *rq = cpu_rq(this_cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005468 struct task_struct *p = rq->idle;
5469 unsigned long flags;
5470
5471 /* cpu has to be offline */
Ingo Molnar48f24c42006-07-03 00:25:40 -07005472 BUG_ON(cpu_online(this_cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005473
Ingo Molnar48f24c42006-07-03 00:25:40 -07005474 /*
5475 * Strictly not necessary since rest of the CPUs are stopped by now
5476 * and interrupts disabled on the current cpu.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005477 */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005478 raw_spin_lock_irqsave(&rq->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005479
Ingo Molnardd41f592007-07-09 18:51:59 +02005480 __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1);
Ingo Molnar48f24c42006-07-03 00:25:40 -07005481
Dmitry Adamushko94bc9a72007-11-15 20:57:40 +01005482 activate_task(rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005483
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005484 raw_spin_unlock_irqrestore(&rq->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005485}
5486
Ingo Molnar48f24c42006-07-03 00:25:40 -07005487/*
5488 * Ensures that the idle task is using init_mm right before its cpu goes
Linus Torvalds1da177e2005-04-16 15:20:36 -07005489 * offline.
5490 */
5491void idle_task_exit(void)
5492{
5493 struct mm_struct *mm = current->active_mm;
5494
5495 BUG_ON(cpu_online(smp_processor_id()));
5496
5497 if (mm != &init_mm)
5498 switch_mm(mm, &init_mm, current);
5499 mmdrop(mm);
5500}
5501
Kirill Korotaev054b9102006-12-10 02:20:11 -08005502/* called under rq->lock with disabled interrupts */
Ingo Molnar36c8b582006-07-03 00:25:41 -07005503static void migrate_dead(unsigned int dead_cpu, struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005504{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005505 struct rq *rq = cpu_rq(dead_cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005506
5507 /* Must be exiting, otherwise would be on tasklist. */
Eugene Teo270f7222007-10-18 23:40:38 -07005508 BUG_ON(!p->exit_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005509
5510 /* Cannot have done final schedule yet: would have vanished. */
Oleg Nesterovc394cc92006-09-29 02:01:11 -07005511 BUG_ON(p->state == TASK_DEAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005512
Ingo Molnar48f24c42006-07-03 00:25:40 -07005513 get_task_struct(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005514
5515 /*
5516 * Drop lock around migration; if someone else moves it,
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005517 * that's OK. No task can be added to this CPU, so iteration is
Linus Torvalds1da177e2005-04-16 15:20:36 -07005518 * fine.
5519 */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005520 raw_spin_unlock_irq(&rq->lock);
Ingo Molnar48f24c42006-07-03 00:25:40 -07005521 move_task_off_dead_cpu(dead_cpu, p);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005522 raw_spin_lock_irq(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005523
Ingo Molnar48f24c42006-07-03 00:25:40 -07005524 put_task_struct(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005525}
5526
5527/* release_task() removes task from tasklist, so we won't find dead tasks. */
5528static void migrate_dead_tasks(unsigned int dead_cpu)
5529{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005530 struct rq *rq = cpu_rq(dead_cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02005531 struct task_struct *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005532
Ingo Molnardd41f592007-07-09 18:51:59 +02005533 for ( ; ; ) {
5534 if (!rq->nr_running)
5535 break;
Wang Chenb67802e2009-03-02 13:55:26 +08005536 next = pick_next_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02005537 if (!next)
5538 break;
Dmitry Adamushko79c53792008-06-29 00:16:56 +02005539 next->sched_class->put_prev_task(rq, next);
Ingo Molnardd41f592007-07-09 18:51:59 +02005540 migrate_dead(dead_cpu, next);
Nick Piggine692ab52007-07-26 13:40:43 +02005541
Linus Torvalds1da177e2005-04-16 15:20:36 -07005542 }
5543}
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02005544
5545/*
5546 * remove the tasks which were accounted by rq from calc_load_tasks.
5547 */
5548static void calc_global_load_remove(struct rq *rq)
5549{
5550 atomic_long_sub(rq->calc_load_active, &calc_load_tasks);
Thomas Gleixnera468d382009-07-17 14:15:46 +02005551 rq->calc_load_active = 0;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02005552}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005553#endif /* CONFIG_HOTPLUG_CPU */
5554
Nick Piggine692ab52007-07-26 13:40:43 +02005555#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
5556
5557static struct ctl_table sd_ctl_dir[] = {
Alexey Dobriyane0361852007-08-09 11:16:46 +02005558 {
5559 .procname = "sched_domain",
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005560 .mode = 0555,
Alexey Dobriyane0361852007-08-09 11:16:46 +02005561 },
Eric W. Biederman56992302009-11-05 15:38:40 -08005562 {}
Nick Piggine692ab52007-07-26 13:40:43 +02005563};
5564
5565static struct ctl_table sd_ctl_root[] = {
Alexey Dobriyane0361852007-08-09 11:16:46 +02005566 {
5567 .procname = "kernel",
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005568 .mode = 0555,
Alexey Dobriyane0361852007-08-09 11:16:46 +02005569 .child = sd_ctl_dir,
5570 },
Eric W. Biederman56992302009-11-05 15:38:40 -08005571 {}
Nick Piggine692ab52007-07-26 13:40:43 +02005572};
5573
5574static struct ctl_table *sd_alloc_ctl_entry(int n)
5575{
5576 struct ctl_table *entry =
Milton Miller5cf9f062007-10-15 17:00:19 +02005577 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
Nick Piggine692ab52007-07-26 13:40:43 +02005578
Nick Piggine692ab52007-07-26 13:40:43 +02005579 return entry;
5580}
5581
Milton Miller6382bc92007-10-15 17:00:19 +02005582static void sd_free_ctl_entry(struct ctl_table **tablep)
5583{
Milton Millercd7900762007-10-17 16:55:11 +02005584 struct ctl_table *entry;
Milton Miller6382bc92007-10-15 17:00:19 +02005585
Milton Millercd7900762007-10-17 16:55:11 +02005586 /*
5587 * In the intermediate directories, both the child directory and
5588 * procname are dynamically allocated and could fail but the mode
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01005589 * will always be set. In the lowest directory the names are
Milton Millercd7900762007-10-17 16:55:11 +02005590 * static strings and all have proc handlers.
5591 */
5592 for (entry = *tablep; entry->mode; entry++) {
Milton Miller6382bc92007-10-15 17:00:19 +02005593 if (entry->child)
5594 sd_free_ctl_entry(&entry->child);
Milton Millercd7900762007-10-17 16:55:11 +02005595 if (entry->proc_handler == NULL)
5596 kfree(entry->procname);
5597 }
Milton Miller6382bc92007-10-15 17:00:19 +02005598
5599 kfree(*tablep);
5600 *tablep = NULL;
5601}
5602
Nick Piggine692ab52007-07-26 13:40:43 +02005603static void
Alexey Dobriyane0361852007-08-09 11:16:46 +02005604set_table_entry(struct ctl_table *entry,
Nick Piggine692ab52007-07-26 13:40:43 +02005605 const char *procname, void *data, int maxlen,
5606 mode_t mode, proc_handler *proc_handler)
5607{
Nick Piggine692ab52007-07-26 13:40:43 +02005608 entry->procname = procname;
5609 entry->data = data;
5610 entry->maxlen = maxlen;
5611 entry->mode = mode;
5612 entry->proc_handler = proc_handler;
5613}
5614
5615static struct ctl_table *
5616sd_alloc_ctl_domain_table(struct sched_domain *sd)
5617{
Ingo Molnara5d8c342008-10-09 11:35:51 +02005618 struct ctl_table *table = sd_alloc_ctl_entry(13);
Nick Piggine692ab52007-07-26 13:40:43 +02005619
Milton Millerad1cdc12007-10-15 17:00:19 +02005620 if (table == NULL)
5621 return NULL;
5622
Alexey Dobriyane0361852007-08-09 11:16:46 +02005623 set_table_entry(&table[0], "min_interval", &sd->min_interval,
Nick Piggine692ab52007-07-26 13:40:43 +02005624 sizeof(long), 0644, proc_doulongvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005625 set_table_entry(&table[1], "max_interval", &sd->max_interval,
Nick Piggine692ab52007-07-26 13:40:43 +02005626 sizeof(long), 0644, proc_doulongvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005627 set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005628 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005629 set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005630 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005631 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005632 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005633 set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005634 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005635 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
Nick Piggine692ab52007-07-26 13:40:43 +02005636 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005637 set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
Nick Piggine692ab52007-07-26 13:40:43 +02005638 sizeof(int), 0644, proc_dointvec_minmax);
Alexey Dobriyane0361852007-08-09 11:16:46 +02005639 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
Nick Piggine692ab52007-07-26 13:40:43 +02005640 sizeof(int), 0644, proc_dointvec_minmax);
Zou Nan haiace8b3d2007-10-15 17:00:14 +02005641 set_table_entry(&table[9], "cache_nice_tries",
Nick Piggine692ab52007-07-26 13:40:43 +02005642 &sd->cache_nice_tries,
5643 sizeof(int), 0644, proc_dointvec_minmax);
Zou Nan haiace8b3d2007-10-15 17:00:14 +02005644 set_table_entry(&table[10], "flags", &sd->flags,
Nick Piggine692ab52007-07-26 13:40:43 +02005645 sizeof(int), 0644, proc_dointvec_minmax);
Ingo Molnara5d8c342008-10-09 11:35:51 +02005646 set_table_entry(&table[11], "name", sd->name,
5647 CORENAME_MAX_SIZE, 0444, proc_dostring);
5648 /* &table[12] is terminator */
Nick Piggine692ab52007-07-26 13:40:43 +02005649
5650 return table;
5651}
5652
Ingo Molnar9a4e7152007-11-28 15:52:56 +01005653static ctl_table *sd_alloc_ctl_cpu_table(int cpu)
Nick Piggine692ab52007-07-26 13:40:43 +02005654{
5655 struct ctl_table *entry, *table;
5656 struct sched_domain *sd;
5657 int domain_num = 0, i;
5658 char buf[32];
5659
5660 for_each_domain(cpu, sd)
5661 domain_num++;
5662 entry = table = sd_alloc_ctl_entry(domain_num + 1);
Milton Millerad1cdc12007-10-15 17:00:19 +02005663 if (table == NULL)
5664 return NULL;
Nick Piggine692ab52007-07-26 13:40:43 +02005665
5666 i = 0;
5667 for_each_domain(cpu, sd) {
5668 snprintf(buf, 32, "domain%d", i);
Nick Piggine692ab52007-07-26 13:40:43 +02005669 entry->procname = kstrdup(buf, GFP_KERNEL);
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005670 entry->mode = 0555;
Nick Piggine692ab52007-07-26 13:40:43 +02005671 entry->child = sd_alloc_ctl_domain_table(sd);
5672 entry++;
5673 i++;
5674 }
5675 return table;
5676}
5677
5678static struct ctl_table_header *sd_sysctl_header;
Milton Miller6382bc92007-10-15 17:00:19 +02005679static void register_sched_domain_sysctl(void)
Nick Piggine692ab52007-07-26 13:40:43 +02005680{
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01005681 int i, cpu_num = num_possible_cpus();
Nick Piggine692ab52007-07-26 13:40:43 +02005682 struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
5683 char buf[32];
5684
Milton Miller73785472007-10-24 18:23:48 +02005685 WARN_ON(sd_ctl_dir[0].child);
5686 sd_ctl_dir[0].child = entry;
5687
Milton Millerad1cdc12007-10-15 17:00:19 +02005688 if (entry == NULL)
5689 return;
5690
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01005691 for_each_possible_cpu(i) {
Nick Piggine692ab52007-07-26 13:40:43 +02005692 snprintf(buf, 32, "cpu%d", i);
Nick Piggine692ab52007-07-26 13:40:43 +02005693 entry->procname = kstrdup(buf, GFP_KERNEL);
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02005694 entry->mode = 0555;
Nick Piggine692ab52007-07-26 13:40:43 +02005695 entry->child = sd_alloc_ctl_cpu_table(i);
Milton Miller97b6ea72007-10-15 17:00:19 +02005696 entry++;
Nick Piggine692ab52007-07-26 13:40:43 +02005697 }
Milton Miller73785472007-10-24 18:23:48 +02005698
5699 WARN_ON(sd_sysctl_header);
Nick Piggine692ab52007-07-26 13:40:43 +02005700 sd_sysctl_header = register_sysctl_table(sd_ctl_root);
5701}
Milton Miller6382bc92007-10-15 17:00:19 +02005702
Milton Miller73785472007-10-24 18:23:48 +02005703/* may be called multiple times per register */
Milton Miller6382bc92007-10-15 17:00:19 +02005704static void unregister_sched_domain_sysctl(void)
5705{
Milton Miller73785472007-10-24 18:23:48 +02005706 if (sd_sysctl_header)
5707 unregister_sysctl_table(sd_sysctl_header);
Milton Miller6382bc92007-10-15 17:00:19 +02005708 sd_sysctl_header = NULL;
Milton Miller73785472007-10-24 18:23:48 +02005709 if (sd_ctl_dir[0].child)
5710 sd_free_ctl_entry(&sd_ctl_dir[0].child);
Milton Miller6382bc92007-10-15 17:00:19 +02005711}
Nick Piggine692ab52007-07-26 13:40:43 +02005712#else
Milton Miller6382bc92007-10-15 17:00:19 +02005713static void register_sched_domain_sysctl(void)
5714{
5715}
5716static void unregister_sched_domain_sysctl(void)
Nick Piggine692ab52007-07-26 13:40:43 +02005717{
5718}
5719#endif
5720
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005721static void set_rq_online(struct rq *rq)
5722{
5723 if (!rq->online) {
5724 const struct sched_class *class;
5725
Rusty Russellc6c49272008-11-25 02:35:05 +10305726 cpumask_set_cpu(rq->cpu, rq->rd->online);
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005727 rq->online = 1;
5728
5729 for_each_class(class) {
5730 if (class->rq_online)
5731 class->rq_online(rq);
5732 }
5733 }
5734}
5735
5736static void set_rq_offline(struct rq *rq)
5737{
5738 if (rq->online) {
5739 const struct sched_class *class;
5740
5741 for_each_class(class) {
5742 if (class->rq_offline)
5743 class->rq_offline(rq);
5744 }
5745
Rusty Russellc6c49272008-11-25 02:35:05 +10305746 cpumask_clear_cpu(rq->cpu, rq->rd->online);
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005747 rq->online = 0;
5748 }
5749}
5750
Linus Torvalds1da177e2005-04-16 15:20:36 -07005751/*
5752 * migration_call - callback that gets triggered when a CPU is added.
5753 * Here we can start up the necessary migration thread for the new CPU.
