blob: 6271b89f87acb4eb6ad834732256a62317d14566 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Peter Zijlstra391e43d2011-11-15 17:14:39 +01002 * kernel/sched/core.c
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/mmu_context.h>
36#include <linux/interrupt.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080037#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <linux/completion.h>
39#include <linux/kernel_stat.h>
Ingo Molnar9a11b49a2006-07-03 00:24:33 -070040#include <linux/debug_locks.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020041#include <linux/perf_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/security.h>
43#include <linux/notifier.h>
44#include <linux/profile.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080045#include <linux/freezer.h>
akpm@osdl.org198e2f12006-01-12 01:05:30 -080046#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/blkdev.h>
48#include <linux/delay.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070049#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/smp.h>
51#include <linux/threads.h>
52#include <linux/timer.h>
53#include <linux/rcupdate.h>
54#include <linux/cpu.h>
55#include <linux/cpuset.h>
56#include <linux/percpu.h>
Alexey Dobriyanb5aadf72008-10-06 13:23:43 +040057#include <linux/proc_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#include <linux/seq_file.h>
Nick Piggine692ab52007-07-26 13:40:43 +020059#include <linux/sysctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include <linux/syscalls.h>
61#include <linux/times.h>
Jay Lan8f0ab512006-09-30 23:28:59 -070062#include <linux/tsacct_kern.h>
bibo maoc6fd91f2006-03-26 01:38:20 -080063#include <linux/kprobes.h>
Shailabh Nagar0ff92242006-07-14 00:24:37 -070064#include <linux/delayacct.h>
Ingo Molnardff06c12007-07-09 18:52:00 +020065#include <linux/unistd.h>
Jens Axboef5ff8422007-09-21 09:19:54 +020066#include <linux/pagemap.h>
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +010067#include <linux/hrtimer.h>
Reynes Philippe30914a52008-03-17 16:19:05 -070068#include <linux/tick.h>
Peter Zijlstraf00b45c2008-04-19 19:45:00 +020069#include <linux/debugfs.h>
70#include <linux/ctype.h>
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +020071#include <linux/ftrace.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090072#include <linux/slab.h>
Carsten Emdef1c6f1a2011-10-26 23:14:16 +020073#include <linux/init_task.h>
Al Viro40401532012-02-13 03:58:52 +000074#include <linux/binfmts.h>
Frederic Weisbecker91d1aa432012-11-27 19:33:25 +010075#include <linux/context_tracking.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
David Howells96f951e2012-03-28 18:30:03 +010077#include <asm/switch_to.h>
Eric Dumazet5517d862007-05-08 00:32:57 -070078#include <asm/tlb.h>
Satyam Sharma838225b2007-10-24 18:23:50 +020079#include <asm/irq_regs.h>
Christian Borntraegerdb7e5272012-01-11 08:58:16 +010080#include <asm/mutex.h>
Glauber Costae6e66852011-07-11 15:28:17 -040081#ifdef CONFIG_PARAVIRT
82#include <asm/paravirt.h>
83#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Peter Zijlstra029632f2011-10-25 10:00:11 +020085#include "sched.h"
Peter Zijlstra391e43d2011-11-15 17:14:39 +010086#include "../workqueue_sched.h"
Thomas Gleixner29d5e042012-04-20 13:05:45 +000087#include "../smpboot.h"
Gregory Haskins6e0534f2008-05-12 21:21:01 +020088
Steven Rostedta8d154b2009-04-10 09:36:00 -040089#define CREATE_TRACE_POINTS
Steven Rostedtad8d75f2009-04-14 19:39:12 -040090#include <trace/events/sched.h>
Steven Rostedta8d154b2009-04-10 09:36:00 -040091
Peter Zijlstra029632f2011-10-25 10:00:11 +020092void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period)
Paul Turner58088ad2011-07-21 09:43:31 -070093{
94 unsigned long delta;
95 ktime_t soft, hard, now;
96
97 for (;;) {
98 if (hrtimer_active(period_timer))
99 break;
100
101 now = hrtimer_cb_get_time(period_timer);
102 hrtimer_forward(period_timer, now, period);
103
104 soft = hrtimer_get_softexpires(period_timer);
105 hard = hrtimer_get_expires(period_timer);
106 delta = ktime_to_ns(ktime_sub(hard, soft));
107 __hrtimer_start_range_ns(period_timer, soft, delta,
108 HRTIMER_MODE_ABS_PINNED, 0);
109 }
110}
111
Peter Zijlstra029632f2011-10-25 10:00:11 +0200112DEFINE_MUTEX(sched_domains_mutex);
113DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200114
Peter Zijlstrafe44d622010-12-09 14:15:34 +0100115static void update_rq_clock_task(struct rq *rq, s64 delta);
Venkatesh Pallipadi305e6832010-10-04 17:03:21 -0700116
Peter Zijlstra029632f2011-10-25 10:00:11 +0200117void update_rq_clock(struct rq *rq)
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200118{
Peter Zijlstrafe44d622010-12-09 14:15:34 +0100119 s64 delta;
Venkatesh Pallipadi305e6832010-10-04 17:03:21 -0700120
Mike Galbraith61eadef2011-04-29 08:36:50 +0200121 if (rq->skip_clock_update > 0)
Mike Galbraithf26f9af2010-12-08 11:05:42 +0100122 return;
Venkatesh Pallipadiaa483802010-10-04 17:03:22 -0700123
Peter Zijlstrafe44d622010-12-09 14:15:34 +0100124 delta = sched_clock_cpu(cpu_of(rq)) - rq->clock;
125 rq->clock += delta;
126 update_rq_clock_task(rq, delta);
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200127}
128
Ingo Molnare436d802007-07-19 21:28:35 +0200129/*
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200130 * Debugging: various feature bits
131 */
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200132
133#define SCHED_FEAT(name, enabled) \
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200134 (1UL << __SCHED_FEAT_##name) * enabled |
135
136const_debug unsigned int sysctl_sched_features =
Peter Zijlstra391e43d2011-11-15 17:14:39 +0100137#include "features.h"
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200138 0;
139
140#undef SCHED_FEAT
141
142#ifdef CONFIG_SCHED_DEBUG
143#define SCHED_FEAT(name, enabled) \
144 #name ,
145
Hiroshi Shimamoto12925312012-05-25 15:41:54 +0900146static const char * const sched_feat_names[] = {
Peter Zijlstra391e43d2011-11-15 17:14:39 +0100147#include "features.h"
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200148};
149
150#undef SCHED_FEAT
151
Li Zefan34f3a812008-10-30 15:23:32 +0800152static int sched_feat_show(struct seq_file *m, void *v)
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200153{
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200154 int i;
155
Peter Zijlstraf8b6d1c2011-07-06 14:20:14 +0200156 for (i = 0; i < __SCHED_FEAT_NR; i++) {
Li Zefan34f3a812008-10-30 15:23:32 +0800157 if (!(sysctl_sched_features & (1UL << i)))
158 seq_puts(m, "NO_");
159 seq_printf(m, "%s ", sched_feat_names[i]);
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200160 }
Li Zefan34f3a812008-10-30 15:23:32 +0800161 seq_puts(m, "\n");
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200162
Li Zefan34f3a812008-10-30 15:23:32 +0800163 return 0;
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200164}
165
Peter Zijlstraf8b6d1c2011-07-06 14:20:14 +0200166#ifdef HAVE_JUMP_LABEL
167
Ingo Molnarc5905af2012-02-24 08:31:31 +0100168#define jump_label_key__true STATIC_KEY_INIT_TRUE
169#define jump_label_key__false STATIC_KEY_INIT_FALSE
Peter Zijlstraf8b6d1c2011-07-06 14:20:14 +0200170
171#define SCHED_FEAT(name, enabled) \
172 jump_label_key__##enabled ,
173
Ingo Molnarc5905af2012-02-24 08:31:31 +0100174struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
Peter Zijlstraf8b6d1c2011-07-06 14:20:14 +0200175#include "features.h"
176};
177
178#undef SCHED_FEAT
179
180static void sched_feat_disable(int i)
181{
Ingo Molnarc5905af2012-02-24 08:31:31 +0100182 if (static_key_enabled(&sched_feat_keys[i]))
183 static_key_slow_dec(&sched_feat_keys[i]);
Peter Zijlstraf8b6d1c2011-07-06 14:20:14 +0200184}
185
186static void sched_feat_enable(int i)
187{
Ingo Molnarc5905af2012-02-24 08:31:31 +0100188 if (!static_key_enabled(&sched_feat_keys[i]))
189 static_key_slow_inc(&sched_feat_keys[i]);
Peter Zijlstraf8b6d1c2011-07-06 14:20:14 +0200190}
191#else
192static void sched_feat_disable(int i) { };
193static void sched_feat_enable(int i) { };
194#endif /* HAVE_JUMP_LABEL */
195
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200196static ssize_t
197sched_feat_write(struct file *filp, const char __user *ubuf,
198 size_t cnt, loff_t *ppos)
199{
200 char buf[64];
Mathieu Desnoyers77401912010-09-13 17:47:00 -0400201 char *cmp;
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200202 int neg = 0;
203 int i;
204
205 if (cnt > 63)
206 cnt = 63;
207
208 if (copy_from_user(&buf, ubuf, cnt))
209 return -EFAULT;
210
211 buf[cnt] = 0;
Mathieu Desnoyers77401912010-09-13 17:47:00 -0400212 cmp = strstrip(buf);
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200213
Hillf Danton524429c2011-01-06 20:58:12 +0800214 if (strncmp(cmp, "NO_", 3) == 0) {
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200215 neg = 1;
216 cmp += 3;
217 }
218
Peter Zijlstraf8b6d1c2011-07-06 14:20:14 +0200219 for (i = 0; i < __SCHED_FEAT_NR; i++) {
Mathieu Desnoyers77401912010-09-13 17:47:00 -0400220 if (strcmp(cmp, sched_feat_names[i]) == 0) {
Peter Zijlstraf8b6d1c2011-07-06 14:20:14 +0200221 if (neg) {
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200222 sysctl_sched_features &= ~(1UL << i);
Peter Zijlstraf8b6d1c2011-07-06 14:20:14 +0200223 sched_feat_disable(i);
224 } else {
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200225 sysctl_sched_features |= (1UL << i);
Peter Zijlstraf8b6d1c2011-07-06 14:20:14 +0200226 sched_feat_enable(i);
227 }
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200228 break;
229 }
230 }
231
Peter Zijlstraf8b6d1c2011-07-06 14:20:14 +0200232 if (i == __SCHED_FEAT_NR)
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200233 return -EINVAL;
234
Jan Blunck42994722009-11-20 17:40:37 +0100235 *ppos += cnt;
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200236
237 return cnt;
238}
239
Li Zefan34f3a812008-10-30 15:23:32 +0800240static int sched_feat_open(struct inode *inode, struct file *filp)
241{
242 return single_open(filp, sched_feat_show, NULL);
243}
244
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700245static const struct file_operations sched_feat_fops = {
Li Zefan34f3a812008-10-30 15:23:32 +0800246 .open = sched_feat_open,
247 .write = sched_feat_write,
248 .read = seq_read,
249 .llseek = seq_lseek,
250 .release = single_release,
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200251};
252
253static __init int sched_init_debug(void)
254{
Peter Zijlstraf00b45c2008-04-19 19:45:00 +0200255 debugfs_create_file("sched_features", 0644, NULL, NULL,
256 &sched_feat_fops);
257
258 return 0;
259}
260late_initcall(sched_init_debug);
Peter Zijlstraf8b6d1c2011-07-06 14:20:14 +0200261#endif /* CONFIG_SCHED_DEBUG */
Ingo Molnarbf5c91b2007-10-15 17:00:04 +0200262
263/*
Peter Zijlstrab82d9fd2007-11-09 22:39:39 +0100264 * Number of tasks to iterate in a single balance run.
265 * Limited because this is done with IRQs disabled.
266 */
267const_debug unsigned int sysctl_sched_nr_migrate = 32;
268
269/*
Peter Zijlstrae9e92502009-09-01 10:34:37 +0200270 * period over which we average the RT time consumption, measured
271 * in ms.
272 *
273 * default: 1s
274 */
275const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC;
276
277/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100278 * period over which we measure -rt task cpu usage in us.
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100279 * default: 1s
280 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100281unsigned int sysctl_sched_rt_period = 1000000;
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100282
Peter Zijlstra029632f2011-10-25 10:00:11 +0200283__read_mostly int scheduler_running;
Ingo Molnar6892b752008-02-13 14:02:36 +0100284
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100285/*
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100286 * part of the period that we allow rt tasks to run in us.
287 * default: 0.95s
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100288 */
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100289int sysctl_sched_rt_runtime = 950000;
290
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200291
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
293/*
Peter Zijlstra0122ec52011-04-05 17:23:51 +0200294 * __task_rq_lock - lock the rq @p resides on.
Ingo Molnarb29739f2006-06-27 02:54:51 -0700295 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700296static inline struct rq *__task_rq_lock(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -0700297 __acquires(rq->lock)
298{
Peter Zijlstra0970d292010-02-15 14:45:54 +0100299 struct rq *rq;
300
Peter Zijlstra0122ec52011-04-05 17:23:51 +0200301 lockdep_assert_held(&p->pi_lock);
302
Andi Kleen3a5c3592007-10-15 17:00:14 +0200303 for (;;) {
Peter Zijlstra0970d292010-02-15 14:45:54 +0100304 rq = task_rq(p);
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100305 raw_spin_lock(&rq->lock);
Peter Zijlstra65cc8e42010-03-25 21:05:16 +0100306 if (likely(rq == task_rq(p)))
Andi Kleen3a5c3592007-10-15 17:00:14 +0200307 return rq;
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100308 raw_spin_unlock(&rq->lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -0700309 }
Ingo Molnarb29739f2006-06-27 02:54:51 -0700310}
311
312/*
Peter Zijlstra0122ec52011-04-05 17:23:51 +0200313 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 */
Ingo Molnar70b97a72006-07-03 00:25:42 -0700315static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
Peter Zijlstra0122ec52011-04-05 17:23:51 +0200316 __acquires(p->pi_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 __acquires(rq->lock)
318{
Ingo Molnar70b97a72006-07-03 00:25:42 -0700319 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Andi Kleen3a5c3592007-10-15 17:00:14 +0200321 for (;;) {
Peter Zijlstra0122ec52011-04-05 17:23:51 +0200322 raw_spin_lock_irqsave(&p->pi_lock, *flags);
Andi Kleen3a5c3592007-10-15 17:00:14 +0200323 rq = task_rq(p);
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100324 raw_spin_lock(&rq->lock);
Peter Zijlstra65cc8e42010-03-25 21:05:16 +0100325 if (likely(rq == task_rq(p)))
Andi Kleen3a5c3592007-10-15 17:00:14 +0200326 return rq;
Peter Zijlstra0122ec52011-04-05 17:23:51 +0200327 raw_spin_unlock(&rq->lock);
328 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330}
331
Alexey Dobriyana9957442007-10-15 17:00:13 +0200332static void __task_rq_unlock(struct rq *rq)
Ingo Molnarb29739f2006-06-27 02:54:51 -0700333 __releases(rq->lock)
334{
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100335 raw_spin_unlock(&rq->lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -0700336}
337
Peter Zijlstra0122ec52011-04-05 17:23:51 +0200338static inline void
339task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 __releases(rq->lock)
Peter Zijlstra0122ec52011-04-05 17:23:51 +0200341 __releases(p->pi_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
Peter Zijlstra0122ec52011-04-05 17:23:51 +0200343 raw_spin_unlock(&rq->lock);
344 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345}
346
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347/*
Robert P. J. Daycc2a73b2006-12-10 02:20:00 -0800348 * this_rq_lock - lock this runqueue and disable interrupts.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 */
Alexey Dobriyana9957442007-10-15 17:00:13 +0200350static struct rq *this_rq_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 __acquires(rq->lock)
352{
Ingo Molnar70b97a72006-07-03 00:25:42 -0700353 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
355 local_irq_disable();
356 rq = this_rq();
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100357 raw_spin_lock(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
359 return rq;
360}
361
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100362#ifdef CONFIG_SCHED_HRTICK
363/*
364 * Use HR-timers to deliver accurate preemption points.
365 *
366 * Its all a bit involved since we cannot program an hrt while holding the
367 * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a
368 * reschedule event.
369 *
370 * When we get rescheduled we reprogram the hrtick_timer outside of the
371 * rq->lock.
372 */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100373
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100374static void hrtick_clear(struct rq *rq)
375{
376 if (hrtimer_active(&rq->hrtick_timer))
377 hrtimer_cancel(&rq->hrtick_timer);
378}
379
380/*
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100381 * High-resolution timer tick.
382 * Runs from hardirq context with interrupts disabled.
383 */
384static enum hrtimer_restart hrtick(struct hrtimer *timer)
385{
386 struct rq *rq = container_of(timer, struct rq, hrtick_timer);
387
388 WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
389
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100390 raw_spin_lock(&rq->lock);
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200391 update_rq_clock(rq);
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100392 rq->curr->sched_class->task_tick(rq, rq->curr, 1);
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100393 raw_spin_unlock(&rq->lock);
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100394
395 return HRTIMER_NORESTART;
396}
397
Rabin Vincent95e904c2008-05-11 05:55:33 +0530398#ifdef CONFIG_SMP
Peter Zijlstra31656512008-07-18 18:01:23 +0200399/*
400 * called from hardirq (IPI) context
401 */
402static void __hrtick_start(void *arg)
Peter Zijlstrab328ca12008-04-29 10:02:46 +0200403{
Peter Zijlstra31656512008-07-18 18:01:23 +0200404 struct rq *rq = arg;
Peter Zijlstrab328ca12008-04-29 10:02:46 +0200405
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100406 raw_spin_lock(&rq->lock);
Peter Zijlstra31656512008-07-18 18:01:23 +0200407 hrtimer_restart(&rq->hrtick_timer);
408 rq->hrtick_csd_pending = 0;
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100409 raw_spin_unlock(&rq->lock);
Peter Zijlstrab328ca12008-04-29 10:02:46 +0200410}
411
Peter Zijlstra31656512008-07-18 18:01:23 +0200412/*
413 * Called to set the hrtick timer state.
414 *
415 * called with rq->lock held and irqs disabled
416 */
Peter Zijlstra029632f2011-10-25 10:00:11 +0200417void hrtick_start(struct rq *rq, u64 delay)
Peter Zijlstrab328ca12008-04-29 10:02:46 +0200418{
Peter Zijlstra31656512008-07-18 18:01:23 +0200419 struct hrtimer *timer = &rq->hrtick_timer;
420 ktime_t time = ktime_add_ns(timer->base->get_time(), delay);
Peter Zijlstrab328ca12008-04-29 10:02:46 +0200421
Arjan van de Vencc584b22008-09-01 15:02:30 -0700422 hrtimer_set_expires(timer, time);
Peter Zijlstra31656512008-07-18 18:01:23 +0200423
424 if (rq == this_rq()) {
425 hrtimer_restart(timer);
426 } else if (!rq->hrtick_csd_pending) {
Peter Zijlstra6e275632009-02-25 13:59:48 +0100427 __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0);
Peter Zijlstra31656512008-07-18 18:01:23 +0200428 rq->hrtick_csd_pending = 1;
429 }
Peter Zijlstrab328ca12008-04-29 10:02:46 +0200430}
431
432static int
433hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu)
434{
435 int cpu = (int)(long)hcpu;
436
437 switch (action) {
438 case CPU_UP_CANCELED:
439 case CPU_UP_CANCELED_FROZEN:
440 case CPU_DOWN_PREPARE:
441 case CPU_DOWN_PREPARE_FROZEN:
442 case CPU_DEAD:
443 case CPU_DEAD_FROZEN:
Peter Zijlstra31656512008-07-18 18:01:23 +0200444 hrtick_clear(cpu_rq(cpu));
Peter Zijlstrab328ca12008-04-29 10:02:46 +0200445 return NOTIFY_OK;
446 }
447
448 return NOTIFY_DONE;
449}
450
Rakib Mullickfa748202008-09-22 14:55:45 -0700451static __init void init_hrtick(void)
Peter Zijlstrab328ca12008-04-29 10:02:46 +0200452{
453 hotcpu_notifier(hotplug_hrtick, 0);
454}
Peter Zijlstra31656512008-07-18 18:01:23 +0200455#else
456/*
457 * Called to set the hrtick timer state.
458 *
459 * called with rq->lock held and irqs disabled
460 */
Peter Zijlstra029632f2011-10-25 10:00:11 +0200461void hrtick_start(struct rq *rq, u64 delay)
Peter Zijlstra31656512008-07-18 18:01:23 +0200462{
Peter Zijlstra7f1e2ca2009-03-13 12:21:27 +0100463 __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0,
Arun R Bharadwaj5c333862009-04-16 12:14:37 +0530464 HRTIMER_MODE_REL_PINNED, 0);
Peter Zijlstra31656512008-07-18 18:01:23 +0200465}
466
Andrew Morton006c75f2008-09-22 14:55:46 -0700467static inline void init_hrtick(void)
Peter Zijlstra31656512008-07-18 18:01:23 +0200468{
469}
Rabin Vincent95e904c2008-05-11 05:55:33 +0530470#endif /* CONFIG_SMP */
Peter Zijlstrab328ca12008-04-29 10:02:46 +0200471
472static void init_rq_hrtick(struct rq *rq)
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100473{
Peter Zijlstra31656512008-07-18 18:01:23 +0200474#ifdef CONFIG_SMP
475 rq->hrtick_csd_pending = 0;
476
477 rq->hrtick_csd.flags = 0;
478 rq->hrtick_csd.func = __hrtick_start;
479 rq->hrtick_csd.info = rq;
480#endif
481
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100482 hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
483 rq->hrtick_timer.function = hrtick;
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100484}
Andrew Morton006c75f2008-09-22 14:55:46 -0700485#else /* CONFIG_SCHED_HRTICK */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100486static inline void hrtick_clear(struct rq *rq)
487{
488}
489
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100490static inline void init_rq_hrtick(struct rq *rq)
491{
492}
493
Peter Zijlstrab328ca12008-04-29 10:02:46 +0200494static inline void init_hrtick(void)
495{
496}
Andrew Morton006c75f2008-09-22 14:55:46 -0700497#endif /* CONFIG_SCHED_HRTICK */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +0100498
Ingo Molnar1b9f19c2007-07-09 18:51:59 +0200499/*
Ingo Molnarc24d20d2007-07-09 18:51:59 +0200500 * resched_task - mark a task 'to be rescheduled now'.
501 *
502 * On UP this means the setting of the need_resched flag, on SMP it
503 * might also involve a cross-CPU call to trigger the scheduler on
504 * the target CPU.
505 */
506#ifdef CONFIG_SMP
507
508#ifndef tsk_is_polling
Al Viro16a80162012-06-01 14:22:01 -0400509#define tsk_is_polling(t) 0
Ingo Molnarc24d20d2007-07-09 18:51:59 +0200510#endif
511
Peter Zijlstra029632f2011-10-25 10:00:11 +0200512void resched_task(struct task_struct *p)
Ingo Molnarc24d20d2007-07-09 18:51:59 +0200513{
514 int cpu;
515
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100516 assert_raw_spin_locked(&task_rq(p)->lock);
Ingo Molnarc24d20d2007-07-09 18:51:59 +0200517
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +0800518 if (test_tsk_need_resched(p))
Ingo Molnarc24d20d2007-07-09 18:51:59 +0200519 return;
520
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +0800521 set_tsk_need_resched(p);
Ingo Molnarc24d20d2007-07-09 18:51:59 +0200522
523 cpu = task_cpu(p);
524 if (cpu == smp_processor_id())
525 return;
526
527 /* NEED_RESCHED must be visible before we test polling */
528 smp_mb();
529 if (!tsk_is_polling(p))
530 smp_send_reschedule(cpu);
531}
532
Peter Zijlstra029632f2011-10-25 10:00:11 +0200533void resched_cpu(int cpu)
Ingo Molnarc24d20d2007-07-09 18:51:59 +0200534{
535 struct rq *rq = cpu_rq(cpu);
536 unsigned long flags;
537
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100538 if (!raw_spin_trylock_irqsave(&rq->lock, flags))
Ingo Molnarc24d20d2007-07-09 18:51:59 +0200539 return;
540 resched_task(cpu_curr(cpu));
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100541 raw_spin_unlock_irqrestore(&rq->lock, flags);
Ingo Molnarc24d20d2007-07-09 18:51:59 +0200542}
Thomas Gleixner06d83082008-03-22 09:20:24 +0100543
544#ifdef CONFIG_NO_HZ
545/*
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -0700546 * In the semi idle case, use the nearest busy cpu for migrating timers
547 * from an idle cpu. This is good for power-savings.
548 *
549 * We don't do similar optimization for completely idle system, as
550 * selecting an idle cpu will add more delays to the timers than intended
551 * (as that cpu's timer base may not be uptodate wrt jiffies etc).
552 */
553int get_nohz_timer_target(void)
554{
555 int cpu = smp_processor_id();
556 int i;
557 struct sched_domain *sd;
558
Peter Zijlstra057f3fa2011-04-18 11:24:34 +0200559 rcu_read_lock();
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -0700560 for_each_domain(cpu, sd) {
Peter Zijlstra057f3fa2011-04-18 11:24:34 +0200561 for_each_cpu(i, sched_domain_span(sd)) {
562 if (!idle_cpu(i)) {
563 cpu = i;
564 goto unlock;
565 }
566 }
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -0700567 }
Peter Zijlstra057f3fa2011-04-18 11:24:34 +0200568unlock:
569 rcu_read_unlock();
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -0700570 return cpu;
571}
572/*
Thomas Gleixner06d83082008-03-22 09:20:24 +0100573 * When add_timer_on() enqueues a timer into the timer wheel of an
574 * idle CPU then this timer might expire before the next timer event
575 * which is scheduled to wake up that CPU. In case of a completely
576 * idle system the next event might even be infinite time into the
577 * future. wake_up_idle_cpu() ensures that the CPU is woken up and
578 * leaves the inner idle loop so the newly added timer is taken into
579 * account when the CPU goes back to idle and evaluates the timer
580 * wheel for the next timer event.
581 */
582void wake_up_idle_cpu(int cpu)
583{
584 struct rq *rq = cpu_rq(cpu);
585
586 if (cpu == smp_processor_id())
587 return;
588
589 /*
590 * This is safe, as this function is called with the timer
591 * wheel base lock of (cpu) held. When the CPU is on the way
592 * to idle and has not yet set rq->curr to idle then it will
593 * be serialized on the timer wheel base lock and take the new
594 * timer into account automatically.
595 */
596 if (rq->curr != rq->idle)
597 return;
598
599 /*
600 * We can set TIF_RESCHED on the idle task of the other CPU
601 * lockless. The worst case is that the other CPU runs the
602 * idle task through an additional NOOP schedule()
603 */
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +0800604 set_tsk_need_resched(rq->idle);
Thomas Gleixner06d83082008-03-22 09:20:24 +0100605
606 /* NEED_RESCHED must be visible before we test polling */
607 smp_mb();
608 if (!tsk_is_polling(rq->idle))
609 smp_send_reschedule(cpu);
610}
Mike Galbraith39c0cbe2010-03-11 17:17:13 +0100611
Suresh Siddhaca380622011-10-03 15:09:00 -0700612static inline bool got_nohz_idle_kick(void)
613{
Suresh Siddha1c792db2011-12-01 17:07:32 -0800614 int cpu = smp_processor_id();
615 return idle_cpu(cpu) && test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu));
Suresh Siddhaca380622011-10-03 15:09:00 -0700616}
617
618#else /* CONFIG_NO_HZ */
619
620static inline bool got_nohz_idle_kick(void)
621{
622 return false;
623}
624
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +0200625#endif /* CONFIG_NO_HZ */
Thomas Gleixner06d83082008-03-22 09:20:24 +0100626
Peter Zijlstra029632f2011-10-25 10:00:11 +0200627void sched_avg_update(struct rq *rq)
Peter Zijlstrae9e92502009-09-01 10:34:37 +0200628{
629 s64 period = sched_avg_period();
630
631 while ((s64)(rq->clock - rq->age_stamp) > period) {
Will Deacon0d98bb22010-05-24 12:11:43 -0700632 /*
633 * Inline assembly required to prevent the compiler
634 * optimising this loop into a divmod call.
635 * See __iter_div_u64_rem() for another example of this.
636 */
637 asm("" : "+rm" (rq->age_stamp));
Peter Zijlstrae9e92502009-09-01 10:34:37 +0200638 rq->age_stamp += period;
639 rq->rt_avg /= 2;
640 }
641}
642
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +0200643#else /* !CONFIG_SMP */
Peter Zijlstra029632f2011-10-25 10:00:11 +0200644void resched_task(struct task_struct *p)
Ingo Molnarc24d20d2007-07-09 18:51:59 +0200645{
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100646 assert_raw_spin_locked(&task_rq(p)->lock);
Peter Zijlstra31656512008-07-18 18:01:23 +0200647 set_tsk_need_resched(p);
Ingo Molnarc24d20d2007-07-09 18:51:59 +0200648}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +0200649#endif /* CONFIG_SMP */
Ingo Molnarc24d20d2007-07-09 18:51:59 +0200650
Paul Turnera790de92011-07-21 09:43:29 -0700651#if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
652 (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
Peter Zijlstraeb755802008-08-19 12:33:05 +0200653/*
Paul Turner82774342011-07-21 09:43:35 -0700654 * Iterate task_group tree rooted at *from, calling @down when first entering a
655 * node and @up when leaving it for the final time.
656 *
657 * Caller must hold rcu_lock or sufficient equivalent.
Peter Zijlstraeb755802008-08-19 12:33:05 +0200658 */
Peter Zijlstra029632f2011-10-25 10:00:11 +0200659int walk_tg_tree_from(struct task_group *from,
Paul Turner82774342011-07-21 09:43:35 -0700660 tg_visitor down, tg_visitor up, void *data)
Peter Zijlstraeb755802008-08-19 12:33:05 +0200661{
662 struct task_group *parent, *child;
663 int ret;
664
Paul Turner82774342011-07-21 09:43:35 -0700665 parent = from;
666
Peter Zijlstraeb755802008-08-19 12:33:05 +0200667down:
668 ret = (*down)(parent, data);
669 if (ret)
Paul Turner82774342011-07-21 09:43:35 -0700670 goto out;
Peter Zijlstraeb755802008-08-19 12:33:05 +0200671 list_for_each_entry_rcu(child, &parent->children, siblings) {
672 parent = child;
673 goto down;
674
675up:
676 continue;
677 }
678 ret = (*up)(parent, data);
Paul Turner82774342011-07-21 09:43:35 -0700679 if (ret || parent == from)
680 goto out;
Peter Zijlstraeb755802008-08-19 12:33:05 +0200681
682 child = parent;
683 parent = parent->parent;
684 if (parent)
685 goto up;
Paul Turner82774342011-07-21 09:43:35 -0700686out:
Peter Zijlstraeb755802008-08-19 12:33:05 +0200687 return ret;
688}
689
Peter Zijlstra029632f2011-10-25 10:00:11 +0200690int tg_nop(struct task_group *tg, void *data)
Peter Zijlstraeb755802008-08-19 12:33:05 +0200691{
692 return 0;
693}
694#endif
695
Ingo Molnar45bf76d2007-07-09 18:51:59 +0200696static void set_load_weight(struct task_struct *p)
697{
Nikhil Raof05998d2011-05-18 10:09:38 -0700698 int prio = p->static_prio - MAX_RT_PRIO;
699 struct load_weight *load = &p->se.load;
700
Ingo Molnardd41f592007-07-09 18:51:59 +0200701 /*
702 * SCHED_IDLE tasks get minimal weight:
703 */
704 if (p->policy == SCHED_IDLE) {
Nikhil Raoc8b28112011-05-18 14:37:48 -0700705 load->weight = scale_load(WEIGHT_IDLEPRIO);
Nikhil Raof05998d2011-05-18 10:09:38 -0700706 load->inv_weight = WMULT_IDLEPRIO;
Ingo Molnardd41f592007-07-09 18:51:59 +0200707 return;
708 }
709
Nikhil Raoc8b28112011-05-18 14:37:48 -0700710 load->weight = scale_load(prio_to_weight[prio]);
Nikhil Raof05998d2011-05-18 10:09:38 -0700711 load->inv_weight = prio_to_wmult[prio];
Ingo Molnar45bf76d2007-07-09 18:51:59 +0200712}
713
Peter Zijlstra371fd7e2010-03-24 16:38:48 +0100714static void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
Gregory Haskins2087a1a2008-06-27 14:30:00 -0600715{
Mike Galbraitha64692a2010-03-11 17:16:20 +0100716 update_rq_clock(rq);
Ingo Molnar71f8bd42007-07-09 18:51:59 +0200717 sched_info_queued(p);
Peter Zijlstra371fd7e2010-03-24 16:38:48 +0100718 p->sched_class->enqueue_task(rq, p, flags);
Ingo Molnardd41f592007-07-09 18:51:59 +0200719}
720
Peter Zijlstra371fd7e2010-03-24 16:38:48 +0100721static void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
Ingo Molnardd41f592007-07-09 18:51:59 +0200722{
Mike Galbraitha64692a2010-03-11 17:16:20 +0100723 update_rq_clock(rq);
Ankita Garg46ac22b2008-07-01 14:30:06 +0530724 sched_info_dequeued(p);
Peter Zijlstra371fd7e2010-03-24 16:38:48 +0100725 p->sched_class->dequeue_task(rq, p, flags);
Ingo Molnar71f8bd42007-07-09 18:51:59 +0200726}
727
Peter Zijlstra029632f2011-10-25 10:00:11 +0200728void activate_task(struct rq *rq, struct task_struct *p, int flags)
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +0100729{
730 if (task_contributes_to_load(p))
731 rq->nr_uninterruptible--;
732
Peter Zijlstra371fd7e2010-03-24 16:38:48 +0100733 enqueue_task(rq, p, flags);
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +0100734}
735
Peter Zijlstra029632f2011-10-25 10:00:11 +0200736void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +0100737{
738 if (task_contributes_to_load(p))
739 rq->nr_uninterruptible++;
740
Peter Zijlstra371fd7e2010-03-24 16:38:48 +0100741 dequeue_task(rq, p, flags);
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +0100742}
743
Peter Zijlstrafe44d622010-12-09 14:15:34 +0100744static void update_rq_clock_task(struct rq *rq, s64 delta)
Venkatesh Pallipadiaa483802010-10-04 17:03:22 -0700745{
Glauber Costa095c0aa2011-07-11 15:28:18 -0400746/*
747 * In theory, the compile should just see 0 here, and optimize out the call
748 * to sched_rt_avg_update. But I don't trust it...
749 */
750#if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
751 s64 steal = 0, irq_delta = 0;
752#endif
753#ifdef CONFIG_IRQ_TIME_ACCOUNTING
Peter Zijlstra8e92c202010-12-09 14:15:34 +0100754 irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;
Peter Zijlstrafe44d622010-12-09 14:15:34 +0100755
756 /*
757 * Since irq_time is only updated on {soft,}irq_exit, we might run into
758 * this case when a previous update_rq_clock() happened inside a
759 * {soft,}irq region.
760 *
761 * When this happens, we stop ->clock_task and only update the
762 * prev_irq_time stamp to account for the part that fit, so that a next
763 * update will consume the rest. This ensures ->clock_task is
764 * monotonic.
765 *
766 * It does however cause some slight miss-attribution of {soft,}irq
767 * time, a more accurate solution would be to update the irq_time using
768 * the current rq->clock timestamp, except that would require using
769 * atomic ops.
770 */
771 if (irq_delta > delta)
772 irq_delta = delta;
773
774 rq->prev_irq_time += irq_delta;
775 delta -= irq_delta;
Glauber Costa095c0aa2011-07-11 15:28:18 -0400776#endif
777#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
Ingo Molnarc5905af2012-02-24 08:31:31 +0100778 if (static_key_false((&paravirt_steal_rq_enabled))) {
Glauber Costa095c0aa2011-07-11 15:28:18 -0400779 u64 st;
780
781 steal = paravirt_steal_clock(cpu_of(rq));
782 steal -= rq->prev_steal_time_rq;
783
784 if (unlikely(steal > delta))
785 steal = delta;
786
787 st = steal_ticks(steal);
788 steal = st * TICK_NSEC;
789
790 rq->prev_steal_time_rq += steal;
791
792 delta -= steal;
793 }
794#endif
795
Peter Zijlstrafe44d622010-12-09 14:15:34 +0100796 rq->clock_task += delta;
797
Glauber Costa095c0aa2011-07-11 15:28:18 -0400798#if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
799 if ((irq_delta + steal) && sched_feat(NONTASK_POWER))
800 sched_rt_avg_update(rq, irq_delta + steal);
801#endif
Venkatesh Pallipadiaa483802010-10-04 17:03:22 -0700802}
803
Peter Zijlstra34f971f2010-09-22 13:53:15 +0200804void sched_set_stop_task(int cpu, struct task_struct *stop)
805{
806 struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
807 struct task_struct *old_stop = cpu_rq(cpu)->stop;
808
809 if (stop) {
810 /*
811 * Make it appear like a SCHED_FIFO task, its something
812 * userspace knows about and won't get confused about.
813 *
814 * Also, it will make PI more or less work without too
815 * much confusion -- but then, stop work should not
816 * rely on PI working anyway.
817 */
818 sched_setscheduler_nocheck(stop, SCHED_FIFO, &param);
819
820 stop->sched_class = &stop_sched_class;
821 }
822
823 cpu_rq(cpu)->stop = stop;
824
825 if (old_stop) {
826 /*
827 * Reset it back to a normal scheduling class so that
828 * it can die in pieces.
829 */
830 old_stop->sched_class = &rt_sched_class;
831 }
832}
833
Peter Zijlstra1e3c88b2009-12-17 17:00:43 +0100834/*
Ingo Molnardd41f592007-07-09 18:51:59 +0200835 * __normal_prio - return the priority that is based on the static prio
Ingo Molnar71f8bd42007-07-09 18:51:59 +0200836 */
Ingo Molnar14531182007-07-09 18:51:59 +0200837static inline int __normal_prio(struct task_struct *p)
838{
Ingo Molnardd41f592007-07-09 18:51:59 +0200839 return p->static_prio;
Ingo Molnar14531182007-07-09 18:51:59 +0200840}
841
842/*
Ingo Molnarb29739f2006-06-27 02:54:51 -0700843 * Calculate the expected normal priority: i.e. priority
844 * without taking RT-inheritance into account. Might be
845 * boosted by interactivity modifiers. Changes upon fork,
846 * setprio syscalls, and whenever the interactivity
847 * estimator recalculates.
848 */
Ingo Molnar36c8b582006-07-03 00:25:41 -0700849static inline int normal_prio(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -0700850{
851 int prio;
852
Ingo Molnare05606d2007-07-09 18:51:59 +0200853 if (task_has_rt_policy(p))
Ingo Molnarb29739f2006-06-27 02:54:51 -0700854 prio = MAX_RT_PRIO-1 - p->rt_priority;
855 else
856 prio = __normal_prio(p);
857 return prio;
858}
859
860/*
861 * Calculate the current priority, i.e. the priority
862 * taken into account by the scheduler. This value might
863 * be boosted by RT tasks, or might be boosted by
864 * interactivity modifiers. Will be RT if the task got
865 * RT-boosted. If not then it returns p->normal_prio.
866 */
Ingo Molnar36c8b582006-07-03 00:25:41 -0700867static int effective_prio(struct task_struct *p)
Ingo Molnarb29739f2006-06-27 02:54:51 -0700868{
869 p->normal_prio = normal_prio(p);
870 /*
871 * If we are RT tasks or we were boosted to RT priority,
872 * keep the priority unchanged. Otherwise, update priority
873 * to the normal priority:
874 */
875 if (!rt_prio(p->prio))
876 return p->normal_prio;
877 return p->prio;
878}
879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880/**
881 * task_curr - is this task currently executing on a CPU?
882 * @p: the task in question.
883 */
Ingo Molnar36c8b582006-07-03 00:25:41 -0700884inline int task_curr(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885{
886 return cpu_curr(task_cpu(p)) == p;
887}
888
Steven Rostedtcb469842008-01-25 21:08:22 +0100889static inline void check_class_changed(struct rq *rq, struct task_struct *p,
890 const struct sched_class *prev_class,
Peter Zijlstrada7a7352011-01-17 17:03:27 +0100891 int oldprio)
Steven Rostedtcb469842008-01-25 21:08:22 +0100892{
893 if (prev_class != p->sched_class) {
894 if (prev_class->switched_from)
Peter Zijlstrada7a7352011-01-17 17:03:27 +0100895 prev_class->switched_from(rq, p);
896 p->sched_class->switched_to(rq, p);
897 } else if (oldprio != p->prio)
898 p->sched_class->prio_changed(rq, p, oldprio);
Steven Rostedtcb469842008-01-25 21:08:22 +0100899}
900
Peter Zijlstra029632f2011-10-25 10:00:11 +0200901void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
Peter Zijlstra1e5a7402010-10-31 12:37:04 +0100902{
903 const struct sched_class *class;
904
905 if (p->sched_class == rq->curr->sched_class) {
906 rq->curr->sched_class->check_preempt_curr(rq, p, flags);
907 } else {
908 for_each_class(class) {
909 if (class == rq->curr->sched_class)
910 break;
911 if (class == p->sched_class) {
912 resched_task(rq->curr);
913 break;
914 }
915 }
916 }
917
918 /*
919 * A queue event has occurred, and we're going to schedule. In
920 * this case, we can save a useless back to back clock update.
921 */
Peter Zijlstrafd2f4412011-04-05 17:23:44 +0200922 if (rq->curr->on_rq && test_tsk_need_resched(rq->curr))
Peter Zijlstra1e5a7402010-10-31 12:37:04 +0100923 rq->skip_clock_update = 1;
924}
925
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926#ifdef CONFIG_SMP
Ingo Molnardd41f592007-07-09 18:51:59 +0200927void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
Ingo Molnarc65cc872007-07-09 18:51:58 +0200928{
Peter Zijlstrae2912002009-12-16 18:04:36 +0100929#ifdef CONFIG_SCHED_DEBUG
930 /*
931 * We should never call set_task_cpu() on a blocked task,
932 * ttwu() will sort out the placement.
933 */
Peter Zijlstra077614e2009-12-17 13:16:31 +0100934 WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
935 !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE));
Peter Zijlstra0122ec52011-04-05 17:23:51 +0200936
937#ifdef CONFIG_LOCKDEP
Peter Zijlstra6c6c54e2011-06-03 17:37:07 +0200938 /*
939 * The caller should hold either p->pi_lock or rq->lock, when changing
940 * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks.
941 *
942 * sched_move_task() holds both and thus holding either pins the cgroup,
Peter Zijlstra8323f262012-06-22 13:36:05 +0200943 * see task_group().
Peter Zijlstra6c6c54e2011-06-03 17:37:07 +0200944 *
945 * Furthermore, all task_rq users should acquire both locks, see
946 * task_rq_lock().
947 */
Peter Zijlstra0122ec52011-04-05 17:23:51 +0200948 WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
949 lockdep_is_held(&task_rq(p)->lock)));
950#endif
Peter Zijlstrae2912002009-12-16 18:04:36 +0100951#endif
952
Mathieu Desnoyersde1d7282009-05-05 16:49:59 +0800953 trace_sched_migrate_task(p, new_cpu);
Peter Zijlstracbc34ed2008-12-10 08:08:22 +0100954
Peter Zijlstra0c697742009-12-22 15:43:19 +0100955 if (task_cpu(p) != new_cpu) {
Paul Turner0a74bef2012-10-04 13:18:30 +0200956 if (p->sched_class->migrate_task_rq)
957 p->sched_class->migrate_task_rq(p, new_cpu);
Peter Zijlstra0c697742009-12-22 15:43:19 +0100958 p->se.nr_migrations++;
Peter Zijlstraa8b0ca12011-06-27 14:41:57 +0200959 perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0);
Peter Zijlstra0c697742009-12-22 15:43:19 +0100960 }
Ingo Molnardd41f592007-07-09 18:51:59 +0200961
962 __set_task_cpu(p, new_cpu);
Ingo Molnarc65cc872007-07-09 18:51:58 +0200963}
964
Tejun Heo969c7922010-05-06 18:49:21 +0200965struct migration_arg {
Ingo Molnar36c8b582006-07-03 00:25:41 -0700966 struct task_struct *task;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 int dest_cpu;
Ingo Molnar70b97a72006-07-03 00:25:42 -0700968};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
Tejun Heo969c7922010-05-06 18:49:21 +0200970static int migration_cpu_stop(void *data);
971
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 * wait_task_inactive - wait for a thread to unschedule.
974 *
Roland McGrath85ba2d82008-07-25 19:45:58 -0700975 * If @match_state is nonzero, it's the @p->state value just checked and
976 * not expected to change. If it changes, i.e. @p might have woken up,
977 * then return zero. When we succeed in waiting for @p to be off its CPU,
978 * we return a positive number (its total switch count). If a second call
979 * a short while later returns the same number, the caller can be sure that
980 * @p has remained unscheduled the whole time.
981 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 * The caller must ensure that the task *will* unschedule sometime soon,
983 * else this function might spin for a *long* time. This function can't
984 * be called with interrupts off, or it may introduce deadlock with
985 * smp_call_function() if an IPI is sent by the same process we are
986 * waiting to become inactive.
987 */
Roland McGrath85ba2d82008-07-25 19:45:58 -0700988unsigned long wait_task_inactive(struct task_struct *p, long match_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989{
990 unsigned long flags;
Ingo Molnardd41f592007-07-09 18:51:59 +0200991 int running, on_rq;
Roland McGrath85ba2d82008-07-25 19:45:58 -0700992 unsigned long ncsw;
Ingo Molnar70b97a72006-07-03 00:25:42 -0700993 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
Andi Kleen3a5c3592007-10-15 17:00:14 +0200995 for (;;) {
996 /*
997 * We do the initial early heuristics without holding
998 * any task-queue locks at all. We'll only try to get
999 * the runqueue lock when things look like they will
1000 * work out!
1001 */
1002 rq = task_rq(p);
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07001003
Andi Kleen3a5c3592007-10-15 17:00:14 +02001004 /*
1005 * If the task is actively running on another CPU
1006 * still, just relax and busy-wait without holding
1007 * any locks.
1008 *
1009 * NOTE! Since we don't hold any locks, it's not
1010 * even sure that "rq" stays as the right runqueue!
1011 * But we don't care, since "task_running()" will
1012 * return false if the runqueue has changed and p
1013 * is actually now running somewhere else!
1014 */
Roland McGrath85ba2d82008-07-25 19:45:58 -07001015 while (task_running(rq, p)) {
1016 if (match_state && unlikely(p->state != match_state))
1017 return 0;
Andi Kleen3a5c3592007-10-15 17:00:14 +02001018 cpu_relax();
Roland McGrath85ba2d82008-07-25 19:45:58 -07001019 }
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07001020
Andi Kleen3a5c3592007-10-15 17:00:14 +02001021 /*
1022 * Ok, time to look more closely! We need the rq
1023 * lock now, to be *sure*. If we're wrong, we'll
1024 * just go back and repeat.
1025 */
1026 rq = task_rq_lock(p, &flags);
Peter Zijlstra27a9da62010-05-04 20:36:56 +02001027 trace_sched_wait_task(p);
Andi Kleen3a5c3592007-10-15 17:00:14 +02001028 running = task_running(rq, p);
Peter Zijlstrafd2f4412011-04-05 17:23:44 +02001029 on_rq = p->on_rq;
Roland McGrath85ba2d82008-07-25 19:45:58 -07001030 ncsw = 0;
Oleg Nesterovf31e11d2008-08-20 16:54:44 -07001031 if (!match_state || p->state == match_state)
Oleg Nesterov93dcf552008-08-20 16:54:44 -07001032 ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
Peter Zijlstra0122ec52011-04-05 17:23:51 +02001033 task_rq_unlock(rq, p, &flags);
Linus Torvaldsfa490cf2007-06-18 09:34:40 -07001034
Andi Kleen3a5c3592007-10-15 17:00:14 +02001035 /*
Roland McGrath85ba2d82008-07-25 19:45:58 -07001036 * If it changed from the expected state, bail out now.
1037 */
1038 if (unlikely(!ncsw))
1039 break;
1040
1041 /*
Andi Kleen3a5c3592007-10-15 17:00:14 +02001042 * Was it really running after all now that we
1043 * checked with the proper locks actually held?
1044 *
1045 * Oops. Go back and try again..
