]>
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 | 165 | struct rt_bandwidth { |
ea736ed5 IM |
166 | /* nests inside the rq lock: */ |
167 | spinlock_t rt_runtime_lock; | |
168 | ktime_t rt_period; | |
169 | u64 rt_runtime; | |
170 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
171 | }; |
172 | ||
173 | static struct rt_bandwidth def_rt_bandwidth; | |
174 | ||
175 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
176 | ||
177 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
178 | { | |
179 | struct rt_bandwidth *rt_b = | |
180 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
181 | ktime_t now; | |
182 | int overrun; | |
183 | int idle = 0; | |
184 | ||
185 | for (;;) { | |
186 | now = hrtimer_cb_get_time(timer); | |
187 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
188 | ||
189 | if (!overrun) | |
190 | break; | |
191 | ||
192 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
193 | } | |
194 | ||
195 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
196 | } | |
197 | ||
198 | static | |
199 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
200 | { | |
201 | rt_b->rt_period = ns_to_ktime(period); | |
202 | rt_b->rt_runtime = runtime; | |
203 | ||
ac086bc2 PZ |
204 | spin_lock_init(&rt_b->rt_runtime_lock); |
205 | ||
d0b27fa7 PZ |
206 | hrtimer_init(&rt_b->rt_period_timer, |
207 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
208 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
209 | rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
210 | } | |
211 | ||
212 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
213 | { | |
214 | ktime_t now; | |
215 | ||
216 | if (rt_b->rt_runtime == RUNTIME_INF) | |
217 | return; | |
218 | ||
219 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
220 | return; | |
221 | ||
222 | spin_lock(&rt_b->rt_runtime_lock); | |
223 | for (;;) { | |
224 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
225 | break; | |
226 | ||
227 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
228 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
229 | hrtimer_start(&rt_b->rt_period_timer, | |
230 | rt_b->rt_period_timer.expires, | |
231 | HRTIMER_MODE_ABS); | |
232 | } | |
233 | spin_unlock(&rt_b->rt_runtime_lock); | |
234 | } | |
235 | ||
236 | #ifdef CONFIG_RT_GROUP_SCHED | |
237 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
238 | { | |
239 | hrtimer_cancel(&rt_b->rt_period_timer); | |
240 | } | |
241 | #endif | |
242 | ||
052f1dc7 | 243 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 244 | |
68318b8e SV |
245 | #include <linux/cgroup.h> |
246 | ||
29f59db3 SV |
247 | struct cfs_rq; |
248 | ||
6f505b16 PZ |
249 | static LIST_HEAD(task_groups); |
250 | ||
29f59db3 | 251 | /* task group related information */ |
4cf86d77 | 252 | struct task_group { |
052f1dc7 | 253 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
254 | struct cgroup_subsys_state css; |
255 | #endif | |
052f1dc7 PZ |
256 | |
257 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
258 | /* schedulable entities of this group on each cpu */ |
259 | struct sched_entity **se; | |
260 | /* runqueue "owned" by this group on each cpu */ | |
261 | struct cfs_rq **cfs_rq; | |
262 | unsigned long shares; | |
052f1dc7 PZ |
263 | #endif |
264 | ||
265 | #ifdef CONFIG_RT_GROUP_SCHED | |
266 | struct sched_rt_entity **rt_se; | |
267 | struct rt_rq **rt_rq; | |
268 | ||
d0b27fa7 | 269 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 270 | #endif |
6b2d7700 | 271 | |
ae8393e5 | 272 | struct rcu_head rcu; |
6f505b16 | 273 | struct list_head list; |
f473aa5e PZ |
274 | |
275 | struct task_group *parent; | |
276 | struct list_head siblings; | |
277 | struct list_head children; | |
29f59db3 SV |
278 | }; |
279 | ||
354d60c2 | 280 | #ifdef CONFIG_USER_SCHED |
eff766a6 PZ |
281 | |
282 | /* | |
283 | * Root task group. | |
284 | * Every UID task group (including init_task_group aka UID-0) will | |
285 | * be a child to this group. | |
286 | */ | |
287 | struct task_group root_task_group; | |
288 | ||
052f1dc7 | 289 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
290 | /* Default task group's sched entity on each cpu */ |
291 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
292 | /* Default task group's cfs_rq on each cpu */ | |
293 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
052f1dc7 PZ |
294 | #endif |
295 | ||
296 | #ifdef CONFIG_RT_GROUP_SCHED | |
297 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
298 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
052f1dc7 | 299 | #endif |
eff766a6 PZ |
300 | #else |
301 | #define root_task_group init_task_group | |
354d60c2 | 302 | #endif |
6f505b16 | 303 | |
8ed36996 | 304 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
305 | * a task group's cpu shares. |
306 | */ | |
8ed36996 | 307 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 308 | |
a1835615 SV |
309 | /* doms_cur_mutex serializes access to doms_cur[] array */ |
310 | static DEFINE_MUTEX(doms_cur_mutex); | |
311 | ||
052f1dc7 | 312 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
313 | #ifdef CONFIG_USER_SCHED |
314 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
315 | #else | |
316 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD | |
317 | #endif | |
318 | ||
052f1dc7 PZ |
319 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
320 | #endif | |
321 | ||
29f59db3 | 322 | /* Default task group. |
3a252015 | 323 | * Every task in system belong to this group at bootup. |
29f59db3 | 324 | */ |
434d53b0 | 325 | struct task_group init_task_group; |
29f59db3 SV |
326 | |
327 | /* return group to which a task belongs */ | |
4cf86d77 | 328 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 329 | { |
4cf86d77 | 330 | struct task_group *tg; |
9b5b7751 | 331 | |
052f1dc7 | 332 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 333 | tg = p->user->tg; |
052f1dc7 | 334 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
335 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
336 | struct task_group, css); | |
24e377a8 | 337 | #else |
41a2d6cf | 338 | tg = &init_task_group; |
24e377a8 | 339 | #endif |
9b5b7751 | 340 | return tg; |
29f59db3 SV |
341 | } |
342 | ||
343 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 344 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 345 | { |
052f1dc7 | 346 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
347 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
348 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 349 | #endif |
6f505b16 | 350 | |
052f1dc7 | 351 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
352 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
353 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 354 | #endif |
29f59db3 SV |
355 | } |
356 | ||
a1835615 SV |
357 | static inline void lock_doms_cur(void) |
358 | { | |
359 | mutex_lock(&doms_cur_mutex); | |
360 | } | |
361 | ||
362 | static inline void unlock_doms_cur(void) | |
363 | { | |
364 | mutex_unlock(&doms_cur_mutex); | |
365 | } | |
366 | ||
29f59db3 SV |
367 | #else |
368 | ||
6f505b16 | 369 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
a1835615 SV |
370 | static inline void lock_doms_cur(void) { } |
371 | static inline void unlock_doms_cur(void) { } | |
29f59db3 | 372 | |
052f1dc7 | 373 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 374 | |
6aa645ea IM |
375 | /* CFS-related fields in a runqueue */ |
376 | struct cfs_rq { | |
377 | struct load_weight load; | |
378 | unsigned long nr_running; | |
379 | ||
6aa645ea | 380 | u64 exec_clock; |
e9acbff6 | 381 | u64 min_vruntime; |
6aa645ea IM |
382 | |
383 | struct rb_root tasks_timeline; | |
384 | struct rb_node *rb_leftmost; | |
385 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
386 | /* 'curr' points to currently running entity on this cfs_rq. |
387 | * It is set to NULL otherwise (i.e when none are currently running). | |
388 | */ | |
aa2ac252 | 389 | struct sched_entity *curr, *next; |
ddc97297 PZ |
390 | |
391 | unsigned long nr_spread_over; | |
392 | ||
62160e3f | 393 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
394 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
395 | ||
41a2d6cf IM |
396 | /* |
397 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
398 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
399 | * (like users, containers etc.) | |
400 | * | |
401 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
402 | * list is used during load balance. | |
403 | */ | |
41a2d6cf IM |
404 | struct list_head leaf_cfs_rq_list; |
405 | struct task_group *tg; /* group that "owns" this runqueue */ | |
6aa645ea IM |
406 | #endif |
407 | }; | |
1da177e4 | 408 | |
6aa645ea IM |
409 | /* Real-Time classes' related field in a runqueue: */ |
410 | struct rt_rq { | |
411 | struct rt_prio_array active; | |
63489e45 | 412 | unsigned long rt_nr_running; |
052f1dc7 | 413 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
414 | int highest_prio; /* highest queued rt task prio */ |
415 | #endif | |
fa85ae24 | 416 | #ifdef CONFIG_SMP |
73fe6aae | 417 | unsigned long rt_nr_migratory; |
a22d7fc1 | 418 | int overloaded; |
fa85ae24 | 419 | #endif |
6f505b16 | 420 | int rt_throttled; |
fa85ae24 | 421 | u64 rt_time; |
ac086bc2 | 422 | u64 rt_runtime; |
ea736ed5 | 423 | /* Nests inside the rq lock: */ |
ac086bc2 | 424 | spinlock_t rt_runtime_lock; |
6f505b16 | 425 | |
052f1dc7 | 426 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
427 | unsigned long rt_nr_boosted; |
428 | ||
6f505b16 PZ |
429 | struct rq *rq; |
430 | struct list_head leaf_rt_rq_list; | |
431 | struct task_group *tg; | |
432 | struct sched_rt_entity *rt_se; | |
433 | #endif | |
6aa645ea IM |
434 | }; |
435 | ||
57d885fe GH |
436 | #ifdef CONFIG_SMP |
437 | ||
438 | /* | |
439 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
440 | * variables. Each exclusive cpuset essentially defines an island domain by |
441 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
442 | * exclusive cpuset is created, we also create and attach a new root-domain |
443 | * object. | |
444 | * | |
57d885fe GH |
445 | */ |
446 | struct root_domain { | |
447 | atomic_t refcount; | |
448 | cpumask_t span; | |
449 | cpumask_t online; | |
637f5085 | 450 | |
0eab9146 | 451 | /* |
637f5085 GH |
452 | * The "RT overload" flag: it gets set if a CPU has more than |
453 | * one runnable RT task. | |
454 | */ | |
455 | cpumask_t rto_mask; | |
0eab9146 | 456 | atomic_t rto_count; |
57d885fe GH |
457 | }; |
458 | ||
dc938520 GH |
459 | /* |
460 | * By default the system creates a single root-domain with all cpus as | |
461 | * members (mimicking the global state we have today). | |
462 | */ | |
57d885fe GH |
463 | static struct root_domain def_root_domain; |
464 | ||
465 | #endif | |
466 | ||
1da177e4 LT |
467 | /* |
468 | * This is the main, per-CPU runqueue data structure. | |
469 | * | |
470 | * Locking rule: those places that want to lock multiple runqueues | |
471 | * (such as the load balancing or the thread migration code), lock | |
472 | * acquire operations must be ordered by ascending &runqueue. | |
473 | */ | |
70b97a7f | 474 | struct rq { |
d8016491 IM |
475 | /* runqueue lock: */ |
476 | spinlock_t lock; | |
1da177e4 LT |
477 | |
478 | /* | |
479 | * nr_running and cpu_load should be in the same cacheline because | |
480 | * remote CPUs use both these fields when doing load calculation. | |
481 | */ | |
482 | unsigned long nr_running; | |
6aa645ea IM |
483 | #define CPU_LOAD_IDX_MAX 5 |
484 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 485 | unsigned char idle_at_tick; |
46cb4b7c | 486 | #ifdef CONFIG_NO_HZ |
15934a37 | 487 | unsigned long last_tick_seen; |
46cb4b7c SS |
488 | unsigned char in_nohz_recently; |
489 | #endif | |
d8016491 IM |
490 | /* capture load from *all* tasks on this cpu: */ |
491 | struct load_weight load; | |
6aa645ea IM |
492 | unsigned long nr_load_updates; |
493 | u64 nr_switches; | |
494 | ||
495 | struct cfs_rq cfs; | |
6f505b16 | 496 | struct rt_rq rt; |
6f505b16 | 497 | |
6aa645ea | 498 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
499 | /* list of leaf cfs_rq on this cpu: */ |
500 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
501 | #endif |
502 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 503 | struct list_head leaf_rt_rq_list; |
1da177e4 | 504 | #endif |
1da177e4 LT |
505 | |
506 | /* | |
507 | * This is part of a global counter where only the total sum | |
508 | * over all CPUs matters. A task can increase this counter on | |
509 | * one CPU and if it got migrated afterwards it may decrease | |
510 | * it on another CPU. Always updated under the runqueue lock: | |
511 | */ | |
512 | unsigned long nr_uninterruptible; | |
513 | ||
36c8b586 | 514 | struct task_struct *curr, *idle; |
c9819f45 | 515 | unsigned long next_balance; |
1da177e4 | 516 | struct mm_struct *prev_mm; |
6aa645ea | 517 | |
6aa645ea IM |
518 | u64 clock, prev_clock_raw; |
519 | s64 clock_max_delta; | |
520 | ||
cc203d24 | 521 | unsigned int clock_warps, clock_overflows, clock_underflows; |
2aa44d05 IM |
522 | u64 idle_clock; |
523 | unsigned int clock_deep_idle_events; | |
529c7726 | 524 | u64 tick_timestamp; |
6aa645ea | 525 | |
1da177e4 LT |
526 | atomic_t nr_iowait; |
527 | ||
528 | #ifdef CONFIG_SMP | |
0eab9146 | 529 | struct root_domain *rd; |
1da177e4 LT |
530 | struct sched_domain *sd; |
531 | ||
532 | /* For active balancing */ | |
533 | int active_balance; | |
534 | int push_cpu; | |
d8016491 IM |
535 | /* cpu of this runqueue: */ |
536 | int cpu; | |
1da177e4 | 537 | |
36c8b586 | 538 | struct task_struct *migration_thread; |
1da177e4 LT |
539 | struct list_head migration_queue; |
540 | #endif | |
541 | ||
8f4d37ec PZ |
542 | #ifdef CONFIG_SCHED_HRTICK |
543 | unsigned long hrtick_flags; | |
544 | ktime_t hrtick_expire; | |
545 | struct hrtimer hrtick_timer; | |
546 | #endif | |
547 | ||
1da177e4 LT |
548 | #ifdef CONFIG_SCHEDSTATS |
549 | /* latency stats */ | |
550 | struct sched_info rq_sched_info; | |
551 | ||
552 | /* sys_sched_yield() stats */ | |
480b9434 KC |
553 | unsigned int yld_exp_empty; |
554 | unsigned int yld_act_empty; | |
555 | unsigned int yld_both_empty; | |
556 | unsigned int yld_count; | |
1da177e4 LT |
557 | |
558 | /* schedule() stats */ | |
480b9434 KC |
559 | unsigned int sched_switch; |
560 | unsigned int sched_count; | |
561 | unsigned int sched_goidle; | |
1da177e4 LT |
562 | |
563 | /* try_to_wake_up() stats */ | |
480b9434 KC |
564 | unsigned int ttwu_count; |
565 | unsigned int ttwu_local; | |
b8efb561 IM |
566 | |
567 | /* BKL stats */ | |
480b9434 | 568 | unsigned int bkl_count; |
1da177e4 | 569 | #endif |
fcb99371 | 570 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
571 | }; |
572 | ||
f34e3b61 | 573 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 574 | |
dd41f596 IM |
575 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
576 | { | |
577 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
578 | } | |
579 | ||
0a2966b4 CL |
580 | static inline int cpu_of(struct rq *rq) |
581 | { | |
582 | #ifdef CONFIG_SMP | |
583 | return rq->cpu; | |
584 | #else | |
585 | return 0; | |
586 | #endif | |
587 | } | |
588 | ||
15934a37 GC |
589 | #ifdef CONFIG_NO_HZ |
590 | static inline bool nohz_on(int cpu) | |
591 | { | |
592 | return tick_get_tick_sched(cpu)->nohz_mode != NOHZ_MODE_INACTIVE; | |
593 | } | |
594 | ||
595 | static inline u64 max_skipped_ticks(struct rq *rq) | |
596 | { | |
597 | return nohz_on(cpu_of(rq)) ? jiffies - rq->last_tick_seen + 2 : 1; | |
598 | } | |
599 | ||
600 | static inline void update_last_tick_seen(struct rq *rq) | |
601 | { | |
602 | rq->last_tick_seen = jiffies; | |
603 | } | |
604 | #else | |
605 | static inline u64 max_skipped_ticks(struct rq *rq) | |
606 | { | |
607 | return 1; | |
608 | } | |
609 | ||
610 | static inline void update_last_tick_seen(struct rq *rq) | |
611 | { | |
612 | } | |
613 | #endif | |
614 | ||
20d315d4 | 615 | /* |
b04a0f4c IM |
616 | * Update the per-runqueue clock, as finegrained as the platform can give |
617 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 618 | */ |
b04a0f4c | 619 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
620 | { |
621 | u64 prev_raw = rq->prev_clock_raw; | |
622 | u64 now = sched_clock(); | |
623 | s64 delta = now - prev_raw; | |
624 | u64 clock = rq->clock; | |
625 | ||
b04a0f4c IM |
626 | #ifdef CONFIG_SCHED_DEBUG |
627 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
628 | #endif | |
20d315d4 IM |
629 | /* |
630 | * Protect against sched_clock() occasionally going backwards: | |
631 | */ | |
632 | if (unlikely(delta < 0)) { | |
633 | clock++; | |
634 | rq->clock_warps++; | |
635 | } else { | |
636 | /* | |
637 | * Catch too large forward jumps too: | |
638 | */ | |
15934a37 GC |
639 | u64 max_jump = max_skipped_ticks(rq) * TICK_NSEC; |
640 | u64 max_time = rq->tick_timestamp + max_jump; | |
641 | ||
642 | if (unlikely(clock + delta > max_time)) { | |
643 | if (clock < max_time) | |
644 | clock = max_time; | |
529c7726 IM |
645 | else |
646 | clock++; | |
20d315d4 IM |
647 | rq->clock_overflows++; |
648 | } else { | |
649 | if (unlikely(delta > rq->clock_max_delta)) | |
650 | rq->clock_max_delta = delta; | |
651 | clock += delta; | |
652 | } | |
653 | } | |
654 | ||
655 | rq->prev_clock_raw = now; | |
656 | rq->clock = clock; | |
b04a0f4c | 657 | } |
20d315d4 | 658 | |
b04a0f4c IM |
659 | static void update_rq_clock(struct rq *rq) |
660 | { | |
661 | if (likely(smp_processor_id() == cpu_of(rq))) | |
662 | __update_rq_clock(rq); | |
20d315d4 IM |
663 | } |
664 | ||
674311d5 NP |
665 | /* |
666 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 667 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
668 | * |
669 | * The domain tree of any CPU may only be accessed from within | |
670 | * preempt-disabled sections. | |
671 | */ | |
48f24c4d IM |
672 | #define for_each_domain(cpu, __sd) \ |
673 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
674 | |
675 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
676 | #define this_rq() (&__get_cpu_var(runqueues)) | |
677 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
678 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
679 | ||
bf5c91ba IM |
680 | /* |
681 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
682 | */ | |
683 | #ifdef CONFIG_SCHED_DEBUG | |
684 | # define const_debug __read_mostly | |
685 | #else | |
686 | # define const_debug static const | |
687 | #endif | |
688 | ||
689 | /* | |
690 | * Debugging: various feature bits | |
691 | */ | |
692 | enum { | |
bbdba7c0 | 693 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
9612633a IM |
694 | SCHED_FEAT_WAKEUP_PREEMPT = 2, |
695 | SCHED_FEAT_START_DEBIT = 4, | |
d25ce4cd IM |
696 | SCHED_FEAT_AFFINE_WAKEUPS = 8, |
697 | SCHED_FEAT_CACHE_HOT_BUDDY = 16, | |
02e2b83b IM |
698 | SCHED_FEAT_SYNC_WAKEUPS = 32, |
699 | SCHED_FEAT_HRTICK = 64, | |
700 | SCHED_FEAT_DOUBLE_TICK = 128, | |
112f53f5 | 701 | SCHED_FEAT_NORMALIZED_SLEEPER = 256, |
bf5c91ba IM |
702 | }; |
703 | ||
704 | const_debug unsigned int sysctl_sched_features = | |
8401f775 | 705 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
9612633a | 706 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | |
8401f775 | 707 | SCHED_FEAT_START_DEBIT * 1 | |
d25ce4cd IM |
708 | SCHED_FEAT_AFFINE_WAKEUPS * 1 | |
709 | SCHED_FEAT_CACHE_HOT_BUDDY * 1 | | |
02e2b83b | 710 | SCHED_FEAT_SYNC_WAKEUPS * 1 | |
8f4d37ec | 711 | SCHED_FEAT_HRTICK * 1 | |
112f53f5 PZ |
712 | SCHED_FEAT_DOUBLE_TICK * 0 | |
713 | SCHED_FEAT_NORMALIZED_SLEEPER * 1; | |
bf5c91ba IM |
714 | |
715 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
716 | ||
b82d9fdd PZ |
717 | /* |
718 | * Number of tasks to iterate in a single balance run. | |
719 | * Limited because this is done with IRQs disabled. | |
720 | */ | |
721 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
722 | ||
fa85ae24 | 723 | /* |
9f0c1e56 | 724 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
725 | * default: 1s |
726 | */ | |
9f0c1e56 | 727 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 728 | |
6892b75e IM |
729 | static __read_mostly int scheduler_running; |
730 | ||
9f0c1e56 PZ |
731 | /* |
732 | * part of the period that we allow rt tasks to run in us. | |
733 | * default: 0.95s | |
734 | */ | |
735 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 736 | |
d0b27fa7 PZ |
737 | static inline u64 global_rt_period(void) |
738 | { | |
739 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
740 | } | |
741 | ||
742 | static inline u64 global_rt_runtime(void) | |
743 | { | |
744 | if (sysctl_sched_rt_period < 0) | |
745 | return RUNTIME_INF; | |
746 | ||
747 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
748 | } | |
fa85ae24 | 749 | |
27ec4407 IM |
750 | static const unsigned long long time_sync_thresh = 100000; |
751 | ||
752 | static DEFINE_PER_CPU(unsigned long long, time_offset); | |
753 | static DEFINE_PER_CPU(unsigned long long, prev_cpu_time); | |
754 | ||
e436d800 | 755 | /* |
27ec4407 IM |
756 | * Global lock which we take every now and then to synchronize |
757 | * the CPUs time. This method is not warp-safe, but it's good | |
758 | * enough to synchronize slowly diverging time sources and thus | |
759 | * it's good enough for tracing: | |
e436d800 | 760 | */ |
27ec4407 IM |
761 | static DEFINE_SPINLOCK(time_sync_lock); |
762 | static unsigned long long prev_global_time; | |
763 | ||
764 | static unsigned long long __sync_cpu_clock(cycles_t time, int cpu) | |
765 | { | |
766 | unsigned long flags; | |
767 | ||
768 | spin_lock_irqsave(&time_sync_lock, flags); | |
769 | ||
770 | if (time < prev_global_time) { | |
771 | per_cpu(time_offset, cpu) += prev_global_time - time; | |
772 | time = prev_global_time; | |
773 | } else { | |
774 | prev_global_time = time; | |
775 | } | |
776 | ||
777 | spin_unlock_irqrestore(&time_sync_lock, flags); | |
778 | ||
779 | return time; | |
780 | } | |
781 | ||
782 | static unsigned long long __cpu_clock(int cpu) | |
e436d800 | 783 | { |
e436d800 IM |
784 | unsigned long long now; |
785 | unsigned long flags; | |
b04a0f4c | 786 | struct rq *rq; |
e436d800 | 787 | |
8ced5f69 IM |
788 | /* |
789 | * Only call sched_clock() if the scheduler has already been | |
790 | * initialized (some code might call cpu_clock() very early): | |
791 | */ | |
6892b75e IM |
792 | if (unlikely(!scheduler_running)) |
793 | return 0; | |
794 | ||
795 | local_irq_save(flags); | |
796 | rq = cpu_rq(cpu); | |
797 | update_rq_clock(rq); | |
b04a0f4c | 798 | now = rq->clock; |
2cd4d0ea | 799 | local_irq_restore(flags); |
e436d800 IM |
800 | |
801 | return now; | |
802 | } | |
27ec4407 IM |
803 | |
804 | /* | |
805 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
806 | * clock constructed from sched_clock(): | |
807 | */ | |
808 | unsigned long long cpu_clock(int cpu) | |
809 | { | |
810 | unsigned long long prev_cpu_time, time, delta_time; | |
811 | ||
812 | prev_cpu_time = per_cpu(prev_cpu_time, cpu); | |
813 | time = __cpu_clock(cpu) + per_cpu(time_offset, cpu); | |
814 | delta_time = time-prev_cpu_time; | |
815 | ||
816 | if (unlikely(delta_time > time_sync_thresh)) | |
817 | time = __sync_cpu_clock(time, cpu); | |
818 | ||
819 | return time; | |
820 | } | |
a58f6f25 | 821 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 822 | |
1da177e4 | 823 | #ifndef prepare_arch_switch |
4866cde0 NP |
824 | # define prepare_arch_switch(next) do { } while (0) |
825 | #endif | |
826 | #ifndef finish_arch_switch | |
827 | # define finish_arch_switch(prev) do { } while (0) | |
828 | #endif | |
829 | ||
051a1d1a DA |
830 | static inline int task_current(struct rq *rq, struct task_struct *p) |
831 | { | |
832 | return rq->curr == p; | |
833 | } | |
834 | ||
4866cde0 | 835 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 836 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 837 | { |
051a1d1a | 838 | return task_current(rq, p); |
4866cde0 NP |
839 | } |
840 | ||
70b97a7f | 841 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
842 | { |
843 | } | |
844 | ||
70b97a7f | 845 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 846 | { |
da04c035 IM |
847 | #ifdef CONFIG_DEBUG_SPINLOCK |
848 | /* this is a valid case when another task releases the spinlock */ | |
849 | rq->lock.owner = current; | |
850 | #endif | |
8a25d5de IM |
851 | /* |
852 | * If we are tracking spinlock dependencies then we have to | |
853 | * fix up the runqueue lock - which gets 'carried over' from | |
854 | * prev into current: | |
855 | */ | |
856 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
857 | ||
4866cde0 NP |
858 | spin_unlock_irq(&rq->lock); |
859 | } | |
860 | ||
861 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 862 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
863 | { |
864 | #ifdef CONFIG_SMP | |
865 | return p->oncpu; | |
866 | #else | |
051a1d1a | 867 | return task_current(rq, p); |
4866cde0 NP |
868 | #endif |
869 | } | |
870 | ||
70b97a7f | 871 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
872 | { |
873 | #ifdef CONFIG_SMP | |
874 | /* | |
875 | * We can optimise this out completely for !SMP, because the | |
876 | * SMP rebalancing from interrupt is the only thing that cares | |
877 | * here. | |
878 | */ | |
879 | next->oncpu = 1; | |
880 | #endif | |
881 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
882 | spin_unlock_irq(&rq->lock); | |
883 | #else | |
884 | spin_unlock(&rq->lock); | |
885 | #endif | |
886 | } | |
887 | ||
70b97a7f | 888 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
889 | { |
890 | #ifdef CONFIG_SMP | |
891 | /* | |
892 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
893 | * We must ensure this doesn't happen until the switch is completely | |
894 | * finished. | |
895 | */ | |
896 | smp_wmb(); | |
897 | prev->oncpu = 0; | |
898 | #endif | |
899 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
900 | local_irq_enable(); | |
1da177e4 | 901 | #endif |
4866cde0 NP |
902 | } |
903 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 904 | |
b29739f9 IM |
905 | /* |
906 | * __task_rq_lock - lock the runqueue a given task resides on. | |
907 | * Must be called interrupts disabled. | |
908 | */ | |
70b97a7f | 909 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
910 | __acquires(rq->lock) |
911 | { | |
3a5c359a AK |
912 | for (;;) { |
913 | struct rq *rq = task_rq(p); | |
914 | spin_lock(&rq->lock); | |
915 | if (likely(rq == task_rq(p))) | |
916 | return rq; | |
b29739f9 | 917 | spin_unlock(&rq->lock); |
b29739f9 | 918 | } |
b29739f9 IM |
919 | } |
920 | ||
1da177e4 LT |
921 | /* |
922 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 923 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
924 | * explicitly disabling preemption. |
925 | */ | |
70b97a7f | 926 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
927 | __acquires(rq->lock) |
928 | { | |
70b97a7f | 929 | struct rq *rq; |
1da177e4 | 930 | |
3a5c359a AK |
931 | for (;;) { |
932 | local_irq_save(*flags); | |
933 | rq = task_rq(p); | |
934 | spin_lock(&rq->lock); | |
935 | if (likely(rq == task_rq(p))) | |
936 | return rq; | |
1da177e4 | 937 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 938 | } |
1da177e4 LT |
939 | } |
940 | ||
a9957449 | 941 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
942 | __releases(rq->lock) |
943 | { | |
944 | spin_unlock(&rq->lock); | |
945 | } | |
946 | ||
70b97a7f | 947 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
948 | __releases(rq->lock) |
949 | { | |
950 | spin_unlock_irqrestore(&rq->lock, *flags); | |
951 | } | |
952 | ||
1da177e4 | 953 | /* |
cc2a73b5 | 954 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 955 | */ |
a9957449 | 956 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
957 | __acquires(rq->lock) |
958 | { | |
70b97a7f | 959 | struct rq *rq; |
1da177e4 LT |
960 | |
961 | local_irq_disable(); | |
962 | rq = this_rq(); | |
963 | spin_lock(&rq->lock); | |
964 | ||
965 | return rq; | |
966 | } | |
967 | ||
1b9f19c2 | 968 | /* |
2aa44d05 | 969 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 970 | */ |
2aa44d05 | 971 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 972 | { |
2aa44d05 IM |
973 | struct rq *rq = cpu_rq(smp_processor_id()); |
974 | ||
975 | spin_lock(&rq->lock); | |
976 | __update_rq_clock(rq); | |
977 | spin_unlock(&rq->lock); | |
978 | rq->clock_deep_idle_events++; | |
979 | } | |
980 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
981 | ||
982 | /* | |
983 | * We just idled delta nanoseconds (called with irqs disabled): | |
984 | */ | |
985 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
986 | { | |
987 | struct rq *rq = cpu_rq(smp_processor_id()); | |
988 | u64 now = sched_clock(); | |
1b9f19c2 | 989 | |
2aa44d05 IM |
990 | rq->idle_clock += delta_ns; |
991 | /* | |
992 | * Override the previous timestamp and ignore all | |
993 | * sched_clock() deltas that occured while we idled, | |
994 | * and use the PM-provided delta_ns to advance the | |
995 | * rq clock: | |
996 | */ | |
997 | spin_lock(&rq->lock); | |
998 | rq->prev_clock_raw = now; | |
999 | rq->clock += delta_ns; | |
1000 | spin_unlock(&rq->lock); | |
782daeee | 1001 | touch_softlockup_watchdog(); |
1b9f19c2 | 1002 | } |
2aa44d05 | 1003 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 1004 | |
8f4d37ec PZ |
1005 | static void __resched_task(struct task_struct *p, int tif_bit); |
1006 | ||
1007 | static inline void resched_task(struct task_struct *p) | |
1008 | { | |
1009 | __resched_task(p, TIF_NEED_RESCHED); | |
1010 | } | |
1011 | ||
1012 | #ifdef CONFIG_SCHED_HRTICK | |
1013 | /* | |
1014 | * Use HR-timers to deliver accurate preemption points. | |
1015 | * | |
1016 | * Its all a bit involved since we cannot program an hrt while holding the | |
1017 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1018 | * reschedule event. | |
1019 | * | |
1020 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1021 | * rq->lock. | |
1022 | */ | |
1023 | static inline void resched_hrt(struct task_struct *p) | |
1024 | { | |
1025 | __resched_task(p, TIF_HRTICK_RESCHED); | |
1026 | } | |
1027 | ||
1028 | static inline void resched_rq(struct rq *rq) | |
1029 | { | |
1030 | unsigned long flags; | |
1031 | ||
1032 | spin_lock_irqsave(&rq->lock, flags); | |
1033 | resched_task(rq->curr); | |
1034 | spin_unlock_irqrestore(&rq->lock, flags); | |
1035 | } | |
1036 | ||
1037 | enum { | |
1038 | HRTICK_SET, /* re-programm hrtick_timer */ | |
1039 | HRTICK_RESET, /* not a new slice */ | |
1040 | }; | |
1041 | ||
1042 | /* | |
1043 | * Use hrtick when: | |
1044 | * - enabled by features | |
1045 | * - hrtimer is actually high res | |
1046 | */ | |
1047 | static inline int hrtick_enabled(struct rq *rq) | |
1048 | { | |
1049 | if (!sched_feat(HRTICK)) | |
1050 | return 0; | |
1051 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
1052 | } | |
1053 | ||
1054 | /* | |
1055 | * Called to set the hrtick timer state. | |
1056 | * | |
1057 | * called with rq->lock held and irqs disabled | |
1058 | */ | |
1059 | static void hrtick_start(struct rq *rq, u64 delay, int reset) | |
1060 | { | |
1061 | assert_spin_locked(&rq->lock); | |
1062 | ||
1063 | /* | |
1064 | * preempt at: now + delay | |
1065 | */ | |
1066 | rq->hrtick_expire = | |
1067 | ktime_add_ns(rq->hrtick_timer.base->get_time(), delay); | |
1068 | /* | |
1069 | * indicate we need to program the timer | |
1070 | */ | |
1071 | __set_bit(HRTICK_SET, &rq->hrtick_flags); | |
1072 | if (reset) | |
1073 | __set_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1074 | ||
1075 | /* | |
1076 | * New slices are called from the schedule path and don't need a | |
1077 | * forced reschedule. | |
1078 | */ | |
1079 | if (reset) | |
1080 | resched_hrt(rq->curr); | |
1081 | } | |
1082 | ||
1083 | static void hrtick_clear(struct rq *rq) | |
1084 | { | |
1085 | if (hrtimer_active(&rq->hrtick_timer)) | |
1086 | hrtimer_cancel(&rq->hrtick_timer); | |
1087 | } | |
1088 | ||
1089 | /* | |
1090 | * Update the timer from the possible pending state. | |
1091 | */ | |
1092 | static void hrtick_set(struct rq *rq) | |
1093 | { | |
1094 | ktime_t time; | |
1095 | int set, reset; | |
1096 | unsigned long flags; | |
1097 | ||
1098 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1099 | ||
1100 | spin_lock_irqsave(&rq->lock, flags); | |
1101 | set = __test_and_clear_bit(HRTICK_SET, &rq->hrtick_flags); | |
1102 | reset = __test_and_clear_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1103 | time = rq->hrtick_expire; | |
1104 | clear_thread_flag(TIF_HRTICK_RESCHED); | |
1105 | spin_unlock_irqrestore(&rq->lock, flags); | |
1106 | ||
1107 | if (set) { | |
1108 | hrtimer_start(&rq->hrtick_timer, time, HRTIMER_MODE_ABS); | |
1109 | if (reset && !hrtimer_active(&rq->hrtick_timer)) | |
1110 | resched_rq(rq); | |
1111 | } else | |
1112 | hrtick_clear(rq); | |
1113 | } | |
1114 | ||
1115 | /* | |
1116 | * High-resolution timer tick. | |
1117 | * Runs from hardirq context with interrupts disabled. | |
1118 | */ | |
1119 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1120 | { | |
1121 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1122 | ||
1123 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1124 | ||
1125 | spin_lock(&rq->lock); | |
1126 | __update_rq_clock(rq); | |
1127 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); | |
1128 | spin_unlock(&rq->lock); | |
1129 | ||
1130 | return HRTIMER_NORESTART; | |
1131 | } | |
1132 | ||
1133 | static inline void init_rq_hrtick(struct rq *rq) | |
1134 | { | |
1135 | rq->hrtick_flags = 0; | |
1136 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1137 | rq->hrtick_timer.function = hrtick; | |
1138 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
1139 | } | |
1140 | ||
1141 | void hrtick_resched(void) | |
1142 | { | |
1143 | struct rq *rq; | |
1144 | unsigned long flags; | |
1145 | ||
1146 | if (!test_thread_flag(TIF_HRTICK_RESCHED)) | |
1147 | return; | |
1148 | ||
1149 | local_irq_save(flags); | |
1150 | rq = cpu_rq(smp_processor_id()); | |
1151 | hrtick_set(rq); | |
1152 | local_irq_restore(flags); | |
1153 | } | |
1154 | #else | |
1155 | static inline void hrtick_clear(struct rq *rq) | |
1156 | { | |
1157 | } | |
1158 | ||
1159 | static inline void hrtick_set(struct rq *rq) | |
1160 | { | |
1161 | } | |
1162 | ||
1163 | static inline void init_rq_hrtick(struct rq *rq) | |
1164 | { | |
1165 | } | |
1166 | ||
1167 | void hrtick_resched(void) | |
1168 | { | |
1169 | } | |
1170 | #endif | |
1171 | ||
c24d20db IM |
1172 | /* |
1173 | * resched_task - mark a task 'to be rescheduled now'. | |
1174 | * | |
1175 | * On UP this means the setting of the need_resched flag, on SMP it | |
1176 | * might also involve a cross-CPU call to trigger the scheduler on | |
1177 | * the target CPU. | |
1178 | */ | |
1179 | #ifdef CONFIG_SMP | |
1180 | ||
1181 | #ifndef tsk_is_polling | |
1182 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1183 | #endif | |
1184 | ||
8f4d37ec | 1185 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1186 | { |
1187 | int cpu; | |
1188 | ||
1189 | assert_spin_locked(&task_rq(p)->lock); | |
1190 | ||
8f4d37ec | 1191 | if (unlikely(test_tsk_thread_flag(p, tif_bit))) |
c24d20db IM |
1192 | return; |
1193 | ||
8f4d37ec | 1194 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1195 | |
1196 | cpu = task_cpu(p); | |
1197 | if (cpu == smp_processor_id()) | |
1198 | return; | |
1199 | ||
1200 | /* NEED_RESCHED must be visible before we test polling */ | |
1201 | smp_mb(); | |
1202 | if (!tsk_is_polling(p)) | |
1203 | smp_send_reschedule(cpu); | |
1204 | } | |
1205 | ||
1206 | static void resched_cpu(int cpu) | |
1207 | { | |
1208 | struct rq *rq = cpu_rq(cpu); | |
1209 | unsigned long flags; | |
1210 | ||
1211 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1212 | return; | |
1213 | resched_task(cpu_curr(cpu)); | |
1214 | spin_unlock_irqrestore(&rq->lock, flags); | |
1215 | } | |
06d8308c TG |
1216 | |
1217 | #ifdef CONFIG_NO_HZ | |
1218 | /* | |
1219 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1220 | * idle CPU then this timer might expire before the next timer event | |
1221 | * which is scheduled to wake up that CPU. In case of a completely | |
1222 | * idle system the next event might even be infinite time into the | |
1223 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1224 | * leaves the inner idle loop so the newly added timer is taken into | |
1225 | * account when the CPU goes back to idle and evaluates the timer | |
1226 | * wheel for the next timer event. | |
1227 | */ | |
1228 | void wake_up_idle_cpu(int cpu) | |
1229 | { | |
1230 | struct rq *rq = cpu_rq(cpu); | |
1231 | ||
1232 | if (cpu == smp_processor_id()) | |
1233 | return; | |
1234 | ||
1235 | /* | |
1236 | * This is safe, as this function is called with the timer | |
1237 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1238 | * to idle and has not yet set rq->curr to idle then it will | |
1239 | * be serialized on the timer wheel base lock and take the new | |
1240 | * timer into account automatically. | |
1241 | */ | |
1242 | if (rq->curr != rq->idle) | |
1243 | return; | |
1244 | ||
1245 | /* | |
1246 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1247 | * lockless. The worst case is that the other CPU runs the | |
1248 | * idle task through an additional NOOP schedule() | |
1249 | */ | |
1250 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1251 | ||
1252 | /* NEED_RESCHED must be visible before we test polling */ | |
1253 | smp_mb(); | |
1254 | if (!tsk_is_polling(rq->idle)) | |
1255 | smp_send_reschedule(cpu); | |
1256 | } | |
1257 | #endif | |
1258 | ||
c24d20db | 1259 | #else |
8f4d37ec | 1260 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1261 | { |
1262 | assert_spin_locked(&task_rq(p)->lock); | |
8f4d37ec | 1263 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1264 | } |
1265 | #endif | |
1266 | ||
45bf76df IM |
1267 | #if BITS_PER_LONG == 32 |
1268 | # define WMULT_CONST (~0UL) | |
1269 | #else | |
1270 | # define WMULT_CONST (1UL << 32) | |
1271 | #endif | |
1272 | ||
1273 | #define WMULT_SHIFT 32 | |
1274 | ||
194081eb IM |
1275 | /* |
1276 | * Shift right and round: | |
1277 | */ | |
cf2ab469 | 1278 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1279 | |
cb1c4fc9 | 1280 | static unsigned long |
45bf76df IM |
1281 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1282 | struct load_weight *lw) | |
1283 | { | |
1284 | u64 tmp; | |
1285 | ||
1286 | if (unlikely(!lw->inv_weight)) | |
27d11726 | 1287 | lw->inv_weight = (WMULT_CONST-lw->weight/2) / (lw->weight+1); |
45bf76df IM |
1288 | |
1289 | tmp = (u64)delta_exec * weight; | |
1290 | /* | |
1291 | * Check whether we'd overflow the 64-bit multiplication: | |
1292 | */ | |