5754 */
Ingo Molnar48f24c42006-07-03 00:25:40 -07005755static int __cpuinit
5756migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005757{
Ingo Molnar48f24c42006-07-03 00:25:40 -07005758 int cpu = (long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005759 unsigned long flags;
Tejun Heo969c7922010-05-06 18:49:21 +02005760 struct rq *rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005761
5762 switch (action) {
Gautham R Shenoy5be93612007-05-09 02:34:04 -07005763
Linus Torvalds1da177e2005-04-16 15:20:36 -07005764 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005765 case CPU_UP_PREPARE_FROZEN:
Thomas Gleixnera468d382009-07-17 14:15:46 +02005766 rq->calc_load_update = calc_load_update;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005767 break;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005768
Linus Torvalds1da177e2005-04-16 15:20:36 -07005769 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005770 case CPU_ONLINE_FROZEN:
Gregory Haskins1f94ef52008-03-10 16:52:41 -04005771 /* Update our root-domain */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005772 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins1f94ef52008-03-10 16:52:41 -04005773 if (rq->rd) {
Rusty Russellc6c49272008-11-25 02:35:05 +10305774 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005775
5776 set_rq_online(rq);
Gregory Haskins1f94ef52008-03-10 16:52:41 -04005777 }
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005778 raw_spin_unlock_irqrestore(&rq->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005779 break;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005780
Linus Torvalds1da177e2005-04-16 15:20:36 -07005781#ifdef CONFIG_HOTPLUG_CPU
Linus Torvalds1da177e2005-04-16 15:20:36 -07005782 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005783 case CPU_DEAD_FROZEN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005784 migrate_live_tasks(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005785 /* Idle task back to normal (off runqueue, low prio) */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005786 raw_spin_lock_irq(&rq->lock);
Ingo Molnar2e1cb742007-08-09 11:16:49 +02005787 deactivate_task(rq, rq->idle, 0);
Ingo Molnardd41f592007-07-09 18:51:59 +02005788 __setscheduler(rq, rq->idle, SCHED_NORMAL, 0);
5789 rq->idle->sched_class = &idle_sched_class;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005790 migrate_dead_tasks(cpu);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005791 raw_spin_unlock_irq(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005792 migrate_nr_uninterruptible(rq);
5793 BUG_ON(rq->nr_running != 0);
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02005794 calc_global_load_remove(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005795 break;
Gregory Haskins57d885f2008-01-25 21:08:18 +01005796
Gregory Haskins08f503b2008-03-10 17:59:11 -04005797 case CPU_DYING:
5798 case CPU_DYING_FROZEN:
Gregory Haskins57d885f2008-01-25 21:08:18 +01005799 /* Update our root-domain */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005800 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005801 if (rq->rd) {
Rusty Russellc6c49272008-11-25 02:35:05 +10305802 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005803 set_rq_offline(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005804 }
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005805 raw_spin_unlock_irqrestore(&rq->lock, flags);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005806 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005807#endif
5808 }
5809 return NOTIFY_OK;
5810}
5811
Paul Mackerrasf38b0822009-06-02 21:05:16 +10005812/*
5813 * Register at high priority so that task migration (migrate_all_tasks)
5814 * happens before everything else. This has to be lower priority than
Ingo Molnarcdd6c482009-09-21 12:02:48 +02005815 * the notifier in the perf_event subsystem, though.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005816 */
Chandra Seetharaman26c21432006-06-27 02:54:10 -07005817static struct notifier_block __cpuinitdata migration_notifier = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005818 .notifier_call = migration_call,
5819 .priority = 10
5820};
5821
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -07005822static int __init migration_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005823{
5824 void *cpu = (void *)(long)smp_processor_id();
Akinobu Mita07dccf32006-09-29 02:00:22 -07005825 int err;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005826
5827 /* Start one for the boot CPU: */
Akinobu Mita07dccf32006-09-29 02:00:22 -07005828 err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
5829 BUG_ON(err == NOTIFY_BAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005830 migration_call(&migration_notifier, CPU_ONLINE, cpu);
5831 register_cpu_notifier(&migration_notifier);
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -07005832
Thomas Gleixnera004cd42009-07-21 09:54:05 +02005833 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005834}
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -07005835early_initcall(migration_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005836#endif
5837
5838#ifdef CONFIG_SMP
Christoph Lameter476f3532007-05-06 14:48:58 -07005839
Ingo Molnar3e9830d2007-10-15 17:00:13 +02005840#ifdef CONFIG_SCHED_DEBUG
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005841
Mike Travisf6630112009-11-17 18:22:15 -06005842static __read_mostly int sched_domain_debug_enabled;
5843
5844static int __init sched_domain_debug_setup(char *str)
5845{
5846 sched_domain_debug_enabled = 1;
5847
5848 return 0;
5849}
5850early_param("sched_debug", sched_domain_debug_setup);
5851
Mike Travis7c16ec52008-04-04 18:11:11 -07005852static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
Rusty Russell96f874e2008-11-25 02:35:14 +10305853 struct cpumask *groupmask)
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005854{
5855 struct sched_group *group = sd->groups;
Mike Travis434d53b2008-04-04 18:11:04 -07005856 char str[256];
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005857
Rusty Russell968ea6d2008-12-13 21:55:51 +10305858 cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd));
Rusty Russell96f874e2008-11-25 02:35:14 +10305859 cpumask_clear(groupmask);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005860
5861 printk(KERN_DEBUG "%*s domain %d: ", level, "", level);
5862
5863 if (!(sd->flags & SD_LOAD_BALANCE)) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005864 printk("does not load-balance\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005865 if (sd->parent)
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005866 printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
5867 " has parent");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005868 return -1;
5869 }
5870
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005871 printk(KERN_CONT "span %s level %s\n", str, sd->name);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005872
Rusty Russell758b2cd2008-11-25 02:35:04 +10305873 if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005874 printk(KERN_ERR "ERROR: domain->span does not contain "
5875 "CPU%d\n", cpu);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005876 }
Rusty Russell758b2cd2008-11-25 02:35:04 +10305877 if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005878 printk(KERN_ERR "ERROR: domain->groups does not contain"
5879 " CPU%d\n", cpu);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005880 }
5881
5882 printk(KERN_DEBUG "%*s groups:", level + 1, "");
5883 do {
5884 if (!group) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005885 printk("\n");
5886 printk(KERN_ERR "ERROR: group is NULL\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005887 break;
5888 }
5889
Peter Zijlstra18a38852009-09-01 10:34:39 +02005890 if (!group->cpu_power) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005891 printk(KERN_CONT "\n");
5892 printk(KERN_ERR "ERROR: domain->cpu_power not "
5893 "set\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005894 break;
5895 }
5896
Rusty Russell758b2cd2008-11-25 02:35:04 +10305897 if (!cpumask_weight(sched_group_cpus(group))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005898 printk(KERN_CONT "\n");
5899 printk(KERN_ERR "ERROR: empty group\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005900 break;
5901 }
5902
Rusty Russell758b2cd2008-11-25 02:35:04 +10305903 if (cpumask_intersects(groupmask, sched_group_cpus(group))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005904 printk(KERN_CONT "\n");
5905 printk(KERN_ERR "ERROR: repeated CPUs\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005906 break;
5907 }
5908
Rusty Russell758b2cd2008-11-25 02:35:04 +10305909 cpumask_or(groupmask, groupmask, sched_group_cpus(group));
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005910
Rusty Russell968ea6d2008-12-13 21:55:51 +10305911 cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
Gautham R Shenoy381512c2009-04-14 09:09:36 +05305912
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005913 printk(KERN_CONT " %s", str);
Peter Zijlstra18a38852009-09-01 10:34:39 +02005914 if (group->cpu_power != SCHED_LOAD_SCALE) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005915 printk(KERN_CONT " (cpu_power = %d)",
5916 group->cpu_power);
Gautham R Shenoy381512c2009-04-14 09:09:36 +05305917 }
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005918
5919 group = group->next;
5920 } while (group != sd->groups);
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005921 printk(KERN_CONT "\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005922
Rusty Russell758b2cd2008-11-25 02:35:04 +10305923 if (!cpumask_equal(sched_domain_span(sd), groupmask))
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005924 printk(KERN_ERR "ERROR: groups don't span domain->span\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005925
Rusty Russell758b2cd2008-11-25 02:35:04 +10305926 if (sd->parent &&
5927 !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005928 printk(KERN_ERR "ERROR: parent span is not a superset "
5929 "of domain->span\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005930 return 0;
5931}
5932
Linus Torvalds1da177e2005-04-16 15:20:36 -07005933static void sched_domain_debug(struct sched_domain *sd, int cpu)
5934{
Rusty Russelld5dd3db2008-11-25 02:35:12 +10305935 cpumask_var_t groupmask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005936 int level = 0;
5937
Mike Travisf6630112009-11-17 18:22:15 -06005938 if (!sched_domain_debug_enabled)
5939 return;
5940
Nick Piggin41c7ce92005-06-25 14:57:24 -07005941 if (!sd) {
5942 printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
5943 return;
5944 }
5945
Linus Torvalds1da177e2005-04-16 15:20:36 -07005946 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
5947
Rusty Russelld5dd3db2008-11-25 02:35:12 +10305948 if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) {
Mike Travis7c16ec52008-04-04 18:11:11 -07005949 printk(KERN_DEBUG "Cannot load-balance (out of memory)\n");
5950 return;
5951 }
5952
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005953 for (;;) {
Mike Travis7c16ec52008-04-04 18:11:11 -07005954 if (sched_domain_debug_one(sd, cpu, level, groupmask))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005955 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005956 level++;
5957 sd = sd->parent;
Miguel Ojeda Sandonis33859f72006-12-10 02:20:38 -08005958 if (!sd)
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005959 break;
5960 }
Rusty Russelld5dd3db2008-11-25 02:35:12 +10305961 free_cpumask_var(groupmask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005962}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02005963#else /* !CONFIG_SCHED_DEBUG */
Ingo Molnar48f24c42006-07-03 00:25:40 -07005964# define sched_domain_debug(sd, cpu) do { } while (0)
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02005965#endif /* CONFIG_SCHED_DEBUG */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005966
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07005967static int sd_degenerate(struct sched_domain *sd)
Suresh Siddha245af2c2005-06-25 14:57:25 -07005968{
Rusty Russell758b2cd2008-11-25 02:35:04 +10305969 if (cpumask_weight(sched_domain_span(sd)) == 1)
Suresh Siddha245af2c2005-06-25 14:57:25 -07005970 return 1;
5971
5972 /* Following flags need at least 2 groups */
5973 if (sd->flags & (SD_LOAD_BALANCE |
5974 SD_BALANCE_NEWIDLE |
5975 SD_BALANCE_FORK |
Siddha, Suresh B89c47102006-10-03 01:14:09 -07005976 SD_BALANCE_EXEC |
5977 SD_SHARE_CPUPOWER |
5978 SD_SHARE_PKG_RESOURCES)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07005979 if (sd->groups != sd->groups->next)
5980 return 0;
5981 }
5982
5983 /* Following flags don't use groups */
Peter Zijlstrac88d5912009-09-10 13:50:02 +02005984 if (sd->flags & (SD_WAKE_AFFINE))
Suresh Siddha245af2c2005-06-25 14:57:25 -07005985 return 0;
5986
5987 return 1;
5988}
5989
Ingo Molnar48f24c42006-07-03 00:25:40 -07005990static int
5991sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
Suresh Siddha245af2c2005-06-25 14:57:25 -07005992{
5993 unsigned long cflags = sd->flags, pflags = parent->flags;
5994
5995 if (sd_degenerate(parent))
5996 return 1;
5997
Rusty Russell758b2cd2008-11-25 02:35:04 +10305998 if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
Suresh Siddha245af2c2005-06-25 14:57:25 -07005999 return 0;
6000
Suresh Siddha245af2c2005-06-25 14:57:25 -07006001 /* Flags needing groups don't count if only 1 group in parent */
6002 if (parent->groups == parent->groups->next) {
6003 pflags &= ~(SD_LOAD_BALANCE |
6004 SD_BALANCE_NEWIDLE |
6005 SD_BALANCE_FORK |
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006006 SD_BALANCE_EXEC |
6007 SD_SHARE_CPUPOWER |
6008 SD_SHARE_PKG_RESOURCES);
Ken Chen54364992008-12-07 18:47:37 -08006009 if (nr_node_ids == 1)
6010 pflags &= ~SD_SERIALIZE;
Suresh Siddha245af2c2005-06-25 14:57:25 -07006011 }
6012 if (~cflags & pflags)
6013 return 0;
6014
6015 return 1;
6016}
6017
Rusty Russellc6c49272008-11-25 02:35:05 +10306018static void free_rootdomain(struct root_domain *rd)
6019{
Peter Zijlstra047106a2009-11-16 10:28:09 +01006020 synchronize_sched();
6021
Rusty Russell68e74562008-11-25 02:35:13 +10306022 cpupri_cleanup(&rd->cpupri);
6023
Rusty Russellc6c49272008-11-25 02:35:05 +10306024 free_cpumask_var(rd->rto_mask);
6025 free_cpumask_var(rd->online);
6026 free_cpumask_var(rd->span);
6027 kfree(rd);
6028}
6029
Gregory Haskins57d885f2008-01-25 21:08:18 +01006030static void rq_attach_root(struct rq *rq, struct root_domain *rd)
6031{
Ingo Molnara0490fa2009-02-12 11:35:40 +01006032 struct root_domain *old_rd = NULL;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006033 unsigned long flags;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006034
Thomas Gleixner05fa7852009-11-17 14:28:38 +01006035 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006036
6037 if (rq->rd) {
Ingo Molnara0490fa2009-02-12 11:35:40 +01006038 old_rd = rq->rd;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006039
Rusty Russellc6c49272008-11-25 02:35:05 +10306040 if (cpumask_test_cpu(rq->cpu, old_rd->online))
Gregory Haskins1f11eb62008-06-04 15:04:05 -04006041 set_rq_offline(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006042
Rusty Russellc6c49272008-11-25 02:35:05 +10306043 cpumask_clear_cpu(rq->cpu, old_rd->span);
Gregory Haskinsdc938522008-01-25 21:08:26 +01006044
Ingo Molnara0490fa2009-02-12 11:35:40 +01006045 /*
6046 * If we dont want to free the old_rt yet then
6047 * set old_rd to NULL to skip the freeing later
6048 * in this function:
6049 */
6050 if (!atomic_dec_and_test(&old_rd->refcount))
6051 old_rd = NULL;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006052 }
6053
6054 atomic_inc(&rd->refcount);
6055 rq->rd = rd;
6056
Rusty Russellc6c49272008-11-25 02:35:05 +10306057 cpumask_set_cpu(rq->cpu, rd->span);
Gregory Haskins00aec932009-07-30 10:57:23 -04006058 if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
Gregory Haskins1f11eb62008-06-04 15:04:05 -04006059 set_rq_online(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006060
Thomas Gleixner05fa7852009-11-17 14:28:38 +01006061 raw_spin_unlock_irqrestore(&rq->lock, flags);
Ingo Molnara0490fa2009-02-12 11:35:40 +01006062
6063 if (old_rd)
6064 free_rootdomain(old_rd);
Gregory Haskins57d885f2008-01-25 21:08:18 +01006065}
6066
Li Zefanfd5e1b52009-06-15 13:34:19 +08006067static int init_rootdomain(struct root_domain *rd, bool bootmem)
Gregory Haskins57d885f2008-01-25 21:08:18 +01006068{
Pekka Enberg36b7b6d2009-06-10 23:42:36 +03006069 gfp_t gfp = GFP_KERNEL;
6070
Gregory Haskins57d885f2008-01-25 21:08:18 +01006071 memset(rd, 0, sizeof(*rd));
6072
Pekka Enberg36b7b6d2009-06-10 23:42:36 +03006073 if (bootmem)
6074 gfp = GFP_NOWAIT;
Gregory Haskins6e0534f2008-05-12 21:21:01 +02006075
Pekka Enberg36b7b6d2009-06-10 23:42:36 +03006076 if (!alloc_cpumask_var(&rd->span, gfp))
Li Zefan0c910d22009-01-06 17:39:06 +08006077 goto out;
Pekka Enberg36b7b6d2009-06-10 23:42:36 +03006078 if (!alloc_cpumask_var(&rd->online, gfp))
Rusty Russellc6c49272008-11-25 02:35:05 +10306079 goto free_span;
Pekka Enberg36b7b6d2009-06-10 23:42:36 +03006080 if (!alloc_cpumask_var(&rd->rto_mask, gfp))
Rusty Russellc6c49272008-11-25 02:35:05 +10306081 goto free_online;
Gregory Haskins6e0534f2008-05-12 21:21:01 +02006082
Pekka Enberg0fb53022009-06-11 08:41:22 +03006083 if (cpupri_init(&rd->cpupri, bootmem) != 0)
Rusty Russell68e74562008-11-25 02:35:13 +10306084 goto free_rto_mask;
Rusty Russellc6c49272008-11-25 02:35:05 +10306085 return 0;
6086
Rusty Russell68e74562008-11-25 02:35:13 +10306087free_rto_mask:
6088 free_cpumask_var(rd->rto_mask);
Rusty Russellc6c49272008-11-25 02:35:05 +10306089free_online:
6090 free_cpumask_var(rd->online);
6091free_span:
6092 free_cpumask_var(rd->span);
Li Zefan0c910d22009-01-06 17:39:06 +08006093out:
Rusty Russellc6c49272008-11-25 02:35:05 +10306094 return -ENOMEM;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006095}
6096
6097static void init_defrootdomain(void)
6098{
Rusty Russellc6c49272008-11-25 02:35:05 +10306099 init_rootdomain(&def_root_domain, true);
6100
Gregory Haskins57d885f2008-01-25 21:08:18 +01006101 atomic_set(&def_root_domain.refcount, 1);
6102}
6103
Gregory Haskinsdc938522008-01-25 21:08:26 +01006104static struct root_domain *alloc_rootdomain(void)
Gregory Haskins57d885f2008-01-25 21:08:18 +01006105{
6106 struct root_domain *rd;
6107
6108 rd = kmalloc(sizeof(*rd), GFP_KERNEL);
6109 if (!rd)
6110 return NULL;
6111
Rusty Russellc6c49272008-11-25 02:35:05 +10306112 if (init_rootdomain(rd, false) != 0) {
6113 kfree(rd);
6114 return NULL;
6115 }
Gregory Haskins57d885f2008-01-25 21:08:18 +01006116
6117 return rd;
6118}
6119
Linus Torvalds1da177e2005-04-16 15:20:36 -07006120/*
Ingo Molnar0eab9142008-01-25 21:08:19 +01006121 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
Linus Torvalds1da177e2005-04-16 15:20:36 -07006122 * hold the hotplug lock.
6123 */
Ingo Molnar0eab9142008-01-25 21:08:19 +01006124static void
6125cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006126{
Ingo Molnar70b97a72006-07-03 00:25:42 -07006127 struct rq *rq = cpu_rq(cpu);
Suresh Siddha245af2c2005-06-25 14:57:25 -07006128 struct sched_domain *tmp;
6129
Peter Zijlstra669c55e2010-04-16 14:59:29 +02006130 for (tmp = sd; tmp; tmp = tmp->parent)
6131 tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
6132
Suresh Siddha245af2c2005-06-25 14:57:25 -07006133 /* Remove the sched domains which do not contribute to scheduling. */
Li Zefanf29c9b12008-11-06 09:45:16 +08006134 for (tmp = sd; tmp; ) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006135 struct sched_domain *parent = tmp->parent;
6136 if (!parent)
6137 break;
Li Zefanf29c9b12008-11-06 09:45:16 +08006138
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006139 if (sd_parent_degenerate(tmp, parent)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006140 tmp->parent = parent->parent;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006141 if (parent->parent)
6142 parent->parent->child = tmp;
Li Zefanf29c9b12008-11-06 09:45:16 +08006143 } else
6144 tmp = tmp->parent;
Suresh Siddha245af2c2005-06-25 14:57:25 -07006145 }
6146
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006147 if (sd && sd_degenerate(sd)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07006148 sd = sd->parent;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07006149 if (sd)
6150 sd->child = NULL;
6151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006152
6153 sched_domain_debug(sd, cpu);
6154
Gregory Haskins57d885f2008-01-25 21:08:18 +01006155 rq_attach_root(rq, rd);
Nick Piggin674311d2005-06-25 14:57:27 -07006156 rcu_assign_pointer(rq->sd, sd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006157}
6158
6159/* cpus with isolated domains */
Rusty Russelldcc30a32008-11-25 02:35:12 +10306160static cpumask_var_t cpu_isolated_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006161
6162/* Setup the mask of cpus configured for isolated domains */
6163static int __init isolated_cpu_setup(char *str)
6164{
Rusty Russellbdddd292009-12-02 14:09:16 +10306165 alloc_bootmem_cpumask_var(&cpu_isolated_map);
Rusty Russell968ea6d2008-12-13 21:55:51 +10306166 cpulist_parse(str, cpu_isolated_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006167 return 1;
6168}
6169
Ingo Molnar8927f492007-10-15 17:00:13 +02006170__setup("isolcpus=", isolated_cpu_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006171
6172/*
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006173 * init_sched_build_groups takes the cpumask we wish to span, and a pointer
6174 * to a function which identifies what group(along with sched group) a CPU
Rusty Russell96f874e2008-11-25 02:35:14 +10306175 * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids
6176 * (due to the fact that we keep track of groups covered with a struct cpumask).
Linus Torvalds1da177e2005-04-16 15:20:36 -07006177 *
6178 * init_sched_build_groups will build a circular linked list of the groups
6179 * covered by the given span, and will set each group's ->cpumask correctly,
6180 * and ->cpu_power to 0.
6181 */
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006182static void
Rusty Russell96f874e2008-11-25 02:35:14 +10306183init_sched_build_groups(const struct cpumask *span,
6184 const struct cpumask *cpu_map,
6185 int (*group_fn)(int cpu, const struct cpumask *cpu_map,
Mike Travis7c16ec52008-04-04 18:11:11 -07006186 struct sched_group **sg,
Rusty Russell96f874e2008-11-25 02:35:14 +10306187 struct cpumask *tmpmask),
6188 struct cpumask *covered, struct cpumask *tmpmask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006189{
6190 struct sched_group *first = NULL, *last = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006191 int i;
6192
Rusty Russell96f874e2008-11-25 02:35:14 +10306193 cpumask_clear(covered);
Mike Travis7c16ec52008-04-04 18:11:11 -07006194
Rusty Russellabcd0832008-11-25 02:35:02 +10306195 for_each_cpu(i, span) {
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006196 struct sched_group *sg;
Mike Travis7c16ec52008-04-04 18:11:11 -07006197 int group = group_fn(i, cpu_map, &sg, tmpmask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006198 int j;
6199
Rusty Russell758b2cd2008-11-25 02:35:04 +10306200 if (cpumask_test_cpu(i, covered))
Linus Torvalds1da177e2005-04-16 15:20:36 -07006201 continue;
6202
Rusty Russell758b2cd2008-11-25 02:35:04 +10306203 cpumask_clear(sched_group_cpus(sg));
Peter Zijlstra18a38852009-09-01 10:34:39 +02006204 sg->cpu_power = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006205
Rusty Russellabcd0832008-11-25 02:35:02 +10306206 for_each_cpu(j, span) {
Mike Travis7c16ec52008-04-04 18:11:11 -07006207 if (group_fn(j, cpu_map, NULL, tmpmask) != group)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006208 continue;
6209
Rusty Russell96f874e2008-11-25 02:35:14 +10306210 cpumask_set_cpu(j, covered);
Rusty Russell758b2cd2008-11-25 02:35:04 +10306211 cpumask_set_cpu(j, sched_group_cpus(sg));
Linus Torvalds1da177e2005-04-16 15:20:36 -07006212 }
6213 if (!first)
6214 first = sg;
6215 if (last)
6216 last->next = sg;
6217 last = sg;
6218 }
6219 last->next = first;
6220}
6221
John Hawkes9c1cfda2005-09-06 15:18:14 -07006222#define SD_NODES_PER_DOMAIN 16
Linus Torvalds1da177e2005-04-16 15:20:36 -07006223
John Hawkes9c1cfda2005-09-06 15:18:14 -07006224#ifdef CONFIG_NUMA
akpm@osdl.org198e2f12006-01-12 01:05:30 -08006225
John Hawkes9c1cfda2005-09-06 15:18:14 -07006226/**
6227 * find_next_best_node - find the next node to include in a sched_domain
6228 * @node: node whose sched_domain we're building
6229 * @used_nodes: nodes already in the sched_domain
6230 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006231 * Find the next node to include in a given scheduling domain. Simply
John Hawkes9c1cfda2005-09-06 15:18:14 -07006232 * finds the closest node not already in the @used_nodes map.
6233 *
6234 * Should use nodemask_t.
6235 */
Mike Travisc5f59f02008-04-04 18:11:10 -07006236static int find_next_best_node(int node, nodemask_t *used_nodes)
John Hawkes9c1cfda2005-09-06 15:18:14 -07006237{
6238 int i, n, val, min_val, best_node = 0;
6239
6240 min_val = INT_MAX;
6241
Mike Travis076ac2a2008-05-12 21:21:12 +02006242 for (i = 0; i < nr_node_ids; i++) {
John Hawkes9c1cfda2005-09-06 15:18:14 -07006243 /* Start at @node */
Mike Travis076ac2a2008-05-12 21:21:12 +02006244 n = (node + i) % nr_node_ids;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006245
6246 if (!nr_cpus_node(n))
6247 continue;
6248
6249 /* Skip already used nodes */
Mike Travisc5f59f02008-04-04 18:11:10 -07006250 if (node_isset(n, *used_nodes))
John Hawkes9c1cfda2005-09-06 15:18:14 -07006251 continue;
6252
6253 /* Simple min distance search */
6254 val = node_distance(node, n);
6255
6256 if (val < min_val) {
6257 min_val = val;
6258 best_node = n;
6259 }
6260 }
6261
Mike Travisc5f59f02008-04-04 18:11:10 -07006262 node_set(best_node, *used_nodes);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006263 return best_node;
6264}
6265
6266/**
6267 * sched_domain_node_span - get a cpumask for a node's sched_domain
6268 * @node: node whose cpumask we're constructing
Randy Dunlap73486722008-04-22 10:07:22 -07006269 * @span: resulting cpumask
John Hawkes9c1cfda2005-09-06 15:18:14 -07006270 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006271 * Given a node, construct a good cpumask for its sched_domain to span. It
John Hawkes9c1cfda2005-09-06 15:18:14 -07006272 * should be one that prevents unnecessary balancing, but also spreads tasks
6273 * out optimally.
