]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 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 | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
57 | #include <linux/kthread.h> | |
58 | #include <linux/seq_file.h> | |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
5517d86b | 65 | #include <linux/reciprocal_div.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
434d53b0 | 70 | #include <linux/bootmem.h> |
1da177e4 | 71 | |
5517d86b | 72 | #include <asm/tlb.h> |
838225b4 | 73 | #include <asm/irq_regs.h> |
1da177e4 | 74 | |
b035b6de AD |
75 | /* |
76 | * Scheduler clock - returns current time in nanosec units. | |
77 | * This is default implementation. | |
78 | * Architectures and sub-architectures can override this. | |
79 | */ | |
80 | unsigned long long __attribute__((weak)) sched_clock(void) | |
81 | { | |
d6322faf | 82 | return (unsigned long long)jiffies * (NSEC_PER_SEC / HZ); |
b035b6de AD |
83 | } |
84 | ||
1da177e4 LT |
85 | /* |
86 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
87 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
88 | * and back. | |
89 | */ | |
90 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
91 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
92 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
93 | ||
94 | /* | |
95 | * 'User priority' is the nice value converted to something we | |
96 | * can work with better when scaling various scheduler parameters, | |
97 | * it's a [ 0 ... 39 ] range. | |
98 | */ | |
99 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
100 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
101 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
102 | ||
103 | /* | |
d7876a08 | 104 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 105 | */ |
d6322faf | 106 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 107 | |
6aa645ea IM |
108 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
109 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
110 | ||
1da177e4 LT |
111 | /* |
112 | * These are the 'tuning knobs' of the scheduler: | |
113 | * | |
a4ec24b4 | 114 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
115 | * Timeslices get refilled after they expire. |
116 | */ | |
1da177e4 | 117 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 118 | |
d0b27fa7 PZ |
119 | /* |
120 | * single value that denotes runtime == period, ie unlimited time. | |
121 | */ | |
122 | #define RUNTIME_INF ((u64)~0ULL) | |
123 | ||
5517d86b ED |
124 | #ifdef CONFIG_SMP |
125 | /* | |
126 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
127 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
128 | */ | |
129 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
130 | { | |
131 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
132 | } | |
133 | ||
134 | /* | |
135 | * Each time a sched group cpu_power is changed, | |
136 | * we must compute its reciprocal value | |
137 | */ | |
138 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
139 | { | |
140 | sg->__cpu_power += val; | |
141 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
142 | } | |
143 | #endif | |
144 | ||
e05606d3 IM |
145 | static inline int rt_policy(int policy) |
146 | { | |
147 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
148 | return 1; | |
149 | return 0; | |
150 | } | |
151 | ||
152 | static inline int task_has_rt_policy(struct task_struct *p) | |
153 | { | |
154 | return rt_policy(p->policy); | |
155 | } | |
156 | ||
1da177e4 | 157 | /* |
6aa645ea | 158 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 159 | */ |
6aa645ea IM |
160 | struct rt_prio_array { |
161 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
162 | struct list_head queue[MAX_RT_PRIO]; | |
163 | }; | |
164 | ||
d0b27fa7 PZ |
165 | struct rt_bandwidth { |
166 | ktime_t rt_period; | |
167 | u64 rt_runtime; | |
ac086bc2 | 168 | spinlock_t rt_runtime_lock; |
d0b27fa7 PZ |
169 | struct hrtimer rt_period_timer; |
170 | }; | |
171 | ||
172 | static struct rt_bandwidth def_rt_bandwidth; | |
173 | ||
174 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
175 | ||
176 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
177 | { | |
178 | struct rt_bandwidth *rt_b = | |
179 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
180 | ktime_t now; | |
181 | int overrun; | |
182 | int idle = 0; | |
183 | ||
184 | for (;;) { | |
185 | now = hrtimer_cb_get_time(timer); | |
186 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
187 | ||
188 | if (!overrun) | |
189 | break; | |
190 | ||
191 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
192 | } | |
193 | ||
194 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
195 | } | |
196 | ||
197 | static | |
198 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
199 | { | |
200 | rt_b->rt_period = ns_to_ktime(period); | |
201 | rt_b->rt_runtime = runtime; | |
202 | ||
ac086bc2 PZ |
203 | spin_lock_init(&rt_b->rt_runtime_lock); |
204 | ||
d0b27fa7 PZ |
205 | hrtimer_init(&rt_b->rt_period_timer, |
206 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
207 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
208 | rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
209 | } | |
210 | ||
211 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
212 | { | |
213 | ktime_t now; | |
214 | ||
215 | if (rt_b->rt_runtime == RUNTIME_INF) | |
216 | return; | |
217 | ||
218 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
219 | return; | |
220 | ||
221 | spin_lock(&rt_b->rt_runtime_lock); | |
222 | for (;;) { | |
223 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
224 | break; | |
225 | ||
226 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
227 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
228 | hrtimer_start(&rt_b->rt_period_timer, | |
229 | rt_b->rt_period_timer.expires, | |
230 | HRTIMER_MODE_ABS); | |
231 | } | |
232 | spin_unlock(&rt_b->rt_runtime_lock); | |
233 | } | |
234 | ||
235 | #ifdef CONFIG_RT_GROUP_SCHED | |
236 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
237 | { | |
238 | hrtimer_cancel(&rt_b->rt_period_timer); | |
239 | } | |
240 | #endif | |
241 | ||
052f1dc7 | 242 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 243 | |
68318b8e SV |
244 | #include <linux/cgroup.h> |
245 | ||
29f59db3 SV |
246 | struct cfs_rq; |
247 | ||
6f505b16 PZ |
248 | static LIST_HEAD(task_groups); |
249 | ||
29f59db3 | 250 | /* task group related information */ |
4cf86d77 | 251 | struct task_group { |
052f1dc7 | 252 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
253 | struct cgroup_subsys_state css; |
254 | #endif | |
052f1dc7 PZ |
255 | |
256 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
257 | /* schedulable entities of this group on each cpu */ |
258 | struct sched_entity **se; | |
259 | /* runqueue "owned" by this group on each cpu */ | |
260 | struct cfs_rq **cfs_rq; | |
261 | unsigned long shares; | |
052f1dc7 PZ |
262 | #endif |
263 | ||
264 | #ifdef CONFIG_RT_GROUP_SCHED | |
265 | struct sched_rt_entity **rt_se; | |
266 | struct rt_rq **rt_rq; | |
267 | ||
d0b27fa7 | 268 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 269 | #endif |
6b2d7700 | 270 | |
ae8393e5 | 271 | struct rcu_head rcu; |
6f505b16 | 272 | struct list_head list; |
29f59db3 SV |
273 | }; |
274 | ||
052f1dc7 | 275 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
276 | /* Default task group's sched entity on each cpu */ |
277 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
278 | /* Default task group's cfs_rq on each cpu */ | |
279 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
052f1dc7 PZ |
280 | #endif |
281 | ||
282 | #ifdef CONFIG_RT_GROUP_SCHED | |
283 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
284 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
052f1dc7 | 285 | #endif |
6f505b16 | 286 | |
8ed36996 | 287 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
288 | * a task group's cpu shares. |
289 | */ | |
8ed36996 | 290 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 291 | |
a1835615 SV |
292 | /* doms_cur_mutex serializes access to doms_cur[] array */ |
293 | static DEFINE_MUTEX(doms_cur_mutex); | |
294 | ||
052f1dc7 | 295 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
296 | #ifdef CONFIG_USER_SCHED |
297 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
298 | #else | |
299 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD | |
300 | #endif | |
301 | ||
052f1dc7 PZ |
302 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
303 | #endif | |
304 | ||
29f59db3 | 305 | /* Default task group. |
3a252015 | 306 | * Every task in system belong to this group at bootup. |
29f59db3 | 307 | */ |
434d53b0 | 308 | struct task_group init_task_group; |
29f59db3 SV |
309 | |
310 | /* return group to which a task belongs */ | |
4cf86d77 | 311 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 312 | { |
4cf86d77 | 313 | struct task_group *tg; |
9b5b7751 | 314 | |
052f1dc7 | 315 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 316 | tg = p->user->tg; |
052f1dc7 | 317 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
318 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
319 | struct task_group, css); | |
24e377a8 | 320 | #else |
41a2d6cf | 321 | tg = &init_task_group; |
24e377a8 | 322 | #endif |
9b5b7751 | 323 | return tg; |
29f59db3 SV |
324 | } |
325 | ||
326 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 327 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 328 | { |
052f1dc7 | 329 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
330 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
331 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 332 | #endif |
6f505b16 | 333 | |
052f1dc7 | 334 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
335 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
336 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 337 | #endif |
29f59db3 SV |
338 | } |
339 | ||
a1835615 SV |
340 | static inline void lock_doms_cur(void) |
341 | { | |
342 | mutex_lock(&doms_cur_mutex); | |
343 | } | |
344 | ||
345 | static inline void unlock_doms_cur(void) | |
346 | { | |
347 | mutex_unlock(&doms_cur_mutex); | |
348 | } | |
349 | ||
29f59db3 SV |
350 | #else |
351 | ||
6f505b16 | 352 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
a1835615 SV |
353 | static inline void lock_doms_cur(void) { } |
354 | static inline void unlock_doms_cur(void) { } | |
29f59db3 | 355 | |
052f1dc7 | 356 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 357 | |
6aa645ea IM |
358 | /* CFS-related fields in a runqueue */ |
359 | struct cfs_rq { | |
360 | struct load_weight load; | |
361 | unsigned long nr_running; | |
362 | ||
6aa645ea | 363 | u64 exec_clock; |
e9acbff6 | 364 | u64 min_vruntime; |
6aa645ea IM |
365 | |
366 | struct rb_root tasks_timeline; | |
367 | struct rb_node *rb_leftmost; | |
368 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
369 | /* 'curr' points to currently running entity on this cfs_rq. |
370 | * It is set to NULL otherwise (i.e when none are currently running). | |
371 | */ | |
aa2ac252 | 372 | struct sched_entity *curr, *next; |
ddc97297 PZ |
373 | |
374 | unsigned long nr_spread_over; | |
375 | ||
62160e3f | 376 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
377 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
378 | ||
41a2d6cf IM |
379 | /* |
380 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
381 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
382 | * (like users, containers etc.) | |
383 | * | |
384 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
385 | * list is used during load balance. | |
386 | */ | |
41a2d6cf IM |
387 | struct list_head leaf_cfs_rq_list; |
388 | struct task_group *tg; /* group that "owns" this runqueue */ | |
6aa645ea IM |
389 | #endif |
390 | }; | |
1da177e4 | 391 | |
6aa645ea IM |
392 | /* Real-Time classes' related field in a runqueue: */ |
393 | struct rt_rq { | |
394 | struct rt_prio_array active; | |
63489e45 | 395 | unsigned long rt_nr_running; |
052f1dc7 | 396 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
397 | int highest_prio; /* highest queued rt task prio */ |
398 | #endif | |
fa85ae24 | 399 | #ifdef CONFIG_SMP |
73fe6aae | 400 | unsigned long rt_nr_migratory; |
a22d7fc1 | 401 | int overloaded; |
fa85ae24 | 402 | #endif |
6f505b16 | 403 | int rt_throttled; |
fa85ae24 | 404 | u64 rt_time; |
ac086bc2 PZ |
405 | u64 rt_runtime; |
406 | spinlock_t rt_runtime_lock; | |
6f505b16 | 407 | |
052f1dc7 | 408 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
409 | unsigned long rt_nr_boosted; |
410 | ||
6f505b16 PZ |
411 | struct rq *rq; |
412 | struct list_head leaf_rt_rq_list; | |
413 | struct task_group *tg; | |
414 | struct sched_rt_entity *rt_se; | |
415 | #endif | |
6aa645ea IM |
416 | }; |
417 | ||
57d885fe GH |
418 | #ifdef CONFIG_SMP |
419 | ||
420 | /* | |
421 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
422 | * variables. Each exclusive cpuset essentially defines an island domain by |
423 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
424 | * exclusive cpuset is created, we also create and attach a new root-domain |
425 | * object. | |
426 | * | |
57d885fe GH |
427 | */ |
428 | struct root_domain { | |
429 | atomic_t refcount; | |
430 | cpumask_t span; | |
431 | cpumask_t online; | |
637f5085 | 432 | |
0eab9146 | 433 | /* |
637f5085 GH |
434 | * The "RT overload" flag: it gets set if a CPU has more than |
435 | * one runnable RT task. | |
436 | */ | |
437 | cpumask_t rto_mask; | |
0eab9146 | 438 | atomic_t rto_count; |
57d885fe GH |
439 | }; |
440 | ||
dc938520 GH |
441 | /* |
442 | * By default the system creates a single root-domain with all cpus as | |
443 | * members (mimicking the global state we have today). | |
444 | */ | |
57d885fe GH |
445 | static struct root_domain def_root_domain; |
446 | ||
447 | #endif | |
448 | ||
1da177e4 LT |
449 | /* |
450 | * This is the main, per-CPU runqueue data structure. | |
451 | * | |
452 | * Locking rule: those places that want to lock multiple runqueues | |
453 | * (such as the load balancing or the thread migration code), lock | |
454 | * acquire operations must be ordered by ascending &runqueue. | |
455 | */ | |
70b97a7f | 456 | struct rq { |
d8016491 IM |
457 | /* runqueue lock: */ |
458 | spinlock_t lock; | |
1da177e4 LT |
459 | |
460 | /* | |
461 | * nr_running and cpu_load should be in the same cacheline because | |
462 | * remote CPUs use both these fields when doing load calculation. | |
463 | */ | |
464 | unsigned long nr_running; | |
6aa645ea IM |
465 | #define CPU_LOAD_IDX_MAX 5 |
466 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 467 | unsigned char idle_at_tick; |
46cb4b7c | 468 | #ifdef CONFIG_NO_HZ |
15934a37 | 469 | unsigned long last_tick_seen; |
46cb4b7c SS |
470 | unsigned char in_nohz_recently; |
471 | #endif | |
d8016491 IM |
472 | /* capture load from *all* tasks on this cpu: */ |
473 | struct load_weight load; | |
6aa645ea IM |
474 | unsigned long nr_load_updates; |
475 | u64 nr_switches; | |
476 | ||
477 | struct cfs_rq cfs; | |
6f505b16 | 478 | struct rt_rq rt; |
6f505b16 | 479 | |
6aa645ea | 480 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
481 | /* list of leaf cfs_rq on this cpu: */ |
482 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
483 | #endif |
484 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 485 | struct list_head leaf_rt_rq_list; |
1da177e4 | 486 | #endif |
1da177e4 LT |
487 | |
488 | /* | |
489 | * This is part of a global counter where only the total sum | |
490 | * over all CPUs matters. A task can increase this counter on | |
491 | * one CPU and if it got migrated afterwards it may decrease | |
492 | * it on another CPU. Always updated under the runqueue lock: | |
493 | */ | |
494 | unsigned long nr_uninterruptible; | |
495 | ||
36c8b586 | 496 | struct task_struct *curr, *idle; |
c9819f45 | 497 | unsigned long next_balance; |
1da177e4 | 498 | struct mm_struct *prev_mm; |
6aa645ea | 499 | |
6aa645ea IM |
500 | u64 clock, prev_clock_raw; |
501 | s64 clock_max_delta; | |
502 | ||
cc203d24 | 503 | unsigned int clock_warps, clock_overflows, clock_underflows; |
2aa44d05 IM |
504 | u64 idle_clock; |
505 | unsigned int clock_deep_idle_events; | |
529c7726 | 506 | u64 tick_timestamp; |
6aa645ea | 507 | |
1da177e4 LT |
508 | atomic_t nr_iowait; |
509 | ||
510 | #ifdef CONFIG_SMP | |
0eab9146 | 511 | struct root_domain *rd; |
1da177e4 LT |
512 | struct sched_domain *sd; |
513 | ||
514 | /* For active balancing */ | |
515 | int active_balance; | |
516 | int push_cpu; | |
d8016491 IM |
517 | /* cpu of this runqueue: */ |
518 | int cpu; | |
1da177e4 | 519 | |
36c8b586 | 520 | struct task_struct *migration_thread; |
1da177e4 LT |
521 | struct list_head migration_queue; |
522 | #endif | |
523 | ||
8f4d37ec PZ |
524 | #ifdef CONFIG_SCHED_HRTICK |
525 | unsigned long hrtick_flags; | |
526 | ktime_t hrtick_expire; | |
527 | struct hrtimer hrtick_timer; | |
528 | #endif | |
529 | ||
1da177e4 LT |
530 | #ifdef CONFIG_SCHEDSTATS |
531 | /* latency stats */ | |
532 | struct sched_info rq_sched_info; | |
533 | ||
534 | /* sys_sched_yield() stats */ | |
480b9434 KC |
535 | unsigned int yld_exp_empty; |
536 | unsigned int yld_act_empty; | |
537 | unsigned int yld_both_empty; | |
538 | unsigned int yld_count; | |
1da177e4 LT |
539 | |
540 | /* schedule() stats */ | |
480b9434 KC |
541 | unsigned int sched_switch; |
542 | unsigned int sched_count; | |
543 | unsigned int sched_goidle; | |
1da177e4 LT |
544 | |
545 | /* try_to_wake_up() stats */ | |
480b9434 KC |
546 | unsigned int ttwu_count; |
547 | unsigned int ttwu_local; | |
b8efb561 IM |
548 | |
549 | /* BKL stats */ | |
480b9434 | 550 | unsigned int bkl_count; |
1da177e4 | 551 | #endif |
fcb99371 | 552 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
553 | }; |
554 | ||
f34e3b61 | 555 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 556 | |
dd41f596 IM |
557 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
558 | { | |
559 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
560 | } | |
561 | ||
0a2966b4 CL |
562 | static inline int cpu_of(struct rq *rq) |
563 | { | |
564 | #ifdef CONFIG_SMP | |
565 | return rq->cpu; | |
566 | #else | |
567 | return 0; | |
568 | #endif | |
569 | } | |
570 | ||
15934a37 GC |
571 | #ifdef CONFIG_NO_HZ |
572 | static inline bool nohz_on(int cpu) | |
573 | { | |
574 | return tick_get_tick_sched(cpu)->nohz_mode != NOHZ_MODE_INACTIVE; | |
575 | } | |
576 | ||
577 | static inline u64 max_skipped_ticks(struct rq *rq) | |
578 | { | |
579 | return nohz_on(cpu_of(rq)) ? jiffies - rq->last_tick_seen + 2 : 1; | |
580 | } | |
581 | ||
582 | static inline void update_last_tick_seen(struct rq *rq) | |
583 | { | |
584 | rq->last_tick_seen = jiffies; | |
585 | } | |
586 | #else | |
587 | static inline u64 max_skipped_ticks(struct rq *rq) | |
588 | { | |
589 | return 1; | |
590 | } | |
591 | ||
592 | static inline void update_last_tick_seen(struct rq *rq) | |
593 | { | |
594 | } | |
595 | #endif | |
596 | ||
20d315d4 | 597 | /* |
b04a0f4c IM |
598 | * Update the per-runqueue clock, as finegrained as the platform can give |
599 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 600 | */ |
b04a0f4c | 601 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
602 | { |
603 | u64 prev_raw = rq->prev_clock_raw; | |
604 | u64 now = sched_clock(); | |
605 | s64 delta = now - prev_raw; | |
606 | u64 clock = rq->clock; | |
607 | ||
b04a0f4c IM |
608 | #ifdef CONFIG_SCHED_DEBUG |
609 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
610 | #endif | |
20d315d4 IM |
611 | /* |
612 | * Protect against sched_clock() occasionally going backwards: | |
613 | */ | |
614 | if (unlikely(delta < 0)) { | |
615 | clock++; | |
616 | rq->clock_warps++; | |
617 | } else { | |
618 | /* | |
619 | * Catch too large forward jumps too: | |
620 | */ | |
15934a37 GC |
621 | u64 max_jump = max_skipped_ticks(rq) * TICK_NSEC; |
622 | u64 max_time = rq->tick_timestamp + max_jump; | |
623 | ||
624 | if (unlikely(clock + delta > max_time)) { | |
625 | if (clock < max_time) | |
626 | clock = max_time; | |
529c7726 IM |
627 | else |
628 | clock++; | |
20d315d4 IM |
629 | rq->clock_overflows++; |
630 | } else { | |
631 | if (unlikely(delta > rq->clock_max_delta)) | |
632 | rq->clock_max_delta = delta; | |
633 | clock += delta; | |
634 | } | |
635 | } | |
636 | ||
637 | rq->prev_clock_raw = now; | |
638 | rq->clock = clock; | |
b04a0f4c | 639 | } |
20d315d4 | 640 | |
b04a0f4c IM |
641 | static void update_rq_clock(struct rq *rq) |
642 | { | |
643 | if (likely(smp_processor_id() == cpu_of(rq))) | |
644 | __update_rq_clock(rq); | |
20d315d4 IM |
645 | } |
646 | ||
674311d5 NP |
647 | /* |
648 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 649 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
650 | * |
651 | * The domain tree of any CPU may only be accessed from within | |
652 | * preempt-disabled sections. | |
653 | */ | |
48f24c4d IM |
654 | #define for_each_domain(cpu, __sd) \ |
655 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
656 | |
657 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
658 | #define this_rq() (&__get_cpu_var(runqueues)) | |
659 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
660 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
661 | ||
bf5c91ba IM |
662 | /* |
663 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
664 | */ | |
665 | #ifdef CONFIG_SCHED_DEBUG | |
666 | # define const_debug __read_mostly | |
667 | #else | |
668 | # define const_debug static const | |
669 | #endif | |
670 | ||
671 | /* | |
672 | * Debugging: various feature bits | |
673 | */ | |
674 | enum { | |
bbdba7c0 | 675 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
9612633a IM |
676 | SCHED_FEAT_WAKEUP_PREEMPT = 2, |
677 | SCHED_FEAT_START_DEBIT = 4, | |
d25ce4cd IM |
678 | SCHED_FEAT_AFFINE_WAKEUPS = 8, |
679 | SCHED_FEAT_CACHE_HOT_BUDDY = 16, | |
02e2b83b IM |
680 | SCHED_FEAT_SYNC_WAKEUPS = 32, |
681 | SCHED_FEAT_HRTICK = 64, | |
682 | SCHED_FEAT_DOUBLE_TICK = 128, | |
bf5c91ba IM |
683 | }; |
684 | ||
685 | const_debug unsigned int sysctl_sched_features = | |
8401f775 | 686 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
9612633a | 687 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | |
8401f775 | 688 | SCHED_FEAT_START_DEBIT * 1 | |
d25ce4cd IM |
689 | SCHED_FEAT_AFFINE_WAKEUPS * 1 | |
690 | SCHED_FEAT_CACHE_HOT_BUDDY * 1 | | |
02e2b83b | 691 | SCHED_FEAT_SYNC_WAKEUPS * 1 | |
8f4d37ec | 692 | SCHED_FEAT_HRTICK * 1 | |
02e2b83b | 693 | SCHED_FEAT_DOUBLE_TICK * 0; |
bf5c91ba IM |
694 | |
695 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
696 | ||
b82d9fdd PZ |
697 | /* |
698 | * Number of tasks to iterate in a single balance run. | |
699 | * Limited because this is done with IRQs disabled. | |
700 | */ | |
701 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
702 | ||
fa85ae24 | 703 | /* |
9f0c1e56 | 704 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
705 | * default: 1s |
706 | */ | |
9f0c1e56 | 707 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 708 | |
6892b75e IM |
709 | static __read_mostly int scheduler_running; |
710 | ||
9f0c1e56 PZ |
711 | /* |
712 | * part of the period that we allow rt tasks to run in us. | |
713 | * default: 0.95s | |
714 | */ | |
715 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 716 | |
d0b27fa7 PZ |
717 | static inline u64 global_rt_period(void) |
718 | { | |
719 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
720 | } | |
721 | ||
722 | static inline u64 global_rt_runtime(void) | |
723 | { | |
724 | if (sysctl_sched_rt_period < 0) | |
725 | return RUNTIME_INF; | |
726 | ||
727 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
728 | } | |
fa85ae24 | 729 | |
27ec4407 IM |
730 | static const unsigned long long time_sync_thresh = 100000; |
731 | ||
732 | static DEFINE_PER_CPU(unsigned long long, time_offset); | |
733 | static DEFINE_PER_CPU(unsigned long long, prev_cpu_time); | |
734 | ||
e436d800 | 735 | /* |
27ec4407 IM |
736 | * Global lock which we take every now and then to synchronize |
737 | * the CPUs time. This method is not warp-safe, but it's good | |
738 | * enough to synchronize slowly diverging time sources and thus | |
739 | * it's good enough for tracing: | |
e436d800 | 740 | */ |
27ec4407 IM |
741 | static DEFINE_SPINLOCK(time_sync_lock); |
742 | static unsigned long long prev_global_time; | |
743 | ||
744 | static unsigned long long __sync_cpu_clock(cycles_t time, int cpu) | |
745 | { | |
746 | unsigned long flags; | |
747 | ||
748 | spin_lock_irqsave(&time_sync_lock, flags); | |
749 | ||
750 | if (time < prev_global_time) { | |
751 | per_cpu(time_offset, cpu) += prev_global_time - time; | |
752 | time = prev_global_time; | |
753 | } else { | |
754 | prev_global_time = time; | |
755 | } | |
756 | ||
757 | spin_unlock_irqrestore(&time_sync_lock, flags); | |
758 | ||
759 | return time; | |
760 | } | |
761 | ||
762 | static unsigned long long __cpu_clock(int cpu) | |
e436d800 | 763 | { |
e436d800 IM |
764 | unsigned long long now; |
765 | unsigned long flags; | |
b04a0f4c | 766 | struct rq *rq; |
e436d800 | 767 | |
8ced5f69 IM |
768 | /* |
769 | * Only call sched_clock() if the scheduler has already been | |
770 | * initialized (some code might call cpu_clock() very early): | |
771 | */ | |
6892b75e IM |
772 | if (unlikely(!scheduler_running)) |
773 | return 0; | |
774 | ||
775 | local_irq_save(flags); | |
776 | rq = cpu_rq(cpu); | |
777 | update_rq_clock(rq); | |
b04a0f4c | 778 | now = rq->clock; |
2cd4d0ea | 779 | local_irq_restore(flags); |
e436d800 IM |
780 | |
781 | return now; | |
782 | } | |
27ec4407 IM |
783 | |
784 | /* | |
785 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
786 | * clock constructed from sched_clock(): | |
787 | */ | |
788 | unsigned long long cpu_clock(int cpu) | |
789 | { | |
790 | unsigned long long prev_cpu_time, time, delta_time; | |
791 | ||
792 | prev_cpu_time = per_cpu(prev_cpu_time, cpu); | |
793 | time = __cpu_clock(cpu) + per_cpu(time_offset, cpu); | |
794 | delta_time = time-prev_cpu_time; | |
795 | ||
796 | if (unlikely(delta_time > time_sync_thresh)) | |
797 | time = __sync_cpu_clock(time, cpu); | |
798 | ||
799 | return time; | |
800 | } | |
a58f6f25 | 801 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 802 | |
1da177e4 | 803 | #ifndef prepare_arch_switch |
4866cde0 NP |
804 | # define prepare_arch_switch(next) do { } while (0) |
805 | #endif | |
806 | #ifndef finish_arch_switch | |
807 | # define finish_arch_switch(prev) do { } while (0) | |
808 | #endif | |
809 | ||
051a1d1a DA |
810 | static inline int task_current(struct rq *rq, struct task_struct *p) |
811 | { | |
812 | return rq->curr == p; | |
813 | } | |
814 | ||
4866cde0 | 815 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 816 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 817 | { |
051a1d1a | 818 | return task_current(rq, p); |
4866cde0 NP |
819 | } |
820 | ||
70b97a7f | 821 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
822 | { |
823 | } | |
824 | ||
70b97a7f | 825 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 826 | { |
da04c035 IM |
827 | #ifdef CONFIG_DEBUG_SPINLOCK |
828 | /* this is a valid case when another task releases the spinlock */ | |
829 | rq->lock.owner = current; | |
830 | #endif | |
8a25d5de IM |
831 | /* |
832 | * If we are tracking spinlock dependencies then we have to | |
833 | * fix up the runqueue lock - which gets 'carried over' from | |
834 | * prev into current: | |
835 | */ | |
836 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
837 | ||
4866cde0 NP |
838 | spin_unlock_irq(&rq->lock); |
839 | } | |
840 | ||
841 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 842 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
843 | { |
844 | #ifdef CONFIG_SMP | |
845 | return p->oncpu; | |
846 | #else | |
051a1d1a | 847 | return task_current(rq, p); |
4866cde0 NP |
848 | #endif |
849 | } | |
850 | ||
70b97a7f | 851 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
852 | { |
853 | #ifdef CONFIG_SMP | |
854 | /* | |
855 | * We can optimise this out completely for !SMP, because the | |
856 | * SMP rebalancing from interrupt is the only thing that cares | |
857 | * here. | |
858 | */ | |
859 | next->oncpu = 1; | |
860 | #endif | |
861 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
862 | spin_unlock_irq(&rq->lock); | |
863 | #else | |
864 | spin_unlock(&rq->lock); | |
865 | #endif | |
866 | } | |
867 | ||
70b97a7f | 868 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
869 | { |
870 | #ifdef CONFIG_SMP | |
871 | /* | |
872 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
873 | * We must ensure this doesn't happen until the switch is completely | |
874 | * finished. | |
875 | */ | |
876 | smp_wmb(); | |
877 | prev->oncpu = 0; | |
878 | #endif | |
879 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
880 | local_irq_enable(); | |
1da177e4 | 881 | #endif |
4866cde0 NP |
882 | } |
883 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 884 | |
b29739f9 IM |
885 | /* |
886 | * __task_rq_lock - lock the runqueue a given task resides on. | |
887 | * Must be called interrupts disabled. | |
888 | */ | |
70b97a7f | 889 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
890 | __acquires(rq->lock) |
891 | { | |
3a5c359a AK |
892 | for (;;) { |
893 | struct rq *rq = task_rq(p); | |
894 | spin_lock(&rq->lock); | |
895 | if (likely(rq == task_rq(p))) | |
896 | return rq; | |
b29739f9 | 897 | spin_unlock(&rq->lock); |
b29739f9 | 898 | } |
b29739f9 IM |
899 | } |
900 | ||
1da177e4 LT |
901 | /* |
902 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 903 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
904 | * explicitly disabling preemption. |
905 | */ | |
70b97a7f | 906 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
907 | __acquires(rq->lock) |
908 | { | |
70b97a7f | 909 | struct rq *rq; |
1da177e4 | 910 | |
3a5c359a AK |
911 | for (;;) { |
912 | local_irq_save(*flags); | |
913 | rq = task_rq(p); | |
914 | spin_lock(&rq->lock); | |
915 | if (likely(rq == task_rq(p))) | |
916 | return rq; | |
1da177e4 | 917 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 918 | } |
1da177e4 LT |
919 | } |
920 | ||
a9957449 | 921 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
922 | __releases(rq->lock) |
923 | { | |
924 | spin_unlock(&rq->lock); | |
925 | } | |
926 | ||
70b97a7f | 927 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
928 | __releases(rq->lock) |
929 | { | |
930 | spin_unlock_irqrestore(&rq->lock, *flags); | |
931 | } | |
932 | ||
1da177e4 | 933 | /* |
cc2a73b5 | 934 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 935 | */ |
a9957449 | 936 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
937 | __acquires(rq->lock) |
938 | { | |
70b97a7f | 939 | struct rq *rq; |
1da177e4 LT |
940 | |
941 | local_irq_disable(); | |
942 | rq = this_rq(); | |
943 | spin_lock(&rq->lock); | |
944 | ||
945 | return rq; | |
946 | } | |
947 | ||
1b9f19c2 | 948 | /* |
2aa44d05 | 949 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 950 | */ |
2aa44d05 | 951 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 952 | { |
2aa44d05 IM |
953 | struct rq *rq = cpu_rq(smp_processor_id()); |
954 | ||
955 | spin_lock(&rq->lock); | |
956 | __update_rq_clock(rq); | |
957 | spin_unlock(&rq->lock); | |
958 | rq->clock_deep_idle_events++; | |
959 | } | |
960 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
961 | ||
962 | /* | |
963 | * We just idled delta nanoseconds (called with irqs disabled): | |
964 | */ | |
965 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
966 | { | |
967 | struct rq *rq = cpu_rq(smp_processor_id()); | |
968 | u64 now = sched_clock(); | |
1b9f19c2 | 969 | |
2aa44d05 IM |
970 | rq->idle_clock += delta_ns; |
971 | /* | |
972 | * Override the previous timestamp and ignore all | |
973 | * sched_clock() deltas that occured while we idled, | |
974 | * and use the PM-provided delta_ns to advance the | |
975 | * rq clock: | |
976 | */ | |
977 | spin_lock(&rq->lock); | |
978 | rq->prev_clock_raw = now; | |
979 | rq->clock += delta_ns; | |
980 | spin_unlock(&rq->lock); | |
782daeee | 981 | touch_softlockup_watchdog(); |
1b9f19c2 | 982 | } |
2aa44d05 | 983 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 984 | |
8f4d37ec PZ |
985 | static void __resched_task(struct task_struct *p, int tif_bit); |
986 | ||
987 | static inline void resched_task(struct task_struct *p) | |
988 | { | |
989 | __resched_task(p, TIF_NEED_RESCHED); | |
990 | } | |
991 | ||
992 | #ifdef CONFIG_SCHED_HRTICK | |
993 | /* | |
994 | * Use HR-timers to deliver accurate preemption points. | |
995 | * | |
996 | * Its all a bit involved since we cannot program an hrt while holding the | |
997 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
998 | * reschedule event. | |
999 | * | |
1000 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1001 | * rq->lock. | |
1002 | */ | |
1003 | static inline void resched_hrt(struct task_struct *p) | |
1004 | { | |
1005 | __resched_task(p, TIF_HRTICK_RESCHED); | |
1006 | } | |
1007 | ||
1008 | static inline void resched_rq(struct rq *rq) | |
1009 | { | |
1010 | unsigned long flags; | |
1011 | ||
1012 | spin_lock_irqsave(&rq->lock, flags); | |
1013 | resched_task(rq->curr); | |
1014 | spin_unlock_irqrestore(&rq->lock, flags); | |
1015 | } | |
1016 | ||
1017 | enum { | |
1018 | HRTICK_SET, /* re-programm hrtick_timer */ | |
1019 | HRTICK_RESET, /* not a new slice */ | |
1020 | }; | |
1021 | ||
1022 | /* | |
1023 | * Use hrtick when: | |
1024 | * - enabled by features | |
1025 | * - hrtimer is actually high res | |
1026 | */ | |
1027 | static inline int hrtick_enabled(struct rq *rq) | |
1028 | { | |
1029 | if (!sched_feat(HRTICK)) | |
1030 | return 0; | |
1031 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
1032 | } | |
1033 | ||
1034 | /* | |
1035 | * Called to set the hrtick timer state. | |
1036 | * | |
1037 | * called with rq->lock held and irqs disabled | |
1038 | */ | |
1039 | static void hrtick_start(struct rq *rq, u64 delay, int reset) | |
1040 | { | |
1041 | assert_spin_locked(&rq->lock); | |
1042 | ||
1043 | /* | |
1044 | * preempt at: now + delay | |
1045 | */ | |
1046 | rq->hrtick_expire = | |
1047 | ktime_add_ns(rq->hrtick_timer.base->get_time(), delay); | |
1048 | /* | |
1049 | * indicate we need to program the timer | |
1050 | */ | |
1051 | __set_bit(HRTICK_SET, &rq->hrtick_flags); | |
1052 | if (reset) | |
1053 | __set_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1054 | ||
1055 | /* | |
1056 | * New slices are called from the schedule path and don't need a | |
1057 | * forced reschedule. | |
1058 | */ | |
1059 | if (reset) | |
1060 | resched_hrt(rq->curr); | |
1061 | } | |
1062 | ||
1063 | static void hrtick_clear(struct rq *rq) | |
1064 | { | |
1065 | if (hrtimer_active(&rq->hrtick_timer)) | |
1066 | hrtimer_cancel(&rq->hrtick_timer); | |
1067 | } | |
1068 | ||
1069 | /* | |
1070 | * Update the timer from the possible pending state. | |
1071 | */ | |
1072 | static void hrtick_set(struct rq *rq) | |
1073 | { | |
1074 | ktime_t time; | |
1075 | int set, reset; | |
1076 | unsigned long flags; | |
1077 | ||
1078 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1079 | ||
1080 | spin_lock_irqsave(&rq->lock, flags); | |
1081 | set = __test_and_clear_bit(HRTICK_SET, &rq->hrtick_flags); | |
1082 | reset = __test_and_clear_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1083 | time = rq->hrtick_expire; | |
1084 | clear_thread_flag(TIF_HRTICK_RESCHED); | |
1085 | spin_unlock_irqrestore(&rq->lock, flags); | |
1086 | ||
1087 | if (set) { | |
1088 | hrtimer_start(&rq->hrtick_timer, time, HRTIMER_MODE_ABS); | |
1089 | if (reset && !hrtimer_active(&rq->hrtick_timer)) | |
1090 | resched_rq(rq); | |
1091 | } else | |
1092 | hrtick_clear(rq); | |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * High-resolution timer tick. | |
1097 | * Runs from hardirq context with interrupts disabled. | |
1098 | */ | |
1099 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1100 | { | |
1101 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1102 | ||
1103 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1104 | ||
1105 | spin_lock(&rq->lock); | |
1106 | __update_rq_clock(rq); | |
1107 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); | |
1108 | spin_unlock(&rq->lock); | |
1109 | ||
1110 | return HRTIMER_NORESTART; | |
1111 | } | |
1112 | ||
1113 | static inline void init_rq_hrtick(struct rq *rq) | |
1114 | { | |
1115 | rq->hrtick_flags = 0; | |
1116 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1117 | rq->hrtick_timer.function = hrtick; | |
1118 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
1119 | } | |
1120 | ||
1121 | void hrtick_resched(void) | |
1122 | { | |
1123 | struct rq *rq; | |
1124 | unsigned long flags; | |
1125 | ||
1126 | if (!test_thread_flag(TIF_HRTICK_RESCHED)) | |
1127 | return; | |
1128 | ||
1129 | local_irq_save(flags); | |
1130 | rq = cpu_rq(smp_processor_id()); | |
1131 | hrtick_set(rq); | |
1132 | local_irq_restore(flags); | |
1133 | } | |
1134 | #else | |
1135 | static inline void hrtick_clear(struct rq *rq) | |
1136 | { | |
1137 | } | |
1138 | ||
1139 | static inline void hrtick_set(struct rq *rq) | |
1140 | { | |
1141 | } | |
1142 | ||
1143 | static inline void init_rq_hrtick(struct rq *rq) | |
1144 | { | |
1145 | } | |
1146 | ||
1147 | void hrtick_resched(void) | |
1148 | { | |
1149 | } | |
1150 | #endif | |
1151 | ||
c24d20db IM |
1152 | /* |
1153 | * resched_task - mark a task 'to be rescheduled now'. | |
1154 | * | |
1155 | * On UP this means the setting of the need_resched flag, on SMP it | |
1156 | * might also involve a cross-CPU call to trigger the scheduler on | |
1157 | * the target CPU. | |
1158 | */ | |
1159 | #ifdef CONFIG_SMP | |