194081eb | 1293 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1294 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1295 | WMULT_SHIFT/2); |
1296 | else | |
cf2ab469 | 1297 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1298 | |
ecf691da | 1299 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1300 | } |
1301 | ||
1302 | static inline unsigned long | |
1303 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
1304 | { | |
1305 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
1306 | } | |
1307 | ||
1091985b | 1308 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1309 | { |
1310 | lw->weight += inc; | |
e89996ae | 1311 | lw->inv_weight = 0; |
45bf76df IM |
1312 | } |
1313 | ||
1091985b | 1314 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1315 | { |
1316 | lw->weight -= dec; | |
e89996ae | 1317 | lw->inv_weight = 0; |
45bf76df IM |
1318 | } |
1319 | ||
2dd73a4f PW |
1320 | /* |
1321 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1322 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1323 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1324 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1325 | * scaled version of the new time slice allocation that they receive on time |
1326 | * slice expiry etc. | |
1327 | */ | |
1328 | ||
dd41f596 IM |
1329 | #define WEIGHT_IDLEPRIO 2 |
1330 | #define WMULT_IDLEPRIO (1 << 31) | |
1331 | ||
1332 | /* | |
1333 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1334 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1335 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1336 | * that remained on nice 0. | |
1337 | * | |
1338 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1339 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1340 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1341 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1342 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1343 | */ |
1344 | static const int prio_to_weight[40] = { | |
254753dc IM |
1345 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1346 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1347 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1348 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1349 | /* 0 */ 1024, 820, 655, 526, 423, | |
1350 | /* 5 */ 335, 272, 215, 172, 137, | |
1351 | /* 10 */ 110, 87, 70, 56, 45, | |
1352 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1353 | }; |
1354 | ||
5714d2de IM |
1355 | /* |
1356 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1357 | * | |
1358 | * In cases where the weight does not change often, we can use the | |
1359 | * precalculated inverse to speed up arithmetics by turning divisions | |
1360 | * into multiplications: | |
1361 | */ | |
dd41f596 | 1362 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1363 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1364 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1365 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1366 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1367 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1368 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1369 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1370 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1371 | }; |
2dd73a4f | 1372 | |
dd41f596 IM |
1373 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1374 | ||
1375 | /* | |
1376 | * runqueue iterator, to support SMP load-balancing between different | |
1377 | * scheduling classes, without having to expose their internal data | |
1378 | * structures to the load-balancing proper: | |
1379 | */ | |
1380 | struct rq_iterator { | |
1381 | void *arg; | |
1382 | struct task_struct *(*start)(void *); | |
1383 | struct task_struct *(*next)(void *); | |
1384 | }; | |
1385 | ||
e1d1484f PW |
1386 | #ifdef CONFIG_SMP |
1387 | static unsigned long | |
1388 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1389 | unsigned long max_load_move, struct sched_domain *sd, | |
1390 | enum cpu_idle_type idle, int *all_pinned, | |
1391 | int *this_best_prio, struct rq_iterator *iterator); | |
1392 | ||
1393 | static int | |
1394 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1395 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1396 | struct rq_iterator *iterator); | |
e1d1484f | 1397 | #endif |
dd41f596 | 1398 | |
d842de87 SV |
1399 | #ifdef CONFIG_CGROUP_CPUACCT |
1400 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1401 | #else | |
1402 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1403 | #endif | |
1404 | ||
e7693a36 GH |
1405 | #ifdef CONFIG_SMP |
1406 | static unsigned long source_load(int cpu, int type); | |
1407 | static unsigned long target_load(int cpu, int type); | |
1408 | static unsigned long cpu_avg_load_per_task(int cpu); | |
1409 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1410 | #endif /* CONFIG_SMP */ | |
1411 | ||
dd41f596 | 1412 | #include "sched_stats.h" |
dd41f596 | 1413 | #include "sched_idletask.c" |
5522d5d5 IM |
1414 | #include "sched_fair.c" |
1415 | #include "sched_rt.c" | |
dd41f596 IM |
1416 | #ifdef CONFIG_SCHED_DEBUG |
1417 | # include "sched_debug.c" | |
1418 | #endif | |
1419 | ||
1420 | #define sched_class_highest (&rt_sched_class) | |
1421 | ||
62fb1851 PZ |
1422 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
1423 | { | |
1424 | update_load_add(&rq->load, p->se.load.weight); | |
1425 | } | |
1426 | ||
1427 | static inline void dec_load(struct rq *rq, const struct task_struct *p) | |
1428 | { | |
1429 | update_load_sub(&rq->load, p->se.load.weight); | |
1430 | } | |
1431 | ||
1432 | static void inc_nr_running(struct task_struct *p, struct rq *rq) | |
9c217245 IM |
1433 | { |
1434 | rq->nr_running++; | |
62fb1851 | 1435 | inc_load(rq, p); |
9c217245 IM |
1436 | } |
1437 | ||
62fb1851 | 1438 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
1439 | { |
1440 | rq->nr_running--; | |
62fb1851 | 1441 | dec_load(rq, p); |
9c217245 IM |
1442 | } |
1443 | ||
45bf76df IM |
1444 | static void set_load_weight(struct task_struct *p) |
1445 | { | |
1446 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1447 | p->se.load.weight = prio_to_weight[0] * 2; |
1448 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1449 | return; | |
1450 | } | |
45bf76df | 1451 | |
dd41f596 IM |
1452 | /* |
1453 | * SCHED_IDLE tasks get minimal weight: | |
1454 | */ | |
1455 | if (p->policy == SCHED_IDLE) { | |
1456 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1457 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1458 | return; | |
1459 | } | |
71f8bd46 | 1460 | |
dd41f596 IM |
1461 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1462 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1463 | } |
1464 | ||
8159f87e | 1465 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1466 | { |
dd41f596 | 1467 | sched_info_queued(p); |
fd390f6a | 1468 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1469 | p->se.on_rq = 1; |
71f8bd46 IM |
1470 | } |
1471 | ||
69be72c1 | 1472 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1473 | { |
f02231e5 | 1474 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1475 | p->se.on_rq = 0; |
71f8bd46 IM |
1476 | } |
1477 | ||
14531189 | 1478 | /* |
dd41f596 | 1479 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1480 | */ |
14531189 IM |
1481 | static inline int __normal_prio(struct task_struct *p) |
1482 | { | |
dd41f596 | 1483 | return p->static_prio; |
14531189 IM |
1484 | } |
1485 | ||
b29739f9 IM |
1486 | /* |
1487 | * Calculate the expected normal priority: i.e. priority | |
1488 | * without taking RT-inheritance into account. Might be | |
1489 | * boosted by interactivity modifiers. Changes upon fork, | |
1490 | * setprio syscalls, and whenever the interactivity | |
1491 | * estimator recalculates. | |
1492 | */ | |
36c8b586 | 1493 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1494 | { |
1495 | int prio; | |
1496 | ||
e05606d3 | 1497 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1498 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1499 | else | |
1500 | prio = __normal_prio(p); | |
1501 | return prio; | |
1502 | } | |
1503 | ||
1504 | /* | |
1505 | * Calculate the current priority, i.e. the priority | |
1506 | * taken into account by the scheduler. This value might | |
1507 | * be boosted by RT tasks, or might be boosted by | |
1508 | * interactivity modifiers. Will be RT if the task got | |
1509 | * RT-boosted. If not then it returns p->normal_prio. | |
1510 | */ | |
36c8b586 | 1511 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1512 | { |
1513 | p->normal_prio = normal_prio(p); | |
1514 | /* | |
1515 | * If we are RT tasks or we were boosted to RT priority, | |
1516 | * keep the priority unchanged. Otherwise, update priority | |
1517 | * to the normal priority: | |
1518 | */ | |
1519 | if (!rt_prio(p->prio)) | |
1520 | return p->normal_prio; | |
1521 | return p->prio; | |
1522 | } | |
1523 | ||
1da177e4 | 1524 | /* |
dd41f596 | 1525 | * activate_task - move a task to the runqueue. |
1da177e4 | 1526 | */ |
dd41f596 | 1527 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1528 | { |
d9514f6c | 1529 | if (task_contributes_to_load(p)) |
dd41f596 | 1530 | rq->nr_uninterruptible--; |
1da177e4 | 1531 | |
8159f87e | 1532 | enqueue_task(rq, p, wakeup); |
62fb1851 | 1533 | inc_nr_running(p, rq); |
1da177e4 LT |
1534 | } |
1535 | ||
1da177e4 LT |
1536 | /* |
1537 | * deactivate_task - remove a task from the runqueue. | |
1538 | */ | |
2e1cb74a | 1539 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1540 | { |
d9514f6c | 1541 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1542 | rq->nr_uninterruptible++; |
1543 | ||
69be72c1 | 1544 | dequeue_task(rq, p, sleep); |
62fb1851 | 1545 | dec_nr_running(p, rq); |
1da177e4 LT |
1546 | } |
1547 | ||
1da177e4 LT |
1548 | /** |
1549 | * task_curr - is this task currently executing on a CPU? | |
1550 | * @p: the task in question. | |
1551 | */ | |
36c8b586 | 1552 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1553 | { |
1554 | return cpu_curr(task_cpu(p)) == p; | |
1555 | } | |
1556 | ||
2dd73a4f PW |
1557 | /* Used instead of source_load when we know the type == 0 */ |
1558 | unsigned long weighted_cpuload(const int cpu) | |
1559 | { | |
495eca49 | 1560 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1561 | } |
1562 | ||
1563 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1564 | { | |
6f505b16 | 1565 | set_task_rq(p, cpu); |
dd41f596 | 1566 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1567 | /* |
1568 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1569 | * successfuly executed on another CPU. We must ensure that updates of | |
1570 | * per-task data have been completed by this moment. | |
1571 | */ | |
1572 | smp_wmb(); | |
dd41f596 | 1573 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1574 | #endif |
2dd73a4f PW |
1575 | } |
1576 | ||
cb469845 SR |
1577 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1578 | const struct sched_class *prev_class, | |
1579 | int oldprio, int running) | |
1580 | { | |
1581 | if (prev_class != p->sched_class) { | |
1582 | if (prev_class->switched_from) | |
1583 | prev_class->switched_from(rq, p, running); | |
1584 | p->sched_class->switched_to(rq, p, running); | |
1585 | } else | |
1586 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1587 | } | |
1588 | ||
1da177e4 | 1589 | #ifdef CONFIG_SMP |
c65cc870 | 1590 | |
cc367732 IM |
1591 | /* |
1592 | * Is this task likely cache-hot: | |
1593 | */ | |
e7693a36 | 1594 | static int |
cc367732 IM |
1595 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1596 | { | |
1597 | s64 delta; | |
1598 | ||
f540a608 IM |
1599 | /* |
1600 | * Buddy candidates are cache hot: | |
1601 | */ | |
d25ce4cd | 1602 | if (sched_feat(CACHE_HOT_BUDDY) && (&p->se == cfs_rq_of(&p->se)->next)) |
f540a608 IM |
1603 | return 1; |
1604 | ||
cc367732 IM |
1605 | if (p->sched_class != &fair_sched_class) |
1606 | return 0; | |
1607 | ||
6bc1665b IM |
1608 | if (sysctl_sched_migration_cost == -1) |
1609 | return 1; | |
1610 | if (sysctl_sched_migration_cost == 0) | |
1611 | return 0; | |
1612 | ||
cc367732 IM |
1613 | delta = now - p->se.exec_start; |
1614 | ||
1615 | return delta < (s64)sysctl_sched_migration_cost; | |
1616 | } | |
1617 | ||
1618 | ||
dd41f596 | 1619 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1620 | { |
dd41f596 IM |
1621 | int old_cpu = task_cpu(p); |
1622 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1623 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1624 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1625 | u64 clock_offset; |
dd41f596 IM |
1626 | |
1627 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1628 | |
1629 | #ifdef CONFIG_SCHEDSTATS | |
1630 | if (p->se.wait_start) | |
1631 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1632 | if (p->se.sleep_start) |
1633 | p->se.sleep_start -= clock_offset; | |
1634 | if (p->se.block_start) | |
1635 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1636 | if (old_cpu != new_cpu) { |
1637 | schedstat_inc(p, se.nr_migrations); | |
1638 | if (task_hot(p, old_rq->clock, NULL)) | |
1639 | schedstat_inc(p, se.nr_forced2_migrations); | |
1640 | } | |
6cfb0d5d | 1641 | #endif |
2830cf8c SV |
1642 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1643 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1644 | |
1645 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1646 | } |
1647 | ||
70b97a7f | 1648 | struct migration_req { |
1da177e4 | 1649 | struct list_head list; |
1da177e4 | 1650 | |
36c8b586 | 1651 | struct task_struct *task; |
1da177e4 LT |
1652 | int dest_cpu; |
1653 | ||
1da177e4 | 1654 | struct completion done; |
70b97a7f | 1655 | }; |
1da177e4 LT |
1656 | |
1657 | /* | |
1658 | * The task's runqueue lock must be held. | |
1659 | * Returns true if you have to wait for migration thread. | |
1660 | */ | |
36c8b586 | 1661 | static int |
70b97a7f | 1662 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1663 | { |
70b97a7f | 1664 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1665 | |
1666 | /* | |
1667 | * If the task is not on a runqueue (and not running), then | |
1668 | * it is sufficient to simply update the task's cpu field. | |
1669 | */ | |
dd41f596 | 1670 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1671 | set_task_cpu(p, dest_cpu); |
1672 | return 0; | |
1673 | } | |
1674 | ||
1675 | init_completion(&req->done); | |
1da177e4 LT |
1676 | req->task = p; |
1677 | req->dest_cpu = dest_cpu; | |
1678 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1679 | |
1da177e4 LT |
1680 | return 1; |
1681 | } | |
1682 | ||
1683 | /* | |
1684 | * wait_task_inactive - wait for a thread to unschedule. | |
1685 | * | |
1686 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1687 | * else this function might spin for a *long* time. This function can't | |
1688 | * be called with interrupts off, or it may introduce deadlock with | |
1689 | * smp_call_function() if an IPI is sent by the same process we are | |
1690 | * waiting to become inactive. | |
1691 | */ | |
36c8b586 | 1692 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1693 | { |
1694 | unsigned long flags; | |
dd41f596 | 1695 | int running, on_rq; |
70b97a7f | 1696 | struct rq *rq; |
1da177e4 | 1697 | |
3a5c359a AK |
1698 | for (;;) { |
1699 | /* | |
1700 | * We do the initial early heuristics without holding | |
1701 | * any task-queue locks at all. We'll only try to get | |
1702 | * the runqueue lock when things look like they will | |
1703 | * work out! | |
1704 | */ | |
1705 | rq = task_rq(p); | |
fa490cfd | 1706 | |
3a5c359a AK |
1707 | /* |
1708 | * If the task is actively running on another CPU | |
1709 | * still, just relax and busy-wait without holding | |
1710 | * any locks. | |
1711 | * | |
1712 | * NOTE! Since we don't hold any locks, it's not | |
1713 | * even sure that "rq" stays as the right runqueue! | |
1714 | * But we don't care, since "task_running()" will | |
1715 | * return false if the runqueue has changed and p | |
1716 | * is actually now running somewhere else! | |
1717 | */ | |
1718 | while (task_running(rq, p)) | |
1719 | cpu_relax(); | |
fa490cfd | 1720 | |
3a5c359a AK |
1721 | /* |
1722 | * Ok, time to look more closely! We need the rq | |
1723 | * lock now, to be *sure*. If we're wrong, we'll | |
1724 | * just go back and repeat. | |
1725 | */ | |
1726 | rq = task_rq_lock(p, &flags); | |
1727 | running = task_running(rq, p); | |
1728 | on_rq = p->se.on_rq; | |
1729 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1730 | |
3a5c359a AK |
1731 | /* |
1732 | * Was it really running after all now that we | |
1733 | * checked with the proper locks actually held? | |
1734 | * | |
1735 | * Oops. Go back and try again.. | |
1736 | */ | |
1737 | if (unlikely(running)) { | |
1738 | cpu_relax(); | |
1739 | continue; | |
1740 | } | |
fa490cfd | 1741 | |
3a5c359a AK |
1742 | /* |
1743 | * It's not enough that it's not actively running, | |
1744 | * it must be off the runqueue _entirely_, and not | |
1745 | * preempted! | |
1746 | * | |
1747 | * So if it wa still runnable (but just not actively | |
1748 | * running right now), it's preempted, and we should | |
1749 | * yield - it could be a while. | |
1750 | */ | |
1751 | if (unlikely(on_rq)) { | |
1752 | schedule_timeout_uninterruptible(1); | |
1753 | continue; | |
1754 | } | |
fa490cfd | 1755 | |
3a5c359a AK |
1756 | /* |
1757 | * Ahh, all good. It wasn't running, and it wasn't | |
1758 | * runnable, which means that it will never become | |
1759 | * running in the future either. We're all done! | |
1760 | */ | |
1761 | break; | |
1762 | } | |
1da177e4 LT |
1763 | } |
1764 | ||
1765 | /*** | |
1766 | * kick_process - kick a running thread to enter/exit the kernel | |
1767 | * @p: the to-be-kicked thread | |
1768 | * | |
1769 | * Cause a process which is running on another CPU to enter | |
1770 | * kernel-mode, without any delay. (to get signals handled.) | |
1771 | * | |
1772 | * NOTE: this function doesnt have to take the runqueue lock, | |
1773 | * because all it wants to ensure is that the remote task enters | |
1774 | * the kernel. If the IPI races and the task has been migrated | |
1775 | * to another CPU then no harm is done and the purpose has been | |
1776 | * achieved as well. | |
1777 | */ | |
36c8b586 | 1778 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1779 | { |
1780 | int cpu; | |
1781 | ||
1782 | preempt_disable(); | |
1783 | cpu = task_cpu(p); | |
1784 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1785 | smp_send_reschedule(cpu); | |
1786 | preempt_enable(); | |
1787 | } | |
1788 | ||
1789 | /* | |
2dd73a4f PW |
1790 | * Return a low guess at the load of a migration-source cpu weighted |
1791 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1792 | * |
1793 | * We want to under-estimate the load of migration sources, to | |
1794 | * balance conservatively. | |
1795 | */ | |
a9957449 | 1796 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1797 | { |
70b97a7f | 1798 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1799 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1800 | |
3b0bd9bc | 1801 | if (type == 0) |
dd41f596 | 1802 | return total; |
b910472d | 1803 | |
dd41f596 | 1804 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1805 | } |
1806 | ||
1807 | /* | |
2dd73a4f PW |
1808 | * Return a high guess at the load of a migration-target cpu weighted |
1809 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1810 | */ |
a9957449 | 1811 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1812 | { |
70b97a7f | 1813 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1814 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1815 | |
7897986b | 1816 | if (type == 0) |
dd41f596 | 1817 | return total; |
3b0bd9bc | 1818 | |
dd41f596 | 1819 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1820 | } |
1821 | ||
1822 | /* | |
1823 | * Return the average load per task on the cpu's run queue | |
1824 | */ | |
e7693a36 | 1825 | static unsigned long cpu_avg_load_per_task(int cpu) |
2dd73a4f | 1826 | { |
70b97a7f | 1827 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1828 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1829 | unsigned long n = rq->nr_running; |
1830 | ||
dd41f596 | 1831 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1832 | } |
1833 | ||
147cbb4b NP |
1834 | /* |
1835 | * find_idlest_group finds and returns the least busy CPU group within the | |
1836 | * domain. | |
1837 | */ | |
1838 | static struct sched_group * | |
1839 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1840 | { | |
1841 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1842 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1843 | int load_idx = sd->forkexec_idx; | |
1844 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1845 | ||
1846 | do { | |
1847 | unsigned long load, avg_load; | |
1848 | int local_group; | |
1849 | int i; | |
1850 | ||
da5a5522 BD |
1851 | /* Skip over this group if it has no CPUs allowed */ |
1852 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1853 | continue; |
da5a5522 | 1854 | |
147cbb4b | 1855 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1856 | |
1857 | /* Tally up the load of all CPUs in the group */ | |
1858 | avg_load = 0; | |
1859 | ||
1860 | for_each_cpu_mask(i, group->cpumask) { | |
1861 | /* Bias balancing toward cpus of our domain */ | |
1862 | if (local_group) | |
1863 | load = source_load(i, load_idx); | |
1864 | else | |
1865 | load = target_load(i, load_idx); | |
1866 | ||
1867 | avg_load += load; | |
1868 | } | |
1869 | ||
1870 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1871 | avg_load = sg_div_cpu_power(group, |
1872 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1873 | |
1874 | if (local_group) { | |
1875 | this_load = avg_load; | |
1876 | this = group; | |
1877 | } else if (avg_load < min_load) { | |
1878 | min_load = avg_load; | |
1879 | idlest = group; | |
1880 | } | |
3a5c359a | 1881 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1882 | |
1883 | if (!idlest || 100*this_load < imbalance*min_load) | |
1884 | return NULL; | |
1885 | return idlest; | |
1886 | } | |
1887 | ||
1888 | /* | |
0feaece9 | 1889 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1890 | */ |
95cdf3b7 | 1891 | static int |
7c16ec58 MT |
1892 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
1893 | cpumask_t *tmp) | |
147cbb4b NP |
1894 | { |
1895 | unsigned long load, min_load = ULONG_MAX; | |
1896 | int idlest = -1; | |
1897 | int i; | |
1898 | ||
da5a5522 | 1899 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 1900 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 1901 | |
7c16ec58 | 1902 | for_each_cpu_mask(i, *tmp) { |
2dd73a4f | 1903 | load = weighted_cpuload(i); |
147cbb4b NP |
1904 | |
1905 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1906 | min_load = load; | |
1907 | idlest = i; | |
1908 | } | |
1909 | } | |
1910 | ||
1911 | return idlest; | |
1912 | } | |
1913 | ||
476d139c NP |
1914 | /* |
1915 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1916 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1917 | * SD_BALANCE_EXEC. | |
1918 | * | |
1919 | * Balance, ie. select the least loaded group. | |
1920 | * | |
1921 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1922 | * | |
1923 | * preempt must be disabled. | |
1924 | */ | |
1925 | static int sched_balance_self(int cpu, int flag) | |
1926 | { | |
1927 | struct task_struct *t = current; | |
1928 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1929 | |
c96d145e | 1930 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1931 | /* |
1932 | * If power savings logic is enabled for a domain, stop there. | |
1933 | */ | |
5c45bf27 SS |
1934 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1935 | break; | |
476d139c NP |
1936 | if (tmp->flags & flag) |
1937 | sd = tmp; | |
c96d145e | 1938 | } |
476d139c NP |
1939 | |
1940 | while (sd) { | |
7c16ec58 | 1941 | cpumask_t span, tmpmask; |
476d139c | 1942 | struct sched_group *group; |
1a848870 SS |
1943 | int new_cpu, weight; |
1944 | ||
1945 | if (!(sd->flags & flag)) { | |
1946 | sd = sd->child; | |
1947 | continue; | |
1948 | } | |
476d139c NP |
1949 | |
1950 | span = sd->span; | |
1951 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1952 | if (!group) { |
1953 | sd = sd->child; | |
1954 | continue; | |
1955 | } | |
476d139c | 1956 | |
7c16ec58 | 1957 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
1958 | if (new_cpu == -1 || new_cpu == cpu) { |
1959 | /* Now try balancing at a lower domain level of cpu */ | |
1960 | sd = sd->child; | |
1961 | continue; | |
1962 | } | |
476d139c | 1963 | |
1a848870 | 1964 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1965 | cpu = new_cpu; |
476d139c NP |
1966 | sd = NULL; |
1967 | weight = cpus_weight(span); | |
1968 | for_each_domain(cpu, tmp) { | |
1969 | if (weight <= cpus_weight(tmp->span)) | |
1970 | break; | |
1971 | if (tmp->flags & flag) | |
1972 | sd = tmp; | |
1973 | } | |
1974 | /* while loop will break here if sd == NULL */ | |
1975 | } | |
1976 | ||
1977 | return cpu; | |
1978 | } | |
1979 | ||
1980 | #endif /* CONFIG_SMP */ | |
1da177e4 | 1981 | |
1da177e4 LT |
1982 | /*** |
1983 | * try_to_wake_up - wake up a thread | |
1984 | * @p: the to-be-woken-up thread | |
1985 | * @state: the mask of task states that can be woken | |
1986 | * @sync: do a synchronous wakeup? | |
1987 | * | |
1988 | * Put it on the run-queue if it's not already there. The "current" | |
1989 | * thread is always on the run-queue (except when the actual | |
1990 | * re-schedule is in progress), and as such you're allowed to do | |
1991 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1992 | * runnable without the overhead of this. | |
1993 | * | |
1994 | * returns failure only if the task is already active. | |
1995 | */ | |
36c8b586 | 1996 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1997 | { |
cc367732 | 1998 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1999 | unsigned long flags; |
2000 | long old_state; | |
70b97a7f | 2001 | struct rq *rq; |
1da177e4 | 2002 | |
b85d0667 IM |
2003 | if (!sched_feat(SYNC_WAKEUPS)) |
2004 | sync = 0; | |
2005 | ||
04e2f174 | 2006 | smp_wmb(); |
1da177e4 LT |
2007 | rq = task_rq_lock(p, &flags); |
2008 | old_state = p->state; | |
2009 | if (!(old_state & state)) | |
2010 | goto out; | |
2011 | ||
dd41f596 | 2012 | if (p->se.on_rq) |
1da177e4 LT |
2013 | goto out_running; |
2014 | ||
2015 | cpu = task_cpu(p); | |
cc367732 | 2016 | orig_cpu = cpu; |
1da177e4 LT |
2017 | this_cpu = smp_processor_id(); |
2018 | ||
2019 | #ifdef CONFIG_SMP | |
2020 | if (unlikely(task_running(rq, p))) | |
2021 | goto out_activate; | |
2022 | ||
5d2f5a61 DA |
2023 | cpu = p->sched_class->select_task_rq(p, sync); |
2024 | if (cpu != orig_cpu) { | |
2025 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2026 | task_rq_unlock(rq, &flags); |
2027 | /* might preempt at this point */ | |
2028 | rq = task_rq_lock(p, &flags); | |
2029 | old_state = p->state; | |
2030 | if (!(old_state & state)) | |
2031 | goto out; | |
dd41f596 | 2032 | if (p->se.on_rq) |
1da177e4 LT |
2033 | goto out_running; |
2034 | ||
2035 | this_cpu = smp_processor_id(); | |
2036 | cpu = task_cpu(p); | |
2037 | } | |
2038 | ||
e7693a36 GH |
2039 | #ifdef CONFIG_SCHEDSTATS |
2040 | schedstat_inc(rq, ttwu_count); | |
2041 | if (cpu == this_cpu) | |
2042 | schedstat_inc(rq, ttwu_local); | |
2043 | else { | |
2044 | struct sched_domain *sd; | |
2045 | for_each_domain(this_cpu, sd) { | |
2046 | if (cpu_isset(cpu, sd->span)) { | |
2047 | schedstat_inc(sd, ttwu_wake_remote); | |
2048 | break; | |
2049 | } | |
2050 | } | |
2051 | } | |
e7693a36 GH |
2052 | #endif |
2053 | ||
1da177e4 LT |
2054 | out_activate: |
2055 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2056 | schedstat_inc(p, se.nr_wakeups); |
2057 | if (sync) | |
2058 | schedstat_inc(p, se.nr_wakeups_sync); | |
2059 | if (orig_cpu != cpu) | |
2060 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2061 | if (cpu == this_cpu) | |
2062 | schedstat_inc(p, se.nr_wakeups_local); | |
2063 | else | |
2064 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2065 | update_rq_clock(rq); |
dd41f596 | 2066 | activate_task(rq, p, 1); |
1da177e4 LT |
2067 | success = 1; |
2068 | ||
2069 | out_running: | |
4ae7d5ce IM |
2070 | check_preempt_curr(rq, p); |
2071 | ||
1da177e4 | 2072 | p->state = TASK_RUNNING; |
9a897c5a SR |
2073 | #ifdef CONFIG_SMP |
2074 | if (p->sched_class->task_wake_up) | |
2075 | p->sched_class->task_wake_up(rq, p); | |
2076 | #endif | |
1da177e4 LT |
2077 | out: |
2078 | task_rq_unlock(rq, &flags); | |
2079 | ||
2080 | return success; | |
2081 | } | |
2082 | ||
7ad5b3a5 | 2083 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2084 | { |
d9514f6c | 2085 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2086 | } |
1da177e4 LT |
2087 | EXPORT_SYMBOL(wake_up_process); |
2088 | ||
7ad5b3a5 | 2089 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2090 | { |
2091 | return try_to_wake_up(p, state, 0); | |
2092 | } | |
2093 | ||
1da177e4 LT |
2094 | /* |
2095 | * Perform scheduler related setup for a newly forked process p. | |
2096 | * p is forked by current. | |
dd41f596 IM |
2097 | * |
2098 | * __sched_fork() is basic setup used by init_idle() too: | |
2099 | */ | |
2100 | static void __sched_fork(struct task_struct *p) | |
2101 | { | |
dd41f596 IM |
2102 | p->se.exec_start = 0; |
2103 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2104 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2105 | p->se.last_wakeup = 0; |
2106 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2107 | |
2108 | #ifdef CONFIG_SCHEDSTATS | |
2109 | p->se.wait_start = 0; | |
dd41f596 IM |
2110 | p->se.sum_sleep_runtime = 0; |
2111 | p->se.sleep_start = 0; | |
dd41f596 IM |
2112 | p->se.block_start = 0; |
2113 | p->se.sleep_max = 0; | |
2114 | p->se.block_max = 0; | |
2115 | p->se.exec_max = 0; | |
eba1ed4b | 2116 | p->se.slice_max = 0; |
dd41f596 | 2117 | p->se.wait_max = 0; |
6cfb0d5d | 2118 | #endif |
476d139c | 2119 | |
fa717060 | 2120 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2121 | p->se.on_rq = 0; |
476d139c | 2122 | |
e107be36 AK |
2123 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2124 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2125 | #endif | |
2126 | ||
1da177e4 LT |
2127 | /* |
2128 | * We mark the process as running here, but have not actually | |
2129 | * inserted it onto the runqueue yet. This guarantees that | |
2130 | * nobody will actually run it, and a signal or other external | |
2131 | * event cannot wake it up and insert it on the runqueue either. | |
2132 | */ | |
2133 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2134 | } |
2135 | ||
2136 | /* | |
2137 | * fork()/clone()-time setup: | |
2138 | */ | |
2139 | void sched_fork(struct task_struct *p, int clone_flags) | |
2140 | { | |
2141 | int cpu = get_cpu(); | |
2142 | ||
2143 | __sched_fork(p); | |
2144 | ||
2145 | #ifdef CONFIG_SMP | |
2146 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2147 | #endif | |
02e4bac2 | 2148 | set_task_cpu(p, cpu); |
b29739f9 IM |
2149 | |
2150 | /* | |
2151 | * Make sure we do not leak PI boosting priority to the child: | |
2152 | */ | |
2153 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2154 | if (!rt_prio(p->prio)) |
2155 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2156 | |
52f17b6c | 2157 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2158 | if (likely(sched_info_on())) |
52f17b6c | 2159 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2160 | #endif |
d6077cb8 | 2161 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2162 | p->oncpu = 0; |
2163 | #endif | |
1da177e4 | 2164 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2165 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2166 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2167 | #endif |
476d139c | 2168 | put_cpu(); |
1da177e4 LT |
2169 | } |
2170 | ||
2171 | /* | |
2172 | * wake_up_new_task - wake up a newly created task for the first time. | |
2173 | * | |
2174 | * This function will do some initial scheduler statistics housekeeping | |
2175 | * that must be done for every newly created context, then puts the task | |
2176 | * on the runqueue and wakes it. | |
2177 | */ | |
7ad5b3a5 | 2178 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2179 | { |
2180 | unsigned long flags; | |
dd41f596 | 2181 | struct rq *rq; |
1da177e4 LT |
2182 | |
2183 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2184 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2185 | update_rq_clock(rq); |
1da177e4 LT |
2186 | |
2187 | p->prio = effective_prio(p); | |
2188 | ||
b9dca1e0 | 2189 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2190 | activate_task(rq, p, 0); |
1da177e4 | 2191 | } else { |
1da177e4 | 2192 | /* |
dd41f596 IM |
2193 | * Let the scheduling class do new task startup |
2194 | * management (if any): | |
1da177e4 | 2195 | */ |
ee0827d8 | 2196 | p->sched_class->task_new(rq, p); |
62fb1851 | 2197 | inc_nr_running(p, rq); |
1da177e4 | 2198 | } |
dd41f596 | 2199 | check_preempt_curr(rq, p); |
9a897c5a SR |
2200 | #ifdef CONFIG_SMP |
2201 | if (p->sched_class->task_wake_up) | |
2202 | p->sched_class->task_wake_up(rq, p); | |
2203 | #endif | |
dd41f596 | 2204 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2205 | } |
2206 | ||
e107be36 AK |
2207 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2208 | ||
2209 | /** | |
421cee29 RD |
2210 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2211 | * @notifier: notifier struct to register | |
e107be36 AK |
2212 | */ |
2213 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2214 | { | |
2215 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2216 | } | |
2217 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2218 | ||
2219 | /** | |
2220 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2221 | * @notifier: notifier struct to unregister |
e107be36 AK |
2222 | * |
2223 | * This is safe to call from within a preemption notifier. | |
2224 | */ | |
2225 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2226 | { | |
2227 | hlist_del(¬ifier->link); | |
2228 | } | |
2229 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2230 | ||
2231 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2232 | { | |
2233 | struct preempt_notifier *notifier; | |
2234 | struct hlist_node *node; | |
2235 | ||
2236 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2237 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2238 | } | |
2239 | ||
2240 | static void | |
2241 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2242 | struct task_struct *next) | |
2243 | { | |
2244 | struct preempt_notifier *notifier; | |
2245 | struct hlist_node *node; | |
2246 | ||
2247 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2248 | notifier->ops->sched_out(notifier, next); | |
2249 | } | |
2250 | ||
2251 | #else | |
2252 | ||
2253 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2254 | { | |
2255 | } | |
2256 | ||
2257 | static void | |
2258 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2259 | struct task_struct *next) | |
2260 | { | |
2261 | } | |
2262 | ||
2263 | #endif | |
2264 | ||
4866cde0 NP |
2265 | /** |
2266 | * prepare_task_switch - prepare to switch tasks | |
2267 | * @rq: the runqueue preparing to switch | |
421cee29 | 2268 | * @prev: the current task that is being switched out |
4866cde0 NP |
2269 | * @next: the task we are going to switch to. |
2270 | * | |
2271 | * This is called with the rq lock held and interrupts off. It must | |
2272 | * be paired with a subsequent finish_task_switch after the context | |
2273 | * switch. | |
2274 | * | |
2275 | * prepare_task_switch sets up locking and calls architecture specific | |
2276 | * hooks. | |
2277 | */ | |
e107be36 AK |
2278 | static inline void |
2279 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2280 | struct task_struct *next) | |
4866cde0 | 2281 | { |
e107be36 | 2282 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2283 | prepare_lock_switch(rq, next); |
2284 | prepare_arch_switch(next); | |
2285 | } | |
2286 | ||
1da177e4 LT |
2287 | /** |
2288 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2289 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2290 | * @prev: the thread we just switched away from. |
2291 | * | |
4866cde0 NP |
2292 | * finish_task_switch must be called after the context switch, paired |
2293 | * with a prepare_task_switch call before the context switch. | |
2294 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2295 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2296 | * |
2297 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2298 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2299 | * with the lock held can cause deadlocks; see schedule() for |
2300 | * details.) | |
2301 | */ | |
a9957449 | 2302 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2303 | __releases(rq->lock) |
2304 | { | |
1da177e4 | 2305 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2306 | long prev_state; |
1da177e4 LT |
2307 | |
2308 | rq->prev_mm = NULL; | |
2309 | ||
2310 | /* | |
2311 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2312 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2313 | * schedule one last time. The schedule call will never return, and |
2314 | * the scheduled task must drop that reference. | |
c394cc9f | 2315 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2316 | * still held, otherwise prev could be scheduled on another cpu, die |
2317 | * there before we look at prev->state, and then the reference would | |
2318 | * be dropped twice. | |
2319 | * Manfred Spraul <manfred@colorfullife.com> | |
2320 | */ | |
55a101f8 | 2321 | prev_state = prev->state; |
4866cde0 NP |
2322 | finish_arch_switch(prev); |
2323 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2324 | #ifdef CONFIG_SMP |
2325 | if (current->sched_class->post_schedule) | |
2326 | current->sched_class->post_schedule(rq); | |
2327 | #endif | |
e8fa1362 | 2328 | |
e107be36 | 2329 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2330 | if (mm) |
2331 | mmdrop(mm); | |
c394cc9f | 2332 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2333 | /* |
2334 | * Remove function-return probe instances associated with this | |
2335 | * task and put them back on the free list. | |
9761eea8 | 2336 | */ |
c6fd91f0 | 2337 | kprobe_flush_task(prev); |
1da177e4 | 2338 | put_task_struct(prev); |
c6fd91f0 | 2339 | } |
1da177e4 LT |
2340 | } |
2341 | ||
2342 | /** | |
2343 | * schedule_tail - first thing a freshly forked thread must call. | |
2344 | * @prev: the thread we just switched away from. | |