1046 */
1047 if (unlikely(running)) {
1048 cpu_relax();
1049 continue;
1050 }
1051
1052 /*
1053 * It's not enough that it's not actively running,
1054 * it must be off the runqueue _entirely_, and not
1055 * preempted!
1056 *
Luis Henriques80dd99b2009-03-16 19:58:09 +00001057 * So if it was still runnable (but just not actively
Andi Kleen3a5c3592007-10-15 17:00:14 +02001058 * running right now), it's preempted, and we should
1059 * yield - it could be a while.
1060 */
1061 if (unlikely(on_rq)) {
Thomas Gleixner8eb90c32011-02-23 23:52:21 +00001062 ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ);
1063
1064 set_current_state(TASK_UNINTERRUPTIBLE);
1065 schedule_hrtimeout(&to, HRTIMER_MODE_REL);
Andi Kleen3a5c3592007-10-15 17:00:14 +02001066 continue;
1067 }
1068
1069 /*
1070 * Ahh, all good. It wasn't running, and it wasn't
1071 * runnable, which means that it will never become
1072 * running in the future either. We're all done!
1073 */
1074 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 }
Roland McGrath85ba2d82008-07-25 19:45:58 -07001076
1077 return ncsw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078}
1079
1080/***
1081 * kick_process - kick a running thread to enter/exit the kernel
1082 * @p: the to-be-kicked thread
1083 *
1084 * Cause a process which is running on another CPU to enter
1085 * kernel-mode, without any delay. (to get signals handled.)
1086 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001087 * NOTE: this function doesn't have to take the runqueue lock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 * because all it wants to ensure is that the remote task enters
1089 * the kernel. If the IPI races and the task has been migrated
1090 * to another CPU then no harm is done and the purpose has been
1091 * achieved as well.
1092 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001093void kick_process(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094{
1095 int cpu;
1096
1097 preempt_disable();
1098 cpu = task_cpu(p);
1099 if ((cpu != smp_processor_id()) && task_curr(p))
1100 smp_send_reschedule(cpu);
1101 preempt_enable();
1102}
Rusty Russellb43e3522009-06-12 22:27:00 -06001103EXPORT_SYMBOL_GPL(kick_process);
Nick Piggin476d1392005-06-25 14:57:29 -07001104#endif /* CONFIG_SMP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Peter Zijlstra970b13b2009-11-25 13:31:39 +01001106#ifdef CONFIG_SMP
Oleg Nesterov30da6882010-03-15 10:10:19 +01001107/*
Peter Zijlstra013fdb82011-04-05 17:23:45 +02001108 * ->cpus_allowed is protected by both rq->lock and p->pi_lock
Oleg Nesterov30da6882010-03-15 10:10:19 +01001109 */
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01001110static int select_fallback_rq(int cpu, struct task_struct *p)
1111{
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01001112 const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu));
Peter Zijlstra2baab4e2012-03-20 15:57:01 +01001113 enum { cpuset, possible, fail } state = cpuset;
1114 int dest_cpu;
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01001115
1116 /* Look for allowed, online CPU in same node. */
Srivatsa S. Bhate3831ed2012-03-30 19:40:28 +05301117 for_each_cpu(dest_cpu, nodemask) {
Peter Zijlstra2baab4e2012-03-20 15:57:01 +01001118 if (!cpu_online(dest_cpu))
1119 continue;
1120 if (!cpu_active(dest_cpu))
1121 continue;
Peter Zijlstrafa17b502011-06-16 12:23:22 +02001122 if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p)))
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01001123 return dest_cpu;
Peter Zijlstra2baab4e2012-03-20 15:57:01 +01001124 }
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01001125
Peter Zijlstra2baab4e2012-03-20 15:57:01 +01001126 for (;;) {
1127 /* Any allowed, online CPU? */
Srivatsa S. Bhate3831ed2012-03-30 19:40:28 +05301128 for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) {
Peter Zijlstra2baab4e2012-03-20 15:57:01 +01001129 if (!cpu_online(dest_cpu))
1130 continue;
1131 if (!cpu_active(dest_cpu))
1132 continue;
1133 goto out;
1134 }
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01001135
Peter Zijlstra2baab4e2012-03-20 15:57:01 +01001136 switch (state) {
1137 case cpuset:
1138 /* No more Mr. Nice Guy. */
1139 cpuset_cpus_allowed_fallback(p);
1140 state = possible;
1141 break;
1142
1143 case possible:
1144 do_set_cpus_allowed(p, cpu_possible_mask);
1145 state = fail;
1146 break;
1147
1148 case fail:
1149 BUG();
1150 break;
1151 }
1152 }
1153
1154out:
1155 if (state != cpuset) {
1156 /*
1157 * Don't tell them about moving exiting tasks or
1158 * kernel threads (both mm NULL), since they never
1159 * leave kernel.
1160 */
1161 if (p->mm && printk_ratelimit()) {
1162 printk_sched("process %d (%s) no longer affine to cpu%d\n",
1163 task_pid_nr(p), p->comm, cpu);
1164 }
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01001165 }
1166
1167 return dest_cpu;
1168}
1169
Peter Zijlstrae2912002009-12-16 18:04:36 +01001170/*
Peter Zijlstra013fdb82011-04-05 17:23:45 +02001171 * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
Peter Zijlstrae2912002009-12-16 18:04:36 +01001172 */
Peter Zijlstra970b13b2009-11-25 13:31:39 +01001173static inline
Peter Zijlstra7608dec2011-04-05 17:23:46 +02001174int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags)
Peter Zijlstra970b13b2009-11-25 13:31:39 +01001175{
Peter Zijlstra7608dec2011-04-05 17:23:46 +02001176 int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags);
Peter Zijlstrae2912002009-12-16 18:04:36 +01001177
1178 /*
1179 * In order not to call set_task_cpu() on a blocking task we need
1180 * to rely on ttwu() to place the task on a valid ->cpus_allowed
1181 * cpu.
1182 *
1183 * Since this is common to all placement strategies, this lives here.
1184 *
1185 * [ this allows ->select_task() to simply return task_cpu(p) and
1186 * not worry about this generic constraint ]
1187 */
Peter Zijlstrafa17b502011-06-16 12:23:22 +02001188 if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) ||
Peter Zijlstra70f11202009-12-20 17:36:27 +01001189 !cpu_online(cpu)))
Peter Zijlstra5da9a0f2009-12-16 18:04:38 +01001190 cpu = select_fallback_rq(task_cpu(p), p);
Peter Zijlstrae2912002009-12-16 18:04:36 +01001191
1192 return cpu;
Peter Zijlstra970b13b2009-11-25 13:31:39 +01001193}
Mike Galbraith09a40af2010-04-15 07:29:59 +02001194
1195static void update_avg(u64 *avg, u64 sample)
1196{
1197 s64 diff = sample - *avg;
1198 *avg += diff >> 3;
1199}
Peter Zijlstra970b13b2009-11-25 13:31:39 +01001200#endif
1201
Peter Zijlstrad7c01d22011-04-05 17:23:43 +02001202static void
Peter Zijlstrab84cb5d2011-04-05 17:23:55 +02001203ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
Tejun Heo9ed38112009-12-03 15:08:03 +09001204{
Peter Zijlstrad7c01d22011-04-05 17:23:43 +02001205#ifdef CONFIG_SCHEDSTATS
Peter Zijlstrab84cb5d2011-04-05 17:23:55 +02001206 struct rq *rq = this_rq();
Tejun Heo9ed38112009-12-03 15:08:03 +09001207
Peter Zijlstrad7c01d22011-04-05 17:23:43 +02001208#ifdef CONFIG_SMP
1209 int this_cpu = smp_processor_id();
Tejun Heo9ed38112009-12-03 15:08:03 +09001210
Peter Zijlstrad7c01d22011-04-05 17:23:43 +02001211 if (cpu == this_cpu) {
1212 schedstat_inc(rq, ttwu_local);
1213 schedstat_inc(p, se.statistics.nr_wakeups_local);
1214 } else {
1215 struct sched_domain *sd;
1216
1217 schedstat_inc(p, se.statistics.nr_wakeups_remote);
Peter Zijlstra057f3fa2011-04-18 11:24:34 +02001218 rcu_read_lock();
Peter Zijlstrad7c01d22011-04-05 17:23:43 +02001219 for_each_domain(this_cpu, sd) {
1220 if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
1221 schedstat_inc(sd, ttwu_wake_remote);
1222 break;
1223 }
1224 }
Peter Zijlstra057f3fa2011-04-18 11:24:34 +02001225 rcu_read_unlock();
Peter Zijlstrad7c01d22011-04-05 17:23:43 +02001226 }
Peter Zijlstraf339b9d2011-05-31 10:49:20 +02001227
1228 if (wake_flags & WF_MIGRATED)
1229 schedstat_inc(p, se.statistics.nr_wakeups_migrate);
1230
Peter Zijlstrad7c01d22011-04-05 17:23:43 +02001231#endif /* CONFIG_SMP */
1232
1233 schedstat_inc(rq, ttwu_count);
1234 schedstat_inc(p, se.statistics.nr_wakeups);
1235
1236 if (wake_flags & WF_SYNC)
1237 schedstat_inc(p, se.statistics.nr_wakeups_sync);
1238
Peter Zijlstrad7c01d22011-04-05 17:23:43 +02001239#endif /* CONFIG_SCHEDSTATS */
Tejun Heo9ed38112009-12-03 15:08:03 +09001240}
1241
Peter Zijlstrad7c01d22011-04-05 17:23:43 +02001242static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
Tejun Heo9ed38112009-12-03 15:08:03 +09001243{
Tejun Heo9ed38112009-12-03 15:08:03 +09001244 activate_task(rq, p, en_flags);
Peter Zijlstrafd2f4412011-04-05 17:23:44 +02001245 p->on_rq = 1;
Peter Zijlstrac2f71152011-04-13 13:28:56 +02001246
1247 /* if a worker is waking up, notify workqueue */
1248 if (p->flags & PF_WQ_WORKER)
1249 wq_worker_waking_up(p, cpu_of(rq));
Tejun Heo9ed38112009-12-03 15:08:03 +09001250}
1251
Peter Zijlstra23f41ee2011-04-05 17:23:56 +02001252/*
1253 * Mark the task runnable and perform wakeup-preemption.
1254 */
Peter Zijlstra89363382011-04-05 17:23:42 +02001255static void
Peter Zijlstra23f41ee2011-04-05 17:23:56 +02001256ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
Tejun Heo9ed38112009-12-03 15:08:03 +09001257{
Peter Zijlstra89363382011-04-05 17:23:42 +02001258 trace_sched_wakeup(p, true);
Tejun Heo9ed38112009-12-03 15:08:03 +09001259 check_preempt_curr(rq, p, wake_flags);
1260
1261 p->state = TASK_RUNNING;
1262#ifdef CONFIG_SMP
1263 if (p->sched_class->task_woken)
1264 p->sched_class->task_woken(rq, p);
1265
Steven Rostedte69c6342010-12-06 17:10:31 -05001266 if (rq->idle_stamp) {
Tejun Heo9ed38112009-12-03 15:08:03 +09001267 u64 delta = rq->clock - rq->idle_stamp;
1268 u64 max = 2*sysctl_sched_migration_cost;
1269
1270 if (delta > max)
1271 rq->avg_idle = max;
1272 else
1273 update_avg(&rq->avg_idle, delta);
1274 rq->idle_stamp = 0;
1275 }
1276#endif
1277}
1278
Peter Zijlstrac05fbaf2011-04-05 17:23:57 +02001279static void
1280ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags)
1281{
1282#ifdef CONFIG_SMP
1283 if (p->sched_contributes_to_load)
1284 rq->nr_uninterruptible--;
1285#endif
1286
1287 ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING);
1288 ttwu_do_wakeup(rq, p, wake_flags);
1289}
1290
1291/*
1292 * Called in case the task @p isn't fully descheduled from its runqueue,
1293 * in this case we must do a remote wakeup. Its a 'light' wakeup though,
1294 * since all we need to do is flip p->state to TASK_RUNNING, since
1295 * the task is still ->on_rq.
1296 */
1297static int ttwu_remote(struct task_struct *p, int wake_flags)
1298{
1299 struct rq *rq;
1300 int ret = 0;
1301
1302 rq = __task_rq_lock(p);
1303 if (p->on_rq) {
1304 ttwu_do_wakeup(rq, p, wake_flags);
1305 ret = 1;
1306 }
1307 __task_rq_unlock(rq);
1308
1309 return ret;
1310}
1311
Peter Zijlstra317f3942011-04-05 17:23:58 +02001312#ifdef CONFIG_SMP
Peter Zijlstrafa14ff42011-09-12 13:06:17 +02001313static void sched_ttwu_pending(void)
Peter Zijlstra317f3942011-04-05 17:23:58 +02001314{
1315 struct rq *rq = this_rq();
Peter Zijlstrafa14ff42011-09-12 13:06:17 +02001316 struct llist_node *llist = llist_del_all(&rq->wake_list);
1317 struct task_struct *p;
Peter Zijlstra317f3942011-04-05 17:23:58 +02001318
1319 raw_spin_lock(&rq->lock);
1320
Peter Zijlstrafa14ff42011-09-12 13:06:17 +02001321 while (llist) {
1322 p = llist_entry(llist, struct task_struct, wake_entry);
1323 llist = llist_next(llist);
Peter Zijlstra317f3942011-04-05 17:23:58 +02001324 ttwu_do_activate(rq, p, 0);
1325 }
1326
1327 raw_spin_unlock(&rq->lock);
1328}
1329
1330void scheduler_ipi(void)
1331{
Suresh Siddhaca380622011-10-03 15:09:00 -07001332 if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick())
Peter Zijlstrac5d753a2011-07-19 15:07:25 -07001333 return;
1334
1335 /*
1336 * Not all reschedule IPI handlers call irq_enter/irq_exit, since
1337 * traditionally all their work was done from the interrupt return
1338 * path. Now that we actually do some work, we need to make sure
1339 * we do call them.
1340 *
1341 * Some archs already do call them, luckily irq_enter/exit nest
1342 * properly.
1343 *
1344 * Arguably we should visit all archs and update all handlers,
1345 * however a fair share of IPIs are still resched only so this would
1346 * somewhat pessimize the simple resched case.
1347 */
1348 irq_enter();
Peter Zijlstrafa14ff42011-09-12 13:06:17 +02001349 sched_ttwu_pending();
Suresh Siddhaca380622011-10-03 15:09:00 -07001350
1351 /*
1352 * Check if someone kicked us for doing the nohz idle load balance.
1353 */
Suresh Siddha6eb57e02011-10-03 15:09:01 -07001354 if (unlikely(got_nohz_idle_kick() && !need_resched())) {
1355 this_rq()->idle_balance = 1;
Suresh Siddhaca380622011-10-03 15:09:00 -07001356 raise_softirq_irqoff(SCHED_SOFTIRQ);
Suresh Siddha6eb57e02011-10-03 15:09:01 -07001357 }
Peter Zijlstrac5d753a2011-07-19 15:07:25 -07001358 irq_exit();
Peter Zijlstra317f3942011-04-05 17:23:58 +02001359}
1360
1361static void ttwu_queue_remote(struct task_struct *p, int cpu)
1362{
Peter Zijlstrafa14ff42011-09-12 13:06:17 +02001363 if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list))
Peter Zijlstra317f3942011-04-05 17:23:58 +02001364 smp_send_reschedule(cpu);
1365}
Peter Zijlstrad6aa8f82011-05-26 14:21:33 +02001366
Peter Zijlstra39be3502012-01-26 12:44:34 +01001367bool cpus_share_cache(int this_cpu, int that_cpu)
Peter Zijlstra518cd622011-12-07 15:07:31 +01001368{
1369 return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
1370}
Peter Zijlstrad6aa8f82011-05-26 14:21:33 +02001371#endif /* CONFIG_SMP */
Peter Zijlstra317f3942011-04-05 17:23:58 +02001372
Peter Zijlstrac05fbaf2011-04-05 17:23:57 +02001373static void ttwu_queue(struct task_struct *p, int cpu)
1374{
1375 struct rq *rq = cpu_rq(cpu);
1376
Daniel Hellstrom17d9f312011-05-20 04:01:10 +00001377#if defined(CONFIG_SMP)
Peter Zijlstra39be3502012-01-26 12:44:34 +01001378 if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
Peter Zijlstraf01114c2011-05-31 12:26:55 +02001379 sched_clock_cpu(cpu); /* sync clocks x-cpu */
Peter Zijlstra317f3942011-04-05 17:23:58 +02001380 ttwu_queue_remote(p, cpu);
1381 return;
1382 }
1383#endif
1384
Peter Zijlstrac05fbaf2011-04-05 17:23:57 +02001385 raw_spin_lock(&rq->lock);
1386 ttwu_do_activate(rq, p, 0);
1387 raw_spin_unlock(&rq->lock);
Tejun Heo9ed38112009-12-03 15:08:03 +09001388}
1389
1390/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 * try_to_wake_up - wake up a thread
Tejun Heo9ed38112009-12-03 15:08:03 +09001392 * @p: the thread to be awakened
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 * @state: the mask of task states that can be woken
Tejun Heo9ed38112009-12-03 15:08:03 +09001394 * @wake_flags: wake modifier flags (WF_*)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 *
1396 * Put it on the run-queue if it's not already there. The "current"
1397 * thread is always on the run-queue (except when the actual
1398 * re-schedule is in progress), and as such you're allowed to do
1399 * the simpler "current->state = TASK_RUNNING" to mark yourself
1400 * runnable without the overhead of this.
1401 *
Tejun Heo9ed38112009-12-03 15:08:03 +09001402 * Returns %true if @p was woken up, %false if it was already running
1403 * or @state didn't match @p's state.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 */
Peter Zijlstrae4a52bc2011-04-05 17:23:54 +02001405static int
1406try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 unsigned long flags;
Peter Zijlstrac05fbaf2011-04-05 17:23:57 +02001409 int cpu, success = 0;
Peter Zijlstra2398f2c2008-06-27 13:41:35 +02001410
Linus Torvalds04e2f172008-02-23 18:05:03 -08001411 smp_wmb();
Peter Zijlstra013fdb82011-04-05 17:23:45 +02001412 raw_spin_lock_irqsave(&p->pi_lock, flags);
Peter Zijlstrae9c84312009-09-15 14:43:03 +02001413 if (!(p->state & state))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 goto out;
1415
Peter Zijlstrac05fbaf2011-04-05 17:23:57 +02001416 success = 1; /* we're going to change ->state */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 cpu = task_cpu(p);
Peter Zijlstrad7c01d22011-04-05 17:23:43 +02001418
Peter Zijlstrac05fbaf2011-04-05 17:23:57 +02001419 if (p->on_rq && ttwu_remote(p, wake_flags))
1420 goto stat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
1422#ifdef CONFIG_SMP
Peter Zijlstrae9c84312009-09-15 14:43:03 +02001423 /*
Peter Zijlstrac05fbaf2011-04-05 17:23:57 +02001424 * If the owning (remote) cpu is still in the middle of schedule() with
1425 * this task as prev, wait until its done referencing the task.
Peter Zijlstrae9c84312009-09-15 14:43:03 +02001426 */
Peter Zijlstraf3e94782012-09-12 11:22:00 +02001427 while (p->on_cpu)
Peter Zijlstrae4a52bc2011-04-05 17:23:54 +02001428 cpu_relax();
Peter Zijlstrae4a52bc2011-04-05 17:23:54 +02001429 /*
1430 * Pairs with the smp_wmb() in finish_lock_switch().
1431 */
1432 smp_rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
Peter Zijlstraa8e4f2e2011-04-05 17:23:49 +02001434 p->sched_contributes_to_load = !!task_contributes_to_load(p);
Peter Zijlstrae9c84312009-09-15 14:43:03 +02001435 p->state = TASK_WAKING;
Peter Zijlstraefbbd052009-12-16 18:04:40 +01001436
Peter Zijlstrae4a52bc2011-04-05 17:23:54 +02001437 if (p->sched_class->task_waking)
Peter Zijlstra74f8e4b2011-04-05 17:23:47 +02001438 p->sched_class->task_waking(p);
Peter Zijlstraab19cb22009-11-27 15:44:43 +01001439
Peter Zijlstra7608dec2011-04-05 17:23:46 +02001440 cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags);
Peter Zijlstraf339b9d2011-05-31 10:49:20 +02001441 if (task_cpu(p) != cpu) {
1442 wake_flags |= WF_MIGRATED;
Mike Galbraithf5dc3752009-10-09 08:35:03 +02001443 set_task_cpu(p, cpu);
Peter Zijlstraf339b9d2011-05-31 10:49:20 +02001444 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445#endif /* CONFIG_SMP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
Peter Zijlstrac05fbaf2011-04-05 17:23:57 +02001447 ttwu_queue(p, cpu);
1448stat:
Peter Zijlstrab84cb5d2011-04-05 17:23:55 +02001449 ttwu_stat(p, cpu, wake_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450out:
Peter Zijlstra013fdb82011-04-05 17:23:45 +02001451 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
1453 return success;
1454}
1455
David Howells50fa6102009-04-28 15:01:38 +01001456/**
Tejun Heo21aa9af2010-06-08 21:40:37 +02001457 * try_to_wake_up_local - try to wake up a local task with rq lock held
1458 * @p: the thread to be awakened
1459 *
Peter Zijlstra2acca552011-04-05 17:23:50 +02001460 * Put @p on the run-queue if it's not already there. The caller must
Tejun Heo21aa9af2010-06-08 21:40:37 +02001461 * ensure that this_rq() is locked, @p is bound to this_rq() and not
Peter Zijlstra2acca552011-04-05 17:23:50 +02001462 * the current task.
Tejun Heo21aa9af2010-06-08 21:40:37 +02001463 */
1464static void try_to_wake_up_local(struct task_struct *p)
1465{
1466 struct rq *rq = task_rq(p);
Tejun Heo21aa9af2010-06-08 21:40:37 +02001467
1468 BUG_ON(rq != this_rq());
1469 BUG_ON(p == current);
1470 lockdep_assert_held(&rq->lock);
1471
Peter Zijlstra2acca552011-04-05 17:23:50 +02001472 if (!raw_spin_trylock(&p->pi_lock)) {
1473 raw_spin_unlock(&rq->lock);
1474 raw_spin_lock(&p->pi_lock);
1475 raw_spin_lock(&rq->lock);
Tejun Heo21aa9af2010-06-08 21:40:37 +02001476 }
Peter Zijlstra2acca552011-04-05 17:23:50 +02001477
Tejun Heo21aa9af2010-06-08 21:40:37 +02001478 if (!(p->state & TASK_NORMAL))
Peter Zijlstra2acca552011-04-05 17:23:50 +02001479 goto out;
Tejun Heo21aa9af2010-06-08 21:40:37 +02001480
Peter Zijlstrafd2f4412011-04-05 17:23:44 +02001481 if (!p->on_rq)
Peter Zijlstrad7c01d22011-04-05 17:23:43 +02001482 ttwu_activate(rq, p, ENQUEUE_WAKEUP);
1483
Peter Zijlstra23f41ee2011-04-05 17:23:56 +02001484 ttwu_do_wakeup(rq, p, 0);
Peter Zijlstrab84cb5d2011-04-05 17:23:55 +02001485 ttwu_stat(p, smp_processor_id(), 0);
Peter Zijlstra2acca552011-04-05 17:23:50 +02001486out:
1487 raw_spin_unlock(&p->pi_lock);
Tejun Heo21aa9af2010-06-08 21:40:37 +02001488}
1489
1490/**
David Howells50fa6102009-04-28 15:01:38 +01001491 * wake_up_process - Wake up a specific process
1492 * @p: The process to be woken up.
1493 *
1494 * Attempt to wake up the nominated process and move it to the set of runnable
1495 * processes. Returns 1 if the process was woken up, 0 if it was already
1496 * running.
1497 *
1498 * It may be assumed that this function implies a write memory barrier before
1499 * changing the task state if and only if any tasks are woken up.
1500 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08001501int wake_up_process(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502{
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05001503 return try_to_wake_up(p, TASK_ALL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505EXPORT_SYMBOL(wake_up_process);
1506
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08001507int wake_up_state(struct task_struct *p, unsigned int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508{
1509 return try_to_wake_up(p, state, 0);
1510}
1511
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512/*
1513 * Perform scheduler related setup for a newly forked process p.
1514 * p is forked by current.
Ingo Molnardd41f592007-07-09 18:51:59 +02001515 *
1516 * __sched_fork() is basic setup used by init_idle() too:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 */
Ingo Molnardd41f592007-07-09 18:51:59 +02001518static void __sched_fork(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519{
Peter Zijlstrafd2f4412011-04-05 17:23:44 +02001520 p->on_rq = 0;
1521
1522 p->se.on_rq = 0;
Ingo Molnardd41f592007-07-09 18:51:59 +02001523 p->se.exec_start = 0;
1524 p->se.sum_exec_runtime = 0;
Ingo Molnarf6cf8912007-08-28 12:53:24 +02001525 p->se.prev_sum_exec_runtime = 0;
Ingo Molnar6c594c22008-12-14 12:34:15 +01001526 p->se.nr_migrations = 0;
Peter Zijlstrada7a7352011-01-17 17:03:27 +01001527 p->se.vruntime = 0;
Peter Zijlstrafd2f4412011-04-05 17:23:44 +02001528 INIT_LIST_HEAD(&p->se.group_node);
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02001529
Paul Turnerf4e26b12012-10-04 13:18:32 +02001530/*
1531 * Load-tracking only depends on SMP, FAIR_GROUP_SCHED dependency below may be
1532 * removed when useful for applications beyond shares distribution (e.g.
1533 * load-balance).
1534 */
1535#if defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)
Paul Turner9d85f212012-10-04 13:18:29 +02001536 p->se.avg.runnable_avg_period = 0;
1537 p->se.avg.runnable_avg_sum = 0;
1538#endif
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02001539#ifdef CONFIG_SCHEDSTATS
Lucas De Marchi41acab82010-03-10 23:37:45 -03001540 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02001541#endif
Nick Piggin476d1392005-06-25 14:57:29 -07001542
Peter Zijlstrafa717062008-01-25 21:08:27 +01001543 INIT_LIST_HEAD(&p->rt.run_list);
Nick Piggin476d1392005-06-25 14:57:29 -07001544
Avi Kivitye107be32007-07-26 13:40:43 +02001545#ifdef CONFIG_PREEMPT_NOTIFIERS
1546 INIT_HLIST_HEAD(&p->preempt_notifiers);
1547#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02001548}
1549
1550/*
1551 * fork()/clone()-time setup:
1552 */
Samir Bellabes3e51e3e2011-05-11 18:18:05 +02001553void sched_fork(struct task_struct *p)
Ingo Molnardd41f592007-07-09 18:51:59 +02001554{
Peter Zijlstra0122ec52011-04-05 17:23:51 +02001555 unsigned long flags;
Ingo Molnardd41f592007-07-09 18:51:59 +02001556 int cpu = get_cpu();
1557
1558 __sched_fork(p);
Peter Zijlstra06b83b52009-12-16 18:04:35 +01001559 /*
Peter Zijlstra0017d732010-03-24 18:34:10 +01001560 * We mark the process as running here. This guarantees that
Peter Zijlstra06b83b52009-12-16 18:04:35 +01001561 * nobody will actually run it, and a signal or other external
1562 * event cannot wake it up and insert it on the runqueue either.
1563 */
Peter Zijlstra0017d732010-03-24 18:34:10 +01001564 p->state = TASK_RUNNING;
Ingo Molnardd41f592007-07-09 18:51:59 +02001565
Ingo Molnarb29739f2006-06-27 02:54:51 -07001566 /*
Mike Galbraithc350a042011-07-27 17:14:55 +02001567 * Make sure we do not leak PI boosting priority to the child.
1568 */
1569 p->prio = current->normal_prio;
1570
1571 /*
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02001572 * Revert to default priority/policy on fork if requested.
1573 */
1574 if (unlikely(p->sched_reset_on_fork)) {
Mike Galbraithc350a042011-07-27 17:14:55 +02001575 if (task_has_rt_policy(p)) {
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02001576 p->policy = SCHED_NORMAL;
Mike Galbraith6c697bdf2009-06-17 10:48:02 +02001577 p->static_prio = NICE_TO_PRIO(0);
Mike Galbraithc350a042011-07-27 17:14:55 +02001578 p->rt_priority = 0;
1579 } else if (PRIO_TO_NICE(p->static_prio) < 0)
1580 p->static_prio = NICE_TO_PRIO(0);
1581
1582 p->prio = p->normal_prio = __normal_prio(p);
1583 set_load_weight(p);
Mike Galbraith6c697bdf2009-06-17 10:48:02 +02001584
Mike Galbraithb9dc29e2009-06-17 10:46:01 +02001585 /*
1586 * We don't need the reset flag anymore after the fork. It has
1587 * fulfilled its duty:
1588 */
1589 p->sched_reset_on_fork = 0;
1590 }
Lennart Poetteringca94c442009-06-15 17:17:47 +02001591
Hiroshi Shimamoto2ddbf952007-10-15 17:00:11 +02001592 if (!rt_prio(p->prio))
1593 p->sched_class = &fair_sched_class;
Ingo Molnarb29739f2006-06-27 02:54:51 -07001594
Peter Zijlstracd29fe62009-11-27 17:32:46 +01001595 if (p->sched_class->task_fork)
1596 p->sched_class->task_fork(p);
1597
Peter Zijlstra86951592010-06-22 11:44:53 +02001598 /*
1599 * The child is not yet in the pid-hash so no cgroup attach races,
1600 * and the cgroup is pinned to this child due to cgroup_fork()
1601 * is ran before sched_fork().
1602 *
1603 * Silence PROVE_RCU.
1604 */
Peter Zijlstra0122ec52011-04-05 17:23:51 +02001605 raw_spin_lock_irqsave(&p->pi_lock, flags);
Peter Zijlstra5f3edc12009-09-10 13:42:00 +02001606 set_task_cpu(p, cpu);
Peter Zijlstra0122ec52011-04-05 17:23:51 +02001607 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
Peter Zijlstra5f3edc12009-09-10 13:42:00 +02001608
Chandra Seetharaman52f17b62006-07-14 00:24:38 -07001609#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
Ingo Molnardd41f592007-07-09 18:51:59 +02001610 if (likely(sched_info_on()))
Chandra Seetharaman52f17b62006-07-14 00:24:38 -07001611 memset(&p->sched_info, 0, sizeof(p->sched_info));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612#endif
Peter Zijlstra3ca7a442011-04-05 17:23:40 +02001613#if defined(CONFIG_SMP)
1614 p->on_cpu = 0;
Nick Piggin4866cde2005-06-25 14:57:23 -07001615#endif
Frederic Weisbeckerbdd4e852011-06-08 01:13:27 +02001616#ifdef CONFIG_PREEMPT_COUNT
Nick Piggin4866cde2005-06-25 14:57:23 -07001617 /* Want to start with kernel preemption disabled. */
Al Viroa1261f52005-11-13 16:06:55 -08001618 task_thread_info(p)->preempt_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619#endif
Dario Faggioli806c09a2010-11-30 19:51:33 +01001620#ifdef CONFIG_SMP
Gregory Haskins917b6272008-12-29 09:39:53 -05001621 plist_node_init(&p->pushable_tasks, MAX_PRIO);
Dario Faggioli806c09a2010-11-30 19:51:33 +01001622#endif
Gregory Haskins917b6272008-12-29 09:39:53 -05001623
Nick Piggin476d1392005-06-25 14:57:29 -07001624 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625}
1626
1627/*
1628 * wake_up_new_task - wake up a newly created task for the first time.
1629 *
1630 * This function will do some initial scheduler statistics housekeeping
1631 * that must be done for every newly created context, then puts the task
1632 * on the runqueue and wakes it.
1633 */
Samir Bellabes3e51e3e2011-05-11 18:18:05 +02001634void wake_up_new_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635{
1636 unsigned long flags;
Ingo Molnardd41f592007-07-09 18:51:59 +02001637 struct rq *rq;
Peter Zijlstrafabf3182010-01-21 21:04:57 +01001638
Peter Zijlstraab2515c2011-04-05 17:23:52 +02001639 raw_spin_lock_irqsave(&p->pi_lock, flags);
Peter Zijlstrafabf3182010-01-21 21:04:57 +01001640#ifdef CONFIG_SMP
1641 /*
1642 * Fork balancing, do it here and not earlier because:
1643 * - cpus_allowed can change in the fork path
1644 * - any previously selected cpu might disappear through hotplug
Peter Zijlstrafabf3182010-01-21 21:04:57 +01001645 */
Peter Zijlstraab2515c2011-04-05 17:23:52 +02001646 set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0));
Peter Zijlstrafabf3182010-01-21 21:04:57 +01001647#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648
Peter Zijlstraab2515c2011-04-05 17:23:52 +02001649 rq = __task_rq_lock(p);
Peter Zijlstracd29fe62009-11-27 17:32:46 +01001650 activate_task(rq, p, 0);
Peter Zijlstrafd2f4412011-04-05 17:23:44 +02001651 p->on_rq = 1;
Peter Zijlstra89363382011-04-05 17:23:42 +02001652 trace_sched_wakeup_new(p, true);
Peter Zijlstraa7558e02009-09-14 20:02:34 +02001653 check_preempt_curr(rq, p, WF_FORK);
Steven Rostedt9a897c52008-01-25 21:08:22 +01001654#ifdef CONFIG_SMP
Peter Zijlstraefbbd052009-12-16 18:04:40 +01001655 if (p->sched_class->task_woken)
1656 p->sched_class->task_woken(rq, p);
Steven Rostedt9a897c52008-01-25 21:08:22 +01001657#endif
Peter Zijlstra0122ec52011-04-05 17:23:51 +02001658 task_rq_unlock(rq, p, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659}
1660
Avi Kivitye107be32007-07-26 13:40:43 +02001661#ifdef CONFIG_PREEMPT_NOTIFIERS
1662
1663/**
Luis Henriques80dd99b2009-03-16 19:58:09 +00001664 * preempt_notifier_register - tell me when current is being preempted & rescheduled
Randy Dunlap421cee22007-07-31 00:37:50 -07001665 * @notifier: notifier struct to register
Avi Kivitye107be32007-07-26 13:40:43 +02001666 */
1667void preempt_notifier_register(struct preempt_notifier *notifier)
1668{
1669 hlist_add_head(&notifier->link, &current->preempt_notifiers);
1670}
1671EXPORT_SYMBOL_GPL(preempt_notifier_register);
1672
1673/**
1674 * preempt_notifier_unregister - no longer interested in preemption notifications
Randy Dunlap421cee22007-07-31 00:37:50 -07001675 * @notifier: notifier struct to unregister
Avi Kivitye107be32007-07-26 13:40:43 +02001676 *
1677 * This is safe to call from within a preemption notifier.
1678 */
1679void preempt_notifier_unregister(struct preempt_notifier *notifier)
1680{
1681 hlist_del(&notifier->link);
1682}
1683EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
1684
1685static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
1686{
1687 struct preempt_notifier *notifier;
1688 struct hlist_node *node;
1689
1690 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
1691 notifier->ops->sched_in(notifier, raw_smp_processor_id());
1692}
1693
1694static void
1695fire_sched_out_preempt_notifiers(struct task_struct *curr,
1696 struct task_struct *next)
1697{
1698 struct preempt_notifier *notifier;
1699 struct hlist_node *node;
1700
1701 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
1702 notifier->ops->sched_out(notifier, next);
1703}
1704
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02001705#else /* !CONFIG_PREEMPT_NOTIFIERS */
Avi Kivitye107be32007-07-26 13:40:43 +02001706
1707static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
1708{
1709}
1710
1711static void
1712fire_sched_out_preempt_notifiers(struct task_struct *curr,
1713 struct task_struct *next)
1714{
1715}
1716
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02001717#endif /* CONFIG_PREEMPT_NOTIFIERS */
Avi Kivitye107be32007-07-26 13:40:43 +02001718
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719/**
Nick Piggin4866cde2005-06-25 14:57:23 -07001720 * prepare_task_switch - prepare to switch tasks
1721 * @rq: the runqueue preparing to switch
Randy Dunlap421cee22007-07-31 00:37:50 -07001722 * @prev: the current task that is being switched out
Nick Piggin4866cde2005-06-25 14:57:23 -07001723 * @next: the task we are going to switch to.
1724 *
1725 * This is called with the rq lock held and interrupts off. It must
1726 * be paired with a subsequent finish_task_switch after the context
1727 * switch.
1728 *
1729 * prepare_task_switch sets up locking and calls architecture specific
1730 * hooks.
1731 */
Avi Kivitye107be32007-07-26 13:40:43 +02001732static inline void
1733prepare_task_switch(struct rq *rq, struct task_struct *prev,
1734 struct task_struct *next)
Nick Piggin4866cde2005-06-25 14:57:23 -07001735{
Andrew Vagin895dd922012-07-12 14:14:29 +04001736 trace_sched_switch(prev, next);
Peter Zijlstrafe4b04f2011-02-02 13:19:09 +01001737 sched_info_switch(prev, next);
1738 perf_event_task_sched_out(prev, next);
Avi Kivitye107be32007-07-26 13:40:43 +02001739 fire_sched_out_preempt_notifiers(prev, next);
Nick Piggin4866cde2005-06-25 14:57:23 -07001740 prepare_lock_switch(rq, next);
1741 prepare_arch_switch(next);
1742}
1743
1744/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 * finish_task_switch - clean up after a task-switch
Jeff Garzik344baba2005-09-07 01:15:17 -04001746 * @rq: runqueue associated with task-switch
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 * @prev: the thread we just switched away from.
1748 *
Nick Piggin4866cde2005-06-25 14:57:23 -07001749 * finish_task_switch must be called after the context switch, paired
1750 * with a prepare_task_switch call before the context switch.
1751 * finish_task_switch will reconcile locking set up by prepare_task_switch,
1752 * and do any other architecture-specific cleanup actions.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 *
1754 * Note that we may have delayed dropping an mm in context_switch(). If
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01001755 * so, we finish that here outside of the runqueue lock. (Doing it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 * with the lock held can cause deadlocks; see schedule() for
1757 * details.)
1758 */
Alexey Dobriyana9957442007-10-15 17:00:13 +02001759static void finish_task_switch(struct rq *rq, struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 __releases(rq->lock)
1761{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 struct mm_struct *mm = rq->prev_mm;
Oleg Nesterov55a101f2006-09-29 02:01:10 -07001763 long prev_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764
1765 rq->prev_mm = NULL;
1766
1767 /*
1768 * A task struct has one reference for the use as "current".
Oleg Nesterovc394cc92006-09-29 02:01:11 -07001769 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
Oleg Nesterov55a101f2006-09-29 02:01:10 -07001770 * schedule one last time. The schedule call will never return, and
1771 * the scheduled task must drop that reference.
Oleg Nesterovc394cc92006-09-29 02:01:11 -07001772 * The test for TASK_DEAD must occur while the runqueue locks are
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 * still held, otherwise prev could be scheduled on another cpu, die
1774 * there before we look at prev->state, and then the reference would
1775 * be dropped twice.
1776 * Manfred Spraul <manfred@colorfullife.com>
1777 */
Oleg Nesterov55a101f2006-09-29 02:01:10 -07001778 prev_state = prev->state;
Frederic Weisbeckerbf9fae92012-09-08 15:23:11 +02001779 vtime_task_switch(prev);
Nick Piggin4866cde2005-06-25 14:57:23 -07001780 finish_arch_switch(prev);
Stephane Eraniana8d757e2011-08-25 15:58:03 +02001781 perf_event_task_sched_in(prev, current);
Nick Piggin4866cde2005-06-25 14:57:23 -07001782 finish_lock_switch(rq, prev);
Catalin Marinas01f23e12011-11-27 21:43:10 +00001783 finish_arch_post_lock_switch();
Steven Rostedte8fa1362008-01-25 21:08:05 +01001784
Avi Kivitye107be32007-07-26 13:40:43 +02001785 fire_sched_in_preempt_notifiers(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 if (mm)
1787 mmdrop(mm);
Oleg Nesterovc394cc92006-09-29 02:01:11 -07001788 if (unlikely(prev_state == TASK_DEAD)) {
bibo maoc6fd91f2006-03-26 01:38:20 -08001789 /*
1790 * Remove function-return probe instances associated with this
1791 * task and put them back on the free list.
Ingo Molnar9761eea2007-07-09 18:52:00 +02001792 */
bibo maoc6fd91f2006-03-26 01:38:20 -08001793 kprobe_flush_task(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 put_task_struct(prev);
bibo maoc6fd91f2006-03-26 01:38:20 -08001795 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796}
1797
Gregory Haskins3f029d32009-07-29 11:08:47 -04001798#ifdef CONFIG_SMP
1799
1800/* assumes rq->lock is held */
1801static inline void pre_schedule(struct rq *rq, struct task_struct *prev)
1802{
1803 if (prev->sched_class->pre_schedule)
1804 prev->sched_class->pre_schedule(rq, prev);
1805}
1806
1807/* rq->lock is NOT held, but preemption is disabled */
1808static inline void post_schedule(struct rq *rq)
1809{
1810 if (rq->post_schedule) {
1811 unsigned long flags;
1812
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001813 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins3f029d32009-07-29 11:08:47 -04001814 if (rq->curr->sched_class->post_schedule)
1815 rq->curr->sched_class->post_schedule(rq);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001816 raw_spin_unlock_irqrestore(&rq->lock, flags);
Gregory Haskins3f029d32009-07-29 11:08:47 -04001817
1818 rq->post_schedule = 0;
1819 }
1820}
1821
1822#else
1823
1824static inline void pre_schedule(struct rq *rq, struct task_struct *p)
1825{
1826}
1827
1828static inline void post_schedule(struct rq *rq)
1829{
1830}
1831
1832#endif
1833
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834/**
1835 * schedule_tail - first thing a freshly forked thread must call.
1836 * @prev: the thread we just switched away from.
1837 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001838asmlinkage void schedule_tail(struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 __releases(rq->lock)
1840{
Ingo Molnar70b97a72006-07-03 00:25:42 -07001841 struct rq *rq = this_rq();
1842
Nick Piggin4866cde2005-06-25 14:57:23 -07001843 finish_task_switch(rq, prev);
Steven Rostedtda19ab52009-07-29 00:21:22 -04001844
Gregory Haskins3f029d32009-07-29 11:08:47 -04001845 /*
1846 * FIXME: do we need to worry about rq being invalidated by the
1847 * task_switch?
1848 */
1849 post_schedule(rq);
Steven Rostedtda19ab52009-07-29 00:21:22 -04001850
Nick Piggin4866cde2005-06-25 14:57:23 -07001851#ifdef __ARCH_WANT_UNLOCKED_CTXSW
1852 /* In this case, finish_task_switch does not reenable preemption */
1853 preempt_enable();
1854#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 if (current->set_child_tid)
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001856 put_user(task_pid_vnr(current), current->set_child_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857}
1858
1859/*
1860 * context_switch - switch to the new MM and the new
1861 * thread's register state.
1862 */
Ingo Molnardd41f592007-07-09 18:51:59 +02001863static inline void
Ingo Molnar70b97a72006-07-03 00:25:42 -07001864context_switch(struct rq *rq, struct task_struct *prev,
Ingo Molnar36c8b582006-07-03 00:25:41 -07001865 struct task_struct *next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866{
Ingo Molnardd41f592007-07-09 18:51:59 +02001867 struct mm_struct *mm, *oldmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868
Avi Kivitye107be32007-07-26 13:40:43 +02001869 prepare_task_switch(rq, prev, next);
Peter Zijlstrafe4b04f2011-02-02 13:19:09 +01001870
Ingo Molnardd41f592007-07-09 18:51:59 +02001871 mm = next->mm;
1872 oldmm = prev->active_mm;
Zachary Amsden9226d122007-02-13 13:26:21 +01001873 /*
1874 * For paravirt, this is coupled with an exit in switch_to to
1875 * combine the page table reload and the switch backend into
1876 * one hypercall.
1877 */
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -08001878 arch_start_context_switch(prev);
Zachary Amsden9226d122007-02-13 13:26:21 +01001879
Heiko Carstens31915ab2010-09-16 14:42:25 +02001880 if (!mm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 next->active_mm = oldmm;
1882 atomic_inc(&oldmm->mm_count);
1883 enter_lazy_tlb(oldmm, next);
1884 } else
1885 switch_mm(oldmm, mm, next);
1886
Heiko Carstens31915ab2010-09-16 14:42:25 +02001887 if (!prev->mm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 prev->active_mm = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 rq->prev_mm = oldmm;
1890 }
Ingo Molnar3a5f5e42006-07-14 00:24:27 -07001891 /*
1892 * Since the runqueue lock will be released by the next
1893 * task (which is an invalid locking op but in the case
1894 * of the scheduler it's an obvious special-case), so we
1895 * do an early lockdep release here:
1896 */
1897#ifndef __ARCH_WANT_UNLOCKED_CTXSW
Ingo Molnar8a25d5d2006-07-03 00:24:54 -07001898 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
Ingo Molnar3a5f5e42006-07-14 00:24:27 -07001899#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
Frederic Weisbecker91d1aa432012-11-27 19:33:25 +01001901 context_tracking_task_switch(prev, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 /* Here we just switch the register state and the stack. */
1903 switch_to(prev, next, prev);
1904
Ingo Molnardd41f592007-07-09 18:51:59 +02001905 barrier();
1906 /*
1907 * this_rq must be evaluated again because prev may have moved
1908 * CPUs since it called schedule(), thus the 'rq' on its stack
1909 * frame will be invalid.
1910 */
1911 finish_task_switch(this_rq(), prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912}
1913
1914/*
1915 * nr_running, nr_uninterruptible and nr_context_switches:
1916 *
1917 * externally visible scheduler statistics: current number of runnable
1918 * threads, current number of uninterruptible-sleeping threads, total
1919 * number of context switches performed since bootup.
1920 */
1921unsigned long nr_running(void)
1922{
1923 unsigned long i, sum = 0;
1924
1925 for_each_online_cpu(i)
1926 sum += cpu_rq(i)->nr_running;
1927
1928 return sum;
1929}
1930
1931unsigned long nr_uninterruptible(void)
1932{
1933 unsigned long i, sum = 0;
1934
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08001935 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 sum += cpu_rq(i)->nr_uninterruptible;
1937
1938 /*
1939 * Since we read the counters lockless, it might be slightly
1940 * inaccurate. Do not allow it to go below zero though:
1941 */
1942 if (unlikely((long)sum < 0))
1943 sum = 0;
1944
1945 return sum;
1946}
1947
1948unsigned long long nr_context_switches(void)
1949{
Steven Rostedtcc94abf2006-06-27 02:54:31 -07001950 int i;
1951 unsigned long long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08001953 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 sum += cpu_rq(i)->nr_switches;
1955
1956 return sum;
1957}
1958
1959unsigned long nr_iowait(void)
1960{
1961 unsigned long i, sum = 0;
1962
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08001963 for_each_possible_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 sum += atomic_read(&cpu_rq(i)->nr_iowait);
1965
1966 return sum;
1967}
1968
Peter Zijlstra8c215bd2010-07-01 09:07:17 +02001969unsigned long nr_iowait_cpu(int cpu)
Arjan van de Ven69d25872009-09-21 17:04:08 -07001970{
Peter Zijlstra8c215bd2010-07-01 09:07:17 +02001971 struct rq *this = cpu_rq(cpu);
Arjan van de Ven69d25872009-09-21 17:04:08 -07001972 return atomic_read(&this->nr_iowait);
1973}
1974
1975unsigned long this_cpu_load(void)
1976{
1977 struct rq *this = this_rq();
1978 return this->cpu_load[0];
1979}
1980
1981
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02001982/*
1983 * Global load-average calculations
1984 *
1985 * We take a distributed and async approach to calculating the global load-avg
1986 * in order to minimize overhead.
1987 *
1988 * The global load average is an exponentially decaying average of nr_running +
1989 * nr_uninterruptible.
1990 *
1991 * Once every LOAD_FREQ:
1992 *
1993 * nr_active = 0;
1994 * for_each_possible_cpu(cpu)
1995 * nr_active += cpu_of(cpu)->nr_running + cpu_of(cpu)->nr_uninterruptible;
1996 *
1997 * avenrun[n] = avenrun[0] * exp_n + nr_active * (1 - exp_n)
1998 *
1999 * Due to a number of reasons the above turns in the mess below:
2000 *
2001 * - for_each_possible_cpu() is prohibitively expensive on machines with
2002 * serious number of cpus, therefore we need to take a distributed approach
2003 * to calculating nr_active.