6274 */
Rusty Russell96f874e2008-11-25 02:35:14 +10306275static void sched_domain_node_span(int node, struct cpumask *span)
John Hawkes9c1cfda2005-09-06 15:18:14 -07006276{
Mike Travisc5f59f02008-04-04 18:11:10 -07006277 nodemask_t used_nodes;
Ingo Molnar48f24c42006-07-03 00:25:40 -07006278 int i;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006279
Mike Travis6ca09df2008-12-31 18:08:45 -08006280 cpumask_clear(span);
Mike Travisc5f59f02008-04-04 18:11:10 -07006281 nodes_clear(used_nodes);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006282
Mike Travis6ca09df2008-12-31 18:08:45 -08006283 cpumask_or(span, span, cpumask_of_node(node));
Mike Travisc5f59f02008-04-04 18:11:10 -07006284 node_set(node, used_nodes);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006285
6286 for (i = 1; i < SD_NODES_PER_DOMAIN; i++) {
Mike Travisc5f59f02008-04-04 18:11:10 -07006287 int next_node = find_next_best_node(node, &used_nodes);
Ingo Molnar48f24c42006-07-03 00:25:40 -07006288
Mike Travis6ca09df2008-12-31 18:08:45 -08006289 cpumask_or(span, span, cpumask_of_node(next_node));
John Hawkes9c1cfda2005-09-06 15:18:14 -07006290 }
John Hawkes9c1cfda2005-09-06 15:18:14 -07006291}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006292#endif /* CONFIG_NUMA */
John Hawkes9c1cfda2005-09-06 15:18:14 -07006293
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006294int sched_smt_power_savings = 0, sched_mc_power_savings = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07006295
John Hawkes9c1cfda2005-09-06 15:18:14 -07006296/*
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306297 * The cpus mask in sched_group and sched_domain hangs off the end.
Ingo Molnar4200efd2009-05-19 09:22:19 +02006298 *
6299 * ( See the the comments in include/linux/sched.h:struct sched_group
6300 * and struct sched_domain. )
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306301 */
6302struct static_sched_group {
6303 struct sched_group sg;
6304 DECLARE_BITMAP(cpus, CONFIG_NR_CPUS);
6305};
6306
6307struct static_sched_domain {
6308 struct sched_domain sd;
6309 DECLARE_BITMAP(span, CONFIG_NR_CPUS);
6310};
6311
Andreas Herrmann49a02c52009-08-18 12:51:52 +02006312struct s_data {
6313#ifdef CONFIG_NUMA
6314 int sd_allnodes;
6315 cpumask_var_t domainspan;
6316 cpumask_var_t covered;
6317 cpumask_var_t notcovered;
6318#endif
6319 cpumask_var_t nodemask;
6320 cpumask_var_t this_sibling_map;
6321 cpumask_var_t this_core_map;
6322 cpumask_var_t send_covered;
6323 cpumask_var_t tmpmask;
6324 struct sched_group **sched_group_nodes;
6325 struct root_domain *rd;
6326};
6327
Andreas Herrmann2109b992009-08-18 12:53:00 +02006328enum s_alloc {
6329 sa_sched_groups = 0,
6330 sa_rootdomain,
6331 sa_tmpmask,
6332 sa_send_covered,
6333 sa_this_core_map,
6334 sa_this_sibling_map,
6335 sa_nodemask,
6336 sa_sched_group_nodes,
6337#ifdef CONFIG_NUMA
6338 sa_notcovered,
6339 sa_covered,
6340 sa_domainspan,
6341#endif
6342 sa_none,
6343};
6344
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306345/*
Ingo Molnar48f24c42006-07-03 00:25:40 -07006346 * SMT sched-domains:
John Hawkes9c1cfda2005-09-06 15:18:14 -07006347 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006348#ifdef CONFIG_SCHED_SMT
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306349static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains);
Tejun Heo1871e522009-10-29 22:34:13 +09006350static DEFINE_PER_CPU(struct static_sched_group, sched_groups);
Ingo Molnar48f24c42006-07-03 00:25:40 -07006351
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006352static int
Rusty Russell96f874e2008-11-25 02:35:14 +10306353cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map,
6354 struct sched_group **sg, struct cpumask *unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006355{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006356 if (sg)
Tejun Heo1871e522009-10-29 22:34:13 +09006357 *sg = &per_cpu(sched_groups, cpu).sg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006358 return cpu;
6359}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006360#endif /* CONFIG_SCHED_SMT */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006361
Ingo Molnar48f24c42006-07-03 00:25:40 -07006362/*
6363 * multi-core sched-domains:
6364 */
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006365#ifdef CONFIG_SCHED_MC
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306366static DEFINE_PER_CPU(struct static_sched_domain, core_domains);
6367static DEFINE_PER_CPU(struct static_sched_group, sched_group_core);
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006368#endif /* CONFIG_SCHED_MC */
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006369
6370#if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT)
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006371static int
Rusty Russell96f874e2008-11-25 02:35:14 +10306372cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
6373 struct sched_group **sg, struct cpumask *mask)
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006374{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006375 int group;
Mike Travis7c16ec52008-04-04 18:11:11 -07006376
Rusty Russellc69fc562009-03-13 14:49:46 +10306377 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
Rusty Russell96f874e2008-11-25 02:35:14 +10306378 group = cpumask_first(mask);
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006379 if (sg)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306380 *sg = &per_cpu(sched_group_core, group).sg;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006381 return group;
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006382}
6383#elif defined(CONFIG_SCHED_MC)
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006384static int
Rusty Russell96f874e2008-11-25 02:35:14 +10306385cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
6386 struct sched_group **sg, struct cpumask *unused)
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006387{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006388 if (sg)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306389 *sg = &per_cpu(sched_group_core, cpu).sg;
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006390 return cpu;
6391}
6392#endif
6393
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306394static DEFINE_PER_CPU(struct static_sched_domain, phys_domains);
6395static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys);
Ingo Molnar48f24c42006-07-03 00:25:40 -07006396
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006397static int
Rusty Russell96f874e2008-11-25 02:35:14 +10306398cpu_to_phys_group(int cpu, const struct cpumask *cpu_map,
6399 struct sched_group **sg, struct cpumask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006400{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006401 int group;
Ingo Molnar48f24c42006-07-03 00:25:40 -07006402#ifdef CONFIG_SCHED_MC
Mike Travis6ca09df2008-12-31 18:08:45 -08006403 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
Rusty Russell96f874e2008-11-25 02:35:14 +10306404 group = cpumask_first(mask);
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006405#elif defined(CONFIG_SCHED_SMT)
Rusty Russellc69fc562009-03-13 14:49:46 +10306406 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
Rusty Russell96f874e2008-11-25 02:35:14 +10306407 group = cpumask_first(mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006408#else
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006409 group = cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006410#endif
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006411 if (sg)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306412 *sg = &per_cpu(sched_group_phys, group).sg;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006413 return group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006414}
6415
6416#ifdef CONFIG_NUMA
John Hawkes9c1cfda2005-09-06 15:18:14 -07006417/*
6418 * The init_sched_build_groups can't handle what we want to do with node
6419 * groups, so roll our own. Now each node has its own list of groups which
6420 * gets dynamically allocated.
6421 */
Rusty Russell62ea9ce2009-01-11 01:04:16 +01006422static DEFINE_PER_CPU(struct static_sched_domain, node_domains);
Mike Travis434d53b2008-04-04 18:11:04 -07006423static struct sched_group ***sched_group_nodes_bycpu;
John Hawkes9c1cfda2005-09-06 15:18:14 -07006424
Rusty Russell62ea9ce2009-01-11 01:04:16 +01006425static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains);
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306426static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006427
Rusty Russell96f874e2008-11-25 02:35:14 +10306428static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map,
6429 struct sched_group **sg,
6430 struct cpumask *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006431{
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006432 int group;
6433
Mike Travis6ca09df2008-12-31 18:08:45 -08006434 cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map);
Rusty Russell96f874e2008-11-25 02:35:14 +10306435 group = cpumask_first(nodemask);
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006436
6437 if (sg)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306438 *sg = &per_cpu(sched_group_allnodes, group).sg;
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006439 return group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006440}
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006441
Siddha, Suresh B08069032006-03-27 01:15:23 -08006442static void init_numa_sched_groups_power(struct sched_group *group_head)
6443{
6444 struct sched_group *sg = group_head;
6445 int j;
6446
6447 if (!sg)
6448 return;
Andi Kleen3a5c3592007-10-15 17:00:14 +02006449 do {
Rusty Russell758b2cd2008-11-25 02:35:04 +10306450 for_each_cpu(j, sched_group_cpus(sg)) {
Andi Kleen3a5c3592007-10-15 17:00:14 +02006451 struct sched_domain *sd;
Siddha, Suresh B08069032006-03-27 01:15:23 -08006452
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306453 sd = &per_cpu(phys_domains, j).sd;
Miao Xie13318a72009-04-15 09:59:10 +08006454 if (j != group_first_cpu(sd->groups)) {
Andi Kleen3a5c3592007-10-15 17:00:14 +02006455 /*
6456 * Only add "power" once for each
6457 * physical package.
6458 */
6459 continue;
6460 }
6461
Peter Zijlstra18a38852009-09-01 10:34:39 +02006462 sg->cpu_power += sd->groups->cpu_power;
Siddha, Suresh B08069032006-03-27 01:15:23 -08006463 }
Andi Kleen3a5c3592007-10-15 17:00:14 +02006464 sg = sg->next;
6465 } while (sg != group_head);
Siddha, Suresh B08069032006-03-27 01:15:23 -08006466}
Andreas Herrmann0601a882009-08-18 13:01:11 +02006467
6468static int build_numa_sched_groups(struct s_data *d,
6469 const struct cpumask *cpu_map, int num)
6470{
6471 struct sched_domain *sd;
6472 struct sched_group *sg, *prev;
6473 int n, j;
6474
6475 cpumask_clear(d->covered);
6476 cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map);
6477 if (cpumask_empty(d->nodemask)) {
6478 d->sched_group_nodes[num] = NULL;
6479 goto out;
6480 }
6481
6482 sched_domain_node_span(num, d->domainspan);
6483 cpumask_and(d->domainspan, d->domainspan, cpu_map);
6484
6485 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
6486 GFP_KERNEL, num);
6487 if (!sg) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006488 printk(KERN_WARNING "Can not alloc domain group for node %d\n",
6489 num);
Andreas Herrmann0601a882009-08-18 13:01:11 +02006490 return -ENOMEM;
6491 }
6492 d->sched_group_nodes[num] = sg;
6493
6494 for_each_cpu(j, d->nodemask) {
6495 sd = &per_cpu(node_domains, j).sd;
6496 sd->groups = sg;
6497 }
6498
Peter Zijlstra18a38852009-09-01 10:34:39 +02006499 sg->cpu_power = 0;
Andreas Herrmann0601a882009-08-18 13:01:11 +02006500 cpumask_copy(sched_group_cpus(sg), d->nodemask);
6501 sg->next = sg;
6502 cpumask_or(d->covered, d->covered, d->nodemask);
6503
6504 prev = sg;
6505 for (j = 0; j < nr_node_ids; j++) {
6506 n = (num + j) % nr_node_ids;
6507 cpumask_complement(d->notcovered, d->covered);
6508 cpumask_and(d->tmpmask, d->notcovered, cpu_map);
6509 cpumask_and(d->tmpmask, d->tmpmask, d->domainspan);
6510 if (cpumask_empty(d->tmpmask))
6511 break;
6512 cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n));
6513 if (cpumask_empty(d->tmpmask))
6514 continue;
6515 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
6516 GFP_KERNEL, num);
6517 if (!sg) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006518 printk(KERN_WARNING
6519 "Can not alloc domain group for node %d\n", j);
Andreas Herrmann0601a882009-08-18 13:01:11 +02006520 return -ENOMEM;
6521 }
Peter Zijlstra18a38852009-09-01 10:34:39 +02006522 sg->cpu_power = 0;
Andreas Herrmann0601a882009-08-18 13:01:11 +02006523 cpumask_copy(sched_group_cpus(sg), d->tmpmask);
6524 sg->next = prev->next;
6525 cpumask_or(d->covered, d->covered, d->tmpmask);
6526 prev->next = sg;
6527 prev = sg;
6528 }
6529out:
6530 return 0;
6531}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006532#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006533
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006534#ifdef CONFIG_NUMA
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006535/* Free memory allocated for various sched_group structures */
Rusty Russell96f874e2008-11-25 02:35:14 +10306536static void free_sched_groups(const struct cpumask *cpu_map,
6537 struct cpumask *nodemask)
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006538{
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006539 int cpu, i;
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006540
Rusty Russellabcd0832008-11-25 02:35:02 +10306541 for_each_cpu(cpu, cpu_map) {
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006542 struct sched_group **sched_group_nodes
6543 = sched_group_nodes_bycpu[cpu];
6544
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006545 if (!sched_group_nodes)
6546 continue;
6547
Mike Travis076ac2a2008-05-12 21:21:12 +02006548 for (i = 0; i < nr_node_ids; i++) {
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006549 struct sched_group *oldsg, *sg = sched_group_nodes[i];
6550
Mike Travis6ca09df2008-12-31 18:08:45 -08006551 cpumask_and(nodemask, cpumask_of_node(i), cpu_map);
Rusty Russell96f874e2008-11-25 02:35:14 +10306552 if (cpumask_empty(nodemask))
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006553 continue;
6554
6555 if (sg == NULL)
6556 continue;
6557 sg = sg->next;
6558next_sg:
6559 oldsg = sg;
6560 sg = sg->next;
6561 kfree(oldsg);
6562 if (oldsg != sched_group_nodes[i])
6563 goto next_sg;
6564 }
6565 kfree(sched_group_nodes);
6566 sched_group_nodes_bycpu[cpu] = NULL;
6567 }
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006568}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006569#else /* !CONFIG_NUMA */
Rusty Russell96f874e2008-11-25 02:35:14 +10306570static void free_sched_groups(const struct cpumask *cpu_map,
6571 struct cpumask *nodemask)
Siddha, Suresh Ba6160582006-10-03 01:14:06 -07006572{
6573}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006574#endif /* CONFIG_NUMA */
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006575
Linus Torvalds1da177e2005-04-16 15:20:36 -07006576/*
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006577 * Initialize sched groups cpu_power.
6578 *
6579 * cpu_power indicates the capacity of sched group, which is used while
6580 * distributing the load between different sched groups in a sched domain.
6581 * Typically cpu_power for all the groups in a sched domain will be same unless
6582 * there are asymmetries in the topology. If there are asymmetries, group
6583 * having more cpu_power will pickup more load compared to the group having
6584 * less cpu_power.
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006585 */
6586static void init_sched_groups_power(int cpu, struct sched_domain *sd)
6587{
6588 struct sched_domain *child;
6589 struct sched_group *group;
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006590 long power;
6591 int weight;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006592
6593 WARN_ON(!sd || !sd->groups);
6594
Miao Xie13318a72009-04-15 09:59:10 +08006595 if (cpu != group_first_cpu(sd->groups))
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006596 return;
6597
6598 child = sd->child;
6599
Peter Zijlstra18a38852009-09-01 10:34:39 +02006600 sd->groups->cpu_power = 0;
Eric Dumazet5517d862007-05-08 00:32:57 -07006601
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006602 if (!child) {
6603 power = SCHED_LOAD_SCALE;
6604 weight = cpumask_weight(sched_domain_span(sd));
6605 /*
6606 * SMT siblings share the power of a single core.
Peter Zijlstraa52bfd732009-09-01 10:34:35 +02006607 * Usually multiple threads get a better yield out of
6608 * that one core than a single thread would have,
6609 * reflect that in sd->smt_gain.
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006610 */
Peter Zijlstraa52bfd732009-09-01 10:34:35 +02006611 if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) {
6612 power *= sd->smt_gain;
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006613 power /= weight;
Peter Zijlstraa52bfd732009-09-01 10:34:35 +02006614 power >>= SCHED_LOAD_SHIFT;
6615 }
Peter Zijlstra18a38852009-09-01 10:34:39 +02006616 sd->groups->cpu_power += power;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006617 return;
6618 }
6619
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006620 /*
Peter Zijlstraf93e65c2009-09-01 10:34:32 +02006621 * Add cpu_power of each child group to this groups cpu_power.