2345 | */ | |
36c8b586 | 2346 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2347 | __releases(rq->lock) |
2348 | { | |
70b97a7f IM |
2349 | struct rq *rq = this_rq(); |
2350 | ||
4866cde0 NP |
2351 | finish_task_switch(rq, prev); |
2352 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2353 | /* In this case, finish_task_switch does not reenable preemption */ | |
2354 | preempt_enable(); | |
2355 | #endif | |
1da177e4 | 2356 | if (current->set_child_tid) |
b488893a | 2357 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2358 | } |
2359 | ||
2360 | /* | |
2361 | * context_switch - switch to the new MM and the new | |
2362 | * thread's register state. | |
2363 | */ | |
dd41f596 | 2364 | static inline void |
70b97a7f | 2365 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2366 | struct task_struct *next) |
1da177e4 | 2367 | { |
dd41f596 | 2368 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2369 | |
e107be36 | 2370 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
2371 | mm = next->mm; |
2372 | oldmm = prev->active_mm; | |
9226d125 ZA |
2373 | /* |
2374 | * For paravirt, this is coupled with an exit in switch_to to | |
2375 | * combine the page table reload and the switch backend into | |
2376 | * one hypercall. | |
2377 | */ | |
2378 | arch_enter_lazy_cpu_mode(); | |
2379 | ||
dd41f596 | 2380 | if (unlikely(!mm)) { |
1da177e4 LT |
2381 | next->active_mm = oldmm; |
2382 | atomic_inc(&oldmm->mm_count); | |
2383 | enter_lazy_tlb(oldmm, next); | |
2384 | } else | |
2385 | switch_mm(oldmm, mm, next); | |
2386 | ||
dd41f596 | 2387 | if (unlikely(!prev->mm)) { |
1da177e4 | 2388 | prev->active_mm = NULL; |
1da177e4 LT |
2389 | rq->prev_mm = oldmm; |
2390 | } | |
3a5f5e48 IM |
2391 | /* |
2392 | * Since the runqueue lock will be released by the next | |
2393 | * task (which is an invalid locking op but in the case | |
2394 | * of the scheduler it's an obvious special-case), so we | |
2395 | * do an early lockdep release here: | |
2396 | */ | |
2397 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2398 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2399 | #endif |
1da177e4 LT |
2400 | |
2401 | /* Here we just switch the register state and the stack. */ | |
2402 | switch_to(prev, next, prev); | |
2403 | ||
dd41f596 IM |
2404 | barrier(); |
2405 | /* | |
2406 | * this_rq must be evaluated again because prev may have moved | |
2407 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2408 | * frame will be invalid. | |
2409 | */ | |
2410 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2411 | } |
2412 | ||
2413 | /* | |
2414 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2415 | * | |
2416 | * externally visible scheduler statistics: current number of runnable | |
2417 | * threads, current number of uninterruptible-sleeping threads, total | |
2418 | * number of context switches performed since bootup. | |
2419 | */ | |
2420 | unsigned long nr_running(void) | |
2421 | { | |
2422 | unsigned long i, sum = 0; | |
2423 | ||
2424 | for_each_online_cpu(i) | |
2425 | sum += cpu_rq(i)->nr_running; | |
2426 | ||
2427 | return sum; | |
2428 | } | |
2429 | ||
2430 | unsigned long nr_uninterruptible(void) | |
2431 | { | |
2432 | unsigned long i, sum = 0; | |
2433 | ||
0a945022 | 2434 | for_each_possible_cpu(i) |
1da177e4 LT |
2435 | sum += cpu_rq(i)->nr_uninterruptible; |
2436 | ||
2437 | /* | |
2438 | * Since we read the counters lockless, it might be slightly | |
2439 | * inaccurate. Do not allow it to go below zero though: | |
2440 | */ | |
2441 | if (unlikely((long)sum < 0)) | |
2442 | sum = 0; | |
2443 | ||
2444 | return sum; | |
2445 | } | |
2446 | ||
2447 | unsigned long long nr_context_switches(void) | |
2448 | { | |
cc94abfc SR |
2449 | int i; |
2450 | unsigned long long sum = 0; | |
1da177e4 | 2451 | |
0a945022 | 2452 | for_each_possible_cpu(i) |
1da177e4 LT |
2453 | sum += cpu_rq(i)->nr_switches; |
2454 | ||
2455 | return sum; | |
2456 | } | |
2457 | ||
2458 | unsigned long nr_iowait(void) | |
2459 | { | |
2460 | unsigned long i, sum = 0; | |
2461 | ||
0a945022 | 2462 | for_each_possible_cpu(i) |
1da177e4 LT |
2463 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2464 | ||
2465 | return sum; | |
2466 | } | |
2467 | ||
db1b1fef JS |
2468 | unsigned long nr_active(void) |
2469 | { | |
2470 | unsigned long i, running = 0, uninterruptible = 0; | |
2471 | ||
2472 | for_each_online_cpu(i) { | |
2473 | running += cpu_rq(i)->nr_running; | |
2474 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2475 | } | |
2476 | ||
2477 | if (unlikely((long)uninterruptible < 0)) | |
2478 | uninterruptible = 0; | |
2479 | ||
2480 | return running + uninterruptible; | |
2481 | } | |
2482 | ||
48f24c4d | 2483 | /* |
dd41f596 IM |
2484 | * Update rq->cpu_load[] statistics. This function is usually called every |
2485 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2486 | */ |
dd41f596 | 2487 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2488 | { |
495eca49 | 2489 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2490 | int i, scale; |
2491 | ||
2492 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2493 | |
2494 | /* Update our load: */ | |
2495 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2496 | unsigned long old_load, new_load; | |
2497 | ||
2498 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2499 | ||
2500 | old_load = this_rq->cpu_load[i]; | |
2501 | new_load = this_load; | |
a25707f3 IM |
2502 | /* |
2503 | * Round up the averaging division if load is increasing. This | |
2504 | * prevents us from getting stuck on 9 if the load is 10, for | |
2505 | * example. | |
2506 | */ | |
2507 | if (new_load > old_load) | |
2508 | new_load += scale-1; | |
dd41f596 IM |
2509 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2510 | } | |
48f24c4d IM |
2511 | } |
2512 | ||
dd41f596 IM |
2513 | #ifdef CONFIG_SMP |
2514 | ||
1da177e4 LT |
2515 | /* |
2516 | * double_rq_lock - safely lock two runqueues | |
2517 | * | |
2518 | * Note this does not disable interrupts like task_rq_lock, | |
2519 | * you need to do so manually before calling. | |
2520 | */ | |
70b97a7f | 2521 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2522 | __acquires(rq1->lock) |
2523 | __acquires(rq2->lock) | |
2524 | { | |
054b9108 | 2525 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2526 | if (rq1 == rq2) { |
2527 | spin_lock(&rq1->lock); | |
2528 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2529 | } else { | |
c96d145e | 2530 | if (rq1 < rq2) { |
1da177e4 LT |
2531 | spin_lock(&rq1->lock); |
2532 | spin_lock(&rq2->lock); | |
2533 | } else { | |
2534 | spin_lock(&rq2->lock); | |
2535 | spin_lock(&rq1->lock); | |
2536 | } | |
2537 | } | |
6e82a3be IM |
2538 | update_rq_clock(rq1); |
2539 | update_rq_clock(rq2); | |
1da177e4 LT |
2540 | } |
2541 | ||
2542 | /* | |
2543 | * double_rq_unlock - safely unlock two runqueues | |
2544 | * | |
2545 | * Note this does not restore interrupts like task_rq_unlock, | |
2546 | * you need to do so manually after calling. | |
2547 | */ | |
70b97a7f | 2548 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2549 | __releases(rq1->lock) |
2550 | __releases(rq2->lock) | |
2551 | { | |
2552 | spin_unlock(&rq1->lock); | |
2553 | if (rq1 != rq2) | |
2554 | spin_unlock(&rq2->lock); | |
2555 | else | |
2556 | __release(rq2->lock); | |
2557 | } | |
2558 | ||
2559 | /* | |
2560 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2561 | */ | |
e8fa1362 | 2562 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2563 | __releases(this_rq->lock) |
2564 | __acquires(busiest->lock) | |
2565 | __acquires(this_rq->lock) | |
2566 | { | |
e8fa1362 SR |
2567 | int ret = 0; |
2568 | ||
054b9108 KK |
2569 | if (unlikely(!irqs_disabled())) { |
2570 | /* printk() doesn't work good under rq->lock */ | |
2571 | spin_unlock(&this_rq->lock); | |
2572 | BUG_ON(1); | |
2573 | } | |
1da177e4 | 2574 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2575 | if (busiest < this_rq) { |
1da177e4 LT |
2576 | spin_unlock(&this_rq->lock); |
2577 | spin_lock(&busiest->lock); | |
2578 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2579 | ret = 1; |
1da177e4 LT |
2580 | } else |
2581 | spin_lock(&busiest->lock); | |
2582 | } | |
e8fa1362 | 2583 | return ret; |
1da177e4 LT |
2584 | } |
2585 | ||
1da177e4 LT |
2586 | /* |
2587 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2588 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2589 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2590 | * the cpu_allowed mask is restored. |
2591 | */ | |
36c8b586 | 2592 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2593 | { |
70b97a7f | 2594 | struct migration_req req; |
1da177e4 | 2595 | unsigned long flags; |
70b97a7f | 2596 | struct rq *rq; |
1da177e4 LT |
2597 | |
2598 | rq = task_rq_lock(p, &flags); | |
2599 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2600 | || unlikely(cpu_is_offline(dest_cpu))) | |
2601 | goto out; | |
2602 | ||
2603 | /* force the process onto the specified CPU */ | |
2604 | if (migrate_task(p, dest_cpu, &req)) { | |
2605 | /* Need to wait for migration thread (might exit: take ref). */ | |
2606 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2607 | |
1da177e4 LT |
2608 | get_task_struct(mt); |
2609 | task_rq_unlock(rq, &flags); | |
2610 | wake_up_process(mt); | |
2611 | put_task_struct(mt); | |
2612 | wait_for_completion(&req.done); | |
36c8b586 | 2613 | |
1da177e4 LT |
2614 | return; |
2615 | } | |
2616 | out: | |
2617 | task_rq_unlock(rq, &flags); | |
2618 | } | |
2619 | ||
2620 | /* | |
476d139c NP |
2621 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2622 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2623 | */ |
2624 | void sched_exec(void) | |
2625 | { | |
1da177e4 | 2626 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2627 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2628 | put_cpu(); |
476d139c NP |
2629 | if (new_cpu != this_cpu) |
2630 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2631 | } |
2632 | ||
2633 | /* | |
2634 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2635 | * Both runqueues must be locked. | |
2636 | */ | |
dd41f596 IM |
2637 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2638 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2639 | { |
2e1cb74a | 2640 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2641 | set_task_cpu(p, this_cpu); |
dd41f596 | 2642 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2643 | /* |
2644 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2645 | * to be always true for them. | |
2646 | */ | |
dd41f596 | 2647 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2648 | } |
2649 | ||
2650 | /* | |
2651 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2652 | */ | |
858119e1 | 2653 | static |
70b97a7f | 2654 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2655 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2656 | int *all_pinned) |
1da177e4 LT |
2657 | { |
2658 | /* | |
2659 | * We do not migrate tasks that are: | |
2660 | * 1) running (obviously), or | |
2661 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2662 | * 3) are cache-hot on their current CPU. | |
2663 | */ | |
cc367732 IM |
2664 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2665 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2666 | return 0; |
cc367732 | 2667 | } |
81026794 NP |
2668 | *all_pinned = 0; |
2669 | ||
cc367732 IM |
2670 | if (task_running(rq, p)) { |
2671 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2672 | return 0; |
cc367732 | 2673 | } |
1da177e4 | 2674 | |
da84d961 IM |
2675 | /* |
2676 | * Aggressive migration if: | |
2677 | * 1) task is cache cold, or | |
2678 | * 2) too many balance attempts have failed. | |
2679 | */ | |
2680 | ||
6bc1665b IM |
2681 | if (!task_hot(p, rq->clock, sd) || |
2682 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2683 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2684 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2685 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2686 | schedstat_inc(p, se.nr_forced_migrations); |
2687 | } | |
da84d961 IM |
2688 | #endif |
2689 | return 1; | |
2690 | } | |
2691 | ||
cc367732 IM |
2692 | if (task_hot(p, rq->clock, sd)) { |
2693 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2694 | return 0; |
cc367732 | 2695 | } |
1da177e4 LT |
2696 | return 1; |
2697 | } | |
2698 | ||
e1d1484f PW |
2699 | static unsigned long |
2700 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2701 | unsigned long max_load_move, struct sched_domain *sd, | |
2702 | enum cpu_idle_type idle, int *all_pinned, | |
2703 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2704 | { |
b82d9fdd | 2705 | int loops = 0, pulled = 0, pinned = 0, skip_for_load; |
dd41f596 IM |
2706 | struct task_struct *p; |
2707 | long rem_load_move = max_load_move; | |
1da177e4 | 2708 | |
e1d1484f | 2709 | if (max_load_move == 0) |
1da177e4 LT |
2710 | goto out; |
2711 | ||
81026794 NP |
2712 | pinned = 1; |
2713 | ||
1da177e4 | 2714 | /* |
dd41f596 | 2715 | * Start the load-balancing iterator: |
1da177e4 | 2716 | */ |
dd41f596 IM |
2717 | p = iterator->start(iterator->arg); |
2718 | next: | |
b82d9fdd | 2719 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2720 | goto out; |
50ddd969 | 2721 | /* |
b82d9fdd | 2722 | * To help distribute high priority tasks across CPUs we don't |
50ddd969 PW |
2723 | * skip a task if it will be the highest priority task (i.e. smallest |
2724 | * prio value) on its new queue regardless of its load weight | |
2725 | */ | |
dd41f596 IM |
2726 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2727 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2728 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2729 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2730 | p = iterator->next(iterator->arg); |
2731 | goto next; | |
1da177e4 LT |
2732 | } |
2733 | ||
dd41f596 | 2734 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2735 | pulled++; |
dd41f596 | 2736 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2737 | |
2dd73a4f | 2738 | /* |
b82d9fdd | 2739 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2740 | */ |
e1d1484f | 2741 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2742 | if (p->prio < *this_best_prio) |
2743 | *this_best_prio = p->prio; | |
dd41f596 IM |
2744 | p = iterator->next(iterator->arg); |
2745 | goto next; | |
1da177e4 LT |
2746 | } |
2747 | out: | |
2748 | /* | |
e1d1484f | 2749 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2750 | * so we can safely collect pull_task() stats here rather than |
2751 | * inside pull_task(). | |
2752 | */ | |
2753 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2754 | |
2755 | if (all_pinned) | |
2756 | *all_pinned = pinned; | |
e1d1484f PW |
2757 | |
2758 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2759 | } |
2760 | ||
dd41f596 | 2761 | /* |
43010659 PW |
2762 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2763 | * this_rq, as part of a balancing operation within domain "sd". | |
2764 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2765 | * |
2766 | * Called with both runqueues locked. | |
2767 | */ | |
2768 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2769 | unsigned long max_load_move, |
dd41f596 IM |
2770 | struct sched_domain *sd, enum cpu_idle_type idle, |
2771 | int *all_pinned) | |
2772 | { | |
5522d5d5 | 2773 | const struct sched_class *class = sched_class_highest; |
43010659 | 2774 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2775 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2776 | |
2777 | do { | |
43010659 PW |
2778 | total_load_moved += |
2779 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2780 | max_load_move - total_load_moved, |
a4ac01c3 | 2781 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2782 | class = class->next; |
43010659 | 2783 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2784 | |
43010659 PW |
2785 | return total_load_moved > 0; |
2786 | } | |
2787 | ||
e1d1484f PW |
2788 | static int |
2789 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2790 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2791 | struct rq_iterator *iterator) | |
2792 | { | |
2793 | struct task_struct *p = iterator->start(iterator->arg); | |
2794 | int pinned = 0; | |
2795 | ||
2796 | while (p) { | |
2797 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
2798 | pull_task(busiest, p, this_rq, this_cpu); | |
2799 | /* | |
2800 | * Right now, this is only the second place pull_task() | |
2801 | * is called, so we can safely collect pull_task() | |
2802 | * stats here rather than inside pull_task(). | |
2803 | */ | |
2804 | schedstat_inc(sd, lb_gained[idle]); | |
2805 | ||
2806 | return 1; | |
2807 | } | |
2808 | p = iterator->next(iterator->arg); | |
2809 | } | |
2810 | ||
2811 | return 0; | |
2812 | } | |
2813 | ||
43010659 PW |
2814 | /* |
2815 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2816 | * part of active balancing operations within "domain". | |
2817 | * Returns 1 if successful and 0 otherwise. | |
2818 | * | |
2819 | * Called with both runqueues locked. | |
2820 | */ | |
2821 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2822 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2823 | { | |
5522d5d5 | 2824 | const struct sched_class *class; |
43010659 PW |
2825 | |
2826 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 2827 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
2828 | return 1; |
2829 | ||
2830 | return 0; | |
dd41f596 IM |
2831 | } |
2832 | ||
1da177e4 LT |
2833 | /* |
2834 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2835 | * domain. It calculates and returns the amount of weighted load which |
2836 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2837 | */ |
2838 | static struct sched_group * | |
2839 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 2840 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 2841 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
2842 | { |
2843 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2844 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2845 | unsigned long max_pull; |
2dd73a4f PW |
2846 | unsigned long busiest_load_per_task, busiest_nr_running; |
2847 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2848 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2849 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2850 | int power_savings_balance = 1; | |
2851 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2852 | unsigned long min_nr_running = ULONG_MAX; | |
2853 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2854 | #endif | |
1da177e4 LT |
2855 | |
2856 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2857 | busiest_load_per_task = busiest_nr_running = 0; |
2858 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2859 | if (idle == CPU_NOT_IDLE) |
7897986b | 2860 | load_idx = sd->busy_idx; |
d15bcfdb | 2861 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2862 | load_idx = sd->newidle_idx; |
2863 | else | |
2864 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2865 | |
2866 | do { | |
908a7c1b | 2867 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2868 | int local_group; |
2869 | int i; | |
908a7c1b | 2870 | int __group_imb = 0; |
783609c6 | 2871 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2872 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2873 | |
2874 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2875 | ||
783609c6 SS |
2876 | if (local_group) |
2877 | balance_cpu = first_cpu(group->cpumask); | |
2878 | ||
1da177e4 | 2879 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2880 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2881 | max_cpu_load = 0; |
2882 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2883 | |
2884 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2885 | struct rq *rq; |
2886 | ||
2887 | if (!cpu_isset(i, *cpus)) | |
2888 | continue; | |
2889 | ||
2890 | rq = cpu_rq(i); | |
2dd73a4f | 2891 | |
9439aab8 | 2892 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2893 | *sd_idle = 0; |
2894 | ||
1da177e4 | 2895 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2896 | if (local_group) { |
2897 | if (idle_cpu(i) && !first_idle_cpu) { | |
2898 | first_idle_cpu = 1; | |
2899 | balance_cpu = i; | |
2900 | } | |
2901 | ||
a2000572 | 2902 | load = target_load(i, load_idx); |
908a7c1b | 2903 | } else { |
a2000572 | 2904 | load = source_load(i, load_idx); |
908a7c1b KC |
2905 | if (load > max_cpu_load) |
2906 | max_cpu_load = load; | |
2907 | if (min_cpu_load > load) | |
2908 | min_cpu_load = load; | |
2909 | } | |
1da177e4 LT |
2910 | |
2911 | avg_load += load; | |
2dd73a4f | 2912 | sum_nr_running += rq->nr_running; |
dd41f596 | 2913 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2914 | } |
2915 | ||
783609c6 SS |
2916 | /* |
2917 | * First idle cpu or the first cpu(busiest) in this sched group | |
2918 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2919 | * domains. In the newly idle case, we will allow all the cpu's |
2920 | * to do the newly idle load balance. | |
783609c6 | 2921 | */ |
9439aab8 SS |
2922 | if (idle != CPU_NEWLY_IDLE && local_group && |
2923 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2924 | *balance = 0; |
2925 | goto ret; | |
2926 | } | |
2927 | ||
1da177e4 | 2928 | total_load += avg_load; |
5517d86b | 2929 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2930 | |
2931 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2932 | avg_load = sg_div_cpu_power(group, |
2933 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2934 | |
908a7c1b KC |
2935 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2936 | __group_imb = 1; | |
2937 | ||
5517d86b | 2938 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2939 | |
1da177e4 LT |
2940 | if (local_group) { |
2941 | this_load = avg_load; | |
2942 | this = group; | |
2dd73a4f PW |
2943 | this_nr_running = sum_nr_running; |
2944 | this_load_per_task = sum_weighted_load; | |
2945 | } else if (avg_load > max_load && | |
908a7c1b | 2946 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2947 | max_load = avg_load; |
2948 | busiest = group; | |
2dd73a4f PW |
2949 | busiest_nr_running = sum_nr_running; |
2950 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2951 | group_imb = __group_imb; |
1da177e4 | 2952 | } |
5c45bf27 SS |
2953 | |
2954 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2955 | /* | |
2956 | * Busy processors will not participate in power savings | |
2957 | * balance. | |
2958 | */ | |
dd41f596 IM |
2959 | if (idle == CPU_NOT_IDLE || |
2960 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2961 | goto group_next; | |
5c45bf27 SS |
2962 | |
2963 | /* | |
2964 | * If the local group is idle or completely loaded | |
2965 | * no need to do power savings balance at this domain | |
2966 | */ | |
2967 | if (local_group && (this_nr_running >= group_capacity || | |
2968 | !this_nr_running)) | |
2969 | power_savings_balance = 0; | |
2970 | ||
dd41f596 | 2971 | /* |
5c45bf27 SS |
2972 | * If a group is already running at full capacity or idle, |
2973 | * don't include that group in power savings calculations | |
dd41f596 IM |
2974 | */ |
2975 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2976 | || !sum_nr_running) |
dd41f596 | 2977 | goto group_next; |
5c45bf27 | 2978 | |
dd41f596 | 2979 | /* |
5c45bf27 | 2980 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2981 | * This is the group from where we need to pick up the load |
2982 | * for saving power | |
2983 | */ | |
2984 | if ((sum_nr_running < min_nr_running) || | |
2985 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2986 | first_cpu(group->cpumask) < |
2987 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2988 | group_min = group; |
2989 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2990 | min_load_per_task = sum_weighted_load / |
2991 | sum_nr_running; | |
dd41f596 | 2992 | } |
5c45bf27 | 2993 | |
dd41f596 | 2994 | /* |
5c45bf27 | 2995 | * Calculate the group which is almost near its |
dd41f596 IM |
2996 | * capacity but still has some space to pick up some load |
2997 | * from other group and save more power | |
2998 | */ | |
2999 | if (sum_nr_running <= group_capacity - 1) { | |
3000 | if (sum_nr_running > leader_nr_running || | |
3001 | (sum_nr_running == leader_nr_running && | |
3002 | first_cpu(group->cpumask) > | |
3003 | first_cpu(group_leader->cpumask))) { | |
3004 | group_leader = group; | |
3005 | leader_nr_running = sum_nr_running; | |
3006 | } | |
48f24c4d | 3007 | } |
5c45bf27 SS |
3008 | group_next: |
3009 | #endif | |
1da177e4 LT |
3010 | group = group->next; |
3011 | } while (group != sd->groups); | |
3012 | ||
2dd73a4f | 3013 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3014 | goto out_balanced; |
3015 | ||
3016 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3017 | ||
3018 | if (this_load >= avg_load || | |
3019 | 100*max_load <= sd->imbalance_pct*this_load) | |
3020 | goto out_balanced; | |
3021 | ||
2dd73a4f | 3022 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3023 | if (group_imb) |
3024 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3025 | ||
1da177e4 LT |
3026 | /* |
3027 | * We're trying to get all the cpus to the average_load, so we don't | |
3028 | * want to push ourselves above the average load, nor do we wish to | |
3029 | * reduce the max loaded cpu below the average load, as either of these | |
3030 | * actions would just result in more rebalancing later, and ping-pong | |
3031 | * tasks around. Thus we look for the minimum possible imbalance. | |
3032 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3033 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3034 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3035 | * appear as very large values with unsigned longs. |
3036 | */ | |
2dd73a4f PW |
3037 | if (max_load <= busiest_load_per_task) |
3038 | goto out_balanced; | |
3039 | ||
3040 | /* | |
3041 | * In the presence of smp nice balancing, certain scenarios can have | |
3042 | * max load less than avg load(as we skip the groups at or below | |
3043 | * its cpu_power, while calculating max_load..) | |
3044 | */ | |
3045 | if (max_load < avg_load) { | |
3046 | *imbalance = 0; | |
3047 | goto small_imbalance; | |
3048 | } | |
0c117f1b SS |
3049 | |
3050 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3051 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3052 | |
1da177e4 | 3053 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3054 | *imbalance = min(max_pull * busiest->__cpu_power, |
3055 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3056 | / SCHED_LOAD_SCALE; |
3057 | ||
2dd73a4f PW |
3058 | /* |
3059 | * if *imbalance is less than the average load per runnable task | |
3060 | * there is no gaurantee that any tasks will be moved so we'll have | |
3061 | * a think about bumping its value to force at least one task to be | |
3062 | * moved | |
3063 | */ | |
7fd0d2dd | 3064 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3065 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3066 | unsigned int imbn; |
3067 | ||
3068 | small_imbalance: | |
3069 | pwr_move = pwr_now = 0; | |
3070 | imbn = 2; | |
3071 | if (this_nr_running) { | |
3072 | this_load_per_task /= this_nr_running; | |
3073 | if (busiest_load_per_task > this_load_per_task) | |
3074 | imbn = 1; | |
3075 | } else | |
3076 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 3077 | |
dd41f596 IM |
3078 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
3079 | busiest_load_per_task * imbn) { | |
2dd73a4f | 3080 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3081 | return busiest; |
3082 | } | |
3083 | ||
3084 | /* | |
3085 | * OK, we don't have enough imbalance to justify moving tasks, | |
3086 | * however we may be able to increase total CPU power used by | |
3087 | * moving them. | |
3088 | */ | |
3089 | ||
5517d86b ED |
3090 | pwr_now += busiest->__cpu_power * |
3091 | min(busiest_load_per_task, max_load); | |
3092 | pwr_now += this->__cpu_power * | |
3093 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3094 | pwr_now /= SCHED_LOAD_SCALE; |
3095 | ||
3096 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3097 | tmp = sg_div_cpu_power(busiest, |
3098 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3099 | if (max_load > tmp) |
5517d86b | 3100 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3101 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3102 | |
3103 | /* Amount of load we'd add */ | |
5517d86b | 3104 | if (max_load * busiest->__cpu_power < |
33859f7f | 3105 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3106 | tmp = sg_div_cpu_power(this, |
3107 | max_load * busiest->__cpu_power); | |
1da177e4 | 3108 | else |
5517d86b ED |
3109 | tmp = sg_div_cpu_power(this, |
3110 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3111 | pwr_move += this->__cpu_power * | |
3112 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3113 | pwr_move /= SCHED_LOAD_SCALE; |
3114 | ||
3115 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3116 | if (pwr_move > pwr_now) |
3117 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3118 | } |
3119 | ||
1da177e4 LT |
3120 | return busiest; |
3121 | ||
3122 | out_balanced: | |
5c45bf27 | 3123 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3124 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3125 | goto ret; |
1da177e4 | 3126 | |
5c45bf27 SS |
3127 | if (this == group_leader && group_leader != group_min) { |
3128 | *imbalance = min_load_per_task; | |
3129 | return group_min; | |
3130 | } | |
5c45bf27 | 3131 | #endif |
783609c6 | 3132 | ret: |
1da177e4 LT |
3133 | *imbalance = 0; |
3134 | return NULL; | |
3135 | } | |
3136 | ||
3137 | /* | |
3138 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3139 | */ | |
70b97a7f | 3140 | static struct rq * |
d15bcfdb | 3141 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3142 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3143 | { |
70b97a7f | 3144 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3145 | unsigned long max_load = 0; |
1da177e4 LT |
3146 | int i; |
3147 | ||
3148 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 3149 | unsigned long wl; |
0a2966b4 CL |
3150 | |
3151 | if (!cpu_isset(i, *cpus)) | |
3152 | continue; | |
3153 | ||
48f24c4d | 3154 | rq = cpu_rq(i); |
dd41f596 | 3155 | wl = weighted_cpuload(i); |
2dd73a4f | 3156 | |
dd41f596 | 3157 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3158 | continue; |
1da177e4 | 3159 | |
dd41f596 IM |
3160 | if (wl > max_load) { |
3161 | max_load = wl; | |
48f24c4d | 3162 | busiest = rq; |
1da177e4 LT |
3163 | } |
3164 | } | |
3165 | ||
3166 | return busiest; | |
3167 | } | |
3168 | ||
77391d71 NP |
3169 | /* |
3170 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3171 | * so long as it is large enough. | |
3172 | */ | |
3173 | #define MAX_PINNED_INTERVAL 512 | |
3174 | ||
1da177e4 LT |
3175 | /* |
3176 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3177 | * tasks if there is an imbalance. | |
1da177e4 | 3178 | */ |
70b97a7f | 3179 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3180 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3181 | int *balance, cpumask_t *cpus) |
1da177e4 | 3182 | { |
43010659 | 3183 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3184 | struct sched_group *group; |
1da177e4 | 3185 | unsigned long imbalance; |
70b97a7f | 3186 | struct rq *busiest; |
fe2eea3f | 3187 | unsigned long flags; |
5969fe06 | 3188 | |
7c16ec58 MT |
3189 | cpus_setall(*cpus); |
3190 | ||
89c4710e SS |
3191 | /* |
3192 | * When power savings policy is enabled for the parent domain, idle | |
3193 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3194 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3195 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3196 | */ |
d15bcfdb | 3197 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3198 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3199 | sd_idle = 1; |
1da177e4 | 3200 | |
2d72376b | 3201 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3202 | |
0a2966b4 CL |
3203 | redo: |
3204 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
7c16ec58 | 3205 | cpus, balance); |
783609c6 | 3206 | |
06066714 | 3207 | if (*balance == 0) |
783609c6 | 3208 | goto out_balanced; |
783609c6 | 3209 | |
1da177e4 LT |
3210 | if (!group) { |
3211 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3212 | goto out_balanced; | |
3213 | } | |
3214 | ||
7c16ec58 | 3215 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3216 | if (!busiest) { |
3217 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3218 | goto out_balanced; | |
3219 | } | |
3220 | ||
db935dbd | 3221 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3222 | |
3223 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3224 | ||
43010659 | 3225 | ld_moved = 0; |
1da177e4 LT |
3226 | if (busiest->nr_running > 1) { |
3227 | /* | |
3228 | * Attempt to move tasks. If find_busiest_group has found | |
3229 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3230 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3231 | * correctly treated as an imbalance. |
3232 | */ | |
fe2eea3f | 3233 | local_irq_save(flags); |
e17224bf | 3234 | double_rq_lock(this_rq, busiest); |
43010659 | 3235 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3236 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3237 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3238 | local_irq_restore(flags); |
81026794 | 3239 | |
46cb4b7c SS |
3240 | /* |
3241 | * some other cpu did the load balance for us. | |
3242 | */ | |
43010659 | 3243 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3244 | resched_cpu(this_cpu); |
3245 | ||
81026794 | 3246 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3247 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3248 | cpu_clear(cpu_of(busiest), *cpus); |
3249 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3250 | goto redo; |
81026794 | 3251 | goto out_balanced; |
0a2966b4 | 3252 | } |
1da177e4 | 3253 | } |
81026794 | 3254 | |
43010659 | 3255 | if (!ld_moved) { |
1da177e4 LT |
3256 | schedstat_inc(sd, lb_failed[idle]); |
3257 | sd->nr_balance_failed++; | |
3258 | ||
3259 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3260 | |
fe2eea3f | 3261 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3262 | |
3263 | /* don't kick the migration_thread, if the curr | |
3264 | * task on busiest cpu can't be moved to this_cpu | |
3265 | */ | |
3266 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3267 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3268 | all_pinned = 1; |
3269 | goto out_one_pinned; | |
3270 | } | |
3271 | ||
1da177e4 LT |
3272 | if (!busiest->active_balance) { |
3273 | busiest->active_balance = 1; | |
3274 | busiest->push_cpu = this_cpu; | |
81026794 | 3275 | active_balance = 1; |
1da177e4 | 3276 | } |
fe2eea3f | 3277 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3278 | if (active_balance) |
1da177e4 LT |
3279 | wake_up_process(busiest->migration_thread); |
3280 | ||
3281 | /* | |
3282 | * We've kicked active balancing, reset the failure | |
3283 | * counter. | |
3284 | */ | |
39507451 | 3285 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3286 | } |
81026794 | 3287 | } else |
1da177e4 LT |
3288 | sd->nr_balance_failed = 0; |
3289 | ||
81026794 | 3290 | if (likely(!active_balance)) { |
1da177e4 LT |
3291 | /* We were unbalanced, so reset the balancing interval */ |
3292 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3293 | } else { |
3294 | /* | |
3295 | * If we've begun active balancing, start to back off. This | |
3296 | * case may not be covered by the all_pinned logic if there | |
3297 | * is only 1 task on the busy runqueue (because we don't call | |
3298 | * move_tasks). | |
3299 | */ | |
3300 | if (sd->balance_interval < sd->max_interval) | |
3301 | sd->balance_interval *= 2; | |
1da177e4 LT |
3302 | } |
3303 | ||
43010659 | 3304 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3305 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3306 | return -1; |
43010659 | 3307 | return ld_moved; |
1da177e4 LT |
3308 | |
3309 | out_balanced: | |
1da177e4 LT |
3310 | schedstat_inc(sd, lb_balanced[idle]); |
3311 | ||
16cfb1c0 | 3312 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3313 | |
3314 | out_one_pinned: | |
1da177e4 | 3315 | /* tune up the balancing interval */ |
77391d71 NP |
3316 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3317 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3318 | sd->balance_interval *= 2; |
3319 | ||
48f24c4d | 3320 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3321 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3322 | return -1; |
1da177e4 LT |
3323 | return 0; |
3324 | } | |
3325 | ||
3326 | /* | |
3327 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3328 | * tasks if there is an imbalance. | |
3329 | * | |
d15bcfdb | 3330 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3331 | * this_rq is locked. |
3332 | */ | |
48f24c4d | 3333 | static int |
7c16ec58 MT |
3334 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3335 | cpumask_t *cpus) | |
1da177e4 LT |
3336 | { |
3337 | struct sched_group *group; | |
70b97a7f | 3338 | struct rq *busiest = NULL; |
1da177e4 | 3339 | unsigned long imbalance; |
43010659 | 3340 | int ld_moved = 0; |
5969fe06 | 3341 | int sd_idle = 0; |
969bb4e4 | 3342 | int all_pinned = 0; |
7c16ec58 MT |
3343 | |
3344 | cpus_setall(*cpus); | |
5969fe06 | 3345 | |
89c4710e SS |
3346 | /* |
3347 | * When power savings policy is enabled for the parent domain, idle | |
3348 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3349 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3350 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3351 | */ |
3352 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3353 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3354 | sd_idle = 1; |
1da177e4 | 3355 | |
2d72376b | 3356 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3357 | redo: |
d15bcfdb | 3358 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3359 | &sd_idle, cpus, NULL); |
1da177e4 | 3360 | if (!group) { |
d15bcfdb | 3361 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3362 | goto out_balanced; |
1da177e4 LT |
3363 | } |
3364 | ||
7c16ec58 | 3365 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3366 | if (!busiest) { |
d15bcfdb | 3367 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3368 | goto out_balanced; |
1da177e4 LT |
3369 | } |
3370 | ||
db935dbd NP |
3371 | BUG_ON(busiest == this_rq); |
3372 | ||
d15bcfdb | 3373 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3374 | |