1160 | ||
1161 | #ifndef tsk_is_polling | |
1162 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1163 | #endif | |
1164 | ||
8f4d37ec | 1165 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1166 | { |
1167 | int cpu; | |
1168 | ||
1169 | assert_spin_locked(&task_rq(p)->lock); | |
1170 | ||
8f4d37ec | 1171 | if (unlikely(test_tsk_thread_flag(p, tif_bit))) |
c24d20db IM |
1172 | return; |
1173 | ||
8f4d37ec | 1174 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1175 | |
1176 | cpu = task_cpu(p); | |
1177 | if (cpu == smp_processor_id()) | |
1178 | return; | |
1179 | ||
1180 | /* NEED_RESCHED must be visible before we test polling */ | |
1181 | smp_mb(); | |
1182 | if (!tsk_is_polling(p)) | |
1183 | smp_send_reschedule(cpu); | |
1184 | } | |
1185 | ||
1186 | static void resched_cpu(int cpu) | |
1187 | { | |
1188 | struct rq *rq = cpu_rq(cpu); | |
1189 | unsigned long flags; | |
1190 | ||
1191 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1192 | return; | |
1193 | resched_task(cpu_curr(cpu)); | |
1194 | spin_unlock_irqrestore(&rq->lock, flags); | |
1195 | } | |
06d8308c TG |
1196 | |
1197 | #ifdef CONFIG_NO_HZ | |
1198 | /* | |
1199 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1200 | * idle CPU then this timer might expire before the next timer event | |
1201 | * which is scheduled to wake up that CPU. In case of a completely | |
1202 | * idle system the next event might even be infinite time into the | |
1203 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1204 | * leaves the inner idle loop so the newly added timer is taken into | |
1205 | * account when the CPU goes back to idle and evaluates the timer | |
1206 | * wheel for the next timer event. | |
1207 | */ | |
1208 | void wake_up_idle_cpu(int cpu) | |
1209 | { | |
1210 | struct rq *rq = cpu_rq(cpu); | |
1211 | ||
1212 | if (cpu == smp_processor_id()) | |
1213 | return; | |
1214 | ||
1215 | /* | |
1216 | * This is safe, as this function is called with the timer | |
1217 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1218 | * to idle and has not yet set rq->curr to idle then it will | |
1219 | * be serialized on the timer wheel base lock and take the new | |
1220 | * timer into account automatically. | |
1221 | */ | |
1222 | if (rq->curr != rq->idle) | |
1223 | return; | |
1224 | ||
1225 | /* | |
1226 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1227 | * lockless. The worst case is that the other CPU runs the | |
1228 | * idle task through an additional NOOP schedule() | |
1229 | */ | |
1230 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1231 | ||
1232 | /* NEED_RESCHED must be visible before we test polling */ | |
1233 | smp_mb(); | |
1234 | if (!tsk_is_polling(rq->idle)) | |
1235 | smp_send_reschedule(cpu); | |
1236 | } | |
1237 | #endif | |
1238 | ||
c24d20db | 1239 | #else |
8f4d37ec | 1240 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1241 | { |
1242 | assert_spin_locked(&task_rq(p)->lock); | |
8f4d37ec | 1243 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1244 | } |
1245 | #endif | |
1246 | ||
45bf76df IM |
1247 | #if BITS_PER_LONG == 32 |
1248 | # define WMULT_CONST (~0UL) | |
1249 | #else | |
1250 | # define WMULT_CONST (1UL << 32) | |
1251 | #endif | |
1252 | ||
1253 | #define WMULT_SHIFT 32 | |
1254 | ||
194081eb IM |
1255 | /* |
1256 | * Shift right and round: | |
1257 | */ | |
cf2ab469 | 1258 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1259 | |
cb1c4fc9 | 1260 | static unsigned long |
45bf76df IM |
1261 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1262 | struct load_weight *lw) | |
1263 | { | |
1264 | u64 tmp; | |
1265 | ||
1266 | if (unlikely(!lw->inv_weight)) | |
27d11726 | 1267 | lw->inv_weight = (WMULT_CONST-lw->weight/2) / (lw->weight+1); |
45bf76df IM |
1268 | |
1269 | tmp = (u64)delta_exec * weight; | |
1270 | /* | |
1271 | * Check whether we'd overflow the 64-bit multiplication: | |
1272 | */ | |
194081eb | 1273 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1274 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1275 | WMULT_SHIFT/2); |
1276 | else | |
cf2ab469 | 1277 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1278 | |
ecf691da | 1279 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1280 | } |
1281 | ||
1282 | static inline unsigned long | |
1283 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
1284 | { | |
1285 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
1286 | } | |
1287 | ||
1091985b | 1288 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1289 | { |
1290 | lw->weight += inc; | |
e89996ae | 1291 | lw->inv_weight = 0; |
45bf76df IM |
1292 | } |
1293 | ||
1091985b | 1294 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1295 | { |
1296 | lw->weight -= dec; | |
e89996ae | 1297 | lw->inv_weight = 0; |
45bf76df IM |
1298 | } |
1299 | ||
2dd73a4f PW |
1300 | /* |
1301 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1302 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1303 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1304 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1305 | * scaled version of the new time slice allocation that they receive on time |
1306 | * slice expiry etc. | |
1307 | */ | |
1308 | ||
dd41f596 IM |
1309 | #define WEIGHT_IDLEPRIO 2 |
1310 | #define WMULT_IDLEPRIO (1 << 31) | |
1311 | ||
1312 | /* | |
1313 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1314 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1315 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1316 | * that remained on nice 0. | |
1317 | * | |
1318 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1319 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1320 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1321 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1322 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1323 | */ |
1324 | static const int prio_to_weight[40] = { | |
254753dc IM |
1325 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1326 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1327 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1328 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1329 | /* 0 */ 1024, 820, 655, 526, 423, | |
1330 | /* 5 */ 335, 272, 215, 172, 137, | |
1331 | /* 10 */ 110, 87, 70, 56, 45, | |
1332 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1333 | }; |
1334 | ||
5714d2de IM |
1335 | /* |
1336 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1337 | * | |
1338 | * In cases where the weight does not change often, we can use the | |
1339 | * precalculated inverse to speed up arithmetics by turning divisions | |
1340 | * into multiplications: | |
1341 | */ | |
dd41f596 | 1342 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1343 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1344 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1345 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1346 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1347 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1348 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1349 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1350 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1351 | }; |
2dd73a4f | 1352 | |
dd41f596 IM |
1353 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1354 | ||
1355 | /* | |
1356 | * runqueue iterator, to support SMP load-balancing between different | |
1357 | * scheduling classes, without having to expose their internal data | |
1358 | * structures to the load-balancing proper: | |
1359 | */ | |
1360 | struct rq_iterator { | |
1361 | void *arg; | |
1362 | struct task_struct *(*start)(void *); | |
1363 | struct task_struct *(*next)(void *); | |
1364 | }; | |
1365 | ||
e1d1484f PW |
1366 | #ifdef CONFIG_SMP |
1367 | static unsigned long | |
1368 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1369 | unsigned long max_load_move, struct sched_domain *sd, | |
1370 | enum cpu_idle_type idle, int *all_pinned, | |
1371 | int *this_best_prio, struct rq_iterator *iterator); | |
1372 | ||
1373 | static int | |
1374 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1375 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1376 | struct rq_iterator *iterator); | |
e1d1484f | 1377 | #endif |
dd41f596 | 1378 | |
d842de87 SV |
1379 | #ifdef CONFIG_CGROUP_CPUACCT |
1380 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1381 | #else | |
1382 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1383 | #endif | |
1384 | ||
e7693a36 GH |
1385 | #ifdef CONFIG_SMP |
1386 | static unsigned long source_load(int cpu, int type); | |
1387 | static unsigned long target_load(int cpu, int type); | |
1388 | static unsigned long cpu_avg_load_per_task(int cpu); | |
1389 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1390 | #endif /* CONFIG_SMP */ | |
1391 | ||
dd41f596 | 1392 | #include "sched_stats.h" |
dd41f596 | 1393 | #include "sched_idletask.c" |
5522d5d5 IM |
1394 | #include "sched_fair.c" |
1395 | #include "sched_rt.c" | |
dd41f596 IM |
1396 | #ifdef CONFIG_SCHED_DEBUG |
1397 | # include "sched_debug.c" | |
1398 | #endif | |
1399 | ||
1400 | #define sched_class_highest (&rt_sched_class) | |
1401 | ||
62fb1851 PZ |
1402 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
1403 | { | |
1404 | update_load_add(&rq->load, p->se.load.weight); | |
1405 | } | |
1406 | ||
1407 | static inline void dec_load(struct rq *rq, const struct task_struct *p) | |
1408 | { | |
1409 | update_load_sub(&rq->load, p->se.load.weight); | |
1410 | } | |
1411 | ||
1412 | static void inc_nr_running(struct task_struct *p, struct rq *rq) | |
9c217245 IM |
1413 | { |
1414 | rq->nr_running++; | |
62fb1851 | 1415 | inc_load(rq, p); |
9c217245 IM |
1416 | } |
1417 | ||
62fb1851 | 1418 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
1419 | { |
1420 | rq->nr_running--; | |
62fb1851 | 1421 | dec_load(rq, p); |
9c217245 IM |
1422 | } |
1423 | ||
45bf76df IM |
1424 | static void set_load_weight(struct task_struct *p) |
1425 | { | |
1426 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1427 | p->se.load.weight = prio_to_weight[0] * 2; |
1428 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1429 | return; | |
1430 | } | |
45bf76df | 1431 | |
dd41f596 IM |
1432 | /* |
1433 | * SCHED_IDLE tasks get minimal weight: | |
1434 | */ | |
1435 | if (p->policy == SCHED_IDLE) { | |
1436 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1437 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1438 | return; | |
1439 | } | |
71f8bd46 | 1440 | |
dd41f596 IM |
1441 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1442 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1443 | } |
1444 | ||
8159f87e | 1445 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1446 | { |
dd41f596 | 1447 | sched_info_queued(p); |
fd390f6a | 1448 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1449 | p->se.on_rq = 1; |
71f8bd46 IM |
1450 | } |
1451 | ||
69be72c1 | 1452 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1453 | { |
f02231e5 | 1454 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1455 | p->se.on_rq = 0; |
71f8bd46 IM |
1456 | } |
1457 | ||
14531189 | 1458 | /* |
dd41f596 | 1459 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1460 | */ |
14531189 IM |
1461 | static inline int __normal_prio(struct task_struct *p) |
1462 | { | |
dd41f596 | 1463 | return p->static_prio; |
14531189 IM |
1464 | } |
1465 | ||
b29739f9 IM |
1466 | /* |
1467 | * Calculate the expected normal priority: i.e. priority | |
1468 | * without taking RT-inheritance into account. Might be | |
1469 | * boosted by interactivity modifiers. Changes upon fork, | |
1470 | * setprio syscalls, and whenever the interactivity | |
1471 | * estimator recalculates. | |
1472 | */ | |
36c8b586 | 1473 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1474 | { |
1475 | int prio; | |
1476 | ||
e05606d3 | 1477 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1478 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1479 | else | |
1480 | prio = __normal_prio(p); | |
1481 | return prio; | |
1482 | } | |
1483 | ||
1484 | /* | |
1485 | * Calculate the current priority, i.e. the priority | |
1486 | * taken into account by the scheduler. This value might | |
1487 | * be boosted by RT tasks, or might be boosted by | |
1488 | * interactivity modifiers. Will be RT if the task got | |
1489 | * RT-boosted. If not then it returns p->normal_prio. | |
1490 | */ | |
36c8b586 | 1491 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1492 | { |
1493 | p->normal_prio = normal_prio(p); | |
1494 | /* | |
1495 | * If we are RT tasks or we were boosted to RT priority, | |
1496 | * keep the priority unchanged. Otherwise, update priority | |
1497 | * to the normal priority: | |
1498 | */ | |
1499 | if (!rt_prio(p->prio)) | |
1500 | return p->normal_prio; | |
1501 | return p->prio; | |
1502 | } | |
1503 | ||
1da177e4 | 1504 | /* |
dd41f596 | 1505 | * activate_task - move a task to the runqueue. |
1da177e4 | 1506 | */ |
dd41f596 | 1507 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1508 | { |
d9514f6c | 1509 | if (task_contributes_to_load(p)) |
dd41f596 | 1510 | rq->nr_uninterruptible--; |
1da177e4 | 1511 | |
8159f87e | 1512 | enqueue_task(rq, p, wakeup); |
62fb1851 | 1513 | inc_nr_running(p, rq); |
1da177e4 LT |
1514 | } |
1515 | ||
1da177e4 LT |
1516 | /* |
1517 | * deactivate_task - remove a task from the runqueue. | |
1518 | */ | |
2e1cb74a | 1519 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1520 | { |
d9514f6c | 1521 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1522 | rq->nr_uninterruptible++; |
1523 | ||
69be72c1 | 1524 | dequeue_task(rq, p, sleep); |
62fb1851 | 1525 | dec_nr_running(p, rq); |
1da177e4 LT |
1526 | } |
1527 | ||
1da177e4 LT |
1528 | /** |
1529 | * task_curr - is this task currently executing on a CPU? | |
1530 | * @p: the task in question. | |
1531 | */ | |
36c8b586 | 1532 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1533 | { |
1534 | return cpu_curr(task_cpu(p)) == p; | |
1535 | } | |
1536 | ||
2dd73a4f PW |
1537 | /* Used instead of source_load when we know the type == 0 */ |
1538 | unsigned long weighted_cpuload(const int cpu) | |
1539 | { | |
495eca49 | 1540 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1541 | } |
1542 | ||
1543 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1544 | { | |
6f505b16 | 1545 | set_task_rq(p, cpu); |
dd41f596 | 1546 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1547 | /* |
1548 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1549 | * successfuly executed on another CPU. We must ensure that updates of | |
1550 | * per-task data have been completed by this moment. | |
1551 | */ | |
1552 | smp_wmb(); | |
dd41f596 | 1553 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1554 | #endif |
2dd73a4f PW |
1555 | } |
1556 | ||
cb469845 SR |
1557 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1558 | const struct sched_class *prev_class, | |
1559 | int oldprio, int running) | |
1560 | { | |
1561 | if (prev_class != p->sched_class) { | |
1562 | if (prev_class->switched_from) | |
1563 | prev_class->switched_from(rq, p, running); | |
1564 | p->sched_class->switched_to(rq, p, running); | |
1565 | } else | |
1566 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1567 | } | |
1568 | ||
1da177e4 | 1569 | #ifdef CONFIG_SMP |
c65cc870 | 1570 | |
cc367732 IM |
1571 | /* |
1572 | * Is this task likely cache-hot: | |
1573 | */ | |
e7693a36 | 1574 | static int |
cc367732 IM |
1575 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1576 | { | |
1577 | s64 delta; | |
1578 | ||
f540a608 IM |
1579 | /* |
1580 | * Buddy candidates are cache hot: | |
1581 | */ | |
d25ce4cd | 1582 | if (sched_feat(CACHE_HOT_BUDDY) && (&p->se == cfs_rq_of(&p->se)->next)) |
f540a608 IM |
1583 | return 1; |
1584 | ||
cc367732 IM |
1585 | if (p->sched_class != &fair_sched_class) |
1586 | return 0; | |
1587 | ||
6bc1665b IM |
1588 | if (sysctl_sched_migration_cost == -1) |
1589 | return 1; | |
1590 | if (sysctl_sched_migration_cost == 0) | |
1591 | return 0; | |
1592 | ||
cc367732 IM |
1593 | delta = now - p->se.exec_start; |
1594 | ||
1595 | return delta < (s64)sysctl_sched_migration_cost; | |
1596 | } | |
1597 | ||
1598 | ||
dd41f596 | 1599 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1600 | { |
dd41f596 IM |
1601 | int old_cpu = task_cpu(p); |
1602 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1603 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1604 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1605 | u64 clock_offset; |
dd41f596 IM |
1606 | |
1607 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1608 | |
1609 | #ifdef CONFIG_SCHEDSTATS | |
1610 | if (p->se.wait_start) | |
1611 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1612 | if (p->se.sleep_start) |
1613 | p->se.sleep_start -= clock_offset; | |
1614 | if (p->se.block_start) | |
1615 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1616 | if (old_cpu != new_cpu) { |
1617 | schedstat_inc(p, se.nr_migrations); | |
1618 | if (task_hot(p, old_rq->clock, NULL)) | |
1619 | schedstat_inc(p, se.nr_forced2_migrations); | |
1620 | } | |
6cfb0d5d | 1621 | #endif |
2830cf8c SV |
1622 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1623 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1624 | |
1625 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1626 | } |
1627 | ||
70b97a7f | 1628 | struct migration_req { |
1da177e4 | 1629 | struct list_head list; |
1da177e4 | 1630 | |
36c8b586 | 1631 | struct task_struct *task; |
1da177e4 LT |
1632 | int dest_cpu; |
1633 | ||
1da177e4 | 1634 | struct completion done; |
70b97a7f | 1635 | }; |
1da177e4 LT |
1636 | |
1637 | /* | |
1638 | * The task's runqueue lock must be held. | |
1639 | * Returns true if you have to wait for migration thread. | |
1640 | */ | |
36c8b586 | 1641 | static int |
70b97a7f | 1642 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1643 | { |
70b97a7f | 1644 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1645 | |
1646 | /* | |
1647 | * If the task is not on a runqueue (and not running), then | |
1648 | * it is sufficient to simply update the task's cpu field. | |
1649 | */ | |
dd41f596 | 1650 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1651 | set_task_cpu(p, dest_cpu); |
1652 | return 0; | |
1653 | } | |
1654 | ||
1655 | init_completion(&req->done); | |
1da177e4 LT |
1656 | req->task = p; |
1657 | req->dest_cpu = dest_cpu; | |
1658 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1659 | |
1da177e4 LT |
1660 | return 1; |
1661 | } | |
1662 | ||
1663 | /* | |
1664 | * wait_task_inactive - wait for a thread to unschedule. | |
1665 | * | |
1666 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1667 | * else this function might spin for a *long* time. This function can't | |
1668 | * be called with interrupts off, or it may introduce deadlock with | |
1669 | * smp_call_function() if an IPI is sent by the same process we are | |
1670 | * waiting to become inactive. | |
1671 | */ | |
36c8b586 | 1672 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1673 | { |
1674 | unsigned long flags; | |
dd41f596 | 1675 | int running, on_rq; |
70b97a7f | 1676 | struct rq *rq; |
1da177e4 | 1677 | |
3a5c359a AK |
1678 | for (;;) { |
1679 | /* | |
1680 | * We do the initial early heuristics without holding | |
1681 | * any task-queue locks at all. We'll only try to get | |
1682 | * the runqueue lock when things look like they will | |
1683 | * work out! | |
1684 | */ | |
1685 | rq = task_rq(p); | |
fa490cfd | 1686 | |
3a5c359a AK |
1687 | /* |
1688 | * If the task is actively running on another CPU | |
1689 | * still, just relax and busy-wait without holding | |
1690 | * any locks. | |
1691 | * | |
1692 | * NOTE! Since we don't hold any locks, it's not | |
1693 | * even sure that "rq" stays as the right runqueue! | |
1694 | * But we don't care, since "task_running()" will | |
1695 | * return false if the runqueue has changed and p | |
1696 | * is actually now running somewhere else! | |
1697 | */ | |
1698 | while (task_running(rq, p)) | |
1699 | cpu_relax(); | |
fa490cfd | 1700 | |
3a5c359a AK |
1701 | /* |
1702 | * Ok, time to look more closely! We need the rq | |
1703 | * lock now, to be *sure*. If we're wrong, we'll | |
1704 | * just go back and repeat. | |
1705 | */ | |
1706 | rq = task_rq_lock(p, &flags); | |
1707 | running = task_running(rq, p); | |
1708 | on_rq = p->se.on_rq; | |
1709 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1710 | |
3a5c359a AK |
1711 | /* |
1712 | * Was it really running after all now that we | |
1713 | * checked with the proper locks actually held? | |
1714 | * | |
1715 | * Oops. Go back and try again.. | |
1716 | */ | |
1717 | if (unlikely(running)) { | |
1718 | cpu_relax(); | |
1719 | continue; | |
1720 | } | |
fa490cfd | 1721 | |
3a5c359a AK |
1722 | /* |
1723 | * It's not enough that it's not actively running, | |
1724 | * it must be off the runqueue _entirely_, and not | |
1725 | * preempted! | |
1726 | * | |
1727 | * So if it wa still runnable (but just not actively | |
1728 | * running right now), it's preempted, and we should | |
1729 | * yield - it could be a while. | |
1730 | */ | |
1731 | if (unlikely(on_rq)) { | |
1732 | schedule_timeout_uninterruptible(1); | |
1733 | continue; | |
1734 | } | |
fa490cfd | 1735 | |
3a5c359a AK |
1736 | /* |
1737 | * Ahh, all good. It wasn't running, and it wasn't | |
1738 | * runnable, which means that it will never become | |
1739 | * running in the future either. We're all done! | |
1740 | */ | |
1741 | break; | |
1742 | } | |
1da177e4 LT |
1743 | } |
1744 | ||
1745 | /*** | |
1746 | * kick_process - kick a running thread to enter/exit the kernel | |
1747 | * @p: the to-be-kicked thread | |
1748 | * | |
1749 | * Cause a process which is running on another CPU to enter | |
1750 | * kernel-mode, without any delay. (to get signals handled.) | |
1751 | * | |
1752 | * NOTE: this function doesnt have to take the runqueue lock, | |
1753 | * because all it wants to ensure is that the remote task enters | |
1754 | * the kernel. If the IPI races and the task has been migrated | |
1755 | * to another CPU then no harm is done and the purpose has been | |
1756 | * achieved as well. | |
1757 | */ | |
36c8b586 | 1758 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1759 | { |
1760 | int cpu; | |
1761 | ||
1762 | preempt_disable(); | |
1763 | cpu = task_cpu(p); | |
1764 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1765 | smp_send_reschedule(cpu); | |
1766 | preempt_enable(); | |
1767 | } | |
1768 | ||
1769 | /* | |
2dd73a4f PW |
1770 | * Return a low guess at the load of a migration-source cpu weighted |
1771 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1772 | * |
1773 | * We want to under-estimate the load of migration sources, to | |
1774 | * balance conservatively. | |
1775 | */ | |
a9957449 | 1776 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1777 | { |
70b97a7f | 1778 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1779 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1780 | |
3b0bd9bc | 1781 | if (type == 0) |
dd41f596 | 1782 | return total; |
b910472d | 1783 | |
dd41f596 | 1784 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1785 | } |
1786 | ||
1787 | /* | |
2dd73a4f PW |
1788 | * Return a high guess at the load of a migration-target cpu weighted |
1789 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1790 | */ |
a9957449 | 1791 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1792 | { |
70b97a7f | 1793 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1794 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1795 | |
7897986b | 1796 | if (type == 0) |
dd41f596 | 1797 | return total; |
3b0bd9bc | 1798 | |
dd41f596 | 1799 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1800 | } |
1801 | ||
1802 | /* | |
1803 | * Return the average load per task on the cpu's run queue | |
1804 | */ | |
e7693a36 | 1805 | static unsigned long cpu_avg_load_per_task(int cpu) |
2dd73a4f | 1806 | { |
70b97a7f | 1807 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1808 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1809 | unsigned long n = rq->nr_running; |
1810 | ||
dd41f596 | 1811 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1812 | } |
1813 | ||
147cbb4b NP |
1814 | /* |
1815 | * find_idlest_group finds and returns the least busy CPU group within the | |
1816 | * domain. | |
1817 | */ | |
1818 | static struct sched_group * | |
1819 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1820 | { | |
1821 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1822 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1823 | int load_idx = sd->forkexec_idx; | |
1824 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1825 | ||
1826 | do { | |
1827 | unsigned long load, avg_load; | |
1828 | int local_group; | |
1829 | int i; | |
1830 | ||
da5a5522 BD |
1831 | /* Skip over this group if it has no CPUs allowed */ |
1832 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1833 | continue; |
da5a5522 | 1834 | |
147cbb4b | 1835 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1836 | |
1837 | /* Tally up the load of all CPUs in the group */ | |
1838 | avg_load = 0; | |
1839 | ||
1840 | for_each_cpu_mask(i, group->cpumask) { | |
1841 | /* Bias balancing toward cpus of our domain */ | |
1842 | if (local_group) | |
1843 | load = source_load(i, load_idx); | |
1844 | else | |
1845 | load = target_load(i, load_idx); | |
1846 | ||
1847 | avg_load += load; | |
1848 | } | |
1849 | ||
1850 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1851 | avg_load = sg_div_cpu_power(group, |
1852 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1853 | |
1854 | if (local_group) { | |
1855 | this_load = avg_load; | |
1856 | this = group; | |
1857 | } else if (avg_load < min_load) { | |
1858 | min_load = avg_load; | |
1859 | idlest = group; | |
1860 | } | |
3a5c359a | 1861 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1862 | |
1863 | if (!idlest || 100*this_load < imbalance*min_load) | |
1864 | return NULL; | |
1865 | return idlest; | |
1866 | } | |
1867 | ||
1868 | /* | |
0feaece9 | 1869 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1870 | */ |
95cdf3b7 IM |
1871 | static int |
1872 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1873 | { |
da5a5522 | 1874 | cpumask_t tmp; |
147cbb4b NP |
1875 | unsigned long load, min_load = ULONG_MAX; |
1876 | int idlest = -1; | |
1877 | int i; | |
1878 | ||
da5a5522 BD |
1879 | /* Traverse only the allowed CPUs */ |
1880 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1881 | ||
1882 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1883 | load = weighted_cpuload(i); |
147cbb4b NP |
1884 | |
1885 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1886 | min_load = load; | |
1887 | idlest = i; | |
1888 | } | |
1889 | } | |
1890 | ||
1891 | return idlest; | |
1892 | } | |
1893 | ||
476d139c NP |
1894 | /* |
1895 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1896 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1897 | * SD_BALANCE_EXEC. | |
1898 | * | |
1899 | * Balance, ie. select the least loaded group. | |
1900 | * | |
1901 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1902 | * | |
1903 | * preempt must be disabled. | |
1904 | */ | |
1905 | static int sched_balance_self(int cpu, int flag) | |
1906 | { | |
1907 | struct task_struct *t = current; | |
1908 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1909 | |
c96d145e | 1910 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1911 | /* |
1912 | * If power savings logic is enabled for a domain, stop there. | |
1913 | */ | |
5c45bf27 SS |
1914 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1915 | break; | |
476d139c NP |
1916 | if (tmp->flags & flag) |
1917 | sd = tmp; | |
c96d145e | 1918 | } |
476d139c NP |
1919 | |
1920 | while (sd) { | |
1921 | cpumask_t span; | |
1922 | struct sched_group *group; | |
1a848870 SS |
1923 | int new_cpu, weight; |
1924 | ||
1925 | if (!(sd->flags & flag)) { | |
1926 | sd = sd->child; | |
1927 | continue; | |
1928 | } | |
476d139c NP |
1929 | |
1930 | span = sd->span; | |
1931 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1932 | if (!group) { |
1933 | sd = sd->child; | |
1934 | continue; | |
1935 | } | |
476d139c | 1936 | |
da5a5522 | 1937 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1938 | if (new_cpu == -1 || new_cpu == cpu) { |
1939 | /* Now try balancing at a lower domain level of cpu */ | |
1940 | sd = sd->child; | |
1941 | continue; | |
1942 | } | |
476d139c | 1943 | |
1a848870 | 1944 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1945 | cpu = new_cpu; |
476d139c NP |
1946 | sd = NULL; |
1947 | weight = cpus_weight(span); | |
1948 | for_each_domain(cpu, tmp) { | |
1949 | if (weight <= cpus_weight(tmp->span)) | |
1950 | break; | |
1951 | if (tmp->flags & flag) | |
1952 | sd = tmp; | |
1953 | } | |
1954 | /* while loop will break here if sd == NULL */ | |
1955 | } | |
1956 | ||
1957 | return cpu; | |
1958 | } | |
1959 | ||
1960 | #endif /* CONFIG_SMP */ | |
1da177e4 | 1961 | |
1da177e4 LT |
1962 | /*** |
1963 | * try_to_wake_up - wake up a thread | |
1964 | * @p: the to-be-woken-up thread | |
1965 | * @state: the mask of task states that can be woken | |
1966 | * @sync: do a synchronous wakeup? | |
1967 | * | |
1968 | * Put it on the run-queue if it's not already there. The "current" | |
1969 | * thread is always on the run-queue (except when the actual | |
1970 | * re-schedule is in progress), and as such you're allowed to do | |
1971 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1972 | * runnable without the overhead of this. | |
1973 | * | |
1974 | * returns failure only if the task is already active. | |
1975 | */ | |
36c8b586 | 1976 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1977 | { |
cc367732 | 1978 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1979 | unsigned long flags; |
1980 | long old_state; | |
70b97a7f | 1981 | struct rq *rq; |
1da177e4 | 1982 | |
b85d0667 IM |
1983 | if (!sched_feat(SYNC_WAKEUPS)) |
1984 | sync = 0; | |
1985 | ||
04e2f174 | 1986 | smp_wmb(); |
1da177e4 LT |
1987 | rq = task_rq_lock(p, &flags); |
1988 | old_state = p->state; | |
1989 | if (!(old_state & state)) | |
1990 | goto out; | |
1991 | ||
dd41f596 | 1992 | if (p->se.on_rq) |
1da177e4 LT |
1993 | goto out_running; |
1994 | ||
1995 | cpu = task_cpu(p); | |
cc367732 | 1996 | orig_cpu = cpu; |
1da177e4 LT |
1997 | this_cpu = smp_processor_id(); |
1998 | ||
1999 | #ifdef CONFIG_SMP | |
2000 | if (unlikely(task_running(rq, p))) | |
2001 | goto out_activate; | |
2002 | ||
5d2f5a61 DA |
2003 | cpu = p->sched_class->select_task_rq(p, sync); |
2004 | if (cpu != orig_cpu) { | |
2005 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2006 | task_rq_unlock(rq, &flags); |
2007 | /* might preempt at this point */ | |
2008 | rq = task_rq_lock(p, &flags); | |
2009 | old_state = p->state; | |
2010 | if (!(old_state & state)) | |
2011 | goto out; | |
dd41f596 | 2012 | if (p->se.on_rq) |
1da177e4 LT |
2013 | goto out_running; |
2014 | ||
2015 | this_cpu = smp_processor_id(); | |
2016 | cpu = task_cpu(p); | |
2017 | } | |
2018 | ||
e7693a36 GH |
2019 | #ifdef CONFIG_SCHEDSTATS |
2020 | schedstat_inc(rq, ttwu_count); | |
2021 | if (cpu == this_cpu) | |
2022 | schedstat_inc(rq, ttwu_local); | |
2023 | else { | |
2024 | struct sched_domain *sd; | |
2025 | for_each_domain(this_cpu, sd) { | |
2026 | if (cpu_isset(cpu, sd->span)) { | |
2027 | schedstat_inc(sd, ttwu_wake_remote); | |
2028 | break; | |
2029 | } | |
2030 | } | |
2031 | } | |
e7693a36 GH |
2032 | #endif |
2033 | ||
1da177e4 LT |
2034 | out_activate: |
2035 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2036 | schedstat_inc(p, se.nr_wakeups); |
2037 | if (sync) | |
2038 | schedstat_inc(p, se.nr_wakeups_sync); | |
2039 | if (orig_cpu != cpu) | |
2040 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2041 | if (cpu == this_cpu) | |
2042 | schedstat_inc(p, se.nr_wakeups_local); | |
2043 | else | |
2044 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2045 | update_rq_clock(rq); |
dd41f596 | 2046 | activate_task(rq, p, 1); |
1da177e4 LT |
2047 | success = 1; |
2048 | ||
2049 | out_running: | |
4ae7d5ce IM |
2050 | check_preempt_curr(rq, p); |
2051 | ||
1da177e4 | 2052 | p->state = TASK_RUNNING; |
9a897c5a SR |
2053 | #ifdef CONFIG_SMP |
2054 | if (p->sched_class->task_wake_up) | |
2055 | p->sched_class->task_wake_up(rq, p); | |
2056 | #endif | |
1da177e4 LT |
2057 | out: |
2058 | task_rq_unlock(rq, &flags); | |
2059 | ||
2060 | return success; | |
2061 | } | |
2062 | ||
7ad5b3a5 | 2063 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2064 | { |
d9514f6c | 2065 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2066 | } |
1da177e4 LT |
2067 | EXPORT_SYMBOL(wake_up_process); |
2068 | ||
7ad5b3a5 | 2069 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2070 | { |
2071 | return try_to_wake_up(p, state, 0); | |
2072 | } | |
2073 | ||
1da177e4 LT |
2074 | /* |
2075 | * Perform scheduler related setup for a newly forked process p. | |
2076 | * p is forked by current. | |
dd41f596 IM |
2077 | * |
2078 | * __sched_fork() is basic setup used by init_idle() too: | |
2079 | */ | |
2080 | static void __sched_fork(struct task_struct *p) | |
2081 | { | |
dd41f596 IM |
2082 | p->se.exec_start = 0; |
2083 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2084 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2085 | p->se.last_wakeup = 0; |
2086 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2087 | |
2088 | #ifdef CONFIG_SCHEDSTATS | |
2089 | p->se.wait_start = 0; | |
dd41f596 IM |
2090 | p->se.sum_sleep_runtime = 0; |
2091 | p->se.sleep_start = 0; | |
dd41f596 IM |
2092 | p->se.block_start = 0; |
2093 | p->se.sleep_max = 0; | |
2094 | p->se.block_max = 0; | |
2095 | p->se.exec_max = 0; | |
eba1ed4b | 2096 | p->se.slice_max = 0; |
dd41f596 | 2097 | p->se.wait_max = 0; |
6cfb0d5d | 2098 | #endif |
476d139c | 2099 | |
fa717060 | 2100 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2101 | p->se.on_rq = 0; |
476d139c | 2102 | |
e107be36 AK |
2103 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2104 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2105 | #endif | |
2106 | ||
1da177e4 LT |
2107 | /* |
2108 | * We mark the process as running here, but have not actually | |
2109 | * inserted it onto the runqueue yet. This guarantees that | |
2110 | * nobody will actually run it, and a signal or other external | |
2111 | * event cannot wake it up and insert it on the runqueue either. | |
2112 | */ | |
2113 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2114 | } |
2115 | ||
2116 | /* | |
2117 | * fork()/clone()-time setup: | |
2118 | */ | |
2119 | void sched_fork(struct task_struct *p, int clone_flags) | |
2120 | { | |
2121 | int cpu = get_cpu(); | |
2122 | ||
2123 | __sched_fork(p); | |
2124 | ||
2125 | #ifdef CONFIG_SMP | |
2126 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2127 | #endif | |
02e4bac2 | 2128 | set_task_cpu(p, cpu); |
b29739f9 IM |
2129 | |
2130 | /* | |
2131 | * Make sure we do not leak PI boosting priority to the child: | |
2132 | */ | |
2133 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2134 | if (!rt_prio(p->prio)) |
2135 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2136 | |
52f17b6c | 2137 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2138 | if (likely(sched_info_on())) |
52f17b6c | 2139 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2140 | #endif |
d6077cb8 | 2141 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2142 | p->oncpu = 0; |
2143 | #endif | |
1da177e4 | 2144 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2145 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2146 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2147 | #endif |
476d139c | 2148 | put_cpu(); |
1da177e4 LT |
2149 | } |
2150 | ||
2151 | /* | |
2152 | * wake_up_new_task - wake up a newly created task for the first time. | |
2153 | * | |
2154 | * This function will do some initial scheduler statistics housekeeping | |
2155 | * that must be done for every newly created context, then puts the task | |
2156 | * on the runqueue and wakes it. | |
2157 | */ | |
7ad5b3a5 | 2158 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2159 | { |
2160 | unsigned long flags; | |
dd41f596 | 2161 | struct rq *rq; |
1da177e4 LT |
2162 | |
2163 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2164 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2165 | update_rq_clock(rq); |
1da177e4 LT |
2166 | |
2167 | p->prio = effective_prio(p); | |
2168 | ||
b9dca1e0 | 2169 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2170 | activate_task(rq, p, 0); |
1da177e4 | 2171 | } else { |
1da177e4 | 2172 | /* |
dd41f596 IM |
2173 | * Let the scheduling class do new task startup |
2174 | * management (if any): | |
1da177e4 | 2175 | */ |
ee0827d8 | 2176 | p->sched_class->task_new(rq, p); |
62fb1851 | 2177 | inc_nr_running(p, rq); |
1da177e4 | 2178 | } |
dd41f596 | 2179 | check_preempt_curr(rq, p); |
9a897c5a SR |
2180 | #ifdef CONFIG_SMP |
2181 | if (p->sched_class->task_wake_up) | |
2182 | p->sched_class->task_wake_up(rq, p); | |
2183 | #endif | |
dd41f596 | 2184 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2185 | } |
2186 | ||
e107be36 AK |
2187 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2188 | ||
2189 | /** | |
421cee29 RD |
2190 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2191 | * @notifier: notifier struct to register | |
e107be36 AK |
2192 | */ |
2193 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2194 | { | |
2195 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2196 | } | |
2197 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2198 | ||