2004 *
2005 * \Sum_i x_i(t) = \Sum_i x_i(t) - x_i(t_0) | x_i(t_0) := 0
2006 * = \Sum_i { \Sum_j=1 x_i(t_j) - x_i(t_j-1) }
2007 *
2008 * So assuming nr_active := 0 when we start out -- true per definition, we
2009 * can simply take per-cpu deltas and fold those into a global accumulate
2010 * to obtain the same result. See calc_load_fold_active().
2011 *
2012 * Furthermore, in order to avoid synchronizing all per-cpu delta folding
2013 * across the machine, we assume 10 ticks is sufficient time for every
2014 * cpu to have completed this task.
2015 *
2016 * This places an upper-bound on the IRQ-off latency of the machine. Then
2017 * again, being late doesn't loose the delta, just wrecks the sample.
2018 *
2019 * - cpu_rq()->nr_uninterruptible isn't accurately tracked per-cpu because
2020 * this would add another cross-cpu cacheline miss and atomic operation
2021 * to the wakeup path. Instead we increment on whatever cpu the task ran
2022 * when it went into uninterruptible state and decrement on whatever cpu
2023 * did the wakeup. This means that only the sum of nr_uninterruptible over
2024 * all cpus yields the correct result.
2025 *
2026 * This covers the NO_HZ=n code, for extra head-aches, see the comment below.
2027 */
2028
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002029/* Variables and functions for calc_load */
2030static atomic_long_t calc_load_tasks;
2031static unsigned long calc_load_update;
2032unsigned long avenrun[3];
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002033EXPORT_SYMBOL(avenrun); /* should be removed */
2034
2035/**
2036 * get_avenrun - get the load average array
2037 * @loads: pointer to dest load array
2038 * @offset: offset to add
2039 * @shift: shift count to shift the result left
2040 *
2041 * These values are estimates at best, so no need for locking.
2042 */
2043void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
2044{
2045 loads[0] = (avenrun[0] + offset) << shift;
2046 loads[1] = (avenrun[1] + offset) << shift;
2047 loads[2] = (avenrun[2] + offset) << shift;
2048}
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002049
Peter Zijlstra74f51872010-04-22 21:50:19 +02002050static long calc_load_fold_active(struct rq *this_rq)
2051{
2052 long nr_active, delta = 0;
2053
2054 nr_active = this_rq->nr_running;
2055 nr_active += (long) this_rq->nr_uninterruptible;
2056
2057 if (nr_active != this_rq->calc_load_active) {
2058 delta = nr_active - this_rq->calc_load_active;
2059 this_rq->calc_load_active = nr_active;
2060 }
2061
2062 return delta;
2063}
2064
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002065/*
2066 * a1 = a0 * e + a * (1 - e)
2067 */
Peter Zijlstra0f004f52010-11-30 19:48:45 +01002068static unsigned long
2069calc_load(unsigned long load, unsigned long exp, unsigned long active)
2070{
2071 load *= exp;
2072 load += active * (FIXED_1 - exp);
2073 load += 1UL << (FSHIFT - 1);
2074 return load >> FSHIFT;
2075}
2076
Peter Zijlstra74f51872010-04-22 21:50:19 +02002077#ifdef CONFIG_NO_HZ
2078/*
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002079 * Handle NO_HZ for the global load-average.
2080 *
2081 * Since the above described distributed algorithm to compute the global
2082 * load-average relies on per-cpu sampling from the tick, it is affected by
2083 * NO_HZ.
2084 *
2085 * The basic idea is to fold the nr_active delta into a global idle-delta upon
2086 * entering NO_HZ state such that we can include this as an 'extra' cpu delta
2087 * when we read the global state.
2088 *
2089 * Obviously reality has to ruin such a delightfully simple scheme:
2090 *
2091 * - When we go NO_HZ idle during the window, we can negate our sample
2092 * contribution, causing under-accounting.
2093 *
2094 * We avoid this by keeping two idle-delta counters and flipping them
2095 * when the window starts, thus separating old and new NO_HZ load.
2096 *
2097 * The only trick is the slight shift in index flip for read vs write.
2098 *
2099 * 0s 5s 10s 15s
2100 * +10 +10 +10 +10
2101 * |-|-----------|-|-----------|-|-----------|-|
2102 * r:0 0 1 1 0 0 1 1 0
2103 * w:0 1 1 0 0 1 1 0 0
2104 *
2105 * This ensures we'll fold the old idle contribution in this window while
2106 * accumlating the new one.
2107 *
2108 * - When we wake up from NO_HZ idle during the window, we push up our
2109 * contribution, since we effectively move our sample point to a known
2110 * busy state.
2111 *
2112 * This is solved by pushing the window forward, and thus skipping the
2113 * sample, for this cpu (effectively using the idle-delta for this cpu which
2114 * was in effect at the time the window opened). This also solves the issue
2115 * of having to deal with a cpu having been in NOHZ idle for multiple
2116 * LOAD_FREQ intervals.
Peter Zijlstra74f51872010-04-22 21:50:19 +02002117 *
2118 * When making the ILB scale, we should try to pull this in as well.
2119 */
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002120static atomic_long_t calc_load_idle[2];
2121static int calc_load_idx;
Peter Zijlstra74f51872010-04-22 21:50:19 +02002122
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002123static inline int calc_load_write_idx(void)
Peter Zijlstra74f51872010-04-22 21:50:19 +02002124{
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002125 int idx = calc_load_idx;
2126
2127 /*
2128 * See calc_global_nohz(), if we observe the new index, we also
2129 * need to observe the new update time.
2130 */
2131 smp_rmb();
2132
2133 /*
2134 * If the folding window started, make sure we start writing in the
2135 * next idle-delta.
2136 */
2137 if (!time_before(jiffies, calc_load_update))
2138 idx++;
2139
2140 return idx & 1;
2141}
2142
2143static inline int calc_load_read_idx(void)
2144{
2145 return calc_load_idx & 1;
2146}
2147
2148void calc_load_enter_idle(void)
2149{
2150 struct rq *this_rq = this_rq();
Peter Zijlstra74f51872010-04-22 21:50:19 +02002151 long delta;
2152
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002153 /*
2154 * We're going into NOHZ mode, if there's any pending delta, fold it
2155 * into the pending idle delta.
2156 */
Peter Zijlstra74f51872010-04-22 21:50:19 +02002157 delta = calc_load_fold_active(this_rq);
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002158 if (delta) {
2159 int idx = calc_load_write_idx();
2160 atomic_long_add(delta, &calc_load_idle[idx]);
2161 }
2162}
2163
2164void calc_load_exit_idle(void)
2165{
2166 struct rq *this_rq = this_rq();
2167
2168 /*
2169 * If we're still before the sample window, we're done.
2170 */
2171 if (time_before(jiffies, this_rq->calc_load_update))
2172 return;
2173
2174 /*
2175 * We woke inside or after the sample window, this means we're already
2176 * accounted through the nohz accounting, so skip the entire deal and
2177 * sync up for the next window.
2178 */
2179 this_rq->calc_load_update = calc_load_update;
2180 if (time_before(jiffies, this_rq->calc_load_update + 10))
2181 this_rq->calc_load_update += LOAD_FREQ;
Peter Zijlstra74f51872010-04-22 21:50:19 +02002182}
2183
2184static long calc_load_fold_idle(void)
2185{
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002186 int idx = calc_load_read_idx();
Peter Zijlstra74f51872010-04-22 21:50:19 +02002187 long delta = 0;
2188
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002189 if (atomic_long_read(&calc_load_idle[idx]))
2190 delta = atomic_long_xchg(&calc_load_idle[idx], 0);
Peter Zijlstra74f51872010-04-22 21:50:19 +02002191
2192 return delta;
2193}
Peter Zijlstra0f004f52010-11-30 19:48:45 +01002194
2195/**
2196 * fixed_power_int - compute: x^n, in O(log n) time
2197 *
2198 * @x: base of the power
2199 * @frac_bits: fractional bits of @x
2200 * @n: power to raise @x to.
2201 *
2202 * By exploiting the relation between the definition of the natural power
2203 * function: x^n := x*x*...*x (x multiplied by itself for n times), and
2204 * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i,
2205 * (where: n_i \elem {0, 1}, the binary vector representing n),
2206 * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is
2207 * of course trivially computable in O(log_2 n), the length of our binary
2208 * vector.
2209 */
2210static unsigned long
2211fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n)
2212{
2213 unsigned long result = 1UL << frac_bits;
2214
2215 if (n) for (;;) {
2216 if (n & 1) {
2217 result *= x;
2218 result += 1UL << (frac_bits - 1);
2219 result >>= frac_bits;
2220 }
2221 n >>= 1;
2222 if (!n)
2223 break;
2224 x *= x;
2225 x += 1UL << (frac_bits - 1);
2226 x >>= frac_bits;
2227 }
2228
2229 return result;
2230}
2231
2232/*
2233 * a1 = a0 * e + a * (1 - e)
2234 *
2235 * a2 = a1 * e + a * (1 - e)
2236 * = (a0 * e + a * (1 - e)) * e + a * (1 - e)
2237 * = a0 * e^2 + a * (1 - e) * (1 + e)
2238 *
2239 * a3 = a2 * e + a * (1 - e)
2240 * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e)
2241 * = a0 * e^3 + a * (1 - e) * (1 + e + e^2)
2242 *
2243 * ...
2244 *
2245 * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1]
2246 * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e)
2247 * = a0 * e^n + a * (1 - e^n)
2248 *
2249 * [1] application of the geometric series:
2250 *
2251 * n 1 - x^(n+1)
2252 * S_n := \Sum x^i = -------------
2253 * i=0 1 - x
2254 */
2255static unsigned long
2256calc_load_n(unsigned long load, unsigned long exp,
2257 unsigned long active, unsigned int n)
2258{
2259
2260 return calc_load(load, fixed_power_int(exp, FSHIFT, n), active);
2261}
2262
2263/*
2264 * NO_HZ can leave us missing all per-cpu ticks calling
2265 * calc_load_account_active(), but since an idle CPU folds its delta into
2266 * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold
2267 * in the pending idle delta if our idle period crossed a load cycle boundary.
2268 *
2269 * Once we've updated the global active value, we need to apply the exponential
2270 * weights adjusted to the number of cycles missed.
2271 */
Peter Zijlstrac308b562012-03-01 15:04:46 +01002272static void calc_global_nohz(void)
Peter Zijlstra0f004f52010-11-30 19:48:45 +01002273{
2274 long delta, active, n;
2275
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002276 if (!time_before(jiffies, calc_load_update + 10)) {
2277 /*
2278 * Catch-up, fold however many we are behind still
2279 */
2280 delta = jiffies - calc_load_update - 10;
2281 n = 1 + (delta / LOAD_FREQ);
2282
2283 active = atomic_long_read(&calc_load_tasks);
2284 active = active > 0 ? active * FIXED_1 : 0;
2285
2286 avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n);
2287 avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n);
2288 avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n);
2289
2290 calc_load_update += n * LOAD_FREQ;
2291 }
Peter Zijlstra0f004f52010-11-30 19:48:45 +01002292
2293 /*
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002294 * Flip the idle index...
2295 *
2296 * Make sure we first write the new time then flip the index, so that
2297 * calc_load_write_idx() will see the new time when it reads the new
2298 * index, this avoids a double flip messing things up.
Peter Zijlstra0f004f52010-11-30 19:48:45 +01002299 */
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002300 smp_wmb();
2301 calc_load_idx++;
Peter Zijlstra0f004f52010-11-30 19:48:45 +01002302}
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002303#else /* !CONFIG_NO_HZ */
Peter Zijlstra74f51872010-04-22 21:50:19 +02002304
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002305static inline long calc_load_fold_idle(void) { return 0; }
2306static inline void calc_global_nohz(void) { }
Peter Zijlstra0f004f52010-11-30 19:48:45 +01002307
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002308#endif /* CONFIG_NO_HZ */
Thomas Gleixner2d024942009-05-02 20:08:52 +02002309
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002310/*
2311 * calc_load - update the avenrun load estimates 10 ticks after the
2312 * CPUs have updated calc_load_tasks.
2313 */
Peter Zijlstra0f004f52010-11-30 19:48:45 +01002314void calc_global_load(unsigned long ticks)
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002315{
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002316 long active, delta;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002317
Peter Zijlstra0f004f52010-11-30 19:48:45 +01002318 if (time_before(jiffies, calc_load_update + 10))
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002319 return;
2320
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002321 /*
2322 * Fold the 'old' idle-delta to include all NO_HZ cpus.
2323 */
2324 delta = calc_load_fold_idle();
2325 if (delta)
2326 atomic_long_add(delta, &calc_load_tasks);
2327
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002328 active = atomic_long_read(&calc_load_tasks);
2329 active = active > 0 ? active * FIXED_1 : 0;
2330
2331 avenrun[0] = calc_load(avenrun[0], EXP_1, active);
2332 avenrun[1] = calc_load(avenrun[1], EXP_5, active);
2333 avenrun[2] = calc_load(avenrun[2], EXP_15, active);
2334
2335 calc_load_update += LOAD_FREQ;
Peter Zijlstrac308b562012-03-01 15:04:46 +01002336
2337 /*
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002338 * In case we idled for multiple LOAD_FREQ intervals, catch up in bulk.
Peter Zijlstrac308b562012-03-01 15:04:46 +01002339 */
2340 calc_global_nohz();
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002341}
2342
2343/*
Peter Zijlstra74f51872010-04-22 21:50:19 +02002344 * Called from update_cpu_load() to periodically update this CPU's
2345 * active count.
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002346 */
2347static void calc_load_account_active(struct rq *this_rq)
2348{
Peter Zijlstra74f51872010-04-22 21:50:19 +02002349 long delta;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002350
Peter Zijlstra74f51872010-04-22 21:50:19 +02002351 if (time_before(jiffies, this_rq->calc_load_update))
2352 return;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002353
Peter Zijlstra74f51872010-04-22 21:50:19 +02002354 delta = calc_load_fold_active(this_rq);
Peter Zijlstra74f51872010-04-22 21:50:19 +02002355 if (delta)
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002356 atomic_long_add(delta, &calc_load_tasks);
Peter Zijlstra74f51872010-04-22 21:50:19 +02002357
2358 this_rq->calc_load_update += LOAD_FREQ;
Jack Steinerdb1b1fe2006-03-31 02:31:21 -08002359}
2360
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361/*
Peter Zijlstra5167e8d2012-06-22 15:52:09 +02002362 * End of global load-average stuff
2363 */
2364
2365/*
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07002366 * The exact cpuload at various idx values, calculated at every tick would be
2367 * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load
2368 *
2369 * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called
2370 * on nth tick when cpu may be busy, then we have:
2371 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
2372 * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load
2373 *
2374 * decay_load_missed() below does efficient calculation of
2375 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
2376 * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load
2377 *
2378 * The calculation is approximated on a 128 point scale.
2379 * degrade_zero_ticks is the number of ticks after which load at any
2380 * particular idx is approximated to be zero.
2381 * degrade_factor is a precomputed table, a row for each load idx.
2382 * Each column corresponds to degradation factor for a power of two ticks,
2383 * based on 128 point scale.
2384 * Example:
2385 * row 2, col 3 (=12) says that the degradation at load idx 2 after
2386 * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8).
2387 *
2388 * With this power of 2 load factors, we can degrade the load n times
2389 * by looking at 1 bits in n and doing as many mult/shift instead of
2390 * n mult/shifts needed by the exact degradation.
2391 */
2392#define DEGRADE_SHIFT 7
2393static const unsigned char
2394 degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128};
2395static const unsigned char
2396 degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = {
2397 {0, 0, 0, 0, 0, 0, 0, 0},
2398 {64, 32, 8, 0, 0, 0, 0, 0},
2399 {96, 72, 40, 12, 1, 0, 0},
2400 {112, 98, 75, 43, 15, 1, 0},
2401 {120, 112, 98, 76, 45, 16, 2} };
2402
2403/*
2404 * Update cpu_load for any missed ticks, due to tickless idle. The backlog
2405 * would be when CPU is idle and so we just decay the old load without
2406 * adding any new load.
2407 */
2408static unsigned long
2409decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
2410{
2411 int j = 0;
2412
2413 if (!missed_updates)
2414 return load;
2415
2416 if (missed_updates >= degrade_zero_ticks[idx])
2417 return 0;
2418
2419 if (idx == 1)
2420 return load >> missed_updates;
2421
2422 while (missed_updates) {
2423 if (missed_updates % 2)
2424 load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT;
2425
2426 missed_updates >>= 1;
2427 j++;
2428 }
2429 return load;
2430}
2431
2432/*
Ingo Molnardd41f592007-07-09 18:51:59 +02002433 * Update rq->cpu_load[] statistics. This function is usually called every
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07002434 * scheduler tick (TICK_NSEC). With tickless idle this will not be called
2435 * every tick. We fix it up based on jiffies.
Ingo Molnar48f24c42006-07-03 00:25:40 -07002436 */
Peter Zijlstra556061b2012-05-11 17:31:26 +02002437static void __update_cpu_load(struct rq *this_rq, unsigned long this_load,
2438 unsigned long pending_updates)
Ingo Molnar48f24c42006-07-03 00:25:40 -07002439{
Ingo Molnardd41f592007-07-09 18:51:59 +02002440 int i, scale;
2441
2442 this_rq->nr_load_updates++;
Ingo Molnardd41f592007-07-09 18:51:59 +02002443
2444 /* Update our load: */
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07002445 this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */
2446 for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
Ingo Molnardd41f592007-07-09 18:51:59 +02002447 unsigned long old_load, new_load;
2448
2449 /* scale is effectively 1 << i now, and >> i divides by scale */
2450
2451 old_load = this_rq->cpu_load[i];
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07002452 old_load = decay_load_missed(old_load, pending_updates - 1, i);
Ingo Molnardd41f592007-07-09 18:51:59 +02002453 new_load = this_load;
Ingo Molnara25707f2007-10-15 17:00:03 +02002454 /*
2455 * Round up the averaging division if load is increasing. This
2456 * prevents us from getting stuck on 9 if the load is 10, for
2457 * example.
2458 */
2459 if (new_load > old_load)
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07002460 new_load += scale - 1;
2461
2462 this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i;
Ingo Molnardd41f592007-07-09 18:51:59 +02002463 }
Suresh Siddhada2b71e2010-08-23 13:42:51 -07002464
2465 sched_avg_update(this_rq);
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07002466}
2467
Peter Zijlstra5aaa0b72012-05-17 17:15:29 +02002468#ifdef CONFIG_NO_HZ
2469/*
2470 * There is no sane way to deal with nohz on smp when using jiffies because the
2471 * cpu doing the jiffies update might drift wrt the cpu doing the jiffy reading
2472 * causing off-by-one errors in observed deltas; {0,2} instead of {1,1}.
2473 *
2474 * Therefore we cannot use the delta approach from the regular tick since that
2475 * would seriously skew the load calculation. However we'll make do for those
2476 * updates happening while idle (nohz_idle_balance) or coming out of idle
2477 * (tick_nohz_idle_exit).
2478 *
2479 * This means we might still be one tick off for nohz periods.
2480 */
2481
Peter Zijlstra556061b2012-05-11 17:31:26 +02002482/*
2483 * Called from nohz_idle_balance() to update the load ratings before doing the
2484 * idle balance.
2485 */
2486void update_idle_cpu_load(struct rq *this_rq)
2487{
Peter Zijlstra5aaa0b72012-05-17 17:15:29 +02002488 unsigned long curr_jiffies = ACCESS_ONCE(jiffies);
Peter Zijlstra556061b2012-05-11 17:31:26 +02002489 unsigned long load = this_rq->load.weight;
2490 unsigned long pending_updates;
2491
2492 /*
Peter Zijlstra5aaa0b72012-05-17 17:15:29 +02002493 * bail if there's load or we're actually up-to-date.
Peter Zijlstra556061b2012-05-11 17:31:26 +02002494 */
2495 if (load || curr_jiffies == this_rq->last_load_update_tick)
2496 return;
2497
2498 pending_updates = curr_jiffies - this_rq->last_load_update_tick;
2499 this_rq->last_load_update_tick = curr_jiffies;
2500
2501 __update_cpu_load(this_rq, load, pending_updates);
2502}
2503
2504/*
Peter Zijlstra5aaa0b72012-05-17 17:15:29 +02002505 * Called from tick_nohz_idle_exit() -- try and fix up the ticks we missed.
2506 */
2507void update_cpu_load_nohz(void)
2508{
2509 struct rq *this_rq = this_rq();
2510 unsigned long curr_jiffies = ACCESS_ONCE(jiffies);
2511 unsigned long pending_updates;
2512
2513 if (curr_jiffies == this_rq->last_load_update_tick)
2514 return;
2515
2516 raw_spin_lock(&this_rq->lock);
2517 pending_updates = curr_jiffies - this_rq->last_load_update_tick;
2518 if (pending_updates) {
2519 this_rq->last_load_update_tick = curr_jiffies;
2520 /*
2521 * We were idle, this means load 0, the current load might be
2522 * !0 due to remote wakeups and the sort.
2523 */
2524 __update_cpu_load(this_rq, 0, pending_updates);
2525 }
2526 raw_spin_unlock(&this_rq->lock);
2527}
2528#endif /* CONFIG_NO_HZ */
2529
2530/*
Peter Zijlstra556061b2012-05-11 17:31:26 +02002531 * Called from scheduler_tick()
2532 */
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07002533static void update_cpu_load_active(struct rq *this_rq)
2534{
Peter Zijlstra556061b2012-05-11 17:31:26 +02002535 /*
Peter Zijlstra5aaa0b72012-05-17 17:15:29 +02002536 * See the mess around update_idle_cpu_load() / update_cpu_load_nohz().
Peter Zijlstra556061b2012-05-11 17:31:26 +02002537 */
2538 this_rq->last_load_update_tick = jiffies;
2539 __update_cpu_load(this_rq, this_rq->load.weight, 1);
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02002540
Peter Zijlstra74f51872010-04-22 21:50:19 +02002541 calc_load_account_active(this_rq);
Ingo Molnar48f24c42006-07-03 00:25:40 -07002542}
2543
Ingo Molnardd41f592007-07-09 18:51:59 +02002544#ifdef CONFIG_SMP
2545
Ingo Molnar48f24c42006-07-03 00:25:40 -07002546/*
Peter Zijlstra38022902009-12-16 18:04:37 +01002547 * sched_exec - execve() is a valuable balancing opportunity, because at
2548 * this point the task has the smallest effective memory and cache footprint.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 */
Peter Zijlstra38022902009-12-16 18:04:37 +01002550void sched_exec(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551{
Peter Zijlstra38022902009-12-16 18:04:37 +01002552 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 unsigned long flags;
Peter Zijlstra0017d732010-03-24 18:34:10 +01002554 int dest_cpu;
Peter Zijlstra38022902009-12-16 18:04:37 +01002555
Peter Zijlstra8f42ced2011-04-05 17:23:53 +02002556 raw_spin_lock_irqsave(&p->pi_lock, flags);
Peter Zijlstra7608dec2011-04-05 17:23:46 +02002557 dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0);
Peter Zijlstra0017d732010-03-24 18:34:10 +01002558 if (dest_cpu == smp_processor_id())
2559 goto unlock;
Peter Zijlstra38022902009-12-16 18:04:37 +01002560
Peter Zijlstra8f42ced2011-04-05 17:23:53 +02002561 if (likely(cpu_active(dest_cpu))) {
Tejun Heo969c7922010-05-06 18:49:21 +02002562 struct migration_arg arg = { p, dest_cpu };
Ingo Molnar36c8b582006-07-03 00:25:41 -07002563
Peter Zijlstra8f42ced2011-04-05 17:23:53 +02002564 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2565 stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 return;
2567 }
Peter Zijlstra0017d732010-03-24 18:34:10 +01002568unlock:
Peter Zijlstra8f42ced2011-04-05 17:23:53 +02002569 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570}
2571
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572#endif
2573
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574DEFINE_PER_CPU(struct kernel_stat, kstat);
Glauber Costa3292beb2011-11-28 14:45:17 -02002575DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576
2577EXPORT_PER_CPU_SYMBOL(kstat);
Glauber Costa3292beb2011-11-28 14:45:17 -02002578EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579
2580/*
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09002581 * Return any ns on the sched_clock that have not yet been accounted in
Frank Mayharf06febc2008-09-12 09:54:39 -07002582 * @p in case that task is currently running.
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09002583 *
2584 * Called with task_rq_lock() held on @rq.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 */
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09002586static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq)
2587{
2588 u64 ns = 0;
2589
2590 if (task_current(rq, p)) {
2591 update_rq_clock(rq);
Venkatesh Pallipadi305e6832010-10-04 17:03:21 -07002592 ns = rq->clock_task - p->se.exec_start;
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09002593 if ((s64)ns < 0)
2594 ns = 0;
2595 }
2596
2597 return ns;
2598}
2599
Frank Mayharbb34d922008-09-12 09:54:39 -07002600unsigned long long task_delta_exec(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 unsigned long flags;
Ingo Molnar41b86e92007-07-09 18:51:58 +02002603 struct rq *rq;
Frank Mayharbb34d922008-09-12 09:54:39 -07002604 u64 ns = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07002605
Ingo Molnar41b86e92007-07-09 18:51:58 +02002606 rq = task_rq_lock(p, &flags);
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09002607 ns = do_task_delta_exec(p, rq);
Peter Zijlstra0122ec52011-04-05 17:23:51 +02002608 task_rq_unlock(rq, p, &flags);
Ingo Molnar15084872008-09-30 08:28:17 +02002609
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09002610 return ns;
2611}
Frank Mayharf06febc2008-09-12 09:54:39 -07002612
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09002613/*
2614 * Return accounted runtime for the task.
2615 * In case the task is currently running, return the runtime plus current's
2616 * pending runtime that have not been accounted yet.
2617 */
2618unsigned long long task_sched_runtime(struct task_struct *p)
2619{
2620 unsigned long flags;
2621 struct rq *rq;
2622 u64 ns = 0;
Ingo Molnar48f24c42006-07-03 00:25:40 -07002623
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09002624 rq = task_rq_lock(p, &flags);
2625 ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq);
Peter Zijlstra0122ec52011-04-05 17:23:51 +02002626 task_rq_unlock(rq, p, &flags);
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +09002627
2628 return ns;
2629}
2630
Balbir Singh49048622008-09-05 18:12:23 +02002631/*
Christoph Lameter7835b982006-12-10 02:20:22 -08002632 * This function gets called by the timer code, with HZ frequency.
2633 * We call it with interrupts disabled.
Christoph Lameter7835b982006-12-10 02:20:22 -08002634 */
2635void scheduler_tick(void)
2636{
Christoph Lameter7835b982006-12-10 02:20:22 -08002637 int cpu = smp_processor_id();
2638 struct rq *rq = cpu_rq(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02002639 struct task_struct *curr = rq->curr;
Peter Zijlstra3e51f332008-05-03 18:29:28 +02002640
2641 sched_clock_tick();
Christoph Lameter7835b982006-12-10 02:20:22 -08002642
Thomas Gleixner05fa7852009-11-17 14:28:38 +01002643 raw_spin_lock(&rq->lock);
Peter Zijlstra3e51f332008-05-03 18:29:28 +02002644 update_rq_clock(rq);
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07002645 update_cpu_load_active(rq);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01002646 curr->sched_class->task_tick(rq, curr, 0);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01002647 raw_spin_unlock(&rq->lock);
Ingo Molnardd41f592007-07-09 18:51:59 +02002648
Peter Zijlstrae9d2b062010-09-17 11:28:50 +02002649 perf_event_task_tick();
Peter Zijlstrae220d2d2009-05-23 18:28:55 +02002650
Christoph Lametere418e1c2006-12-10 02:20:23 -08002651#ifdef CONFIG_SMP
Suresh Siddha6eb57e02011-10-03 15:09:01 -07002652 rq->idle_balance = idle_cpu(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02002653 trigger_load_balance(rq, cpu);
Christoph Lametere418e1c2006-12-10 02:20:23 -08002654#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655}
2656
Lai Jiangshan132380a2009-04-02 14:18:25 +08002657notrace unsigned long get_parent_ip(unsigned long addr)
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02002658{
2659 if (in_lock_functions(addr)) {
2660 addr = CALLER_ADDR2;
2661 if (in_lock_functions(addr))
2662 addr = CALLER_ADDR3;
2663 }
2664 return addr;
2665}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666
Steven Rostedt7e49fcc2009-01-22 19:01:40 -05002667#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
2668 defined(CONFIG_PREEMPT_TRACER))
2669
Srinivasa Ds43627582008-02-23 15:24:04 -08002670void __kprobes add_preempt_count(int val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671{
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02002672#ifdef CONFIG_DEBUG_PREEMPT
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 /*
2674 * Underflow?
2675 */
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07002676 if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
2677 return;
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02002678#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 preempt_count() += val;
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02002680#ifdef CONFIG_DEBUG_PREEMPT
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 /*
2682 * Spinlock count overflowing soon?
2683 */
Miguel Ojeda Sandonis33859f72006-12-10 02:20:38 -08002684 DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
2685 PREEMPT_MASK - 10);
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02002686#endif
2687 if (preempt_count() == val)
2688 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689}
2690EXPORT_SYMBOL(add_preempt_count);
2691
Srinivasa Ds43627582008-02-23 15:24:04 -08002692void __kprobes sub_preempt_count(int val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693{
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02002694#ifdef CONFIG_DEBUG_PREEMPT
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 /*
2696 * Underflow?
2697 */
Ingo Molnar01e3eb82009-01-12 13:00:50 +01002698 if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07002699 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 /*
2701 * Is the spinlock portion underflowing?
2702 */
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07002703 if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
2704 !(preempt_count() & PREEMPT_MASK)))
2705 return;
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02002706#endif
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07002707
Steven Rostedt6cd8a4b2008-05-12 21:20:42 +02002708 if (preempt_count() == val)
2709 trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 preempt_count() -= val;
2711}
2712EXPORT_SYMBOL(sub_preempt_count);
2713
2714#endif
2715
2716/*
Ingo Molnardd41f592007-07-09 18:51:59 +02002717 * Print scheduling while atomic bug:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 */
Ingo Molnardd41f592007-07-09 18:51:59 +02002719static noinline void __schedule_bug(struct task_struct *prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720{
Dave Jones664dfa62011-12-22 16:39:30 -05002721 if (oops_in_progress)
2722 return;
2723
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01002724 printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
2725 prev->comm, prev->pid, preempt_count());
Satyam Sharma838225b2007-10-24 18:23:50 +02002726
Ingo Molnardd41f592007-07-09 18:51:59 +02002727 debug_show_held_locks(prev);
Arjan van de Vene21f5b12008-05-23 09:05:58 -07002728 print_modules();
Ingo Molnardd41f592007-07-09 18:51:59 +02002729 if (irqs_disabled())
2730 print_irqtrace_events(prev);
Stephen Boyd6135fc12012-03-28 17:10:47 -07002731 dump_stack();
Konstantin Khlebnikov1c2927f2012-05-10 16:20:04 +04002732 add_taint(TAINT_WARN);
Ingo Molnardd41f592007-07-09 18:51:59 +02002733}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734
Ingo Molnardd41f592007-07-09 18:51:59 +02002735/*
2736 * Various schedule()-time debugging checks and statistics:
2737 */
2738static inline void schedule_debug(struct task_struct *prev)
2739{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 /*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01002741 * Test if we are atomic. Since do_exit() needs to call into
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 * schedule() atomically, we ignore that path for now.
2743 * Otherwise, whine if we are scheduling when we should not be.
2744 */
Roel Kluin3f33a7c2008-05-13 23:44:11 +02002745 if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
Ingo Molnardd41f592007-07-09 18:51:59 +02002746 __schedule_bug(prev);
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -07002747 rcu_sleep_check();
Ingo Molnardd41f592007-07-09 18:51:59 +02002748
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 profile_hit(SCHED_PROFILING, __builtin_return_address(0));
2750
Ingo Molnar2d723762007-10-15 17:00:12 +02002751 schedstat_inc(this_rq(), sched_count);
Ingo Molnardd41f592007-07-09 18:51:59 +02002752}
2753
Peter Zijlstra6cecd082009-11-30 13:00:37 +01002754static void put_prev_task(struct rq *rq, struct task_struct *prev)
Mike Galbraithdf1c99d2009-03-10 19:08:11 +01002755{
Mike Galbraith61eadef2011-04-29 08:36:50 +02002756 if (prev->on_rq || rq->skip_clock_update < 0)
Mike Galbraitha64692a2010-03-11 17:16:20 +01002757 update_rq_clock(rq);
Peter Zijlstra6cecd082009-11-30 13:00:37 +01002758 prev->sched_class->put_prev_task(rq, prev);
Mike Galbraithdf1c99d2009-03-10 19:08:11 +01002759}
2760
Ingo Molnardd41f592007-07-09 18:51:59 +02002761/*
2762 * Pick up the highest-prio task:
2763 */
2764static inline struct task_struct *
Wang Chenb67802e2009-03-02 13:55:26 +08002765pick_next_task(struct rq *rq)
Ingo Molnardd41f592007-07-09 18:51:59 +02002766{
Ingo Molnar5522d5d2007-10-15 17:00:12 +02002767 const struct sched_class *class;
Ingo Molnardd41f592007-07-09 18:51:59 +02002768 struct task_struct *p;
2769
2770 /*
2771 * Optimization: we know that if all tasks are in
2772 * the fair class we can call that function directly:
2773 */
Paul Turner953bfcd2011-07-21 09:43:27 -07002774 if (likely(rq->nr_running == rq->cfs.h_nr_running)) {
Ingo Molnarfb8d4722007-08-09 11:16:48 +02002775 p = fair_sched_class.pick_next_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02002776 if (likely(p))
2777 return p;
2778 }
2779
Peter Zijlstra34f971f2010-09-22 13:53:15 +02002780 for_each_class(class) {
Ingo Molnarfb8d4722007-08-09 11:16:48 +02002781 p = class->pick_next_task(rq);
Ingo Molnardd41f592007-07-09 18:51:59 +02002782 if (p)
2783 return p;
Ingo Molnardd41f592007-07-09 18:51:59 +02002784 }
Peter Zijlstra34f971f2010-09-22 13:53:15 +02002785
2786 BUG(); /* the idle class will always have a runnable task */
Ingo Molnardd41f592007-07-09 18:51:59 +02002787}
2788
2789/*
Thomas Gleixnerc259e012011-06-22 19:47:00 +02002790 * __schedule() is the main scheduler function.
Pekka Enbergedde96e2012-08-04 11:49:47 +03002791 *
2792 * The main means of driving the scheduler and thus entering this function are:
2793 *
2794 * 1. Explicit blocking: mutex, semaphore, waitqueue, etc.
2795 *
2796 * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return
2797 * paths. For example, see arch/x86/entry_64.S.
2798 *
2799 * To drive preemption between tasks, the scheduler sets the flag in timer
2800 * interrupt handler scheduler_tick().
2801 *
2802 * 3. Wakeups don't really cause entry into schedule(). They add a
2803 * task to the run-queue and that's it.
2804 *
2805 * Now, if the new task added to the run-queue preempts the current
2806 * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets
2807 * called on the nearest possible occasion:
2808 *
2809 * - If the kernel is preemptible (CONFIG_PREEMPT=y):
2810 *
2811 * - in syscall or exception context, at the next outmost
2812 * preempt_enable(). (this might be as soon as the wake_up()'s
2813 * spin_unlock()!)
2814 *
2815 * - in IRQ context, return from interrupt-handler to
2816 * preemptible context
2817 *
2818 * - If the kernel is not preemptible (CONFIG_PREEMPT is not set)
2819 * then at the next:
2820 *
2821 * - cond_resched() call
2822 * - explicit schedule() call
2823 * - return from syscall or exception to user-space
2824 * - return from interrupt-handler to user-space
Ingo Molnardd41f592007-07-09 18:51:59 +02002825 */
Thomas Gleixnerc259e012011-06-22 19:47:00 +02002826static void __sched __schedule(void)
Ingo Molnardd41f592007-07-09 18:51:59 +02002827{
2828 struct task_struct *prev, *next;
Harvey Harrison67ca7bd2008-02-15 09:56:36 -08002829 unsigned long *switch_count;
Ingo Molnardd41f592007-07-09 18:51:59 +02002830 struct rq *rq;
Peter Zijlstra31656512008-07-18 18:01:23 +02002831 int cpu;
Ingo Molnardd41f592007-07-09 18:51:59 +02002832
Peter Zijlstraff743342009-03-13 12:21:26 +01002833need_resched:
2834 preempt_disable();
Ingo Molnardd41f592007-07-09 18:51:59 +02002835 cpu = smp_processor_id();
2836 rq = cpu_rq(cpu);
Paul E. McKenney25502a62010-04-01 17:37:01 -07002837 rcu_note_context_switch(cpu);
Ingo Molnardd41f592007-07-09 18:51:59 +02002838 prev = rq->curr;
Ingo Molnardd41f592007-07-09 18:51:59 +02002839
Ingo Molnardd41f592007-07-09 18:51:59 +02002840 schedule_debug(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841
Peter Zijlstra31656512008-07-18 18:01:23 +02002842 if (sched_feat(HRTICK))
Mike Galbraithf333fdc2008-05-12 21:20:55 +02002843 hrtick_clear(rq);
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01002844
Thomas Gleixner05fa7852009-11-17 14:28:38 +01002845 raw_spin_lock_irq(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846
Oleg Nesterov246d86b2010-05-19 14:57:11 +02002847 switch_count = &prev->nivcsw;
Ingo Molnardd41f592007-07-09 18:51:59 +02002848 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
Tejun Heo21aa9af2010-06-08 21:40:37 +02002849 if (unlikely(signal_pending_state(prev->state, prev))) {
Ingo Molnardd41f592007-07-09 18:51:59 +02002850 prev->state = TASK_RUNNING;
Tejun Heo21aa9af2010-06-08 21:40:37 +02002851 } else {
Peter Zijlstra2acca552011-04-05 17:23:50 +02002852 deactivate_task(rq, prev, DEQUEUE_SLEEP);
2853 prev->on_rq = 0;
2854
Tejun Heo21aa9af2010-06-08 21:40:37 +02002855 /*
Peter Zijlstra2acca552011-04-05 17:23:50 +02002856 * If a worker went to sleep, notify and ask workqueue
2857 * whether it wants to wake up a task to maintain
2858 * concurrency.
Tejun Heo21aa9af2010-06-08 21:40:37 +02002859 */
2860 if (prev->flags & PF_WQ_WORKER) {
2861 struct task_struct *to_wakeup;
2862
2863 to_wakeup = wq_worker_sleeping(prev, cpu);
2864 if (to_wakeup)
2865 try_to_wake_up_local(to_wakeup);
2866 }
Tejun Heo21aa9af2010-06-08 21:40:37 +02002867 }
Ingo Molnardd41f592007-07-09 18:51:59 +02002868 switch_count = &prev->nvcsw;
2869 }
2870
Gregory Haskins3f029d32009-07-29 11:08:47 -04002871 pre_schedule(rq, prev);
Steven Rostedtf65eda42008-01-25 21:08:07 +01002872
Ingo Molnardd41f592007-07-09 18:51:59 +02002873 if (unlikely(!rq->nr_running))
2874 idle_balance(cpu, rq);
2875
Mike Galbraithdf1c99d2009-03-10 19:08:11 +01002876 put_prev_task(rq, prev);
Wang Chenb67802e2009-03-02 13:55:26 +08002877 next = pick_next_task(rq);
Mike Galbraithf26f9af2010-12-08 11:05:42 +01002878 clear_tsk_need_resched(prev);
2879 rq->skip_clock_update = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 if (likely(prev != next)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 rq->nr_switches++;
2883 rq->curr = next;
2884 ++*switch_count;
2885
Ingo Molnardd41f592007-07-09 18:51:59 +02002886 context_switch(rq, prev, next); /* unlocks the rq */
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01002887 /*
Oleg Nesterov246d86b2010-05-19 14:57:11 +02002888 * The context switch have flipped the stack from under us
2889 * and restored the local variables which were saved when
2890 * this task called schedule() in the past. prev == current
2891 * is still correct, but it can be moved to another cpu/rq.
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01002892 */
2893 cpu = smp_processor_id();
2894 rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895 } else
Thomas Gleixner05fa7852009-11-17 14:28:38 +01002896 raw_spin_unlock_irq(&rq->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897
Gregory Haskins3f029d32009-07-29 11:08:47 -04002898 post_schedule(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899
Thomas Gleixnerba74c142011-03-21 13:32:17 +01002900 sched_preempt_enable_no_resched();
Peter Zijlstraff743342009-03-13 12:21:26 +01002901 if (need_resched())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 goto need_resched;
2903}
Thomas Gleixnerc259e012011-06-22 19:47:00 +02002904
Thomas Gleixner9c40cef2011-06-22 19:47:01 +02002905static inline void sched_submit_work(struct task_struct *tsk)
2906{
Thomas Gleixner3c7d5182011-07-17 20:46:52 +02002907 if (!tsk->state || tsk_is_pi_blocked(tsk))
Thomas Gleixner9c40cef2011-06-22 19:47:01 +02002908 return;
2909 /*
2910 * If we are going to sleep and we have plugged IO queued,
2911 * make sure to submit it to avoid deadlocks.
2912 */
2913 if (blk_needs_flush_plug(tsk))
2914 blk_schedule_flush_plug(tsk);
2915}
2916
Simon Kirby6ebbe7a2011-09-22 17:03:46 -07002917asmlinkage void __sched schedule(void)
Thomas Gleixnerc259e012011-06-22 19:47:00 +02002918{
Thomas Gleixner9c40cef2011-06-22 19:47:01 +02002919 struct task_struct *tsk = current;
2920
2921 sched_submit_work(tsk);
Thomas Gleixnerc259e012011-06-22 19:47:00 +02002922 __schedule();
2923}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924EXPORT_SYMBOL(schedule);
2925
Frederic Weisbecker91d1aa432012-11-27 19:33:25 +01002926#ifdef CONFIG_CONTEXT_TRACKING
Frederic Weisbecker20ab65e32012-07-11 20:26:37 +02002927asmlinkage void __sched schedule_user(void)
2928{
2929 /*
2930 * If we come here after a random call to set_need_resched(),
2931 * or we have been woken up remotely but the IPI has not yet arrived,
2932 * we haven't yet exited the RCU idle mode. Do it here manually until
2933 * we find a better solution.
2934 */
Frederic Weisbecker91d1aa432012-11-27 19:33:25 +01002935 user_exit();
Frederic Weisbecker20ab65e32012-07-11 20:26:37 +02002936 schedule();
Frederic Weisbecker91d1aa432012-11-27 19:33:25 +01002937 user_enter();
Frederic Weisbecker20ab65e32012-07-11 20:26:37 +02002938}
2939#endif
2940
Thomas Gleixnerc5491ea2011-03-21 12:09:35 +01002941/**
2942 * schedule_preempt_disabled - called with preemption disabled
2943 *
2944 * Returns with preemption disabled. Note: preempt_count must be 1
2945 */
2946void __sched schedule_preempt_disabled(void)
2947{
Thomas Gleixnerba74c142011-03-21 13:32:17 +01002948 sched_preempt_enable_no_resched();
Thomas Gleixnerc5491ea2011-03-21 12:09:35 +01002949 schedule();
2950 preempt_disable();
2951}
2952
Frederic Weisbeckerc08f7822009-12-02 20:49:17 +01002953#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
Peter Zijlstrac6eb3dd2011-04-05 17:23:41 +02002954
2955static inline bool owner_running(struct mutex *lock, struct task_struct *owner)
2956{
Peter Zijlstrac6eb3dd2011-04-05 17:23:41 +02002957 if (lock->owner != owner)
Thomas Gleixner307bf982011-06-10 15:08:55 +02002958 return false;
Peter Zijlstrac6eb3dd2011-04-05 17:23:41 +02002959
2960 /*
2961 * Ensure we emit the owner->on_cpu, dereference _after_ checking
2962 * lock->owner still matches owner, if that fails, owner might
2963 * point to free()d memory, if it still matches, the rcu_read_lock()
2964 * ensures the memory stays valid.
2965 */
2966 barrier();
2967
Thomas Gleixner307bf982011-06-10 15:08:55 +02002968 return owner->on_cpu;
Peter Zijlstrac6eb3dd2011-04-05 17:23:41 +02002969}
2970
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01002971/*
2972 * Look out! "owner" is an entirely speculative pointer
2973 * access and not reliable.
2974 */
Peter Zijlstrac6eb3dd2011-04-05 17:23:41 +02002975int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner)
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01002976{
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01002977 if (!sched_feat(OWNER_SPIN))
2978 return 0;
2979
Thomas Gleixner307bf982011-06-10 15:08:55 +02002980 rcu_read_lock();
Peter Zijlstrac6eb3dd2011-04-05 17:23:41 +02002981 while (owner_running(lock, owner)) {
2982 if (need_resched())
Thomas Gleixner307bf982011-06-10 15:08:55 +02002983 break;
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01002984
Gerald Schaefer335d7af2010-11-22 15:47:36 +01002985 arch_mutex_cpu_relax();
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01002986 }
Thomas Gleixner307bf982011-06-10 15:08:55 +02002987 rcu_read_unlock();
Benjamin Herrenschmidt4b402212010-04-16 23:20:00 +02002988
Peter Zijlstrac6eb3dd2011-04-05 17:23:41 +02002989 /*
Thomas Gleixner307bf982011-06-10 15:08:55 +02002990 * We break out the loop above on need_resched() and when the
2991 * owner changed, which is a sign for heavy contention. Return
2992 * success only when lock->owner is NULL.
Peter Zijlstrac6eb3dd2011-04-05 17:23:41 +02002993 */
Thomas Gleixner307bf982011-06-10 15:08:55 +02002994 return lock->owner == NULL;
Peter Zijlstra0d66bf62009-01-12 14:01:47 +01002995}
2996#endif
2997
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998#ifdef CONFIG_PREEMPT
2999/*
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003000 * this is the entry point to schedule() from in-kernel preemption
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003001 * off of preempt_enable. Kernel preemptions off return from interrupt
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 * occur there and call schedule directly.
3003 */
Steven Rostedtd1f74e22010-06-02 21:52:29 -04003004asmlinkage void __sched notrace preempt_schedule(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005{
3006 struct thread_info *ti = current_thread_info();
Ingo Molnar6478d882008-01-25 21:08:33 +01003007
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 /*
3009 * If there is a non-zero preempt_count or interrupts are disabled,
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003010 * we do not want to preempt the current task. Just return..
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 */
Nick Pigginbeed33a2006-10-11 01:21:52 -07003012 if (likely(ti->preempt_count || irqs_disabled()))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013 return;
3014
Andi Kleen3a5c3592007-10-15 17:00:14 +02003015 do {
Steven Rostedtd1f74e22010-06-02 21:52:29 -04003016 add_preempt_count_notrace(PREEMPT_ACTIVE);
Thomas Gleixnerc259e012011-06-22 19:47:00 +02003017 __schedule();
Steven Rostedtd1f74e22010-06-02 21:52:29 -04003018 sub_preempt_count_notrace(PREEMPT_ACTIVE);
Andi Kleen3a5c3592007-10-15 17:00:14 +02003019
3020 /*
3021 * Check again in case we missed a preemption opportunity
3022 * between schedule and now.
3023 */
3024 barrier();
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08003025 } while (need_resched());
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027EXPORT_SYMBOL(preempt_schedule);
3028
3029/*
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003030 * this is the entry point to schedule() from kernel preemption
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 * off of irq context.
3032 * Note, that this is called and return with irqs disabled. This will
3033 * protect us against recursive calling from irq.
3034 */
3035asmlinkage void __sched preempt_schedule_irq(void)
3036{
3037 struct thread_info *ti = current_thread_info();
Ingo Molnar6478d882008-01-25 21:08:33 +01003038
Andreas Mohr2ed6e342006-07-10 04:43:52 -07003039 /* Catch callers which need to be fixed */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 BUG_ON(ti->preempt_count || !irqs_disabled());
3041
Frederic Weisbecker91d1aa432012-11-27 19:33:25 +01003042 user_exit();
Andi Kleen3a5c3592007-10-15 17:00:14 +02003043 do {
3044 add_preempt_count(PREEMPT_ACTIVE);
Andi Kleen3a5c3592007-10-15 17:00:14 +02003045 local_irq_enable();
Thomas Gleixnerc259e012011-06-22 19:47:00 +02003046 __schedule();
Andi Kleen3a5c3592007-10-15 17:00:14 +02003047 local_irq_disable();
Andi Kleen3a5c3592007-10-15 17:00:14 +02003048 sub_preempt_count(PREEMPT_ACTIVE);
3049
3050 /*
3051 * Check again in case we missed a preemption opportunity
3052 * between schedule and now.