43010659 | 3375 | ld_moved = 0; |
d6d5cfaf NP |
3376 | if (busiest->nr_running > 1) { |
3377 | /* Attempt to move tasks */ | |
3378 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3379 | /* this_rq->clock is already updated */ |
3380 | update_rq_clock(busiest); | |
43010659 | 3381 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3382 | imbalance, sd, CPU_NEWLY_IDLE, |
3383 | &all_pinned); | |
d6d5cfaf | 3384 | spin_unlock(&busiest->lock); |
0a2966b4 | 3385 | |
969bb4e4 | 3386 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3387 | cpu_clear(cpu_of(busiest), *cpus); |
3388 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3389 | goto redo; |
3390 | } | |
d6d5cfaf NP |
3391 | } |
3392 | ||
43010659 | 3393 | if (!ld_moved) { |
d15bcfdb | 3394 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3395 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3396 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3397 | return -1; |
3398 | } else | |
16cfb1c0 | 3399 | sd->nr_balance_failed = 0; |
1da177e4 | 3400 | |
43010659 | 3401 | return ld_moved; |
16cfb1c0 NP |
3402 | |
3403 | out_balanced: | |
d15bcfdb | 3404 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3405 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3406 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3407 | return -1; |
16cfb1c0 | 3408 | sd->nr_balance_failed = 0; |
48f24c4d | 3409 | |
16cfb1c0 | 3410 | return 0; |
1da177e4 LT |
3411 | } |
3412 | ||
3413 | /* | |
3414 | * idle_balance is called by schedule() if this_cpu is about to become | |
3415 | * idle. Attempts to pull tasks from other CPUs. | |
3416 | */ | |
70b97a7f | 3417 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3418 | { |
3419 | struct sched_domain *sd; | |
dd41f596 IM |
3420 | int pulled_task = -1; |
3421 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3422 | cpumask_t tmpmask; |
1da177e4 LT |
3423 | |
3424 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3425 | unsigned long interval; |
3426 | ||
3427 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3428 | continue; | |
3429 | ||
3430 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3431 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3432 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3433 | sd, &tmpmask); | |
92c4ca5c CL |
3434 | |
3435 | interval = msecs_to_jiffies(sd->balance_interval); | |
3436 | if (time_after(next_balance, sd->last_balance + interval)) | |
3437 | next_balance = sd->last_balance + interval; | |
3438 | if (pulled_task) | |
3439 | break; | |
1da177e4 | 3440 | } |
dd41f596 | 3441 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3442 | /* |
3443 | * We are going idle. next_balance may be set based on | |
3444 | * a busy processor. So reset next_balance. | |
3445 | */ | |
3446 | this_rq->next_balance = next_balance; | |
dd41f596 | 3447 | } |
1da177e4 LT |
3448 | } |
3449 | ||
3450 | /* | |
3451 | * active_load_balance is run by migration threads. It pushes running tasks | |
3452 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3453 | * running on each physical CPU where possible, and avoids physical / | |
3454 | * logical imbalances. | |
3455 | * | |
3456 | * Called with busiest_rq locked. | |
3457 | */ | |
70b97a7f | 3458 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3459 | { |
39507451 | 3460 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3461 | struct sched_domain *sd; |
3462 | struct rq *target_rq; | |
39507451 | 3463 | |
48f24c4d | 3464 | /* Is there any task to move? */ |
39507451 | 3465 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3466 | return; |
3467 | ||
3468 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3469 | |
3470 | /* | |
39507451 | 3471 | * This condition is "impossible", if it occurs |
41a2d6cf | 3472 | * we need to fix it. Originally reported by |
39507451 | 3473 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3474 | */ |
39507451 | 3475 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3476 | |
39507451 NP |
3477 | /* move a task from busiest_rq to target_rq */ |
3478 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3479 | update_rq_clock(busiest_rq); |
3480 | update_rq_clock(target_rq); | |
39507451 NP |
3481 | |
3482 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3483 | for_each_domain(target_cpu, sd) { |
39507451 | 3484 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3485 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3486 | break; |
c96d145e | 3487 | } |
39507451 | 3488 | |
48f24c4d | 3489 | if (likely(sd)) { |
2d72376b | 3490 | schedstat_inc(sd, alb_count); |
39507451 | 3491 | |
43010659 PW |
3492 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3493 | sd, CPU_IDLE)) | |
48f24c4d IM |
3494 | schedstat_inc(sd, alb_pushed); |
3495 | else | |
3496 | schedstat_inc(sd, alb_failed); | |
3497 | } | |
39507451 | 3498 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3499 | } |
3500 | ||
46cb4b7c SS |
3501 | #ifdef CONFIG_NO_HZ |
3502 | static struct { | |
3503 | atomic_t load_balancer; | |
41a2d6cf | 3504 | cpumask_t cpu_mask; |
46cb4b7c SS |
3505 | } nohz ____cacheline_aligned = { |
3506 | .load_balancer = ATOMIC_INIT(-1), | |
3507 | .cpu_mask = CPU_MASK_NONE, | |
3508 | }; | |
3509 | ||
7835b98b | 3510 | /* |
46cb4b7c SS |
3511 | * This routine will try to nominate the ilb (idle load balancing) |
3512 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3513 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3514 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3515 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3516 | * arrives... | |
3517 | * | |
3518 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3519 | * for idle load balancing. ilb owner will still be part of | |
3520 | * nohz.cpu_mask.. | |
7835b98b | 3521 | * |
46cb4b7c SS |
3522 | * While stopping the tick, this cpu will become the ilb owner if there |
3523 | * is no other owner. And will be the owner till that cpu becomes busy | |
3524 | * or if all cpus in the system stop their ticks at which point | |
3525 | * there is no need for ilb owner. | |
3526 | * | |
3527 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3528 | * next busy scheduler_tick() | |
3529 | */ | |
3530 | int select_nohz_load_balancer(int stop_tick) | |
3531 | { | |
3532 | int cpu = smp_processor_id(); | |
3533 | ||
3534 | if (stop_tick) { | |
3535 | cpu_set(cpu, nohz.cpu_mask); | |
3536 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3537 | ||
3538 | /* | |
3539 | * If we are going offline and still the leader, give up! | |
3540 | */ | |
3541 | if (cpu_is_offline(cpu) && | |
3542 | atomic_read(&nohz.load_balancer) == cpu) { | |
3543 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3544 | BUG(); | |
3545 | return 0; | |
3546 | } | |
3547 | ||
3548 | /* time for ilb owner also to sleep */ | |
3549 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3550 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3551 | atomic_set(&nohz.load_balancer, -1); | |
3552 | return 0; | |
3553 | } | |
3554 | ||
3555 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3556 | /* make me the ilb owner */ | |
3557 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3558 | return 1; | |
3559 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3560 | return 1; | |
3561 | } else { | |
3562 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3563 | return 0; | |
3564 | ||
3565 | cpu_clear(cpu, nohz.cpu_mask); | |
3566 | ||
3567 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3568 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3569 | BUG(); | |
3570 | } | |
3571 | return 0; | |
3572 | } | |
3573 | #endif | |
3574 | ||
3575 | static DEFINE_SPINLOCK(balancing); | |
3576 | ||
3577 | /* | |
7835b98b CL |
3578 | * It checks each scheduling domain to see if it is due to be balanced, |
3579 | * and initiates a balancing operation if so. | |
3580 | * | |
3581 | * Balancing parameters are set up in arch_init_sched_domains. | |
3582 | */ | |
a9957449 | 3583 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3584 | { |
46cb4b7c SS |
3585 | int balance = 1; |
3586 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3587 | unsigned long interval; |
3588 | struct sched_domain *sd; | |
46cb4b7c | 3589 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3590 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3591 | int update_next_balance = 0; |
7c16ec58 | 3592 | cpumask_t tmp; |
1da177e4 | 3593 | |
46cb4b7c | 3594 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3595 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3596 | continue; | |
3597 | ||
3598 | interval = sd->balance_interval; | |
d15bcfdb | 3599 | if (idle != CPU_IDLE) |
1da177e4 LT |
3600 | interval *= sd->busy_factor; |
3601 | ||
3602 | /* scale ms to jiffies */ | |
3603 | interval = msecs_to_jiffies(interval); | |
3604 | if (unlikely(!interval)) | |
3605 | interval = 1; | |
dd41f596 IM |
3606 | if (interval > HZ*NR_CPUS/10) |
3607 | interval = HZ*NR_CPUS/10; | |
3608 | ||
1da177e4 | 3609 | |
08c183f3 CL |
3610 | if (sd->flags & SD_SERIALIZE) { |
3611 | if (!spin_trylock(&balancing)) | |
3612 | goto out; | |
3613 | } | |
3614 | ||
c9819f45 | 3615 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3616 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3617 | /* |
3618 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3619 | * longer idle, or one of our SMT siblings is |
3620 | * not idle. | |
3621 | */ | |
d15bcfdb | 3622 | idle = CPU_NOT_IDLE; |
1da177e4 | 3623 | } |
1bd77f2d | 3624 | sd->last_balance = jiffies; |
1da177e4 | 3625 | } |
08c183f3 CL |
3626 | if (sd->flags & SD_SERIALIZE) |
3627 | spin_unlock(&balancing); | |
3628 | out: | |
f549da84 | 3629 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3630 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3631 | update_next_balance = 1; |
3632 | } | |
783609c6 SS |
3633 | |
3634 | /* | |
3635 | * Stop the load balance at this level. There is another | |
3636 | * CPU in our sched group which is doing load balancing more | |
3637 | * actively. | |
3638 | */ | |
3639 | if (!balance) | |
3640 | break; | |
1da177e4 | 3641 | } |
f549da84 SS |
3642 | |
3643 | /* | |
3644 | * next_balance will be updated only when there is a need. | |
3645 | * When the cpu is attached to null domain for ex, it will not be | |
3646 | * updated. | |
3647 | */ | |
3648 | if (likely(update_next_balance)) | |
3649 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3650 | } |
3651 | ||
3652 | /* | |
3653 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3654 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3655 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3656 | */ | |
3657 | static void run_rebalance_domains(struct softirq_action *h) | |
3658 | { | |
dd41f596 IM |
3659 | int this_cpu = smp_processor_id(); |
3660 | struct rq *this_rq = cpu_rq(this_cpu); | |
3661 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3662 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3663 | |
dd41f596 | 3664 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3665 | |
3666 | #ifdef CONFIG_NO_HZ | |
3667 | /* | |
3668 | * If this cpu is the owner for idle load balancing, then do the | |
3669 | * balancing on behalf of the other idle cpus whose ticks are | |
3670 | * stopped. | |
3671 | */ | |
dd41f596 IM |
3672 | if (this_rq->idle_at_tick && |
3673 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3674 | cpumask_t cpus = nohz.cpu_mask; |
3675 | struct rq *rq; | |
3676 | int balance_cpu; | |
3677 | ||
dd41f596 | 3678 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3679 | for_each_cpu_mask(balance_cpu, cpus) { |
3680 | /* | |
3681 | * If this cpu gets work to do, stop the load balancing | |
3682 | * work being done for other cpus. Next load | |
3683 | * balancing owner will pick it up. | |
3684 | */ | |
3685 | if (need_resched()) | |
3686 | break; | |
3687 | ||
de0cf899 | 3688 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3689 | |
3690 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3691 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3692 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3693 | } |
3694 | } | |
3695 | #endif | |
3696 | } | |
3697 | ||
3698 | /* | |
3699 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3700 | * | |
3701 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3702 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3703 | * if the whole system is idle. | |
3704 | */ | |
dd41f596 | 3705 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3706 | { |
46cb4b7c SS |
3707 | #ifdef CONFIG_NO_HZ |
3708 | /* | |
3709 | * If we were in the nohz mode recently and busy at the current | |
3710 | * scheduler tick, then check if we need to nominate new idle | |
3711 | * load balancer. | |
3712 | */ | |
3713 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3714 | rq->in_nohz_recently = 0; | |
3715 | ||
3716 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3717 | cpu_clear(cpu, nohz.cpu_mask); | |
3718 | atomic_set(&nohz.load_balancer, -1); | |
3719 | } | |
3720 | ||
3721 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3722 | /* | |
3723 | * simple selection for now: Nominate the | |
3724 | * first cpu in the nohz list to be the next | |
3725 | * ilb owner. | |
3726 | * | |
3727 | * TBD: Traverse the sched domains and nominate | |
3728 | * the nearest cpu in the nohz.cpu_mask. | |
3729 | */ | |
3730 | int ilb = first_cpu(nohz.cpu_mask); | |
3731 | ||
434d53b0 | 3732 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
3733 | resched_cpu(ilb); |
3734 | } | |
3735 | } | |
3736 | ||
3737 | /* | |
3738 | * If this cpu is idle and doing idle load balancing for all the | |
3739 | * cpus with ticks stopped, is it time for that to stop? | |
3740 | */ | |
3741 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3742 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3743 | resched_cpu(cpu); | |
3744 | return; | |
3745 | } | |
3746 | ||
3747 | /* | |
3748 | * If this cpu is idle and the idle load balancing is done by | |
3749 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3750 | */ | |
3751 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3752 | cpu_isset(cpu, nohz.cpu_mask)) | |
3753 | return; | |
3754 | #endif | |
3755 | if (time_after_eq(jiffies, rq->next_balance)) | |
3756 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3757 | } |
dd41f596 IM |
3758 | |
3759 | #else /* CONFIG_SMP */ | |
3760 | ||
1da177e4 LT |
3761 | /* |
3762 | * on UP we do not need to balance between CPUs: | |
3763 | */ | |
70b97a7f | 3764 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3765 | { |
3766 | } | |
dd41f596 | 3767 | |
1da177e4 LT |
3768 | #endif |
3769 | ||
1da177e4 LT |
3770 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3771 | ||
3772 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3773 | ||
3774 | /* | |
41b86e9c IM |
3775 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3776 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3777 | */ |
41b86e9c | 3778 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3779 | { |
1da177e4 | 3780 | unsigned long flags; |
41b86e9c IM |
3781 | u64 ns, delta_exec; |
3782 | struct rq *rq; | |
48f24c4d | 3783 | |
41b86e9c IM |
3784 | rq = task_rq_lock(p, &flags); |
3785 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 3786 | if (task_current(rq, p)) { |
a8e504d2 IM |
3787 | update_rq_clock(rq); |
3788 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3789 | if ((s64)delta_exec > 0) |
3790 | ns += delta_exec; | |
3791 | } | |
3792 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3793 | |
1da177e4 LT |
3794 | return ns; |
3795 | } | |
3796 | ||
1da177e4 LT |
3797 | /* |
3798 | * Account user cpu time to a process. | |
3799 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3800 | * @cputime: the cpu time spent in user space since the last update |
3801 | */ | |
3802 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3803 | { | |
3804 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3805 | cputime64_t tmp; | |
3806 | ||
3807 | p->utime = cputime_add(p->utime, cputime); | |
3808 | ||
3809 | /* Add user time to cpustat. */ | |
3810 | tmp = cputime_to_cputime64(cputime); | |
3811 | if (TASK_NICE(p) > 0) | |
3812 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3813 | else | |
3814 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3815 | } | |
3816 | ||
94886b84 LV |
3817 | /* |
3818 | * Account guest cpu time to a process. | |
3819 | * @p: the process that the cpu time gets accounted to | |
3820 | * @cputime: the cpu time spent in virtual machine since the last update | |
3821 | */ | |
f7402e03 | 3822 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
3823 | { |
3824 | cputime64_t tmp; | |
3825 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3826 | ||
3827 | tmp = cputime_to_cputime64(cputime); | |
3828 | ||
3829 | p->utime = cputime_add(p->utime, cputime); | |
3830 | p->gtime = cputime_add(p->gtime, cputime); | |
3831 | ||
3832 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3833 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3834 | } | |
3835 | ||
c66f08be MN |
3836 | /* |
3837 | * Account scaled user cpu time to a process. | |
3838 | * @p: the process that the cpu time gets accounted to | |
3839 | * @cputime: the cpu time spent in user space since the last update | |
3840 | */ | |
3841 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3842 | { | |
3843 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3844 | } | |
3845 | ||
1da177e4 LT |
3846 | /* |
3847 | * Account system cpu time to a process. | |
3848 | * @p: the process that the cpu time gets accounted to | |
3849 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3850 | * @cputime: the cpu time spent in kernel space since the last update | |
3851 | */ | |
3852 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3853 | cputime_t cputime) | |
3854 | { | |
3855 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3856 | struct rq *rq = this_rq(); |
1da177e4 LT |
3857 | cputime64_t tmp; |
3858 | ||
9778385d CB |
3859 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) |
3860 | return account_guest_time(p, cputime); | |
94886b84 | 3861 | |
1da177e4 LT |
3862 | p->stime = cputime_add(p->stime, cputime); |
3863 | ||
3864 | /* Add system time to cpustat. */ | |
3865 | tmp = cputime_to_cputime64(cputime); | |
3866 | if (hardirq_count() - hardirq_offset) | |
3867 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3868 | else if (softirq_count()) | |
3869 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 3870 | else if (p != rq->idle) |
1da177e4 | 3871 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 3872 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
3873 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3874 | else | |
3875 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3876 | /* Account for system time used */ | |
3877 | acct_update_integrals(p); | |
1da177e4 LT |
3878 | } |
3879 | ||
c66f08be MN |
3880 | /* |
3881 | * Account scaled system cpu time to a process. | |
3882 | * @p: the process that the cpu time gets accounted to | |
3883 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3884 | * @cputime: the cpu time spent in kernel space since the last update | |
3885 | */ | |
3886 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3887 | { | |
3888 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3889 | } | |
3890 | ||
1da177e4 LT |
3891 | /* |
3892 | * Account for involuntary wait time. | |
3893 | * @p: the process from which the cpu time has been stolen | |
3894 | * @steal: the cpu time spent in involuntary wait | |
3895 | */ | |
3896 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3897 | { | |
3898 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3899 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3900 | struct rq *rq = this_rq(); |
1da177e4 LT |
3901 | |
3902 | if (p == rq->idle) { | |
3903 | p->stime = cputime_add(p->stime, steal); | |
3904 | if (atomic_read(&rq->nr_iowait) > 0) | |
3905 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3906 | else | |
3907 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 3908 | } else |
1da177e4 LT |
3909 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
3910 | } | |
3911 | ||
7835b98b CL |
3912 | /* |
3913 | * This function gets called by the timer code, with HZ frequency. | |
3914 | * We call it with interrupts disabled. | |
3915 | * | |
3916 | * It also gets called by the fork code, when changing the parent's | |
3917 | * timeslices. | |
3918 | */ | |
3919 | void scheduler_tick(void) | |
3920 | { | |
7835b98b CL |
3921 | int cpu = smp_processor_id(); |
3922 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3923 | struct task_struct *curr = rq->curr; |
529c7726 | 3924 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3925 | |
3926 | spin_lock(&rq->lock); | |
546fe3c9 | 3927 | __update_rq_clock(rq); |
529c7726 IM |
3928 | /* |
3929 | * Let rq->clock advance by at least TICK_NSEC: | |
3930 | */ | |
cc203d24 | 3931 | if (unlikely(rq->clock < next_tick)) { |
529c7726 | 3932 | rq->clock = next_tick; |
cc203d24 GC |
3933 | rq->clock_underflows++; |
3934 | } | |
529c7726 | 3935 | rq->tick_timestamp = rq->clock; |
15934a37 | 3936 | update_last_tick_seen(rq); |
f1a438d8 | 3937 | update_cpu_load(rq); |
fa85ae24 | 3938 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 3939 | spin_unlock(&rq->lock); |
7835b98b | 3940 | |
e418e1c2 | 3941 | #ifdef CONFIG_SMP |
dd41f596 IM |
3942 | rq->idle_at_tick = idle_cpu(cpu); |
3943 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3944 | #endif |
1da177e4 LT |
3945 | } |
3946 | ||
1da177e4 LT |
3947 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3948 | ||
43627582 | 3949 | void __kprobes add_preempt_count(int val) |
1da177e4 LT |
3950 | { |
3951 | /* | |
3952 | * Underflow? | |
3953 | */ | |
9a11b49a IM |
3954 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3955 | return; | |
1da177e4 LT |
3956 | preempt_count() += val; |
3957 | /* | |
3958 | * Spinlock count overflowing soon? | |
3959 | */ | |
33859f7f MOS |
3960 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3961 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3962 | } |
3963 | EXPORT_SYMBOL(add_preempt_count); | |
3964 | ||
43627582 | 3965 | void __kprobes sub_preempt_count(int val) |
1da177e4 LT |
3966 | { |
3967 | /* | |
3968 | * Underflow? | |
3969 | */ | |
9a11b49a IM |
3970 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3971 | return; | |
1da177e4 LT |
3972 | /* |
3973 | * Is the spinlock portion underflowing? | |
3974 | */ | |
9a11b49a IM |
3975 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3976 | !(preempt_count() & PREEMPT_MASK))) | |
3977 | return; | |
3978 | ||
1da177e4 LT |
3979 | preempt_count() -= val; |
3980 | } | |
3981 | EXPORT_SYMBOL(sub_preempt_count); | |
3982 | ||
3983 | #endif | |
3984 | ||
3985 | /* | |
dd41f596 | 3986 | * Print scheduling while atomic bug: |
1da177e4 | 3987 | */ |
dd41f596 | 3988 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3989 | { |
838225b4 SS |
3990 | struct pt_regs *regs = get_irq_regs(); |
3991 | ||
3992 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
3993 | prev->comm, prev->pid, preempt_count()); | |
3994 | ||
dd41f596 IM |
3995 | debug_show_held_locks(prev); |
3996 | if (irqs_disabled()) | |
3997 | print_irqtrace_events(prev); | |
838225b4 SS |
3998 | |
3999 | if (regs) | |
4000 | show_regs(regs); | |
4001 | else | |
4002 | dump_stack(); | |
dd41f596 | 4003 | } |
1da177e4 | 4004 | |
dd41f596 IM |
4005 | /* |
4006 | * Various schedule()-time debugging checks and statistics: | |
4007 | */ | |
4008 | static inline void schedule_debug(struct task_struct *prev) | |
4009 | { | |
1da177e4 | 4010 | /* |
41a2d6cf | 4011 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4012 | * schedule() atomically, we ignore that path for now. |
4013 | * Otherwise, whine if we are scheduling when we should not be. | |
4014 | */ | |
dd41f596 IM |
4015 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
4016 | __schedule_bug(prev); | |
4017 | ||
1da177e4 LT |
4018 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4019 | ||
2d72376b | 4020 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4021 | #ifdef CONFIG_SCHEDSTATS |
4022 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4023 | schedstat_inc(this_rq(), bkl_count); |
4024 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4025 | } |
4026 | #endif | |
dd41f596 IM |
4027 | } |
4028 | ||
4029 | /* | |
4030 | * Pick up the highest-prio task: | |
4031 | */ | |
4032 | static inline struct task_struct * | |
ff95f3df | 4033 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4034 | { |
5522d5d5 | 4035 | const struct sched_class *class; |
dd41f596 | 4036 | struct task_struct *p; |
1da177e4 LT |
4037 | |
4038 | /* | |
dd41f596 IM |
4039 | * Optimization: we know that if all tasks are in |
4040 | * the fair class we can call that function directly: | |
1da177e4 | 4041 | */ |
dd41f596 | 4042 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4043 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4044 | if (likely(p)) |
4045 | return p; | |
1da177e4 LT |
4046 | } |
4047 | ||
dd41f596 IM |
4048 | class = sched_class_highest; |
4049 | for ( ; ; ) { | |
fb8d4724 | 4050 | p = class->pick_next_task(rq); |
dd41f596 IM |
4051 | if (p) |
4052 | return p; | |
4053 | /* | |
4054 | * Will never be NULL as the idle class always | |
4055 | * returns a non-NULL p: | |
4056 | */ | |
4057 | class = class->next; | |
4058 | } | |
4059 | } | |
1da177e4 | 4060 | |
dd41f596 IM |
4061 | /* |
4062 | * schedule() is the main scheduler function. | |
4063 | */ | |
4064 | asmlinkage void __sched schedule(void) | |
4065 | { | |
4066 | struct task_struct *prev, *next; | |
67ca7bde | 4067 | unsigned long *switch_count; |
dd41f596 | 4068 | struct rq *rq; |
dd41f596 IM |
4069 | int cpu; |
4070 | ||
4071 | need_resched: | |
4072 | preempt_disable(); | |
4073 | cpu = smp_processor_id(); | |
4074 | rq = cpu_rq(cpu); | |
4075 | rcu_qsctr_inc(cpu); | |
4076 | prev = rq->curr; | |
4077 | switch_count = &prev->nivcsw; | |
4078 | ||
4079 | release_kernel_lock(prev); | |
4080 | need_resched_nonpreemptible: | |
4081 | ||
4082 | schedule_debug(prev); | |
1da177e4 | 4083 | |
8f4d37ec PZ |
4084 | hrtick_clear(rq); |
4085 | ||
1e819950 IM |
4086 | /* |
4087 | * Do the rq-clock update outside the rq lock: | |
4088 | */ | |
4089 | local_irq_disable(); | |
c1b3da3e | 4090 | __update_rq_clock(rq); |
1e819950 IM |
4091 | spin_lock(&rq->lock); |
4092 | clear_tsk_need_resched(prev); | |
1da177e4 | 4093 | |
1da177e4 | 4094 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 4095 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
23e3c3cd | 4096 | signal_pending(prev))) { |
1da177e4 | 4097 | prev->state = TASK_RUNNING; |
dd41f596 | 4098 | } else { |
2e1cb74a | 4099 | deactivate_task(rq, prev, 1); |
1da177e4 | 4100 | } |
dd41f596 | 4101 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4102 | } |
4103 | ||
9a897c5a SR |
4104 | #ifdef CONFIG_SMP |
4105 | if (prev->sched_class->pre_schedule) | |
4106 | prev->sched_class->pre_schedule(rq, prev); | |
4107 | #endif | |
f65eda4f | 4108 | |
dd41f596 | 4109 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4110 | idle_balance(cpu, rq); |
1da177e4 | 4111 | |
31ee529c | 4112 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4113 | next = pick_next_task(rq, prev); |
1da177e4 LT |
4114 | |
4115 | sched_info_switch(prev, next); | |
dd41f596 | 4116 | |
1da177e4 | 4117 | if (likely(prev != next)) { |
1da177e4 LT |
4118 | rq->nr_switches++; |
4119 | rq->curr = next; | |
4120 | ++*switch_count; | |
4121 | ||
dd41f596 | 4122 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4123 | /* |
4124 | * the context switch might have flipped the stack from under | |
4125 | * us, hence refresh the local variables. | |
4126 | */ | |
4127 | cpu = smp_processor_id(); | |
4128 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4129 | } else |
4130 | spin_unlock_irq(&rq->lock); | |
4131 | ||
8f4d37ec PZ |
4132 | hrtick_set(rq); |
4133 | ||
4134 | if (unlikely(reacquire_kernel_lock(current) < 0)) | |
1da177e4 | 4135 | goto need_resched_nonpreemptible; |
8f4d37ec | 4136 | |
1da177e4 LT |
4137 | preempt_enable_no_resched(); |
4138 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4139 | goto need_resched; | |
4140 | } | |
1da177e4 LT |
4141 | EXPORT_SYMBOL(schedule); |
4142 | ||
4143 | #ifdef CONFIG_PREEMPT | |
4144 | /* | |
2ed6e34f | 4145 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4146 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4147 | * occur there and call schedule directly. |
4148 | */ | |
4149 | asmlinkage void __sched preempt_schedule(void) | |
4150 | { | |
4151 | struct thread_info *ti = current_thread_info(); | |
1da177e4 LT |
4152 | struct task_struct *task = current; |
4153 | int saved_lock_depth; | |
6478d880 | 4154 | |
1da177e4 LT |
4155 | /* |
4156 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4157 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4158 | */ |
beed33a8 | 4159 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4160 | return; |
4161 | ||
3a5c359a AK |
4162 | do { |
4163 | add_preempt_count(PREEMPT_ACTIVE); | |
4164 | ||
4165 | /* | |
4166 | * We keep the big kernel semaphore locked, but we | |
4167 | * clear ->lock_depth so that schedule() doesnt | |
4168 | * auto-release the semaphore: | |
4169 | */ | |
3a5c359a AK |
4170 | saved_lock_depth = task->lock_depth; |
4171 | task->lock_depth = -1; | |
3a5c359a | 4172 | schedule(); |
3a5c359a | 4173 | task->lock_depth = saved_lock_depth; |
3a5c359a | 4174 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4175 | |
3a5c359a AK |
4176 | /* |
4177 | * Check again in case we missed a preemption opportunity | |
4178 | * between schedule and now. | |
4179 | */ | |
4180 | barrier(); | |
4181 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4182 | } |
1da177e4 LT |
4183 | EXPORT_SYMBOL(preempt_schedule); |
4184 | ||
4185 | /* | |
2ed6e34f | 4186 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4187 | * off of irq context. |
4188 | * Note, that this is called and return with irqs disabled. This will | |
4189 | * protect us against recursive calling from irq. | |
4190 | */ | |
4191 | asmlinkage void __sched preempt_schedule_irq(void) | |
4192 | { | |
4193 | struct thread_info *ti = current_thread_info(); | |
1da177e4 LT |
4194 | struct task_struct *task = current; |
4195 | int saved_lock_depth; | |
6478d880 | 4196 | |
2ed6e34f | 4197 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4198 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4199 | ||
3a5c359a AK |
4200 | do { |
4201 | add_preempt_count(PREEMPT_ACTIVE); | |
4202 | ||
4203 | /* | |
4204 | * We keep the big kernel semaphore locked, but we | |
4205 | * clear ->lock_depth so that schedule() doesnt | |
4206 | * auto-release the semaphore: | |
4207 | */ | |
3a5c359a AK |
4208 | saved_lock_depth = task->lock_depth; |
4209 | task->lock_depth = -1; | |
3a5c359a AK |
4210 | local_irq_enable(); |
4211 | schedule(); | |
4212 | local_irq_disable(); | |
3a5c359a | 4213 | task->lock_depth = saved_lock_depth; |
3a5c359a | 4214 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4215 | |
3a5c359a AK |
4216 | /* |
4217 | * Check again in case we missed a preemption opportunity | |
4218 | * between schedule and now. | |
4219 | */ | |
4220 | barrier(); | |
4221 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4222 | } |
4223 | ||
4224 | #endif /* CONFIG_PREEMPT */ | |
4225 | ||
95cdf3b7 IM |
4226 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4227 | void *key) | |
1da177e4 | 4228 | { |
48f24c4d | 4229 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4230 | } |
1da177e4 LT |
4231 | EXPORT_SYMBOL(default_wake_function); |
4232 | ||
4233 | /* | |
41a2d6cf IM |
4234 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4235 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4236 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4237 | * | |
4238 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4239 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4240 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4241 | */ | |
4242 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4243 | int nr_exclusive, int sync, void *key) | |
4244 | { | |
2e45874c | 4245 | wait_queue_t *curr, *next; |
1da177e4 | 4246 | |
2e45874c | 4247 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4248 | unsigned flags = curr->flags; |
4249 | ||
1da177e4 | 4250 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4251 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4252 | break; |
4253 | } | |
4254 | } | |
4255 | ||
4256 | /** | |
4257 | * __wake_up - wake up threads blocked on a waitqueue. | |
4258 | * @q: the waitqueue | |
4259 | * @mode: which threads | |
4260 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4261 | * @key: is directly passed to the wakeup function |
1da177e4 | 4262 | */ |
7ad5b3a5 | 4263 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4264 | int nr_exclusive, void *key) |
1da177e4 LT |
4265 | { |
4266 | unsigned long flags; | |
4267 | ||
4268 | spin_lock_irqsave(&q->lock, flags); | |
4269 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4270 | spin_unlock_irqrestore(&q->lock, flags); | |
4271 | } | |
1da177e4 LT |
4272 | EXPORT_SYMBOL(__wake_up); |
4273 | ||
4274 | /* | |
4275 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4276 | */ | |
7ad5b3a5 | 4277 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4278 | { |
4279 | __wake_up_common(q, mode, 1, 0, NULL); | |
4280 | } | |
4281 | ||
4282 | /** | |
67be2dd1 | 4283 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4284 | * @q: the waitqueue |
4285 | * @mode: which threads | |
4286 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4287 | * | |
4288 | * The sync wakeup differs that the waker knows that it will schedule | |
4289 | * away soon, so while the target thread will be woken up, it will not | |
4290 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4291 | * with each other. This can prevent needless bouncing between CPUs. | |
4292 | * | |
4293 | * On UP it can prevent extra preemption. | |
4294 | */ | |
7ad5b3a5 | 4295 | void |
95cdf3b7 | 4296 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4297 | { |
4298 | unsigned long flags; | |
4299 | int sync = 1; | |
4300 | ||
4301 | if (unlikely(!q)) | |
4302 | return; | |
4303 | ||
4304 | if (unlikely(!nr_exclusive)) | |
4305 | sync = 0; | |
4306 | ||
4307 | spin_lock_irqsave(&q->lock, flags); | |
4308 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4309 | spin_unlock_irqrestore(&q->lock, flags); | |
4310 | } | |
4311 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4312 | ||
b15136e9 | 4313 | void complete(struct completion *x) |
1da177e4 LT |
4314 | { |
4315 | unsigned long flags; | |
4316 | ||
4317 | spin_lock_irqsave(&x->wait.lock, flags); | |
4318 | x->done++; | |
d9514f6c | 4319 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4320 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4321 | } | |
4322 | EXPORT_SYMBOL(complete); | |
4323 | ||
b15136e9 | 4324 | void complete_all(struct completion *x) |
1da177e4 LT |
4325 | { |
4326 | unsigned long flags; | |
4327 | ||
4328 | spin_lock_irqsave(&x->wait.lock, flags); | |
4329 | x->done += UINT_MAX/2; | |
d9514f6c | 4330 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4331 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4332 | } | |
4333 | EXPORT_SYMBOL(complete_all); | |
4334 | ||
8cbbe86d AK |
4335 | static inline long __sched |
4336 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4337 | { |
1da177e4 LT |
4338 | if (!x->done) { |
4339 | DECLARE_WAITQUEUE(wait, current); | |
4340 | ||
4341 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4342 | __add_wait_queue_tail(&x->wait, &wait); | |
4343 | do { | |
009e577e MW |
4344 | if ((state == TASK_INTERRUPTIBLE && |
4345 | signal_pending(current)) || | |
4346 | (state == TASK_KILLABLE && | |
4347 | fatal_signal_pending(current))) { | |
8cbbe86d AK |
4348 | __remove_wait_queue(&x->wait, &wait); |
4349 | return -ERESTARTSYS; | |
4350 | } | |
4351 | __set_current_state(state); | |
1da177e4 LT |
4352 | spin_unlock_irq(&x->wait.lock); |
4353 | timeout = schedule_timeout(timeout); | |
4354 | spin_lock_irq(&x->wait.lock); | |
4355 | if (!timeout) { | |
4356 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 4357 | return timeout; |
1da177e4 LT |
4358 | } |
4359 | } while (!x->done); | |
4360 | __remove_wait_queue(&x->wait, &wait); | |
4361 | } | |
4362 | x->done--; | |
1da177e4 LT |
4363 | return timeout; |
4364 | } | |
1da177e4 | 4365 | |
8cbbe86d AK |
4366 | static long __sched |
4367 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4368 | { |
1da177e4 LT |
4369 | might_sleep(); |
4370 | ||
4371 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4372 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4373 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4374 | return timeout; |
4375 | } | |
1da177e4 | 4376 | |
b15136e9 | 4377 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4378 | { |
4379 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4380 | } |
8cbbe86d | 4381 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4382 | |
b15136e9 | 4383 | unsigned long __sched |
8cbbe86d | 4384 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4385 | { |
8cbbe86d | 4386 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4387 | } |
8cbbe86d | 4388 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4389 | |
8cbbe86d | 4390 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4391 | { |
51e97990 AK |
4392 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4393 | if (t == -ERESTARTSYS) | |
4394 | return t; | |
4395 | return 0; | |
0fec171c | 4396 | } |
8cbbe86d | 4397 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4398 | |
b15136e9 | 4399 | unsigned long __sched |
8cbbe86d AK |
4400 | wait_for_completion_interruptible_timeout(struct completion *x, |
4401 | unsigned long timeout) | |
0fec171c | 4402 | { |
8cbbe86d | 4403 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4404 | } |
8cbbe86d | 4405 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4406 | |
009e577e MW |
4407 | int __sched wait_for_completion_killable(struct completion *x) |
4408 | { | |
4409 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4410 | if (t == -ERESTARTSYS) | |
4411 | return t; | |
4412 | return 0; | |
4413 | } | |
4414 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4415 | ||
8cbbe86d AK |
4416 | static long __sched |
4417 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4418 | { |
0fec171c IM |
4419 | unsigned long flags; |