2199 | /** | |
2200 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2201 | * @notifier: notifier struct to unregister |
e107be36 AK |
2202 | * |
2203 | * This is safe to call from within a preemption notifier. | |
2204 | */ | |
2205 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2206 | { | |
2207 | hlist_del(¬ifier->link); | |
2208 | } | |
2209 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2210 | ||
2211 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2212 | { | |
2213 | struct preempt_notifier *notifier; | |
2214 | struct hlist_node *node; | |
2215 | ||
2216 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2217 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2218 | } | |
2219 | ||
2220 | static void | |
2221 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2222 | struct task_struct *next) | |
2223 | { | |
2224 | struct preempt_notifier *notifier; | |
2225 | struct hlist_node *node; | |
2226 | ||
2227 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2228 | notifier->ops->sched_out(notifier, next); | |
2229 | } | |
2230 | ||
2231 | #else | |
2232 | ||
2233 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2234 | { | |
2235 | } | |
2236 | ||
2237 | static void | |
2238 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2239 | struct task_struct *next) | |
2240 | { | |
2241 | } | |
2242 | ||
2243 | #endif | |
2244 | ||
4866cde0 NP |
2245 | /** |
2246 | * prepare_task_switch - prepare to switch tasks | |
2247 | * @rq: the runqueue preparing to switch | |
421cee29 | 2248 | * @prev: the current task that is being switched out |
4866cde0 NP |
2249 | * @next: the task we are going to switch to. |
2250 | * | |
2251 | * This is called with the rq lock held and interrupts off. It must | |
2252 | * be paired with a subsequent finish_task_switch after the context | |
2253 | * switch. | |
2254 | * | |
2255 | * prepare_task_switch sets up locking and calls architecture specific | |
2256 | * hooks. | |
2257 | */ | |
e107be36 AK |
2258 | static inline void |
2259 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2260 | struct task_struct *next) | |
4866cde0 | 2261 | { |
e107be36 | 2262 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2263 | prepare_lock_switch(rq, next); |
2264 | prepare_arch_switch(next); | |
2265 | } | |
2266 | ||
1da177e4 LT |
2267 | /** |
2268 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2269 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2270 | * @prev: the thread we just switched away from. |
2271 | * | |
4866cde0 NP |
2272 | * finish_task_switch must be called after the context switch, paired |
2273 | * with a prepare_task_switch call before the context switch. | |
2274 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2275 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2276 | * |
2277 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2278 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2279 | * with the lock held can cause deadlocks; see schedule() for |
2280 | * details.) | |
2281 | */ | |
a9957449 | 2282 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2283 | __releases(rq->lock) |
2284 | { | |
1da177e4 | 2285 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2286 | long prev_state; |
1da177e4 LT |
2287 | |
2288 | rq->prev_mm = NULL; | |
2289 | ||
2290 | /* | |
2291 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2292 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2293 | * schedule one last time. The schedule call will never return, and |
2294 | * the scheduled task must drop that reference. | |
c394cc9f | 2295 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2296 | * still held, otherwise prev could be scheduled on another cpu, die |
2297 | * there before we look at prev->state, and then the reference would | |
2298 | * be dropped twice. | |
2299 | * Manfred Spraul <manfred@colorfullife.com> | |
2300 | */ | |
55a101f8 | 2301 | prev_state = prev->state; |
4866cde0 NP |
2302 | finish_arch_switch(prev); |
2303 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2304 | #ifdef CONFIG_SMP |
2305 | if (current->sched_class->post_schedule) | |
2306 | current->sched_class->post_schedule(rq); | |
2307 | #endif | |
e8fa1362 | 2308 | |
e107be36 | 2309 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2310 | if (mm) |
2311 | mmdrop(mm); | |
c394cc9f | 2312 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2313 | /* |
2314 | * Remove function-return probe instances associated with this | |
2315 | * task and put them back on the free list. | |
9761eea8 | 2316 | */ |
c6fd91f0 | 2317 | kprobe_flush_task(prev); |
1da177e4 | 2318 | put_task_struct(prev); |
c6fd91f0 | 2319 | } |
1da177e4 LT |
2320 | } |
2321 | ||
2322 | /** | |
2323 | * schedule_tail - first thing a freshly forked thread must call. | |
2324 | * @prev: the thread we just switched away from. | |
2325 | */ | |
36c8b586 | 2326 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2327 | __releases(rq->lock) |
2328 | { | |
70b97a7f IM |
2329 | struct rq *rq = this_rq(); |
2330 | ||
4866cde0 NP |
2331 | finish_task_switch(rq, prev); |
2332 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2333 | /* In this case, finish_task_switch does not reenable preemption */ | |
2334 | preempt_enable(); | |
2335 | #endif | |
1da177e4 | 2336 | if (current->set_child_tid) |
b488893a | 2337 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2338 | } |
2339 | ||
2340 | /* | |
2341 | * context_switch - switch to the new MM and the new | |
2342 | * thread's register state. | |
2343 | */ | |
dd41f596 | 2344 | static inline void |
70b97a7f | 2345 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2346 | struct task_struct *next) |
1da177e4 | 2347 | { |
dd41f596 | 2348 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2349 | |
e107be36 | 2350 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
2351 | mm = next->mm; |
2352 | oldmm = prev->active_mm; | |
9226d125 ZA |
2353 | /* |
2354 | * For paravirt, this is coupled with an exit in switch_to to | |
2355 | * combine the page table reload and the switch backend into | |
2356 | * one hypercall. | |
2357 | */ | |
2358 | arch_enter_lazy_cpu_mode(); | |
2359 | ||
dd41f596 | 2360 | if (unlikely(!mm)) { |
1da177e4 LT |
2361 | next->active_mm = oldmm; |
2362 | atomic_inc(&oldmm->mm_count); | |
2363 | enter_lazy_tlb(oldmm, next); | |
2364 | } else | |
2365 | switch_mm(oldmm, mm, next); | |
2366 | ||
dd41f596 | 2367 | if (unlikely(!prev->mm)) { |
1da177e4 | 2368 | prev->active_mm = NULL; |
1da177e4 LT |
2369 | rq->prev_mm = oldmm; |
2370 | } | |
3a5f5e48 IM |
2371 | /* |
2372 | * Since the runqueue lock will be released by the next | |
2373 | * task (which is an invalid locking op but in the case | |
2374 | * of the scheduler it's an obvious special-case), so we | |
2375 | * do an early lockdep release here: | |
2376 | */ | |
2377 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2378 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2379 | #endif |
1da177e4 LT |
2380 | |
2381 | /* Here we just switch the register state and the stack. */ | |
2382 | switch_to(prev, next, prev); | |
2383 | ||
dd41f596 IM |
2384 | barrier(); |
2385 | /* | |
2386 | * this_rq must be evaluated again because prev may have moved | |
2387 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2388 | * frame will be invalid. | |
2389 | */ | |
2390 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2391 | } |
2392 | ||
2393 | /* | |
2394 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2395 | * | |
2396 | * externally visible scheduler statistics: current number of runnable | |
2397 | * threads, current number of uninterruptible-sleeping threads, total | |
2398 | * number of context switches performed since bootup. | |
2399 | */ | |
2400 | unsigned long nr_running(void) | |
2401 | { | |
2402 | unsigned long i, sum = 0; | |
2403 | ||
2404 | for_each_online_cpu(i) | |
2405 | sum += cpu_rq(i)->nr_running; | |
2406 | ||
2407 | return sum; | |
2408 | } | |
2409 | ||
2410 | unsigned long nr_uninterruptible(void) | |
2411 | { | |
2412 | unsigned long i, sum = 0; | |
2413 | ||
0a945022 | 2414 | for_each_possible_cpu(i) |
1da177e4 LT |
2415 | sum += cpu_rq(i)->nr_uninterruptible; |
2416 | ||
2417 | /* | |
2418 | * Since we read the counters lockless, it might be slightly | |
2419 | * inaccurate. Do not allow it to go below zero though: | |
2420 | */ | |
2421 | if (unlikely((long)sum < 0)) | |
2422 | sum = 0; | |
2423 | ||
2424 | return sum; | |
2425 | } | |
2426 | ||
2427 | unsigned long long nr_context_switches(void) | |
2428 | { | |
cc94abfc SR |
2429 | int i; |
2430 | unsigned long long sum = 0; | |
1da177e4 | 2431 | |
0a945022 | 2432 | for_each_possible_cpu(i) |
1da177e4 LT |
2433 | sum += cpu_rq(i)->nr_switches; |
2434 | ||
2435 | return sum; | |
2436 | } | |
2437 | ||
2438 | unsigned long nr_iowait(void) | |
2439 | { | |
2440 | unsigned long i, sum = 0; | |
2441 | ||
0a945022 | 2442 | for_each_possible_cpu(i) |
1da177e4 LT |
2443 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2444 | ||
2445 | return sum; | |
2446 | } | |
2447 | ||
db1b1fef JS |
2448 | unsigned long nr_active(void) |
2449 | { | |
2450 | unsigned long i, running = 0, uninterruptible = 0; | |
2451 | ||
2452 | for_each_online_cpu(i) { | |
2453 | running += cpu_rq(i)->nr_running; | |
2454 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2455 | } | |
2456 | ||
2457 | if (unlikely((long)uninterruptible < 0)) | |
2458 | uninterruptible = 0; | |
2459 | ||
2460 | return running + uninterruptible; | |
2461 | } | |
2462 | ||
48f24c4d | 2463 | /* |
dd41f596 IM |
2464 | * Update rq->cpu_load[] statistics. This function is usually called every |
2465 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2466 | */ |
dd41f596 | 2467 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2468 | { |
495eca49 | 2469 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2470 | int i, scale; |
2471 | ||
2472 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2473 | |
2474 | /* Update our load: */ | |
2475 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2476 | unsigned long old_load, new_load; | |
2477 | ||
2478 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2479 | ||
2480 | old_load = this_rq->cpu_load[i]; | |
2481 | new_load = this_load; | |
a25707f3 IM |
2482 | /* |
2483 | * Round up the averaging division if load is increasing. This | |
2484 | * prevents us from getting stuck on 9 if the load is 10, for | |
2485 | * example. | |
2486 | */ | |
2487 | if (new_load > old_load) | |
2488 | new_load += scale-1; | |
dd41f596 IM |
2489 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2490 | } | |
48f24c4d IM |
2491 | } |
2492 | ||
dd41f596 IM |
2493 | #ifdef CONFIG_SMP |
2494 | ||
1da177e4 LT |
2495 | /* |
2496 | * double_rq_lock - safely lock two runqueues | |
2497 | * | |
2498 | * Note this does not disable interrupts like task_rq_lock, | |
2499 | * you need to do so manually before calling. | |
2500 | */ | |
70b97a7f | 2501 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2502 | __acquires(rq1->lock) |
2503 | __acquires(rq2->lock) | |
2504 | { | |
054b9108 | 2505 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2506 | if (rq1 == rq2) { |
2507 | spin_lock(&rq1->lock); | |
2508 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2509 | } else { | |
c96d145e | 2510 | if (rq1 < rq2) { |
1da177e4 LT |
2511 | spin_lock(&rq1->lock); |
2512 | spin_lock(&rq2->lock); | |
2513 | } else { | |
2514 | spin_lock(&rq2->lock); | |
2515 | spin_lock(&rq1->lock); | |
2516 | } | |
2517 | } | |
6e82a3be IM |
2518 | update_rq_clock(rq1); |
2519 | update_rq_clock(rq2); | |
1da177e4 LT |
2520 | } |
2521 | ||
2522 | /* | |
2523 | * double_rq_unlock - safely unlock two runqueues | |
2524 | * | |
2525 | * Note this does not restore interrupts like task_rq_unlock, | |
2526 | * you need to do so manually after calling. | |
2527 | */ | |
70b97a7f | 2528 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2529 | __releases(rq1->lock) |
2530 | __releases(rq2->lock) | |
2531 | { | |
2532 | spin_unlock(&rq1->lock); | |
2533 | if (rq1 != rq2) | |
2534 | spin_unlock(&rq2->lock); | |
2535 | else | |
2536 | __release(rq2->lock); | |
2537 | } | |
2538 | ||
2539 | /* | |
2540 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2541 | */ | |
e8fa1362 | 2542 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2543 | __releases(this_rq->lock) |
2544 | __acquires(busiest->lock) | |
2545 | __acquires(this_rq->lock) | |
2546 | { | |
e8fa1362 SR |
2547 | int ret = 0; |
2548 | ||
054b9108 KK |
2549 | if (unlikely(!irqs_disabled())) { |
2550 | /* printk() doesn't work good under rq->lock */ | |
2551 | spin_unlock(&this_rq->lock); | |
2552 | BUG_ON(1); | |
2553 | } | |
1da177e4 | 2554 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2555 | if (busiest < this_rq) { |
1da177e4 LT |
2556 | spin_unlock(&this_rq->lock); |
2557 | spin_lock(&busiest->lock); | |
2558 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2559 | ret = 1; |
1da177e4 LT |
2560 | } else |
2561 | spin_lock(&busiest->lock); | |
2562 | } | |
e8fa1362 | 2563 | return ret; |
1da177e4 LT |
2564 | } |
2565 | ||
1da177e4 LT |
2566 | /* |
2567 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2568 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2569 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2570 | * the cpu_allowed mask is restored. |
2571 | */ | |
36c8b586 | 2572 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2573 | { |
70b97a7f | 2574 | struct migration_req req; |
1da177e4 | 2575 | unsigned long flags; |
70b97a7f | 2576 | struct rq *rq; |
1da177e4 LT |
2577 | |
2578 | rq = task_rq_lock(p, &flags); | |
2579 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2580 | || unlikely(cpu_is_offline(dest_cpu))) | |
2581 | goto out; | |
2582 | ||
2583 | /* force the process onto the specified CPU */ | |
2584 | if (migrate_task(p, dest_cpu, &req)) { | |
2585 | /* Need to wait for migration thread (might exit: take ref). */ | |
2586 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2587 | |
1da177e4 LT |
2588 | get_task_struct(mt); |
2589 | task_rq_unlock(rq, &flags); | |
2590 | wake_up_process(mt); | |
2591 | put_task_struct(mt); | |
2592 | wait_for_completion(&req.done); | |
36c8b586 | 2593 | |
1da177e4 LT |
2594 | return; |
2595 | } | |
2596 | out: | |
2597 | task_rq_unlock(rq, &flags); | |
2598 | } | |
2599 | ||
2600 | /* | |
476d139c NP |
2601 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2602 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2603 | */ |
2604 | void sched_exec(void) | |
2605 | { | |
1da177e4 | 2606 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2607 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2608 | put_cpu(); |
476d139c NP |
2609 | if (new_cpu != this_cpu) |
2610 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2611 | } |
2612 | ||
2613 | /* | |
2614 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2615 | * Both runqueues must be locked. | |
2616 | */ | |
dd41f596 IM |
2617 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2618 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2619 | { |
2e1cb74a | 2620 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2621 | set_task_cpu(p, this_cpu); |
dd41f596 | 2622 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2623 | /* |
2624 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2625 | * to be always true for them. | |
2626 | */ | |
dd41f596 | 2627 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2628 | } |
2629 | ||
2630 | /* | |
2631 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2632 | */ | |
858119e1 | 2633 | static |
70b97a7f | 2634 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2635 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2636 | int *all_pinned) |
1da177e4 LT |
2637 | { |
2638 | /* | |
2639 | * We do not migrate tasks that are: | |
2640 | * 1) running (obviously), or | |
2641 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2642 | * 3) are cache-hot on their current CPU. | |
2643 | */ | |
cc367732 IM |
2644 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2645 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2646 | return 0; |
cc367732 | 2647 | } |
81026794 NP |
2648 | *all_pinned = 0; |
2649 | ||
cc367732 IM |
2650 | if (task_running(rq, p)) { |
2651 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2652 | return 0; |
cc367732 | 2653 | } |
1da177e4 | 2654 | |
da84d961 IM |
2655 | /* |
2656 | * Aggressive migration if: | |
2657 | * 1) task is cache cold, or | |
2658 | * 2) too many balance attempts have failed. | |
2659 | */ | |
2660 | ||
6bc1665b IM |
2661 | if (!task_hot(p, rq->clock, sd) || |
2662 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2663 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2664 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2665 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2666 | schedstat_inc(p, se.nr_forced_migrations); |
2667 | } | |
da84d961 IM |
2668 | #endif |
2669 | return 1; | |
2670 | } | |
2671 | ||
cc367732 IM |
2672 | if (task_hot(p, rq->clock, sd)) { |
2673 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2674 | return 0; |
cc367732 | 2675 | } |
1da177e4 LT |
2676 | return 1; |
2677 | } | |
2678 | ||
e1d1484f PW |
2679 | static unsigned long |
2680 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2681 | unsigned long max_load_move, struct sched_domain *sd, | |
2682 | enum cpu_idle_type idle, int *all_pinned, | |
2683 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2684 | { |
b82d9fdd | 2685 | int loops = 0, pulled = 0, pinned = 0, skip_for_load; |
dd41f596 IM |
2686 | struct task_struct *p; |
2687 | long rem_load_move = max_load_move; | |
1da177e4 | 2688 | |
e1d1484f | 2689 | if (max_load_move == 0) |
1da177e4 LT |
2690 | goto out; |
2691 | ||
81026794 NP |
2692 | pinned = 1; |
2693 | ||
1da177e4 | 2694 | /* |
dd41f596 | 2695 | * Start the load-balancing iterator: |
1da177e4 | 2696 | */ |
dd41f596 IM |
2697 | p = iterator->start(iterator->arg); |
2698 | next: | |
b82d9fdd | 2699 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2700 | goto out; |
50ddd969 | 2701 | /* |
b82d9fdd | 2702 | * To help distribute high priority tasks across CPUs we don't |
50ddd969 PW |
2703 | * skip a task if it will be the highest priority task (i.e. smallest |
2704 | * prio value) on its new queue regardless of its load weight | |
2705 | */ | |
dd41f596 IM |
2706 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2707 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2708 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2709 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2710 | p = iterator->next(iterator->arg); |
2711 | goto next; | |
1da177e4 LT |
2712 | } |
2713 | ||
dd41f596 | 2714 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2715 | pulled++; |
dd41f596 | 2716 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2717 | |
2dd73a4f | 2718 | /* |
b82d9fdd | 2719 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2720 | */ |
e1d1484f | 2721 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2722 | if (p->prio < *this_best_prio) |
2723 | *this_best_prio = p->prio; | |
dd41f596 IM |
2724 | p = iterator->next(iterator->arg); |
2725 | goto next; | |
1da177e4 LT |
2726 | } |
2727 | out: | |
2728 | /* | |
e1d1484f | 2729 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2730 | * so we can safely collect pull_task() stats here rather than |
2731 | * inside pull_task(). | |
2732 | */ | |
2733 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2734 | |
2735 | if (all_pinned) | |
2736 | *all_pinned = pinned; | |
e1d1484f PW |
2737 | |
2738 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2739 | } |
2740 | ||
dd41f596 | 2741 | /* |
43010659 PW |
2742 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2743 | * this_rq, as part of a balancing operation within domain "sd". | |
2744 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2745 | * |
2746 | * Called with both runqueues locked. | |
2747 | */ | |
2748 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2749 | unsigned long max_load_move, |
dd41f596 IM |
2750 | struct sched_domain *sd, enum cpu_idle_type idle, |
2751 | int *all_pinned) | |
2752 | { | |
5522d5d5 | 2753 | const struct sched_class *class = sched_class_highest; |
43010659 | 2754 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2755 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2756 | |
2757 | do { | |
43010659 PW |
2758 | total_load_moved += |
2759 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2760 | max_load_move - total_load_moved, |
a4ac01c3 | 2761 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2762 | class = class->next; |
43010659 | 2763 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2764 | |
43010659 PW |
2765 | return total_load_moved > 0; |
2766 | } | |
2767 | ||
e1d1484f PW |
2768 | static int |
2769 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2770 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2771 | struct rq_iterator *iterator) | |
2772 | { | |
2773 | struct task_struct *p = iterator->start(iterator->arg); | |
2774 | int pinned = 0; | |
2775 | ||
2776 | while (p) { | |
2777 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
2778 | pull_task(busiest, p, this_rq, this_cpu); | |
2779 | /* | |
2780 | * Right now, this is only the second place pull_task() | |
2781 | * is called, so we can safely collect pull_task() | |
2782 | * stats here rather than inside pull_task(). | |
2783 | */ | |
2784 | schedstat_inc(sd, lb_gained[idle]); | |
2785 | ||
2786 | return 1; | |
2787 | } | |
2788 | p = iterator->next(iterator->arg); | |
2789 | } | |
2790 | ||
2791 | return 0; | |
2792 | } | |
2793 | ||
43010659 PW |
2794 | /* |
2795 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2796 | * part of active balancing operations within "domain". | |
2797 | * Returns 1 if successful and 0 otherwise. | |
2798 | * | |
2799 | * Called with both runqueues locked. | |
2800 | */ | |
2801 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2802 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2803 | { | |
5522d5d5 | 2804 | const struct sched_class *class; |
43010659 PW |
2805 | |
2806 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 2807 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
2808 | return 1; |
2809 | ||
2810 | return 0; | |
dd41f596 IM |
2811 | } |
2812 | ||
1da177e4 LT |
2813 | /* |
2814 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2815 | * domain. It calculates and returns the amount of weighted load which |
2816 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2817 | */ |
2818 | static struct sched_group * | |
2819 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2820 | unsigned long *imbalance, enum cpu_idle_type idle, |
2821 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2822 | { |
2823 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2824 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2825 | unsigned long max_pull; |
2dd73a4f PW |
2826 | unsigned long busiest_load_per_task, busiest_nr_running; |
2827 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2828 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2829 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2830 | int power_savings_balance = 1; | |
2831 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2832 | unsigned long min_nr_running = ULONG_MAX; | |
2833 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2834 | #endif | |
1da177e4 LT |
2835 | |
2836 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2837 | busiest_load_per_task = busiest_nr_running = 0; |
2838 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2839 | if (idle == CPU_NOT_IDLE) |
7897986b | 2840 | load_idx = sd->busy_idx; |
d15bcfdb | 2841 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2842 | load_idx = sd->newidle_idx; |
2843 | else | |
2844 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2845 | |
2846 | do { | |
908a7c1b | 2847 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2848 | int local_group; |
2849 | int i; | |
908a7c1b | 2850 | int __group_imb = 0; |
783609c6 | 2851 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2852 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2853 | |
2854 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2855 | ||
783609c6 SS |
2856 | if (local_group) |
2857 | balance_cpu = first_cpu(group->cpumask); | |
2858 | ||
1da177e4 | 2859 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2860 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2861 | max_cpu_load = 0; |
2862 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2863 | |
2864 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2865 | struct rq *rq; |
2866 | ||
2867 | if (!cpu_isset(i, *cpus)) | |
2868 | continue; | |
2869 | ||
2870 | rq = cpu_rq(i); | |
2dd73a4f | 2871 | |
9439aab8 | 2872 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2873 | *sd_idle = 0; |
2874 | ||
1da177e4 | 2875 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2876 | if (local_group) { |
2877 | if (idle_cpu(i) && !first_idle_cpu) { | |
2878 | first_idle_cpu = 1; | |
2879 | balance_cpu = i; | |
2880 | } | |
2881 | ||
a2000572 | 2882 | load = target_load(i, load_idx); |
908a7c1b | 2883 | } else { |
a2000572 | 2884 | load = source_load(i, load_idx); |
908a7c1b KC |
2885 | if (load > max_cpu_load) |
2886 | max_cpu_load = load; | |
2887 | if (min_cpu_load > load) | |
2888 | min_cpu_load = load; | |
2889 | } | |
1da177e4 LT |
2890 | |
2891 | avg_load += load; | |
2dd73a4f | 2892 | sum_nr_running += rq->nr_running; |
dd41f596 | 2893 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2894 | } |
2895 | ||
783609c6 SS |
2896 | /* |
2897 | * First idle cpu or the first cpu(busiest) in this sched group | |
2898 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2899 | * domains. In the newly idle case, we will allow all the cpu's |
2900 | * to do the newly idle load balance. | |
783609c6 | 2901 | */ |
9439aab8 SS |
2902 | if (idle != CPU_NEWLY_IDLE && local_group && |
2903 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2904 | *balance = 0; |
2905 | goto ret; | |
2906 | } | |
2907 | ||
1da177e4 | 2908 | total_load += avg_load; |
5517d86b | 2909 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2910 | |
2911 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2912 | avg_load = sg_div_cpu_power(group, |
2913 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2914 | |
908a7c1b KC |
2915 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2916 | __group_imb = 1; | |
2917 | ||
5517d86b | 2918 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2919 | |
1da177e4 LT |
2920 | if (local_group) { |
2921 | this_load = avg_load; | |
2922 | this = group; | |
2dd73a4f PW |
2923 | this_nr_running = sum_nr_running; |
2924 | this_load_per_task = sum_weighted_load; | |
2925 | } else if (avg_load > max_load && | |
908a7c1b | 2926 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2927 | max_load = avg_load; |
2928 | busiest = group; | |
2dd73a4f PW |
2929 | busiest_nr_running = sum_nr_running; |
2930 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2931 | group_imb = __group_imb; |
1da177e4 | 2932 | } |
5c45bf27 SS |
2933 | |
2934 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2935 | /* | |
2936 | * Busy processors will not participate in power savings | |
2937 | * balance. | |
2938 | */ | |
dd41f596 IM |
2939 | if (idle == CPU_NOT_IDLE || |
2940 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2941 | goto group_next; | |
5c45bf27 SS |
2942 | |
2943 | /* | |
2944 | * If the local group is idle or completely loaded | |
2945 | * no need to do power savings balance at this domain | |
2946 | */ | |
2947 | if (local_group && (this_nr_running >= group_capacity || | |
2948 | !this_nr_running)) | |
2949 | power_savings_balance = 0; | |
2950 | ||
dd41f596 | 2951 | /* |
5c45bf27 SS |
2952 | * If a group is already running at full capacity or idle, |
2953 | * don't include that group in power savings calculations | |
dd41f596 IM |
2954 | */ |
2955 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2956 | || !sum_nr_running) |
dd41f596 | 2957 | goto group_next; |
5c45bf27 | 2958 | |
dd41f596 | 2959 | /* |
5c45bf27 | 2960 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2961 | * This is the group from where we need to pick up the load |
2962 | * for saving power | |
2963 | */ | |
2964 | if ((sum_nr_running < min_nr_running) || | |
2965 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2966 | first_cpu(group->cpumask) < |
2967 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2968 | group_min = group; |
2969 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2970 | min_load_per_task = sum_weighted_load / |
2971 | sum_nr_running; | |
dd41f596 | 2972 | } |
5c45bf27 | 2973 | |
dd41f596 | 2974 | /* |
5c45bf27 | 2975 | * Calculate the group which is almost near its |
dd41f596 IM |
2976 | * capacity but still has some space to pick up some load |
2977 | * from other group and save more power | |
2978 | */ | |
2979 | if (sum_nr_running <= group_capacity - 1) { | |
2980 | if (sum_nr_running > leader_nr_running || | |
2981 | (sum_nr_running == leader_nr_running && | |
2982 | first_cpu(group->cpumask) > | |
2983 | first_cpu(group_leader->cpumask))) { | |
2984 | group_leader = group; | |
2985 | leader_nr_running = sum_nr_running; | |
2986 | } | |
48f24c4d | 2987 | } |
5c45bf27 SS |
2988 | group_next: |
2989 | #endif | |
1da177e4 LT |
2990 | group = group->next; |
2991 | } while (group != sd->groups); | |
2992 | ||
2dd73a4f | 2993 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2994 | goto out_balanced; |
2995 | ||
2996 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2997 | ||
2998 | if (this_load >= avg_load || | |
2999 | 100*max_load <= sd->imbalance_pct*this_load) | |
3000 | goto out_balanced; | |
3001 | ||
2dd73a4f | 3002 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3003 | if (group_imb) |
3004 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3005 | ||
1da177e4 LT |
3006 | /* |
3007 | * We're trying to get all the cpus to the average_load, so we don't | |
3008 | * want to push ourselves above the average load, nor do we wish to | |
3009 | * reduce the max loaded cpu below the average load, as either of these | |
3010 | * actions would just result in more rebalancing later, and ping-pong | |
3011 | * tasks around. Thus we look for the minimum possible imbalance. | |
3012 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3013 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3014 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3015 | * appear as very large values with unsigned longs. |
3016 | */ | |
2dd73a4f PW |
3017 | if (max_load <= busiest_load_per_task) |
3018 | goto out_balanced; | |
3019 | ||
3020 | /* | |
3021 | * In the presence of smp nice balancing, certain scenarios can have | |
3022 | * max load less than avg load(as we skip the groups at or below | |
3023 | * its cpu_power, while calculating max_load..) | |
3024 | */ | |
3025 | if (max_load < avg_load) { | |
3026 | *imbalance = 0; | |
3027 | goto small_imbalance; | |
3028 | } | |
0c117f1b SS |
3029 | |
3030 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3031 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3032 | |
1da177e4 | 3033 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3034 | *imbalance = min(max_pull * busiest->__cpu_power, |
3035 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3036 | / SCHED_LOAD_SCALE; |
3037 | ||
2dd73a4f PW |
3038 | /* |
3039 | * if *imbalance is less than the average load per runnable task | |
3040 | * there is no gaurantee that any tasks will be moved so we'll have | |
3041 | * a think about bumping its value to force at least one task to be | |
3042 | * moved | |
3043 | */ | |
7fd0d2dd | 3044 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3045 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3046 | unsigned int imbn; |
3047 | ||
3048 | small_imbalance: | |
3049 | pwr_move = pwr_now = 0; | |
3050 | imbn = 2; | |
3051 | if (this_nr_running) { | |
3052 | this_load_per_task /= this_nr_running; | |
3053 | if (busiest_load_per_task > this_load_per_task) | |
3054 | imbn = 1; | |
3055 | } else | |
3056 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 3057 | |
dd41f596 IM |
3058 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
3059 | busiest_load_per_task * imbn) { | |
2dd73a4f | 3060 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3061 | return busiest; |
3062 | } | |
3063 | ||
3064 | /* | |
3065 | * OK, we don't have enough imbalance to justify moving tasks, | |
3066 | * however we may be able to increase total CPU power used by | |
3067 | * moving them. | |
3068 | */ | |
3069 | ||
5517d86b ED |
3070 | pwr_now += busiest->__cpu_power * |
3071 | min(busiest_load_per_task, max_load); | |
3072 | pwr_now += this->__cpu_power * | |
3073 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3074 | pwr_now /= SCHED_LOAD_SCALE; |
3075 | ||
3076 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3077 | tmp = sg_div_cpu_power(busiest, |
3078 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3079 | if (max_load > tmp) |
5517d86b | 3080 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3081 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3082 | |
3083 | /* Amount of load we'd add */ | |
5517d86b | 3084 | if (max_load * busiest->__cpu_power < |
33859f7f | 3085 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3086 | tmp = sg_div_cpu_power(this, |
3087 | max_load * busiest->__cpu_power); | |
1da177e4 | 3088 | else |
5517d86b ED |
3089 | tmp = sg_div_cpu_power(this, |
3090 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3091 | pwr_move += this->__cpu_power * | |
3092 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3093 | pwr_move /= SCHED_LOAD_SCALE; |
3094 | ||
3095 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3096 | if (pwr_move > pwr_now) |
3097 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3098 | } |
3099 | ||
1da177e4 LT |
3100 | return busiest; |
3101 | ||
3102 | out_balanced: | |
5c45bf27 | 3103 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3104 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3105 | goto ret; |
1da177e4 | 3106 | |
5c45bf27 SS |
3107 | if (this == group_leader && group_leader != group_min) { |
3108 | *imbalance = min_load_per_task; | |
3109 | return group_min; | |
3110 | } | |
5c45bf27 | 3111 | #endif |
783609c6 | 3112 | ret: |
1da177e4 LT |
3113 | *imbalance = 0; |
3114 | return NULL; | |
3115 | } | |
3116 | ||
3117 | /* | |
3118 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3119 | */ | |
70b97a7f | 3120 | static struct rq * |
d15bcfdb | 3121 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 3122 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 3123 | { |
70b97a7f | 3124 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3125 | unsigned long max_load = 0; |
1da177e4 LT |
3126 | int i; |
3127 | ||
3128 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 3129 | unsigned long wl; |
0a2966b4 CL |
3130 | |
3131 | if (!cpu_isset(i, *cpus)) | |
3132 | continue; | |
3133 | ||
48f24c4d | 3134 | rq = cpu_rq(i); |
dd41f596 | 3135 | wl = weighted_cpuload(i); |
2dd73a4f | 3136 | |
dd41f596 | 3137 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3138 | continue; |
1da177e4 | 3139 | |
dd41f596 IM |
3140 | if (wl > max_load) { |
3141 | max_load = wl; | |
48f24c4d | 3142 | busiest = rq; |
1da177e4 LT |
3143 | } |
3144 | } | |
3145 | ||
3146 | return busiest; | |
3147 | } | |
3148 | ||
77391d71 NP |
3149 | /* |
3150 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3151 | * so long as it is large enough. | |
3152 | */ | |
3153 | #define MAX_PINNED_INTERVAL 512 | |
3154 | ||
1da177e4 LT |
3155 | /* |
3156 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3157 | * tasks if there is an imbalance. | |
1da177e4 | 3158 | */ |
70b97a7f | 3159 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3160 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 3161 | int *balance) |
1da177e4 | 3162 | { |
43010659 | 3163 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3164 | struct sched_group *group; |
1da177e4 | 3165 | unsigned long imbalance; |
70b97a7f | 3166 | struct rq *busiest; |
0a2966b4 | 3167 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 3168 | unsigned long flags; |
5969fe06 | 3169 | |
89c4710e SS |
3170 | /* |
3171 | * When power savings policy is enabled for the parent domain, idle | |
3172 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3173 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3174 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3175 | */ |
d15bcfdb | 3176 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3177 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3178 | sd_idle = 1; |
1da177e4 | 3179 | |
2d72376b | 3180 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3181 | |
0a2966b4 CL |
3182 | redo: |
3183 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
3184 | &cpus, balance); |
3185 | ||
06066714 | 3186 | if (*balance == 0) |
783609c6 | 3187 | goto out_balanced; |
783609c6 | 3188 | |
1da177e4 LT |
3189 | if (!group) { |
3190 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3191 | goto out_balanced; | |
3192 | } | |
3193 | ||
0a2966b4 | 3194 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
3195 | if (!busiest) { |
3196 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3197 | goto out_balanced; | |
3198 | } | |
3199 | ||
db935dbd | 3200 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3201 | |
3202 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3203 | ||
43010659 | 3204 | ld_moved = 0; |
1da177e4 LT |
3205 | if (busiest->nr_running > 1) { |
