]>
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> |
cdd6c482 | 42 | #include <linux/perf_event.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
58 | #include <linux/kthread.h> | |
b5aadf7f | 59 | #include <linux/proc_fs.h> |
1da177e4 | 60 | #include <linux/seq_file.h> |
e692ab53 | 61 | #include <linux/sysctl.h> |
1da177e4 LT |
62 | #include <linux/syscalls.h> |
63 | #include <linux/times.h> | |
8f0ab514 | 64 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 65 | #include <linux/kprobes.h> |
0ff92245 | 66 | #include <linux/delayacct.h> |
dff06c15 | 67 | #include <linux/unistd.h> |
f5ff8422 | 68 | #include <linux/pagemap.h> |
8f4d37ec | 69 | #include <linux/hrtimer.h> |
30914a58 | 70 | #include <linux/tick.h> |
f00b45c1 PZ |
71 | #include <linux/debugfs.h> |
72 | #include <linux/ctype.h> | |
6cd8a4bb | 73 | #include <linux/ftrace.h> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
1da177e4 | 77 | |
6e0534f2 GH |
78 | #include "sched_cpupri.h" |
79 | ||
a8d154b0 | 80 | #define CREATE_TRACE_POINTS |
ad8d75ff | 81 | #include <trace/events/sched.h> |
a8d154b0 | 82 | |
1da177e4 LT |
83 | /* |
84 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
85 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
86 | * and back. | |
87 | */ | |
88 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
89 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
90 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
91 | ||
92 | /* | |
93 | * 'User priority' is the nice value converted to something we | |
94 | * can work with better when scaling various scheduler parameters, | |
95 | * it's a [ 0 ... 39 ] range. | |
96 | */ | |
97 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
98 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
99 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
100 | ||
101 | /* | |
d7876a08 | 102 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 103 | */ |
d6322faf | 104 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 105 | |
6aa645ea IM |
106 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
107 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
108 | ||
1da177e4 LT |
109 | /* |
110 | * These are the 'tuning knobs' of the scheduler: | |
111 | * | |
a4ec24b4 | 112 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
113 | * Timeslices get refilled after they expire. |
114 | */ | |
1da177e4 | 115 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 116 | |
d0b27fa7 PZ |
117 | /* |
118 | * single value that denotes runtime == period, ie unlimited time. | |
119 | */ | |
120 | #define RUNTIME_INF ((u64)~0ULL) | |
121 | ||
e05606d3 IM |
122 | static inline int rt_policy(int policy) |
123 | { | |
3f33a7ce | 124 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
125 | return 1; |
126 | return 0; | |
127 | } | |
128 | ||
129 | static inline int task_has_rt_policy(struct task_struct *p) | |
130 | { | |
131 | return rt_policy(p->policy); | |
132 | } | |
133 | ||
1da177e4 | 134 | /* |
6aa645ea | 135 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 136 | */ |
6aa645ea IM |
137 | struct rt_prio_array { |
138 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
139 | struct list_head queue[MAX_RT_PRIO]; | |
140 | }; | |
141 | ||
d0b27fa7 | 142 | struct rt_bandwidth { |
ea736ed5 | 143 | /* nests inside the rq lock: */ |
0986b11b | 144 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
145 | ktime_t rt_period; |
146 | u64 rt_runtime; | |
147 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
148 | }; |
149 | ||
150 | static struct rt_bandwidth def_rt_bandwidth; | |
151 | ||
152 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
153 | ||
154 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
155 | { | |
156 | struct rt_bandwidth *rt_b = | |
157 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
158 | ktime_t now; | |
159 | int overrun; | |
160 | int idle = 0; | |
161 | ||
162 | for (;;) { | |
163 | now = hrtimer_cb_get_time(timer); | |
164 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
165 | ||
166 | if (!overrun) | |
167 | break; | |
168 | ||
169 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
170 | } | |
171 | ||
172 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
173 | } | |
174 | ||
175 | static | |
176 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
177 | { | |
178 | rt_b->rt_period = ns_to_ktime(period); | |
179 | rt_b->rt_runtime = runtime; | |
180 | ||
0986b11b | 181 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 182 | |
d0b27fa7 PZ |
183 | hrtimer_init(&rt_b->rt_period_timer, |
184 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
185 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
186 | } |
187 | ||
c8bfff6d KH |
188 | static inline int rt_bandwidth_enabled(void) |
189 | { | |
190 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
191 | } |
192 | ||
193 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
194 | { | |
195 | ktime_t now; | |
196 | ||
cac64d00 | 197 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
198 | return; |
199 | ||
200 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
201 | return; | |
202 | ||
0986b11b | 203 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 204 | for (;;) { |
7f1e2ca9 PZ |
205 | unsigned long delta; |
206 | ktime_t soft, hard; | |
207 | ||
d0b27fa7 PZ |
208 | if (hrtimer_active(&rt_b->rt_period_timer)) |
209 | break; | |
210 | ||
211 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
212 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
213 | |
214 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
215 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
216 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
217 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 218 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 219 | } |
0986b11b | 220 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
221 | } |
222 | ||
223 | #ifdef CONFIG_RT_GROUP_SCHED | |
224 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
225 | { | |
226 | hrtimer_cancel(&rt_b->rt_period_timer); | |
227 | } | |
228 | #endif | |
229 | ||
712555ee HC |
230 | /* |
231 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
232 | * detach_destroy_domains and partition_sched_domains. | |
233 | */ | |
234 | static DEFINE_MUTEX(sched_domains_mutex); | |
235 | ||
7c941438 | 236 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 237 | |
68318b8e SV |
238 | #include <linux/cgroup.h> |
239 | ||
29f59db3 SV |
240 | struct cfs_rq; |
241 | ||
6f505b16 PZ |
242 | static LIST_HEAD(task_groups); |
243 | ||
29f59db3 | 244 | /* task group related information */ |
4cf86d77 | 245 | struct task_group { |
68318b8e | 246 | struct cgroup_subsys_state css; |
6c415b92 | 247 | |
052f1dc7 | 248 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
249 | /* schedulable entities of this group on each cpu */ |
250 | struct sched_entity **se; | |
251 | /* runqueue "owned" by this group on each cpu */ | |
252 | struct cfs_rq **cfs_rq; | |
253 | unsigned long shares; | |
052f1dc7 PZ |
254 | #endif |
255 | ||
256 | #ifdef CONFIG_RT_GROUP_SCHED | |
257 | struct sched_rt_entity **rt_se; | |
258 | struct rt_rq **rt_rq; | |
259 | ||
d0b27fa7 | 260 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 261 | #endif |
6b2d7700 | 262 | |
ae8393e5 | 263 | struct rcu_head rcu; |
6f505b16 | 264 | struct list_head list; |
f473aa5e PZ |
265 | |
266 | struct task_group *parent; | |
267 | struct list_head siblings; | |
268 | struct list_head children; | |
29f59db3 SV |
269 | }; |
270 | ||
eff766a6 | 271 | #define root_task_group init_task_group |
6f505b16 | 272 | |
8ed36996 | 273 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
274 | * a task group's cpu shares. |
275 | */ | |
8ed36996 | 276 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 277 | |
e9036b36 CG |
278 | #ifdef CONFIG_FAIR_GROUP_SCHED |
279 | ||
57310a98 PZ |
280 | #ifdef CONFIG_SMP |
281 | static int root_task_group_empty(void) | |
282 | { | |
283 | return list_empty(&root_task_group.children); | |
284 | } | |
285 | #endif | |
286 | ||
052f1dc7 | 287 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 288 | |
cb4ad1ff | 289 | /* |
2e084786 LJ |
290 | * A weight of 0 or 1 can cause arithmetics problems. |
291 | * A weight of a cfs_rq is the sum of weights of which entities | |
292 | * are queued on this cfs_rq, so a weight of a entity should not be | |
293 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
294 | * (The default weight is 1024 - so there's no practical |
295 | * limitation from this.) | |
296 | */ | |
18d95a28 | 297 | #define MIN_SHARES 2 |
2e084786 | 298 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 299 | |
052f1dc7 PZ |
300 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
301 | #endif | |
302 | ||
29f59db3 | 303 | /* Default task group. |
3a252015 | 304 | * Every task in system belong to this group at bootup. |
29f59db3 | 305 | */ |
434d53b0 | 306 | struct task_group init_task_group; |
29f59db3 SV |
307 | |
308 | /* return group to which a task belongs */ | |
4cf86d77 | 309 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 310 | { |
4cf86d77 | 311 | struct task_group *tg; |
9b5b7751 | 312 | |
7c941438 | 313 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
314 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
315 | struct task_group, css); | |
24e377a8 | 316 | #else |
41a2d6cf | 317 | tg = &init_task_group; |
24e377a8 | 318 | #endif |
9b5b7751 | 319 | return tg; |
29f59db3 SV |
320 | } |
321 | ||
322 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 323 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 324 | { |
052f1dc7 | 325 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
326 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
327 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 328 | #endif |
6f505b16 | 329 | |
052f1dc7 | 330 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
331 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
332 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 333 | #endif |
29f59db3 SV |
334 | } |
335 | ||
336 | #else | |
337 | ||
6f505b16 | 338 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
339 | static inline struct task_group *task_group(struct task_struct *p) |
340 | { | |
341 | return NULL; | |
342 | } | |
29f59db3 | 343 | |
7c941438 | 344 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 345 | |
6aa645ea IM |
346 | /* CFS-related fields in a runqueue */ |
347 | struct cfs_rq { | |
348 | struct load_weight load; | |
349 | unsigned long nr_running; | |
350 | ||
6aa645ea | 351 | u64 exec_clock; |
e9acbff6 | 352 | u64 min_vruntime; |
6aa645ea IM |
353 | |
354 | struct rb_root tasks_timeline; | |
355 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
356 | |
357 | struct list_head tasks; | |
358 | struct list_head *balance_iterator; | |
359 | ||
360 | /* | |
361 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
362 | * It is set to NULL otherwise (i.e when none are currently running). |
363 | */ | |
4793241b | 364 | struct sched_entity *curr, *next, *last; |
ddc97297 | 365 | |
5ac5c4d6 | 366 | unsigned int nr_spread_over; |
ddc97297 | 367 | |
62160e3f | 368 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
369 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
370 | ||
41a2d6cf IM |
371 | /* |
372 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
373 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
374 | * (like users, containers etc.) | |
375 | * | |
376 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
377 | * list is used during load balance. | |
378 | */ | |
41a2d6cf IM |
379 | struct list_head leaf_cfs_rq_list; |
380 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
381 | |
382 | #ifdef CONFIG_SMP | |
c09595f6 | 383 | /* |
c8cba857 | 384 | * the part of load.weight contributed by tasks |
c09595f6 | 385 | */ |
c8cba857 | 386 | unsigned long task_weight; |
c09595f6 | 387 | |
c8cba857 PZ |
388 | /* |
389 | * h_load = weight * f(tg) | |
390 | * | |
391 | * Where f(tg) is the recursive weight fraction assigned to | |
392 | * this group. | |
393 | */ | |
394 | unsigned long h_load; | |
c09595f6 | 395 | |
c8cba857 PZ |
396 | /* |
397 | * this cpu's part of tg->shares | |
398 | */ | |
399 | unsigned long shares; | |
f1d239f7 PZ |
400 | |
401 | /* | |
402 | * load.weight at the time we set shares | |
403 | */ | |
404 | unsigned long rq_weight; | |
c09595f6 | 405 | #endif |
6aa645ea IM |
406 | #endif |
407 | }; | |
1da177e4 | 408 | |
6aa645ea IM |
409 | /* Real-Time classes' related field in a runqueue: */ |
410 | struct rt_rq { | |
411 | struct rt_prio_array active; | |
63489e45 | 412 | unsigned long rt_nr_running; |
052f1dc7 | 413 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
414 | struct { |
415 | int curr; /* highest queued rt task prio */ | |
398a153b | 416 | #ifdef CONFIG_SMP |
e864c499 | 417 | int next; /* next highest */ |
398a153b | 418 | #endif |
e864c499 | 419 | } highest_prio; |
6f505b16 | 420 | #endif |
fa85ae24 | 421 | #ifdef CONFIG_SMP |
73fe6aae | 422 | unsigned long rt_nr_migratory; |
a1ba4d8b | 423 | unsigned long rt_nr_total; |
a22d7fc1 | 424 | int overloaded; |
917b627d | 425 | struct plist_head pushable_tasks; |
fa85ae24 | 426 | #endif |
6f505b16 | 427 | int rt_throttled; |
fa85ae24 | 428 | u64 rt_time; |
ac086bc2 | 429 | u64 rt_runtime; |
ea736ed5 | 430 | /* Nests inside the rq lock: */ |
0986b11b | 431 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 432 | |
052f1dc7 | 433 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
434 | unsigned long rt_nr_boosted; |
435 | ||
6f505b16 PZ |
436 | struct rq *rq; |
437 | struct list_head leaf_rt_rq_list; | |
438 | struct task_group *tg; | |
6f505b16 | 439 | #endif |
6aa645ea IM |
440 | }; |
441 | ||
57d885fe GH |
442 | #ifdef CONFIG_SMP |
443 | ||
444 | /* | |
445 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
446 | * variables. Each exclusive cpuset essentially defines an island domain by |
447 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
448 | * exclusive cpuset is created, we also create and attach a new root-domain |
449 | * object. | |
450 | * | |
57d885fe GH |
451 | */ |
452 | struct root_domain { | |
453 | atomic_t refcount; | |
c6c4927b RR |
454 | cpumask_var_t span; |
455 | cpumask_var_t online; | |
637f5085 | 456 | |
0eab9146 | 457 | /* |
637f5085 GH |
458 | * The "RT overload" flag: it gets set if a CPU has more than |
459 | * one runnable RT task. | |
460 | */ | |
c6c4927b | 461 | cpumask_var_t rto_mask; |
0eab9146 | 462 | atomic_t rto_count; |
6e0534f2 GH |
463 | #ifdef CONFIG_SMP |
464 | struct cpupri cpupri; | |
465 | #endif | |
57d885fe GH |
466 | }; |
467 | ||
dc938520 GH |
468 | /* |
469 | * By default the system creates a single root-domain with all cpus as | |
470 | * members (mimicking the global state we have today). | |
471 | */ | |
57d885fe GH |
472 | static struct root_domain def_root_domain; |
473 | ||
474 | #endif | |
475 | ||
1da177e4 LT |
476 | /* |
477 | * This is the main, per-CPU runqueue data structure. | |
478 | * | |
479 | * Locking rule: those places that want to lock multiple runqueues | |
480 | * (such as the load balancing or the thread migration code), lock | |
481 | * acquire operations must be ordered by ascending &runqueue. | |
482 | */ | |
70b97a7f | 483 | struct rq { |
d8016491 | 484 | /* runqueue lock: */ |
05fa785c | 485 | raw_spinlock_t lock; |
1da177e4 LT |
486 | |
487 | /* | |
488 | * nr_running and cpu_load should be in the same cacheline because | |
489 | * remote CPUs use both these fields when doing load calculation. | |
490 | */ | |
491 | unsigned long nr_running; | |
6aa645ea IM |
492 | #define CPU_LOAD_IDX_MAX 5 |
493 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 494 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 495 | u64 nohz_stamp; |
46cb4b7c SS |
496 | unsigned char in_nohz_recently; |
497 | #endif | |
a64692a3 MG |
498 | unsigned int skip_clock_update; |
499 | ||
d8016491 IM |
500 | /* capture load from *all* tasks on this cpu: */ |
501 | struct load_weight load; | |
6aa645ea IM |
502 | unsigned long nr_load_updates; |
503 | u64 nr_switches; | |
504 | ||
505 | struct cfs_rq cfs; | |
6f505b16 | 506 | struct rt_rq rt; |
6f505b16 | 507 | |
6aa645ea | 508 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
509 | /* list of leaf cfs_rq on this cpu: */ |
510 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
511 | #endif |
512 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 513 | struct list_head leaf_rt_rq_list; |
1da177e4 | 514 | #endif |
1da177e4 LT |
515 | |
516 | /* | |
517 | * This is part of a global counter where only the total sum | |
518 | * over all CPUs matters. A task can increase this counter on | |
519 | * one CPU and if it got migrated afterwards it may decrease | |
520 | * it on another CPU. Always updated under the runqueue lock: | |
521 | */ | |
522 | unsigned long nr_uninterruptible; | |
523 | ||
36c8b586 | 524 | struct task_struct *curr, *idle; |
c9819f45 | 525 | unsigned long next_balance; |
1da177e4 | 526 | struct mm_struct *prev_mm; |
6aa645ea | 527 | |
3e51f33f | 528 | u64 clock; |
6aa645ea | 529 | |
1da177e4 LT |
530 | atomic_t nr_iowait; |
531 | ||
532 | #ifdef CONFIG_SMP | |
0eab9146 | 533 | struct root_domain *rd; |
1da177e4 LT |
534 | struct sched_domain *sd; |
535 | ||
a0a522ce | 536 | unsigned char idle_at_tick; |
1da177e4 | 537 | /* For active balancing */ |
3f029d3c | 538 | int post_schedule; |
1da177e4 LT |
539 | int active_balance; |
540 | int push_cpu; | |
d8016491 IM |
541 | /* cpu of this runqueue: */ |
542 | int cpu; | |
1f11eb6a | 543 | int online; |
1da177e4 | 544 | |
a8a51d5e | 545 | unsigned long avg_load_per_task; |
1da177e4 | 546 | |
36c8b586 | 547 | struct task_struct *migration_thread; |
1da177e4 | 548 | struct list_head migration_queue; |
e9e9250b PZ |
549 | |
550 | u64 rt_avg; | |
551 | u64 age_stamp; | |
1b9508f6 MG |
552 | u64 idle_stamp; |
553 | u64 avg_idle; | |
1da177e4 LT |
554 | #endif |
555 | ||
dce48a84 TG |
556 | /* calc_load related fields */ |
557 | unsigned long calc_load_update; | |
558 | long calc_load_active; | |
559 | ||
8f4d37ec | 560 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
561 | #ifdef CONFIG_SMP |
562 | int hrtick_csd_pending; | |
563 | struct call_single_data hrtick_csd; | |
564 | #endif | |
8f4d37ec PZ |
565 | struct hrtimer hrtick_timer; |
566 | #endif | |
567 | ||
1da177e4 LT |
568 | #ifdef CONFIG_SCHEDSTATS |
569 | /* latency stats */ | |
570 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
571 | unsigned long long rq_cpu_time; |
572 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
573 | |
574 | /* sys_sched_yield() stats */ | |
480b9434 | 575 | unsigned int yld_count; |
1da177e4 LT |
576 | |
577 | /* schedule() stats */ | |
480b9434 KC |
578 | unsigned int sched_switch; |
579 | unsigned int sched_count; | |
580 | unsigned int sched_goidle; | |
1da177e4 LT |
581 | |
582 | /* try_to_wake_up() stats */ | |
480b9434 KC |
583 | unsigned int ttwu_count; |
584 | unsigned int ttwu_local; | |
b8efb561 IM |
585 | |
586 | /* BKL stats */ | |
480b9434 | 587 | unsigned int bkl_count; |
1da177e4 LT |
588 | #endif |
589 | }; | |
590 | ||
f34e3b61 | 591 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 592 | |
7d478721 PZ |
593 | static inline |
594 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 595 | { |
7d478721 | 596 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
a64692a3 MG |
597 | |
598 | /* | |
599 | * A queue event has occurred, and we're going to schedule. In | |
600 | * this case, we can save a useless back to back clock update. | |
601 | */ | |
602 | if (test_tsk_need_resched(p)) | |
603 | rq->skip_clock_update = 1; | |
dd41f596 IM |
604 | } |
605 | ||
0a2966b4 CL |
606 | static inline int cpu_of(struct rq *rq) |
607 | { | |
608 | #ifdef CONFIG_SMP | |
609 | return rq->cpu; | |
610 | #else | |
611 | return 0; | |
612 | #endif | |
613 | } | |
614 | ||
497f0ab3 | 615 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
616 | rcu_dereference_check((p), \ |
617 | rcu_read_lock_sched_held() || \ | |
618 | lockdep_is_held(&sched_domains_mutex)) | |
619 | ||
674311d5 NP |
620 | /* |
621 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 622 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
623 | * |
624 | * The domain tree of any CPU may only be accessed from within | |
625 | * preempt-disabled sections. | |
626 | */ | |
48f24c4d | 627 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 628 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
629 | |
630 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
631 | #define this_rq() (&__get_cpu_var(runqueues)) | |
632 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
633 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 634 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 635 | |
aa9c4c0f | 636 | inline void update_rq_clock(struct rq *rq) |
3e51f33f | 637 | { |
a64692a3 MG |
638 | if (!rq->skip_clock_update) |
639 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
3e51f33f PZ |
640 | } |
641 | ||
bf5c91ba IM |
642 | /* |
643 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
644 | */ | |
645 | #ifdef CONFIG_SCHED_DEBUG | |
646 | # define const_debug __read_mostly | |
647 | #else | |
648 | # define const_debug static const | |
649 | #endif | |
650 | ||
017730c1 IM |
651 | /** |
652 | * runqueue_is_locked | |
e17b38bf | 653 | * @cpu: the processor in question. |
017730c1 IM |
654 | * |
655 | * Returns true if the current cpu runqueue is locked. | |
656 | * This interface allows printk to be called with the runqueue lock | |
657 | * held and know whether or not it is OK to wake up the klogd. | |
658 | */ | |
89f19f04 | 659 | int runqueue_is_locked(int cpu) |
017730c1 | 660 | { |
05fa785c | 661 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
662 | } |
663 | ||
bf5c91ba IM |
664 | /* |
665 | * Debugging: various feature bits | |
666 | */ | |
f00b45c1 PZ |
667 | |
668 | #define SCHED_FEAT(name, enabled) \ | |
669 | __SCHED_FEAT_##name , | |
670 | ||
bf5c91ba | 671 | enum { |
f00b45c1 | 672 | #include "sched_features.h" |
bf5c91ba IM |
673 | }; |
674 | ||
f00b45c1 PZ |
675 | #undef SCHED_FEAT |
676 | ||
677 | #define SCHED_FEAT(name, enabled) \ | |
678 | (1UL << __SCHED_FEAT_##name) * enabled | | |
679 | ||
bf5c91ba | 680 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
681 | #include "sched_features.h" |
682 | 0; | |
683 | ||
684 | #undef SCHED_FEAT | |
685 | ||
686 | #ifdef CONFIG_SCHED_DEBUG | |
687 | #define SCHED_FEAT(name, enabled) \ | |
688 | #name , | |
689 | ||
983ed7a6 | 690 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
691 | #include "sched_features.h" |
692 | NULL | |
693 | }; | |
694 | ||
695 | #undef SCHED_FEAT | |
696 | ||
34f3a814 | 697 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 698 | { |
f00b45c1 PZ |
699 | int i; |
700 | ||
701 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
702 | if (!(sysctl_sched_features & (1UL << i))) |
703 | seq_puts(m, "NO_"); | |
704 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 705 | } |
34f3a814 | 706 | seq_puts(m, "\n"); |
f00b45c1 | 707 | |
34f3a814 | 708 | return 0; |
f00b45c1 PZ |
709 | } |
710 | ||
711 | static ssize_t | |
712 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
713 | size_t cnt, loff_t *ppos) | |
714 | { | |
715 | char buf[64]; | |
716 | char *cmp = buf; | |
717 | int neg = 0; | |
718 | int i; | |
719 | ||
720 | if (cnt > 63) | |
721 | cnt = 63; | |
722 | ||
723 | if (copy_from_user(&buf, ubuf, cnt)) | |
724 | return -EFAULT; | |
725 | ||
726 | buf[cnt] = 0; | |
727 | ||
c24b7c52 | 728 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
729 | neg = 1; |
730 | cmp += 3; | |
731 | } | |
732 | ||
733 | for (i = 0; sched_feat_names[i]; i++) { | |
734 | int len = strlen(sched_feat_names[i]); | |
735 | ||
736 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
737 | if (neg) | |
738 | sysctl_sched_features &= ~(1UL << i); | |
739 | else | |
740 | sysctl_sched_features |= (1UL << i); | |
741 | break; | |
742 | } | |
743 | } | |
744 | ||
745 | if (!sched_feat_names[i]) | |
746 | return -EINVAL; | |
747 | ||
42994724 | 748 | *ppos += cnt; |
f00b45c1 PZ |
749 | |
750 | return cnt; | |
751 | } | |
752 | ||
34f3a814 LZ |
753 | static int sched_feat_open(struct inode *inode, struct file *filp) |
754 | { | |
755 | return single_open(filp, sched_feat_show, NULL); | |
756 | } | |
757 | ||
828c0950 | 758 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
759 | .open = sched_feat_open, |
760 | .write = sched_feat_write, | |
761 | .read = seq_read, | |
762 | .llseek = seq_lseek, | |
763 | .release = single_release, | |
f00b45c1 PZ |
764 | }; |
765 | ||
766 | static __init int sched_init_debug(void) | |
767 | { | |
f00b45c1 PZ |
768 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
769 | &sched_feat_fops); | |
770 | ||
771 | return 0; | |
772 | } | |
773 | late_initcall(sched_init_debug); | |
774 | ||
775 | #endif | |
776 | ||
777 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 778 | |
b82d9fdd PZ |
779 | /* |
780 | * Number of tasks to iterate in a single balance run. | |
781 | * Limited because this is done with IRQs disabled. | |
782 | */ | |
783 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
784 | ||
2398f2c6 PZ |
785 | /* |
786 | * ratelimit for updating the group shares. | |
55cd5340 | 787 | * default: 0.25ms |
2398f2c6 | 788 | */ |
55cd5340 | 789 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
0bcdcf28 | 790 | unsigned int normalized_sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 791 | |
ffda12a1 PZ |
792 | /* |
793 | * Inject some fuzzyness into changing the per-cpu group shares | |
794 | * this avoids remote rq-locks at the expense of fairness. | |
795 | * default: 4 | |
796 | */ | |
797 | unsigned int sysctl_sched_shares_thresh = 4; | |
798 | ||
e9e9250b PZ |
799 | /* |
800 | * period over which we average the RT time consumption, measured | |
801 | * in ms. | |
802 | * | |
803 | * default: 1s | |
804 | */ | |
805 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
806 | ||
fa85ae24 | 807 | /* |
9f0c1e56 | 808 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
809 | * default: 1s |
810 | */ | |
9f0c1e56 | 811 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 812 | |
6892b75e IM |
813 | static __read_mostly int scheduler_running; |
814 | ||
9f0c1e56 PZ |
815 | /* |
816 | * part of the period that we allow rt tasks to run in us. | |
817 | * default: 0.95s | |
818 | */ | |
819 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 820 | |
d0b27fa7 PZ |
821 | static inline u64 global_rt_period(void) |
822 | { | |
823 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
824 | } | |
825 | ||
826 | static inline u64 global_rt_runtime(void) | |
827 | { | |
e26873bb | 828 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
829 | return RUNTIME_INF; |
830 | ||
831 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
832 | } | |
fa85ae24 | 833 | |
1da177e4 | 834 | #ifndef prepare_arch_switch |
4866cde0 NP |
835 | # define prepare_arch_switch(next) do { } while (0) |
836 | #endif | |
837 | #ifndef finish_arch_switch | |
838 | # define finish_arch_switch(prev) do { } while (0) | |
839 | #endif | |
840 | ||
051a1d1a DA |
841 | static inline int task_current(struct rq *rq, struct task_struct *p) |
842 | { | |
843 | return rq->curr == p; | |
844 | } | |
845 | ||
4866cde0 | 846 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 847 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 848 | { |
051a1d1a | 849 | return task_current(rq, p); |
4866cde0 NP |
850 | } |
851 | ||
70b97a7f | 852 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
853 | { |
854 | } | |
855 | ||
70b97a7f | 856 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 857 | { |
da04c035 IM |
858 | #ifdef CONFIG_DEBUG_SPINLOCK |
859 | /* this is a valid case when another task releases the spinlock */ | |
860 | rq->lock.owner = current; | |
861 | #endif | |
8a25d5de IM |
862 | /* |
863 | * If we are tracking spinlock dependencies then we have to | |
864 | * fix up the runqueue lock - which gets 'carried over' from | |
865 | * prev into current: | |
866 | */ | |
867 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
868 | ||
05fa785c | 869 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
870 | } |
871 | ||
872 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 873 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
874 | { |
875 | #ifdef CONFIG_SMP | |
876 | return p->oncpu; | |
877 | #else | |
051a1d1a | 878 | return task_current(rq, p); |
4866cde0 NP |
879 | #endif |
880 | } | |
881 | ||
70b97a7f | 882 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
883 | { |
884 | #ifdef CONFIG_SMP | |
885 | /* | |
886 | * We can optimise this out completely for !SMP, because the | |
887 | * SMP rebalancing from interrupt is the only thing that cares | |
888 | * here. | |
889 | */ | |
890 | next->oncpu = 1; | |
891 | #endif | |
892 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 893 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 894 | #else |
05fa785c | 895 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
896 | #endif |
897 | } | |
898 | ||
70b97a7f | 899 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
900 | { |
901 | #ifdef CONFIG_SMP | |
902 | /* | |
903 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
904 | * We must ensure this doesn't happen until the switch is completely | |
905 | * finished. | |
906 | */ | |
907 | smp_wmb(); | |
908 | prev->oncpu = 0; | |
909 | #endif | |
910 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
911 | local_irq_enable(); | |
1da177e4 | 912 | #endif |
4866cde0 NP |
913 | } |
914 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 915 | |
0970d299 PZ |
916 | /* |
917 | * Check whether the task is waking, we use this to synchronize against | |
918 | * ttwu() so that task_cpu() reports a stable number. | |
919 | * | |
920 | * We need to make an exception for PF_STARTING tasks because the fork | |
921 | * path might require task_rq_lock() to work, eg. it can call | |
922 | * set_cpus_allowed_ptr() from the cpuset clone_ns code. | |
923 | */ | |
924 | static inline int task_is_waking(struct task_struct *p) | |
925 | { | |
926 | return unlikely((p->state == TASK_WAKING) && !(p->flags & PF_STARTING)); | |
927 | } | |
928 | ||
b29739f9 IM |
929 | /* |
930 | * __task_rq_lock - lock the runqueue a given task resides on. | |
931 | * Must be called interrupts disabled. | |
932 | */ | |
70b97a7f | 933 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
934 | __acquires(rq->lock) |
935 | { | |
0970d299 PZ |
936 | struct rq *rq; |
937 | ||
3a5c359a | 938 | for (;;) { |
0970d299 PZ |
939 | while (task_is_waking(p)) |
940 | cpu_relax(); | |
941 | rq = task_rq(p); | |
05fa785c | 942 | raw_spin_lock(&rq->lock); |
0970d299 | 943 | if (likely(rq == task_rq(p) && !task_is_waking(p))) |
3a5c359a | 944 | return rq; |
05fa785c | 945 | raw_spin_unlock(&rq->lock); |
b29739f9 | 946 | } |
b29739f9 IM |
947 | } |
948 | ||
1da177e4 LT |
949 | /* |
950 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 951 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
952 | * explicitly disabling preemption. |
953 | */ | |
70b97a7f | 954 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
955 | __acquires(rq->lock) |
956 | { | |
70b97a7f | 957 | struct rq *rq; |
1da177e4 | 958 | |
3a5c359a | 959 | for (;;) { |
0970d299 PZ |
960 | while (task_is_waking(p)) |
961 | cpu_relax(); | |
3a5c359a AK |
962 | local_irq_save(*flags); |
963 | rq = task_rq(p); | |
05fa785c | 964 | raw_spin_lock(&rq->lock); |
0970d299 | 965 | if (likely(rq == task_rq(p) && !task_is_waking(p))) |
3a5c359a | 966 | return rq; |
05fa785c | 967 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 968 | } |
1da177e4 LT |
969 | } |
970 | ||
ad474cac ON |
971 | void task_rq_unlock_wait(struct task_struct *p) |
972 | { | |
973 | struct rq *rq = task_rq(p); | |
974 | ||
975 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
05fa785c | 976 | raw_spin_unlock_wait(&rq->lock); |
ad474cac ON |
977 | } |
978 | ||
a9957449 | 979 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
980 | __releases(rq->lock) |
981 | { | |
05fa785c | 982 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
983 | } |
984 | ||
70b97a7f | 985 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
986 | __releases(rq->lock) |
987 | { | |
05fa785c | 988 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
989 | } |
990 | ||
1da177e4 | 991 | /* |
cc2a73b5 | 992 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 993 | */ |
a9957449 | 994 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
995 | __acquires(rq->lock) |
996 | { | |
70b97a7f | 997 | struct rq *rq; |
1da177e4 LT |
998 | |
999 | local_irq_disable(); | |
1000 | rq = this_rq(); | |
05fa785c | 1001 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
1002 | |
1003 | return rq; | |
1004 | } | |
1005 | ||
8f4d37ec PZ |
1006 | #ifdef CONFIG_SCHED_HRTICK |
1007 | /* | |
1008 | * Use HR-timers to deliver accurate preemption points. | |
1009 | * | |
1010 | * Its all a bit involved since we cannot program an hrt while holding the | |
1011 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1012 | * reschedule event. | |
1013 | * | |
1014 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1015 | * rq->lock. | |
1016 | */ | |
8f4d37ec PZ |
1017 | |
1018 | /* | |
1019 | * Use hrtick when: | |
1020 | * - enabled by features | |
1021 | * - hrtimer is actually high res | |
1022 | */ | |
1023 | static inline int hrtick_enabled(struct rq *rq) | |
1024 | { | |
1025 | if (!sched_feat(HRTICK)) | |
1026 | return 0; | |
ba42059f | 1027 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1028 | return 0; |
8f4d37ec PZ |
1029 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1030 | } | |
1031 | ||
8f4d37ec PZ |
1032 | static void hrtick_clear(struct rq *rq) |
1033 | { | |
1034 | if (hrtimer_active(&rq->hrtick_timer)) | |
1035 | hrtimer_cancel(&rq->hrtick_timer); | |
1036 | } | |
1037 | ||
8f4d37ec PZ |
1038 | /* |
1039 | * High-resolution timer tick. | |
1040 | * Runs from hardirq context with interrupts disabled. | |
1041 | */ | |
1042 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1043 | { | |
1044 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1045 | ||
1046 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1047 | ||
05fa785c | 1048 | raw_spin_lock(&rq->lock); |
3e51f33f | 1049 | update_rq_clock(rq); |
8f4d37ec | 1050 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1051 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1052 | |
1053 | return HRTIMER_NORESTART; | |
1054 | } | |
1055 | ||
95e904c7 | 1056 | #ifdef CONFIG_SMP |
31656519 PZ |
1057 | /* |
1058 | * called from hardirq (IPI) context | |
1059 | */ | |
1060 | static void __hrtick_start(void *arg) | |
b328ca18 | 1061 | { |
31656519 | 1062 | struct rq *rq = arg; |
b328ca18 | 1063 | |
05fa785c | 1064 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1065 | hrtimer_restart(&rq->hrtick_timer); |
1066 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1067 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1068 | } |
1069 | ||
31656519 PZ |
1070 | /* |
1071 | * Called to set the hrtick timer state. | |
1072 | * | |
1073 | * called with rq->lock held and irqs disabled | |
1074 | */ | |
1075 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1076 | { |
31656519 PZ |
1077 | struct hrtimer *timer = &rq->hrtick_timer; |
1078 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1079 | |
cc584b21 | 1080 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1081 | |
1082 | if (rq == this_rq()) { | |
1083 | hrtimer_restart(timer); | |
1084 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1085 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1086 | rq->hrtick_csd_pending = 1; |
1087 | } | |
b328ca18 PZ |
1088 | } |
1089 | ||
1090 | static int | |
1091 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1092 | { | |
1093 | int cpu = (int)(long)hcpu; | |
1094 | ||
1095 | switch (action) { | |
1096 | case CPU_UP_CANCELED: | |
1097 | case CPU_UP_CANCELED_FROZEN: | |
1098 | case CPU_DOWN_PREPARE: | |
1099 | case CPU_DOWN_PREPARE_FROZEN: | |
1100 | case CPU_DEAD: | |
1101 | case CPU_DEAD_FROZEN: | |
31656519 | 1102 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1103 | return NOTIFY_OK; |
1104 | } | |
1105 | ||
1106 | return NOTIFY_DONE; | |
1107 | } | |
1108 | ||
fa748203 | 1109 | static __init void init_hrtick(void) |
b328ca18 PZ |
1110 | { |
1111 | hotcpu_notifier(hotplug_hrtick, 0); | |
1112 | } | |
31656519 PZ |
1113 | #else |
1114 | /* | |
1115 | * Called to set the hrtick timer state. | |
1116 | * | |
1117 | * called with rq->lock held and irqs disabled | |
1118 | */ | |
1119 | static void hrtick_start(struct rq *rq, u64 delay) | |
1120 | { | |
7f1e2ca9 | 1121 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1122 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1123 | } |
b328ca18 | 1124 | |
006c75f1 | 1125 | static inline void init_hrtick(void) |
8f4d37ec | 1126 | { |
8f4d37ec | 1127 | } |
31656519 | 1128 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1129 | |
31656519 | 1130 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1131 | { |
31656519 PZ |
1132 | #ifdef CONFIG_SMP |
1133 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1134 | |
31656519 PZ |
1135 | rq->hrtick_csd.flags = 0; |
1136 | rq->hrtick_csd.func = __hrtick_start; | |
1137 | rq->hrtick_csd.info = rq; | |
1138 | #endif | |
8f4d37ec | 1139 | |
31656519 PZ |
1140 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1141 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1142 | } |
006c75f1 | 1143 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1144 | static inline void hrtick_clear(struct rq *rq) |
1145 | { | |
1146 | } | |
1147 | ||
8f4d37ec PZ |
1148 | static inline void init_rq_hrtick(struct rq *rq) |
1149 | { | |
1150 | } | |
1151 | ||
b328ca18 PZ |
1152 | static inline void init_hrtick(void) |
1153 | { | |
1154 | } | |
006c75f1 | 1155 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1156 | |
c24d20db IM |
1157 | /* |
1158 | * resched_task - mark a task 'to be rescheduled now'. | |
1159 | * | |
1160 | * On UP this means the setting of the need_resched flag, on SMP it | |
1161 | * might also involve a cross-CPU call to trigger the scheduler on | |
1162 | * the target CPU. | |
1163 | */ | |
1164 | #ifdef CONFIG_SMP | |
1165 | ||
1166 | #ifndef tsk_is_polling | |
1167 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1168 | #endif | |
1169 | ||
31656519 | 1170 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1171 | { |
1172 | int cpu; | |
1173 | ||
05fa785c | 1174 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1175 | |
5ed0cec0 | 1176 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1177 | return; |
1178 | ||
5ed0cec0 | 1179 | set_tsk_need_resched(p); |
c24d20db IM |
1180 | |
1181 | cpu = task_cpu(p); | |
1182 | if (cpu == smp_processor_id()) | |
1183 | return; | |
1184 | ||
1185 | /* NEED_RESCHED must be visible before we test polling */ | |
1186 | smp_mb(); | |
1187 | if (!tsk_is_polling(p)) | |
1188 | smp_send_reschedule(cpu); | |
1189 | } | |
1190 | ||
1191 | static void resched_cpu(int cpu) | |
1192 | { | |
1193 | struct rq *rq = cpu_rq(cpu); | |
1194 | unsigned long flags; | |
1195 | ||
05fa785c | 1196 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1197 | return; |
1198 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1199 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1200 | } |
06d8308c TG |
1201 | |
1202 | #ifdef CONFIG_NO_HZ | |
1203 | /* | |
1204 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1205 | * idle CPU then this timer might expire before the next timer event | |
1206 | * which is scheduled to wake up that CPU. In case of a completely | |
1207 | * idle system the next event might even be infinite time into the | |
1208 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1209 | * leaves the inner idle loop so the newly added timer is taken into | |
1210 | * account when the CPU goes back to idle and evaluates the timer | |
1211 | * wheel for the next timer event. | |
1212 | */ | |
1213 | void wake_up_idle_cpu(int cpu) | |
1214 | { | |
1215 | struct rq *rq = cpu_rq(cpu); | |
1216 | ||
1217 | if (cpu == smp_processor_id()) | |
1218 | return; | |
1219 | ||
1220 | /* | |
1221 | * This is safe, as this function is called with the timer | |
1222 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1223 | * to idle and has not yet set rq->curr to idle then it will | |
1224 | * be serialized on the timer wheel base lock and take the new | |
1225 | * timer into account automatically. | |
1226 | */ | |
1227 | if (rq->curr != rq->idle) | |
1228 | return; | |
1229 | ||
1230 | /* | |
1231 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1232 | * lockless. The worst case is that the other CPU runs the | |
1233 | * idle task through an additional NOOP schedule() | |
1234 | */ | |
5ed0cec0 | 1235 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1236 | |
1237 | /* NEED_RESCHED must be visible before we test polling */ | |