4420 | wait_queue_t wait; | |
4421 | ||
4422 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4423 | |
8cbbe86d | 4424 | __set_current_state(state); |
1da177e4 | 4425 | |
8cbbe86d AK |
4426 | spin_lock_irqsave(&q->lock, flags); |
4427 | __add_wait_queue(q, &wait); | |
4428 | spin_unlock(&q->lock); | |
4429 | timeout = schedule_timeout(timeout); | |
4430 | spin_lock_irq(&q->lock); | |
4431 | __remove_wait_queue(q, &wait); | |
4432 | spin_unlock_irqrestore(&q->lock, flags); | |
4433 | ||
4434 | return timeout; | |
4435 | } | |
4436 | ||
4437 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4438 | { | |
4439 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4440 | } |
1da177e4 LT |
4441 | EXPORT_SYMBOL(interruptible_sleep_on); |
4442 | ||
0fec171c | 4443 | long __sched |
95cdf3b7 | 4444 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4445 | { |
8cbbe86d | 4446 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4447 | } |
1da177e4 LT |
4448 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4449 | ||
0fec171c | 4450 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4451 | { |
8cbbe86d | 4452 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4453 | } |
1da177e4 LT |
4454 | EXPORT_SYMBOL(sleep_on); |
4455 | ||
0fec171c | 4456 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4457 | { |
8cbbe86d | 4458 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4459 | } |
1da177e4 LT |
4460 | EXPORT_SYMBOL(sleep_on_timeout); |
4461 | ||
b29739f9 IM |
4462 | #ifdef CONFIG_RT_MUTEXES |
4463 | ||
4464 | /* | |
4465 | * rt_mutex_setprio - set the current priority of a task | |
4466 | * @p: task | |
4467 | * @prio: prio value (kernel-internal form) | |
4468 | * | |
4469 | * This function changes the 'effective' priority of a task. It does | |
4470 | * not touch ->normal_prio like __setscheduler(). | |
4471 | * | |
4472 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4473 | */ | |
36c8b586 | 4474 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4475 | { |
4476 | unsigned long flags; | |
83b699ed | 4477 | int oldprio, on_rq, running; |
70b97a7f | 4478 | struct rq *rq; |
cb469845 | 4479 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4480 | |
4481 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4482 | ||
4483 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4484 | update_rq_clock(rq); |
b29739f9 | 4485 | |
d5f9f942 | 4486 | oldprio = p->prio; |
dd41f596 | 4487 | on_rq = p->se.on_rq; |
051a1d1a | 4488 | running = task_current(rq, p); |
0e1f3483 | 4489 | if (on_rq) |
69be72c1 | 4490 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4491 | if (running) |
4492 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4493 | |
4494 | if (rt_prio(prio)) | |
4495 | p->sched_class = &rt_sched_class; | |
4496 | else | |
4497 | p->sched_class = &fair_sched_class; | |
4498 | ||
b29739f9 IM |
4499 | p->prio = prio; |
4500 | ||
0e1f3483 HS |
4501 | if (running) |
4502 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4503 | if (on_rq) { |
8159f87e | 4504 | enqueue_task(rq, p, 0); |
cb469845 SR |
4505 | |
4506 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4507 | } |
4508 | task_rq_unlock(rq, &flags); | |
4509 | } | |
4510 | ||
4511 | #endif | |
4512 | ||
36c8b586 | 4513 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4514 | { |
dd41f596 | 4515 | int old_prio, delta, on_rq; |
1da177e4 | 4516 | unsigned long flags; |
70b97a7f | 4517 | struct rq *rq; |
1da177e4 LT |
4518 | |
4519 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4520 | return; | |
4521 | /* | |
4522 | * We have to be careful, if called from sys_setpriority(), | |
4523 | * the task might be in the middle of scheduling on another CPU. | |
4524 | */ | |
4525 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4526 | update_rq_clock(rq); |
1da177e4 LT |
4527 | /* |
4528 | * The RT priorities are set via sched_setscheduler(), but we still | |
4529 | * allow the 'normal' nice value to be set - but as expected | |
4530 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4531 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4532 | */ |
e05606d3 | 4533 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4534 | p->static_prio = NICE_TO_PRIO(nice); |
4535 | goto out_unlock; | |
4536 | } | |
dd41f596 | 4537 | on_rq = p->se.on_rq; |
62fb1851 | 4538 | if (on_rq) { |
69be72c1 | 4539 | dequeue_task(rq, p, 0); |
62fb1851 PZ |
4540 | dec_load(rq, p); |
4541 | } | |
1da177e4 | 4542 | |
1da177e4 | 4543 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4544 | set_load_weight(p); |
b29739f9 IM |
4545 | old_prio = p->prio; |
4546 | p->prio = effective_prio(p); | |
4547 | delta = p->prio - old_prio; | |
1da177e4 | 4548 | |
dd41f596 | 4549 | if (on_rq) { |
8159f87e | 4550 | enqueue_task(rq, p, 0); |
62fb1851 | 4551 | inc_load(rq, p); |
1da177e4 | 4552 | /* |
d5f9f942 AM |
4553 | * If the task increased its priority or is running and |
4554 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4555 | */ |
d5f9f942 | 4556 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4557 | resched_task(rq->curr); |
4558 | } | |
4559 | out_unlock: | |
4560 | task_rq_unlock(rq, &flags); | |
4561 | } | |
1da177e4 LT |
4562 | EXPORT_SYMBOL(set_user_nice); |
4563 | ||
e43379f1 MM |
4564 | /* |
4565 | * can_nice - check if a task can reduce its nice value | |
4566 | * @p: task | |
4567 | * @nice: nice value | |
4568 | */ | |
36c8b586 | 4569 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4570 | { |
024f4747 MM |
4571 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4572 | int nice_rlim = 20 - nice; | |
48f24c4d | 4573 | |
e43379f1 MM |
4574 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4575 | capable(CAP_SYS_NICE)); | |
4576 | } | |
4577 | ||
1da177e4 LT |
4578 | #ifdef __ARCH_WANT_SYS_NICE |
4579 | ||
4580 | /* | |
4581 | * sys_nice - change the priority of the current process. | |
4582 | * @increment: priority increment | |
4583 | * | |
4584 | * sys_setpriority is a more generic, but much slower function that | |
4585 | * does similar things. | |
4586 | */ | |
4587 | asmlinkage long sys_nice(int increment) | |
4588 | { | |
48f24c4d | 4589 | long nice, retval; |
1da177e4 LT |
4590 | |
4591 | /* | |
4592 | * Setpriority might change our priority at the same moment. | |
4593 | * We don't have to worry. Conceptually one call occurs first | |
4594 | * and we have a single winner. | |
4595 | */ | |
e43379f1 MM |
4596 | if (increment < -40) |
4597 | increment = -40; | |
1da177e4 LT |
4598 | if (increment > 40) |
4599 | increment = 40; | |
4600 | ||
4601 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4602 | if (nice < -20) | |
4603 | nice = -20; | |
4604 | if (nice > 19) | |
4605 | nice = 19; | |
4606 | ||
e43379f1 MM |
4607 | if (increment < 0 && !can_nice(current, nice)) |
4608 | return -EPERM; | |
4609 | ||
1da177e4 LT |
4610 | retval = security_task_setnice(current, nice); |
4611 | if (retval) | |
4612 | return retval; | |
4613 | ||
4614 | set_user_nice(current, nice); | |
4615 | return 0; | |
4616 | } | |
4617 | ||
4618 | #endif | |
4619 | ||
4620 | /** | |
4621 | * task_prio - return the priority value of a given task. | |
4622 | * @p: the task in question. | |
4623 | * | |
4624 | * This is the priority value as seen by users in /proc. | |
4625 | * RT tasks are offset by -200. Normal tasks are centered | |
4626 | * around 0, value goes from -16 to +15. | |
4627 | */ | |
36c8b586 | 4628 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4629 | { |
4630 | return p->prio - MAX_RT_PRIO; | |
4631 | } | |
4632 | ||
4633 | /** | |
4634 | * task_nice - return the nice value of a given task. | |
4635 | * @p: the task in question. | |
4636 | */ | |
36c8b586 | 4637 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4638 | { |
4639 | return TASK_NICE(p); | |
4640 | } | |
150d8bed | 4641 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4642 | |
4643 | /** | |
4644 | * idle_cpu - is a given cpu idle currently? | |
4645 | * @cpu: the processor in question. | |
4646 | */ | |
4647 | int idle_cpu(int cpu) | |
4648 | { | |
4649 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4650 | } | |
4651 | ||
1da177e4 LT |
4652 | /** |
4653 | * idle_task - return the idle task for a given cpu. | |
4654 | * @cpu: the processor in question. | |
4655 | */ | |
36c8b586 | 4656 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4657 | { |
4658 | return cpu_rq(cpu)->idle; | |
4659 | } | |
4660 | ||
4661 | /** | |
4662 | * find_process_by_pid - find a process with a matching PID value. | |
4663 | * @pid: the pid in question. | |
4664 | */ | |
a9957449 | 4665 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4666 | { |
228ebcbe | 4667 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4668 | } |
4669 | ||
4670 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4671 | static void |
4672 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4673 | { |
dd41f596 | 4674 | BUG_ON(p->se.on_rq); |
48f24c4d | 4675 | |
1da177e4 | 4676 | p->policy = policy; |
dd41f596 IM |
4677 | switch (p->policy) { |
4678 | case SCHED_NORMAL: | |
4679 | case SCHED_BATCH: | |
4680 | case SCHED_IDLE: | |
4681 | p->sched_class = &fair_sched_class; | |
4682 | break; | |
4683 | case SCHED_FIFO: | |
4684 | case SCHED_RR: | |
4685 | p->sched_class = &rt_sched_class; | |
4686 | break; | |
4687 | } | |
4688 | ||
1da177e4 | 4689 | p->rt_priority = prio; |
b29739f9 IM |
4690 | p->normal_prio = normal_prio(p); |
4691 | /* we are holding p->pi_lock already */ | |
4692 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4693 | set_load_weight(p); |
1da177e4 LT |
4694 | } |
4695 | ||
4696 | /** | |
72fd4a35 | 4697 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4698 | * @p: the task in question. |
4699 | * @policy: new policy. | |
4700 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4701 | * |
72fd4a35 | 4702 | * NOTE that the task may be already dead. |
1da177e4 | 4703 | */ |
95cdf3b7 IM |
4704 | int sched_setscheduler(struct task_struct *p, int policy, |
4705 | struct sched_param *param) | |
1da177e4 | 4706 | { |
83b699ed | 4707 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4708 | unsigned long flags; |
cb469845 | 4709 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4710 | struct rq *rq; |
1da177e4 | 4711 | |
66e5393a SR |
4712 | /* may grab non-irq protected spin_locks */ |
4713 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4714 | recheck: |
4715 | /* double check policy once rq lock held */ | |
4716 | if (policy < 0) | |
4717 | policy = oldpolicy = p->policy; | |
4718 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4719 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4720 | policy != SCHED_IDLE) | |
b0a9499c | 4721 | return -EINVAL; |
1da177e4 LT |
4722 | /* |
4723 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4724 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4725 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4726 | */ |
4727 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4728 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4729 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4730 | return -EINVAL; |
e05606d3 | 4731 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4732 | return -EINVAL; |
4733 | ||
37e4ab3f OC |
4734 | /* |
4735 | * Allow unprivileged RT tasks to decrease priority: | |
4736 | */ | |
4737 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4738 | if (rt_policy(policy)) { |
8dc3e909 | 4739 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4740 | |
4741 | if (!lock_task_sighand(p, &flags)) | |
4742 | return -ESRCH; | |
4743 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4744 | unlock_task_sighand(p, &flags); | |
4745 | ||
4746 | /* can't set/change the rt policy */ | |
4747 | if (policy != p->policy && !rlim_rtprio) | |
4748 | return -EPERM; | |
4749 | ||
4750 | /* can't increase priority */ | |
4751 | if (param->sched_priority > p->rt_priority && | |
4752 | param->sched_priority > rlim_rtprio) | |
4753 | return -EPERM; | |
4754 | } | |
dd41f596 IM |
4755 | /* |
4756 | * Like positive nice levels, dont allow tasks to | |
4757 | * move out of SCHED_IDLE either: | |
4758 | */ | |
4759 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4760 | return -EPERM; | |
5fe1d75f | 4761 | |
37e4ab3f OC |
4762 | /* can't change other user's priorities */ |
4763 | if ((current->euid != p->euid) && | |
4764 | (current->euid != p->uid)) | |
4765 | return -EPERM; | |
4766 | } | |
1da177e4 | 4767 | |
b68aa230 PZ |
4768 | #ifdef CONFIG_RT_GROUP_SCHED |
4769 | /* | |
4770 | * Do not allow realtime tasks into groups that have no runtime | |
4771 | * assigned. | |
4772 | */ | |
d0b27fa7 | 4773 | if (rt_policy(policy) && task_group(p)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
4774 | return -EPERM; |
4775 | #endif | |
4776 | ||
1da177e4 LT |
4777 | retval = security_task_setscheduler(p, policy, param); |
4778 | if (retval) | |
4779 | return retval; | |
b29739f9 IM |
4780 | /* |
4781 | * make sure no PI-waiters arrive (or leave) while we are | |
4782 | * changing the priority of the task: | |
4783 | */ | |
4784 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4785 | /* |
4786 | * To be able to change p->policy safely, the apropriate | |
4787 | * runqueue lock must be held. | |
4788 | */ | |
b29739f9 | 4789 | rq = __task_rq_lock(p); |
1da177e4 LT |
4790 | /* recheck policy now with rq lock held */ |
4791 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4792 | policy = oldpolicy = -1; | |
b29739f9 IM |
4793 | __task_rq_unlock(rq); |
4794 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4795 | goto recheck; |
4796 | } | |
2daa3577 | 4797 | update_rq_clock(rq); |
dd41f596 | 4798 | on_rq = p->se.on_rq; |
051a1d1a | 4799 | running = task_current(rq, p); |
0e1f3483 | 4800 | if (on_rq) |
2e1cb74a | 4801 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4802 | if (running) |
4803 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4804 | |
1da177e4 | 4805 | oldprio = p->prio; |
dd41f596 | 4806 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4807 | |
0e1f3483 HS |
4808 | if (running) |
4809 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4810 | if (on_rq) { |
4811 | activate_task(rq, p, 0); | |
cb469845 SR |
4812 | |
4813 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4814 | } |
b29739f9 IM |
4815 | __task_rq_unlock(rq); |
4816 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4817 | ||
95e02ca9 TG |
4818 | rt_mutex_adjust_pi(p); |
4819 | ||
1da177e4 LT |
4820 | return 0; |
4821 | } | |
4822 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4823 | ||
95cdf3b7 IM |
4824 | static int |
4825 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4826 | { |
1da177e4 LT |
4827 | struct sched_param lparam; |
4828 | struct task_struct *p; | |
36c8b586 | 4829 | int retval; |
1da177e4 LT |
4830 | |
4831 | if (!param || pid < 0) | |
4832 | return -EINVAL; | |
4833 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4834 | return -EFAULT; | |
5fe1d75f ON |
4835 | |
4836 | rcu_read_lock(); | |
4837 | retval = -ESRCH; | |
1da177e4 | 4838 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4839 | if (p != NULL) |
4840 | retval = sched_setscheduler(p, policy, &lparam); | |
4841 | rcu_read_unlock(); | |
36c8b586 | 4842 | |
1da177e4 LT |
4843 | return retval; |
4844 | } | |
4845 | ||
4846 | /** | |
4847 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4848 | * @pid: the pid in question. | |
4849 | * @policy: new policy. | |
4850 | * @param: structure containing the new RT priority. | |
4851 | */ | |
41a2d6cf IM |
4852 | asmlinkage long |
4853 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4854 | { |
c21761f1 JB |
4855 | /* negative values for policy are not valid */ |
4856 | if (policy < 0) | |
4857 | return -EINVAL; | |
4858 | ||
1da177e4 LT |
4859 | return do_sched_setscheduler(pid, policy, param); |
4860 | } | |
4861 | ||
4862 | /** | |
4863 | * sys_sched_setparam - set/change the RT priority of a thread | |
4864 | * @pid: the pid in question. | |
4865 | * @param: structure containing the new RT priority. | |
4866 | */ | |
4867 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4868 | { | |
4869 | return do_sched_setscheduler(pid, -1, param); | |
4870 | } | |
4871 | ||
4872 | /** | |
4873 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4874 | * @pid: the pid in question. | |
4875 | */ | |
4876 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4877 | { | |
36c8b586 | 4878 | struct task_struct *p; |
3a5c359a | 4879 | int retval; |
1da177e4 LT |
4880 | |
4881 | if (pid < 0) | |
3a5c359a | 4882 | return -EINVAL; |
1da177e4 LT |
4883 | |
4884 | retval = -ESRCH; | |
4885 | read_lock(&tasklist_lock); | |
4886 | p = find_process_by_pid(pid); | |
4887 | if (p) { | |
4888 | retval = security_task_getscheduler(p); | |
4889 | if (!retval) | |
4890 | retval = p->policy; | |
4891 | } | |
4892 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4893 | return retval; |
4894 | } | |
4895 | ||
4896 | /** | |
4897 | * sys_sched_getscheduler - get the RT priority of a thread | |
4898 | * @pid: the pid in question. | |
4899 | * @param: structure containing the RT priority. | |
4900 | */ | |
4901 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4902 | { | |
4903 | struct sched_param lp; | |
36c8b586 | 4904 | struct task_struct *p; |
3a5c359a | 4905 | int retval; |
1da177e4 LT |
4906 | |
4907 | if (!param || pid < 0) | |
3a5c359a | 4908 | return -EINVAL; |
1da177e4 LT |
4909 | |
4910 | read_lock(&tasklist_lock); | |
4911 | p = find_process_by_pid(pid); | |
4912 | retval = -ESRCH; | |
4913 | if (!p) | |
4914 | goto out_unlock; | |
4915 | ||
4916 | retval = security_task_getscheduler(p); | |
4917 | if (retval) | |
4918 | goto out_unlock; | |
4919 | ||
4920 | lp.sched_priority = p->rt_priority; | |
4921 | read_unlock(&tasklist_lock); | |
4922 | ||
4923 | /* | |
4924 | * This one might sleep, we cannot do it with a spinlock held ... | |
4925 | */ | |
4926 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4927 | ||
1da177e4 LT |
4928 | return retval; |
4929 | ||
4930 | out_unlock: | |
4931 | read_unlock(&tasklist_lock); | |
4932 | return retval; | |
4933 | } | |
4934 | ||
b53e921b | 4935 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 4936 | { |
1da177e4 | 4937 | cpumask_t cpus_allowed; |
b53e921b | 4938 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
4939 | struct task_struct *p; |
4940 | int retval; | |
1da177e4 | 4941 | |
95402b38 | 4942 | get_online_cpus(); |
1da177e4 LT |
4943 | read_lock(&tasklist_lock); |
4944 | ||
4945 | p = find_process_by_pid(pid); | |
4946 | if (!p) { | |
4947 | read_unlock(&tasklist_lock); | |
95402b38 | 4948 | put_online_cpus(); |
1da177e4 LT |
4949 | return -ESRCH; |
4950 | } | |
4951 | ||
4952 | /* | |
4953 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 4954 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
4955 | * usage count and then drop tasklist_lock. |
4956 | */ | |
4957 | get_task_struct(p); | |
4958 | read_unlock(&tasklist_lock); | |
4959 | ||
4960 | retval = -EPERM; | |
4961 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4962 | !capable(CAP_SYS_NICE)) | |
4963 | goto out_unlock; | |
4964 | ||
e7834f8f DQ |
4965 | retval = security_task_setscheduler(p, 0, NULL); |
4966 | if (retval) | |
4967 | goto out_unlock; | |
4968 | ||
f9a86fcb | 4969 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 4970 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 4971 | again: |
7c16ec58 | 4972 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 4973 | |
8707d8b8 | 4974 | if (!retval) { |
f9a86fcb | 4975 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
4976 | if (!cpus_subset(new_mask, cpus_allowed)) { |
4977 | /* | |
4978 | * We must have raced with a concurrent cpuset | |
4979 | * update. Just reset the cpus_allowed to the | |
4980 | * cpuset's cpus_allowed | |
4981 | */ | |
4982 | new_mask = cpus_allowed; | |
4983 | goto again; | |
4984 | } | |
4985 | } | |
1da177e4 LT |
4986 | out_unlock: |
4987 | put_task_struct(p); | |
95402b38 | 4988 | put_online_cpus(); |
1da177e4 LT |
4989 | return retval; |
4990 | } | |
4991 | ||
4992 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4993 | cpumask_t *new_mask) | |
4994 | { | |
4995 | if (len < sizeof(cpumask_t)) { | |
4996 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4997 | } else if (len > sizeof(cpumask_t)) { | |
4998 | len = sizeof(cpumask_t); | |
4999 | } | |
5000 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5001 | } | |
5002 | ||
5003 | /** | |
5004 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5005 | * @pid: pid of the process | |
5006 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5007 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5008 | */ | |
5009 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5010 | unsigned long __user *user_mask_ptr) | |
5011 | { | |
5012 | cpumask_t new_mask; | |
5013 | int retval; | |
5014 | ||
5015 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5016 | if (retval) | |
5017 | return retval; | |
5018 | ||
b53e921b | 5019 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5020 | } |
5021 | ||
5022 | /* | |
5023 | * Represents all cpu's present in the system | |
5024 | * In systems capable of hotplug, this map could dynamically grow | |
5025 | * as new cpu's are detected in the system via any platform specific | |
5026 | * method, such as ACPI for e.g. | |
5027 | */ | |
5028 | ||
4cef0c61 | 5029 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
5030 | EXPORT_SYMBOL(cpu_present_map); |
5031 | ||
5032 | #ifndef CONFIG_SMP | |
4cef0c61 | 5033 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
5034 | EXPORT_SYMBOL(cpu_online_map); |
5035 | ||
4cef0c61 | 5036 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 5037 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
5038 | #endif |
5039 | ||
5040 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
5041 | { | |
36c8b586 | 5042 | struct task_struct *p; |
1da177e4 | 5043 | int retval; |
1da177e4 | 5044 | |
95402b38 | 5045 | get_online_cpus(); |
1da177e4 LT |
5046 | read_lock(&tasklist_lock); |
5047 | ||
5048 | retval = -ESRCH; | |
5049 | p = find_process_by_pid(pid); | |
5050 | if (!p) | |
5051 | goto out_unlock; | |
5052 | ||
e7834f8f DQ |
5053 | retval = security_task_getscheduler(p); |
5054 | if (retval) | |
5055 | goto out_unlock; | |
5056 | ||
2f7016d9 | 5057 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5058 | |
5059 | out_unlock: | |
5060 | read_unlock(&tasklist_lock); | |
95402b38 | 5061 | put_online_cpus(); |
1da177e4 | 5062 | |
9531b62f | 5063 | return retval; |
1da177e4 LT |
5064 | } |
5065 | ||
5066 | /** | |
5067 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5068 | * @pid: pid of the process | |
5069 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5070 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5071 | */ | |
5072 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5073 | unsigned long __user *user_mask_ptr) | |
5074 | { | |
5075 | int ret; | |
5076 | cpumask_t mask; | |
5077 | ||
5078 | if (len < sizeof(cpumask_t)) | |
5079 | return -EINVAL; | |
5080 | ||
5081 | ret = sched_getaffinity(pid, &mask); | |
5082 | if (ret < 0) | |
5083 | return ret; | |
5084 | ||
5085 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5086 | return -EFAULT; | |
5087 | ||
5088 | return sizeof(cpumask_t); | |
5089 | } | |
5090 | ||
5091 | /** | |
5092 | * sys_sched_yield - yield the current processor to other threads. | |
5093 | * | |
dd41f596 IM |
5094 | * This function yields the current CPU to other tasks. If there are no |
5095 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5096 | */ |
5097 | asmlinkage long sys_sched_yield(void) | |
5098 | { | |
70b97a7f | 5099 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5100 | |
2d72376b | 5101 | schedstat_inc(rq, yld_count); |
4530d7ab | 5102 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5103 | |
5104 | /* | |
5105 | * Since we are going to call schedule() anyway, there's | |
5106 | * no need to preempt or enable interrupts: | |
5107 | */ | |
5108 | __release(rq->lock); | |
8a25d5de | 5109 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5110 | _raw_spin_unlock(&rq->lock); |
5111 | preempt_enable_no_resched(); | |
5112 | ||
5113 | schedule(); | |
5114 | ||
5115 | return 0; | |
5116 | } | |
5117 | ||
e7b38404 | 5118 | static void __cond_resched(void) |
1da177e4 | 5119 | { |
8e0a43d8 IM |
5120 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5121 | __might_sleep(__FILE__, __LINE__); | |
5122 | #endif | |
5bbcfd90 IM |
5123 | /* |
5124 | * The BKS might be reacquired before we have dropped | |
5125 | * PREEMPT_ACTIVE, which could trigger a second | |
5126 | * cond_resched() call. | |
5127 | */ | |
1da177e4 LT |
5128 | do { |
5129 | add_preempt_count(PREEMPT_ACTIVE); | |
5130 | schedule(); | |
5131 | sub_preempt_count(PREEMPT_ACTIVE); | |
5132 | } while (need_resched()); | |
5133 | } | |
5134 | ||
02b67cc3 HX |
5135 | #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PREEMPT_VOLUNTARY) |
5136 | int __sched _cond_resched(void) | |
1da177e4 | 5137 | { |
9414232f IM |
5138 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5139 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5140 | __cond_resched(); |
5141 | return 1; | |
5142 | } | |
5143 | return 0; | |
5144 | } | |
02b67cc3 HX |
5145 | EXPORT_SYMBOL(_cond_resched); |
5146 | #endif | |
1da177e4 LT |
5147 | |
5148 | /* | |
5149 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5150 | * call schedule, and on return reacquire the lock. | |
5151 | * | |
41a2d6cf | 5152 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5153 | * operations here to prevent schedule() from being called twice (once via |
5154 | * spin_unlock(), once by hand). | |
5155 | */ | |
95cdf3b7 | 5156 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5157 | { |
95c354fe | 5158 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5159 | int ret = 0; |
5160 | ||
95c354fe | 5161 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5162 | spin_unlock(lock); |
95c354fe NP |
5163 | if (resched && need_resched()) |
5164 | __cond_resched(); | |
5165 | else | |
5166 | cpu_relax(); | |
6df3cecb | 5167 | ret = 1; |
1da177e4 | 5168 | spin_lock(lock); |
1da177e4 | 5169 | } |
6df3cecb | 5170 | return ret; |
1da177e4 | 5171 | } |
1da177e4 LT |
5172 | EXPORT_SYMBOL(cond_resched_lock); |
5173 | ||
5174 | int __sched cond_resched_softirq(void) | |
5175 | { | |
5176 | BUG_ON(!in_softirq()); | |
5177 | ||
9414232f | 5178 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5179 | local_bh_enable(); |
1da177e4 LT |
5180 | __cond_resched(); |
5181 | local_bh_disable(); | |
5182 | return 1; | |
5183 | } | |
5184 | return 0; | |
5185 | } | |
1da177e4 LT |
5186 | EXPORT_SYMBOL(cond_resched_softirq); |
5187 | ||
1da177e4 LT |
5188 | /** |
5189 | * yield - yield the current processor to other threads. | |
5190 | * | |
72fd4a35 | 5191 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5192 | * thread runnable and calls sys_sched_yield(). |
5193 | */ | |
5194 | void __sched yield(void) | |
5195 | { | |
5196 | set_current_state(TASK_RUNNING); | |
5197 | sys_sched_yield(); | |
5198 | } | |
1da177e4 LT |
5199 | EXPORT_SYMBOL(yield); |
5200 | ||
5201 | /* | |
41a2d6cf | 5202 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5203 | * that process accounting knows that this is a task in IO wait state. |
5204 | * | |
5205 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5206 | * has set its backing_dev_info: the queue against which it should throttle) | |
5207 | */ | |
5208 | void __sched io_schedule(void) | |
5209 | { | |
70b97a7f | 5210 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5211 | |
0ff92245 | 5212 | delayacct_blkio_start(); |
1da177e4 LT |
5213 | atomic_inc(&rq->nr_iowait); |
5214 | schedule(); | |
5215 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5216 | delayacct_blkio_end(); |
1da177e4 | 5217 | } |
1da177e4 LT |
5218 | EXPORT_SYMBOL(io_schedule); |
5219 | ||
5220 | long __sched io_schedule_timeout(long timeout) | |
5221 | { | |
70b97a7f | 5222 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5223 | long ret; |
5224 | ||
0ff92245 | 5225 | delayacct_blkio_start(); |
1da177e4 LT |
5226 | atomic_inc(&rq->nr_iowait); |
5227 | ret = schedule_timeout(timeout); | |
5228 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5229 | delayacct_blkio_end(); |
1da177e4 LT |
5230 | return ret; |
5231 | } | |
5232 | ||
5233 | /** | |
5234 | * sys_sched_get_priority_max - return maximum RT priority. | |
5235 | * @policy: scheduling class. | |
5236 | * | |
5237 | * this syscall returns the maximum rt_priority that can be used | |
5238 | * by a given scheduling class. | |
5239 | */ | |
5240 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5241 | { | |
5242 | int ret = -EINVAL; | |
5243 | ||
5244 | switch (policy) { | |
5245 | case SCHED_FIFO: | |
5246 | case SCHED_RR: | |
5247 | ret = MAX_USER_RT_PRIO-1; | |
5248 | break; | |
5249 | case SCHED_NORMAL: | |
b0a9499c | 5250 | case SCHED_BATCH: |
dd41f596 | 5251 | case SCHED_IDLE: |
1da177e4 LT |
5252 | ret = 0; |
5253 | break; | |
5254 | } | |
5255 | return ret; | |
5256 | } | |
5257 | ||
5258 | /** | |
5259 | * sys_sched_get_priority_min - return minimum RT priority. | |
5260 | * @policy: scheduling class. | |
5261 | * | |
5262 | * this syscall returns the minimum rt_priority that can be used | |
5263 | * by a given scheduling class. | |
5264 | */ | |
5265 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5266 | { | |
5267 | int ret = -EINVAL; | |
5268 | ||
5269 | switch (policy) { | |
5270 | case SCHED_FIFO: | |
5271 | case SCHED_RR: | |
5272 | ret = 1; | |
5273 | break; | |
5274 | case SCHED_NORMAL: | |
b0a9499c | 5275 | case SCHED_BATCH: |
dd41f596 | 5276 | case SCHED_IDLE: |
1da177e4 LT |
5277 | ret = 0; |
5278 | } | |
5279 | return ret; | |
5280 | } | |
5281 | ||
5282 | /** | |
5283 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5284 | * @pid: pid of the process. | |
5285 | * @interval: userspace pointer to the timeslice value. | |
5286 | * | |
5287 | * this syscall writes the default timeslice value of a given process | |
5288 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5289 | */ | |
5290 | asmlinkage | |
5291 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5292 | { | |
36c8b586 | 5293 | struct task_struct *p; |
a4ec24b4 | 5294 | unsigned int time_slice; |
3a5c359a | 5295 | int retval; |
1da177e4 | 5296 | struct timespec t; |
1da177e4 LT |
5297 | |
5298 | if (pid < 0) | |
3a5c359a | 5299 | return -EINVAL; |
1da177e4 LT |
5300 | |
5301 | retval = -ESRCH; | |
5302 | read_lock(&tasklist_lock); | |
5303 | p = find_process_by_pid(pid); | |
5304 | if (!p) | |
5305 | goto out_unlock; | |
5306 | ||
5307 | retval = security_task_getscheduler(p); | |
5308 | if (retval) | |
5309 | goto out_unlock; | |
5310 | ||
77034937 IM |
5311 | /* |
5312 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5313 | * tasks that are on an otherwise idle runqueue: | |
5314 | */ | |
5315 | time_slice = 0; | |
5316 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5317 | time_slice = DEF_TIMESLICE; |
1868f958 | 5318 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5319 | struct sched_entity *se = &p->se; |
5320 | unsigned long flags; | |
5321 | struct rq *rq; | |
5322 | ||
5323 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5324 | if (rq->cfs.load.weight) |
5325 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5326 | task_rq_unlock(rq, &flags); |
5327 | } | |
1da177e4 | 5328 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5329 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5330 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5331 | return retval; |
3a5c359a | 5332 | |
1da177e4 LT |
5333 | out_unlock: |
5334 | read_unlock(&tasklist_lock); | |
5335 | return retval; | |
5336 | } | |
5337 | ||
2ed6e34f | 5338 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 | 5339 | |
82a1fcb9 | 5340 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5341 | { |
1da177e4 | 5342 | unsigned long free = 0; |
36c8b586 | 5343 | unsigned state; |
1da177e4 | 5344 | |
1da177e4 | 5345 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5346 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5347 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5348 | #if BITS_PER_LONG == 32 |
1da177e4 | 5349 | if (state == TASK_RUNNING) |
cc4ea795 | 5350 | printk(KERN_CONT " running "); |
1da177e4 | 5351 | else |
cc4ea795 | 5352 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5353 | #else |
5354 | if (state == TASK_RUNNING) | |
cc4ea795 | 5355 | printk(KERN_CONT " running task "); |
1da177e4 | 5356 | else |
cc4ea795 | 5357 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5358 | #endif |
5359 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5360 | { | |
10ebffde | 5361 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5362 | while (!*n) |
5363 | n++; | |
10ebffde | 5364 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5365 | } |
5366 | #endif | |
ba25f9dc | 5367 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5368 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5369 | |
5fb5e6de | 5370 | show_stack(p, NULL); |
1da177e4 LT |
5371 | } |
5372 | ||
e59e2ae2 | 5373 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5374 | { |
36c8b586 | 5375 | struct task_struct *g, *p; |
1da177e4 | 5376 | |
4bd77321 IM |
5377 | #if BITS_PER_LONG == 32 |
5378 | printk(KERN_INFO | |
5379 | " task PC stack pid father\n"); | |
1da177e4 | 5380 | #else |
4bd77321 IM |
5381 | printk(KERN_INFO |
5382 | " task PC stack pid father\n"); | |
1da177e4 LT |
5383 | #endif |
5384 | read_lock(&tasklist_lock); | |
5385 | do_each_thread(g, p) { | |
5386 | /* | |
5387 | * reset the NMI-timeout, listing all files on a slow | |
5388 | * console might take alot of time: | |
5389 | */ | |
5390 | touch_nmi_watchdog(); | |
39bc89fd | 5391 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5392 | sched_show_task(p); |
1da177e4 LT |
5393 | } while_each_thread(g, p); |
5394 | ||
04c9167f JF |
5395 | touch_all_softlockup_watchdogs(); |
5396 | ||
dd41f596 IM |
5397 | #ifdef CONFIG_SCHED_DEBUG |
5398 | sysrq_sched_debug_show(); | |
5399 | #endif | |
1da177e4 | 5400 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5401 | /* |
5402 | * Only show locks if all tasks are dumped: | |
5403 | */ | |
5404 | if (state_filter == -1) | |
5405 | debug_show_all_locks(); | |
1da177e4 LT |
5406 | } |
5407 | ||
1df21055 IM |
5408 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5409 | { | |
dd41f596 | 5410 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5411 | } |
5412 | ||
f340c0d1 IM |
5413 | /** |
5414 | * init_idle - set up an idle thread for a given CPU | |
5415 | * @idle: task in question | |
5416 | * @cpu: cpu the idle task belongs to | |
5417 | * | |
5418 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5419 | * flag, to make booting more robust. | |
5420 | */ | |
5c1e1767 | 5421 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5422 | { |
70b97a7f | 5423 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5424 | unsigned long flags; |
5425 | ||
dd41f596 IM |
5426 | __sched_fork(idle); |
5427 | idle->se.exec_start = sched_clock(); | |
5428 | ||
b29739f9 | 5429 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5430 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5431 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5432 | |
5433 | spin_lock_irqsave(&rq->lock, flags); | |
5434 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5435 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5436 | idle->oncpu = 1; | |
5437 | #endif | |
1da177e4 LT |
5438 | spin_unlock_irqrestore(&rq->lock, flags); |
5439 | ||
5440 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 5441 | task_thread_info(idle)->preempt_count = 0; |
6478d880 | 5442 | |
dd41f596 IM |
5443 | /* |
5444 | * The idle tasks have their own, simple scheduling class: | |
5445 | */ | |
5446 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5447 | } |
5448 | ||
5449 | /* | |
5450 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5451 | * indicates which cpus entered this state. This is used | |
5452 | * in the rcu update to wait only for active cpus. For system | |
5453 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5454 | * always be CPU_MASK_NONE. | |
5455 | */ | |
5456 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5457 | ||
19978ca6 IM |
5458 | /* |
5459 | * Increase the granularity value when there are more CPUs, | |
5460 | * because with more CPUs the 'effective latency' as visible | |
5461 | * to users decreases. But the relationship is not linear, | |
5462 | * so pick a second-best guess by going with the log2 of the | |
5463 | * number of CPUs. | |
5464 | * | |
5465 | * This idea comes from the SD scheduler of Con Kolivas: | |
5466 | */ | |
5467 | static inline void sched_init_granularity(void) | |
5468 | { | |
5469 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5470 | const unsigned long limit = 200000000; | |
5471 | ||
5472 | sysctl_sched_min_granularity *= factor; | |
5473 | if (sysctl_sched_min_granularity > limit) | |
5474 | sysctl_sched_min_granularity = limit; | |
5475 | ||
5476 | sysctl_sched_latency *= factor; | |
5477 | if (sysctl_sched_latency > limit) | |
5478 | sysctl_sched_latency = limit; | |
5479 | ||
5480 | sysctl_sched_wakeup_granularity *= factor; | |
19978ca6 IM |
5481 | } |
5482 | ||
1da177e4 LT |
5483 | #ifdef CONFIG_SMP |
5484 | /* | |
5485 | * This is how migration works: | |
5486 | * | |
70b97a7f | 5487 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5488 | * runqueue and wake up that CPU's migration thread. |
5489 | * 2) we down() the locked semaphore => thread blocks. | |
5490 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5491 | * thread off the CPU) | |
5492 | * 4) it gets the migration request and checks whether the migrated | |
5493 | * task is still in the wrong runqueue. | |
5494 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5495 | * it and puts it into the right queue. | |
5496 | * 6) migration thread up()s the semaphore. | |
5497 | * 7) we wake up and the migration is done. | |
5498 | */ | |
5499 | ||
5500 | /* | |
5501 | * Change a given task's CPU affinity. Migrate the thread to a | |
5502 | * proper CPU and schedule it away if the CPU it's executing on | |
5503 | * is removed from the allowed bitmask. | |
5504 | * | |
5505 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5506 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5507 | * call is not atomic; no spinlocks may be held. |
5508 | */ | |
cd8ba7cd | 5509 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5510 | { |