3053 */
3054 barrier();
Lai Jiangshan5ed0cec2009-03-06 19:40:20 +08003055 } while (need_resched());
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056}
3057
3058#endif /* CONFIG_PREEMPT */
3059
Peter Zijlstra63859d42009-09-15 19:14:42 +02003060int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07003061 void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062{
Peter Zijlstra63859d42009-09-15 19:14:42 +02003063 return try_to_wake_up(curr->private, mode, wake_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065EXPORT_SYMBOL(default_wake_function);
3066
3067/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003068 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
3069 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070 * number) then we wake all the non-exclusive tasks and one exclusive task.
3071 *
3072 * There are circumstances in which we can try to wake a task which has already
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01003073 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074 * zero in this (rare) case, and we handle it by continuing to scan the queue.
3075 */
Johannes Weiner78ddb082009-04-14 16:53:05 +02003076static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
Peter Zijlstra63859d42009-09-15 19:14:42 +02003077 int nr_exclusive, int wake_flags, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078{
Matthias Kaehlcke2e458742007-10-15 17:00:02 +02003079 wait_queue_t *curr, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080
Matthias Kaehlcke2e458742007-10-15 17:00:02 +02003081 list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
Ingo Molnar48f24c42006-07-03 00:25:40 -07003082 unsigned flags = curr->flags;
3083
Peter Zijlstra63859d42009-09-15 19:14:42 +02003084 if (curr->func(curr, mode, wake_flags, key) &&
Ingo Molnar48f24c42006-07-03 00:25:40 -07003085 (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 break;
3087 }
3088}
3089
3090/**
3091 * __wake_up - wake up threads blocked on a waitqueue.
3092 * @q: the waitqueue
3093 * @mode: which threads
3094 * @nr_exclusive: how many wake-one or wake-many threads to wake up
Martin Waitz67be2dd2005-05-01 08:59:26 -07003095 * @key: is directly passed to the wakeup function
David Howells50fa6102009-04-28 15:01:38 +01003096 *
3097 * It may be assumed that this function implies a write memory barrier before
3098 * changing the task state if and only if any tasks are woken up.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08003100void __wake_up(wait_queue_head_t *q, unsigned int mode,
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07003101 int nr_exclusive, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102{
3103 unsigned long flags;
3104
3105 spin_lock_irqsave(&q->lock, flags);
3106 __wake_up_common(q, mode, nr_exclusive, 0, key);
3107 spin_unlock_irqrestore(&q->lock, flags);
3108}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109EXPORT_SYMBOL(__wake_up);
3110
3111/*
3112 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
3113 */
Thomas Gleixner63b20012011-12-01 00:04:00 +01003114void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115{
Thomas Gleixner63b20012011-12-01 00:04:00 +01003116 __wake_up_common(q, mode, nr, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117}
Michal Nazarewicz22c43c82010-05-05 12:53:11 +02003118EXPORT_SYMBOL_GPL(__wake_up_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119
Davide Libenzi4ede8162009-03-31 15:24:20 -07003120void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
3121{
3122 __wake_up_common(q, mode, 1, 0, key);
3123}
Trond Myklebustbf294b42011-02-21 11:05:41 -08003124EXPORT_SYMBOL_GPL(__wake_up_locked_key);
Davide Libenzi4ede8162009-03-31 15:24:20 -07003125
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126/**
Davide Libenzi4ede8162009-03-31 15:24:20 -07003127 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128 * @q: the waitqueue
3129 * @mode: which threads
3130 * @nr_exclusive: how many wake-one or wake-many threads to wake up
Davide Libenzi4ede8162009-03-31 15:24:20 -07003131 * @key: opaque value to be passed to wakeup targets
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132 *
3133 * The sync wakeup differs that the waker knows that it will schedule
3134 * away soon, so while the target thread will be woken up, it will not
3135 * be migrated to another CPU - ie. the two threads are 'synchronized'
3136 * with each other. This can prevent needless bouncing between CPUs.
3137 *
3138 * On UP it can prevent extra preemption.
David Howells50fa6102009-04-28 15:01:38 +01003139 *
3140 * It may be assumed that this function implies a write memory barrier before
3141 * changing the task state if and only if any tasks are woken up.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 */
Davide Libenzi4ede8162009-03-31 15:24:20 -07003143void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
3144 int nr_exclusive, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145{
3146 unsigned long flags;
Peter Zijlstra7d478722009-09-14 19:55:44 +02003147 int wake_flags = WF_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148
3149 if (unlikely(!q))
3150 return;
3151
3152 if (unlikely(!nr_exclusive))
Peter Zijlstra7d478722009-09-14 19:55:44 +02003153 wake_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154
3155 spin_lock_irqsave(&q->lock, flags);
Peter Zijlstra7d478722009-09-14 19:55:44 +02003156 __wake_up_common(q, mode, nr_exclusive, wake_flags, key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157 spin_unlock_irqrestore(&q->lock, flags);
3158}
Davide Libenzi4ede8162009-03-31 15:24:20 -07003159EXPORT_SYMBOL_GPL(__wake_up_sync_key);
3160
3161/*
3162 * __wake_up_sync - see __wake_up_sync_key()
3163 */
3164void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
3165{
3166 __wake_up_sync_key(q, mode, nr_exclusive, NULL);
3167}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
3169
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003170/**
3171 * complete: - signals a single thread waiting on this completion
3172 * @x: holds the state of this particular completion
3173 *
3174 * This will wake up a single thread waiting on this completion. Threads will be
3175 * awakened in the same order in which they were queued.
3176 *
3177 * See also complete_all(), wait_for_completion() and related routines.
David Howells50fa6102009-04-28 15:01:38 +01003178 *
3179 * It may be assumed that this function implies a write memory barrier before
3180 * changing the task state if and only if any tasks are woken up.
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003181 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02003182void complete(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183{
3184 unsigned long flags;
3185
3186 spin_lock_irqsave(&x->wait.lock, flags);
3187 x->done++;
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05003188 __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189 spin_unlock_irqrestore(&x->wait.lock, flags);
3190}
3191EXPORT_SYMBOL(complete);
3192
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003193/**
3194 * complete_all: - signals all threads waiting on this completion
3195 * @x: holds the state of this particular completion
3196 *
3197 * This will wake up all threads waiting on this particular completion event.
David Howells50fa6102009-04-28 15:01:38 +01003198 *
3199 * It may be assumed that this function implies a write memory barrier before
3200 * changing the task state if and only if any tasks are woken up.
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003201 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02003202void complete_all(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203{
3204 unsigned long flags;
3205
3206 spin_lock_irqsave(&x->wait.lock, flags);
3207 x->done += UINT_MAX/2;
Matthew Wilcoxd9514f62007-12-06 11:07:07 -05003208 __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003209 spin_unlock_irqrestore(&x->wait.lock, flags);
3210}
3211EXPORT_SYMBOL(complete_all);
3212
Andi Kleen8cbbe862007-10-15 17:00:14 +02003213static inline long __sched
3214do_wait_for_common(struct completion *x, long timeout, int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216 if (!x->done) {
3217 DECLARE_WAITQUEUE(wait, current);
3218
Changli Gaoa93d2f12010-05-07 14:33:26 +08003219 __add_wait_queue_tail_exclusive(&x->wait, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220 do {
Oleg Nesterov94d3d822008-08-20 16:54:41 -07003221 if (signal_pending_state(state, current)) {
Oleg Nesterovea71a542008-06-20 18:32:20 +04003222 timeout = -ERESTARTSYS;
3223 break;
Andi Kleen8cbbe862007-10-15 17:00:14 +02003224 }
3225 __set_current_state(state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226 spin_unlock_irq(&x->wait.lock);
Andi Kleen8cbbe862007-10-15 17:00:14 +02003227 timeout = schedule_timeout(timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003228 spin_lock_irq(&x->wait.lock);
Oleg Nesterovea71a542008-06-20 18:32:20 +04003229 } while (!x->done && timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230 __remove_wait_queue(&x->wait, &wait);
Oleg Nesterovea71a542008-06-20 18:32:20 +04003231 if (!x->done)
3232 return timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233 }
3234 x->done--;
Oleg Nesterovea71a542008-06-20 18:32:20 +04003235 return timeout ?: 1;
Andi Kleen8cbbe862007-10-15 17:00:14 +02003236}
3237
3238static long __sched
3239wait_for_common(struct completion *x, long timeout, int state)
3240{
3241 might_sleep();
3242
3243 spin_lock_irq(&x->wait.lock);
3244 timeout = do_wait_for_common(x, timeout, state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003245 spin_unlock_irq(&x->wait.lock);
Andi Kleen8cbbe862007-10-15 17:00:14 +02003246 return timeout;
3247}
3248
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003249/**
3250 * wait_for_completion: - waits for completion of a task
3251 * @x: holds the state of this particular completion
3252 *
3253 * This waits to be signaled for completion of a specific task. It is NOT
3254 * interruptible and there is no timeout.
3255 *
3256 * See also similar routines (i.e. wait_for_completion_timeout()) with timeout
3257 * and interrupt capability. Also see complete().
3258 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02003259void __sched wait_for_completion(struct completion *x)
Andi Kleen8cbbe862007-10-15 17:00:14 +02003260{
3261 wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262}
3263EXPORT_SYMBOL(wait_for_completion);
3264
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003265/**
3266 * wait_for_completion_timeout: - waits for completion of a task (w/timeout)
3267 * @x: holds the state of this particular completion
3268 * @timeout: timeout value in jiffies
3269 *
3270 * This waits for either a completion of a specific task to be signaled or for a
3271 * specified timeout to expire. The timeout is in jiffies. It is not
3272 * interruptible.
J. Bruce Fieldsc6dc7f02011-10-06 15:22:46 -04003273 *
3274 * The return value is 0 if timed out, and positive (at least 1, or number of
3275 * jiffies left till timeout) if completed.
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003276 */
Ingo Molnarb15136e2007-10-24 18:23:48 +02003277unsigned long __sched
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278wait_for_completion_timeout(struct completion *x, unsigned long timeout)
3279{
Andi Kleen8cbbe862007-10-15 17:00:14 +02003280 return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281}
3282EXPORT_SYMBOL(wait_for_completion_timeout);
3283
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003284/**
3285 * wait_for_completion_interruptible: - waits for completion of a task (w/intr)
3286 * @x: holds the state of this particular completion
3287 *
3288 * This waits for completion of a specific task to be signaled. It is
3289 * interruptible.
J. Bruce Fieldsc6dc7f02011-10-06 15:22:46 -04003290 *
3291 * The return value is -ERESTARTSYS if interrupted, 0 if completed.
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003292 */
Andi Kleen8cbbe862007-10-15 17:00:14 +02003293int __sched wait_for_completion_interruptible(struct completion *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294{
Andi Kleen51e97992007-10-18 21:32:55 +02003295 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
3296 if (t == -ERESTARTSYS)
3297 return t;
3298 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299}
3300EXPORT_SYMBOL(wait_for_completion_interruptible);
3301
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003302/**
3303 * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr))
3304 * @x: holds the state of this particular completion
3305 * @timeout: timeout value in jiffies
3306 *
3307 * This waits for either a completion of a specific task to be signaled or for a
3308 * specified timeout to expire. It is interruptible. The timeout is in jiffies.
J. Bruce Fieldsc6dc7f02011-10-06 15:22:46 -04003309 *
3310 * The return value is -ERESTARTSYS if interrupted, 0 if timed out,
3311 * positive (at least 1, or number of jiffies left till timeout) if completed.
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003312 */
NeilBrown6bf41232011-01-05 12:50:16 +11003313long __sched
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314wait_for_completion_interruptible_timeout(struct completion *x,
3315 unsigned long timeout)
3316{
Andi Kleen8cbbe862007-10-15 17:00:14 +02003317 return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318}
3319EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
3320
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003321/**
3322 * wait_for_completion_killable: - waits for completion of a task (killable)
3323 * @x: holds the state of this particular completion
3324 *
3325 * This waits to be signaled for completion of a specific task. It can be
3326 * interrupted by a kill signal.
J. Bruce Fieldsc6dc7f02011-10-06 15:22:46 -04003327 *
3328 * The return value is -ERESTARTSYS if interrupted, 0 if completed.
Kevin Diggs65eb3dc2008-08-26 10:26:54 +02003329 */
Matthew Wilcox009e5772007-12-06 12:29:54 -05003330int __sched wait_for_completion_killable(struct completion *x)
3331{
3332 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
3333 if (t == -ERESTARTSYS)
3334 return t;
3335 return 0;
3336}
3337EXPORT_SYMBOL(wait_for_completion_killable);
3338
Dave Chinnerbe4de352008-08-15 00:40:44 -07003339/**
Sage Weil0aa12fb2010-05-29 09:12:30 -07003340 * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable))
3341 * @x: holds the state of this particular completion
3342 * @timeout: timeout value in jiffies
3343 *
3344 * This waits for either a completion of a specific task to be
3345 * signaled or for a specified timeout to expire. It can be
3346 * interrupted by a kill signal. The timeout is in jiffies.
J. Bruce Fieldsc6dc7f02011-10-06 15:22:46 -04003347 *
3348 * The return value is -ERESTARTSYS if interrupted, 0 if timed out,
3349 * positive (at least 1, or number of jiffies left till timeout) if completed.
Sage Weil0aa12fb2010-05-29 09:12:30 -07003350 */
NeilBrown6bf41232011-01-05 12:50:16 +11003351long __sched
Sage Weil0aa12fb2010-05-29 09:12:30 -07003352wait_for_completion_killable_timeout(struct completion *x,
3353 unsigned long timeout)
3354{
3355 return wait_for_common(x, timeout, TASK_KILLABLE);
3356}
3357EXPORT_SYMBOL(wait_for_completion_killable_timeout);
3358
3359/**
Dave Chinnerbe4de352008-08-15 00:40:44 -07003360 * try_wait_for_completion - try to decrement a completion without blocking
3361 * @x: completion structure
3362 *
3363 * Returns: 0 if a decrement cannot be done without blocking
3364 * 1 if a decrement succeeded.
3365 *
3366 * If a completion is being used as a counting completion,
3367 * attempt to decrement the counter without blocking. This
3368 * enables us to avoid waiting if the resource the completion
3369 * is protecting is not available.
3370 */
3371bool try_wait_for_completion(struct completion *x)
3372{
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01003373 unsigned long flags;
Dave Chinnerbe4de352008-08-15 00:40:44 -07003374 int ret = 1;
3375
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01003376 spin_lock_irqsave(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07003377 if (!x->done)
3378 ret = 0;
3379 else
3380 x->done--;
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01003381 spin_unlock_irqrestore(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07003382 return ret;
3383}
3384EXPORT_SYMBOL(try_wait_for_completion);
3385
3386/**
3387 * completion_done - Test to see if a completion has any waiters
3388 * @x: completion structure
3389 *
3390 * Returns: 0 if there are waiters (wait_for_completion() in progress)
3391 * 1 if there are no waiters.
3392 *
3393 */
3394bool completion_done(struct completion *x)
3395{
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01003396 unsigned long flags;
Dave Chinnerbe4de352008-08-15 00:40:44 -07003397 int ret = 1;
3398
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01003399 spin_lock_irqsave(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07003400 if (!x->done)
3401 ret = 0;
Rafael J. Wysocki7539a3b2009-12-13 00:07:30 +01003402 spin_unlock_irqrestore(&x->wait.lock, flags);
Dave Chinnerbe4de352008-08-15 00:40:44 -07003403 return ret;
3404}
3405EXPORT_SYMBOL(completion_done);
3406
Andi Kleen8cbbe862007-10-15 17:00:14 +02003407static long __sched
3408sleep_on_common(wait_queue_head_t *q, int state, long timeout)
Ingo Molnar0fec1712007-07-09 18:52:01 +02003409{
3410 unsigned long flags;
3411 wait_queue_t wait;
3412
3413 init_waitqueue_entry(&wait, current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414
Andi Kleen8cbbe862007-10-15 17:00:14 +02003415 __set_current_state(state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003416
Andi Kleen8cbbe862007-10-15 17:00:14 +02003417 spin_lock_irqsave(&q->lock, flags);
3418 __add_wait_queue(q, &wait);
3419 spin_unlock(&q->lock);
3420 timeout = schedule_timeout(timeout);
3421 spin_lock_irq(&q->lock);
3422 __remove_wait_queue(q, &wait);
3423 spin_unlock_irqrestore(&q->lock, flags);
3424
3425 return timeout;
3426}
3427
3428void __sched interruptible_sleep_on(wait_queue_head_t *q)
3429{
3430 sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432EXPORT_SYMBOL(interruptible_sleep_on);
3433
Ingo Molnar0fec1712007-07-09 18:52:01 +02003434long __sched
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07003435interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436{
Andi Kleen8cbbe862007-10-15 17:00:14 +02003437 return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003439EXPORT_SYMBOL(interruptible_sleep_on_timeout);
3440
Ingo Molnar0fec1712007-07-09 18:52:01 +02003441void __sched sleep_on(wait_queue_head_t *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442{
Andi Kleen8cbbe862007-10-15 17:00:14 +02003443 sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445EXPORT_SYMBOL(sleep_on);
3446
Ingo Molnar0fec1712007-07-09 18:52:01 +02003447long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003448{
Andi Kleen8cbbe862007-10-15 17:00:14 +02003449 return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003451EXPORT_SYMBOL(sleep_on_timeout);
3452
Ingo Molnarb29739f2006-06-27 02:54:51 -07003453#ifdef CONFIG_RT_MUTEXES
3454
3455/*
3456 * rt_mutex_setprio - set the current priority of a task
3457 * @p: task
3458 * @prio: prio value (kernel-internal form)
3459 *
3460 * This function changes the 'effective' priority of a task. It does
3461 * not touch ->normal_prio like __setscheduler().
3462 *
3463 * Used by the rt_mutex code to implement priority inheritance logic.
3464 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07003465void rt_mutex_setprio(struct task_struct *p, int prio)
Ingo Molnarb29739f2006-06-27 02:54:51 -07003466{
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02003467 int oldprio, on_rq, running;
Ingo Molnar70b97a72006-07-03 00:25:42 -07003468 struct rq *rq;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01003469 const struct sched_class *prev_class;
Ingo Molnarb29739f2006-06-27 02:54:51 -07003470
3471 BUG_ON(prio < 0 || prio > MAX_PRIO);
3472
Peter Zijlstra0122ec52011-04-05 17:23:51 +02003473 rq = __task_rq_lock(p);
Ingo Molnarb29739f2006-06-27 02:54:51 -07003474
Thomas Gleixner1c4dd992011-06-06 20:07:38 +02003475 /*
3476 * Idle task boosting is a nono in general. There is one
3477 * exception, when PREEMPT_RT and NOHZ is active:
3478 *
3479 * The idle task calls get_next_timer_interrupt() and holds
3480 * the timer wheel base->lock on the CPU and another CPU wants
3481 * to access the timer (probably to cancel it). We can safely
3482 * ignore the boosting request, as the idle CPU runs this code
3483 * with interrupts disabled and will complete the lock
3484 * protected section without being interrupted. So there is no
3485 * real need to boost.
3486 */
3487 if (unlikely(p == rq->idle)) {
3488 WARN_ON(p != rq->curr);
3489 WARN_ON(p->pi_blocked_on);
3490 goto out_unlock;
3491 }
3492
Steven Rostedta8027072010-09-20 15:13:34 -04003493 trace_sched_pi_setprio(p, prio);
Andrew Mortond5f9f942007-05-08 20:27:06 -07003494 oldprio = p->prio;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01003495 prev_class = p->sched_class;
Peter Zijlstrafd2f4412011-04-05 17:23:44 +02003496 on_rq = p->on_rq;
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01003497 running = task_current(rq, p);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07003498 if (on_rq)
Ingo Molnar69be72c2007-08-09 11:16:49 +02003499 dequeue_task(rq, p, 0);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07003500 if (running)
3501 p->sched_class->put_prev_task(rq, p);
Ingo Molnardd41f592007-07-09 18:51:59 +02003502
3503 if (rt_prio(prio))
3504 p->sched_class = &rt_sched_class;
3505 else
3506 p->sched_class = &fair_sched_class;
3507
Ingo Molnarb29739f2006-06-27 02:54:51 -07003508 p->prio = prio;
3509
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07003510 if (running)
3511 p->sched_class->set_curr_task(rq);
Peter Zijlstrada7a7352011-01-17 17:03:27 +01003512 if (on_rq)
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01003513 enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0);
Steven Rostedtcb469842008-01-25 21:08:22 +01003514
Peter Zijlstrada7a7352011-01-17 17:03:27 +01003515 check_class_changed(rq, p, prev_class, oldprio);
Thomas Gleixner1c4dd992011-06-06 20:07:38 +02003516out_unlock:
Peter Zijlstra0122ec52011-04-05 17:23:51 +02003517 __task_rq_unlock(rq);
Ingo Molnarb29739f2006-06-27 02:54:51 -07003518}
Ingo Molnarb29739f2006-06-27 02:54:51 -07003519#endif
Ingo Molnar36c8b582006-07-03 00:25:41 -07003520void set_user_nice(struct task_struct *p, long nice)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003521{
Ingo Molnardd41f592007-07-09 18:51:59 +02003522 int old_prio, delta, on_rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003523 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07003524 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003525
3526 if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
3527 return;
3528 /*
3529 * We have to be careful, if called from sys_setpriority(),
3530 * the task might be in the middle of scheduling on another CPU.
3531 */
3532 rq = task_rq_lock(p, &flags);
3533 /*
3534 * The RT priorities are set via sched_setscheduler(), but we still
3535 * allow the 'normal' nice value to be set - but as expected
3536 * it wont have any effect on scheduling until the task is
Ingo Molnardd41f592007-07-09 18:51:59 +02003537 * SCHED_FIFO/SCHED_RR:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003538 */
Ingo Molnare05606d2007-07-09 18:51:59 +02003539 if (task_has_rt_policy(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540 p->static_prio = NICE_TO_PRIO(nice);
3541 goto out_unlock;
3542 }
Peter Zijlstrafd2f4412011-04-05 17:23:44 +02003543 on_rq = p->on_rq;
Peter Zijlstrac09595f2008-06-27 13:41:14 +02003544 if (on_rq)
Ingo Molnar69be72c2007-08-09 11:16:49 +02003545 dequeue_task(rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547 p->static_prio = NICE_TO_PRIO(nice);
Peter Williams2dd73a42006-06-27 02:54:34 -07003548 set_load_weight(p);
Ingo Molnarb29739f2006-06-27 02:54:51 -07003549 old_prio = p->prio;
3550 p->prio = effective_prio(p);
3551 delta = p->prio - old_prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552
Ingo Molnardd41f592007-07-09 18:51:59 +02003553 if (on_rq) {
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01003554 enqueue_task(rq, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003555 /*
Andrew Mortond5f9f942007-05-08 20:27:06 -07003556 * If the task increased its priority or is running and
3557 * lowered its priority, then reschedule its CPU:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003558 */
Andrew Mortond5f9f942007-05-08 20:27:06 -07003559 if (delta < 0 || (delta > 0 && task_running(rq, p)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560 resched_task(rq->curr);
3561 }
3562out_unlock:
Peter Zijlstra0122ec52011-04-05 17:23:51 +02003563 task_rq_unlock(rq, p, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003564}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565EXPORT_SYMBOL(set_user_nice);
3566
Matt Mackalle43379f2005-05-01 08:59:00 -07003567/*
3568 * can_nice - check if a task can reduce its nice value
3569 * @p: task
3570 * @nice: nice value
3571 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07003572int can_nice(const struct task_struct *p, const int nice)
Matt Mackalle43379f2005-05-01 08:59:00 -07003573{
Matt Mackall024f4742005-08-18 11:24:19 -07003574 /* convert nice value [19,-20] to rlimit style value [1,40] */
3575 int nice_rlim = 20 - nice;
Ingo Molnar48f24c42006-07-03 00:25:40 -07003576
Jiri Slaby78d7d402010-03-05 13:42:54 -08003577 return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
Matt Mackalle43379f2005-05-01 08:59:00 -07003578 capable(CAP_SYS_NICE));
3579}
3580
Linus Torvalds1da177e2005-04-16 15:20:36 -07003581#ifdef __ARCH_WANT_SYS_NICE
3582
3583/*
3584 * sys_nice - change the priority of the current process.
3585 * @increment: priority increment
3586 *
3587 * sys_setpriority is a more generic, but much slower function that
3588 * does similar things.
3589 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01003590SYSCALL_DEFINE1(nice, int, increment)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591{
Ingo Molnar48f24c42006-07-03 00:25:40 -07003592 long nice, retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593
3594 /*
3595 * Setpriority might change our priority at the same moment.
3596 * We don't have to worry. Conceptually one call occurs first
3597 * and we have a single winner.
3598 */
Matt Mackalle43379f2005-05-01 08:59:00 -07003599 if (increment < -40)
3600 increment = -40;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003601 if (increment > 40)
3602 increment = 40;
3603
Américo Wang2b8f8362009-02-16 18:54:21 +08003604 nice = TASK_NICE(current) + increment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003605 if (nice < -20)
3606 nice = -20;
3607 if (nice > 19)
3608 nice = 19;
3609
Matt Mackalle43379f2005-05-01 08:59:00 -07003610 if (increment < 0 && !can_nice(current, nice))
3611 return -EPERM;
3612
Linus Torvalds1da177e2005-04-16 15:20:36 -07003613 retval = security_task_setnice(current, nice);
3614 if (retval)
3615 return retval;
3616
3617 set_user_nice(current, nice);
3618 return 0;
3619}
3620
3621#endif
3622
3623/**
3624 * task_prio - return the priority value of a given task.
3625 * @p: the task in question.
3626 *
3627 * This is the priority value as seen by users in /proc.
3628 * RT tasks are offset by -200. Normal tasks are centered
3629 * around 0, value goes from -16 to +15.
3630 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07003631int task_prio(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003632{
3633 return p->prio - MAX_RT_PRIO;
3634}
3635
3636/**
3637 * task_nice - return the nice value of a given task.
3638 * @p: the task in question.
3639 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07003640int task_nice(const struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641{
3642 return TASK_NICE(p);
3643}
Pavel Roskin150d8be2008-03-05 16:56:37 -05003644EXPORT_SYMBOL(task_nice);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645
3646/**
3647 * idle_cpu - is a given cpu idle currently?
3648 * @cpu: the processor in question.
3649 */
3650int idle_cpu(int cpu)
3651{
Thomas Gleixner908a3282011-09-15 15:32:06 +02003652 struct rq *rq = cpu_rq(cpu);
3653
3654 if (rq->curr != rq->idle)
3655 return 0;
3656
3657 if (rq->nr_running)
3658 return 0;
3659
3660#ifdef CONFIG_SMP
3661 if (!llist_empty(&rq->wake_list))
3662 return 0;
3663#endif
3664
3665 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003666}
3667
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668/**
3669 * idle_task - return the idle task for a given cpu.
3670 * @cpu: the processor in question.
3671 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07003672struct task_struct *idle_task(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003673{
3674 return cpu_rq(cpu)->idle;
3675}
3676
3677/**
3678 * find_process_by_pid - find a process with a matching PID value.
3679 * @pid: the pid in question.
3680 */
Alexey Dobriyana9957442007-10-15 17:00:13 +02003681static struct task_struct *find_process_by_pid(pid_t pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682{
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07003683 return pid ? find_task_by_vpid(pid) : current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684}
3685
3686/* Actually do priority change: must hold rq lock. */
Ingo Molnardd41f592007-07-09 18:51:59 +02003687static void
3688__setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003690 p->policy = policy;
3691 p->rt_priority = prio;
Ingo Molnarb29739f2006-06-27 02:54:51 -07003692 p->normal_prio = normal_prio(p);
3693 /* we are holding p->pi_lock already */
3694 p->prio = rt_mutex_getprio(p);
Peter Zijlstraffd44db2009-11-10 20:12:01 +01003695 if (rt_prio(p->prio))
3696 p->sched_class = &rt_sched_class;
3697 else
3698 p->sched_class = &fair_sched_class;
Peter Williams2dd73a42006-06-27 02:54:34 -07003699 set_load_weight(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003700}
3701
David Howellsc69e8d92008-11-14 10:39:19 +11003702/*
3703 * check the target process has a UID that matches the current process's
3704 */
3705static bool check_same_owner(struct task_struct *p)
3706{
3707 const struct cred *cred = current_cred(), *pcred;
3708 bool match;
3709
3710 rcu_read_lock();
3711 pcred = __task_cred(p);
Eric W. Biederman9c806aa2012-02-02 18:54:02 -08003712 match = (uid_eq(cred->euid, pcred->euid) ||
3713 uid_eq(cred->euid, pcred->uid));
David Howellsc69e8d92008-11-14 10:39:19 +11003714 rcu_read_unlock();
3715 return match;
3716}
3717
Rusty Russell961ccdd2008-06-23 13:55:38 +10003718static int __sched_setscheduler(struct task_struct *p, int policy,
KOSAKI Motohirofe7de492010-10-20 16:01:12 -07003719 const struct sched_param *param, bool user)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720{
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02003721 int retval, oldprio, oldpolicy = -1, on_rq, running;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722 unsigned long flags;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01003723 const struct sched_class *prev_class;
Ingo Molnar70b97a72006-07-03 00:25:42 -07003724 struct rq *rq;
Lennart Poetteringca94c442009-06-15 17:17:47 +02003725 int reset_on_fork;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726
Steven Rostedt66e53932006-06-27 02:54:44 -07003727 /* may grab non-irq protected spin_locks */
3728 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07003729recheck:
3730 /* double check policy once rq lock held */
Lennart Poetteringca94c442009-06-15 17:17:47 +02003731 if (policy < 0) {
3732 reset_on_fork = p->sched_reset_on_fork;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003733 policy = oldpolicy = p->policy;
Lennart Poetteringca94c442009-06-15 17:17:47 +02003734 } else {
3735 reset_on_fork = !!(policy & SCHED_RESET_ON_FORK);
3736 policy &= ~SCHED_RESET_ON_FORK;
3737
3738 if (policy != SCHED_FIFO && policy != SCHED_RR &&
3739 policy != SCHED_NORMAL && policy != SCHED_BATCH &&
3740 policy != SCHED_IDLE)
3741 return -EINVAL;
3742 }
3743
Linus Torvalds1da177e2005-04-16 15:20:36 -07003744 /*
3745 * Valid priorities for SCHED_FIFO and SCHED_RR are
Ingo Molnardd41f592007-07-09 18:51:59 +02003746 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
3747 * SCHED_BATCH and SCHED_IDLE is 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748 */
3749 if (param->sched_priority < 0 ||
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07003750 (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
Steven Rostedtd46523e2005-07-25 16:28:39 -04003751 (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003752 return -EINVAL;
Ingo Molnare05606d2007-07-09 18:51:59 +02003753 if (rt_policy(policy) != (param->sched_priority != 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754 return -EINVAL;
3755
Olivier Croquette37e4ab32005-06-25 14:57:32 -07003756 /*
3757 * Allow unprivileged RT tasks to decrease priority:
3758 */
Rusty Russell961ccdd2008-06-23 13:55:38 +10003759 if (user && !capable(CAP_SYS_NICE)) {
Ingo Molnare05606d2007-07-09 18:51:59 +02003760 if (rt_policy(policy)) {
Oleg Nesterova44702e2010-06-11 01:09:44 +02003761 unsigned long rlim_rtprio =
3762 task_rlimit(p, RLIMIT_RTPRIO);
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07003763
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07003764 /* can't set/change the rt policy */
3765 if (policy != p->policy && !rlim_rtprio)
3766 return -EPERM;
3767
3768 /* can't increase priority */
3769 if (param->sched_priority > p->rt_priority &&
3770 param->sched_priority > rlim_rtprio)
3771 return -EPERM;
3772 }
Darren Hartc02aa732011-02-17 15:37:07 -08003773
Ingo Molnardd41f592007-07-09 18:51:59 +02003774 /*
Darren Hartc02aa732011-02-17 15:37:07 -08003775 * Treat SCHED_IDLE as nice 20. Only allow a switch to
3776 * SCHED_NORMAL if the RLIMIT_NICE would normally permit it.
Ingo Molnardd41f592007-07-09 18:51:59 +02003777 */
Darren Hartc02aa732011-02-17 15:37:07 -08003778 if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) {
3779 if (!can_nice(p, TASK_NICE(p)))
3780 return -EPERM;
3781 }
Oleg Nesterov8dc3e902006-09-29 02:00:50 -07003782
Olivier Croquette37e4ab32005-06-25 14:57:32 -07003783 /* can't change other user's priorities */
David Howellsc69e8d92008-11-14 10:39:19 +11003784 if (!check_same_owner(p))
Olivier Croquette37e4ab32005-06-25 14:57:32 -07003785 return -EPERM;
Lennart Poetteringca94c442009-06-15 17:17:47 +02003786
3787 /* Normal users shall not reset the sched_reset_on_fork flag */
3788 if (p->sched_reset_on_fork && !reset_on_fork)
3789 return -EPERM;
Olivier Croquette37e4ab32005-06-25 14:57:32 -07003790 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791
Jeremy Fitzhardinge725aad22008-08-03 09:33:03 -07003792 if (user) {
KOSAKI Motohirob0ae1982010-10-15 04:21:18 +09003793 retval = security_task_setscheduler(p);
Jeremy Fitzhardinge725aad22008-08-03 09:33:03 -07003794 if (retval)
3795 return retval;
3796 }
3797
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 /*
Ingo Molnarb29739f2006-06-27 02:54:51 -07003799 * make sure no PI-waiters arrive (or leave) while we are
3800 * changing the priority of the task:
Peter Zijlstra0122ec52011-04-05 17:23:51 +02003801 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003802 * To be able to change p->policy safely, the appropriate
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803 * runqueue lock must be held.
3804 */
Peter Zijlstra0122ec52011-04-05 17:23:51 +02003805 rq = task_rq_lock(p, &flags);
Peter Zijlstradc61b1d2010-06-08 11:40:42 +02003806
Peter Zijlstra34f971f2010-09-22 13:53:15 +02003807 /*
3808 * Changing the policy of the stop threads its a very bad idea
3809 */
3810 if (p == rq->stop) {
Peter Zijlstra0122ec52011-04-05 17:23:51 +02003811 task_rq_unlock(rq, p, &flags);
Peter Zijlstra34f971f2010-09-22 13:53:15 +02003812 return -EINVAL;
3813 }
3814
Dario Faggiolia51e9192011-03-24 14:00:18 +01003815 /*
3816 * If not changing anything there's no need to proceed further:
3817 */
3818 if (unlikely(policy == p->policy && (!rt_policy(policy) ||
3819 param->sched_priority == p->rt_priority))) {
Namhyung Kim45afb172012-07-07 16:49:02 +09003820 task_rq_unlock(rq, p, &flags);
Dario Faggiolia51e9192011-03-24 14:00:18 +01003821 return 0;
3822 }
3823
Peter Zijlstradc61b1d2010-06-08 11:40:42 +02003824#ifdef CONFIG_RT_GROUP_SCHED
3825 if (user) {
3826 /*
3827 * Do not allow realtime tasks into groups that have no runtime
3828 * assigned.
3829 */
3830 if (rt_bandwidth_enabled() && rt_policy(policy) &&
Mike Galbraithf4493772011-01-13 04:54:50 +01003831 task_group(p)->rt_bandwidth.rt_runtime == 0 &&
3832 !task_group_is_autogroup(task_group(p))) {
Peter Zijlstra0122ec52011-04-05 17:23:51 +02003833 task_rq_unlock(rq, p, &flags);
Peter Zijlstradc61b1d2010-06-08 11:40:42 +02003834 return -EPERM;
3835 }
3836 }
3837#endif
3838
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839 /* recheck policy now with rq lock held */
3840 if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
3841 policy = oldpolicy = -1;
Peter Zijlstra0122ec52011-04-05 17:23:51 +02003842 task_rq_unlock(rq, p, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003843 goto recheck;
3844 }
Peter Zijlstrafd2f4412011-04-05 17:23:44 +02003845 on_rq = p->on_rq;
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01003846 running = task_current(rq, p);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07003847 if (on_rq)
Peter Zijlstra4ca9b722012-01-25 11:50:51 +01003848 dequeue_task(rq, p, 0);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07003849 if (running)
3850 p->sched_class->put_prev_task(rq, p);
Dmitry Adamushkof6b53202007-10-15 17:00:08 +02003851
Lennart Poetteringca94c442009-06-15 17:17:47 +02003852 p->sched_reset_on_fork = reset_on_fork;
3853
Linus Torvalds1da177e2005-04-16 15:20:36 -07003854 oldprio = p->prio;
Thomas Gleixner83ab0aa2010-02-17 09:05:48 +01003855 prev_class = p->sched_class;
Ingo Molnardd41f592007-07-09 18:51:59 +02003856 __setscheduler(rq, p, policy, param->sched_priority);
Dmitry Adamushkof6b53202007-10-15 17:00:08 +02003857
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07003858 if (running)
3859 p->sched_class->set_curr_task(rq);
Peter Zijlstrada7a7352011-01-17 17:03:27 +01003860 if (on_rq)
Peter Zijlstra4ca9b722012-01-25 11:50:51 +01003861 enqueue_task(rq, p, 0);
Steven Rostedtcb469842008-01-25 21:08:22 +01003862
Peter Zijlstrada7a7352011-01-17 17:03:27 +01003863 check_class_changed(rq, p, prev_class, oldprio);
Peter Zijlstra0122ec52011-04-05 17:23:51 +02003864 task_rq_unlock(rq, p, &flags);
Ingo Molnarb29739f2006-06-27 02:54:51 -07003865
Thomas Gleixner95e02ca2006-06-27 02:55:02 -07003866 rt_mutex_adjust_pi(p);
3867
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868 return 0;
3869}
Rusty Russell961ccdd2008-06-23 13:55:38 +10003870
3871/**
3872 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
3873 * @p: the task in question.
3874 * @policy: new policy.
3875 * @param: structure containing the new RT priority.
3876 *
3877 * NOTE that the task may be already dead.
3878 */
3879int sched_setscheduler(struct task_struct *p, int policy,
KOSAKI Motohirofe7de492010-10-20 16:01:12 -07003880 const struct sched_param *param)
Rusty Russell961ccdd2008-06-23 13:55:38 +10003881{
3882 return __sched_setscheduler(p, policy, param, true);
3883}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884EXPORT_SYMBOL_GPL(sched_setscheduler);
3885
Rusty Russell961ccdd2008-06-23 13:55:38 +10003886/**
3887 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
3888 * @p: the task in question.
3889 * @policy: new policy.
3890 * @param: structure containing the new RT priority.
3891 *
3892 * Just like sched_setscheduler, only don't bother checking if the
3893 * current context has permission. For example, this is needed in
3894 * stop_machine(): we create temporary high priority worker threads,
3895 * but our caller might not have that capability.
3896 */
3897int sched_setscheduler_nocheck(struct task_struct *p, int policy,
KOSAKI Motohirofe7de492010-10-20 16:01:12 -07003898 const struct sched_param *param)
Rusty Russell961ccdd2008-06-23 13:55:38 +10003899{
3900 return __sched_setscheduler(p, policy, param, false);
3901}
3902
Ingo Molnar95cdf3b2005-09-10 00:26:11 -07003903static int
3904do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003906 struct sched_param lparam;
3907 struct task_struct *p;
Ingo Molnar36c8b582006-07-03 00:25:41 -07003908 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003909
3910 if (!param || pid < 0)
3911 return -EINVAL;
3912 if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
3913 return -EFAULT;
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07003914
3915 rcu_read_lock();
3916 retval = -ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003917 p = find_process_by_pid(pid);
Oleg Nesterov5fe1d752006-09-29 02:00:48 -07003918 if (p != NULL)
3919 retval = sched_setscheduler(p, policy, &lparam);
3920 rcu_read_unlock();
Ingo Molnar36c8b582006-07-03 00:25:41 -07003921
Linus Torvalds1da177e2005-04-16 15:20:36 -07003922 return retval;
3923}
3924
3925/**
3926 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
3927 * @pid: the pid in question.
3928 * @policy: new policy.
3929 * @param: structure containing the new RT priority.
3930 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01003931SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
3932 struct sched_param __user *, param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003933{
Jason Baronc21761f2006-01-18 17:43:03 -08003934 /* negative values for policy are not valid */
3935 if (policy < 0)
3936 return -EINVAL;
3937
Linus Torvalds1da177e2005-04-16 15:20:36 -07003938 return do_sched_setscheduler(pid, policy, param);
3939}
3940
3941/**
3942 * sys_sched_setparam - set/change the RT priority of a thread
3943 * @pid: the pid in question.
3944 * @param: structure containing the new RT priority.
3945 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01003946SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003947{
3948 return do_sched_setscheduler(pid, -1, param);
3949}
3950
3951/**
3952 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
3953 * @pid: the pid in question.
3954 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01003955SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003956{
Ingo Molnar36c8b582006-07-03 00:25:41 -07003957 struct task_struct *p;
Andi Kleen3a5c3592007-10-15 17:00:14 +02003958 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959
3960 if (pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02003961 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003962
3963 retval = -ESRCH;
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00003964 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965 p = find_process_by_pid(pid);
3966 if (p) {
3967 retval = security_task_getscheduler(p);
3968 if (!retval)
Lennart Poetteringca94c442009-06-15 17:17:47 +02003969 retval = p->policy
3970 | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003971 }
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00003972 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003973 return retval;
3974}
3975
3976/**
Lennart Poetteringca94c442009-06-15 17:17:47 +02003977 * sys_sched_getparam - get the RT priority of a thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07003978 * @pid: the pid in question.
3979 * @param: structure containing the RT priority.
3980 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01003981SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003982{
3983 struct sched_param lp;
Ingo Molnar36c8b582006-07-03 00:25:41 -07003984 struct task_struct *p;
Andi Kleen3a5c3592007-10-15 17:00:14 +02003985 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986
3987 if (!param || pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02003988 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00003990 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991 p = find_process_by_pid(pid);
3992 retval = -ESRCH;
3993 if (!p)
3994 goto out_unlock;
3995
3996 retval = security_task_getscheduler(p);
3997 if (retval)
3998 goto out_unlock;
3999
4000 lp.sched_priority = p->rt_priority;
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004001 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004002
4003 /*
4004 * This one might sleep, we cannot do it with a spinlock held ...
4005 */
4006 retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
4007
Linus Torvalds1da177e2005-04-16 15:20:36 -07004008 return retval;
4009
4010out_unlock:
Thomas Gleixner5fe85be2009-12-09 10:14:58 +00004011 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004012 return retval;
4013}
4014
Rusty Russell96f874e2008-11-25 02:35:14 +10304015long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004016{
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304017 cpumask_var_t cpus_allowed, new_mask;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004018 struct task_struct *p;
4019 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004020
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004021 get_online_cpus();
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004022 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004023
4024 p = find_process_by_pid(pid);
4025 if (!p) {
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004026 rcu_read_unlock();
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004027 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028 return -ESRCH;
4029 }
4030
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004031 /* Prevent p going away */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004032 get_task_struct(p);
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004033 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004034
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304035 if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
4036 retval = -ENOMEM;
4037 goto out_put_task;
4038 }
4039 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
4040 retval = -ENOMEM;
4041 goto out_free_cpus_allowed;
4042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004043 retval = -EPERM;
Linus Torvaldsc49c41a2012-01-14 18:36:33 -08004044 if (!check_same_owner(p) && !ns_capable(task_user_ns(p), CAP_SYS_NICE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045 goto out_unlock;
4046
KOSAKI Motohirob0ae1982010-10-15 04:21:18 +09004047 retval = security_task_setscheduler(p);
David Quigleye7834f82006-06-23 02:03:59 -07004048 if (retval)
4049 goto out_unlock;
4050
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304051 cpuset_cpus_allowed(p, cpus_allowed);
4052 cpumask_and(new_mask, in_mask, cpus_allowed);
Peter Zijlstra49246272010-10-17 21:46:10 +02004053again:
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304054 retval = set_cpus_allowed_ptr(p, new_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004055
Paul Menage8707d8b2007-10-18 23:40:22 -07004056 if (!retval) {
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304057 cpuset_cpus_allowed(p, cpus_allowed);
4058 if (!cpumask_subset(new_mask, cpus_allowed)) {
Paul Menage8707d8b2007-10-18 23:40:22 -07004059 /*
4060 * We must have raced with a concurrent cpuset
4061 * update. Just reset the cpus_allowed to the
4062 * cpuset's cpus_allowed
4063 */
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304064 cpumask_copy(new_mask, cpus_allowed);
Paul Menage8707d8b2007-10-18 23:40:22 -07004065 goto again;
4066 }
4067 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004068out_unlock:
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304069 free_cpumask_var(new_mask);
4070out_free_cpus_allowed:
4071 free_cpumask_var(cpus_allowed);
4072out_put_task:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004073 put_task_struct(p);
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004074 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075 return retval;
4076}
4077
4078static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
Rusty Russell96f874e2008-11-25 02:35:14 +10304079 struct cpumask *new_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004080{
Rusty Russell96f874e2008-11-25 02:35:14 +10304081 if (len < cpumask_size())
4082 cpumask_clear(new_mask);
4083 else if (len > cpumask_size())
4084 len = cpumask_size();
4085
Linus Torvalds1da177e2005-04-16 15:20:36 -07004086 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
4087}
4088
4089/**
4090 * sys_sched_setaffinity - set the cpu affinity of a process
4091 * @pid: pid of the process
4092 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4093 * @user_mask_ptr: user-space pointer to the new cpu mask
4094 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004095SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
4096 unsigned long __user *, user_mask_ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004097{
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304098 cpumask_var_t new_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004099 int retval;
4100
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304101 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
4102 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004103
Rusty Russell5a16f3d2008-11-25 02:35:11 +10304104 retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
4105 if (retval == 0)
4106 retval = sched_setaffinity(pid, new_mask);
4107 free_cpumask_var(new_mask);
4108 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109}
4110
Rusty Russell96f874e2008-11-25 02:35:14 +10304111long sched_getaffinity(pid_t pid, struct cpumask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004112{
Ingo Molnar36c8b582006-07-03 00:25:41 -07004113 struct task_struct *p;
Thomas Gleixner31605682009-12-08 20:24:16 +00004114 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004115 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004116
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004117 get_online_cpus();
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004118 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004119
4120 retval = -ESRCH;
4121 p = find_process_by_pid(pid);
4122 if (!p)
4123 goto out_unlock;
4124
David Quigleye7834f82006-06-23 02:03:59 -07004125 retval = security_task_getscheduler(p);
4126 if (retval)
4127 goto out_unlock;
4128
Peter Zijlstra013fdb82011-04-05 17:23:45 +02004129 raw_spin_lock_irqsave(&p->pi_lock, flags);
Rusty Russell96f874e2008-11-25 02:35:14 +10304130 cpumask_and(mask, &p->cpus_allowed, cpu_online_mask);
Peter Zijlstra013fdb82011-04-05 17:23:45 +02004131 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004132
4133out_unlock:
Thomas Gleixner23f5d142009-12-09 10:15:01 +00004134 rcu_read_unlock();
Gautham R Shenoy95402b32008-01-25 21:08:02 +01004135 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004136
Ulrich Drepper9531b622007-08-09 11:16:46 +02004137 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004138}
4139
4140/**
4141 * sys_sched_getaffinity - get the cpu affinity of a process
4142 * @pid: pid of the process
4143 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4144 * @user_mask_ptr: user-space pointer to hold the current cpu mask
4145 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004146SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
4147 unsigned long __user *, user_mask_ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004148{
4149 int ret;
Rusty Russellf17c8602008-11-25 02:35:11 +10304150 cpumask_var_t mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004151
Anton Blanchard84fba5e2010-04-06 17:02:19 +10004152 if ((len * BITS_PER_BYTE) < nr_cpu_ids)
KOSAKI Motohirocd3d8032010-03-12 16:15:36 +09004153 return -EINVAL;
4154 if (len & (sizeof(unsigned long)-1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004155 return -EINVAL;
4156
Rusty Russellf17c8602008-11-25 02:35:11 +10304157 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
4158 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004159
Rusty Russellf17c8602008-11-25 02:35:11 +10304160 ret = sched_getaffinity(pid, mask);
4161 if (ret == 0) {
KOSAKI Motohiro8bc037f2010-03-17 09:36:58 +09004162 size_t retlen = min_t(size_t, len, cpumask_size());
KOSAKI Motohirocd3d8032010-03-12 16:15:36 +09004163
4164 if (copy_to_user(user_mask_ptr, mask, retlen))
Rusty Russellf17c8602008-11-25 02:35:11 +10304165 ret = -EFAULT;
4166 else
KOSAKI Motohirocd3d8032010-03-12 16:15:36 +09004167 ret = retlen;
Rusty Russellf17c8602008-11-25 02:35:11 +10304168 }
4169 free_cpumask_var(mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004170
Rusty Russellf17c8602008-11-25 02:35:11 +10304171 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004172}
4173
4174/**
4175 * sys_sched_yield - yield the current processor to other threads.