1238 | smp_mb(); | |
1239 | if (!tsk_is_polling(rq->idle)) | |
1240 | smp_send_reschedule(cpu); | |
1241 | } | |
39c0cbe2 MG |
1242 | |
1243 | int nohz_ratelimit(int cpu) | |
1244 | { | |
1245 | struct rq *rq = cpu_rq(cpu); | |
1246 | u64 diff = rq->clock - rq->nohz_stamp; | |
1247 | ||
1248 | rq->nohz_stamp = rq->clock; | |
1249 | ||
1250 | return diff < (NSEC_PER_SEC / HZ) >> 1; | |
1251 | } | |
1252 | ||
6d6bc0ad | 1253 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1254 | |
e9e9250b PZ |
1255 | static u64 sched_avg_period(void) |
1256 | { | |
1257 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1258 | } | |
1259 | ||
1260 | static void sched_avg_update(struct rq *rq) | |
1261 | { | |
1262 | s64 period = sched_avg_period(); | |
1263 | ||
1264 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1265 | rq->age_stamp += period; | |
1266 | rq->rt_avg /= 2; | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1271 | { | |
1272 | rq->rt_avg += rt_delta; | |
1273 | sched_avg_update(rq); | |
1274 | } | |
1275 | ||
6d6bc0ad | 1276 | #else /* !CONFIG_SMP */ |
31656519 | 1277 | static void resched_task(struct task_struct *p) |
c24d20db | 1278 | { |
05fa785c | 1279 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1280 | set_tsk_need_resched(p); |
c24d20db | 1281 | } |
e9e9250b PZ |
1282 | |
1283 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1284 | { | |
1285 | } | |
6d6bc0ad | 1286 | #endif /* CONFIG_SMP */ |
c24d20db | 1287 | |
45bf76df IM |
1288 | #if BITS_PER_LONG == 32 |
1289 | # define WMULT_CONST (~0UL) | |
1290 | #else | |
1291 | # define WMULT_CONST (1UL << 32) | |
1292 | #endif | |
1293 | ||
1294 | #define WMULT_SHIFT 32 | |
1295 | ||
194081eb IM |
1296 | /* |
1297 | * Shift right and round: | |
1298 | */ | |
cf2ab469 | 1299 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1300 | |
a7be37ac PZ |
1301 | /* |
1302 | * delta *= weight / lw | |
1303 | */ | |
cb1c4fc9 | 1304 | static unsigned long |
45bf76df IM |
1305 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1306 | struct load_weight *lw) | |
1307 | { | |
1308 | u64 tmp; | |
1309 | ||
7a232e03 LJ |
1310 | if (!lw->inv_weight) { |
1311 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1312 | lw->inv_weight = 1; | |
1313 | else | |
1314 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1315 | / (lw->weight+1); | |
1316 | } | |
45bf76df IM |
1317 | |
1318 | tmp = (u64)delta_exec * weight; | |
1319 | /* | |
1320 | * Check whether we'd overflow the 64-bit multiplication: | |
1321 | */ | |
194081eb | 1322 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1323 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1324 | WMULT_SHIFT/2); |
1325 | else | |
cf2ab469 | 1326 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1327 | |
ecf691da | 1328 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1329 | } |
1330 | ||
1091985b | 1331 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1332 | { |
1333 | lw->weight += inc; | |
e89996ae | 1334 | lw->inv_weight = 0; |
45bf76df IM |
1335 | } |
1336 | ||
1091985b | 1337 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1338 | { |
1339 | lw->weight -= dec; | |
e89996ae | 1340 | lw->inv_weight = 0; |
45bf76df IM |
1341 | } |
1342 | ||
2dd73a4f PW |
1343 | /* |
1344 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1345 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1346 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1347 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1348 | * scaled version of the new time slice allocation that they receive on time |
1349 | * slice expiry etc. | |
1350 | */ | |
1351 | ||
cce7ade8 PZ |
1352 | #define WEIGHT_IDLEPRIO 3 |
1353 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1354 | |
1355 | /* | |
1356 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1357 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1358 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1359 | * that remained on nice 0. | |
1360 | * | |
1361 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1362 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1363 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1364 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1365 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1366 | */ |
1367 | static const int prio_to_weight[40] = { | |
254753dc IM |
1368 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1369 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1370 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1371 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1372 | /* 0 */ 1024, 820, 655, 526, 423, | |
1373 | /* 5 */ 335, 272, 215, 172, 137, | |
1374 | /* 10 */ 110, 87, 70, 56, 45, | |
1375 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1376 | }; |
1377 | ||
5714d2de IM |
1378 | /* |
1379 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1380 | * | |
1381 | * In cases where the weight does not change often, we can use the | |
1382 | * precalculated inverse to speed up arithmetics by turning divisions | |
1383 | * into multiplications: | |
1384 | */ | |
dd41f596 | 1385 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1386 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1387 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1388 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1389 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1390 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1391 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1392 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1393 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1394 | }; |
2dd73a4f | 1395 | |
ef12fefa BR |
1396 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1397 | enum cpuacct_stat_index { | |
1398 | CPUACCT_STAT_USER, /* ... user mode */ | |
1399 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1400 | ||
1401 | CPUACCT_STAT_NSTATS, | |
1402 | }; | |
1403 | ||
d842de87 SV |
1404 | #ifdef CONFIG_CGROUP_CPUACCT |
1405 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1406 | static void cpuacct_update_stats(struct task_struct *tsk, |
1407 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1408 | #else |
1409 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1410 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1411 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1412 | #endif |
1413 | ||
18d95a28 PZ |
1414 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1415 | { | |
1416 | update_load_add(&rq->load, load); | |
1417 | } | |
1418 | ||
1419 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1420 | { | |
1421 | update_load_sub(&rq->load, load); | |
1422 | } | |
1423 | ||
7940ca36 | 1424 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1425 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1426 | |
1427 | /* | |
1428 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1429 | * leaving it for the final time. | |
1430 | */ | |
eb755805 | 1431 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1432 | { |
1433 | struct task_group *parent, *child; | |
eb755805 | 1434 | int ret; |
c09595f6 PZ |
1435 | |
1436 | rcu_read_lock(); | |
1437 | parent = &root_task_group; | |
1438 | down: | |
eb755805 PZ |
1439 | ret = (*down)(parent, data); |
1440 | if (ret) | |
1441 | goto out_unlock; | |
c09595f6 PZ |
1442 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1443 | parent = child; | |
1444 | goto down; | |
1445 | ||
1446 | up: | |
1447 | continue; | |
1448 | } | |
eb755805 PZ |
1449 | ret = (*up)(parent, data); |
1450 | if (ret) | |
1451 | goto out_unlock; | |
c09595f6 PZ |
1452 | |
1453 | child = parent; | |
1454 | parent = parent->parent; | |
1455 | if (parent) | |
1456 | goto up; | |
eb755805 | 1457 | out_unlock: |
c09595f6 | 1458 | rcu_read_unlock(); |
eb755805 PZ |
1459 | |
1460 | return ret; | |
c09595f6 PZ |
1461 | } |
1462 | ||
eb755805 PZ |
1463 | static int tg_nop(struct task_group *tg, void *data) |
1464 | { | |
1465 | return 0; | |
c09595f6 | 1466 | } |
eb755805 PZ |
1467 | #endif |
1468 | ||
1469 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1470 | /* Used instead of source_load when we know the type == 0 */ |
1471 | static unsigned long weighted_cpuload(const int cpu) | |
1472 | { | |
1473 | return cpu_rq(cpu)->load.weight; | |
1474 | } | |
1475 | ||
1476 | /* | |
1477 | * Return a low guess at the load of a migration-source cpu weighted | |
1478 | * according to the scheduling class and "nice" value. | |
1479 | * | |
1480 | * We want to under-estimate the load of migration sources, to | |
1481 | * balance conservatively. | |
1482 | */ | |
1483 | static unsigned long source_load(int cpu, int type) | |
1484 | { | |
1485 | struct rq *rq = cpu_rq(cpu); | |
1486 | unsigned long total = weighted_cpuload(cpu); | |
1487 | ||
1488 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1489 | return total; | |
1490 | ||
1491 | return min(rq->cpu_load[type-1], total); | |
1492 | } | |
1493 | ||
1494 | /* | |
1495 | * Return a high guess at the load of a migration-target cpu weighted | |
1496 | * according to the scheduling class and "nice" value. | |
1497 | */ | |
1498 | static unsigned long target_load(int cpu, int type) | |
1499 | { | |
1500 | struct rq *rq = cpu_rq(cpu); | |
1501 | unsigned long total = weighted_cpuload(cpu); | |
1502 | ||
1503 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1504 | return total; | |
1505 | ||
1506 | return max(rq->cpu_load[type-1], total); | |
1507 | } | |
1508 | ||
ae154be1 PZ |
1509 | static struct sched_group *group_of(int cpu) |
1510 | { | |
d11c563d | 1511 | struct sched_domain *sd = rcu_dereference_sched(cpu_rq(cpu)->sd); |
ae154be1 PZ |
1512 | |
1513 | if (!sd) | |
1514 | return NULL; | |
1515 | ||
1516 | return sd->groups; | |
1517 | } | |
1518 | ||
1519 | static unsigned long power_of(int cpu) | |
1520 | { | |
1521 | struct sched_group *group = group_of(cpu); | |
1522 | ||
1523 | if (!group) | |
1524 | return SCHED_LOAD_SCALE; | |
1525 | ||
1526 | return group->cpu_power; | |
1527 | } | |
1528 | ||
eb755805 PZ |
1529 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1530 | ||
1531 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1532 | { | |
1533 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1534 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1535 | |
4cd42620 SR |
1536 | if (nr_running) |
1537 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1538 | else |
1539 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1540 | |
1541 | return rq->avg_load_per_task; | |
1542 | } | |
1543 | ||
1544 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1545 | |
43cf38eb | 1546 | static __read_mostly unsigned long __percpu *update_shares_data; |
34d76c41 | 1547 | |
c09595f6 PZ |
1548 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1549 | ||
1550 | /* | |
1551 | * Calculate and set the cpu's group shares. | |
1552 | */ | |
34d76c41 PZ |
1553 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1554 | unsigned long sd_shares, | |
1555 | unsigned long sd_rq_weight, | |
4a6cc4bd | 1556 | unsigned long *usd_rq_weight) |
18d95a28 | 1557 | { |
34d76c41 | 1558 | unsigned long shares, rq_weight; |
a5004278 | 1559 | int boost = 0; |
c09595f6 | 1560 | |
4a6cc4bd | 1561 | rq_weight = usd_rq_weight[cpu]; |
a5004278 PZ |
1562 | if (!rq_weight) { |
1563 | boost = 1; | |
1564 | rq_weight = NICE_0_LOAD; | |
1565 | } | |
c8cba857 | 1566 | |
c09595f6 | 1567 | /* |
a8af7246 PZ |
1568 | * \Sum_j shares_j * rq_weight_i |
1569 | * shares_i = ----------------------------- | |
1570 | * \Sum_j rq_weight_j | |
c09595f6 | 1571 | */ |
ec4e0e2f | 1572 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1573 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1574 | |
ffda12a1 PZ |
1575 | if (abs(shares - tg->se[cpu]->load.weight) > |
1576 | sysctl_sched_shares_thresh) { | |
1577 | struct rq *rq = cpu_rq(cpu); | |
1578 | unsigned long flags; | |
c09595f6 | 1579 | |
05fa785c | 1580 | raw_spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1581 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1582 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 | 1583 | __set_se_shares(tg->se[cpu], shares); |
05fa785c | 1584 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
ffda12a1 | 1585 | } |
18d95a28 | 1586 | } |
c09595f6 PZ |
1587 | |
1588 | /* | |
c8cba857 PZ |
1589 | * Re-compute the task group their per cpu shares over the given domain. |
1590 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1591 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1592 | */ |
eb755805 | 1593 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1594 | { |
cd8ad40d | 1595 | unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0; |
4a6cc4bd | 1596 | unsigned long *usd_rq_weight; |
eb755805 | 1597 | struct sched_domain *sd = data; |
34d76c41 | 1598 | unsigned long flags; |
c8cba857 | 1599 | int i; |
c09595f6 | 1600 | |
34d76c41 PZ |
1601 | if (!tg->se[0]) |
1602 | return 0; | |
1603 | ||
1604 | local_irq_save(flags); | |
4a6cc4bd | 1605 | usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id()); |
34d76c41 | 1606 | |
758b2cdc | 1607 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 | 1608 | weight = tg->cfs_rq[i]->load.weight; |
4a6cc4bd | 1609 | usd_rq_weight[i] = weight; |
34d76c41 | 1610 | |
cd8ad40d | 1611 | rq_weight += weight; |
ec4e0e2f KC |
1612 | /* |
1613 | * If there are currently no tasks on the cpu pretend there | |
1614 | * is one of average load so that when a new task gets to | |
1615 | * run here it will not get delayed by group starvation. | |
1616 | */ | |
ec4e0e2f KC |
1617 | if (!weight) |
1618 | weight = NICE_0_LOAD; | |
1619 | ||
cd8ad40d | 1620 | sum_weight += weight; |
c8cba857 | 1621 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1622 | } |
c09595f6 | 1623 | |
cd8ad40d PZ |
1624 | if (!rq_weight) |
1625 | rq_weight = sum_weight; | |
1626 | ||
c8cba857 PZ |
1627 | if ((!shares && rq_weight) || shares > tg->shares) |
1628 | shares = tg->shares; | |
1629 | ||
1630 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1631 | shares = tg->shares; | |
c09595f6 | 1632 | |
758b2cdc | 1633 | for_each_cpu(i, sched_domain_span(sd)) |
4a6cc4bd | 1634 | update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight); |
34d76c41 PZ |
1635 | |
1636 | local_irq_restore(flags); | |
eb755805 PZ |
1637 | |
1638 | return 0; | |
c09595f6 PZ |
1639 | } |
1640 | ||
1641 | /* | |
c8cba857 PZ |
1642 | * Compute the cpu's hierarchical load factor for each task group. |
1643 | * This needs to be done in a top-down fashion because the load of a child | |
1644 | * group is a fraction of its parents load. | |
c09595f6 | 1645 | */ |
eb755805 | 1646 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1647 | { |
c8cba857 | 1648 | unsigned long load; |
eb755805 | 1649 | long cpu = (long)data; |
c09595f6 | 1650 | |
c8cba857 PZ |
1651 | if (!tg->parent) { |
1652 | load = cpu_rq(cpu)->load.weight; | |
1653 | } else { | |
1654 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1655 | load *= tg->cfs_rq[cpu]->shares; | |
1656 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1657 | } | |
c09595f6 | 1658 | |
c8cba857 | 1659 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1660 | |
eb755805 | 1661 | return 0; |
c09595f6 PZ |
1662 | } |
1663 | ||
c8cba857 | 1664 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1665 | { |
e7097159 PZ |
1666 | s64 elapsed; |
1667 | u64 now; | |
1668 | ||
1669 | if (root_task_group_empty()) | |
1670 | return; | |
1671 | ||
1672 | now = cpu_clock(raw_smp_processor_id()); | |
1673 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1674 | |
1675 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1676 | sd->last_update = now; | |
eb755805 | 1677 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1678 | } |
4d8d595d PZ |
1679 | } |
1680 | ||
eb755805 | 1681 | static void update_h_load(long cpu) |
c09595f6 | 1682 | { |
e7097159 PZ |
1683 | if (root_task_group_empty()) |
1684 | return; | |
1685 | ||
eb755805 | 1686 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1687 | } |
1688 | ||
c09595f6 PZ |
1689 | #else |
1690 | ||
c8cba857 | 1691 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1692 | { |
1693 | } | |
1694 | ||
18d95a28 PZ |
1695 | #endif |
1696 | ||
8f45e2b5 GH |
1697 | #ifdef CONFIG_PREEMPT |
1698 | ||
b78bb868 PZ |
1699 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1700 | ||
70574a99 | 1701 | /* |
8f45e2b5 GH |
1702 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1703 | * way at the expense of forcing extra atomic operations in all | |
1704 | * invocations. This assures that the double_lock is acquired using the | |
1705 | * same underlying policy as the spinlock_t on this architecture, which | |
1706 | * reduces latency compared to the unfair variant below. However, it | |
1707 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1708 | */ |
8f45e2b5 GH |
1709 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1710 | __releases(this_rq->lock) | |
1711 | __acquires(busiest->lock) | |
1712 | __acquires(this_rq->lock) | |
1713 | { | |
05fa785c | 1714 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1715 | double_rq_lock(this_rq, busiest); |
1716 | ||
1717 | return 1; | |
1718 | } | |
1719 | ||
1720 | #else | |
1721 | /* | |
1722 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1723 | * latency by eliminating extra atomic operations when the locks are | |
1724 | * already in proper order on entry. This favors lower cpu-ids and will | |
1725 | * grant the double lock to lower cpus over higher ids under contention, | |
1726 | * regardless of entry order into the function. | |
1727 | */ | |
1728 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1729 | __releases(this_rq->lock) |
1730 | __acquires(busiest->lock) | |
1731 | __acquires(this_rq->lock) | |
1732 | { | |
1733 | int ret = 0; | |
1734 | ||
05fa785c | 1735 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1736 | if (busiest < this_rq) { |
05fa785c TG |
1737 | raw_spin_unlock(&this_rq->lock); |
1738 | raw_spin_lock(&busiest->lock); | |
1739 | raw_spin_lock_nested(&this_rq->lock, | |
1740 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1741 | ret = 1; |
1742 | } else | |
05fa785c TG |
1743 | raw_spin_lock_nested(&busiest->lock, |
1744 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1745 | } |
1746 | return ret; | |
1747 | } | |
1748 | ||
8f45e2b5 GH |
1749 | #endif /* CONFIG_PREEMPT */ |
1750 | ||
1751 | /* | |
1752 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1753 | */ | |
1754 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1755 | { | |
1756 | if (unlikely(!irqs_disabled())) { | |
1757 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1758 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1759 | BUG_ON(1); |
1760 | } | |
1761 | ||
1762 | return _double_lock_balance(this_rq, busiest); | |
1763 | } | |
1764 | ||
70574a99 AD |
1765 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1766 | __releases(busiest->lock) | |
1767 | { | |
05fa785c | 1768 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1769 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1770 | } | |
1e3c88bd PZ |
1771 | |
1772 | /* | |
1773 | * double_rq_lock - safely lock two runqueues | |
1774 | * | |
1775 | * Note this does not disable interrupts like task_rq_lock, | |
1776 | * you need to do so manually before calling. | |
1777 | */ | |
1778 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1779 | __acquires(rq1->lock) | |
1780 | __acquires(rq2->lock) | |
1781 | { | |
1782 | BUG_ON(!irqs_disabled()); | |
1783 | if (rq1 == rq2) { | |
1784 | raw_spin_lock(&rq1->lock); | |
1785 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1786 | } else { | |
1787 | if (rq1 < rq2) { | |
1788 | raw_spin_lock(&rq1->lock); | |
1789 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1790 | } else { | |
1791 | raw_spin_lock(&rq2->lock); | |
1792 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1793 | } | |
1794 | } | |
1e3c88bd PZ |
1795 | } |
1796 | ||
1797 | /* | |
1798 | * double_rq_unlock - safely unlock two runqueues | |
1799 | * | |
1800 | * Note this does not restore interrupts like task_rq_unlock, | |
1801 | * you need to do so manually after calling. | |
1802 | */ | |
1803 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1804 | __releases(rq1->lock) | |
1805 | __releases(rq2->lock) | |
1806 | { | |
1807 | raw_spin_unlock(&rq1->lock); | |
1808 | if (rq1 != rq2) | |
1809 | raw_spin_unlock(&rq2->lock); | |
1810 | else | |
1811 | __release(rq2->lock); | |
1812 | } | |
1813 | ||
18d95a28 PZ |
1814 | #endif |
1815 | ||
30432094 | 1816 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1817 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1818 | { | |
30432094 | 1819 | #ifdef CONFIG_SMP |
34e83e85 IM |
1820 | cfs_rq->shares = shares; |
1821 | #endif | |
1822 | } | |
30432094 | 1823 | #endif |
e7693a36 | 1824 | |
dce48a84 | 1825 | static void calc_load_account_active(struct rq *this_rq); |
0bcdcf28 | 1826 | static void update_sysctl(void); |
acb4a848 | 1827 | static int get_update_sysctl_factor(void); |
dce48a84 | 1828 | |
cd29fe6f PZ |
1829 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1830 | { | |
1831 | set_task_rq(p, cpu); | |
1832 | #ifdef CONFIG_SMP | |
1833 | /* | |
1834 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1835 | * successfuly executed on another CPU. We must ensure that updates of | |
1836 | * per-task data have been completed by this moment. | |
1837 | */ | |
1838 | smp_wmb(); | |
1839 | task_thread_info(p)->cpu = cpu; | |
1840 | #endif | |
1841 | } | |
dce48a84 | 1842 | |
1e3c88bd | 1843 | static const struct sched_class rt_sched_class; |
dd41f596 IM |
1844 | |
1845 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1846 | #define for_each_class(class) \ |
1847 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1848 | |
1e3c88bd PZ |
1849 | #include "sched_stats.h" |
1850 | ||
c09595f6 | 1851 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1852 | { |
1853 | rq->nr_running++; | |
9c217245 IM |
1854 | } |
1855 | ||
c09595f6 | 1856 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1857 | { |
1858 | rq->nr_running--; | |
9c217245 IM |
1859 | } |
1860 | ||
45bf76df IM |
1861 | static void set_load_weight(struct task_struct *p) |
1862 | { | |
1863 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1864 | p->se.load.weight = prio_to_weight[0] * 2; |
1865 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1866 | return; | |
1867 | } | |
45bf76df | 1868 | |
dd41f596 IM |
1869 | /* |
1870 | * SCHED_IDLE tasks get minimal weight: | |
1871 | */ | |
1872 | if (p->policy == SCHED_IDLE) { | |
1873 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1874 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1875 | return; | |
1876 | } | |
71f8bd46 | 1877 | |
dd41f596 IM |
1878 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1879 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1880 | } |
1881 | ||
2087a1ad GH |
1882 | static void update_avg(u64 *avg, u64 sample) |
1883 | { | |
1884 | s64 diff = sample - *avg; | |
1885 | *avg += diff >> 3; | |
1886 | } | |
1887 | ||
ea87bb78 TG |
1888 | static void |
1889 | enqueue_task(struct rq *rq, struct task_struct *p, int wakeup, bool head) | |
71f8bd46 | 1890 | { |
a64692a3 | 1891 | update_rq_clock(rq); |
dd41f596 | 1892 | sched_info_queued(p); |
ea87bb78 | 1893 | p->sched_class->enqueue_task(rq, p, wakeup, head); |
dd41f596 | 1894 | p->se.on_rq = 1; |
71f8bd46 IM |
1895 | } |
1896 | ||
69be72c1 | 1897 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1898 | { |
a64692a3 | 1899 | update_rq_clock(rq); |
46ac22ba | 1900 | sched_info_dequeued(p); |
f02231e5 | 1901 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1902 | p->se.on_rq = 0; |
71f8bd46 IM |
1903 | } |
1904 | ||
1e3c88bd PZ |
1905 | /* |
1906 | * activate_task - move a task to the runqueue. | |
1907 | */ | |
1908 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) | |
1909 | { | |
1910 | if (task_contributes_to_load(p)) | |
1911 | rq->nr_uninterruptible--; | |
1912 | ||
ea87bb78 | 1913 | enqueue_task(rq, p, wakeup, false); |
1e3c88bd PZ |
1914 | inc_nr_running(rq); |
1915 | } | |
1916 | ||
1917 | /* | |
1918 | * deactivate_task - remove a task from the runqueue. | |
1919 | */ | |
1920 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) | |
1921 | { | |
1922 | if (task_contributes_to_load(p)) | |
1923 | rq->nr_uninterruptible++; | |
1924 | ||
1925 | dequeue_task(rq, p, sleep); | |
1926 | dec_nr_running(rq); | |
1927 | } | |
1928 | ||
1929 | #include "sched_idletask.c" | |
1930 | #include "sched_fair.c" | |
1931 | #include "sched_rt.c" | |
1932 | #ifdef CONFIG_SCHED_DEBUG | |
1933 | # include "sched_debug.c" | |
1934 | #endif | |
1935 | ||
14531189 | 1936 | /* |
dd41f596 | 1937 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1938 | */ |
14531189 IM |
1939 | static inline int __normal_prio(struct task_struct *p) |
1940 | { | |
dd41f596 | 1941 | return p->static_prio; |
14531189 IM |
1942 | } |
1943 | ||
b29739f9 IM |
1944 | /* |
1945 | * Calculate the expected normal priority: i.e. priority | |
1946 | * without taking RT-inheritance into account. Might be | |
1947 | * boosted by interactivity modifiers. Changes upon fork, | |
1948 | * setprio syscalls, and whenever the interactivity | |
1949 | * estimator recalculates. | |
1950 | */ | |
36c8b586 | 1951 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1952 | { |
1953 | int prio; | |
1954 | ||
e05606d3 | 1955 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1956 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1957 | else | |
1958 | prio = __normal_prio(p); | |
1959 | return prio; | |
1960 | } | |
1961 | ||
1962 | /* | |
1963 | * Calculate the current priority, i.e. the priority | |
1964 | * taken into account by the scheduler. This value might | |
1965 | * be boosted by RT tasks, or might be boosted by | |
1966 | * interactivity modifiers. Will be RT if the task got | |
1967 | * RT-boosted. If not then it returns p->normal_prio. | |
1968 | */ | |
36c8b586 | 1969 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1970 | { |
1971 | p->normal_prio = normal_prio(p); | |
1972 | /* | |
1973 | * If we are RT tasks or we were boosted to RT priority, | |
1974 | * keep the priority unchanged. Otherwise, update priority | |
1975 | * to the normal priority: | |
1976 | */ | |
1977 | if (!rt_prio(p->prio)) | |
1978 | return p->normal_prio; | |
1979 | return p->prio; | |
1980 | } | |
1981 | ||
1da177e4 LT |
1982 | /** |
1983 | * task_curr - is this task currently executing on a CPU? | |
1984 | * @p: the task in question. | |
1985 | */ | |
36c8b586 | 1986 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1987 | { |
1988 | return cpu_curr(task_cpu(p)) == p; | |
1989 | } | |
1990 | ||
cb469845 SR |
1991 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1992 | const struct sched_class *prev_class, | |
1993 | int oldprio, int running) | |
1994 | { | |
1995 | if (prev_class != p->sched_class) { | |
1996 | if (prev_class->switched_from) | |
1997 | prev_class->switched_from(rq, p, running); | |
1998 | p->sched_class->switched_to(rq, p, running); | |
1999 | } else | |
2000 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
2001 | } | |
2002 | ||
1da177e4 | 2003 | #ifdef CONFIG_SMP |
cc367732 IM |
2004 | /* |
2005 | * Is this task likely cache-hot: | |
2006 | */ | |
e7693a36 | 2007 | static int |
cc367732 IM |
2008 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2009 | { | |
2010 | s64 delta; | |
2011 | ||
e6c8fba7 PZ |
2012 | if (p->sched_class != &fair_sched_class) |
2013 | return 0; | |
2014 | ||
f540a608 IM |
2015 | /* |
2016 | * Buddy candidates are cache hot: | |
2017 | */ | |
f685ceac | 2018 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2019 | (&p->se == cfs_rq_of(&p->se)->next || |
2020 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2021 | return 1; |
2022 | ||
6bc1665b IM |
2023 | if (sysctl_sched_migration_cost == -1) |
2024 | return 1; | |
2025 | if (sysctl_sched_migration_cost == 0) | |
2026 | return 0; | |
2027 | ||
cc367732 IM |
2028 | delta = now - p->se.exec_start; |
2029 | ||
2030 | return delta < (s64)sysctl_sched_migration_cost; | |
2031 | } | |
2032 | ||
dd41f596 | 2033 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2034 | { |
e2912009 PZ |
2035 | #ifdef CONFIG_SCHED_DEBUG |
2036 | /* | |
2037 | * We should never call set_task_cpu() on a blocked task, | |
2038 | * ttwu() will sort out the placement. | |
2039 | */ | |
077614ee PZ |
2040 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2041 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2042 | #endif |
2043 | ||
de1d7286 | 2044 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2045 | |
0c69774e PZ |
2046 | if (task_cpu(p) != new_cpu) { |
2047 | p->se.nr_migrations++; | |
2048 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2049 | } | |
dd41f596 IM |
2050 | |
2051 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2052 | } |
2053 | ||
70b97a7f | 2054 | struct migration_req { |
1da177e4 | 2055 | struct list_head list; |
1da177e4 | 2056 | |
36c8b586 | 2057 | struct task_struct *task; |
1da177e4 LT |
2058 | int dest_cpu; |
2059 | ||
1da177e4 | 2060 | struct completion done; |
70b97a7f | 2061 | }; |
1da177e4 LT |
2062 | |
2063 | /* | |
2064 | * The task's runqueue lock must be held. | |
2065 | * Returns true if you have to wait for migration thread. | |
2066 | */ | |
36c8b586 | 2067 | static int |
70b97a7f | 2068 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2069 | { |
70b97a7f | 2070 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2071 | |
2072 | /* | |
2073 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2074 | * the next wake-up will properly place the task. |
1da177e4 | 2075 | */ |
e2912009 | 2076 | if (!p->se.on_rq && !task_running(rq, p)) |
1da177e4 | 2077 | return 0; |
1da177e4 LT |
2078 | |
2079 | init_completion(&req->done); | |
1da177e4 LT |
2080 | req->task = p; |
2081 | req->dest_cpu = dest_cpu; | |
2082 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2083 | |
1da177e4 LT |
2084 | return 1; |
2085 | } | |
2086 | ||
a26b89f0 MM |
2087 | /* |
2088 | * wait_task_context_switch - wait for a thread to complete at least one | |
2089 | * context switch. | |
2090 | * | |
2091 | * @p must not be current. | |
2092 | */ | |
2093 | void wait_task_context_switch(struct task_struct *p) | |
2094 | { | |
2095 | unsigned long nvcsw, nivcsw, flags; | |
2096 | int running; | |
2097 | struct rq *rq; | |
2098 | ||
2099 | nvcsw = p->nvcsw; | |
2100 | nivcsw = p->nivcsw; | |
2101 | for (;;) { | |
2102 | /* | |
2103 | * The runqueue is assigned before the actual context | |
2104 | * switch. We need to take the runqueue lock. | |
2105 | * | |
2106 | * We could check initially without the lock but it is | |
2107 | * very likely that we need to take the lock in every | |
2108 | * iteration. | |
2109 | */ | |
2110 | rq = task_rq_lock(p, &flags); | |
2111 | running = task_running(rq, p); | |
2112 | task_rq_unlock(rq, &flags); | |
2113 | ||
2114 | if (likely(!running)) | |
2115 | break; | |
2116 | /* | |
2117 | * The switch count is incremented before the actual | |
2118 | * context switch. We thus wait for two switches to be | |
2119 | * sure at least one completed. | |
2120 | */ | |
2121 | if ((p->nvcsw - nvcsw) > 1) | |
2122 | break; | |
2123 | if ((p->nivcsw - nivcsw) > 1) | |
2124 | break; | |
2125 | ||
2126 | cpu_relax(); | |
2127 | } | |
2128 | } | |
2129 | ||
1da177e4 LT |
2130 | /* |
2131 | * wait_task_inactive - wait for a thread to unschedule. | |
2132 | * | |
85ba2d86 RM |
2133 | * If @match_state is nonzero, it's the @p->state value just checked and |
2134 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2135 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2136 | * we return a positive number (its total switch count). If a second call | |
2137 | * a short while later returns the same number, the caller can be sure that | |
2138 | * @p has remained unscheduled the whole time. | |
2139 | * | |
1da177e4 LT |
2140 | * The caller must ensure that the task *will* unschedule sometime soon, |
2141 | * else this function might spin for a *long* time. This function can't | |
2142 | * be called with interrupts off, or it may introduce deadlock with | |
2143 | * smp_call_function() if an IPI is sent by the same process we are | |
2144 | * waiting to become inactive. | |
2145 | */ | |
85ba2d86 | 2146 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2147 | { |
2148 | unsigned long flags; | |
dd41f596 | 2149 | int running, on_rq; |
85ba2d86 | 2150 | unsigned long ncsw; |
70b97a7f | 2151 | struct rq *rq; |
1da177e4 | 2152 | |
3a5c359a AK |
2153 | for (;;) { |
2154 | /* | |
2155 | * We do the initial early heuristics without holding | |
2156 | * any task-queue locks at all. We'll only try to get | |
2157 | * the runqueue lock when things look like they will | |
2158 | * work out! | |
2159 | */ | |
2160 | rq = task_rq(p); | |
fa490cfd | 2161 | |
3a5c359a AK |
2162 | /* |
2163 | * If the task is actively running on another CPU | |
2164 | * still, just relax and busy-wait without holding | |
2165 | * any locks. | |
2166 | * | |
2167 | * NOTE! Since we don't hold any locks, it's not | |
2168 | * even sure that "rq" stays as the right runqueue! | |
2169 | * But we don't care, since "task_running()" will | |
2170 | * return false if the runqueue has changed and p | |
2171 | * is actually now running somewhere else! | |
2172 | */ | |
85ba2d86 RM |
2173 | while (task_running(rq, p)) { |
2174 | if (match_state && unlikely(p->state != match_state)) | |
2175 | return 0; | |
3a5c359a | 2176 | cpu_relax(); |
85ba2d86 | 2177 | } |
fa490cfd | 2178 | |
3a5c359a AK |
2179 | /* |
2180 | * Ok, time to look more closely! We need the rq | |
2181 | * lock now, to be *sure*. If we're wrong, we'll | |
2182 | * just go back and repeat. | |
2183 | */ | |
2184 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2185 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2186 | running = task_running(rq, p); |
2187 | on_rq = p->se.on_rq; | |
85ba2d86 | 2188 | ncsw = 0; |
f31e11d8 | 2189 | if (!match_state || p->state == match_state) |
93dcf55f | 2190 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2191 | task_rq_unlock(rq, &flags); |
fa490cfd | 2192 | |
85ba2d86 RM |
2193 | /* |
2194 | * If it changed from the expected state, bail out now. | |
2195 | */ | |
2196 | if (unlikely(!ncsw)) | |
2197 | break; | |
2198 | ||
3a5c359a AK |
2199 | /* |
2200 | * Was it really running after all now that we | |
2201 | * checked with the proper locks actually held? | |
2202 | * | |
2203 | * Oops. Go back and try again.. | |
2204 | */ | |
2205 | if (unlikely(running)) { | |
2206 | cpu_relax(); | |
2207 | continue; | |
2208 | } | |
fa490cfd | 2209 | |
3a5c359a AK |
2210 | /* |
2211 | * It's not enough that it's not actively running, | |
2212 | * it must be off the runqueue _entirely_, and not | |
2213 | * preempted! | |
2214 | * | |
80dd99b3 | 2215 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2216 | * running right now), it's preempted, and we should |
2217 | * yield - it could be a while. | |
2218 | */ | |
2219 | if (unlikely(on_rq)) { | |
2220 | schedule_timeout_uninterruptible(1); | |
2221 | continue; | |
2222 | } | |
fa490cfd | 2223 | |
3a5c359a AK |
2224 | /* |
2225 | * Ahh, all good. It wasn't running, and it wasn't | |
2226 | * runnable, which means that it will never become | |
2227 | * running in the future either. We're all done! | |
2228 | */ | |
2229 | break; | |
2230 | } | |
85ba2d86 RM |
2231 | |
2232 | return ncsw; | |
1da177e4 LT |
2233 | } |
2234 | ||
2235 | /*** | |
2236 | * kick_process - kick a running thread to enter/exit the kernel | |
2237 | * @p: the to-be-kicked thread | |
2238 | * | |
2239 | * Cause a process which is running on another CPU to enter | |
2240 | * kernel-mode, without any delay. (to get signals handled.) | |
2241 | * | |
2242 | * NOTE: this function doesnt have to take the runqueue lock, | |
2243 | * because all it wants to ensure is that the remote task enters | |
2244 | * the kernel. If the IPI races and the task has been migrated | |
2245 | * to another CPU then no harm is done and the purpose has been | |
2246 | * achieved as well. | |
2247 | */ | |
36c8b586 | 2248 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2249 | { |
2250 | int cpu; | |
2251 | ||
2252 | preempt_disable(); | |
2253 | cpu = task_cpu(p); | |
2254 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2255 | smp_send_reschedule(cpu); | |
2256 | preempt_enable(); | |
2257 | } | |
b43e3521 | 2258 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2259 | #endif /* CONFIG_SMP */ |
1da177e4 | 2260 | |
0793a61d TG |
2261 | /** |
2262 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2263 | * @p: the task to evaluate | |
2264 | * @func: the function to be called | |
2265 | * @info: the function call argument | |
2266 | * | |
2267 | * Calls the function @func when the task is currently running. This might | |
2268 | * be on the current CPU, which just calls the function directly | |
2269 | */ | |
2270 | void task_oncpu_function_call(struct task_struct *p, | |
2271 | void (*func) (void *info), void *info) | |
2272 | { | |
2273 | int cpu; | |
2274 | ||
2275 | preempt_disable(); | |
2276 | cpu = task_cpu(p); | |
2277 | if (task_curr(p)) | |
2278 | smp_call_function_single(cpu, func, info, 1); | |
2279 | preempt_enable(); | |
2280 | } | |
2281 | ||
970b13ba | 2282 | #ifdef CONFIG_SMP |
5da9a0fb PZ |
2283 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2284 | { | |
2285 | int dest_cpu; | |
2286 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2287 | ||
2288 | /* Look for allowed, online CPU in same node. */ | |
2289 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2290 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2291 | return dest_cpu; | |
2292 | ||
2293 | /* Any allowed, online CPU? */ | |
2294 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2295 | if (dest_cpu < nr_cpu_ids) | |
2296 | return dest_cpu; | |
2297 | ||
2298 | /* No more Mr. Nice Guy. */ | |
2299 | if (dest_cpu >= nr_cpu_ids) { | |
2300 | rcu_read_lock(); | |
2301 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); | |
2302 | rcu_read_unlock(); | |
2303 | dest_cpu = cpumask_any_and(cpu_active_mask, &p->cpus_allowed); | |
2304 | ||
2305 | /* | |
2306 | * Don't tell them about moving exiting tasks or | |
2307 | * kernel threads (both mm NULL), since they never | |
2308 | * leave kernel. | |
2309 | */ | |
2310 | if (p->mm && printk_ratelimit()) { | |
2311 | printk(KERN_INFO "process %d (%s) no " | |
2312 | "longer affine to cpu%d\n", | |
2313 | task_pid_nr(p), p->comm, cpu); | |
2314 | } | |
2315 | } | |
2316 | ||
2317 | return dest_cpu; | |
2318 | } | |
2319 | ||
e2912009 | 2320 | /* |
fabf318e PZ |
2321 | * Gets called from 3 sites (exec, fork, wakeup), since it is called without |
2322 | * holding rq->lock we need to ensure ->cpus_allowed is stable, this is done | |
2323 | * by: | |
e2912009 | 2324 | * |
fabf318e PZ |
2325 | * exec: is unstable, retry loop |
2326 | * fork & wake-up: serialize ->cpus_allowed against TASK_WAKING | |
e2912009 | 2327 | */ |
970b13ba PZ |
2328 | static inline |
2329 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) | |
2330 | { | |
e2912009 PZ |
2331 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
2332 | ||
2333 | /* | |
2334 | * In order not to call set_task_cpu() on a blocking task we need | |
2335 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2336 | * cpu. | |
2337 | * | |
2338 | * Since this is common to all placement strategies, this lives here. | |
2339 | * | |
2340 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2341 | * not worry about this generic constraint ] | |
2342 | */ | |
2343 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2344 | !cpu_online(cpu))) |
5da9a0fb | 2345 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2346 | |
2347 | return cpu; | |
970b13ba PZ |
2348 | } |
2349 | #endif | |
2350 | ||
1da177e4 LT |
2351 | /*** |
2352 | * try_to_wake_up - wake up a thread | |
2353 | * @p: the to-be-woken-up thread | |
2354 | * @state: the mask of task states that can be woken | |
2355 | * @sync: do a synchronous wakeup? | |
2356 | * | |
2357 | * Put it on the run-queue if it's not already there. The "current" | |
2358 | * thread is always on the run-queue (except when the actual | |
2359 | * re-schedule is in progress), and as such you're allowed to do | |
2360 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2361 | * runnable without the overhead of this. | |
2362 | * | |
2363 | * returns failure only if the task is already active. | |
2364 | */ | |
7d478721 PZ |
2365 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2366 | int wake_flags) | |
1da177e4 | 2367 | { |
cc367732 | 2368 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2369 | unsigned long flags; |
ab3b3aa5 | 2370 | struct rq *rq; |
1da177e4 | 2371 | |
e9c84311 | 2372 | this_cpu = get_cpu(); |
2398f2c6 | 2373 | |
04e2f174 | 2374 | smp_wmb(); |
ab3b3aa5 | 2375 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2376 | if (!(p->state & state)) |