3206 | /* | |
3207 | * Attempt to move tasks. If find_busiest_group has found | |
3208 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3209 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3210 | * correctly treated as an imbalance. |
3211 | */ | |
fe2eea3f | 3212 | local_irq_save(flags); |
e17224bf | 3213 | double_rq_lock(this_rq, busiest); |
43010659 | 3214 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3215 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3216 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3217 | local_irq_restore(flags); |
81026794 | 3218 | |
46cb4b7c SS |
3219 | /* |
3220 | * some other cpu did the load balance for us. | |
3221 | */ | |
43010659 | 3222 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3223 | resched_cpu(this_cpu); |
3224 | ||
81026794 | 3225 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
3226 | if (unlikely(all_pinned)) { |
3227 | cpu_clear(cpu_of(busiest), cpus); | |
3228 | if (!cpus_empty(cpus)) | |
3229 | goto redo; | |
81026794 | 3230 | goto out_balanced; |
0a2966b4 | 3231 | } |
1da177e4 | 3232 | } |
81026794 | 3233 | |
43010659 | 3234 | if (!ld_moved) { |
1da177e4 LT |
3235 | schedstat_inc(sd, lb_failed[idle]); |
3236 | sd->nr_balance_failed++; | |
3237 | ||
3238 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3239 | |
fe2eea3f | 3240 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3241 | |
3242 | /* don't kick the migration_thread, if the curr | |
3243 | * task on busiest cpu can't be moved to this_cpu | |
3244 | */ | |
3245 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3246 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3247 | all_pinned = 1; |
3248 | goto out_one_pinned; | |
3249 | } | |
3250 | ||
1da177e4 LT |
3251 | if (!busiest->active_balance) { |
3252 | busiest->active_balance = 1; | |
3253 | busiest->push_cpu = this_cpu; | |
81026794 | 3254 | active_balance = 1; |
1da177e4 | 3255 | } |
fe2eea3f | 3256 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3257 | if (active_balance) |
1da177e4 LT |
3258 | wake_up_process(busiest->migration_thread); |
3259 | ||
3260 | /* | |
3261 | * We've kicked active balancing, reset the failure | |
3262 | * counter. | |
3263 | */ | |
39507451 | 3264 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3265 | } |
81026794 | 3266 | } else |
1da177e4 LT |
3267 | sd->nr_balance_failed = 0; |
3268 | ||
81026794 | 3269 | if (likely(!active_balance)) { |
1da177e4 LT |
3270 | /* We were unbalanced, so reset the balancing interval */ |
3271 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3272 | } else { |
3273 | /* | |
3274 | * If we've begun active balancing, start to back off. This | |
3275 | * case may not be covered by the all_pinned logic if there | |
3276 | * is only 1 task on the busy runqueue (because we don't call | |
3277 | * move_tasks). | |
3278 | */ | |
3279 | if (sd->balance_interval < sd->max_interval) | |
3280 | sd->balance_interval *= 2; | |
1da177e4 LT |
3281 | } |
3282 | ||
43010659 | 3283 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3284 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3285 | return -1; |
43010659 | 3286 | return ld_moved; |
1da177e4 LT |
3287 | |
3288 | out_balanced: | |
1da177e4 LT |
3289 | schedstat_inc(sd, lb_balanced[idle]); |
3290 | ||
16cfb1c0 | 3291 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3292 | |
3293 | out_one_pinned: | |
1da177e4 | 3294 | /* tune up the balancing interval */ |
77391d71 NP |
3295 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3296 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3297 | sd->balance_interval *= 2; |
3298 | ||
48f24c4d | 3299 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3300 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3301 | return -1; |
1da177e4 LT |
3302 | return 0; |
3303 | } | |
3304 | ||
3305 | /* | |
3306 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3307 | * tasks if there is an imbalance. | |
3308 | * | |
d15bcfdb | 3309 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3310 | * this_rq is locked. |
3311 | */ | |
48f24c4d | 3312 | static int |
70b97a7f | 3313 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
3314 | { |
3315 | struct sched_group *group; | |
70b97a7f | 3316 | struct rq *busiest = NULL; |
1da177e4 | 3317 | unsigned long imbalance; |
43010659 | 3318 | int ld_moved = 0; |
5969fe06 | 3319 | int sd_idle = 0; |
969bb4e4 | 3320 | int all_pinned = 0; |
0a2966b4 | 3321 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 3322 | |
89c4710e SS |
3323 | /* |
3324 | * When power savings policy is enabled for the parent domain, idle | |
3325 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3326 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3327 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3328 | */ |
3329 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3330 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3331 | sd_idle = 1; |
1da177e4 | 3332 | |
2d72376b | 3333 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3334 | redo: |
d15bcfdb | 3335 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 3336 | &sd_idle, &cpus, NULL); |
1da177e4 | 3337 | if (!group) { |
d15bcfdb | 3338 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3339 | goto out_balanced; |
1da177e4 LT |
3340 | } |
3341 | ||
d15bcfdb | 3342 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 3343 | &cpus); |
db935dbd | 3344 | if (!busiest) { |
d15bcfdb | 3345 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3346 | goto out_balanced; |
1da177e4 LT |
3347 | } |
3348 | ||
db935dbd NP |
3349 | BUG_ON(busiest == this_rq); |
3350 | ||
d15bcfdb | 3351 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3352 | |
43010659 | 3353 | ld_moved = 0; |
d6d5cfaf NP |
3354 | if (busiest->nr_running > 1) { |
3355 | /* Attempt to move tasks */ | |
3356 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3357 | /* this_rq->clock is already updated */ |
3358 | update_rq_clock(busiest); | |
43010659 | 3359 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3360 | imbalance, sd, CPU_NEWLY_IDLE, |
3361 | &all_pinned); | |
d6d5cfaf | 3362 | spin_unlock(&busiest->lock); |
0a2966b4 | 3363 | |
969bb4e4 | 3364 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
3365 | cpu_clear(cpu_of(busiest), cpus); |
3366 | if (!cpus_empty(cpus)) | |
3367 | goto redo; | |
3368 | } | |
d6d5cfaf NP |
3369 | } |
3370 | ||
43010659 | 3371 | if (!ld_moved) { |
d15bcfdb | 3372 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3373 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3374 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3375 | return -1; |
3376 | } else | |
16cfb1c0 | 3377 | sd->nr_balance_failed = 0; |
1da177e4 | 3378 | |
43010659 | 3379 | return ld_moved; |
16cfb1c0 NP |
3380 | |
3381 | out_balanced: | |
d15bcfdb | 3382 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3383 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3384 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3385 | return -1; |
16cfb1c0 | 3386 | sd->nr_balance_failed = 0; |
48f24c4d | 3387 | |
16cfb1c0 | 3388 | return 0; |
1da177e4 LT |
3389 | } |
3390 | ||
3391 | /* | |
3392 | * idle_balance is called by schedule() if this_cpu is about to become | |
3393 | * idle. Attempts to pull tasks from other CPUs. | |
3394 | */ | |
70b97a7f | 3395 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3396 | { |
3397 | struct sched_domain *sd; | |
dd41f596 IM |
3398 | int pulled_task = -1; |
3399 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
3400 | |
3401 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3402 | unsigned long interval; |
3403 | ||
3404 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3405 | continue; | |
3406 | ||
3407 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3408 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 3409 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
3410 | this_rq, sd); |
3411 | ||
3412 | interval = msecs_to_jiffies(sd->balance_interval); | |
3413 | if (time_after(next_balance, sd->last_balance + interval)) | |
3414 | next_balance = sd->last_balance + interval; | |
3415 | if (pulled_task) | |
3416 | break; | |
1da177e4 | 3417 | } |
dd41f596 | 3418 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3419 | /* |
3420 | * We are going idle. next_balance may be set based on | |
3421 | * a busy processor. So reset next_balance. | |
3422 | */ | |
3423 | this_rq->next_balance = next_balance; | |
dd41f596 | 3424 | } |
1da177e4 LT |
3425 | } |
3426 | ||
3427 | /* | |
3428 | * active_load_balance is run by migration threads. It pushes running tasks | |
3429 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3430 | * running on each physical CPU where possible, and avoids physical / | |
3431 | * logical imbalances. | |
3432 | * | |
3433 | * Called with busiest_rq locked. | |
3434 | */ | |
70b97a7f | 3435 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3436 | { |
39507451 | 3437 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3438 | struct sched_domain *sd; |
3439 | struct rq *target_rq; | |
39507451 | 3440 | |
48f24c4d | 3441 | /* Is there any task to move? */ |
39507451 | 3442 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3443 | return; |
3444 | ||
3445 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3446 | |
3447 | /* | |
39507451 | 3448 | * This condition is "impossible", if it occurs |
41a2d6cf | 3449 | * we need to fix it. Originally reported by |
39507451 | 3450 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3451 | */ |
39507451 | 3452 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3453 | |
39507451 NP |
3454 | /* move a task from busiest_rq to target_rq */ |
3455 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3456 | update_rq_clock(busiest_rq); |
3457 | update_rq_clock(target_rq); | |
39507451 NP |
3458 | |
3459 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3460 | for_each_domain(target_cpu, sd) { |
39507451 | 3461 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3462 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3463 | break; |
c96d145e | 3464 | } |
39507451 | 3465 | |
48f24c4d | 3466 | if (likely(sd)) { |
2d72376b | 3467 | schedstat_inc(sd, alb_count); |
39507451 | 3468 | |
43010659 PW |
3469 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3470 | sd, CPU_IDLE)) | |
48f24c4d IM |
3471 | schedstat_inc(sd, alb_pushed); |
3472 | else | |
3473 | schedstat_inc(sd, alb_failed); | |
3474 | } | |
39507451 | 3475 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3476 | } |
3477 | ||
46cb4b7c SS |
3478 | #ifdef CONFIG_NO_HZ |
3479 | static struct { | |
3480 | atomic_t load_balancer; | |
41a2d6cf | 3481 | cpumask_t cpu_mask; |
46cb4b7c SS |
3482 | } nohz ____cacheline_aligned = { |
3483 | .load_balancer = ATOMIC_INIT(-1), | |
3484 | .cpu_mask = CPU_MASK_NONE, | |
3485 | }; | |
3486 | ||
7835b98b | 3487 | /* |
46cb4b7c SS |
3488 | * This routine will try to nominate the ilb (idle load balancing) |
3489 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3490 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3491 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3492 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3493 | * arrives... | |
3494 | * | |
3495 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3496 | * for idle load balancing. ilb owner will still be part of | |
3497 | * nohz.cpu_mask.. | |
7835b98b | 3498 | * |
46cb4b7c SS |
3499 | * While stopping the tick, this cpu will become the ilb owner if there |
3500 | * is no other owner. And will be the owner till that cpu becomes busy | |
3501 | * or if all cpus in the system stop their ticks at which point | |
3502 | * there is no need for ilb owner. | |
3503 | * | |
3504 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3505 | * next busy scheduler_tick() | |
3506 | */ | |
3507 | int select_nohz_load_balancer(int stop_tick) | |
3508 | { | |
3509 | int cpu = smp_processor_id(); | |
3510 | ||
3511 | if (stop_tick) { | |
3512 | cpu_set(cpu, nohz.cpu_mask); | |
3513 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3514 | ||
3515 | /* | |
3516 | * If we are going offline and still the leader, give up! | |
3517 | */ | |
3518 | if (cpu_is_offline(cpu) && | |
3519 | atomic_read(&nohz.load_balancer) == cpu) { | |
3520 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3521 | BUG(); | |
3522 | return 0; | |
3523 | } | |
3524 | ||
3525 | /* time for ilb owner also to sleep */ | |
3526 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3527 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3528 | atomic_set(&nohz.load_balancer, -1); | |
3529 | return 0; | |
3530 | } | |
3531 | ||
3532 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3533 | /* make me the ilb owner */ | |
3534 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3535 | return 1; | |
3536 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3537 | return 1; | |
3538 | } else { | |
3539 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3540 | return 0; | |
3541 | ||
3542 | cpu_clear(cpu, nohz.cpu_mask); | |
3543 | ||
3544 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3545 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3546 | BUG(); | |
3547 | } | |
3548 | return 0; | |
3549 | } | |
3550 | #endif | |
3551 | ||
3552 | static DEFINE_SPINLOCK(balancing); | |
3553 | ||
3554 | /* | |
7835b98b CL |
3555 | * It checks each scheduling domain to see if it is due to be balanced, |
3556 | * and initiates a balancing operation if so. | |
3557 | * | |
3558 | * Balancing parameters are set up in arch_init_sched_domains. | |
3559 | */ | |
a9957449 | 3560 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3561 | { |
46cb4b7c SS |
3562 | int balance = 1; |
3563 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3564 | unsigned long interval; |
3565 | struct sched_domain *sd; | |
46cb4b7c | 3566 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3567 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3568 | int update_next_balance = 0; |
1da177e4 | 3569 | |
46cb4b7c | 3570 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3571 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3572 | continue; | |
3573 | ||
3574 | interval = sd->balance_interval; | |
d15bcfdb | 3575 | if (idle != CPU_IDLE) |
1da177e4 LT |
3576 | interval *= sd->busy_factor; |
3577 | ||
3578 | /* scale ms to jiffies */ | |
3579 | interval = msecs_to_jiffies(interval); | |
3580 | if (unlikely(!interval)) | |
3581 | interval = 1; | |
dd41f596 IM |
3582 | if (interval > HZ*NR_CPUS/10) |
3583 | interval = HZ*NR_CPUS/10; | |
3584 | ||
1da177e4 | 3585 | |
08c183f3 CL |
3586 | if (sd->flags & SD_SERIALIZE) { |
3587 | if (!spin_trylock(&balancing)) | |
3588 | goto out; | |
3589 | } | |
3590 | ||
c9819f45 | 3591 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3592 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3593 | /* |
3594 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3595 | * longer idle, or one of our SMT siblings is |
3596 | * not idle. | |
3597 | */ | |
d15bcfdb | 3598 | idle = CPU_NOT_IDLE; |
1da177e4 | 3599 | } |
1bd77f2d | 3600 | sd->last_balance = jiffies; |
1da177e4 | 3601 | } |
08c183f3 CL |
3602 | if (sd->flags & SD_SERIALIZE) |
3603 | spin_unlock(&balancing); | |
3604 | out: | |
f549da84 | 3605 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3606 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3607 | update_next_balance = 1; |
3608 | } | |
783609c6 SS |
3609 | |
3610 | /* | |
3611 | * Stop the load balance at this level. There is another | |
3612 | * CPU in our sched group which is doing load balancing more | |
3613 | * actively. | |
3614 | */ | |
3615 | if (!balance) | |
3616 | break; | |
1da177e4 | 3617 | } |
f549da84 SS |
3618 | |
3619 | /* | |
3620 | * next_balance will be updated only when there is a need. | |
3621 | * When the cpu is attached to null domain for ex, it will not be | |
3622 | * updated. | |
3623 | */ | |
3624 | if (likely(update_next_balance)) | |
3625 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3626 | } |
3627 | ||
3628 | /* | |
3629 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3630 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3631 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3632 | */ | |
3633 | static void run_rebalance_domains(struct softirq_action *h) | |
3634 | { | |
dd41f596 IM |
3635 | int this_cpu = smp_processor_id(); |
3636 | struct rq *this_rq = cpu_rq(this_cpu); | |
3637 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3638 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3639 | |
dd41f596 | 3640 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3641 | |
3642 | #ifdef CONFIG_NO_HZ | |
3643 | /* | |
3644 | * If this cpu is the owner for idle load balancing, then do the | |
3645 | * balancing on behalf of the other idle cpus whose ticks are | |
3646 | * stopped. | |
3647 | */ | |
dd41f596 IM |
3648 | if (this_rq->idle_at_tick && |
3649 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3650 | cpumask_t cpus = nohz.cpu_mask; |
3651 | struct rq *rq; | |
3652 | int balance_cpu; | |
3653 | ||
dd41f596 | 3654 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3655 | for_each_cpu_mask(balance_cpu, cpus) { |
3656 | /* | |
3657 | * If this cpu gets work to do, stop the load balancing | |
3658 | * work being done for other cpus. Next load | |
3659 | * balancing owner will pick it up. | |
3660 | */ | |
3661 | if (need_resched()) | |
3662 | break; | |
3663 | ||
de0cf899 | 3664 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3665 | |
3666 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3667 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3668 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3669 | } |
3670 | } | |
3671 | #endif | |
3672 | } | |
3673 | ||
3674 | /* | |
3675 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3676 | * | |
3677 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3678 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3679 | * if the whole system is idle. | |
3680 | */ | |
dd41f596 | 3681 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3682 | { |
46cb4b7c SS |
3683 | #ifdef CONFIG_NO_HZ |
3684 | /* | |
3685 | * If we were in the nohz mode recently and busy at the current | |
3686 | * scheduler tick, then check if we need to nominate new idle | |
3687 | * load balancer. | |
3688 | */ | |
3689 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3690 | rq->in_nohz_recently = 0; | |
3691 | ||
3692 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3693 | cpu_clear(cpu, nohz.cpu_mask); | |
3694 | atomic_set(&nohz.load_balancer, -1); | |
3695 | } | |
3696 | ||
3697 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3698 | /* | |
3699 | * simple selection for now: Nominate the | |
3700 | * first cpu in the nohz list to be the next | |
3701 | * ilb owner. | |
3702 | * | |
3703 | * TBD: Traverse the sched domains and nominate | |
3704 | * the nearest cpu in the nohz.cpu_mask. | |
3705 | */ | |
3706 | int ilb = first_cpu(nohz.cpu_mask); | |
3707 | ||
434d53b0 | 3708 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
3709 | resched_cpu(ilb); |
3710 | } | |
3711 | } | |
3712 | ||
3713 | /* | |
3714 | * If this cpu is idle and doing idle load balancing for all the | |
3715 | * cpus with ticks stopped, is it time for that to stop? | |
3716 | */ | |
3717 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3718 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3719 | resched_cpu(cpu); | |
3720 | return; | |
3721 | } | |
3722 | ||
3723 | /* | |
3724 | * If this cpu is idle and the idle load balancing is done by | |
3725 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3726 | */ | |
3727 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3728 | cpu_isset(cpu, nohz.cpu_mask)) | |
3729 | return; | |
3730 | #endif | |
3731 | if (time_after_eq(jiffies, rq->next_balance)) | |
3732 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3733 | } |
dd41f596 IM |
3734 | |
3735 | #else /* CONFIG_SMP */ | |
3736 | ||
1da177e4 LT |
3737 | /* |
3738 | * on UP we do not need to balance between CPUs: | |
3739 | */ | |
70b97a7f | 3740 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3741 | { |
3742 | } | |
dd41f596 | 3743 | |
1da177e4 LT |
3744 | #endif |
3745 | ||
1da177e4 LT |
3746 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3747 | ||
3748 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3749 | ||
3750 | /* | |
41b86e9c IM |
3751 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3752 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3753 | */ |
41b86e9c | 3754 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3755 | { |
1da177e4 | 3756 | unsigned long flags; |
41b86e9c IM |
3757 | u64 ns, delta_exec; |
3758 | struct rq *rq; | |
48f24c4d | 3759 | |
41b86e9c IM |
3760 | rq = task_rq_lock(p, &flags); |
3761 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 3762 | if (task_current(rq, p)) { |
a8e504d2 IM |
3763 | update_rq_clock(rq); |
3764 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3765 | if ((s64)delta_exec > 0) |
3766 | ns += delta_exec; | |
3767 | } | |
3768 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3769 | |
1da177e4 LT |
3770 | return ns; |
3771 | } | |
3772 | ||
1da177e4 LT |
3773 | /* |
3774 | * Account user cpu time to a process. | |
3775 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3776 | * @cputime: the cpu time spent in user space since the last update |
3777 | */ | |
3778 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3779 | { | |
3780 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3781 | cputime64_t tmp; | |
3782 | ||
3783 | p->utime = cputime_add(p->utime, cputime); | |
3784 | ||
3785 | /* Add user time to cpustat. */ | |
3786 | tmp = cputime_to_cputime64(cputime); | |
3787 | if (TASK_NICE(p) > 0) | |
3788 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3789 | else | |
3790 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3791 | } | |
3792 | ||
94886b84 LV |
3793 | /* |
3794 | * Account guest cpu time to a process. | |
3795 | * @p: the process that the cpu time gets accounted to | |
3796 | * @cputime: the cpu time spent in virtual machine since the last update | |
3797 | */ | |
f7402e03 | 3798 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
3799 | { |
3800 | cputime64_t tmp; | |
3801 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3802 | ||
3803 | tmp = cputime_to_cputime64(cputime); | |
3804 | ||
3805 | p->utime = cputime_add(p->utime, cputime); | |
3806 | p->gtime = cputime_add(p->gtime, cputime); | |
3807 | ||
3808 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3809 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3810 | } | |
3811 | ||
c66f08be MN |
3812 | /* |
3813 | * Account scaled user cpu time to a process. | |
3814 | * @p: the process that the cpu time gets accounted to | |
3815 | * @cputime: the cpu time spent in user space since the last update | |
3816 | */ | |
3817 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3818 | { | |
3819 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3820 | } | |
3821 | ||
1da177e4 LT |
3822 | /* |
3823 | * Account system cpu time to a process. | |
3824 | * @p: the process that the cpu time gets accounted to | |
3825 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3826 | * @cputime: the cpu time spent in kernel space since the last update | |
3827 | */ | |
3828 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3829 | cputime_t cputime) | |
3830 | { | |
3831 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3832 | struct rq *rq = this_rq(); |
1da177e4 LT |
3833 | cputime64_t tmp; |
3834 | ||
9778385d CB |
3835 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) |
3836 | return account_guest_time(p, cputime); | |
94886b84 | 3837 | |
1da177e4 LT |
3838 | p->stime = cputime_add(p->stime, cputime); |
3839 | ||
3840 | /* Add system time to cpustat. */ | |
3841 | tmp = cputime_to_cputime64(cputime); | |
3842 | if (hardirq_count() - hardirq_offset) | |
3843 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3844 | else if (softirq_count()) | |
3845 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 3846 | else if (p != rq->idle) |
1da177e4 | 3847 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 3848 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
3849 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3850 | else | |
3851 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3852 | /* Account for system time used */ | |
3853 | acct_update_integrals(p); | |
1da177e4 LT |
3854 | } |
3855 | ||
c66f08be MN |
3856 | /* |
3857 | * Account scaled system cpu time to a process. | |
3858 | * @p: the process that the cpu time gets accounted to | |
3859 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3860 | * @cputime: the cpu time spent in kernel space since the last update | |
3861 | */ | |
3862 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3863 | { | |
3864 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3865 | } | |
3866 | ||
1da177e4 LT |
3867 | /* |
3868 | * Account for involuntary wait time. | |
3869 | * @p: the process from which the cpu time has been stolen | |
3870 | * @steal: the cpu time spent in involuntary wait | |
3871 | */ | |
3872 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3873 | { | |
3874 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3875 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3876 | struct rq *rq = this_rq(); |
1da177e4 LT |
3877 | |
3878 | if (p == rq->idle) { | |
3879 | p->stime = cputime_add(p->stime, steal); | |
3880 | if (atomic_read(&rq->nr_iowait) > 0) | |
3881 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3882 | else | |
3883 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 3884 | } else |
1da177e4 LT |
3885 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
3886 | } | |
3887 | ||
7835b98b CL |
3888 | /* |
3889 | * This function gets called by the timer code, with HZ frequency. | |
3890 | * We call it with interrupts disabled. | |
3891 | * | |
3892 | * It also gets called by the fork code, when changing the parent's | |
3893 | * timeslices. | |
3894 | */ | |
3895 | void scheduler_tick(void) | |
3896 | { | |
7835b98b CL |
3897 | int cpu = smp_processor_id(); |
3898 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3899 | struct task_struct *curr = rq->curr; |
529c7726 | 3900 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3901 | |
3902 | spin_lock(&rq->lock); | |
546fe3c9 | 3903 | __update_rq_clock(rq); |
529c7726 IM |
3904 | /* |
3905 | * Let rq->clock advance by at least TICK_NSEC: | |
3906 | */ | |
cc203d24 | 3907 | if (unlikely(rq->clock < next_tick)) { |
529c7726 | 3908 | rq->clock = next_tick; |
cc203d24 GC |
3909 | rq->clock_underflows++; |
3910 | } | |
529c7726 | 3911 | rq->tick_timestamp = rq->clock; |
15934a37 | 3912 | update_last_tick_seen(rq); |
f1a438d8 | 3913 | update_cpu_load(rq); |
fa85ae24 | 3914 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 3915 | spin_unlock(&rq->lock); |
7835b98b | 3916 | |
e418e1c2 | 3917 | #ifdef CONFIG_SMP |
dd41f596 IM |
3918 | rq->idle_at_tick = idle_cpu(cpu); |
3919 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3920 | #endif |
1da177e4 LT |
3921 | } |
3922 | ||
1da177e4 LT |
3923 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3924 | ||
43627582 | 3925 | void __kprobes add_preempt_count(int val) |
1da177e4 LT |
3926 | { |
3927 | /* | |
3928 | * Underflow? | |
3929 | */ | |
9a11b49a IM |
3930 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3931 | return; | |
1da177e4 LT |
3932 | preempt_count() += val; |
3933 | /* | |
3934 | * Spinlock count overflowing soon? | |
3935 | */ | |
33859f7f MOS |
3936 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3937 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3938 | } |
3939 | EXPORT_SYMBOL(add_preempt_count); | |
3940 | ||
43627582 | 3941 | void __kprobes sub_preempt_count(int val) |
1da177e4 LT |
3942 | { |
3943 | /* | |
3944 | * Underflow? | |
3945 | */ | |
9a11b49a IM |
3946 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3947 | return; | |
1da177e4 LT |
3948 | /* |
3949 | * Is the spinlock portion underflowing? | |
3950 | */ | |
9a11b49a IM |
3951 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3952 | !(preempt_count() & PREEMPT_MASK))) | |
3953 | return; | |
3954 | ||
1da177e4 LT |
3955 | preempt_count() -= val; |
3956 | } | |
3957 | EXPORT_SYMBOL(sub_preempt_count); | |
3958 | ||
3959 | #endif | |
3960 | ||
3961 | /* | |
dd41f596 | 3962 | * Print scheduling while atomic bug: |
1da177e4 | 3963 | */ |
dd41f596 | 3964 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3965 | { |
838225b4 SS |
3966 | struct pt_regs *regs = get_irq_regs(); |
3967 | ||
3968 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
3969 | prev->comm, prev->pid, preempt_count()); | |
3970 | ||
dd41f596 IM |
3971 | debug_show_held_locks(prev); |
3972 | if (irqs_disabled()) | |
3973 | print_irqtrace_events(prev); | |
838225b4 SS |
3974 | |
3975 | if (regs) | |
3976 | show_regs(regs); | |
3977 | else | |
3978 | dump_stack(); | |
dd41f596 | 3979 | } |
1da177e4 | 3980 | |
dd41f596 IM |
3981 | /* |
3982 | * Various schedule()-time debugging checks and statistics: | |
3983 | */ | |
3984 | static inline void schedule_debug(struct task_struct *prev) | |
3985 | { | |
1da177e4 | 3986 | /* |
41a2d6cf | 3987 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3988 | * schedule() atomically, we ignore that path for now. |
3989 | * Otherwise, whine if we are scheduling when we should not be. | |
3990 | */ | |
dd41f596 IM |
3991 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3992 | __schedule_bug(prev); | |
3993 | ||
1da177e4 LT |
3994 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3995 | ||
2d72376b | 3996 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3997 | #ifdef CONFIG_SCHEDSTATS |
3998 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3999 | schedstat_inc(this_rq(), bkl_count); |
4000 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4001 | } |
4002 | #endif | |
dd41f596 IM |
4003 | } |
4004 | ||
4005 | /* | |
4006 | * Pick up the highest-prio task: | |
4007 | */ | |
4008 | static inline struct task_struct * | |
ff95f3df | 4009 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4010 | { |
5522d5d5 | 4011 | const struct sched_class *class; |
dd41f596 | 4012 | struct task_struct *p; |
1da177e4 LT |
4013 | |
4014 | /* | |
dd41f596 IM |
4015 | * Optimization: we know that if all tasks are in |
4016 | * the fair class we can call that function directly: | |
1da177e4 | 4017 | */ |
dd41f596 | 4018 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4019 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4020 | if (likely(p)) |
4021 | return p; | |
1da177e4 LT |
4022 | } |
4023 | ||
dd41f596 IM |
4024 | class = sched_class_highest; |
4025 | for ( ; ; ) { | |
fb8d4724 | 4026 | p = class->pick_next_task(rq); |
dd41f596 IM |
4027 | if (p) |
4028 | return p; | |
4029 | /* | |
4030 | * Will never be NULL as the idle class always | |
4031 | * returns a non-NULL p: | |
4032 | */ | |
4033 | class = class->next; | |
4034 | } | |
4035 | } | |
1da177e4 | 4036 | |
dd41f596 IM |
4037 | /* |
4038 | * schedule() is the main scheduler function. | |
4039 | */ | |
4040 | asmlinkage void __sched schedule(void) | |
4041 | { | |
4042 | struct task_struct *prev, *next; | |
67ca7bde | 4043 | unsigned long *switch_count; |
dd41f596 | 4044 | struct rq *rq; |
dd41f596 IM |
4045 | int cpu; |
4046 | ||
4047 | need_resched: | |
4048 | preempt_disable(); | |
4049 | cpu = smp_processor_id(); | |
4050 | rq = cpu_rq(cpu); | |
4051 | rcu_qsctr_inc(cpu); | |
4052 | prev = rq->curr; | |
4053 | switch_count = &prev->nivcsw; | |
4054 | ||
4055 | release_kernel_lock(prev); | |
4056 | need_resched_nonpreemptible: | |
4057 | ||
4058 | schedule_debug(prev); | |
1da177e4 | 4059 | |
8f4d37ec PZ |
4060 | hrtick_clear(rq); |
4061 | ||
1e819950 IM |
4062 | /* |
4063 | * Do the rq-clock update outside the rq lock: | |
4064 | */ | |
4065 | local_irq_disable(); | |
c1b3da3e | 4066 | __update_rq_clock(rq); |
1e819950 IM |
4067 | spin_lock(&rq->lock); |
4068 | clear_tsk_need_resched(prev); | |
1da177e4 | 4069 | |
1da177e4 | 4070 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 4071 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
23e3c3cd | 4072 | signal_pending(prev))) { |
1da177e4 | 4073 | prev->state = TASK_RUNNING; |
dd41f596 | 4074 | } else { |
2e1cb74a | 4075 | deactivate_task(rq, prev, 1); |
1da177e4 | 4076 | } |
dd41f596 | 4077 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4078 | } |
4079 | ||
9a897c5a SR |
4080 | #ifdef CONFIG_SMP |
4081 | if (prev->sched_class->pre_schedule) | |
4082 | prev->sched_class->pre_schedule(rq, prev); | |
4083 | #endif | |
f65eda4f | 4084 | |
dd41f596 | 4085 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4086 | idle_balance(cpu, rq); |
1da177e4 | 4087 | |
31ee529c | 4088 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4089 | next = pick_next_task(rq, prev); |
1da177e4 LT |
4090 | |
4091 | sched_info_switch(prev, next); | |
dd41f596 | 4092 | |
1da177e4 | 4093 | if (likely(prev != next)) { |
1da177e4 LT |
4094 | rq->nr_switches++; |
4095 | rq->curr = next; | |
4096 | ++*switch_count; | |
4097 | ||
dd41f596 | 4098 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4099 | /* |
4100 | * the context switch might have flipped the stack from under | |
4101 | * us, hence refresh the local variables. | |
4102 | */ | |
4103 | cpu = smp_processor_id(); | |
4104 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4105 | } else |
4106 | spin_unlock_irq(&rq->lock); | |
4107 | ||
8f4d37ec PZ |
4108 | hrtick_set(rq); |
4109 | ||
4110 | if (unlikely(reacquire_kernel_lock(current) < 0)) | |
1da177e4 | 4111 | goto need_resched_nonpreemptible; |
8f4d37ec | 4112 | |
1da177e4 LT |
4113 | preempt_enable_no_resched(); |
4114 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4115 | goto need_resched; | |
4116 | } | |
1da177e4 LT |
4117 | EXPORT_SYMBOL(schedule); |
4118 | ||
4119 | #ifdef CONFIG_PREEMPT | |
4120 | /* | |
2ed6e34f | 4121 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4122 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4123 | * occur there and call schedule directly. |
4124 | */ | |
4125 | asmlinkage void __sched preempt_schedule(void) | |
4126 | { | |
4127 | struct thread_info *ti = current_thread_info(); | |
1da177e4 LT |
4128 | struct task_struct *task = current; |
4129 | int saved_lock_depth; | |
6478d880 | 4130 | |
1da177e4 LT |
4131 | /* |
4132 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4133 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4134 | */ |
beed33a8 | 4135 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4136 | return; |
4137 | ||
3a5c359a AK |
4138 | do { |
4139 | add_preempt_count(PREEMPT_ACTIVE); | |
4140 | ||
4141 | /* | |
4142 | * We keep the big kernel semaphore locked, but we | |
4143 | * clear ->lock_depth so that schedule() doesnt | |
4144 | * auto-release the semaphore: | |
4145 | */ | |
3a5c359a AK |
4146 | saved_lock_depth = task->lock_depth; |
4147 | task->lock_depth = -1; | |
3a5c359a | 4148 | schedule(); |
3a5c359a | 4149 | task->lock_depth = saved_lock_depth; |
3a5c359a | 4150 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4151 | |
3a5c359a AK |
4152 | /* |
4153 | * Check again in case we missed a preemption opportunity | |
4154 | * between schedule and now. | |
4155 | */ | |
4156 | barrier(); | |
4157 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4158 | } |
1da177e4 LT |
4159 | EXPORT_SYMBOL(preempt_schedule); |
4160 | ||
4161 | /* | |
2ed6e34f | 4162 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4163 | * off of irq context. |
4164 | * Note, that this is called and return with irqs disabled. This will | |
4165 | * protect us against recursive calling from irq. | |
4166 | */ | |
4167 | asmlinkage void __sched preempt_schedule_irq(void) | |
4168 | { | |
4169 | struct thread_info *ti = current_thread_info(); | |
1da177e4 LT |
4170 | struct task_struct *task = current; |
4171 | int saved_lock_depth; | |
6478d880 | 4172 | |
2ed6e34f | 4173 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4174 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4175 | ||
3a5c359a AK |
4176 | do { |
4177 | add_preempt_count(PREEMPT_ACTIVE); | |
4178 | ||
4179 | /* | |
4180 | * We keep the big kernel semaphore locked, but we | |
4181 | * clear ->lock_depth so that schedule() doesnt | |
4182 | * auto-release the semaphore: | |
4183 | */ | |
3a5c359a AK |
4184 | saved_lock_depth = task->lock_depth; |
4185 | task->lock_depth = -1; | |
3a5c359a AK |
4186 | local_irq_enable(); |
4187 | schedule(); | |
4188 | local_irq_disable(); | |
3a5c359a | 4189 | task->lock_depth = saved_lock_depth; |
3a5c359a | 4190 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4191 | |
3a5c359a AK |
4192 | /* |
4193 | * Check again in case we missed a preemption opportunity | |
4194 | * between schedule and now. | |
4195 | */ | |
4196 | barrier(); | |
4197 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4198 | } |
4199 | ||
4200 | #endif /* CONFIG_PREEMPT */ | |
4201 | ||
95cdf3b7 IM |
4202 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4203 | void *key) | |
1da177e4 | 4204 | { |
48f24c4d | 4205 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4206 | } |
1da177e4 LT |
4207 | EXPORT_SYMBOL(default_wake_function); |
4208 | ||
4209 | /* | |
41a2d6cf IM |
4210 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4211 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4212 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4213 | * | |
4214 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4215 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4216 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4217 | */ | |