4176 *
Ingo Molnardd41f592007-07-09 18:51:59 +02004177 * This function yields the current CPU to other tasks. If there are no
4178 * other threads running on this CPU then this function will return.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004179 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004180SYSCALL_DEFINE0(sched_yield)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181{
Ingo Molnar70b97a72006-07-03 00:25:42 -07004182 struct rq *rq = this_rq_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183
Ingo Molnar2d723762007-10-15 17:00:12 +02004184 schedstat_inc(rq, yld_count);
Dmitry Adamushko4530d7a2007-10-15 17:00:08 +02004185 current->sched_class->yield_task(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004186
4187 /*
4188 * Since we are going to call schedule() anyway, there's
4189 * no need to preempt or enable interrupts:
4190 */
4191 __release(rq->lock);
Ingo Molnar8a25d5d2006-07-03 00:24:54 -07004192 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
Thomas Gleixner9828ea92009-12-03 20:55:53 +01004193 do_raw_spin_unlock(&rq->lock);
Thomas Gleixnerba74c142011-03-21 13:32:17 +01004194 sched_preempt_enable_no_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004195
4196 schedule();
4197
4198 return 0;
4199}
4200
Peter Zijlstrad86ee482009-07-10 14:57:57 +02004201static inline int should_resched(void)
4202{
4203 return need_resched() && !(preempt_count() & PREEMPT_ACTIVE);
4204}
4205
Andrew Mortone7b38402006-06-30 01:56:00 -07004206static void __cond_resched(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004207{
Frederic Weisbeckere7aaaa62009-07-16 15:44:29 +02004208 add_preempt_count(PREEMPT_ACTIVE);
Thomas Gleixnerc259e012011-06-22 19:47:00 +02004209 __schedule();
Frederic Weisbeckere7aaaa62009-07-16 15:44:29 +02004210 sub_preempt_count(PREEMPT_ACTIVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004211}
4212
Herbert Xu02b67cc32008-01-25 21:08:28 +01004213int __sched _cond_resched(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004214{
Peter Zijlstrad86ee482009-07-10 14:57:57 +02004215 if (should_resched()) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004216 __cond_resched();
4217 return 1;
4218 }
4219 return 0;
4220}
Herbert Xu02b67cc32008-01-25 21:08:28 +01004221EXPORT_SYMBOL(_cond_resched);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004222
4223/*
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02004224 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004225 * call schedule, and on return reacquire the lock.
4226 *
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01004227 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
Linus Torvalds1da177e2005-04-16 15:20:36 -07004228 * operations here to prevent schedule() from being called twice (once via
4229 * spin_unlock(), once by hand).
4230 */
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02004231int __cond_resched_lock(spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004232{
Peter Zijlstrad86ee482009-07-10 14:57:57 +02004233 int resched = should_resched();
Jan Kara6df3cec2005-06-13 15:52:32 -07004234 int ret = 0;
4235
Peter Zijlstraf607c662009-07-20 19:16:29 +02004236 lockdep_assert_held(lock);
4237
Nick Piggin95c354f2008-01-30 13:31:20 +01004238 if (spin_needbreak(lock) || resched) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004239 spin_unlock(lock);
Peter Zijlstrad86ee482009-07-10 14:57:57 +02004240 if (resched)
Nick Piggin95c354f2008-01-30 13:31:20 +01004241 __cond_resched();
4242 else
4243 cpu_relax();
Jan Kara6df3cec2005-06-13 15:52:32 -07004244 ret = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004245 spin_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004246 }
Jan Kara6df3cec2005-06-13 15:52:32 -07004247 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248}
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02004249EXPORT_SYMBOL(__cond_resched_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004250
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02004251int __sched __cond_resched_softirq(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004252{
4253 BUG_ON(!in_softirq());
4254
Peter Zijlstrad86ee482009-07-10 14:57:57 +02004255 if (should_resched()) {
Thomas Gleixner98d825672007-05-23 13:58:18 -07004256 local_bh_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004257 __cond_resched();
4258 local_bh_disable();
4259 return 1;
4260 }
4261 return 0;
4262}
Frederic Weisbecker613afbf2009-07-16 15:44:29 +02004263EXPORT_SYMBOL(__cond_resched_softirq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004264
Linus Torvalds1da177e2005-04-16 15:20:36 -07004265/**
4266 * yield - yield the current processor to other threads.
4267 *
Peter Zijlstra8e3fabf2012-03-06 18:54:26 +01004268 * Do not ever use this function, there's a 99% chance you're doing it wrong.
4269 *
4270 * The scheduler is at all times free to pick the calling task as the most
4271 * eligible task to run, if removing the yield() call from your code breaks
4272 * it, its already broken.
4273 *
4274 * Typical broken usage is:
4275 *
4276 * while (!event)
4277 * yield();
4278 *
4279 * where one assumes that yield() will let 'the other' process run that will
4280 * make event true. If the current task is a SCHED_FIFO task that will never
4281 * happen. Never use yield() as a progress guarantee!!
4282 *
4283 * If you want to use yield() to wait for something, use wait_event().
4284 * If you want to use yield() to be 'nice' for others, use cond_resched().
4285 * If you still want to use yield(), do not!
Linus Torvalds1da177e2005-04-16 15:20:36 -07004286 */
4287void __sched yield(void)
4288{
4289 set_current_state(TASK_RUNNING);
4290 sys_sched_yield();
4291}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004292EXPORT_SYMBOL(yield);
4293
Mike Galbraithd95f4122011-02-01 09:50:51 -05004294/**
4295 * yield_to - yield the current processor to another thread in
4296 * your thread group, or accelerate that thread toward the
4297 * processor it's on.
Randy Dunlap16addf92011-03-18 09:34:53 -07004298 * @p: target task
4299 * @preempt: whether task preemption is allowed or not
Mike Galbraithd95f4122011-02-01 09:50:51 -05004300 *
4301 * It's the caller's job to ensure that the target task struct
4302 * can't go away on us before we can do any checks.
4303 *
4304 * Returns true if we indeed boosted the target task.
4305 */
4306bool __sched yield_to(struct task_struct *p, bool preempt)
4307{
4308 struct task_struct *curr = current;
4309 struct rq *rq, *p_rq;
4310 unsigned long flags;
4311 bool yielded = 0;
4312
4313 local_irq_save(flags);
4314 rq = this_rq();
4315
4316again:
4317 p_rq = task_rq(p);
4318 double_rq_lock(rq, p_rq);
4319 while (task_rq(p) != p_rq) {
4320 double_rq_unlock(rq, p_rq);
4321 goto again;
4322 }
4323
4324 if (!curr->sched_class->yield_to_task)
4325 goto out;
4326
4327 if (curr->sched_class != p->sched_class)
4328 goto out;
4329
4330 if (task_running(p_rq, p) || p->state)
4331 goto out;
4332
4333 yielded = curr->sched_class->yield_to_task(rq, p, preempt);
Venkatesh Pallipadi6d1cafd2011-03-01 16:28:21 -08004334 if (yielded) {
Mike Galbraithd95f4122011-02-01 09:50:51 -05004335 schedstat_inc(rq, yld_count);
Venkatesh Pallipadi6d1cafd2011-03-01 16:28:21 -08004336 /*
4337 * Make p's CPU reschedule; pick_next_entity takes care of
4338 * fairness.
4339 */
4340 if (preempt && rq != p_rq)
4341 resched_task(p_rq->curr);
4342 }
Mike Galbraithd95f4122011-02-01 09:50:51 -05004343
4344out:
4345 double_rq_unlock(rq, p_rq);
4346 local_irq_restore(flags);
4347
4348 if (yielded)
4349 schedule();
4350
4351 return yielded;
4352}
4353EXPORT_SYMBOL_GPL(yield_to);
4354
Linus Torvalds1da177e2005-04-16 15:20:36 -07004355/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01004356 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
Linus Torvalds1da177e2005-04-16 15:20:36 -07004357 * that process accounting knows that this is a task in IO wait state.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004358 */
4359void __sched io_schedule(void)
4360{
Hitoshi Mitake54d35f22009-06-29 14:44:57 +09004361 struct rq *rq = raw_rq();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004362
Shailabh Nagar0ff92242006-07-14 00:24:37 -07004363 delayacct_blkio_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004364 atomic_inc(&rq->nr_iowait);
Jens Axboe73c10102011-03-08 13:19:51 +01004365 blk_flush_plug(current);
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07004366 current->in_iowait = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004367 schedule();
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07004368 current->in_iowait = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004369 atomic_dec(&rq->nr_iowait);
Shailabh Nagar0ff92242006-07-14 00:24:37 -07004370 delayacct_blkio_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004371}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004372EXPORT_SYMBOL(io_schedule);
4373
4374long __sched io_schedule_timeout(long timeout)
4375{
Hitoshi Mitake54d35f22009-06-29 14:44:57 +09004376 struct rq *rq = raw_rq();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004377 long ret;
4378
Shailabh Nagar0ff92242006-07-14 00:24:37 -07004379 delayacct_blkio_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004380 atomic_inc(&rq->nr_iowait);
Jens Axboe73c10102011-03-08 13:19:51 +01004381 blk_flush_plug(current);
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07004382 current->in_iowait = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004383 ret = schedule_timeout(timeout);
Arjan van de Ven8f0dfc32009-07-20 11:26:58 -07004384 current->in_iowait = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004385 atomic_dec(&rq->nr_iowait);
Shailabh Nagar0ff92242006-07-14 00:24:37 -07004386 delayacct_blkio_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004387 return ret;
4388}
4389
4390/**
4391 * sys_sched_get_priority_max - return maximum RT priority.
4392 * @policy: scheduling class.
4393 *
4394 * this syscall returns the maximum rt_priority that can be used
4395 * by a given scheduling class.
4396 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004397SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004398{
4399 int ret = -EINVAL;
4400
4401 switch (policy) {
4402 case SCHED_FIFO:
4403 case SCHED_RR:
4404 ret = MAX_USER_RT_PRIO-1;
4405 break;
4406 case SCHED_NORMAL:
Ingo Molnarb0a94992006-01-14 13:20:41 -08004407 case SCHED_BATCH:
Ingo Molnardd41f592007-07-09 18:51:59 +02004408 case SCHED_IDLE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004409 ret = 0;
4410 break;
4411 }
4412 return ret;
4413}
4414
4415/**
4416 * sys_sched_get_priority_min - return minimum RT priority.
4417 * @policy: scheduling class.
4418 *
4419 * this syscall returns the minimum rt_priority that can be used
4420 * by a given scheduling class.
4421 */
Heiko Carstens5add95d2009-01-14 14:14:08 +01004422SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004423{
4424 int ret = -EINVAL;
4425
4426 switch (policy) {
4427 case SCHED_FIFO:
4428 case SCHED_RR:
4429 ret = 1;
4430 break;
4431 case SCHED_NORMAL:
Ingo Molnarb0a94992006-01-14 13:20:41 -08004432 case SCHED_BATCH:
Ingo Molnardd41f592007-07-09 18:51:59 +02004433 case SCHED_IDLE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004434 ret = 0;
4435 }
4436 return ret;
4437}
4438
4439/**
4440 * sys_sched_rr_get_interval - return the default timeslice of a process.
4441 * @pid: pid of the process.
4442 * @interval: userspace pointer to the timeslice value.
4443 *
4444 * this syscall writes the default timeslice value of a given process
4445 * into the user-space timespec buffer. A value of '0' means infinity.
4446 */
Heiko Carstens17da2bd2009-01-14 14:14:10 +01004447SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
Heiko Carstens754fe8d2009-01-14 14:14:09 +01004448 struct timespec __user *, interval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004449{
Ingo Molnar36c8b582006-07-03 00:25:41 -07004450 struct task_struct *p;
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02004451 unsigned int time_slice;
Thomas Gleixnerdba091b2009-12-09 09:32:03 +01004452 unsigned long flags;
4453 struct rq *rq;
Andi Kleen3a5c3592007-10-15 17:00:14 +02004454 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004455 struct timespec t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004456
4457 if (pid < 0)
Andi Kleen3a5c3592007-10-15 17:00:14 +02004458 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004459
4460 retval = -ESRCH;
Thomas Gleixner1a551ae2009-12-09 10:15:11 +00004461 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004462 p = find_process_by_pid(pid);
4463 if (!p)
4464 goto out_unlock;
4465
4466 retval = security_task_getscheduler(p);
4467 if (retval)
4468 goto out_unlock;
4469
Thomas Gleixnerdba091b2009-12-09 09:32:03 +01004470 rq = task_rq_lock(p, &flags);
4471 time_slice = p->sched_class->get_rr_interval(rq, p);
Peter Zijlstra0122ec52011-04-05 17:23:51 +02004472 task_rq_unlock(rq, p, &flags);
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02004473
Thomas Gleixner1a551ae2009-12-09 10:15:11 +00004474 rcu_read_unlock();
Dmitry Adamushkoa4ec24b2007-10-15 17:00:13 +02004475 jiffies_to_timespec(time_slice, &t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004476 retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004477 return retval;
Andi Kleen3a5c3592007-10-15 17:00:14 +02004478
Linus Torvalds1da177e2005-04-16 15:20:36 -07004479out_unlock:
Thomas Gleixner1a551ae2009-12-09 10:15:11 +00004480 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004481 return retval;
4482}
4483
Steven Rostedt7c731e02008-05-12 21:20:41 +02004484static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004485
Ingo Molnar82a1fcb2008-01-25 21:08:02 +01004486void sched_show_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004487{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004488 unsigned long free = 0;
Paul E. McKenney4e797522012-11-07 13:35:32 -08004489 int ppid;
Ingo Molnar36c8b582006-07-03 00:25:41 -07004490 unsigned state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004491
Linus Torvalds1da177e2005-04-16 15:20:36 -07004492 state = p->state ? __ffs(p->state) + 1 : 0;
Erik Gilling28d06862010-11-19 18:08:51 -08004493 printk(KERN_INFO "%-15.15s %c", p->comm,
Andreas Mohr2ed6e342006-07-10 04:43:52 -07004494 state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
Ingo Molnar4bd77322007-07-11 21:21:47 +02004495#if BITS_PER_LONG == 32
Linus Torvalds1da177e2005-04-16 15:20:36 -07004496 if (state == TASK_RUNNING)
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01004497 printk(KERN_CONT " running ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004498 else
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01004499 printk(KERN_CONT " %08lx ", thread_saved_pc(p));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004500#else
4501 if (state == TASK_RUNNING)
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01004502 printk(KERN_CONT " running task ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004503 else
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01004504 printk(KERN_CONT " %016lx ", thread_saved_pc(p));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004505#endif
4506#ifdef CONFIG_DEBUG_STACK_USAGE
Eric Sandeen7c9f8862008-04-22 16:38:23 -05004507 free = stack_not_used(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004508#endif
Paul E. McKenney4e797522012-11-07 13:35:32 -08004509 rcu_read_lock();
4510 ppid = task_pid_nr(rcu_dereference(p->real_parent));
4511 rcu_read_unlock();
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01004512 printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
Paul E. McKenney4e797522012-11-07 13:35:32 -08004513 task_pid_nr(p), ppid,
David Rientjesaa47b7e2009-05-04 01:38:05 -07004514 (unsigned long)task_thread_info(p)->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004515
Nick Piggin5fb5e6d2008-01-25 21:08:34 +01004516 show_stack(p, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004517}
4518
Ingo Molnare59e2ae2006-12-06 20:35:59 -08004519void show_state_filter(unsigned long state_filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004520{
Ingo Molnar36c8b582006-07-03 00:25:41 -07004521 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004522
Ingo Molnar4bd77322007-07-11 21:21:47 +02004523#if BITS_PER_LONG == 32
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01004524 printk(KERN_INFO
4525 " task PC stack pid father\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004526#else
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01004527 printk(KERN_INFO
4528 " task PC stack pid father\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004529#endif
Thomas Gleixner510f5ac2011-07-17 20:47:54 +02004530 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004531 do_each_thread(g, p) {
4532 /*
4533 * reset the NMI-timeout, listing all files on a slow
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004534 * console might take a lot of time:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004535 */
4536 touch_nmi_watchdog();
Ingo Molnar39bc89f2007-04-25 20:50:03 -07004537 if (!state_filter || (p->state & state_filter))
Ingo Molnar82a1fcb2008-01-25 21:08:02 +01004538 sched_show_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004539 } while_each_thread(g, p);
4540
Jeremy Fitzhardinge04c91672007-05-08 00:28:05 -07004541 touch_all_softlockup_watchdogs();
4542
Ingo Molnardd41f592007-07-09 18:51:59 +02004543#ifdef CONFIG_SCHED_DEBUG
4544 sysrq_sched_debug_show();
4545#endif
Thomas Gleixner510f5ac2011-07-17 20:47:54 +02004546 rcu_read_unlock();
Ingo Molnare59e2ae2006-12-06 20:35:59 -08004547 /*
4548 * Only show locks if all tasks are dumped:
4549 */
Shmulik Ladkani93335a22009-11-25 15:23:41 +02004550 if (!state_filter)
Ingo Molnare59e2ae2006-12-06 20:35:59 -08004551 debug_show_all_locks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004552}
4553
Ingo Molnar1df21052007-07-09 18:51:58 +02004554void __cpuinit init_idle_bootup_task(struct task_struct *idle)
4555{
Ingo Molnardd41f592007-07-09 18:51:59 +02004556 idle->sched_class = &idle_sched_class;
Ingo Molnar1df21052007-07-09 18:51:58 +02004557}
4558
Ingo Molnarf340c0d2005-06-28 16:40:42 +02004559/**
4560 * init_idle - set up an idle thread for a given CPU
4561 * @idle: task in question
4562 * @cpu: cpu the idle task belongs to
4563 *
4564 * NOTE: this function does not set the idle thread's NEED_RESCHED
4565 * flag, to make booting more robust.
4566 */
Nick Piggin5c1e1762006-10-03 01:14:04 -07004567void __cpuinit init_idle(struct task_struct *idle, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004568{
Ingo Molnar70b97a72006-07-03 00:25:42 -07004569 struct rq *rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004570 unsigned long flags;
4571
Thomas Gleixner05fa7852009-11-17 14:28:38 +01004572 raw_spin_lock_irqsave(&rq->lock, flags);
Ingo Molnar5cbd54e2008-11-12 20:05:50 +01004573
Ingo Molnardd41f592007-07-09 18:51:59 +02004574 __sched_fork(idle);
Peter Zijlstra06b83b52009-12-16 18:04:35 +01004575 idle->state = TASK_RUNNING;
Ingo Molnardd41f592007-07-09 18:51:59 +02004576 idle->se.exec_start = sched_clock();
4577
KOSAKI Motohiro1e1b6c52011-05-19 15:08:58 +09004578 do_set_cpus_allowed(idle, cpumask_of(cpu));
Peter Zijlstra6506cf6c2010-09-16 17:50:31 +02004579 /*
4580 * We're having a chicken and egg problem, even though we are
4581 * holding rq->lock, the cpu isn't yet set to this cpu so the
4582 * lockdep check in task_group() will fail.
4583 *
4584 * Similar case to sched_fork(). / Alternatively we could
4585 * use task_rq_lock() here and obtain the other rq->lock.
4586 *
4587 * Silence PROVE_RCU
4588 */
4589 rcu_read_lock();
Ingo Molnardd41f592007-07-09 18:51:59 +02004590 __set_task_cpu(idle, cpu);
Peter Zijlstra6506cf6c2010-09-16 17:50:31 +02004591 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004592
Linus Torvalds1da177e2005-04-16 15:20:36 -07004593 rq->curr = rq->idle = idle;
Peter Zijlstra3ca7a442011-04-05 17:23:40 +02004594#if defined(CONFIG_SMP)
4595 idle->on_cpu = 1;
Nick Piggin4866cde2005-06-25 14:57:23 -07004596#endif
Thomas Gleixner05fa7852009-11-17 14:28:38 +01004597 raw_spin_unlock_irqrestore(&rq->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004598
4599 /* Set the preempt count _outside_ the spinlocks! */
Al Viroa1261f52005-11-13 16:06:55 -08004600 task_thread_info(idle)->preempt_count = 0;
Jonathan Corbet625f2a32011-04-22 11:19:10 -06004601
Ingo Molnardd41f592007-07-09 18:51:59 +02004602 /*
4603 * The idle tasks have their own, simple scheduling class:
4604 */
4605 idle->sched_class = &idle_sched_class;
Steven Rostedt868baf02011-02-10 21:26:13 -05004606 ftrace_graph_init_idle_task(idle, cpu);
Carsten Emdef1c6f1a2011-10-26 23:14:16 +02004607#if defined(CONFIG_SMP)
4608 sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
4609#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004610}
4611
Linus Torvalds1da177e2005-04-16 15:20:36 -07004612#ifdef CONFIG_SMP
KOSAKI Motohiro1e1b6c52011-05-19 15:08:58 +09004613void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
4614{
4615 if (p->sched_class && p->sched_class->set_cpus_allowed)
4616 p->sched_class->set_cpus_allowed(p, new_mask);
Peter Zijlstra49396022011-06-25 15:45:46 +02004617
4618 cpumask_copy(&p->cpus_allowed, new_mask);
Peter Zijlstra29baa742012-04-23 12:11:21 +02004619 p->nr_cpus_allowed = cpumask_weight(new_mask);
KOSAKI Motohiro1e1b6c52011-05-19 15:08:58 +09004620}
4621
Linus Torvalds1da177e2005-04-16 15:20:36 -07004622/*
4623 * This is how migration works:
4624 *
Tejun Heo969c7922010-05-06 18:49:21 +02004625 * 1) we invoke migration_cpu_stop() on the target CPU using
4626 * stop_one_cpu().
4627 * 2) stopper starts to run (implicitly forcing the migrated thread
4628 * off the CPU)
4629 * 3) it checks whether the migrated task is still in the wrong runqueue.
4630 * 4) if it's in the wrong runqueue then the migration thread removes
Linus Torvalds1da177e2005-04-16 15:20:36 -07004631 * it and puts it into the right queue.
Tejun Heo969c7922010-05-06 18:49:21 +02004632 * 5) stopper completes and stop_one_cpu() returns and the migration
4633 * is done.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004634 */
4635
4636/*
4637 * Change a given task's CPU affinity. Migrate the thread to a
4638 * proper CPU and schedule it away if the CPU it's executing on
4639 * is removed from the allowed bitmask.
4640 *
4641 * NOTE: the caller must have a valid reference to the task, the
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01004642 * task must not exit() & deallocate itself prematurely. The
Linus Torvalds1da177e2005-04-16 15:20:36 -07004643 * call is not atomic; no spinlocks may be held.
4644 */
Rusty Russell96f874e2008-11-25 02:35:14 +10304645int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004646{
4647 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07004648 struct rq *rq;
Tejun Heo969c7922010-05-06 18:49:21 +02004649 unsigned int dest_cpu;
Ingo Molnar48f24c42006-07-03 00:25:40 -07004650 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004651
4652 rq = task_rq_lock(p, &flags);
Peter Zijlstrae2912002009-12-16 18:04:36 +01004653
Yong Zhangdb44fc02011-05-09 22:07:05 +08004654 if (cpumask_equal(&p->cpus_allowed, new_mask))
4655 goto out;
4656
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01004657 if (!cpumask_intersects(new_mask, cpu_active_mask)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004658 ret = -EINVAL;
4659 goto out;
4660 }
4661
Yong Zhangdb44fc02011-05-09 22:07:05 +08004662 if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) {
David Rientjes9985b0b2008-06-05 12:57:11 -07004663 ret = -EINVAL;
4664 goto out;
4665 }
4666
KOSAKI Motohiro1e1b6c52011-05-19 15:08:58 +09004667 do_set_cpus_allowed(p, new_mask);
Gregory Haskins73fe6aa2008-01-25 21:08:07 +01004668
Linus Torvalds1da177e2005-04-16 15:20:36 -07004669 /* Can the task run on the task's current CPU? If so, we're done */
Rusty Russell96f874e2008-11-25 02:35:14 +10304670 if (cpumask_test_cpu(task_cpu(p), new_mask))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004671 goto out;
4672
Tejun Heo969c7922010-05-06 18:49:21 +02004673 dest_cpu = cpumask_any_and(cpu_active_mask, new_mask);
Peter Zijlstrabd8e7dd2011-04-05 17:23:59 +02004674 if (p->on_rq) {
Tejun Heo969c7922010-05-06 18:49:21 +02004675 struct migration_arg arg = { p, dest_cpu };
Linus Torvalds1da177e2005-04-16 15:20:36 -07004676 /* Need help from migration thread: drop lock and wait. */
Peter Zijlstra0122ec52011-04-05 17:23:51 +02004677 task_rq_unlock(rq, p, &flags);
Tejun Heo969c7922010-05-06 18:49:21 +02004678 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004679 tlb_migrate_finish(p->mm);
4680 return 0;
4681 }
4682out:
Peter Zijlstra0122ec52011-04-05 17:23:51 +02004683 task_rq_unlock(rq, p, &flags);
Ingo Molnar48f24c42006-07-03 00:25:40 -07004684
Linus Torvalds1da177e2005-04-16 15:20:36 -07004685 return ret;
4686}
Mike Traviscd8ba7c2008-03-26 14:23:49 -07004687EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004688
4689/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01004690 * Move (not current) task off this cpu, onto dest cpu. We're doing
Linus Torvalds1da177e2005-04-16 15:20:36 -07004691 * this because either it can't run here any more (set_cpus_allowed()
4692 * away from this CPU, or CPU going down), or because we're
4693 * attempting to rebalance this task on exec (sched_exec).
4694 *
4695 * So we race with normal scheduler movements, but that's OK, as long
4696 * as the task is no longer on this CPU.
Kirill Korotaevefc30812006-06-27 02:54:32 -07004697 *
4698 * Returns non-zero if task was successfully migrated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004699 */
Kirill Korotaevefc30812006-06-27 02:54:32 -07004700static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004701{
Ingo Molnar70b97a72006-07-03 00:25:42 -07004702 struct rq *rq_dest, *rq_src;
Peter Zijlstrae2912002009-12-16 18:04:36 +01004703 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004704
Max Krasnyanskye761b772008-07-15 04:43:49 -07004705 if (unlikely(!cpu_active(dest_cpu)))
Kirill Korotaevefc30812006-06-27 02:54:32 -07004706 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004707
4708 rq_src = cpu_rq(src_cpu);
4709 rq_dest = cpu_rq(dest_cpu);
4710
Peter Zijlstra0122ec52011-04-05 17:23:51 +02004711 raw_spin_lock(&p->pi_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004712 double_rq_lock(rq_src, rq_dest);
4713 /* Already moved. */
4714 if (task_cpu(p) != src_cpu)
Linus Torvaldsb1e38732008-07-10 11:25:03 -07004715 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004716 /* Affinity changed (again). */
Peter Zijlstrafa17b502011-06-16 12:23:22 +02004717 if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p)))
Linus Torvaldsb1e38732008-07-10 11:25:03 -07004718 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004719
Peter Zijlstrae2912002009-12-16 18:04:36 +01004720 /*
4721 * If we're not on a rq, the next wake-up will ensure we're
4722 * placed properly.
4723 */
Peter Zijlstrafd2f4412011-04-05 17:23:44 +02004724 if (p->on_rq) {
Peter Zijlstra4ca9b722012-01-25 11:50:51 +01004725 dequeue_task(rq_src, p, 0);
Peter Zijlstrae2912002009-12-16 18:04:36 +01004726 set_task_cpu(p, dest_cpu);
Peter Zijlstra4ca9b722012-01-25 11:50:51 +01004727 enqueue_task(rq_dest, p, 0);
Peter Zijlstra15afe092008-09-20 23:38:02 +02004728 check_preempt_curr(rq_dest, p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004729 }
Linus Torvaldsb1e38732008-07-10 11:25:03 -07004730done:
Kirill Korotaevefc30812006-06-27 02:54:32 -07004731 ret = 1;
Linus Torvaldsb1e38732008-07-10 11:25:03 -07004732fail:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004733 double_rq_unlock(rq_src, rq_dest);
Peter Zijlstra0122ec52011-04-05 17:23:51 +02004734 raw_spin_unlock(&p->pi_lock);
Kirill Korotaevefc30812006-06-27 02:54:32 -07004735 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004736}
4737
4738/*
Tejun Heo969c7922010-05-06 18:49:21 +02004739 * migration_cpu_stop - this will be executed by a highprio stopper thread
4740 * and performs thread migration by bumping thread off CPU then
4741 * 'pushing' onto another runqueue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004742 */
Tejun Heo969c7922010-05-06 18:49:21 +02004743static int migration_cpu_stop(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004744{
Tejun Heo969c7922010-05-06 18:49:21 +02004745 struct migration_arg *arg = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004746
Tejun Heo969c7922010-05-06 18:49:21 +02004747 /*
4748 * The original target cpu might have gone down and we might
4749 * be on another cpu but it doesn't matter.
4750 */
4751 local_irq_disable();
4752 __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu);
4753 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004754 return 0;
4755}
4756
4757#ifdef CONFIG_HOTPLUG_CPU
Linus Torvalds1da177e2005-04-16 15:20:36 -07004758
Ingo Molnar48f24c42006-07-03 00:25:40 -07004759/*
4760 * Ensures that the idle task is using init_mm right before its cpu goes
Linus Torvalds1da177e2005-04-16 15:20:36 -07004761 * offline.
4762 */
4763void idle_task_exit(void)
4764{
4765 struct mm_struct *mm = current->active_mm;
4766
4767 BUG_ON(cpu_online(smp_processor_id()));
4768
4769 if (mm != &init_mm)
4770 switch_mm(mm, &init_mm, current);
4771 mmdrop(mm);
4772}
4773
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01004774/*
Peter Zijlstra5d180232012-08-20 11:26:57 +02004775 * Since this CPU is going 'away' for a while, fold any nr_active delta
4776 * we might have. Assumes we're called after migrate_tasks() so that the
4777 * nr_active count is stable.
4778 *
4779 * Also see the comment "Global load-average calculations".
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01004780 */
Peter Zijlstra5d180232012-08-20 11:26:57 +02004781static void calc_load_migrate(struct rq *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004782{
Peter Zijlstra5d180232012-08-20 11:26:57 +02004783 long delta = calc_load_fold_active(rq);
4784 if (delta)
4785 atomic_long_add(delta, &calc_load_tasks);
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02004786}
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01004787
4788/*
4789 * Migrate all tasks from the rq, sleeping tasks will be migrated by
4790 * try_to_wake_up()->select_task_rq().
4791 *
4792 * Called with rq->lock held even though we'er in stop_machine() and
4793 * there's no concurrency possible, we hold the required locks anyway
4794 * because of lock validation efforts.
4795 */
4796static void migrate_tasks(unsigned int dead_cpu)
4797{
4798 struct rq *rq = cpu_rq(dead_cpu);
4799 struct task_struct *next, *stop = rq->stop;
4800 int dest_cpu;
4801
4802 /*
4803 * Fudge the rq selection such that the below task selection loop
4804 * doesn't get stuck on the currently eligible stop task.
4805 *
4806 * We're currently inside stop_machine() and the rq is either stuck
4807 * in the stop_machine_cpu_stop() loop, or we're executing this code,
4808 * either way we should never end up calling schedule() until we're
4809 * done here.
4810 */
4811 rq->stop = NULL;
4812
4813 for ( ; ; ) {
4814 /*
4815 * There's this thread running, bail when that's the only
4816 * remaining thread.
4817 */
4818 if (rq->nr_running == 1)
4819 break;
4820
4821 next = pick_next_task(rq);
4822 BUG_ON(!next);
4823 next->sched_class->put_prev_task(rq, next);
4824
4825 /* Find suitable destination for @next, with force if needed. */
4826 dest_cpu = select_fallback_rq(dead_cpu, next);
4827 raw_spin_unlock(&rq->lock);
4828
4829 __migrate_task(next, dead_cpu, dest_cpu);
4830
4831 raw_spin_lock(&rq->lock);
4832 }
4833
4834 rq->stop = stop;
4835}
4836
Linus Torvalds1da177e2005-04-16 15:20:36 -07004837#endif /* CONFIG_HOTPLUG_CPU */
4838
Nick Piggine692ab52007-07-26 13:40:43 +02004839#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
4840
4841static struct ctl_table sd_ctl_dir[] = {
Alexey Dobriyane0361852007-08-09 11:16:46 +02004842 {
4843 .procname = "sched_domain",
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02004844 .mode = 0555,
Alexey Dobriyane0361852007-08-09 11:16:46 +02004845 },
Eric W. Biederman56992302009-11-05 15:38:40 -08004846 {}
Nick Piggine692ab52007-07-26 13:40:43 +02004847};
4848
4849static struct ctl_table sd_ctl_root[] = {
Alexey Dobriyane0361852007-08-09 11:16:46 +02004850 {
4851 .procname = "kernel",
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02004852 .mode = 0555,
Alexey Dobriyane0361852007-08-09 11:16:46 +02004853 .child = sd_ctl_dir,
4854 },
Eric W. Biederman56992302009-11-05 15:38:40 -08004855 {}
Nick Piggine692ab52007-07-26 13:40:43 +02004856};
4857
4858static struct ctl_table *sd_alloc_ctl_entry(int n)
4859{
4860 struct ctl_table *entry =
Milton Miller5cf9f062007-10-15 17:00:19 +02004861 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
Nick Piggine692ab52007-07-26 13:40:43 +02004862
Nick Piggine692ab52007-07-26 13:40:43 +02004863 return entry;
4864}
4865
Milton Miller6382bc92007-10-15 17:00:19 +02004866static void sd_free_ctl_entry(struct ctl_table **tablep)
4867{
Milton Millercd7900762007-10-17 16:55:11 +02004868 struct ctl_table *entry;
Milton Miller6382bc92007-10-15 17:00:19 +02004869
Milton Millercd7900762007-10-17 16:55:11 +02004870 /*
4871 * In the intermediate directories, both the child directory and
4872 * procname are dynamically allocated and could fail but the mode
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01004873 * will always be set. In the lowest directory the names are
Milton Millercd7900762007-10-17 16:55:11 +02004874 * static strings and all have proc handlers.
4875 */
4876 for (entry = *tablep; entry->mode; entry++) {
Milton Miller6382bc92007-10-15 17:00:19 +02004877 if (entry->child)
4878 sd_free_ctl_entry(&entry->child);
Milton Millercd7900762007-10-17 16:55:11 +02004879 if (entry->proc_handler == NULL)
4880 kfree(entry->procname);
4881 }
Milton Miller6382bc92007-10-15 17:00:19 +02004882
4883 kfree(*tablep);
4884 *tablep = NULL;
4885}
4886
Namhyung Kim201c3732012-08-16 17:03:24 +09004887static int min_load_idx = 0;
4888static int max_load_idx = CPU_LOAD_IDX_MAX;
4889
Nick Piggine692ab52007-07-26 13:40:43 +02004890static void
Alexey Dobriyane0361852007-08-09 11:16:46 +02004891set_table_entry(struct ctl_table *entry,
Nick Piggine692ab52007-07-26 13:40:43 +02004892 const char *procname, void *data, int maxlen,
Namhyung Kim201c3732012-08-16 17:03:24 +09004893 umode_t mode, proc_handler *proc_handler,
4894 bool load_idx)
Nick Piggine692ab52007-07-26 13:40:43 +02004895{
Nick Piggine692ab52007-07-26 13:40:43 +02004896 entry->procname = procname;
4897 entry->data = data;
4898 entry->maxlen = maxlen;
4899 entry->mode = mode;
4900 entry->proc_handler = proc_handler;
Namhyung Kim201c3732012-08-16 17:03:24 +09004901
4902 if (load_idx) {
4903 entry->extra1 = &min_load_idx;
4904 entry->extra2 = &max_load_idx;
4905 }
Nick Piggine692ab52007-07-26 13:40:43 +02004906}
4907
4908static struct ctl_table *
4909sd_alloc_ctl_domain_table(struct sched_domain *sd)
4910{
Ingo Molnara5d8c342008-10-09 11:35:51 +02004911 struct ctl_table *table = sd_alloc_ctl_entry(13);
Nick Piggine692ab52007-07-26 13:40:43 +02004912
Milton Millerad1cdc12007-10-15 17:00:19 +02004913 if (table == NULL)
4914 return NULL;
4915
Alexey Dobriyane0361852007-08-09 11:16:46 +02004916 set_table_entry(&table[0], "min_interval", &sd->min_interval,
Namhyung Kim201c3732012-08-16 17:03:24 +09004917 sizeof(long), 0644, proc_doulongvec_minmax, false);
Alexey Dobriyane0361852007-08-09 11:16:46 +02004918 set_table_entry(&table[1], "max_interval", &sd->max_interval,
Namhyung Kim201c3732012-08-16 17:03:24 +09004919 sizeof(long), 0644, proc_doulongvec_minmax, false);
Alexey Dobriyane0361852007-08-09 11:16:46 +02004920 set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
Namhyung Kim201c3732012-08-16 17:03:24 +09004921 sizeof(int), 0644, proc_dointvec_minmax, true);
Alexey Dobriyane0361852007-08-09 11:16:46 +02004922 set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
Namhyung Kim201c3732012-08-16 17:03:24 +09004923 sizeof(int), 0644, proc_dointvec_minmax, true);
Alexey Dobriyane0361852007-08-09 11:16:46 +02004924 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
Namhyung Kim201c3732012-08-16 17:03:24 +09004925 sizeof(int), 0644, proc_dointvec_minmax, true);
Alexey Dobriyane0361852007-08-09 11:16:46 +02004926 set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
Namhyung Kim201c3732012-08-16 17:03:24 +09004927 sizeof(int), 0644, proc_dointvec_minmax, true);
Alexey Dobriyane0361852007-08-09 11:16:46 +02004928 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
Namhyung Kim201c3732012-08-16 17:03:24 +09004929 sizeof(int), 0644, proc_dointvec_minmax, true);
Alexey Dobriyane0361852007-08-09 11:16:46 +02004930 set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
Namhyung Kim201c3732012-08-16 17:03:24 +09004931 sizeof(int), 0644, proc_dointvec_minmax, false);
Alexey Dobriyane0361852007-08-09 11:16:46 +02004932 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
Namhyung Kim201c3732012-08-16 17:03:24 +09004933 sizeof(int), 0644, proc_dointvec_minmax, false);
Zou Nan haiace8b3d2007-10-15 17:00:14 +02004934 set_table_entry(&table[9], "cache_nice_tries",
Nick Piggine692ab52007-07-26 13:40:43 +02004935 &sd->cache_nice_tries,
Namhyung Kim201c3732012-08-16 17:03:24 +09004936 sizeof(int), 0644, proc_dointvec_minmax, false);
Zou Nan haiace8b3d2007-10-15 17:00:14 +02004937 set_table_entry(&table[10], "flags", &sd->flags,
Namhyung Kim201c3732012-08-16 17:03:24 +09004938 sizeof(int), 0644, proc_dointvec_minmax, false);
Ingo Molnara5d8c342008-10-09 11:35:51 +02004939 set_table_entry(&table[11], "name", sd->name,
Namhyung Kim201c3732012-08-16 17:03:24 +09004940 CORENAME_MAX_SIZE, 0444, proc_dostring, false);
Ingo Molnara5d8c342008-10-09 11:35:51 +02004941 /* &table[12] is terminator */
Nick Piggine692ab52007-07-26 13:40:43 +02004942
4943 return table;
4944}
4945
Ingo Molnar9a4e7152007-11-28 15:52:56 +01004946static ctl_table *sd_alloc_ctl_cpu_table(int cpu)
Nick Piggine692ab52007-07-26 13:40:43 +02004947{
4948 struct ctl_table *entry, *table;
4949 struct sched_domain *sd;
4950 int domain_num = 0, i;
4951 char buf[32];
4952
4953 for_each_domain(cpu, sd)
4954 domain_num++;
4955 entry = table = sd_alloc_ctl_entry(domain_num + 1);
Milton Millerad1cdc12007-10-15 17:00:19 +02004956 if (table == NULL)
4957 return NULL;
Nick Piggine692ab52007-07-26 13:40:43 +02004958
4959 i = 0;
4960 for_each_domain(cpu, sd) {
4961 snprintf(buf, 32, "domain%d", i);
Nick Piggine692ab52007-07-26 13:40:43 +02004962 entry->procname = kstrdup(buf, GFP_KERNEL);
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02004963 entry->mode = 0555;
Nick Piggine692ab52007-07-26 13:40:43 +02004964 entry->child = sd_alloc_ctl_domain_table(sd);
4965 entry++;
4966 i++;
4967 }
4968 return table;
4969}
4970
4971static struct ctl_table_header *sd_sysctl_header;
Milton Miller6382bc92007-10-15 17:00:19 +02004972static void register_sched_domain_sysctl(void)
Nick Piggine692ab52007-07-26 13:40:43 +02004973{
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01004974 int i, cpu_num = num_possible_cpus();
Nick Piggine692ab52007-07-26 13:40:43 +02004975 struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
4976 char buf[32];
4977
Milton Miller73785472007-10-24 18:23:48 +02004978 WARN_ON(sd_ctl_dir[0].child);
4979 sd_ctl_dir[0].child = entry;
4980
Milton Millerad1cdc12007-10-15 17:00:19 +02004981 if (entry == NULL)
4982 return;
4983
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01004984 for_each_possible_cpu(i) {
Nick Piggine692ab52007-07-26 13:40:43 +02004985 snprintf(buf, 32, "cpu%d", i);
Nick Piggine692ab52007-07-26 13:40:43 +02004986 entry->procname = kstrdup(buf, GFP_KERNEL);
Eric W. Biedermanc57baf12007-08-23 15:18:02 +02004987 entry->mode = 0555;
Nick Piggine692ab52007-07-26 13:40:43 +02004988 entry->child = sd_alloc_ctl_cpu_table(i);
Milton Miller97b6ea72007-10-15 17:00:19 +02004989 entry++;
Nick Piggine692ab52007-07-26 13:40:43 +02004990 }
Milton Miller73785472007-10-24 18:23:48 +02004991
4992 WARN_ON(sd_sysctl_header);
Nick Piggine692ab52007-07-26 13:40:43 +02004993 sd_sysctl_header = register_sysctl_table(sd_ctl_root);
4994}
Milton Miller6382bc92007-10-15 17:00:19 +02004995
Milton Miller73785472007-10-24 18:23:48 +02004996/* may be called multiple times per register */
Milton Miller6382bc92007-10-15 17:00:19 +02004997static void unregister_sched_domain_sysctl(void)
4998{
Milton Miller73785472007-10-24 18:23:48 +02004999 if (sd_sysctl_header)
5000 unregister_sysctl_table(sd_sysctl_header);
Milton Miller6382bc92007-10-15 17:00:19 +02005001 sd_sysctl_header = NULL;
Milton Miller73785472007-10-24 18:23:48 +02005002 if (sd_ctl_dir[0].child)
5003 sd_free_ctl_entry(&sd_ctl_dir[0].child);
Milton Miller6382bc92007-10-15 17:00:19 +02005004}
Nick Piggine692ab52007-07-26 13:40:43 +02005005#else
Milton Miller6382bc92007-10-15 17:00:19 +02005006static void register_sched_domain_sysctl(void)
5007{
5008}
5009static void unregister_sched_domain_sysctl(void)
Nick Piggine692ab52007-07-26 13:40:43 +02005010{
5011}
5012#endif
5013
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005014static void set_rq_online(struct rq *rq)
5015{
5016 if (!rq->online) {
5017 const struct sched_class *class;
5018
Rusty Russellc6c49272008-11-25 02:35:05 +10305019 cpumask_set_cpu(rq->cpu, rq->rd->online);
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005020 rq->online = 1;
5021
5022 for_each_class(class) {
5023 if (class->rq_online)
5024 class->rq_online(rq);
5025 }
5026 }
5027}
5028
5029static void set_rq_offline(struct rq *rq)
5030{
5031 if (rq->online) {
5032 const struct sched_class *class;
5033
5034 for_each_class(class) {
5035 if (class->rq_offline)
5036 class->rq_offline(rq);
5037 }
5038
Rusty Russellc6c49272008-11-25 02:35:05 +10305039 cpumask_clear_cpu(rq->cpu, rq->rd->online);
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005040 rq->online = 0;
5041 }
5042}
5043
Linus Torvalds1da177e2005-04-16 15:20:36 -07005044/*
5045 * migration_call - callback that gets triggered when a CPU is added.
5046 * Here we can start up the necessary migration thread for the new CPU.
