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