4218 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4219 | int nr_exclusive, int sync, void *key) | |
4220 | { | |
2e45874c | 4221 | wait_queue_t *curr, *next; |
1da177e4 | 4222 | |
2e45874c | 4223 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4224 | unsigned flags = curr->flags; |
4225 | ||
1da177e4 | 4226 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4227 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4228 | break; |
4229 | } | |
4230 | } | |
4231 | ||
4232 | /** | |
4233 | * __wake_up - wake up threads blocked on a waitqueue. | |
4234 | * @q: the waitqueue | |
4235 | * @mode: which threads | |
4236 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4237 | * @key: is directly passed to the wakeup function |
1da177e4 | 4238 | */ |
7ad5b3a5 | 4239 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4240 | int nr_exclusive, void *key) |
1da177e4 LT |
4241 | { |
4242 | unsigned long flags; | |
4243 | ||
4244 | spin_lock_irqsave(&q->lock, flags); | |
4245 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4246 | spin_unlock_irqrestore(&q->lock, flags); | |
4247 | } | |
1da177e4 LT |
4248 | EXPORT_SYMBOL(__wake_up); |
4249 | ||
4250 | /* | |
4251 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4252 | */ | |
7ad5b3a5 | 4253 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4254 | { |
4255 | __wake_up_common(q, mode, 1, 0, NULL); | |
4256 | } | |
4257 | ||
4258 | /** | |
67be2dd1 | 4259 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4260 | * @q: the waitqueue |
4261 | * @mode: which threads | |
4262 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4263 | * | |
4264 | * The sync wakeup differs that the waker knows that it will schedule | |
4265 | * away soon, so while the target thread will be woken up, it will not | |
4266 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4267 | * with each other. This can prevent needless bouncing between CPUs. | |
4268 | * | |
4269 | * On UP it can prevent extra preemption. | |
4270 | */ | |
7ad5b3a5 | 4271 | void |
95cdf3b7 | 4272 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4273 | { |
4274 | unsigned long flags; | |
4275 | int sync = 1; | |
4276 | ||
4277 | if (unlikely(!q)) | |
4278 | return; | |
4279 | ||
4280 | if (unlikely(!nr_exclusive)) | |
4281 | sync = 0; | |
4282 | ||
4283 | spin_lock_irqsave(&q->lock, flags); | |
4284 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4285 | spin_unlock_irqrestore(&q->lock, flags); | |
4286 | } | |
4287 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4288 | ||
b15136e9 | 4289 | void complete(struct completion *x) |
1da177e4 LT |
4290 | { |
4291 | unsigned long flags; | |
4292 | ||
4293 | spin_lock_irqsave(&x->wait.lock, flags); | |
4294 | x->done++; | |
d9514f6c | 4295 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4296 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4297 | } | |
4298 | EXPORT_SYMBOL(complete); | |
4299 | ||
b15136e9 | 4300 | void complete_all(struct completion *x) |
1da177e4 LT |
4301 | { |
4302 | unsigned long flags; | |
4303 | ||
4304 | spin_lock_irqsave(&x->wait.lock, flags); | |
4305 | x->done += UINT_MAX/2; | |
d9514f6c | 4306 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4307 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4308 | } | |
4309 | EXPORT_SYMBOL(complete_all); | |
4310 | ||
8cbbe86d AK |
4311 | static inline long __sched |
4312 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4313 | { |
1da177e4 LT |
4314 | if (!x->done) { |
4315 | DECLARE_WAITQUEUE(wait, current); | |
4316 | ||
4317 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4318 | __add_wait_queue_tail(&x->wait, &wait); | |
4319 | do { | |
009e577e MW |
4320 | if ((state == TASK_INTERRUPTIBLE && |
4321 | signal_pending(current)) || | |
4322 | (state == TASK_KILLABLE && | |
4323 | fatal_signal_pending(current))) { | |
8cbbe86d AK |
4324 | __remove_wait_queue(&x->wait, &wait); |
4325 | return -ERESTARTSYS; | |
4326 | } | |
4327 | __set_current_state(state); | |
1da177e4 LT |
4328 | spin_unlock_irq(&x->wait.lock); |
4329 | timeout = schedule_timeout(timeout); | |
4330 | spin_lock_irq(&x->wait.lock); | |
4331 | if (!timeout) { | |
4332 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 4333 | return timeout; |
1da177e4 LT |
4334 | } |
4335 | } while (!x->done); | |
4336 | __remove_wait_queue(&x->wait, &wait); | |
4337 | } | |
4338 | x->done--; | |
1da177e4 LT |
4339 | return timeout; |
4340 | } | |
1da177e4 | 4341 | |
8cbbe86d AK |
4342 | static long __sched |
4343 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4344 | { |
1da177e4 LT |
4345 | might_sleep(); |
4346 | ||
4347 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4348 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4349 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4350 | return timeout; |
4351 | } | |
1da177e4 | 4352 | |
b15136e9 | 4353 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4354 | { |
4355 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4356 | } |
8cbbe86d | 4357 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4358 | |
b15136e9 | 4359 | unsigned long __sched |
8cbbe86d | 4360 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4361 | { |
8cbbe86d | 4362 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4363 | } |
8cbbe86d | 4364 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4365 | |
8cbbe86d | 4366 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4367 | { |
51e97990 AK |
4368 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4369 | if (t == -ERESTARTSYS) | |
4370 | return t; | |
4371 | return 0; | |
0fec171c | 4372 | } |
8cbbe86d | 4373 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4374 | |
b15136e9 | 4375 | unsigned long __sched |
8cbbe86d AK |
4376 | wait_for_completion_interruptible_timeout(struct completion *x, |
4377 | unsigned long timeout) | |
0fec171c | 4378 | { |
8cbbe86d | 4379 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4380 | } |
8cbbe86d | 4381 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4382 | |
009e577e MW |
4383 | int __sched wait_for_completion_killable(struct completion *x) |
4384 | { | |
4385 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4386 | if (t == -ERESTARTSYS) | |
4387 | return t; | |
4388 | return 0; | |
4389 | } | |
4390 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4391 | ||
8cbbe86d AK |
4392 | static long __sched |
4393 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4394 | { |
0fec171c IM |
4395 | unsigned long flags; |
4396 | wait_queue_t wait; | |
4397 | ||
4398 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4399 | |
8cbbe86d | 4400 | __set_current_state(state); |
1da177e4 | 4401 | |
8cbbe86d AK |
4402 | spin_lock_irqsave(&q->lock, flags); |
4403 | __add_wait_queue(q, &wait); | |
4404 | spin_unlock(&q->lock); | |
4405 | timeout = schedule_timeout(timeout); | |
4406 | spin_lock_irq(&q->lock); | |
4407 | __remove_wait_queue(q, &wait); | |
4408 | spin_unlock_irqrestore(&q->lock, flags); | |
4409 | ||
4410 | return timeout; | |
4411 | } | |
4412 | ||
4413 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4414 | { | |
4415 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4416 | } |
1da177e4 LT |
4417 | EXPORT_SYMBOL(interruptible_sleep_on); |
4418 | ||
0fec171c | 4419 | long __sched |
95cdf3b7 | 4420 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4421 | { |
8cbbe86d | 4422 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4423 | } |
1da177e4 LT |
4424 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4425 | ||
0fec171c | 4426 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4427 | { |
8cbbe86d | 4428 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4429 | } |
1da177e4 LT |
4430 | EXPORT_SYMBOL(sleep_on); |
4431 | ||
0fec171c | 4432 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4433 | { |
8cbbe86d | 4434 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4435 | } |
1da177e4 LT |
4436 | EXPORT_SYMBOL(sleep_on_timeout); |
4437 | ||
b29739f9 IM |
4438 | #ifdef CONFIG_RT_MUTEXES |
4439 | ||
4440 | /* | |
4441 | * rt_mutex_setprio - set the current priority of a task | |
4442 | * @p: task | |
4443 | * @prio: prio value (kernel-internal form) | |
4444 | * | |
4445 | * This function changes the 'effective' priority of a task. It does | |
4446 | * not touch ->normal_prio like __setscheduler(). | |
4447 | * | |
4448 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4449 | */ | |
36c8b586 | 4450 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4451 | { |
4452 | unsigned long flags; | |
83b699ed | 4453 | int oldprio, on_rq, running; |
70b97a7f | 4454 | struct rq *rq; |
cb469845 | 4455 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4456 | |
4457 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4458 | ||
4459 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4460 | update_rq_clock(rq); |
b29739f9 | 4461 | |
d5f9f942 | 4462 | oldprio = p->prio; |
dd41f596 | 4463 | on_rq = p->se.on_rq; |
051a1d1a | 4464 | running = task_current(rq, p); |
0e1f3483 | 4465 | if (on_rq) |
69be72c1 | 4466 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4467 | if (running) |
4468 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4469 | |
4470 | if (rt_prio(prio)) | |
4471 | p->sched_class = &rt_sched_class; | |
4472 | else | |
4473 | p->sched_class = &fair_sched_class; | |
4474 | ||
b29739f9 IM |
4475 | p->prio = prio; |
4476 | ||
0e1f3483 HS |
4477 | if (running) |
4478 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4479 | if (on_rq) { |
8159f87e | 4480 | enqueue_task(rq, p, 0); |
cb469845 SR |
4481 | |
4482 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4483 | } |
4484 | task_rq_unlock(rq, &flags); | |
4485 | } | |
4486 | ||
4487 | #endif | |
4488 | ||
36c8b586 | 4489 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4490 | { |
dd41f596 | 4491 | int old_prio, delta, on_rq; |
1da177e4 | 4492 | unsigned long flags; |
70b97a7f | 4493 | struct rq *rq; |
1da177e4 LT |
4494 | |
4495 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4496 | return; | |
4497 | /* | |
4498 | * We have to be careful, if called from sys_setpriority(), | |
4499 | * the task might be in the middle of scheduling on another CPU. | |
4500 | */ | |
4501 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4502 | update_rq_clock(rq); |
1da177e4 LT |
4503 | /* |
4504 | * The RT priorities are set via sched_setscheduler(), but we still | |
4505 | * allow the 'normal' nice value to be set - but as expected | |
4506 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4507 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4508 | */ |
e05606d3 | 4509 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4510 | p->static_prio = NICE_TO_PRIO(nice); |
4511 | goto out_unlock; | |
4512 | } | |
dd41f596 | 4513 | on_rq = p->se.on_rq; |
62fb1851 | 4514 | if (on_rq) { |
69be72c1 | 4515 | dequeue_task(rq, p, 0); |
62fb1851 PZ |
4516 | dec_load(rq, p); |
4517 | } | |
1da177e4 | 4518 | |
1da177e4 | 4519 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4520 | set_load_weight(p); |
b29739f9 IM |
4521 | old_prio = p->prio; |
4522 | p->prio = effective_prio(p); | |
4523 | delta = p->prio - old_prio; | |
1da177e4 | 4524 | |
dd41f596 | 4525 | if (on_rq) { |
8159f87e | 4526 | enqueue_task(rq, p, 0); |
62fb1851 | 4527 | inc_load(rq, p); |
1da177e4 | 4528 | /* |
d5f9f942 AM |
4529 | * If the task increased its priority or is running and |
4530 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4531 | */ |
d5f9f942 | 4532 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4533 | resched_task(rq->curr); |
4534 | } | |
4535 | out_unlock: | |
4536 | task_rq_unlock(rq, &flags); | |
4537 | } | |
1da177e4 LT |
4538 | EXPORT_SYMBOL(set_user_nice); |
4539 | ||
e43379f1 MM |
4540 | /* |
4541 | * can_nice - check if a task can reduce its nice value | |
4542 | * @p: task | |
4543 | * @nice: nice value | |
4544 | */ | |
36c8b586 | 4545 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4546 | { |
024f4747 MM |
4547 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4548 | int nice_rlim = 20 - nice; | |
48f24c4d | 4549 | |
e43379f1 MM |
4550 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4551 | capable(CAP_SYS_NICE)); | |
4552 | } | |
4553 | ||
1da177e4 LT |
4554 | #ifdef __ARCH_WANT_SYS_NICE |
4555 | ||
4556 | /* | |
4557 | * sys_nice - change the priority of the current process. | |
4558 | * @increment: priority increment | |
4559 | * | |
4560 | * sys_setpriority is a more generic, but much slower function that | |
4561 | * does similar things. | |
4562 | */ | |
4563 | asmlinkage long sys_nice(int increment) | |
4564 | { | |
48f24c4d | 4565 | long nice, retval; |
1da177e4 LT |
4566 | |
4567 | /* | |
4568 | * Setpriority might change our priority at the same moment. | |
4569 | * We don't have to worry. Conceptually one call occurs first | |
4570 | * and we have a single winner. | |
4571 | */ | |
e43379f1 MM |
4572 | if (increment < -40) |
4573 | increment = -40; | |
1da177e4 LT |
4574 | if (increment > 40) |
4575 | increment = 40; | |
4576 | ||
4577 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4578 | if (nice < -20) | |
4579 | nice = -20; | |
4580 | if (nice > 19) | |
4581 | nice = 19; | |
4582 | ||
e43379f1 MM |
4583 | if (increment < 0 && !can_nice(current, nice)) |
4584 | return -EPERM; | |
4585 | ||
1da177e4 LT |
4586 | retval = security_task_setnice(current, nice); |
4587 | if (retval) | |
4588 | return retval; | |
4589 | ||
4590 | set_user_nice(current, nice); | |
4591 | return 0; | |
4592 | } | |
4593 | ||
4594 | #endif | |
4595 | ||
4596 | /** | |
4597 | * task_prio - return the priority value of a given task. | |
4598 | * @p: the task in question. | |
4599 | * | |
4600 | * This is the priority value as seen by users in /proc. | |
4601 | * RT tasks are offset by -200. Normal tasks are centered | |
4602 | * around 0, value goes from -16 to +15. | |
4603 | */ | |
36c8b586 | 4604 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4605 | { |
4606 | return p->prio - MAX_RT_PRIO; | |
4607 | } | |
4608 | ||
4609 | /** | |
4610 | * task_nice - return the nice value of a given task. | |
4611 | * @p: the task in question. | |
4612 | */ | |
36c8b586 | 4613 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4614 | { |
4615 | return TASK_NICE(p); | |
4616 | } | |
150d8bed | 4617 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4618 | |
4619 | /** | |
4620 | * idle_cpu - is a given cpu idle currently? | |
4621 | * @cpu: the processor in question. | |
4622 | */ | |
4623 | int idle_cpu(int cpu) | |
4624 | { | |
4625 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4626 | } | |
4627 | ||
1da177e4 LT |
4628 | /** |
4629 | * idle_task - return the idle task for a given cpu. | |
4630 | * @cpu: the processor in question. | |
4631 | */ | |
36c8b586 | 4632 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4633 | { |
4634 | return cpu_rq(cpu)->idle; | |
4635 | } | |
4636 | ||
4637 | /** | |
4638 | * find_process_by_pid - find a process with a matching PID value. | |
4639 | * @pid: the pid in question. | |
4640 | */ | |
a9957449 | 4641 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4642 | { |
228ebcbe | 4643 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4644 | } |
4645 | ||
4646 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4647 | static void |
4648 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4649 | { |
dd41f596 | 4650 | BUG_ON(p->se.on_rq); |
48f24c4d | 4651 | |
1da177e4 | 4652 | p->policy = policy; |
dd41f596 IM |
4653 | switch (p->policy) { |
4654 | case SCHED_NORMAL: | |
4655 | case SCHED_BATCH: | |
4656 | case SCHED_IDLE: | |
4657 | p->sched_class = &fair_sched_class; | |
4658 | break; | |
4659 | case SCHED_FIFO: | |
4660 | case SCHED_RR: | |
4661 | p->sched_class = &rt_sched_class; | |
4662 | break; | |
4663 | } | |
4664 | ||
1da177e4 | 4665 | p->rt_priority = prio; |
b29739f9 IM |
4666 | p->normal_prio = normal_prio(p); |
4667 | /* we are holding p->pi_lock already */ | |
4668 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4669 | set_load_weight(p); |
1da177e4 LT |
4670 | } |
4671 | ||
4672 | /** | |
72fd4a35 | 4673 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4674 | * @p: the task in question. |
4675 | * @policy: new policy. | |
4676 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4677 | * |
72fd4a35 | 4678 | * NOTE that the task may be already dead. |
1da177e4 | 4679 | */ |
95cdf3b7 IM |
4680 | int sched_setscheduler(struct task_struct *p, int policy, |
4681 | struct sched_param *param) | |
1da177e4 | 4682 | { |
83b699ed | 4683 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4684 | unsigned long flags; |
cb469845 | 4685 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4686 | struct rq *rq; |
1da177e4 | 4687 | |
66e5393a SR |
4688 | /* may grab non-irq protected spin_locks */ |
4689 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4690 | recheck: |
4691 | /* double check policy once rq lock held */ | |
4692 | if (policy < 0) | |
4693 | policy = oldpolicy = p->policy; | |
4694 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4695 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4696 | policy != SCHED_IDLE) | |
b0a9499c | 4697 | return -EINVAL; |
1da177e4 LT |
4698 | /* |
4699 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4700 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4701 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4702 | */ |
4703 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4704 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4705 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4706 | return -EINVAL; |
e05606d3 | 4707 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4708 | return -EINVAL; |
4709 | ||
37e4ab3f OC |
4710 | /* |
4711 | * Allow unprivileged RT tasks to decrease priority: | |
4712 | */ | |
4713 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4714 | if (rt_policy(policy)) { |
8dc3e909 | 4715 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4716 | |
4717 | if (!lock_task_sighand(p, &flags)) | |
4718 | return -ESRCH; | |
4719 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4720 | unlock_task_sighand(p, &flags); | |
4721 | ||
4722 | /* can't set/change the rt policy */ | |
4723 | if (policy != p->policy && !rlim_rtprio) | |
4724 | return -EPERM; | |
4725 | ||
4726 | /* can't increase priority */ | |
4727 | if (param->sched_priority > p->rt_priority && | |
4728 | param->sched_priority > rlim_rtprio) | |
4729 | return -EPERM; | |
4730 | } | |
dd41f596 IM |
4731 | /* |
4732 | * Like positive nice levels, dont allow tasks to | |
4733 | * move out of SCHED_IDLE either: | |
4734 | */ | |
4735 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4736 | return -EPERM; | |
5fe1d75f | 4737 | |
37e4ab3f OC |
4738 | /* can't change other user's priorities */ |
4739 | if ((current->euid != p->euid) && | |
4740 | (current->euid != p->uid)) | |
4741 | return -EPERM; | |
4742 | } | |
1da177e4 | 4743 | |
b68aa230 PZ |
4744 | #ifdef CONFIG_RT_GROUP_SCHED |
4745 | /* | |
4746 | * Do not allow realtime tasks into groups that have no runtime | |
4747 | * assigned. | |
4748 | */ | |
d0b27fa7 | 4749 | if (rt_policy(policy) && task_group(p)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
4750 | return -EPERM; |
4751 | #endif | |
4752 | ||
1da177e4 LT |
4753 | retval = security_task_setscheduler(p, policy, param); |
4754 | if (retval) | |
4755 | return retval; | |
b29739f9 IM |
4756 | /* |
4757 | * make sure no PI-waiters arrive (or leave) while we are | |
4758 | * changing the priority of the task: | |
4759 | */ | |
4760 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4761 | /* |
4762 | * To be able to change p->policy safely, the apropriate | |
4763 | * runqueue lock must be held. | |
4764 | */ | |
b29739f9 | 4765 | rq = __task_rq_lock(p); |
1da177e4 LT |
4766 | /* recheck policy now with rq lock held */ |
4767 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4768 | policy = oldpolicy = -1; | |
b29739f9 IM |
4769 | __task_rq_unlock(rq); |
4770 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4771 | goto recheck; |
4772 | } | |
2daa3577 | 4773 | update_rq_clock(rq); |
dd41f596 | 4774 | on_rq = p->se.on_rq; |
051a1d1a | 4775 | running = task_current(rq, p); |
0e1f3483 | 4776 | if (on_rq) |
2e1cb74a | 4777 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4778 | if (running) |
4779 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4780 | |
1da177e4 | 4781 | oldprio = p->prio; |
dd41f596 | 4782 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4783 | |
0e1f3483 HS |
4784 | if (running) |
4785 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4786 | if (on_rq) { |
4787 | activate_task(rq, p, 0); | |
cb469845 SR |
4788 | |
4789 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4790 | } |
b29739f9 IM |
4791 | __task_rq_unlock(rq); |
4792 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4793 | ||
95e02ca9 TG |
4794 | rt_mutex_adjust_pi(p); |
4795 | ||
1da177e4 LT |
4796 | return 0; |
4797 | } | |
4798 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4799 | ||
95cdf3b7 IM |
4800 | static int |
4801 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4802 | { |
1da177e4 LT |
4803 | struct sched_param lparam; |
4804 | struct task_struct *p; | |
36c8b586 | 4805 | int retval; |
1da177e4 LT |
4806 | |
4807 | if (!param || pid < 0) | |
4808 | return -EINVAL; | |
4809 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4810 | return -EFAULT; | |
5fe1d75f ON |
4811 | |
4812 | rcu_read_lock(); | |
4813 | retval = -ESRCH; | |
1da177e4 | 4814 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4815 | if (p != NULL) |
4816 | retval = sched_setscheduler(p, policy, &lparam); | |
4817 | rcu_read_unlock(); | |
36c8b586 | 4818 | |
1da177e4 LT |
4819 | return retval; |
4820 | } | |
4821 | ||
4822 | /** | |
4823 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4824 | * @pid: the pid in question. | |
4825 | * @policy: new policy. | |
4826 | * @param: structure containing the new RT priority. | |
4827 | */ | |
41a2d6cf IM |
4828 | asmlinkage long |
4829 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4830 | { |
c21761f1 JB |
4831 | /* negative values for policy are not valid */ |
4832 | if (policy < 0) | |
4833 | return -EINVAL; | |
4834 | ||
1da177e4 LT |
4835 | return do_sched_setscheduler(pid, policy, param); |
4836 | } | |
4837 | ||
4838 | /** | |
4839 | * sys_sched_setparam - set/change the RT priority of a thread | |
4840 | * @pid: the pid in question. | |
4841 | * @param: structure containing the new RT priority. | |
4842 | */ | |
4843 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4844 | { | |
4845 | return do_sched_setscheduler(pid, -1, param); | |
4846 | } | |
4847 | ||
4848 | /** | |
4849 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4850 | * @pid: the pid in question. | |
4851 | */ | |
4852 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4853 | { | |
36c8b586 | 4854 | struct task_struct *p; |
3a5c359a | 4855 | int retval; |
1da177e4 LT |
4856 | |
4857 | if (pid < 0) | |
3a5c359a | 4858 | return -EINVAL; |
1da177e4 LT |
4859 | |
4860 | retval = -ESRCH; | |
4861 | read_lock(&tasklist_lock); | |
4862 | p = find_process_by_pid(pid); | |
4863 | if (p) { | |
4864 | retval = security_task_getscheduler(p); | |
4865 | if (!retval) | |
4866 | retval = p->policy; | |
4867 | } | |
4868 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4869 | return retval; |
4870 | } | |
4871 | ||
4872 | /** | |
4873 | * sys_sched_getscheduler - get the RT priority of a thread | |
4874 | * @pid: the pid in question. | |
4875 | * @param: structure containing the RT priority. | |
4876 | */ | |
4877 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4878 | { | |
4879 | struct sched_param lp; | |
36c8b586 | 4880 | struct task_struct *p; |
3a5c359a | 4881 | int retval; |
1da177e4 LT |
4882 | |
4883 | if (!param || pid < 0) | |
3a5c359a | 4884 | return -EINVAL; |
1da177e4 LT |
4885 | |
4886 | read_lock(&tasklist_lock); | |
4887 | p = find_process_by_pid(pid); | |
4888 | retval = -ESRCH; | |
4889 | if (!p) | |
4890 | goto out_unlock; | |
4891 | ||
4892 | retval = security_task_getscheduler(p); | |
4893 | if (retval) | |
4894 | goto out_unlock; | |
4895 | ||
4896 | lp.sched_priority = p->rt_priority; | |
4897 | read_unlock(&tasklist_lock); | |
4898 | ||
4899 | /* | |
4900 | * This one might sleep, we cannot do it with a spinlock held ... | |
4901 | */ | |
4902 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4903 | ||
1da177e4 LT |
4904 | return retval; |
4905 | ||
4906 | out_unlock: | |
4907 | read_unlock(&tasklist_lock); | |
4908 | return retval; | |
4909 | } | |
4910 | ||
b53e921b | 4911 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 4912 | { |
1da177e4 | 4913 | cpumask_t cpus_allowed; |
b53e921b | 4914 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
4915 | struct task_struct *p; |
4916 | int retval; | |
1da177e4 | 4917 | |
95402b38 | 4918 | get_online_cpus(); |
1da177e4 LT |
4919 | read_lock(&tasklist_lock); |
4920 | ||
4921 | p = find_process_by_pid(pid); | |
4922 | if (!p) { | |
4923 | read_unlock(&tasklist_lock); | |
95402b38 | 4924 | put_online_cpus(); |
1da177e4 LT |
4925 | return -ESRCH; |
4926 | } | |
4927 | ||
4928 | /* | |
4929 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 4930 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
4931 | * usage count and then drop tasklist_lock. |
4932 | */ | |
4933 | get_task_struct(p); | |
4934 | read_unlock(&tasklist_lock); | |
4935 | ||
4936 | retval = -EPERM; | |
4937 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4938 | !capable(CAP_SYS_NICE)) | |
4939 | goto out_unlock; | |
4940 | ||
e7834f8f DQ |
4941 | retval = security_task_setscheduler(p, 0, NULL); |
4942 | if (retval) | |
4943 | goto out_unlock; | |
4944 | ||
f9a86fcb | 4945 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 4946 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 4947 | again: |
1da177e4 LT |
4948 | retval = set_cpus_allowed(p, new_mask); |
4949 | ||
8707d8b8 | 4950 | if (!retval) { |
f9a86fcb | 4951 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
4952 | if (!cpus_subset(new_mask, cpus_allowed)) { |
4953 | /* | |
4954 | * We must have raced with a concurrent cpuset | |
4955 | * update. Just reset the cpus_allowed to the | |
4956 | * cpuset's cpus_allowed | |
4957 | */ | |
4958 | new_mask = cpus_allowed; | |
4959 | goto again; | |
4960 | } | |
4961 | } | |
1da177e4 LT |
4962 | out_unlock: |
4963 | put_task_struct(p); | |
95402b38 | 4964 | put_online_cpus(); |
1da177e4 LT |
4965 | return retval; |
4966 | } | |
4967 | ||
4968 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4969 | cpumask_t *new_mask) | |
4970 | { | |
4971 | if (len < sizeof(cpumask_t)) { | |
4972 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4973 | } else if (len > sizeof(cpumask_t)) { | |
4974 | len = sizeof(cpumask_t); | |
4975 | } | |
4976 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4977 | } | |
4978 | ||
4979 | /** | |
4980 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4981 | * @pid: pid of the process | |
4982 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4983 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4984 | */ | |
4985 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4986 | unsigned long __user *user_mask_ptr) | |
4987 | { | |
4988 | cpumask_t new_mask; | |
4989 | int retval; | |
4990 | ||
4991 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4992 | if (retval) | |
4993 | return retval; | |
4994 | ||
b53e921b | 4995 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
4996 | } |
4997 | ||
4998 | /* | |
4999 | * Represents all cpu's present in the system | |
5000 | * In systems capable of hotplug, this map could dynamically grow | |
5001 | * as new cpu's are detected in the system via any platform specific | |
5002 | * method, such as ACPI for e.g. | |
5003 | */ | |
5004 | ||
4cef0c61 | 5005 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
5006 | EXPORT_SYMBOL(cpu_present_map); |
5007 | ||
5008 | #ifndef CONFIG_SMP | |
4cef0c61 | 5009 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
5010 | EXPORT_SYMBOL(cpu_online_map); |
5011 | ||
4cef0c61 | 5012 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 5013 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
5014 | #endif |
5015 | ||
5016 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
5017 | { | |
36c8b586 | 5018 | struct task_struct *p; |
1da177e4 | 5019 | int retval; |
1da177e4 | 5020 | |
95402b38 | 5021 | get_online_cpus(); |
1da177e4 LT |
5022 | read_lock(&tasklist_lock); |
5023 | ||
5024 | retval = -ESRCH; | |
5025 | p = find_process_by_pid(pid); | |
5026 | if (!p) | |
5027 | goto out_unlock; | |
5028 | ||
e7834f8f DQ |
5029 | retval = security_task_getscheduler(p); |
5030 | if (retval) | |
5031 | goto out_unlock; | |
5032 | ||
2f7016d9 | 5033 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5034 | |
5035 | out_unlock: | |
5036 | read_unlock(&tasklist_lock); | |
95402b38 | 5037 | put_online_cpus(); |
1da177e4 | 5038 | |
9531b62f | 5039 | return retval; |
1da177e4 LT |
5040 | } |
5041 | ||
5042 | /** | |
5043 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5044 | * @pid: pid of the process | |
5045 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5046 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5047 | */ | |
5048 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5049 | unsigned long __user *user_mask_ptr) | |
5050 | { | |
5051 | int ret; | |
5052 | cpumask_t mask; | |
5053 | ||
5054 | if (len < sizeof(cpumask_t)) | |
5055 | return -EINVAL; | |
5056 | ||
5057 | ret = sched_getaffinity(pid, &mask); | |
5058 | if (ret < 0) | |
5059 | return ret; | |
5060 | ||
5061 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5062 | return -EFAULT; | |
5063 | ||
5064 | return sizeof(cpumask_t); | |
5065 | } | |
5066 | ||
5067 | /** | |
5068 | * sys_sched_yield - yield the current processor to other threads. | |
5069 | * | |
dd41f596 IM |
5070 | * This function yields the current CPU to other tasks. If there are no |
5071 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5072 | */ |
5073 | asmlinkage long sys_sched_yield(void) | |
5074 | { | |
70b97a7f | 5075 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5076 | |
2d72376b | 5077 | schedstat_inc(rq, yld_count); |
4530d7ab | 5078 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5079 | |
5080 | /* | |
5081 | * Since we are going to call schedule() anyway, there's | |
5082 | * no need to preempt or enable interrupts: | |
5083 | */ | |
5084 | __release(rq->lock); | |
8a25d5de | 5085 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5086 | _raw_spin_unlock(&rq->lock); |
5087 | preempt_enable_no_resched(); | |
5088 | ||
5089 | schedule(); | |
5090 | ||
5091 | return 0; | |
5092 | } | |
5093 | ||
e7b38404 | 5094 | static void __cond_resched(void) |
1da177e4 | 5095 | { |
8e0a43d8 IM |
5096 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5097 | __might_sleep(__FILE__, __LINE__); | |
5098 | #endif | |
5bbcfd90 IM |
5099 | /* |
5100 | * The BKS might be reacquired before we have dropped | |
5101 | * PREEMPT_ACTIVE, which could trigger a second | |
5102 | * cond_resched() call. | |
5103 | */ | |
1da177e4 LT |
5104 | do { |
5105 | add_preempt_count(PREEMPT_ACTIVE); | |
5106 | schedule(); | |
5107 | sub_preempt_count(PREEMPT_ACTIVE); | |
5108 | } while (need_resched()); | |
5109 | } | |
5110 | ||
02b67cc3 HX |
5111 | #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PREEMPT_VOLUNTARY) |
5112 | int __sched _cond_resched(void) | |
1da177e4 | 5113 | { |
9414232f IM |
5114 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5115 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5116 | __cond_resched(); |
5117 | return 1; | |
5118 | } | |
5119 | return 0; | |
5120 | } | |
02b67cc3 HX |
5121 | EXPORT_SYMBOL(_cond_resched); |
5122 | #endif | |
1da177e4 LT |
5123 | |
5124 | /* | |
5125 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5126 | * call schedule, and on return reacquire the lock. | |
5127 | * | |
41a2d6cf | 5128 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5129 | * operations here to prevent schedule() from being called twice (once via |
5130 | * spin_unlock(), once by hand). | |
5131 | */ | |
95cdf3b7 | 5132 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5133 | { |
95c354fe | 5134 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5135 | int ret = 0; |
5136 | ||
95c354fe | 5137 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5138 | spin_unlock(lock); |
95c354fe NP |
5139 | if (resched && need_resched()) |
5140 | __cond_resched(); | |
5141 | else | |
5142 | cpu_relax(); | |
6df3cecb | 5143 | ret = 1; |
1da177e4 | 5144 | spin_lock(lock); |
1da177e4 | 5145 | } |
6df3cecb | 5146 | return ret; |
1da177e4 | 5147 | } |
1da177e4 LT |
5148 | EXPORT_SYMBOL(cond_resched_lock); |
5149 | ||
5150 | int __sched cond_resched_softirq(void) | |
5151 | { | |
5152 | BUG_ON(!in_softirq()); | |
5153 | ||
9414232f | 5154 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5155 | local_bh_enable(); |
1da177e4 LT |
5156 | __cond_resched(); |
5157 | local_bh_disable(); | |
5158 | return 1; | |
5159 | } | |
5160 | return 0; | |
5161 | } | |
1da177e4 LT |
5162 | EXPORT_SYMBOL(cond_resched_softirq); |
5163 | ||
1da177e4 LT |
5164 | /** |
5165 | * yield - yield the current processor to other threads. | |
5166 | * | |
72fd4a35 | 5167 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5168 | * thread runnable and calls sys_sched_yield(). |
5169 | */ | |
5170 | void __sched yield(void) | |
5171 | { | |
5172 | set_current_state(TASK_RUNNING); | |
5173 | sys_sched_yield(); | |
5174 | } | |
1da177e4 LT |
5175 | EXPORT_SYMBOL(yield); |
5176 | ||
5177 | /* | |
41a2d6cf | 5178 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5179 | * that process accounting knows that this is a task in IO wait state. |
5180 | * | |
5181 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5182 | * has set its backing_dev_info: the queue against which it should throttle) | |
5183 | */ | |
5184 | void __sched io_schedule(void) | |
5185 | { | |
70b97a7f | 5186 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5187 | |
0ff92245 | 5188 | delayacct_blkio_start(); |
1da177e4 LT |
5189 | atomic_inc(&rq->nr_iowait); |
5190 | schedule(); | |
5191 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5192 | delayacct_blkio_end(); |
1da177e4 | 5193 | } |
1da177e4 LT |
5194 | EXPORT_SYMBOL(io_schedule); |
5195 | ||
5196 | long __sched io_schedule_timeout(long timeout) | |
5197 | { | |
70b97a7f | 5198 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5199 | long ret; |
5200 | ||
0ff92245 | 5201 | delayacct_blkio_start(); |
1da177e4 LT |
5202 | atomic_inc(&rq->nr_iowait); |
5203 | ret = schedule_timeout(timeout); | |
5204 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5205 | delayacct_blkio_end(); |
1da177e4 LT |
5206 | return ret; |
5207 | } | |
5208 | ||
5209 | /** | |
5210 | * sys_sched_get_priority_max - return maximum RT priority. | |
5211 | * @policy: scheduling class. | |
5212 | * | |
5213 | * this syscall returns the maximum rt_priority that can be used | |
5214 | * by a given scheduling class. | |
5215 | */ | |
5216 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5217 | { | |
5218 | int ret = -EINVAL; | |
5219 | ||
5220 | switch (policy) { | |
5221 | case SCHED_FIFO: | |
5222 | case SCHED_RR: | |
5223 | ret = MAX_USER_RT_PRIO-1; | |
5224 | break; | |
5225 | case SCHED_NORMAL: | |
b0a9499c | 5226 | case SCHED_BATCH: |
dd41f596 | 5227 | case SCHED_IDLE: |
1da177e4 LT |
5228 | ret = 0; |
5229 | break; | |
5230 | } | |
5231 | return ret; | |
5232 | } | |
5233 | ||
5234 | /** | |
5235 | * sys_sched_get_priority_min - return minimum RT priority. | |
5236 | * @policy: scheduling class. | |
5237 | * | |
5238 | * this syscall returns the minimum rt_priority that can be used | |
5239 | * by a given scheduling class. | |
5240 | */ | |
5241 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5242 | { | |
5243 | int ret = -EINVAL; | |
5244 | ||
5245 | switch (policy) { | |
5246 | case SCHED_FIFO: | |
5247 | case SCHED_RR: | |
5248 | ret = 1; | |
5249 | break; | |
5250 | case SCHED_NORMAL: | |
b0a9499c | 5251 | case SCHED_BATCH: |
dd41f596 | 5252 | case SCHED_IDLE: |
1da177e4 LT |
5253 | ret = 0; |
5254 | } | |
5255 | return ret; | |
5256 | } | |
5257 | ||
5258 | /** | |
5259 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5260 | * @pid: pid of the process. | |
5261 | * @interval: userspace pointer to the timeslice value. | |
5262 | * | |
5263 | * this syscall writes the default timeslice value of a given process | |
5264 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5265 | */ | |
5266 | asmlinkage | |
5267 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5268 | { | |
36c8b586 | 5269 | struct task_struct *p; |
a4ec24b4 | 5270 | unsigned int time_slice; |
3a5c359a | 5271 | int retval; |
1da177e4 | 5272 | struct timespec t; |
1da177e4 LT |
5273 | |
5274 | if (pid < 0) | |
3a5c359a | 5275 | return -EINVAL; |
1da177e4 LT |
5276 | |
5277 | retval = -ESRCH; | |
5278 | read_lock(&tasklist_lock); | |
5279 | p = find_process_by_pid(pid); | |
5280 | if (!p) | |
5281 | goto out_unlock; | |
5282 | ||
5283 | retval = security_task_getscheduler(p); | |
5284 | if (retval) | |
5285 | goto out_unlock; | |
5286 | ||
77034937 IM |
5287 | /* |
5288 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5289 | * tasks that are on an otherwise idle runqueue: | |
5290 | */ | |
5291 | time_slice = 0; | |
5292 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5293 | time_slice = DEF_TIMESLICE; |
1868f958 | 5294 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5295 | struct sched_entity *se = &p->se; |