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006622 */
6623 group = child->groups;
6624 do {
Peter Zijlstra18a38852009-09-01 10:34:39 +02006625 sd->groups->cpu_power += group->cpu_power;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006626 group = group->next;
6627 } while (group != child->groups);
6628}
6629
6630/*
Mike Travis7c16ec52008-04-04 18:11:11 -07006631 * Initializers for schedule domains
6632 * Non-inlined to reduce accumulated stack pressure in build_sched_domains()
6633 */
6634
Ingo Molnara5d8c342008-10-09 11:35:51 +02006635#ifdef CONFIG_SCHED_DEBUG
6636# define SD_INIT_NAME(sd, type) sd->name = #type
6637#else
6638# define SD_INIT_NAME(sd, type) do { } while (0)
6639#endif
6640
Mike Travis7c16ec52008-04-04 18:11:11 -07006641#define SD_INIT(sd, type) sd_init_##type(sd)
Ingo Molnara5d8c342008-10-09 11:35:51 +02006642
Mike Travis7c16ec52008-04-04 18:11:11 -07006643#define SD_INIT_FUNC(type) \
6644static noinline void sd_init_##type(struct sched_domain *sd) \
6645{ \
6646 memset(sd, 0, sizeof(*sd)); \
6647 *sd = SD_##type##_INIT; \
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006648 sd->level = SD_LV_##type; \
Ingo Molnara5d8c342008-10-09 11:35:51 +02006649 SD_INIT_NAME(sd, type); \
Mike Travis7c16ec52008-04-04 18:11:11 -07006650}
6651
6652SD_INIT_FUNC(CPU)
6653#ifdef CONFIG_NUMA
6654 SD_INIT_FUNC(ALLNODES)
6655 SD_INIT_FUNC(NODE)
6656#endif
6657#ifdef CONFIG_SCHED_SMT
6658 SD_INIT_FUNC(SIBLING)
6659#endif
6660#ifdef CONFIG_SCHED_MC
6661 SD_INIT_FUNC(MC)
6662#endif
6663
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006664static int default_relax_domain_level = -1;
6665
6666static int __init setup_relax_domain_level(char *str)
6667{
Li Zefan30e0e172008-05-13 10:27:17 +08006668 unsigned long val;
6669
6670 val = simple_strtoul(str, NULL, 0);
6671 if (val < SD_LV_MAX)
6672 default_relax_domain_level = val;
6673
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006674 return 1;
6675}
6676__setup("relax_domain_level=", setup_relax_domain_level);
6677
6678static void set_domain_attribute(struct sched_domain *sd,
6679 struct sched_domain_attr *attr)
6680{
6681 int request;
6682
6683 if (!attr || attr->relax_domain_level < 0) {
6684 if (default_relax_domain_level < 0)
6685 return;
6686 else
6687 request = default_relax_domain_level;
6688 } else
6689 request = attr->relax_domain_level;
6690 if (request < sd->level) {
6691 /* turn off idle balance on this domain */
Peter Zijlstrac88d5912009-09-10 13:50:02 +02006692 sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006693 } else {
6694 /* turn on idle balance on this domain */
Peter Zijlstrac88d5912009-09-10 13:50:02 +02006695 sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006696 }
6697}
6698
Andreas Herrmann2109b992009-08-18 12:53:00 +02006699static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
6700 const struct cpumask *cpu_map)
6701{
6702 switch (what) {
6703 case sa_sched_groups:
6704 free_sched_groups(cpu_map, d->tmpmask); /* fall through */
6705 d->sched_group_nodes = NULL;
6706 case sa_rootdomain:
6707 free_rootdomain(d->rd); /* fall through */
6708 case sa_tmpmask:
6709 free_cpumask_var(d->tmpmask); /* fall through */
6710 case sa_send_covered:
6711 free_cpumask_var(d->send_covered); /* fall through */
6712 case sa_this_core_map:
6713 free_cpumask_var(d->this_core_map); /* fall through */
6714 case sa_this_sibling_map:
6715 free_cpumask_var(d->this_sibling_map); /* fall through */
6716 case sa_nodemask:
6717 free_cpumask_var(d->nodemask); /* fall through */
6718 case sa_sched_group_nodes:
6719#ifdef CONFIG_NUMA
6720 kfree(d->sched_group_nodes); /* fall through */
6721 case sa_notcovered:
6722 free_cpumask_var(d->notcovered); /* fall through */
6723 case sa_covered:
6724 free_cpumask_var(d->covered); /* fall through */
6725 case sa_domainspan:
6726 free_cpumask_var(d->domainspan); /* fall through */
6727#endif
6728 case sa_none:
6729 break;
6730 }
6731}
6732
6733static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
6734 const struct cpumask *cpu_map)
6735{
6736#ifdef CONFIG_NUMA
6737 if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL))
6738 return sa_none;
6739 if (!alloc_cpumask_var(&d->covered, GFP_KERNEL))
6740 return sa_domainspan;
6741 if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL))
6742 return sa_covered;
6743 /* Allocate the per-node list of sched groups */
6744 d->sched_group_nodes = kcalloc(nr_node_ids,
6745 sizeof(struct sched_group *), GFP_KERNEL);
6746 if (!d->sched_group_nodes) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006747 printk(KERN_WARNING "Can not alloc sched group node list\n");
Andreas Herrmann2109b992009-08-18 12:53:00 +02006748 return sa_notcovered;
6749 }
6750 sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes;
6751#endif
6752 if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL))
6753 return sa_sched_group_nodes;
6754 if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL))
6755 return sa_nodemask;
6756 if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL))
6757 return sa_this_sibling_map;
6758 if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL))
6759 return sa_this_core_map;
6760 if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL))
6761 return sa_send_covered;
6762 d->rd = alloc_rootdomain();
6763 if (!d->rd) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006764 printk(KERN_WARNING "Cannot alloc root domain\n");
Andreas Herrmann2109b992009-08-18 12:53:00 +02006765 return sa_tmpmask;
6766 }
6767 return sa_rootdomain;
6768}
6769
Andreas Herrmann7f4588f2009-08-18 12:54:06 +02006770static struct sched_domain *__build_numa_sched_domains(struct s_data *d,
6771 const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i)
6772{
6773 struct sched_domain *sd = NULL;
6774#ifdef CONFIG_NUMA
6775 struct sched_domain *parent;
6776
6777 d->sd_allnodes = 0;
6778 if (cpumask_weight(cpu_map) >
6779 SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) {
6780 sd = &per_cpu(allnodes_domains, i).sd;
6781 SD_INIT(sd, ALLNODES);
6782 set_domain_attribute(sd, attr);
6783 cpumask_copy(sched_domain_span(sd), cpu_map);
6784 cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask);
6785 d->sd_allnodes = 1;
6786 }
6787 parent = sd;
6788
6789 sd = &per_cpu(node_domains, i).sd;
6790 SD_INIT(sd, NODE);
6791 set_domain_attribute(sd, attr);
6792 sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd));
6793 sd->parent = parent;
6794 if (parent)
6795 parent->child = sd;
6796 cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map);
6797#endif
6798 return sd;
6799}
6800
Andreas Herrmann87cce662009-08-18 12:54:55 +02006801static struct sched_domain *__build_cpu_sched_domain(struct s_data *d,
6802 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
6803 struct sched_domain *parent, int i)
6804{
6805 struct sched_domain *sd;
6806 sd = &per_cpu(phys_domains, i).sd;
6807 SD_INIT(sd, CPU);
6808 set_domain_attribute(sd, attr);
6809 cpumask_copy(sched_domain_span(sd), d->nodemask);
6810 sd->parent = parent;
6811 if (parent)
6812 parent->child = sd;
6813 cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask);
6814 return sd;
6815}
6816
Andreas Herrmann410c4082009-08-18 12:56:14 +02006817static struct sched_domain *__build_mc_sched_domain(struct s_data *d,
6818 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
6819 struct sched_domain *parent, int i)
6820{
6821 struct sched_domain *sd = parent;
6822#ifdef CONFIG_SCHED_MC
6823 sd = &per_cpu(core_domains, i).sd;
6824 SD_INIT(sd, MC);
6825 set_domain_attribute(sd, attr);
6826 cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i));
6827 sd->parent = parent;
6828 parent->child = sd;
6829 cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask);
6830#endif
6831 return sd;
6832}
6833
Andreas Herrmannd8173532009-08-18 12:57:03 +02006834static struct sched_domain *__build_smt_sched_domain(struct s_data *d,
6835 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
6836 struct sched_domain *parent, int i)
6837{
6838 struct sched_domain *sd = parent;
6839#ifdef CONFIG_SCHED_SMT
6840 sd = &per_cpu(cpu_domains, i).sd;
6841 SD_INIT(sd, SIBLING);
6842 set_domain_attribute(sd, attr);
6843 cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i));
6844 sd->parent = parent;
6845 parent->child = sd;
6846 cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask);
6847#endif
6848 return sd;
6849}
6850
Andreas Herrmann0e8e85c2009-08-18 12:57:51 +02006851static void build_sched_groups(struct s_data *d, enum sched_domain_level l,
6852 const struct cpumask *cpu_map, int cpu)
6853{
6854 switch (l) {
6855#ifdef CONFIG_SCHED_SMT
6856 case SD_LV_SIBLING: /* set up CPU (sibling) groups */
6857 cpumask_and(d->this_sibling_map, cpu_map,
6858 topology_thread_cpumask(cpu));
6859 if (cpu == cpumask_first(d->this_sibling_map))
6860 init_sched_build_groups(d->this_sibling_map, cpu_map,
6861 &cpu_to_cpu_group,
6862 d->send_covered, d->tmpmask);
6863 break;
6864#endif
Andreas Herrmanna2af04c2009-08-18 12:58:38 +02006865#ifdef CONFIG_SCHED_MC
6866 case SD_LV_MC: /* set up multi-core groups */
6867 cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu));
6868 if (cpu == cpumask_first(d->this_core_map))
6869 init_sched_build_groups(d->this_core_map, cpu_map,
6870 &cpu_to_core_group,
6871 d->send_covered, d->tmpmask);
6872 break;
6873#endif
Andreas Herrmann86548092009-08-18 12:59:28 +02006874 case SD_LV_CPU: /* set up physical groups */
6875 cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map);
6876 if (!cpumask_empty(d->nodemask))
6877 init_sched_build_groups(d->nodemask, cpu_map,
6878 &cpu_to_phys_group,
6879 d->send_covered, d->tmpmask);
6880 break;
Andreas Herrmannde616e32009-08-18 13:00:13 +02006881#ifdef CONFIG_NUMA
6882 case SD_LV_ALLNODES:
6883 init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group,
6884 d->send_covered, d->tmpmask);
6885 break;
6886#endif
Andreas Herrmann0e8e85c2009-08-18 12:57:51 +02006887 default:
6888 break;
6889 }
6890}
6891
Mike Travis7c16ec52008-04-04 18:11:11 -07006892/*
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006893 * Build sched domains for a given set of cpus and attach the sched domains
6894 * to the individual cpus
Linus Torvalds1da177e2005-04-16 15:20:36 -07006895 */
Rusty Russell96f874e2008-11-25 02:35:14 +10306896static int __build_sched_domains(const struct cpumask *cpu_map,
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006897 struct sched_domain_attr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006898{
Andreas Herrmann2109b992009-08-18 12:53:00 +02006899 enum s_alloc alloc_state = sa_none;
Andreas Herrmann49a02c52009-08-18 12:51:52 +02006900 struct s_data d;
Andreas Herrmann294b0c92009-08-18 13:02:29 +02006901 struct sched_domain *sd;
Andreas Herrmann2109b992009-08-18 12:53:00 +02006902 int i;
John Hawkesd1b55132005-09-06 15:18:14 -07006903#ifdef CONFIG_NUMA
Andreas Herrmann49a02c52009-08-18 12:51:52 +02006904 d.sd_allnodes = 0;
Rusty Russell3404c8d2008-11-25 02:35:03 +10306905#endif
6906
Andreas Herrmann2109b992009-08-18 12:53:00 +02006907 alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
6908 if (alloc_state != sa_rootdomain)
6909 goto error;
6910 alloc_state = sa_sched_groups;
Mike Travis7c16ec52008-04-04 18:11:11 -07006911
Linus Torvalds1da177e2005-04-16 15:20:36 -07006912 /*
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006913 * Set up domains for cpus specified by the cpu_map.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006914 */
Rusty Russellabcd0832008-11-25 02:35:02 +10306915 for_each_cpu(i, cpu_map) {
Andreas Herrmann49a02c52009-08-18 12:51:52 +02006916 cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)),
6917 cpu_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006918
Andreas Herrmann7f4588f2009-08-18 12:54:06 +02006919 sd = __build_numa_sched_domains(&d, cpu_map, attr, i);
Andreas Herrmann87cce662009-08-18 12:54:55 +02006920 sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i);
Andreas Herrmann410c4082009-08-18 12:56:14 +02006921 sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i);
Andreas Herrmannd8173532009-08-18 12:57:03 +02006922 sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006923 }
6924
Rusty Russellabcd0832008-11-25 02:35:02 +10306925 for_each_cpu(i, cpu_map) {
Andreas Herrmann0e8e85c2009-08-18 12:57:51 +02006926 build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i);
Andreas Herrmanna2af04c2009-08-18 12:58:38 +02006927 build_sched_groups(&d, SD_LV_MC, cpu_map, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006928 }
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006929
Linus Torvalds1da177e2005-04-16 15:20:36 -07006930 /* Set up physical groups */
Andreas Herrmann86548092009-08-18 12:59:28 +02006931 for (i = 0; i < nr_node_ids; i++)
6932 build_sched_groups(&d, SD_LV_CPU, cpu_map, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006933
6934#ifdef CONFIG_NUMA
6935 /* Set up node groups */
Andreas Herrmannde616e32009-08-18 13:00:13 +02006936 if (d.sd_allnodes)
6937 build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006938
Andreas Herrmann0601a882009-08-18 13:01:11 +02006939 for (i = 0; i < nr_node_ids; i++)
6940 if (build_numa_sched_groups(&d, cpu_map, i))
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006941 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006942#endif
6943
6944 /* Calculate CPU power for physical packages and nodes */
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006945#ifdef CONFIG_SCHED_SMT
Rusty Russellabcd0832008-11-25 02:35:02 +10306946 for_each_cpu(i, cpu_map) {
Andreas Herrmann294b0c92009-08-18 13:02:29 +02006947 sd = &per_cpu(cpu_domains, i).sd;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006948 init_sched_groups_power(i, sd);
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006949 }
6950#endif
6951#ifdef CONFIG_SCHED_MC
Rusty Russellabcd0832008-11-25 02:35:02 +10306952 for_each_cpu(i, cpu_map) {
Andreas Herrmann294b0c92009-08-18 13:02:29 +02006953 sd = &per_cpu(core_domains, i).sd;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006954 init_sched_groups_power(i, sd);
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07006955 }
6956#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07006957
Rusty Russellabcd0832008-11-25 02:35:02 +10306958 for_each_cpu(i, cpu_map) {
Andreas Herrmann294b0c92009-08-18 13:02:29 +02006959 sd = &per_cpu(phys_domains, i).sd;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07006960 init_sched_groups_power(i, sd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006961 }
6962
John Hawkes9c1cfda2005-09-06 15:18:14 -07006963#ifdef CONFIG_NUMA
Mike Travis076ac2a2008-05-12 21:21:12 +02006964 for (i = 0; i < nr_node_ids; i++)
Andreas Herrmann49a02c52009-08-18 12:51:52 +02006965 init_numa_sched_groups_power(d.sched_group_nodes[i]);
John Hawkes9c1cfda2005-09-06 15:18:14 -07006966
Andreas Herrmann49a02c52009-08-18 12:51:52 +02006967 if (d.sd_allnodes) {
Siddha, Suresh B6711cab2006-12-10 02:20:07 -08006968 struct sched_group *sg;
Siddha, Suresh Bf712c0c2006-07-30 03:02:59 -07006969
Rusty Russell96f874e2008-11-25 02:35:14 +10306970 cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg,
Andreas Herrmann49a02c52009-08-18 12:51:52 +02006971 d.tmpmask);
Siddha, Suresh Bf712c0c2006-07-30 03:02:59 -07006972 init_numa_sched_groups_power(sg);
6973 }
John Hawkes9c1cfda2005-09-06 15:18:14 -07006974#endif
6975
Linus Torvalds1da177e2005-04-16 15:20:36 -07006976 /* Attach the domains */
Rusty Russellabcd0832008-11-25 02:35:02 +10306977 for_each_cpu(i, cpu_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006978#ifdef CONFIG_SCHED_SMT
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306979 sd = &per_cpu(cpu_domains, i).sd;
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08006980#elif defined(CONFIG_SCHED_MC)
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306981 sd = &per_cpu(core_domains, i).sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006982#else
Rusty Russell6c99e9a2008-11-25 02:35:04 +10306983 sd = &per_cpu(phys_domains, i).sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006984#endif
Andreas Herrmann49a02c52009-08-18 12:51:52 +02006985 cpu_attach_domain(sd, d.rd, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006986 }
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006987
Andreas Herrmann2109b992009-08-18 12:53:00 +02006988 d.sched_group_nodes = NULL; /* don't free this we still need it */
6989 __free_domain_allocs(&d, sa_tmpmask, cpu_map);
6990 return 0;
Rusty Russell3404c8d2008-11-25 02:35:03 +10306991
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006992error:
Andreas Herrmann2109b992009-08-18 12:53:00 +02006993 __free_domain_allocs(&d, alloc_state, cpu_map);
6994 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006995}
Paul Jackson029190c2007-10-18 23:40:20 -07006996
Rusty Russell96f874e2008-11-25 02:35:14 +10306997static int build_sched_domains(const struct cpumask *cpu_map)
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006998{
6999 return __build_sched_domains(cpu_map, NULL);
7000}
7001
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307002static cpumask_var_t *doms_cur; /* current sched domains */
Paul Jackson029190c2007-10-18 23:40:20 -07007003static int ndoms_cur; /* number of sched domains in 'doms_cur' */
Ingo Molnar4285f5942008-05-16 17:47:14 +02007004static struct sched_domain_attr *dattr_cur;
7005 /* attribues of custom domains in 'doms_cur' */
Paul Jackson029190c2007-10-18 23:40:20 -07007006
7007/*
7008 * Special case: If a kmalloc of a doms_cur partition (array of
Rusty Russell42128232008-11-25 02:35:12 +10307009 * cpumask) fails, then fallback to a single sched domain,
7010 * as determined by the single cpumask fallback_doms.
Paul Jackson029190c2007-10-18 23:40:20 -07007011 */
Rusty Russell42128232008-11-25 02:35:12 +10307012static cpumask_var_t fallback_doms;
Paul Jackson029190c2007-10-18 23:40:20 -07007013
Heiko Carstensee79d1b2008-12-09 18:49:50 +01007014/*
7015 * arch_update_cpu_topology lets virtualized architectures update the
7016 * cpu core maps. It is supposed to return 1 if the topology changed
7017 * or 0 if it stayed the same.
7018 */
7019int __attribute__((weak)) arch_update_cpu_topology(void)
Heiko Carstens22e52b02008-03-12 18:31:59 +01007020{
Heiko Carstensee79d1b2008-12-09 18:49:50 +01007021 return 0;
Heiko Carstens22e52b02008-03-12 18:31:59 +01007022}
7023
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307024cpumask_var_t *alloc_sched_domains(unsigned int ndoms)
7025{
7026 int i;
7027 cpumask_var_t *doms;
7028
7029 doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL);
7030 if (!doms)
7031 return NULL;
7032 for (i = 0; i < ndoms; i++) {
7033 if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) {
7034 free_sched_domains(doms, i);
7035 return NULL;
7036 }
7037 }
7038 return doms;
7039}
7040
7041void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms)
7042{
7043 unsigned int i;
7044 for (i = 0; i < ndoms; i++)
7045 free_cpumask_var(doms[i]);
7046 kfree(doms);
7047}
7048
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007049/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007050 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
Paul Jackson029190c2007-10-18 23:40:20 -07007051 * For now this just excludes isolated cpus, but could be used to
7052 * exclude other special cases in the future.
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007053 */
Rusty Russell96f874e2008-11-25 02:35:14 +10307054static int arch_init_sched_domains(const struct cpumask *cpu_map)
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007055{
Milton Miller73785472007-10-24 18:23:48 +02007056 int err;
7057
Heiko Carstens22e52b02008-03-12 18:31:59 +01007058 arch_update_cpu_topology();
Paul Jackson029190c2007-10-18 23:40:20 -07007059 ndoms_cur = 1;
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307060 doms_cur = alloc_sched_domains(ndoms_cur);
Paul Jackson029190c2007-10-18 23:40:20 -07007061 if (!doms_cur)
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307062 doms_cur = &fallback_doms;
7063 cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007064 dattr_cur = NULL;
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307065 err = build_sched_domains(doms_cur[0]);
Milton Miller6382bc92007-10-15 17:00:19 +02007066 register_sched_domain_sysctl();
Milton Miller73785472007-10-24 18:23:48 +02007067
7068 return err;
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007069}
7070
Rusty Russell96f874e2008-11-25 02:35:14 +10307071static void arch_destroy_sched_domains(const struct cpumask *cpu_map,
7072 struct cpumask *tmpmask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007073{
Mike Travis7c16ec52008-04-04 18:11:11 -07007074 free_sched_groups(cpu_map, tmpmask);
John Hawkes9c1cfda2005-09-06 15:18:14 -07007075}
Linus Torvalds1da177e2005-04-16 15:20:36 -07007076
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007077/*
7078 * Detach sched domains from a group of cpus specified in cpu_map
7079 * These cpus will now be attached to the NULL domain
7080 */
Rusty Russell96f874e2008-11-25 02:35:14 +10307081static void detach_destroy_domains(const struct cpumask *cpu_map)
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007082{
Rusty Russell96f874e2008-11-25 02:35:14 +10307083 /* Save because hotplug lock held. */
7084 static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS);
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007085 int i;
7086
Rusty Russellabcd0832008-11-25 02:35:02 +10307087 for_each_cpu(i, cpu_map)
Gregory Haskins57d885f2008-01-25 21:08:18 +01007088 cpu_attach_domain(NULL, &def_root_domain, i);
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007089 synchronize_sched();
Rusty Russell96f874e2008-11-25 02:35:14 +10307090 arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask));
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07007091}
7092
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007093/* handle null as "default" */
7094static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
7095 struct sched_domain_attr *new, int idx_new)
7096{
7097 struct sched_domain_attr tmp;
7098
7099 /* fast path */
7100 if (!new && !cur)
7101 return 1;
7102
7103 tmp = SD_ATTR_INIT;
7104 return !memcmp(cur ? (cur + idx_cur) : &tmp,
7105 new ? (new + idx_new) : &tmp,
7106 sizeof(struct sched_domain_attr));
7107}
7108
Paul Jackson029190c2007-10-18 23:40:20 -07007109/*
7110 * Partition sched domains as specified by the 'ndoms_new'
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007111 * cpumasks in the array doms_new[] of cpumasks. This compares
Paul Jackson029190c2007-10-18 23:40:20 -07007112 * doms_new[] to the current sched domain partitioning, doms_cur[].
7113 * It destroys each deleted domain and builds each new domain.
7114 *
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307115 * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'.
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007116 * The masks don't intersect (don't overlap.) We should setup one
7117 * sched domain for each mask. CPUs not in any of the cpumasks will
7118 * not be load balanced. If the same cpumask appears both in the
Paul Jackson029190c2007-10-18 23:40:20 -07007119 * current 'doms_cur' domains and in the new 'doms_new', we can leave
7120 * it as it is.