1da177e4 LT |
2377 | goto out; |
2378 | ||
dd41f596 | 2379 | if (p->se.on_rq) |
1da177e4 LT |
2380 | goto out_running; |
2381 | ||
2382 | cpu = task_cpu(p); | |
cc367732 | 2383 | orig_cpu = cpu; |
1da177e4 LT |
2384 | |
2385 | #ifdef CONFIG_SMP | |
2386 | if (unlikely(task_running(rq, p))) | |
2387 | goto out_activate; | |
2388 | ||
e9c84311 PZ |
2389 | /* |
2390 | * In order to handle concurrent wakeups and release the rq->lock | |
2391 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2392 | * |
2393 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2394 | */ |
eb24073b IM |
2395 | if (task_contributes_to_load(p)) |
2396 | rq->nr_uninterruptible--; | |
e9c84311 | 2397 | p->state = TASK_WAKING; |
efbbd05a PZ |
2398 | |
2399 | if (p->sched_class->task_waking) | |
2400 | p->sched_class->task_waking(rq, p); | |
2401 | ||
ab19cb23 | 2402 | __task_rq_unlock(rq); |
e9c84311 | 2403 | |
970b13ba | 2404 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
0970d299 PZ |
2405 | if (cpu != orig_cpu) { |
2406 | /* | |
2407 | * Since we migrate the task without holding any rq->lock, | |
2408 | * we need to be careful with task_rq_lock(), since that | |
2409 | * might end up locking an invalid rq. | |
2410 | */ | |
5d2f5a61 | 2411 | set_task_cpu(p, cpu); |
0970d299 | 2412 | } |
ab19cb23 | 2413 | |
0970d299 PZ |
2414 | rq = cpu_rq(cpu); |
2415 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2416 | |
0970d299 PZ |
2417 | /* |
2418 | * We migrated the task without holding either rq->lock, however | |
2419 | * since the task is not on the task list itself, nobody else | |
2420 | * will try and migrate the task, hence the rq should match the | |
2421 | * cpu we just moved it to. | |
2422 | */ | |
2423 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2424 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2425 | |
e7693a36 GH |
2426 | #ifdef CONFIG_SCHEDSTATS |
2427 | schedstat_inc(rq, ttwu_count); | |
2428 | if (cpu == this_cpu) | |
2429 | schedstat_inc(rq, ttwu_local); | |
2430 | else { | |
2431 | struct sched_domain *sd; | |
2432 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2433 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2434 | schedstat_inc(sd, ttwu_wake_remote); |
2435 | break; | |
2436 | } | |
2437 | } | |
2438 | } | |
6d6bc0ad | 2439 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2440 | |
1da177e4 LT |
2441 | out_activate: |
2442 | #endif /* CONFIG_SMP */ | |
41acab88 | 2443 | schedstat_inc(p, se.statistics.nr_wakeups); |
7d478721 | 2444 | if (wake_flags & WF_SYNC) |
41acab88 | 2445 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
cc367732 | 2446 | if (orig_cpu != cpu) |
41acab88 | 2447 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); |
cc367732 | 2448 | if (cpu == this_cpu) |
41acab88 | 2449 | schedstat_inc(p, se.statistics.nr_wakeups_local); |
cc367732 | 2450 | else |
41acab88 | 2451 | schedstat_inc(p, se.statistics.nr_wakeups_remote); |
dd41f596 | 2452 | activate_task(rq, p, 1); |
1da177e4 LT |
2453 | success = 1; |
2454 | ||
2455 | out_running: | |
468a15bb | 2456 | trace_sched_wakeup(rq, p, success); |
7d478721 | 2457 | check_preempt_curr(rq, p, wake_flags); |
4ae7d5ce | 2458 | |
1da177e4 | 2459 | p->state = TASK_RUNNING; |
9a897c5a | 2460 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2461 | if (p->sched_class->task_woken) |
2462 | p->sched_class->task_woken(rq, p); | |
eae0c9df MG |
2463 | |
2464 | if (unlikely(rq->idle_stamp)) { | |
2465 | u64 delta = rq->clock - rq->idle_stamp; | |
2466 | u64 max = 2*sysctl_sched_migration_cost; | |
2467 | ||
2468 | if (delta > max) | |
2469 | rq->avg_idle = max; | |
2470 | else | |
2471 | update_avg(&rq->avg_idle, delta); | |
2472 | rq->idle_stamp = 0; | |
2473 | } | |
9a897c5a | 2474 | #endif |
1da177e4 LT |
2475 | out: |
2476 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2477 | put_cpu(); |
1da177e4 LT |
2478 | |
2479 | return success; | |
2480 | } | |
2481 | ||
50fa610a DH |
2482 | /** |
2483 | * wake_up_process - Wake up a specific process | |
2484 | * @p: The process to be woken up. | |
2485 | * | |
2486 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2487 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2488 | * running. | |
2489 | * | |
2490 | * It may be assumed that this function implies a write memory barrier before | |
2491 | * changing the task state if and only if any tasks are woken up. | |
2492 | */ | |
7ad5b3a5 | 2493 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2494 | { |
d9514f6c | 2495 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2496 | } |
1da177e4 LT |
2497 | EXPORT_SYMBOL(wake_up_process); |
2498 | ||
7ad5b3a5 | 2499 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2500 | { |
2501 | return try_to_wake_up(p, state, 0); | |
2502 | } | |
2503 | ||
1da177e4 LT |
2504 | /* |
2505 | * Perform scheduler related setup for a newly forked process p. | |
2506 | * p is forked by current. | |
dd41f596 IM |
2507 | * |
2508 | * __sched_fork() is basic setup used by init_idle() too: | |
2509 | */ | |
2510 | static void __sched_fork(struct task_struct *p) | |
2511 | { | |
dd41f596 IM |
2512 | p->se.exec_start = 0; |
2513 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2514 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2515 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2516 | |
2517 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2518 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2519 | #endif |
476d139c | 2520 | |
fa717060 | 2521 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2522 | p->se.on_rq = 0; |
4a55bd5e | 2523 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2524 | |
e107be36 AK |
2525 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2526 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2527 | #endif | |
dd41f596 IM |
2528 | } |
2529 | ||
2530 | /* | |
2531 | * fork()/clone()-time setup: | |
2532 | */ | |
2533 | void sched_fork(struct task_struct *p, int clone_flags) | |
2534 | { | |
2535 | int cpu = get_cpu(); | |
2536 | ||
2537 | __sched_fork(p); | |
06b83b5f PZ |
2538 | /* |
2539 | * We mark the process as waking here. This guarantees that | |
2540 | * nobody will actually run it, and a signal or other external | |
2541 | * event cannot wake it up and insert it on the runqueue either. | |
2542 | */ | |
2543 | p->state = TASK_WAKING; | |
dd41f596 | 2544 | |
b9dc29e7 MG |
2545 | /* |
2546 | * Revert to default priority/policy on fork if requested. | |
2547 | */ | |
2548 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2549 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2550 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2551 | p->normal_prio = p->static_prio; |
2552 | } | |
b9dc29e7 | 2553 | |
6c697bdf MG |
2554 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2555 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2556 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2557 | set_load_weight(p); |
2558 | } | |
2559 | ||
b9dc29e7 MG |
2560 | /* |
2561 | * We don't need the reset flag anymore after the fork. It has | |
2562 | * fulfilled its duty: | |
2563 | */ | |
2564 | p->sched_reset_on_fork = 0; | |
2565 | } | |
ca94c442 | 2566 | |
f83f9ac2 PW |
2567 | /* |
2568 | * Make sure we do not leak PI boosting priority to the child. | |
2569 | */ | |
2570 | p->prio = current->normal_prio; | |
2571 | ||
2ddbf952 HS |
2572 | if (!rt_prio(p->prio)) |
2573 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2574 | |
cd29fe6f PZ |
2575 | if (p->sched_class->task_fork) |
2576 | p->sched_class->task_fork(p); | |
2577 | ||
5f3edc1b PZ |
2578 | set_task_cpu(p, cpu); |
2579 | ||
52f17b6c | 2580 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2581 | if (likely(sched_info_on())) |
52f17b6c | 2582 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2583 | #endif |
d6077cb8 | 2584 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2585 | p->oncpu = 0; |
2586 | #endif | |
1da177e4 | 2587 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2588 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2589 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2590 | #endif |
917b627d GH |
2591 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2592 | ||
476d139c | 2593 | put_cpu(); |
1da177e4 LT |
2594 | } |
2595 | ||
2596 | /* | |
2597 | * wake_up_new_task - wake up a newly created task for the first time. | |
2598 | * | |
2599 | * This function will do some initial scheduler statistics housekeeping | |
2600 | * that must be done for every newly created context, then puts the task | |
2601 | * on the runqueue and wakes it. | |
2602 | */ | |
7ad5b3a5 | 2603 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2604 | { |
2605 | unsigned long flags; | |
dd41f596 | 2606 | struct rq *rq; |
c890692b | 2607 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2608 | |
2609 | #ifdef CONFIG_SMP | |
2610 | /* | |
2611 | * Fork balancing, do it here and not earlier because: | |
2612 | * - cpus_allowed can change in the fork path | |
2613 | * - any previously selected cpu might disappear through hotplug | |
2614 | * | |
2615 | * We still have TASK_WAKING but PF_STARTING is gone now, meaning | |
2616 | * ->cpus_allowed is stable, we have preemption disabled, meaning | |
2617 | * cpu_online_mask is stable. | |
2618 | */ | |
2619 | cpu = select_task_rq(p, SD_BALANCE_FORK, 0); | |
2620 | set_task_cpu(p, cpu); | |
2621 | #endif | |
1da177e4 | 2622 | |
0970d299 PZ |
2623 | /* |
2624 | * Since the task is not on the rq and we still have TASK_WAKING set | |
2625 | * nobody else will migrate this task. | |
2626 | */ | |
2627 | rq = cpu_rq(cpu); | |
2628 | raw_spin_lock_irqsave(&rq->lock, flags); | |
2629 | ||
06b83b5f PZ |
2630 | BUG_ON(p->state != TASK_WAKING); |
2631 | p->state = TASK_RUNNING; | |
cd29fe6f | 2632 | activate_task(rq, p, 0); |
c71dd42d | 2633 | trace_sched_wakeup_new(rq, p, 1); |
a7558e01 | 2634 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2635 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2636 | if (p->sched_class->task_woken) |
2637 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2638 | #endif |
dd41f596 | 2639 | task_rq_unlock(rq, &flags); |
fabf318e | 2640 | put_cpu(); |
1da177e4 LT |
2641 | } |
2642 | ||
e107be36 AK |
2643 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2644 | ||
2645 | /** | |
80dd99b3 | 2646 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2647 | * @notifier: notifier struct to register |
e107be36 AK |
2648 | */ |
2649 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2650 | { | |
2651 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2652 | } | |
2653 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2654 | ||
2655 | /** | |
2656 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2657 | * @notifier: notifier struct to unregister |
e107be36 AK |
2658 | * |
2659 | * This is safe to call from within a preemption notifier. | |
2660 | */ | |
2661 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2662 | { | |
2663 | hlist_del(¬ifier->link); | |
2664 | } | |
2665 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2666 | ||
2667 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2668 | { | |
2669 | struct preempt_notifier *notifier; | |
2670 | struct hlist_node *node; | |
2671 | ||
2672 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2673 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2674 | } | |
2675 | ||
2676 | static void | |
2677 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2678 | struct task_struct *next) | |
2679 | { | |
2680 | struct preempt_notifier *notifier; | |
2681 | struct hlist_node *node; | |
2682 | ||
2683 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2684 | notifier->ops->sched_out(notifier, next); | |
2685 | } | |
2686 | ||
6d6bc0ad | 2687 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2688 | |
2689 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2690 | { | |
2691 | } | |
2692 | ||
2693 | static void | |
2694 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2695 | struct task_struct *next) | |
2696 | { | |
2697 | } | |
2698 | ||
6d6bc0ad | 2699 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2700 | |
4866cde0 NP |
2701 | /** |
2702 | * prepare_task_switch - prepare to switch tasks | |
2703 | * @rq: the runqueue preparing to switch | |
421cee29 | 2704 | * @prev: the current task that is being switched out |
4866cde0 NP |
2705 | * @next: the task we are going to switch to. |
2706 | * | |
2707 | * This is called with the rq lock held and interrupts off. It must | |
2708 | * be paired with a subsequent finish_task_switch after the context | |
2709 | * switch. | |
2710 | * | |
2711 | * prepare_task_switch sets up locking and calls architecture specific | |
2712 | * hooks. | |
2713 | */ | |
e107be36 AK |
2714 | static inline void |
2715 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2716 | struct task_struct *next) | |
4866cde0 | 2717 | { |
e107be36 | 2718 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2719 | prepare_lock_switch(rq, next); |
2720 | prepare_arch_switch(next); | |
2721 | } | |
2722 | ||
1da177e4 LT |
2723 | /** |
2724 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2725 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2726 | * @prev: the thread we just switched away from. |
2727 | * | |
4866cde0 NP |
2728 | * finish_task_switch must be called after the context switch, paired |
2729 | * with a prepare_task_switch call before the context switch. | |
2730 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2731 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2732 | * |
2733 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2734 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2735 | * with the lock held can cause deadlocks; see schedule() for |
2736 | * details.) | |
2737 | */ | |
a9957449 | 2738 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2739 | __releases(rq->lock) |
2740 | { | |
1da177e4 | 2741 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2742 | long prev_state; |
1da177e4 LT |
2743 | |
2744 | rq->prev_mm = NULL; | |
2745 | ||
2746 | /* | |
2747 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2748 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2749 | * schedule one last time. The schedule call will never return, and |
2750 | * the scheduled task must drop that reference. | |
c394cc9f | 2751 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2752 | * still held, otherwise prev could be scheduled on another cpu, die |
2753 | * there before we look at prev->state, and then the reference would | |
2754 | * be dropped twice. | |
2755 | * Manfred Spraul <manfred@colorfullife.com> | |
2756 | */ | |
55a101f8 | 2757 | prev_state = prev->state; |
4866cde0 | 2758 | finish_arch_switch(prev); |
8381f65d JI |
2759 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2760 | local_irq_disable(); | |
2761 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2762 | perf_event_task_sched_in(current); |
8381f65d JI |
2763 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2764 | local_irq_enable(); | |
2765 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2766 | finish_lock_switch(rq, prev); |
e8fa1362 | 2767 | |
e107be36 | 2768 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2769 | if (mm) |
2770 | mmdrop(mm); | |
c394cc9f | 2771 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2772 | /* |
2773 | * Remove function-return probe instances associated with this | |
2774 | * task and put them back on the free list. | |
9761eea8 | 2775 | */ |
c6fd91f0 | 2776 | kprobe_flush_task(prev); |
1da177e4 | 2777 | put_task_struct(prev); |
c6fd91f0 | 2778 | } |
1da177e4 LT |
2779 | } |
2780 | ||
3f029d3c GH |
2781 | #ifdef CONFIG_SMP |
2782 | ||
2783 | /* assumes rq->lock is held */ | |
2784 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2785 | { | |
2786 | if (prev->sched_class->pre_schedule) | |
2787 | prev->sched_class->pre_schedule(rq, prev); | |
2788 | } | |
2789 | ||
2790 | /* rq->lock is NOT held, but preemption is disabled */ | |
2791 | static inline void post_schedule(struct rq *rq) | |
2792 | { | |
2793 | if (rq->post_schedule) { | |
2794 | unsigned long flags; | |
2795 | ||
05fa785c | 2796 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2797 | if (rq->curr->sched_class->post_schedule) |
2798 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2799 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2800 | |
2801 | rq->post_schedule = 0; | |
2802 | } | |
2803 | } | |
2804 | ||
2805 | #else | |
da19ab51 | 2806 | |
3f029d3c GH |
2807 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2808 | { | |
2809 | } | |
2810 | ||
2811 | static inline void post_schedule(struct rq *rq) | |
2812 | { | |
1da177e4 LT |
2813 | } |
2814 | ||
3f029d3c GH |
2815 | #endif |
2816 | ||
1da177e4 LT |
2817 | /** |
2818 | * schedule_tail - first thing a freshly forked thread must call. | |
2819 | * @prev: the thread we just switched away from. | |
2820 | */ | |
36c8b586 | 2821 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2822 | __releases(rq->lock) |
2823 | { | |
70b97a7f IM |
2824 | struct rq *rq = this_rq(); |
2825 | ||
4866cde0 | 2826 | finish_task_switch(rq, prev); |
da19ab51 | 2827 | |
3f029d3c GH |
2828 | /* |
2829 | * FIXME: do we need to worry about rq being invalidated by the | |
2830 | * task_switch? | |
2831 | */ | |
2832 | post_schedule(rq); | |
70b97a7f | 2833 | |
4866cde0 NP |
2834 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2835 | /* In this case, finish_task_switch does not reenable preemption */ | |
2836 | preempt_enable(); | |
2837 | #endif | |
1da177e4 | 2838 | if (current->set_child_tid) |
b488893a | 2839 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2840 | } |
2841 | ||
2842 | /* | |
2843 | * context_switch - switch to the new MM and the new | |
2844 | * thread's register state. | |
2845 | */ | |
dd41f596 | 2846 | static inline void |
70b97a7f | 2847 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2848 | struct task_struct *next) |
1da177e4 | 2849 | { |
dd41f596 | 2850 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2851 | |
e107be36 | 2852 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2853 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2854 | mm = next->mm; |
2855 | oldmm = prev->active_mm; | |
9226d125 ZA |
2856 | /* |
2857 | * For paravirt, this is coupled with an exit in switch_to to | |
2858 | * combine the page table reload and the switch backend into | |
2859 | * one hypercall. | |
2860 | */ | |
224101ed | 2861 | arch_start_context_switch(prev); |
9226d125 | 2862 | |
710390d9 | 2863 | if (likely(!mm)) { |
1da177e4 LT |
2864 | next->active_mm = oldmm; |
2865 | atomic_inc(&oldmm->mm_count); | |
2866 | enter_lazy_tlb(oldmm, next); | |
2867 | } else | |
2868 | switch_mm(oldmm, mm, next); | |
2869 | ||
710390d9 | 2870 | if (likely(!prev->mm)) { |
1da177e4 | 2871 | prev->active_mm = NULL; |
1da177e4 LT |
2872 | rq->prev_mm = oldmm; |
2873 | } | |
3a5f5e48 IM |
2874 | /* |
2875 | * Since the runqueue lock will be released by the next | |
2876 | * task (which is an invalid locking op but in the case | |
2877 | * of the scheduler it's an obvious special-case), so we | |
2878 | * do an early lockdep release here: | |
2879 | */ | |
2880 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2881 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2882 | #endif |
1da177e4 LT |
2883 | |
2884 | /* Here we just switch the register state and the stack. */ | |
2885 | switch_to(prev, next, prev); | |
2886 | ||
dd41f596 IM |
2887 | barrier(); |
2888 | /* | |
2889 | * this_rq must be evaluated again because prev may have moved | |
2890 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2891 | * frame will be invalid. | |
2892 | */ | |
2893 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2894 | } |
2895 | ||
2896 | /* | |
2897 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2898 | * | |
2899 | * externally visible scheduler statistics: current number of runnable | |
2900 | * threads, current number of uninterruptible-sleeping threads, total | |
2901 | * number of context switches performed since bootup. | |
2902 | */ | |
2903 | unsigned long nr_running(void) | |
2904 | { | |
2905 | unsigned long i, sum = 0; | |
2906 | ||
2907 | for_each_online_cpu(i) | |
2908 | sum += cpu_rq(i)->nr_running; | |
2909 | ||
2910 | return sum; | |
f711f609 | 2911 | } |
1da177e4 LT |
2912 | |
2913 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2914 | { |
1da177e4 | 2915 | unsigned long i, sum = 0; |
f711f609 | 2916 | |
0a945022 | 2917 | for_each_possible_cpu(i) |
1da177e4 | 2918 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2919 | |
2920 | /* | |
1da177e4 LT |
2921 | * Since we read the counters lockless, it might be slightly |
2922 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2923 | */ |
1da177e4 LT |
2924 | if (unlikely((long)sum < 0)) |
2925 | sum = 0; | |
f711f609 | 2926 | |
1da177e4 | 2927 | return sum; |
f711f609 | 2928 | } |
f711f609 | 2929 | |
1da177e4 | 2930 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2931 | { |
cc94abfc SR |
2932 | int i; |
2933 | unsigned long long sum = 0; | |
46cb4b7c | 2934 | |
0a945022 | 2935 | for_each_possible_cpu(i) |
1da177e4 | 2936 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2937 | |
1da177e4 LT |
2938 | return sum; |
2939 | } | |
483b4ee6 | 2940 | |
1da177e4 LT |
2941 | unsigned long nr_iowait(void) |
2942 | { | |
2943 | unsigned long i, sum = 0; | |
483b4ee6 | 2944 | |
0a945022 | 2945 | for_each_possible_cpu(i) |
1da177e4 | 2946 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2947 | |
1da177e4 LT |
2948 | return sum; |
2949 | } | |
483b4ee6 | 2950 | |
69d25870 AV |
2951 | unsigned long nr_iowait_cpu(void) |
2952 | { | |
2953 | struct rq *this = this_rq(); | |
2954 | return atomic_read(&this->nr_iowait); | |
2955 | } | |
46cb4b7c | 2956 | |
69d25870 AV |
2957 | unsigned long this_cpu_load(void) |
2958 | { | |
2959 | struct rq *this = this_rq(); | |
2960 | return this->cpu_load[0]; | |
2961 | } | |
e790fb0b | 2962 | |
46cb4b7c | 2963 | |
dce48a84 TG |
2964 | /* Variables and functions for calc_load */ |
2965 | static atomic_long_t calc_load_tasks; | |
2966 | static unsigned long calc_load_update; | |
2967 | unsigned long avenrun[3]; | |
2968 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2969 | |
2d02494f TG |
2970 | /** |
2971 | * get_avenrun - get the load average array | |
2972 | * @loads: pointer to dest load array | |
2973 | * @offset: offset to add | |
2974 | * @shift: shift count to shift the result left | |
2975 | * | |
2976 | * These values are estimates at best, so no need for locking. | |
2977 | */ | |
2978 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2979 | { | |
2980 | loads[0] = (avenrun[0] + offset) << shift; | |
2981 | loads[1] = (avenrun[1] + offset) << shift; | |
2982 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 2983 | } |
46cb4b7c | 2984 | |
dce48a84 TG |
2985 | static unsigned long |
2986 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2987 | { |
dce48a84 TG |
2988 | load *= exp; |
2989 | load += active * (FIXED_1 - exp); | |
2990 | return load >> FSHIFT; | |
2991 | } | |
46cb4b7c SS |
2992 | |
2993 | /* | |
dce48a84 TG |
2994 | * calc_load - update the avenrun load estimates 10 ticks after the |
2995 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2996 | */ |
dce48a84 | 2997 | void calc_global_load(void) |
7835b98b | 2998 | { |
dce48a84 TG |
2999 | unsigned long upd = calc_load_update + 10; |
3000 | long active; | |
1da177e4 | 3001 | |
dce48a84 TG |
3002 | if (time_before(jiffies, upd)) |
3003 | return; | |
1da177e4 | 3004 | |
dce48a84 TG |
3005 | active = atomic_long_read(&calc_load_tasks); |
3006 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3007 | |
dce48a84 TG |
3008 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3009 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3010 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3011 | |
dce48a84 TG |
3012 | calc_load_update += LOAD_FREQ; |
3013 | } | |
1da177e4 | 3014 | |
dce48a84 TG |
3015 | /* |
3016 | * Either called from update_cpu_load() or from a cpu going idle | |
3017 | */ | |
3018 | static void calc_load_account_active(struct rq *this_rq) | |
3019 | { | |
3020 | long nr_active, delta; | |
08c183f3 | 3021 | |
dce48a84 TG |
3022 | nr_active = this_rq->nr_running; |
3023 | nr_active += (long) this_rq->nr_uninterruptible; | |
783609c6 | 3024 | |
dce48a84 TG |
3025 | if (nr_active != this_rq->calc_load_active) { |
3026 | delta = nr_active - this_rq->calc_load_active; | |
3027 | this_rq->calc_load_active = nr_active; | |
3028 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 | 3029 | } |
46cb4b7c SS |
3030 | } |
3031 | ||
3032 | /* | |
dd41f596 IM |
3033 | * Update rq->cpu_load[] statistics. This function is usually called every |
3034 | * scheduler tick (TICK_NSEC). | |
46cb4b7c | 3035 | */ |
dd41f596 | 3036 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3037 | { |
495eca49 | 3038 | unsigned long this_load = this_rq->load.weight; |
dd41f596 | 3039 | int i, scale; |
46cb4b7c | 3040 | |
dd41f596 | 3041 | this_rq->nr_load_updates++; |
46cb4b7c | 3042 | |
dd41f596 IM |
3043 | /* Update our load: */ |
3044 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3045 | unsigned long old_load, new_load; | |
7d1e6a9b | 3046 | |
dd41f596 | 3047 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3048 | |
dd41f596 IM |
3049 | old_load = this_rq->cpu_load[i]; |
3050 | new_load = this_load; | |
a25707f3 IM |
3051 | /* |
3052 | * Round up the averaging division if load is increasing. This | |
3053 | * prevents us from getting stuck on 9 if the load is 10, for | |
3054 | * example. | |
3055 | */ | |
3056 | if (new_load > old_load) | |
3057 | new_load += scale-1; | |
dd41f596 IM |
3058 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3059 | } | |
46cb4b7c | 3060 | |
dce48a84 TG |
3061 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { |
3062 | this_rq->calc_load_update += LOAD_FREQ; | |
3063 | calc_load_account_active(this_rq); | |
46cb4b7c | 3064 | } |
46cb4b7c SS |
3065 | } |
3066 | ||
dd41f596 | 3067 | #ifdef CONFIG_SMP |
8a0be9ef | 3068 | |
46cb4b7c | 3069 | /* |
38022906 PZ |
3070 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3071 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3072 | */ |
38022906 | 3073 | void sched_exec(void) |
46cb4b7c | 3074 | { |
38022906 | 3075 | struct task_struct *p = current; |
70b97a7f | 3076 | struct migration_req req; |
38022906 | 3077 | int dest_cpu, this_cpu; |
1da177e4 | 3078 | unsigned long flags; |
70b97a7f | 3079 | struct rq *rq; |
46cb4b7c | 3080 | |
38022906 PZ |
3081 | again: |
3082 | this_cpu = get_cpu(); | |
3083 | dest_cpu = select_task_rq(p, SD_BALANCE_EXEC, 0); | |
3084 | if (dest_cpu == this_cpu) { | |
3085 | put_cpu(); | |
3086 | return; | |
46cb4b7c SS |
3087 | } |
3088 | ||
1da177e4 | 3089 | rq = task_rq_lock(p, &flags); |
38022906 PZ |
3090 | put_cpu(); |
3091 | ||
46cb4b7c | 3092 | /* |
38022906 | 3093 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3094 | */ |
96f874e2 | 3095 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
38022906 PZ |
3096 | || unlikely(!cpu_active(dest_cpu))) { |
3097 | task_rq_unlock(rq, &flags); | |
3098 | goto again; | |
46cb4b7c SS |
3099 | } |
3100 | ||
1da177e4 LT |
3101 | /* force the process onto the specified CPU */ |
3102 | if (migrate_task(p, dest_cpu, &req)) { | |
3103 | /* Need to wait for migration thread (might exit: take ref). */ | |
3104 | struct task_struct *mt = rq->migration_thread; | |
dd41f596 | 3105 | |
1da177e4 LT |
3106 | get_task_struct(mt); |
3107 | task_rq_unlock(rq, &flags); | |
3108 | wake_up_process(mt); | |
3109 | put_task_struct(mt); | |
3110 | wait_for_completion(&req.done); | |
dd41f596 | 3111 | |
1da177e4 LT |
3112 | return; |
3113 | } | |
1da177e4 | 3114 | task_rq_unlock(rq, &flags); |
1da177e4 | 3115 | } |
dd41f596 | 3116 | |
1da177e4 LT |
3117 | #endif |
3118 | ||
1da177e4 LT |
3119 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3120 | ||
3121 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3122 | ||
3123 | /* | |
c5f8d995 | 3124 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3125 | * @p in case that task is currently running. |
c5f8d995 HS |
3126 | * |
3127 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3128 | */ |
c5f8d995 HS |
3129 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3130 | { | |
3131 | u64 ns = 0; | |
3132 | ||
3133 | if (task_current(rq, p)) { | |
3134 | update_rq_clock(rq); | |
3135 | ns = rq->clock - p->se.exec_start; | |
3136 | if ((s64)ns < 0) | |
3137 | ns = 0; | |
3138 | } | |
3139 | ||
3140 | return ns; | |
3141 | } | |
3142 | ||
bb34d92f | 3143 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3144 | { |
1da177e4 | 3145 | unsigned long flags; |
41b86e9c | 3146 | struct rq *rq; |
bb34d92f | 3147 | u64 ns = 0; |
48f24c4d | 3148 | |
41b86e9c | 3149 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3150 | ns = do_task_delta_exec(p, rq); |
3151 | task_rq_unlock(rq, &flags); | |
1508487e | 3152 | |
c5f8d995 HS |
3153 | return ns; |
3154 | } | |
f06febc9 | 3155 | |
c5f8d995 HS |
3156 | /* |
3157 | * Return accounted runtime for the task. | |
3158 | * In case the task is currently running, return the runtime plus current's | |
3159 | * pending runtime that have not been accounted yet. | |
3160 | */ | |
3161 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3162 | { | |
3163 | unsigned long flags; | |
3164 | struct rq *rq; | |
3165 | u64 ns = 0; | |
3166 | ||
3167 | rq = task_rq_lock(p, &flags); | |
3168 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3169 | task_rq_unlock(rq, &flags); | |
3170 | ||
3171 | return ns; | |
3172 | } | |
48f24c4d | 3173 | |
c5f8d995 HS |
3174 | /* |
3175 | * Return sum_exec_runtime for the thread group. | |
3176 | * In case the task is currently running, return the sum plus current's | |
3177 | * pending runtime that have not been accounted yet. | |
3178 | * | |
3179 | * Note that the thread group might have other running tasks as well, | |
3180 | * so the return value not includes other pending runtime that other | |
3181 | * running tasks might have. | |
3182 | */ | |
3183 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3184 | { | |
3185 | struct task_cputime totals; | |
3186 | unsigned long flags; | |
3187 | struct rq *rq; | |
3188 | u64 ns; | |
3189 | ||
3190 | rq = task_rq_lock(p, &flags); | |
3191 | thread_group_cputime(p, &totals); | |
3192 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3193 | task_rq_unlock(rq, &flags); |
48f24c4d | 3194 | |
1da177e4 LT |
3195 | return ns; |
3196 | } | |
3197 | ||
1da177e4 LT |
3198 | /* |
3199 | * Account user cpu time to a process. | |
3200 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3201 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3202 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3203 | */ |
457533a7 MS |
3204 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3205 | cputime_t cputime_scaled) | |
1da177e4 LT |
3206 | { |
3207 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3208 | cputime64_t tmp; | |
3209 | ||
457533a7 | 3210 | /* Add user time to process. */ |
1da177e4 | 3211 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3212 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3213 | account_group_user_time(p, cputime); |
1da177e4 LT |
3214 | |
3215 | /* Add user time to cpustat. */ | |
3216 | tmp = cputime_to_cputime64(cputime); | |
3217 | if (TASK_NICE(p) > 0) | |
3218 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3219 | else | |
3220 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3221 | |
3222 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3223 | /* Account for user time used */ |
3224 | acct_update_integrals(p); | |
1da177e4 LT |
3225 | } |
3226 | ||
94886b84 LV |
3227 | /* |
3228 | * Account guest cpu time to a process. | |
3229 | * @p: the process that the cpu time gets accounted to | |
3230 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3231 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3232 | */ |
457533a7 MS |
3233 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3234 | cputime_t cputime_scaled) | |
94886b84 LV |
3235 | { |
3236 | cputime64_t tmp; | |
3237 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3238 | ||
3239 | tmp = cputime_to_cputime64(cputime); | |
3240 | ||
457533a7 | 3241 | /* Add guest time to process. */ |
94886b84 | 3242 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3243 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3244 | account_group_user_time(p, cputime); |
94886b84 LV |
3245 | p->gtime = cputime_add(p->gtime, cputime); |
3246 | ||
457533a7 | 3247 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3248 | if (TASK_NICE(p) > 0) { |
3249 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3250 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3251 | } else { | |
3252 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3253 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3254 | } | |
94886b84 LV |
3255 | } |
3256 | ||
1da177e4 LT |
3257 | /* |
3258 | * Account system cpu time to a process. | |
3259 | * @p: the process that the cpu time gets accounted to | |
3260 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3261 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3262 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3263 | */ |
3264 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3265 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3266 | { |
3267 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3268 | cputime64_t tmp; |
3269 | ||
983ed7a6 | 3270 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3271 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3272 | return; |
3273 | } | |
94886b84 | 3274 | |
457533a7 | 3275 | /* Add system time to process. */ |
1da177e4 | 3276 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3277 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3278 | account_group_system_time(p, cputime); |
1da177e4 LT |
3279 | |
3280 | /* Add system time to cpustat. */ | |
3281 | tmp = cputime_to_cputime64(cputime); | |
3282 | if (hardirq_count() - hardirq_offset) | |
3283 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3284 | else if (softirq_count()) | |
3285 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 3286 | else |
79741dd3 MS |
3287 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3288 | ||
ef12fefa BR |
3289 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3290 | ||
1da177e4 LT |
3291 | /* Account for system time used */ |
3292 | acct_update_integrals(p); | |
1da177e4 LT |
3293 | } |
3294 | ||
c66f08be | 3295 | /* |
1da177e4 | 3296 | * Account for involuntary wait time. |
1da177e4 | 3297 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3298 | */ |
79741dd3 | 3299 | void account_steal_time(cputime_t cputime) |
c66f08be | 3300 | { |
79741dd3 MS |
3301 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3302 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3303 | ||
3304 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3305 | } |
3306 | ||
1da177e4 | 3307 | /* |
79741dd3 MS |
3308 | * Account for idle time. |
3309 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3310 | */ |
79741dd3 | 3311 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3312 | { |
3313 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3314 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3315 | struct rq *rq = this_rq(); |
1da177e4 | 3316 | |
79741dd3 MS |
3317 | if (atomic_read(&rq->nr_iowait) > 0) |
3318 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3319 | else | |
3320 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3321 | } |
3322 | ||
79741dd3 MS |
3323 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3324 | ||
3325 | /* | |
3326 | * Account a single tick of cpu time. | |
3327 | * @p: the process that the cpu time gets accounted to | |
3328 | * @user_tick: indicates if the tick is a user or a system tick | |
3329 | */ | |
3330 | void account_process_tick(struct task_struct *p, int user_tick) | |
3331 | { | |
a42548a1 | 3332 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3333 | struct rq *rq = this_rq(); |
3334 | ||
3335 | if (user_tick) | |
a42548a1 | 3336 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3337 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3338 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3339 | one_jiffy_scaled); |
3340 | else | |
a42548a1 | 3341 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3342 | } |
3343 | ||
3344 | /* | |
3345 | * Account multiple ticks of steal time. | |
3346 | * @p: the process from which the cpu time has been stolen | |
3347 | * @ticks: number of stolen ticks | |
3348 | */ | |
3349 | void account_steal_ticks(unsigned long ticks) | |
3350 | { | |
3351 | account_steal_time(jiffies_to_cputime(ticks)); | |
3352 | } | |
3353 | ||
3354 | /* | |
3355 | * Account multiple ticks of idle time. | |
3356 | * @ticks: number of stolen ticks | |
3357 | */ | |
3358 | void account_idle_ticks(unsigned long ticks) | |
3359 | { | |
3360 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3361 | } |
3362 | ||
79741dd3 MS |
3363 | #endif |
3364 | ||
49048622 BS |
3365 | /* |
3366 | * Use precise platform statistics if available: | |
3367 | */ | |
3368 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3369 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3370 | { |
d99ca3b9 HS |
3371 | *ut = p->utime; |
3372 | *st = p->stime; | |
49048622 BS |
3373 | } |
3374 | ||
0cf55e1e | 3375 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3376 | { |
0cf55e1e HS |
3377 | struct task_cputime cputime; |
3378 | ||
3379 | thread_group_cputime(p, &cputime); | |
3380 | ||
3381 | *ut = cputime.utime; | |
3382 | *st = cputime.stime; | |
49048622 BS |
3383 | } |
3384 | #else | |
761b1d26 HS |
3385 | |
3386 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3387 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3388 | #endif |
3389 | ||
d180c5bc | 3390 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3391 | { |
d99ca3b9 | 3392 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3393 | |
3394 | /* | |
3395 | * Use CFS's precise accounting: | |
3396 | */ | |
d180c5bc | 3397 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3398 | |
3399 | if (total) { | |
d180c5bc HS |
3400 | u64 temp; |
3401 | ||
3402 | temp = (u64)(rtime * utime); | |
49048622 | 3403 | do_div(temp, total); |
d180c5bc HS |
3404 | utime = (cputime_t)temp; |
3405 | } else | |
3406 | utime = rtime; | |
49048622 | 3407 | |
d180c5bc HS |
3408 | /* |
3409 | * Compare with previous values, to keep monotonicity: | |
3410 | */ | |
761b1d26 | 3411 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3412 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3413 | |
d99ca3b9 HS |
3414 | *ut = p->prev_utime; |
3415 | *st = p->prev_stime; | |
49048622 BS |
3416 | } |
3417 | ||
0cf55e1e HS |
3418 | /* |
3419 | * Must be called with siglock held. | |
3420 | */ | |
3421 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3422 | { |
0cf55e1e HS |
3423 | struct signal_struct *sig = p->signal; |
3424 | struct task_cputime cputime; | |
3425 | cputime_t rtime, utime, total; | |
49048622 | 3426 | |
0cf55e1e | 3427 | thread_group_cputime(p, &cputime); |
49048622 | 3428 | |
0cf55e1e HS |
3429 | total = cputime_add(cputime.utime, cputime.stime); |
3430 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3431 | |
0cf55e1e HS |
3432 | if (total) { |
3433 | u64 temp; | |
49048622 | 3434 | |
0cf55e1e HS |
3435 | temp = (u64)(rtime * cputime.utime); |
3436 | do_div(temp, total); | |
3437 | utime = (cputime_t)temp; | |
3438 | } else | |
3439 | utime = rtime; | |
3440 | ||
3441 | sig->prev_utime = max(sig->prev_utime, utime); | |
3442 | sig->prev_stime = max(sig->prev_stime, | |
3443 | cputime_sub(rtime, sig->prev_utime)); | |
3444 | ||
3445 | *ut = sig->prev_utime; | |
3446 | *st = sig->prev_stime; | |
49048622 | 3447 | } |
49048622 | 3448 | #endif |
49048622 | 3449 | |
7835b98b CL |
3450 | /* |
3451 | * This function gets called by the timer code, with HZ frequency. | |
3452 | * We call it with interrupts disabled. | |
3453 | * | |
3454 | * It also gets called by the fork code, when changing the parent's | |
3455 | * timeslices. | |
3456 | */ | |
3457 | void scheduler_tick(void) | |
3458 | { | |
7835b98b CL |
3459 | int cpu = smp_processor_id(); |
3460 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3461 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3462 | |
3463 | sched_clock_tick(); | |
dd41f596 | 3464 | |
05fa785c | 3465 | raw_spin_lock(&rq->lock); |
3e51f33f | 3466 | update_rq_clock(rq); |
f1a438d8 | 3467 | update_cpu_load(rq); |
fa85ae24 | 3468 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3469 | raw_spin_unlock(&rq->lock); |
7835b98b | 3470 | |
49f47433 | 3471 | perf_event_task_tick(curr); |
e220d2dc | 3472 | |
e418e1c2 | 3473 | #ifdef CONFIG_SMP |
dd41f596 IM |
3474 | rq->idle_at_tick = idle_cpu(cpu); |
3475 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3476 | #endif |
1da177e4 LT |
3477 | } |