5296 | unsigned long flags; | |
5297 | struct rq *rq; | |
5298 | ||
5299 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5300 | if (rq->cfs.load.weight) |
5301 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5302 | task_rq_unlock(rq, &flags); |
5303 | } | |
1da177e4 | 5304 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5305 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5306 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5307 | return retval; |
3a5c359a | 5308 | |
1da177e4 LT |
5309 | out_unlock: |
5310 | read_unlock(&tasklist_lock); | |
5311 | return retval; | |
5312 | } | |
5313 | ||
2ed6e34f | 5314 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 | 5315 | |
82a1fcb9 | 5316 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5317 | { |
1da177e4 | 5318 | unsigned long free = 0; |
36c8b586 | 5319 | unsigned state; |
1da177e4 | 5320 | |
1da177e4 | 5321 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5322 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5323 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5324 | #if BITS_PER_LONG == 32 |
1da177e4 | 5325 | if (state == TASK_RUNNING) |
cc4ea795 | 5326 | printk(KERN_CONT " running "); |
1da177e4 | 5327 | else |
cc4ea795 | 5328 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5329 | #else |
5330 | if (state == TASK_RUNNING) | |
cc4ea795 | 5331 | printk(KERN_CONT " running task "); |
1da177e4 | 5332 | else |
cc4ea795 | 5333 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5334 | #endif |
5335 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5336 | { | |
10ebffde | 5337 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5338 | while (!*n) |
5339 | n++; | |
10ebffde | 5340 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5341 | } |
5342 | #endif | |
ba25f9dc | 5343 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5344 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5345 | |
5fb5e6de | 5346 | show_stack(p, NULL); |
1da177e4 LT |
5347 | } |
5348 | ||
e59e2ae2 | 5349 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5350 | { |
36c8b586 | 5351 | struct task_struct *g, *p; |
1da177e4 | 5352 | |
4bd77321 IM |
5353 | #if BITS_PER_LONG == 32 |
5354 | printk(KERN_INFO | |
5355 | " task PC stack pid father\n"); | |
1da177e4 | 5356 | #else |
4bd77321 IM |
5357 | printk(KERN_INFO |
5358 | " task PC stack pid father\n"); | |
1da177e4 LT |
5359 | #endif |
5360 | read_lock(&tasklist_lock); | |
5361 | do_each_thread(g, p) { | |
5362 | /* | |
5363 | * reset the NMI-timeout, listing all files on a slow | |
5364 | * console might take alot of time: | |
5365 | */ | |
5366 | touch_nmi_watchdog(); | |
39bc89fd | 5367 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5368 | sched_show_task(p); |
1da177e4 LT |
5369 | } while_each_thread(g, p); |
5370 | ||
04c9167f JF |
5371 | touch_all_softlockup_watchdogs(); |
5372 | ||
dd41f596 IM |
5373 | #ifdef CONFIG_SCHED_DEBUG |
5374 | sysrq_sched_debug_show(); | |
5375 | #endif | |
1da177e4 | 5376 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5377 | /* |
5378 | * Only show locks if all tasks are dumped: | |
5379 | */ | |
5380 | if (state_filter == -1) | |
5381 | debug_show_all_locks(); | |
1da177e4 LT |
5382 | } |
5383 | ||
1df21055 IM |
5384 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5385 | { | |
dd41f596 | 5386 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5387 | } |
5388 | ||
f340c0d1 IM |
5389 | /** |
5390 | * init_idle - set up an idle thread for a given CPU | |
5391 | * @idle: task in question | |
5392 | * @cpu: cpu the idle task belongs to | |
5393 | * | |
5394 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5395 | * flag, to make booting more robust. | |
5396 | */ | |
5c1e1767 | 5397 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5398 | { |
70b97a7f | 5399 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5400 | unsigned long flags; |
5401 | ||
dd41f596 IM |
5402 | __sched_fork(idle); |
5403 | idle->se.exec_start = sched_clock(); | |
5404 | ||
b29739f9 | 5405 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5406 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5407 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5408 | |
5409 | spin_lock_irqsave(&rq->lock, flags); | |
5410 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5411 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5412 | idle->oncpu = 1; | |
5413 | #endif | |
1da177e4 LT |
5414 | spin_unlock_irqrestore(&rq->lock, flags); |
5415 | ||
5416 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 5417 | task_thread_info(idle)->preempt_count = 0; |
6478d880 | 5418 | |
dd41f596 IM |
5419 | /* |
5420 | * The idle tasks have their own, simple scheduling class: | |
5421 | */ | |
5422 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5423 | } |
5424 | ||
5425 | /* | |
5426 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5427 | * indicates which cpus entered this state. This is used | |
5428 | * in the rcu update to wait only for active cpus. For system | |
5429 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5430 | * always be CPU_MASK_NONE. | |
5431 | */ | |
5432 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5433 | ||
19978ca6 IM |
5434 | /* |
5435 | * Increase the granularity value when there are more CPUs, | |
5436 | * because with more CPUs the 'effective latency' as visible | |
5437 | * to users decreases. But the relationship is not linear, | |
5438 | * so pick a second-best guess by going with the log2 of the | |
5439 | * number of CPUs. | |
5440 | * | |
5441 | * This idea comes from the SD scheduler of Con Kolivas: | |
5442 | */ | |
5443 | static inline void sched_init_granularity(void) | |
5444 | { | |
5445 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5446 | const unsigned long limit = 200000000; | |
5447 | ||
5448 | sysctl_sched_min_granularity *= factor; | |
5449 | if (sysctl_sched_min_granularity > limit) | |
5450 | sysctl_sched_min_granularity = limit; | |
5451 | ||
5452 | sysctl_sched_latency *= factor; | |
5453 | if (sysctl_sched_latency > limit) | |
5454 | sysctl_sched_latency = limit; | |
5455 | ||
5456 | sysctl_sched_wakeup_granularity *= factor; | |
19978ca6 IM |
5457 | } |
5458 | ||
1da177e4 LT |
5459 | #ifdef CONFIG_SMP |
5460 | /* | |
5461 | * This is how migration works: | |
5462 | * | |
70b97a7f | 5463 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5464 | * runqueue and wake up that CPU's migration thread. |
5465 | * 2) we down() the locked semaphore => thread blocks. | |
5466 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5467 | * thread off the CPU) | |
5468 | * 4) it gets the migration request and checks whether the migrated | |
5469 | * task is still in the wrong runqueue. | |
5470 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5471 | * it and puts it into the right queue. | |
5472 | * 6) migration thread up()s the semaphore. | |
5473 | * 7) we wake up and the migration is done. | |
5474 | */ | |
5475 | ||
5476 | /* | |
5477 | * Change a given task's CPU affinity. Migrate the thread to a | |
5478 | * proper CPU and schedule it away if the CPU it's executing on | |
5479 | * is removed from the allowed bitmask. | |
5480 | * | |
5481 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5482 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5483 | * call is not atomic; no spinlocks may be held. |
5484 | */ | |
36c8b586 | 5485 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 5486 | { |
70b97a7f | 5487 | struct migration_req req; |
1da177e4 | 5488 | unsigned long flags; |
70b97a7f | 5489 | struct rq *rq; |
48f24c4d | 5490 | int ret = 0; |
1da177e4 LT |
5491 | |
5492 | rq = task_rq_lock(p, &flags); | |
5493 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
5494 | ret = -EINVAL; | |
5495 | goto out; | |
5496 | } | |
5497 | ||
73fe6aae GH |
5498 | if (p->sched_class->set_cpus_allowed) |
5499 | p->sched_class->set_cpus_allowed(p, &new_mask); | |
5500 | else { | |
0eab9146 | 5501 | p->cpus_allowed = new_mask; |
6f505b16 | 5502 | p->rt.nr_cpus_allowed = cpus_weight(new_mask); |
73fe6aae GH |
5503 | } |
5504 | ||
1da177e4 LT |
5505 | /* Can the task run on the task's current CPU? If so, we're done */ |
5506 | if (cpu_isset(task_cpu(p), new_mask)) | |
5507 | goto out; | |
5508 | ||
5509 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
5510 | /* Need help from migration thread: drop lock and wait. */ | |
5511 | task_rq_unlock(rq, &flags); | |
5512 | wake_up_process(rq->migration_thread); | |
5513 | wait_for_completion(&req.done); | |
5514 | tlb_migrate_finish(p->mm); | |
5515 | return 0; | |
5516 | } | |
5517 | out: | |
5518 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5519 | |
1da177e4 LT |
5520 | return ret; |
5521 | } | |
1da177e4 LT |
5522 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5523 | ||
5524 | /* | |
41a2d6cf | 5525 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5526 | * this because either it can't run here any more (set_cpus_allowed() |
5527 | * away from this CPU, or CPU going down), or because we're | |
5528 | * attempting to rebalance this task on exec (sched_exec). | |
5529 | * | |
5530 | * So we race with normal scheduler movements, but that's OK, as long | |
5531 | * as the task is no longer on this CPU. | |
efc30814 KK |
5532 | * |
5533 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5534 | */ |
efc30814 | 5535 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5536 | { |
70b97a7f | 5537 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5538 | int ret = 0, on_rq; |
1da177e4 LT |
5539 | |
5540 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5541 | return ret; |
1da177e4 LT |
5542 | |
5543 | rq_src = cpu_rq(src_cpu); | |
5544 | rq_dest = cpu_rq(dest_cpu); | |
5545 | ||
5546 | double_rq_lock(rq_src, rq_dest); | |
5547 | /* Already moved. */ | |
5548 | if (task_cpu(p) != src_cpu) | |
5549 | goto out; | |
5550 | /* Affinity changed (again). */ | |
5551 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5552 | goto out; | |
5553 | ||
dd41f596 | 5554 | on_rq = p->se.on_rq; |
6e82a3be | 5555 | if (on_rq) |
2e1cb74a | 5556 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5557 | |
1da177e4 | 5558 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5559 | if (on_rq) { |
5560 | activate_task(rq_dest, p, 0); | |
5561 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5562 | } |
efc30814 | 5563 | ret = 1; |
1da177e4 LT |
5564 | out: |
5565 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5566 | return ret; |
1da177e4 LT |
5567 | } |
5568 | ||
5569 | /* | |
5570 | * migration_thread - this is a highprio system thread that performs | |
5571 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5572 | * another runqueue. | |
5573 | */ | |
95cdf3b7 | 5574 | static int migration_thread(void *data) |
1da177e4 | 5575 | { |
1da177e4 | 5576 | int cpu = (long)data; |
70b97a7f | 5577 | struct rq *rq; |
1da177e4 LT |
5578 | |
5579 | rq = cpu_rq(cpu); | |
5580 | BUG_ON(rq->migration_thread != current); | |
5581 | ||
5582 | set_current_state(TASK_INTERRUPTIBLE); | |
5583 | while (!kthread_should_stop()) { | |
70b97a7f | 5584 | struct migration_req *req; |
1da177e4 | 5585 | struct list_head *head; |
1da177e4 | 5586 | |
1da177e4 LT |
5587 | spin_lock_irq(&rq->lock); |
5588 | ||
5589 | if (cpu_is_offline(cpu)) { | |
5590 | spin_unlock_irq(&rq->lock); | |
5591 | goto wait_to_die; | |
5592 | } | |
5593 | ||
5594 | if (rq->active_balance) { | |
5595 | active_load_balance(rq, cpu); | |
5596 | rq->active_balance = 0; | |
5597 | } | |
5598 | ||
5599 | head = &rq->migration_queue; | |
5600 | ||
5601 | if (list_empty(head)) { | |
5602 | spin_unlock_irq(&rq->lock); | |
5603 | schedule(); | |
5604 | set_current_state(TASK_INTERRUPTIBLE); | |
5605 | continue; | |
5606 | } | |
70b97a7f | 5607 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5608 | list_del_init(head->next); |
5609 | ||
674311d5 NP |
5610 | spin_unlock(&rq->lock); |
5611 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5612 | local_irq_enable(); | |
1da177e4 LT |
5613 | |
5614 | complete(&req->done); | |
5615 | } | |
5616 | __set_current_state(TASK_RUNNING); | |
5617 | return 0; | |
5618 | ||
5619 | wait_to_die: | |
5620 | /* Wait for kthread_stop */ | |
5621 | set_current_state(TASK_INTERRUPTIBLE); | |
5622 | while (!kthread_should_stop()) { | |
5623 | schedule(); | |
5624 | set_current_state(TASK_INTERRUPTIBLE); | |
5625 | } | |
5626 | __set_current_state(TASK_RUNNING); | |
5627 | return 0; | |
5628 | } | |
5629 | ||
5630 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5631 | |
5632 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5633 | { | |
5634 | int ret; | |
5635 | ||
5636 | local_irq_disable(); | |
5637 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5638 | local_irq_enable(); | |
5639 | return ret; | |
5640 | } | |
5641 | ||
054b9108 | 5642 | /* |
3a4fa0a2 | 5643 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5644 | * NOTE: interrupts should be disabled by the caller |
5645 | */ | |
48f24c4d | 5646 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5647 | { |
efc30814 | 5648 | unsigned long flags; |
1da177e4 | 5649 | cpumask_t mask; |
70b97a7f IM |
5650 | struct rq *rq; |
5651 | int dest_cpu; | |
1da177e4 | 5652 | |
3a5c359a AK |
5653 | do { |
5654 | /* On same node? */ | |
5655 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5656 | cpus_and(mask, mask, p->cpus_allowed); | |
5657 | dest_cpu = any_online_cpu(mask); | |
5658 | ||
5659 | /* On any allowed CPU? */ | |
434d53b0 | 5660 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
5661 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5662 | ||
5663 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 5664 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
5665 | cpumask_t cpus_allowed; |
5666 | ||
5667 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
5668 | /* |
5669 | * Try to stay on the same cpuset, where the | |
5670 | * current cpuset may be a subset of all cpus. | |
5671 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5672 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5673 | * called within calls to cpuset_lock/cpuset_unlock. |
5674 | */ | |
3a5c359a | 5675 | rq = task_rq_lock(p, &flags); |
470fd646 | 5676 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5677 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5678 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5679 | |
3a5c359a AK |
5680 | /* |
5681 | * Don't tell them about moving exiting tasks or | |
5682 | * kernel threads (both mm NULL), since they never | |
5683 | * leave kernel. | |
5684 | */ | |
41a2d6cf | 5685 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5686 | printk(KERN_INFO "process %d (%s) no " |
5687 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5688 | task_pid_nr(p), p->comm, dead_cpu); |
5689 | } | |
3a5c359a | 5690 | } |
f7b4cddc | 5691 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5692 | } |
5693 | ||
5694 | /* | |
5695 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5696 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5697 | * for performance reasons the counter is not stricly tracking tasks to | |
5698 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5699 | * to keep the global sum constant after CPU-down: | |
5700 | */ | |
70b97a7f | 5701 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5702 | { |
70b97a7f | 5703 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5704 | unsigned long flags; |
5705 | ||
5706 | local_irq_save(flags); | |
5707 | double_rq_lock(rq_src, rq_dest); | |
5708 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5709 | rq_src->nr_uninterruptible = 0; | |
5710 | double_rq_unlock(rq_src, rq_dest); | |
5711 | local_irq_restore(flags); | |
5712 | } | |
5713 | ||
5714 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5715 | static void migrate_live_tasks(int src_cpu) | |
5716 | { | |
48f24c4d | 5717 | struct task_struct *p, *t; |
1da177e4 | 5718 | |
f7b4cddc | 5719 | read_lock(&tasklist_lock); |
1da177e4 | 5720 | |
48f24c4d IM |
5721 | do_each_thread(t, p) { |
5722 | if (p == current) | |
1da177e4 LT |
5723 | continue; |
5724 | ||
48f24c4d IM |
5725 | if (task_cpu(p) == src_cpu) |
5726 | move_task_off_dead_cpu(src_cpu, p); | |
5727 | } while_each_thread(t, p); | |
1da177e4 | 5728 | |
f7b4cddc | 5729 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5730 | } |
5731 | ||
dd41f596 IM |
5732 | /* |
5733 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5734 | * It does so by boosting its priority to highest possible. |
5735 | * Used by CPU offline code. | |
1da177e4 LT |
5736 | */ |
5737 | void sched_idle_next(void) | |
5738 | { | |
48f24c4d | 5739 | int this_cpu = smp_processor_id(); |
70b97a7f | 5740 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5741 | struct task_struct *p = rq->idle; |
5742 | unsigned long flags; | |
5743 | ||
5744 | /* cpu has to be offline */ | |
48f24c4d | 5745 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5746 | |
48f24c4d IM |
5747 | /* |
5748 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5749 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5750 | */ |
5751 | spin_lock_irqsave(&rq->lock, flags); | |
5752 | ||
dd41f596 | 5753 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5754 | |
94bc9a7b DA |
5755 | update_rq_clock(rq); |
5756 | activate_task(rq, p, 0); | |
1da177e4 LT |
5757 | |
5758 | spin_unlock_irqrestore(&rq->lock, flags); | |
5759 | } | |
5760 | ||
48f24c4d IM |
5761 | /* |
5762 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5763 | * offline. |
5764 | */ | |
5765 | void idle_task_exit(void) | |
5766 | { | |
5767 | struct mm_struct *mm = current->active_mm; | |
5768 | ||
5769 | BUG_ON(cpu_online(smp_processor_id())); | |
5770 | ||
5771 | if (mm != &init_mm) | |
5772 | switch_mm(mm, &init_mm, current); | |
5773 | mmdrop(mm); | |
5774 | } | |
5775 | ||
054b9108 | 5776 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5777 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5778 | { |
70b97a7f | 5779 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5780 | |
5781 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5782 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5783 | |
5784 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5785 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5786 | |
48f24c4d | 5787 | get_task_struct(p); |
1da177e4 LT |
5788 | |
5789 | /* | |
5790 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5791 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5792 | * fine. |
5793 | */ | |
f7b4cddc | 5794 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5795 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5796 | spin_lock_irq(&rq->lock); |
1da177e4 | 5797 | |
48f24c4d | 5798 | put_task_struct(p); |
1da177e4 LT |
5799 | } |
5800 | ||
5801 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5802 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5803 | { | |
70b97a7f | 5804 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5805 | struct task_struct *next; |
48f24c4d | 5806 | |
dd41f596 IM |
5807 | for ( ; ; ) { |
5808 | if (!rq->nr_running) | |
5809 | break; | |
a8e504d2 | 5810 | update_rq_clock(rq); |
ff95f3df | 5811 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5812 | if (!next) |
5813 | break; | |
5814 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5815 | |
1da177e4 LT |
5816 | } |
5817 | } | |
5818 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5819 | ||
e692ab53 NP |
5820 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5821 | ||
5822 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5823 | { |
5824 | .procname = "sched_domain", | |
c57baf1e | 5825 | .mode = 0555, |
e0361851 | 5826 | }, |
38605cae | 5827 | {0, }, |
e692ab53 NP |
5828 | }; |
5829 | ||
5830 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5831 | { |
c57baf1e | 5832 | .ctl_name = CTL_KERN, |
e0361851 | 5833 | .procname = "kernel", |
c57baf1e | 5834 | .mode = 0555, |
e0361851 AD |
5835 | .child = sd_ctl_dir, |
5836 | }, | |
38605cae | 5837 | {0, }, |
e692ab53 NP |
5838 | }; |
5839 | ||
5840 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5841 | { | |
5842 | struct ctl_table *entry = | |
5cf9f062 | 5843 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5844 | |
e692ab53 NP |
5845 | return entry; |
5846 | } | |
5847 | ||
6382bc90 MM |
5848 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5849 | { | |
cd790076 | 5850 | struct ctl_table *entry; |
6382bc90 | 5851 | |
cd790076 MM |
5852 | /* |
5853 | * In the intermediate directories, both the child directory and | |
5854 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5855 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5856 | * static strings and all have proc handlers. |
5857 | */ | |
5858 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5859 | if (entry->child) |
5860 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5861 | if (entry->proc_handler == NULL) |
5862 | kfree(entry->procname); | |
5863 | } | |
6382bc90 MM |
5864 | |
5865 | kfree(*tablep); | |
5866 | *tablep = NULL; | |
5867 | } | |
5868 | ||
e692ab53 | 5869 | static void |
e0361851 | 5870 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5871 | const char *procname, void *data, int maxlen, |
5872 | mode_t mode, proc_handler *proc_handler) | |
5873 | { | |
e692ab53 NP |
5874 | entry->procname = procname; |
5875 | entry->data = data; | |
5876 | entry->maxlen = maxlen; | |
5877 | entry->mode = mode; | |
5878 | entry->proc_handler = proc_handler; | |
5879 | } | |
5880 | ||
5881 | static struct ctl_table * | |
5882 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5883 | { | |
ace8b3d6 | 5884 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5885 | |
ad1cdc1d MM |
5886 | if (table == NULL) |
5887 | return NULL; | |
5888 | ||
e0361851 | 5889 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5890 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5891 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5892 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5893 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5894 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5895 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5896 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5897 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5898 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5899 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5900 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5901 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5902 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5903 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5904 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5905 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5906 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5907 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5908 | &sd->cache_nice_tries, |
5909 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5910 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5911 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5912 | /* &table[11] is terminator */ |
e692ab53 NP |
5913 | |
5914 | return table; | |
5915 | } | |
5916 | ||
9a4e7159 | 5917 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5918 | { |
5919 | struct ctl_table *entry, *table; | |
5920 | struct sched_domain *sd; | |
5921 | int domain_num = 0, i; | |
5922 | char buf[32]; | |
5923 | ||
5924 | for_each_domain(cpu, sd) | |
5925 | domain_num++; | |
5926 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5927 | if (table == NULL) |
5928 | return NULL; | |
e692ab53 NP |
5929 | |
5930 | i = 0; | |
5931 | for_each_domain(cpu, sd) { | |
5932 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5933 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5934 | entry->mode = 0555; |
e692ab53 NP |
5935 | entry->child = sd_alloc_ctl_domain_table(sd); |
5936 | entry++; | |
5937 | i++; | |
5938 | } | |
5939 | return table; | |
5940 | } | |
5941 | ||
5942 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5943 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5944 | { |
5945 | int i, cpu_num = num_online_cpus(); | |
5946 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5947 | char buf[32]; | |
5948 | ||
7378547f MM |
5949 | WARN_ON(sd_ctl_dir[0].child); |
5950 | sd_ctl_dir[0].child = entry; | |
5951 | ||
ad1cdc1d MM |
5952 | if (entry == NULL) |
5953 | return; | |
5954 | ||
97b6ea7b | 5955 | for_each_online_cpu(i) { |
e692ab53 | 5956 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5957 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5958 | entry->mode = 0555; |
e692ab53 | 5959 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5960 | entry++; |
e692ab53 | 5961 | } |
7378547f MM |
5962 | |
5963 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5964 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5965 | } | |
6382bc90 | 5966 | |
7378547f | 5967 | /* may be called multiple times per register */ |
6382bc90 MM |
5968 | static void unregister_sched_domain_sysctl(void) |
5969 | { | |
7378547f MM |
5970 | if (sd_sysctl_header) |
5971 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5972 | sd_sysctl_header = NULL; |
7378547f MM |
5973 | if (sd_ctl_dir[0].child) |
5974 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5975 | } |
e692ab53 | 5976 | #else |
6382bc90 MM |
5977 | static void register_sched_domain_sysctl(void) |
5978 | { | |
5979 | } | |
5980 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5981 | { |
5982 | } | |
5983 | #endif | |
5984 | ||
1da177e4 LT |
5985 | /* |
5986 | * migration_call - callback that gets triggered when a CPU is added. | |
5987 | * Here we can start up the necessary migration thread for the new CPU. | |
5988 | */ | |
48f24c4d IM |
5989 | static int __cpuinit |
5990 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5991 | { |
1da177e4 | 5992 | struct task_struct *p; |
48f24c4d | 5993 | int cpu = (long)hcpu; |
1da177e4 | 5994 | unsigned long flags; |
70b97a7f | 5995 | struct rq *rq; |
1da177e4 LT |
5996 | |
5997 | switch (action) { | |
5be9361c | 5998 | |
1da177e4 | 5999 | case CPU_UP_PREPARE: |
8bb78442 | 6000 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6001 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6002 | if (IS_ERR(p)) |
6003 | return NOTIFY_BAD; | |
1da177e4 LT |
6004 | kthread_bind(p, cpu); |
6005 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6006 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6007 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6008 | task_rq_unlock(rq, &flags); |
6009 | cpu_rq(cpu)->migration_thread = p; | |
6010 | break; | |
48f24c4d | 6011 | |
1da177e4 | 6012 | case CPU_ONLINE: |
8bb78442 | 6013 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6014 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6015 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6016 | |
6017 | /* Update our root-domain */ | |
6018 | rq = cpu_rq(cpu); | |
6019 | spin_lock_irqsave(&rq->lock, flags); | |
6020 | if (rq->rd) { | |
6021 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
6022 | cpu_set(cpu, rq->rd->online); | |
6023 | } | |
6024 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6025 | break; |
48f24c4d | 6026 | |
1da177e4 LT |
6027 | #ifdef CONFIG_HOTPLUG_CPU |
6028 | case CPU_UP_CANCELED: | |
8bb78442 | 6029 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6030 | if (!cpu_rq(cpu)->migration_thread) |
6031 | break; | |
41a2d6cf | 6032 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6033 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6034 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6035 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6036 | cpu_rq(cpu)->migration_thread = NULL; | |
6037 | break; | |
48f24c4d | 6038 | |
1da177e4 | 6039 | case CPU_DEAD: |
8bb78442 | 6040 | case CPU_DEAD_FROZEN: |
470fd646 | 6041 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6042 | migrate_live_tasks(cpu); |
6043 | rq = cpu_rq(cpu); | |
6044 | kthread_stop(rq->migration_thread); | |
6045 | rq->migration_thread = NULL; | |
6046 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6047 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6048 | update_rq_clock(rq); |
2e1cb74a | 6049 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6050 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6051 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6052 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6053 | migrate_dead_tasks(cpu); |
d2da272a | 6054 | spin_unlock_irq(&rq->lock); |
470fd646 | 6055 | cpuset_unlock(); |
1da177e4 LT |
6056 | migrate_nr_uninterruptible(rq); |
6057 | BUG_ON(rq->nr_running != 0); | |
6058 | ||
41a2d6cf IM |
6059 | /* |
6060 | * No need to migrate the tasks: it was best-effort if | |
6061 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6062 | * the requestors. | |
6063 | */ | |
1da177e4 LT |
6064 | spin_lock_irq(&rq->lock); |
6065 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6066 | struct migration_req *req; |
6067 | ||
1da177e4 | 6068 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6069 | struct migration_req, list); |
1da177e4 LT |
6070 | list_del_init(&req->list); |
6071 | complete(&req->done); | |
6072 | } | |
6073 | spin_unlock_irq(&rq->lock); | |
6074 | break; | |
57d885fe | 6075 | |
08f503b0 GH |
6076 | case CPU_DYING: |
6077 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6078 | /* Update our root-domain */ |
6079 | rq = cpu_rq(cpu); | |
6080 | spin_lock_irqsave(&rq->lock, flags); | |
6081 | if (rq->rd) { | |
6082 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
6083 | cpu_clear(cpu, rq->rd->online); | |
6084 | } | |
6085 | spin_unlock_irqrestore(&rq->lock, flags); | |
6086 | break; | |
1da177e4 LT |
6087 | #endif |
6088 | } | |
6089 | return NOTIFY_OK; | |
6090 | } | |
6091 | ||
6092 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6093 | * happens before everything else. | |
6094 | */ | |
26c2143b | 6095 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6096 | .notifier_call = migration_call, |
6097 | .priority = 10 | |
6098 | }; | |
6099 | ||
e6fe6649 | 6100 | void __init migration_init(void) |
1da177e4 LT |
6101 | { |
6102 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6103 | int err; |
48f24c4d IM |
6104 | |
6105 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6106 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6107 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6108 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6109 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
6110 | } |
6111 | #endif | |
6112 | ||
6113 | #ifdef CONFIG_SMP | |
476f3534 CL |
6114 | |
6115 | /* Number of possible processor ids */ | |
6116 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
6117 | EXPORT_SYMBOL(nr_cpu_ids); | |
6118 | ||
3e9830dc | 6119 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff IM |
6120 | |
6121 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level) | |
1da177e4 | 6122 | { |
4dcf6aff IM |
6123 | struct sched_group *group = sd->groups; |
6124 | cpumask_t groupmask; | |
434d53b0 | 6125 | char str[256]; |
1da177e4 | 6126 | |
434d53b0 | 6127 | cpulist_scnprintf(str, sizeof(str), sd->span); |
4dcf6aff IM |
6128 | cpus_clear(groupmask); |
6129 | ||
6130 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6131 | ||
6132 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6133 | printk("does not load-balance\n"); | |
6134 | if (sd->parent) | |
6135 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6136 | " has parent"); | |
6137 | return -1; | |
41c7ce9a NP |
6138 | } |
6139 | ||
4dcf6aff IM |
6140 | printk(KERN_CONT "span %s\n", str); |
6141 | ||
6142 | if (!cpu_isset(cpu, sd->span)) { | |
6143 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6144 | "CPU%d\n", cpu); | |
6145 | } | |
6146 | if (!cpu_isset(cpu, group->cpumask)) { | |
6147 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6148 | " CPU%d\n", cpu); | |
6149 | } | |
1da177e4 | 6150 | |
4dcf6aff | 6151 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6152 | do { |
4dcf6aff IM |
6153 | if (!group) { |
6154 | printk("\n"); | |
6155 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6156 | break; |
6157 | } | |
6158 | ||
4dcf6aff IM |
6159 | if (!group->__cpu_power) { |
6160 | printk(KERN_CONT "\n"); | |
6161 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6162 | "set\n"); | |
6163 | break; | |
6164 | } | |
1da177e4 | 6165 | |
4dcf6aff IM |
6166 | if (!cpus_weight(group->cpumask)) { |
6167 | printk(KERN_CONT "\n"); | |
6168 | printk(KERN_ERR "ERROR: empty group\n"); | |
6169 | break; | |
6170 | } | |
1da177e4 | 6171 | |
4dcf6aff IM |
6172 | if (cpus_intersects(groupmask, group->cpumask)) { |
6173 | printk(KERN_CONT "\n"); | |
6174 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6175 | break; | |
6176 | } | |
1da177e4 | 6177 | |
4dcf6aff | 6178 | cpus_or(groupmask, groupmask, group->cpumask); |
1da177e4 | 6179 | |
434d53b0 | 6180 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6181 | printk(KERN_CONT " %s", str); |
1da177e4 | 6182 | |
4dcf6aff IM |
6183 | group = group->next; |
6184 | } while (group != sd->groups); | |
6185 | printk(KERN_CONT "\n"); | |
1da177e4 | 6186 | |
4dcf6aff IM |
6187 | if (!cpus_equal(sd->span, groupmask)) |
6188 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); | |
1da177e4 | 6189 | |
4dcf6aff IM |
6190 | if (sd->parent && !cpus_subset(groupmask, sd->parent->span)) |
6191 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
6192 | "of domain->span\n"); | |
6193 | return 0; | |
6194 | } | |
1da177e4 | 6195 | |
4dcf6aff IM |
6196 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6197 | { | |
6198 | int level = 0; | |
1da177e4 | 6199 | |
4dcf6aff IM |
6200 | if (!sd) { |
6201 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6202 | return; | |
6203 | } | |
1da177e4 | 6204 | |
4dcf6aff IM |
6205 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6206 | ||
6207 | for (;;) { | |
6208 | if (sched_domain_debug_one(sd, cpu, level)) | |
6209 | break; | |
1da177e4 LT |
6210 | level++; |
6211 | sd = sd->parent; | |
33859f7f | 6212 | if (!sd) |
4dcf6aff IM |
6213 | break; |
6214 | } | |
1da177e4 LT |
6215 | } |
6216 | #else | |
48f24c4d | 6217 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
6218 | #endif |
6219 | ||
1a20ff27 | 6220 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6221 | { |
6222 | if (cpus_weight(sd->span) == 1) | |
6223 | return 1; | |
6224 | ||
6225 | /* Following flags need at least 2 groups */ | |
6226 | if (sd->flags & (SD_LOAD_BALANCE | | |
6227 | SD_BALANCE_NEWIDLE | | |
6228 | SD_BALANCE_FORK | | |
89c4710e SS |
6229 | SD_BALANCE_EXEC | |
6230 | SD_SHARE_CPUPOWER | | |
6231 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6232 | if (sd->groups != sd->groups->next) |
6233 | return 0; | |
6234 | } | |
6235 | ||
6236 | /* Following flags don't use groups */ | |
6237 | if (sd->flags & (SD_WAKE_IDLE | | |
6238 | SD_WAKE_AFFINE | | |
6239 | SD_WAKE_BALANCE)) | |
6240 | return 0; | |
6241 | ||
6242 | return 1; | |
6243 | } | |
6244 | ||
48f24c4d IM |
6245 | static int |
6246 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6247 | { |
6248 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6249 | ||
6250 | if (sd_degenerate(parent)) | |
6251 | return 1; | |
6252 | ||
6253 | if (!cpus_equal(sd->span, parent->span)) | |
6254 | return 0; | |
6255 | ||
6256 | /* Does parent contain flags not in child? */ | |
6257 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6258 | if (cflags & SD_WAKE_AFFINE) | |
6259 | pflags &= ~SD_WAKE_BALANCE; | |
6260 | /* Flags needing groups don't count if only 1 group in parent */ | |
6261 | if (parent->groups == parent->groups->next) { | |
6262 | pflags &= ~(SD_LOAD_BALANCE | | |
6263 | SD_BALANCE_NEWIDLE | | |
6264 | SD_BALANCE_FORK | | |
89c4710e SS |
6265 | SD_BALANCE_EXEC | |
6266 | SD_SHARE_CPUPOWER | | |
6267 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6268 | } |
6269 | if (~cflags & pflags) | |
6270 | return 0; | |
6271 | ||
6272 | return 1; | |
6273 | } | |
6274 | ||
57d885fe GH |
6275 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6276 | { | |
6277 | unsigned long flags; | |
6278 | const struct sched_class *class; | |
6279 | ||
6280 | spin_lock_irqsave(&rq->lock, flags); | |
6281 | ||
6282 | if (rq->rd) { | |
6283 | struct root_domain *old_rd = rq->rd; | |
6284 | ||
0eab9146 | 6285 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6286 | if (class->leave_domain) |
6287 | class->leave_domain(rq); | |
0eab9146 | 6288 | } |
57d885fe | 6289 | |
dc938520 GH |
6290 | cpu_clear(rq->cpu, old_rd->span); |
6291 | cpu_clear(rq->cpu, old_rd->online); | |
6292 | ||
57d885fe GH |
6293 | if (atomic_dec_and_test(&old_rd->refcount)) |
6294 | kfree(old_rd); | |
6295 | } | |
6296 | ||
6297 | atomic_inc(&rd->refcount); | |
6298 | rq->rd = rd; | |
6299 | ||
dc938520 | 6300 | cpu_set(rq->cpu, rd->span); |
1f94ef59 GH |
6301 | if (cpu_isset(rq->cpu, cpu_online_map)) |
6302 | cpu_set(rq->cpu, rd->online); | |
dc938520 | 6303 | |
0eab9146 | 6304 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6305 | if (class->join_domain) |
6306 | class->join_domain(rq); | |
0eab9146 | 6307 | } |
57d885fe GH |
6308 | |
6309 | spin_unlock_irqrestore(&rq->lock, flags); | |
6310 | } | |
6311 | ||
dc938520 | 6312 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6313 | { |
6314 | memset(rd, 0, sizeof(*rd)); | |
6315 | ||
dc938520 GH |
6316 | cpus_clear(rd->span); |
6317 | cpus_clear(rd->online); | |
57d885fe GH |
6318 | } |
6319 | ||
6320 | static void init_defrootdomain(void) | |
6321 | { | |
dc938520 | 6322 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6323 | atomic_set(&def_root_domain.refcount, 1); |