70b97a7f | 5511 | struct migration_req req; |
1da177e4 | 5512 | unsigned long flags; |
70b97a7f | 5513 | struct rq *rq; |
48f24c4d | 5514 | int ret = 0; |
1da177e4 LT |
5515 | |
5516 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5517 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5518 | ret = -EINVAL; |
5519 | goto out; | |
5520 | } | |
5521 | ||
73fe6aae | 5522 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5523 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5524 | else { |
cd8ba7cd MT |
5525 | p->cpus_allowed = *new_mask; |
5526 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5527 | } |
5528 | ||
1da177e4 | 5529 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5530 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5531 | goto out; |
5532 | ||
cd8ba7cd | 5533 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5534 | /* Need help from migration thread: drop lock and wait. */ |
5535 | task_rq_unlock(rq, &flags); | |
5536 | wake_up_process(rq->migration_thread); | |
5537 | wait_for_completion(&req.done); | |
5538 | tlb_migrate_finish(p->mm); | |
5539 | return 0; | |
5540 | } | |
5541 | out: | |
5542 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5543 | |
1da177e4 LT |
5544 | return ret; |
5545 | } | |
cd8ba7cd | 5546 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5547 | |
5548 | /* | |
41a2d6cf | 5549 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5550 | * this because either it can't run here any more (set_cpus_allowed() |
5551 | * away from this CPU, or CPU going down), or because we're | |
5552 | * attempting to rebalance this task on exec (sched_exec). | |
5553 | * | |
5554 | * So we race with normal scheduler movements, but that's OK, as long | |
5555 | * as the task is no longer on this CPU. | |
efc30814 KK |
5556 | * |
5557 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5558 | */ |
efc30814 | 5559 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5560 | { |
70b97a7f | 5561 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5562 | int ret = 0, on_rq; |
1da177e4 LT |
5563 | |
5564 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5565 | return ret; |
1da177e4 LT |
5566 | |
5567 | rq_src = cpu_rq(src_cpu); | |
5568 | rq_dest = cpu_rq(dest_cpu); | |
5569 | ||
5570 | double_rq_lock(rq_src, rq_dest); | |
5571 | /* Already moved. */ | |
5572 | if (task_cpu(p) != src_cpu) | |
5573 | goto out; | |
5574 | /* Affinity changed (again). */ | |
5575 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5576 | goto out; | |
5577 | ||
dd41f596 | 5578 | on_rq = p->se.on_rq; |
6e82a3be | 5579 | if (on_rq) |
2e1cb74a | 5580 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5581 | |
1da177e4 | 5582 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5583 | if (on_rq) { |
5584 | activate_task(rq_dest, p, 0); | |
5585 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5586 | } |
efc30814 | 5587 | ret = 1; |
1da177e4 LT |
5588 | out: |
5589 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5590 | return ret; |
1da177e4 LT |
5591 | } |
5592 | ||
5593 | /* | |
5594 | * migration_thread - this is a highprio system thread that performs | |
5595 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5596 | * another runqueue. | |
5597 | */ | |
95cdf3b7 | 5598 | static int migration_thread(void *data) |
1da177e4 | 5599 | { |
1da177e4 | 5600 | int cpu = (long)data; |
70b97a7f | 5601 | struct rq *rq; |
1da177e4 LT |
5602 | |
5603 | rq = cpu_rq(cpu); | |
5604 | BUG_ON(rq->migration_thread != current); | |
5605 | ||
5606 | set_current_state(TASK_INTERRUPTIBLE); | |
5607 | while (!kthread_should_stop()) { | |
70b97a7f | 5608 | struct migration_req *req; |
1da177e4 | 5609 | struct list_head *head; |
1da177e4 | 5610 | |
1da177e4 LT |
5611 | spin_lock_irq(&rq->lock); |
5612 | ||
5613 | if (cpu_is_offline(cpu)) { | |
5614 | spin_unlock_irq(&rq->lock); | |
5615 | goto wait_to_die; | |
5616 | } | |
5617 | ||
5618 | if (rq->active_balance) { | |
5619 | active_load_balance(rq, cpu); | |
5620 | rq->active_balance = 0; | |
5621 | } | |
5622 | ||
5623 | head = &rq->migration_queue; | |
5624 | ||
5625 | if (list_empty(head)) { | |
5626 | spin_unlock_irq(&rq->lock); | |
5627 | schedule(); | |
5628 | set_current_state(TASK_INTERRUPTIBLE); | |
5629 | continue; | |
5630 | } | |
70b97a7f | 5631 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5632 | list_del_init(head->next); |
5633 | ||
674311d5 NP |
5634 | spin_unlock(&rq->lock); |
5635 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5636 | local_irq_enable(); | |
1da177e4 LT |
5637 | |
5638 | complete(&req->done); | |
5639 | } | |
5640 | __set_current_state(TASK_RUNNING); | |
5641 | return 0; | |
5642 | ||
5643 | wait_to_die: | |
5644 | /* Wait for kthread_stop */ | |
5645 | set_current_state(TASK_INTERRUPTIBLE); | |
5646 | while (!kthread_should_stop()) { | |
5647 | schedule(); | |
5648 | set_current_state(TASK_INTERRUPTIBLE); | |
5649 | } | |
5650 | __set_current_state(TASK_RUNNING); | |
5651 | return 0; | |
5652 | } | |
5653 | ||
5654 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5655 | |
5656 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5657 | { | |
5658 | int ret; | |
5659 | ||
5660 | local_irq_disable(); | |
5661 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5662 | local_irq_enable(); | |
5663 | return ret; | |
5664 | } | |
5665 | ||
054b9108 | 5666 | /* |
3a4fa0a2 | 5667 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5668 | * NOTE: interrupts should be disabled by the caller |
5669 | */ | |
48f24c4d | 5670 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5671 | { |
efc30814 | 5672 | unsigned long flags; |
1da177e4 | 5673 | cpumask_t mask; |
70b97a7f IM |
5674 | struct rq *rq; |
5675 | int dest_cpu; | |
1da177e4 | 5676 | |
3a5c359a AK |
5677 | do { |
5678 | /* On same node? */ | |
5679 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5680 | cpus_and(mask, mask, p->cpus_allowed); | |
5681 | dest_cpu = any_online_cpu(mask); | |
5682 | ||
5683 | /* On any allowed CPU? */ | |
434d53b0 | 5684 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
5685 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5686 | ||
5687 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 5688 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
5689 | cpumask_t cpus_allowed; |
5690 | ||
5691 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
5692 | /* |
5693 | * Try to stay on the same cpuset, where the | |
5694 | * current cpuset may be a subset of all cpus. | |
5695 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5696 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5697 | * called within calls to cpuset_lock/cpuset_unlock. |
5698 | */ | |
3a5c359a | 5699 | rq = task_rq_lock(p, &flags); |
470fd646 | 5700 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5701 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5702 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5703 | |
3a5c359a AK |
5704 | /* |
5705 | * Don't tell them about moving exiting tasks or | |
5706 | * kernel threads (both mm NULL), since they never | |
5707 | * leave kernel. | |
5708 | */ | |
41a2d6cf | 5709 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5710 | printk(KERN_INFO "process %d (%s) no " |
5711 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5712 | task_pid_nr(p), p->comm, dead_cpu); |
5713 | } | |
3a5c359a | 5714 | } |
f7b4cddc | 5715 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5716 | } |
5717 | ||
5718 | /* | |
5719 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5720 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5721 | * for performance reasons the counter is not stricly tracking tasks to | |
5722 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5723 | * to keep the global sum constant after CPU-down: | |
5724 | */ | |
70b97a7f | 5725 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5726 | { |
7c16ec58 | 5727 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
5728 | unsigned long flags; |
5729 | ||
5730 | local_irq_save(flags); | |
5731 | double_rq_lock(rq_src, rq_dest); | |
5732 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5733 | rq_src->nr_uninterruptible = 0; | |
5734 | double_rq_unlock(rq_src, rq_dest); | |
5735 | local_irq_restore(flags); | |
5736 | } | |
5737 | ||
5738 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5739 | static void migrate_live_tasks(int src_cpu) | |
5740 | { | |
48f24c4d | 5741 | struct task_struct *p, *t; |
1da177e4 | 5742 | |
f7b4cddc | 5743 | read_lock(&tasklist_lock); |
1da177e4 | 5744 | |
48f24c4d IM |
5745 | do_each_thread(t, p) { |
5746 | if (p == current) | |
1da177e4 LT |
5747 | continue; |
5748 | ||
48f24c4d IM |
5749 | if (task_cpu(p) == src_cpu) |
5750 | move_task_off_dead_cpu(src_cpu, p); | |
5751 | } while_each_thread(t, p); | |
1da177e4 | 5752 | |
f7b4cddc | 5753 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5754 | } |
5755 | ||
dd41f596 IM |
5756 | /* |
5757 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5758 | * It does so by boosting its priority to highest possible. |
5759 | * Used by CPU offline code. | |
1da177e4 LT |
5760 | */ |
5761 | void sched_idle_next(void) | |
5762 | { | |
48f24c4d | 5763 | int this_cpu = smp_processor_id(); |
70b97a7f | 5764 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5765 | struct task_struct *p = rq->idle; |
5766 | unsigned long flags; | |
5767 | ||
5768 | /* cpu has to be offline */ | |
48f24c4d | 5769 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5770 | |
48f24c4d IM |
5771 | /* |
5772 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5773 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5774 | */ |
5775 | spin_lock_irqsave(&rq->lock, flags); | |
5776 | ||
dd41f596 | 5777 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5778 | |
94bc9a7b DA |
5779 | update_rq_clock(rq); |
5780 | activate_task(rq, p, 0); | |
1da177e4 LT |
5781 | |
5782 | spin_unlock_irqrestore(&rq->lock, flags); | |
5783 | } | |
5784 | ||
48f24c4d IM |
5785 | /* |
5786 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5787 | * offline. |
5788 | */ | |
5789 | void idle_task_exit(void) | |
5790 | { | |
5791 | struct mm_struct *mm = current->active_mm; | |
5792 | ||
5793 | BUG_ON(cpu_online(smp_processor_id())); | |
5794 | ||
5795 | if (mm != &init_mm) | |
5796 | switch_mm(mm, &init_mm, current); | |
5797 | mmdrop(mm); | |
5798 | } | |
5799 | ||
054b9108 | 5800 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5801 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5802 | { |
70b97a7f | 5803 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5804 | |
5805 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5806 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5807 | |
5808 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5809 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5810 | |
48f24c4d | 5811 | get_task_struct(p); |
1da177e4 LT |
5812 | |
5813 | /* | |
5814 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5815 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5816 | * fine. |
5817 | */ | |
f7b4cddc | 5818 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5819 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5820 | spin_lock_irq(&rq->lock); |
1da177e4 | 5821 | |
48f24c4d | 5822 | put_task_struct(p); |
1da177e4 LT |
5823 | } |
5824 | ||
5825 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5826 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5827 | { | |
70b97a7f | 5828 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5829 | struct task_struct *next; |
48f24c4d | 5830 | |
dd41f596 IM |
5831 | for ( ; ; ) { |
5832 | if (!rq->nr_running) | |
5833 | break; | |
a8e504d2 | 5834 | update_rq_clock(rq); |
ff95f3df | 5835 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5836 | if (!next) |
5837 | break; | |
5838 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5839 | |
1da177e4 LT |
5840 | } |
5841 | } | |
5842 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5843 | ||
e692ab53 NP |
5844 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5845 | ||
5846 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5847 | { |
5848 | .procname = "sched_domain", | |
c57baf1e | 5849 | .mode = 0555, |
e0361851 | 5850 | }, |
38605cae | 5851 | {0, }, |
e692ab53 NP |
5852 | }; |
5853 | ||
5854 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5855 | { |
c57baf1e | 5856 | .ctl_name = CTL_KERN, |
e0361851 | 5857 | .procname = "kernel", |
c57baf1e | 5858 | .mode = 0555, |
e0361851 AD |
5859 | .child = sd_ctl_dir, |
5860 | }, | |
38605cae | 5861 | {0, }, |
e692ab53 NP |
5862 | }; |
5863 | ||
5864 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5865 | { | |
5866 | struct ctl_table *entry = | |
5cf9f062 | 5867 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5868 | |
e692ab53 NP |
5869 | return entry; |
5870 | } | |
5871 | ||
6382bc90 MM |
5872 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5873 | { | |
cd790076 | 5874 | struct ctl_table *entry; |
6382bc90 | 5875 | |
cd790076 MM |
5876 | /* |
5877 | * In the intermediate directories, both the child directory and | |
5878 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5879 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5880 | * static strings and all have proc handlers. |
5881 | */ | |
5882 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5883 | if (entry->child) |
5884 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5885 | if (entry->proc_handler == NULL) |
5886 | kfree(entry->procname); | |
5887 | } | |
6382bc90 MM |
5888 | |
5889 | kfree(*tablep); | |
5890 | *tablep = NULL; | |
5891 | } | |
5892 | ||
e692ab53 | 5893 | static void |
e0361851 | 5894 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5895 | const char *procname, void *data, int maxlen, |
5896 | mode_t mode, proc_handler *proc_handler) | |
5897 | { | |
e692ab53 NP |
5898 | entry->procname = procname; |
5899 | entry->data = data; | |
5900 | entry->maxlen = maxlen; | |
5901 | entry->mode = mode; | |
5902 | entry->proc_handler = proc_handler; | |
5903 | } | |
5904 | ||
5905 | static struct ctl_table * | |
5906 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5907 | { | |
ace8b3d6 | 5908 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5909 | |
ad1cdc1d MM |
5910 | if (table == NULL) |
5911 | return NULL; | |
5912 | ||
e0361851 | 5913 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5914 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5915 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5916 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5917 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5918 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5919 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5920 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5921 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5922 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5923 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5924 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5925 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5926 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5927 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5928 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5929 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5930 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5931 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5932 | &sd->cache_nice_tries, |
5933 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5934 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5935 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5936 | /* &table[11] is terminator */ |
e692ab53 NP |
5937 | |
5938 | return table; | |
5939 | } | |
5940 | ||
9a4e7159 | 5941 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5942 | { |
5943 | struct ctl_table *entry, *table; | |
5944 | struct sched_domain *sd; | |
5945 | int domain_num = 0, i; | |
5946 | char buf[32]; | |
5947 | ||
5948 | for_each_domain(cpu, sd) | |
5949 | domain_num++; | |
5950 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5951 | if (table == NULL) |
5952 | return NULL; | |
e692ab53 NP |
5953 | |
5954 | i = 0; | |
5955 | for_each_domain(cpu, sd) { | |
5956 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5957 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5958 | entry->mode = 0555; |
e692ab53 NP |
5959 | entry->child = sd_alloc_ctl_domain_table(sd); |
5960 | entry++; | |
5961 | i++; | |
5962 | } | |
5963 | return table; | |
5964 | } | |
5965 | ||
5966 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5967 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5968 | { |
5969 | int i, cpu_num = num_online_cpus(); | |
5970 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5971 | char buf[32]; | |
5972 | ||
7378547f MM |
5973 | WARN_ON(sd_ctl_dir[0].child); |
5974 | sd_ctl_dir[0].child = entry; | |
5975 | ||
ad1cdc1d MM |
5976 | if (entry == NULL) |
5977 | return; | |
5978 | ||
97b6ea7b | 5979 | for_each_online_cpu(i) { |
e692ab53 | 5980 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5981 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5982 | entry->mode = 0555; |
e692ab53 | 5983 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5984 | entry++; |
e692ab53 | 5985 | } |
7378547f MM |
5986 | |
5987 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5988 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5989 | } | |
6382bc90 | 5990 | |
7378547f | 5991 | /* may be called multiple times per register */ |
6382bc90 MM |
5992 | static void unregister_sched_domain_sysctl(void) |
5993 | { | |
7378547f MM |
5994 | if (sd_sysctl_header) |
5995 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5996 | sd_sysctl_header = NULL; |
7378547f MM |
5997 | if (sd_ctl_dir[0].child) |
5998 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5999 | } |
e692ab53 | 6000 | #else |
6382bc90 MM |
6001 | static void register_sched_domain_sysctl(void) |
6002 | { | |
6003 | } | |
6004 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6005 | { |
6006 | } | |
6007 | #endif | |
6008 | ||
1da177e4 LT |
6009 | /* |
6010 | * migration_call - callback that gets triggered when a CPU is added. | |
6011 | * Here we can start up the necessary migration thread for the new CPU. | |
6012 | */ | |
48f24c4d IM |
6013 | static int __cpuinit |
6014 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6015 | { |
1da177e4 | 6016 | struct task_struct *p; |
48f24c4d | 6017 | int cpu = (long)hcpu; |
1da177e4 | 6018 | unsigned long flags; |
70b97a7f | 6019 | struct rq *rq; |
1da177e4 LT |
6020 | |
6021 | switch (action) { | |
5be9361c | 6022 | |
1da177e4 | 6023 | case CPU_UP_PREPARE: |
8bb78442 | 6024 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6025 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6026 | if (IS_ERR(p)) |
6027 | return NOTIFY_BAD; | |
1da177e4 LT |
6028 | kthread_bind(p, cpu); |
6029 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6030 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6031 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6032 | task_rq_unlock(rq, &flags); |
6033 | cpu_rq(cpu)->migration_thread = p; | |
6034 | break; | |
48f24c4d | 6035 | |
1da177e4 | 6036 | case CPU_ONLINE: |
8bb78442 | 6037 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6038 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6039 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6040 | |
6041 | /* Update our root-domain */ | |
6042 | rq = cpu_rq(cpu); | |
6043 | spin_lock_irqsave(&rq->lock, flags); | |
6044 | if (rq->rd) { | |
6045 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
6046 | cpu_set(cpu, rq->rd->online); | |
6047 | } | |
6048 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6049 | break; |
48f24c4d | 6050 | |
1da177e4 LT |
6051 | #ifdef CONFIG_HOTPLUG_CPU |
6052 | case CPU_UP_CANCELED: | |
8bb78442 | 6053 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6054 | if (!cpu_rq(cpu)->migration_thread) |
6055 | break; | |
41a2d6cf | 6056 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6057 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6058 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6059 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6060 | cpu_rq(cpu)->migration_thread = NULL; | |
6061 | break; | |
48f24c4d | 6062 | |
1da177e4 | 6063 | case CPU_DEAD: |
8bb78442 | 6064 | case CPU_DEAD_FROZEN: |
470fd646 | 6065 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6066 | migrate_live_tasks(cpu); |
6067 | rq = cpu_rq(cpu); | |
6068 | kthread_stop(rq->migration_thread); | |
6069 | rq->migration_thread = NULL; | |
6070 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6071 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6072 | update_rq_clock(rq); |
2e1cb74a | 6073 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6074 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6075 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6076 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6077 | migrate_dead_tasks(cpu); |
d2da272a | 6078 | spin_unlock_irq(&rq->lock); |
470fd646 | 6079 | cpuset_unlock(); |
1da177e4 LT |
6080 | migrate_nr_uninterruptible(rq); |
6081 | BUG_ON(rq->nr_running != 0); | |
6082 | ||
41a2d6cf IM |
6083 | /* |
6084 | * No need to migrate the tasks: it was best-effort if | |
6085 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6086 | * the requestors. | |
6087 | */ | |
1da177e4 LT |
6088 | spin_lock_irq(&rq->lock); |
6089 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6090 | struct migration_req *req; |
6091 | ||
1da177e4 | 6092 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6093 | struct migration_req, list); |
1da177e4 LT |
6094 | list_del_init(&req->list); |
6095 | complete(&req->done); | |
6096 | } | |
6097 | spin_unlock_irq(&rq->lock); | |
6098 | break; | |
57d885fe | 6099 | |
08f503b0 GH |
6100 | case CPU_DYING: |
6101 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6102 | /* Update our root-domain */ |
6103 | rq = cpu_rq(cpu); | |
6104 | spin_lock_irqsave(&rq->lock, flags); | |
6105 | if (rq->rd) { | |
6106 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
6107 | cpu_clear(cpu, rq->rd->online); | |
6108 | } | |
6109 | spin_unlock_irqrestore(&rq->lock, flags); | |
6110 | break; | |
1da177e4 LT |
6111 | #endif |
6112 | } | |
6113 | return NOTIFY_OK; | |
6114 | } | |
6115 | ||
6116 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6117 | * happens before everything else. | |
6118 | */ | |
26c2143b | 6119 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6120 | .notifier_call = migration_call, |
6121 | .priority = 10 | |
6122 | }; | |
6123 | ||
e6fe6649 | 6124 | void __init migration_init(void) |
1da177e4 LT |
6125 | { |
6126 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6127 | int err; |
48f24c4d IM |
6128 | |
6129 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6130 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6131 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6132 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6133 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
6134 | } |
6135 | #endif | |
6136 | ||
6137 | #ifdef CONFIG_SMP | |
476f3534 | 6138 | |
3e9830dc | 6139 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6140 | |
7c16ec58 MT |
6141 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6142 | cpumask_t *groupmask) | |
1da177e4 | 6143 | { |
4dcf6aff | 6144 | struct sched_group *group = sd->groups; |
434d53b0 | 6145 | char str[256]; |
1da177e4 | 6146 | |
434d53b0 | 6147 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6148 | cpus_clear(*groupmask); |
4dcf6aff IM |
6149 | |
6150 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6151 | ||
6152 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6153 | printk("does not load-balance\n"); | |
6154 | if (sd->parent) | |
6155 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6156 | " has parent"); | |
6157 | return -1; | |
41c7ce9a NP |
6158 | } |
6159 | ||
4dcf6aff IM |
6160 | printk(KERN_CONT "span %s\n", str); |
6161 | ||
6162 | if (!cpu_isset(cpu, sd->span)) { | |
6163 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6164 | "CPU%d\n", cpu); | |
6165 | } | |
6166 | if (!cpu_isset(cpu, group->cpumask)) { | |
6167 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6168 | " CPU%d\n", cpu); | |
6169 | } | |
1da177e4 | 6170 | |
4dcf6aff | 6171 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6172 | do { |
4dcf6aff IM |
6173 | if (!group) { |
6174 | printk("\n"); | |
6175 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6176 | break; |
6177 | } | |
6178 | ||
4dcf6aff IM |
6179 | if (!group->__cpu_power) { |
6180 | printk(KERN_CONT "\n"); | |
6181 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6182 | "set\n"); | |
6183 | break; | |
6184 | } | |
1da177e4 | 6185 | |
4dcf6aff IM |
6186 | if (!cpus_weight(group->cpumask)) { |
6187 | printk(KERN_CONT "\n"); | |
6188 | printk(KERN_ERR "ERROR: empty group\n"); | |
6189 | break; | |
6190 | } | |
1da177e4 | 6191 | |
7c16ec58 | 6192 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6193 | printk(KERN_CONT "\n"); |
6194 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6195 | break; | |
6196 | } | |
1da177e4 | 6197 | |
7c16ec58 | 6198 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6199 | |
434d53b0 | 6200 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6201 | printk(KERN_CONT " %s", str); |
1da177e4 | 6202 | |
4dcf6aff IM |
6203 | group = group->next; |
6204 | } while (group != sd->groups); | |
6205 | printk(KERN_CONT "\n"); | |
1da177e4 | 6206 | |
7c16ec58 | 6207 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6208 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6209 | |
7c16ec58 | 6210 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6211 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6212 | "of domain->span\n"); | |
6213 | return 0; | |
6214 | } | |
1da177e4 | 6215 | |
4dcf6aff IM |
6216 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6217 | { | |
7c16ec58 | 6218 | cpumask_t *groupmask; |
4dcf6aff | 6219 | int level = 0; |
1da177e4 | 6220 | |
4dcf6aff IM |
6221 | if (!sd) { |
6222 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6223 | return; | |
6224 | } | |
1da177e4 | 6225 | |
4dcf6aff IM |
6226 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6227 | ||
7c16ec58 MT |
6228 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6229 | if (!groupmask) { | |
6230 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6231 | return; | |
6232 | } | |
6233 | ||
4dcf6aff | 6234 | for (;;) { |
7c16ec58 | 6235 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6236 | break; |
1da177e4 LT |
6237 | level++; |
6238 | sd = sd->parent; | |
33859f7f | 6239 | if (!sd) |
4dcf6aff IM |
6240 | break; |
6241 | } | |
7c16ec58 | 6242 | kfree(groupmask); |
1da177e4 LT |
6243 | } |
6244 | #else | |
48f24c4d | 6245 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
6246 | #endif |
6247 | ||
1a20ff27 | 6248 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6249 | { |
6250 | if (cpus_weight(sd->span) == 1) | |
6251 | return 1; | |
6252 | ||
6253 | /* Following flags need at least 2 groups */ | |
6254 | if (sd->flags & (SD_LOAD_BALANCE | | |
6255 | SD_BALANCE_NEWIDLE | | |
6256 | SD_BALANCE_FORK | | |
89c4710e SS |
6257 | SD_BALANCE_EXEC | |
6258 | SD_SHARE_CPUPOWER | | |
6259 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6260 | if (sd->groups != sd->groups->next) |
6261 | return 0; | |
6262 | } | |
6263 | ||
6264 | /* Following flags don't use groups */ | |
6265 | if (sd->flags & (SD_WAKE_IDLE | | |
6266 | SD_WAKE_AFFINE | | |
6267 | SD_WAKE_BALANCE)) | |
6268 | return 0; | |
6269 | ||
6270 | return 1; | |
6271 | } | |
6272 | ||
48f24c4d IM |
6273 | static int |
6274 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6275 | { |
6276 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6277 | ||
6278 | if (sd_degenerate(parent)) | |
6279 | return 1; | |
6280 | ||
6281 | if (!cpus_equal(sd->span, parent->span)) | |
6282 | return 0; | |
6283 | ||
6284 | /* Does parent contain flags not in child? */ | |
6285 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6286 | if (cflags & SD_WAKE_AFFINE) | |
6287 | pflags &= ~SD_WAKE_BALANCE; | |
6288 | /* Flags needing groups don't count if only 1 group in parent */ | |
6289 | if (parent->groups == parent->groups->next) { | |
6290 | pflags &= ~(SD_LOAD_BALANCE | | |
6291 | SD_BALANCE_NEWIDLE | | |
6292 | SD_BALANCE_FORK | | |
89c4710e SS |
6293 | SD_BALANCE_EXEC | |
6294 | SD_SHARE_CPUPOWER | | |
6295 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6296 | } |
6297 | if (~cflags & pflags) | |
6298 | return 0; | |
6299 | ||
6300 | return 1; | |
6301 | } | |
6302 | ||
57d885fe GH |
6303 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6304 | { | |
6305 | unsigned long flags; | |
6306 | const struct sched_class *class; | |
6307 | ||
6308 | spin_lock_irqsave(&rq->lock, flags); | |
6309 | ||
6310 | if (rq->rd) { | |
6311 | struct root_domain *old_rd = rq->rd; | |
6312 | ||
0eab9146 | 6313 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6314 | if (class->leave_domain) |
6315 | class->leave_domain(rq); | |
0eab9146 | 6316 | } |
57d885fe | 6317 | |
dc938520 GH |
6318 | cpu_clear(rq->cpu, old_rd->span); |
6319 | cpu_clear(rq->cpu, old_rd->online); | |
6320 | ||
57d885fe GH |
6321 | if (atomic_dec_and_test(&old_rd->refcount)) |
6322 | kfree(old_rd); | |
6323 | } | |
6324 | ||
6325 | atomic_inc(&rd->refcount); | |
6326 | rq->rd = rd; | |
6327 | ||
dc938520 | 6328 | cpu_set(rq->cpu, rd->span); |
1f94ef59 GH |
6329 | if (cpu_isset(rq->cpu, cpu_online_map)) |
6330 | cpu_set(rq->cpu, rd->online); | |
dc938520 | 6331 | |
0eab9146 | 6332 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6333 | if (class->join_domain) |
6334 | class->join_domain(rq); | |
0eab9146 | 6335 | } |
57d885fe GH |
6336 | |
6337 | spin_unlock_irqrestore(&rq->lock, flags); | |
6338 | } | |
6339 | ||
dc938520 | 6340 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6341 | { |
6342 | memset(rd, 0, sizeof(*rd)); | |
6343 | ||
dc938520 GH |
6344 | cpus_clear(rd->span); |
6345 | cpus_clear(rd->online); | |
57d885fe GH |
6346 | } |
6347 | ||
6348 | static void init_defrootdomain(void) | |
6349 | { | |
dc938520 | 6350 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6351 | atomic_set(&def_root_domain.refcount, 1); |
6352 | } | |
6353 | ||
dc938520 | 6354 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6355 | { |
6356 | struct root_domain *rd; | |
6357 | ||
6358 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6359 | if (!rd) | |
6360 | return NULL; | |
6361 | ||
dc938520 | 6362 | init_rootdomain(rd); |
57d885fe GH |
6363 | |
6364 | return rd; | |
6365 | } | |
6366 | ||
1da177e4 | 6367 | /* |
0eab9146 | 6368 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6369 | * hold the hotplug lock. |
6370 | */ | |
0eab9146 IM |
6371 | static void |
6372 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6373 | { |
70b97a7f | 6374 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6375 | struct sched_domain *tmp; |
6376 | ||
6377 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6378 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6379 | struct sched_domain *parent = tmp->parent; | |
6380 | if (!parent) | |
6381 | break; | |
1a848870 | 6382 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6383 | tmp->parent = parent->parent; |
1a848870 SS |
6384 | if (parent->parent) |
6385 | parent->parent->child = tmp; | |
6386 | } | |
245af2c7 SS |
6387 | } |
6388 | ||
1a848870 | 6389 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6390 | sd = sd->parent; |
1a848870 SS |
6391 | if (sd) |
6392 | sd->child = NULL; | |
6393 | } | |
1da177e4 LT |
6394 | |
6395 | sched_domain_debug(sd, cpu); | |
6396 | ||
57d885fe | 6397 | rq_attach_root(rq, rd); |
674311d5 | 6398 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6399 | } |
6400 | ||
6401 | /* cpus with isolated domains */ | |
67af63a6 | 6402 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6403 | |
6404 | /* Setup the mask of cpus configured for isolated domains */ | |
6405 | static int __init isolated_cpu_setup(char *str) | |
6406 | { | |
6407 | int ints[NR_CPUS], i; | |
6408 | ||
6409 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6410 | cpus_clear(cpu_isolated_map); | |
6411 | for (i = 1; i <= ints[0]; i++) | |
6412 | if (ints[i] < NR_CPUS) | |
6413 | cpu_set(ints[i], cpu_isolated_map); | |
6414 | return 1; | |
6415 | } | |
6416 | ||
8927f494 | 6417 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6418 | |
6419 | /* | |
6711cab4 SS |
6420 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6421 | * to a function which identifies what group(along with sched group) a CPU | |
6422 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6423 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6424 | * |
6425 | * init_sched_build_groups will build a circular linked list of the groups | |
6426 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6427 | * and ->cpu_power to 0. | |
6428 | */ | |
a616058b | 6429 | static void |
7c16ec58 | 6430 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6431 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6432 | struct sched_group **sg, |
6433 | cpumask_t *tmpmask), | |
6434 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6435 | { |
6436 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6437 | int i; |
6438 | ||
7c16ec58 MT |
6439 | cpus_clear(*covered); |
6440 | ||
6441 | for_each_cpu_mask(i, *span) { | |
6711cab4 | 6442 | struct sched_group *sg; |
7c16ec58 | 6443 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6444 | int j; |
6445 | ||
7c16ec58 | 6446 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6447 | continue; |
6448 | ||
7c16ec58 | 6449 | cpus_clear(sg->cpumask); |
5517d86b | 6450 | sg->__cpu_power = 0; |
1da177e4 | 6451 | |
7c16ec58 MT |
6452 | for_each_cpu_mask(j, *span) { |
6453 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) | |
1da177e4 LT |
6454 | continue; |
6455 | ||
7c16ec58 | 6456 | cpu_set(j, *covered); |
1da177e4 LT |
6457 | cpu_set(j, sg->cpumask); |
6458 | } | |
6459 | if (!first) | |
6460 | first = sg; | |
6461 | if (last) | |
6462 | last->next = sg; | |
6463 | last = sg; | |
6464 | } | |
6465 | last->next = first; | |
6466 | } | |
6467 | ||
9c1cfda2 | 6468 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6469 | |
9c1cfda2 | 6470 | #ifdef CONFIG_NUMA |
198e2f18 | 6471 | |
9c1cfda2 JH |
6472 | /** |
6473 | * find_next_best_node - find the next node to include in a sched_domain | |
6474 | * @node: node whose sched_domain we're building | |
6475 | * @used_nodes: nodes already in the sched_domain | |
6476 | * | |
41a2d6cf | 6477 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6478 | * finds the closest node not already in the @used_nodes map. |
6479 | * | |
6480 | * Should use nodemask_t. | |
6481 | */ | |
c5f59f08 | 6482 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6483 | { |
6484 | int i, n, val, min_val, best_node = 0; | |
6485 | ||
6486 | min_val = INT_MAX; | |
6487 | ||
6488 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6489 | /* Start at @node */ | |
6490 | n = (node + i) % MAX_NUMNODES; | |
6491 | ||
6492 | if (!nr_cpus_node(n)) | |
6493 | continue; | |
6494 | ||
6495 | /* Skip already used nodes */ | |
c5f59f08 | 6496 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6497 | continue; |
6498 | ||
6499 | /* Simple min distance search */ | |
6500 | val = node_distance(node, n); | |
6501 | ||
6502 | if (val < min_val) { | |
6503 | min_val = val; | |
6504 | best_node = n; | |
6505 | } | |
6506 | } | |
6507 | ||
c5f59f08 | 6508 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6509 | return best_node; |
6510 | } | |
6511 | ||
6512 | /** | |
6513 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6514 | * @node: node whose cpumask we're constructing | |
9c1cfda2 | 6515 | * |
41a2d6cf | 6516 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6517 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6518 | * out optimally. | |
6519 | */ | |
4bdbaad3 | 6520 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 6521 | { |
c5f59f08 | 6522 | nodemask_t used_nodes; |
c5f59f08 | 6523 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 6524 | int i; |
9c1cfda2 | 6525 | |
4bdbaad3 | 6526 | cpus_clear(*span); |
c5f59f08 | 6527 | nodes_clear(used_nodes); |
9c1cfda2 | 6528 | |
4bdbaad3 | 6529 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 6530 | node_set(node, used_nodes); |
9c1cfda2 JH |
6531 | |
6532 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6533 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6534 | |
c5f59f08 | 6535 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 6536 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 6537 | } |
9c1cfda2 JH |
6538 | } |
6539 | #endif | |
6540 | ||
5c45bf27 | 6541 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6542 | |
9c1cfda2 | 6543 | /* |
48f24c4d | 6544 | * SMT sched-domains: |
9c1cfda2 | 6545 | */ |
1da177e4 LT |
6546 | #ifdef CONFIG_SCHED_SMT |
6547 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6548 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6549 | |
41a2d6cf | 6550 | static int |
7c16ec58 MT |
6551 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6552 | cpumask_t *unused) | |
1da177e4 | 6553 | { |
6711cab4 SS |
6554 | if (sg) |
6555 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6556 | return cpu; |
6557 | } | |
6558 | #endif | |
6559 | ||
48f24c4d IM |
6560 | /* |
6561 | * multi-core sched-domains: | |
6562 | */ | |
1e9f28fa SS |
6563 | #ifdef CONFIG_SCHED_MC |
6564 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6565 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
6566 | #endif |
6567 | ||
6568 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6569 | static int |
7c16ec58 MT |
6570 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6571 | cpumask_t *mask) | |
1e9f28fa | 6572 | { |
6711cab4 | 6573 | int group; |
7c16ec58 MT |
6574 | |
6575 | *mask = per_cpu(cpu_sibling_map, cpu); | |
6576 | cpus_and(*mask, *mask, *cpu_map); | |
6577 | group = first_cpu(*mask); | |
6711cab4 SS |
6578 | if (sg) |
6579 | *sg = &per_cpu(sched_group_core, group); | |
6580 | return group; | |
1e9f28fa SS |
6581 | } |
6582 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6583 | static int |
7c16ec58 MT |
6584 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6585 | cpumask_t *unused) | |
1e9f28fa | 6586 | { |
6711cab4 SS |
6587 | if (sg) |
6588 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6589 | return cpu; |
6590 | } | |
6591 | #endif | |
6592 | ||