3478 | ||
132380a0 | 3479 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3480 | { |
3481 | if (in_lock_functions(addr)) { | |
3482 | addr = CALLER_ADDR2; | |
3483 | if (in_lock_functions(addr)) | |
3484 | addr = CALLER_ADDR3; | |
3485 | } | |
3486 | return addr; | |
3487 | } | |
1da177e4 | 3488 | |
7e49fcce SR |
3489 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3490 | defined(CONFIG_PREEMPT_TRACER)) | |
3491 | ||
43627582 | 3492 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3493 | { |
6cd8a4bb | 3494 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3495 | /* |
3496 | * Underflow? | |
3497 | */ | |
9a11b49a IM |
3498 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3499 | return; | |
6cd8a4bb | 3500 | #endif |
1da177e4 | 3501 | preempt_count() += val; |
6cd8a4bb | 3502 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3503 | /* |
3504 | * Spinlock count overflowing soon? | |
3505 | */ | |
33859f7f MOS |
3506 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3507 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3508 | #endif |
3509 | if (preempt_count() == val) | |
3510 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3511 | } |
3512 | EXPORT_SYMBOL(add_preempt_count); | |
3513 | ||
43627582 | 3514 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3515 | { |
6cd8a4bb | 3516 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3517 | /* |
3518 | * Underflow? | |
3519 | */ | |
01e3eb82 | 3520 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3521 | return; |
1da177e4 LT |
3522 | /* |
3523 | * Is the spinlock portion underflowing? | |
3524 | */ | |
9a11b49a IM |
3525 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3526 | !(preempt_count() & PREEMPT_MASK))) | |
3527 | return; | |
6cd8a4bb | 3528 | #endif |
9a11b49a | 3529 | |
6cd8a4bb SR |
3530 | if (preempt_count() == val) |
3531 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3532 | preempt_count() -= val; |
3533 | } | |
3534 | EXPORT_SYMBOL(sub_preempt_count); | |
3535 | ||
3536 | #endif | |
3537 | ||
3538 | /* | |
dd41f596 | 3539 | * Print scheduling while atomic bug: |
1da177e4 | 3540 | */ |
dd41f596 | 3541 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3542 | { |
838225b4 SS |
3543 | struct pt_regs *regs = get_irq_regs(); |
3544 | ||
3df0fc5b PZ |
3545 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3546 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3547 | |
dd41f596 | 3548 | debug_show_held_locks(prev); |
e21f5b15 | 3549 | print_modules(); |
dd41f596 IM |
3550 | if (irqs_disabled()) |
3551 | print_irqtrace_events(prev); | |
838225b4 SS |
3552 | |
3553 | if (regs) | |
3554 | show_regs(regs); | |
3555 | else | |
3556 | dump_stack(); | |
dd41f596 | 3557 | } |
1da177e4 | 3558 | |
dd41f596 IM |
3559 | /* |
3560 | * Various schedule()-time debugging checks and statistics: | |
3561 | */ | |
3562 | static inline void schedule_debug(struct task_struct *prev) | |
3563 | { | |
1da177e4 | 3564 | /* |
41a2d6cf | 3565 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3566 | * schedule() atomically, we ignore that path for now. |
3567 | * Otherwise, whine if we are scheduling when we should not be. | |
3568 | */ | |
3f33a7ce | 3569 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3570 | __schedule_bug(prev); |
3571 | ||
1da177e4 LT |
3572 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3573 | ||
2d72376b | 3574 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3575 | #ifdef CONFIG_SCHEDSTATS |
3576 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3577 | schedstat_inc(this_rq(), bkl_count); |
3578 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3579 | } |
3580 | #endif | |
dd41f596 IM |
3581 | } |
3582 | ||
6cecd084 | 3583 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3584 | { |
a64692a3 MG |
3585 | if (prev->se.on_rq) |
3586 | update_rq_clock(rq); | |
3587 | rq->skip_clock_update = 0; | |
6cecd084 | 3588 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3589 | } |
3590 | ||
dd41f596 IM |
3591 | /* |
3592 | * Pick up the highest-prio task: | |
3593 | */ | |
3594 | static inline struct task_struct * | |
b67802ea | 3595 | pick_next_task(struct rq *rq) |
dd41f596 | 3596 | { |
5522d5d5 | 3597 | const struct sched_class *class; |
dd41f596 | 3598 | struct task_struct *p; |
1da177e4 LT |
3599 | |
3600 | /* | |
dd41f596 IM |
3601 | * Optimization: we know that if all tasks are in |
3602 | * the fair class we can call that function directly: | |
1da177e4 | 3603 | */ |
dd41f596 | 3604 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3605 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3606 | if (likely(p)) |
3607 | return p; | |
1da177e4 LT |
3608 | } |
3609 | ||
dd41f596 IM |
3610 | class = sched_class_highest; |
3611 | for ( ; ; ) { | |
fb8d4724 | 3612 | p = class->pick_next_task(rq); |
dd41f596 IM |
3613 | if (p) |
3614 | return p; | |
3615 | /* | |
3616 | * Will never be NULL as the idle class always | |
3617 | * returns a non-NULL p: | |
3618 | */ | |
3619 | class = class->next; | |
3620 | } | |
3621 | } | |
1da177e4 | 3622 | |
dd41f596 IM |
3623 | /* |
3624 | * schedule() is the main scheduler function. | |
3625 | */ | |
ff743345 | 3626 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3627 | { |
3628 | struct task_struct *prev, *next; | |
67ca7bde | 3629 | unsigned long *switch_count; |
dd41f596 | 3630 | struct rq *rq; |
31656519 | 3631 | int cpu; |
dd41f596 | 3632 | |
ff743345 PZ |
3633 | need_resched: |
3634 | preempt_disable(); | |
dd41f596 IM |
3635 | cpu = smp_processor_id(); |
3636 | rq = cpu_rq(cpu); | |
d6714c22 | 3637 | rcu_sched_qs(cpu); |
dd41f596 IM |
3638 | prev = rq->curr; |
3639 | switch_count = &prev->nivcsw; | |
3640 | ||
3641 | release_kernel_lock(prev); | |
3642 | need_resched_nonpreemptible: | |
3643 | ||
3644 | schedule_debug(prev); | |
1da177e4 | 3645 | |
31656519 | 3646 | if (sched_feat(HRTICK)) |
f333fdc9 | 3647 | hrtick_clear(rq); |
8f4d37ec | 3648 | |
05fa785c | 3649 | raw_spin_lock_irq(&rq->lock); |
1e819950 | 3650 | clear_tsk_need_resched(prev); |
1da177e4 | 3651 | |
1da177e4 | 3652 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 3653 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 3654 | prev->state = TASK_RUNNING; |
16882c1e | 3655 | else |
2e1cb74a | 3656 | deactivate_task(rq, prev, 1); |
dd41f596 | 3657 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3658 | } |
3659 | ||
3f029d3c | 3660 | pre_schedule(rq, prev); |
f65eda4f | 3661 | |
dd41f596 | 3662 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3663 | idle_balance(cpu, rq); |
1da177e4 | 3664 | |
df1c99d4 | 3665 | put_prev_task(rq, prev); |
b67802ea | 3666 | next = pick_next_task(rq); |
1da177e4 | 3667 | |
1da177e4 | 3668 | if (likely(prev != next)) { |
673a90a1 | 3669 | sched_info_switch(prev, next); |
49f47433 | 3670 | perf_event_task_sched_out(prev, next); |
673a90a1 | 3671 | |
1da177e4 LT |
3672 | rq->nr_switches++; |
3673 | rq->curr = next; | |
3674 | ++*switch_count; | |
3675 | ||
dd41f596 | 3676 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
3677 | /* |
3678 | * the context switch might have flipped the stack from under | |
3679 | * us, hence refresh the local variables. | |
3680 | */ | |
3681 | cpu = smp_processor_id(); | |
3682 | rq = cpu_rq(cpu); | |
1da177e4 | 3683 | } else |
05fa785c | 3684 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3685 | |
3f029d3c | 3686 | post_schedule(rq); |
1da177e4 | 3687 | |
6d558c3a YZ |
3688 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3689 | prev = rq->curr; | |
3690 | switch_count = &prev->nivcsw; | |
1da177e4 | 3691 | goto need_resched_nonpreemptible; |
6d558c3a | 3692 | } |
8f4d37ec | 3693 | |
1da177e4 | 3694 | preempt_enable_no_resched(); |
ff743345 | 3695 | if (need_resched()) |
1da177e4 LT |
3696 | goto need_resched; |
3697 | } | |
1da177e4 LT |
3698 | EXPORT_SYMBOL(schedule); |
3699 | ||
c08f7829 | 3700 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
3701 | /* |
3702 | * Look out! "owner" is an entirely speculative pointer | |
3703 | * access and not reliable. | |
3704 | */ | |
3705 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
3706 | { | |
3707 | unsigned int cpu; | |
3708 | struct rq *rq; | |
3709 | ||
3710 | if (!sched_feat(OWNER_SPIN)) | |
3711 | return 0; | |
3712 | ||
3713 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3714 | /* | |
3715 | * Need to access the cpu field knowing that | |
3716 | * DEBUG_PAGEALLOC could have unmapped it if | |
3717 | * the mutex owner just released it and exited. | |
3718 | */ | |
3719 | if (probe_kernel_address(&owner->cpu, cpu)) | |
3720 | goto out; | |
3721 | #else | |
3722 | cpu = owner->cpu; | |
3723 | #endif | |
3724 | ||
3725 | /* | |
3726 | * Even if the access succeeded (likely case), | |
3727 | * the cpu field may no longer be valid. | |
3728 | */ | |
3729 | if (cpu >= nr_cpumask_bits) | |
3730 | goto out; | |
3731 | ||
3732 | /* | |
3733 | * We need to validate that we can do a | |
3734 | * get_cpu() and that we have the percpu area. | |
3735 | */ | |
3736 | if (!cpu_online(cpu)) | |
3737 | goto out; | |
3738 | ||
3739 | rq = cpu_rq(cpu); | |
3740 | ||
3741 | for (;;) { | |
3742 | /* | |
3743 | * Owner changed, break to re-assess state. | |
3744 | */ | |
3745 | if (lock->owner != owner) | |
3746 | break; | |
3747 | ||
3748 | /* | |
3749 | * Is that owner really running on that cpu? | |
3750 | */ | |
3751 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
3752 | return 0; | |
3753 | ||
3754 | cpu_relax(); | |
3755 | } | |
3756 | out: | |
3757 | return 1; | |
3758 | } | |
3759 | #endif | |
3760 | ||
1da177e4 LT |
3761 | #ifdef CONFIG_PREEMPT |
3762 | /* | |
2ed6e34f | 3763 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3764 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3765 | * occur there and call schedule directly. |
3766 | */ | |
3767 | asmlinkage void __sched preempt_schedule(void) | |
3768 | { | |
3769 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3770 | |
1da177e4 LT |
3771 | /* |
3772 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3773 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3774 | */ |
beed33a8 | 3775 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3776 | return; |
3777 | ||
3a5c359a AK |
3778 | do { |
3779 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3780 | schedule(); |
3a5c359a | 3781 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3782 | |
3a5c359a AK |
3783 | /* |
3784 | * Check again in case we missed a preemption opportunity | |
3785 | * between schedule and now. | |
3786 | */ | |
3787 | barrier(); | |
5ed0cec0 | 3788 | } while (need_resched()); |
1da177e4 | 3789 | } |
1da177e4 LT |
3790 | EXPORT_SYMBOL(preempt_schedule); |
3791 | ||
3792 | /* | |
2ed6e34f | 3793 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3794 | * off of irq context. |
3795 | * Note, that this is called and return with irqs disabled. This will | |
3796 | * protect us against recursive calling from irq. | |
3797 | */ | |
3798 | asmlinkage void __sched preempt_schedule_irq(void) | |
3799 | { | |
3800 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3801 | |
2ed6e34f | 3802 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3803 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3804 | ||
3a5c359a AK |
3805 | do { |
3806 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
3807 | local_irq_enable(); |
3808 | schedule(); | |
3809 | local_irq_disable(); | |
3a5c359a | 3810 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3811 | |
3a5c359a AK |
3812 | /* |
3813 | * Check again in case we missed a preemption opportunity | |
3814 | * between schedule and now. | |
3815 | */ | |
3816 | barrier(); | |
5ed0cec0 | 3817 | } while (need_resched()); |
1da177e4 LT |
3818 | } |
3819 | ||
3820 | #endif /* CONFIG_PREEMPT */ | |
3821 | ||
63859d4f | 3822 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3823 | void *key) |
1da177e4 | 3824 | { |
63859d4f | 3825 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3826 | } |
1da177e4 LT |
3827 | EXPORT_SYMBOL(default_wake_function); |
3828 | ||
3829 | /* | |
41a2d6cf IM |
3830 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3831 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3832 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3833 | * | |
3834 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3835 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3836 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3837 | */ | |
78ddb08f | 3838 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3839 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3840 | { |
2e45874c | 3841 | wait_queue_t *curr, *next; |
1da177e4 | 3842 | |
2e45874c | 3843 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3844 | unsigned flags = curr->flags; |
3845 | ||
63859d4f | 3846 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3847 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3848 | break; |
3849 | } | |
3850 | } | |
3851 | ||
3852 | /** | |
3853 | * __wake_up - wake up threads blocked on a waitqueue. | |
3854 | * @q: the waitqueue | |
3855 | * @mode: which threads | |
3856 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3857 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3858 | * |
3859 | * It may be assumed that this function implies a write memory barrier before | |
3860 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3861 | */ |
7ad5b3a5 | 3862 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3863 | int nr_exclusive, void *key) |
1da177e4 LT |
3864 | { |
3865 | unsigned long flags; | |
3866 | ||
3867 | spin_lock_irqsave(&q->lock, flags); | |
3868 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3869 | spin_unlock_irqrestore(&q->lock, flags); | |
3870 | } | |
1da177e4 LT |
3871 | EXPORT_SYMBOL(__wake_up); |
3872 | ||
3873 | /* | |
3874 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3875 | */ | |
7ad5b3a5 | 3876 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3877 | { |
3878 | __wake_up_common(q, mode, 1, 0, NULL); | |
3879 | } | |
3880 | ||
4ede816a DL |
3881 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3882 | { | |
3883 | __wake_up_common(q, mode, 1, 0, key); | |
3884 | } | |
3885 | ||
1da177e4 | 3886 | /** |
4ede816a | 3887 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3888 | * @q: the waitqueue |
3889 | * @mode: which threads | |
3890 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3891 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3892 | * |
3893 | * The sync wakeup differs that the waker knows that it will schedule | |
3894 | * away soon, so while the target thread will be woken up, it will not | |
3895 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3896 | * with each other. This can prevent needless bouncing between CPUs. | |
3897 | * | |
3898 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3899 | * |
3900 | * It may be assumed that this function implies a write memory barrier before | |
3901 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3902 | */ |
4ede816a DL |
3903 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3904 | int nr_exclusive, void *key) | |
1da177e4 LT |
3905 | { |
3906 | unsigned long flags; | |
7d478721 | 3907 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3908 | |
3909 | if (unlikely(!q)) | |
3910 | return; | |
3911 | ||
3912 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3913 | wake_flags = 0; |
1da177e4 LT |
3914 | |
3915 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3916 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3917 | spin_unlock_irqrestore(&q->lock, flags); |
3918 | } | |
4ede816a DL |
3919 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3920 | ||
3921 | /* | |
3922 | * __wake_up_sync - see __wake_up_sync_key() | |
3923 | */ | |
3924 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3925 | { | |
3926 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3927 | } | |
1da177e4 LT |
3928 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3929 | ||
65eb3dc6 KD |
3930 | /** |
3931 | * complete: - signals a single thread waiting on this completion | |
3932 | * @x: holds the state of this particular completion | |
3933 | * | |
3934 | * This will wake up a single thread waiting on this completion. Threads will be | |
3935 | * awakened in the same order in which they were queued. | |
3936 | * | |
3937 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3938 | * |
3939 | * It may be assumed that this function implies a write memory barrier before | |
3940 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3941 | */ |
b15136e9 | 3942 | void complete(struct completion *x) |
1da177e4 LT |
3943 | { |
3944 | unsigned long flags; | |
3945 | ||
3946 | spin_lock_irqsave(&x->wait.lock, flags); | |
3947 | x->done++; | |
d9514f6c | 3948 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3949 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3950 | } | |
3951 | EXPORT_SYMBOL(complete); | |
3952 | ||
65eb3dc6 KD |
3953 | /** |
3954 | * complete_all: - signals all threads waiting on this completion | |
3955 | * @x: holds the state of this particular completion | |
3956 | * | |
3957 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3958 | * |
3959 | * It may be assumed that this function implies a write memory barrier before | |
3960 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3961 | */ |
b15136e9 | 3962 | void complete_all(struct completion *x) |
1da177e4 LT |
3963 | { |
3964 | unsigned long flags; | |
3965 | ||
3966 | spin_lock_irqsave(&x->wait.lock, flags); | |
3967 | x->done += UINT_MAX/2; | |
d9514f6c | 3968 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3969 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3970 | } | |
3971 | EXPORT_SYMBOL(complete_all); | |
3972 | ||
8cbbe86d AK |
3973 | static inline long __sched |
3974 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3975 | { |
1da177e4 LT |
3976 | if (!x->done) { |
3977 | DECLARE_WAITQUEUE(wait, current); | |
3978 | ||
3979 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3980 | __add_wait_queue_tail(&x->wait, &wait); | |
3981 | do { | |
94d3d824 | 3982 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3983 | timeout = -ERESTARTSYS; |
3984 | break; | |
8cbbe86d AK |
3985 | } |
3986 | __set_current_state(state); | |
1da177e4 LT |
3987 | spin_unlock_irq(&x->wait.lock); |
3988 | timeout = schedule_timeout(timeout); | |
3989 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3990 | } while (!x->done && timeout); |
1da177e4 | 3991 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3992 | if (!x->done) |
3993 | return timeout; | |
1da177e4 LT |
3994 | } |
3995 | x->done--; | |
ea71a546 | 3996 | return timeout ?: 1; |
1da177e4 | 3997 | } |
1da177e4 | 3998 | |
8cbbe86d AK |
3999 | static long __sched |
4000 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4001 | { |
1da177e4 LT |
4002 | might_sleep(); |
4003 | ||
4004 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4005 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4006 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4007 | return timeout; |
4008 | } | |
1da177e4 | 4009 | |
65eb3dc6 KD |
4010 | /** |
4011 | * wait_for_completion: - waits for completion of a task | |
4012 | * @x: holds the state of this particular completion | |
4013 | * | |
4014 | * This waits to be signaled for completion of a specific task. It is NOT | |
4015 | * interruptible and there is no timeout. | |
4016 | * | |
4017 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4018 | * and interrupt capability. Also see complete(). | |
4019 | */ | |
b15136e9 | 4020 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4021 | { |
4022 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4023 | } |
8cbbe86d | 4024 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4025 | |
65eb3dc6 KD |
4026 | /** |
4027 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4028 | * @x: holds the state of this particular completion | |
4029 | * @timeout: timeout value in jiffies | |
4030 | * | |
4031 | * This waits for either a completion of a specific task to be signaled or for a | |
4032 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4033 | * interruptible. | |
4034 | */ | |
b15136e9 | 4035 | unsigned long __sched |
8cbbe86d | 4036 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4037 | { |
8cbbe86d | 4038 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4039 | } |
8cbbe86d | 4040 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4041 | |
65eb3dc6 KD |
4042 | /** |
4043 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4044 | * @x: holds the state of this particular completion | |
4045 | * | |
4046 | * This waits for completion of a specific task to be signaled. It is | |
4047 | * interruptible. | |
4048 | */ | |
8cbbe86d | 4049 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4050 | { |
51e97990 AK |
4051 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4052 | if (t == -ERESTARTSYS) | |
4053 | return t; | |
4054 | return 0; | |
0fec171c | 4055 | } |
8cbbe86d | 4056 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4057 | |
65eb3dc6 KD |
4058 | /** |
4059 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4060 | * @x: holds the state of this particular completion | |
4061 | * @timeout: timeout value in jiffies | |
4062 | * | |
4063 | * This waits for either a completion of a specific task to be signaled or for a | |
4064 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4065 | */ | |
b15136e9 | 4066 | unsigned long __sched |
8cbbe86d AK |
4067 | wait_for_completion_interruptible_timeout(struct completion *x, |
4068 | unsigned long timeout) | |
0fec171c | 4069 | { |
8cbbe86d | 4070 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4071 | } |
8cbbe86d | 4072 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4073 | |
65eb3dc6 KD |
4074 | /** |
4075 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4076 | * @x: holds the state of this particular completion | |
4077 | * | |
4078 | * This waits to be signaled for completion of a specific task. It can be | |
4079 | * interrupted by a kill signal. | |
4080 | */ | |
009e577e MW |
4081 | int __sched wait_for_completion_killable(struct completion *x) |
4082 | { | |
4083 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4084 | if (t == -ERESTARTSYS) | |
4085 | return t; | |
4086 | return 0; | |
4087 | } | |
4088 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4089 | ||
be4de352 DC |
4090 | /** |
4091 | * try_wait_for_completion - try to decrement a completion without blocking | |
4092 | * @x: completion structure | |
4093 | * | |
4094 | * Returns: 0 if a decrement cannot be done without blocking | |
4095 | * 1 if a decrement succeeded. | |
4096 | * | |
4097 | * If a completion is being used as a counting completion, | |
4098 | * attempt to decrement the counter without blocking. This | |
4099 | * enables us to avoid waiting if the resource the completion | |
4100 | * is protecting is not available. | |
4101 | */ | |
4102 | bool try_wait_for_completion(struct completion *x) | |
4103 | { | |
7539a3b3 | 4104 | unsigned long flags; |
be4de352 DC |
4105 | int ret = 1; |
4106 | ||
7539a3b3 | 4107 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4108 | if (!x->done) |
4109 | ret = 0; | |
4110 | else | |
4111 | x->done--; | |
7539a3b3 | 4112 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4113 | return ret; |
4114 | } | |
4115 | EXPORT_SYMBOL(try_wait_for_completion); | |
4116 | ||
4117 | /** | |
4118 | * completion_done - Test to see if a completion has any waiters | |
4119 | * @x: completion structure | |
4120 | * | |
4121 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4122 | * 1 if there are no waiters. | |
4123 | * | |
4124 | */ | |
4125 | bool completion_done(struct completion *x) | |
4126 | { | |
7539a3b3 | 4127 | unsigned long flags; |
be4de352 DC |
4128 | int ret = 1; |
4129 | ||
7539a3b3 | 4130 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4131 | if (!x->done) |
4132 | ret = 0; | |
7539a3b3 | 4133 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4134 | return ret; |
4135 | } | |
4136 | EXPORT_SYMBOL(completion_done); | |
4137 | ||
8cbbe86d AK |
4138 | static long __sched |
4139 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4140 | { |
0fec171c IM |
4141 | unsigned long flags; |
4142 | wait_queue_t wait; | |
4143 | ||
4144 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4145 | |
8cbbe86d | 4146 | __set_current_state(state); |
1da177e4 | 4147 | |
8cbbe86d AK |
4148 | spin_lock_irqsave(&q->lock, flags); |
4149 | __add_wait_queue(q, &wait); | |
4150 | spin_unlock(&q->lock); | |
4151 | timeout = schedule_timeout(timeout); | |
4152 | spin_lock_irq(&q->lock); | |
4153 | __remove_wait_queue(q, &wait); | |
4154 | spin_unlock_irqrestore(&q->lock, flags); | |
4155 | ||
4156 | return timeout; | |
4157 | } | |
4158 | ||
4159 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4160 | { | |
4161 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4162 | } |
1da177e4 LT |
4163 | EXPORT_SYMBOL(interruptible_sleep_on); |
4164 | ||
0fec171c | 4165 | long __sched |
95cdf3b7 | 4166 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4167 | { |
8cbbe86d | 4168 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4169 | } |
1da177e4 LT |
4170 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4171 | ||
0fec171c | 4172 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4173 | { |
8cbbe86d | 4174 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4175 | } |
1da177e4 LT |
4176 | EXPORT_SYMBOL(sleep_on); |
4177 | ||
0fec171c | 4178 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4179 | { |
8cbbe86d | 4180 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4181 | } |
1da177e4 LT |
4182 | EXPORT_SYMBOL(sleep_on_timeout); |
4183 | ||
b29739f9 IM |
4184 | #ifdef CONFIG_RT_MUTEXES |
4185 | ||
4186 | /* | |
4187 | * rt_mutex_setprio - set the current priority of a task | |
4188 | * @p: task | |
4189 | * @prio: prio value (kernel-internal form) | |
4190 | * | |
4191 | * This function changes the 'effective' priority of a task. It does | |
4192 | * not touch ->normal_prio like __setscheduler(). | |
4193 | * | |
4194 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4195 | */ | |
36c8b586 | 4196 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4197 | { |
4198 | unsigned long flags; | |
83b699ed | 4199 | int oldprio, on_rq, running; |
70b97a7f | 4200 | struct rq *rq; |
83ab0aa0 | 4201 | const struct sched_class *prev_class; |
b29739f9 IM |
4202 | |
4203 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4204 | ||
4205 | rq = task_rq_lock(p, &flags); | |
4206 | ||
d5f9f942 | 4207 | oldprio = p->prio; |
83ab0aa0 | 4208 | prev_class = p->sched_class; |
dd41f596 | 4209 | on_rq = p->se.on_rq; |
051a1d1a | 4210 | running = task_current(rq, p); |
0e1f3483 | 4211 | if (on_rq) |
69be72c1 | 4212 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4213 | if (running) |
4214 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4215 | |
4216 | if (rt_prio(prio)) | |
4217 | p->sched_class = &rt_sched_class; | |
4218 | else | |
4219 | p->sched_class = &fair_sched_class; | |
4220 | ||
b29739f9 IM |
4221 | p->prio = prio; |
4222 | ||
0e1f3483 HS |
4223 | if (running) |
4224 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4225 | if (on_rq) { |
60db48ca | 4226 | enqueue_task(rq, p, 0, oldprio < prio); |
cb469845 SR |
4227 | |
4228 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4229 | } |
4230 | task_rq_unlock(rq, &flags); | |
4231 | } | |
4232 | ||
4233 | #endif | |
4234 | ||
36c8b586 | 4235 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4236 | { |
dd41f596 | 4237 | int old_prio, delta, on_rq; |
1da177e4 | 4238 | unsigned long flags; |
70b97a7f | 4239 | struct rq *rq; |
1da177e4 LT |
4240 | |
4241 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4242 | return; | |
4243 | /* | |
4244 | * We have to be careful, if called from sys_setpriority(), | |
4245 | * the task might be in the middle of scheduling on another CPU. | |
4246 | */ | |
4247 | rq = task_rq_lock(p, &flags); | |
4248 | /* | |
4249 | * The RT priorities are set via sched_setscheduler(), but we still | |
4250 | * allow the 'normal' nice value to be set - but as expected | |
4251 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4252 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4253 | */ |
e05606d3 | 4254 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4255 | p->static_prio = NICE_TO_PRIO(nice); |
4256 | goto out_unlock; | |
4257 | } | |
dd41f596 | 4258 | on_rq = p->se.on_rq; |
c09595f6 | 4259 | if (on_rq) |
69be72c1 | 4260 | dequeue_task(rq, p, 0); |
1da177e4 | 4261 | |
1da177e4 | 4262 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4263 | set_load_weight(p); |
b29739f9 IM |
4264 | old_prio = p->prio; |
4265 | p->prio = effective_prio(p); | |
4266 | delta = p->prio - old_prio; | |
1da177e4 | 4267 | |
dd41f596 | 4268 | if (on_rq) { |
ea87bb78 | 4269 | enqueue_task(rq, p, 0, false); |
1da177e4 | 4270 | /* |
d5f9f942 AM |
4271 | * If the task increased its priority or is running and |
4272 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4273 | */ |
d5f9f942 | 4274 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4275 | resched_task(rq->curr); |
4276 | } | |
4277 | out_unlock: | |
4278 | task_rq_unlock(rq, &flags); | |
4279 | } | |
1da177e4 LT |
4280 | EXPORT_SYMBOL(set_user_nice); |
4281 | ||
e43379f1 MM |
4282 | /* |
4283 | * can_nice - check if a task can reduce its nice value | |
4284 | * @p: task | |
4285 | * @nice: nice value | |
4286 | */ | |
36c8b586 | 4287 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4288 | { |
024f4747 MM |
4289 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4290 | int nice_rlim = 20 - nice; | |
48f24c4d | 4291 | |
78d7d407 | 4292 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4293 | capable(CAP_SYS_NICE)); |
4294 | } | |
4295 | ||
1da177e4 LT |
4296 | #ifdef __ARCH_WANT_SYS_NICE |
4297 | ||
4298 | /* | |
4299 | * sys_nice - change the priority of the current process. | |
4300 | * @increment: priority increment | |
4301 | * | |
4302 | * sys_setpriority is a more generic, but much slower function that | |
4303 | * does similar things. | |
4304 | */ | |
5add95d4 | 4305 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4306 | { |
48f24c4d | 4307 | long nice, retval; |
1da177e4 LT |
4308 | |
4309 | /* | |
4310 | * Setpriority might change our priority at the same moment. | |
4311 | * We don't have to worry. Conceptually one call occurs first | |
4312 | * and we have a single winner. | |
4313 | */ | |
e43379f1 MM |
4314 | if (increment < -40) |
4315 | increment = -40; | |
1da177e4 LT |
4316 | if (increment > 40) |
4317 | increment = 40; | |
4318 | ||
2b8f836f | 4319 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4320 | if (nice < -20) |
4321 | nice = -20; | |
4322 | if (nice > 19) | |
4323 | nice = 19; | |
4324 | ||
e43379f1 MM |
4325 | if (increment < 0 && !can_nice(current, nice)) |
4326 | return -EPERM; | |
4327 | ||
1da177e4 LT |
4328 | retval = security_task_setnice(current, nice); |
4329 | if (retval) | |
4330 | return retval; | |
4331 | ||
4332 | set_user_nice(current, nice); | |
4333 | return 0; | |
4334 | } | |
4335 | ||
4336 | #endif | |
4337 | ||
4338 | /** | |
4339 | * task_prio - return the priority value of a given task. | |
4340 | * @p: the task in question. | |
4341 | * | |
4342 | * This is the priority value as seen by users in /proc. | |
4343 | * RT tasks are offset by -200. Normal tasks are centered | |
4344 | * around 0, value goes from -16 to +15. | |
4345 | */ | |
36c8b586 | 4346 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4347 | { |
4348 | return p->prio - MAX_RT_PRIO; | |
4349 | } | |
4350 | ||
4351 | /** | |
4352 | * task_nice - return the nice value of a given task. | |
4353 | * @p: the task in question. | |
4354 | */ | |
36c8b586 | 4355 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4356 | { |
4357 | return TASK_NICE(p); | |
4358 | } | |
150d8bed | 4359 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4360 | |
4361 | /** | |
4362 | * idle_cpu - is a given cpu idle currently? | |
4363 | * @cpu: the processor in question. | |
4364 | */ | |
4365 | int idle_cpu(int cpu) | |
4366 | { | |
4367 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4368 | } | |
4369 | ||
1da177e4 LT |
4370 | /** |
4371 | * idle_task - return the idle task for a given cpu. | |
4372 | * @cpu: the processor in question. | |
4373 | */ | |
36c8b586 | 4374 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4375 | { |
4376 | return cpu_rq(cpu)->idle; | |
4377 | } | |
4378 | ||
4379 | /** | |
4380 | * find_process_by_pid - find a process with a matching PID value. | |
4381 | * @pid: the pid in question. | |
4382 | */ | |
a9957449 | 4383 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4384 | { |
228ebcbe | 4385 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4386 | } |
4387 | ||
4388 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4389 | static void |
4390 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4391 | { |
dd41f596 | 4392 | BUG_ON(p->se.on_rq); |
48f24c4d | 4393 | |
1da177e4 LT |
4394 | p->policy = policy; |
4395 | p->rt_priority = prio; | |
b29739f9 IM |
4396 | p->normal_prio = normal_prio(p); |
4397 | /* we are holding p->pi_lock already */ | |
4398 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4399 | if (rt_prio(p->prio)) |
4400 | p->sched_class = &rt_sched_class; | |
4401 | else | |
4402 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4403 | set_load_weight(p); |
1da177e4 LT |
4404 | } |
4405 | ||
c69e8d9c DH |
4406 | /* |
4407 | * check the target process has a UID that matches the current process's | |
4408 | */ | |
4409 | static bool check_same_owner(struct task_struct *p) | |
4410 | { | |
4411 | const struct cred *cred = current_cred(), *pcred; | |
4412 | bool match; | |
4413 | ||
4414 | rcu_read_lock(); | |
4415 | pcred = __task_cred(p); | |
4416 | match = (cred->euid == pcred->euid || | |
4417 | cred->euid == pcred->uid); | |
4418 | rcu_read_unlock(); | |
4419 | return match; | |
4420 | } | |
4421 | ||
961ccddd RR |
4422 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4423 | struct sched_param *param, bool user) | |
1da177e4 | 4424 | { |
83b699ed | 4425 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4426 | unsigned long flags; |
83ab0aa0 | 4427 | const struct sched_class *prev_class; |
70b97a7f | 4428 | struct rq *rq; |
ca94c442 | 4429 | int reset_on_fork; |
1da177e4 | 4430 | |
66e5393a SR |
4431 | /* may grab non-irq protected spin_locks */ |
4432 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4433 | recheck: |
4434 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4435 | if (policy < 0) { |
4436 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4437 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4438 | } else { |
4439 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4440 | policy &= ~SCHED_RESET_ON_FORK; | |
4441 | ||
4442 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4443 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4444 | policy != SCHED_IDLE) | |
4445 | return -EINVAL; | |
4446 | } | |
4447 | ||
1da177e4 LT |
4448 | /* |
4449 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4450 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4451 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4452 | */ |
4453 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4454 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4455 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4456 | return -EINVAL; |
e05606d3 | 4457 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4458 | return -EINVAL; |
4459 | ||
37e4ab3f OC |
4460 | /* |
4461 | * Allow unprivileged RT tasks to decrease priority: | |
4462 | */ | |
961ccddd | 4463 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4464 | if (rt_policy(policy)) { |
8dc3e909 | 4465 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4466 | |
4467 | if (!lock_task_sighand(p, &flags)) | |
4468 | return -ESRCH; | |
78d7d407 | 4469 | rlim_rtprio = task_rlimit(p, RLIMIT_RTPRIO); |
8dc3e909 ON |
4470 | unlock_task_sighand(p, &flags); |
4471 | ||
4472 | /* can't set/change the rt policy */ | |
4473 | if (policy != p->policy && !rlim_rtprio) | |
4474 | return -EPERM; | |
4475 | ||
4476 | /* can't increase priority */ | |
4477 | if (param->sched_priority > p->rt_priority && | |
4478 | param->sched_priority > rlim_rtprio) | |
4479 | return -EPERM; | |
4480 | } | |
dd41f596 IM |
4481 | /* |
4482 | * Like positive nice levels, dont allow tasks to | |
4483 | * move out of SCHED_IDLE either: | |
4484 | */ | |
4485 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4486 | return -EPERM; | |
5fe1d75f | 4487 | |
37e4ab3f | 4488 | /* can't change other user's priorities */ |
c69e8d9c | 4489 | if (!check_same_owner(p)) |
37e4ab3f | 4490 | return -EPERM; |
ca94c442 LP |
4491 | |
4492 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4493 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4494 | return -EPERM; | |
37e4ab3f | 4495 | } |
1da177e4 | 4496 | |
725aad24 | 4497 | if (user) { |
b68aa230 | 4498 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
4499 | /* |
4500 | * Do not allow realtime tasks into groups that have no runtime | |
4501 | * assigned. | |
4502 | */ | |
9a7e0b18 PZ |
4503 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
4504 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 4505 | return -EPERM; |
b68aa230 PZ |
4506 | #endif |
4507 | ||
725aad24 JF |
4508 | retval = security_task_setscheduler(p, policy, param); |
4509 | if (retval) | |
4510 | return retval; | |
4511 | } | |
4512 | ||
b29739f9 IM |
4513 | /* |
4514 | * make sure no PI-waiters arrive (or leave) while we are | |
4515 | * changing the priority of the task: | |
4516 | */ | |
1d615482 | 4517 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4518 | /* |
4519 | * To be able to change p->policy safely, the apropriate | |
4520 | * runqueue lock must be held. | |
4521 | */ | |
b29739f9 | 4522 | rq = __task_rq_lock(p); |
1da177e4 LT |
4523 | /* recheck policy now with rq lock held */ |
4524 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4525 | policy = oldpolicy = -1; | |
b29739f9 | 4526 | __task_rq_unlock(rq); |
1d615482 | 4527 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4528 | goto recheck; |
4529 | } | |
dd41f596 | 4530 | on_rq = p->se.on_rq; |
051a1d1a | 4531 | running = task_current(rq, p); |
0e1f3483 | 4532 | if (on_rq) |
2e1cb74a | 4533 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4534 | if (running) |
4535 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4536 | |
ca94c442 LP |
4537 | p->sched_reset_on_fork = reset_on_fork; |
4538 | ||
1da177e4 | 4539 | oldprio = p->prio; |
83ab0aa0 | 4540 | prev_class = p->sched_class; |
dd41f596 | 4541 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4542 | |
0e1f3483 HS |
4543 | if (running) |
4544 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4545 | if (on_rq) { |
4546 | activate_task(rq, p, 0); | |
cb469845 SR |
4547 | |
4548 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4549 | } |
b29739f9 | 4550 | __task_rq_unlock(rq); |
1d615482 | 4551 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4552 | |
95e02ca9 TG |
4553 | rt_mutex_adjust_pi(p); |
4554 | ||
1da177e4 LT |
4555 | return 0; |
4556 | } | |
961ccddd RR |
4557 | |
4558 | /** | |
4559 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4560 | * @p: the task in question. | |
4561 | * @policy: new policy. | |
4562 | * @param: structure containing the new RT priority. | |
4563 | * | |
4564 | * NOTE that the task may be already dead. | |
4565 | */ | |
4566 | int sched_setscheduler(struct task_struct *p, int policy, | |
4567 | struct sched_param *param) | |
4568 | { | |
4569 | return __sched_setscheduler(p, policy, param, true); | |
4570 | } | |
1da177e4 LT |
4571 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4572 | ||
961ccddd RR |
4573 | /** |
4574 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4575 | * @p: the task in question. | |
4576 | * @policy: new policy. | |
4577 | * @param: structure containing the new RT priority. | |
4578 | * | |
4579 | * Just like sched_setscheduler, only don't bother checking if the | |
4580 | * current context has permission. For example, this is needed in | |
4581 | * stop_machine(): we create temporary high priority worker threads, | |
4582 | * but our caller might not have that capability. | |
4583 | */ | |
4584 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
4585 | struct sched_param *param) | |
4586 | { | |
4587 | return __sched_setscheduler(p, policy, param, false); | |
4588 | } | |
4589 | ||
95cdf3b7 IM |
4590 | static int |
4591 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4592 | { |
1da177e4 LT |
4593 | struct sched_param lparam; |
4594 | struct task_struct *p; | |
36c8b586 | 4595 | int retval; |
1da177e4 LT |
4596 | |
4597 | if (!param || pid < 0) | |
4598 | return -EINVAL; | |
4599 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4600 | return -EFAULT; | |
5fe1d75f ON |
4601 | |
4602 | rcu_read_lock(); | |