7121 *
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307122 * The passed in 'doms_new' should be allocated using
7123 * alloc_sched_domains. This routine takes ownership of it and will
7124 * free_sched_domains it when done with it. If the caller failed the
7125 * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1,
7126 * and partition_sched_domains() will fallback to the single partition
7127 * 'fallback_doms', it also forces the domains to be rebuilt.
Paul Jackson029190c2007-10-18 23:40:20 -07007128 *
Rusty Russell96f874e2008-11-25 02:35:14 +10307129 * If doms_new == NULL it will be replaced with cpu_online_mask.
Li Zefan700018e2008-11-18 14:02:03 +08007130 * ndoms_new == 0 is a special case for destroying existing domains,
7131 * and it will not create the default domain.
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07007132 *
Paul Jackson029190c2007-10-18 23:40:20 -07007133 * Call with hotplug lock held
7134 */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307135void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007136 struct sched_domain_attr *dattr_new)
Paul Jackson029190c2007-10-18 23:40:20 -07007137{
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07007138 int i, j, n;
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01007139 int new_topology;
Paul Jackson029190c2007-10-18 23:40:20 -07007140
Heiko Carstens712555e2008-04-28 11:33:07 +02007141 mutex_lock(&sched_domains_mutex);
Srivatsa Vaddagiria1835612008-01-25 21:08:00 +01007142
Milton Miller73785472007-10-24 18:23:48 +02007143 /* always unregister in case we don't destroy any domains */
7144 unregister_sched_domain_sysctl();
7145
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01007146 /* Let architecture update cpu core mappings. */
7147 new_topology = arch_update_cpu_topology();
7148
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07007149 n = doms_new ? ndoms_new : 0;
Paul Jackson029190c2007-10-18 23:40:20 -07007150
7151 /* Destroy deleted domains */
7152 for (i = 0; i < ndoms_cur; i++) {
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01007153 for (j = 0; j < n && !new_topology; j++) {
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307154 if (cpumask_equal(doms_cur[i], doms_new[j])
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007155 && dattrs_equal(dattr_cur, i, dattr_new, j))
Paul Jackson029190c2007-10-18 23:40:20 -07007156 goto match1;
7157 }
7158 /* no match - a current sched domain not in new doms_new[] */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307159 detach_destroy_domains(doms_cur[i]);
Paul Jackson029190c2007-10-18 23:40:20 -07007160match1:
7161 ;
7162 }
7163
Max Krasnyanskye761b772008-07-15 04:43:49 -07007164 if (doms_new == NULL) {
7165 ndoms_cur = 0;
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307166 doms_new = &fallback_doms;
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01007167 cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
Li Zefanfaa2f982008-11-04 16:20:23 +08007168 WARN_ON_ONCE(dattr_new);
Max Krasnyanskye761b772008-07-15 04:43:49 -07007169 }
7170
Paul Jackson029190c2007-10-18 23:40:20 -07007171 /* Build new domains */
7172 for (i = 0; i < ndoms_new; i++) {
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01007173 for (j = 0; j < ndoms_cur && !new_topology; j++) {
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307174 if (cpumask_equal(doms_new[i], doms_cur[j])
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007175 && dattrs_equal(dattr_new, i, dattr_cur, j))
Paul Jackson029190c2007-10-18 23:40:20 -07007176 goto match2;
7177 }
7178 /* no match - add a new doms_new */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307179 __build_sched_domains(doms_new[i],
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007180 dattr_new ? dattr_new + i : NULL);
Paul Jackson029190c2007-10-18 23:40:20 -07007181match2:
7182 ;
7183 }
7184
7185 /* Remember the new sched domains */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10307186 if (doms_cur != &fallback_doms)
7187 free_sched_domains(doms_cur, ndoms_cur);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007188 kfree(dattr_cur); /* kfree(NULL) is safe */
Paul Jackson029190c2007-10-18 23:40:20 -07007189 doms_cur = doms_new;
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09007190 dattr_cur = dattr_new;
Paul Jackson029190c2007-10-18 23:40:20 -07007191 ndoms_cur = ndoms_new;
Milton Miller73785472007-10-24 18:23:48 +02007192
7193 register_sched_domain_sysctl();
Srivatsa Vaddagiria1835612008-01-25 21:08:00 +01007194
Heiko Carstens712555e2008-04-28 11:33:07 +02007195 mutex_unlock(&sched_domains_mutex);
Paul Jackson029190c2007-10-18 23:40:20 -07007196}
7197
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007198#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
Li Zefanc70f22d2009-01-05 19:07:50 +08007199static void arch_reinit_sched_domains(void)
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007200{
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007201 get_online_cpus();
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07007202
7203 /* Destroy domains first to force the rebuild */
7204 partition_sched_domains(0, NULL, NULL);
7205
Max Krasnyanskye761b772008-07-15 04:43:49 -07007206 rebuild_sched_domains();
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007207 put_online_cpus();
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007208}
7209
7210static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt)
7211{
Gautham R Shenoyafb8a9b2008-12-18 23:26:09 +05307212 unsigned int level = 0;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007213
Gautham R Shenoyafb8a9b2008-12-18 23:26:09 +05307214 if (sscanf(buf, "%u", &level) != 1)
7215 return -EINVAL;
7216
7217 /*
7218 * level is always be positive so don't check for
7219 * level < POWERSAVINGS_BALANCE_NONE which is 0
7220 * What happens on 0 or 1 byte write,
7221 * need to check for count as well?
7222 */
7223
7224 if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS)
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007225 return -EINVAL;
7226
7227 if (smt)
Gautham R Shenoyafb8a9b2008-12-18 23:26:09 +05307228 sched_smt_power_savings = level;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007229 else
Gautham R Shenoyafb8a9b2008-12-18 23:26:09 +05307230 sched_mc_power_savings = level;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007231
Li Zefanc70f22d2009-01-05 19:07:50 +08007232 arch_reinit_sched_domains();
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007233
Li Zefanc70f22d2009-01-05 19:07:50 +08007234 return count;
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007235}
7236
Adrian Bunk6707de002007-08-12 18:08:19 +02007237#ifdef CONFIG_SCHED_MC
Andi Kleenf718cd42008-07-29 22:33:52 -07007238static ssize_t sched_mc_power_savings_show(struct sysdev_class *class,
Andi Kleenc9be0a32010-01-05 12:47:58 +01007239 struct sysdev_class_attribute *attr,
Andi Kleenf718cd42008-07-29 22:33:52 -07007240 char *page)
Adrian Bunk6707de002007-08-12 18:08:19 +02007241{
7242 return sprintf(page, "%u\n", sched_mc_power_savings);
7243}
Andi Kleenf718cd42008-07-29 22:33:52 -07007244static ssize_t sched_mc_power_savings_store(struct sysdev_class *class,
Andi Kleenc9be0a32010-01-05 12:47:58 +01007245 struct sysdev_class_attribute *attr,
Adrian Bunk6707de002007-08-12 18:08:19 +02007246 const char *buf, size_t count)
7247{
7248 return sched_power_savings_store(buf, count, 0);
7249}
Andi Kleenf718cd42008-07-29 22:33:52 -07007250static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644,
7251 sched_mc_power_savings_show,
7252 sched_mc_power_savings_store);
Adrian Bunk6707de002007-08-12 18:08:19 +02007253#endif
7254
7255#ifdef CONFIG_SCHED_SMT
Andi Kleenf718cd42008-07-29 22:33:52 -07007256static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev,
Andi Kleenc9be0a32010-01-05 12:47:58 +01007257 struct sysdev_class_attribute *attr,
Andi Kleenf718cd42008-07-29 22:33:52 -07007258 char *page)
Adrian Bunk6707de002007-08-12 18:08:19 +02007259{
7260 return sprintf(page, "%u\n", sched_smt_power_savings);
7261}
Andi Kleenf718cd42008-07-29 22:33:52 -07007262static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev,
Andi Kleenc9be0a32010-01-05 12:47:58 +01007263 struct sysdev_class_attribute *attr,
Adrian Bunk6707de002007-08-12 18:08:19 +02007264 const char *buf, size_t count)
7265{
7266 return sched_power_savings_store(buf, count, 1);
7267}
Andi Kleenf718cd42008-07-29 22:33:52 -07007268static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644,
7269 sched_smt_power_savings_show,
Adrian Bunk6707de002007-08-12 18:08:19 +02007270 sched_smt_power_savings_store);
7271#endif
7272
Li Zefan39aac642009-01-05 19:18:02 +08007273int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls)
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007274{
7275 int err = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07007276
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007277#ifdef CONFIG_SCHED_SMT
7278 if (smt_capable())
7279 err = sysfs_create_file(&cls->kset.kobj,
7280 &attr_sched_smt_power_savings.attr);
7281#endif
7282#ifdef CONFIG_SCHED_MC
7283 if (!err && mc_capable())
7284 err = sysfs_create_file(&cls->kset.kobj,
7285 &attr_sched_mc_power_savings.attr);
7286#endif
7287 return err;
7288}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007289#endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */
Siddha, Suresh B5c45bf22006-06-27 02:54:42 -07007290
Max Krasnyanskye761b772008-07-15 04:43:49 -07007291#ifndef CONFIG_CPUSETS
Linus Torvalds1da177e2005-04-16 15:20:36 -07007292/*
Max Krasnyanskye761b772008-07-15 04:43:49 -07007293 * Add online and remove offline CPUs from the scheduler domains.
7294 * When cpusets are enabled they take over this function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07007295 */
7296static int update_sched_domains(struct notifier_block *nfb,
7297 unsigned long action, void *hcpu)
7298{
Max Krasnyanskye761b772008-07-15 04:43:49 -07007299 switch (action) {
7300 case CPU_ONLINE:
7301 case CPU_ONLINE_FROZEN:
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01007302 case CPU_DOWN_PREPARE:
7303 case CPU_DOWN_PREPARE_FROZEN:
7304 case CPU_DOWN_FAILED:
7305 case CPU_DOWN_FAILED_FROZEN:
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07007306 partition_sched_domains(1, NULL, NULL);
Max Krasnyanskye761b772008-07-15 04:43:49 -07007307 return NOTIFY_OK;
7308
7309 default:
7310 return NOTIFY_DONE;
7311 }
7312}
7313#endif
7314
7315static int update_runtime(struct notifier_block *nfb,
7316 unsigned long action, void *hcpu)
7317{
Peter Zijlstra7def2be2008-06-05 14:49:58 +02007318 int cpu = (int)(long)hcpu;
7319
Linus Torvalds1da177e2005-04-16 15:20:36 -07007320 switch (action) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007321 case CPU_DOWN_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007322 case CPU_DOWN_PREPARE_FROZEN:
Peter Zijlstra7def2be2008-06-05 14:49:58 +02007323 disable_runtime(cpu_rq(cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -07007324 return NOTIFY_OK;
7325
Linus Torvalds1da177e2005-04-16 15:20:36 -07007326 case CPU_DOWN_FAILED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007327 case CPU_DOWN_FAILED_FROZEN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07007328 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07007329 case CPU_ONLINE_FROZEN:
Peter Zijlstra7def2be2008-06-05 14:49:58 +02007330 enable_runtime(cpu_rq(cpu));
Max Krasnyanskye761b772008-07-15 04:43:49 -07007331 return NOTIFY_OK;
7332
Linus Torvalds1da177e2005-04-16 15:20:36 -07007333 default:
7334 return NOTIFY_DONE;
7335 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007336}
Linus Torvalds1da177e2005-04-16 15:20:36 -07007337
7338void __init sched_init_smp(void)
7339{
Rusty Russelldcc30a32008-11-25 02:35:12 +10307340 cpumask_var_t non_isolated_cpus;
7341
7342 alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
Yong Zhangcb5fd132009-09-14 20:20:16 +08007343 alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
Nick Piggin5c1e1762006-10-03 01:14:04 -07007344
Mike Travis434d53b2008-04-04 18:11:04 -07007345#if defined(CONFIG_NUMA)
7346 sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **),
7347 GFP_KERNEL);
7348 BUG_ON(sched_group_nodes_bycpu == NULL);
7349#endif
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007350 get_online_cpus();
Heiko Carstens712555e2008-04-28 11:33:07 +02007351 mutex_lock(&sched_domains_mutex);
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01007352 arch_init_sched_domains(cpu_active_mask);
Rusty Russelldcc30a32008-11-25 02:35:12 +10307353 cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map);
7354 if (cpumask_empty(non_isolated_cpus))
7355 cpumask_set_cpu(smp_processor_id(), non_isolated_cpus);
Heiko Carstens712555e2008-04-28 11:33:07 +02007356 mutex_unlock(&sched_domains_mutex);
Gautham R Shenoy95402b32008-01-25 21:08:02 +01007357 put_online_cpus();
Max Krasnyanskye761b772008-07-15 04:43:49 -07007358
7359#ifndef CONFIG_CPUSETS
Linus Torvalds1da177e2005-04-16 15:20:36 -07007360 /* XXX: Theoretical race here - CPU may be hotplugged now */
7361 hotcpu_notifier(update_sched_domains, 0);
Max Krasnyanskye761b772008-07-15 04:43:49 -07007362#endif
7363
7364 /* RT runtime code needs to handle some hotplug events */
7365 hotcpu_notifier(update_runtime, 0);
7366
Peter Zijlstrab328ca12008-04-29 10:02:46 +02007367 init_hrtick();
Nick Piggin5c1e1762006-10-03 01:14:04 -07007368
7369 /* Move init over to a non-isolated CPU */
Rusty Russelldcc30a32008-11-25 02:35:12 +10307370 if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
Nick Piggin5c1e1762006-10-03 01:14:04 -07007371 BUG();
Ingo Molnar19978ca2007-11-09 22:39:38 +01007372 sched_init_granularity();
Rusty Russelldcc30a32008-11-25 02:35:12 +10307373 free_cpumask_var(non_isolated_cpus);
Rusty Russell42128232008-11-25 02:35:12 +10307374
Rusty Russell0e3900e2008-11-25 02:35:13 +10307375 init_sched_rt_class();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007376}
7377#else
7378void __init sched_init_smp(void)
7379{
Ingo Molnar19978ca2007-11-09 22:39:38 +01007380 sched_init_granularity();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007381}
7382#endif /* CONFIG_SMP */
7383
Arun R Bharadwajcd1bb942009-04-16 12:15:34 +05307384const_debug unsigned int sysctl_timer_migration = 1;
7385
Linus Torvalds1da177e2005-04-16 15:20:36 -07007386int in_sched_functions(unsigned long addr)
7387{
Linus Torvalds1da177e2005-04-16 15:20:36 -07007388 return in_lock_functions(addr) ||
7389 (addr >= (unsigned long)__sched_text_start
7390 && addr < (unsigned long)__sched_text_end);
7391}
7392
Alexey Dobriyana9957442007-10-15 17:00:13 +02007393static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq)
Ingo Molnardd41f592007-07-09 18:51:59 +02007394{
7395 cfs_rq->tasks_timeline = RB_ROOT;
Peter Zijlstra4a55bd52008-04-19 19:45:00 +02007396 INIT_LIST_HEAD(&cfs_rq->tasks);
Ingo Molnardd41f592007-07-09 18:51:59 +02007397#ifdef CONFIG_FAIR_GROUP_SCHED
7398 cfs_rq->rq = rq;
7399#endif
Peter Zijlstra67e9fb22007-10-15 17:00:10 +02007400 cfs_rq->min_vruntime = (u64)(-(1LL << 20));
Ingo Molnardd41f592007-07-09 18:51:59 +02007401}
7402
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007403static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq)
7404{
7405 struct rt_prio_array *array;
7406 int i;
7407
7408 array = &rt_rq->active;
7409 for (i = 0; i < MAX_RT_PRIO; i++) {
7410 INIT_LIST_HEAD(array->queue + i);
7411 __clear_bit(i, array->bitmap);
7412 }
7413 /* delimiter for bitsearch: */
7414 __set_bit(MAX_RT_PRIO, array->bitmap);
7415
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007416#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
Gregory Haskinse864c492008-12-29 09:39:49 -05007417 rt_rq->highest_prio.curr = MAX_RT_PRIO;
Gregory Haskins398a1532009-01-14 09:10:04 -05007418#ifdef CONFIG_SMP
Gregory Haskinse864c492008-12-29 09:39:49 -05007419 rt_rq->highest_prio.next = MAX_RT_PRIO;
Peter Zijlstra48d5e252008-01-25 21:08:31 +01007420#endif
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007421#endif
7422#ifdef CONFIG_SMP
7423 rt_rq->rt_nr_migratory = 0;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007424 rt_rq->overloaded = 0;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01007425 plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007426#endif
7427
7428 rt_rq->rt_time = 0;
7429 rt_rq->rt_throttled = 0;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007430 rt_rq->rt_runtime = 0;
Thomas Gleixner0986b112009-11-17 15:32:06 +01007431 raw_spin_lock_init(&rt_rq->rt_runtime_lock);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007432
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007433#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +01007434 rt_rq->rt_nr_boosted = 0;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007435 rt_rq->rq = rq;
7436#endif
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007437}
7438
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007439#ifdef CONFIG_FAIR_GROUP_SCHED
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007440static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
7441 struct sched_entity *se, int cpu, int add,
7442 struct sched_entity *parent)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007443{
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007444 struct rq *rq = cpu_rq(cpu);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007445 tg->cfs_rq[cpu] = cfs_rq;
7446 init_cfs_rq(cfs_rq, rq);
7447 cfs_rq->tg = tg;
7448 if (add)
7449 list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list);
7450
7451 tg->se[cpu] = se;
Dhaval Giani354d60c2008-04-19 19:44:59 +02007452 /* se could be NULL for init_task_group */
7453 if (!se)
7454 return;
7455
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007456 if (!parent)
7457 se->cfs_rq = &rq->cfs;
7458 else
7459 se->cfs_rq = parent->my_q;
7460
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007461 se->my_q = cfs_rq;
7462 se->load.weight = tg->shares;
Peter Zijlstrae05510d2008-05-05 23:56:17 +02007463 se->load.inv_weight = 0;
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007464 se->parent = parent;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007465}
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007466#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007467
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007468#ifdef CONFIG_RT_GROUP_SCHED
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007469static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
7470 struct sched_rt_entity *rt_se, int cpu, int add,
7471 struct sched_rt_entity *parent)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007472{
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007473 struct rq *rq = cpu_rq(cpu);
7474
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007475 tg->rt_rq[cpu] = rt_rq;
7476 init_rt_rq(rt_rq, rq);
7477 rt_rq->tg = tg;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007478 rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007479 if (add)
7480 list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list);
7481
7482 tg->rt_se[cpu] = rt_se;
Dhaval Giani354d60c2008-04-19 19:44:59 +02007483 if (!rt_se)
7484 return;
7485
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007486 if (!parent)
7487 rt_se->rt_rq = &rq->rt;
7488 else
7489 rt_se->rt_rq = parent->my_q;
7490
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007491 rt_se->my_q = rt_rq;
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007492 rt_se->parent = parent;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007493 INIT_LIST_HEAD(&rt_se->run_list);
7494}
7495#endif
7496
Linus Torvalds1da177e2005-04-16 15:20:36 -07007497void __init sched_init(void)
7498{
Ingo Molnardd41f592007-07-09 18:51:59 +02007499 int i, j;
Mike Travis434d53b2008-04-04 18:11:04 -07007500 unsigned long alloc_size = 0, ptr;
7501
7502#ifdef CONFIG_FAIR_GROUP_SCHED
7503 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
7504#endif
7505#ifdef CONFIG_RT_GROUP_SCHED
7506 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
7507#endif
Rusty Russelldf7c8e82009-03-19 15:22:20 +10307508#ifdef CONFIG_CPUMASK_OFFSTACK
Rusty Russell8c083f02009-03-19 15:22:20 +10307509 alloc_size += num_possible_cpus() * cpumask_size();
Rusty Russelldf7c8e82009-03-19 15:22:20 +10307510#endif
Mike Travis434d53b2008-04-04 18:11:04 -07007511 if (alloc_size) {
Pekka Enberg36b7b6d2009-06-10 23:42:36 +03007512 ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
Mike Travis434d53b2008-04-04 18:11:04 -07007513
7514#ifdef CONFIG_FAIR_GROUP_SCHED
7515 init_task_group.se = (struct sched_entity **)ptr;
7516 ptr += nr_cpu_ids * sizeof(void **);
7517
7518 init_task_group.cfs_rq = (struct cfs_rq **)ptr;
7519 ptr += nr_cpu_ids * sizeof(void **);
Peter Zijlstraeff766a2008-04-19 19:45:00 +02007520
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007521#endif /* CONFIG_FAIR_GROUP_SCHED */
Mike Travis434d53b2008-04-04 18:11:04 -07007522#ifdef CONFIG_RT_GROUP_SCHED
7523 init_task_group.rt_se = (struct sched_rt_entity **)ptr;
7524 ptr += nr_cpu_ids * sizeof(void **);
7525
7526 init_task_group.rt_rq = (struct rt_rq **)ptr;
Peter Zijlstraeff766a2008-04-19 19:45:00 +02007527 ptr += nr_cpu_ids * sizeof(void **);
7528
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007529#endif /* CONFIG_RT_GROUP_SCHED */
Rusty Russelldf7c8e82009-03-19 15:22:20 +10307530#ifdef CONFIG_CPUMASK_OFFSTACK
7531 for_each_possible_cpu(i) {
7532 per_cpu(load_balance_tmpmask, i) = (void *)ptr;
7533 ptr += cpumask_size();
7534 }
7535#endif /* CONFIG_CPUMASK_OFFSTACK */
Mike Travis434d53b2008-04-04 18:11:04 -07007536 }
Ingo Molnardd41f592007-07-09 18:51:59 +02007537
Gregory Haskins57d885f2008-01-25 21:08:18 +01007538#ifdef CONFIG_SMP
7539 init_defrootdomain();
7540#endif
7541
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007542 init_rt_bandwidth(&def_rt_bandwidth,
7543 global_rt_period(), global_rt_runtime());
7544
7545#ifdef CONFIG_RT_GROUP_SCHED
7546 init_rt_bandwidth(&init_task_group.rt_bandwidth,
7547 global_rt_period(), global_rt_runtime());
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007548#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007549
Dhaval Giani7c941432010-01-20 13:26:18 +01007550#ifdef CONFIG_CGROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007551 list_add(&init_task_group.list, &task_groups);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02007552 INIT_LIST_HEAD(&init_task_group.children);
7553
Dhaval Giani7c941432010-01-20 13:26:18 +01007554#endif /* CONFIG_CGROUP_SCHED */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007555
Jiri Kosina4a6cc4b2009-10-29 00:26:00 +09007556#if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP
7557 update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long),
7558 __alignof__(unsigned long));
7559#endif
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08007560 for_each_possible_cpu(i) {
Ingo Molnar70b97a72006-07-03 00:25:42 -07007561 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007562
7563 rq = cpu_rq(i);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01007564 raw_spin_lock_init(&rq->lock);
Nick Piggin78979862005-06-25 14:57:13 -07007565 rq->nr_running = 0;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02007566 rq->calc_load_active = 0;
7567 rq->calc_load_update = jiffies + LOAD_FREQ;
Ingo Molnardd41f592007-07-09 18:51:59 +02007568 init_cfs_rq(&rq->cfs, rq);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01007569 init_rt_rq(&rq->rt, rq);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007570#ifdef CONFIG_FAIR_GROUP_SCHED
7571 init_task_group.shares = init_task_group_load;
7572 INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
Dhaval Giani354d60c2008-04-19 19:44:59 +02007573#ifdef CONFIG_CGROUP_SCHED
7574 /*
7575 * How much cpu bandwidth does init_task_group get?