1da177e4 | 6593 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6594 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6595 | |
41a2d6cf | 6596 | static int |
7c16ec58 MT |
6597 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6598 | cpumask_t *mask) | |
1da177e4 | 6599 | { |
6711cab4 | 6600 | int group; |
48f24c4d | 6601 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
6602 | *mask = cpu_coregroup_map(cpu); |
6603 | cpus_and(*mask, *mask, *cpu_map); | |
6604 | group = first_cpu(*mask); | |
1e9f28fa | 6605 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
6606 | *mask = per_cpu(cpu_sibling_map, cpu); |
6607 | cpus_and(*mask, *mask, *cpu_map); | |
6608 | group = first_cpu(*mask); | |
1da177e4 | 6609 | #else |
6711cab4 | 6610 | group = cpu; |
1da177e4 | 6611 | #endif |
6711cab4 SS |
6612 | if (sg) |
6613 | *sg = &per_cpu(sched_group_phys, group); | |
6614 | return group; | |
1da177e4 LT |
6615 | } |
6616 | ||
6617 | #ifdef CONFIG_NUMA | |
1da177e4 | 6618 | /* |
9c1cfda2 JH |
6619 | * The init_sched_build_groups can't handle what we want to do with node |
6620 | * groups, so roll our own. Now each node has its own list of groups which | |
6621 | * gets dynamically allocated. | |
1da177e4 | 6622 | */ |
9c1cfda2 | 6623 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 6624 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6625 | |
9c1cfda2 | 6626 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6627 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6628 | |
6711cab4 | 6629 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 6630 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 6631 | { |
6711cab4 SS |
6632 | int group; |
6633 | ||
7c16ec58 MT |
6634 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6635 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6636 | group = first_cpu(*nodemask); | |
6711cab4 SS |
6637 | |
6638 | if (sg) | |
6639 | *sg = &per_cpu(sched_group_allnodes, group); | |
6640 | return group; | |
1da177e4 | 6641 | } |
6711cab4 | 6642 | |
08069033 SS |
6643 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6644 | { | |
6645 | struct sched_group *sg = group_head; | |
6646 | int j; | |
6647 | ||
6648 | if (!sg) | |
6649 | return; | |
3a5c359a AK |
6650 | do { |
6651 | for_each_cpu_mask(j, sg->cpumask) { | |
6652 | struct sched_domain *sd; | |
08069033 | 6653 | |
3a5c359a AK |
6654 | sd = &per_cpu(phys_domains, j); |
6655 | if (j != first_cpu(sd->groups->cpumask)) { | |
6656 | /* | |
6657 | * Only add "power" once for each | |
6658 | * physical package. | |
6659 | */ | |
6660 | continue; | |
6661 | } | |
08069033 | 6662 | |
3a5c359a AK |
6663 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6664 | } | |
6665 | sg = sg->next; | |
6666 | } while (sg != group_head); | |
08069033 | 6667 | } |
1da177e4 LT |
6668 | #endif |
6669 | ||
a616058b | 6670 | #ifdef CONFIG_NUMA |
51888ca2 | 6671 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 6672 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 6673 | { |
a616058b | 6674 | int cpu, i; |
51888ca2 SV |
6675 | |
6676 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6677 | struct sched_group **sched_group_nodes |
6678 | = sched_group_nodes_bycpu[cpu]; | |
6679 | ||
51888ca2 SV |
6680 | if (!sched_group_nodes) |
6681 | continue; | |
6682 | ||
6683 | for (i = 0; i < MAX_NUMNODES; i++) { | |
51888ca2 SV |
6684 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6685 | ||
7c16ec58 MT |
6686 | *nodemask = node_to_cpumask(i); |
6687 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6688 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
6689 | continue; |
6690 | ||
6691 | if (sg == NULL) | |
6692 | continue; | |
6693 | sg = sg->next; | |
6694 | next_sg: | |
6695 | oldsg = sg; | |
6696 | sg = sg->next; | |
6697 | kfree(oldsg); | |
6698 | if (oldsg != sched_group_nodes[i]) | |
6699 | goto next_sg; | |
6700 | } | |
6701 | kfree(sched_group_nodes); | |
6702 | sched_group_nodes_bycpu[cpu] = NULL; | |
6703 | } | |
51888ca2 | 6704 | } |
a616058b | 6705 | #else |
7c16ec58 | 6706 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
6707 | { |
6708 | } | |
6709 | #endif | |
51888ca2 | 6710 | |
89c4710e SS |
6711 | /* |
6712 | * Initialize sched groups cpu_power. | |
6713 | * | |
6714 | * cpu_power indicates the capacity of sched group, which is used while | |
6715 | * distributing the load between different sched groups in a sched domain. | |
6716 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6717 | * there are asymmetries in the topology. If there are asymmetries, group | |
6718 | * having more cpu_power will pickup more load compared to the group having | |
6719 | * less cpu_power. | |
6720 | * | |
6721 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6722 | * the maximum number of tasks a group can handle in the presence of other idle | |
6723 | * or lightly loaded groups in the same sched domain. | |
6724 | */ | |
6725 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6726 | { | |
6727 | struct sched_domain *child; | |
6728 | struct sched_group *group; | |
6729 | ||
6730 | WARN_ON(!sd || !sd->groups); | |
6731 | ||
6732 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6733 | return; | |
6734 | ||
6735 | child = sd->child; | |
6736 | ||
5517d86b ED |
6737 | sd->groups->__cpu_power = 0; |
6738 | ||
89c4710e SS |
6739 | /* |
6740 | * For perf policy, if the groups in child domain share resources | |
6741 | * (for example cores sharing some portions of the cache hierarchy | |
6742 | * or SMT), then set this domain groups cpu_power such that each group | |
6743 | * can handle only one task, when there are other idle groups in the | |
6744 | * same sched domain. | |
6745 | */ | |
6746 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6747 | (child->flags & | |
6748 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6749 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6750 | return; |
6751 | } | |
6752 | ||
89c4710e SS |
6753 | /* |
6754 | * add cpu_power of each child group to this groups cpu_power | |
6755 | */ | |
6756 | group = child->groups; | |
6757 | do { | |
5517d86b | 6758 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6759 | group = group->next; |
6760 | } while (group != child->groups); | |
6761 | } | |
6762 | ||
7c16ec58 MT |
6763 | /* |
6764 | * Initializers for schedule domains | |
6765 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6766 | */ | |
6767 | ||
6768 | #define SD_INIT(sd, type) sd_init_##type(sd) | |
6769 | #define SD_INIT_FUNC(type) \ | |
6770 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6771 | { \ | |
6772 | memset(sd, 0, sizeof(*sd)); \ | |
6773 | *sd = SD_##type##_INIT; \ | |
6774 | } | |
6775 | ||
6776 | SD_INIT_FUNC(CPU) | |
6777 | #ifdef CONFIG_NUMA | |
6778 | SD_INIT_FUNC(ALLNODES) | |
6779 | SD_INIT_FUNC(NODE) | |
6780 | #endif | |
6781 | #ifdef CONFIG_SCHED_SMT | |
6782 | SD_INIT_FUNC(SIBLING) | |
6783 | #endif | |
6784 | #ifdef CONFIG_SCHED_MC | |
6785 | SD_INIT_FUNC(MC) | |
6786 | #endif | |
6787 | ||
6788 | /* | |
6789 | * To minimize stack usage kmalloc room for cpumasks and share the | |
6790 | * space as the usage in build_sched_domains() dictates. Used only | |
6791 | * if the amount of space is significant. | |
6792 | */ | |
6793 | struct allmasks { | |
6794 | cpumask_t tmpmask; /* make this one first */ | |
6795 | union { | |
6796 | cpumask_t nodemask; | |
6797 | cpumask_t this_sibling_map; | |
6798 | cpumask_t this_core_map; | |
6799 | }; | |
6800 | cpumask_t send_covered; | |
6801 | ||
6802 | #ifdef CONFIG_NUMA | |
6803 | cpumask_t domainspan; | |
6804 | cpumask_t covered; | |
6805 | cpumask_t notcovered; | |
6806 | #endif | |
6807 | }; | |
6808 | ||
6809 | #if NR_CPUS > 128 | |
6810 | #define SCHED_CPUMASK_ALLOC 1 | |
6811 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
6812 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
6813 | #else | |
6814 | #define SCHED_CPUMASK_ALLOC 0 | |
6815 | #define SCHED_CPUMASK_FREE(v) | |
6816 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
6817 | #endif | |
6818 | ||
6819 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
6820 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
6821 | ||
1da177e4 | 6822 | /* |
1a20ff27 DG |
6823 | * Build sched domains for a given set of cpus and attach the sched domains |
6824 | * to the individual cpus | |
1da177e4 | 6825 | */ |
51888ca2 | 6826 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6827 | { |
6828 | int i; | |
57d885fe | 6829 | struct root_domain *rd; |
7c16ec58 MT |
6830 | SCHED_CPUMASK_DECLARE(allmasks); |
6831 | cpumask_t *tmpmask; | |
d1b55138 JH |
6832 | #ifdef CONFIG_NUMA |
6833 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6834 | int sd_allnodes = 0; |
d1b55138 JH |
6835 | |
6836 | /* | |
6837 | * Allocate the per-node list of sched groups | |
6838 | */ | |
5cf9f062 | 6839 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 6840 | GFP_KERNEL); |
d1b55138 JH |
6841 | if (!sched_group_nodes) { |
6842 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6843 | return -ENOMEM; |
d1b55138 | 6844 | } |
d1b55138 | 6845 | #endif |
1da177e4 | 6846 | |
dc938520 | 6847 | rd = alloc_rootdomain(); |
57d885fe GH |
6848 | if (!rd) { |
6849 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
6850 | #ifdef CONFIG_NUMA |
6851 | kfree(sched_group_nodes); | |
6852 | #endif | |
57d885fe GH |
6853 | return -ENOMEM; |
6854 | } | |
6855 | ||
7c16ec58 MT |
6856 | #if SCHED_CPUMASK_ALLOC |
6857 | /* get space for all scratch cpumask variables */ | |
6858 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
6859 | if (!allmasks) { | |
6860 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
6861 | kfree(rd); | |
6862 | #ifdef CONFIG_NUMA | |
6863 | kfree(sched_group_nodes); | |
6864 | #endif | |
6865 | return -ENOMEM; | |
6866 | } | |
6867 | #endif | |
6868 | tmpmask = (cpumask_t *)allmasks; | |
6869 | ||
6870 | ||
6871 | #ifdef CONFIG_NUMA | |
6872 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6873 | #endif | |
6874 | ||
1da177e4 | 6875 | /* |
1a20ff27 | 6876 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6877 | */ |
1a20ff27 | 6878 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6879 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 6880 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 6881 | |
7c16ec58 MT |
6882 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
6883 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
6884 | |
6885 | #ifdef CONFIG_NUMA | |
dd41f596 | 6886 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 6887 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 6888 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 6889 | SD_INIT(sd, ALLNODES); |
9c1cfda2 | 6890 | sd->span = *cpu_map; |
7c16ec58 | 6891 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 6892 | p = sd; |
6711cab4 | 6893 | sd_allnodes = 1; |
9c1cfda2 JH |
6894 | } else |
6895 | p = NULL; | |
6896 | ||
1da177e4 | 6897 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 6898 | SD_INIT(sd, NODE); |
4bdbaad3 | 6899 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 6900 | sd->parent = p; |
1a848870 SS |
6901 | if (p) |
6902 | p->child = sd; | |
9c1cfda2 | 6903 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6904 | #endif |
6905 | ||
6906 | p = sd; | |
6907 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 MT |
6908 | SD_INIT(sd, CPU); |
6909 | sd->span = *nodemask; | |
1da177e4 | 6910 | sd->parent = p; |
1a848870 SS |
6911 | if (p) |
6912 | p->child = sd; | |
7c16ec58 | 6913 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 6914 | |
1e9f28fa SS |
6915 | #ifdef CONFIG_SCHED_MC |
6916 | p = sd; | |
6917 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 6918 | SD_INIT(sd, MC); |
1e9f28fa SS |
6919 | sd->span = cpu_coregroup_map(i); |
6920 | cpus_and(sd->span, sd->span, *cpu_map); | |
6921 | sd->parent = p; | |
1a848870 | 6922 | p->child = sd; |
7c16ec58 | 6923 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
6924 | #endif |
6925 | ||
1da177e4 LT |
6926 | #ifdef CONFIG_SCHED_SMT |
6927 | p = sd; | |
6928 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 6929 | SD_INIT(sd, SIBLING); |
d5a7430d | 6930 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6931 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6932 | sd->parent = p; |
1a848870 | 6933 | p->child = sd; |
7c16ec58 | 6934 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
6935 | #endif |
6936 | } | |
6937 | ||
6938 | #ifdef CONFIG_SCHED_SMT | |
6939 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6940 | for_each_cpu_mask(i, *cpu_map) { |
7c16ec58 MT |
6941 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
6942 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
6943 | ||
6944 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
6945 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
6946 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
6947 | continue; |
6948 | ||
dd41f596 | 6949 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
6950 | &cpu_to_cpu_group, |
6951 | send_covered, tmpmask); | |
1da177e4 LT |
6952 | } |
6953 | #endif | |
6954 | ||
1e9f28fa SS |
6955 | #ifdef CONFIG_SCHED_MC |
6956 | /* Set up multi-core groups */ | |
6957 | for_each_cpu_mask(i, *cpu_map) { | |
7c16ec58 MT |
6958 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
6959 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
6960 | ||
6961 | *this_core_map = cpu_coregroup_map(i); | |
6962 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
6963 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 6964 | continue; |
7c16ec58 | 6965 | |
dd41f596 | 6966 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
6967 | &cpu_to_core_group, |
6968 | send_covered, tmpmask); | |
1e9f28fa SS |
6969 | } |
6970 | #endif | |
6971 | ||
1da177e4 LT |
6972 | /* Set up physical groups */ |
6973 | for (i = 0; i < MAX_NUMNODES; i++) { | |
7c16ec58 MT |
6974 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
6975 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 6976 | |
7c16ec58 MT |
6977 | *nodemask = node_to_cpumask(i); |
6978 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6979 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
6980 | continue; |
6981 | ||
7c16ec58 MT |
6982 | init_sched_build_groups(nodemask, cpu_map, |
6983 | &cpu_to_phys_group, | |
6984 | send_covered, tmpmask); | |
1da177e4 LT |
6985 | } |
6986 | ||
6987 | #ifdef CONFIG_NUMA | |
6988 | /* Set up node groups */ | |
7c16ec58 MT |
6989 | if (sd_allnodes) { |
6990 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
6991 | ||
6992 | init_sched_build_groups(cpu_map, cpu_map, | |
6993 | &cpu_to_allnodes_group, | |
6994 | send_covered, tmpmask); | |
6995 | } | |
9c1cfda2 JH |
6996 | |
6997 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6998 | /* Set up node groups */ | |
6999 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7000 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7001 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7002 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7003 | int j; |
7004 | ||
7c16ec58 MT |
7005 | *nodemask = node_to_cpumask(i); |
7006 | cpus_clear(*covered); | |
7007 | ||
7008 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7009 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7010 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7011 | continue; |
d1b55138 | 7012 | } |
9c1cfda2 | 7013 | |
4bdbaad3 | 7014 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7015 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7016 | |
15f0b676 | 7017 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7018 | if (!sg) { |
7019 | printk(KERN_WARNING "Can not alloc domain group for " | |
7020 | "node %d\n", i); | |
7021 | goto error; | |
7022 | } | |
9c1cfda2 | 7023 | sched_group_nodes[i] = sg; |
7c16ec58 | 7024 | for_each_cpu_mask(j, *nodemask) { |
9c1cfda2 | 7025 | struct sched_domain *sd; |
9761eea8 | 7026 | |
9c1cfda2 JH |
7027 | sd = &per_cpu(node_domains, j); |
7028 | sd->groups = sg; | |
9c1cfda2 | 7029 | } |
5517d86b | 7030 | sg->__cpu_power = 0; |
7c16ec58 | 7031 | sg->cpumask = *nodemask; |
51888ca2 | 7032 | sg->next = sg; |
7c16ec58 | 7033 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7034 | prev = sg; |
7035 | ||
7036 | for (j = 0; j < MAX_NUMNODES; j++) { | |
7c16ec58 | 7037 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
9c1cfda2 | 7038 | int n = (i + j) % MAX_NUMNODES; |
c5f59f08 | 7039 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7040 | |
7c16ec58 MT |
7041 | cpus_complement(*notcovered, *covered); |
7042 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7043 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7044 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7045 | break; |
7046 | ||
7c16ec58 MT |
7047 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7048 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7049 | continue; |
7050 | ||
15f0b676 SV |
7051 | sg = kmalloc_node(sizeof(struct sched_group), |
7052 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7053 | if (!sg) { |
7054 | printk(KERN_WARNING | |
7055 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7056 | goto error; |
9c1cfda2 | 7057 | } |
5517d86b | 7058 | sg->__cpu_power = 0; |
7c16ec58 | 7059 | sg->cpumask = *tmpmask; |
51888ca2 | 7060 | sg->next = prev->next; |
7c16ec58 | 7061 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7062 | prev->next = sg; |
7063 | prev = sg; | |
7064 | } | |
9c1cfda2 | 7065 | } |
1da177e4 LT |
7066 | #endif |
7067 | ||
7068 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7069 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 7070 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7071 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7072 | ||
89c4710e | 7073 | init_sched_groups_power(i, sd); |
5c45bf27 | 7074 | } |
1da177e4 | 7075 | #endif |
1e9f28fa | 7076 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 7077 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7078 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7079 | ||
89c4710e | 7080 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7081 | } |
7082 | #endif | |
1e9f28fa | 7083 | |
5c45bf27 | 7084 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7085 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7086 | ||
89c4710e | 7087 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7088 | } |
7089 | ||
9c1cfda2 | 7090 | #ifdef CONFIG_NUMA |
08069033 SS |
7091 | for (i = 0; i < MAX_NUMNODES; i++) |
7092 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 7093 | |
6711cab4 SS |
7094 | if (sd_allnodes) { |
7095 | struct sched_group *sg; | |
f712c0c7 | 7096 | |
7c16ec58 MT |
7097 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7098 | tmpmask); | |
f712c0c7 SS |
7099 | init_numa_sched_groups_power(sg); |
7100 | } | |
9c1cfda2 JH |
7101 | #endif |
7102 | ||
1da177e4 | 7103 | /* Attach the domains */ |
1a20ff27 | 7104 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
7105 | struct sched_domain *sd; |
7106 | #ifdef CONFIG_SCHED_SMT | |
7107 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7108 | #elif defined(CONFIG_SCHED_MC) |
7109 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7110 | #else |
7111 | sd = &per_cpu(phys_domains, i); | |
7112 | #endif | |
57d885fe | 7113 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7114 | } |
51888ca2 | 7115 | |
7c16ec58 | 7116 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7117 | return 0; |
7118 | ||
a616058b | 7119 | #ifdef CONFIG_NUMA |
51888ca2 | 7120 | error: |
7c16ec58 MT |
7121 | free_sched_groups(cpu_map, tmpmask); |
7122 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7123 | return -ENOMEM; |
a616058b | 7124 | #endif |
1da177e4 | 7125 | } |
029190c5 PJ |
7126 | |
7127 | static cpumask_t *doms_cur; /* current sched domains */ | |
7128 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
7129 | ||
7130 | /* | |
7131 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7132 | * cpumask_t) fails, then fallback to a single sched domain, | |
7133 | * as determined by the single cpumask_t fallback_doms. | |
7134 | */ | |
7135 | static cpumask_t fallback_doms; | |
7136 | ||
22e52b07 HC |
7137 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7138 | { | |
7139 | } | |
7140 | ||
1a20ff27 | 7141 | /* |
41a2d6cf | 7142 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7143 | * For now this just excludes isolated cpus, but could be used to |
7144 | * exclude other special cases in the future. | |
1a20ff27 | 7145 | */ |
51888ca2 | 7146 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7147 | { |
7378547f MM |
7148 | int err; |
7149 | ||
22e52b07 | 7150 | arch_update_cpu_topology(); |
029190c5 PJ |
7151 | ndoms_cur = 1; |
7152 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7153 | if (!doms_cur) | |
7154 | doms_cur = &fallback_doms; | |
7155 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
7378547f | 7156 | err = build_sched_domains(doms_cur); |
6382bc90 | 7157 | register_sched_domain_sysctl(); |
7378547f MM |
7158 | |
7159 | return err; | |
1a20ff27 DG |
7160 | } |
7161 | ||
7c16ec58 MT |
7162 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7163 | cpumask_t *tmpmask) | |
1da177e4 | 7164 | { |
7c16ec58 | 7165 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7166 | } |
1da177e4 | 7167 | |
1a20ff27 DG |
7168 | /* |
7169 | * Detach sched domains from a group of cpus specified in cpu_map | |
7170 | * These cpus will now be attached to the NULL domain | |
7171 | */ | |
858119e1 | 7172 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7173 | { |
7c16ec58 | 7174 | cpumask_t tmpmask; |
1a20ff27 DG |
7175 | int i; |
7176 | ||
6382bc90 MM |
7177 | unregister_sched_domain_sysctl(); |
7178 | ||
1a20ff27 | 7179 | for_each_cpu_mask(i, *cpu_map) |
57d885fe | 7180 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7181 | synchronize_sched(); |
7c16ec58 | 7182 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7183 | } |
7184 | ||
029190c5 PJ |
7185 | /* |
7186 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7187 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7188 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7189 | * It destroys each deleted domain and builds each new domain. | |
7190 | * | |
7191 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7192 | * The masks don't intersect (don't overlap.) We should setup one |
7193 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7194 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7195 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7196 | * it as it is. | |
7197 | * | |
41a2d6cf IM |
7198 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7199 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7200 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7201 | * and partition_sched_domains() will fallback to the single partition | |
7202 | * 'fallback_doms'. | |
7203 | * | |
7204 | * Call with hotplug lock held | |
7205 | */ | |
7206 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
7207 | { | |
7208 | int i, j; | |
7209 | ||
a1835615 SV |
7210 | lock_doms_cur(); |
7211 | ||
7378547f MM |
7212 | /* always unregister in case we don't destroy any domains */ |
7213 | unregister_sched_domain_sysctl(); | |
7214 | ||
029190c5 PJ |
7215 | if (doms_new == NULL) { |
7216 | ndoms_new = 1; | |
7217 | doms_new = &fallback_doms; | |
7218 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
7219 | } | |
7220 | ||
7221 | /* Destroy deleted domains */ | |
7222 | for (i = 0; i < ndoms_cur; i++) { | |
7223 | for (j = 0; j < ndoms_new; j++) { | |
7224 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
7225 | goto match1; | |
7226 | } | |
7227 | /* no match - a current sched domain not in new doms_new[] */ | |
7228 | detach_destroy_domains(doms_cur + i); | |
7229 | match1: | |
7230 | ; | |
7231 | } | |
7232 | ||
7233 | /* Build new domains */ | |
7234 | for (i = 0; i < ndoms_new; i++) { | |
7235 | for (j = 0; j < ndoms_cur; j++) { | |
7236 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
7237 | goto match2; | |
7238 | } | |
7239 | /* no match - add a new doms_new */ | |
7240 | build_sched_domains(doms_new + i); | |
7241 | match2: | |
7242 | ; | |
7243 | } | |
7244 | ||
7245 | /* Remember the new sched domains */ | |
7246 | if (doms_cur != &fallback_doms) | |
7247 | kfree(doms_cur); | |
7248 | doms_cur = doms_new; | |
7249 | ndoms_cur = ndoms_new; | |
7378547f MM |
7250 | |
7251 | register_sched_domain_sysctl(); | |
a1835615 SV |
7252 | |
7253 | unlock_doms_cur(); | |
029190c5 PJ |
7254 | } |
7255 | ||
5c45bf27 | 7256 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7257 | int arch_reinit_sched_domains(void) |
5c45bf27 SS |
7258 | { |
7259 | int err; | |
7260 | ||
95402b38 | 7261 | get_online_cpus(); |
5c45bf27 SS |
7262 | detach_destroy_domains(&cpu_online_map); |
7263 | err = arch_init_sched_domains(&cpu_online_map); | |
95402b38 | 7264 | put_online_cpus(); |
5c45bf27 SS |
7265 | |
7266 | return err; | |
7267 | } | |
7268 | ||
7269 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7270 | { | |
7271 | int ret; | |
7272 | ||
7273 | if (buf[0] != '0' && buf[0] != '1') | |
7274 | return -EINVAL; | |
7275 | ||
7276 | if (smt) | |
7277 | sched_smt_power_savings = (buf[0] == '1'); | |
7278 | else | |
7279 | sched_mc_power_savings = (buf[0] == '1'); | |
7280 | ||
7281 | ret = arch_reinit_sched_domains(); | |
7282 | ||
7283 | return ret ? ret : count; | |
7284 | } | |
7285 | ||
5c45bf27 SS |
7286 | #ifdef CONFIG_SCHED_MC |
7287 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
7288 | { | |
7289 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7290 | } | |
48f24c4d IM |
7291 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
7292 | const char *buf, size_t count) | |
5c45bf27 SS |
7293 | { |
7294 | return sched_power_savings_store(buf, count, 0); | |
7295 | } | |
6707de00 AB |
7296 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
7297 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7298 | #endif |
7299 | ||
7300 | #ifdef CONFIG_SCHED_SMT | |
7301 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
7302 | { | |
7303 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7304 | } | |
48f24c4d IM |
7305 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
7306 | const char *buf, size_t count) | |
5c45bf27 SS |
7307 | { |
7308 | return sched_power_savings_store(buf, count, 1); | |
7309 | } | |
6707de00 AB |
7310 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
7311 | sched_smt_power_savings_store); | |
7312 | #endif | |
7313 | ||
7314 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7315 | { | |
7316 | int err = 0; | |
7317 | ||
7318 | #ifdef CONFIG_SCHED_SMT | |
7319 | if (smt_capable()) | |
7320 | err = sysfs_create_file(&cls->kset.kobj, | |
7321 | &attr_sched_smt_power_savings.attr); | |
7322 | #endif | |
7323 | #ifdef CONFIG_SCHED_MC | |
7324 | if (!err && mc_capable()) | |
7325 | err = sysfs_create_file(&cls->kset.kobj, | |
7326 | &attr_sched_mc_power_savings.attr); | |
7327 | #endif | |
7328 | return err; | |
7329 | } | |
5c45bf27 SS |
7330 | #endif |
7331 | ||
1da177e4 | 7332 | /* |
41a2d6cf | 7333 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 7334 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 7335 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
7336 | * which will prevent rebalancing while the sched domains are recalculated. |
7337 | */ | |
7338 | static int update_sched_domains(struct notifier_block *nfb, | |
7339 | unsigned long action, void *hcpu) | |
7340 | { | |
1da177e4 LT |
7341 | switch (action) { |
7342 | case CPU_UP_PREPARE: | |
8bb78442 | 7343 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 7344 | case CPU_DOWN_PREPARE: |
8bb78442 | 7345 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 7346 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
7347 | return NOTIFY_OK; |
7348 | ||
7349 | case CPU_UP_CANCELED: | |
8bb78442 | 7350 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 7351 | case CPU_DOWN_FAILED: |
8bb78442 | 7352 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7353 | case CPU_ONLINE: |
8bb78442 | 7354 | case CPU_ONLINE_FROZEN: |
1da177e4 | 7355 | case CPU_DEAD: |
8bb78442 | 7356 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
7357 | /* |
7358 | * Fall through and re-initialise the domains. | |
7359 | */ | |
7360 | break; | |
7361 | default: | |
7362 | return NOTIFY_DONE; | |
7363 | } | |
7364 | ||
7365 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 7366 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
7367 | |
7368 | return NOTIFY_OK; | |
7369 | } | |
1da177e4 LT |
7370 | |
7371 | void __init sched_init_smp(void) | |
7372 | { | |
5c1e1767 NP |
7373 | cpumask_t non_isolated_cpus; |
7374 | ||
434d53b0 MT |
7375 | #if defined(CONFIG_NUMA) |
7376 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7377 | GFP_KERNEL); | |
7378 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7379 | #endif | |
95402b38 | 7380 | get_online_cpus(); |
1a20ff27 | 7381 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7382 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7383 | if (cpus_empty(non_isolated_cpus)) |
7384 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
95402b38 | 7385 | put_online_cpus(); |
1da177e4 LT |
7386 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7387 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
7388 | |
7389 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 7390 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 7391 | BUG(); |
19978ca6 | 7392 | sched_init_granularity(); |
1da177e4 LT |
7393 | } |
7394 | #else | |
7395 | void __init sched_init_smp(void) | |
7396 | { | |
434d53b0 MT |
7397 | #if defined(CONFIG_NUMA) |
7398 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7399 | GFP_KERNEL); | |
7400 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7401 | #endif | |
19978ca6 | 7402 | sched_init_granularity(); |
1da177e4 LT |
7403 | } |
7404 | #endif /* CONFIG_SMP */ | |
7405 | ||
7406 | int in_sched_functions(unsigned long addr) | |
7407 | { | |
1da177e4 LT |
7408 | return in_lock_functions(addr) || |
7409 | (addr >= (unsigned long)__sched_text_start | |
7410 | && addr < (unsigned long)__sched_text_end); | |
7411 | } | |
7412 | ||
a9957449 | 7413 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7414 | { |
7415 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
7416 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7417 | cfs_rq->rq = rq; | |
7418 | #endif | |
67e9fb2a | 7419 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7420 | } |
7421 | ||
fa85ae24 PZ |
7422 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7423 | { | |
7424 | struct rt_prio_array *array; | |
7425 | int i; | |
7426 | ||
7427 | array = &rt_rq->active; | |
7428 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7429 | INIT_LIST_HEAD(array->queue + i); | |
7430 | __clear_bit(i, array->bitmap); | |
7431 | } | |
7432 | /* delimiter for bitsearch: */ | |
7433 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7434 | ||
052f1dc7 | 7435 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7436 | rt_rq->highest_prio = MAX_RT_PRIO; |
7437 | #endif | |
fa85ae24 PZ |
7438 | #ifdef CONFIG_SMP |
7439 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
7440 | rt_rq->overloaded = 0; |
7441 | #endif | |
7442 | ||
7443 | rt_rq->rt_time = 0; | |
7444 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
7445 | rt_rq->rt_runtime = 0; |
7446 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 7447 | |
052f1dc7 | 7448 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7449 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7450 | rt_rq->rq = rq; |
7451 | #endif | |
fa85ae24 PZ |
7452 | } |
7453 | ||
6f505b16 | 7454 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7455 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7456 | struct sched_entity *se, int cpu, int add, | |
7457 | struct sched_entity *parent) | |
6f505b16 | 7458 | { |
ec7dc8ac | 7459 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7460 | tg->cfs_rq[cpu] = cfs_rq; |
7461 | init_cfs_rq(cfs_rq, rq); | |
7462 | cfs_rq->tg = tg; | |
7463 | if (add) | |
7464 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7465 | ||
7466 | tg->se[cpu] = se; | |
354d60c2 DG |
7467 | /* se could be NULL for init_task_group */ |
7468 | if (!se) | |
7469 | return; | |
7470 | ||
ec7dc8ac DG |
7471 | if (!parent) |
7472 | se->cfs_rq = &rq->cfs; | |
7473 | else | |
7474 | se->cfs_rq = parent->my_q; | |
7475 | ||
6f505b16 PZ |
7476 | se->my_q = cfs_rq; |
7477 | se->load.weight = tg->shares; | |
7478 | se->load.inv_weight = div64_64(1ULL<<32, se->load.weight); | |
ec7dc8ac | 7479 | se->parent = parent; |
6f505b16 | 7480 | } |
052f1dc7 | 7481 | #endif |
6f505b16 | 7482 | |
052f1dc7 | 7483 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7484 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7485 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7486 | struct sched_rt_entity *parent) | |
6f505b16 | 7487 | { |
ec7dc8ac DG |
7488 | struct rq *rq = cpu_rq(cpu); |
7489 | ||
6f505b16 PZ |
7490 | tg->rt_rq[cpu] = rt_rq; |
7491 | init_rt_rq(rt_rq, rq); | |
7492 | rt_rq->tg = tg; | |
7493 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 7494 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7495 | if (add) |
7496 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7497 | ||
7498 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7499 | if (!rt_se) |
7500 | return; | |
7501 | ||
ec7dc8ac DG |
7502 | if (!parent) |
7503 | rt_se->rt_rq = &rq->rt; | |
7504 | else | |
7505 | rt_se->rt_rq = parent->my_q; | |
7506 | ||
6f505b16 PZ |
7507 | rt_se->rt_rq = &rq->rt; |
7508 | rt_se->my_q = rt_rq; | |
ec7dc8ac | 7509 | rt_se->parent = parent; |
6f505b16 PZ |
7510 | INIT_LIST_HEAD(&rt_se->run_list); |
7511 | } | |
7512 | #endif | |
7513 | ||
1da177e4 LT |
7514 | void __init sched_init(void) |
7515 | { | |
dd41f596 | 7516 | int i, j; |
434d53b0 MT |
7517 | unsigned long alloc_size = 0, ptr; |
7518 | ||
7519 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7520 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7521 | #endif | |
7522 | #ifdef CONFIG_RT_GROUP_SCHED | |
7523 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7524 | #endif |
7525 | #ifdef CONFIG_USER_SCHED | |
7526 | alloc_size *= 2; | |
434d53b0 MT |
7527 | #endif |
7528 | /* | |
7529 | * As sched_init() is called before page_alloc is setup, | |
7530 | * we use alloc_bootmem(). | |
7531 | */ | |
7532 | if (alloc_size) { | |
7533 | ptr = (unsigned long)alloc_bootmem_low(alloc_size); | |
7534 | ||
7535 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7536 | init_task_group.se = (struct sched_entity **)ptr; | |
7537 | ptr += nr_cpu_ids * sizeof(void **); | |
7538 | ||
7539 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7540 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7541 | |
7542 | #ifdef CONFIG_USER_SCHED | |
7543 | root_task_group.se = (struct sched_entity **)ptr; | |
7544 | ptr += nr_cpu_ids * sizeof(void **); | |
7545 | ||
7546 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7547 | ptr += nr_cpu_ids * sizeof(void **); | |
7548 | #endif | |
434d53b0 MT |
7549 | #endif |
7550 | #ifdef CONFIG_RT_GROUP_SCHED | |
7551 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7552 | ptr += nr_cpu_ids * sizeof(void **); | |
7553 | ||
7554 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7555 | ptr += nr_cpu_ids * sizeof(void **); |
7556 | ||
7557 | #ifdef CONFIG_USER_SCHED | |
7558 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7559 | ptr += nr_cpu_ids * sizeof(void **); | |
7560 | ||
7561 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
7562 | ptr += nr_cpu_ids * sizeof(void **); | |
7563 | #endif | |
434d53b0 MT |
7564 | #endif |
7565 | } | |
dd41f596 | 7566 | |
57d885fe GH |
7567 | #ifdef CONFIG_SMP |
7568 | init_defrootdomain(); | |
7569 | #endif | |
7570 | ||
d0b27fa7 PZ |
7571 | init_rt_bandwidth(&def_rt_bandwidth, |
7572 | global_rt_period(), global_rt_runtime()); | |
7573 | ||
7574 | #ifdef CONFIG_RT_GROUP_SCHED | |
7575 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7576 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
7577 | #ifdef CONFIG_USER_SCHED |
7578 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
7579 | global_rt_period(), RUNTIME_INF); | |
7580 | #endif | |
d0b27fa7 PZ |
7581 | #endif |
7582 | ||
052f1dc7 | 7583 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 7584 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7585 | INIT_LIST_HEAD(&init_task_group.children); |
7586 | ||
7587 | #ifdef CONFIG_USER_SCHED | |
7588 | INIT_LIST_HEAD(&root_task_group.children); | |
7589 | init_task_group.parent = &root_task_group; | |
7590 | list_add(&init_task_group.siblings, &root_task_group.children); | |
7591 | #endif | |
6f505b16 PZ |
7592 | #endif |
7593 | ||
0a945022 | 7594 | for_each_possible_cpu(i) { |
70b97a7f | 7595 | struct rq *rq; |
1da177e4 LT |
7596 | |
7597 | rq = cpu_rq(i); | |
7598 | spin_lock_init(&rq->lock); | |
fcb99371 | 7599 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 7600 | rq->nr_running = 0; |
dd41f596 | 7601 | rq->clock = 1; |
15934a37 | 7602 | update_last_tick_seen(rq); |
dd41f596 | 7603 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7604 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7605 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7606 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7607 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7608 | #ifdef CONFIG_CGROUP_SCHED |
7609 | /* | |
7610 | * How much cpu bandwidth does init_task_group get? | |
7611 | * | |
7612 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7613 | * gets 100% of the cpu resources in the system. This overall | |
7614 | * system cpu resource is divided among the tasks of | |
7615 | * init_task_group and its child task-groups in a fair manner, | |
7616 | * based on each entity's (task or task-group's) weight | |
7617 | * (se->load.weight). | |
7618 | * | |
7619 | * In other words, if init_task_group has 10 tasks of weight | |
7620 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7621 | * then A0's share of the cpu resource is: | |
7622 | * | |
7623 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
7624 | * | |
7625 | * We achieve this by letting init_task_group's tasks sit | |
7626 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7627 | */ | |
ec7dc8ac | 7628 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 7629 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
7630 | root_task_group.shares = NICE_0_LOAD; |
7631 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
7632 | /* |