6324 | } | |
6325 | ||
dc938520 | 6326 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6327 | { |
6328 | struct root_domain *rd; | |
6329 | ||
6330 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6331 | if (!rd) | |
6332 | return NULL; | |
6333 | ||
dc938520 | 6334 | init_rootdomain(rd); |
57d885fe GH |
6335 | |
6336 | return rd; | |
6337 | } | |
6338 | ||
1da177e4 | 6339 | /* |
0eab9146 | 6340 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6341 | * hold the hotplug lock. |
6342 | */ | |
0eab9146 IM |
6343 | static void |
6344 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6345 | { |
70b97a7f | 6346 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6347 | struct sched_domain *tmp; |
6348 | ||
6349 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6350 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6351 | struct sched_domain *parent = tmp->parent; | |
6352 | if (!parent) | |
6353 | break; | |
1a848870 | 6354 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6355 | tmp->parent = parent->parent; |
1a848870 SS |
6356 | if (parent->parent) |
6357 | parent->parent->child = tmp; | |
6358 | } | |
245af2c7 SS |
6359 | } |
6360 | ||
1a848870 | 6361 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6362 | sd = sd->parent; |
1a848870 SS |
6363 | if (sd) |
6364 | sd->child = NULL; | |
6365 | } | |
1da177e4 LT |
6366 | |
6367 | sched_domain_debug(sd, cpu); | |
6368 | ||
57d885fe | 6369 | rq_attach_root(rq, rd); |
674311d5 | 6370 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6371 | } |
6372 | ||
6373 | /* cpus with isolated domains */ | |
67af63a6 | 6374 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6375 | |
6376 | /* Setup the mask of cpus configured for isolated domains */ | |
6377 | static int __init isolated_cpu_setup(char *str) | |
6378 | { | |
6379 | int ints[NR_CPUS], i; | |
6380 | ||
6381 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6382 | cpus_clear(cpu_isolated_map); | |
6383 | for (i = 1; i <= ints[0]; i++) | |
6384 | if (ints[i] < NR_CPUS) | |
6385 | cpu_set(ints[i], cpu_isolated_map); | |
6386 | return 1; | |
6387 | } | |
6388 | ||
8927f494 | 6389 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6390 | |
6391 | /* | |
6711cab4 SS |
6392 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6393 | * to a function which identifies what group(along with sched group) a CPU | |
6394 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6395 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6396 | * |
6397 | * init_sched_build_groups will build a circular linked list of the groups | |
6398 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6399 | * and ->cpu_power to 0. | |
6400 | */ | |
a616058b | 6401 | static void |
6711cab4 SS |
6402 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
6403 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
6404 | struct sched_group **sg)) | |
1da177e4 LT |
6405 | { |
6406 | struct sched_group *first = NULL, *last = NULL; | |
6407 | cpumask_t covered = CPU_MASK_NONE; | |
6408 | int i; | |
6409 | ||
6410 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
6411 | struct sched_group *sg; |
6412 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
6413 | int j; |
6414 | ||
6415 | if (cpu_isset(i, covered)) | |
6416 | continue; | |
6417 | ||
6418 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 6419 | sg->__cpu_power = 0; |
1da177e4 LT |
6420 | |
6421 | for_each_cpu_mask(j, span) { | |
6711cab4 | 6422 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
6423 | continue; |
6424 | ||
6425 | cpu_set(j, covered); | |
6426 | cpu_set(j, sg->cpumask); | |
6427 | } | |
6428 | if (!first) | |
6429 | first = sg; | |
6430 | if (last) | |
6431 | last->next = sg; | |
6432 | last = sg; | |
6433 | } | |
6434 | last->next = first; | |
6435 | } | |
6436 | ||
9c1cfda2 | 6437 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6438 | |
9c1cfda2 | 6439 | #ifdef CONFIG_NUMA |
198e2f18 | 6440 | |
9c1cfda2 JH |
6441 | /** |
6442 | * find_next_best_node - find the next node to include in a sched_domain | |
6443 | * @node: node whose sched_domain we're building | |
6444 | * @used_nodes: nodes already in the sched_domain | |
6445 | * | |
41a2d6cf | 6446 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6447 | * finds the closest node not already in the @used_nodes map. |
6448 | * | |
6449 | * Should use nodemask_t. | |
6450 | */ | |
6451 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
6452 | { | |
6453 | int i, n, val, min_val, best_node = 0; | |
6454 | ||
6455 | min_val = INT_MAX; | |
6456 | ||
6457 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6458 | /* Start at @node */ | |
6459 | n = (node + i) % MAX_NUMNODES; | |
6460 | ||
6461 | if (!nr_cpus_node(n)) | |
6462 | continue; | |
6463 | ||
6464 | /* Skip already used nodes */ | |
6465 | if (test_bit(n, used_nodes)) | |
6466 | continue; | |
6467 | ||
6468 | /* Simple min distance search */ | |
6469 | val = node_distance(node, n); | |
6470 | ||
6471 | if (val < min_val) { | |
6472 | min_val = val; | |
6473 | best_node = n; | |
6474 | } | |
6475 | } | |
6476 | ||
6477 | set_bit(best_node, used_nodes); | |
6478 | return best_node; | |
6479 | } | |
6480 | ||
6481 | /** | |
6482 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6483 | * @node: node whose cpumask we're constructing | |
6484 | * @size: number of nodes to include in this span | |
6485 | * | |
41a2d6cf | 6486 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6487 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6488 | * out optimally. | |
6489 | */ | |
6490 | static cpumask_t sched_domain_node_span(int node) | |
6491 | { | |
9c1cfda2 | 6492 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
6493 | cpumask_t span, nodemask; |
6494 | int i; | |
9c1cfda2 JH |
6495 | |
6496 | cpus_clear(span); | |
6497 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
6498 | ||
6499 | nodemask = node_to_cpumask(node); | |
6500 | cpus_or(span, span, nodemask); | |
6501 | set_bit(node, used_nodes); | |
6502 | ||
6503 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
6504 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 6505 | |
9c1cfda2 JH |
6506 | nodemask = node_to_cpumask(next_node); |
6507 | cpus_or(span, span, nodemask); | |
6508 | } | |
6509 | ||
6510 | return span; | |
6511 | } | |
6512 | #endif | |
6513 | ||
5c45bf27 | 6514 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6515 | |
9c1cfda2 | 6516 | /* |
48f24c4d | 6517 | * SMT sched-domains: |
9c1cfda2 | 6518 | */ |
1da177e4 LT |
6519 | #ifdef CONFIG_SCHED_SMT |
6520 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6521 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6522 | |
41a2d6cf IM |
6523 | static int |
6524 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6525 | { |
6711cab4 SS |
6526 | if (sg) |
6527 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6528 | return cpu; |
6529 | } | |
6530 | #endif | |
6531 | ||
48f24c4d IM |
6532 | /* |
6533 | * multi-core sched-domains: | |
6534 | */ | |
1e9f28fa SS |
6535 | #ifdef CONFIG_SCHED_MC |
6536 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6537 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
6538 | #endif |
6539 | ||
6540 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf IM |
6541 | static int |
6542 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6543 | { |
6711cab4 | 6544 | int group; |
d5a7430d | 6545 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6546 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
6547 | group = first_cpu(mask); |
6548 | if (sg) | |
6549 | *sg = &per_cpu(sched_group_core, group); | |
6550 | return group; | |
1e9f28fa SS |
6551 | } |
6552 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf IM |
6553 | static int |
6554 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6555 | { |
6711cab4 SS |
6556 | if (sg) |
6557 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6558 | return cpu; |
6559 | } | |
6560 | #endif | |
6561 | ||
1da177e4 | 6562 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6563 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6564 | |
41a2d6cf IM |
6565 | static int |
6566 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6567 | { |
6711cab4 | 6568 | int group; |
48f24c4d | 6569 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 6570 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 6571 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6572 | group = first_cpu(mask); |
1e9f28fa | 6573 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 6574 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6575 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6576 | group = first_cpu(mask); |
1da177e4 | 6577 | #else |
6711cab4 | 6578 | group = cpu; |
1da177e4 | 6579 | #endif |
6711cab4 SS |
6580 | if (sg) |
6581 | *sg = &per_cpu(sched_group_phys, group); | |
6582 | return group; | |
1da177e4 LT |
6583 | } |
6584 | ||
6585 | #ifdef CONFIG_NUMA | |
1da177e4 | 6586 | /* |
9c1cfda2 JH |
6587 | * The init_sched_build_groups can't handle what we want to do with node |
6588 | * groups, so roll our own. Now each node has its own list of groups which | |
6589 | * gets dynamically allocated. | |
1da177e4 | 6590 | */ |
9c1cfda2 | 6591 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 6592 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6593 | |
9c1cfda2 | 6594 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6595 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6596 | |
6711cab4 SS |
6597 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
6598 | struct sched_group **sg) | |
9c1cfda2 | 6599 | { |
6711cab4 SS |
6600 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6601 | int group; | |
6602 | ||
6603 | cpus_and(nodemask, nodemask, *cpu_map); | |
6604 | group = first_cpu(nodemask); | |
6605 | ||
6606 | if (sg) | |
6607 | *sg = &per_cpu(sched_group_allnodes, group); | |
6608 | return group; | |
1da177e4 | 6609 | } |
6711cab4 | 6610 | |
08069033 SS |
6611 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6612 | { | |
6613 | struct sched_group *sg = group_head; | |
6614 | int j; | |
6615 | ||
6616 | if (!sg) | |
6617 | return; | |
3a5c359a AK |
6618 | do { |
6619 | for_each_cpu_mask(j, sg->cpumask) { | |
6620 | struct sched_domain *sd; | |
08069033 | 6621 | |
3a5c359a AK |
6622 | sd = &per_cpu(phys_domains, j); |
6623 | if (j != first_cpu(sd->groups->cpumask)) { | |
6624 | /* | |
6625 | * Only add "power" once for each | |
6626 | * physical package. | |
6627 | */ | |
6628 | continue; | |
6629 | } | |
08069033 | 6630 | |
3a5c359a AK |
6631 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6632 | } | |
6633 | sg = sg->next; | |
6634 | } while (sg != group_head); | |
08069033 | 6635 | } |
1da177e4 LT |
6636 | #endif |
6637 | ||
a616058b | 6638 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6639 | /* Free memory allocated for various sched_group structures */ |
6640 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6641 | { | |
a616058b | 6642 | int cpu, i; |
51888ca2 SV |
6643 | |
6644 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6645 | struct sched_group **sched_group_nodes |
6646 | = sched_group_nodes_bycpu[cpu]; | |
6647 | ||
51888ca2 SV |
6648 | if (!sched_group_nodes) |
6649 | continue; | |
6650 | ||
6651 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6652 | cpumask_t nodemask = node_to_cpumask(i); | |
6653 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6654 | ||
6655 | cpus_and(nodemask, nodemask, *cpu_map); | |
6656 | if (cpus_empty(nodemask)) | |
6657 | continue; | |
6658 | ||
6659 | if (sg == NULL) | |
6660 | continue; | |
6661 | sg = sg->next; | |
6662 | next_sg: | |
6663 | oldsg = sg; | |
6664 | sg = sg->next; | |
6665 | kfree(oldsg); | |
6666 | if (oldsg != sched_group_nodes[i]) | |
6667 | goto next_sg; | |
6668 | } | |
6669 | kfree(sched_group_nodes); | |
6670 | sched_group_nodes_bycpu[cpu] = NULL; | |
6671 | } | |
51888ca2 | 6672 | } |
a616058b SS |
6673 | #else |
6674 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6675 | { | |
6676 | } | |
6677 | #endif | |
51888ca2 | 6678 | |
89c4710e SS |
6679 | /* |
6680 | * Initialize sched groups cpu_power. | |
6681 | * | |
6682 | * cpu_power indicates the capacity of sched group, which is used while | |
6683 | * distributing the load between different sched groups in a sched domain. | |
6684 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6685 | * there are asymmetries in the topology. If there are asymmetries, group | |
6686 | * having more cpu_power will pickup more load compared to the group having | |
6687 | * less cpu_power. | |
6688 | * | |
6689 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6690 | * the maximum number of tasks a group can handle in the presence of other idle | |
6691 | * or lightly loaded groups in the same sched domain. | |
6692 | */ | |
6693 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6694 | { | |
6695 | struct sched_domain *child; | |
6696 | struct sched_group *group; | |
6697 | ||
6698 | WARN_ON(!sd || !sd->groups); | |
6699 | ||
6700 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6701 | return; | |
6702 | ||
6703 | child = sd->child; | |
6704 | ||
5517d86b ED |
6705 | sd->groups->__cpu_power = 0; |
6706 | ||
89c4710e SS |
6707 | /* |
6708 | * For perf policy, if the groups in child domain share resources | |
6709 | * (for example cores sharing some portions of the cache hierarchy | |
6710 | * or SMT), then set this domain groups cpu_power such that each group | |
6711 | * can handle only one task, when there are other idle groups in the | |
6712 | * same sched domain. | |
6713 | */ | |
6714 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6715 | (child->flags & | |
6716 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6717 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6718 | return; |
6719 | } | |
6720 | ||
89c4710e SS |
6721 | /* |
6722 | * add cpu_power of each child group to this groups cpu_power | |
6723 | */ | |
6724 | group = child->groups; | |
6725 | do { | |
5517d86b | 6726 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6727 | group = group->next; |
6728 | } while (group != child->groups); | |
6729 | } | |
6730 | ||
1da177e4 | 6731 | /* |
1a20ff27 DG |
6732 | * Build sched domains for a given set of cpus and attach the sched domains |
6733 | * to the individual cpus | |
1da177e4 | 6734 | */ |
51888ca2 | 6735 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6736 | { |
6737 | int i; | |
57d885fe | 6738 | struct root_domain *rd; |
d1b55138 JH |
6739 | #ifdef CONFIG_NUMA |
6740 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6741 | int sd_allnodes = 0; |
d1b55138 JH |
6742 | |
6743 | /* | |
6744 | * Allocate the per-node list of sched groups | |
6745 | */ | |
5cf9f062 | 6746 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 6747 | GFP_KERNEL); |
d1b55138 JH |
6748 | if (!sched_group_nodes) { |
6749 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6750 | return -ENOMEM; |
d1b55138 JH |
6751 | } |
6752 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6753 | #endif | |
1da177e4 | 6754 | |
dc938520 | 6755 | rd = alloc_rootdomain(); |
57d885fe GH |
6756 | if (!rd) { |
6757 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
6758 | return -ENOMEM; | |
6759 | } | |
6760 | ||
1da177e4 | 6761 | /* |
1a20ff27 | 6762 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6763 | */ |
1a20ff27 | 6764 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6765 | struct sched_domain *sd = NULL, *p; |
6766 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6767 | ||
1a20ff27 | 6768 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6769 | |
6770 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6771 | if (cpus_weight(*cpu_map) > |
6772 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6773 | sd = &per_cpu(allnodes_domains, i); |
6774 | *sd = SD_ALLNODES_INIT; | |
6775 | sd->span = *cpu_map; | |
6711cab4 | 6776 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6777 | p = sd; |
6711cab4 | 6778 | sd_allnodes = 1; |
9c1cfda2 JH |
6779 | } else |
6780 | p = NULL; | |
6781 | ||
1da177e4 | 6782 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6783 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6784 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6785 | sd->parent = p; | |
1a848870 SS |
6786 | if (p) |
6787 | p->child = sd; | |
9c1cfda2 | 6788 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6789 | #endif |
6790 | ||
6791 | p = sd; | |
6792 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6793 | *sd = SD_CPU_INIT; |
6794 | sd->span = nodemask; | |
6795 | sd->parent = p; | |
1a848870 SS |
6796 | if (p) |
6797 | p->child = sd; | |
6711cab4 | 6798 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6799 | |
1e9f28fa SS |
6800 | #ifdef CONFIG_SCHED_MC |
6801 | p = sd; | |
6802 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6803 | *sd = SD_MC_INIT; |
6804 | sd->span = cpu_coregroup_map(i); | |
6805 | cpus_and(sd->span, sd->span, *cpu_map); | |
6806 | sd->parent = p; | |
1a848870 | 6807 | p->child = sd; |
6711cab4 | 6808 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6809 | #endif |
6810 | ||
1da177e4 LT |
6811 | #ifdef CONFIG_SCHED_SMT |
6812 | p = sd; | |
6813 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6814 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6815 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6816 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6817 | sd->parent = p; |
1a848870 | 6818 | p->child = sd; |
6711cab4 | 6819 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6820 | #endif |
6821 | } | |
6822 | ||
6823 | #ifdef CONFIG_SCHED_SMT | |
6824 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6825 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6826 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6827 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6828 | if (i != first_cpu(this_sibling_map)) |
6829 | continue; | |
6830 | ||
dd41f596 IM |
6831 | init_sched_build_groups(this_sibling_map, cpu_map, |
6832 | &cpu_to_cpu_group); | |
1da177e4 LT |
6833 | } |
6834 | #endif | |
6835 | ||
1e9f28fa SS |
6836 | #ifdef CONFIG_SCHED_MC |
6837 | /* Set up multi-core groups */ | |
6838 | for_each_cpu_mask(i, *cpu_map) { | |
6839 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6840 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6841 | if (i != first_cpu(this_core_map)) | |
6842 | continue; | |
dd41f596 IM |
6843 | init_sched_build_groups(this_core_map, cpu_map, |
6844 | &cpu_to_core_group); | |
1e9f28fa SS |
6845 | } |
6846 | #endif | |
6847 | ||
1da177e4 LT |
6848 | /* Set up physical groups */ |
6849 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6850 | cpumask_t nodemask = node_to_cpumask(i); | |
6851 | ||
1a20ff27 | 6852 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6853 | if (cpus_empty(nodemask)) |
6854 | continue; | |
6855 | ||
6711cab4 | 6856 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6857 | } |
6858 | ||
6859 | #ifdef CONFIG_NUMA | |
6860 | /* Set up node groups */ | |
6711cab4 | 6861 | if (sd_allnodes) |
dd41f596 IM |
6862 | init_sched_build_groups(*cpu_map, cpu_map, |
6863 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6864 | |
6865 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6866 | /* Set up node groups */ | |
6867 | struct sched_group *sg, *prev; | |
6868 | cpumask_t nodemask = node_to_cpumask(i); | |
6869 | cpumask_t domainspan; | |
6870 | cpumask_t covered = CPU_MASK_NONE; | |
6871 | int j; | |
6872 | ||
6873 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6874 | if (cpus_empty(nodemask)) { |
6875 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6876 | continue; |
d1b55138 | 6877 | } |
9c1cfda2 JH |
6878 | |
6879 | domainspan = sched_domain_node_span(i); | |
6880 | cpus_and(domainspan, domainspan, *cpu_map); | |
6881 | ||
15f0b676 | 6882 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6883 | if (!sg) { |
6884 | printk(KERN_WARNING "Can not alloc domain group for " | |
6885 | "node %d\n", i); | |
6886 | goto error; | |
6887 | } | |
9c1cfda2 JH |
6888 | sched_group_nodes[i] = sg; |
6889 | for_each_cpu_mask(j, nodemask) { | |
6890 | struct sched_domain *sd; | |
9761eea8 | 6891 | |
9c1cfda2 JH |
6892 | sd = &per_cpu(node_domains, j); |
6893 | sd->groups = sg; | |
9c1cfda2 | 6894 | } |
5517d86b | 6895 | sg->__cpu_power = 0; |
9c1cfda2 | 6896 | sg->cpumask = nodemask; |
51888ca2 | 6897 | sg->next = sg; |
9c1cfda2 JH |
6898 | cpus_or(covered, covered, nodemask); |
6899 | prev = sg; | |
6900 | ||
6901 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6902 | cpumask_t tmp, notcovered; | |
6903 | int n = (i + j) % MAX_NUMNODES; | |
6904 | ||
6905 | cpus_complement(notcovered, covered); | |
6906 | cpus_and(tmp, notcovered, *cpu_map); | |
6907 | cpus_and(tmp, tmp, domainspan); | |
6908 | if (cpus_empty(tmp)) | |
6909 | break; | |
6910 | ||
6911 | nodemask = node_to_cpumask(n); | |
6912 | cpus_and(tmp, tmp, nodemask); | |
6913 | if (cpus_empty(tmp)) | |
6914 | continue; | |
6915 | ||
15f0b676 SV |
6916 | sg = kmalloc_node(sizeof(struct sched_group), |
6917 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6918 | if (!sg) { |
6919 | printk(KERN_WARNING | |
6920 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6921 | goto error; |
9c1cfda2 | 6922 | } |
5517d86b | 6923 | sg->__cpu_power = 0; |
9c1cfda2 | 6924 | sg->cpumask = tmp; |
51888ca2 | 6925 | sg->next = prev->next; |
9c1cfda2 JH |
6926 | cpus_or(covered, covered, tmp); |
6927 | prev->next = sg; | |
6928 | prev = sg; | |
6929 | } | |
9c1cfda2 | 6930 | } |
1da177e4 LT |
6931 | #endif |
6932 | ||
6933 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6934 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6935 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6936 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6937 | ||
89c4710e | 6938 | init_sched_groups_power(i, sd); |
5c45bf27 | 6939 | } |
1da177e4 | 6940 | #endif |
1e9f28fa | 6941 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6942 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6943 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6944 | ||
89c4710e | 6945 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6946 | } |
6947 | #endif | |
1e9f28fa | 6948 | |
5c45bf27 | 6949 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6950 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6951 | ||
89c4710e | 6952 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6953 | } |
6954 | ||
9c1cfda2 | 6955 | #ifdef CONFIG_NUMA |
08069033 SS |
6956 | for (i = 0; i < MAX_NUMNODES; i++) |
6957 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6958 | |
6711cab4 SS |
6959 | if (sd_allnodes) { |
6960 | struct sched_group *sg; | |
f712c0c7 | 6961 | |
6711cab4 | 6962 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6963 | init_numa_sched_groups_power(sg); |
6964 | } | |
9c1cfda2 JH |
6965 | #endif |
6966 | ||
1da177e4 | 6967 | /* Attach the domains */ |
1a20ff27 | 6968 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6969 | struct sched_domain *sd; |
6970 | #ifdef CONFIG_SCHED_SMT | |
6971 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6972 | #elif defined(CONFIG_SCHED_MC) |
6973 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6974 | #else |
6975 | sd = &per_cpu(phys_domains, i); | |
6976 | #endif | |
57d885fe | 6977 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 6978 | } |
51888ca2 SV |
6979 | |
6980 | return 0; | |
6981 | ||
a616058b | 6982 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6983 | error: |
6984 | free_sched_groups(cpu_map); | |
6985 | return -ENOMEM; | |
a616058b | 6986 | #endif |
1da177e4 | 6987 | } |
029190c5 PJ |
6988 | |
6989 | static cpumask_t *doms_cur; /* current sched domains */ | |
6990 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
6991 | ||
6992 | /* | |
6993 | * Special case: If a kmalloc of a doms_cur partition (array of | |
6994 | * cpumask_t) fails, then fallback to a single sched domain, | |
6995 | * as determined by the single cpumask_t fallback_doms. | |
6996 | */ | |
6997 | static cpumask_t fallback_doms; | |
6998 | ||
22e52b07 HC |
6999 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7000 | { | |
7001 | } | |
7002 | ||
1a20ff27 | 7003 | /* |
41a2d6cf | 7004 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7005 | * For now this just excludes isolated cpus, but could be used to |
7006 | * exclude other special cases in the future. | |
1a20ff27 | 7007 | */ |
51888ca2 | 7008 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7009 | { |
7378547f MM |
7010 | int err; |
7011 | ||
22e52b07 | 7012 | arch_update_cpu_topology(); |
029190c5 PJ |
7013 | ndoms_cur = 1; |
7014 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7015 | if (!doms_cur) | |
7016 | doms_cur = &fallback_doms; | |
7017 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
7378547f | 7018 | err = build_sched_domains(doms_cur); |
6382bc90 | 7019 | register_sched_domain_sysctl(); |
7378547f MM |
7020 | |
7021 | return err; | |
1a20ff27 DG |
7022 | } |
7023 | ||
7024 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 7025 | { |
51888ca2 | 7026 | free_sched_groups(cpu_map); |
9c1cfda2 | 7027 | } |
1da177e4 | 7028 | |
1a20ff27 DG |
7029 | /* |
7030 | * Detach sched domains from a group of cpus specified in cpu_map | |
7031 | * These cpus will now be attached to the NULL domain | |
7032 | */ | |
858119e1 | 7033 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
7034 | { |
7035 | int i; | |
7036 | ||
6382bc90 MM |
7037 | unregister_sched_domain_sysctl(); |
7038 | ||
1a20ff27 | 7039 | for_each_cpu_mask(i, *cpu_map) |
57d885fe | 7040 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 DG |
7041 | synchronize_sched(); |
7042 | arch_destroy_sched_domains(cpu_map); | |
7043 | } | |
7044 | ||
029190c5 PJ |
7045 | /* |
7046 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7047 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7048 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7049 | * It destroys each deleted domain and builds each new domain. | |
7050 | * | |
7051 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7052 | * The masks don't intersect (don't overlap.) We should setup one |
7053 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7054 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7055 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7056 | * it as it is. | |
7057 | * | |
41a2d6cf IM |
7058 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7059 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7060 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7061 | * and partition_sched_domains() will fallback to the single partition | |
7062 | * 'fallback_doms'. | |
7063 | * | |
7064 | * Call with hotplug lock held | |
7065 | */ | |
7066 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
7067 | { | |
7068 | int i, j; | |
7069 | ||
a1835615 SV |
7070 | lock_doms_cur(); |
7071 | ||
7378547f MM |
7072 | /* always unregister in case we don't destroy any domains */ |
7073 | unregister_sched_domain_sysctl(); | |
7074 | ||
029190c5 PJ |
7075 | if (doms_new == NULL) { |
7076 | ndoms_new = 1; | |
7077 | doms_new = &fallback_doms; | |
7078 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
7079 | } | |
7080 | ||
7081 | /* Destroy deleted domains */ | |
7082 | for (i = 0; i < ndoms_cur; i++) { | |
7083 | for (j = 0; j < ndoms_new; j++) { | |
7084 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
7085 | goto match1; | |
7086 | } | |
7087 | /* no match - a current sched domain not in new doms_new[] */ | |
7088 | detach_destroy_domains(doms_cur + i); | |
7089 | match1: | |
7090 | ; | |
7091 | } | |
7092 | ||
7093 | /* Build new domains */ | |
7094 | for (i = 0; i < ndoms_new; i++) { | |
7095 | for (j = 0; j < ndoms_cur; j++) { | |
7096 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
7097 | goto match2; | |
7098 | } | |
7099 | /* no match - add a new doms_new */ | |
7100 | build_sched_domains(doms_new + i); | |
7101 | match2: | |
7102 | ; | |
7103 | } | |
7104 | ||
7105 | /* Remember the new sched domains */ | |
7106 | if (doms_cur != &fallback_doms) | |
7107 | kfree(doms_cur); | |
7108 | doms_cur = doms_new; | |
7109 | ndoms_cur = ndoms_new; | |
7378547f MM |
7110 | |
7111 | register_sched_domain_sysctl(); | |
a1835615 SV |
7112 | |
7113 | unlock_doms_cur(); | |
029190c5 PJ |
7114 | } |
7115 | ||
5c45bf27 | 7116 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7117 | int arch_reinit_sched_domains(void) |
5c45bf27 SS |
7118 | { |
7119 | int err; | |
7120 | ||
95402b38 | 7121 | get_online_cpus(); |
5c45bf27 SS |
7122 | detach_destroy_domains(&cpu_online_map); |
7123 | err = arch_init_sched_domains(&cpu_online_map); | |
95402b38 | 7124 | put_online_cpus(); |
5c45bf27 SS |
7125 | |
7126 | return err; | |
7127 | } | |
7128 | ||
7129 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7130 | { | |
7131 | int ret; | |
7132 | ||
7133 | if (buf[0] != '0' && buf[0] != '1') | |
7134 | return -EINVAL; | |
7135 | ||
7136 | if (smt) | |
7137 | sched_smt_power_savings = (buf[0] == '1'); | |
7138 | else | |
7139 | sched_mc_power_savings = (buf[0] == '1'); | |
7140 | ||
7141 | ret = arch_reinit_sched_domains(); | |
7142 | ||
7143 | return ret ? ret : count; | |
7144 | } | |
7145 | ||
5c45bf27 SS |
7146 | #ifdef CONFIG_SCHED_MC |
7147 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
7148 | { | |
7149 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7150 | } | |
48f24c4d IM |
7151 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
7152 | const char *buf, size_t count) | |
5c45bf27 SS |
7153 | { |
7154 | return sched_power_savings_store(buf, count, 0); | |
7155 | } | |
6707de00 AB |
7156 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
7157 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7158 | #endif |
7159 | ||
7160 | #ifdef CONFIG_SCHED_SMT | |
7161 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
7162 | { | |
7163 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7164 | } | |
48f24c4d IM |
7165 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
7166 | const char *buf, size_t count) | |
5c45bf27 SS |
7167 | { |
7168 | return sched_power_savings_store(buf, count, 1); | |
7169 | } | |
6707de00 AB |
7170 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
7171 | sched_smt_power_savings_store); | |
7172 | #endif | |
7173 | ||
7174 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7175 | { | |
7176 | int err = 0; | |
7177 | ||
7178 | #ifdef CONFIG_SCHED_SMT | |
7179 | if (smt_capable()) | |
7180 | err = sysfs_create_file(&cls->kset.kobj, | |
7181 | &attr_sched_smt_power_savings.attr); | |
7182 | #endif | |
7183 | #ifdef CONFIG_SCHED_MC | |
7184 | if (!err && mc_capable()) | |
7185 | err = sysfs_create_file(&cls->kset.kobj, | |
7186 | &attr_sched_mc_power_savings.attr); | |
7187 | #endif | |
7188 | return err; | |
7189 | } | |
5c45bf27 SS |
7190 | #endif |
7191 | ||
1da177e4 | 7192 | /* |
41a2d6cf | 7193 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 7194 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 7195 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
7196 | * which will prevent rebalancing while the sched domains are recalculated. |
7197 | */ | |
7198 | static int update_sched_domains(struct notifier_block *nfb, | |
7199 | unsigned long action, void *hcpu) | |
7200 | { | |
1da177e4 LT |
7201 | switch (action) { |
7202 | case CPU_UP_PREPARE: | |
8bb78442 | 7203 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 7204 | case CPU_DOWN_PREPARE: |
8bb78442 | 7205 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 7206 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
7207 | return NOTIFY_OK; |
7208 | ||
7209 | case CPU_UP_CANCELED: | |
8bb78442 | 7210 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 7211 | case CPU_DOWN_FAILED: |
8bb78442 | 7212 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7213 | case CPU_ONLINE: |
8bb78442 | 7214 | case CPU_ONLINE_FROZEN: |
1da177e4 | 7215 | case CPU_DEAD: |
8bb78442 | 7216 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
7217 | /* |
7218 | * Fall through and re-initialise the domains. | |
7219 | */ | |
7220 | break; | |
7221 | default: | |
7222 | return NOTIFY_DONE; | |
7223 | } | |
7224 | ||
7225 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 7226 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
7227 | |
7228 | return NOTIFY_OK; | |
7229 | } | |
1da177e4 LT |
7230 | |
7231 | void __init sched_init_smp(void) | |
7232 | { | |
5c1e1767 NP |
7233 | cpumask_t non_isolated_cpus; |
7234 | ||
434d53b0 MT |
7235 | #if defined(CONFIG_NUMA) |
7236 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7237 | GFP_KERNEL); | |
7238 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7239 | #endif | |
95402b38 | 7240 | get_online_cpus(); |
1a20ff27 | 7241 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7242 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7243 | if (cpus_empty(non_isolated_cpus)) |
7244 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
95402b38 | 7245 | put_online_cpus(); |
1da177e4 LT |
7246 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7247 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
7248 | |
7249 | /* Move init over to a non-isolated CPU */ | |
7250 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
7251 | BUG(); | |
19978ca6 | 7252 | sched_init_granularity(); |
1da177e4 LT |
7253 | } |
7254 | #else | |
7255 | void __init sched_init_smp(void) | |
7256 | { | |
434d53b0 MT |
7257 | #if defined(CONFIG_NUMA) |
7258 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7259 | GFP_KERNEL); | |
7260 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7261 | #endif | |
19978ca6 | 7262 | sched_init_granularity(); |
1da177e4 LT |
7263 | } |
7264 | #endif /* CONFIG_SMP */ | |
7265 | ||
7266 | int in_sched_functions(unsigned long addr) | |
7267 | { | |
1da177e4 LT |
7268 | return in_lock_functions(addr) || |
7269 | (addr >= (unsigned long)__sched_text_start | |
7270 | && addr < (unsigned long)__sched_text_end); | |
7271 | } | |
7272 | ||
a9957449 | 7273 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7274 | { |
7275 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
7276 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7277 | cfs_rq->rq = rq; | |
7278 | #endif | |
67e9fb2a | 7279 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7280 | } |
7281 | ||
fa85ae24 PZ |
7282 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7283 | { | |
7284 | struct rt_prio_array *array; | |
7285 | int i; | |
7286 | ||
7287 | array = &rt_rq->active; | |
7288 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7289 | INIT_LIST_HEAD(array->queue + i); | |
7290 | __clear_bit(i, array->bitmap); | |
7291 | } | |
7292 | /* delimiter for bitsearch: */ | |
7293 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7294 | ||
052f1dc7 | 7295 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7296 | rt_rq->highest_prio = MAX_RT_PRIO; |
7297 | #endif | |
fa85ae24 PZ |
7298 | #ifdef CONFIG_SMP |
7299 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
7300 | rt_rq->overloaded = 0; |
7301 | #endif | |
7302 | ||
7303 | rt_rq->rt_time = 0; | |
7304 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
7305 | rt_rq->rt_runtime = 0; |
7306 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 7307 | |
052f1dc7 | 7308 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7309 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7310 | rt_rq->rq = rq; |
7311 | #endif | |
fa85ae24 PZ |
7312 | } |
7313 | ||
6f505b16 PZ |
7314 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7315 | static void init_tg_cfs_entry(struct rq *rq, struct task_group *tg, | |
7316 | struct cfs_rq *cfs_rq, struct sched_entity *se, | |
7317 | int cpu, int add) | |
7318 | { | |
7319 | tg->cfs_rq[cpu] = cfs_rq; | |
7320 | init_cfs_rq(cfs_rq, rq); | |
7321 | cfs_rq->tg = tg; | |
7322 | if (add) | |
7323 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7324 | ||
7325 | tg->se[cpu] = se; | |
7326 | se->cfs_rq = &rq->cfs; | |
7327 | se->my_q = cfs_rq; | |
7328 | se->load.weight = tg->shares; | |
7329 | se->load.inv_weight = div64_64(1ULL<<32, se->load.weight); | |
7330 | se->parent = NULL; | |
7331 | } | |
052f1dc7 | 7332 | #endif |
6f505b16 | 7333 | |
052f1dc7 | 7334 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
7335 | static void init_tg_rt_entry(struct rq *rq, struct task_group *tg, |
7336 | struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, | |
7337 | int cpu, int add) | |
7338 | { | |
7339 | tg->rt_rq[cpu] = rt_rq; | |
7340 | init_rt_rq(rt_rq, rq); | |
7341 | rt_rq->tg = tg; | |
7342 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 7343 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7344 | if (add) |
7345 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7346 | ||
7347 | tg->rt_se[cpu] = rt_se; | |
7348 | rt_se->rt_rq = &rq->rt; | |
7349 | rt_se->my_q = rt_rq; | |
7350 | rt_se->parent = NULL; | |
7351 | INIT_LIST_HEAD(&rt_se->run_list); | |
7352 | } | |
7353 | #endif | |
7354 | ||
1da177e4 LT |
7355 | void __init sched_init(void) |
7356 | { | |
476f3534 | 7357 | int highest_cpu = 0; |
dd41f596 | 7358 | int i, j; |
434d53b0 MT |
7359 | unsigned long alloc_size = 0, ptr; |
7360 | ||
7361 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7362 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7363 | #endif | |
7364 | #ifdef CONFIG_RT_GROUP_SCHED | |
7365 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7366 | #endif | |
7367 | /* | |
7368 | * As sched_init() is called before page_alloc is setup, | |
7369 | * we use alloc_bootmem(). | |
7370 | */ | |
7371 | if (alloc_size) { | |