4603 | retval = -ESRCH; | |
1da177e4 | 4604 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4605 | if (p != NULL) |
4606 | retval = sched_setscheduler(p, policy, &lparam); | |
4607 | rcu_read_unlock(); | |
36c8b586 | 4608 | |
1da177e4 LT |
4609 | return retval; |
4610 | } | |
4611 | ||
4612 | /** | |
4613 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4614 | * @pid: the pid in question. | |
4615 | * @policy: new policy. | |
4616 | * @param: structure containing the new RT priority. | |
4617 | */ | |
5add95d4 HC |
4618 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4619 | struct sched_param __user *, param) | |
1da177e4 | 4620 | { |
c21761f1 JB |
4621 | /* negative values for policy are not valid */ |
4622 | if (policy < 0) | |
4623 | return -EINVAL; | |
4624 | ||
1da177e4 LT |
4625 | return do_sched_setscheduler(pid, policy, param); |
4626 | } | |
4627 | ||
4628 | /** | |
4629 | * sys_sched_setparam - set/change the RT priority of a thread | |
4630 | * @pid: the pid in question. | |
4631 | * @param: structure containing the new RT priority. | |
4632 | */ | |
5add95d4 | 4633 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4634 | { |
4635 | return do_sched_setscheduler(pid, -1, param); | |
4636 | } | |
4637 | ||
4638 | /** | |
4639 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4640 | * @pid: the pid in question. | |
4641 | */ | |
5add95d4 | 4642 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4643 | { |
36c8b586 | 4644 | struct task_struct *p; |
3a5c359a | 4645 | int retval; |
1da177e4 LT |
4646 | |
4647 | if (pid < 0) | |
3a5c359a | 4648 | return -EINVAL; |
1da177e4 LT |
4649 | |
4650 | retval = -ESRCH; | |
5fe85be0 | 4651 | rcu_read_lock(); |
1da177e4 LT |
4652 | p = find_process_by_pid(pid); |
4653 | if (p) { | |
4654 | retval = security_task_getscheduler(p); | |
4655 | if (!retval) | |
ca94c442 LP |
4656 | retval = p->policy |
4657 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4658 | } |
5fe85be0 | 4659 | rcu_read_unlock(); |
1da177e4 LT |
4660 | return retval; |
4661 | } | |
4662 | ||
4663 | /** | |
ca94c442 | 4664 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4665 | * @pid: the pid in question. |
4666 | * @param: structure containing the RT priority. | |
4667 | */ | |
5add95d4 | 4668 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4669 | { |
4670 | struct sched_param lp; | |
36c8b586 | 4671 | struct task_struct *p; |
3a5c359a | 4672 | int retval; |
1da177e4 LT |
4673 | |
4674 | if (!param || pid < 0) | |
3a5c359a | 4675 | return -EINVAL; |
1da177e4 | 4676 | |
5fe85be0 | 4677 | rcu_read_lock(); |
1da177e4 LT |
4678 | p = find_process_by_pid(pid); |
4679 | retval = -ESRCH; | |
4680 | if (!p) | |
4681 | goto out_unlock; | |
4682 | ||
4683 | retval = security_task_getscheduler(p); | |
4684 | if (retval) | |
4685 | goto out_unlock; | |
4686 | ||
4687 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4688 | rcu_read_unlock(); |
1da177e4 LT |
4689 | |
4690 | /* | |
4691 | * This one might sleep, we cannot do it with a spinlock held ... | |
4692 | */ | |
4693 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4694 | ||
1da177e4 LT |
4695 | return retval; |
4696 | ||
4697 | out_unlock: | |
5fe85be0 | 4698 | rcu_read_unlock(); |
1da177e4 LT |
4699 | return retval; |
4700 | } | |
4701 | ||
96f874e2 | 4702 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4703 | { |
5a16f3d3 | 4704 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4705 | struct task_struct *p; |
4706 | int retval; | |
1da177e4 | 4707 | |
95402b38 | 4708 | get_online_cpus(); |
23f5d142 | 4709 | rcu_read_lock(); |
1da177e4 LT |
4710 | |
4711 | p = find_process_by_pid(pid); | |
4712 | if (!p) { | |
23f5d142 | 4713 | rcu_read_unlock(); |
95402b38 | 4714 | put_online_cpus(); |
1da177e4 LT |
4715 | return -ESRCH; |
4716 | } | |
4717 | ||
23f5d142 | 4718 | /* Prevent p going away */ |
1da177e4 | 4719 | get_task_struct(p); |
23f5d142 | 4720 | rcu_read_unlock(); |
1da177e4 | 4721 | |
5a16f3d3 RR |
4722 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4723 | retval = -ENOMEM; | |
4724 | goto out_put_task; | |
4725 | } | |
4726 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4727 | retval = -ENOMEM; | |
4728 | goto out_free_cpus_allowed; | |
4729 | } | |
1da177e4 | 4730 | retval = -EPERM; |
c69e8d9c | 4731 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
4732 | goto out_unlock; |
4733 | ||
e7834f8f DQ |
4734 | retval = security_task_setscheduler(p, 0, NULL); |
4735 | if (retval) | |
4736 | goto out_unlock; | |
4737 | ||
5a16f3d3 RR |
4738 | cpuset_cpus_allowed(p, cpus_allowed); |
4739 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 4740 | again: |
5a16f3d3 | 4741 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4742 | |
8707d8b8 | 4743 | if (!retval) { |
5a16f3d3 RR |
4744 | cpuset_cpus_allowed(p, cpus_allowed); |
4745 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4746 | /* |
4747 | * We must have raced with a concurrent cpuset | |
4748 | * update. Just reset the cpus_allowed to the | |
4749 | * cpuset's cpus_allowed | |
4750 | */ | |
5a16f3d3 | 4751 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4752 | goto again; |
4753 | } | |
4754 | } | |
1da177e4 | 4755 | out_unlock: |
5a16f3d3 RR |
4756 | free_cpumask_var(new_mask); |
4757 | out_free_cpus_allowed: | |
4758 | free_cpumask_var(cpus_allowed); | |
4759 | out_put_task: | |
1da177e4 | 4760 | put_task_struct(p); |
95402b38 | 4761 | put_online_cpus(); |
1da177e4 LT |
4762 | return retval; |
4763 | } | |
4764 | ||
4765 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4766 | struct cpumask *new_mask) |
1da177e4 | 4767 | { |
96f874e2 RR |
4768 | if (len < cpumask_size()) |
4769 | cpumask_clear(new_mask); | |
4770 | else if (len > cpumask_size()) | |
4771 | len = cpumask_size(); | |
4772 | ||
1da177e4 LT |
4773 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4774 | } | |
4775 | ||
4776 | /** | |
4777 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4778 | * @pid: pid of the process | |
4779 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4780 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4781 | */ | |
5add95d4 HC |
4782 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4783 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4784 | { |
5a16f3d3 | 4785 | cpumask_var_t new_mask; |
1da177e4 LT |
4786 | int retval; |
4787 | ||
5a16f3d3 RR |
4788 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4789 | return -ENOMEM; | |
1da177e4 | 4790 | |
5a16f3d3 RR |
4791 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4792 | if (retval == 0) | |
4793 | retval = sched_setaffinity(pid, new_mask); | |
4794 | free_cpumask_var(new_mask); | |
4795 | return retval; | |
1da177e4 LT |
4796 | } |
4797 | ||
96f874e2 | 4798 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4799 | { |
36c8b586 | 4800 | struct task_struct *p; |
31605683 TG |
4801 | unsigned long flags; |
4802 | struct rq *rq; | |
1da177e4 | 4803 | int retval; |
1da177e4 | 4804 | |
95402b38 | 4805 | get_online_cpus(); |
23f5d142 | 4806 | rcu_read_lock(); |
1da177e4 LT |
4807 | |
4808 | retval = -ESRCH; | |
4809 | p = find_process_by_pid(pid); | |
4810 | if (!p) | |
4811 | goto out_unlock; | |
4812 | ||
e7834f8f DQ |
4813 | retval = security_task_getscheduler(p); |
4814 | if (retval) | |
4815 | goto out_unlock; | |
4816 | ||
31605683 | 4817 | rq = task_rq_lock(p, &flags); |
96f874e2 | 4818 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 4819 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
4820 | |
4821 | out_unlock: | |
23f5d142 | 4822 | rcu_read_unlock(); |
95402b38 | 4823 | put_online_cpus(); |
1da177e4 | 4824 | |
9531b62f | 4825 | return retval; |
1da177e4 LT |
4826 | } |
4827 | ||
4828 | /** | |
4829 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4830 | * @pid: pid of the process | |
4831 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4832 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4833 | */ | |
5add95d4 HC |
4834 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4835 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4836 | { |
4837 | int ret; | |
f17c8607 | 4838 | cpumask_var_t mask; |
1da177e4 | 4839 | |
cd3d8031 KM |
4840 | if (len < nr_cpu_ids) |
4841 | return -EINVAL; | |
4842 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4843 | return -EINVAL; |
4844 | ||
f17c8607 RR |
4845 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4846 | return -ENOMEM; | |
1da177e4 | 4847 | |
f17c8607 RR |
4848 | ret = sched_getaffinity(pid, mask); |
4849 | if (ret == 0) { | |
8bc037fb | 4850 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4851 | |
4852 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4853 | ret = -EFAULT; |
4854 | else | |
cd3d8031 | 4855 | ret = retlen; |
f17c8607 RR |
4856 | } |
4857 | free_cpumask_var(mask); | |
1da177e4 | 4858 | |
f17c8607 | 4859 | return ret; |
1da177e4 LT |
4860 | } |
4861 | ||
4862 | /** | |
4863 | * sys_sched_yield - yield the current processor to other threads. | |
4864 | * | |
dd41f596 IM |
4865 | * This function yields the current CPU to other tasks. If there are no |
4866 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4867 | */ |
5add95d4 | 4868 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4869 | { |
70b97a7f | 4870 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4871 | |
2d72376b | 4872 | schedstat_inc(rq, yld_count); |
4530d7ab | 4873 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4874 | |
4875 | /* | |
4876 | * Since we are going to call schedule() anyway, there's | |
4877 | * no need to preempt or enable interrupts: | |
4878 | */ | |
4879 | __release(rq->lock); | |
8a25d5de | 4880 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4881 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
4882 | preempt_enable_no_resched(); |
4883 | ||
4884 | schedule(); | |
4885 | ||
4886 | return 0; | |
4887 | } | |
4888 | ||
d86ee480 PZ |
4889 | static inline int should_resched(void) |
4890 | { | |
4891 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4892 | } | |
4893 | ||
e7b38404 | 4894 | static void __cond_resched(void) |
1da177e4 | 4895 | { |
e7aaaa69 FW |
4896 | add_preempt_count(PREEMPT_ACTIVE); |
4897 | schedule(); | |
4898 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
4899 | } |
4900 | ||
02b67cc3 | 4901 | int __sched _cond_resched(void) |
1da177e4 | 4902 | { |
d86ee480 | 4903 | if (should_resched()) { |
1da177e4 LT |
4904 | __cond_resched(); |
4905 | return 1; | |
4906 | } | |
4907 | return 0; | |
4908 | } | |
02b67cc3 | 4909 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4910 | |
4911 | /* | |
613afbf8 | 4912 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4913 | * call schedule, and on return reacquire the lock. |
4914 | * | |
41a2d6cf | 4915 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4916 | * operations here to prevent schedule() from being called twice (once via |
4917 | * spin_unlock(), once by hand). | |
4918 | */ | |
613afbf8 | 4919 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4920 | { |
d86ee480 | 4921 | int resched = should_resched(); |
6df3cecb JK |
4922 | int ret = 0; |
4923 | ||
f607c668 PZ |
4924 | lockdep_assert_held(lock); |
4925 | ||
95c354fe | 4926 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4927 | spin_unlock(lock); |
d86ee480 | 4928 | if (resched) |
95c354fe NP |
4929 | __cond_resched(); |
4930 | else | |
4931 | cpu_relax(); | |
6df3cecb | 4932 | ret = 1; |
1da177e4 | 4933 | spin_lock(lock); |
1da177e4 | 4934 | } |
6df3cecb | 4935 | return ret; |
1da177e4 | 4936 | } |
613afbf8 | 4937 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4938 | |
613afbf8 | 4939 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4940 | { |
4941 | BUG_ON(!in_softirq()); | |
4942 | ||
d86ee480 | 4943 | if (should_resched()) { |
98d82567 | 4944 | local_bh_enable(); |
1da177e4 LT |
4945 | __cond_resched(); |
4946 | local_bh_disable(); | |
4947 | return 1; | |
4948 | } | |
4949 | return 0; | |
4950 | } | |
613afbf8 | 4951 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4952 | |
1da177e4 LT |
4953 | /** |
4954 | * yield - yield the current processor to other threads. | |
4955 | * | |
72fd4a35 | 4956 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4957 | * thread runnable and calls sys_sched_yield(). |
4958 | */ | |
4959 | void __sched yield(void) | |
4960 | { | |
4961 | set_current_state(TASK_RUNNING); | |
4962 | sys_sched_yield(); | |
4963 | } | |
1da177e4 LT |
4964 | EXPORT_SYMBOL(yield); |
4965 | ||
4966 | /* | |
41a2d6cf | 4967 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4968 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4969 | */ |
4970 | void __sched io_schedule(void) | |
4971 | { | |
54d35f29 | 4972 | struct rq *rq = raw_rq(); |
1da177e4 | 4973 | |
0ff92245 | 4974 | delayacct_blkio_start(); |
1da177e4 | 4975 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 4976 | current->in_iowait = 1; |
1da177e4 | 4977 | schedule(); |
8f0dfc34 | 4978 | current->in_iowait = 0; |
1da177e4 | 4979 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4980 | delayacct_blkio_end(); |
1da177e4 | 4981 | } |
1da177e4 LT |
4982 | EXPORT_SYMBOL(io_schedule); |
4983 | ||
4984 | long __sched io_schedule_timeout(long timeout) | |
4985 | { | |
54d35f29 | 4986 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4987 | long ret; |
4988 | ||
0ff92245 | 4989 | delayacct_blkio_start(); |
1da177e4 | 4990 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 4991 | current->in_iowait = 1; |
1da177e4 | 4992 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4993 | current->in_iowait = 0; |
1da177e4 | 4994 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4995 | delayacct_blkio_end(); |
1da177e4 LT |
4996 | return ret; |
4997 | } | |
4998 | ||
4999 | /** | |
5000 | * sys_sched_get_priority_max - return maximum RT priority. | |
5001 | * @policy: scheduling class. | |
5002 | * | |
5003 | * this syscall returns the maximum rt_priority that can be used | |
5004 | * by a given scheduling class. | |
5005 | */ | |
5add95d4 | 5006 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5007 | { |
5008 | int ret = -EINVAL; | |
5009 | ||
5010 | switch (policy) { | |
5011 | case SCHED_FIFO: | |
5012 | case SCHED_RR: | |
5013 | ret = MAX_USER_RT_PRIO-1; | |
5014 | break; | |
5015 | case SCHED_NORMAL: | |
b0a9499c | 5016 | case SCHED_BATCH: |
dd41f596 | 5017 | case SCHED_IDLE: |
1da177e4 LT |
5018 | ret = 0; |
5019 | break; | |
5020 | } | |
5021 | return ret; | |
5022 | } | |
5023 | ||
5024 | /** | |
5025 | * sys_sched_get_priority_min - return minimum RT priority. | |
5026 | * @policy: scheduling class. | |
5027 | * | |
5028 | * this syscall returns the minimum rt_priority that can be used | |
5029 | * by a given scheduling class. | |
5030 | */ | |
5add95d4 | 5031 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5032 | { |
5033 | int ret = -EINVAL; | |
5034 | ||
5035 | switch (policy) { | |
5036 | case SCHED_FIFO: | |
5037 | case SCHED_RR: | |
5038 | ret = 1; | |
5039 | break; | |
5040 | case SCHED_NORMAL: | |
b0a9499c | 5041 | case SCHED_BATCH: |
dd41f596 | 5042 | case SCHED_IDLE: |
1da177e4 LT |
5043 | ret = 0; |
5044 | } | |
5045 | return ret; | |
5046 | } | |
5047 | ||
5048 | /** | |
5049 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5050 | * @pid: pid of the process. | |
5051 | * @interval: userspace pointer to the timeslice value. | |
5052 | * | |
5053 | * this syscall writes the default timeslice value of a given process | |
5054 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5055 | */ | |
17da2bd9 | 5056 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5057 | struct timespec __user *, interval) |
1da177e4 | 5058 | { |
36c8b586 | 5059 | struct task_struct *p; |
a4ec24b4 | 5060 | unsigned int time_slice; |
dba091b9 TG |
5061 | unsigned long flags; |
5062 | struct rq *rq; | |
3a5c359a | 5063 | int retval; |
1da177e4 | 5064 | struct timespec t; |
1da177e4 LT |
5065 | |
5066 | if (pid < 0) | |
3a5c359a | 5067 | return -EINVAL; |
1da177e4 LT |
5068 | |
5069 | retval = -ESRCH; | |
1a551ae7 | 5070 | rcu_read_lock(); |
1da177e4 LT |
5071 | p = find_process_by_pid(pid); |
5072 | if (!p) | |
5073 | goto out_unlock; | |
5074 | ||
5075 | retval = security_task_getscheduler(p); | |
5076 | if (retval) | |
5077 | goto out_unlock; | |
5078 | ||
dba091b9 TG |
5079 | rq = task_rq_lock(p, &flags); |
5080 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5081 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5082 | |
1a551ae7 | 5083 | rcu_read_unlock(); |
a4ec24b4 | 5084 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5085 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5086 | return retval; |
3a5c359a | 5087 | |
1da177e4 | 5088 | out_unlock: |
1a551ae7 | 5089 | rcu_read_unlock(); |
1da177e4 LT |
5090 | return retval; |
5091 | } | |
5092 | ||
7c731e0a | 5093 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5094 | |
82a1fcb9 | 5095 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5096 | { |
1da177e4 | 5097 | unsigned long free = 0; |
36c8b586 | 5098 | unsigned state; |
1da177e4 | 5099 | |
1da177e4 | 5100 | state = p->state ? __ffs(p->state) + 1 : 0; |
3df0fc5b | 5101 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5102 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5103 | #if BITS_PER_LONG == 32 |
1da177e4 | 5104 | if (state == TASK_RUNNING) |
3df0fc5b | 5105 | printk(KERN_CONT " running "); |
1da177e4 | 5106 | else |
3df0fc5b | 5107 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5108 | #else |
5109 | if (state == TASK_RUNNING) | |
3df0fc5b | 5110 | printk(KERN_CONT " running task "); |
1da177e4 | 5111 | else |
3df0fc5b | 5112 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5113 | #endif |
5114 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5115 | free = stack_not_used(p); |
1da177e4 | 5116 | #endif |
3df0fc5b | 5117 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5118 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5119 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5120 | |
5fb5e6de | 5121 | show_stack(p, NULL); |
1da177e4 LT |
5122 | } |
5123 | ||
e59e2ae2 | 5124 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5125 | { |
36c8b586 | 5126 | struct task_struct *g, *p; |
1da177e4 | 5127 | |
4bd77321 | 5128 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5129 | printk(KERN_INFO |
5130 | " task PC stack pid father\n"); | |
1da177e4 | 5131 | #else |
3df0fc5b PZ |
5132 | printk(KERN_INFO |
5133 | " task PC stack pid father\n"); | |
1da177e4 LT |
5134 | #endif |
5135 | read_lock(&tasklist_lock); | |
5136 | do_each_thread(g, p) { | |
5137 | /* | |
5138 | * reset the NMI-timeout, listing all files on a slow | |
5139 | * console might take alot of time: | |
5140 | */ | |
5141 | touch_nmi_watchdog(); | |
39bc89fd | 5142 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5143 | sched_show_task(p); |
1da177e4 LT |
5144 | } while_each_thread(g, p); |
5145 | ||
04c9167f JF |
5146 | touch_all_softlockup_watchdogs(); |
5147 | ||
dd41f596 IM |
5148 | #ifdef CONFIG_SCHED_DEBUG |
5149 | sysrq_sched_debug_show(); | |
5150 | #endif | |
1da177e4 | 5151 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5152 | /* |
5153 | * Only show locks if all tasks are dumped: | |
5154 | */ | |
93335a21 | 5155 | if (!state_filter) |
e59e2ae2 | 5156 | debug_show_all_locks(); |
1da177e4 LT |
5157 | } |
5158 | ||
1df21055 IM |
5159 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5160 | { | |
dd41f596 | 5161 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5162 | } |
5163 | ||
f340c0d1 IM |
5164 | /** |
5165 | * init_idle - set up an idle thread for a given CPU | |
5166 | * @idle: task in question | |
5167 | * @cpu: cpu the idle task belongs to | |
5168 | * | |
5169 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5170 | * flag, to make booting more robust. | |
5171 | */ | |
5c1e1767 | 5172 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5173 | { |
70b97a7f | 5174 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5175 | unsigned long flags; |
5176 | ||
05fa785c | 5177 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5178 | |
dd41f596 | 5179 | __sched_fork(idle); |
06b83b5f | 5180 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5181 | idle->se.exec_start = sched_clock(); |
5182 | ||
96f874e2 | 5183 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 5184 | __set_task_cpu(idle, cpu); |
1da177e4 | 5185 | |
1da177e4 | 5186 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5187 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5188 | idle->oncpu = 1; | |
5189 | #endif | |
05fa785c | 5190 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5191 | |
5192 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5193 | #if defined(CONFIG_PREEMPT) |
5194 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5195 | #else | |
a1261f54 | 5196 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5197 | #endif |
dd41f596 IM |
5198 | /* |
5199 | * The idle tasks have their own, simple scheduling class: | |
5200 | */ | |
5201 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5202 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5203 | } |
5204 | ||
5205 | /* | |
5206 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5207 | * indicates which cpus entered this state. This is used | |
5208 | * in the rcu update to wait only for active cpus. For system | |
5209 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5210 | * always be CPU_BITS_NONE. |
1da177e4 | 5211 | */ |
6a7b3dc3 | 5212 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5213 | |
19978ca6 IM |
5214 | /* |
5215 | * Increase the granularity value when there are more CPUs, | |
5216 | * because with more CPUs the 'effective latency' as visible | |
5217 | * to users decreases. But the relationship is not linear, | |
5218 | * so pick a second-best guess by going with the log2 of the | |
5219 | * number of CPUs. | |
5220 | * | |
5221 | * This idea comes from the SD scheduler of Con Kolivas: | |
5222 | */ | |
acb4a848 | 5223 | static int get_update_sysctl_factor(void) |
19978ca6 | 5224 | { |
4ca3ef71 | 5225 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5226 | unsigned int factor; |
5227 | ||
5228 | switch (sysctl_sched_tunable_scaling) { | |
5229 | case SCHED_TUNABLESCALING_NONE: | |
5230 | factor = 1; | |
5231 | break; | |
5232 | case SCHED_TUNABLESCALING_LINEAR: | |
5233 | factor = cpus; | |
5234 | break; | |
5235 | case SCHED_TUNABLESCALING_LOG: | |
5236 | default: | |
5237 | factor = 1 + ilog2(cpus); | |
5238 | break; | |
5239 | } | |
19978ca6 | 5240 | |
acb4a848 CE |
5241 | return factor; |
5242 | } | |
19978ca6 | 5243 | |
acb4a848 CE |
5244 | static void update_sysctl(void) |
5245 | { | |
5246 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5247 | |
0bcdcf28 CE |
5248 | #define SET_SYSCTL(name) \ |
5249 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5250 | SET_SYSCTL(sched_min_granularity); | |
5251 | SET_SYSCTL(sched_latency); | |
5252 | SET_SYSCTL(sched_wakeup_granularity); | |
5253 | SET_SYSCTL(sched_shares_ratelimit); | |
5254 | #undef SET_SYSCTL | |
5255 | } | |
55cd5340 | 5256 | |
0bcdcf28 CE |
5257 | static inline void sched_init_granularity(void) |
5258 | { | |
5259 | update_sysctl(); | |
19978ca6 IM |
5260 | } |
5261 | ||
1da177e4 LT |
5262 | #ifdef CONFIG_SMP |
5263 | /* | |
5264 | * This is how migration works: | |
5265 | * | |
70b97a7f | 5266 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5267 | * runqueue and wake up that CPU's migration thread. |
5268 | * 2) we down() the locked semaphore => thread blocks. | |
5269 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5270 | * thread off the CPU) | |
5271 | * 4) it gets the migration request and checks whether the migrated | |
5272 | * task is still in the wrong runqueue. | |
5273 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5274 | * it and puts it into the right queue. | |
5275 | * 6) migration thread up()s the semaphore. | |
5276 | * 7) we wake up and the migration is done. | |
5277 | */ | |
5278 | ||
5279 | /* | |
5280 | * Change a given task's CPU affinity. Migrate the thread to a | |
5281 | * proper CPU and schedule it away if the CPU it's executing on | |
5282 | * is removed from the allowed bitmask. | |
5283 | * | |
5284 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5285 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5286 | * call is not atomic; no spinlocks may be held. |
5287 | */ | |
96f874e2 | 5288 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 5289 | { |
70b97a7f | 5290 | struct migration_req req; |
1da177e4 | 5291 | unsigned long flags; |
70b97a7f | 5292 | struct rq *rq; |
48f24c4d | 5293 | int ret = 0; |
1da177e4 LT |
5294 | |
5295 | rq = task_rq_lock(p, &flags); | |
e2912009 | 5296 | |
6ad4c188 | 5297 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5298 | ret = -EINVAL; |
5299 | goto out; | |
5300 | } | |
5301 | ||
9985b0ba | 5302 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5303 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5304 | ret = -EINVAL; |
5305 | goto out; | |
5306 | } | |
5307 | ||
73fe6aae | 5308 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5309 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5310 | else { |
96f874e2 RR |
5311 | cpumask_copy(&p->cpus_allowed, new_mask); |
5312 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5313 | } |
5314 | ||
1da177e4 | 5315 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5316 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5317 | goto out; |
5318 | ||
6ad4c188 | 5319 | if (migrate_task(p, cpumask_any_and(cpu_active_mask, new_mask), &req)) { |
1da177e4 | 5320 | /* Need help from migration thread: drop lock and wait. */ |
693525e3 PZ |
5321 | struct task_struct *mt = rq->migration_thread; |
5322 | ||
5323 | get_task_struct(mt); | |
1da177e4 LT |
5324 | task_rq_unlock(rq, &flags); |
5325 | wake_up_process(rq->migration_thread); | |
693525e3 | 5326 | put_task_struct(mt); |
1da177e4 LT |
5327 | wait_for_completion(&req.done); |
5328 | tlb_migrate_finish(p->mm); | |
5329 | return 0; | |
5330 | } | |
5331 | out: | |
5332 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5333 | |
1da177e4 LT |
5334 | return ret; |
5335 | } | |
cd8ba7cd | 5336 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5337 | |
5338 | /* | |
41a2d6cf | 5339 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5340 | * this because either it can't run here any more (set_cpus_allowed() |
5341 | * away from this CPU, or CPU going down), or because we're | |
5342 | * attempting to rebalance this task on exec (sched_exec). | |
5343 | * | |
5344 | * So we race with normal scheduler movements, but that's OK, as long | |
5345 | * as the task is no longer on this CPU. | |
efc30814 KK |
5346 | * |
5347 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5348 | */ |
efc30814 | 5349 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5350 | { |
70b97a7f | 5351 | struct rq *rq_dest, *rq_src; |
e2912009 | 5352 | int ret = 0; |
1da177e4 | 5353 | |
e761b772 | 5354 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5355 | return ret; |
1da177e4 LT |
5356 | |
5357 | rq_src = cpu_rq(src_cpu); | |
5358 | rq_dest = cpu_rq(dest_cpu); | |
5359 | ||
5360 | double_rq_lock(rq_src, rq_dest); | |
5361 | /* Already moved. */ | |
5362 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5363 | goto done; |
1da177e4 | 5364 | /* Affinity changed (again). */ |
96f874e2 | 5365 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5366 | goto fail; |
1da177e4 | 5367 | |
e2912009 PZ |
5368 | /* |
5369 | * If we're not on a rq, the next wake-up will ensure we're | |
5370 | * placed properly. | |
5371 | */ | |
5372 | if (p->se.on_rq) { | |
2e1cb74a | 5373 | deactivate_task(rq_src, p, 0); |
e2912009 | 5374 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5375 | activate_task(rq_dest, p, 0); |
15afe09b | 5376 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5377 | } |
b1e38734 | 5378 | done: |
efc30814 | 5379 | ret = 1; |
b1e38734 | 5380 | fail: |
1da177e4 | 5381 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5382 | return ret; |
1da177e4 LT |
5383 | } |
5384 | ||
03b042bf PM |
5385 | #define RCU_MIGRATION_IDLE 0 |
5386 | #define RCU_MIGRATION_NEED_QS 1 | |
5387 | #define RCU_MIGRATION_GOT_QS 2 | |
5388 | #define RCU_MIGRATION_MUST_SYNC 3 | |
5389 | ||
1da177e4 LT |
5390 | /* |
5391 | * migration_thread - this is a highprio system thread that performs | |
5392 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5393 | * another runqueue. | |
5394 | */ | |
95cdf3b7 | 5395 | static int migration_thread(void *data) |
1da177e4 | 5396 | { |
03b042bf | 5397 | int badcpu; |
1da177e4 | 5398 | int cpu = (long)data; |
70b97a7f | 5399 | struct rq *rq; |
1da177e4 LT |
5400 | |
5401 | rq = cpu_rq(cpu); | |
5402 | BUG_ON(rq->migration_thread != current); | |
5403 | ||
5404 | set_current_state(TASK_INTERRUPTIBLE); | |
5405 | while (!kthread_should_stop()) { | |
70b97a7f | 5406 | struct migration_req *req; |
1da177e4 | 5407 | struct list_head *head; |
1da177e4 | 5408 | |
05fa785c | 5409 | raw_spin_lock_irq(&rq->lock); |
1da177e4 LT |
5410 | |
5411 | if (cpu_is_offline(cpu)) { | |
05fa785c | 5412 | raw_spin_unlock_irq(&rq->lock); |
371cbb38 | 5413 | break; |
1da177e4 LT |
5414 | } |
5415 | ||
5416 | if (rq->active_balance) { | |
5417 | active_load_balance(rq, cpu); | |
5418 | rq->active_balance = 0; | |
5419 | } | |
5420 | ||
5421 | head = &rq->migration_queue; | |
5422 | ||
5423 | if (list_empty(head)) { | |
05fa785c | 5424 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5425 | schedule(); |
5426 | set_current_state(TASK_INTERRUPTIBLE); | |
5427 | continue; | |
5428 | } | |
70b97a7f | 5429 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5430 | list_del_init(head->next); |
5431 | ||
03b042bf | 5432 | if (req->task != NULL) { |
05fa785c | 5433 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5434 | __migrate_task(req->task, cpu, req->dest_cpu); |
5435 | } else if (likely(cpu == (badcpu = smp_processor_id()))) { | |
5436 | req->dest_cpu = RCU_MIGRATION_GOT_QS; | |
05fa785c | 5437 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5438 | } else { |
5439 | req->dest_cpu = RCU_MIGRATION_MUST_SYNC; | |
05fa785c | 5440 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5441 | WARN_ONCE(1, "migration_thread() on CPU %d, expected %d\n", badcpu, cpu); |
5442 | } | |
674311d5 | 5443 | local_irq_enable(); |
1da177e4 LT |
5444 | |
5445 | complete(&req->done); | |
5446 | } | |
5447 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 5448 | |
1da177e4 LT |
5449 | return 0; |
5450 | } | |
5451 | ||
5452 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5453 | |
5454 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5455 | { | |
5456 | int ret; | |
5457 | ||
5458 | local_irq_disable(); | |
5459 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5460 | local_irq_enable(); | |
5461 | return ret; | |
5462 | } | |
5463 | ||
054b9108 | 5464 | /* |
3a4fa0a2 | 5465 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 5466 | */ |
48f24c4d | 5467 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5468 | { |
70b97a7f | 5469 | int dest_cpu; |
e76bd8d9 RR |
5470 | |
5471 | again: | |
5da9a0fb | 5472 | dest_cpu = select_fallback_rq(dead_cpu, p); |
e76bd8d9 | 5473 | |
e76bd8d9 RR |
5474 | /* It can have affinity changed while we were choosing. */ |
5475 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
5476 | goto again; | |
1da177e4 LT |
5477 | } |
5478 | ||
5479 | /* | |
5480 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5481 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5482 | * for performance reasons the counter is not stricly tracking tasks to | |
5483 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5484 | * to keep the global sum constant after CPU-down: | |
5485 | */ | |
70b97a7f | 5486 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5487 | { |
6ad4c188 | 5488 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
5489 | unsigned long flags; |
5490 | ||
5491 | local_irq_save(flags); | |
5492 | double_rq_lock(rq_src, rq_dest); | |
5493 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5494 | rq_src->nr_uninterruptible = 0; | |
5495 | double_rq_unlock(rq_src, rq_dest); | |
5496 | local_irq_restore(flags); | |
5497 | } | |
5498 | ||
5499 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5500 | static void migrate_live_tasks(int src_cpu) | |
5501 | { | |
48f24c4d | 5502 | struct task_struct *p, *t; |
1da177e4 | 5503 | |
f7b4cddc | 5504 | read_lock(&tasklist_lock); |
1da177e4 | 5505 | |
48f24c4d IM |
5506 | do_each_thread(t, p) { |
5507 | if (p == current) | |
1da177e4 LT |
5508 | continue; |
5509 | ||
48f24c4d IM |
5510 | if (task_cpu(p) == src_cpu) |
5511 | move_task_off_dead_cpu(src_cpu, p); | |
5512 | } while_each_thread(t, p); | |
1da177e4 | 5513 | |
f7b4cddc | 5514 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5515 | } |
5516 | ||
dd41f596 IM |
5517 | /* |
5518 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5519 | * It does so by boosting its priority to highest possible. |
5520 | * Used by CPU offline code. | |
1da177e4 LT |
5521 | */ |
5522 | void sched_idle_next(void) | |
5523 | { | |
48f24c4d | 5524 | int this_cpu = smp_processor_id(); |
70b97a7f | 5525 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5526 | struct task_struct *p = rq->idle; |
5527 | unsigned long flags; | |
5528 | ||
5529 | /* cpu has to be offline */ | |
48f24c4d | 5530 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5531 | |
48f24c4d IM |
5532 | /* |
5533 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5534 | * and interrupts disabled on the current cpu. | |
1da177e4 | 5535 | */ |
05fa785c | 5536 | raw_spin_lock_irqsave(&rq->lock, flags); |
1da177e4 | 5537 | |
dd41f596 | 5538 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5539 | |
94bc9a7b | 5540 | activate_task(rq, p, 0); |
1da177e4 | 5541 | |
05fa785c | 5542 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5543 | } |
5544 | ||
48f24c4d IM |
5545 | /* |
5546 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5547 | * offline. |
5548 | */ | |
5549 | void idle_task_exit(void) | |
5550 | { | |
5551 | struct mm_struct *mm = current->active_mm; | |
5552 | ||
5553 | BUG_ON(cpu_online(smp_processor_id())); | |
5554 | ||
5555 | if (mm != &init_mm) | |
5556 | switch_mm(mm, &init_mm, current); | |
5557 | mmdrop(mm); | |
5558 | } | |
5559 | ||
054b9108 | 5560 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5561 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5562 | { |
70b97a7f | 5563 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5564 | |
5565 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5566 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5567 | |
5568 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5569 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5570 | |
48f24c4d | 5571 | get_task_struct(p); |
1da177e4 LT |
5572 | |
5573 | /* | |
5574 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5575 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5576 | * fine. |
5577 | */ | |
05fa785c | 5578 | raw_spin_unlock_irq(&rq->lock); |
48f24c4d | 5579 | move_task_off_dead_cpu(dead_cpu, p); |
05fa785c | 5580 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5581 | |
48f24c4d | 5582 | put_task_struct(p); |
1da177e4 LT |
5583 | } |
5584 | ||
5585 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5586 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5587 | { | |
70b97a7f | 5588 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5589 | struct task_struct *next; |
48f24c4d | 5590 | |
dd41f596 IM |
5591 | for ( ; ; ) { |
5592 | if (!rq->nr_running) | |
5593 | break; | |
b67802ea | 5594 | next = pick_next_task(rq); |
dd41f596 IM |
5595 | if (!next) |
5596 | break; | |
79c53799 | 5597 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 5598 | migrate_dead(dead_cpu, next); |
e692ab53 | 5599 | |
1da177e4 LT |
5600 | } |
5601 | } | |
dce48a84 TG |
5602 | |
5603 | /* | |
5604 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
5605 | */ | |
5606 | static void calc_global_load_remove(struct rq *rq) | |
5607 | { | |
5608 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 5609 | rq->calc_load_active = 0; |
dce48a84 | 5610 | } |
1da177e4 LT |
5611 | #endif /* CONFIG_HOTPLUG_CPU */ |
5612 | ||
e692ab53 NP |
5613 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5614 | ||
5615 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5616 | { |
5617 | .procname = "sched_domain", | |
c57baf1e | 5618 | .mode = 0555, |
e0361851 | 5619 | }, |
56992309 | 5620 | {} |
e692ab53 NP |
5621 | }; |
5622 | ||
5623 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5624 | { |
5625 | .procname = "kernel", | |
c57baf1e | 5626 | .mode = 0555, |
e0361851 AD |
5627 | .child = sd_ctl_dir, |
5628 | }, | |
56992309 | 5629 | {} |
e692ab53 NP |
5630 | }; |
5631 | ||
5632 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5633 | { | |
5634 | struct ctl_table *entry = | |
5cf9f062 | 5635 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5636 | |
e692ab53 NP |
5637 | return entry; |
5638 | } | |
5639 | ||
6382bc90 MM |
5640 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5641 | { | |
cd790076 | 5642 | struct ctl_table *entry; |
6382bc90 | 5643 | |
cd790076 MM |
5644 | /* |
5645 | * In the intermediate directories, both the child directory and | |
5646 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5647 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5648 | * static strings and all have proc handlers. |
5649 | */ | |
5650 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5651 | if (entry->child) |
5652 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5653 | if (entry->proc_handler == NULL) |
5654 | kfree(entry->procname); | |
5655 | } | |
6382bc90 MM |
5656 | |
5657 | kfree(*tablep); | |
5658 | *tablep = NULL; | |
5659 | } | |
5660 | ||
e692ab53 | 5661 | static void |
e0361851 | 5662 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5663 | const char *procname, void *data, int maxlen, |
5664 | mode_t mode, proc_handler *proc_handler) | |
5665 | { | |
e692ab53 NP |
5666 | entry->procname = procname; |
5667 | entry->data = data; | |
5668 | entry->maxlen = maxlen; | |
5669 | entry->mode = mode; | |
5670 | entry->proc_handler = proc_handler; | |
5671 | } | |
5672 | ||
5673 | static struct ctl_table * | |
5674 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5675 | { | |
a5d8c348 | 5676 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5677 | |
ad1cdc1d MM |
5678 | if (table == NULL) |
5679 | return NULL; | |
5680 | ||
e0361851 | 5681 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5682 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5683 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5684 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5685 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5686 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5687 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5688 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5689 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5690 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5691 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5692 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5693 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5694 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5695 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5696 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5697 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5698 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5699 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5700 | &sd->cache_nice_tries, |
5701 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5702 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5703 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5704 | set_table_entry(&table[11], "name", sd->name, |
5705 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5706 | /* &table[12] is terminator */ | |
e692ab53 NP |
5707 | |
5708 | return table; | |
5709 | } | |
5710 | ||
9a4e7159 | 5711 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5712 | { |
5713 | struct ctl_table *entry, *table; | |
5714 | struct sched_domain *sd; | |
5715 | int domain_num = 0, i; | |
5716 | char buf[32]; | |
5717 | ||
5718 | for_each_domain(cpu, sd) | |
5719 | domain_num++; | |
5720 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5721 | if (table == NULL) |
5722 | return NULL; | |
e692ab53 NP |
5723 | |
5724 | i = 0; | |
5725 | for_each_domain(cpu, sd) { | |
5726 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5727 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5728 | entry->mode = 0555; |
e692ab53 NP |
5729 | entry->child = sd_alloc_ctl_domain_table(sd); |
5730 | entry++; | |
5731 | i++; | |
5732 | } | |
5733 | return table; | |
5734 | } | |
5735 | ||
5736 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5737 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5738 | { |