5047 */
Ingo Molnar48f24c42006-07-03 00:25:40 -07005048static int __cpuinit
5049migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005050{
Ingo Molnar48f24c42006-07-03 00:25:40 -07005051 int cpu = (long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005052 unsigned long flags;
Tejun Heo969c7922010-05-06 18:49:21 +02005053 struct rq *rq = cpu_rq(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005054
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01005055 switch (action & ~CPU_TASKS_FROZEN) {
Gautham R Shenoy5be93612007-05-09 02:34:04 -07005056
Linus Torvalds1da177e2005-04-16 15:20:36 -07005057 case CPU_UP_PREPARE:
Thomas Gleixnera468d382009-07-17 14:15:46 +02005058 rq->calc_load_update = calc_load_update;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005059 break;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005060
Linus Torvalds1da177e2005-04-16 15:20:36 -07005061 case CPU_ONLINE:
Gregory Haskins1f94ef52008-03-10 16:52:41 -04005062 /* Update our root-domain */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005063 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins1f94ef52008-03-10 16:52:41 -04005064 if (rq->rd) {
Rusty Russellc6c49272008-11-25 02:35:05 +10305065 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005066
5067 set_rq_online(rq);
Gregory Haskins1f94ef52008-03-10 16:52:41 -04005068 }
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005069 raw_spin_unlock_irqrestore(&rq->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005070 break;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005071
Linus Torvalds1da177e2005-04-16 15:20:36 -07005072#ifdef CONFIG_HOTPLUG_CPU
Gregory Haskins08f503b2008-03-10 17:59:11 -04005073 case CPU_DYING:
Peter Zijlstra317f3942011-04-05 17:23:58 +02005074 sched_ttwu_pending();
Gregory Haskins57d885f2008-01-25 21:08:18 +01005075 /* Update our root-domain */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005076 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005077 if (rq->rd) {
Rusty Russellc6c49272008-11-25 02:35:05 +10305078 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005079 set_rq_offline(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005080 }
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01005081 migrate_tasks(cpu);
5082 BUG_ON(rq->nr_running != 1); /* the migration thread */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005083 raw_spin_unlock_irqrestore(&rq->lock, flags);
Peter Zijlstra5d180232012-08-20 11:26:57 +02005084 break;
Peter Zijlstra48c5cca2010-11-13 19:32:29 +01005085
Peter Zijlstra5d180232012-08-20 11:26:57 +02005086 case CPU_DEAD:
Peter Zijlstraf319da02012-08-20 11:26:57 +02005087 calc_load_migrate(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005088 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005089#endif
5090 }
Peter Zijlstra49c022e2011-04-05 10:14:25 +02005091
5092 update_max_interval();
5093
Linus Torvalds1da177e2005-04-16 15:20:36 -07005094 return NOTIFY_OK;
5095}
5096
Paul Mackerrasf38b0822009-06-02 21:05:16 +10005097/*
5098 * Register at high priority so that task migration (migrate_all_tasks)
5099 * happens before everything else. This has to be lower priority than
Ingo Molnarcdd6c482009-09-21 12:02:48 +02005100 * the notifier in the perf_event subsystem, though.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005101 */
Chandra Seetharaman26c21432006-06-27 02:54:10 -07005102static struct notifier_block __cpuinitdata migration_notifier = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005103 .notifier_call = migration_call,
Tejun Heo50a323b2010-06-08 21:40:36 +02005104 .priority = CPU_PRI_MIGRATION,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005105};
5106
Tejun Heo3a101d02010-06-08 21:40:36 +02005107static int __cpuinit sched_cpu_active(struct notifier_block *nfb,
5108 unsigned long action, void *hcpu)
5109{
5110 switch (action & ~CPU_TASKS_FROZEN) {
Peter Zijlstra5fbd0362011-12-15 17:09:22 +01005111 case CPU_STARTING:
Tejun Heo3a101d02010-06-08 21:40:36 +02005112 case CPU_DOWN_FAILED:
5113 set_cpu_active((long)hcpu, true);
5114 return NOTIFY_OK;
5115 default:
5116 return NOTIFY_DONE;
5117 }
5118}
5119
5120static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb,
5121 unsigned long action, void *hcpu)
5122{
5123 switch (action & ~CPU_TASKS_FROZEN) {
5124 case CPU_DOWN_PREPARE:
5125 set_cpu_active((long)hcpu, false);
5126 return NOTIFY_OK;
5127 default:
5128 return NOTIFY_DONE;
5129 }
5130}
5131
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -07005132static int __init migration_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005133{
5134 void *cpu = (void *)(long)smp_processor_id();
Akinobu Mita07dccf32006-09-29 02:00:22 -07005135 int err;
Ingo Molnar48f24c42006-07-03 00:25:40 -07005136
Tejun Heo3a101d02010-06-08 21:40:36 +02005137 /* Initialize migration for the boot CPU */
Akinobu Mita07dccf32006-09-29 02:00:22 -07005138 err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
5139 BUG_ON(err == NOTIFY_BAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005140 migration_call(&migration_notifier, CPU_ONLINE, cpu);
5141 register_cpu_notifier(&migration_notifier);
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -07005142
Tejun Heo3a101d02010-06-08 21:40:36 +02005143 /* Register cpu active notifiers */
5144 cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE);
5145 cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE);
5146
Thomas Gleixnera004cd42009-07-21 09:54:05 +02005147 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005148}
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -07005149early_initcall(migration_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005150#endif
5151
5152#ifdef CONFIG_SMP
Christoph Lameter476f3532007-05-06 14:48:58 -07005153
Peter Zijlstra4cb98832011-04-07 14:09:58 +02005154static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */
5155
Ingo Molnar3e9830d2007-10-15 17:00:13 +02005156#ifdef CONFIG_SCHED_DEBUG
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005157
Peter Zijlstrad039ac62012-05-31 21:20:16 +02005158static __read_mostly int sched_debug_enabled;
Mike Travisf6630112009-11-17 18:22:15 -06005159
Peter Zijlstrad039ac62012-05-31 21:20:16 +02005160static int __init sched_debug_setup(char *str)
Mike Travisf6630112009-11-17 18:22:15 -06005161{
Peter Zijlstrad039ac62012-05-31 21:20:16 +02005162 sched_debug_enabled = 1;
Mike Travisf6630112009-11-17 18:22:15 -06005163
5164 return 0;
5165}
Peter Zijlstrad039ac62012-05-31 21:20:16 +02005166early_param("sched_debug", sched_debug_setup);
5167
5168static inline bool sched_debug(void)
5169{
5170 return sched_debug_enabled;
5171}
Mike Travisf6630112009-11-17 18:22:15 -06005172
Mike Travis7c16ec52008-04-04 18:11:11 -07005173static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
Rusty Russell96f874e2008-11-25 02:35:14 +10305174 struct cpumask *groupmask)
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005175{
5176 struct sched_group *group = sd->groups;
Mike Travis434d53b2008-04-04 18:11:04 -07005177 char str[256];
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005178
Rusty Russell968ea6d2008-12-13 21:55:51 +10305179 cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd));
Rusty Russell96f874e2008-11-25 02:35:14 +10305180 cpumask_clear(groupmask);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005181
5182 printk(KERN_DEBUG "%*s domain %d: ", level, "", level);
5183
5184 if (!(sd->flags & SD_LOAD_BALANCE)) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005185 printk("does not load-balance\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005186 if (sd->parent)
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005187 printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
5188 " has parent");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005189 return -1;
5190 }
5191
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005192 printk(KERN_CONT "span %s level %s\n", str, sd->name);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005193
Rusty Russell758b2cd2008-11-25 02:35:04 +10305194 if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005195 printk(KERN_ERR "ERROR: domain->span does not contain "
5196 "CPU%d\n", cpu);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005197 }
Rusty Russell758b2cd2008-11-25 02:35:04 +10305198 if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005199 printk(KERN_ERR "ERROR: domain->groups does not contain"
5200 " CPU%d\n", cpu);
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005201 }
5202
5203 printk(KERN_DEBUG "%*s groups:", level + 1, "");
5204 do {
5205 if (!group) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005206 printk("\n");
5207 printk(KERN_ERR "ERROR: group is NULL\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005208 break;
5209 }
5210
Peter Zijlstrac3decf02012-05-31 12:05:32 +02005211 /*
5212 * Even though we initialize ->power to something semi-sane,
5213 * we leave power_orig unset. This allows us to detect if
5214 * domain iteration is still funny without causing /0 traps.
5215 */
5216 if (!group->sgp->power_orig) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005217 printk(KERN_CONT "\n");
5218 printk(KERN_ERR "ERROR: domain->cpu_power not "
5219 "set\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005220 break;
5221 }
5222
Rusty Russell758b2cd2008-11-25 02:35:04 +10305223 if (!cpumask_weight(sched_group_cpus(group))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005224 printk(KERN_CONT "\n");
5225 printk(KERN_ERR "ERROR: empty group\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005226 break;
5227 }
5228
Peter Zijlstracb83b622012-04-17 15:49:36 +02005229 if (!(sd->flags & SD_OVERLAP) &&
5230 cpumask_intersects(groupmask, sched_group_cpus(group))) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005231 printk(KERN_CONT "\n");
5232 printk(KERN_ERR "ERROR: repeated CPUs\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005233 break;
5234 }
5235
Rusty Russell758b2cd2008-11-25 02:35:04 +10305236 cpumask_or(groupmask, groupmask, sched_group_cpus(group));
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005237
Rusty Russell968ea6d2008-12-13 21:55:51 +10305238 cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
Gautham R Shenoy381512c2009-04-14 09:09:36 +05305239
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005240 printk(KERN_CONT " %s", str);
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02005241 if (group->sgp->power != SCHED_POWER_SCALE) {
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005242 printk(KERN_CONT " (cpu_power = %d)",
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02005243 group->sgp->power);
Gautham R Shenoy381512c2009-04-14 09:09:36 +05305244 }
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005245
5246 group = group->next;
5247 } while (group != sd->groups);
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005248 printk(KERN_CONT "\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005249
Rusty Russell758b2cd2008-11-25 02:35:04 +10305250 if (!cpumask_equal(sched_domain_span(sd), groupmask))
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005251 printk(KERN_ERR "ERROR: groups don't span domain->span\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005252
Rusty Russell758b2cd2008-11-25 02:35:04 +10305253 if (sd->parent &&
5254 !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01005255 printk(KERN_ERR "ERROR: parent span is not a superset "
5256 "of domain->span\n");
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005257 return 0;
5258}
5259
Linus Torvalds1da177e2005-04-16 15:20:36 -07005260static void sched_domain_debug(struct sched_domain *sd, int cpu)
5261{
5262 int level = 0;
5263
Peter Zijlstrad039ac62012-05-31 21:20:16 +02005264 if (!sched_debug_enabled)
Mike Travisf6630112009-11-17 18:22:15 -06005265 return;
5266
Nick Piggin41c7ce92005-06-25 14:57:24 -07005267 if (!sd) {
5268 printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
5269 return;
5270 }
5271
Linus Torvalds1da177e2005-04-16 15:20:36 -07005272 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
5273
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005274 for (;;) {
Peter Zijlstra4cb98832011-04-07 14:09:58 +02005275 if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005276 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005277 level++;
5278 sd = sd->parent;
Miguel Ojeda Sandonis33859f72006-12-10 02:20:38 -08005279 if (!sd)
Ingo Molnar4dcf6af2007-10-24 18:23:48 +02005280 break;
5281 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005282}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02005283#else /* !CONFIG_SCHED_DEBUG */
Ingo Molnar48f24c42006-07-03 00:25:40 -07005284# define sched_domain_debug(sd, cpu) do { } while (0)
Peter Zijlstrad039ac62012-05-31 21:20:16 +02005285static inline bool sched_debug(void)
5286{
5287 return false;
5288}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02005289#endif /* CONFIG_SCHED_DEBUG */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005290
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07005291static int sd_degenerate(struct sched_domain *sd)
Suresh Siddha245af2c2005-06-25 14:57:25 -07005292{
Rusty Russell758b2cd2008-11-25 02:35:04 +10305293 if (cpumask_weight(sched_domain_span(sd)) == 1)
Suresh Siddha245af2c2005-06-25 14:57:25 -07005294 return 1;
5295
5296 /* Following flags need at least 2 groups */
5297 if (sd->flags & (SD_LOAD_BALANCE |
5298 SD_BALANCE_NEWIDLE |
5299 SD_BALANCE_FORK |
Siddha, Suresh B89c47102006-10-03 01:14:09 -07005300 SD_BALANCE_EXEC |
5301 SD_SHARE_CPUPOWER |
5302 SD_SHARE_PKG_RESOURCES)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07005303 if (sd->groups != sd->groups->next)
5304 return 0;
5305 }
5306
5307 /* Following flags don't use groups */
Peter Zijlstrac88d5912009-09-10 13:50:02 +02005308 if (sd->flags & (SD_WAKE_AFFINE))
Suresh Siddha245af2c2005-06-25 14:57:25 -07005309 return 0;
5310
5311 return 1;
5312}
5313
Ingo Molnar48f24c42006-07-03 00:25:40 -07005314static int
5315sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
Suresh Siddha245af2c2005-06-25 14:57:25 -07005316{
5317 unsigned long cflags = sd->flags, pflags = parent->flags;
5318
5319 if (sd_degenerate(parent))
5320 return 1;
5321
Rusty Russell758b2cd2008-11-25 02:35:04 +10305322 if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
Suresh Siddha245af2c2005-06-25 14:57:25 -07005323 return 0;
5324
Suresh Siddha245af2c2005-06-25 14:57:25 -07005325 /* Flags needing groups don't count if only 1 group in parent */
5326 if (parent->groups == parent->groups->next) {
5327 pflags &= ~(SD_LOAD_BALANCE |
5328 SD_BALANCE_NEWIDLE |
5329 SD_BALANCE_FORK |
Siddha, Suresh B89c47102006-10-03 01:14:09 -07005330 SD_BALANCE_EXEC |
5331 SD_SHARE_CPUPOWER |
5332 SD_SHARE_PKG_RESOURCES);
Ken Chen54364992008-12-07 18:47:37 -08005333 if (nr_node_ids == 1)
5334 pflags &= ~SD_SERIALIZE;
Suresh Siddha245af2c2005-06-25 14:57:25 -07005335 }
5336 if (~cflags & pflags)
5337 return 0;
5338
5339 return 1;
5340}
5341
Peter Zijlstradce840a2011-04-07 14:09:50 +02005342static void free_rootdomain(struct rcu_head *rcu)
Rusty Russellc6c49272008-11-25 02:35:05 +10305343{
Peter Zijlstradce840a2011-04-07 14:09:50 +02005344 struct root_domain *rd = container_of(rcu, struct root_domain, rcu);
Peter Zijlstra047106a2009-11-16 10:28:09 +01005345
Rusty Russell68e74562008-11-25 02:35:13 +10305346 cpupri_cleanup(&rd->cpupri);
Rusty Russellc6c49272008-11-25 02:35:05 +10305347 free_cpumask_var(rd->rto_mask);
5348 free_cpumask_var(rd->online);
5349 free_cpumask_var(rd->span);
5350 kfree(rd);
5351}
5352
Gregory Haskins57d885f2008-01-25 21:08:18 +01005353static void rq_attach_root(struct rq *rq, struct root_domain *rd)
5354{
Ingo Molnara0490fa2009-02-12 11:35:40 +01005355 struct root_domain *old_rd = NULL;
Gregory Haskins57d885f2008-01-25 21:08:18 +01005356 unsigned long flags;
Gregory Haskins57d885f2008-01-25 21:08:18 +01005357
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005358 raw_spin_lock_irqsave(&rq->lock, flags);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005359
5360 if (rq->rd) {
Ingo Molnara0490fa2009-02-12 11:35:40 +01005361 old_rd = rq->rd;
Gregory Haskins57d885f2008-01-25 21:08:18 +01005362
Rusty Russellc6c49272008-11-25 02:35:05 +10305363 if (cpumask_test_cpu(rq->cpu, old_rd->online))
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005364 set_rq_offline(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005365
Rusty Russellc6c49272008-11-25 02:35:05 +10305366 cpumask_clear_cpu(rq->cpu, old_rd->span);
Gregory Haskinsdc938522008-01-25 21:08:26 +01005367
Ingo Molnara0490fa2009-02-12 11:35:40 +01005368 /*
5369 * If we dont want to free the old_rt yet then
5370 * set old_rd to NULL to skip the freeing later
5371 * in this function:
5372 */
5373 if (!atomic_dec_and_test(&old_rd->refcount))
5374 old_rd = NULL;
Gregory Haskins57d885f2008-01-25 21:08:18 +01005375 }
5376
5377 atomic_inc(&rd->refcount);
5378 rq->rd = rd;
5379
Rusty Russellc6c49272008-11-25 02:35:05 +10305380 cpumask_set_cpu(rq->cpu, rd->span);
Gregory Haskins00aec932009-07-30 10:57:23 -04005381 if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
Gregory Haskins1f11eb62008-06-04 15:04:05 -04005382 set_rq_online(rq);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005383
Thomas Gleixner05fa7852009-11-17 14:28:38 +01005384 raw_spin_unlock_irqrestore(&rq->lock, flags);
Ingo Molnara0490fa2009-02-12 11:35:40 +01005385
5386 if (old_rd)
Peter Zijlstradce840a2011-04-07 14:09:50 +02005387 call_rcu_sched(&old_rd->rcu, free_rootdomain);
Gregory Haskins57d885f2008-01-25 21:08:18 +01005388}
5389
Pekka Enberg68c38fc2010-07-15 23:18:22 +03005390static int init_rootdomain(struct root_domain *rd)
Gregory Haskins57d885f2008-01-25 21:08:18 +01005391{
5392 memset(rd, 0, sizeof(*rd));
5393
Pekka Enberg68c38fc2010-07-15 23:18:22 +03005394 if (!alloc_cpumask_var(&rd->span, GFP_KERNEL))
Li Zefan0c910d22009-01-06 17:39:06 +08005395 goto out;
Pekka Enberg68c38fc2010-07-15 23:18:22 +03005396 if (!alloc_cpumask_var(&rd->online, GFP_KERNEL))
Rusty Russellc6c49272008-11-25 02:35:05 +10305397 goto free_span;
Pekka Enberg68c38fc2010-07-15 23:18:22 +03005398 if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
Rusty Russellc6c49272008-11-25 02:35:05 +10305399 goto free_online;
Gregory Haskins6e0534f2008-05-12 21:21:01 +02005400
Pekka Enberg68c38fc2010-07-15 23:18:22 +03005401 if (cpupri_init(&rd->cpupri) != 0)
Rusty Russell68e74562008-11-25 02:35:13 +10305402 goto free_rto_mask;
Rusty Russellc6c49272008-11-25 02:35:05 +10305403 return 0;
5404
Rusty Russell68e74562008-11-25 02:35:13 +10305405free_rto_mask:
5406 free_cpumask_var(rd->rto_mask);
Rusty Russellc6c49272008-11-25 02:35:05 +10305407free_online:
5408 free_cpumask_var(rd->online);
5409free_span:
5410 free_cpumask_var(rd->span);
Li Zefan0c910d22009-01-06 17:39:06 +08005411out:
Rusty Russellc6c49272008-11-25 02:35:05 +10305412 return -ENOMEM;
Gregory Haskins57d885f2008-01-25 21:08:18 +01005413}
5414
Peter Zijlstra029632f2011-10-25 10:00:11 +02005415/*
5416 * By default the system creates a single root-domain with all cpus as
5417 * members (mimicking the global state we have today).
5418 */
5419struct root_domain def_root_domain;
5420
Gregory Haskins57d885f2008-01-25 21:08:18 +01005421static void init_defrootdomain(void)
5422{
Pekka Enberg68c38fc2010-07-15 23:18:22 +03005423 init_rootdomain(&def_root_domain);
Rusty Russellc6c49272008-11-25 02:35:05 +10305424
Gregory Haskins57d885f2008-01-25 21:08:18 +01005425 atomic_set(&def_root_domain.refcount, 1);
5426}
5427
Gregory Haskinsdc938522008-01-25 21:08:26 +01005428static struct root_domain *alloc_rootdomain(void)
Gregory Haskins57d885f2008-01-25 21:08:18 +01005429{
5430 struct root_domain *rd;
5431
5432 rd = kmalloc(sizeof(*rd), GFP_KERNEL);
5433 if (!rd)
5434 return NULL;
5435
Pekka Enberg68c38fc2010-07-15 23:18:22 +03005436 if (init_rootdomain(rd) != 0) {
Rusty Russellc6c49272008-11-25 02:35:05 +10305437 kfree(rd);
5438 return NULL;
5439 }
Gregory Haskins57d885f2008-01-25 21:08:18 +01005440
5441 return rd;
5442}
5443
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005444static void free_sched_groups(struct sched_group *sg, int free_sgp)
5445{
5446 struct sched_group *tmp, *first;
5447
5448 if (!sg)
5449 return;
5450
5451 first = sg;
5452 do {
5453 tmp = sg->next;
5454
5455 if (free_sgp && atomic_dec_and_test(&sg->sgp->ref))
5456 kfree(sg->sgp);
5457
5458 kfree(sg);
5459 sg = tmp;
5460 } while (sg != first);
5461}
5462
Peter Zijlstradce840a2011-04-07 14:09:50 +02005463static void free_sched_domain(struct rcu_head *rcu)
5464{
5465 struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu);
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005466
5467 /*
5468 * If its an overlapping domain it has private groups, iterate and
5469 * nuke them all.
5470 */
5471 if (sd->flags & SD_OVERLAP) {
5472 free_sched_groups(sd->groups, 1);
5473 } else if (atomic_dec_and_test(&sd->groups->ref)) {
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02005474 kfree(sd->groups->sgp);
Peter Zijlstradce840a2011-04-07 14:09:50 +02005475 kfree(sd->groups);
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02005476 }
Peter Zijlstradce840a2011-04-07 14:09:50 +02005477 kfree(sd);
5478}
5479
5480static void destroy_sched_domain(struct sched_domain *sd, int cpu)
5481{
5482 call_rcu(&sd->rcu, free_sched_domain);
5483}
5484
5485static void destroy_sched_domains(struct sched_domain *sd, int cpu)
5486{
5487 for (; sd; sd = sd->parent)
5488 destroy_sched_domain(sd, cpu);
5489}
5490
Linus Torvalds1da177e2005-04-16 15:20:36 -07005491/*
Peter Zijlstra518cd622011-12-07 15:07:31 +01005492 * Keep a special pointer to the highest sched_domain that has
5493 * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this
5494 * allows us to avoid some pointer chasing select_idle_sibling().
5495 *
5496 * Also keep a unique ID per domain (we use the first cpu number in
5497 * the cpumask of the domain), this allows us to quickly tell if
Peter Zijlstra39be3502012-01-26 12:44:34 +01005498 * two cpus are in the same cache domain, see cpus_share_cache().
Peter Zijlstra518cd622011-12-07 15:07:31 +01005499 */
5500DEFINE_PER_CPU(struct sched_domain *, sd_llc);
5501DEFINE_PER_CPU(int, sd_llc_id);
5502
5503static void update_top_cache_domain(int cpu)
5504{
5505 struct sched_domain *sd;
5506 int id = cpu;
5507
5508 sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES);
Linus Torvalds37407ea2012-09-16 12:29:43 -07005509 if (sd)
Peter Zijlstra518cd622011-12-07 15:07:31 +01005510 id = cpumask_first(sched_domain_span(sd));
5511
5512 rcu_assign_pointer(per_cpu(sd_llc, cpu), sd);
5513 per_cpu(sd_llc_id, cpu) = id;
5514}
5515
5516/*
Ingo Molnar0eab9142008-01-25 21:08:19 +01005517 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
Linus Torvalds1da177e2005-04-16 15:20:36 -07005518 * hold the hotplug lock.
5519 */
Ingo Molnar0eab9142008-01-25 21:08:19 +01005520static void
5521cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005522{
Ingo Molnar70b97a72006-07-03 00:25:42 -07005523 struct rq *rq = cpu_rq(cpu);
Suresh Siddha245af2c2005-06-25 14:57:25 -07005524 struct sched_domain *tmp;
5525
5526 /* Remove the sched domains which do not contribute to scheduling. */
Li Zefanf29c9b12008-11-06 09:45:16 +08005527 for (tmp = sd; tmp; ) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07005528 struct sched_domain *parent = tmp->parent;
5529 if (!parent)
5530 break;
Li Zefanf29c9b12008-11-06 09:45:16 +08005531
Siddha, Suresh B1a848872006-10-03 01:14:08 -07005532 if (sd_parent_degenerate(tmp, parent)) {
Suresh Siddha245af2c2005-06-25 14:57:25 -07005533 tmp->parent = parent->parent;
Siddha, Suresh B1a848872006-10-03 01:14:08 -07005534 if (parent->parent)
5535 parent->parent->child = tmp;
Peter Zijlstradce840a2011-04-07 14:09:50 +02005536 destroy_sched_domain(parent, cpu);
Li Zefanf29c9b12008-11-06 09:45:16 +08005537 } else
5538 tmp = tmp->parent;
Suresh Siddha245af2c2005-06-25 14:57:25 -07005539 }
5540
Siddha, Suresh B1a848872006-10-03 01:14:08 -07005541 if (sd && sd_degenerate(sd)) {
Peter Zijlstradce840a2011-04-07 14:09:50 +02005542 tmp = sd;
Suresh Siddha245af2c2005-06-25 14:57:25 -07005543 sd = sd->parent;
Peter Zijlstradce840a2011-04-07 14:09:50 +02005544 destroy_sched_domain(tmp, cpu);
Siddha, Suresh B1a848872006-10-03 01:14:08 -07005545 if (sd)
5546 sd->child = NULL;
5547 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005548
Peter Zijlstra4cb98832011-04-07 14:09:58 +02005549 sched_domain_debug(sd, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005550
Gregory Haskins57d885f2008-01-25 21:08:18 +01005551 rq_attach_root(rq, rd);
Peter Zijlstradce840a2011-04-07 14:09:50 +02005552 tmp = rq->sd;
Nick Piggin674311d2005-06-25 14:57:27 -07005553 rcu_assign_pointer(rq->sd, sd);
Peter Zijlstradce840a2011-04-07 14:09:50 +02005554 destroy_sched_domains(tmp, cpu);
Peter Zijlstra518cd622011-12-07 15:07:31 +01005555
5556 update_top_cache_domain(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005557}
5558
5559/* cpus with isolated domains */
Rusty Russelldcc30a32008-11-25 02:35:12 +10305560static cpumask_var_t cpu_isolated_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005561
5562/* Setup the mask of cpus configured for isolated domains */
5563static int __init isolated_cpu_setup(char *str)
5564{
Rusty Russellbdddd292009-12-02 14:09:16 +10305565 alloc_bootmem_cpumask_var(&cpu_isolated_map);
Rusty Russell968ea6d2008-12-13 21:55:51 +10305566 cpulist_parse(str, cpu_isolated_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005567 return 1;
5568}
5569
Ingo Molnar8927f492007-10-15 17:00:13 +02005570__setup("isolcpus=", isolated_cpu_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005571
Peter Zijlstrad3081f52011-04-07 14:09:59 +02005572static const struct cpumask *cpu_cpu_mask(int cpu)
5573{
5574 return cpumask_of_node(cpu_to_node(cpu));
5575}
5576
Peter Zijlstradce840a2011-04-07 14:09:50 +02005577struct sd_data {
5578 struct sched_domain **__percpu sd;
5579 struct sched_group **__percpu sg;
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02005580 struct sched_group_power **__percpu sgp;
Peter Zijlstradce840a2011-04-07 14:09:50 +02005581};
5582
Andreas Herrmann49a02c52009-08-18 12:51:52 +02005583struct s_data {
Peter Zijlstra21d42cc2011-04-07 14:09:48 +02005584 struct sched_domain ** __percpu sd;
Andreas Herrmann49a02c52009-08-18 12:51:52 +02005585 struct root_domain *rd;
5586};
5587
Andreas Herrmann2109b992009-08-18 12:53:00 +02005588enum s_alloc {
Andreas Herrmann2109b992009-08-18 12:53:00 +02005589 sa_rootdomain,
Peter Zijlstra21d42cc2011-04-07 14:09:48 +02005590 sa_sd,
Peter Zijlstradce840a2011-04-07 14:09:50 +02005591 sa_sd_storage,
Andreas Herrmann2109b992009-08-18 12:53:00 +02005592 sa_none,
5593};
5594
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02005595struct sched_domain_topology_level;
5596
5597typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu);
Peter Zijlstraeb7a74e62011-04-07 14:10:00 +02005598typedef const struct cpumask *(*sched_domain_mask_f)(int cpu);
5599
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005600#define SDTL_OVERLAP 0x01
5601
Peter Zijlstraeb7a74e62011-04-07 14:10:00 +02005602struct sched_domain_topology_level {
Peter Zijlstra2c402dc2011-04-07 14:10:01 +02005603 sched_domain_init_f init;
5604 sched_domain_mask_f mask;
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005605 int flags;
Peter Zijlstracb83b622012-04-17 15:49:36 +02005606 int numa_level;
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02005607 struct sd_data data;
Peter Zijlstraeb7a74e62011-04-07 14:10:00 +02005608};
5609
Peter Zijlstrac1174872012-05-31 14:47:33 +02005610/*
5611 * Build an iteration mask that can exclude certain CPUs from the upwards
5612 * domain traversal.
5613 *
5614 * Asymmetric node setups can result in situations where the domain tree is of
5615 * unequal depth, make sure to skip domains that already cover the entire
5616 * range.
5617 *
5618 * In that case build_sched_domains() will have terminated the iteration early
5619 * and our sibling sd spans will be empty. Domains should always include the
5620 * cpu they're built on, so check that.
5621 *
5622 */
5623static void build_group_mask(struct sched_domain *sd, struct sched_group *sg)
5624{
5625 const struct cpumask *span = sched_domain_span(sd);
5626 struct sd_data *sdd = sd->private;
5627 struct sched_domain *sibling;
5628 int i;
5629
5630 for_each_cpu(i, span) {
5631 sibling = *per_cpu_ptr(sdd->sd, i);
5632 if (!cpumask_test_cpu(i, sched_domain_span(sibling)))
5633 continue;
5634
5635 cpumask_set_cpu(i, sched_group_mask(sg));
5636 }
5637}
5638
5639/*
5640 * Return the canonical balance cpu for this group, this is the first cpu
5641 * of this group that's also in the iteration mask.
5642 */
5643int group_balance_cpu(struct sched_group *sg)
5644{
5645 return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg));
5646}
5647
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005648static int
5649build_overlap_sched_groups(struct sched_domain *sd, int cpu)
5650{
5651 struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg;
5652 const struct cpumask *span = sched_domain_span(sd);
5653 struct cpumask *covered = sched_domains_tmpmask;
5654 struct sd_data *sdd = sd->private;
5655 struct sched_domain *child;
5656 int i;
5657
5658 cpumask_clear(covered);
5659
5660 for_each_cpu(i, span) {
5661 struct cpumask *sg_span;
5662
5663 if (cpumask_test_cpu(i, covered))
5664 continue;
5665
Peter Zijlstrac1174872012-05-31 14:47:33 +02005666 child = *per_cpu_ptr(sdd->sd, i);
5667
5668 /* See the comment near build_group_mask(). */
5669 if (!cpumask_test_cpu(i, sched_domain_span(child)))
5670 continue;
5671
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005672 sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
Suresh Siddha4d78a222011-11-18 15:03:29 -08005673 GFP_KERNEL, cpu_to_node(cpu));
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005674
5675 if (!sg)
5676 goto fail;
5677
5678 sg_span = sched_group_cpus(sg);
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005679 if (child->child) {
5680 child = child->child;
5681 cpumask_copy(sg_span, sched_domain_span(child));
5682 } else
5683 cpumask_set_cpu(i, sg_span);
5684
5685 cpumask_or(covered, covered, sg_span);
5686
Peter Zijlstra74a5ce22012-05-23 18:00:43 +02005687 sg->sgp = *per_cpu_ptr(sdd->sgp, i);
Peter Zijlstrac1174872012-05-31 14:47:33 +02005688 if (atomic_inc_return(&sg->sgp->ref) == 1)
5689 build_group_mask(sd, sg);
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005690
Peter Zijlstrac3decf02012-05-31 12:05:32 +02005691 /*
5692 * Initialize sgp->power such that even if we mess up the
5693 * domains and no possible iteration will get us here, we won't
5694 * die on a /0 trap.
5695 */
5696 sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span);
Peter Zijlstrac1174872012-05-31 14:47:33 +02005697
5698 /*
5699 * Make sure the first group of this domain contains the
5700 * canonical balance cpu. Otherwise the sched_domain iteration
5701 * breaks. See update_sg_lb_stats().
5702 */
Peter Zijlstra74a5ce22012-05-23 18:00:43 +02005703 if ((!groups && cpumask_test_cpu(cpu, sg_span)) ||
Peter Zijlstrac1174872012-05-31 14:47:33 +02005704 group_balance_cpu(sg) == cpu)
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005705 groups = sg;
5706
5707 if (!first)
5708 first = sg;
5709 if (last)
5710 last->next = sg;
5711 last = sg;
5712 last->next = first;
5713 }
5714 sd->groups = groups;
5715
5716 return 0;
5717
5718fail:
5719 free_sched_groups(first, 0);
5720
5721 return -ENOMEM;
5722}
5723
Peter Zijlstradce840a2011-04-07 14:09:50 +02005724static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005725{
Peter Zijlstradce840a2011-04-07 14:09:50 +02005726 struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu);
5727 struct sched_domain *child = sd->child;
5728
5729 if (child)
5730 cpu = cpumask_first(sched_domain_span(child));
5731
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02005732 if (sg) {
Peter Zijlstradce840a2011-04-07 14:09:50 +02005733 *sg = *per_cpu_ptr(sdd->sg, cpu);
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02005734 (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu);
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005735 atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02005736 }
Peter Zijlstradce840a2011-04-07 14:09:50 +02005737
Linus Torvalds1da177e2005-04-16 15:20:36 -07005738 return cpu;
5739}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005740
Ingo Molnar48f24c42006-07-03 00:25:40 -07005741/*
Peter Zijlstradce840a2011-04-07 14:09:50 +02005742 * build_sched_groups will build a circular linked list of the groups
5743 * covered by the given span, and will set each group's ->cpumask correctly,
5744 * and ->cpu_power to 0.
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005745 *
5746 * Assumes the sched_domain tree is fully constructed
Ingo Molnar48f24c42006-07-03 00:25:40 -07005747 */
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005748static int
5749build_sched_groups(struct sched_domain *sd, int cpu)
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08005750{
Peter Zijlstradce840a2011-04-07 14:09:50 +02005751 struct sched_group *first = NULL, *last = NULL;
5752 struct sd_data *sdd = sd->private;
5753 const struct cpumask *span = sched_domain_span(sd);
Peter Zijlstraf96225f2011-04-07 14:09:57 +02005754 struct cpumask *covered;
Peter Zijlstradce840a2011-04-07 14:09:50 +02005755 int i;
5756
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005757 get_group(cpu, sdd, &sd->groups);
5758 atomic_inc(&sd->groups->ref);
5759
5760 if (cpu != cpumask_first(sched_domain_span(sd)))
5761 return 0;
5762
Peter Zijlstraf96225f2011-04-07 14:09:57 +02005763 lockdep_assert_held(&sched_domains_mutex);
5764 covered = sched_domains_tmpmask;
5765
Peter Zijlstradce840a2011-04-07 14:09:50 +02005766 cpumask_clear(covered);
5767
5768 for_each_cpu(i, span) {
5769 struct sched_group *sg;
5770 int group = get_group(i, sdd, &sg);
5771 int j;
5772
5773 if (cpumask_test_cpu(i, covered))
5774 continue;
5775
5776 cpumask_clear(sched_group_cpus(sg));
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02005777 sg->sgp->power = 0;
Peter Zijlstrac1174872012-05-31 14:47:33 +02005778 cpumask_setall(sched_group_mask(sg));
Peter Zijlstradce840a2011-04-07 14:09:50 +02005779
5780 for_each_cpu(j, span) {
5781 if (get_group(j, sdd, NULL) != group)
5782 continue;
5783
5784 cpumask_set_cpu(j, covered);
5785 cpumask_set_cpu(j, sched_group_cpus(sg));
5786 }
5787
5788 if (!first)
5789 first = sg;
5790 if (last)
5791 last->next = sg;
5792 last = sg;
5793 }
5794 last->next = first;
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005795
5796 return 0;
Siddha, Suresh B1e9f28f2006-03-27 01:15:22 -08005797}
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07005798
Linus Torvalds1da177e2005-04-16 15:20:36 -07005799/*
Siddha, Suresh B89c47102006-10-03 01:14:09 -07005800 * Initialize sched groups cpu_power.
5801 *
5802 * cpu_power indicates the capacity of sched group, which is used while
5803 * distributing the load between different sched groups in a sched domain.
5804 * Typically cpu_power for all the groups in a sched domain will be same unless
5805 * there are asymmetries in the topology. If there are asymmetries, group
5806 * having more cpu_power will pickup more load compared to the group having
5807 * less cpu_power.
Siddha, Suresh B89c47102006-10-03 01:14:09 -07005808 */
5809static void init_sched_groups_power(int cpu, struct sched_domain *sd)
5810{
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005811 struct sched_group *sg = sd->groups;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07005812
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005813 WARN_ON(!sd || !sg);
5814
5815 do {
5816 sg->group_weight = cpumask_weight(sched_group_cpus(sg));
5817 sg = sg->next;
5818 } while (sg != sd->groups);
5819
Peter Zijlstrac1174872012-05-31 14:47:33 +02005820 if (cpu != group_balance_cpu(sg))
Siddha, Suresh B89c47102006-10-03 01:14:09 -07005821 return;
5822
Peter Zijlstrad274cb32011-04-07 14:09:43 +02005823 update_group_power(sd, cpu);
Suresh Siddha69e1e812011-12-01 17:07:33 -08005824 atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight);
Siddha, Suresh B89c47102006-10-03 01:14:09 -07005825}
5826
Peter Zijlstra029632f2011-10-25 10:00:11 +02005827int __weak arch_sd_sibling_asym_packing(void)
5828{
5829 return 0*SD_ASYM_PACKING;
Siddha, Suresh B89c47102006-10-03 01:14:09 -07005830}
5831
5832/*
Mike Travis7c16ec52008-04-04 18:11:11 -07005833 * Initializers for schedule domains
5834 * Non-inlined to reduce accumulated stack pressure in build_sched_domains()
5835 */
5836
Ingo Molnara5d8c342008-10-09 11:35:51 +02005837#ifdef CONFIG_SCHED_DEBUG
5838# define SD_INIT_NAME(sd, type) sd->name = #type
5839#else
5840# define SD_INIT_NAME(sd, type) do { } while (0)
5841#endif
5842
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02005843#define SD_INIT_FUNC(type) \
5844static noinline struct sched_domain * \
5845sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \
5846{ \
5847 struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \
5848 *sd = SD_##type##_INIT; \
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02005849 SD_INIT_NAME(sd, type); \
5850 sd->private = &tl->data; \
5851 return sd; \
Mike Travis7c16ec52008-04-04 18:11:11 -07005852}
5853
5854SD_INIT_FUNC(CPU)
Mike Travis7c16ec52008-04-04 18:11:11 -07005855#ifdef CONFIG_SCHED_SMT
5856 SD_INIT_FUNC(SIBLING)
5857#endif
5858#ifdef CONFIG_SCHED_MC
5859 SD_INIT_FUNC(MC)
5860#endif
Heiko Carstens01a08542010-08-31 10:28:16 +02005861#ifdef CONFIG_SCHED_BOOK
5862 SD_INIT_FUNC(BOOK)
5863#endif
Mike Travis7c16ec52008-04-04 18:11:11 -07005864
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09005865static int default_relax_domain_level = -1;
Peter Zijlstra60495e72011-04-07 14:10:04 +02005866int sched_domain_level_max;
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09005867
5868static int __init setup_relax_domain_level(char *str)
5869{
Dimitri Sivanicha841f8c2012-06-05 13:44:36 -05005870 if (kstrtoint(str, 0, &default_relax_domain_level))
5871 pr_warn("Unable to set relax_domain_level\n");
Li Zefan30e0e172008-05-13 10:27:17 +08005872
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09005873 return 1;
5874}
5875__setup("relax_domain_level=", setup_relax_domain_level);
5876
5877static void set_domain_attribute(struct sched_domain *sd,
5878 struct sched_domain_attr *attr)
5879{
5880 int request;
5881
5882 if (!attr || attr->relax_domain_level < 0) {
5883 if (default_relax_domain_level < 0)
5884 return;
5885 else
5886 request = default_relax_domain_level;
5887 } else
5888 request = attr->relax_domain_level;
5889 if (request < sd->level) {
5890 /* turn off idle balance on this domain */
Peter Zijlstrac88d5912009-09-10 13:50:02 +02005891 sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09005892 } else {
5893 /* turn on idle balance on this domain */
Peter Zijlstrac88d5912009-09-10 13:50:02 +02005894 sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09005895 }
5896}
5897
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02005898static void __sdt_free(const struct cpumask *cpu_map);
5899static int __sdt_alloc(const struct cpumask *cpu_map);
5900
Andreas Herrmann2109b992009-08-18 12:53:00 +02005901static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
5902 const struct cpumask *cpu_map)
5903{
5904 switch (what) {
Andreas Herrmann2109b992009-08-18 12:53:00 +02005905 case sa_rootdomain:
Peter Zijlstra822ff792011-04-07 14:09:51 +02005906 if (!atomic_read(&d->rd->refcount))
5907 free_rootdomain(&d->rd->rcu); /* fall through */
Peter Zijlstra21d42cc2011-04-07 14:09:48 +02005908 case sa_sd:
5909 free_percpu(d->sd); /* fall through */
Peter Zijlstradce840a2011-04-07 14:09:50 +02005910 case sa_sd_storage:
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02005911 __sdt_free(cpu_map); /* fall through */
Andreas Herrmann2109b992009-08-18 12:53:00 +02005912 case sa_none:
5913 break;
5914 }
5915}
5916
5917static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
5918 const struct cpumask *cpu_map)
5919{
Peter Zijlstradce840a2011-04-07 14:09:50 +02005920 memset(d, 0, sizeof(*d));
5921
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02005922 if (__sdt_alloc(cpu_map))
5923 return sa_sd_storage;
Peter Zijlstra21d42cc2011-04-07 14:09:48 +02005924 d->sd = alloc_percpu(struct sched_domain *);
Peter Zijlstradce840a2011-04-07 14:09:50 +02005925 if (!d->sd)
5926 return sa_sd_storage;
Andreas Herrmann2109b992009-08-18 12:53:00 +02005927 d->rd = alloc_rootdomain();
Peter Zijlstradce840a2011-04-07 14:09:50 +02005928 if (!d->rd)
Peter Zijlstra21d42cc2011-04-07 14:09:48 +02005929 return sa_sd;
Andreas Herrmann2109b992009-08-18 12:53:00 +02005930 return sa_rootdomain;
5931}
5932
Peter Zijlstradce840a2011-04-07 14:09:50 +02005933/*
5934 * NULL the sd_data elements we've used to build the sched_domain and
5935 * sched_group structure so that the subsequent __free_domain_allocs()
5936 * will not free the data we're using.
5937 */
5938static void claim_allocations(int cpu, struct sched_domain *sd)
5939{
5940 struct sd_data *sdd = sd->private;
Peter Zijlstradce840a2011-04-07 14:09:50 +02005941
5942 WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd);
5943 *per_cpu_ptr(sdd->sd, cpu) = NULL;
5944
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005945 if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref))
Peter Zijlstradce840a2011-04-07 14:09:50 +02005946 *per_cpu_ptr(sdd->sg, cpu) = NULL;
Peter Zijlstrae3589f62011-07-15 10:35:52 +02005947
5948 if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref))
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02005949 *per_cpu_ptr(sdd->sgp, cpu) = NULL;
Peter Zijlstradce840a2011-04-07 14:09:50 +02005950}
5951
Andreas Herrmannd8173532009-08-18 12:57:03 +02005952#ifdef CONFIG_SCHED_SMT
Peter Zijlstra2c402dc2011-04-07 14:10:01 +02005953static const struct cpumask *cpu_smt_mask(int cpu)
5954{
5955 return topology_thread_cpumask(cpu);
Andreas Herrmannd8173532009-08-18 12:57:03 +02005956}
Peter Zijlstra2c402dc2011-04-07 14:10:01 +02005957#endif
Andreas Herrmannd8173532009-08-18 12:57:03 +02005958
Peter Zijlstrad069b912011-04-07 14:10:02 +02005959/*
5960 * Topology list, bottom-up.
5961 */
Peter Zijlstraeb7a74e62011-04-07 14:10:00 +02005962static struct sched_domain_topology_level default_topology[] = {
Peter Zijlstrad069b912011-04-07 14:10:02 +02005963#ifdef CONFIG_SCHED_SMT
5964 { sd_init_SIBLING, cpu_smt_mask, },
Peter Zijlstra2c402dc2011-04-07 14:10:01 +02005965#endif
5966#ifdef CONFIG_SCHED_MC
5967 { sd_init_MC, cpu_coregroup_mask, },
5968#endif
Peter Zijlstrad069b912011-04-07 14:10:02 +02005969#ifdef CONFIG_SCHED_BOOK
5970 { sd_init_BOOK, cpu_book_mask, },
5971#endif
5972 { sd_init_CPU, cpu_cpu_mask, },
Peter Zijlstraeb7a74e62011-04-07 14:10:00 +02005973 { NULL, },
5974};
5975
5976static struct sched_domain_topology_level *sched_domain_topology = default_topology;
5977
Peter Zijlstracb83b622012-04-17 15:49:36 +02005978#ifdef CONFIG_NUMA
5979
5980static int sched_domains_numa_levels;
Peter Zijlstracb83b622012-04-17 15:49:36 +02005981static int *sched_domains_numa_distance;
5982static struct cpumask ***sched_domains_numa_masks;
5983static int sched_domains_curr_level;
5984
Peter Zijlstracb83b622012-04-17 15:49:36 +02005985static inline int sd_local_flags(int level)
5986{
Alex Shi10717dc2012-06-06 14:52:51 +08005987 if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE)
Peter Zijlstracb83b622012-04-17 15:49:36 +02005988 return 0;
5989
5990 return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE;
5991}
5992
5993static struct sched_domain *
5994sd_numa_init(struct sched_domain_topology_level *tl, int cpu)
5995{
5996 struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu);
5997 int level = tl->numa_level;
5998 int sd_weight = cpumask_weight(
5999 sched_domains_numa_masks[level][cpu_to_node(cpu)]);
6000
6001 *sd = (struct sched_domain){
6002 .min_interval = sd_weight,
6003 .max_interval = 2*sd_weight,
6004 .busy_factor = 32,
Peter Zijlstra870a0bb2012-05-11 00:26:27 +02006005 .imbalance_pct = 125,
Peter Zijlstracb83b622012-04-17 15:49:36 +02006006 .cache_nice_tries = 2,
6007 .busy_idx = 3,
6008 .idle_idx = 2,
6009 .newidle_idx = 0,
6010 .wake_idx = 0,
6011 .forkexec_idx = 0,
6012
6013 .flags = 1*SD_LOAD_BALANCE
6014 | 1*SD_BALANCE_NEWIDLE
6015 | 0*SD_BALANCE_EXEC
6016 | 0*SD_BALANCE_FORK
6017 | 0*SD_BALANCE_WAKE
6018 | 0*SD_WAKE_AFFINE
Peter Zijlstracb83b622012-04-17 15:49:36 +02006019 | 0*SD_SHARE_CPUPOWER
Peter Zijlstracb83b622012-04-17 15:49:36 +02006020 | 0*SD_SHARE_PKG_RESOURCES
6021 | 1*SD_SERIALIZE
6022 | 0*SD_PREFER_SIBLING
6023 | sd_local_flags(level)
6024 ,
6025 .last_balance = jiffies,
6026 .balance_interval = sd_weight,
6027 };
6028 SD_INIT_NAME(sd, NUMA);
6029 sd->private = &tl->data;
6030
6031 /*
6032 * Ugly hack to pass state to sd_numa_mask()...
6033 */
6034 sched_domains_curr_level = tl->numa_level;
6035
6036 return sd;
6037}
6038
6039static const struct cpumask *sd_numa_mask(int cpu)
6040{
6041 return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)];
6042}
6043
Peter Zijlstrad039ac62012-05-31 21:20:16 +02006044static void sched_numa_warn(const char *str)
6045{
6046 static int done = false;
6047 int i,j;
6048
6049 if (done)
6050 return;
6051
6052 done = true;
6053
6054 printk(KERN_WARNING "ERROR: %s\n\n", str);
6055
6056 for (i = 0; i < nr_node_ids; i++) {
6057 printk(KERN_WARNING " ");
6058 for (j = 0; j < nr_node_ids; j++)
6059 printk(KERN_CONT "%02d ", node_distance(i,j));
6060 printk(KERN_CONT "\n");
6061 }
6062 printk(KERN_WARNING "\n");
6063}
6064
6065static bool find_numa_distance(int distance)
6066{
6067 int i;
6068
6069 if (distance == node_distance(0, 0))
6070 return true;
6071
6072 for (i = 0; i < sched_domains_numa_levels; i++) {
6073 if (sched_domains_numa_distance[i] == distance)
6074 return true;
6075 }
6076
6077 return false;
6078}
6079
Peter Zijlstracb83b622012-04-17 15:49:36 +02006080static void sched_init_numa(void)
6081{
6082 int next_distance, curr_distance = node_distance(0, 0);
6083 struct sched_domain_topology_level *tl;
6084 int level = 0;
6085 int i, j, k;
6086
Peter Zijlstracb83b622012-04-17 15:49:36 +02006087 sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL);
6088 if (!sched_domains_numa_distance)
6089 return;
6090
6091 /*
6092 * O(nr_nodes^2) deduplicating selection sort -- in order to find the
6093 * unique distances in the node_distance() table.
6094 *
6095 * Assumes node_distance(0,j) includes all distances in
6096 * node_distance(i,j) in order to avoid cubic time.
Peter Zijlstracb83b622012-04-17 15:49:36 +02006097 */
6098 next_distance = curr_distance;
6099 for (i = 0; i < nr_node_ids; i++) {
6100 for (j = 0; j < nr_node_ids; j++) {
Peter Zijlstrad039ac62012-05-31 21:20:16 +02006101 for (k = 0; k < nr_node_ids; k++) {
6102 int distance = node_distance(i, k);
6103
6104 if (distance > curr_distance &&
6105 (distance < next_distance ||
6106 next_distance == curr_distance))
6107 next_distance = distance;
6108
6109 /*
6110 * While not a strong assumption it would be nice to know
6111 * about cases where if node A is connected to B, B is not
6112 * equally connected to A.
6113 */
6114 if (sched_debug() && node_distance(k, i) != distance)
6115 sched_numa_warn("Node-distance not symmetric");
6116
6117 if (sched_debug() && i && !find_numa_distance(distance))
6118 sched_numa_warn("Node-0 not representative");
6119 }
6120 if (next_distance != curr_distance) {
6121 sched_domains_numa_distance[level++] = next_distance;
6122 sched_domains_numa_levels = level;
6123 curr_distance = next_distance;
6124 } else break;
Peter Zijlstracb83b622012-04-17 15:49:36 +02006125 }
Peter Zijlstrad039ac62012-05-31 21:20:16 +02006126
6127 /*
6128 * In case of sched_debug() we verify the above assumption.
6129 */
6130 if (!sched_debug())
6131 break;
Peter Zijlstracb83b622012-04-17 15:49:36 +02006132 }
6133 /*
6134 * 'level' contains the number of unique distances, excluding the
6135 * identity distance node_distance(i,i).
6136 *
6137 * The sched_domains_nume_distance[] array includes the actual distance
6138 * numbers.