7576 *
7577 * In case of task-groups formed thr' the cgroup filesystem, it
7578 * gets 100% of the cpu resources in the system. This overall
7579 * system cpu resource is divided among the tasks of
7580 * init_task_group and its child task-groups in a fair manner,
7581 * based on each entity's (task or task-group's) weight
7582 * (se->load.weight).
7583 *
7584 * In other words, if init_task_group has 10 tasks of weight
7585 * 1024) and two child groups A0 and A1 (of weight 1024 each),
7586 * then A0's share of the cpu resource is:
7587 *
Ingo Molnar0d905bc2009-05-04 19:13:30 +02007588 * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
Dhaval Giani354d60c2008-04-19 19:44:59 +02007589 *
7590 * We achieve this by letting init_task_group's tasks sit
7591 * directly in rq->cfs (i.e init_task_group->se[] = NULL).
7592 */
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007593 init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL);
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007594#endif
Dhaval Giani354d60c2008-04-19 19:44:59 +02007595#endif /* CONFIG_FAIR_GROUP_SCHED */
7596
7597 rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007598#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007599 INIT_LIST_HEAD(&rq->leaf_rt_rq_list);
Dhaval Giani354d60c2008-04-19 19:44:59 +02007600#ifdef CONFIG_CGROUP_SCHED
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007601 init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL);
Dhaval Giani354d60c2008-04-19 19:44:59 +02007602#endif
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007603#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07007604
Ingo Molnardd41f592007-07-09 18:51:59 +02007605 for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
7606 rq->cpu_load[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007607#ifdef CONFIG_SMP
Nick Piggin41c7ce92005-06-25 14:57:24 -07007608 rq->sd = NULL;
Gregory Haskins57d885f2008-01-25 21:08:18 +01007609 rq->rd = NULL;
Peter Zijlstrae51fd5e2010-05-31 12:37:30 +02007610 rq->cpu_power = SCHED_LOAD_SCALE;
Gregory Haskins3f029d32009-07-29 11:08:47 -04007611 rq->post_schedule = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007612 rq->active_balance = 0;
Ingo Molnardd41f592007-07-09 18:51:59 +02007613 rq->next_balance = jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007614 rq->push_cpu = 0;
Christoph Lameter0a2966b2006-09-25 23:30:51 -07007615 rq->cpu = i;
Gregory Haskins1f11eb62008-06-04 15:04:05 -04007616 rq->online = 0;
Mike Galbraitheae0c9d2009-11-10 03:50:02 +01007617 rq->idle_stamp = 0;
7618 rq->avg_idle = 2*sysctl_sched_migration_cost;
Gregory Haskinsdc938522008-01-25 21:08:26 +01007619 rq_attach_root(rq, &def_root_domain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007620#endif
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01007621 init_rq_hrtick(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007622 atomic_set(&rq->nr_iowait, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007623 }
7624
Peter Williams2dd73a42006-06-27 02:54:34 -07007625 set_load_weight(&init_task);
Heiko Carstensb50f60c2006-07-30 03:03:52 -07007626
Avi Kivitye107be32007-07-26 13:40:43 +02007627#ifdef CONFIG_PREEMPT_NOTIFIERS
7628 INIT_HLIST_HEAD(&init_task.preempt_notifiers);
7629#endif
7630
Christoph Lameterc9819f42006-12-10 02:20:25 -08007631#ifdef CONFIG_SMP
Carlos R. Mafra962cf362008-05-15 11:15:37 -03007632 open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);
Christoph Lameterc9819f42006-12-10 02:20:25 -08007633#endif
7634
Heiko Carstensb50f60c2006-07-30 03:03:52 -07007635#ifdef CONFIG_RT_MUTEXES
Thomas Gleixner1d615482009-11-17 14:54:03 +01007636 plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock);
Heiko Carstensb50f60c2006-07-30 03:03:52 -07007637#endif
7638
Linus Torvalds1da177e2005-04-16 15:20:36 -07007639 /*
7640 * The boot idle thread does lazy MMU switching as well:
7641 */
7642 atomic_inc(&init_mm.mm_count);
7643 enter_lazy_tlb(&init_mm, current);
7644
7645 /*
7646 * Make us the idle thread. Technically, schedule() should not be
7647 * called from this thread, however somewhere below it might be,
7648 * but because we are the idle thread, we just pick up running again
7649 * when this runqueue becomes "idle".
7650 */
7651 init_idle(current, smp_processor_id());
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02007652
7653 calc_load_update = jiffies + LOAD_FREQ;
7654
Ingo Molnardd41f592007-07-09 18:51:59 +02007655 /*
7656 * During early bootup we pretend to be a normal task:
7657 */
7658 current->sched_class = &fair_sched_class;
Ingo Molnar6892b752008-02-13 14:02:36 +01007659
Rusty Russell6a7b3dc2008-11-25 02:35:04 +10307660 /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */
Rusty Russell49557e62009-11-02 20:37:20 +10307661 zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT);
Rusty Russellbf4d83f2008-11-25 09:57:51 +10307662#ifdef CONFIG_SMP
Rusty Russell7d1e6a92008-11-25 02:35:09 +10307663#ifdef CONFIG_NO_HZ
Rusty Russell49557e62009-11-02 20:37:20 +10307664 zalloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT);
Pekka Enberg4bdddf82009-06-11 08:35:27 +03007665 alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT);
Rusty Russell7d1e6a92008-11-25 02:35:09 +10307666#endif
Rusty Russellbdddd292009-12-02 14:09:16 +10307667 /* May be allocated at isolcpus cmdline parse time */
7668 if (cpu_isolated_map == NULL)
7669 zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT);
Rusty Russellbf4d83f2008-11-25 09:57:51 +10307670#endif /* SMP */
Rusty Russell6a7b3dc2008-11-25 02:35:04 +10307671
Ingo Molnarcdd6c482009-09-21 12:02:48 +02007672 perf_event_init();
Ingo Molnar0d905bc2009-05-04 19:13:30 +02007673
Ingo Molnar6892b752008-02-13 14:02:36 +01007674 scheduler_running = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007675}
7676
7677#ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
Frederic Weisbeckere4aafea2009-07-16 15:44:29 +02007678static inline int preempt_count_equals(int preempt_offset)
7679{
Frederic Weisbecker234da7b2009-12-16 20:21:05 +01007680 int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth();
Frederic Weisbeckere4aafea2009-07-16 15:44:29 +02007681
7682 return (nested == PREEMPT_INATOMIC_BASE + preempt_offset);
7683}
7684
Simon Kagstromd8948372009-12-23 11:08:18 +01007685void __might_sleep(const char *file, int line, int preempt_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007686{
Ingo Molnar48f24c42006-07-03 00:25:40 -07007687#ifdef in_atomic
Linus Torvalds1da177e2005-04-16 15:20:36 -07007688 static unsigned long prev_jiffy; /* ratelimiting */
7689
Frederic Weisbeckere4aafea2009-07-16 15:44:29 +02007690 if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) ||
7691 system_state != SYSTEM_RUNNING || oops_in_progress)
Ingo Molnaraef745f2008-08-28 11:34:43 +02007692 return;
7693 if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
7694 return;
7695 prev_jiffy = jiffies;
7696
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01007697 printk(KERN_ERR
7698 "BUG: sleeping function called from invalid context at %s:%d\n",
7699 file, line);
7700 printk(KERN_ERR
7701 "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n",
7702 in_atomic(), irqs_disabled(),
7703 current->pid, current->comm);
Ingo Molnaraef745f2008-08-28 11:34:43 +02007704
7705 debug_show_held_locks(current);
7706 if (irqs_disabled())
7707 print_irqtrace_events(current);
7708 dump_stack();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007709#endif
7710}
7711EXPORT_SYMBOL(__might_sleep);
7712#endif
7713
7714#ifdef CONFIG_MAGIC_SYSRQ
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02007715static void normalize_task(struct rq *rq, struct task_struct *p)
7716{
7717 int on_rq;
Peter Zijlstra3e51f332008-05-03 18:29:28 +02007718
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02007719 on_rq = p->se.on_rq;
7720 if (on_rq)
7721 deactivate_task(rq, p, 0);
7722 __setscheduler(rq, p, SCHED_NORMAL, 0);
7723 if (on_rq) {
7724 activate_task(rq, p, 0);
7725 resched_task(rq->curr);
7726 }
7727}
7728
Linus Torvalds1da177e2005-04-16 15:20:36 -07007729void normalize_rt_tasks(void)
7730{
Ingo Molnara0f98a12007-06-17 18:37:45 +02007731 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007732 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07007733 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007734
Peter Zijlstra4cf5d772008-02-13 15:45:39 +01007735 read_lock_irqsave(&tasklist_lock, flags);
Ingo Molnara0f98a12007-06-17 18:37:45 +02007736 do_each_thread(g, p) {
Ingo Molnar178be792007-10-15 17:00:18 +02007737 /*
7738 * Only normalize user tasks:
7739 */
7740 if (!p->mm)
7741 continue;
7742
Ingo Molnardd41f592007-07-09 18:51:59 +02007743 p->se.exec_start = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02007744#ifdef CONFIG_SCHEDSTATS
Lucas De Marchi41acab82010-03-10 23:37:45 -03007745 p->se.statistics.wait_start = 0;
7746 p->se.statistics.sleep_start = 0;
7747 p->se.statistics.block_start = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02007748#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02007749
7750 if (!rt_task(p)) {
7751 /*
7752 * Renice negative nice level userspace
7753 * tasks back to 0:
7754 */
7755 if (TASK_NICE(p) < 0 && p->mm)
7756 set_user_nice(p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007757 continue;
Ingo Molnardd41f592007-07-09 18:51:59 +02007758 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007759
Thomas Gleixner1d615482009-11-17 14:54:03 +01007760 raw_spin_lock(&p->pi_lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -07007761 rq = __task_rq_lock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007762
Ingo Molnar178be792007-10-15 17:00:18 +02007763 normalize_task(rq, p);
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02007764
Ingo Molnarb29739f2006-06-27 02:54:51 -07007765 __task_rq_unlock(rq);
Thomas Gleixner1d615482009-11-17 14:54:03 +01007766 raw_spin_unlock(&p->pi_lock);
Ingo Molnara0f98a12007-06-17 18:37:45 +02007767 } while_each_thread(g, p);
7768
Peter Zijlstra4cf5d772008-02-13 15:45:39 +01007769 read_unlock_irqrestore(&tasklist_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007770}
7771
7772#endif /* CONFIG_MAGIC_SYSRQ */
Linus Torvalds1df5c102005-09-12 07:59:21 -07007773
Jason Wessel67fc4e02010-05-20 21:04:21 -05007774#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
Linus Torvalds1df5c102005-09-12 07:59:21 -07007775/*
Jason Wessel67fc4e02010-05-20 21:04:21 -05007776 * These functions are only useful for the IA64 MCA handling, or kdb.
Linus Torvalds1df5c102005-09-12 07:59:21 -07007777 *
7778 * They can only be called when the whole system has been
7779 * stopped - every CPU needs to be quiescent, and no scheduling
7780 * activity can take place. Using them for anything else would
7781 * be a serious bug, and as a result, they aren't even visible
7782 * under any other configuration.
7783 */
7784
7785/**
7786 * curr_task - return the current task for a given cpu.
7787 * @cpu: the processor in question.
7788 *
7789 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
7790 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07007791struct task_struct *curr_task(int cpu)
Linus Torvalds1df5c102005-09-12 07:59:21 -07007792{
7793 return cpu_curr(cpu);
7794}
7795
Jason Wessel67fc4e02010-05-20 21:04:21 -05007796#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */
7797
7798#ifdef CONFIG_IA64
Linus Torvalds1df5c102005-09-12 07:59:21 -07007799/**
7800 * set_curr_task - set the current task for a given cpu.
7801 * @cpu: the processor in question.
7802 * @p: the task pointer to set.
7803 *
7804 * Description: This function must only be used when non-maskable interrupts
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007805 * are serviced on a separate stack. It allows the architecture to switch the
7806 * notion of the current task on a cpu in a non-blocking manner. This function
Linus Torvalds1df5c102005-09-12 07:59:21 -07007807 * must be called with all CPU's synchronized, and interrupts disabled, the
7808 * and caller must save the original value of the current task (see
7809 * curr_task() above) and restore that value before reenabling interrupts and
7810 * re-starting the system.
7811 *
7812 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
7813 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07007814void set_curr_task(int cpu, struct task_struct *p)
Linus Torvalds1df5c102005-09-12 07:59:21 -07007815{
7816 cpu_curr(cpu) = p;
7817}
7818
7819#endif
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007820
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007821#ifdef CONFIG_FAIR_GROUP_SCHED
7822static void free_fair_sched_group(struct task_group *tg)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007823{
7824 int i;
7825
7826 for_each_possible_cpu(i) {
7827 if (tg->cfs_rq)
7828 kfree(tg->cfs_rq[i]);
7829 if (tg->se)
7830 kfree(tg->se[i]);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007831 }
7832
7833 kfree(tg->cfs_rq);
7834 kfree(tg->se);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007835}
7836
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007837static
7838int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007839{
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007840 struct cfs_rq *cfs_rq;
Li Zefaneab17222008-10-29 17:03:22 +08007841 struct sched_entity *se;
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007842 struct rq *rq;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007843 int i;
7844
Mike Travis434d53b2008-04-04 18:11:04 -07007845 tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007846 if (!tg->cfs_rq)
7847 goto err;
Mike Travis434d53b2008-04-04 18:11:04 -07007848 tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007849 if (!tg->se)
7850 goto err;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007851
7852 tg->shares = NICE_0_LOAD;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007853
7854 for_each_possible_cpu(i) {
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007855 rq = cpu_rq(i);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007856
Li Zefaneab17222008-10-29 17:03:22 +08007857 cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
7858 GFP_KERNEL, cpu_to_node(i));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007859 if (!cfs_rq)
7860 goto err;
7861
Li Zefaneab17222008-10-29 17:03:22 +08007862 se = kzalloc_node(sizeof(struct sched_entity),
7863 GFP_KERNEL, cpu_to_node(i));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007864 if (!se)
Phil Carmodydfc12eb2009-12-10 14:29:37 +02007865 goto err_free_rq;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007866
Li Zefaneab17222008-10-29 17:03:22 +08007867 init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007868 }
7869
7870 return 1;
7871
Phil Carmodydfc12eb2009-12-10 14:29:37 +02007872 err_free_rq:
7873 kfree(cfs_rq);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007874 err:
7875 return 0;
7876}
7877
7878static inline void register_fair_sched_group(struct task_group *tg, int cpu)
7879{
7880 list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list,
7881 &cpu_rq(cpu)->leaf_cfs_rq_list);
7882}
7883
7884static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
7885{
7886 list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list);
7887}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007888#else /* !CONFG_FAIR_GROUP_SCHED */
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007889static inline void free_fair_sched_group(struct task_group *tg)
7890{
7891}
7892
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007893static inline
7894int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007895{
7896 return 1;
7897}
7898
7899static inline void register_fair_sched_group(struct task_group *tg, int cpu)
7900{
7901}
7902
7903static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
7904{
7905}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007906#endif /* CONFIG_FAIR_GROUP_SCHED */
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007907
7908#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007909static void free_rt_sched_group(struct task_group *tg)
7910{
7911 int i;
7912
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007913 destroy_rt_bandwidth(&tg->rt_bandwidth);
7914
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007915 for_each_possible_cpu(i) {
7916 if (tg->rt_rq)
7917 kfree(tg->rt_rq[i]);
7918 if (tg->rt_se)
7919 kfree(tg->rt_se[i]);
7920 }
7921
7922 kfree(tg->rt_rq);
7923 kfree(tg->rt_se);
7924}
7925
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007926static
7927int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007928{
7929 struct rt_rq *rt_rq;
Li Zefaneab17222008-10-29 17:03:22 +08007930 struct sched_rt_entity *rt_se;
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007931 struct rq *rq;
7932 int i;
7933
Mike Travis434d53b2008-04-04 18:11:04 -07007934 tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007935 if (!tg->rt_rq)
7936 goto err;
Mike Travis434d53b2008-04-04 18:11:04 -07007937 tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007938 if (!tg->rt_se)
7939 goto err;
7940
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007941 init_rt_bandwidth(&tg->rt_bandwidth,
7942 ktime_to_ns(def_rt_bandwidth.rt_period), 0);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007943
7944 for_each_possible_cpu(i) {
7945 rq = cpu_rq(i);
7946
Li Zefaneab17222008-10-29 17:03:22 +08007947 rt_rq = kzalloc_node(sizeof(struct rt_rq),
7948 GFP_KERNEL, cpu_to_node(i));
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007949 if (!rt_rq)
7950 goto err;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007951
Li Zefaneab17222008-10-29 17:03:22 +08007952 rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
7953 GFP_KERNEL, cpu_to_node(i));
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007954 if (!rt_se)
Phil Carmodydfc12eb2009-12-10 14:29:37 +02007955 goto err_free_rq;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007956
Li Zefaneab17222008-10-29 17:03:22 +08007957 init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007958 }
7959
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007960 return 1;
7961
Phil Carmodydfc12eb2009-12-10 14:29:37 +02007962 err_free_rq:
7963 kfree(rt_rq);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007964 err:
7965 return 0;
7966}
7967
7968static inline void register_rt_sched_group(struct task_group *tg, int cpu)
7969{
7970 list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list,
7971 &cpu_rq(cpu)->leaf_rt_rq_list);
7972}
7973
7974static inline void unregister_rt_sched_group(struct task_group *tg, int cpu)
7975{
7976 list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list);
7977}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007978#else /* !CONFIG_RT_GROUP_SCHED */
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007979static inline void free_rt_sched_group(struct task_group *tg)
7980{
7981}
7982
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007983static inline
7984int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007985{
7986 return 1;
7987}
7988
7989static inline void register_rt_sched_group(struct task_group *tg, int cpu)
7990{
7991}
7992
7993static inline void unregister_rt_sched_group(struct task_group *tg, int cpu)
7994{
7995}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007996#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007997
Dhaval Giani7c941432010-01-20 13:26:18 +01007998#ifdef CONFIG_CGROUP_SCHED
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007999static void free_sched_group(struct task_group *tg)
8000{
8001 free_fair_sched_group(tg);
8002 free_rt_sched_group(tg);
8003 kfree(tg);
8004}
8005
8006/* allocate runqueue etc for a new task group */
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008007struct task_group *sched_create_group(struct task_group *parent)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008008{
8009 struct task_group *tg;
8010 unsigned long flags;
8011 int i;
8012
8013 tg = kzalloc(sizeof(*tg), GFP_KERNEL);
8014 if (!tg)
8015 return ERR_PTR(-ENOMEM);
8016
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008017 if (!alloc_fair_sched_group(tg, parent))
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008018 goto err;
8019
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008020 if (!alloc_rt_sched_group(tg, parent))
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008021 goto err;
8022
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008023 spin_lock_irqsave(&task_group_lock, flags);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008024 for_each_possible_cpu(i) {
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008025 register_fair_sched_group(tg, i);
8026 register_rt_sched_group(tg, i);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008027 }
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008028 list_add_rcu(&tg->list, &task_groups);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02008029
8030 WARN_ON(!parent); /* root should already exist */
8031
8032 tg->parent = parent;
Peter Zijlstraf473aa52008-04-19 19:45:00 +02008033 INIT_LIST_HEAD(&tg->children);
Zhang, Yanmin09f27242030-08-14 15:56:40 +08008034 list_add_rcu(&tg->siblings, &parent->children);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008035 spin_unlock_irqrestore(&task_group_lock, flags);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008036
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008037 return tg;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008038
8039err:
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008040 free_sched_group(tg);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008041 return ERR_PTR(-ENOMEM);
8042}
8043
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008044/* rcu callback to free various structures associated with a task group */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008045static void free_sched_group_rcu(struct rcu_head *rhp)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008046{
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008047 /* now it should be safe to free those cfs_rqs */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008048 free_sched_group(container_of(rhp, struct task_group, rcu));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008049}
8050
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008051/* Destroy runqueue etc associated with a task group */
Ingo Molnar4cf86d72007-10-15 17:00:14 +02008052void sched_destroy_group(struct task_group *tg)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008053{
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008054 unsigned long flags;
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008055 int i;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008056
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008057 spin_lock_irqsave(&task_group_lock, flags);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008058 for_each_possible_cpu(i) {
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008059 unregister_fair_sched_group(tg, i);
8060 unregister_rt_sched_group(tg, i);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008061 }
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008062 list_del_rcu(&tg->list);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02008063 list_del_rcu(&tg->siblings);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008064 spin_unlock_irqrestore(&task_group_lock, flags);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008065
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008066 /* wait for possible concurrent references to cfs_rqs complete */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008067 call_rcu(&tg->rcu, free_sched_group_rcu);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008068}
8069
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008070/* change task's runqueue when it moves between groups.