6ad4c188 | 5739 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5740 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5741 | char buf[32]; | |
5742 | ||
7378547f MM |
5743 | WARN_ON(sd_ctl_dir[0].child); |
5744 | sd_ctl_dir[0].child = entry; | |
5745 | ||
ad1cdc1d MM |
5746 | if (entry == NULL) |
5747 | return; | |
5748 | ||
6ad4c188 | 5749 | for_each_possible_cpu(i) { |
e692ab53 | 5750 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5751 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5752 | entry->mode = 0555; |
e692ab53 | 5753 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5754 | entry++; |
e692ab53 | 5755 | } |
7378547f MM |
5756 | |
5757 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5758 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5759 | } | |
6382bc90 | 5760 | |
7378547f | 5761 | /* may be called multiple times per register */ |
6382bc90 MM |
5762 | static void unregister_sched_domain_sysctl(void) |
5763 | { | |
7378547f MM |
5764 | if (sd_sysctl_header) |
5765 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5766 | sd_sysctl_header = NULL; |
7378547f MM |
5767 | if (sd_ctl_dir[0].child) |
5768 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5769 | } |
e692ab53 | 5770 | #else |
6382bc90 MM |
5771 | static void register_sched_domain_sysctl(void) |
5772 | { | |
5773 | } | |
5774 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5775 | { |
5776 | } | |
5777 | #endif | |
5778 | ||
1f11eb6a GH |
5779 | static void set_rq_online(struct rq *rq) |
5780 | { | |
5781 | if (!rq->online) { | |
5782 | const struct sched_class *class; | |
5783 | ||
c6c4927b | 5784 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5785 | rq->online = 1; |
5786 | ||
5787 | for_each_class(class) { | |
5788 | if (class->rq_online) | |
5789 | class->rq_online(rq); | |
5790 | } | |
5791 | } | |
5792 | } | |
5793 | ||
5794 | static void set_rq_offline(struct rq *rq) | |
5795 | { | |
5796 | if (rq->online) { | |
5797 | const struct sched_class *class; | |
5798 | ||
5799 | for_each_class(class) { | |
5800 | if (class->rq_offline) | |
5801 | class->rq_offline(rq); | |
5802 | } | |
5803 | ||
c6c4927b | 5804 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5805 | rq->online = 0; |
5806 | } | |
5807 | } | |
5808 | ||
1da177e4 LT |
5809 | /* |
5810 | * migration_call - callback that gets triggered when a CPU is added. | |
5811 | * Here we can start up the necessary migration thread for the new CPU. | |
5812 | */ | |
48f24c4d IM |
5813 | static int __cpuinit |
5814 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5815 | { |
1da177e4 | 5816 | struct task_struct *p; |
48f24c4d | 5817 | int cpu = (long)hcpu; |
1da177e4 | 5818 | unsigned long flags; |
70b97a7f | 5819 | struct rq *rq; |
1da177e4 LT |
5820 | |
5821 | switch (action) { | |
5be9361c | 5822 | |
1da177e4 | 5823 | case CPU_UP_PREPARE: |
8bb78442 | 5824 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5825 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5826 | if (IS_ERR(p)) |
5827 | return NOTIFY_BAD; | |
1da177e4 LT |
5828 | kthread_bind(p, cpu); |
5829 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5830 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5831 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 5832 | task_rq_unlock(rq, &flags); |
371cbb38 | 5833 | get_task_struct(p); |
1da177e4 | 5834 | cpu_rq(cpu)->migration_thread = p; |
a468d389 | 5835 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5836 | break; |
48f24c4d | 5837 | |
1da177e4 | 5838 | case CPU_ONLINE: |
8bb78442 | 5839 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5840 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 5841 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
5842 | |
5843 | /* Update our root-domain */ | |
5844 | rq = cpu_rq(cpu); | |
05fa785c | 5845 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5846 | if (rq->rd) { |
c6c4927b | 5847 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5848 | |
5849 | set_rq_online(rq); | |
1f94ef59 | 5850 | } |
05fa785c | 5851 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5852 | break; |
48f24c4d | 5853 | |
1da177e4 LT |
5854 | #ifdef CONFIG_HOTPLUG_CPU |
5855 | case CPU_UP_CANCELED: | |
8bb78442 | 5856 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5857 | if (!cpu_rq(cpu)->migration_thread) |
5858 | break; | |
41a2d6cf | 5859 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 5860 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 5861 | cpumask_any(cpu_online_mask)); |
1da177e4 | 5862 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 5863 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
5864 | cpu_rq(cpu)->migration_thread = NULL; |
5865 | break; | |
48f24c4d | 5866 | |
1da177e4 | 5867 | case CPU_DEAD: |
8bb78442 | 5868 | case CPU_DEAD_FROZEN: |
470fd646 | 5869 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
5870 | migrate_live_tasks(cpu); |
5871 | rq = cpu_rq(cpu); | |
5872 | kthread_stop(rq->migration_thread); | |
371cbb38 | 5873 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
5874 | rq->migration_thread = NULL; |
5875 | /* Idle task back to normal (off runqueue, low prio) */ | |
05fa785c | 5876 | raw_spin_lock_irq(&rq->lock); |
2e1cb74a | 5877 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
5878 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5879 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5880 | migrate_dead_tasks(cpu); |
05fa785c | 5881 | raw_spin_unlock_irq(&rq->lock); |
470fd646 | 5882 | cpuset_unlock(); |
1da177e4 LT |
5883 | migrate_nr_uninterruptible(rq); |
5884 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 5885 | calc_global_load_remove(rq); |
41a2d6cf IM |
5886 | /* |
5887 | * No need to migrate the tasks: it was best-effort if | |
5888 | * they didn't take sched_hotcpu_mutex. Just wake up | |
5889 | * the requestors. | |
5890 | */ | |
05fa785c | 5891 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5892 | while (!list_empty(&rq->migration_queue)) { |
70b97a7f IM |
5893 | struct migration_req *req; |
5894 | ||
1da177e4 | 5895 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5896 | struct migration_req, list); |
1da177e4 | 5897 | list_del_init(&req->list); |
05fa785c | 5898 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 5899 | complete(&req->done); |
05fa785c | 5900 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5901 | } |
05fa785c | 5902 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 5903 | break; |
57d885fe | 5904 | |
08f503b0 GH |
5905 | case CPU_DYING: |
5906 | case CPU_DYING_FROZEN: | |
57d885fe GH |
5907 | /* Update our root-domain */ |
5908 | rq = cpu_rq(cpu); | |
05fa785c | 5909 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5910 | if (rq->rd) { |
c6c4927b | 5911 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5912 | set_rq_offline(rq); |
57d885fe | 5913 | } |
05fa785c | 5914 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
57d885fe | 5915 | break; |
1da177e4 LT |
5916 | #endif |
5917 | } | |
5918 | return NOTIFY_OK; | |
5919 | } | |
5920 | ||
f38b0820 PM |
5921 | /* |
5922 | * Register at high priority so that task migration (migrate_all_tasks) | |
5923 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5924 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5925 | */ |
26c2143b | 5926 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5927 | .notifier_call = migration_call, |
5928 | .priority = 10 | |
5929 | }; | |
5930 | ||
7babe8db | 5931 | static int __init migration_init(void) |
1da177e4 LT |
5932 | { |
5933 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5934 | int err; |
48f24c4d IM |
5935 | |
5936 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5937 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5938 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5939 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5940 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5941 | |
a004cd42 | 5942 | return 0; |
1da177e4 | 5943 | } |
7babe8db | 5944 | early_initcall(migration_init); |
1da177e4 LT |
5945 | #endif |
5946 | ||
5947 | #ifdef CONFIG_SMP | |
476f3534 | 5948 | |
3e9830dc | 5949 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5950 | |
f6630114 MT |
5951 | static __read_mostly int sched_domain_debug_enabled; |
5952 | ||
5953 | static int __init sched_domain_debug_setup(char *str) | |
5954 | { | |
5955 | sched_domain_debug_enabled = 1; | |
5956 | ||
5957 | return 0; | |
5958 | } | |
5959 | early_param("sched_debug", sched_domain_debug_setup); | |
5960 | ||
7c16ec58 | 5961 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5962 | struct cpumask *groupmask) |
1da177e4 | 5963 | { |
4dcf6aff | 5964 | struct sched_group *group = sd->groups; |
434d53b0 | 5965 | char str[256]; |
1da177e4 | 5966 | |
968ea6d8 | 5967 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5968 | cpumask_clear(groupmask); |
4dcf6aff IM |
5969 | |
5970 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5971 | ||
5972 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5973 | printk("does not load-balance\n"); |
4dcf6aff | 5974 | if (sd->parent) |
3df0fc5b PZ |
5975 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5976 | " has parent"); | |
4dcf6aff | 5977 | return -1; |
41c7ce9a NP |
5978 | } |
5979 | ||
3df0fc5b | 5980 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5981 | |
758b2cdc | 5982 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5983 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5984 | "CPU%d\n", cpu); | |
4dcf6aff | 5985 | } |
758b2cdc | 5986 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5987 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5988 | " CPU%d\n", cpu); | |
4dcf6aff | 5989 | } |
1da177e4 | 5990 | |
4dcf6aff | 5991 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5992 | do { |
4dcf6aff | 5993 | if (!group) { |
3df0fc5b PZ |
5994 | printk("\n"); |
5995 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5996 | break; |
5997 | } | |
5998 | ||
18a3885f | 5999 | if (!group->cpu_power) { |
3df0fc5b PZ |
6000 | printk(KERN_CONT "\n"); |
6001 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6002 | "set\n"); | |
4dcf6aff IM |
6003 | break; |
6004 | } | |
1da177e4 | 6005 | |
758b2cdc | 6006 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6007 | printk(KERN_CONT "\n"); |
6008 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6009 | break; |
6010 | } | |
1da177e4 | 6011 | |
758b2cdc | 6012 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6013 | printk(KERN_CONT "\n"); |
6014 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6015 | break; |
6016 | } | |
1da177e4 | 6017 | |
758b2cdc | 6018 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6019 | |
968ea6d8 | 6020 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6021 | |
3df0fc5b | 6022 | printk(KERN_CONT " %s", str); |
18a3885f | 6023 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6024 | printk(KERN_CONT " (cpu_power = %d)", |
6025 | group->cpu_power); | |
381512cf | 6026 | } |
1da177e4 | 6027 | |
4dcf6aff IM |
6028 | group = group->next; |
6029 | } while (group != sd->groups); | |
3df0fc5b | 6030 | printk(KERN_CONT "\n"); |
1da177e4 | 6031 | |
758b2cdc | 6032 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6033 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6034 | |
758b2cdc RR |
6035 | if (sd->parent && |
6036 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6037 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6038 | "of domain->span\n"); | |
4dcf6aff IM |
6039 | return 0; |
6040 | } | |
1da177e4 | 6041 | |
4dcf6aff IM |
6042 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6043 | { | |
d5dd3db1 | 6044 | cpumask_var_t groupmask; |
4dcf6aff | 6045 | int level = 0; |
1da177e4 | 6046 | |
f6630114 MT |
6047 | if (!sched_domain_debug_enabled) |
6048 | return; | |
6049 | ||
4dcf6aff IM |
6050 | if (!sd) { |
6051 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6052 | return; | |
6053 | } | |
1da177e4 | 6054 | |
4dcf6aff IM |
6055 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6056 | ||
d5dd3db1 | 6057 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6058 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6059 | return; | |
6060 | } | |
6061 | ||
4dcf6aff | 6062 | for (;;) { |
7c16ec58 | 6063 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6064 | break; |
1da177e4 LT |
6065 | level++; |
6066 | sd = sd->parent; | |
33859f7f | 6067 | if (!sd) |
4dcf6aff IM |
6068 | break; |
6069 | } | |
d5dd3db1 | 6070 | free_cpumask_var(groupmask); |
1da177e4 | 6071 | } |
6d6bc0ad | 6072 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6073 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6074 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6075 | |
1a20ff27 | 6076 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6077 | { |
758b2cdc | 6078 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6079 | return 1; |
6080 | ||
6081 | /* Following flags need at least 2 groups */ | |
6082 | if (sd->flags & (SD_LOAD_BALANCE | | |
6083 | SD_BALANCE_NEWIDLE | | |
6084 | SD_BALANCE_FORK | | |
89c4710e SS |
6085 | SD_BALANCE_EXEC | |
6086 | SD_SHARE_CPUPOWER | | |
6087 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6088 | if (sd->groups != sd->groups->next) |
6089 | return 0; | |
6090 | } | |
6091 | ||
6092 | /* Following flags don't use groups */ | |
c88d5910 | 6093 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6094 | return 0; |
6095 | ||
6096 | return 1; | |
6097 | } | |
6098 | ||
48f24c4d IM |
6099 | static int |
6100 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6101 | { |
6102 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6103 | ||
6104 | if (sd_degenerate(parent)) | |
6105 | return 1; | |
6106 | ||
758b2cdc | 6107 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6108 | return 0; |
6109 | ||
245af2c7 SS |
6110 | /* Flags needing groups don't count if only 1 group in parent */ |
6111 | if (parent->groups == parent->groups->next) { | |
6112 | pflags &= ~(SD_LOAD_BALANCE | | |
6113 | SD_BALANCE_NEWIDLE | | |
6114 | SD_BALANCE_FORK | | |
89c4710e SS |
6115 | SD_BALANCE_EXEC | |
6116 | SD_SHARE_CPUPOWER | | |
6117 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6118 | if (nr_node_ids == 1) |
6119 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6120 | } |
6121 | if (~cflags & pflags) | |
6122 | return 0; | |
6123 | ||
6124 | return 1; | |
6125 | } | |
6126 | ||
c6c4927b RR |
6127 | static void free_rootdomain(struct root_domain *rd) |
6128 | { | |
047106ad PZ |
6129 | synchronize_sched(); |
6130 | ||
68e74568 RR |
6131 | cpupri_cleanup(&rd->cpupri); |
6132 | ||
c6c4927b RR |
6133 | free_cpumask_var(rd->rto_mask); |
6134 | free_cpumask_var(rd->online); | |
6135 | free_cpumask_var(rd->span); | |
6136 | kfree(rd); | |
6137 | } | |
6138 | ||
57d885fe GH |
6139 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6140 | { | |
a0490fa3 | 6141 | struct root_domain *old_rd = NULL; |
57d885fe | 6142 | unsigned long flags; |
57d885fe | 6143 | |
05fa785c | 6144 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6145 | |
6146 | if (rq->rd) { | |
a0490fa3 | 6147 | old_rd = rq->rd; |
57d885fe | 6148 | |
c6c4927b | 6149 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6150 | set_rq_offline(rq); |
57d885fe | 6151 | |
c6c4927b | 6152 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6153 | |
a0490fa3 IM |
6154 | /* |
6155 | * If we dont want to free the old_rt yet then | |
6156 | * set old_rd to NULL to skip the freeing later | |
6157 | * in this function: | |
6158 | */ | |
6159 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6160 | old_rd = NULL; | |
57d885fe GH |
6161 | } |
6162 | ||
6163 | atomic_inc(&rd->refcount); | |
6164 | rq->rd = rd; | |
6165 | ||
c6c4927b | 6166 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6167 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6168 | set_rq_online(rq); |
57d885fe | 6169 | |
05fa785c | 6170 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6171 | |
6172 | if (old_rd) | |
6173 | free_rootdomain(old_rd); | |
57d885fe GH |
6174 | } |
6175 | ||
fd5e1b5d | 6176 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 6177 | { |
36b7b6d4 PE |
6178 | gfp_t gfp = GFP_KERNEL; |
6179 | ||
57d885fe GH |
6180 | memset(rd, 0, sizeof(*rd)); |
6181 | ||
36b7b6d4 PE |
6182 | if (bootmem) |
6183 | gfp = GFP_NOWAIT; | |
c6c4927b | 6184 | |
36b7b6d4 | 6185 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 6186 | goto out; |
36b7b6d4 | 6187 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 6188 | goto free_span; |
36b7b6d4 | 6189 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 6190 | goto free_online; |
6e0534f2 | 6191 | |
0fb53029 | 6192 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 6193 | goto free_rto_mask; |
c6c4927b | 6194 | return 0; |
6e0534f2 | 6195 | |
68e74568 RR |
6196 | free_rto_mask: |
6197 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6198 | free_online: |
6199 | free_cpumask_var(rd->online); | |
6200 | free_span: | |
6201 | free_cpumask_var(rd->span); | |
0c910d28 | 6202 | out: |
c6c4927b | 6203 | return -ENOMEM; |
57d885fe GH |
6204 | } |
6205 | ||
6206 | static void init_defrootdomain(void) | |
6207 | { | |
c6c4927b RR |
6208 | init_rootdomain(&def_root_domain, true); |
6209 | ||
57d885fe GH |
6210 | atomic_set(&def_root_domain.refcount, 1); |
6211 | } | |
6212 | ||
dc938520 | 6213 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6214 | { |
6215 | struct root_domain *rd; | |
6216 | ||
6217 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6218 | if (!rd) | |
6219 | return NULL; | |
6220 | ||
c6c4927b RR |
6221 | if (init_rootdomain(rd, false) != 0) { |
6222 | kfree(rd); | |
6223 | return NULL; | |
6224 | } | |
57d885fe GH |
6225 | |
6226 | return rd; | |
6227 | } | |
6228 | ||
1da177e4 | 6229 | /* |
0eab9146 | 6230 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6231 | * hold the hotplug lock. |
6232 | */ | |
0eab9146 IM |
6233 | static void |
6234 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6235 | { |
70b97a7f | 6236 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6237 | struct sched_domain *tmp; |
6238 | ||
6239 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6240 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6241 | struct sched_domain *parent = tmp->parent; |
6242 | if (!parent) | |
6243 | break; | |
f29c9b1c | 6244 | |
1a848870 | 6245 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6246 | tmp->parent = parent->parent; |
1a848870 SS |
6247 | if (parent->parent) |
6248 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6249 | } else |
6250 | tmp = tmp->parent; | |
245af2c7 SS |
6251 | } |
6252 | ||
1a848870 | 6253 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6254 | sd = sd->parent; |
1a848870 SS |
6255 | if (sd) |
6256 | sd->child = NULL; | |
6257 | } | |
1da177e4 LT |
6258 | |
6259 | sched_domain_debug(sd, cpu); | |
6260 | ||
57d885fe | 6261 | rq_attach_root(rq, rd); |
674311d5 | 6262 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6263 | } |
6264 | ||
6265 | /* cpus with isolated domains */ | |
dcc30a35 | 6266 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6267 | |
6268 | /* Setup the mask of cpus configured for isolated domains */ | |
6269 | static int __init isolated_cpu_setup(char *str) | |
6270 | { | |
bdddd296 | 6271 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6272 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6273 | return 1; |
6274 | } | |
6275 | ||
8927f494 | 6276 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6277 | |
6278 | /* | |
6711cab4 SS |
6279 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6280 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6281 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6282 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6283 | * |
6284 | * init_sched_build_groups will build a circular linked list of the groups | |
6285 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6286 | * and ->cpu_power to 0. | |
6287 | */ | |
a616058b | 6288 | static void |
96f874e2 RR |
6289 | init_sched_build_groups(const struct cpumask *span, |
6290 | const struct cpumask *cpu_map, | |
6291 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6292 | struct sched_group **sg, |
96f874e2 RR |
6293 | struct cpumask *tmpmask), |
6294 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6295 | { |
6296 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6297 | int i; |
6298 | ||
96f874e2 | 6299 | cpumask_clear(covered); |
7c16ec58 | 6300 | |
abcd083a | 6301 | for_each_cpu(i, span) { |
6711cab4 | 6302 | struct sched_group *sg; |
7c16ec58 | 6303 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6304 | int j; |
6305 | ||
758b2cdc | 6306 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6307 | continue; |
6308 | ||
758b2cdc | 6309 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6310 | sg->cpu_power = 0; |
1da177e4 | 6311 | |
abcd083a | 6312 | for_each_cpu(j, span) { |
7c16ec58 | 6313 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6314 | continue; |
6315 | ||
96f874e2 | 6316 | cpumask_set_cpu(j, covered); |
758b2cdc | 6317 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6318 | } |
6319 | if (!first) | |
6320 | first = sg; | |
6321 | if (last) | |
6322 | last->next = sg; | |
6323 | last = sg; | |
6324 | } | |
6325 | last->next = first; | |
6326 | } | |
6327 | ||
9c1cfda2 | 6328 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6329 | |
9c1cfda2 | 6330 | #ifdef CONFIG_NUMA |
198e2f18 | 6331 | |
9c1cfda2 JH |
6332 | /** |
6333 | * find_next_best_node - find the next node to include in a sched_domain | |
6334 | * @node: node whose sched_domain we're building | |
6335 | * @used_nodes: nodes already in the sched_domain | |
6336 | * | |
41a2d6cf | 6337 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6338 | * finds the closest node not already in the @used_nodes map. |
6339 | * | |
6340 | * Should use nodemask_t. | |
6341 | */ | |
c5f59f08 | 6342 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6343 | { |
6344 | int i, n, val, min_val, best_node = 0; | |
6345 | ||
6346 | min_val = INT_MAX; | |
6347 | ||
076ac2af | 6348 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6349 | /* Start at @node */ |
076ac2af | 6350 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6351 | |
6352 | if (!nr_cpus_node(n)) | |
6353 | continue; | |
6354 | ||
6355 | /* Skip already used nodes */ | |
c5f59f08 | 6356 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6357 | continue; |
6358 | ||
6359 | /* Simple min distance search */ | |
6360 | val = node_distance(node, n); | |
6361 | ||
6362 | if (val < min_val) { | |
6363 | min_val = val; | |
6364 | best_node = n; | |
6365 | } | |
6366 | } | |
6367 | ||
c5f59f08 | 6368 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6369 | return best_node; |
6370 | } | |
6371 | ||
6372 | /** | |
6373 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6374 | * @node: node whose cpumask we're constructing | |
73486722 | 6375 | * @span: resulting cpumask |
9c1cfda2 | 6376 | * |
41a2d6cf | 6377 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6378 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6379 | * out optimally. | |
6380 | */ | |
96f874e2 | 6381 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6382 | { |
c5f59f08 | 6383 | nodemask_t used_nodes; |
48f24c4d | 6384 | int i; |
9c1cfda2 | 6385 | |
6ca09dfc | 6386 | cpumask_clear(span); |
c5f59f08 | 6387 | nodes_clear(used_nodes); |
9c1cfda2 | 6388 | |
6ca09dfc | 6389 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6390 | node_set(node, used_nodes); |
9c1cfda2 JH |
6391 | |
6392 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6393 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6394 | |
6ca09dfc | 6395 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6396 | } |
9c1cfda2 | 6397 | } |
6d6bc0ad | 6398 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6399 | |
5c45bf27 | 6400 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6401 | |
6c99e9ad RR |
6402 | /* |
6403 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6404 | * |
6405 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6406 | * and struct sched_domain. ) | |
6c99e9ad RR |
6407 | */ |
6408 | struct static_sched_group { | |
6409 | struct sched_group sg; | |
6410 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6411 | }; | |
6412 | ||
6413 | struct static_sched_domain { | |
6414 | struct sched_domain sd; | |
6415 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6416 | }; | |
6417 | ||
49a02c51 AH |
6418 | struct s_data { |
6419 | #ifdef CONFIG_NUMA | |
6420 | int sd_allnodes; | |
6421 | cpumask_var_t domainspan; | |
6422 | cpumask_var_t covered; | |
6423 | cpumask_var_t notcovered; | |
6424 | #endif | |
6425 | cpumask_var_t nodemask; | |
6426 | cpumask_var_t this_sibling_map; | |
6427 | cpumask_var_t this_core_map; | |
6428 | cpumask_var_t send_covered; | |
6429 | cpumask_var_t tmpmask; | |
6430 | struct sched_group **sched_group_nodes; | |
6431 | struct root_domain *rd; | |
6432 | }; | |
6433 | ||
2109b99e AH |
6434 | enum s_alloc { |
6435 | sa_sched_groups = 0, | |
6436 | sa_rootdomain, | |
6437 | sa_tmpmask, | |
6438 | sa_send_covered, | |
6439 | sa_this_core_map, | |
6440 | sa_this_sibling_map, | |
6441 | sa_nodemask, | |
6442 | sa_sched_group_nodes, | |
6443 | #ifdef CONFIG_NUMA | |
6444 | sa_notcovered, | |
6445 | sa_covered, | |
6446 | sa_domainspan, | |
6447 | #endif | |
6448 | sa_none, | |
6449 | }; | |
6450 | ||
9c1cfda2 | 6451 | /* |
48f24c4d | 6452 | * SMT sched-domains: |
9c1cfda2 | 6453 | */ |
1da177e4 | 6454 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6455 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6456 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6457 | |
41a2d6cf | 6458 | static int |
96f874e2 RR |
6459 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6460 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6461 | { |
6711cab4 | 6462 | if (sg) |
1871e52c | 6463 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6464 | return cpu; |
6465 | } | |
6d6bc0ad | 6466 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6467 | |
48f24c4d IM |
6468 | /* |
6469 | * multi-core sched-domains: | |
6470 | */ | |
1e9f28fa | 6471 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6472 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6473 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 6474 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6475 | |
6476 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6477 | static int |
96f874e2 RR |
6478 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6479 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6480 | { |
6711cab4 | 6481 | int group; |
7c16ec58 | 6482 | |
c69fc56d | 6483 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6484 | group = cpumask_first(mask); |
6711cab4 | 6485 | if (sg) |
6c99e9ad | 6486 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6487 | return group; |
1e9f28fa SS |
6488 | } |
6489 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6490 | static int |
96f874e2 RR |
6491 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6492 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 6493 | { |
6711cab4 | 6494 | if (sg) |
6c99e9ad | 6495 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
6496 | return cpu; |
6497 | } | |
6498 | #endif | |
6499 | ||
6c99e9ad RR |
6500 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6501 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6502 | |
41a2d6cf | 6503 | static int |
96f874e2 RR |
6504 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6505 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6506 | { |
6711cab4 | 6507 | int group; |
48f24c4d | 6508 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 6509 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6510 | group = cpumask_first(mask); |
1e9f28fa | 6511 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6512 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6513 | group = cpumask_first(mask); |
1da177e4 | 6514 | #else |
6711cab4 | 6515 | group = cpu; |
1da177e4 | 6516 | #endif |
6711cab4 | 6517 | if (sg) |
6c99e9ad | 6518 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6519 | return group; |
1da177e4 LT |
6520 | } |
6521 | ||
6522 | #ifdef CONFIG_NUMA | |
1da177e4 | 6523 | /* |
9c1cfda2 JH |
6524 | * The init_sched_build_groups can't handle what we want to do with node |
6525 | * groups, so roll our own. Now each node has its own list of groups which | |
6526 | * gets dynamically allocated. | |
1da177e4 | 6527 | */ |
62ea9ceb | 6528 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6529 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6530 | |
62ea9ceb | 6531 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6532 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6533 | |
96f874e2 RR |
6534 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6535 | struct sched_group **sg, | |
6536 | struct cpumask *nodemask) | |
9c1cfda2 | 6537 | { |
6711cab4 SS |
6538 | int group; |
6539 | ||
6ca09dfc | 6540 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6541 | group = cpumask_first(nodemask); |
6711cab4 SS |
6542 | |
6543 | if (sg) | |
6c99e9ad | 6544 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6545 | return group; |
1da177e4 | 6546 | } |
6711cab4 | 6547 | |
08069033 SS |
6548 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6549 | { | |
6550 | struct sched_group *sg = group_head; | |
6551 | int j; | |
6552 | ||
6553 | if (!sg) | |
6554 | return; | |
3a5c359a | 6555 | do { |
758b2cdc | 6556 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6557 | struct sched_domain *sd; |
08069033 | 6558 | |
6c99e9ad | 6559 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6560 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6561 | /* |
6562 | * Only add "power" once for each | |
6563 | * physical package. | |
6564 | */ | |
6565 | continue; | |
6566 | } | |
08069033 | 6567 | |
18a3885f | 6568 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6569 | } |
6570 | sg = sg->next; | |
6571 | } while (sg != group_head); | |
08069033 | 6572 | } |
0601a88d AH |
6573 | |
6574 | static int build_numa_sched_groups(struct s_data *d, | |
6575 | const struct cpumask *cpu_map, int num) | |
6576 | { | |
6577 | struct sched_domain *sd; | |
6578 | struct sched_group *sg, *prev; | |
6579 | int n, j; | |
6580 | ||
6581 | cpumask_clear(d->covered); | |
6582 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6583 | if (cpumask_empty(d->nodemask)) { | |
6584 | d->sched_group_nodes[num] = NULL; | |
6585 | goto out; | |
6586 | } | |
6587 | ||
6588 | sched_domain_node_span(num, d->domainspan); | |
6589 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6590 | ||
6591 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6592 | GFP_KERNEL, num); | |
6593 | if (!sg) { | |
3df0fc5b PZ |
6594 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6595 | num); | |
0601a88d AH |
6596 | return -ENOMEM; |
6597 | } | |
6598 | d->sched_group_nodes[num] = sg; | |
6599 | ||
6600 | for_each_cpu(j, d->nodemask) { | |
6601 | sd = &per_cpu(node_domains, j).sd; | |
6602 | sd->groups = sg; | |
6603 | } | |
6604 | ||
18a3885f | 6605 | sg->cpu_power = 0; |
0601a88d AH |
6606 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6607 | sg->next = sg; | |
6608 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6609 | ||
6610 | prev = sg; | |
6611 | for (j = 0; j < nr_node_ids; j++) { | |
6612 | n = (num + j) % nr_node_ids; | |
6613 | cpumask_complement(d->notcovered, d->covered); | |
6614 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6615 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6616 | if (cpumask_empty(d->tmpmask)) | |
6617 | break; | |
6618 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6619 | if (cpumask_empty(d->tmpmask)) | |
6620 | continue; | |
6621 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6622 | GFP_KERNEL, num); | |
6623 | if (!sg) { | |
3df0fc5b PZ |
6624 | printk(KERN_WARNING |
6625 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6626 | return -ENOMEM; |
6627 | } | |
18a3885f | 6628 | sg->cpu_power = 0; |
0601a88d AH |
6629 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6630 | sg->next = prev->next; | |
6631 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6632 | prev->next = sg; | |
6633 | prev = sg; | |
6634 | } | |
6635 | out: | |
6636 | return 0; | |
6637 | } | |
6d6bc0ad | 6638 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6639 | |
a616058b | 6640 | #ifdef CONFIG_NUMA |
51888ca2 | 6641 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6642 | static void free_sched_groups(const struct cpumask *cpu_map, |
6643 | struct cpumask *nodemask) | |
51888ca2 | 6644 | { |
a616058b | 6645 | int cpu, i; |
51888ca2 | 6646 | |
abcd083a | 6647 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6648 | struct sched_group **sched_group_nodes |
6649 | = sched_group_nodes_bycpu[cpu]; | |
6650 | ||
51888ca2 SV |
6651 | if (!sched_group_nodes) |
6652 | continue; | |
6653 | ||
076ac2af | 6654 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6655 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6656 | ||
6ca09dfc | 6657 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6658 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6659 | continue; |
6660 | ||
6661 | if (sg == NULL) | |
6662 | continue; | |
6663 | sg = sg->next; | |
6664 | next_sg: | |
6665 | oldsg = sg; | |
6666 | sg = sg->next; | |
6667 | kfree(oldsg); | |
6668 | if (oldsg != sched_group_nodes[i]) | |
6669 | goto next_sg; | |
6670 | } | |
6671 | kfree(sched_group_nodes); | |
6672 | sched_group_nodes_bycpu[cpu] = NULL; | |
6673 | } | |
51888ca2 | 6674 | } |
6d6bc0ad | 6675 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6676 | static void free_sched_groups(const struct cpumask *cpu_map, |
6677 | struct cpumask *nodemask) | |
a616058b SS |
6678 | { |
6679 | } | |
6d6bc0ad | 6680 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6681 | |
89c4710e SS |
6682 | /* |
6683 | * Initialize sched groups cpu_power. | |
6684 | * | |
6685 | * cpu_power indicates the capacity of sched group, which is used while | |
6686 | * distributing the load between different sched groups in a sched domain. | |
6687 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6688 | * there are asymmetries in the topology. If there are asymmetries, group | |
6689 | * having more cpu_power will pickup more load compared to the group having | |
6690 | * less cpu_power. | |
89c4710e SS |
6691 | */ |
6692 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6693 | { | |
6694 | struct sched_domain *child; | |
6695 | struct sched_group *group; | |
f93e65c1 PZ |
6696 | long power; |
6697 | int weight; | |
89c4710e SS |
6698 | |
6699 | WARN_ON(!sd || !sd->groups); | |
6700 | ||
13318a71 | 6701 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6702 | return; |
6703 | ||
6704 | child = sd->child; | |
6705 | ||
18a3885f | 6706 | sd->groups->cpu_power = 0; |
5517d86b | 6707 | |
f93e65c1 PZ |
6708 | if (!child) { |
6709 | power = SCHED_LOAD_SCALE; | |
6710 | weight = cpumask_weight(sched_domain_span(sd)); | |
6711 | /* | |
6712 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6713 | * Usually multiple threads get a better yield out of |
6714 | * that one core than a single thread would have, | |
6715 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6716 | */ |
a52bfd73 PZ |
6717 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6718 | power *= sd->smt_gain; | |
f93e65c1 | 6719 | power /= weight; |
a52bfd73 PZ |
6720 | power >>= SCHED_LOAD_SHIFT; |
6721 | } | |
18a3885f | 6722 | sd->groups->cpu_power += power; |
89c4710e SS |
6723 | return; |
6724 | } | |
6725 | ||
89c4710e | 6726 | /* |
f93e65c1 | 6727 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6728 | */ |
6729 | group = child->groups; | |
6730 | do { | |
18a3885f | 6731 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6732 | group = group->next; |
6733 | } while (group != child->groups); | |
6734 | } | |
6735 | ||
7c16ec58 MT |
6736 | /* |
6737 | * Initializers for schedule domains | |
6738 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6739 | */ | |
6740 | ||
a5d8c348 IM |
6741 | #ifdef CONFIG_SCHED_DEBUG |
6742 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6743 | #else | |
6744 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6745 | #endif | |
6746 | ||
7c16ec58 | 6747 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6748 | |
7c16ec58 MT |
6749 | #define SD_INIT_FUNC(type) \ |
6750 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6751 | { \ | |
6752 | memset(sd, 0, sizeof(*sd)); \ | |
6753 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6754 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6755 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
6756 | } |
6757 | ||
6758 | SD_INIT_FUNC(CPU) | |
6759 | #ifdef CONFIG_NUMA | |
6760 | SD_INIT_FUNC(ALLNODES) | |
6761 | SD_INIT_FUNC(NODE) | |
6762 | #endif | |
6763 | #ifdef CONFIG_SCHED_SMT | |
6764 | SD_INIT_FUNC(SIBLING) | |
6765 | #endif | |
6766 | #ifdef CONFIG_SCHED_MC | |
6767 | SD_INIT_FUNC(MC) | |
6768 | #endif | |
6769 | ||
1d3504fc HS |
6770 | static int default_relax_domain_level = -1; |
6771 | ||
6772 | static int __init setup_relax_domain_level(char *str) | |
6773 | { | |
30e0e178 LZ |
6774 | unsigned long val; |
6775 | ||
6776 | val = simple_strtoul(str, NULL, 0); | |
6777 | if (val < SD_LV_MAX) | |
6778 | default_relax_domain_level = val; | |
6779 | ||
1d3504fc HS |
6780 | return 1; |
6781 | } | |
6782 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6783 | ||
6784 | static void set_domain_attribute(struct sched_domain *sd, | |
6785 | struct sched_domain_attr *attr) | |
6786 | { | |
6787 | int request; | |
6788 | ||
6789 | if (!attr || attr->relax_domain_level < 0) { | |
6790 | if (default_relax_domain_level < 0) | |
6791 | return; | |
6792 | else | |
6793 | request = default_relax_domain_level; | |
6794 | } else | |
6795 | request = attr->relax_domain_level; | |
6796 | if (request < sd->level) { | |
6797 | /* turn off idle balance on this domain */ | |
c88d5910 | 6798 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6799 | } else { |
6800 | /* turn on idle balance on this domain */ | |
c88d5910 | 6801 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6802 | } |
6803 | } | |
6804 | ||
2109b99e AH |
6805 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6806 | const struct cpumask *cpu_map) | |
6807 | { | |
6808 | switch (what) { | |
6809 | case sa_sched_groups: | |
6810 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
6811 | d->sched_group_nodes = NULL; | |
6812 | case sa_rootdomain: | |
6813 | free_rootdomain(d->rd); /* fall through */ | |
6814 | case sa_tmpmask: | |
6815 | free_cpumask_var(d->tmpmask); /* fall through */ | |
6816 | case sa_send_covered: | |
6817 | free_cpumask_var(d->send_covered); /* fall through */ | |
6818 | case sa_this_core_map: | |
6819 | free_cpumask_var(d->this_core_map); /* fall through */ | |
6820 | case sa_this_sibling_map: | |
6821 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
6822 | case sa_nodemask: | |
6823 | free_cpumask_var(d->nodemask); /* fall through */ | |
6824 | case sa_sched_group_nodes: | |
d1b55138 | 6825 | #ifdef CONFIG_NUMA |
2109b99e AH |
6826 | kfree(d->sched_group_nodes); /* fall through */ |
6827 | case sa_notcovered: | |
6828 | free_cpumask_var(d->notcovered); /* fall through */ | |
6829 | case sa_covered: | |
6830 | free_cpumask_var(d->covered); /* fall through */ | |
6831 | case sa_domainspan: | |
6832 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 6833 | #endif |
2109b99e AH |
6834 | case sa_none: |
6835 | break; | |
6836 | } | |
6837 | } | |
3404c8d9 | 6838 | |
2109b99e AH |
6839 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6840 | const struct cpumask *cpu_map) | |
6841 | { | |
3404c8d9 | 6842 | #ifdef CONFIG_NUMA |
2109b99e AH |
6843 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
6844 | return sa_none; | |
6845 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
6846 | return sa_domainspan; | |
6847 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
6848 | return sa_covered; | |
6849 | /* Allocate the per-node list of sched groups */ | |
6850 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
6851 | sizeof(struct sched_group *), GFP_KERNEL); | |
6852 | if (!d->sched_group_nodes) { | |
3df0fc5b | 6853 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 6854 | return sa_notcovered; |
d1b55138 | 6855 | } |
2109b99e | 6856 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 6857 | #endif |
2109b99e AH |
6858 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
6859 | return sa_sched_group_nodes; | |
6860 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
6861 | return sa_nodemask; | |
6862 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
6863 | return sa_this_sibling_map; | |
6864 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
6865 | return sa_this_core_map; | |
6866 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
6867 | return sa_send_covered; | |
6868 | d->rd = alloc_rootdomain(); | |
6869 | if (!d->rd) { | |