6139 */
6140
Tang Chen5f7865f2012-09-25 21:12:30 +08006141 /*
6142 * Here, we should temporarily reset sched_domains_numa_levels to 0.
6143 * If it fails to allocate memory for array sched_domains_numa_masks[][],
6144 * the array will contain less then 'level' members. This could be
6145 * dangerous when we use it to iterate array sched_domains_numa_masks[][]
6146 * in other functions.
6147 *
6148 * We reset it to 'level' at the end of this function.
6149 */
6150 sched_domains_numa_levels = 0;
6151
Peter Zijlstracb83b622012-04-17 15:49:36 +02006152 sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL);
6153 if (!sched_domains_numa_masks)
6154 return;
6155
6156 /*
6157 * Now for each level, construct a mask per node which contains all
6158 * cpus of nodes that are that many hops away from us.
6159 */
6160 for (i = 0; i < level; i++) {
6161 sched_domains_numa_masks[i] =
6162 kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL);
6163 if (!sched_domains_numa_masks[i])
6164 return;
6165
6166 for (j = 0; j < nr_node_ids; j++) {
Peter Zijlstra2ea45802012-05-25 09:26:43 +02006167 struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL);
Peter Zijlstracb83b622012-04-17 15:49:36 +02006168 if (!mask)
6169 return;
6170
6171 sched_domains_numa_masks[i][j] = mask;
6172
6173 for (k = 0; k < nr_node_ids; k++) {
Peter Zijlstradd7d8632012-05-11 00:56:20 +02006174 if (node_distance(j, k) > sched_domains_numa_distance[i])
Peter Zijlstracb83b622012-04-17 15:49:36 +02006175 continue;
6176
6177 cpumask_or(mask, mask, cpumask_of_node(k));
6178 }
6179 }
6180 }
6181
6182 tl = kzalloc((ARRAY_SIZE(default_topology) + level) *
6183 sizeof(struct sched_domain_topology_level), GFP_KERNEL);
6184 if (!tl)
6185 return;
6186
6187 /*
6188 * Copy the default topology bits..
6189 */
6190 for (i = 0; default_topology[i].init; i++)
6191 tl[i] = default_topology[i];
6192
6193 /*
6194 * .. and append 'j' levels of NUMA goodness.
6195 */
6196 for (j = 0; j < level; i++, j++) {
6197 tl[i] = (struct sched_domain_topology_level){
6198 .init = sd_numa_init,
6199 .mask = sd_numa_mask,
6200 .flags = SDTL_OVERLAP,
6201 .numa_level = j,
6202 };
6203 }
6204
6205 sched_domain_topology = tl;
Tang Chen5f7865f2012-09-25 21:12:30 +08006206
6207 sched_domains_numa_levels = level;
Peter Zijlstracb83b622012-04-17 15:49:36 +02006208}
Tang Chen301a5cb2012-09-25 21:12:31 +08006209
6210static void sched_domains_numa_masks_set(int cpu)
6211{
6212 int i, j;
6213 int node = cpu_to_node(cpu);
6214
6215 for (i = 0; i < sched_domains_numa_levels; i++) {
6216 for (j = 0; j < nr_node_ids; j++) {
6217 if (node_distance(j, node) <= sched_domains_numa_distance[i])
6218 cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]);
6219 }
6220 }
6221}
6222
6223static void sched_domains_numa_masks_clear(int cpu)
6224{
6225 int i, j;
6226 for (i = 0; i < sched_domains_numa_levels; i++) {
6227 for (j = 0; j < nr_node_ids; j++)
6228 cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]);
6229 }
6230}
6231
6232/*
6233 * Update sched_domains_numa_masks[level][node] array when new cpus
6234 * are onlined.
6235 */
6236static int sched_domains_numa_masks_update(struct notifier_block *nfb,
6237 unsigned long action,
6238 void *hcpu)
6239{
6240 int cpu = (long)hcpu;
6241
6242 switch (action & ~CPU_TASKS_FROZEN) {
6243 case CPU_ONLINE:
6244 sched_domains_numa_masks_set(cpu);
6245 break;
6246
6247 case CPU_DEAD:
6248 sched_domains_numa_masks_clear(cpu);
6249 break;
6250
6251 default:
6252 return NOTIFY_DONE;
6253 }
6254
6255 return NOTIFY_OK;
Peter Zijlstracb83b622012-04-17 15:49:36 +02006256}
6257#else
6258static inline void sched_init_numa(void)
6259{
6260}
Tang Chen301a5cb2012-09-25 21:12:31 +08006261
6262static int sched_domains_numa_masks_update(struct notifier_block *nfb,
6263 unsigned long action,
6264 void *hcpu)
6265{
6266 return 0;
6267}
Peter Zijlstracb83b622012-04-17 15:49:36 +02006268#endif /* CONFIG_NUMA */
6269
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02006270static int __sdt_alloc(const struct cpumask *cpu_map)
6271{
6272 struct sched_domain_topology_level *tl;
6273 int j;
6274
6275 for (tl = sched_domain_topology; tl->init; tl++) {
6276 struct sd_data *sdd = &tl->data;
6277
6278 sdd->sd = alloc_percpu(struct sched_domain *);
6279 if (!sdd->sd)
6280 return -ENOMEM;
6281
6282 sdd->sg = alloc_percpu(struct sched_group *);
6283 if (!sdd->sg)
6284 return -ENOMEM;
6285
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02006286 sdd->sgp = alloc_percpu(struct sched_group_power *);
6287 if (!sdd->sgp)
6288 return -ENOMEM;
6289
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02006290 for_each_cpu(j, cpu_map) {
6291 struct sched_domain *sd;
6292 struct sched_group *sg;
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02006293 struct sched_group_power *sgp;
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02006294
6295 sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(),
6296 GFP_KERNEL, cpu_to_node(j));
6297 if (!sd)
6298 return -ENOMEM;
6299
6300 *per_cpu_ptr(sdd->sd, j) = sd;
6301
6302 sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
6303 GFP_KERNEL, cpu_to_node(j));
6304 if (!sg)
6305 return -ENOMEM;
6306
Igor Mammedov30b4e9e2012-05-09 12:38:28 +02006307 sg->next = sg;
6308
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02006309 *per_cpu_ptr(sdd->sg, j) = sg;
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02006310
Peter Zijlstrac1174872012-05-31 14:47:33 +02006311 sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(),
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02006312 GFP_KERNEL, cpu_to_node(j));
6313 if (!sgp)
6314 return -ENOMEM;
6315
6316 *per_cpu_ptr(sdd->sgp, j) = sgp;
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02006317 }
6318 }
6319
6320 return 0;
6321}
6322
6323static void __sdt_free(const struct cpumask *cpu_map)
6324{
6325 struct sched_domain_topology_level *tl;
6326 int j;
6327
6328 for (tl = sched_domain_topology; tl->init; tl++) {
6329 struct sd_data *sdd = &tl->data;
6330
6331 for_each_cpu(j, cpu_map) {
he, bofb2cf2c2012-04-25 19:59:21 +08006332 struct sched_domain *sd;
6333
6334 if (sdd->sd) {
6335 sd = *per_cpu_ptr(sdd->sd, j);
6336 if (sd && (sd->flags & SD_OVERLAP))
6337 free_sched_groups(sd->groups, 0);
6338 kfree(*per_cpu_ptr(sdd->sd, j));
6339 }
6340
6341 if (sdd->sg)
6342 kfree(*per_cpu_ptr(sdd->sg, j));
6343 if (sdd->sgp)
6344 kfree(*per_cpu_ptr(sdd->sgp, j));
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02006345 }
6346 free_percpu(sdd->sd);
he, bofb2cf2c2012-04-25 19:59:21 +08006347 sdd->sd = NULL;
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02006348 free_percpu(sdd->sg);
he, bofb2cf2c2012-04-25 19:59:21 +08006349 sdd->sg = NULL;
Peter Zijlstra9c3f75c2011-07-14 13:00:06 +02006350 free_percpu(sdd->sgp);
he, bofb2cf2c2012-04-25 19:59:21 +08006351 sdd->sgp = NULL;
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02006352 }
6353}
6354
Peter Zijlstra2c402dc2011-04-07 14:10:01 +02006355struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl,
6356 struct s_data *d, const struct cpumask *cpu_map,
Peter Zijlstrad069b912011-04-07 14:10:02 +02006357 struct sched_domain_attr *attr, struct sched_domain *child,
Peter Zijlstra2c402dc2011-04-07 14:10:01 +02006358 int cpu)
6359{
Peter Zijlstra54ab4ff2011-04-07 14:10:03 +02006360 struct sched_domain *sd = tl->init(tl, cpu);
Peter Zijlstra2c402dc2011-04-07 14:10:01 +02006361 if (!sd)
Peter Zijlstrad069b912011-04-07 14:10:02 +02006362 return child;
Peter Zijlstra2c402dc2011-04-07 14:10:01 +02006363
Peter Zijlstra2c402dc2011-04-07 14:10:01 +02006364 cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu));
Peter Zijlstra60495e72011-04-07 14:10:04 +02006365 if (child) {
6366 sd->level = child->level + 1;
6367 sched_domain_level_max = max(sched_domain_level_max, sd->level);
Peter Zijlstrad069b912011-04-07 14:10:02 +02006368 child->parent = sd;
Peter Zijlstra60495e72011-04-07 14:10:04 +02006369 }
Peter Zijlstrad069b912011-04-07 14:10:02 +02006370 sd->child = child;
Dimitri Sivanicha841f8c2012-06-05 13:44:36 -05006371 set_domain_attribute(sd, attr);
Peter Zijlstra2c402dc2011-04-07 14:10:01 +02006372
6373 return sd;
6374}
6375
Mike Travis7c16ec52008-04-04 18:11:11 -07006376/*
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006377 * Build sched domains for a given set of cpus and attach the sched domains
6378 * to the individual cpus
Linus Torvalds1da177e2005-04-16 15:20:36 -07006379 */
Peter Zijlstradce840a2011-04-07 14:09:50 +02006380static int build_sched_domains(const struct cpumask *cpu_map,
6381 struct sched_domain_attr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006382{
Andreas Herrmann2109b992009-08-18 12:53:00 +02006383 enum s_alloc alloc_state = sa_none;
Peter Zijlstradce840a2011-04-07 14:09:50 +02006384 struct sched_domain *sd;
Andreas Herrmann49a02c52009-08-18 12:51:52 +02006385 struct s_data d;
Peter Zijlstra822ff792011-04-07 14:09:51 +02006386 int i, ret = -ENOMEM;
Rusty Russell3404c8d2008-11-25 02:35:03 +10306387
Andreas Herrmann2109b992009-08-18 12:53:00 +02006388 alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
6389 if (alloc_state != sa_rootdomain)
6390 goto error;
Mike Travis7c16ec52008-04-04 18:11:11 -07006391
Peter Zijlstradce840a2011-04-07 14:09:50 +02006392 /* Set up domains for cpus specified by the cpu_map. */
Rusty Russellabcd0832008-11-25 02:35:02 +10306393 for_each_cpu(i, cpu_map) {
Peter Zijlstraeb7a74e62011-04-07 14:10:00 +02006394 struct sched_domain_topology_level *tl;
6395
Peter Zijlstra3bd65a82011-04-07 14:09:54 +02006396 sd = NULL;
Peter Zijlstrae3589f62011-07-15 10:35:52 +02006397 for (tl = sched_domain_topology; tl->init; tl++) {
Peter Zijlstra2c402dc2011-04-07 14:10:01 +02006398 sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i);
Peter Zijlstrae3589f62011-07-15 10:35:52 +02006399 if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP))
6400 sd->flags |= SD_OVERLAP;
Peter Zijlstrad1102352011-07-20 18:42:57 +02006401 if (cpumask_equal(cpu_map, sched_domain_span(sd)))
6402 break;
Peter Zijlstrae3589f62011-07-15 10:35:52 +02006403 }
Peter Zijlstrad274cb32011-04-07 14:09:43 +02006404
Peter Zijlstrad069b912011-04-07 14:10:02 +02006405 while (sd->child)
6406 sd = sd->child;
6407
Peter Zijlstra21d42cc2011-04-07 14:09:48 +02006408 *per_cpu_ptr(d.sd, i) = sd;
Peter Zijlstradce840a2011-04-07 14:09:50 +02006409 }
Peter Zijlstra21d42cc2011-04-07 14:09:48 +02006410
Peter Zijlstradce840a2011-04-07 14:09:50 +02006411 /* Build the groups for the domains */
6412 for_each_cpu(i, cpu_map) {
6413 for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
6414 sd->span_weight = cpumask_weight(sched_domain_span(sd));
Peter Zijlstrae3589f62011-07-15 10:35:52 +02006415 if (sd->flags & SD_OVERLAP) {
6416 if (build_overlap_sched_groups(sd, i))
6417 goto error;
6418 } else {
6419 if (build_sched_groups(sd, i))
6420 goto error;
6421 }
Peter Zijlstra1cf519022011-04-07 14:09:47 +02006422 }
Peter Zijlstraa06dadb2011-04-07 14:09:44 +02006423 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006424
Linus Torvalds1da177e2005-04-16 15:20:36 -07006425 /* Calculate CPU power for physical packages and nodes */
Peter Zijlstraa9c9a9b2011-04-07 14:09:49 +02006426 for (i = nr_cpumask_bits-1; i >= 0; i--) {
6427 if (!cpumask_test_cpu(i, cpu_map))
6428 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006429
Peter Zijlstradce840a2011-04-07 14:09:50 +02006430 for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
6431 claim_allocations(i, sd);
Peter Zijlstracd4ea6a2011-04-07 14:09:45 +02006432 init_sched_groups_power(i, sd);
Peter Zijlstradce840a2011-04-07 14:09:50 +02006433 }
Siddha, Suresh Bf712c0c2006-07-30 03:02:59 -07006434 }
John Hawkes9c1cfda2005-09-06 15:18:14 -07006435
Linus Torvalds1da177e2005-04-16 15:20:36 -07006436 /* Attach the domains */
Peter Zijlstradce840a2011-04-07 14:09:50 +02006437 rcu_read_lock();
Rusty Russellabcd0832008-11-25 02:35:02 +10306438 for_each_cpu(i, cpu_map) {
Peter Zijlstra21d42cc2011-04-07 14:09:48 +02006439 sd = *per_cpu_ptr(d.sd, i);
Andreas Herrmann49a02c52009-08-18 12:51:52 +02006440 cpu_attach_domain(sd, d.rd, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006441 }
Peter Zijlstradce840a2011-04-07 14:09:50 +02006442 rcu_read_unlock();
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006443
Peter Zijlstra822ff792011-04-07 14:09:51 +02006444 ret = 0;
Srivatsa Vaddagiri51888ca2006-06-27 02:54:38 -07006445error:
Andreas Herrmann2109b992009-08-18 12:53:00 +02006446 __free_domain_allocs(&d, alloc_state, cpu_map);
Peter Zijlstra822ff792011-04-07 14:09:51 +02006447 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006448}
Paul Jackson029190c2007-10-18 23:40:20 -07006449
Rusty Russellacc3f5d2009-11-03 14:53:40 +10306450static cpumask_var_t *doms_cur; /* current sched domains */
Paul Jackson029190c2007-10-18 23:40:20 -07006451static int ndoms_cur; /* number of sched domains in 'doms_cur' */
Ingo Molnar4285f5942008-05-16 17:47:14 +02006452static struct sched_domain_attr *dattr_cur;
6453 /* attribues of custom domains in 'doms_cur' */
Paul Jackson029190c2007-10-18 23:40:20 -07006454
6455/*
6456 * Special case: If a kmalloc of a doms_cur partition (array of
Rusty Russell42128232008-11-25 02:35:12 +10306457 * cpumask) fails, then fallback to a single sched domain,
6458 * as determined by the single cpumask fallback_doms.
Paul Jackson029190c2007-10-18 23:40:20 -07006459 */
Rusty Russell42128232008-11-25 02:35:12 +10306460static cpumask_var_t fallback_doms;
Paul Jackson029190c2007-10-18 23:40:20 -07006461
Heiko Carstensee79d1b2008-12-09 18:49:50 +01006462/*
6463 * arch_update_cpu_topology lets virtualized architectures update the
6464 * cpu core maps. It is supposed to return 1 if the topology changed
6465 * or 0 if it stayed the same.
6466 */
6467int __attribute__((weak)) arch_update_cpu_topology(void)
Heiko Carstens22e52b02008-03-12 18:31:59 +01006468{
Heiko Carstensee79d1b2008-12-09 18:49:50 +01006469 return 0;
Heiko Carstens22e52b02008-03-12 18:31:59 +01006470}
6471
Rusty Russellacc3f5d2009-11-03 14:53:40 +10306472cpumask_var_t *alloc_sched_domains(unsigned int ndoms)
6473{
6474 int i;
6475 cpumask_var_t *doms;
6476
6477 doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL);
6478 if (!doms)
6479 return NULL;
6480 for (i = 0; i < ndoms; i++) {
6481 if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) {
6482 free_sched_domains(doms, i);
6483 return NULL;
6484 }
6485 }
6486 return doms;
6487}
6488
6489void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms)
6490{
6491 unsigned int i;
6492 for (i = 0; i < ndoms; i++)
6493 free_cpumask_var(doms[i]);
6494 kfree(doms);
6495}
6496
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006497/*
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006498 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
Paul Jackson029190c2007-10-18 23:40:20 -07006499 * For now this just excludes isolated cpus, but could be used to
6500 * exclude other special cases in the future.
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006501 */
Peter Zijlstrac4a88492011-04-07 14:09:42 +02006502static int init_sched_domains(const struct cpumask *cpu_map)
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006503{
Milton Miller73785472007-10-24 18:23:48 +02006504 int err;
6505
Heiko Carstens22e52b02008-03-12 18:31:59 +01006506 arch_update_cpu_topology();
Paul Jackson029190c2007-10-18 23:40:20 -07006507 ndoms_cur = 1;
Rusty Russellacc3f5d2009-11-03 14:53:40 +10306508 doms_cur = alloc_sched_domains(ndoms_cur);
Paul Jackson029190c2007-10-18 23:40:20 -07006509 if (!doms_cur)
Rusty Russellacc3f5d2009-11-03 14:53:40 +10306510 doms_cur = &fallback_doms;
6511 cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
Peter Zijlstradce840a2011-04-07 14:09:50 +02006512 err = build_sched_domains(doms_cur[0], NULL);
Milton Miller6382bc92007-10-15 17:00:19 +02006513 register_sched_domain_sysctl();
Milton Miller73785472007-10-24 18:23:48 +02006514
6515 return err;
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006516}
6517
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006518/*
6519 * Detach sched domains from a group of cpus specified in cpu_map
6520 * These cpus will now be attached to the NULL domain
6521 */
Rusty Russell96f874e2008-11-25 02:35:14 +10306522static void detach_destroy_domains(const struct cpumask *cpu_map)
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006523{
6524 int i;
6525
Peter Zijlstradce840a2011-04-07 14:09:50 +02006526 rcu_read_lock();
Rusty Russellabcd0832008-11-25 02:35:02 +10306527 for_each_cpu(i, cpu_map)
Gregory Haskins57d885f2008-01-25 21:08:18 +01006528 cpu_attach_domain(NULL, &def_root_domain, i);
Peter Zijlstradce840a2011-04-07 14:09:50 +02006529 rcu_read_unlock();
Dinakar Guniguntala1a20ff22005-06-25 14:57:33 -07006530}
6531
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006532/* handle null as "default" */
6533static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
6534 struct sched_domain_attr *new, int idx_new)
6535{
6536 struct sched_domain_attr tmp;
6537
6538 /* fast path */
6539 if (!new && !cur)
6540 return 1;
6541
6542 tmp = SD_ATTR_INIT;
6543 return !memcmp(cur ? (cur + idx_cur) : &tmp,
6544 new ? (new + idx_new) : &tmp,
6545 sizeof(struct sched_domain_attr));
6546}
6547
Paul Jackson029190c2007-10-18 23:40:20 -07006548/*
6549 * Partition sched domains as specified by the 'ndoms_new'
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006550 * cpumasks in the array doms_new[] of cpumasks. This compares
Paul Jackson029190c2007-10-18 23:40:20 -07006551 * doms_new[] to the current sched domain partitioning, doms_cur[].
6552 * It destroys each deleted domain and builds each new domain.
6553 *
Rusty Russellacc3f5d2009-11-03 14:53:40 +10306554 * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'.
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01006555 * The masks don't intersect (don't overlap.) We should setup one
6556 * sched domain for each mask. CPUs not in any of the cpumasks will
6557 * not be load balanced. If the same cpumask appears both in the
Paul Jackson029190c2007-10-18 23:40:20 -07006558 * current 'doms_cur' domains and in the new 'doms_new', we can leave
6559 * it as it is.
6560 *
Rusty Russellacc3f5d2009-11-03 14:53:40 +10306561 * The passed in 'doms_new' should be allocated using
6562 * alloc_sched_domains. This routine takes ownership of it and will
6563 * free_sched_domains it when done with it. If the caller failed the
6564 * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1,
6565 * and partition_sched_domains() will fallback to the single partition
6566 * 'fallback_doms', it also forces the domains to be rebuilt.
Paul Jackson029190c2007-10-18 23:40:20 -07006567 *
Rusty Russell96f874e2008-11-25 02:35:14 +10306568 * If doms_new == NULL it will be replaced with cpu_online_mask.
Li Zefan700018e2008-11-18 14:02:03 +08006569 * ndoms_new == 0 is a special case for destroying existing domains,
6570 * and it will not create the default domain.
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07006571 *
Paul Jackson029190c2007-10-18 23:40:20 -07006572 * Call with hotplug lock held
6573 */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10306574void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006575 struct sched_domain_attr *dattr_new)
Paul Jackson029190c2007-10-18 23:40:20 -07006576{
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07006577 int i, j, n;
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01006578 int new_topology;
Paul Jackson029190c2007-10-18 23:40:20 -07006579
Heiko Carstens712555e2008-04-28 11:33:07 +02006580 mutex_lock(&sched_domains_mutex);
Srivatsa Vaddagiria1835612008-01-25 21:08:00 +01006581
Milton Miller73785472007-10-24 18:23:48 +02006582 /* always unregister in case we don't destroy any domains */
6583 unregister_sched_domain_sysctl();
6584
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01006585 /* Let architecture update cpu core mappings. */
6586 new_topology = arch_update_cpu_topology();
6587
Max Krasnyanskydfb512e2008-08-29 13:11:41 -07006588 n = doms_new ? ndoms_new : 0;
Paul Jackson029190c2007-10-18 23:40:20 -07006589
6590 /* Destroy deleted domains */
6591 for (i = 0; i < ndoms_cur; i++) {
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01006592 for (j = 0; j < n && !new_topology; j++) {
Rusty Russellacc3f5d2009-11-03 14:53:40 +10306593 if (cpumask_equal(doms_cur[i], doms_new[j])
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006594 && dattrs_equal(dattr_cur, i, dattr_new, j))
Paul Jackson029190c2007-10-18 23:40:20 -07006595 goto match1;
6596 }
6597 /* no match - a current sched domain not in new doms_new[] */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10306598 detach_destroy_domains(doms_cur[i]);
Paul Jackson029190c2007-10-18 23:40:20 -07006599match1:
6600 ;
6601 }
6602
Max Krasnyanskye761b772008-07-15 04:43:49 -07006603 if (doms_new == NULL) {
6604 ndoms_cur = 0;
Rusty Russellacc3f5d2009-11-03 14:53:40 +10306605 doms_new = &fallback_doms;
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01006606 cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
Li Zefanfaa2f982008-11-04 16:20:23 +08006607 WARN_ON_ONCE(dattr_new);
Max Krasnyanskye761b772008-07-15 04:43:49 -07006608 }
6609
Paul Jackson029190c2007-10-18 23:40:20 -07006610 /* Build new domains */
6611 for (i = 0; i < ndoms_new; i++) {
Heiko Carstensd65bd5e2008-12-09 18:49:51 +01006612 for (j = 0; j < ndoms_cur && !new_topology; j++) {
Rusty Russellacc3f5d2009-11-03 14:53:40 +10306613 if (cpumask_equal(doms_new[i], doms_cur[j])
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006614 && dattrs_equal(dattr_new, i, dattr_cur, j))
Paul Jackson029190c2007-10-18 23:40:20 -07006615 goto match2;
6616 }
6617 /* no match - add a new doms_new */
Peter Zijlstradce840a2011-04-07 14:09:50 +02006618 build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL);
Paul Jackson029190c2007-10-18 23:40:20 -07006619match2:
6620 ;
6621 }
6622
6623 /* Remember the new sched domains */
Rusty Russellacc3f5d2009-11-03 14:53:40 +10306624 if (doms_cur != &fallback_doms)
6625 free_sched_domains(doms_cur, ndoms_cur);
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006626 kfree(dattr_cur); /* kfree(NULL) is safe */
Paul Jackson029190c2007-10-18 23:40:20 -07006627 doms_cur = doms_new;
Hidetoshi Seto1d3504f2008-04-15 14:04:23 +09006628 dattr_cur = dattr_new;
Paul Jackson029190c2007-10-18 23:40:20 -07006629 ndoms_cur = ndoms_new;
Milton Miller73785472007-10-24 18:23:48 +02006630
6631 register_sched_domain_sysctl();
Srivatsa Vaddagiria1835612008-01-25 21:08:00 +01006632
Heiko Carstens712555e2008-04-28 11:33:07 +02006633 mutex_unlock(&sched_domains_mutex);
Paul Jackson029190c2007-10-18 23:40:20 -07006634}
6635
Srivatsa S. Bhatd35be8b2012-05-24 19:46:26 +05306636static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */
6637
Linus Torvalds1da177e2005-04-16 15:20:36 -07006638/*
Tejun Heo3a101d02010-06-08 21:40:36 +02006639 * Update cpusets according to cpu_active mask. If cpusets are
6640 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
6641 * around partition_sched_domains().
Srivatsa S. Bhatd35be8b2012-05-24 19:46:26 +05306642 *
6643 * If we come here as part of a suspend/resume, don't touch cpusets because we
6644 * want to restore it back to its original state upon resume anyway.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006645 */
Tejun Heo0b2e9182010-06-21 23:53:31 +02006646static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action,
6647 void *hcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006648{
Srivatsa S. Bhatd35be8b2012-05-24 19:46:26 +05306649 switch (action) {
6650 case CPU_ONLINE_FROZEN:
6651 case CPU_DOWN_FAILED_FROZEN:
6652
6653 /*
6654 * num_cpus_frozen tracks how many CPUs are involved in suspend
6655 * resume sequence. As long as this is not the last online
6656 * operation in the resume sequence, just build a single sched
6657 * domain, ignoring cpusets.
6658 */
6659 num_cpus_frozen--;
6660 if (likely(num_cpus_frozen)) {
6661 partition_sched_domains(1, NULL, NULL);
6662 break;
6663 }
6664
6665 /*
6666 * This is the last CPU online operation. So fall through and
6667 * restore the original sched domains by considering the
6668 * cpuset configurations.
6669 */
6670
Max Krasnyanskye761b772008-07-15 04:43:49 -07006671 case CPU_ONLINE:
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01006672 case CPU_DOWN_FAILED:
Srivatsa S. Bhat7ddf96b2012-05-24 19:46:55 +05306673 cpuset_update_active_cpus(true);
Srivatsa S. Bhatd35be8b2012-05-24 19:46:26 +05306674 break;
Max Krasnyanskye761b772008-07-15 04:43:49 -07006675 default:
6676 return NOTIFY_DONE;
6677 }
Srivatsa S. Bhatd35be8b2012-05-24 19:46:26 +05306678 return NOTIFY_OK;
Max Krasnyanskye761b772008-07-15 04:43:49 -07006679}
Tejun Heo3a101d02010-06-08 21:40:36 +02006680
Tejun Heo0b2e9182010-06-21 23:53:31 +02006681static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action,
6682 void *hcpu)
Tejun Heo3a101d02010-06-08 21:40:36 +02006683{
Srivatsa S. Bhatd35be8b2012-05-24 19:46:26 +05306684 switch (action) {
Tejun Heo3a101d02010-06-08 21:40:36 +02006685 case CPU_DOWN_PREPARE:
Srivatsa S. Bhat7ddf96b2012-05-24 19:46:55 +05306686 cpuset_update_active_cpus(false);
Srivatsa S. Bhatd35be8b2012-05-24 19:46:26 +05306687 break;
6688 case CPU_DOWN_PREPARE_FROZEN:
6689 num_cpus_frozen++;
6690 partition_sched_domains(1, NULL, NULL);
6691 break;
Tejun Heo3a101d02010-06-08 21:40:36 +02006692 default:
6693 return NOTIFY_DONE;
6694 }
Srivatsa S. Bhatd35be8b2012-05-24 19:46:26 +05306695 return NOTIFY_OK;
Tejun Heo3a101d02010-06-08 21:40:36 +02006696}
Max Krasnyanskye761b772008-07-15 04:43:49 -07006697
Linus Torvalds1da177e2005-04-16 15:20:36 -07006698void __init sched_init_smp(void)
6699{
Rusty Russelldcc30a32008-11-25 02:35:12 +10306700 cpumask_var_t non_isolated_cpus;
6701
6702 alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
Yong Zhangcb5fd132009-09-14 20:20:16 +08006703 alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
Nick Piggin5c1e1762006-10-03 01:14:04 -07006704
Peter Zijlstracb83b622012-04-17 15:49:36 +02006705 sched_init_numa();
6706
Gautham R Shenoy95402b32008-01-25 21:08:02 +01006707 get_online_cpus();
Heiko Carstens712555e2008-04-28 11:33:07 +02006708 mutex_lock(&sched_domains_mutex);
Peter Zijlstrac4a88492011-04-07 14:09:42 +02006709 init_sched_domains(cpu_active_mask);
Rusty Russelldcc30a32008-11-25 02:35:12 +10306710 cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map);
6711 if (cpumask_empty(non_isolated_cpus))
6712 cpumask_set_cpu(smp_processor_id(), non_isolated_cpus);
Heiko Carstens712555e2008-04-28 11:33:07 +02006713 mutex_unlock(&sched_domains_mutex);
Gautham R Shenoy95402b32008-01-25 21:08:02 +01006714 put_online_cpus();
Max Krasnyanskye761b772008-07-15 04:43:49 -07006715
Tang Chen301a5cb2012-09-25 21:12:31 +08006716 hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE);
Tejun Heo3a101d02010-06-08 21:40:36 +02006717 hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE);
6718 hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE);
Max Krasnyanskye761b772008-07-15 04:43:49 -07006719
6720 /* RT runtime code needs to handle some hotplug events */
6721 hotcpu_notifier(update_runtime, 0);
6722
Peter Zijlstrab328ca12008-04-29 10:02:46 +02006723 init_hrtick();
Nick Piggin5c1e1762006-10-03 01:14:04 -07006724
6725 /* Move init over to a non-isolated CPU */
Rusty Russelldcc30a32008-11-25 02:35:12 +10306726 if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
Nick Piggin5c1e1762006-10-03 01:14:04 -07006727 BUG();
Ingo Molnar19978ca2007-11-09 22:39:38 +01006728 sched_init_granularity();
Rusty Russelldcc30a32008-11-25 02:35:12 +10306729 free_cpumask_var(non_isolated_cpus);
Rusty Russell42128232008-11-25 02:35:12 +10306730
Rusty Russell0e3900e2008-11-25 02:35:13 +10306731 init_sched_rt_class();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006732}
6733#else
6734void __init sched_init_smp(void)
6735{
Ingo Molnar19978ca2007-11-09 22:39:38 +01006736 sched_init_granularity();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006737}
6738#endif /* CONFIG_SMP */
6739
Arun R Bharadwajcd1bb942009-04-16 12:15:34 +05306740const_debug unsigned int sysctl_timer_migration = 1;
6741
Linus Torvalds1da177e2005-04-16 15:20:36 -07006742int in_sched_functions(unsigned long addr)
6743{
Linus Torvalds1da177e2005-04-16 15:20:36 -07006744 return in_lock_functions(addr) ||
6745 (addr >= (unsigned long)__sched_text_start
6746 && addr < (unsigned long)__sched_text_end);
6747}
6748
Peter Zijlstra029632f2011-10-25 10:00:11 +02006749#ifdef CONFIG_CGROUP_SCHED
6750struct task_group root_task_group;
Mike Galbraith35cf4e52012-08-07 05:00:13 +02006751LIST_HEAD(task_groups);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01006752#endif
6753
Peter Zijlstra029632f2011-10-25 10:00:11 +02006754DECLARE_PER_CPU(cpumask_var_t, load_balance_tmpmask);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01006755
Linus Torvalds1da177e2005-04-16 15:20:36 -07006756void __init sched_init(void)
6757{
Ingo Molnardd41f592007-07-09 18:51:59 +02006758 int i, j;
Mike Travis434d53b2008-04-04 18:11:04 -07006759 unsigned long alloc_size = 0, ptr;
6760
6761#ifdef CONFIG_FAIR_GROUP_SCHED
6762 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
6763#endif
6764#ifdef CONFIG_RT_GROUP_SCHED
6765 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
6766#endif
Rusty Russelldf7c8e82009-03-19 15:22:20 +10306767#ifdef CONFIG_CPUMASK_OFFSTACK
Rusty Russell8c083f02009-03-19 15:22:20 +10306768 alloc_size += num_possible_cpus() * cpumask_size();
Rusty Russelldf7c8e82009-03-19 15:22:20 +10306769#endif
Mike Travis434d53b2008-04-04 18:11:04 -07006770 if (alloc_size) {
Pekka Enberg36b7b6d2009-06-10 23:42:36 +03006771 ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
Mike Travis434d53b2008-04-04 18:11:04 -07006772
6773#ifdef CONFIG_FAIR_GROUP_SCHED
Yong Zhang07e06b02011-01-07 15:17:36 +08006774 root_task_group.se = (struct sched_entity **)ptr;
Mike Travis434d53b2008-04-04 18:11:04 -07006775 ptr += nr_cpu_ids * sizeof(void **);
6776
Yong Zhang07e06b02011-01-07 15:17:36 +08006777 root_task_group.cfs_rq = (struct cfs_rq **)ptr;
Mike Travis434d53b2008-04-04 18:11:04 -07006778 ptr += nr_cpu_ids * sizeof(void **);
Peter Zijlstraeff766a2008-04-19 19:45:00 +02006779
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006780#endif /* CONFIG_FAIR_GROUP_SCHED */
Mike Travis434d53b2008-04-04 18:11:04 -07006781#ifdef CONFIG_RT_GROUP_SCHED
Yong Zhang07e06b02011-01-07 15:17:36 +08006782 root_task_group.rt_se = (struct sched_rt_entity **)ptr;
Mike Travis434d53b2008-04-04 18:11:04 -07006783 ptr += nr_cpu_ids * sizeof(void **);
6784
Yong Zhang07e06b02011-01-07 15:17:36 +08006785 root_task_group.rt_rq = (struct rt_rq **)ptr;
Peter Zijlstraeff766a2008-04-19 19:45:00 +02006786 ptr += nr_cpu_ids * sizeof(void **);
6787
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006788#endif /* CONFIG_RT_GROUP_SCHED */
Rusty Russelldf7c8e82009-03-19 15:22:20 +10306789#ifdef CONFIG_CPUMASK_OFFSTACK
6790 for_each_possible_cpu(i) {
6791 per_cpu(load_balance_tmpmask, i) = (void *)ptr;
6792 ptr += cpumask_size();
6793 }
6794#endif /* CONFIG_CPUMASK_OFFSTACK */
Mike Travis434d53b2008-04-04 18:11:04 -07006795 }
Ingo Molnardd41f592007-07-09 18:51:59 +02006796
Gregory Haskins57d885f2008-01-25 21:08:18 +01006797#ifdef CONFIG_SMP
6798 init_defrootdomain();
6799#endif
6800
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02006801 init_rt_bandwidth(&def_rt_bandwidth,
6802 global_rt_period(), global_rt_runtime());
6803
6804#ifdef CONFIG_RT_GROUP_SCHED
Yong Zhang07e06b02011-01-07 15:17:36 +08006805 init_rt_bandwidth(&root_task_group.rt_bandwidth,
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02006806 global_rt_period(), global_rt_runtime());
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02006807#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02006808
Dhaval Giani7c941432010-01-20 13:26:18 +01006809#ifdef CONFIG_CGROUP_SCHED
Yong Zhang07e06b02011-01-07 15:17:36 +08006810 list_add(&root_task_group.list, &task_groups);
6811 INIT_LIST_HEAD(&root_task_group.children);
Glauber Costaf4d6f6c2011-11-01 19:19:07 -02006812 INIT_LIST_HEAD(&root_task_group.siblings);
Mike Galbraith5091faa2010-11-30 14:18:03 +01006813 autogroup_init(&init_task);
Glauber Costa54c707e2011-11-28 14:45:19 -02006814
Dhaval Giani7c941432010-01-20 13:26:18 +01006815#endif /* CONFIG_CGROUP_SCHED */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01006816
Glauber Costa54c707e2011-11-28 14:45:19 -02006817#ifdef CONFIG_CGROUP_CPUACCT
6818 root_cpuacct.cpustat = &kernel_cpustat;
6819 root_cpuacct.cpuusage = alloc_percpu(u64);
6820 /* Too early, not expected to fail */
6821 BUG_ON(!root_cpuacct.cpuusage);
6822#endif
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08006823 for_each_possible_cpu(i) {
Ingo Molnar70b97a72006-07-03 00:25:42 -07006824 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006825
6826 rq = cpu_rq(i);
Thomas Gleixner05fa7852009-11-17 14:28:38 +01006827 raw_spin_lock_init(&rq->lock);
Nick Piggin78979862005-06-25 14:57:13 -07006828 rq->nr_running = 0;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02006829 rq->calc_load_active = 0;
6830 rq->calc_load_update = jiffies + LOAD_FREQ;
Jan H. Schönherracb5a9b2011-07-14 18:32:43 +02006831 init_cfs_rq(&rq->cfs);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +01006832 init_rt_rq(&rq->rt, rq);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01006833#ifdef CONFIG_FAIR_GROUP_SCHED
Peter Zijlstra029632f2011-10-25 10:00:11 +02006834 root_task_group.shares = ROOT_TASK_GROUP_LOAD;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01006835 INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
Dhaval Giani354d60c2008-04-19 19:44:59 +02006836 /*
Yong Zhang07e06b02011-01-07 15:17:36 +08006837 * How much cpu bandwidth does root_task_group get?
Dhaval Giani354d60c2008-04-19 19:44:59 +02006838 *
6839 * In case of task-groups formed thr' the cgroup filesystem, it
6840 * gets 100% of the cpu resources in the system. This overall
6841 * system cpu resource is divided among the tasks of
Yong Zhang07e06b02011-01-07 15:17:36 +08006842 * root_task_group and its child task-groups in a fair manner,
Dhaval Giani354d60c2008-04-19 19:44:59 +02006843 * based on each entity's (task or task-group's) weight
6844 * (se->load.weight).
6845 *
Yong Zhang07e06b02011-01-07 15:17:36 +08006846 * In other words, if root_task_group has 10 tasks of weight
Dhaval Giani354d60c2008-04-19 19:44:59 +02006847 * 1024) and two child groups A0 and A1 (of weight 1024 each),
6848 * then A0's share of the cpu resource is:
6849 *
Ingo Molnar0d905bc2009-05-04 19:13:30 +02006850 * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
Dhaval Giani354d60c2008-04-19 19:44:59 +02006851 *
Yong Zhang07e06b02011-01-07 15:17:36 +08006852 * We achieve this by letting root_task_group's tasks sit
6853 * directly in rq->cfs (i.e root_task_group->se[] = NULL).
Dhaval Giani354d60c2008-04-19 19:44:59 +02006854 */
Paul Turnerab84d312011-07-21 09:43:28 -07006855 init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
Yong Zhang07e06b02011-01-07 15:17:36 +08006856 init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
Dhaval Giani354d60c2008-04-19 19:44:59 +02006857#endif /* CONFIG_FAIR_GROUP_SCHED */
6858
6859 rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01006860#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01006861 INIT_LIST_HEAD(&rq->leaf_rt_rq_list);
Yong Zhang07e06b02011-01-07 15:17:36 +08006862 init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01006863#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07006864
Ingo Molnardd41f592007-07-09 18:51:59 +02006865 for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
6866 rq->cpu_load[j] = 0;
Venkatesh Pallipadifdf3e952010-05-17 18:14:43 -07006867
6868 rq->last_load_update_tick = jiffies;
6869
Linus Torvalds1da177e2005-04-16 15:20:36 -07006870#ifdef CONFIG_SMP
Nick Piggin41c7ce92005-06-25 14:57:24 -07006871 rq->sd = NULL;
Gregory Haskins57d885f2008-01-25 21:08:18 +01006872 rq->rd = NULL;
Nikhil Rao1399fa72011-05-18 10:09:39 -07006873 rq->cpu_power = SCHED_POWER_SCALE;
Gregory Haskins3f029d32009-07-29 11:08:47 -04006874 rq->post_schedule = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006875 rq->active_balance = 0;
Ingo Molnardd41f592007-07-09 18:51:59 +02006876 rq->next_balance = jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006877 rq->push_cpu = 0;
Christoph Lameter0a2966b2006-09-25 23:30:51 -07006878 rq->cpu = i;
Gregory Haskins1f11eb62008-06-04 15:04:05 -04006879 rq->online = 0;
Mike Galbraitheae0c9d2009-11-10 03:50:02 +01006880 rq->idle_stamp = 0;
6881 rq->avg_idle = 2*sysctl_sched_migration_cost;
Peter Zijlstra367456c2012-02-20 21:49:09 +01006882
6883 INIT_LIST_HEAD(&rq->cfs_tasks);
6884
Gregory Haskinsdc938522008-01-25 21:08:26 +01006885 rq_attach_root(rq, &def_root_domain);
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -07006886#ifdef CONFIG_NO_HZ
Suresh Siddha1c792db2011-12-01 17:07:32 -08006887 rq->nohz_flags = 0;
Venkatesh Pallipadi83cd4fe2010-05-21 17:09:41 -07006888#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07006889#endif
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01006890 init_rq_hrtick(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006891 atomic_set(&rq->nr_iowait, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006892 }
6893
Peter Williams2dd73a42006-06-27 02:54:34 -07006894 set_load_weight(&init_task);
Heiko Carstensb50f60c2006-07-30 03:03:52 -07006895
Avi Kivitye107be32007-07-26 13:40:43 +02006896#ifdef CONFIG_PREEMPT_NOTIFIERS
6897 INIT_HLIST_HEAD(&init_task.preempt_notifiers);
6898#endif
6899
Heiko Carstensb50f60c2006-07-30 03:03:52 -07006900#ifdef CONFIG_RT_MUTEXES
Dima Zavin732375c2011-07-07 17:27:59 -07006901 plist_head_init(&init_task.pi_waiters);
Heiko Carstensb50f60c2006-07-30 03:03:52 -07006902#endif
6903
Linus Torvalds1da177e2005-04-16 15:20:36 -07006904 /*
6905 * The boot idle thread does lazy MMU switching as well:
6906 */
6907 atomic_inc(&init_mm.mm_count);
6908 enter_lazy_tlb(&init_mm, current);
6909
6910 /*
6911 * Make us the idle thread. Technically, schedule() should not be
6912 * called from this thread, however somewhere below it might be,
6913 * but because we are the idle thread, we just pick up running again
6914 * when this runqueue becomes "idle".
6915 */
6916 init_idle(current, smp_processor_id());
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02006917
6918 calc_load_update = jiffies + LOAD_FREQ;
6919
Ingo Molnardd41f592007-07-09 18:51:59 +02006920 /*
6921 * During early bootup we pretend to be a normal task:
6922 */
6923 current->sched_class = &fair_sched_class;
Ingo Molnar6892b752008-02-13 14:02:36 +01006924
Rusty Russellbf4d83f2008-11-25 09:57:51 +10306925#ifdef CONFIG_SMP
Peter Zijlstra4cb98832011-04-07 14:09:58 +02006926 zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT);
Rusty Russellbdddd292009-12-02 14:09:16 +10306927 /* May be allocated at isolcpus cmdline parse time */
6928 if (cpu_isolated_map == NULL)
6929 zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT);
Thomas Gleixner29d5e042012-04-20 13:05:45 +00006930 idle_thread_set_boot_cpu();
Peter Zijlstra029632f2011-10-25 10:00:11 +02006931#endif
6932 init_sched_fair_class();
Rusty Russell6a7b3dc2008-11-25 02:35:04 +10306933
Ingo Molnar6892b752008-02-13 14:02:36 +01006934 scheduler_running = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006935}
6936
Frederic Weisbeckerd902db12011-06-08 19:31:56 +02006937#ifdef CONFIG_DEBUG_ATOMIC_SLEEP
Frederic Weisbeckere4aafea2009-07-16 15:44:29 +02006938static inline int preempt_count_equals(int preempt_offset)
6939{
Frederic Weisbecker234da7b2009-12-16 20:21:05 +01006940 int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth();
Frederic Weisbeckere4aafea2009-07-16 15:44:29 +02006941
Arnd Bergmann4ba82162011-01-25 22:52:22 +01006942 return (nested == preempt_offset);
Frederic Weisbeckere4aafea2009-07-16 15:44:29 +02006943}
6944
Simon Kagstromd8948372009-12-23 11:08:18 +01006945void __might_sleep(const char *file, int line, int preempt_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006946{
Linus Torvalds1da177e2005-04-16 15:20:36 -07006947 static unsigned long prev_jiffy; /* ratelimiting */
6948
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -07006949 rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */
Frederic Weisbeckere4aafea2009-07-16 15:44:29 +02006950 if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) ||
6951 system_state != SYSTEM_RUNNING || oops_in_progress)
Ingo Molnaraef745f2008-08-28 11:34:43 +02006952 return;
6953 if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
6954 return;
6955 prev_jiffy = jiffies;
6956
Peter Zijlstra3df0fc52009-12-20 14:23:57 +01006957 printk(KERN_ERR
6958 "BUG: sleeping function called from invalid context at %s:%d\n",
6959 file, line);
6960 printk(KERN_ERR
6961 "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n",
6962 in_atomic(), irqs_disabled(),
6963 current->pid, current->comm);
Ingo Molnaraef745f2008-08-28 11:34:43 +02006964
6965 debug_show_held_locks(current);
6966 if (irqs_disabled())
6967 print_irqtrace_events(current);
6968 dump_stack();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006969}
6970EXPORT_SYMBOL(__might_sleep);
6971#endif
6972
6973#ifdef CONFIG_MAGIC_SYSRQ
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02006974static void normalize_task(struct rq *rq, struct task_struct *p)
6975{
Peter Zijlstrada7a7352011-01-17 17:03:27 +01006976 const struct sched_class *prev_class = p->sched_class;
6977 int old_prio = p->prio;
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02006978 int on_rq;
Peter Zijlstra3e51f332008-05-03 18:29:28 +02006979
Peter Zijlstrafd2f4412011-04-05 17:23:44 +02006980 on_rq = p->on_rq;
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02006981 if (on_rq)
Peter Zijlstra4ca9b722012-01-25 11:50:51 +01006982 dequeue_task(rq, p, 0);
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02006983 __setscheduler(rq, p, SCHED_NORMAL, 0);
6984 if (on_rq) {
Peter Zijlstra4ca9b722012-01-25 11:50:51 +01006985 enqueue_task(rq, p, 0);
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02006986 resched_task(rq->curr);
6987 }
Peter Zijlstrada7a7352011-01-17 17:03:27 +01006988
6989 check_class_changed(rq, p, prev_class, old_prio);
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02006990}
6991
Linus Torvalds1da177e2005-04-16 15:20:36 -07006992void normalize_rt_tasks(void)
6993{
Ingo Molnara0f98a12007-06-17 18:37:45 +02006994 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006995 unsigned long flags;
Ingo Molnar70b97a72006-07-03 00:25:42 -07006996 struct rq *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006997
Peter Zijlstra4cf5d772008-02-13 15:45:39 +01006998 read_lock_irqsave(&tasklist_lock, flags);
Ingo Molnara0f98a12007-06-17 18:37:45 +02006999 do_each_thread(g, p) {
Ingo Molnar178be792007-10-15 17:00:18 +02007000 /*
7001 * Only normalize user tasks:
7002 */
7003 if (!p->mm)
7004 continue;
7005
Ingo Molnardd41f592007-07-09 18:51:59 +02007006 p->se.exec_start = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02007007#ifdef CONFIG_SCHEDSTATS
Lucas De Marchi41acab82010-03-10 23:37:45 -03007008 p->se.statistics.wait_start = 0;
7009 p->se.statistics.sleep_start = 0;
7010 p->se.statistics.block_start = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +02007011#endif
Ingo Molnardd41f592007-07-09 18:51:59 +02007012
7013 if (!rt_task(p)) {
7014 /*
7015 * Renice negative nice level userspace
7016 * tasks back to 0:
7017 */
7018 if (TASK_NICE(p) < 0 && p->mm)
7019 set_user_nice(p, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007020 continue;
Ingo Molnardd41f592007-07-09 18:51:59 +02007021 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007022
Thomas Gleixner1d615482009-11-17 14:54:03 +01007023 raw_spin_lock(&p->pi_lock);
Ingo Molnarb29739f2006-06-27 02:54:51 -07007024 rq = __task_rq_lock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007025
Ingo Molnar178be792007-10-15 17:00:18 +02007026 normalize_task(rq, p);
Andi Kleen3a5e4dc2007-10-15 17:00:15 +02007027
Ingo Molnarb29739f2006-06-27 02:54:51 -07007028 __task_rq_unlock(rq);
Thomas Gleixner1d615482009-11-17 14:54:03 +01007029 raw_spin_unlock(&p->pi_lock);
Ingo Molnara0f98a12007-06-17 18:37:45 +02007030 } while_each_thread(g, p);
7031
Peter Zijlstra4cf5d772008-02-13 15:45:39 +01007032 read_unlock_irqrestore(&tasklist_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007033}
7034
7035#endif /* CONFIG_MAGIC_SYSRQ */
Linus Torvalds1df5c102005-09-12 07:59:21 -07007036
Jason Wessel67fc4e02010-05-20 21:04:21 -05007037#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
Linus Torvalds1df5c102005-09-12 07:59:21 -07007038/*
Jason Wessel67fc4e02010-05-20 21:04:21 -05007039 * These functions are only useful for the IA64 MCA handling, or kdb.