Ingo Molnar3a252012007-10-15 17:00:12 +02008071 * The caller of this function should have put the task in its new group
8072 * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to
8073 * reflect its new group.
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008074 */
8075void sched_move_task(struct task_struct *tsk)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008076{
8077 int on_rq, running;
8078 unsigned long flags;
8079 struct rq *rq;
8080
8081 rq = task_rq_lock(tsk, &flags);
8082
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01008083 running = task_current(rq, tsk);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008084 on_rq = tsk->se.on_rq;
8085
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07008086 if (on_rq)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008087 dequeue_task(rq, tsk, 0);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07008088 if (unlikely(running))
8089 tsk->sched_class->put_prev_task(rq, tsk);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008090
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008091 set_task_rq(tsk, task_cpu(tsk));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008092
Peter Zijlstra810b3812008-02-29 15:21:01 -05008093#ifdef CONFIG_FAIR_GROUP_SCHED
8094 if (tsk->sched_class->moved_group)
Peter Zijlstra88ec22d2009-12-16 18:04:41 +01008095 tsk->sched_class->moved_group(tsk, on_rq);
Peter Zijlstra810b3812008-02-29 15:21:01 -05008096#endif
8097
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07008098 if (unlikely(running))
8099 tsk->sched_class->set_curr_task(rq);
8100 if (on_rq)
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01008101 enqueue_task(rq, tsk, 0);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008102
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008103 task_rq_unlock(rq, &flags);
8104}
Dhaval Giani7c941432010-01-20 13:26:18 +01008105#endif /* CONFIG_CGROUP_SCHED */
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008106
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008107#ifdef CONFIG_FAIR_GROUP_SCHED
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008108static void __set_se_shares(struct sched_entity *se, unsigned long shares)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008109{
8110 struct cfs_rq *cfs_rq = se->cfs_rq;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008111 int on_rq;
8112
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008113 on_rq = se->on_rq;
Peter Zijlstra62fb1852008-02-25 17:34:02 +01008114 if (on_rq)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008115 dequeue_entity(cfs_rq, se, 0);
8116
8117 se->load.weight = shares;
Peter Zijlstrae05510d2008-05-05 23:56:17 +02008118 se->load.inv_weight = 0;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008119
Peter Zijlstra62fb1852008-02-25 17:34:02 +01008120 if (on_rq)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008121 enqueue_entity(cfs_rq, se, 0);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008122}
Peter Zijlstra62fb1852008-02-25 17:34:02 +01008123
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008124static void set_se_shares(struct sched_entity *se, unsigned long shares)
8125{
8126 struct cfs_rq *cfs_rq = se->cfs_rq;
8127 struct rq *rq = cfs_rq->rq;
8128 unsigned long flags;
8129
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008130 raw_spin_lock_irqsave(&rq->lock, flags);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008131 __set_se_shares(se, shares);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008132 raw_spin_unlock_irqrestore(&rq->lock, flags);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008133}
8134
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008135static DEFINE_MUTEX(shares_mutex);
8136
Ingo Molnar4cf86d72007-10-15 17:00:14 +02008137int sched_group_set_shares(struct task_group *tg, unsigned long shares)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008138{
8139 int i;
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008140 unsigned long flags;
Ingo Molnarc61935f2008-01-22 11:24:58 +01008141
Peter Zijlstra62fb1852008-02-25 17:34:02 +01008142 /*
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008143 * We can't change the weight of the root cgroup.
8144 */
8145 if (!tg->se[0])
8146 return -EINVAL;
8147
Peter Zijlstra18d95a22008-04-19 19:45:00 +02008148 if (shares < MIN_SHARES)
8149 shares = MIN_SHARES;
Miao Xiecb4ad1f2008-04-28 12:54:56 +08008150 else if (shares > MAX_SHARES)
8151 shares = MAX_SHARES;
Peter Zijlstra62fb1852008-02-25 17:34:02 +01008152
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008153 mutex_lock(&shares_mutex);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008154 if (tg->shares == shares)
Dhaval Giani5cb350b2007-10-15 17:00:14 +02008155 goto done;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008156
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008157 spin_lock_irqsave(&task_group_lock, flags);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008158 for_each_possible_cpu(i)
8159 unregister_fair_sched_group(tg, i);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02008160 list_del_rcu(&tg->siblings);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008161 spin_unlock_irqrestore(&task_group_lock, flags);
Srivatsa Vaddagiri6b2d7702008-01-25 21:08:00 +01008162
8163 /* wait for any ongoing reference to this group to finish */
8164 synchronize_sched();
8165
8166 /*
8167 * Now we are free to modify the group's share on each cpu
8168 * w/o tripping rebalance_share or load_balance_fair.
8169 */
8170 tg->shares = shares;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008171 for_each_possible_cpu(i) {
8172 /*
8173 * force a rebalance
8174 */
8175 cfs_rq_set_shares(tg->cfs_rq[i], 0);
Miao Xiecb4ad1f2008-04-28 12:54:56 +08008176 set_se_shares(tg->se[i], shares);
Peter Zijlstrac09595f2008-06-27 13:41:14 +02008177 }
Srivatsa Vaddagiri6b2d7702008-01-25 21:08:00 +01008178
8179 /*
8180 * Enable load balance activity on this group, by inserting it back on
8181 * each cpu's rq->leaf_cfs_rq_list.
8182 */
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008183 spin_lock_irqsave(&task_group_lock, flags);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01008184 for_each_possible_cpu(i)
8185 register_fair_sched_group(tg, i);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02008186 list_add_rcu(&tg->siblings, &tg->parent->children);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008187 spin_unlock_irqrestore(&task_group_lock, flags);
Dhaval Giani5cb350b2007-10-15 17:00:14 +02008188done:
Peter Zijlstra8ed36992008-02-13 15:45:39 +01008189 mutex_unlock(&shares_mutex);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02008190 return 0;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02008191}
8192
Dhaval Giani5cb350b2007-10-15 17:00:14 +02008193unsigned long sched_group_shares(struct task_group *tg)
8194{
8195 return tg->shares;
8196}
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008197#endif
Dhaval Giani5cb350b2007-10-15 17:00:14 +02008198
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008199#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008200/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008201 * Ensure that the real time constraints are schedulable.
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008202 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008203static DEFINE_MUTEX(rt_constraints_mutex);
8204
8205static unsigned long to_ratio(u64 period, u64 runtime)
8206{
8207 if (runtime == RUNTIME_INF)
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008208 return 1ULL << 20;
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008209
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008210 return div64_u64(runtime << 20, period);
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008211}
8212
Dhaval Giani521f1a242008-02-28 15:21:56 +05308213/* Must be called with tasklist_lock held */
8214static inline int tg_has_rt_tasks(struct task_group *tg)
8215{
8216 struct task_struct *g, *p;
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008217
Dhaval Giani521f1a242008-02-28 15:21:56 +05308218 do_each_thread(g, p) {
8219 if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg)
8220 return 1;
8221 } while_each_thread(g, p);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008222
Dhaval Giani521f1a242008-02-28 15:21:56 +05308223 return 0;
8224}
8225
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008226struct rt_schedulable_data {
8227 struct task_group *tg;
8228 u64 rt_period;
8229 u64 rt_runtime;
8230};
8231
8232static int tg_schedulable(struct task_group *tg, void *data)
8233{
8234 struct rt_schedulable_data *d = data;
8235 struct task_group *child;
8236 unsigned long total, sum = 0;
8237 u64 period, runtime;
8238
8239 period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8240 runtime = tg->rt_bandwidth.rt_runtime;
8241
8242 if (tg == d->tg) {
8243 period = d->rt_period;
8244 runtime = d->rt_runtime;
8245 }
8246
Peter Zijlstra4653f802008-09-23 15:33:44 +02008247 /*
8248 * Cannot have more runtime than the period.
8249 */
8250 if (runtime > period && runtime != RUNTIME_INF)
8251 return -EINVAL;
8252
8253 /*
8254 * Ensure we don't starve existing RT tasks.
8255 */
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008256 if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
8257 return -EBUSY;
8258
8259 total = to_ratio(period, runtime);
8260
Peter Zijlstra4653f802008-09-23 15:33:44 +02008261 /*
8262 * Nobody can have more than the global setting allows.
8263 */
8264 if (total > to_ratio(global_rt_period(), global_rt_runtime()))
8265 return -EINVAL;
8266
8267 /*
8268 * The sum of our children's runtime should not exceed our own.
8269 */
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008270 list_for_each_entry_rcu(child, &tg->children, siblings) {
8271 period = ktime_to_ns(child->rt_bandwidth.rt_period);
8272 runtime = child->rt_bandwidth.rt_runtime;
8273
8274 if (child == d->tg) {
8275 period = d->rt_period;
8276 runtime = d->rt_runtime;
8277 }
8278
8279 sum += to_ratio(period, runtime);
8280 }
8281
8282 if (sum > total)
8283 return -EINVAL;
8284
8285 return 0;
8286}
8287
8288static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
8289{
8290 struct rt_schedulable_data data = {
8291 .tg = tg,
8292 .rt_period = period,
8293 .rt_runtime = runtime,
8294 };
8295
8296 return walk_tg_tree(tg_schedulable, tg_nop, &data);
8297}
8298
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008299static int tg_set_bandwidth(struct task_group *tg,
8300 u64 rt_period, u64 rt_runtime)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008301{
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008302 int i, err = 0;
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008303
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008304 mutex_lock(&rt_constraints_mutex);
Dhaval Giani521f1a242008-02-28 15:21:56 +05308305 read_lock(&tasklist_lock);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008306 err = __rt_schedulable(tg, rt_period, rt_runtime);
8307 if (err)
Dhaval Giani521f1a242008-02-28 15:21:56 +05308308 goto unlock;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008309
Thomas Gleixner0986b112009-11-17 15:32:06 +01008310 raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008311 tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
8312 tg->rt_bandwidth.rt_runtime = rt_runtime;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008313
8314 for_each_possible_cpu(i) {
8315 struct rt_rq *rt_rq = tg->rt_rq[i];
8316
Thomas Gleixner0986b112009-11-17 15:32:06 +01008317 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008318 rt_rq->rt_runtime = rt_runtime;
Thomas Gleixner0986b112009-11-17 15:32:06 +01008319 raw_spin_unlock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008320 }
Thomas Gleixner0986b112009-11-17 15:32:06 +01008321 raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008322 unlock:
Dhaval Giani521f1a242008-02-28 15:21:56 +05308323 read_unlock(&tasklist_lock);
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008324 mutex_unlock(&rt_constraints_mutex);
8325
8326 return err;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008327}
8328
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008329int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
8330{
8331 u64 rt_runtime, rt_period;
8332
8333 rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8334 rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
8335 if (rt_runtime_us < 0)
8336 rt_runtime = RUNTIME_INF;
8337
8338 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8339}
8340
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008341long sched_group_rt_runtime(struct task_group *tg)
8342{
8343 u64 rt_runtime_us;
8344
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008345 if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008346 return -1;
8347
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008348 rt_runtime_us = tg->rt_bandwidth.rt_runtime;
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008349 do_div(rt_runtime_us, NSEC_PER_USEC);
8350 return rt_runtime_us;
8351}
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008352
8353int sched_group_set_rt_period(struct task_group *tg, long rt_period_us)
8354{
8355 u64 rt_runtime, rt_period;
8356
8357 rt_period = (u64)rt_period_us * NSEC_PER_USEC;
8358 rt_runtime = tg->rt_bandwidth.rt_runtime;
8359
Raistlin619b0482008-06-26 18:54:09 +02008360 if (rt_period == 0)
8361 return -EINVAL;
8362
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008363 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8364}
8365
8366long sched_group_rt_period(struct task_group *tg)
8367{
8368 u64 rt_period_us;
8369
8370 rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
8371 do_div(rt_period_us, NSEC_PER_USEC);
8372 return rt_period_us;
8373}
8374
8375static int sched_rt_global_constraints(void)
8376{
Peter Zijlstra4653f802008-09-23 15:33:44 +02008377 u64 runtime, period;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008378 int ret = 0;
8379
Hiroshi Shimamotoec5d4982008-09-10 17:00:19 -07008380 if (sysctl_sched_rt_period <= 0)
8381 return -EINVAL;
8382
Peter Zijlstra4653f802008-09-23 15:33:44 +02008383 runtime = global_rt_runtime();
8384 period = global_rt_period();
8385
8386 /*
8387 * Sanity check on the sysctl variables.
8388 */
8389 if (runtime > period && runtime != RUNTIME_INF)
8390 return -EINVAL;
Peter Zijlstra10b612f2008-06-19 14:22:27 +02008391
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008392 mutex_lock(&rt_constraints_mutex);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008393 read_lock(&tasklist_lock);
Peter Zijlstra4653f802008-09-23 15:33:44 +02008394 ret = __rt_schedulable(NULL, 0, 0);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02008395 read_unlock(&tasklist_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008396 mutex_unlock(&rt_constraints_mutex);
8397
8398 return ret;
8399}
Dhaval Giani54e99122009-02-27 15:13:54 +05308400
8401int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
8402{
8403 /* Don't accept realtime tasks when there is no way for them to run */
8404 if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0)
8405 return 0;
8406
8407 return 1;
8408}
8409
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008410#else /* !CONFIG_RT_GROUP_SCHED */
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008411static int sched_rt_global_constraints(void)
8412{
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008413 unsigned long flags;
8414 int i;
8415
Hiroshi Shimamotoec5d4982008-09-10 17:00:19 -07008416 if (sysctl_sched_rt_period <= 0)
8417 return -EINVAL;
8418
Peter Zijlstra60aa6052009-05-05 17:50:21 +02008419 /*
8420 * There's always some RT tasks in the root group
8421 * -- migration, kstopmachine etc..