7633 | * In case of task-groups formed thr' the user id of tasks, | |
7634 | * init_task_group represents tasks belonging to root user. | |
7635 | * Hence it forms a sibling of all subsequent groups formed. | |
7636 | * In this case, init_task_group gets only a fraction of overall | |
7637 | * system cpu resource, based on the weight assigned to root | |
7638 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
7639 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
7640 | * (init_cfs_rq) and having one entity represent this group of | |
7641 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
7642 | */ | |
ec7dc8ac | 7643 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 7644 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
7645 | &per_cpu(init_sched_entity, i), i, 1, |
7646 | root_task_group.se[i]); | |
6f505b16 | 7647 | |
052f1dc7 | 7648 | #endif |
354d60c2 DG |
7649 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7650 | ||
7651 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7652 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7653 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7654 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 7655 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 7656 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 7657 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 7658 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 7659 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
7660 | &per_cpu(init_sched_rt_entity, i), i, 1, |
7661 | root_task_group.rt_se[i]); | |
354d60c2 | 7662 | #endif |
dd41f596 | 7663 | #endif |
1da177e4 | 7664 | |
dd41f596 IM |
7665 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7666 | rq->cpu_load[j] = 0; | |
1da177e4 | 7667 | #ifdef CONFIG_SMP |
41c7ce9a | 7668 | rq->sd = NULL; |
57d885fe | 7669 | rq->rd = NULL; |
1da177e4 | 7670 | rq->active_balance = 0; |
dd41f596 | 7671 | rq->next_balance = jiffies; |
1da177e4 | 7672 | rq->push_cpu = 0; |
0a2966b4 | 7673 | rq->cpu = i; |
1da177e4 LT |
7674 | rq->migration_thread = NULL; |
7675 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 7676 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7677 | #endif |
8f4d37ec | 7678 | init_rq_hrtick(rq); |
1da177e4 | 7679 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7680 | } |
7681 | ||
2dd73a4f | 7682 | set_load_weight(&init_task); |
b50f60ce | 7683 | |
e107be36 AK |
7684 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7685 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7686 | #endif | |
7687 | ||
c9819f45 CL |
7688 | #ifdef CONFIG_SMP |
7689 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); | |
7690 | #endif | |
7691 | ||
b50f60ce HC |
7692 | #ifdef CONFIG_RT_MUTEXES |
7693 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
7694 | #endif | |
7695 | ||
1da177e4 LT |
7696 | /* |
7697 | * The boot idle thread does lazy MMU switching as well: | |
7698 | */ | |
7699 | atomic_inc(&init_mm.mm_count); | |
7700 | enter_lazy_tlb(&init_mm, current); | |
7701 | ||
7702 | /* | |
7703 | * Make us the idle thread. Technically, schedule() should not be | |
7704 | * called from this thread, however somewhere below it might be, | |
7705 | * but because we are the idle thread, we just pick up running again | |
7706 | * when this runqueue becomes "idle". | |
7707 | */ | |
7708 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
7709 | /* |
7710 | * During early bootup we pretend to be a normal task: | |
7711 | */ | |
7712 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
7713 | |
7714 | scheduler_running = 1; | |
1da177e4 LT |
7715 | } |
7716 | ||
7717 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
7718 | void __might_sleep(char *file, int line) | |
7719 | { | |
48f24c4d | 7720 | #ifdef in_atomic |
1da177e4 LT |
7721 | static unsigned long prev_jiffy; /* ratelimiting */ |
7722 | ||
7723 | if ((in_atomic() || irqs_disabled()) && | |
7724 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
7725 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7726 | return; | |
7727 | prev_jiffy = jiffies; | |
91368d73 | 7728 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
7729 | " context at %s:%d\n", file, line); |
7730 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
7731 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 7732 | debug_show_held_locks(current); |
3117df04 IM |
7733 | if (irqs_disabled()) |
7734 | print_irqtrace_events(current); | |
1da177e4 LT |
7735 | dump_stack(); |
7736 | } | |
7737 | #endif | |
7738 | } | |
7739 | EXPORT_SYMBOL(__might_sleep); | |
7740 | #endif | |
7741 | ||
7742 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7743 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7744 | { | |
7745 | int on_rq; | |
7746 | update_rq_clock(rq); | |
7747 | on_rq = p->se.on_rq; | |
7748 | if (on_rq) | |
7749 | deactivate_task(rq, p, 0); | |
7750 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7751 | if (on_rq) { | |
7752 | activate_task(rq, p, 0); | |
7753 | resched_task(rq->curr); | |
7754 | } | |
7755 | } | |
7756 | ||
1da177e4 LT |
7757 | void normalize_rt_tasks(void) |
7758 | { | |
a0f98a1c | 7759 | struct task_struct *g, *p; |
1da177e4 | 7760 | unsigned long flags; |
70b97a7f | 7761 | struct rq *rq; |
1da177e4 | 7762 | |
4cf5d77a | 7763 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7764 | do_each_thread(g, p) { |
178be793 IM |
7765 | /* |
7766 | * Only normalize user tasks: | |
7767 | */ | |
7768 | if (!p->mm) | |
7769 | continue; | |
7770 | ||
6cfb0d5d | 7771 | p->se.exec_start = 0; |
6cfb0d5d | 7772 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 7773 | p->se.wait_start = 0; |
dd41f596 | 7774 | p->se.sleep_start = 0; |
dd41f596 | 7775 | p->se.block_start = 0; |
6cfb0d5d | 7776 | #endif |
dd41f596 IM |
7777 | task_rq(p)->clock = 0; |
7778 | ||
7779 | if (!rt_task(p)) { | |
7780 | /* | |
7781 | * Renice negative nice level userspace | |
7782 | * tasks back to 0: | |
7783 | */ | |
7784 | if (TASK_NICE(p) < 0 && p->mm) | |
7785 | set_user_nice(p, 0); | |
1da177e4 | 7786 | continue; |
dd41f596 | 7787 | } |
1da177e4 | 7788 | |
4cf5d77a | 7789 | spin_lock(&p->pi_lock); |
b29739f9 | 7790 | rq = __task_rq_lock(p); |
1da177e4 | 7791 | |
178be793 | 7792 | normalize_task(rq, p); |
3a5e4dc1 | 7793 | |
b29739f9 | 7794 | __task_rq_unlock(rq); |
4cf5d77a | 7795 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7796 | } while_each_thread(g, p); |
7797 | ||
4cf5d77a | 7798 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7799 | } |
7800 | ||
7801 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7802 | |
7803 | #ifdef CONFIG_IA64 | |
7804 | /* | |
7805 | * These functions are only useful for the IA64 MCA handling. | |
7806 | * | |
7807 | * They can only be called when the whole system has been | |
7808 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7809 | * activity can take place. Using them for anything else would | |
7810 | * be a serious bug, and as a result, they aren't even visible | |
7811 | * under any other configuration. | |
7812 | */ | |
7813 | ||
7814 | /** | |
7815 | * curr_task - return the current task for a given cpu. | |
7816 | * @cpu: the processor in question. | |
7817 | * | |
7818 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7819 | */ | |
36c8b586 | 7820 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7821 | { |
7822 | return cpu_curr(cpu); | |
7823 | } | |
7824 | ||
7825 | /** | |
7826 | * set_curr_task - set the current task for a given cpu. | |
7827 | * @cpu: the processor in question. | |
7828 | * @p: the task pointer to set. | |
7829 | * | |
7830 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7831 | * are serviced on a separate stack. It allows the architecture to switch the |
7832 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7833 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7834 | * and caller must save the original value of the current task (see | |
7835 | * curr_task() above) and restore that value before reenabling interrupts and | |
7836 | * re-starting the system. | |
7837 | * | |
7838 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7839 | */ | |
36c8b586 | 7840 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7841 | { |
7842 | cpu_curr(cpu) = p; | |
7843 | } | |
7844 | ||
7845 | #endif | |
29f59db3 | 7846 | |
bccbe08a PZ |
7847 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7848 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
7849 | { |
7850 | int i; | |
7851 | ||
7852 | for_each_possible_cpu(i) { | |
7853 | if (tg->cfs_rq) | |
7854 | kfree(tg->cfs_rq[i]); | |
7855 | if (tg->se) | |
7856 | kfree(tg->se[i]); | |
6f505b16 PZ |
7857 | } |
7858 | ||
7859 | kfree(tg->cfs_rq); | |
7860 | kfree(tg->se); | |
6f505b16 PZ |
7861 | } |
7862 | ||
ec7dc8ac DG |
7863 | static |
7864 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 7865 | { |
29f59db3 | 7866 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 7867 | struct sched_entity *se, *parent_se; |
9b5b7751 | 7868 | struct rq *rq; |
29f59db3 SV |
7869 | int i; |
7870 | ||
434d53b0 | 7871 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7872 | if (!tg->cfs_rq) |
7873 | goto err; | |
434d53b0 | 7874 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7875 | if (!tg->se) |
7876 | goto err; | |
052f1dc7 PZ |
7877 | |
7878 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
7879 | |
7880 | for_each_possible_cpu(i) { | |
9b5b7751 | 7881 | rq = cpu_rq(i); |
29f59db3 | 7882 | |
6f505b16 PZ |
7883 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
7884 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
7885 | if (!cfs_rq) |
7886 | goto err; | |
7887 | ||
6f505b16 PZ |
7888 | se = kmalloc_node(sizeof(struct sched_entity), |
7889 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
7890 | if (!se) |
7891 | goto err; | |
7892 | ||
ec7dc8ac DG |
7893 | parent_se = parent ? parent->se[i] : NULL; |
7894 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
7895 | } |
7896 | ||
7897 | return 1; | |
7898 | ||
7899 | err: | |
7900 | return 0; | |
7901 | } | |
7902 | ||
7903 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7904 | { | |
7905 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
7906 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
7907 | } | |
7908 | ||
7909 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7910 | { | |
7911 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
7912 | } | |
7913 | #else | |
7914 | static inline void free_fair_sched_group(struct task_group *tg) | |
7915 | { | |
7916 | } | |
7917 | ||
ec7dc8ac DG |
7918 | static inline |
7919 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7920 | { |
7921 | return 1; | |
7922 | } | |
7923 | ||
7924 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7925 | { | |
7926 | } | |
7927 | ||
7928 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7929 | { | |
7930 | } | |
052f1dc7 PZ |
7931 | #endif |
7932 | ||
7933 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
7934 | static void free_rt_sched_group(struct task_group *tg) |
7935 | { | |
7936 | int i; | |
7937 | ||
d0b27fa7 PZ |
7938 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
7939 | ||
bccbe08a PZ |
7940 | for_each_possible_cpu(i) { |
7941 | if (tg->rt_rq) | |
7942 | kfree(tg->rt_rq[i]); | |
7943 | if (tg->rt_se) | |
7944 | kfree(tg->rt_se[i]); | |
7945 | } | |
7946 | ||
7947 | kfree(tg->rt_rq); | |
7948 | kfree(tg->rt_se); | |
7949 | } | |
7950 | ||
ec7dc8ac DG |
7951 | static |
7952 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7953 | { |
7954 | struct rt_rq *rt_rq; | |
ec7dc8ac | 7955 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
7956 | struct rq *rq; |
7957 | int i; | |
7958 | ||
434d53b0 | 7959 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
7960 | if (!tg->rt_rq) |
7961 | goto err; | |
434d53b0 | 7962 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
7963 | if (!tg->rt_se) |
7964 | goto err; | |
7965 | ||
d0b27fa7 PZ |
7966 | init_rt_bandwidth(&tg->rt_bandwidth, |
7967 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
7968 | |
7969 | for_each_possible_cpu(i) { | |
7970 | rq = cpu_rq(i); | |
7971 | ||
6f505b16 PZ |
7972 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
7973 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
7974 | if (!rt_rq) | |
7975 | goto err; | |
29f59db3 | 7976 | |
6f505b16 PZ |
7977 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
7978 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
7979 | if (!rt_se) | |
7980 | goto err; | |
29f59db3 | 7981 | |
ec7dc8ac DG |
7982 | parent_se = parent ? parent->rt_se[i] : NULL; |
7983 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
7984 | } |
7985 | ||
bccbe08a PZ |
7986 | return 1; |
7987 | ||
7988 | err: | |
7989 | return 0; | |
7990 | } | |
7991 | ||
7992 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7993 | { | |
7994 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
7995 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
7996 | } | |
7997 | ||
7998 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7999 | { | |
8000 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8001 | } | |
8002 | #else | |
8003 | static inline void free_rt_sched_group(struct task_group *tg) | |
8004 | { | |
8005 | } | |
8006 | ||
ec7dc8ac DG |
8007 | static inline |
8008 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8009 | { |
8010 | return 1; | |
8011 | } | |
8012 | ||
8013 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8014 | { | |
8015 | } | |
8016 | ||
8017 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8018 | { | |
8019 | } | |
8020 | #endif | |
8021 | ||
d0b27fa7 | 8022 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8023 | static void free_sched_group(struct task_group *tg) |
8024 | { | |
8025 | free_fair_sched_group(tg); | |
8026 | free_rt_sched_group(tg); | |
8027 | kfree(tg); | |
8028 | } | |
8029 | ||
8030 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8031 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8032 | { |
8033 | struct task_group *tg; | |
8034 | unsigned long flags; | |
8035 | int i; | |
8036 | ||
8037 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8038 | if (!tg) | |
8039 | return ERR_PTR(-ENOMEM); | |
8040 | ||
ec7dc8ac | 8041 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8042 | goto err; |
8043 | ||
ec7dc8ac | 8044 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8045 | goto err; |
8046 | ||
8ed36996 | 8047 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8048 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8049 | register_fair_sched_group(tg, i); |
8050 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8051 | } |
6f505b16 | 8052 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8053 | |
8054 | WARN_ON(!parent); /* root should already exist */ | |
8055 | ||
8056 | tg->parent = parent; | |
8057 | list_add_rcu(&tg->siblings, &parent->children); | |
8058 | INIT_LIST_HEAD(&tg->children); | |
8ed36996 | 8059 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8060 | |
9b5b7751 | 8061 | return tg; |
29f59db3 SV |
8062 | |
8063 | err: | |
6f505b16 | 8064 | free_sched_group(tg); |
29f59db3 SV |
8065 | return ERR_PTR(-ENOMEM); |
8066 | } | |
8067 | ||
9b5b7751 | 8068 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8069 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8070 | { |
29f59db3 | 8071 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8072 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8073 | } |
8074 | ||
9b5b7751 | 8075 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8076 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8077 | { |
8ed36996 | 8078 | unsigned long flags; |
9b5b7751 | 8079 | int i; |
29f59db3 | 8080 | |
8ed36996 | 8081 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8082 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8083 | unregister_fair_sched_group(tg, i); |
8084 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8085 | } |
6f505b16 | 8086 | list_del_rcu(&tg->list); |
f473aa5e | 8087 | list_del_rcu(&tg->siblings); |
8ed36996 | 8088 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8089 | |
9b5b7751 | 8090 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8091 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8092 | } |
8093 | ||
9b5b7751 | 8094 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8095 | * The caller of this function should have put the task in its new group |
8096 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8097 | * reflect its new group. | |
9b5b7751 SV |
8098 | */ |
8099 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8100 | { |
8101 | int on_rq, running; | |
8102 | unsigned long flags; | |
8103 | struct rq *rq; | |
8104 | ||
8105 | rq = task_rq_lock(tsk, &flags); | |
8106 | ||
29f59db3 SV |
8107 | update_rq_clock(rq); |
8108 | ||
051a1d1a | 8109 | running = task_current(rq, tsk); |
29f59db3 SV |
8110 | on_rq = tsk->se.on_rq; |
8111 | ||
0e1f3483 | 8112 | if (on_rq) |
29f59db3 | 8113 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8114 | if (unlikely(running)) |
8115 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8116 | |
6f505b16 | 8117 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8118 | |
810b3817 PZ |
8119 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8120 | if (tsk->sched_class->moved_group) | |
8121 | tsk->sched_class->moved_group(tsk); | |
8122 | #endif | |
8123 | ||
0e1f3483 HS |
8124 | if (unlikely(running)) |
8125 | tsk->sched_class->set_curr_task(rq); | |
8126 | if (on_rq) | |
7074badb | 8127 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8128 | |
29f59db3 SV |
8129 | task_rq_unlock(rq, &flags); |
8130 | } | |
d0b27fa7 | 8131 | #endif |
29f59db3 | 8132 | |
052f1dc7 | 8133 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
8134 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8135 | { | |
8136 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8137 | struct rq *rq = cfs_rq->rq; | |
8138 | int on_rq; | |
8139 | ||
62fb1851 | 8140 | spin_lock_irq(&rq->lock); |
29f59db3 SV |
8141 | |
8142 | on_rq = se->on_rq; | |
62fb1851 | 8143 | if (on_rq) |
29f59db3 SV |
8144 | dequeue_entity(cfs_rq, se, 0); |
8145 | ||
8146 | se->load.weight = shares; | |
8147 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
8148 | ||
62fb1851 | 8149 | if (on_rq) |
29f59db3 | 8150 | enqueue_entity(cfs_rq, se, 0); |
62fb1851 PZ |
8151 | |
8152 | spin_unlock_irq(&rq->lock); | |
29f59db3 SV |
8153 | } |
8154 | ||
8ed36996 PZ |
8155 | static DEFINE_MUTEX(shares_mutex); |
8156 | ||
4cf86d77 | 8157 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8158 | { |
8159 | int i; | |
8ed36996 | 8160 | unsigned long flags; |
c61935fd | 8161 | |
ec7dc8ac DG |
8162 | /* |
8163 | * We can't change the weight of the root cgroup. | |
8164 | */ | |
8165 | if (!tg->se[0]) | |
8166 | return -EINVAL; | |
8167 | ||
62fb1851 PZ |
8168 | /* |
8169 | * A weight of 0 or 1 can cause arithmetics problems. | |
8170 | * (The default weight is 1024 - so there's no practical | |
8171 | * limitation from this.) | |
8172 | */ | |
8173 | if (shares < 2) | |
8174 | shares = 2; | |
8175 | ||
8ed36996 | 8176 | mutex_lock(&shares_mutex); |
9b5b7751 | 8177 | if (tg->shares == shares) |
5cb350ba | 8178 | goto done; |
29f59db3 | 8179 | |
8ed36996 | 8180 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8181 | for_each_possible_cpu(i) |
8182 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8183 | list_del_rcu(&tg->siblings); |
8ed36996 | 8184 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8185 | |
8186 | /* wait for any ongoing reference to this group to finish */ | |
8187 | synchronize_sched(); | |
8188 | ||
8189 | /* | |
8190 | * Now we are free to modify the group's share on each cpu | |
8191 | * w/o tripping rebalance_share or load_balance_fair. | |
8192 | */ | |
9b5b7751 | 8193 | tg->shares = shares; |
62fb1851 | 8194 | for_each_possible_cpu(i) |
9b5b7751 | 8195 | set_se_shares(tg->se[i], shares); |
29f59db3 | 8196 | |
6b2d7700 SV |
8197 | /* |
8198 | * Enable load balance activity on this group, by inserting it back on | |
8199 | * each cpu's rq->leaf_cfs_rq_list. | |
8200 | */ | |
8ed36996 | 8201 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8202 | for_each_possible_cpu(i) |
8203 | register_fair_sched_group(tg, i); | |
f473aa5e | 8204 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8205 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8206 | done: |
8ed36996 | 8207 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8208 | return 0; |
29f59db3 SV |
8209 | } |
8210 | ||
5cb350ba DG |
8211 | unsigned long sched_group_shares(struct task_group *tg) |
8212 | { | |
8213 | return tg->shares; | |
8214 | } | |
052f1dc7 | 8215 | #endif |
5cb350ba | 8216 | |
052f1dc7 | 8217 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8218 | /* |
9f0c1e56 | 8219 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8220 | */ |
9f0c1e56 PZ |
8221 | static DEFINE_MUTEX(rt_constraints_mutex); |
8222 | ||
8223 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8224 | { | |
8225 | if (runtime == RUNTIME_INF) | |
8226 | return 1ULL << 16; | |
8227 | ||
2692a240 | 8228 | return div64_64(runtime << 16, period); |
9f0c1e56 PZ |
8229 | } |
8230 | ||
b40b2e8e PZ |
8231 | #ifdef CONFIG_CGROUP_SCHED |
8232 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
8233 | { | |
8234 | struct task_group *tgi, *parent = tg->parent; | |
8235 | unsigned long total = 0; | |
8236 | ||
8237 | if (!parent) { | |
8238 | if (global_rt_period() < period) | |
8239 | return 0; | |
8240 | ||
8241 | return to_ratio(period, runtime) < | |
8242 | to_ratio(global_rt_period(), global_rt_runtime()); | |
8243 | } | |
8244 | ||
8245 | if (ktime_to_ns(parent->rt_bandwidth.rt_period) < period) | |
8246 | return 0; | |
8247 | ||
8248 | rcu_read_lock(); | |
8249 | list_for_each_entry_rcu(tgi, &parent->children, siblings) { | |
8250 | if (tgi == tg) | |
8251 | continue; | |
8252 | ||
8253 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), | |
8254 | tgi->rt_bandwidth.rt_runtime); | |
8255 | } | |
8256 | rcu_read_unlock(); | |
8257 | ||
8258 | return total + to_ratio(period, runtime) < | |
8259 | to_ratio(ktime_to_ns(parent->rt_bandwidth.rt_period), | |
8260 | parent->rt_bandwidth.rt_runtime); | |
8261 | } | |
8262 | #elif defined CONFIG_USER_SCHED | |
9f0c1e56 | 8263 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
6f505b16 PZ |
8264 | { |
8265 | struct task_group *tgi; | |
8266 | unsigned long total = 0; | |
9f0c1e56 | 8267 | unsigned long global_ratio = |
d0b27fa7 | 8268 | to_ratio(global_rt_period(), global_rt_runtime()); |
6f505b16 PZ |
8269 | |
8270 | rcu_read_lock(); | |
9f0c1e56 PZ |
8271 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
8272 | if (tgi == tg) | |
8273 | continue; | |
6f505b16 | 8274 | |
d0b27fa7 PZ |
8275 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), |
8276 | tgi->rt_bandwidth.rt_runtime); | |
9f0c1e56 PZ |
8277 | } |
8278 | rcu_read_unlock(); | |
6f505b16 | 8279 | |
9f0c1e56 | 8280 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 | 8281 | } |
b40b2e8e | 8282 | #endif |
6f505b16 | 8283 | |
521f1a24 DG |
8284 | /* Must be called with tasklist_lock held */ |
8285 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
8286 | { | |
8287 | struct task_struct *g, *p; | |
8288 | do_each_thread(g, p) { | |
8289 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8290 | return 1; | |
8291 | } while_each_thread(g, p); | |
8292 | return 0; | |
8293 | } | |
8294 | ||
d0b27fa7 PZ |
8295 | static int tg_set_bandwidth(struct task_group *tg, |
8296 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8297 | { |
ac086bc2 | 8298 | int i, err = 0; |
9f0c1e56 | 8299 | |
9f0c1e56 | 8300 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8301 | read_lock(&tasklist_lock); |
ac086bc2 | 8302 | if (rt_runtime == 0 && tg_has_rt_tasks(tg)) { |
521f1a24 DG |
8303 | err = -EBUSY; |
8304 | goto unlock; | |
8305 | } | |
9f0c1e56 PZ |
8306 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
8307 | err = -EINVAL; | |
8308 | goto unlock; | |
8309 | } | |
ac086bc2 PZ |
8310 | |
8311 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8312 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8313 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8314 | |
8315 | for_each_possible_cpu(i) { | |
8316 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8317 | ||
8318 | spin_lock(&rt_rq->rt_runtime_lock); | |
8319 | rt_rq->rt_runtime = rt_runtime; | |
8320 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8321 | } | |
8322 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8323 | unlock: |
521f1a24 | 8324 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8325 | mutex_unlock(&rt_constraints_mutex); |
8326 | ||
8327 | return err; | |
6f505b16 PZ |
8328 | } |
8329 | ||
d0b27fa7 PZ |
8330 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8331 | { | |
8332 | u64 rt_runtime, rt_period; | |
8333 | ||
8334 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8335 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8336 | if (rt_runtime_us < 0) | |
8337 | rt_runtime = RUNTIME_INF; | |
8338 | ||
8339 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8340 | } | |
8341 | ||
9f0c1e56 PZ |
8342 | long sched_group_rt_runtime(struct task_group *tg) |
8343 | { | |
8344 | u64 rt_runtime_us; | |
8345 | ||
d0b27fa7 | 8346 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8347 | return -1; |
8348 | ||
d0b27fa7 | 8349 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8350 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8351 | return rt_runtime_us; | |
8352 | } | |
d0b27fa7 PZ |
8353 | |
8354 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8355 | { | |
8356 | u64 rt_runtime, rt_period; | |
8357 | ||
8358 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8359 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8360 | ||
8361 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8362 | } | |
8363 | ||
8364 | long sched_group_rt_period(struct task_group *tg) | |
8365 | { | |
8366 | u64 rt_period_us; | |
8367 | ||
8368 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8369 | do_div(rt_period_us, NSEC_PER_USEC); | |
8370 | return rt_period_us; | |
8371 | } | |
8372 | ||
8373 | static int sched_rt_global_constraints(void) | |
8374 | { | |
8375 | int ret = 0; | |
8376 | ||
8377 | mutex_lock(&rt_constraints_mutex); | |
8378 | if (!__rt_schedulable(NULL, 1, 0)) | |
8379 | ret = -EINVAL; | |
8380 | mutex_unlock(&rt_constraints_mutex); | |
8381 | ||
8382 | return ret; | |
8383 | } | |
8384 | #else | |
8385 | static int sched_rt_global_constraints(void) | |
8386 | { | |
ac086bc2 PZ |
8387 | unsigned long flags; |
8388 | int i; | |
8389 | ||
8390 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8391 | for_each_possible_cpu(i) { | |
8392 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8393 | ||
8394 | spin_lock(&rt_rq->rt_runtime_lock); | |
8395 | rt_rq->rt_runtime = global_rt_runtime(); | |
8396 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8397 | } | |
8398 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8399 | ||
d0b27fa7 PZ |
8400 | return 0; |
8401 | } | |
052f1dc7 | 8402 | #endif |
d0b27fa7 PZ |
8403 | |
8404 | int sched_rt_handler(struct ctl_table *table, int write, | |
8405 | struct file *filp, void __user *buffer, size_t *lenp, | |
8406 | loff_t *ppos) | |
8407 | { | |
8408 | int ret; | |
8409 | int old_period, old_runtime; | |
8410 | static DEFINE_MUTEX(mutex); | |
8411 | ||
8412 | mutex_lock(&mutex); | |
8413 | old_period = sysctl_sched_rt_period; | |
8414 | old_runtime = sysctl_sched_rt_runtime; | |
8415 | ||
8416 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
8417 | ||
8418 | if (!ret && write) { | |
8419 | ret = sched_rt_global_constraints(); | |
8420 | if (ret) { | |
8421 | sysctl_sched_rt_period = old_period; | |
8422 | sysctl_sched_rt_runtime = old_runtime; | |
8423 | } else { | |
8424 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8425 | def_rt_bandwidth.rt_period = | |
8426 | ns_to_ktime(global_rt_period()); | |
8427 | } | |
8428 | } | |
8429 | mutex_unlock(&mutex); | |
8430 | ||
8431 | return ret; | |
8432 | } | |
68318b8e | 8433 | |
052f1dc7 | 8434 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8435 | |
8436 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8437 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8438 | { |
2b01dfe3 PM |
8439 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8440 | struct task_group, css); | |
68318b8e SV |
8441 | } |
8442 | ||
8443 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8444 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8445 | { |
ec7dc8ac | 8446 | struct task_group *tg, *parent; |
68318b8e | 8447 | |
2b01dfe3 | 8448 | if (!cgrp->parent) { |
68318b8e | 8449 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 8450 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
8451 | return &init_task_group.css; |
8452 | } | |
8453 | ||
ec7dc8ac DG |
8454 | parent = cgroup_tg(cgrp->parent); |
8455 | tg = sched_create_group(parent); | |
68318b8e SV |
8456 | if (IS_ERR(tg)) |
8457 | return ERR_PTR(-ENOMEM); | |
8458 | ||
8459 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 8460 | tg->css.cgroup = cgrp; |
68318b8e SV |
8461 | |
8462 | return &tg->css; | |
8463 | } | |
8464 | ||
41a2d6cf IM |
8465 | static void |
8466 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8467 | { |
2b01dfe3 | 8468 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8469 | |
8470 | sched_destroy_group(tg); | |
8471 | } | |
8472 | ||
41a2d6cf IM |
8473 | static int |
8474 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8475 | struct task_struct *tsk) | |
68318b8e | 8476 | { |
b68aa230 PZ |
8477 | #ifdef CONFIG_RT_GROUP_SCHED |
8478 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 8479 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
8480 | return -EINVAL; |
8481 | #else | |
68318b8e SV |
8482 | /* We don't support RT-tasks being in separate groups */ |
8483 | if (tsk->sched_class != &fair_sched_class) | |
8484 | return -EINVAL; | |
b68aa230 | 8485 | #endif |
68318b8e SV |
8486 | |
8487 | return 0; | |
8488 | } | |
8489 | ||
8490 | static void | |
2b01dfe3 | 8491 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
8492 | struct cgroup *old_cont, struct task_struct *tsk) |
8493 | { | |
8494 | sched_move_task(tsk); | |
8495 | } | |
8496 | ||
052f1dc7 | 8497 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2b01dfe3 PM |
8498 | static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype, |
8499 | u64 shareval) | |
68318b8e | 8500 | { |
2b01dfe3 | 8501 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8502 | } |
8503 | ||
2b01dfe3 | 8504 | static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8505 | { |
2b01dfe3 | 8506 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8507 | |
8508 | return (u64) tg->shares; | |
8509 | } | |
052f1dc7 | 8510 | #endif |
68318b8e | 8511 | |
052f1dc7 | 8512 | #ifdef CONFIG_RT_GROUP_SCHED |
ac086bc2 | 8513 | static ssize_t cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
9f0c1e56 PZ |
8514 | struct file *file, |
8515 | const char __user *userbuf, | |
8516 | size_t nbytes, loff_t *unused_ppos) | |
6f505b16 | 8517 | { |
9f0c1e56 PZ |
8518 | char buffer[64]; |
8519 | int retval = 0; | |
8520 | s64 val; | |
8521 | char *end; | |
8522 | ||
8523 | if (!nbytes) | |
8524 | return -EINVAL; | |
8525 | if (nbytes >= sizeof(buffer)) | |
8526 | return -E2BIG; | |
8527 | if (copy_from_user(buffer, userbuf, nbytes)) | |
8528 | return -EFAULT; | |
8529 | ||
8530 | buffer[nbytes] = 0; /* nul-terminate */ | |
8531 | ||
8532 | /* strip newline if necessary */ | |
8533 | if (nbytes && (buffer[nbytes-1] == '\n')) | |
8534 | buffer[nbytes-1] = 0; | |
8535 | val = simple_strtoll(buffer, &end, 0); | |
8536 | if (*end) | |
8537 | return -EINVAL; | |
8538 | ||
8539 | /* Pass to subsystem */ | |
8540 | retval = sched_group_set_rt_runtime(cgroup_tg(cgrp), val); | |
8541 | if (!retval) | |
8542 | retval = nbytes; | |
8543 | return retval; | |
6f505b16 PZ |
8544 | } |
8545 | ||
9f0c1e56 PZ |
8546 | static ssize_t cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft, |
8547 | struct file *file, | |
8548 | char __user *buf, size_t nbytes, | |
8549 | loff_t *ppos) | |
6f505b16 | 8550 | { |
9f0c1e56 PZ |
8551 | char tmp[64]; |
8552 | long val = sched_group_rt_runtime(cgroup_tg(cgrp)); | |
8553 | int len = sprintf(tmp, "%ld\n", val); | |
6f505b16 | 8554 | |
9f0c1e56 | 8555 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); |
6f505b16 | 8556 | } |
d0b27fa7 PZ |
8557 | |
8558 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8559 | u64 rt_period_us) | |
8560 | { | |
8561 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8562 | } | |
8563 | ||
8564 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8565 | { | |
8566 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8567 | } | |
052f1dc7 | 8568 | #endif |
6f505b16 | 8569 | |
fe5c7cc2 | 8570 | static struct cftype cpu_files[] = { |
052f1dc7 | 8571 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8572 | { |
8573 | .name = "shares", | |
8574 | .read_uint = cpu_shares_read_uint, | |
8575 | .write_uint = cpu_shares_write_uint, | |
8576 | }, | |
052f1dc7 PZ |
8577 | #endif |
8578 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8579 | { |
9f0c1e56 PZ |
8580 | .name = "rt_runtime_us", |
8581 | .read = cpu_rt_runtime_read, | |
8582 | .write = cpu_rt_runtime_write, | |
6f505b16 | 8583 | }, |
d0b27fa7 PZ |
8584 | { |
8585 | .name = "rt_period_us", | |
8586 | .read_uint = cpu_rt_period_read_uint, | |
8587 | .write_uint = cpu_rt_period_write_uint, | |
8588 | }, | |
052f1dc7 | 8589 | #endif |
68318b8e SV |
8590 | }; |
8591 | ||
8592 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8593 | { | |
fe5c7cc2 | 8594 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8595 | } |
8596 | ||
8597 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8598 | .name = "cpu", |
8599 | .create = cpu_cgroup_create, | |
8600 | .destroy = cpu_cgroup_destroy, | |
8601 | .can_attach = cpu_cgroup_can_attach, | |
8602 | .attach = cpu_cgroup_attach, | |
8603 | .populate = cpu_cgroup_populate, | |
8604 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8605 | .early_init = 1, |
8606 | }; | |
8607 | ||
052f1dc7 | 8608 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8609 | |
8610 | #ifdef CONFIG_CGROUP_CPUACCT | |
8611 | ||
8612 | /* | |
8613 | * CPU accounting code for task groups. | |
8614 | * | |
8615 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8616 | * (balbir@in.ibm.com). | |
8617 | */ | |
8618 | ||
8619 | /* track cpu usage of a group of tasks */ | |
8620 | struct cpuacct { | |
8621 | struct cgroup_subsys_state css; | |
8622 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
8623 | u64 *cpuusage; | |
8624 | }; | |
8625 | ||
8626 | struct cgroup_subsys cpuacct_subsys; | |
8627 | ||
8628 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8629 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8630 | { |
32cd756a | 8631 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8632 | struct cpuacct, css); |
8633 | } | |
8634 | ||
8635 | /* return cpu accounting group to which this task belongs */ | |
8636 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8637 | { | |
8638 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8639 | struct cpuacct, css); | |
8640 | } | |
8641 | ||
8642 | /* create a new cpu accounting group */ | |
8643 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8644 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8645 | { |
8646 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
8647 | ||
8648 | if (!ca) | |
8649 | return ERR_PTR(-ENOMEM); | |
8650 | ||
8651 | ca->cpuusage = alloc_percpu(u64); | |
8652 | if (!ca->cpuusage) { | |
8653 | kfree(ca); | |
8654 | return ERR_PTR(-ENOMEM); | |
8655 | } | |
8656 | ||
8657 | return &ca->css; | |
8658 | } | |
8659 | ||
8660 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8661 | static void |
32cd756a | 8662 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8663 | { |
32cd756a | 8664 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8665 | |
8666 | free_percpu(ca->cpuusage); | |
8667 | kfree(ca); | |
8668 | } | |
8669 | ||
8670 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 8671 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8672 | { |
32cd756a | 8673 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8674 | u64 totalcpuusage = 0; |
8675 | int i; | |
8676 | ||
8677 | for_each_possible_cpu(i) { | |
8678 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
8679 | ||
8680 | /* | |
8681 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
8682 | * platforms. | |
8683 | */ | |
8684 | spin_lock_irq(&cpu_rq(i)->lock); | |
8685 | totalcpuusage += *cpuusage; | |
8686 | spin_unlock_irq(&cpu_rq(i)->lock); | |
8687 | } | |
8688 | ||
8689 | return totalcpuusage; | |
8690 | } | |
8691 | ||
0297b803 DG |
8692 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8693 | u64 reset) | |
8694 | { | |
8695 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8696 | int err = 0; | |
8697 | int i; | |
8698 | ||
8699 | if (reset) { | |
8700 | err = -EINVAL; | |
8701 | goto out; | |
8702 | } | |
8703 | ||
8704 | for_each_possible_cpu(i) { | |
8705 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
8706 | ||
8707 | spin_lock_irq(&cpu_rq(i)->lock); | |
8708 | *cpuusage = 0; | |
8709 | spin_unlock_irq(&cpu_rq(i)->lock); | |
8710 | } | |
8711 | out: | |
8712 | return err; | |
8713 | } | |
8714 | ||
d842de87 SV |
8715 | static struct cftype files[] = { |
8716 | { | |
8717 | .name = "usage", | |
8718 | .read_uint = cpuusage_read, | |
0297b803 | 8719 | .write_uint = cpuusage_write, |
d842de87 SV |
8720 | }, |
8721 | }; | |
8722 | ||
32cd756a | 8723 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8724 | { |
32cd756a | 8725 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8726 | } |
8727 | ||
8728 | /* | |
8729 | * charge this task's execution time to its accounting group. | |
8730 | * | |
8731 | * called with rq->lock held. | |
8732 | */ | |
8733 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8734 | { | |
8735 | struct cpuacct *ca; | |
8736 | ||
8737 | if (!cpuacct_subsys.active) | |
8738 | return; | |
8739 | ||
8740 | ca = task_ca(tsk); | |
8741 | if (ca) { | |
8742 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
8743 | ||
8744 | *cpuusage += cputime; | |
8745 | } | |
8746 | } | |
8747 | ||
8748 | struct cgroup_subsys cpuacct_subsys = { | |
8749 | .name = "cpuacct", | |
8750 | .create = cpuacct_create, | |
8751 | .destroy = cpuacct_destroy, | |
8752 | .populate = cpuacct_populate, | |
8753 | .subsys_id = cpuacct_subsys_id, | |
8754 | }; | |
8755 | #endif /* CONFIG_CGROUP_CPUACCT */ |