3df0fc5b | 6870 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 6871 | return sa_tmpmask; |
57d885fe | 6872 | } |
2109b99e AH |
6873 | return sa_rootdomain; |
6874 | } | |
57d885fe | 6875 | |
7f4588f3 AH |
6876 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
6877 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
6878 | { | |
6879 | struct sched_domain *sd = NULL; | |
7c16ec58 | 6880 | #ifdef CONFIG_NUMA |
7f4588f3 | 6881 | struct sched_domain *parent; |
1da177e4 | 6882 | |
7f4588f3 AH |
6883 | d->sd_allnodes = 0; |
6884 | if (cpumask_weight(cpu_map) > | |
6885 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
6886 | sd = &per_cpu(allnodes_domains, i).sd; | |
6887 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 6888 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
6889 | cpumask_copy(sched_domain_span(sd), cpu_map); |
6890 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6891 | d->sd_allnodes = 1; | |
6892 | } | |
6893 | parent = sd; | |
6894 | ||
6895 | sd = &per_cpu(node_domains, i).sd; | |
6896 | SD_INIT(sd, NODE); | |
6897 | set_domain_attribute(sd, attr); | |
6898 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
6899 | sd->parent = parent; | |
6900 | if (parent) | |
6901 | parent->child = sd; | |
6902 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 6903 | #endif |
7f4588f3 AH |
6904 | return sd; |
6905 | } | |
1da177e4 | 6906 | |
87cce662 AH |
6907 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
6908 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6909 | struct sched_domain *parent, int i) | |
6910 | { | |
6911 | struct sched_domain *sd; | |
6912 | sd = &per_cpu(phys_domains, i).sd; | |
6913 | SD_INIT(sd, CPU); | |
6914 | set_domain_attribute(sd, attr); | |
6915 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
6916 | sd->parent = parent; | |
6917 | if (parent) | |
6918 | parent->child = sd; | |
6919 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6920 | return sd; | |
6921 | } | |
1da177e4 | 6922 | |
410c4081 AH |
6923 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
6924 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6925 | struct sched_domain *parent, int i) | |
6926 | { | |
6927 | struct sched_domain *sd = parent; | |
1e9f28fa | 6928 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
6929 | sd = &per_cpu(core_domains, i).sd; |
6930 | SD_INIT(sd, MC); | |
6931 | set_domain_attribute(sd, attr); | |
6932 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
6933 | sd->parent = parent; | |
6934 | parent->child = sd; | |
6935 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 6936 | #endif |
410c4081 AH |
6937 | return sd; |
6938 | } | |
1e9f28fa | 6939 | |
d8173535 AH |
6940 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
6941 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6942 | struct sched_domain *parent, int i) | |
6943 | { | |
6944 | struct sched_domain *sd = parent; | |
1da177e4 | 6945 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
6946 | sd = &per_cpu(cpu_domains, i).sd; |
6947 | SD_INIT(sd, SIBLING); | |
6948 | set_domain_attribute(sd, attr); | |
6949 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
6950 | sd->parent = parent; | |
6951 | parent->child = sd; | |
6952 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 6953 | #endif |
d8173535 AH |
6954 | return sd; |
6955 | } | |
1da177e4 | 6956 | |
0e8e85c9 AH |
6957 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
6958 | const struct cpumask *cpu_map, int cpu) | |
6959 | { | |
6960 | switch (l) { | |
1da177e4 | 6961 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
6962 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
6963 | cpumask_and(d->this_sibling_map, cpu_map, | |
6964 | topology_thread_cpumask(cpu)); | |
6965 | if (cpu == cpumask_first(d->this_sibling_map)) | |
6966 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
6967 | &cpu_to_cpu_group, | |
6968 | d->send_covered, d->tmpmask); | |
6969 | break; | |
1da177e4 | 6970 | #endif |
1e9f28fa | 6971 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
6972 | case SD_LV_MC: /* set up multi-core groups */ |
6973 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
6974 | if (cpu == cpumask_first(d->this_core_map)) | |
6975 | init_sched_build_groups(d->this_core_map, cpu_map, | |
6976 | &cpu_to_core_group, | |
6977 | d->send_covered, d->tmpmask); | |
6978 | break; | |
1e9f28fa | 6979 | #endif |
86548096 AH |
6980 | case SD_LV_CPU: /* set up physical groups */ |
6981 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
6982 | if (!cpumask_empty(d->nodemask)) | |
6983 | init_sched_build_groups(d->nodemask, cpu_map, | |
6984 | &cpu_to_phys_group, | |
6985 | d->send_covered, d->tmpmask); | |
6986 | break; | |
1da177e4 | 6987 | #ifdef CONFIG_NUMA |
de616e36 AH |
6988 | case SD_LV_ALLNODES: |
6989 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
6990 | d->send_covered, d->tmpmask); | |
6991 | break; | |
6992 | #endif | |
0e8e85c9 AH |
6993 | default: |
6994 | break; | |
7c16ec58 | 6995 | } |
0e8e85c9 | 6996 | } |
9c1cfda2 | 6997 | |
2109b99e AH |
6998 | /* |
6999 | * Build sched domains for a given set of cpus and attach the sched domains | |
7000 | * to the individual cpus | |
7001 | */ | |
7002 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7003 | struct sched_domain_attr *attr) | |
7004 | { | |
7005 | enum s_alloc alloc_state = sa_none; | |
7006 | struct s_data d; | |
294b0c96 | 7007 | struct sched_domain *sd; |
2109b99e | 7008 | int i; |
7c16ec58 | 7009 | #ifdef CONFIG_NUMA |
2109b99e | 7010 | d.sd_allnodes = 0; |
7c16ec58 | 7011 | #endif |
9c1cfda2 | 7012 | |
2109b99e AH |
7013 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7014 | if (alloc_state != sa_rootdomain) | |
7015 | goto error; | |
7016 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7017 | |
1da177e4 | 7018 | /* |
1a20ff27 | 7019 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7020 | */ |
abcd083a | 7021 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7022 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7023 | cpu_map); | |
9761eea8 | 7024 | |
7f4588f3 | 7025 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7026 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7027 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7028 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7029 | } |
9c1cfda2 | 7030 | |
abcd083a | 7031 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7032 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 7033 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7034 | } |
9c1cfda2 | 7035 | |
1da177e4 | 7036 | /* Set up physical groups */ |
86548096 AH |
7037 | for (i = 0; i < nr_node_ids; i++) |
7038 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7039 | |
1da177e4 LT |
7040 | #ifdef CONFIG_NUMA |
7041 | /* Set up node groups */ | |
de616e36 AH |
7042 | if (d.sd_allnodes) |
7043 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7044 | |
0601a88d AH |
7045 | for (i = 0; i < nr_node_ids; i++) |
7046 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7047 | goto error; |
1da177e4 LT |
7048 | #endif |
7049 | ||
7050 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7051 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7052 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7053 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7054 | init_sched_groups_power(i, sd); |
5c45bf27 | 7055 | } |
1da177e4 | 7056 | #endif |
1e9f28fa | 7057 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7058 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7059 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7060 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7061 | } |
7062 | #endif | |
1e9f28fa | 7063 | |
abcd083a | 7064 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7065 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7066 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7067 | } |
7068 | ||
9c1cfda2 | 7069 | #ifdef CONFIG_NUMA |
076ac2af | 7070 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7071 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7072 | |
49a02c51 | 7073 | if (d.sd_allnodes) { |
6711cab4 | 7074 | struct sched_group *sg; |
f712c0c7 | 7075 | |
96f874e2 | 7076 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7077 | d.tmpmask); |
f712c0c7 SS |
7078 | init_numa_sched_groups_power(sg); |
7079 | } | |
9c1cfda2 JH |
7080 | #endif |
7081 | ||
1da177e4 | 7082 | /* Attach the domains */ |
abcd083a | 7083 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7084 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7085 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7086 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7087 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 7088 | #else |
6c99e9ad | 7089 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7090 | #endif |
49a02c51 | 7091 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7092 | } |
51888ca2 | 7093 | |
2109b99e AH |
7094 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7095 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7096 | return 0; | |
51888ca2 | 7097 | |
51888ca2 | 7098 | error: |
2109b99e AH |
7099 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7100 | return -ENOMEM; | |
1da177e4 | 7101 | } |
029190c5 | 7102 | |
96f874e2 | 7103 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7104 | { |
7105 | return __build_sched_domains(cpu_map, NULL); | |
7106 | } | |
7107 | ||
acc3f5d7 | 7108 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7109 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7110 | static struct sched_domain_attr *dattr_cur; |
7111 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7112 | |
7113 | /* | |
7114 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7115 | * cpumask) fails, then fallback to a single sched domain, |
7116 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7117 | */ |
4212823f | 7118 | static cpumask_var_t fallback_doms; |
029190c5 | 7119 | |
ee79d1bd HC |
7120 | /* |
7121 | * arch_update_cpu_topology lets virtualized architectures update the | |
7122 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7123 | * or 0 if it stayed the same. | |
7124 | */ | |
7125 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7126 | { |
ee79d1bd | 7127 | return 0; |
22e52b07 HC |
7128 | } |
7129 | ||
acc3f5d7 RR |
7130 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7131 | { | |
7132 | int i; | |
7133 | cpumask_var_t *doms; | |
7134 | ||
7135 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7136 | if (!doms) | |
7137 | return NULL; | |
7138 | for (i = 0; i < ndoms; i++) { | |
7139 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7140 | free_sched_domains(doms, i); | |
7141 | return NULL; | |
7142 | } | |
7143 | } | |
7144 | return doms; | |
7145 | } | |
7146 | ||
7147 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7148 | { | |
7149 | unsigned int i; | |
7150 | for (i = 0; i < ndoms; i++) | |
7151 | free_cpumask_var(doms[i]); | |
7152 | kfree(doms); | |
7153 | } | |
7154 | ||
1a20ff27 | 7155 | /* |
41a2d6cf | 7156 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7157 | * For now this just excludes isolated cpus, but could be used to |
7158 | * exclude other special cases in the future. | |
1a20ff27 | 7159 | */ |
96f874e2 | 7160 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7161 | { |
7378547f MM |
7162 | int err; |
7163 | ||
22e52b07 | 7164 | arch_update_cpu_topology(); |
029190c5 | 7165 | ndoms_cur = 1; |
acc3f5d7 | 7166 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7167 | if (!doms_cur) |
acc3f5d7 RR |
7168 | doms_cur = &fallback_doms; |
7169 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7170 | dattr_cur = NULL; |
acc3f5d7 | 7171 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7172 | register_sched_domain_sysctl(); |
7378547f MM |
7173 | |
7174 | return err; | |
1a20ff27 DG |
7175 | } |
7176 | ||
96f874e2 RR |
7177 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7178 | struct cpumask *tmpmask) | |
1da177e4 | 7179 | { |
7c16ec58 | 7180 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7181 | } |
1da177e4 | 7182 | |
1a20ff27 DG |
7183 | /* |
7184 | * Detach sched domains from a group of cpus specified in cpu_map | |
7185 | * These cpus will now be attached to the NULL domain | |
7186 | */ | |
96f874e2 | 7187 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7188 | { |
96f874e2 RR |
7189 | /* Save because hotplug lock held. */ |
7190 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7191 | int i; |
7192 | ||
abcd083a | 7193 | for_each_cpu(i, cpu_map) |
57d885fe | 7194 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7195 | synchronize_sched(); |
96f874e2 | 7196 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7197 | } |
7198 | ||
1d3504fc HS |
7199 | /* handle null as "default" */ |
7200 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7201 | struct sched_domain_attr *new, int idx_new) | |
7202 | { | |
7203 | struct sched_domain_attr tmp; | |
7204 | ||
7205 | /* fast path */ | |
7206 | if (!new && !cur) | |
7207 | return 1; | |
7208 | ||
7209 | tmp = SD_ATTR_INIT; | |
7210 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7211 | new ? (new + idx_new) : &tmp, | |
7212 | sizeof(struct sched_domain_attr)); | |
7213 | } | |
7214 | ||
029190c5 PJ |
7215 | /* |
7216 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7217 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7218 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7219 | * It destroys each deleted domain and builds each new domain. | |
7220 | * | |
acc3f5d7 | 7221 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7222 | * The masks don't intersect (don't overlap.) We should setup one |
7223 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7224 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7225 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7226 | * it as it is. | |
7227 | * | |
acc3f5d7 RR |
7228 | * The passed in 'doms_new' should be allocated using |
7229 | * alloc_sched_domains. This routine takes ownership of it and will | |
7230 | * free_sched_domains it when done with it. If the caller failed the | |
7231 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7232 | * and partition_sched_domains() will fallback to the single partition | |
7233 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7234 | * |
96f874e2 | 7235 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7236 | * ndoms_new == 0 is a special case for destroying existing domains, |
7237 | * and it will not create the default domain. | |
dfb512ec | 7238 | * |
029190c5 PJ |
7239 | * Call with hotplug lock held |
7240 | */ | |
acc3f5d7 | 7241 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7242 | struct sched_domain_attr *dattr_new) |
029190c5 | 7243 | { |
dfb512ec | 7244 | int i, j, n; |
d65bd5ec | 7245 | int new_topology; |
029190c5 | 7246 | |
712555ee | 7247 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7248 | |
7378547f MM |
7249 | /* always unregister in case we don't destroy any domains */ |
7250 | unregister_sched_domain_sysctl(); | |
7251 | ||
d65bd5ec HC |
7252 | /* Let architecture update cpu core mappings. */ |
7253 | new_topology = arch_update_cpu_topology(); | |
7254 | ||
dfb512ec | 7255 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7256 | |
7257 | /* Destroy deleted domains */ | |
7258 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7259 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7260 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7261 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7262 | goto match1; |
7263 | } | |
7264 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7265 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7266 | match1: |
7267 | ; | |
7268 | } | |
7269 | ||
e761b772 MK |
7270 | if (doms_new == NULL) { |
7271 | ndoms_cur = 0; | |
acc3f5d7 | 7272 | doms_new = &fallback_doms; |
6ad4c188 | 7273 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7274 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7275 | } |
7276 | ||
029190c5 PJ |
7277 | /* Build new domains */ |
7278 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7279 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7280 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7281 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7282 | goto match2; |
7283 | } | |
7284 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7285 | __build_sched_domains(doms_new[i], |
1d3504fc | 7286 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7287 | match2: |
7288 | ; | |
7289 | } | |
7290 | ||
7291 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7292 | if (doms_cur != &fallback_doms) |
7293 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7294 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7295 | doms_cur = doms_new; |
1d3504fc | 7296 | dattr_cur = dattr_new; |
029190c5 | 7297 | ndoms_cur = ndoms_new; |
7378547f MM |
7298 | |
7299 | register_sched_domain_sysctl(); | |
a1835615 | 7300 | |
712555ee | 7301 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7302 | } |
7303 | ||
5c45bf27 | 7304 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7305 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7306 | { |
95402b38 | 7307 | get_online_cpus(); |
dfb512ec MK |
7308 | |
7309 | /* Destroy domains first to force the rebuild */ | |
7310 | partition_sched_domains(0, NULL, NULL); | |
7311 | ||
e761b772 | 7312 | rebuild_sched_domains(); |
95402b38 | 7313 | put_online_cpus(); |
5c45bf27 SS |
7314 | } |
7315 | ||
7316 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7317 | { | |
afb8a9b7 | 7318 | unsigned int level = 0; |
5c45bf27 | 7319 | |
afb8a9b7 GS |
7320 | if (sscanf(buf, "%u", &level) != 1) |
7321 | return -EINVAL; | |
7322 | ||
7323 | /* | |
7324 | * level is always be positive so don't check for | |
7325 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7326 | * What happens on 0 or 1 byte write, | |
7327 | * need to check for count as well? | |
7328 | */ | |
7329 | ||
7330 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7331 | return -EINVAL; |
7332 | ||
7333 | if (smt) | |
afb8a9b7 | 7334 | sched_smt_power_savings = level; |
5c45bf27 | 7335 | else |
afb8a9b7 | 7336 | sched_mc_power_savings = level; |
5c45bf27 | 7337 | |
c70f22d2 | 7338 | arch_reinit_sched_domains(); |
5c45bf27 | 7339 | |
c70f22d2 | 7340 | return count; |
5c45bf27 SS |
7341 | } |
7342 | ||
5c45bf27 | 7343 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7344 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7345 | struct sysdev_class_attribute *attr, |
f718cd4a | 7346 | char *page) |
5c45bf27 SS |
7347 | { |
7348 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7349 | } | |
f718cd4a | 7350 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7351 | struct sysdev_class_attribute *attr, |
48f24c4d | 7352 | const char *buf, size_t count) |
5c45bf27 SS |
7353 | { |
7354 | return sched_power_savings_store(buf, count, 0); | |
7355 | } | |
f718cd4a AK |
7356 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7357 | sched_mc_power_savings_show, | |
7358 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7359 | #endif |
7360 | ||
7361 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7362 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7363 | struct sysdev_class_attribute *attr, |
f718cd4a | 7364 | char *page) |
5c45bf27 SS |
7365 | { |
7366 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7367 | } | |
f718cd4a | 7368 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7369 | struct sysdev_class_attribute *attr, |
48f24c4d | 7370 | const char *buf, size_t count) |
5c45bf27 SS |
7371 | { |
7372 | return sched_power_savings_store(buf, count, 1); | |
7373 | } | |
f718cd4a AK |
7374 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7375 | sched_smt_power_savings_show, | |
6707de00 AB |
7376 | sched_smt_power_savings_store); |
7377 | #endif | |
7378 | ||
39aac648 | 7379 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7380 | { |
7381 | int err = 0; | |
7382 | ||
7383 | #ifdef CONFIG_SCHED_SMT | |
7384 | if (smt_capable()) | |
7385 | err = sysfs_create_file(&cls->kset.kobj, | |
7386 | &attr_sched_smt_power_savings.attr); | |
7387 | #endif | |
7388 | #ifdef CONFIG_SCHED_MC | |
7389 | if (!err && mc_capable()) | |
7390 | err = sysfs_create_file(&cls->kset.kobj, | |
7391 | &attr_sched_mc_power_savings.attr); | |
7392 | #endif | |
7393 | return err; | |
7394 | } | |
6d6bc0ad | 7395 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7396 | |
e761b772 | 7397 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7398 | /* |
e761b772 MK |
7399 | * Add online and remove offline CPUs from the scheduler domains. |
7400 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7401 | */ |
7402 | static int update_sched_domains(struct notifier_block *nfb, | |
7403 | unsigned long action, void *hcpu) | |
e761b772 MK |
7404 | { |
7405 | switch (action) { | |
7406 | case CPU_ONLINE: | |
7407 | case CPU_ONLINE_FROZEN: | |
6ad4c188 PZ |
7408 | case CPU_DOWN_PREPARE: |
7409 | case CPU_DOWN_PREPARE_FROZEN: | |
7410 | case CPU_DOWN_FAILED: | |
7411 | case CPU_DOWN_FAILED_FROZEN: | |
dfb512ec | 7412 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7413 | return NOTIFY_OK; |
7414 | ||
7415 | default: | |
7416 | return NOTIFY_DONE; | |
7417 | } | |
7418 | } | |
7419 | #endif | |
7420 | ||
7421 | static int update_runtime(struct notifier_block *nfb, | |
7422 | unsigned long action, void *hcpu) | |
1da177e4 | 7423 | { |
7def2be1 PZ |
7424 | int cpu = (int)(long)hcpu; |
7425 | ||
1da177e4 | 7426 | switch (action) { |
1da177e4 | 7427 | case CPU_DOWN_PREPARE: |
8bb78442 | 7428 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7429 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7430 | return NOTIFY_OK; |
7431 | ||
1da177e4 | 7432 | case CPU_DOWN_FAILED: |
8bb78442 | 7433 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7434 | case CPU_ONLINE: |
8bb78442 | 7435 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7436 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7437 | return NOTIFY_OK; |
7438 | ||
1da177e4 LT |
7439 | default: |
7440 | return NOTIFY_DONE; | |
7441 | } | |
1da177e4 | 7442 | } |
1da177e4 LT |
7443 | |
7444 | void __init sched_init_smp(void) | |
7445 | { | |
dcc30a35 RR |
7446 | cpumask_var_t non_isolated_cpus; |
7447 | ||
7448 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7449 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7450 | |
434d53b0 MT |
7451 | #if defined(CONFIG_NUMA) |
7452 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7453 | GFP_KERNEL); | |
7454 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7455 | #endif | |
95402b38 | 7456 | get_online_cpus(); |
712555ee | 7457 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7458 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7459 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7460 | if (cpumask_empty(non_isolated_cpus)) | |
7461 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7462 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7463 | put_online_cpus(); |
e761b772 MK |
7464 | |
7465 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7466 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7467 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7468 | #endif |
7469 | ||
7470 | /* RT runtime code needs to handle some hotplug events */ | |
7471 | hotcpu_notifier(update_runtime, 0); | |
7472 | ||
b328ca18 | 7473 | init_hrtick(); |
5c1e1767 NP |
7474 | |
7475 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7476 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7477 | BUG(); |
19978ca6 | 7478 | sched_init_granularity(); |
dcc30a35 | 7479 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7480 | |
0e3900e6 | 7481 | init_sched_rt_class(); |
1da177e4 LT |
7482 | } |
7483 | #else | |
7484 | void __init sched_init_smp(void) | |
7485 | { | |
19978ca6 | 7486 | sched_init_granularity(); |
1da177e4 LT |
7487 | } |
7488 | #endif /* CONFIG_SMP */ | |
7489 | ||
cd1bb94b AB |
7490 | const_debug unsigned int sysctl_timer_migration = 1; |
7491 | ||
1da177e4 LT |
7492 | int in_sched_functions(unsigned long addr) |
7493 | { | |
1da177e4 LT |
7494 | return in_lock_functions(addr) || |
7495 | (addr >= (unsigned long)__sched_text_start | |
7496 | && addr < (unsigned long)__sched_text_end); | |
7497 | } | |
7498 | ||
a9957449 | 7499 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7500 | { |
7501 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7502 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7503 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7504 | cfs_rq->rq = rq; | |
7505 | #endif | |
67e9fb2a | 7506 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7507 | } |
7508 | ||
fa85ae24 PZ |
7509 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7510 | { | |
7511 | struct rt_prio_array *array; | |
7512 | int i; | |
7513 | ||
7514 | array = &rt_rq->active; | |
7515 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7516 | INIT_LIST_HEAD(array->queue + i); | |
7517 | __clear_bit(i, array->bitmap); | |
7518 | } | |
7519 | /* delimiter for bitsearch: */ | |
7520 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7521 | ||
052f1dc7 | 7522 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7523 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7524 | #ifdef CONFIG_SMP |
e864c499 | 7525 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7526 | #endif |
48d5e258 | 7527 | #endif |
fa85ae24 PZ |
7528 | #ifdef CONFIG_SMP |
7529 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7530 | rt_rq->overloaded = 0; |
05fa785c | 7531 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7532 | #endif |
7533 | ||
7534 | rt_rq->rt_time = 0; | |
7535 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7536 | rt_rq->rt_runtime = 0; |
0986b11b | 7537 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7538 | |
052f1dc7 | 7539 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7540 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7541 | rt_rq->rq = rq; |
7542 | #endif | |
fa85ae24 PZ |
7543 | } |
7544 | ||
6f505b16 | 7545 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7546 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7547 | struct sched_entity *se, int cpu, int add, | |
7548 | struct sched_entity *parent) | |
6f505b16 | 7549 | { |
ec7dc8ac | 7550 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7551 | tg->cfs_rq[cpu] = cfs_rq; |
7552 | init_cfs_rq(cfs_rq, rq); | |
7553 | cfs_rq->tg = tg; | |
7554 | if (add) | |
7555 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7556 | ||
7557 | tg->se[cpu] = se; | |
354d60c2 DG |
7558 | /* se could be NULL for init_task_group */ |
7559 | if (!se) | |
7560 | return; | |
7561 | ||
ec7dc8ac DG |
7562 | if (!parent) |
7563 | se->cfs_rq = &rq->cfs; | |
7564 | else | |
7565 | se->cfs_rq = parent->my_q; | |
7566 | ||
6f505b16 PZ |
7567 | se->my_q = cfs_rq; |
7568 | se->load.weight = tg->shares; | |
e05510d0 | 7569 | se->load.inv_weight = 0; |
ec7dc8ac | 7570 | se->parent = parent; |
6f505b16 | 7571 | } |
052f1dc7 | 7572 | #endif |
6f505b16 | 7573 | |
052f1dc7 | 7574 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7575 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7576 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7577 | struct sched_rt_entity *parent) | |
6f505b16 | 7578 | { |
ec7dc8ac DG |
7579 | struct rq *rq = cpu_rq(cpu); |
7580 | ||
6f505b16 PZ |
7581 | tg->rt_rq[cpu] = rt_rq; |
7582 | init_rt_rq(rt_rq, rq); | |
7583 | rt_rq->tg = tg; | |
ac086bc2 | 7584 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7585 | if (add) |
7586 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7587 | ||
7588 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7589 | if (!rt_se) |
7590 | return; | |
7591 | ||
ec7dc8ac DG |
7592 | if (!parent) |
7593 | rt_se->rt_rq = &rq->rt; | |
7594 | else | |
7595 | rt_se->rt_rq = parent->my_q; | |
7596 | ||
6f505b16 | 7597 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7598 | rt_se->parent = parent; |
6f505b16 PZ |
7599 | INIT_LIST_HEAD(&rt_se->run_list); |
7600 | } | |
7601 | #endif | |
7602 | ||
1da177e4 LT |
7603 | void __init sched_init(void) |
7604 | { | |
dd41f596 | 7605 | int i, j; |
434d53b0 MT |
7606 | unsigned long alloc_size = 0, ptr; |
7607 | ||
7608 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7609 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7610 | #endif | |
7611 | #ifdef CONFIG_RT_GROUP_SCHED | |
7612 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7613 | #endif |
df7c8e84 | 7614 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7615 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7616 | #endif |
434d53b0 | 7617 | if (alloc_size) { |
36b7b6d4 | 7618 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7619 | |
7620 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7621 | init_task_group.se = (struct sched_entity **)ptr; | |
7622 | ptr += nr_cpu_ids * sizeof(void **); | |
7623 | ||
7624 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7625 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7626 | |
6d6bc0ad | 7627 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
7628 | #ifdef CONFIG_RT_GROUP_SCHED |
7629 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7630 | ptr += nr_cpu_ids * sizeof(void **); | |
7631 | ||
7632 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7633 | ptr += nr_cpu_ids * sizeof(void **); |
7634 | ||
6d6bc0ad | 7635 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7636 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7637 | for_each_possible_cpu(i) { | |
7638 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7639 | ptr += cpumask_size(); | |
7640 | } | |
7641 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7642 | } |
dd41f596 | 7643 | |
57d885fe GH |
7644 | #ifdef CONFIG_SMP |
7645 | init_defrootdomain(); | |
7646 | #endif | |
7647 | ||
d0b27fa7 PZ |
7648 | init_rt_bandwidth(&def_rt_bandwidth, |
7649 | global_rt_period(), global_rt_runtime()); | |
7650 | ||
7651 | #ifdef CONFIG_RT_GROUP_SCHED | |
7652 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7653 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 7654 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7655 | |
7c941438 | 7656 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 7657 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7658 | INIT_LIST_HEAD(&init_task_group.children); |
7659 | ||
7c941438 | 7660 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7661 | |
4a6cc4bd JK |
7662 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
7663 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
7664 | __alignof__(unsigned long)); | |
7665 | #endif | |
0a945022 | 7666 | for_each_possible_cpu(i) { |
70b97a7f | 7667 | struct rq *rq; |
1da177e4 LT |
7668 | |
7669 | rq = cpu_rq(i); | |
05fa785c | 7670 | raw_spin_lock_init(&rq->lock); |
7897986b | 7671 | rq->nr_running = 0; |
dce48a84 TG |
7672 | rq->calc_load_active = 0; |
7673 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7674 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7675 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7676 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7677 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7678 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7679 | #ifdef CONFIG_CGROUP_SCHED |
7680 | /* | |
7681 | * How much cpu bandwidth does init_task_group get? | |
7682 | * | |
7683 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7684 | * gets 100% of the cpu resources in the system. This overall | |
7685 | * system cpu resource is divided among the tasks of | |
7686 | * init_task_group and its child task-groups in a fair manner, | |
7687 | * based on each entity's (task or task-group's) weight | |
7688 | * (se->load.weight). | |
7689 | * | |
7690 | * In other words, if init_task_group has 10 tasks of weight | |
7691 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7692 | * then A0's share of the cpu resource is: | |
7693 | * | |
0d905bca | 7694 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
7695 | * |
7696 | * We achieve this by letting init_task_group's tasks sit | |
7697 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7698 | */ | |
ec7dc8ac | 7699 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
052f1dc7 | 7700 | #endif |
354d60c2 DG |
7701 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7702 | ||
7703 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7704 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7705 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7706 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 7707 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 7708 | #endif |
dd41f596 | 7709 | #endif |
1da177e4 | 7710 | |
dd41f596 IM |
7711 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7712 | rq->cpu_load[j] = 0; | |
1da177e4 | 7713 | #ifdef CONFIG_SMP |
41c7ce9a | 7714 | rq->sd = NULL; |
57d885fe | 7715 | rq->rd = NULL; |
3f029d3c | 7716 | rq->post_schedule = 0; |
1da177e4 | 7717 | rq->active_balance = 0; |
dd41f596 | 7718 | rq->next_balance = jiffies; |
1da177e4 | 7719 | rq->push_cpu = 0; |
0a2966b4 | 7720 | rq->cpu = i; |
1f11eb6a | 7721 | rq->online = 0; |
1da177e4 | 7722 | rq->migration_thread = NULL; |
eae0c9df MG |
7723 | rq->idle_stamp = 0; |
7724 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
1da177e4 | 7725 | INIT_LIST_HEAD(&rq->migration_queue); |
dc938520 | 7726 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7727 | #endif |
8f4d37ec | 7728 | init_rq_hrtick(rq); |
1da177e4 | 7729 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7730 | } |
7731 | ||
2dd73a4f | 7732 | set_load_weight(&init_task); |
b50f60ce | 7733 | |
e107be36 AK |
7734 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7735 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7736 | #endif | |
7737 | ||
c9819f45 | 7738 | #ifdef CONFIG_SMP |
962cf36c | 7739 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7740 | #endif |
7741 | ||
b50f60ce | 7742 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7743 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7744 | #endif |
7745 | ||
1da177e4 LT |
7746 | /* |
7747 | * The boot idle thread does lazy MMU switching as well: | |
7748 | */ | |
7749 | atomic_inc(&init_mm.mm_count); | |
7750 | enter_lazy_tlb(&init_mm, current); | |
7751 | ||
7752 | /* | |
7753 | * Make us the idle thread. Technically, schedule() should not be | |
7754 | * called from this thread, however somewhere below it might be, | |
7755 | * but because we are the idle thread, we just pick up running again | |
7756 | * when this runqueue becomes "idle". | |
7757 | */ | |
7758 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7759 | |
7760 | calc_load_update = jiffies + LOAD_FREQ; | |
7761 | ||
dd41f596 IM |
7762 | /* |
7763 | * During early bootup we pretend to be a normal task: | |
7764 | */ | |
7765 | current->sched_class = &fair_sched_class; | |
6892b75e | 7766 | |
6a7b3dc3 | 7767 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7768 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7769 | #ifdef CONFIG_SMP |
7d1e6a9b | 7770 | #ifdef CONFIG_NO_HZ |
49557e62 | 7771 | zalloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
4bdddf8f | 7772 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); |
7d1e6a9b | 7773 | #endif |
bdddd296 RR |
7774 | /* May be allocated at isolcpus cmdline parse time */ |
7775 | if (cpu_isolated_map == NULL) | |
7776 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7777 | #endif /* SMP */ |
6a7b3dc3 | 7778 | |
cdd6c482 | 7779 | perf_event_init(); |
0d905bca | 7780 | |
6892b75e | 7781 | scheduler_running = 1; |
1da177e4 LT |
7782 | } |
7783 | ||
7784 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7785 | static inline int preempt_count_equals(int preempt_offset) |
7786 | { | |
234da7bc | 7787 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
7788 | |
7789 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
7790 | } | |
7791 | ||
d894837f | 7792 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7793 | { |
48f24c4d | 7794 | #ifdef in_atomic |
1da177e4 LT |
7795 | static unsigned long prev_jiffy; /* ratelimiting */ |
7796 | ||
e4aafea2 FW |
7797 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7798 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7799 | return; |
7800 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7801 | return; | |
7802 | prev_jiffy = jiffies; | |
7803 | ||
3df0fc5b PZ |
7804 | printk(KERN_ERR |
7805 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7806 | file, line); | |
7807 | printk(KERN_ERR | |
7808 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7809 | in_atomic(), irqs_disabled(), | |
7810 | current->pid, current->comm); | |
aef745fc IM |
7811 | |
7812 | debug_show_held_locks(current); | |
7813 | if (irqs_disabled()) | |
7814 | print_irqtrace_events(current); | |
7815 | dump_stack(); | |
1da177e4 LT |
7816 | #endif |
7817 | } | |
7818 | EXPORT_SYMBOL(__might_sleep); | |
7819 | #endif | |
7820 | ||
7821 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7822 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7823 | { | |
7824 | int on_rq; | |
3e51f33f | 7825 | |
3a5e4dc1 AK |
7826 | on_rq = p->se.on_rq; |
7827 | if (on_rq) | |
7828 | deactivate_task(rq, p, 0); | |
7829 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7830 | if (on_rq) { | |
7831 | activate_task(rq, p, 0); | |
7832 | resched_task(rq->curr); | |
7833 | } | |
7834 | } | |
7835 | ||
1da177e4 LT |
7836 | void normalize_rt_tasks(void) |
7837 | { | |
a0f98a1c | 7838 | struct task_struct *g, *p; |
1da177e4 | 7839 | unsigned long flags; |
70b97a7f | 7840 | struct rq *rq; |
1da177e4 | 7841 | |
4cf5d77a | 7842 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7843 | do_each_thread(g, p) { |
178be793 IM |
7844 | /* |
7845 | * Only normalize user tasks: | |
7846 | */ | |
7847 | if (!p->mm) | |
7848 | continue; | |
7849 | ||
6cfb0d5d | 7850 | p->se.exec_start = 0; |
6cfb0d5d | 7851 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7852 | p->se.statistics.wait_start = 0; |
7853 | p->se.statistics.sleep_start = 0; | |
7854 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7855 | #endif |
dd41f596 IM |
7856 | |
7857 | if (!rt_task(p)) { | |
7858 | /* | |
7859 | * Renice negative nice level userspace | |
7860 | * tasks back to 0: | |
7861 | */ | |
7862 | if (TASK_NICE(p) < 0 && p->mm) | |
7863 | set_user_nice(p, 0); | |
1da177e4 | 7864 | continue; |
dd41f596 | 7865 | } |
1da177e4 | 7866 | |
1d615482 | 7867 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7868 | rq = __task_rq_lock(p); |
1da177e4 | 7869 | |
178be793 | 7870 | normalize_task(rq, p); |
3a5e4dc1 | 7871 | |
b29739f9 | 7872 | __task_rq_unlock(rq); |
1d615482 | 7873 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7874 | } while_each_thread(g, p); |
7875 | ||
4cf5d77a | 7876 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7877 | } |
7878 | ||
7879 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7880 | |
7881 | #ifdef CONFIG_IA64 | |
7882 | /* | |
7883 | * These functions are only useful for the IA64 MCA handling. | |
7884 | * | |
7885 | * They can only be called when the whole system has been | |
7886 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7887 | * activity can take place. Using them for anything else would | |
7888 | * be a serious bug, and as a result, they aren't even visible | |
7889 | * under any other configuration. | |
7890 | */ | |
7891 | ||
7892 | /** | |
7893 | * curr_task - return the current task for a given cpu. | |
7894 | * @cpu: the processor in question. | |
7895 | * | |
7896 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7897 | */ | |
36c8b586 | 7898 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7899 | { |
7900 | return cpu_curr(cpu); | |
7901 | } | |
7902 | ||
7903 | /** | |
7904 | * set_curr_task - set the current task for a given cpu. | |
7905 | * @cpu: the processor in question. | |
7906 | * @p: the task pointer to set. | |
7907 | * | |
7908 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7909 | * are serviced on a separate stack. It allows the architecture to switch the |
7910 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7911 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7912 | * and caller must save the original value of the current task (see | |
7913 | * curr_task() above) and restore that value before reenabling interrupts and | |
7914 | * re-starting the system. | |
7915 | * | |
7916 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7917 | */ | |
36c8b586 | 7918 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7919 | { |
7920 | cpu_curr(cpu) = p; | |
7921 | } | |
7922 | ||
7923 | #endif | |
29f59db3 | 7924 | |
bccbe08a PZ |
7925 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7926 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
7927 | { |
7928 | int i; | |
7929 | ||
7930 | for_each_possible_cpu(i) { | |
7931 | if (tg->cfs_rq) | |
7932 | kfree(tg->cfs_rq[i]); | |
7933 | if (tg->se) | |
7934 | kfree(tg->se[i]); | |