8422 */
8423 if (sysctl_sched_rt_runtime == 0)
8424 return -EBUSY;
8425
Thomas Gleixner0986b112009-11-17 15:32:06 +01008426 raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008427 for_each_possible_cpu(i) {
8428 struct rt_rq *rt_rq = &cpu_rq(i)->rt;
8429
Thomas Gleixner0986b112009-11-17 15:32:06 +01008430 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008431 rt_rq->rt_runtime = global_rt_runtime();
Thomas Gleixner0986b112009-11-17 15:32:06 +01008432 raw_spin_unlock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008433 }
Thomas Gleixner0986b112009-11-17 15:32:06 +01008434 raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02008435
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008436 return 0;
8437}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008438#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008439
8440int sched_rt_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07008441 void __user *buffer, size_t *lenp,
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008442 loff_t *ppos)
8443{
8444 int ret;
8445 int old_period, old_runtime;
8446 static DEFINE_MUTEX(mutex);
8447
8448 mutex_lock(&mutex);
8449 old_period = sysctl_sched_rt_period;
8450 old_runtime = sysctl_sched_rt_runtime;
8451
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07008452 ret = proc_dointvec(table, write, buffer, lenp, ppos);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008453
8454 if (!ret && write) {
8455 ret = sched_rt_global_constraints();
8456 if (ret) {
8457 sysctl_sched_rt_period = old_period;
8458 sysctl_sched_rt_runtime = old_runtime;
8459 } else {
8460 def_rt_bandwidth.rt_runtime = global_rt_runtime();
8461 def_rt_bandwidth.rt_period =
8462 ns_to_ktime(global_rt_period());
8463 }
8464 }
8465 mutex_unlock(&mutex);
8466
8467 return ret;
8468}
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008469
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008470#ifdef CONFIG_CGROUP_SCHED
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008471
8472/* return corresponding task_group object of a cgroup */
Paul Menage2b01dfe2007-10-24 18:23:50 +02008473static inline struct task_group *cgroup_tg(struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008474{
Paul Menage2b01dfe2007-10-24 18:23:50 +02008475 return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id),
8476 struct task_group, css);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008477}
8478
8479static struct cgroup_subsys_state *
Paul Menage2b01dfe2007-10-24 18:23:50 +02008480cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008481{
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008482 struct task_group *tg, *parent;
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008483
Paul Menage2b01dfe2007-10-24 18:23:50 +02008484 if (!cgrp->parent) {
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008485 /* This is early initialization for the top cgroup */
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008486 return &init_task_group.css;
8487 }
8488
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02008489 parent = cgroup_tg(cgrp->parent);
8490 tg = sched_create_group(parent);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008491 if (IS_ERR(tg))
8492 return ERR_PTR(-ENOMEM);
8493
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008494 return &tg->css;
8495}
8496
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01008497static void
8498cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008499{
Paul Menage2b01dfe2007-10-24 18:23:50 +02008500 struct task_group *tg = cgroup_tg(cgrp);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008501
8502 sched_destroy_group(tg);
8503}
8504
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01008505static int
Ben Blumbe367d02009-09-23 15:56:31 -07008506cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008507{
Peter Zijlstrab68aa232008-02-13 15:45:40 +01008508#ifdef CONFIG_RT_GROUP_SCHED
Dhaval Giani54e99122009-02-27 15:13:54 +05308509 if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk))
Peter Zijlstrab68aa232008-02-13 15:45:40 +01008510 return -EINVAL;
8511#else
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008512 /* We don't support RT-tasks being in separate groups */
8513 if (tsk->sched_class != &fair_sched_class)
8514 return -EINVAL;
Peter Zijlstrab68aa232008-02-13 15:45:40 +01008515#endif
Ben Blumbe367d02009-09-23 15:56:31 -07008516 return 0;
8517}
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008518
Ben Blumbe367d02009-09-23 15:56:31 -07008519static int
8520cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
8521 struct task_struct *tsk, bool threadgroup)
8522{
8523 int retval = cpu_cgroup_can_attach_task(cgrp, tsk);
8524 if (retval)
8525 return retval;
8526 if (threadgroup) {
8527 struct task_struct *c;
8528 rcu_read_lock();
8529 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) {
8530 retval = cpu_cgroup_can_attach_task(cgrp, c);
8531 if (retval) {
8532 rcu_read_unlock();
8533 return retval;
8534 }
8535 }
8536 rcu_read_unlock();
8537 }
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008538 return 0;
8539}
8540
8541static void
Paul Menage2b01dfe2007-10-24 18:23:50 +02008542cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
Ben Blumbe367d02009-09-23 15:56:31 -07008543 struct cgroup *old_cont, struct task_struct *tsk,
8544 bool threadgroup)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008545{
8546 sched_move_task(tsk);
Ben Blumbe367d02009-09-23 15:56:31 -07008547 if (threadgroup) {
8548 struct task_struct *c;
8549 rcu_read_lock();
8550 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) {
8551 sched_move_task(c);
8552 }
8553 rcu_read_unlock();
8554 }
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008555}
8556
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008557#ifdef CONFIG_FAIR_GROUP_SCHED
Paul Menagef4c753b2008-04-29 00:59:56 -07008558static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype,
Paul Menage2b01dfe2007-10-24 18:23:50 +02008559 u64 shareval)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008560{
Paul Menage2b01dfe2007-10-24 18:23:50 +02008561 return sched_group_set_shares(cgroup_tg(cgrp), shareval);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008562}
8563
Paul Menagef4c753b2008-04-29 00:59:56 -07008564static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008565{
Paul Menage2b01dfe2007-10-24 18:23:50 +02008566 struct task_group *tg = cgroup_tg(cgrp);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008567
8568 return (u64) tg->shares;
8569}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008570#endif /* CONFIG_FAIR_GROUP_SCHED */
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008571
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008572#ifdef CONFIG_RT_GROUP_SCHED
Mirco Tischler0c708142008-05-14 16:05:46 -07008573static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft,
Paul Menage06ecb272008-04-29 01:00:06 -07008574 s64 val)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008575{
Paul Menage06ecb272008-04-29 01:00:06 -07008576 return sched_group_set_rt_runtime(cgroup_tg(cgrp), val);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008577}
8578
Paul Menage06ecb272008-04-29 01:00:06 -07008579static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008580{
Paul Menage06ecb272008-04-29 01:00:06 -07008581 return sched_group_rt_runtime(cgroup_tg(cgrp));
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008582}
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008583
8584static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype,
8585 u64 rt_period_us)
8586{
8587 return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us);
8588}
8589
8590static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft)
8591{
8592 return sched_group_rt_period(cgroup_tg(cgrp));
8593}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02008594#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008595
Paul Menagefe5c7cc2007-10-29 21:18:11 +01008596static struct cftype cpu_files[] = {
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008597#ifdef CONFIG_FAIR_GROUP_SCHED
Paul Menagefe5c7cc2007-10-29 21:18:11 +01008598 {
8599 .name = "shares",
Paul Menagef4c753b2008-04-29 00:59:56 -07008600 .read_u64 = cpu_shares_read_u64,
8601 .write_u64 = cpu_shares_write_u64,
Paul Menagefe5c7cc2007-10-29 21:18:11 +01008602 },
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008603#endif
8604#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008605 {
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01008606 .name = "rt_runtime_us",
Paul Menage06ecb272008-04-29 01:00:06 -07008607 .read_s64 = cpu_rt_runtime_read,
8608 .write_s64 = cpu_rt_runtime_write,
Peter Zijlstra6f505b12008-01-25 21:08:30 +01008609 },
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008610 {
8611 .name = "rt_period_us",
Paul Menagef4c753b2008-04-29 00:59:56 -07008612 .read_u64 = cpu_rt_period_read_uint,
8613 .write_u64 = cpu_rt_period_write_uint,
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02008614 },
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008615#endif
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008616};
8617
8618static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont)
8619{
Paul Menagefe5c7cc2007-10-29 21:18:11 +01008620 return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files));
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008621}
8622
8623struct cgroup_subsys cpu_cgroup_subsys = {
Ingo Molnar38605ca2007-10-29 21:18:11 +01008624 .name = "cpu",
8625 .create = cpu_cgroup_create,
8626 .destroy = cpu_cgroup_destroy,
8627 .can_attach = cpu_cgroup_can_attach,
8628 .attach = cpu_cgroup_attach,
8629 .populate = cpu_cgroup_populate,
8630 .subsys_id = cpu_cgroup_subsys_id,
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07008631 .early_init = 1,
8632};
8633
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01008634#endif /* CONFIG_CGROUP_SCHED */
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008635
8636#ifdef CONFIG_CGROUP_CPUACCT
8637
8638/*
8639 * CPU accounting code for task groups.
8640 *
8641 * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
8642 * (balbir@in.ibm.com).
8643 */
8644
Bharata B Rao934352f2008-11-10 20:41:13 +05308645/* track cpu usage of a group of tasks and its child groups */
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008646struct cpuacct {
8647 struct cgroup_subsys_state css;
8648 /* cpuusage holds pointer to a u64-type object on every cpu */
Tejun Heo43cf38e2010-02-02 14:38:57 +09008649 u64 __percpu *cpuusage;
Bharata B Raoef12fef2009-03-31 10:02:22 +05308650 struct percpu_counter cpustat[CPUACCT_STAT_NSTATS];
Bharata B Rao934352f2008-11-10 20:41:13 +05308651 struct cpuacct *parent;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008652};
8653
8654struct cgroup_subsys cpuacct_subsys;
8655
8656/* return cpu accounting group corresponding to this container */
Dhaval Giani32cd7562008-02-29 10:02:43 +05308657static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008658{
Dhaval Giani32cd7562008-02-29 10:02:43 +05308659 return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id),
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008660 struct cpuacct, css);
8661}
8662
8663/* return cpu accounting group to which this task belongs */
8664static inline struct cpuacct *task_ca(struct task_struct *tsk)
8665{
8666 return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
8667 struct cpuacct, css);
8668}
8669
8670/* create a new cpu accounting group */
8671static struct cgroup_subsys_state *cpuacct_create(
Dhaval Giani32cd7562008-02-29 10:02:43 +05308672 struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008673{
8674 struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
Bharata B Raoef12fef2009-03-31 10:02:22 +05308675 int i;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008676
8677 if (!ca)
Bharata B Raoef12fef2009-03-31 10:02:22 +05308678 goto out;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008679
8680 ca->cpuusage = alloc_percpu(u64);
Bharata B Raoef12fef2009-03-31 10:02:22 +05308681 if (!ca->cpuusage)
8682 goto out_free_ca;
8683
8684 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
8685 if (percpu_counter_init(&ca->cpustat[i], 0))
8686 goto out_free_counters;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008687
Bharata B Rao934352f2008-11-10 20:41:13 +05308688 if (cgrp->parent)
8689 ca->parent = cgroup_ca(cgrp->parent);
8690
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008691 return &ca->css;
Bharata B Raoef12fef2009-03-31 10:02:22 +05308692
8693out_free_counters:
8694 while (--i >= 0)
8695 percpu_counter_destroy(&ca->cpustat[i]);
8696 free_percpu(ca->cpuusage);
8697out_free_ca:
8698 kfree(ca);
8699out:
8700 return ERR_PTR(-ENOMEM);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008701}
8702
8703/* destroy an existing cpu accounting group */
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01008704static void
Dhaval Giani32cd7562008-02-29 10:02:43 +05308705cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008706{
Dhaval Giani32cd7562008-02-29 10:02:43 +05308707 struct cpuacct *ca = cgroup_ca(cgrp);
Bharata B Raoef12fef2009-03-31 10:02:22 +05308708 int i;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008709
Bharata B Raoef12fef2009-03-31 10:02:22 +05308710 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
8711 percpu_counter_destroy(&ca->cpustat[i]);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008712 free_percpu(ca->cpuusage);
8713 kfree(ca);
8714}
8715
Ken Chen720f5492008-12-15 22:02:01 -08008716static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu)
8717{
Rusty Russellb36128c2009-02-20 16:29:08 +09008718 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
Ken Chen720f5492008-12-15 22:02:01 -08008719 u64 data;
8720
8721#ifndef CONFIG_64BIT
8722 /*
8723 * Take rq->lock to make 64-bit read safe on 32-bit platforms.
8724 */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008725 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08008726 data = *cpuusage;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008727 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08008728#else
8729 data = *cpuusage;
8730#endif
8731
8732 return data;
8733}
8734
8735static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val)
8736{
Rusty Russellb36128c2009-02-20 16:29:08 +09008737 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
Ken Chen720f5492008-12-15 22:02:01 -08008738
8739#ifndef CONFIG_64BIT
8740 /*
8741 * Take rq->lock to make 64-bit write safe on 32-bit platforms.
8742 */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008743 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08008744 *cpuusage = val;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01008745 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08008746#else
8747 *cpuusage = val;
8748#endif
8749}
8750
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008751/* return total cpu usage (in nanoseconds) of a group */
Dhaval Giani32cd7562008-02-29 10:02:43 +05308752static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008753{
Dhaval Giani32cd7562008-02-29 10:02:43 +05308754 struct cpuacct *ca = cgroup_ca(cgrp);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008755 u64 totalcpuusage = 0;
8756 int i;
8757
Ken Chen720f5492008-12-15 22:02:01 -08008758 for_each_present_cpu(i)
8759 totalcpuusage += cpuacct_cpuusage_read(ca, i);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008760
8761 return totalcpuusage;
8762}
8763
Dhaval Giani0297b802008-02-29 10:02:44 +05308764static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype,
8765 u64 reset)
8766{
8767 struct cpuacct *ca = cgroup_ca(cgrp);
8768 int err = 0;
8769 int i;
8770
8771 if (reset) {
8772 err = -EINVAL;
8773 goto out;
8774 }
8775
Ken Chen720f5492008-12-15 22:02:01 -08008776 for_each_present_cpu(i)
8777 cpuacct_cpuusage_write(ca, i, 0);
Dhaval Giani0297b802008-02-29 10:02:44 +05308778
Dhaval Giani0297b802008-02-29 10:02:44 +05308779out:
8780 return err;
8781}
8782
Ken Chene9515c32008-12-15 22:04:15 -08008783static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft,
8784 struct seq_file *m)
8785{
8786 struct cpuacct *ca = cgroup_ca(cgroup);
8787 u64 percpu;
8788 int i;
8789
8790 for_each_present_cpu(i) {
8791 percpu = cpuacct_cpuusage_read(ca, i);
8792 seq_printf(m, "%llu ", (unsigned long long) percpu);
8793 }
8794 seq_printf(m, "\n");
8795 return 0;
8796}
8797
Bharata B Raoef12fef2009-03-31 10:02:22 +05308798static const char *cpuacct_stat_desc[] = {
8799 [CPUACCT_STAT_USER] = "user",
8800 [CPUACCT_STAT_SYSTEM] = "system",
8801};
8802
8803static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft,
8804 struct cgroup_map_cb *cb)
8805{
8806 struct cpuacct *ca = cgroup_ca(cgrp);
8807 int i;
8808
8809 for (i = 0; i < CPUACCT_STAT_NSTATS; i++) {
8810 s64 val = percpu_counter_read(&ca->cpustat[i]);
8811 val = cputime64_to_clock_t(val);
8812 cb->fill(cb, cpuacct_stat_desc[i], val);
8813 }
8814 return 0;
8815}
8816
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008817static struct cftype files[] = {
8818 {
8819 .name = "usage",
Paul Menagef4c753b2008-04-29 00:59:56 -07008820 .read_u64 = cpuusage_read,
8821 .write_u64 = cpuusage_write,
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008822 },
Ken Chene9515c32008-12-15 22:04:15 -08008823 {
8824 .name = "usage_percpu",
8825 .read_seq_string = cpuacct_percpu_seq_read,
8826 },
Bharata B Raoef12fef2009-03-31 10:02:22 +05308827 {
8828 .name = "stat",
8829 .read_map = cpuacct_stats_show,
8830 },
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008831};
8832
Dhaval Giani32cd7562008-02-29 10:02:43 +05308833static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008834{
Dhaval Giani32cd7562008-02-29 10:02:43 +05308835 return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files));
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008836}
8837
8838/*
8839 * charge this task's execution time to its accounting group.
8840 *
8841 * called with rq->lock held.
8842 */
8843static void cpuacct_charge(struct task_struct *tsk, u64 cputime)
8844{
8845 struct cpuacct *ca;
Bharata B Rao934352f2008-11-10 20:41:13 +05308846 int cpu;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008847
Li Zefanc40c6f82009-02-26 15:40:15 +08008848 if (unlikely(!cpuacct_subsys.active))
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008849 return;
8850
Bharata B Rao934352f2008-11-10 20:41:13 +05308851 cpu = task_cpu(tsk);
Bharata B Raoa18b83b2009-03-23 10:02:53 +05308852
8853 rcu_read_lock();
8854
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008855 ca = task_ca(tsk);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008856
Bharata B Rao934352f2008-11-10 20:41:13 +05308857 for (; ca; ca = ca->parent) {
Rusty Russellb36128c2009-02-20 16:29:08 +09008858 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008859 *cpuusage += cputime;
8860 }
Bharata B Raoa18b83b2009-03-23 10:02:53 +05308861
8862 rcu_read_unlock();
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008863}
8864
Bharata B Raoef12fef2009-03-31 10:02:22 +05308865/*
Anton Blanchardfa535a72010-02-02 14:46:13 -08008866 * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large
8867 * in cputime_t units. As a result, cpuacct_update_stats calls
8868 * percpu_counter_add with values large enough to always overflow the
8869 * per cpu batch limit causing bad SMP scalability.
8870 *
8871 * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we
8872 * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled
8873 * and enabled. We cap it at INT_MAX which is the largest allowed batch value.
8874 */
8875#ifdef CONFIG_SMP
8876#define CPUACCT_BATCH \
8877 min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX)
8878#else
8879#define CPUACCT_BATCH 0
8880#endif
8881
8882/*
Bharata B Raoef12fef2009-03-31 10:02:22 +05308883 * Charge the system/user time to the task's accounting group.
8884 */
8885static void cpuacct_update_stats(struct task_struct *tsk,
8886 enum cpuacct_stat_index idx, cputime_t val)
8887{
8888 struct cpuacct *ca;
Anton Blanchardfa535a72010-02-02 14:46:13 -08008889 int batch = CPUACCT_BATCH;
Bharata B Raoef12fef2009-03-31 10:02:22 +05308890
8891 if (unlikely(!cpuacct_subsys.active))
8892 return;
8893
8894 rcu_read_lock();
8895 ca = task_ca(tsk);
8896
8897 do {
Anton Blanchardfa535a72010-02-02 14:46:13 -08008898 __percpu_counter_add(&ca->cpustat[idx], val, batch);
Bharata B Raoef12fef2009-03-31 10:02:22 +05308899 ca = ca->parent;
8900 } while (ca);
8901 rcu_read_unlock();
8902}
8903
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008904struct cgroup_subsys cpuacct_subsys = {
8905 .name = "cpuacct",
8906 .create = cpuacct_create,
8907 .destroy = cpuacct_destroy,
8908 .populate = cpuacct_populate,
8909 .subsys_id = cpuacct_subsys_id,
8910};
8911#endif /* CONFIG_CGROUP_CPUACCT */
Paul E. McKenney03b042b2009-06-25 09:08:16 -07008912
8913#ifndef CONFIG_SMP
8914
Paul E. McKenney03b042b2009-06-25 09:08:16 -07008915void synchronize_sched_expedited(void)
8916{
Paul E. McKenneyfc390cd2010-05-06 11:42:52 -07008917 barrier();
Paul E. McKenney03b042b2009-06-25 09:08:16 -07008918}
8919EXPORT_SYMBOL_GPL(synchronize_sched_expedited);
8920
8921#else /* #ifndef CONFIG_SMP */
8922
Paul E. McKenneycc631fb2010-05-06 18:49:21 +02008923static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0);
Paul E. McKenney03b042b2009-06-25 09:08:16 -07008924
Paul E. McKenneycc631fb2010-05-06 18:49:21 +02008925static int synchronize_sched_expedited_cpu_stop(void *data)
Paul E. McKenney03b042b2009-06-25 09:08:16 -07008926{
Tejun Heo969c7922010-05-06 18:49:21 +02008927 /*
8928 * There must be a full memory barrier on each affected CPU
8929 * between the time that try_stop_cpus() is called and the
8930 * time that it returns.
8931 *
8932 * In the current initial implementation of cpu_stop, the
8933 * above condition is already met when the control reaches
8934 * this point and the following smp_mb() is not strictly
8935 * necessary. Do smp_mb() anyway for documentation and
8936 * robustness against future implementation changes.
8937 */
Paul E. McKenneycc631fb2010-05-06 18:49:21 +02008938 smp_mb(); /* See above comment block. */
Tejun Heo969c7922010-05-06 18:49:21 +02008939 return 0;
Paul E. McKenney03b042b2009-06-25 09:08:16 -07008940}
Paul E. McKenney03b042b2009-06-25 09:08:16 -07008941
8942/*
8943 * Wait for an rcu-sched grace period to elapse, but use "big hammer"
8944 * approach to force grace period to end quickly. This consumes
8945 * significant time on all CPUs, and is thus not recommended for
8946 * any sort of common-case code.
8947 *
8948 * Note that it is illegal to call this function while holding any
8949 * lock that is acquired by a CPU-hotplug notifier. Failing to
8950 * observe this restriction will result in deadlock.
8951 */
8952void synchronize_sched_expedited(void)
8953{
Tejun Heo969c7922010-05-06 18:49:21 +02008954 int snap, trycount = 0;
Paul E. McKenney03b042b2009-06-25 09:08:16 -07008955
8956 smp_mb(); /* ensure prior mod happens before capturing snap. */
Tejun Heo969c7922010-05-06 18:49:21 +02008957 snap = atomic_read(&synchronize_sched_expedited_count) + 1;
Paul E. McKenney03b042b2009-06-25 09:08:16 -07008958 get_online_cpus();
Tejun Heo969c7922010-05-06 18:49:21 +02008959 while (try_stop_cpus(cpu_online_mask,
8960 synchronize_sched_expedited_cpu_stop,
Tejun Heo94458d52010-05-06 18:49:21 +02008961 NULL) == -EAGAIN) {
Paul E. McKenney03b042b2009-06-25 09:08:16 -07008962 put_online_cpus();
8963 if (trycount++ < 10)
8964 udelay(trycount * num_online_cpus());
8965 else {
8966 synchronize_sched();
8967 return;
8968 }
Tejun Heo969c7922010-05-06 18:49:21 +02008969 if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) {
Paul E. McKenney03b042b2009-06-25 09:08:16 -07008970 smp_mb(); /* ensure test happens before caller kfree */
8971 return;
8972 }
8973 get_online_cpus();
8974 }
Tejun Heo969c7922010-05-06 18:49:21 +02008975 atomic_inc(&synchronize_sched_expedited_count);
Paul E. McKenneycc631fb2010-05-06 18:49:21 +02008976 smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */
Paul E. McKenney03b042b2009-06-25 09:08:16 -07008977 put_online_cpus();
Paul E. McKenney03b042b2009-06-25 09:08:16 -07008978}
8979EXPORT_SYMBOL_GPL(synchronize_sched_expedited);
8980
8981#endif /* #else #ifndef CONFIG_SMP */