7372 | ptr = (unsigned long)alloc_bootmem_low(alloc_size); | |
7373 | ||
7374 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7375 | init_task_group.se = (struct sched_entity **)ptr; | |
7376 | ptr += nr_cpu_ids * sizeof(void **); | |
7377 | ||
7378 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7379 | ptr += nr_cpu_ids * sizeof(void **); | |
7380 | #endif | |
7381 | #ifdef CONFIG_RT_GROUP_SCHED | |
7382 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7383 | ptr += nr_cpu_ids * sizeof(void **); | |
7384 | ||
7385 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
7386 | #endif | |
7387 | } | |
dd41f596 | 7388 | |
57d885fe GH |
7389 | #ifdef CONFIG_SMP |
7390 | init_defrootdomain(); | |
7391 | #endif | |
7392 | ||
d0b27fa7 PZ |
7393 | init_rt_bandwidth(&def_rt_bandwidth, |
7394 | global_rt_period(), global_rt_runtime()); | |
7395 | ||
7396 | #ifdef CONFIG_RT_GROUP_SCHED | |
7397 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7398 | global_rt_period(), global_rt_runtime()); | |
7399 | #endif | |
7400 | ||
052f1dc7 | 7401 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 PZ |
7402 | list_add(&init_task_group.list, &task_groups); |
7403 | #endif | |
7404 | ||
0a945022 | 7405 | for_each_possible_cpu(i) { |
70b97a7f | 7406 | struct rq *rq; |
1da177e4 LT |
7407 | |
7408 | rq = cpu_rq(i); | |
7409 | spin_lock_init(&rq->lock); | |
fcb99371 | 7410 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 7411 | rq->nr_running = 0; |
dd41f596 | 7412 | rq->clock = 1; |
15934a37 | 7413 | update_last_tick_seen(rq); |
dd41f596 | 7414 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7415 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7416 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7417 | init_task_group.shares = init_task_group_load; |
6f505b16 PZ |
7418 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
7419 | init_tg_cfs_entry(rq, &init_task_group, | |
7420 | &per_cpu(init_cfs_rq, i), | |
7421 | &per_cpu(init_sched_entity, i), i, 1); | |
7422 | ||
052f1dc7 PZ |
7423 | #endif |
7424 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 PZ |
7425 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
7426 | init_tg_rt_entry(rq, &init_task_group, | |
7427 | &per_cpu(init_rt_rq, i), | |
7428 | &per_cpu(init_sched_rt_entity, i), i, 1); | |
ac086bc2 PZ |
7429 | #else |
7430 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
dd41f596 | 7431 | #endif |
1da177e4 | 7432 | |
dd41f596 IM |
7433 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7434 | rq->cpu_load[j] = 0; | |
1da177e4 | 7435 | #ifdef CONFIG_SMP |
41c7ce9a | 7436 | rq->sd = NULL; |
57d885fe | 7437 | rq->rd = NULL; |
1da177e4 | 7438 | rq->active_balance = 0; |
dd41f596 | 7439 | rq->next_balance = jiffies; |
1da177e4 | 7440 | rq->push_cpu = 0; |
0a2966b4 | 7441 | rq->cpu = i; |
1da177e4 LT |
7442 | rq->migration_thread = NULL; |
7443 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 7444 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7445 | #endif |
8f4d37ec | 7446 | init_rq_hrtick(rq); |
1da177e4 | 7447 | atomic_set(&rq->nr_iowait, 0); |
476f3534 | 7448 | highest_cpu = i; |
1da177e4 LT |
7449 | } |
7450 | ||
2dd73a4f | 7451 | set_load_weight(&init_task); |
b50f60ce | 7452 | |
e107be36 AK |
7453 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7454 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7455 | #endif | |
7456 | ||
c9819f45 | 7457 | #ifdef CONFIG_SMP |
476f3534 | 7458 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
7459 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
7460 | #endif | |
7461 | ||
b50f60ce HC |
7462 | #ifdef CONFIG_RT_MUTEXES |
7463 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
7464 | #endif | |
7465 | ||
1da177e4 LT |
7466 | /* |
7467 | * The boot idle thread does lazy MMU switching as well: | |
7468 | */ | |
7469 | atomic_inc(&init_mm.mm_count); | |
7470 | enter_lazy_tlb(&init_mm, current); | |
7471 | ||
7472 | /* | |
7473 | * Make us the idle thread. Technically, schedule() should not be | |
7474 | * called from this thread, however somewhere below it might be, | |
7475 | * but because we are the idle thread, we just pick up running again | |
7476 | * when this runqueue becomes "idle". | |
7477 | */ | |
7478 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
7479 | /* |
7480 | * During early bootup we pretend to be a normal task: | |
7481 | */ | |
7482 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
7483 | |
7484 | scheduler_running = 1; | |
1da177e4 LT |
7485 | } |
7486 | ||
7487 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
7488 | void __might_sleep(char *file, int line) | |
7489 | { | |
48f24c4d | 7490 | #ifdef in_atomic |
1da177e4 LT |
7491 | static unsigned long prev_jiffy; /* ratelimiting */ |
7492 | ||
7493 | if ((in_atomic() || irqs_disabled()) && | |
7494 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
7495 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7496 | return; | |
7497 | prev_jiffy = jiffies; | |
91368d73 | 7498 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
7499 | " context at %s:%d\n", file, line); |
7500 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
7501 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 7502 | debug_show_held_locks(current); |
3117df04 IM |
7503 | if (irqs_disabled()) |
7504 | print_irqtrace_events(current); | |
1da177e4 LT |
7505 | dump_stack(); |
7506 | } | |
7507 | #endif | |
7508 | } | |
7509 | EXPORT_SYMBOL(__might_sleep); | |
7510 | #endif | |
7511 | ||
7512 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7513 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7514 | { | |
7515 | int on_rq; | |
7516 | update_rq_clock(rq); | |
7517 | on_rq = p->se.on_rq; | |
7518 | if (on_rq) | |
7519 | deactivate_task(rq, p, 0); | |
7520 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7521 | if (on_rq) { | |
7522 | activate_task(rq, p, 0); | |
7523 | resched_task(rq->curr); | |
7524 | } | |
7525 | } | |
7526 | ||
1da177e4 LT |
7527 | void normalize_rt_tasks(void) |
7528 | { | |
a0f98a1c | 7529 | struct task_struct *g, *p; |
1da177e4 | 7530 | unsigned long flags; |
70b97a7f | 7531 | struct rq *rq; |
1da177e4 | 7532 | |
4cf5d77a | 7533 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7534 | do_each_thread(g, p) { |
178be793 IM |
7535 | /* |
7536 | * Only normalize user tasks: | |
7537 | */ | |
7538 | if (!p->mm) | |
7539 | continue; | |
7540 | ||
6cfb0d5d | 7541 | p->se.exec_start = 0; |
6cfb0d5d | 7542 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 7543 | p->se.wait_start = 0; |
dd41f596 | 7544 | p->se.sleep_start = 0; |
dd41f596 | 7545 | p->se.block_start = 0; |
6cfb0d5d | 7546 | #endif |
dd41f596 IM |
7547 | task_rq(p)->clock = 0; |
7548 | ||
7549 | if (!rt_task(p)) { | |
7550 | /* | |
7551 | * Renice negative nice level userspace | |
7552 | * tasks back to 0: | |
7553 | */ | |
7554 | if (TASK_NICE(p) < 0 && p->mm) | |
7555 | set_user_nice(p, 0); | |
1da177e4 | 7556 | continue; |
dd41f596 | 7557 | } |
1da177e4 | 7558 | |
4cf5d77a | 7559 | spin_lock(&p->pi_lock); |
b29739f9 | 7560 | rq = __task_rq_lock(p); |
1da177e4 | 7561 | |
178be793 | 7562 | normalize_task(rq, p); |
3a5e4dc1 | 7563 | |
b29739f9 | 7564 | __task_rq_unlock(rq); |
4cf5d77a | 7565 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7566 | } while_each_thread(g, p); |
7567 | ||
4cf5d77a | 7568 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7569 | } |
7570 | ||
7571 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7572 | |
7573 | #ifdef CONFIG_IA64 | |
7574 | /* | |
7575 | * These functions are only useful for the IA64 MCA handling. | |
7576 | * | |
7577 | * They can only be called when the whole system has been | |
7578 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7579 | * activity can take place. Using them for anything else would | |
7580 | * be a serious bug, and as a result, they aren't even visible | |
7581 | * under any other configuration. | |
7582 | */ | |
7583 | ||
7584 | /** | |
7585 | * curr_task - return the current task for a given cpu. | |
7586 | * @cpu: the processor in question. | |
7587 | * | |
7588 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7589 | */ | |
36c8b586 | 7590 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7591 | { |
7592 | return cpu_curr(cpu); | |
7593 | } | |
7594 | ||
7595 | /** | |
7596 | * set_curr_task - set the current task for a given cpu. | |
7597 | * @cpu: the processor in question. | |
7598 | * @p: the task pointer to set. | |
7599 | * | |
7600 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7601 | * are serviced on a separate stack. It allows the architecture to switch the |
7602 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7603 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7604 | * and caller must save the original value of the current task (see | |
7605 | * curr_task() above) and restore that value before reenabling interrupts and | |
7606 | * re-starting the system. | |
7607 | * | |
7608 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7609 | */ | |
36c8b586 | 7610 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7611 | { |
7612 | cpu_curr(cpu) = p; | |
7613 | } | |
7614 | ||
7615 | #endif | |
29f59db3 | 7616 | |
bccbe08a PZ |
7617 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7618 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
7619 | { |
7620 | int i; | |
7621 | ||
7622 | for_each_possible_cpu(i) { | |
7623 | if (tg->cfs_rq) | |
7624 | kfree(tg->cfs_rq[i]); | |
7625 | if (tg->se) | |
7626 | kfree(tg->se[i]); | |
6f505b16 PZ |
7627 | } |
7628 | ||
7629 | kfree(tg->cfs_rq); | |
7630 | kfree(tg->se); | |
6f505b16 PZ |
7631 | } |
7632 | ||
bccbe08a | 7633 | static int alloc_fair_sched_group(struct task_group *tg) |
29f59db3 | 7634 | { |
29f59db3 SV |
7635 | struct cfs_rq *cfs_rq; |
7636 | struct sched_entity *se; | |
9b5b7751 | 7637 | struct rq *rq; |
29f59db3 SV |
7638 | int i; |
7639 | ||
434d53b0 | 7640 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7641 | if (!tg->cfs_rq) |
7642 | goto err; | |
434d53b0 | 7643 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7644 | if (!tg->se) |
7645 | goto err; | |
052f1dc7 PZ |
7646 | |
7647 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
7648 | |
7649 | for_each_possible_cpu(i) { | |
9b5b7751 | 7650 | rq = cpu_rq(i); |
29f59db3 | 7651 | |
6f505b16 PZ |
7652 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
7653 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
7654 | if (!cfs_rq) |
7655 | goto err; | |
7656 | ||
6f505b16 PZ |
7657 | se = kmalloc_node(sizeof(struct sched_entity), |
7658 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
7659 | if (!se) |
7660 | goto err; | |
7661 | ||
052f1dc7 | 7662 | init_tg_cfs_entry(rq, tg, cfs_rq, se, i, 0); |
bccbe08a PZ |
7663 | } |
7664 | ||
7665 | return 1; | |
7666 | ||
7667 | err: | |
7668 | return 0; | |
7669 | } | |
7670 | ||
7671 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7672 | { | |
7673 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
7674 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
7675 | } | |
7676 | ||
7677 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7678 | { | |
7679 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
7680 | } | |
7681 | #else | |
7682 | static inline void free_fair_sched_group(struct task_group *tg) | |
7683 | { | |
7684 | } | |
7685 | ||
7686 | static inline int alloc_fair_sched_group(struct task_group *tg) | |
7687 | { | |
7688 | return 1; | |
7689 | } | |
7690 | ||
7691 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7692 | { | |
7693 | } | |
7694 | ||
7695 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7696 | { | |
7697 | } | |
052f1dc7 PZ |
7698 | #endif |
7699 | ||
7700 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
7701 | static void free_rt_sched_group(struct task_group *tg) |
7702 | { | |
7703 | int i; | |
7704 | ||
d0b27fa7 PZ |
7705 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
7706 | ||
bccbe08a PZ |
7707 | for_each_possible_cpu(i) { |
7708 | if (tg->rt_rq) | |
7709 | kfree(tg->rt_rq[i]); | |
7710 | if (tg->rt_se) | |
7711 | kfree(tg->rt_se[i]); | |
7712 | } | |
7713 | ||
7714 | kfree(tg->rt_rq); | |
7715 | kfree(tg->rt_se); | |
7716 | } | |
7717 | ||
7718 | static int alloc_rt_sched_group(struct task_group *tg) | |
7719 | { | |
7720 | struct rt_rq *rt_rq; | |
7721 | struct sched_rt_entity *rt_se; | |
7722 | struct rq *rq; | |
7723 | int i; | |
7724 | ||
434d53b0 | 7725 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
7726 | if (!tg->rt_rq) |
7727 | goto err; | |
434d53b0 | 7728 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
7729 | if (!tg->rt_se) |
7730 | goto err; | |
7731 | ||
d0b27fa7 PZ |
7732 | init_rt_bandwidth(&tg->rt_bandwidth, |
7733 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
7734 | |
7735 | for_each_possible_cpu(i) { | |
7736 | rq = cpu_rq(i); | |
7737 | ||
6f505b16 PZ |
7738 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
7739 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
7740 | if (!rt_rq) | |
7741 | goto err; | |
29f59db3 | 7742 | |
6f505b16 PZ |
7743 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
7744 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
7745 | if (!rt_se) | |
7746 | goto err; | |
29f59db3 | 7747 | |
6f505b16 | 7748 | init_tg_rt_entry(rq, tg, rt_rq, rt_se, i, 0); |
29f59db3 SV |
7749 | } |
7750 | ||
bccbe08a PZ |
7751 | return 1; |
7752 | ||
7753 | err: | |
7754 | return 0; | |
7755 | } | |
7756 | ||
7757 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7758 | { | |
7759 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
7760 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
7761 | } | |
7762 | ||
7763 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7764 | { | |
7765 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
7766 | } | |
7767 | #else | |
7768 | static inline void free_rt_sched_group(struct task_group *tg) | |
7769 | { | |
7770 | } | |
7771 | ||
7772 | static inline int alloc_rt_sched_group(struct task_group *tg) | |
7773 | { | |
7774 | return 1; | |
7775 | } | |
7776 | ||
7777 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7778 | { | |
7779 | } | |
7780 | ||
7781 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7782 | { | |
7783 | } | |
7784 | #endif | |
7785 | ||
d0b27fa7 | 7786 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
7787 | static void free_sched_group(struct task_group *tg) |
7788 | { | |
7789 | free_fair_sched_group(tg); | |
7790 | free_rt_sched_group(tg); | |
7791 | kfree(tg); | |
7792 | } | |
7793 | ||
7794 | /* allocate runqueue etc for a new task group */ | |
7795 | struct task_group *sched_create_group(void) | |
7796 | { | |
7797 | struct task_group *tg; | |
7798 | unsigned long flags; | |
7799 | int i; | |
7800 | ||
7801 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7802 | if (!tg) | |
7803 | return ERR_PTR(-ENOMEM); | |
7804 | ||
7805 | if (!alloc_fair_sched_group(tg)) | |
7806 | goto err; | |
7807 | ||
7808 | if (!alloc_rt_sched_group(tg)) | |
7809 | goto err; | |
7810 | ||
8ed36996 | 7811 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 7812 | for_each_possible_cpu(i) { |
bccbe08a PZ |
7813 | register_fair_sched_group(tg, i); |
7814 | register_rt_sched_group(tg, i); | |
9b5b7751 | 7815 | } |
6f505b16 | 7816 | list_add_rcu(&tg->list, &task_groups); |
8ed36996 | 7817 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 7818 | |
9b5b7751 | 7819 | return tg; |
29f59db3 SV |
7820 | |
7821 | err: | |
6f505b16 | 7822 | free_sched_group(tg); |
29f59db3 SV |
7823 | return ERR_PTR(-ENOMEM); |
7824 | } | |
7825 | ||
9b5b7751 | 7826 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7827 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7828 | { |
29f59db3 | 7829 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7830 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7831 | } |
7832 | ||
9b5b7751 | 7833 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7834 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7835 | { |
8ed36996 | 7836 | unsigned long flags; |
9b5b7751 | 7837 | int i; |
29f59db3 | 7838 | |
8ed36996 | 7839 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 7840 | for_each_possible_cpu(i) { |
bccbe08a PZ |
7841 | unregister_fair_sched_group(tg, i); |
7842 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 7843 | } |
6f505b16 | 7844 | list_del_rcu(&tg->list); |
8ed36996 | 7845 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 7846 | |
9b5b7751 | 7847 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 7848 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
7849 | } |
7850 | ||
9b5b7751 | 7851 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7852 | * The caller of this function should have put the task in its new group |
7853 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7854 | * reflect its new group. | |
9b5b7751 SV |
7855 | */ |
7856 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7857 | { |
7858 | int on_rq, running; | |
7859 | unsigned long flags; | |
7860 | struct rq *rq; | |
7861 | ||
7862 | rq = task_rq_lock(tsk, &flags); | |
7863 | ||
29f59db3 SV |
7864 | update_rq_clock(rq); |
7865 | ||
051a1d1a | 7866 | running = task_current(rq, tsk); |
29f59db3 SV |
7867 | on_rq = tsk->se.on_rq; |
7868 | ||
0e1f3483 | 7869 | if (on_rq) |
29f59db3 | 7870 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7871 | if (unlikely(running)) |
7872 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7873 | |
6f505b16 | 7874 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7875 | |
810b3817 PZ |
7876 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7877 | if (tsk->sched_class->moved_group) | |
7878 | tsk->sched_class->moved_group(tsk); | |
7879 | #endif | |
7880 | ||
0e1f3483 HS |
7881 | if (unlikely(running)) |
7882 | tsk->sched_class->set_curr_task(rq); | |
7883 | if (on_rq) | |
7074badb | 7884 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7885 | |
29f59db3 SV |
7886 | task_rq_unlock(rq, &flags); |
7887 | } | |
d0b27fa7 | 7888 | #endif |
29f59db3 | 7889 | |
052f1dc7 | 7890 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
7891 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
7892 | { | |
7893 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
7894 | struct rq *rq = cfs_rq->rq; | |
7895 | int on_rq; | |
7896 | ||
62fb1851 | 7897 | spin_lock_irq(&rq->lock); |
29f59db3 SV |
7898 | |
7899 | on_rq = se->on_rq; | |
62fb1851 | 7900 | if (on_rq) |
29f59db3 SV |
7901 | dequeue_entity(cfs_rq, se, 0); |
7902 | ||
7903 | se->load.weight = shares; | |
7904 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
7905 | ||
62fb1851 | 7906 | if (on_rq) |
29f59db3 | 7907 | enqueue_entity(cfs_rq, se, 0); |
62fb1851 PZ |
7908 | |
7909 | spin_unlock_irq(&rq->lock); | |
29f59db3 SV |
7910 | } |
7911 | ||
8ed36996 PZ |
7912 | static DEFINE_MUTEX(shares_mutex); |
7913 | ||
4cf86d77 | 7914 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
7915 | { |
7916 | int i; | |
8ed36996 | 7917 | unsigned long flags; |
c61935fd | 7918 | |
62fb1851 PZ |
7919 | /* |
7920 | * A weight of 0 or 1 can cause arithmetics problems. | |
7921 | * (The default weight is 1024 - so there's no practical | |
7922 | * limitation from this.) | |
7923 | */ | |
7924 | if (shares < 2) | |
7925 | shares = 2; | |
7926 | ||
8ed36996 | 7927 | mutex_lock(&shares_mutex); |
9b5b7751 | 7928 | if (tg->shares == shares) |
5cb350ba | 7929 | goto done; |
29f59db3 | 7930 | |
8ed36996 | 7931 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
7932 | for_each_possible_cpu(i) |
7933 | unregister_fair_sched_group(tg, i); | |
8ed36996 | 7934 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
7935 | |
7936 | /* wait for any ongoing reference to this group to finish */ | |
7937 | synchronize_sched(); | |
7938 | ||
7939 | /* | |
7940 | * Now we are free to modify the group's share on each cpu | |
7941 | * w/o tripping rebalance_share or load_balance_fair. | |
7942 | */ | |
9b5b7751 | 7943 | tg->shares = shares; |
62fb1851 | 7944 | for_each_possible_cpu(i) |
9b5b7751 | 7945 | set_se_shares(tg->se[i], shares); |
29f59db3 | 7946 | |
6b2d7700 SV |
7947 | /* |
7948 | * Enable load balance activity on this group, by inserting it back on | |
7949 | * each cpu's rq->leaf_cfs_rq_list. | |
7950 | */ | |
8ed36996 | 7951 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
7952 | for_each_possible_cpu(i) |
7953 | register_fair_sched_group(tg, i); | |
8ed36996 | 7954 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 7955 | done: |
8ed36996 | 7956 | mutex_unlock(&shares_mutex); |
9b5b7751 | 7957 | return 0; |
29f59db3 SV |
7958 | } |
7959 | ||
5cb350ba DG |
7960 | unsigned long sched_group_shares(struct task_group *tg) |
7961 | { | |
7962 | return tg->shares; | |
7963 | } | |
052f1dc7 | 7964 | #endif |
5cb350ba | 7965 | |
052f1dc7 | 7966 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7967 | /* |
9f0c1e56 | 7968 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 7969 | */ |
9f0c1e56 PZ |
7970 | static DEFINE_MUTEX(rt_constraints_mutex); |
7971 | ||
7972 | static unsigned long to_ratio(u64 period, u64 runtime) | |
7973 | { | |
7974 | if (runtime == RUNTIME_INF) | |
7975 | return 1ULL << 16; | |
7976 | ||
2692a240 | 7977 | return div64_64(runtime << 16, period); |
9f0c1e56 PZ |
7978 | } |
7979 | ||
7980 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
6f505b16 PZ |
7981 | { |
7982 | struct task_group *tgi; | |
7983 | unsigned long total = 0; | |
9f0c1e56 | 7984 | unsigned long global_ratio = |
d0b27fa7 | 7985 | to_ratio(global_rt_period(), global_rt_runtime()); |
6f505b16 PZ |
7986 | |
7987 | rcu_read_lock(); | |
9f0c1e56 PZ |
7988 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
7989 | if (tgi == tg) | |
7990 | continue; | |
6f505b16 | 7991 | |
d0b27fa7 PZ |
7992 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), |
7993 | tgi->rt_bandwidth.rt_runtime); | |
9f0c1e56 PZ |
7994 | } |
7995 | rcu_read_unlock(); | |
6f505b16 | 7996 | |
9f0c1e56 | 7997 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 PZ |
7998 | } |
7999 | ||
521f1a24 DG |
8000 | /* Must be called with tasklist_lock held */ |
8001 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
8002 | { | |
8003 | struct task_struct *g, *p; | |
8004 | do_each_thread(g, p) { | |
8005 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8006 | return 1; | |
8007 | } while_each_thread(g, p); | |
8008 | return 0; | |
8009 | } | |
8010 | ||
d0b27fa7 PZ |
8011 | static int tg_set_bandwidth(struct task_group *tg, |
8012 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8013 | { |
ac086bc2 | 8014 | int i, err = 0; |
9f0c1e56 | 8015 | |
9f0c1e56 | 8016 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8017 | read_lock(&tasklist_lock); |
ac086bc2 | 8018 | if (rt_runtime == 0 && tg_has_rt_tasks(tg)) { |
521f1a24 DG |
8019 | err = -EBUSY; |
8020 | goto unlock; | |
8021 | } | |
9f0c1e56 PZ |
8022 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
8023 | err = -EINVAL; | |
8024 | goto unlock; | |
8025 | } | |
ac086bc2 PZ |
8026 | |
8027 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8028 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8029 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8030 | |
8031 | for_each_possible_cpu(i) { | |
8032 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8033 | ||
8034 | spin_lock(&rt_rq->rt_runtime_lock); | |
8035 | rt_rq->rt_runtime = rt_runtime; | |
8036 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8037 | } | |
8038 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8039 | unlock: |
521f1a24 | 8040 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8041 | mutex_unlock(&rt_constraints_mutex); |
8042 | ||
8043 | return err; | |
6f505b16 PZ |
8044 | } |
8045 | ||
d0b27fa7 PZ |
8046 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8047 | { | |
8048 | u64 rt_runtime, rt_period; | |
8049 | ||
8050 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8051 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8052 | if (rt_runtime_us < 0) | |
8053 | rt_runtime = RUNTIME_INF; | |
8054 | ||
8055 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8056 | } | |
8057 | ||
9f0c1e56 PZ |
8058 | long sched_group_rt_runtime(struct task_group *tg) |
8059 | { | |
8060 | u64 rt_runtime_us; | |
8061 | ||
d0b27fa7 | 8062 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8063 | return -1; |
8064 | ||
d0b27fa7 | 8065 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8066 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8067 | return rt_runtime_us; | |
8068 | } | |
d0b27fa7 PZ |
8069 | |
8070 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8071 | { | |
8072 | u64 rt_runtime, rt_period; | |
8073 | ||
8074 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8075 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8076 | ||
8077 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8078 | } | |
8079 | ||
8080 | long sched_group_rt_period(struct task_group *tg) | |
8081 | { | |
8082 | u64 rt_period_us; | |
8083 | ||
8084 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8085 | do_div(rt_period_us, NSEC_PER_USEC); | |
8086 | return rt_period_us; | |
8087 | } | |
8088 | ||
8089 | static int sched_rt_global_constraints(void) | |
8090 | { | |
8091 | int ret = 0; | |
8092 | ||
8093 | mutex_lock(&rt_constraints_mutex); | |
8094 | if (!__rt_schedulable(NULL, 1, 0)) | |
8095 | ret = -EINVAL; | |
8096 | mutex_unlock(&rt_constraints_mutex); | |
8097 | ||
8098 | return ret; | |
8099 | } | |
8100 | #else | |
8101 | static int sched_rt_global_constraints(void) | |
8102 | { | |
ac086bc2 PZ |
8103 | unsigned long flags; |
8104 | int i; | |
8105 | ||
8106 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8107 | for_each_possible_cpu(i) { | |
8108 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8109 | ||
8110 | spin_lock(&rt_rq->rt_runtime_lock); | |
8111 | rt_rq->rt_runtime = global_rt_runtime(); | |
8112 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8113 | } | |
8114 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8115 | ||
d0b27fa7 PZ |
8116 | return 0; |
8117 | } | |
052f1dc7 | 8118 | #endif |
d0b27fa7 PZ |
8119 | |
8120 | int sched_rt_handler(struct ctl_table *table, int write, | |
8121 | struct file *filp, void __user *buffer, size_t *lenp, | |
8122 | loff_t *ppos) | |
8123 | { | |
8124 | int ret; | |
8125 | int old_period, old_runtime; | |
8126 | static DEFINE_MUTEX(mutex); | |
8127 | ||
8128 | mutex_lock(&mutex); | |
8129 | old_period = sysctl_sched_rt_period; | |
8130 | old_runtime = sysctl_sched_rt_runtime; | |
8131 | ||
8132 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
8133 | ||
8134 | if (!ret && write) { | |
8135 | ret = sched_rt_global_constraints(); | |
8136 | if (ret) { | |
8137 | sysctl_sched_rt_period = old_period; | |
8138 | sysctl_sched_rt_runtime = old_runtime; | |
8139 | } else { | |
8140 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8141 | def_rt_bandwidth.rt_period = | |
8142 | ns_to_ktime(global_rt_period()); | |
8143 | } | |
8144 | } | |
8145 | mutex_unlock(&mutex); | |
8146 | ||
8147 | return ret; | |
8148 | } | |
68318b8e | 8149 | |
052f1dc7 | 8150 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8151 | |
8152 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8153 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8154 | { |
2b01dfe3 PM |
8155 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8156 | struct task_group, css); | |
68318b8e SV |
8157 | } |
8158 | ||
8159 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8160 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e SV |
8161 | { |
8162 | struct task_group *tg; | |
8163 | ||
2b01dfe3 | 8164 | if (!cgrp->parent) { |
68318b8e | 8165 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 8166 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
8167 | return &init_task_group.css; |
8168 | } | |
8169 | ||
8170 | /* we support only 1-level deep hierarchical scheduler atm */ | |
2b01dfe3 | 8171 | if (cgrp->parent->parent) |
68318b8e SV |
8172 | return ERR_PTR(-EINVAL); |
8173 | ||
8174 | tg = sched_create_group(); | |
8175 | if (IS_ERR(tg)) | |
8176 | return ERR_PTR(-ENOMEM); | |
8177 | ||
8178 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 8179 | tg->css.cgroup = cgrp; |
68318b8e SV |
8180 | |
8181 | return &tg->css; | |
8182 | } | |
8183 | ||
41a2d6cf IM |
8184 | static void |
8185 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8186 | { |
2b01dfe3 | 8187 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8188 | |
8189 | sched_destroy_group(tg); | |
8190 | } | |
8191 | ||
41a2d6cf IM |
8192 | static int |
8193 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8194 | struct task_struct *tsk) | |
68318b8e | 8195 | { |
b68aa230 PZ |
8196 | #ifdef CONFIG_RT_GROUP_SCHED |
8197 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 8198 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
8199 | return -EINVAL; |
8200 | #else | |
68318b8e SV |
8201 | /* We don't support RT-tasks being in separate groups */ |
8202 | if (tsk->sched_class != &fair_sched_class) | |
8203 | return -EINVAL; | |
b68aa230 | 8204 | #endif |
68318b8e SV |
8205 | |
8206 | return 0; | |
8207 | } | |
8208 | ||
8209 | static void | |
2b01dfe3 | 8210 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
8211 | struct cgroup *old_cont, struct task_struct *tsk) |
8212 | { | |
8213 | sched_move_task(tsk); | |
8214 | } | |
8215 | ||
052f1dc7 | 8216 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2b01dfe3 PM |
8217 | static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype, |
8218 | u64 shareval) | |
68318b8e | 8219 | { |
2b01dfe3 | 8220 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8221 | } |
8222 | ||
2b01dfe3 | 8223 | static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8224 | { |
2b01dfe3 | 8225 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8226 | |
8227 | return (u64) tg->shares; | |
8228 | } | |
052f1dc7 | 8229 | #endif |
68318b8e | 8230 | |
052f1dc7 | 8231 | #ifdef CONFIG_RT_GROUP_SCHED |
ac086bc2 | 8232 | static ssize_t cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
9f0c1e56 PZ |
8233 | struct file *file, |
8234 | const char __user *userbuf, | |
8235 | size_t nbytes, loff_t *unused_ppos) | |
6f505b16 | 8236 | { |
9f0c1e56 PZ |
8237 | char buffer[64]; |
8238 | int retval = 0; | |
8239 | s64 val; | |
8240 | char *end; | |
8241 | ||
8242 | if (!nbytes) | |
8243 | return -EINVAL; | |
8244 | if (nbytes >= sizeof(buffer)) | |
8245 | return -E2BIG; | |
8246 | if (copy_from_user(buffer, userbuf, nbytes)) | |
8247 | return -EFAULT; | |
8248 | ||
8249 | buffer[nbytes] = 0; /* nul-terminate */ | |
8250 | ||
8251 | /* strip newline if necessary */ | |
8252 | if (nbytes && (buffer[nbytes-1] == '\n')) | |
8253 | buffer[nbytes-1] = 0; | |
8254 | val = simple_strtoll(buffer, &end, 0); | |
8255 | if (*end) | |
8256 | return -EINVAL; | |
8257 | ||
8258 | /* Pass to subsystem */ | |
8259 | retval = sched_group_set_rt_runtime(cgroup_tg(cgrp), val); | |
8260 | if (!retval) | |
8261 | retval = nbytes; | |
8262 | return retval; | |
6f505b16 PZ |
8263 | } |
8264 | ||
9f0c1e56 PZ |
8265 | static ssize_t cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft, |
8266 | struct file *file, | |
8267 | char __user *buf, size_t nbytes, | |
8268 | loff_t *ppos) | |
6f505b16 | 8269 | { |
9f0c1e56 PZ |
8270 | char tmp[64]; |
8271 | long val = sched_group_rt_runtime(cgroup_tg(cgrp)); | |
8272 | int len = sprintf(tmp, "%ld\n", val); | |
6f505b16 | 8273 | |
9f0c1e56 | 8274 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); |
6f505b16 | 8275 | } |
d0b27fa7 PZ |
8276 | |
8277 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8278 | u64 rt_period_us) | |
8279 | { | |
8280 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8281 | } | |
8282 | ||
8283 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8284 | { | |
8285 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8286 | } | |
052f1dc7 | 8287 | #endif |
6f505b16 | 8288 | |
fe5c7cc2 | 8289 | static struct cftype cpu_files[] = { |
052f1dc7 | 8290 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8291 | { |
8292 | .name = "shares", | |
8293 | .read_uint = cpu_shares_read_uint, | |
8294 | .write_uint = cpu_shares_write_uint, | |
8295 | }, | |
052f1dc7 PZ |
8296 | #endif |
8297 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8298 | { |
9f0c1e56 PZ |
8299 | .name = "rt_runtime_us", |
8300 | .read = cpu_rt_runtime_read, | |
8301 | .write = cpu_rt_runtime_write, | |
6f505b16 | 8302 | }, |
d0b27fa7 PZ |
8303 | { |
8304 | .name = "rt_period_us", | |
8305 | .read_uint = cpu_rt_period_read_uint, | |
8306 | .write_uint = cpu_rt_period_write_uint, | |
8307 | }, | |
052f1dc7 | 8308 | #endif |
68318b8e SV |
8309 | }; |
8310 | ||
8311 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8312 | { | |
fe5c7cc2 | 8313 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8314 | } |
8315 | ||
8316 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8317 | .name = "cpu", |
8318 | .create = cpu_cgroup_create, | |
8319 | .destroy = cpu_cgroup_destroy, | |
8320 | .can_attach = cpu_cgroup_can_attach, | |
8321 | .attach = cpu_cgroup_attach, | |
8322 | .populate = cpu_cgroup_populate, | |
8323 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8324 | .early_init = 1, |
8325 | }; | |
8326 | ||
052f1dc7 | 8327 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8328 | |
8329 | #ifdef CONFIG_CGROUP_CPUACCT | |
8330 | ||
8331 | /* | |
8332 | * CPU accounting code for task groups. | |
8333 | * | |
8334 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8335 | * (balbir@in.ibm.com). | |
8336 | */ | |
8337 | ||
8338 | /* track cpu usage of a group of tasks */ | |
8339 | struct cpuacct { | |
8340 | struct cgroup_subsys_state css; | |
8341 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
8342 | u64 *cpuusage; | |
8343 | }; | |
8344 | ||
8345 | struct cgroup_subsys cpuacct_subsys; | |
8346 | ||
8347 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8348 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8349 | { |
32cd756a | 8350 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8351 | struct cpuacct, css); |
8352 | } | |
8353 | ||
8354 | /* return cpu accounting group to which this task belongs */ | |
8355 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8356 | { | |
8357 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8358 | struct cpuacct, css); | |
8359 | } | |
8360 | ||
8361 | /* create a new cpu accounting group */ | |
8362 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8363 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8364 | { |
8365 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
8366 | ||
8367 | if (!ca) | |
8368 | return ERR_PTR(-ENOMEM); | |
8369 | ||
8370 | ca->cpuusage = alloc_percpu(u64); | |
8371 | if (!ca->cpuusage) { | |
8372 | kfree(ca); | |
8373 | return ERR_PTR(-ENOMEM); | |
8374 | } | |
8375 | ||
8376 | return &ca->css; | |
8377 | } | |
8378 | ||
8379 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8380 | static void |
32cd756a | 8381 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8382 | { |
32cd756a | 8383 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8384 | |
8385 | free_percpu(ca->cpuusage); | |
8386 | kfree(ca); | |
8387 | } | |
8388 | ||
8389 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 8390 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8391 | { |
32cd756a | 8392 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8393 | u64 totalcpuusage = 0; |
8394 | int i; | |
8395 | ||
8396 | for_each_possible_cpu(i) { | |
8397 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
8398 | ||
8399 | /* | |
8400 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
8401 | * platforms. | |
8402 | */ | |
8403 | spin_lock_irq(&cpu_rq(i)->lock); | |
8404 | totalcpuusage += *cpuusage; | |
8405 | spin_unlock_irq(&cpu_rq(i)->lock); | |
8406 | } | |
8407 | ||
8408 | return totalcpuusage; | |
8409 | } | |
8410 | ||
0297b803 DG |
8411 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8412 | u64 reset) | |
8413 | { | |
8414 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8415 | int err = 0; | |
8416 | int i; | |
8417 | ||
8418 | if (reset) { | |
8419 | err = -EINVAL; | |
8420 | goto out; | |
8421 | } | |
8422 | ||
8423 | for_each_possible_cpu(i) { | |
8424 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
8425 | ||
8426 | spin_lock_irq(&cpu_rq(i)->lock); | |
8427 | *cpuusage = 0; | |
8428 | spin_unlock_irq(&cpu_rq(i)->lock); | |
8429 | } | |
8430 | out: | |
8431 | return err; | |
8432 | } | |
8433 | ||
d842de87 SV |
8434 | static struct cftype files[] = { |
8435 | { | |
8436 | .name = "usage", | |
8437 | .read_uint = cpuusage_read, | |
0297b803 | 8438 | .write_uint = cpuusage_write, |
d842de87 SV |
8439 | }, |
8440 | }; | |
8441 | ||
32cd756a | 8442 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8443 | { |
32cd756a | 8444 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8445 | } |
8446 | ||
8447 | /* | |
8448 | * charge this task's execution time to its accounting group. | |
8449 | * | |
8450 | * called with rq->lock held. | |
8451 | */ | |
8452 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8453 | { | |
8454 | struct cpuacct *ca; | |
8455 | ||
8456 | if (!cpuacct_subsys.active) | |
8457 | return; | |
8458 | ||
8459 | ca = task_ca(tsk); | |
8460 | if (ca) { | |
8461 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
8462 | ||
8463 | *cpuusage += cputime; | |
8464 | } | |
8465 | } | |
8466 | ||
8467 | struct cgroup_subsys cpuacct_subsys = { | |
8468 | .name = "cpuacct", | |
8469 | .create = cpuacct_create, | |
8470 | .destroy = cpuacct_destroy, | |
8471 | .populate = cpuacct_populate, | |
8472 | .subsys_id = cpuacct_subsys_id, | |
8473 | }; | |
8474 | #endif /* CONFIG_CGROUP_CPUACCT */ |