6f505b16 PZ |
7935 | } |
7936 | ||
7937 | kfree(tg->cfs_rq); | |
7938 | kfree(tg->se); | |
6f505b16 PZ |
7939 | } |
7940 | ||
ec7dc8ac DG |
7941 | static |
7942 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 7943 | { |
29f59db3 | 7944 | struct cfs_rq *cfs_rq; |
eab17229 | 7945 | struct sched_entity *se; |
9b5b7751 | 7946 | struct rq *rq; |
29f59db3 SV |
7947 | int i; |
7948 | ||
434d53b0 | 7949 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7950 | if (!tg->cfs_rq) |
7951 | goto err; | |
434d53b0 | 7952 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7953 | if (!tg->se) |
7954 | goto err; | |
052f1dc7 PZ |
7955 | |
7956 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
7957 | |
7958 | for_each_possible_cpu(i) { | |
9b5b7751 | 7959 | rq = cpu_rq(i); |
29f59db3 | 7960 | |
eab17229 LZ |
7961 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
7962 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
7963 | if (!cfs_rq) |
7964 | goto err; | |
7965 | ||
eab17229 LZ |
7966 | se = kzalloc_node(sizeof(struct sched_entity), |
7967 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 7968 | if (!se) |
dfc12eb2 | 7969 | goto err_free_rq; |
29f59db3 | 7970 | |
eab17229 | 7971 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
7972 | } |
7973 | ||
7974 | return 1; | |
7975 | ||
dfc12eb2 PC |
7976 | err_free_rq: |
7977 | kfree(cfs_rq); | |
bccbe08a PZ |
7978 | err: |
7979 | return 0; | |
7980 | } | |
7981 | ||
7982 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7983 | { | |
7984 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
7985 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
7986 | } | |
7987 | ||
7988 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7989 | { | |
7990 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
7991 | } | |
6d6bc0ad | 7992 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
7993 | static inline void free_fair_sched_group(struct task_group *tg) |
7994 | { | |
7995 | } | |
7996 | ||
ec7dc8ac DG |
7997 | static inline |
7998 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7999 | { |
8000 | return 1; | |
8001 | } | |
8002 | ||
8003 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8004 | { | |
8005 | } | |
8006 | ||
8007 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8008 | { | |
8009 | } | |
6d6bc0ad | 8010 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8011 | |
8012 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8013 | static void free_rt_sched_group(struct task_group *tg) |
8014 | { | |
8015 | int i; | |
8016 | ||
d0b27fa7 PZ |
8017 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8018 | ||
bccbe08a PZ |
8019 | for_each_possible_cpu(i) { |
8020 | if (tg->rt_rq) | |
8021 | kfree(tg->rt_rq[i]); | |
8022 | if (tg->rt_se) | |
8023 | kfree(tg->rt_se[i]); | |
8024 | } | |
8025 | ||
8026 | kfree(tg->rt_rq); | |
8027 | kfree(tg->rt_se); | |
8028 | } | |
8029 | ||
ec7dc8ac DG |
8030 | static |
8031 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8032 | { |
8033 | struct rt_rq *rt_rq; | |
eab17229 | 8034 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8035 | struct rq *rq; |
8036 | int i; | |
8037 | ||
434d53b0 | 8038 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8039 | if (!tg->rt_rq) |
8040 | goto err; | |
434d53b0 | 8041 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8042 | if (!tg->rt_se) |
8043 | goto err; | |
8044 | ||
d0b27fa7 PZ |
8045 | init_rt_bandwidth(&tg->rt_bandwidth, |
8046 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8047 | |
8048 | for_each_possible_cpu(i) { | |
8049 | rq = cpu_rq(i); | |
8050 | ||
eab17229 LZ |
8051 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8052 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8053 | if (!rt_rq) |
8054 | goto err; | |
29f59db3 | 8055 | |
eab17229 LZ |
8056 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8057 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8058 | if (!rt_se) |
dfc12eb2 | 8059 | goto err_free_rq; |
29f59db3 | 8060 | |
eab17229 | 8061 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
8062 | } |
8063 | ||
bccbe08a PZ |
8064 | return 1; |
8065 | ||
dfc12eb2 PC |
8066 | err_free_rq: |
8067 | kfree(rt_rq); | |
bccbe08a PZ |
8068 | err: |
8069 | return 0; | |
8070 | } | |
8071 | ||
8072 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8073 | { | |
8074 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8075 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8076 | } | |
8077 | ||
8078 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8079 | { | |
8080 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8081 | } | |
6d6bc0ad | 8082 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8083 | static inline void free_rt_sched_group(struct task_group *tg) |
8084 | { | |
8085 | } | |
8086 | ||
ec7dc8ac DG |
8087 | static inline |
8088 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8089 | { |
8090 | return 1; | |
8091 | } | |
8092 | ||
8093 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8094 | { | |
8095 | } | |
8096 | ||
8097 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8098 | { | |
8099 | } | |
6d6bc0ad | 8100 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8101 | |
7c941438 | 8102 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8103 | static void free_sched_group(struct task_group *tg) |
8104 | { | |
8105 | free_fair_sched_group(tg); | |
8106 | free_rt_sched_group(tg); | |
8107 | kfree(tg); | |
8108 | } | |
8109 | ||
8110 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8111 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8112 | { |
8113 | struct task_group *tg; | |
8114 | unsigned long flags; | |
8115 | int i; | |
8116 | ||
8117 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8118 | if (!tg) | |
8119 | return ERR_PTR(-ENOMEM); | |
8120 | ||
ec7dc8ac | 8121 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8122 | goto err; |
8123 | ||
ec7dc8ac | 8124 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8125 | goto err; |
8126 | ||
8ed36996 | 8127 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8128 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8129 | register_fair_sched_group(tg, i); |
8130 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8131 | } |
6f505b16 | 8132 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8133 | |
8134 | WARN_ON(!parent); /* root should already exist */ | |
8135 | ||
8136 | tg->parent = parent; | |
f473aa5e | 8137 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8138 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8139 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8140 | |
9b5b7751 | 8141 | return tg; |
29f59db3 SV |
8142 | |
8143 | err: | |
6f505b16 | 8144 | free_sched_group(tg); |
29f59db3 SV |
8145 | return ERR_PTR(-ENOMEM); |
8146 | } | |
8147 | ||
9b5b7751 | 8148 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8149 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8150 | { |
29f59db3 | 8151 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8152 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8153 | } |
8154 | ||
9b5b7751 | 8155 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8156 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8157 | { |
8ed36996 | 8158 | unsigned long flags; |
9b5b7751 | 8159 | int i; |
29f59db3 | 8160 | |
8ed36996 | 8161 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8162 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8163 | unregister_fair_sched_group(tg, i); |
8164 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8165 | } |
6f505b16 | 8166 | list_del_rcu(&tg->list); |
f473aa5e | 8167 | list_del_rcu(&tg->siblings); |
8ed36996 | 8168 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8169 | |
9b5b7751 | 8170 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8171 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8172 | } |
8173 | ||
9b5b7751 | 8174 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8175 | * The caller of this function should have put the task in its new group |
8176 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8177 | * reflect its new group. | |
9b5b7751 SV |
8178 | */ |
8179 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8180 | { |
8181 | int on_rq, running; | |
8182 | unsigned long flags; | |
8183 | struct rq *rq; | |
8184 | ||
8185 | rq = task_rq_lock(tsk, &flags); | |
8186 | ||
051a1d1a | 8187 | running = task_current(rq, tsk); |
29f59db3 SV |
8188 | on_rq = tsk->se.on_rq; |
8189 | ||
0e1f3483 | 8190 | if (on_rq) |
29f59db3 | 8191 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8192 | if (unlikely(running)) |
8193 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8194 | |
6f505b16 | 8195 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8196 | |
810b3817 PZ |
8197 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8198 | if (tsk->sched_class->moved_group) | |
88ec22d3 | 8199 | tsk->sched_class->moved_group(tsk, on_rq); |
810b3817 PZ |
8200 | #endif |
8201 | ||
0e1f3483 HS |
8202 | if (unlikely(running)) |
8203 | tsk->sched_class->set_curr_task(rq); | |
8204 | if (on_rq) | |
ea87bb78 | 8205 | enqueue_task(rq, tsk, 0, false); |
29f59db3 | 8206 | |
29f59db3 SV |
8207 | task_rq_unlock(rq, &flags); |
8208 | } | |
7c941438 | 8209 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8210 | |
052f1dc7 | 8211 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8212 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8213 | { |
8214 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8215 | int on_rq; |
8216 | ||
29f59db3 | 8217 | on_rq = se->on_rq; |
62fb1851 | 8218 | if (on_rq) |
29f59db3 SV |
8219 | dequeue_entity(cfs_rq, se, 0); |
8220 | ||
8221 | se->load.weight = shares; | |
e05510d0 | 8222 | se->load.inv_weight = 0; |
29f59db3 | 8223 | |
62fb1851 | 8224 | if (on_rq) |
29f59db3 | 8225 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8226 | } |
62fb1851 | 8227 | |
c09595f6 PZ |
8228 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8229 | { | |
8230 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8231 | struct rq *rq = cfs_rq->rq; | |
8232 | unsigned long flags; | |
8233 | ||
05fa785c | 8234 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 8235 | __set_se_shares(se, shares); |
05fa785c | 8236 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
8237 | } |
8238 | ||
8ed36996 PZ |
8239 | static DEFINE_MUTEX(shares_mutex); |
8240 | ||
4cf86d77 | 8241 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8242 | { |
8243 | int i; | |
8ed36996 | 8244 | unsigned long flags; |
c61935fd | 8245 | |
ec7dc8ac DG |
8246 | /* |
8247 | * We can't change the weight of the root cgroup. | |
8248 | */ | |
8249 | if (!tg->se[0]) | |
8250 | return -EINVAL; | |
8251 | ||
18d95a28 PZ |
8252 | if (shares < MIN_SHARES) |
8253 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8254 | else if (shares > MAX_SHARES) |
8255 | shares = MAX_SHARES; | |
62fb1851 | 8256 | |
8ed36996 | 8257 | mutex_lock(&shares_mutex); |
9b5b7751 | 8258 | if (tg->shares == shares) |
5cb350ba | 8259 | goto done; |
29f59db3 | 8260 | |
8ed36996 | 8261 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8262 | for_each_possible_cpu(i) |
8263 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8264 | list_del_rcu(&tg->siblings); |
8ed36996 | 8265 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8266 | |
8267 | /* wait for any ongoing reference to this group to finish */ | |
8268 | synchronize_sched(); | |
8269 | ||
8270 | /* | |
8271 | * Now we are free to modify the group's share on each cpu | |
8272 | * w/o tripping rebalance_share or load_balance_fair. | |
8273 | */ | |
9b5b7751 | 8274 | tg->shares = shares; |
c09595f6 PZ |
8275 | for_each_possible_cpu(i) { |
8276 | /* | |
8277 | * force a rebalance | |
8278 | */ | |
8279 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8280 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8281 | } |
29f59db3 | 8282 | |
6b2d7700 SV |
8283 | /* |
8284 | * Enable load balance activity on this group, by inserting it back on | |
8285 | * each cpu's rq->leaf_cfs_rq_list. | |
8286 | */ | |
8ed36996 | 8287 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8288 | for_each_possible_cpu(i) |
8289 | register_fair_sched_group(tg, i); | |
f473aa5e | 8290 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8291 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8292 | done: |
8ed36996 | 8293 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8294 | return 0; |
29f59db3 SV |
8295 | } |
8296 | ||
5cb350ba DG |
8297 | unsigned long sched_group_shares(struct task_group *tg) |
8298 | { | |
8299 | return tg->shares; | |
8300 | } | |
052f1dc7 | 8301 | #endif |
5cb350ba | 8302 | |
052f1dc7 | 8303 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8304 | /* |
9f0c1e56 | 8305 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8306 | */ |
9f0c1e56 PZ |
8307 | static DEFINE_MUTEX(rt_constraints_mutex); |
8308 | ||
8309 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8310 | { | |
8311 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8312 | return 1ULL << 20; |
9f0c1e56 | 8313 | |
9a7e0b18 | 8314 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8315 | } |
8316 | ||
9a7e0b18 PZ |
8317 | /* Must be called with tasklist_lock held */ |
8318 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8319 | { |
9a7e0b18 | 8320 | struct task_struct *g, *p; |
b40b2e8e | 8321 | |
9a7e0b18 PZ |
8322 | do_each_thread(g, p) { |
8323 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8324 | return 1; | |
8325 | } while_each_thread(g, p); | |
b40b2e8e | 8326 | |
9a7e0b18 PZ |
8327 | return 0; |
8328 | } | |
b40b2e8e | 8329 | |
9a7e0b18 PZ |
8330 | struct rt_schedulable_data { |
8331 | struct task_group *tg; | |
8332 | u64 rt_period; | |
8333 | u64 rt_runtime; | |
8334 | }; | |
b40b2e8e | 8335 | |
9a7e0b18 PZ |
8336 | static int tg_schedulable(struct task_group *tg, void *data) |
8337 | { | |
8338 | struct rt_schedulable_data *d = data; | |
8339 | struct task_group *child; | |
8340 | unsigned long total, sum = 0; | |
8341 | u64 period, runtime; | |
b40b2e8e | 8342 | |
9a7e0b18 PZ |
8343 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8344 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8345 | |
9a7e0b18 PZ |
8346 | if (tg == d->tg) { |
8347 | period = d->rt_period; | |
8348 | runtime = d->rt_runtime; | |
b40b2e8e | 8349 | } |
b40b2e8e | 8350 | |
4653f803 PZ |
8351 | /* |
8352 | * Cannot have more runtime than the period. | |
8353 | */ | |
8354 | if (runtime > period && runtime != RUNTIME_INF) | |
8355 | return -EINVAL; | |
6f505b16 | 8356 | |
4653f803 PZ |
8357 | /* |
8358 | * Ensure we don't starve existing RT tasks. | |
8359 | */ | |
9a7e0b18 PZ |
8360 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8361 | return -EBUSY; | |
6f505b16 | 8362 | |
9a7e0b18 | 8363 | total = to_ratio(period, runtime); |
6f505b16 | 8364 | |
4653f803 PZ |
8365 | /* |
8366 | * Nobody can have more than the global setting allows. | |
8367 | */ | |
8368 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8369 | return -EINVAL; | |
6f505b16 | 8370 | |
4653f803 PZ |
8371 | /* |
8372 | * The sum of our children's runtime should not exceed our own. | |
8373 | */ | |
9a7e0b18 PZ |
8374 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8375 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8376 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8377 | |
9a7e0b18 PZ |
8378 | if (child == d->tg) { |
8379 | period = d->rt_period; | |
8380 | runtime = d->rt_runtime; | |
8381 | } | |
6f505b16 | 8382 | |
9a7e0b18 | 8383 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8384 | } |
6f505b16 | 8385 | |
9a7e0b18 PZ |
8386 | if (sum > total) |
8387 | return -EINVAL; | |
8388 | ||
8389 | return 0; | |
6f505b16 PZ |
8390 | } |
8391 | ||
9a7e0b18 | 8392 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8393 | { |
9a7e0b18 PZ |
8394 | struct rt_schedulable_data data = { |
8395 | .tg = tg, | |
8396 | .rt_period = period, | |
8397 | .rt_runtime = runtime, | |
8398 | }; | |
8399 | ||
8400 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8401 | } |
8402 | ||
d0b27fa7 PZ |
8403 | static int tg_set_bandwidth(struct task_group *tg, |
8404 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8405 | { |
ac086bc2 | 8406 | int i, err = 0; |
9f0c1e56 | 8407 | |
9f0c1e56 | 8408 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8409 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8410 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8411 | if (err) | |
9f0c1e56 | 8412 | goto unlock; |
ac086bc2 | 8413 | |
0986b11b | 8414 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8415 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8416 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8417 | |
8418 | for_each_possible_cpu(i) { | |
8419 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8420 | ||
0986b11b | 8421 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8422 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8423 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8424 | } |
0986b11b | 8425 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
9f0c1e56 | 8426 | unlock: |
521f1a24 | 8427 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8428 | mutex_unlock(&rt_constraints_mutex); |
8429 | ||
8430 | return err; | |
6f505b16 PZ |
8431 | } |
8432 | ||
d0b27fa7 PZ |
8433 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8434 | { | |
8435 | u64 rt_runtime, rt_period; | |
8436 | ||
8437 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8438 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8439 | if (rt_runtime_us < 0) | |
8440 | rt_runtime = RUNTIME_INF; | |
8441 | ||
8442 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8443 | } | |
8444 | ||
9f0c1e56 PZ |
8445 | long sched_group_rt_runtime(struct task_group *tg) |
8446 | { | |
8447 | u64 rt_runtime_us; | |
8448 | ||
d0b27fa7 | 8449 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8450 | return -1; |
8451 | ||
d0b27fa7 | 8452 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8453 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8454 | return rt_runtime_us; | |
8455 | } | |
d0b27fa7 PZ |
8456 | |
8457 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8458 | { | |
8459 | u64 rt_runtime, rt_period; | |
8460 | ||
8461 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8462 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8463 | ||
619b0488 R |
8464 | if (rt_period == 0) |
8465 | return -EINVAL; | |
8466 | ||
d0b27fa7 PZ |
8467 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8468 | } | |
8469 | ||
8470 | long sched_group_rt_period(struct task_group *tg) | |
8471 | { | |
8472 | u64 rt_period_us; | |
8473 | ||
8474 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8475 | do_div(rt_period_us, NSEC_PER_USEC); | |
8476 | return rt_period_us; | |
8477 | } | |
8478 | ||
8479 | static int sched_rt_global_constraints(void) | |
8480 | { | |
4653f803 | 8481 | u64 runtime, period; |
d0b27fa7 PZ |
8482 | int ret = 0; |
8483 | ||
ec5d4989 HS |
8484 | if (sysctl_sched_rt_period <= 0) |
8485 | return -EINVAL; | |
8486 | ||
4653f803 PZ |
8487 | runtime = global_rt_runtime(); |
8488 | period = global_rt_period(); | |
8489 | ||
8490 | /* | |
8491 | * Sanity check on the sysctl variables. | |
8492 | */ | |
8493 | if (runtime > period && runtime != RUNTIME_INF) | |
8494 | return -EINVAL; | |
10b612f4 | 8495 | |
d0b27fa7 | 8496 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8497 | read_lock(&tasklist_lock); |
4653f803 | 8498 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8499 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8500 | mutex_unlock(&rt_constraints_mutex); |
8501 | ||
8502 | return ret; | |
8503 | } | |
54e99124 DG |
8504 | |
8505 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8506 | { | |
8507 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8508 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8509 | return 0; | |
8510 | ||
8511 | return 1; | |
8512 | } | |
8513 | ||
6d6bc0ad | 8514 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8515 | static int sched_rt_global_constraints(void) |
8516 | { | |
ac086bc2 PZ |
8517 | unsigned long flags; |
8518 | int i; | |
8519 | ||
ec5d4989 HS |
8520 | if (sysctl_sched_rt_period <= 0) |
8521 | return -EINVAL; | |
8522 | ||
60aa605d PZ |
8523 | /* |
8524 | * There's always some RT tasks in the root group | |
8525 | * -- migration, kstopmachine etc.. | |
8526 | */ | |
8527 | if (sysctl_sched_rt_runtime == 0) | |
8528 | return -EBUSY; | |
8529 | ||
0986b11b | 8530 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8531 | for_each_possible_cpu(i) { |
8532 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8533 | ||
0986b11b | 8534 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8535 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8536 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8537 | } |
0986b11b | 8538 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8539 | |
d0b27fa7 PZ |
8540 | return 0; |
8541 | } | |
6d6bc0ad | 8542 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8543 | |
8544 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8545 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8546 | loff_t *ppos) |
8547 | { | |
8548 | int ret; | |
8549 | int old_period, old_runtime; | |
8550 | static DEFINE_MUTEX(mutex); | |
8551 | ||
8552 | mutex_lock(&mutex); | |
8553 | old_period = sysctl_sched_rt_period; | |
8554 | old_runtime = sysctl_sched_rt_runtime; | |
8555 | ||
8d65af78 | 8556 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8557 | |
8558 | if (!ret && write) { | |
8559 | ret = sched_rt_global_constraints(); | |
8560 | if (ret) { | |
8561 | sysctl_sched_rt_period = old_period; | |
8562 | sysctl_sched_rt_runtime = old_runtime; | |
8563 | } else { | |
8564 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8565 | def_rt_bandwidth.rt_period = | |
8566 | ns_to_ktime(global_rt_period()); | |
8567 | } | |
8568 | } | |
8569 | mutex_unlock(&mutex); | |
8570 | ||
8571 | return ret; | |
8572 | } | |
68318b8e | 8573 | |
052f1dc7 | 8574 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8575 | |
8576 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8577 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8578 | { |
2b01dfe3 PM |
8579 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8580 | struct task_group, css); | |
68318b8e SV |
8581 | } |
8582 | ||
8583 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8584 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8585 | { |
ec7dc8ac | 8586 | struct task_group *tg, *parent; |
68318b8e | 8587 | |
2b01dfe3 | 8588 | if (!cgrp->parent) { |
68318b8e | 8589 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
8590 | return &init_task_group.css; |
8591 | } | |
8592 | ||
ec7dc8ac DG |
8593 | parent = cgroup_tg(cgrp->parent); |
8594 | tg = sched_create_group(parent); | |
68318b8e SV |
8595 | if (IS_ERR(tg)) |
8596 | return ERR_PTR(-ENOMEM); | |
8597 | ||
68318b8e SV |
8598 | return &tg->css; |
8599 | } | |
8600 | ||
41a2d6cf IM |
8601 | static void |
8602 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8603 | { |
2b01dfe3 | 8604 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8605 | |
8606 | sched_destroy_group(tg); | |
8607 | } | |
8608 | ||
41a2d6cf | 8609 | static int |
be367d09 | 8610 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8611 | { |
b68aa230 | 8612 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8613 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8614 | return -EINVAL; |
8615 | #else | |
68318b8e SV |
8616 | /* We don't support RT-tasks being in separate groups */ |
8617 | if (tsk->sched_class != &fair_sched_class) | |
8618 | return -EINVAL; | |
b68aa230 | 8619 | #endif |
be367d09 BB |
8620 | return 0; |
8621 | } | |
68318b8e | 8622 | |
be367d09 BB |
8623 | static int |
8624 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8625 | struct task_struct *tsk, bool threadgroup) | |
8626 | { | |
8627 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8628 | if (retval) | |
8629 | return retval; | |
8630 | if (threadgroup) { | |
8631 | struct task_struct *c; | |
8632 | rcu_read_lock(); | |
8633 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8634 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8635 | if (retval) { | |
8636 | rcu_read_unlock(); | |
8637 | return retval; | |
8638 | } | |
8639 | } | |
8640 | rcu_read_unlock(); | |
8641 | } | |
68318b8e SV |
8642 | return 0; |
8643 | } | |
8644 | ||
8645 | static void | |
2b01dfe3 | 8646 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8647 | struct cgroup *old_cont, struct task_struct *tsk, |
8648 | bool threadgroup) | |
68318b8e SV |
8649 | { |
8650 | sched_move_task(tsk); | |
be367d09 BB |
8651 | if (threadgroup) { |
8652 | struct task_struct *c; | |
8653 | rcu_read_lock(); | |
8654 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8655 | sched_move_task(c); | |
8656 | } | |
8657 | rcu_read_unlock(); | |
8658 | } | |
68318b8e SV |
8659 | } |
8660 | ||
052f1dc7 | 8661 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8662 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8663 | u64 shareval) |
68318b8e | 8664 | { |
2b01dfe3 | 8665 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8666 | } |
8667 | ||
f4c753b7 | 8668 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8669 | { |
2b01dfe3 | 8670 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8671 | |
8672 | return (u64) tg->shares; | |
8673 | } | |
6d6bc0ad | 8674 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8675 | |
052f1dc7 | 8676 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8677 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8678 | s64 val) |
6f505b16 | 8679 | { |
06ecb27c | 8680 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8681 | } |
8682 | ||
06ecb27c | 8683 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8684 | { |
06ecb27c | 8685 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8686 | } |
d0b27fa7 PZ |
8687 | |
8688 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8689 | u64 rt_period_us) | |
8690 | { | |
8691 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8692 | } | |
8693 | ||
8694 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8695 | { | |
8696 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8697 | } | |
6d6bc0ad | 8698 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8699 | |
fe5c7cc2 | 8700 | static struct cftype cpu_files[] = { |
052f1dc7 | 8701 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8702 | { |
8703 | .name = "shares", | |
f4c753b7 PM |
8704 | .read_u64 = cpu_shares_read_u64, |
8705 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8706 | }, |
052f1dc7 PZ |
8707 | #endif |
8708 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8709 | { |
9f0c1e56 | 8710 | .name = "rt_runtime_us", |
06ecb27c PM |
8711 | .read_s64 = cpu_rt_runtime_read, |
8712 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8713 | }, |
d0b27fa7 PZ |
8714 | { |
8715 | .name = "rt_period_us", | |
f4c753b7 PM |
8716 | .read_u64 = cpu_rt_period_read_uint, |
8717 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8718 | }, |
052f1dc7 | 8719 | #endif |
68318b8e SV |
8720 | }; |
8721 | ||
8722 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8723 | { | |
fe5c7cc2 | 8724 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8725 | } |
8726 | ||
8727 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8728 | .name = "cpu", |
8729 | .create = cpu_cgroup_create, | |
8730 | .destroy = cpu_cgroup_destroy, | |
8731 | .can_attach = cpu_cgroup_can_attach, | |
8732 | .attach = cpu_cgroup_attach, | |
8733 | .populate = cpu_cgroup_populate, | |
8734 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8735 | .early_init = 1, |
8736 | }; | |
8737 | ||
052f1dc7 | 8738 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8739 | |
8740 | #ifdef CONFIG_CGROUP_CPUACCT | |
8741 | ||
8742 | /* | |
8743 | * CPU accounting code for task groups. | |
8744 | * | |
8745 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8746 | * (balbir@in.ibm.com). | |
8747 | */ | |
8748 | ||
934352f2 | 8749 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8750 | struct cpuacct { |
8751 | struct cgroup_subsys_state css; | |
8752 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8753 | u64 __percpu *cpuusage; |
ef12fefa | 8754 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8755 | struct cpuacct *parent; |
d842de87 SV |
8756 | }; |
8757 | ||
8758 | struct cgroup_subsys cpuacct_subsys; | |
8759 | ||
8760 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8761 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8762 | { |
32cd756a | 8763 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8764 | struct cpuacct, css); |
8765 | } | |
8766 | ||
8767 | /* return cpu accounting group to which this task belongs */ | |
8768 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8769 | { | |
8770 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8771 | struct cpuacct, css); | |
8772 | } | |
8773 | ||
8774 | /* create a new cpu accounting group */ | |
8775 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8776 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8777 | { |
8778 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8779 | int i; |
d842de87 SV |
8780 | |
8781 | if (!ca) | |
ef12fefa | 8782 | goto out; |
d842de87 SV |
8783 | |
8784 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8785 | if (!ca->cpuusage) |
8786 | goto out_free_ca; | |
8787 | ||
8788 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8789 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8790 | goto out_free_counters; | |
d842de87 | 8791 | |
934352f2 BR |
8792 | if (cgrp->parent) |
8793 | ca->parent = cgroup_ca(cgrp->parent); | |
8794 | ||
d842de87 | 8795 | return &ca->css; |
ef12fefa BR |
8796 | |
8797 | out_free_counters: | |
8798 | while (--i >= 0) | |
8799 | percpu_counter_destroy(&ca->cpustat[i]); | |
8800 | free_percpu(ca->cpuusage); | |
8801 | out_free_ca: | |
8802 | kfree(ca); | |
8803 | out: | |
8804 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8805 | } |
8806 | ||
8807 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8808 | static void |
32cd756a | 8809 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8810 | { |
32cd756a | 8811 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8812 | int i; |
d842de87 | 8813 | |
ef12fefa BR |
8814 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8815 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8816 | free_percpu(ca->cpuusage); |
8817 | kfree(ca); | |
8818 | } | |
8819 | ||
720f5498 KC |
8820 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8821 | { | |
b36128c8 | 8822 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8823 | u64 data; |
8824 | ||
8825 | #ifndef CONFIG_64BIT | |
8826 | /* | |
8827 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8828 | */ | |
05fa785c | 8829 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8830 | data = *cpuusage; |
05fa785c | 8831 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8832 | #else |
8833 | data = *cpuusage; | |
8834 | #endif | |
8835 | ||
8836 | return data; | |
8837 | } | |
8838 | ||
8839 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8840 | { | |
b36128c8 | 8841 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8842 | |
8843 | #ifndef CONFIG_64BIT | |
8844 | /* | |
8845 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8846 | */ | |
05fa785c | 8847 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8848 | *cpuusage = val; |
05fa785c | 8849 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8850 | #else |
8851 | *cpuusage = val; | |
8852 | #endif | |
8853 | } | |
8854 | ||
d842de87 | 8855 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8856 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8857 | { |
32cd756a | 8858 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8859 | u64 totalcpuusage = 0; |
8860 | int i; | |
8861 | ||
720f5498 KC |
8862 | for_each_present_cpu(i) |
8863 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8864 | |
8865 | return totalcpuusage; | |
8866 | } | |
8867 | ||
0297b803 DG |
8868 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8869 | u64 reset) | |
8870 | { | |
8871 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8872 | int err = 0; | |
8873 | int i; | |
8874 | ||
8875 | if (reset) { | |
8876 | err = -EINVAL; | |
8877 | goto out; | |
8878 | } | |
8879 | ||
720f5498 KC |
8880 | for_each_present_cpu(i) |
8881 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8882 | |
0297b803 DG |
8883 | out: |
8884 | return err; | |
8885 | } | |
8886 | ||
e9515c3c KC |
8887 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8888 | struct seq_file *m) | |
8889 | { | |
8890 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8891 | u64 percpu; | |
8892 | int i; | |
8893 | ||
8894 | for_each_present_cpu(i) { | |
8895 | percpu = cpuacct_cpuusage_read(ca, i); | |
8896 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8897 | } | |
8898 | seq_printf(m, "\n"); | |
8899 | return 0; | |
8900 | } | |
8901 | ||
ef12fefa BR |
8902 | static const char *cpuacct_stat_desc[] = { |
8903 | [CPUACCT_STAT_USER] = "user", | |
8904 | [CPUACCT_STAT_SYSTEM] = "system", | |
8905 | }; | |
8906 | ||
8907 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
8908 | struct cgroup_map_cb *cb) | |
8909 | { | |
8910 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8911 | int i; | |
8912 | ||
8913 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
8914 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
8915 | val = cputime64_to_clock_t(val); | |
8916 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
8917 | } | |
8918 | return 0; | |
8919 | } | |
8920 | ||
d842de87 SV |
8921 | static struct cftype files[] = { |
8922 | { | |
8923 | .name = "usage", | |
f4c753b7 PM |
8924 | .read_u64 = cpuusage_read, |
8925 | .write_u64 = cpuusage_write, | |
d842de87 | 8926 | }, |
e9515c3c KC |
8927 | { |
8928 | .name = "usage_percpu", | |
8929 | .read_seq_string = cpuacct_percpu_seq_read, | |
8930 | }, | |
ef12fefa BR |
8931 | { |
8932 | .name = "stat", | |
8933 | .read_map = cpuacct_stats_show, | |
8934 | }, | |
d842de87 SV |
8935 | }; |
8936 | ||
32cd756a | 8937 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8938 | { |
32cd756a | 8939 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8940 | } |
8941 | ||
8942 | /* | |
8943 | * charge this task's execution time to its accounting group. | |
8944 | * | |
8945 | * called with rq->lock held. | |
8946 | */ | |
8947 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8948 | { | |
8949 | struct cpuacct *ca; | |
934352f2 | 8950 | int cpu; |
d842de87 | 8951 | |
c40c6f85 | 8952 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8953 | return; |
8954 | ||
934352f2 | 8955 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8956 | |
8957 | rcu_read_lock(); | |
8958 | ||
d842de87 | 8959 | ca = task_ca(tsk); |
d842de87 | 8960 | |
934352f2 | 8961 | for (; ca; ca = ca->parent) { |
b36128c8 | 8962 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8963 | *cpuusage += cputime; |
8964 | } | |
a18b83b7 BR |
8965 | |
8966 | rcu_read_unlock(); | |
d842de87 SV |
8967 | } |
8968 | ||
fa535a77 AB |
8969 | /* |
8970 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
8971 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
8972 | * percpu_counter_add with values large enough to always overflow the | |
8973 | * per cpu batch limit causing bad SMP scalability. | |
8974 | * | |
8975 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
8976 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
8977 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
8978 | */ | |
8979 | #ifdef CONFIG_SMP | |
8980 | #define CPUACCT_BATCH \ | |
8981 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
8982 | #else | |
8983 | #define CPUACCT_BATCH 0 | |
8984 | #endif | |
8985 | ||
ef12fefa BR |
8986 | /* |
8987 | * Charge the system/user time to the task's accounting group. | |
8988 | */ | |
8989 | static void cpuacct_update_stats(struct task_struct *tsk, | |
8990 | enum cpuacct_stat_index idx, cputime_t val) | |
8991 | { | |
8992 | struct cpuacct *ca; | |
fa535a77 | 8993 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
8994 | |
8995 | if (unlikely(!cpuacct_subsys.active)) | |
8996 | return; | |
8997 | ||
8998 | rcu_read_lock(); | |
8999 | ca = task_ca(tsk); | |
9000 | ||
9001 | do { | |
fa535a77 | 9002 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9003 | ca = ca->parent; |
9004 | } while (ca); | |
9005 | rcu_read_unlock(); | |
9006 | } | |
9007 | ||
d842de87 SV |
9008 | struct cgroup_subsys cpuacct_subsys = { |
9009 | .name = "cpuacct", | |
9010 | .create = cpuacct_create, | |
9011 | .destroy = cpuacct_destroy, | |
9012 | .populate = cpuacct_populate, | |
9013 | .subsys_id = cpuacct_subsys_id, | |
9014 | }; | |
9015 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
9016 | |
9017 | #ifndef CONFIG_SMP | |
9018 | ||
9019 | int rcu_expedited_torture_stats(char *page) | |
9020 | { | |
9021 | return 0; | |
9022 | } | |
9023 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
9024 | ||
9025 | void synchronize_sched_expedited(void) | |
9026 | { | |
9027 | } | |
9028 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9029 | ||
9030 | #else /* #ifndef CONFIG_SMP */ | |
9031 | ||
9032 | static DEFINE_PER_CPU(struct migration_req, rcu_migration_req); | |
9033 | static DEFINE_MUTEX(rcu_sched_expedited_mutex); | |
9034 | ||
9035 | #define RCU_EXPEDITED_STATE_POST -2 | |
9036 | #define RCU_EXPEDITED_STATE_IDLE -1 | |
9037 | ||
9038 | static int rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
9039 | ||
9040 | int rcu_expedited_torture_stats(char *page) | |
9041 | { | |
9042 | int cnt = 0; | |
9043 | int cpu; | |
9044 | ||
9045 | cnt += sprintf(&page[cnt], "state: %d /", rcu_expedited_state); | |
9046 | for_each_online_cpu(cpu) { | |
9047 | cnt += sprintf(&page[cnt], " %d:%d", | |
9048 | cpu, per_cpu(rcu_migration_req, cpu).dest_cpu); | |
9049 | } | |
9050 | cnt += sprintf(&page[cnt], "\n"); | |
9051 | return cnt; | |
9052 | } | |
9053 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
9054 | ||
9055 | static long synchronize_sched_expedited_count; | |
9056 | ||
9057 | /* | |
9058 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
9059 | * approach to force grace period to end quickly. This consumes | |
9060 | * significant time on all CPUs, and is thus not recommended for | |
9061 | * any sort of common-case code. | |
9062 | * | |
9063 | * Note that it is illegal to call this function while holding any | |
9064 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
9065 | * observe this restriction will result in deadlock. | |
9066 | */ | |
9067 | void synchronize_sched_expedited(void) | |
9068 | { | |
9069 | int cpu; | |
9070 | unsigned long flags; | |
9071 | bool need_full_sync = 0; | |
9072 | struct rq *rq; | |
9073 | struct migration_req *req; | |
9074 | long snap; | |
9075 | int trycount = 0; | |
9076 | ||
9077 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
9078 | snap = ACCESS_ONCE(synchronize_sched_expedited_count) + 1; | |
9079 | get_online_cpus(); | |
9080 | while (!mutex_trylock(&rcu_sched_expedited_mutex)) { | |
9081 | put_online_cpus(); | |
9082 | if (trycount++ < 10) | |
9083 | udelay(trycount * num_online_cpus()); | |
9084 | else { | |
9085 | synchronize_sched(); | |
9086 | return; | |
9087 | } | |
9088 | if (ACCESS_ONCE(synchronize_sched_expedited_count) - snap > 0) { | |
9089 | smp_mb(); /* ensure test happens before caller kfree */ | |
9090 | return; | |
9091 | } | |
9092 | get_online_cpus(); | |
9093 | } | |
9094 | rcu_expedited_state = RCU_EXPEDITED_STATE_POST; | |
9095 | for_each_online_cpu(cpu) { | |
9096 | rq = cpu_rq(cpu); | |
9097 | req = &per_cpu(rcu_migration_req, cpu); | |
9098 | init_completion(&req->done); | |
9099 | req->task = NULL; | |
9100 | req->dest_cpu = RCU_MIGRATION_NEED_QS; | |
05fa785c | 9101 | raw_spin_lock_irqsave(&rq->lock, flags); |
03b042bf | 9102 | list_add(&req->list, &rq->migration_queue); |
05fa785c | 9103 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
03b042bf PM |
9104 | wake_up_process(rq->migration_thread); |
9105 | } | |
9106 | for_each_online_cpu(cpu) { | |
9107 | rcu_expedited_state = cpu; | |
9108 | req = &per_cpu(rcu_migration_req, cpu); | |
9109 | rq = cpu_rq(cpu); | |
9110 | wait_for_completion(&req->done); | |
05fa785c | 9111 | raw_spin_lock_irqsave(&rq->lock, flags); |
03b042bf PM |
9112 | if (unlikely(req->dest_cpu == RCU_MIGRATION_MUST_SYNC)) |
9113 | need_full_sync = 1; | |
9114 | req->dest_cpu = RCU_MIGRATION_IDLE; | |
05fa785c | 9115 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
03b042bf PM |
9116 | } |
9117 | rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
956539b7 | 9118 | synchronize_sched_expedited_count++; |
03b042bf PM |
9119 | mutex_unlock(&rcu_sched_expedited_mutex); |
9120 | put_online_cpus(); | |
9121 | if (need_full_sync) | |
9122 | synchronize_sched(); | |
9123 | } | |
9124 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9125 | ||
9126 | #endif /* #else #ifndef CONFIG_SMP */ |