Linus Torvalds1df5c102005-09-12 07:59:21 -07007040 *
7041 * They can only be called when the whole system has been
7042 * stopped - every CPU needs to be quiescent, and no scheduling
7043 * activity can take place. Using them for anything else would
7044 * be a serious bug, and as a result, they aren't even visible
7045 * under any other configuration.
7046 */
7047
7048/**
7049 * curr_task - return the current task for a given cpu.
7050 * @cpu: the processor in question.
7051 *
7052 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
7053 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07007054struct task_struct *curr_task(int cpu)
Linus Torvalds1df5c102005-09-12 07:59:21 -07007055{
7056 return cpu_curr(cpu);
7057}
7058
Jason Wessel67fc4e02010-05-20 21:04:21 -05007059#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */
7060
7061#ifdef CONFIG_IA64
Linus Torvalds1df5c102005-09-12 07:59:21 -07007062/**
7063 * set_curr_task - set the current task for a given cpu.
7064 * @cpu: the processor in question.
7065 * @p: the task pointer to set.
7066 *
7067 * Description: This function must only be used when non-maskable interrupts
Ingo Molnar41a2d6c2007-12-05 15:46:09 +01007068 * are serviced on a separate stack. It allows the architecture to switch the
7069 * notion of the current task on a cpu in a non-blocking manner. This function
Linus Torvalds1df5c102005-09-12 07:59:21 -07007070 * must be called with all CPU's synchronized, and interrupts disabled, the
7071 * and caller must save the original value of the current task (see
7072 * curr_task() above) and restore that value before reenabling interrupts and
7073 * re-starting the system.
7074 *
7075 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
7076 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07007077void set_curr_task(int cpu, struct task_struct *p)
Linus Torvalds1df5c102005-09-12 07:59:21 -07007078{
7079 cpu_curr(cpu) = p;
7080}
7081
7082#endif
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007083
Dhaval Giani7c941432010-01-20 13:26:18 +01007084#ifdef CONFIG_CGROUP_SCHED
Peter Zijlstra029632f2011-10-25 10:00:11 +02007085/* task_group_lock serializes the addition/removal of task groups */
7086static DEFINE_SPINLOCK(task_group_lock);
7087
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007088static void free_sched_group(struct task_group *tg)
7089{
7090 free_fair_sched_group(tg);
7091 free_rt_sched_group(tg);
Mike Galbraithe9aa1dd2011-01-05 11:11:25 +01007092 autogroup_free(tg);
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007093 kfree(tg);
7094}
7095
7096/* allocate runqueue etc for a new task group */
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007097struct task_group *sched_create_group(struct task_group *parent)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007098{
7099 struct task_group *tg;
7100 unsigned long flags;
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007101
7102 tg = kzalloc(sizeof(*tg), GFP_KERNEL);
7103 if (!tg)
7104 return ERR_PTR(-ENOMEM);
7105
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007106 if (!alloc_fair_sched_group(tg, parent))
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007107 goto err;
7108
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007109 if (!alloc_rt_sched_group(tg, parent))
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007110 goto err;
7111
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007112 spin_lock_irqsave(&task_group_lock, flags);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007113 list_add_rcu(&tg->list, &task_groups);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02007114
7115 WARN_ON(!parent); /* root should already exist */
7116
7117 tg->parent = parent;
Peter Zijlstraf473aa52008-04-19 19:45:00 +02007118 INIT_LIST_HEAD(&tg->children);
Zhang, Yanmin09f27242030-08-14 15:56:40 +08007119 list_add_rcu(&tg->siblings, &parent->children);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007120 spin_unlock_irqrestore(&task_group_lock, flags);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007121
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007122 return tg;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007123
7124err:
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007125 free_sched_group(tg);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007126 return ERR_PTR(-ENOMEM);
7127}
7128
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007129/* rcu callback to free various structures associated with a task group */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007130static void free_sched_group_rcu(struct rcu_head *rhp)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007131{
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007132 /* now it should be safe to free those cfs_rqs */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007133 free_sched_group(container_of(rhp, struct task_group, rcu));
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007134}
7135
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007136/* Destroy runqueue etc associated with a task group */
Ingo Molnar4cf86d72007-10-15 17:00:14 +02007137void sched_destroy_group(struct task_group *tg)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007138{
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007139 unsigned long flags;
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007140 int i;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007141
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -08007142 /* end participation in shares distribution */
7143 for_each_possible_cpu(i)
Peter Zijlstrabccbe082008-02-13 15:45:40 +01007144 unregister_fair_sched_group(tg, i);
Peter Zijlstra3d4b47b2010-11-15 15:47:01 -08007145
7146 spin_lock_irqsave(&task_group_lock, flags);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007147 list_del_rcu(&tg->list);
Peter Zijlstraf473aa52008-04-19 19:45:00 +02007148 list_del_rcu(&tg->siblings);
Peter Zijlstra8ed36992008-02-13 15:45:39 +01007149 spin_unlock_irqrestore(&task_group_lock, flags);
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007150
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007151 /* wait for possible concurrent references to cfs_rqs complete */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007152 call_rcu(&tg->rcu, free_sched_group_rcu);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007153}
7154
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007155/* change task's runqueue when it moves between groups.
Ingo Molnar3a252012007-10-15 17:00:12 +02007156 * The caller of this function should have put the task in its new group
7157 * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to
7158 * reflect its new group.
Srivatsa Vaddagiri9b5b7752007-10-15 17:00:09 +02007159 */
7160void sched_move_task(struct task_struct *tsk)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007161{
Peter Zijlstra8323f262012-06-22 13:36:05 +02007162 struct task_group *tg;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007163 int on_rq, running;
7164 unsigned long flags;
7165 struct rq *rq;
7166
7167 rq = task_rq_lock(tsk, &flags);
7168
Dmitry Adamushko051a1d12007-12-18 15:21:13 +01007169 running = task_current(rq, tsk);
Peter Zijlstrafd2f4412011-04-05 17:23:44 +02007170 on_rq = tsk->on_rq;
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007171
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07007172 if (on_rq)
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007173 dequeue_task(rq, tsk, 0);
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07007174 if (unlikely(running))
7175 tsk->sched_class->put_prev_task(rq, tsk);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007176
Peter Zijlstra8323f262012-06-22 13:36:05 +02007177 tg = container_of(task_subsys_state_check(tsk, cpu_cgroup_subsys_id,
7178 lockdep_is_held(&tsk->sighand->siglock)),
7179 struct task_group, css);
7180 tg = autogroup_task_group(tsk, tg);
7181 tsk->sched_task_group = tg;
7182
Peter Zijlstra810b3812008-02-29 15:21:01 -05007183#ifdef CONFIG_FAIR_GROUP_SCHED
Peter Zijlstrab2b5ce02010-10-15 15:24:15 +02007184 if (tsk->sched_class->task_move_group)
7185 tsk->sched_class->task_move_group(tsk, on_rq);
7186 else
Peter Zijlstra810b3812008-02-29 15:21:01 -05007187#endif
Peter Zijlstrab2b5ce02010-10-15 15:24:15 +02007188 set_task_rq(tsk, task_cpu(tsk));
Peter Zijlstra810b3812008-02-29 15:21:01 -05007189
Hiroshi Shimamoto0e1f3482008-03-10 11:01:20 -07007190 if (unlikely(running))
7191 tsk->sched_class->set_curr_task(rq);
7192 if (on_rq)
Peter Zijlstra371fd7e2010-03-24 16:38:48 +01007193 enqueue_task(rq, tsk, 0);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007194
Peter Zijlstra0122ec52011-04-05 17:23:51 +02007195 task_rq_unlock(rq, tsk, &flags);
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007196}
Dhaval Giani7c941432010-01-20 13:26:18 +01007197#endif /* CONFIG_CGROUP_SCHED */
Srivatsa Vaddagiri29f59db2007-10-15 17:00:07 +02007198
Paul Turnera790de92011-07-21 09:43:29 -07007199#if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH)
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007200static unsigned long to_ratio(u64 period, u64 runtime)
7201{
7202 if (runtime == RUNTIME_INF)
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02007203 return 1ULL << 20;
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007204
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02007205 return div64_u64(runtime << 20, period);
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007206}
Paul Turnera790de92011-07-21 09:43:29 -07007207#endif
7208
7209#ifdef CONFIG_RT_GROUP_SCHED
7210/*
7211 * Ensure that the real time constraints are schedulable.
7212 */
7213static DEFINE_MUTEX(rt_constraints_mutex);
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007214
Dhaval Giani521f1a242008-02-28 15:21:56 +05307215/* Must be called with tasklist_lock held */
7216static inline int tg_has_rt_tasks(struct task_group *tg)
7217{
7218 struct task_struct *g, *p;
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02007219
Dhaval Giani521f1a242008-02-28 15:21:56 +05307220 do_each_thread(g, p) {
Peter Zijlstra029632f2011-10-25 10:00:11 +02007221 if (rt_task(p) && task_rq(p)->rt.tg == tg)
Dhaval Giani521f1a242008-02-28 15:21:56 +05307222 return 1;
7223 } while_each_thread(g, p);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02007224
Dhaval Giani521f1a242008-02-28 15:21:56 +05307225 return 0;
7226}
7227
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02007228struct rt_schedulable_data {
7229 struct task_group *tg;
7230 u64 rt_period;
7231 u64 rt_runtime;
7232};
7233
Paul Turnera790de92011-07-21 09:43:29 -07007234static int tg_rt_schedulable(struct task_group *tg, void *data)
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02007235{
7236 struct rt_schedulable_data *d = data;
7237 struct task_group *child;
7238 unsigned long total, sum = 0;
7239 u64 period, runtime;
7240
7241 period = ktime_to_ns(tg->rt_bandwidth.rt_period);
7242 runtime = tg->rt_bandwidth.rt_runtime;
7243
7244 if (tg == d->tg) {
7245 period = d->rt_period;
7246 runtime = d->rt_runtime;
7247 }
7248
Peter Zijlstra4653f802008-09-23 15:33:44 +02007249 /*
7250 * Cannot have more runtime than the period.
7251 */
7252 if (runtime > period && runtime != RUNTIME_INF)
7253 return -EINVAL;
7254
7255 /*
7256 * Ensure we don't starve existing RT tasks.
7257 */
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02007258 if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
7259 return -EBUSY;
7260
7261 total = to_ratio(period, runtime);
7262
Peter Zijlstra4653f802008-09-23 15:33:44 +02007263 /*
7264 * Nobody can have more than the global setting allows.
7265 */
7266 if (total > to_ratio(global_rt_period(), global_rt_runtime()))
7267 return -EINVAL;
7268
7269 /*
7270 * The sum of our children's runtime should not exceed our own.
7271 */
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02007272 list_for_each_entry_rcu(child, &tg->children, siblings) {
7273 period = ktime_to_ns(child->rt_bandwidth.rt_period);
7274 runtime = child->rt_bandwidth.rt_runtime;
7275
7276 if (child == d->tg) {
7277 period = d->rt_period;
7278 runtime = d->rt_runtime;
7279 }
7280
7281 sum += to_ratio(period, runtime);
7282 }
7283
7284 if (sum > total)
7285 return -EINVAL;
7286
7287 return 0;
7288}
7289
7290static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
7291{
Paul Turner82774342011-07-21 09:43:35 -07007292 int ret;
7293
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02007294 struct rt_schedulable_data data = {
7295 .tg = tg,
7296 .rt_period = period,
7297 .rt_runtime = runtime,
7298 };
7299
Paul Turner82774342011-07-21 09:43:35 -07007300 rcu_read_lock();
7301 ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data);
7302 rcu_read_unlock();
7303
7304 return ret;
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02007305}
7306
Paul Turnerab84d312011-07-21 09:43:28 -07007307static int tg_set_rt_bandwidth(struct task_group *tg,
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007308 u64 rt_period, u64 rt_runtime)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007309{
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007310 int i, err = 0;
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007311
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007312 mutex_lock(&rt_constraints_mutex);
Dhaval Giani521f1a242008-02-28 15:21:56 +05307313 read_lock(&tasklist_lock);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02007314 err = __rt_schedulable(tg, rt_period, rt_runtime);
7315 if (err)
Dhaval Giani521f1a242008-02-28 15:21:56 +05307316 goto unlock;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007317
Thomas Gleixner0986b112009-11-17 15:32:06 +01007318 raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007319 tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
7320 tg->rt_bandwidth.rt_runtime = rt_runtime;
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007321
7322 for_each_possible_cpu(i) {
7323 struct rt_rq *rt_rq = tg->rt_rq[i];
7324
Thomas Gleixner0986b112009-11-17 15:32:06 +01007325 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007326 rt_rq->rt_runtime = rt_runtime;
Thomas Gleixner0986b112009-11-17 15:32:06 +01007327 raw_spin_unlock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007328 }
Thomas Gleixner0986b112009-11-17 15:32:06 +01007329 raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
Peter Zijlstra49246272010-10-17 21:46:10 +02007330unlock:
Dhaval Giani521f1a242008-02-28 15:21:56 +05307331 read_unlock(&tasklist_lock);
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007332 mutex_unlock(&rt_constraints_mutex);
7333
7334 return err;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007335}
7336
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007337int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
7338{
7339 u64 rt_runtime, rt_period;
7340
7341 rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
7342 rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
7343 if (rt_runtime_us < 0)
7344 rt_runtime = RUNTIME_INF;
7345
Paul Turnerab84d312011-07-21 09:43:28 -07007346 return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007347}
7348
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007349long sched_group_rt_runtime(struct task_group *tg)
7350{
7351 u64 rt_runtime_us;
7352
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007353 if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007354 return -1;
7355
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007356 rt_runtime_us = tg->rt_bandwidth.rt_runtime;
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007357 do_div(rt_runtime_us, NSEC_PER_USEC);
7358 return rt_runtime_us;
7359}
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007360
7361int sched_group_set_rt_period(struct task_group *tg, long rt_period_us)
7362{
7363 u64 rt_runtime, rt_period;
7364
7365 rt_period = (u64)rt_period_us * NSEC_PER_USEC;
7366 rt_runtime = tg->rt_bandwidth.rt_runtime;
7367
Raistlin619b0482008-06-26 18:54:09 +02007368 if (rt_period == 0)
7369 return -EINVAL;
7370
Paul Turnerab84d312011-07-21 09:43:28 -07007371 return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007372}
7373
7374long sched_group_rt_period(struct task_group *tg)
7375{
7376 u64 rt_period_us;
7377
7378 rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
7379 do_div(rt_period_us, NSEC_PER_USEC);
7380 return rt_period_us;
7381}
7382
7383static int sched_rt_global_constraints(void)
7384{
Peter Zijlstra4653f802008-09-23 15:33:44 +02007385 u64 runtime, period;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007386 int ret = 0;
7387
Hiroshi Shimamotoec5d4982008-09-10 17:00:19 -07007388 if (sysctl_sched_rt_period <= 0)
7389 return -EINVAL;
7390
Peter Zijlstra4653f802008-09-23 15:33:44 +02007391 runtime = global_rt_runtime();
7392 period = global_rt_period();
7393
7394 /*
7395 * Sanity check on the sysctl variables.
7396 */
7397 if (runtime > period && runtime != RUNTIME_INF)
7398 return -EINVAL;
Peter Zijlstra10b612f2008-06-19 14:22:27 +02007399
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007400 mutex_lock(&rt_constraints_mutex);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02007401 read_lock(&tasklist_lock);
Peter Zijlstra4653f802008-09-23 15:33:44 +02007402 ret = __rt_schedulable(NULL, 0, 0);
Peter Zijlstra9a7e0b12008-08-19 12:33:06 +02007403 read_unlock(&tasklist_lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007404 mutex_unlock(&rt_constraints_mutex);
7405
7406 return ret;
7407}
Dhaval Giani54e99122009-02-27 15:13:54 +05307408
7409int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
7410{
7411 /* Don't accept realtime tasks when there is no way for them to run */
7412 if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0)
7413 return 0;
7414
7415 return 1;
7416}
7417
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007418#else /* !CONFIG_RT_GROUP_SCHED */
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007419static int sched_rt_global_constraints(void)
7420{
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007421 unsigned long flags;
7422 int i;
7423
Hiroshi Shimamotoec5d4982008-09-10 17:00:19 -07007424 if (sysctl_sched_rt_period <= 0)
7425 return -EINVAL;
7426
Peter Zijlstra60aa6052009-05-05 17:50:21 +02007427 /*
7428 * There's always some RT tasks in the root group
7429 * -- migration, kstopmachine etc..
7430 */
7431 if (sysctl_sched_rt_runtime == 0)
7432 return -EBUSY;
7433
Thomas Gleixner0986b112009-11-17 15:32:06 +01007434 raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007435 for_each_possible_cpu(i) {
7436 struct rt_rq *rt_rq = &cpu_rq(i)->rt;
7437
Thomas Gleixner0986b112009-11-17 15:32:06 +01007438 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007439 rt_rq->rt_runtime = global_rt_runtime();
Thomas Gleixner0986b112009-11-17 15:32:06 +01007440 raw_spin_unlock(&rt_rq->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007441 }
Thomas Gleixner0986b112009-11-17 15:32:06 +01007442 raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
Peter Zijlstraac086bc2008-04-19 19:44:58 +02007443
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007444 return 0;
7445}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007446#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007447
7448int sched_rt_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07007449 void __user *buffer, size_t *lenp,
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007450 loff_t *ppos)
7451{
7452 int ret;
7453 int old_period, old_runtime;
7454 static DEFINE_MUTEX(mutex);
7455
7456 mutex_lock(&mutex);
7457 old_period = sysctl_sched_rt_period;
7458 old_runtime = sysctl_sched_rt_runtime;
7459
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07007460 ret = proc_dointvec(table, write, buffer, lenp, ppos);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007461
7462 if (!ret && write) {
7463 ret = sched_rt_global_constraints();
7464 if (ret) {
7465 sysctl_sched_rt_period = old_period;
7466 sysctl_sched_rt_runtime = old_runtime;
7467 } else {
7468 def_rt_bandwidth.rt_runtime = global_rt_runtime();
7469 def_rt_bandwidth.rt_period =
7470 ns_to_ktime(global_rt_period());
7471 }
7472 }
7473 mutex_unlock(&mutex);
7474
7475 return ret;
7476}
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007477
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007478#ifdef CONFIG_CGROUP_SCHED
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007479
7480/* return corresponding task_group object of a cgroup */
Paul Menage2b01dfe2007-10-24 18:23:50 +02007481static inline struct task_group *cgroup_tg(struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007482{
Paul Menage2b01dfe2007-10-24 18:23:50 +02007483 return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id),
7484 struct task_group, css);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007485}
7486
Tejun Heo92fb9742012-11-19 08:13:38 -08007487static struct cgroup_subsys_state *cpu_cgroup_css_alloc(struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007488{
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007489 struct task_group *tg, *parent;
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007490
Paul Menage2b01dfe2007-10-24 18:23:50 +02007491 if (!cgrp->parent) {
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007492 /* This is early initialization for the top cgroup */
Yong Zhang07e06b02011-01-07 15:17:36 +08007493 return &root_task_group.css;
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007494 }
7495
Dhaval Gianiec7dc8a2008-04-19 19:44:59 +02007496 parent = cgroup_tg(cgrp->parent);
7497 tg = sched_create_group(parent);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007498 if (IS_ERR(tg))
7499 return ERR_PTR(-ENOMEM);
7500
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007501 return &tg->css;
7502}
7503
Tejun Heo92fb9742012-11-19 08:13:38 -08007504static void cpu_cgroup_css_free(struct cgroup *cgrp)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007505{
Paul Menage2b01dfe2007-10-24 18:23:50 +02007506 struct task_group *tg = cgroup_tg(cgrp);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007507
7508 sched_destroy_group(tg);
7509}
7510
Li Zefan761b3ef2012-01-31 13:47:36 +08007511static int cpu_cgroup_can_attach(struct cgroup *cgrp,
Tejun Heobb9d97b2011-12-12 18:12:21 -08007512 struct cgroup_taskset *tset)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007513{
Tejun Heobb9d97b2011-12-12 18:12:21 -08007514 struct task_struct *task;
7515
7516 cgroup_taskset_for_each(task, cgrp, tset) {
Peter Zijlstrab68aa232008-02-13 15:45:40 +01007517#ifdef CONFIG_RT_GROUP_SCHED
Tejun Heobb9d97b2011-12-12 18:12:21 -08007518 if (!sched_rt_can_attach(cgroup_tg(cgrp), task))
7519 return -EINVAL;
Peter Zijlstrab68aa232008-02-13 15:45:40 +01007520#else
Tejun Heobb9d97b2011-12-12 18:12:21 -08007521 /* We don't support RT-tasks being in separate groups */
7522 if (task->sched_class != &fair_sched_class)
7523 return -EINVAL;
Peter Zijlstrab68aa232008-02-13 15:45:40 +01007524#endif
Tejun Heobb9d97b2011-12-12 18:12:21 -08007525 }
Ben Blumbe367d02009-09-23 15:56:31 -07007526 return 0;
7527}
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007528
Li Zefan761b3ef2012-01-31 13:47:36 +08007529static void cpu_cgroup_attach(struct cgroup *cgrp,
Tejun Heobb9d97b2011-12-12 18:12:21 -08007530 struct cgroup_taskset *tset)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007531{
Tejun Heobb9d97b2011-12-12 18:12:21 -08007532 struct task_struct *task;
7533
7534 cgroup_taskset_for_each(task, cgrp, tset)
7535 sched_move_task(task);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007536}
7537
Peter Zijlstra068c5cc2011-01-19 12:26:11 +01007538static void
Li Zefan761b3ef2012-01-31 13:47:36 +08007539cpu_cgroup_exit(struct cgroup *cgrp, struct cgroup *old_cgrp,
7540 struct task_struct *task)
Peter Zijlstra068c5cc2011-01-19 12:26:11 +01007541{
7542 /*
7543 * cgroup_exit() is called in the copy_process() failure path.
7544 * Ignore this case since the task hasn't ran yet, this avoids
7545 * trying to poke a half freed task state from generic code.
7546 */
7547 if (!(task->flags & PF_EXITING))
7548 return;
7549
7550 sched_move_task(task);
7551}
7552
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007553#ifdef CONFIG_FAIR_GROUP_SCHED
Paul Menagef4c753b2008-04-29 00:59:56 -07007554static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype,
Paul Menage2b01dfe2007-10-24 18:23:50 +02007555 u64 shareval)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007556{
Nikhil Raoc8b28112011-05-18 14:37:48 -07007557 return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval));
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007558}
7559
Paul Menagef4c753b2008-04-29 00:59:56 -07007560static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft)
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007561{
Paul Menage2b01dfe2007-10-24 18:23:50 +02007562 struct task_group *tg = cgroup_tg(cgrp);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007563
Nikhil Raoc8b28112011-05-18 14:37:48 -07007564 return (u64) scale_load_down(tg->shares);
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007565}
Paul Turnerab84d312011-07-21 09:43:28 -07007566
7567#ifdef CONFIG_CFS_BANDWIDTH
Paul Turnera790de92011-07-21 09:43:29 -07007568static DEFINE_MUTEX(cfs_constraints_mutex);
7569
Paul Turnerab84d312011-07-21 09:43:28 -07007570const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
7571const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */
7572
Paul Turnera790de92011-07-21 09:43:29 -07007573static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);
7574
Paul Turnerab84d312011-07-21 09:43:28 -07007575static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
7576{
Paul Turner56f570e2011-11-07 20:26:33 -08007577 int i, ret = 0, runtime_enabled, runtime_was_enabled;
Peter Zijlstra029632f2011-10-25 10:00:11 +02007578 struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
Paul Turnerab84d312011-07-21 09:43:28 -07007579
7580 if (tg == &root_task_group)
7581 return -EINVAL;
7582
7583 /*
7584 * Ensure we have at some amount of bandwidth every period. This is
7585 * to prevent reaching a state of large arrears when throttled via
7586 * entity_tick() resulting in prolonged exit starvation.
7587 */
7588 if (quota < min_cfs_quota_period || period < min_cfs_quota_period)
7589 return -EINVAL;
7590
7591 /*
7592 * Likewise, bound things on the otherside by preventing insane quota
7593 * periods. This also allows us to normalize in computing quota
7594 * feasibility.
7595 */
7596 if (period > max_cfs_quota_period)
7597 return -EINVAL;
7598
Paul Turnera790de92011-07-21 09:43:29 -07007599 mutex_lock(&cfs_constraints_mutex);
7600 ret = __cfs_schedulable(tg, period, quota);
7601 if (ret)
7602 goto out_unlock;
7603
Paul Turner58088ad2011-07-21 09:43:31 -07007604 runtime_enabled = quota != RUNTIME_INF;
Paul Turner56f570e2011-11-07 20:26:33 -08007605 runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
7606 account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled);
Paul Turnerab84d312011-07-21 09:43:28 -07007607 raw_spin_lock_irq(&cfs_b->lock);
7608 cfs_b->period = ns_to_ktime(period);
7609 cfs_b->quota = quota;
Paul Turner58088ad2011-07-21 09:43:31 -07007610
Paul Turnera9cf55b2011-07-21 09:43:32 -07007611 __refill_cfs_bandwidth_runtime(cfs_b);
Paul Turner58088ad2011-07-21 09:43:31 -07007612 /* restart the period timer (if active) to handle new period expiry */
7613 if (runtime_enabled && cfs_b->timer_active) {
7614 /* force a reprogram */
7615 cfs_b->timer_active = 0;
7616 __start_cfs_bandwidth(cfs_b);
7617 }
Paul Turnerab84d312011-07-21 09:43:28 -07007618 raw_spin_unlock_irq(&cfs_b->lock);
7619
7620 for_each_possible_cpu(i) {
7621 struct cfs_rq *cfs_rq = tg->cfs_rq[i];
Peter Zijlstra029632f2011-10-25 10:00:11 +02007622 struct rq *rq = cfs_rq->rq;
Paul Turnerab84d312011-07-21 09:43:28 -07007623
7624 raw_spin_lock_irq(&rq->lock);
Paul Turner58088ad2011-07-21 09:43:31 -07007625 cfs_rq->runtime_enabled = runtime_enabled;
Paul Turnerab84d312011-07-21 09:43:28 -07007626 cfs_rq->runtime_remaining = 0;
Paul Turner671fd9d2011-07-21 09:43:34 -07007627
Peter Zijlstra029632f2011-10-25 10:00:11 +02007628 if (cfs_rq->throttled)
Paul Turner671fd9d2011-07-21 09:43:34 -07007629 unthrottle_cfs_rq(cfs_rq);
Paul Turnerab84d312011-07-21 09:43:28 -07007630 raw_spin_unlock_irq(&rq->lock);
7631 }
Paul Turnera790de92011-07-21 09:43:29 -07007632out_unlock:
7633 mutex_unlock(&cfs_constraints_mutex);
Paul Turnerab84d312011-07-21 09:43:28 -07007634
Paul Turnera790de92011-07-21 09:43:29 -07007635 return ret;
Paul Turnerab84d312011-07-21 09:43:28 -07007636}
7637
7638int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us)
7639{
7640 u64 quota, period;
7641
Peter Zijlstra029632f2011-10-25 10:00:11 +02007642 period = ktime_to_ns(tg->cfs_bandwidth.period);
Paul Turnerab84d312011-07-21 09:43:28 -07007643 if (cfs_quota_us < 0)
7644 quota = RUNTIME_INF;
7645 else
7646 quota = (u64)cfs_quota_us * NSEC_PER_USEC;
7647
7648 return tg_set_cfs_bandwidth(tg, period, quota);
7649}
7650
7651long tg_get_cfs_quota(struct task_group *tg)
7652{
7653 u64 quota_us;
7654
Peter Zijlstra029632f2011-10-25 10:00:11 +02007655 if (tg->cfs_bandwidth.quota == RUNTIME_INF)
Paul Turnerab84d312011-07-21 09:43:28 -07007656 return -1;
7657
Peter Zijlstra029632f2011-10-25 10:00:11 +02007658 quota_us = tg->cfs_bandwidth.quota;
Paul Turnerab84d312011-07-21 09:43:28 -07007659 do_div(quota_us, NSEC_PER_USEC);
7660
7661 return quota_us;
7662}
7663
7664int tg_set_cfs_period(struct task_group *tg, long cfs_period_us)
7665{
7666 u64 quota, period;
7667
7668 period = (u64)cfs_period_us * NSEC_PER_USEC;
Peter Zijlstra029632f2011-10-25 10:00:11 +02007669 quota = tg->cfs_bandwidth.quota;
Paul Turnerab84d312011-07-21 09:43:28 -07007670
Paul Turnerab84d312011-07-21 09:43:28 -07007671 return tg_set_cfs_bandwidth(tg, period, quota);
7672}
7673
7674long tg_get_cfs_period(struct task_group *tg)
7675{
7676 u64 cfs_period_us;
7677
Peter Zijlstra029632f2011-10-25 10:00:11 +02007678 cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
Paul Turnerab84d312011-07-21 09:43:28 -07007679 do_div(cfs_period_us, NSEC_PER_USEC);
7680
7681 return cfs_period_us;
7682}
7683
7684static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft)
7685{
7686 return tg_get_cfs_quota(cgroup_tg(cgrp));
7687}
7688
7689static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype,
7690 s64 cfs_quota_us)
7691{
7692 return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us);
7693}
7694
7695static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft)
7696{
7697 return tg_get_cfs_period(cgroup_tg(cgrp));
7698}
7699
7700static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype,
7701 u64 cfs_period_us)
7702{
7703 return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us);
7704}
7705
Paul Turnera790de92011-07-21 09:43:29 -07007706struct cfs_schedulable_data {
7707 struct task_group *tg;
7708 u64 period, quota;
7709};
7710
7711/*
7712 * normalize group quota/period to be quota/max_period
7713 * note: units are usecs
7714 */
7715static u64 normalize_cfs_quota(struct task_group *tg,
7716 struct cfs_schedulable_data *d)
7717{
7718 u64 quota, period;
7719
7720 if (tg == d->tg) {
7721 period = d->period;
7722 quota = d->quota;
7723 } else {
7724 period = tg_get_cfs_period(tg);
7725 quota = tg_get_cfs_quota(tg);
7726 }
7727
7728 /* note: these should typically be equivalent */
7729 if (quota == RUNTIME_INF || quota == -1)
7730 return RUNTIME_INF;
7731
7732 return to_ratio(period, quota);
7733}
7734
7735static int tg_cfs_schedulable_down(struct task_group *tg, void *data)
7736{
7737 struct cfs_schedulable_data *d = data;
Peter Zijlstra029632f2011-10-25 10:00:11 +02007738 struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
Paul Turnera790de92011-07-21 09:43:29 -07007739 s64 quota = 0, parent_quota = -1;
7740
7741 if (!tg->parent) {
7742 quota = RUNTIME_INF;
7743 } else {
Peter Zijlstra029632f2011-10-25 10:00:11 +02007744 struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
Paul Turnera790de92011-07-21 09:43:29 -07007745
7746 quota = normalize_cfs_quota(tg, d);
7747 parent_quota = parent_b->hierarchal_quota;
7748
7749 /*
7750 * ensure max(child_quota) <= parent_quota, inherit when no
7751 * limit is set
7752 */
7753 if (quota == RUNTIME_INF)
7754 quota = parent_quota;
7755 else if (parent_quota != RUNTIME_INF && quota > parent_quota)
7756 return -EINVAL;
7757 }
7758 cfs_b->hierarchal_quota = quota;
7759
7760 return 0;
7761}
7762
7763static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
7764{
Paul Turner82774342011-07-21 09:43:35 -07007765 int ret;
Paul Turnera790de92011-07-21 09:43:29 -07007766 struct cfs_schedulable_data data = {
7767 .tg = tg,
7768 .period = period,
7769 .quota = quota,
7770 };
7771
7772 if (quota != RUNTIME_INF) {
7773 do_div(data.period, NSEC_PER_USEC);
7774 do_div(data.quota, NSEC_PER_USEC);
7775 }
7776
Paul Turner82774342011-07-21 09:43:35 -07007777 rcu_read_lock();
7778 ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
7779 rcu_read_unlock();
7780
7781 return ret;
Paul Turnera790de92011-07-21 09:43:29 -07007782}
Nikhil Raoe8da1b12011-07-21 09:43:40 -07007783
7784static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft,
7785 struct cgroup_map_cb *cb)
7786{
7787 struct task_group *tg = cgroup_tg(cgrp);
Peter Zijlstra029632f2011-10-25 10:00:11 +02007788 struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
Nikhil Raoe8da1b12011-07-21 09:43:40 -07007789
7790 cb->fill(cb, "nr_periods", cfs_b->nr_periods);
7791 cb->fill(cb, "nr_throttled", cfs_b->nr_throttled);
7792 cb->fill(cb, "throttled_time", cfs_b->throttled_time);
7793
7794 return 0;
7795}
Paul Turnerab84d312011-07-21 09:43:28 -07007796#endif /* CONFIG_CFS_BANDWIDTH */
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007797#endif /* CONFIG_FAIR_GROUP_SCHED */
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007798
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007799#ifdef CONFIG_RT_GROUP_SCHED
Mirco Tischler0c708142008-05-14 16:05:46 -07007800static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft,
Paul Menage06ecb272008-04-29 01:00:06 -07007801 s64 val)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007802{
Paul Menage06ecb272008-04-29 01:00:06 -07007803 return sched_group_set_rt_runtime(cgroup_tg(cgrp), val);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007804}
7805
Paul Menage06ecb272008-04-29 01:00:06 -07007806static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007807{
Paul Menage06ecb272008-04-29 01:00:06 -07007808 return sched_group_rt_runtime(cgroup_tg(cgrp));
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007809}
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007810
7811static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype,
7812 u64 rt_period_us)
7813{
7814 return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us);
7815}
7816
7817static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft)
7818{
7819 return sched_group_rt_period(cgroup_tg(cgrp));
7820}
Dhaval Giani6d6bc0a2008-05-30 14:23:45 +02007821#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007822
Paul Menagefe5c7cc2007-10-29 21:18:11 +01007823static struct cftype cpu_files[] = {
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007824#ifdef CONFIG_FAIR_GROUP_SCHED
Paul Menagefe5c7cc2007-10-29 21:18:11 +01007825 {
7826 .name = "shares",
Paul Menagef4c753b2008-04-29 00:59:56 -07007827 .read_u64 = cpu_shares_read_u64,
7828 .write_u64 = cpu_shares_write_u64,
Paul Menagefe5c7cc2007-10-29 21:18:11 +01007829 },
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007830#endif
Paul Turnerab84d312011-07-21 09:43:28 -07007831#ifdef CONFIG_CFS_BANDWIDTH
7832 {
7833 .name = "cfs_quota_us",
7834 .read_s64 = cpu_cfs_quota_read_s64,
7835 .write_s64 = cpu_cfs_quota_write_s64,
7836 },
7837 {
7838 .name = "cfs_period_us",
7839 .read_u64 = cpu_cfs_period_read_u64,
7840 .write_u64 = cpu_cfs_period_write_u64,
7841 },
Nikhil Raoe8da1b12011-07-21 09:43:40 -07007842 {
7843 .name = "stat",
7844 .read_map = cpu_stats_show,
7845 },
Paul Turnerab84d312011-07-21 09:43:28 -07007846#endif
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007847#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007848 {
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +01007849 .name = "rt_runtime_us",
Paul Menage06ecb272008-04-29 01:00:06 -07007850 .read_s64 = cpu_rt_runtime_read,
7851 .write_s64 = cpu_rt_runtime_write,
Peter Zijlstra6f505b12008-01-25 21:08:30 +01007852 },
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007853 {
7854 .name = "rt_period_us",
Paul Menagef4c753b2008-04-29 00:59:56 -07007855 .read_u64 = cpu_rt_period_read_uint,
7856 .write_u64 = cpu_rt_period_write_uint,
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +02007857 },
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007858#endif
Tejun Heo4baf6e32012-04-01 12:09:55 -07007859 { } /* terminate */
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007860};
7861
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007862struct cgroup_subsys cpu_cgroup_subsys = {
Ingo Molnar38605ca2007-10-29 21:18:11 +01007863 .name = "cpu",
Tejun Heo92fb9742012-11-19 08:13:38 -08007864 .css_alloc = cpu_cgroup_css_alloc,
7865 .css_free = cpu_cgroup_css_free,
Tejun Heobb9d97b2011-12-12 18:12:21 -08007866 .can_attach = cpu_cgroup_can_attach,
7867 .attach = cpu_cgroup_attach,
Peter Zijlstra068c5cc2011-01-19 12:26:11 +01007868 .exit = cpu_cgroup_exit,
Ingo Molnar38605ca2007-10-29 21:18:11 +01007869 .subsys_id = cpu_cgroup_subsys_id,
Tejun Heo4baf6e32012-04-01 12:09:55 -07007870 .base_cftypes = cpu_files,
Srivatsa Vaddagiri68318b82007-10-18 23:41:03 -07007871 .early_init = 1,
7872};
7873
Peter Zijlstra052f1dc2008-02-13 15:45:40 +01007874#endif /* CONFIG_CGROUP_SCHED */
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007875
7876#ifdef CONFIG_CGROUP_CPUACCT
7877
7878/*
7879 * CPU accounting code for task groups.
7880 *
7881 * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
7882 * (balbir@in.ibm.com).
7883 */
7884
Frederic Weisbecker73fbec62012-06-16 15:57:37 +02007885struct cpuacct root_cpuacct;
7886
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007887/* create a new cpu accounting group */
Tejun Heo92fb9742012-11-19 08:13:38 -08007888static struct cgroup_subsys_state *cpuacct_css_alloc(struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007889{
Glauber Costa54c707e2011-11-28 14:45:19 -02007890 struct cpuacct *ca;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007891
Glauber Costa54c707e2011-11-28 14:45:19 -02007892 if (!cgrp->parent)
7893 return &root_cpuacct.css;
7894
7895 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007896 if (!ca)
Bharata B Raoef12fef2009-03-31 10:02:22 +05307897 goto out;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007898
7899 ca->cpuusage = alloc_percpu(u64);
Bharata B Raoef12fef2009-03-31 10:02:22 +05307900 if (!ca->cpuusage)
7901 goto out_free_ca;
7902
Glauber Costa54c707e2011-11-28 14:45:19 -02007903 ca->cpustat = alloc_percpu(struct kernel_cpustat);
7904 if (!ca->cpustat)
7905 goto out_free_cpuusage;
Bharata B Rao934352f2008-11-10 20:41:13 +05307906
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007907 return &ca->css;
Bharata B Raoef12fef2009-03-31 10:02:22 +05307908
Glauber Costa54c707e2011-11-28 14:45:19 -02007909out_free_cpuusage:
Bharata B Raoef12fef2009-03-31 10:02:22 +05307910 free_percpu(ca->cpuusage);
7911out_free_ca:
7912 kfree(ca);
7913out:
7914 return ERR_PTR(-ENOMEM);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007915}
7916
7917/* destroy an existing cpu accounting group */
Tejun Heo92fb9742012-11-19 08:13:38 -08007918static void cpuacct_css_free(struct cgroup *cgrp)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007919{
Dhaval Giani32cd7562008-02-29 10:02:43 +05307920 struct cpuacct *ca = cgroup_ca(cgrp);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007921
Glauber Costa54c707e2011-11-28 14:45:19 -02007922 free_percpu(ca->cpustat);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007923 free_percpu(ca->cpuusage);
7924 kfree(ca);
7925}
7926
Ken Chen720f5492008-12-15 22:02:01 -08007927static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu)
7928{
Rusty Russellb36128c2009-02-20 16:29:08 +09007929 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
Ken Chen720f5492008-12-15 22:02:01 -08007930 u64 data;
7931
7932#ifndef CONFIG_64BIT
7933 /*
7934 * Take rq->lock to make 64-bit read safe on 32-bit platforms.
7935 */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01007936 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08007937 data = *cpuusage;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01007938 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08007939#else
7940 data = *cpuusage;
7941#endif
7942
7943 return data;
7944}
7945
7946static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val)
7947{
Rusty Russellb36128c2009-02-20 16:29:08 +09007948 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
Ken Chen720f5492008-12-15 22:02:01 -08007949
7950#ifndef CONFIG_64BIT
7951 /*
7952 * Take rq->lock to make 64-bit write safe on 32-bit platforms.
7953 */
Thomas Gleixner05fa7852009-11-17 14:28:38 +01007954 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08007955 *cpuusage = val;
Thomas Gleixner05fa7852009-11-17 14:28:38 +01007956 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
Ken Chen720f5492008-12-15 22:02:01 -08007957#else
7958 *cpuusage = val;
7959#endif
7960}
7961
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007962/* return total cpu usage (in nanoseconds) of a group */
Dhaval Giani32cd7562008-02-29 10:02:43 +05307963static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007964{
Dhaval Giani32cd7562008-02-29 10:02:43 +05307965 struct cpuacct *ca = cgroup_ca(cgrp);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007966 u64 totalcpuusage = 0;
7967 int i;
7968
Ken Chen720f5492008-12-15 22:02:01 -08007969 for_each_present_cpu(i)
7970 totalcpuusage += cpuacct_cpuusage_read(ca, i);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01007971
7972 return totalcpuusage;
7973}
7974
Dhaval Giani0297b802008-02-29 10:02:44 +05307975static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype,
7976 u64 reset)
7977{
7978 struct cpuacct *ca = cgroup_ca(cgrp);
7979 int err = 0;
7980 int i;
7981
7982 if (reset) {
7983 err = -EINVAL;
7984 goto out;
7985 }
7986
Ken Chen720f5492008-12-15 22:02:01 -08007987 for_each_present_cpu(i)
7988 cpuacct_cpuusage_write(ca, i, 0);
Dhaval Giani0297b802008-02-29 10:02:44 +05307989
Dhaval Giani0297b802008-02-29 10:02:44 +05307990out:
7991 return err;
7992}
7993
Ken Chene9515c32008-12-15 22:04:15 -08007994static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft,
7995 struct seq_file *m)
7996{
7997 struct cpuacct *ca = cgroup_ca(cgroup);
7998 u64 percpu;
7999 int i;
8000
8001 for_each_present_cpu(i) {
8002 percpu = cpuacct_cpuusage_read(ca, i);
8003 seq_printf(m, "%llu ", (unsigned long long) percpu);
8004 }
8005 seq_printf(m, "\n");
8006 return 0;
8007}
8008
Bharata B Raoef12fef2009-03-31 10:02:22 +05308009static const char *cpuacct_stat_desc[] = {
8010 [CPUACCT_STAT_USER] = "user",
8011 [CPUACCT_STAT_SYSTEM] = "system",
8012};
8013
8014static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft,
Glauber Costa54c707e2011-11-28 14:45:19 -02008015 struct cgroup_map_cb *cb)
Bharata B Raoef12fef2009-03-31 10:02:22 +05308016{
8017 struct cpuacct *ca = cgroup_ca(cgrp);
Glauber Costa54c707e2011-11-28 14:45:19 -02008018 int cpu;
8019 s64 val = 0;
Bharata B Raoef12fef2009-03-31 10:02:22 +05308020
Glauber Costa54c707e2011-11-28 14:45:19 -02008021 for_each_online_cpu(cpu) {
8022 struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu);
8023 val += kcpustat->cpustat[CPUTIME_USER];
8024 val += kcpustat->cpustat[CPUTIME_NICE];
Bharata B Raoef12fef2009-03-31 10:02:22 +05308025 }
Glauber Costa54c707e2011-11-28 14:45:19 -02008026 val = cputime64_to_clock_t(val);
8027 cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_USER], val);
8028
8029 val = 0;
8030 for_each_online_cpu(cpu) {
8031 struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu);
8032 val += kcpustat->cpustat[CPUTIME_SYSTEM];
8033 val += kcpustat->cpustat[CPUTIME_IRQ];
8034 val += kcpustat->cpustat[CPUTIME_SOFTIRQ];
8035 }
8036
8037 val = cputime64_to_clock_t(val);
8038 cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val);
8039
Bharata B Raoef12fef2009-03-31 10:02:22 +05308040 return 0;
8041}
8042
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008043static struct cftype files[] = {
8044 {
8045 .name = "usage",
Paul Menagef4c753b2008-04-29 00:59:56 -07008046 .read_u64 = cpuusage_read,
8047 .write_u64 = cpuusage_write,
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008048 },
Ken Chene9515c32008-12-15 22:04:15 -08008049 {
8050 .name = "usage_percpu",
8051 .read_seq_string = cpuacct_percpu_seq_read,
8052 },
Bharata B Raoef12fef2009-03-31 10:02:22 +05308053 {
8054 .name = "stat",
8055 .read_map = cpuacct_stats_show,
8056 },
Tejun Heo4baf6e32012-04-01 12:09:55 -07008057 { } /* terminate */
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008058};
8059
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008060/*
8061 * charge this task's execution time to its accounting group.
8062 *
8063 * called with rq->lock held.
8064 */
Peter Zijlstra029632f2011-10-25 10:00:11 +02008065void cpuacct_charge(struct task_struct *tsk, u64 cputime)
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008066{
8067 struct cpuacct *ca;
Bharata B Rao934352f2008-11-10 20:41:13 +05308068 int cpu;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008069
Li Zefanc40c6f82009-02-26 15:40:15 +08008070 if (unlikely(!cpuacct_subsys.active))
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008071 return;
8072
Bharata B Rao934352f2008-11-10 20:41:13 +05308073 cpu = task_cpu(tsk);
Bharata B Raoa18b83b2009-03-23 10:02:53 +05308074
8075 rcu_read_lock();
8076
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008077 ca = task_ca(tsk);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008078
Glauber Costa44252e42011-11-28 14:45:18 -02008079 for (; ca; ca = parent_ca(ca)) {
Rusty Russellb36128c2009-02-20 16:29:08 +09008080 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008081 *cpuusage += cputime;
8082 }
Bharata B Raoa18b83b2009-03-23 10:02:53 +05308083
8084 rcu_read_unlock();
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008085}
8086
8087struct cgroup_subsys cpuacct_subsys = {
8088 .name = "cpuacct",
Tejun Heo92fb9742012-11-19 08:13:38 -08008089 .css_alloc = cpuacct_css_alloc,
8090 .css_free = cpuacct_css_free,
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008091 .subsys_id = cpuacct_subsys_id,
Tejun Heo4baf6e32012-04-01 12:09:55 -07008092 .base_cftypes = files,
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +01008093};
8094#endif /* CONFIG_CGROUP_CPUACCT */
Paul E. McKenneyb637a322012-09-19 16:58:38 -07008095
8096void dump_cpu_task(int cpu)
8097{
8098 pr_info("Task dump for CPU %d:\n", cpu);
8099 sched_show_task(cpu_curr(cpu));
8100}